Anderson, Lightfoot 2002 The Language Organ

The Language Organ
Linguistics as Cognitive Physiology
Challenging and original, The Language Organ treats human language as the
manifestation of a faculty of the mind, a mental organ whose nature is deter-
mined by human biology. Its functional properties should be explored just
as physiology explores the functional properties of physical organs. The
authors argue that linguistics investigates cognition, taking as its object mental
representations and processes rather than externally observed grammatical pat-
terns (which constitute evidence, but are not themselves the object of study).
Such a view has untraditional consequences for the kinds of argument and
analysis that can be offered in understanding the nature of language. The book
surveys the nature of the language faculty in its various aspects: the systems
of sounds, words, and syntax; the development of language in the child and
historically; and what is known about its relation to the brain. It discusses the
kinds of work that can be carried out in these areas that will contribute to an
understanding of the human language organ. The book will appeal to students
and researchers in linguistics, and is written to be accessible to colleagues in
other disciplines dealing with language as well as to readers with an interest
in general science and the nature of the human mind.
stephen r. anderson is Professor of Linguistics and Cognitive Science

at Yale University. He is the author of The Organization of Phonology (1974),
Phonology in the Twentieth Century (1985), and A-Morphous Morphology
(1992).
david w. lightfoot is Dean of the Graduate School, Georgetown

University. He is the author of eight books, most recently The Development
of Language (1999).
The Language Organ
Linguistics as Cognitive Physiology
Stephen R. Anderson
Yale University
David W. Lightfoot
Georgetown University
         
The Pitt Building, Trumpington Street, Cambridge, United Kingdom
  

The Edinburgh Building, Cambridge CB2 2RU, UK
40 West 20th Street, New York, NY 10011-4211, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
Ruiz de Alarcón 13, 28014 Madrid, Spain
Dock House, The Waterfront, Cape Town 8001, South Africa
http://www.cambridge.org
© Stephen R. Anderson and David W. Lightfoot 2004
First published in printed format 2002
ISBN 0-511-04210-8 eBook (netLibrary)

ISBN 0-521-80994-0 hardback
ISBN 0-521-00783-6 paperback
Dedicated to the memory of
Victoria Fromkin,
who was way ahead of us in
the investigation of language
as an aspect of the mind/brain
[H]uman cognitive systems, when seriously investigated, prove to be no less
marvelous and intricate than the physical structures that develop in the life of
the organism. Why, then, should we not study the acquisition of a cognitive
structure such as language more or less as we study some complex bodily
organ?
Noam Chomsky 1975, p. 10
Contents
Preface page ix
1 Studying the human language faculty 1

1.1 Linguistics and the mind/brain 3
1.2 Linguistics as history 5
1.3 Linguistics as the study of E-language 12
1.4 Linguistics as the study of I-language 14
2 Language as a mental organ 18

2.1 We know more than we learn 18
2.2 The nature of grammars 23
2.3 Back to the puzzles 25
2.4 The acquisition problem 34
2.5 Conclusion 40
3 Syntax 41
3.1 The emergence of syntax within linguistics 41
3.2 Successive merger and deletion 43
3.3 Case 61
3.4 Conclusion 66
4 Sound patterns in language 67

4.1 Phonetics as theory 68
4.2 Phonology: language-particular structure 73
4.3 Morphophonemics and I-language 83
5 Describing linguistic knowledge 92

5.1 Phonological knowledge as it appears in borrowing 93
5.2 Can rules express phonological knowledge? 96
5.3 Constraint-based theories of phonological knowledge 99
5.4 The extension of constraint-based description 109
6 Phonetics and the I-linguistics of speech 111

6.1 Representations and the study of sound structure 111
6.2 A linguistic basis for phonetic representation 114
6.3 Speech microprosody: a research program 127
6.4 Conclusion 129
vii
viii Contents
7 Morphology 131
7.1 The lexicon 132
7.2 Words and “morphemes” 134
7.3 Productivity 152
7.4 Conclusions about lexical organization 155
8 Language change 157

8.1 Long-term directionality in the twentieth century 157
8.2 Grammars and time 161
8.3 English auxiliary verbs 162
8.4 Syntactic effects of the loss of case 175
8.5 Chaos 183
9 “Growing” a language 186

9.1 Principles of Universal Grammar: active early 186
9.2 New phenomena 191
9.3 Experimental technique 194
9.4 Nature of the trigger 198
9.5 Acquiring sound patterns 206
9.6 Conclusion 214
10 The organic basis of language 216

10.1 Only humans have language organs 219
10.2 Language is a function of the brain 221
10.3 Language is a particular faculty 236
10.4 Conclusions 238
References 244
Index 257
Preface
One of the great success stories of post-Second-World-War intellectual inquiry

has been the extent to which linguists have been able to make the syntactic
and phonological structure of natural language into a serious object of explicit
formal study. This work has uncovered principles of surprising subtlety, ab-
stractness, and deductive richness; it has also raised fundamental questions
concerning the ontogenetic and phylogenetic developments by which know-
ledge of this kind could develop in the organism. Much of this progress results
fairly directly from the adoption of an explicitly biological perspective on the
subject: instead of seeing language as an external phenomenon, as a collection
of sounds, words, texts, etc. that exists apart from any particular individual,
contemporary linguistics increasingly concerns itself with the internal organi-
zation and ontogeny of a special kind of knowledge. The specific form that this
aspect of human cognition takes appears, we will argue, to be a species-specific
property of human beings, and thus rooted in our biological nature.
As our subtitle promises, we will describe linguistics, the scientific study
of (human natural) language, as cognitive physiology. An individual’s use of
language involves that person’s brain: the way this brain works depends at
least in part on childhood influences and whether the person was raised in New
Haven, New Delhi, or New Guinea. The relevant aspect of the brain’s structure
that responds to these differences in experience is the person’s language organ,
but to characterize it we need to take seriously the notion of physiology as the
study of functions.
Webster’s Second International Dictionary tells us that physiology is
[t]he branch of biology dealing with the processes, activities, and phenomena incidental
to and characteristic of life or of living organisms; the study of the functions of the
organs, tissues, cells, etc. during life, as distinct from anatomy, which deals with their
structure. The final analysis of these processes and phenomena is mainly physical and
chemical. The phenomena of mental life are usually regarded as outside the ordinary
scope of physiology (see p s y c h o l o g y).
Like many dictionary definitions, this one combines the central conceptual
content of the word with other characteristics that are as much the product of
ix
x Preface
historical accident as of basic meaning. The core notion seems to be that of the
study of functions and processes as opposed to static structure, but that by itself
does not tell us much about just what sorts of object might properly fall within
the scope of this science.
A central aim of this book is to explore the notion that physiology should
extend, if it is to be productive, beyond the study of isolable physical structures
and organs. As we see it, the range of “phenomena . . . characteristic of life”
whose functional organization can be usefully attributed to the biology of the
organism exhibiting them, and substantively studied, goes well beyond those
for which we have a “final analysis” which is “mainly physical and chemical.”
Excluding “the phenomena of mental life” from physiology surely reflects, at
least in part, the ancient Cartesian dualism of mind and body in rejecting the
study of cognition from the biological domain. However, this dualism was soon
undermined, largely because of what we came to learn about the “body” not long
after Descartes. Newton reintroduced “occult” causes and qualities, to his own
dismay, and the Cartesian theory of body was shown to be untenable. Hence,
there is no absolute distinction between body and mind, because our notions
of the body need to be enriched beyond the simple mechanisms envisioned
by Descartes.
Accepting the notion that “the mind is what the brain does,” as much mod-
ern cognitive science does, entails that cognitive phenomena essential to the
life of an organism are good candidates for biological study: as with other or-
gans, we can study them at an appropriate functional level and not just in the
narrow terms of a physical or chemical analysis of living tissue. There is no
longer any motivation for a sharp delineation between functional organization
that is associated with discrete anatomy and that whose relation to physical
implementation is less obvious and discrete. In all cases, we need to identify
an appropriate level of abstraction at which significant generalizations about
biologically determined function can be stated. We shall illustrate such a level
of abstraction: the study of “I-language.”
Does it make sense to take linguistics as a paradigm for this sort of analysis,
though? In modern academic life, linguistics as a discipline suffers from a
lack of clear general perception. All too commonly, the notion that there is
anything very complex or technical about the study of language seems contrary
to common sense: after all, just about everyone (even little kids) can speak
a language perfectly well, so what could be that intricate about it? Of course,
when you learn a new language you have to learn a lot of words, and little details
like where the verb goes in a German sentence, but surely that’s not science. If
there is something to study there, this attitude continues, it must be things about
the histories of languages (Which ones are related to which others? Where did
the word boondocks come from?), or else about the way we use language in
society (Does bilingual education make sense? How do men and women do
different things with the same words?).
Preface xi
As a result, linguistics departments sometimes have a bit of a hard time mak-

ing ends meet. Often enough, when universities feel the need to save some
money, linguistics is one of the first departments to feel the impact, since
surely what people do there is either trivial or else could be perfectly well
done elsewhere (in the language departments, or in Anthropology, Sociology,
Psychology, etc. – “real” disciplines).
Even the location of the linguistics department is often problematic. Some
places have linguistics in the humanities, some in the social sciences; but the
things linguists do on a day-to-day basis seem ill suited, at least in many cases,
to either of these labels. Our own personal inclination is to locate the core
focus of linguistics in the biological (or at least natural) sciences, along with the
rest of cognitive science, as is indeed the case in a few enlightened institutions.
Of course, linguistics is construed in different ways by different scholars; the
particular view we take in this book would surely not be shared by all faculty
in all linguistics departments.
We will argue (in chapter 1 below) that this disciplinary ambiguity actually
reflects the historical development of our understanding of what is essential in
the study of language. The first (non-speculative) studies of language in the
nineteenth century really were focused, at least originally, on the detailed study
of words and texts, their histories, and their importance for an understanding of
the history of individual languages and language families. This kind of activity
falls squarely within the domain of the humanities, and much historical work
on language today continues to fit that characterization.
Around the turn of the twentieth century, however, the ideas of Ferdinand
de Saussure caused students of language to pay much more attention to the
synchronic realities of language, and less to its history. An initial attempt to
understand the nature of this object took language to be fundamentally a set
of social conventions for conveying meaning, and its reality comparable to
other aspects of society and culture and thus transcending any individual to
reside in the entire community of language users. The modern discipline of
sociolinguistics continues to study the relation between language and society,
in ways including the correlation of social and linguistic differences, the relation
between discourse structure and social goals, the formulation of rational national
language policies, and many other things.
In the middle years of the twentieth century the externalization of language
through its construal as a social object, entailed by Saussure’s views, was re-
inforced by behaviorist ideas that treated all study of strictly internal, mental
phenomena as unscientific at best, even nonsensical. Surely some of that at-
titude lingers on in the definition provided by Webster’s Second above, along
with simple Cartesian dualism. At any rate, the activity of linguists at this time
was largely devoted to the development of techniques for reducing the data of
unfamiliar or “exotic” languages to manageable form, an effort that continues
to be of great interest to anthropologists among others.
xii Preface
With the work of Noam Chomsky, and his colleagues and students, however,
a completely new focus arose. Chomsky stressed from an early point that if
we want to understand the essential nature of language, a study of its external
manifestations in speech, texts, communicative behavior, etc. may provide rel-
evant evidence, but that should not be the fundamental object of inquiry in the
field. What we want to understand, that is, is not a text or a spoken utterance,
but the system of knowledge within a speaker that underlies (significant and
well-defined aspects of) it. Linguistics is thus in its most essential aspect a
component of the study of the mind, the study of a system of human know-
ledge (the language organ) that forms a part of our larger cognitive organization.
It has developed a level of abstraction that appears to be appropriate for the
formulation of significant generalizations about language: what we will call
“I-language” below, an individual, internal system. We know something of the
properties of I-language, though certainly not all, and an appropriate level of
abstraction seems to be emerging which permits productive research. Efforts
toward detailing the properties of I-language in a variety of areas constitute the
subject matter of this book.
As modern cognitive science has come increasingly to reject the dualism of
mind and body, and to see our mental life as the product of physical processes
and events, it has become possible at least to consider the possibility that aspects
of cognition have a structure in themselves that is determined as much in its
form by our biological nature as is the structure and functioning of physical
organs like the liver or the skin. And when we look into the matter with care,
we find that there are indeed many strong reasons to consider that the cognitive
organization underlying our ability to acquire and use a language is as much
a part of our genetically determined biology as the structure of our eyes or
skeleton.
This might lead us to expect that the right way to study language might be
to identify and characterize discrete, separable regions of the brain that are
responsible for it, and then to explore the operation of this tissue in precise
detail. It would surely be at least premature to take this approach, however.
We are far from being able to associate specific neural tissue, except at the
very grossest levels, with specific cognitive functions. We can observe that
injuries to certain brain regions result in particular cognitive deficits, and that
shows that the tissue involved must indeed be serving some purpose that is
essential to the cognitive function in question, but that is not at all the same
thing as saying that this is the specific region which “computes” the relevant
cognitive property. Even when rapidly developing imaging techniques allow
us to identify metabolic activity in specific regions of the (normal) brain that
is associated with language-related activities, we are still far from being able
to relate those activities in any direct way to molecular or cellular events and
processes.
Preface xiii
In truth, our understanding of the overall architecture of cognition is still very

sketchy, imprecise, and surely inaccurate. Before we can even pose serious
questions about the specific physical underpinnings of cognitive activity, we
need to know much more about just what that activity consists in. We can
provide inferential, limiting arguments that show beyond a doubt that there is a
large biologically determined component to cognitive abilities such as language,
but that is a long way from showing exactly how neurophysiological activity in
specific regions of the brain (and elsewhere) corresponds to the production and
understanding of language.
While this conclusion may well imply that it will be some time before lin-
guistics is moved definitively out of humanities and social sciences and into the
biology building, we are encouraged to pursue the prerequisites to such an un-
derstanding. In particular, we need to understand the “processes, activities, and
phenomena” characteristic of our mental life, the organization, development,
and interaction of cognition. Of course, to do that, we have to look at the right
things; and in the case of language, that means studying language as a highly
structured kind of knowledge rather than as a collection of external events. And
we submit that this study of the organization and functioning of the human
language capacity is every bit as “physiological” as the study of the processes
that take place in the kidney, even though the “organ” involved is defined (for
the present, at least) in functional rather than anatomical terms.
The study of language over the past half century or so has provided cognitive
scientists with models for how one might go about studying cognition more
generally. In this, linguists have been distinctly lucky in comparison with other
students of the mind, precisely because our knowledge of language has so many
evident and observable consequences, and is so much more open to study than
other domains. Historically, linguistics has provided a model for other fields
on more than one occasion: witness the basic importance of linguistic notions
in the development of “structuralist” analysis quite broadly in the social and
human sciences. More recently, linguistics has also provided a prototype for
the study of cognitive systems in general (even if the ultimate application of
this analogy should prove to be an understanding of just how different cognitive
systems, including language, can be from one another), as well as a focus for
the application of valuable techniques such as the cognitive neuropsychological
exploration of (normal) cognitive architecture on the basis of functional deficits
consequent to brain damage.
Linguists in fact have quite a bit to contribute to the general study of cognition.
In the study of action and motor control, for example, the results of phoneticians
are potentially an incredibly rich source of evidence. There is probably no
single area in which we have a finer grained understanding of the details of
muscular activity and movement in time and in space than in the case of speech
articulation; or in which such low-level detail can be put into correspondence
xiv Preface
with so specific an understanding of higher-level abstract representations of

the intentionality underlying the movements in question. Yet it is only quite
recently that work in phonetics and in the general study of action has begun to
converge and “cross-pollinate.”
A central problem in understanding any cognitive system is that of seeing
how its structure can develop within the individual organism. In this study, the
language organ offers quite unparalleled advantages. We can detail with great
precision the evidence available in the course of language acquisition; we have
a reasonably clear notion of the character of the system which is eventually at-
tained, and we can formulate precise hypotheses about the intermediate stages
through which the organism passes on the way to this knowledge. As such, we
can study the interplay between considerations particular to a single cognitive
system (generally, the content of “Universal Grammar”) and those that gener-
alize across cognition in general. The study of the development of linguistic
competence in the child is an enormously important exemplar for more general
studies of cognitive development.
Linguists have not always gone to enough trouble to inform a broader scien-
tific public about the implications of our work. Surely the huge expansion
of the field within academia during the 1960s and 1970s was largely due
to a perception that this discipline did potentially offer a privileged kind of
“window on the mind,” but in actual practice linguists have tended to neglect
the obligation to make good on the promissory notes thus issued. The technical
details of the field are indeed fascinating to pursue, as well as specialized, com-
plex, and fairly opaque to outsiders. When linguists pursue these technicalities
(in the same way that physicists pursue the details of physics without feeling
a need continually to inform the outside world about the importance of their
discipline), the result is a continuing lack of understanding on the part of the
general public of just what the science of language is all about, and why it
matters.
This book has several aims, then. One is to establish for the non-specialist the
biological nature of the human language faculty, and to give an account of the
organization of that faculty to the extent we can discern it. Another is to give
these same non-specialist readers a sense of the complex interplay between
the empirical phenomena of language and the basic questions of cognitive
organization we would like to be able to ask.
We have writen primarily for our colleagues in anthropology, history,
German, biology, physiology, and a variety of other fields in which they ap-
proach language from their own disciplinary perspective and on the basis of
their own expertise. They sometimes ask us in the Faculty Club about how lin-
guists view things: this book is an extended answer to those questions, among
others. Scholars with different backgrounds – physiologists and historians, for
example – will disagree on which parts of the book are easy and which are
Preface xv
harder reading. Like restaurateurs, we aim to cater to a range of different

tastes within an overall framework.
In the process, we introduce a good deal of the specialized apparatus of
modern linguistics, which displays an intricacy that may surprise some. Indeed,
one of our fondest memories is of the time when one of us, at a rather boring
meeting, was sitting next to Dr. Rita Colwell, the distinguished biologist who
would later become director of the National Science Foundation. Dr. Colwell
idly picked up a book on morphosyntactic theory from her neighbor, leafing
slowly through it with apparently mounting interest. “Well !” she eventually
remarked, handing the book back. “You really do have to be a rocket scientist
to be a linguist !” Certainly no more so than to be a microbiologist, but perhaps
more than is generally appreciated.
For our linguist colleagues, we also have some important goals. Apart from
encouraging them to make the more general implications of their work known
beyond our own small community, we want to stress the significance of see-
ing language as an internal fact about speakers, a form of knowledge, or
“I-language” as Chomsky has called it, as opposed to its external manifestation
as utterances, texts, sets of sentences, or social conventions – “E-language” in
Chomsky’s terms. While a growing part of the rhetoric of the field adopts a
position similar to ours, actual practice often devotes more significance to these
external phenomena than might be scientifically justified. Beginning graduate
students need to understand this, and some of our tables are set for them.
We try, in parts of this book, to explore some of the intellectual history of
the discipline which might be responsible for wrong emphases. This historical
orientation serves at least two goals. On the one hand, it allows us to contrast
a contemporary understanding of important issues with earlier alternatives,
plausible in themselves but positions on which the views presented can (we
hope) be seen as improvements. And on the other hand, it is important not
to neglect the role of inertia in science. In fact, many attitudes of scientists
(linguists included) turn out on closer analysis to be simply the unexamined
residue of earlier stages of their science, positions that have survived major
changes of outlook simply because they did not directly engage the central
issues at play in periods of substantial scientific change. Understanding this
fact, together with the true nature of the basic scientific questions that need
to be asked about language, will perhaps help us to arrive at a more coherent
account of this fascinating part of our nature as humans.
There is a fair amount of technical material in this book: details about exotic
languages, formal apparatus for describing parts of linguistic knowledge, etc.
This is partly in service of introducing serious readers to the real nature of
the field, and partly because it is unavoidable if the fundamental points are
to be made in a serious way. Despite advice from many that the inclusion
of many references in the text will put off non-specialist readers, we have
xvi Preface
tried to document our discussion from the standard literature. These references
are of two general sorts: some are intended to make it possible for skeptics,
or for those who simply want to know more about a particular point, to see
where our position is grounded. The reader may in general presume that such
additional exploration is not crucial to understanding the point at issue, and
regard these citations as so much baroque ornamentation. Other references are
provided specifically to point readers to fuller discussions of areas which we
cannot go into here in adequate detail; these are generally signalled as such,
and should again be considered as supplementary rather than required reading
assignments.
We have tried, overall, to make this book as self-contained as possible, defin-
ing our terms at least to an extent that should make our usage comprehensible to
a non-specialist who is willing to make the effort to understand us. The reader
who has read carefully Steven Pinker’s very useful book The Language Instinct
should have all of the tools that are necessary to make it through this one. Some
parts will be slower going than others, however. The reader may perhaps take
comfort in the fact that most chapters can be appreciated in the absence of a
complete understanding of some of the others.
Linguistics is a field with many subareas, and the basic concepts and methods
of each are to some extent distinct. We have not by any means attempted to
cover the entire field (lexical semantics, for example, is not treated at all, nor
is parsing, or computational analysis, etc.), but we have tried to provide a
representative piece from several of the major areas of grammatical theory, and
from some other areas that are of special importance for the kind of questions
we wish to address. In all cases, we see that we need to get beyond the everyday
notions of “languages,” “words,” “sounds,” and the like – external, extensional
notions from E-language – and to work at a higher level of abstraction in order
to capture significant generalizations.
Chapter 1 summarizes the way ideas have changed over time concerning the
basic object of inquiry in linguistics. Philologists thought of texts as the essen-
tial reality; in the nineteenth century, the neogrammarians looked at individual
sounds and words; structuralists assumed that language should be studied in
itself, but thought of structure as somehow immanent in an external, social real-
ity. American structuralists thought they could characterize linguistic behavior
from a purely external point of view, until Chomsky’s review of Skinner and
the subsequent triumph of mentalism over behaviorism in the study of cog-
nition showed us that the real object we ought to be studying is the nature,
development, and organization of linguistic knowledge – the language organ.
Against the background of this understanding of what linguistics is centrally
concerned with, chapter 2 sets the stage for more detailed consideration of the
various subparts of the study of the language organ. We argue that the properties
of language cannot be understood if we take it to develop by the inductive
Preface xvii
application of general learning strategies to the data available to the child, and
present evidence that the language faculty is a species-specific, biologically
determined capacity.
Chapter 3 deals in more detail with the domain where the cognitive nature of
the subject was clearest from the start, syntax. Prior to the late 1950s, linguistics
focused almost entirely on the smallest units of language, sounds, words, and
minimal meaningful elements (“morphemes”), where the model of the Saus-
surian sign has most plausibility. “Syntax” was largely a promissory note to
the effect that such sign-based analysis would eventually encompass the larger
units of phrases, sentences, etc. When the productive mechanisms of syntac-
tic formation came under scrutiny with the rise of transformational generative
grammar, however, the challenge to the notion that language is essentially an
inventory of signs became apparent. The resulting insights had profound effects
in all areas of the field.
Chapter 4 deals with phonology, the study of what Edward Sapir meant by
the title of his paper “Sound Patterns in Language” (Sapir 1925). The study
of the linguistic organization of sound led, in structuralist linguistics, to the
notion of the “phoneme” as a minimal unit of contrast. This idea, in turn, was
seen as fundamental by students of several other disciplines, and for many
constitutes the main claim of linguistics to scientific status. A focus on the
definition of this unit, however, resulted in a resolutely surface-oriented view
of language, which was only with great difficulty replaced by a more internalist
picture. In fact, generative phonology more or less accidentally discovered the
right alternative to the externalism of phonemics. In what is generally seen to
be the foundational moment in the development of a generative approach to
phonology, Morris Halle offered an argument against classical phonemics that
had profound effects on the field. What is of interest here is the fact that the true
force of Halle’s argument against phonemes only becomes apparent when the
object of inquiry in the study of language is taken to be a form of knowledge
(I-language), not the properties of sounds, words, sentences, etc.
The concern of chapter 4 is with the nature of representations in phonol-
ogy, the description of the way languages use sound properties to distinguish
linguistic elements from one another and the changes that have taken place
in our conception of the relation between phonological form and surface pho-
netic form. Chapter 5 continues the discussion of phonology by asking how
the regularities of linguistic knowledge in this domain are to be characterized.
We discuss reasons to believe that the kind of configuration-specific rules in
terms of which early generative phonology operated can profitably be replaced
with systems of more general constraints. We view the rise of theories of this
sort (such as “Optimality Theory”) as being of a piece with the more general
abandonment in syntax and elsewhere of construction-specific rules in favor of
general principles.
xviii Preface
Chapter 6 deals with phonetics, often thought of not as a theoretical aspect of

the study of language, but rather as just a set of tools for objectively observing
and recording external facts. We argue, in contrast, that phonetics as it forms
part of linguistics is necessarily based on the development and articulation of a
highly abstract theory of the organization of speech production and perception –
a theory that identifies and characterizes the linguistic aspects of speech events
to the exclusion of their other physical and physiological correlates. There is no
external, language-independent basis for this abstraction; and indeed some of
the “phonetic” aspects of speech may not be overtly realized in any measurable
way, but must be inferred from linguistic organization. Again, the coherent
nature of this aspect of language only makes sense in terms of the cognitive
construal of language as a form of knowledge, rather than as E-language.
Chapter 7 introduces the field of morphology. A language’s stock of words
is usually seen as merely an inventory of signs – some of which may be in-
ternally complex, and to which additions may be made in regular ways, but
fundamentally a collection. We argue that the nature of word structure necessi-
tates a conception of the lexicon instead as a form of knowledge, of a kind not
reducible to syntax and/or phonology.
Chapters 8 and 9 concern the related subjects of linguistic change and the
acquisition of grammatical knowledge by the child. Language change is argued
not to be driven by the imperatives of external forces acting on linguistic objects,
as the founders of modern historical linguistics (the neogrammarians) saw it,
but rather as a working out of the possibilities made available by the human
language faculty in the presence of limited and often ambiguous data. Similarly,
the acquisition of a first language by the child is better seen as the growth
of a cognitive capacity in a way sensitive to (but not simply determined by)
the partial and degenerate data available in the environment. In both of these
(tightly interconnected) respects, a coherent view of the regularities involved
necessitates a cognitive, rather than an E-language perspective.
In the concluding chapter, we return to the problem of the organic basis
of language as it develops in humans. Abundant evidence shows that this is an
aspect of the organization of the mind/brain, which naturally raises the question
of where in the brain this faculty is to be found. While some partial answers
in the form of apparently concrete loci for certain aspects of the language
capacity have been proposed, it is reasonably clear that the brain’s organization
is subtler and more complex than this. If language is indeed an “organ,” it is one
in a functional sense, not localized in the manner of the kidney. In this respect,
however, language is simply typical of the functional organs that, collectively,
make up the mind.
The original idea for this book arose when Dr. Joseph Hoffmann, editor of
the Annual Review of Physiology, asked us to prepare an article for that journal
that would suggest to physiologists what linguists were talking about when they
Preface xix
referred to a language “organ.” The result was Anderson and Lightfoot 1999,
which serves as the basis for chapter 2 and much of chapter 10 of the present
book. We are grateful to Dr. Hoffmann not only for giving us an opportunity to
discuss this issue for a discerning and sophisticated audience of non-linguists,
but also for making us think about the importance of the basic issues raised by
that question and the need to present them to a broader audience.
We have received useful comments on various drafts of parts of this book
from Dana Boatman, Norbert Hornstein, Ray Jackendoff, and Charles Yang, as
well as from referees for Cambridge University Press. Some of the research on
which chapters below are based was supported by awards (SBR 9514682 and
BCS 9876456) from the National Science Foundation to Yale University.
Figure 6.1, is reprinted from J.-M. Hombert, ‘Consonant types, vowel height
and tone in Yoruba,’ UCLA Working Papers in Phonetics 33 (1976), 40–54.
Figure 10.1, is reprinted from Richard L. Gregory (ed.) The Oxford Compan-
ion to the Mind (Oxford, Oxford University Press: 1987), 620.
Figure 10.2, is reprinted from D. Purves et al. (eds.), Neuroscience
(Sunderland, MA: Sinauer), 159.
1 Studying the human language faculty
If you meet someone at a cocktail party and tell them you are a carpenter, or
a veterinarian, or an astronomer, they are likely to be quite satisfied with that,
and the subsequent evolution of the conversation will depend, at least in part, on
the depth of their interest in woodworking, animals, or the universe. But if you
tell them you are a linguist, this is unlikely to satisfy whatever curiosity they
may have about you : “Oh, so how many languages can you speak?” is the most
common reply at this point. But in fact, many – probably even most – linguists
actually speak few if any languages in addition to their native tongue, in any
practical sense. A “linguist,” at least in academic disciplinary terms, is not a
person who speaks many languages, but rather someone concerned with the
scientific study of language more generally.
That still doesn’t settle matters, though. As we will discuss below, different
generations of scholars have had rather different notions of what was important
enough about language to warrant study. Languages have histories, and relation-
ships with one another that at least superficially parallel genetic connections,
and one can study those things. Most often, languages are spoken, and it is
possible to study the anatomical, acoustic, and perceptual aspects of speech.
Different spoken forms can mean different things, and we might study the kinds
of things we can “mean” and the ways differences in the forms of words are
related to differences in their meanings. Literature, rhetoric, and the texture of
ordinary verbal interchange show us that we can do many different things with
words, and we might take an understanding of these various potential uses as
the goal of a scientific study of language.
All of these approaches to language, however, assume that language is an
essentially ubiquitous activity of human beings. As an infant, every human
being normally acquires a knowledge of (at least one) language, knowledge
that is acquired while exposed to a limited amount of speech in the child’s
surroundings and that allows him or her to participate in the verbal life of
the community within a relatively short time. Surely the most basic questions
to ask if one wants to understand the phenomenon of language concern the
nature and form of that knowledge, the way it arises and the way it relates to
other aspects of human cognition.
1
2 1 Studying the human language faculty
The study of language and cognition during the past several decades has given
increasing credibility to the view that human knowledge of natural language
results from – and is made possible by – a biologically determined capacity spe-
cific both to this domain and to our species. An exploration of the functional
properties of this capacity is the basic program of the present book. These de-
velop along a regular maturational path, such that it seems appropriate to speak
of our knowledge of our language as “growing” rather than as “being learned.”
As with the visual system, much of the detailed structure that we find seems to
be “wired in,” though interaction with relevant experience is necessary to set
the system in operation and to determine some of its specific properties. We
can refer to experience that plays such a role as triggering the organization of
the system, exploiting a term taken from ethology.
The path of development which we observe suggests that the growth of
language results from a specific innate capacity rather than emerging on a purely
inductive basis from observation of the language around us. The mature system
incorporates properties that could not have been learned from observation or any
plausibly available teaching. The deep similarity among the world’s languages
also provides support for the notion that they are the product of a common
human faculty.
These fundamental, apparently native properties are shared by the gestural
languages which develop spontaneously in Deaf communities, quite indepen-
dently of one another or of the language of the surrounding hearing community.
We must conclude that they are neither the result of simple shared history nor
necessary consequences of the articulatory/acoustic/auditory modality of lan-
guage in its most familiar form, spoken language. The development of struc-
turally deficient pidgins into the essentially normal linguistic systems found
in creoles, as a result of transmission through the natural language learning
process in new generations of children, provides additional evidence for the
richness of that process.
The domain-specificity of the language faculty is supported by the many
dissociations that can be observed between control of language structure and
other cognitive functions. Focal brain lesions can result in quite specific lan-
guage impairments in the presence of otherwise normal cognitive abilities;
and vice versa. Natural as well as acquired disorders of language also support
the proposal that the human language faculty is a product of our genetically
determined biological nature: there is evidence that certain language deficits
show a clear distribution within families, patterns that epidemiological and
other studies show to be just what would be predicted of relatively simple
heritable traits (Gopnik and Crago 1991, Tomblin 1997).
Finally, the species-specificity of the human language faculty is supported by
the very fact that (absent severe pathology) every human child exposed in even
1.1 Linguistics and the mind/brain 3
limited ways to the triggering experience of linguistic data develops a full, rich
capacity which is usually more or less homogeneous with that of the surrounding
community. Meanwhile, efforts to teach human languages to individuals of
other species, even those closest to us, have uniformly failed. While a certain
capacity for arbitrary symbolic reference can be elicited in some higher apes
(and perhaps even in other animals, such as parrots), syntactic systems even
remotely comparable to those of human languages seem to be quite outside the
capacity of non-humans, despite intensive and highly directed training.
These considerations make it plausible that human language arises in biologi-
cally based ways that are quite comparable to those directing other aspects of the
structure of the organism. The language organ, though, is not to be interpreted
as having an anatomical localization comparable to that of, say, the kidney. Our
understanding of the localization of cognitive function in brain tissue is much
too fragmentary and rudimentary. Certain cortical and subcortical areas can be
shown to subserve functions essential to language, in the sense that lesions in
these regions disrupt language functioning (sometimes in remarkably specific
ways), but an inference from this evidence to a claim that “language is located
in Broca’s (and/or Wernicke’s) area” is quite unwarranted. The linguistic capac-
ity which develops naturally in every normal human being appears to be best
understood in functional and not anatomical terms, at least for the time being.
We will return to these issues of the physical basis of linguistic knowledge in
chapter 10; until then, let us take it as given that some such physical basis must
exist, and concentrate on the nature of linguistic capacities.
1.1 Linguistics and the mind/brain

The major puzzles in linguistics concern the fact that our biological nature
makes it possible to acquire a richly structured kind of knowledge, much of
which is only indirectly evidenced (if at all) in the data available to the learner.
This is the “poverty-of-stimulus” problem, which will occupy us in chapters
below. It follows that any sensible student of language would want to take as
his or her basic agenda the investigation of the details of this knowledge, and
the foundation for its growth in the individual. This understanding of what
linguistics is all about, however, with the focus on language organs, is actually
a relatively recent one.
In the Introduction, we noted the problem of where linguistics departments
fall in the geography of modern academia. As the central concerns of the field
have moved away from historical issues and the study of languages for their
own sake, and toward a more general understanding of the phenomenon of lan-
guage, the reaction has often been to move linguistics out of the humanities and
into the social sciences. This has not been particularly comfortable, however,
since the most important questions about the nature of language do not really re-
spond to the assumptions and methods of the disciplines concerned with group
phenomena that arise fundamentally as a part of social reality (anthropology,
economics, sociology, political science, etc.). The problem of relating group
behavior to the properties of the individual, and its appropriate resolution, was
already prefigured some time ago in an even broader form by Edward Sapir, a
notable visionary in his approach to language and to culture more generally.
In his classic article in the first number of the journal Psychiatry (Sapir 1938)
and in many other writings, Sapir urged that a true understanding of the nature
and effects of culture must necessarily be founded on an understanding of the
individuals who participate in culture and society: “In the psychological sense,
culture is not the thing that is given us. The culture of a group as a whole is not
a true reality. What is given – what we do start with – is the individual and his
behavior.”1 And the central term in this understanding is the nature of the mind
and personality of the individual, not an external characterization of his actions
and responses or some system that somehow exists outside of any particular
person.
Trained by Franz Boas as a cultural anthropologist, Sapir devoted most of his
professional life to the study of language and the development of the nascent
discipline of linguistics (see Anderson 1985, chap. 9 and Darnell 1990 for
sketches of his personal and professional life). For Boas, as for Sapir, language
was a key to all other understanding of cultural realities, since it is only through
an appreciation of the particularities of an individual’s language that we can
hope to gain access to his thoughts and conception of the world, both natural
and social. Sapir, indeed, is widely associated with the notion of “linguistic
relativity,” according to which the structure of an individual’s language not only
reflects but even contributes to determining the ways in which he construes his
world.2 Language thus occupies a central place among the phenomena that can
lead us to an understanding of culture; and it must follow that the way to study
language is in terms of the knowledge developed in individual speakers, not
in terms of such externalities as collections of recorded linguistic acts. In the
history of linguistics, Sapir is remembered especially as one who emphasized
the need to study what speakers know and believe (perhaps unconsciously)
about their language, not simply what they do when they speak.
In addition to his primary focus on linguistics, Sapir also wrote widely of more
general issues in the nature of society, culture, and personality. His Psychiatry
piece was far from isolated in the sympathy it showed with the project of psy-
chiatry and psychoanalysis. This interest in psychiatric issues and approaches
1 Sapir 1994, p. 139. The quotation is from a recent reconstruction of Sapir’s lectures on these
topics.
2 This is the so-called “Sapir–Whorf Hypothesis.”
was certainly not isolated from his work as a linguist and anthropologist. On
the contrary, as reflected in the title of his article (“Why cultural anthropology
needs the psychiatrist”), he felt that the mode of understanding essayed by
the psychiatrist was really the only path to a true appreciation of cultural phe-
nomena, given the claim above that the individual is the basic reality in this study.
Why the psychiatrist, in particular? One must remember that the 1930s, when
Sapir was active in general studies of personality and culture, was the time when
ideas from the positivist and behaviorist traditions predominated in scientific
investigations of psychological questions, and of course these were precisely
antagonistic to a sympathetic investigation of the nature and contents of the mind
and personality. Psychiatry, in contrast, was centrally occupied with exactly this,
and so the fundamental place of the individual in language and culture entailed a
need for the kind of light that could only be shed on core issues by psychiatric
investigation and understanding.
While no one denied Sapir’s stunning brilliance as a linguist, both as a
theoretician and as an analyst, many of his colleagues at the time considered
this “mentalist” aspect of his thought to be an eccentricity – even an aberration –
something to be excused rather than imitated. After all, linguistics was on its
way to attaining genuine status as a science precisely through adopting the be-
haviorism of the day, focusing on purely mechanical methods for collecting and
arranging linguistic data so as to arrive at a purely external analysis of linguistic
behavior, eschewing all metaphysical talk about “minds” and such-like unob-
servables (cf. Bloomfield 1933). Over time, however, the field of linguistics
has arrived at essentially the same conclusion Sapir did, by its own path and
making only a little use of the insight he had to offer.
1.2 Linguistics as history

While the study of language up through about the 1870s was as much a branch of
philosophy as an empirical science, scholars began to adopt genuinely scientific
methods around that time. Linguistics grew, in part, out of the study of philology:
how do we establish the correct form of an ancient text, given that we have
perhaps only several corrupted manuscripts to go on? A method developed for
comparing manuscripts to establish the most likely ancestral form from which
they all derived.3
When nineteenth-century scholars came to the conclusion that whole lan-
guages, not just individual texts, could be regarded as related in this way,
the comparative method came to be applied to the problem of how to
3 The connection between philological methods in dealing with ancient manuscripts and the Stamm-
baum or “tree” model of inter-linguistic relationships is discussed by Hoenigswald 1963, who
suggests that the method of reconstructing the geneology of manuscripts developed by F. Ritschl
may have influenced the conceptions of his student A. Schleicher.
reconstruct an ancestral language (e.g., Proto-Indo-European) from which a

number of attested modern languages could be derived. In the case of texts, it
is clear how an original form comes to be corrupted: scribes make mistakes.
But what is the analog in the case of whole languages? Nineteenth-century
linguists came up with the conception of “sound change”: systematically over
time within part of a speech community, certain sounds change into others.
Thus, voiced stops like b, d, g may be uniformly replaced by the corresponding
voiceless stops p, t, k, for example (see (1.2) below). When we reconstruct an
earlier form of language, what we are trying to do is to undo the sequence of
sound changes by which the modern language came about.
Where this became scientific was when people suggested that there was a gen-
eral lawfulness to these changes, and proposed the principle of regularity
of sound change. That is, unlike the random pattern of scribal errors, sound
changes operate in a uniform way, and do not affect individual words idiosyn-
cratically. Furthermore, scholars hoped that it would be possible to derive the
range of possible sound changes from a study of the physics and physiology
of speech, assuming that sound changes reflected regularities that could ulti-
mately be deduced from natural laws of these domains, and not purely arbitrary,
random events.
The point of these observations is that at this stage, the object of scientific
inquiry about language was pretty much defined by the scope of the available
genuinely explanatory principles, and this meant that linguistics was “about”
the histories of languages (and of individual words within those languages). The
basic question of the field could be put as: how do languages come to be as they
are through (diachronic) change over time? As a result, historical linguistics
was the only kind of linguistics there was at the time. Towards the end of the
nineteenth century, Hermann Paul (1880, p. 20) was able to pontificate that
“Es ist eingewendet, dass es noch eine ander wissenschaftliche Betrachtung
der Sprache gäbe, als die geschichtliche. Ich muss das in Abrede stellen”
(It has been objected that there is another scientific view of language possible
besides the historical. I must contradict this).
Linguistics was thus resolutely about change, and for nineteenth-century lin-
guistics, what changed was a “language”: essentially, an inventory of words.
The systematicity of linguistic change arose from the nature of its basic mecha-
nism: change of sounds, which thereby affected all words in which the relevant
sounds (or sound combinations) occurred. There was more to language than
just words, of course, but everything else was attributable either to a universal
“logic” or to individually variable “habits,” and these matters did not greatly
interest the linguists of the time. The job of the linguist, then, was to write the
history of words, their pronunciations, and their meanings. This understanding
of what linguistics is about still characterizes some parts of the field, though
classical historical linguistics is a relatively small subdiscipline today.
Words are transmitted from one generation to the next, and they may change
their form in the course of that transmission. Latin pater “father” became padre,
père, patre, pai, etc. in the modern Romance languages. One could characterize
such changes by writing “sound laws” such as the principle that a dental stop
came to be pronounced with vocal fold vibration (that is, [t] came to be pro-
nounced as [d]) between a vowel and a vocalic r at some point in the transition
from Latin to Italian and Spanish. The important observation which made it
possible to envision a science of this sort of thing was the fact that while such
changes affected particular words, they were in principle formulable in terms
of phonetic environments alone, without direct reference to the words them-
selves. Changes in the set of words, then, were the consequence of the working
of sound laws, and linguistics could potentially develop a science of these that
could hope to explain the development of languages over time.
In this view, languages are the basic objects of reality, entities existing in their
own right “out there,” waiting to be acquired by speakers. Linguists sought to
determine and quantify the degree of historical relatedness among sets of lan-
guages, and this relatedness was expressed through tree diagrams or cladograms
such as that in (1.1), introduced by August Schleicher (1861– 62).
(1.1) Proto-Indo-European
Aryan--Greek--Italo-Celtic Germano-Slavic
Greek--Italo-Celtic
Aryan Italo-Celtic Balto-Slavic
Indic Iranian Greek Albanian Italic Celtic Slavic Baltic Germanic
(After Schleicher1861–62; note that many of the details in this model would be contested
by scholars today.)
As has often been pointed out, the Schleicher-style cladograms, familiar

from all textbooks on historical linguistics, capture only what biologists call
“homologies,” similarities which result from common historical origins, and
not “analogies,” similarities which arise from common developments. This was
recognized very early, and nineteenth-century linguists supplemented the “tree-
theory” picture with “wave-theory” models representing common innovations
shared by geographically contiguous languages.
These cladograms assume that languages are like species, as articulated

by Schleicher 1863, and that they derive from one another in some coherent
historical or evolutionary development. But we should realize some of the ide-
alizations that are involved. Trees of this kind idealize away from the fact that
languages do not split sharply at some specific point and suddenly emerge in
their full individuality. The splitting process is more gradual and is initiated by
minor divergences. We might say that the first change which affected, say, Latin
and not any of the other languages, is the bifurcation point, the point at which
Latin suddenly splits away. That is not enough, though: saying that French and
Italian are both (separately) descended from Latin glosses over the fact that they
are descended from different forms of Latin, and that “Latin” is a cover term
for many different forms of speech. In fact, the analogy between languages and
species is not ultimately a very useful one, and the conventional tree models of
historical linguists would require vast elaboration to be equivalent to modern
biological cladograms based on the molecular genetics of organisms.
Schleicher-style family tree models also do not provide a very insightful
way of talking about relatedness, since they deal only with historically derived
similarities. These models were developed to capture relatedness among lex-
icons, which were taken to be pretty much unstructured inventories of words
(see chapter 7 below). If one goes beyond the mere word-stock, however, and
envisions relatedness more broadly than just in terms of similarities resulting
from a common history, very different relations would emerge.
If one thinks of grammars, individual systems of knowledge, language organs,
developed on the basis of exposure to some relevant linguistic experience and
emerging according to the requirements of the linguistic genotype (universal
grammar) as children set predefined parameters in appropriate ways, one can
then compare the grammars of English speakers with those of German speakers,
and ask whether those grammars are more or less similar to each other than
either are to the grammars of Italian speakers. After all, the child acquiring a
language has no access to the historical sources of the linguistic forms in the
input, and accordingly no bias toward having linguistic ontogeny recapitulate
phylogeny, in Ernst Hackel’s famous phrase.
German grammars are quite different from English grammars: they have
the object preceding rather than following the verb (Ich glaube dass Hans
Maria liebt “I believe that Hans loves Maria”); they require the finite verb to
appear in second position in main clauses (Im Dorf gibt es nicht zu viele Hunde
“In the town there are not many dogs”); they have very different word order
possibilities within the verb phrase. In fact it is quite possible – even likely –
that English grammars might be more similar to grammars with which there
is less historical connection. From this perspective, looking at the parameters
in the current linguistic literature, English grammars may be more similar to
Italian than to German, and French grammars may be more similar to German
than to Spanish.4 There is no reason to believe that structural similarity should
be even an approximate function of historical relatedness – assuming that there
is in fact a non-circular notion of historical relatedness to be discovered.
Nonetheless, nineteenth-century linguists focused on languages, seen as in-
ventories of words composed from basic inventories of sounds, which could
change over time. Languages so conceived appeared to change in systematic
ways which could be formulated as regular correspondences, each the product
of some combination of sound changes that could be expressed as sound laws
independent of the identities of the individual words affected. By the end of
the nineteenth century, linguists knew that this was not the entire story: there
were other regularities of language change which could not be stated in purely
phonetic terms, suggesting that at least in these cases it was not the language
(construed as an inventory) or its sounds that were changing, but rather some
kind of more abstract system. This was dealt with in a terminological move:
there were regularities of “sound change,” but there could be other sorts of
change that worked differently. These were called “analogical change,” and
were assumed to be governed by quite a different, more mysterious, kind of
regularity.
Nineteenth-century linguists focused on the surface forms of words, which
are the products of human behavior, rather than on internal processes that un-
derlie and shape that behavior. They thus dealt with E-language rather than
I-language in the terminology of Chomsky 1986. Not all aspects of language
have satisfying accounts in these terms, though, and it is often necessary to
invoke underlying processes and abstract systems that are not manifest parts of
the observable facts of E-language. This is true for such famous instances of
regular sound change as Grimm’s Law, which affected many types of conso-
nants in a related way in a sort of cycle (cf. (1.2a)); or the Great Vowel Shift
in English (cf. (1.2b)), which changed all of the long vowels in another sort of
cycle, raising all vowels one step and making diphthongs of the highest vowels.
Thus, [swe:t] “sweet” (pronounced with a vowel similar to that of modern skate)
became modern English [swi:t]; [ti:m] “time” (pronounced rather like modern
team) became [taim]; [hu:s] “house” (with a vowel similar to that of modern
loose) became [haus], etc.). Grimm’s Law and the Great Vowel Shift affect
many sounds at the same time, and represent changes in systems, not simply in
inventories.
4 Linguists idealize and speak of French grammars, but a “French grammar” has much the same
status as a “French liver,” a convenient fiction. Individuals have livers and individuals have
grammars. Grammars may be similar to one another, and we may seek similarities between the
grammars of “English speakers,” whoever they may be exactly. In doing so, we must be prepared
to find differences between the grammars of a man in Houston and a woman in Leeds.
(1.2) a. Grimm’s Law

voiced
aspirated voiceless
fricative
b. The Great English Vowel Shift

i: u:
ai au
e: o:
ε: ɔ:
a:
Because the historical linguists of this period were working with the external
forms of language, rather than with internal, underlying processes and abstract
systems, it makes sense that their attention was confined to phonological and
morphological aspects of language, and they paid little attention to change in
syntactic systems. It makes no sense to think of (sets of) sentences, products of
behavior, as being transmitted from one generation to another, because language
acquisition is clearly not a matter of learning sets of sentences. This limitation
was not seen as a matter of concern: they simply worked on what they felt they
had useful tools to elucidate. The debates of the time were primarily about the
nature and causes of sound change.
A part of what made the study of sound change appealingly scientific was
the fact that its systematicities could be related to those of another domain:
the physics and physiology of speech. When we describe Proto-Indo-European
/p,t,k/ as becoming /f,θ,χ/ in the Germanic languages (e.g., prehistoric patr.
becoming modern English father, tres becoming three, etc.), we can unify
these by saying that one kind of formational occlusion (complete blockage
of the oral cavity followed by a rapid release, accompanied by a brief burst of
noisy airflow) is replaced by another (incomplete blockage of airflow, allowing
for noisy, turbulent flow during the period of occlusion). With this unitary
formulation in hand, we can hope to find reasons why one sort of occlusion
should be replaced by another, a rather more hopeful task than that of finding
reasons why one arbitrary set of symbols should be replaced by another equally
arbitrary one.
It is natural that the development of rigorous methods in historical linguistics
more or less coincided with the beginnings of serious investigation of the
phonetics of speech using instrumental techniques that were precise and

quantifiable (if quite primitive by today’s standards). It was widely assumed that
the terms of an explanatory theory of speech production and perception would
serve as the underpinnings of an equally explanatory theory of sound change,
and it is no accident that among the important figures in the late-nineteenth-
century development of comparative Indo-European studies was Eduard
Sievers, author of the basic text of the period on the science of phonetics,
Sievers 1881.
But in spite of the promising start provided by increasingly precise accounts
of speech, sound laws were clearly not reducible to measurable aspects of
the physical world. Grimm’s, Grassmann’s, and Verner’s Laws (three changes
whose discovery and formulation played particularly important roles in the
development of the doctrine of the regularity of sound change) were not general
laws like Boyle’s Law: they were rather particular contingent events that had
taken place at a given point in history, but were not universally valid. As such,
they required a deeper explanation. Changes were taken to be directional, as
in biology. Rask (1818) held that languages became simpler. Schleicher (1848)
identified a progression from isolating to agglutinating to inflectional types,
although this was said to hold for preliterate societies, whereas Rask’s drive to
simplicity was relevant for literate societies. Darwin (1874, pp. 88–92) thought
that languages changed in the direction of having shorter, “easier” forms.
There was widespread agreement that language change followed fixed de-
velopmental laws, and that there was a direction to change, but there was active
disagreement about what that direction was. By the end of the nineteenth cen-
tury there was an enormous body of work on sound correspondences among
historically related languages, and vast compendia of changes that had taken
place, especially in the Indo-European languages. There were, however, few
ideas about why those changes had taken place.
First, the notion that languages became simpler, more natural, easier to pro-
nounce, etc. was quite circular: “simpler,” etc. was what languages change
to, and there was no independent definition in a framework dealing entirely
with historical change. Second, the idea that languages change toward greater
simplicity (or whatever) gives no account of why a given change takes place
when it does – as opposed to the laws of gravity, which apply to all objects at
all times. To that extent, invoking notions of directionality was no more law-
like than the laws of Grimm, Grassmann, and Verner, which directionality was
intended to explain.
There were attempts to break out of this circularity, by invoking psychological
causes, but the psychology itself was implausible and unexplanatory. Grimm
(1848) attributed his consonant changes to the courage and pride associated with
the German race’s advance over all parts of Europe. Jespersen (1909) ascribed
the shift in the meanings of verbs like like, repent etc. (whereby Middle English
sentences like pears like me became modern I like pears) to the greater interest
taken in persons than in things at one stage in the history of English. For other
examples, see Lightfoot 1999. Hardly anyone was satisfied that the result of
such speculations provided a genuine science of (the history of) language.
1.3 Linguistics as the study of E-language

In the early years of the twentieth century, then, the data of linguistics consisted
of an inventory of sound changes occurring for no good reason and tending
in no particular direction. The historical approach had not brought a scientific,
Newtonian-style analysis of language, of the kind that had been hoped for,
and there was no predictability to the changes. One of the greatest figures in
historical linguistics at the time was Ferdinand de Saussure, whose revolutionary
work on the vowel system of Indo-European (de Saussure 1879) established
much of the research agenda in historical linguistics for the twentieth century;
and it is something of an irony that he was the first to argue forcefully that these
historical issues were only a very small part of the story, and that the science
of language would have to provide an account of the nature of the linguistic
systems that we find today in their own terms, not just their histories.
Saussure (de Saussure 1974) argued that when we look at language, we must
distinguish the system of a language (called “langue”) from concrete acts of
speaking or language use (“parole”). What is systematic about language is to be
sought in an analysis of langue, but for Saussure, the locus of linguistic change
is not there, but rather in usage, or parole. It follows that linguistic change itself
cannot be the object of a fundamental science of language.
Following Saussure’s lead, the historicist paradigm – the notion that there are
principles of history to be discovered, which would account for a language’s
development – was largely abandoned by the 1920s (see Lass 1980 for thor-
ough discussion). Indeed, there was a virulent anti-historicism in the writing of
several major figures of this period, including not only Saussure but also Franz
Boas, Edward Sapir, and Leonard Bloomfield. They worked to some extent on
language change, but abandoned historicism, and with it the earlier program of
seeking to explain how it was that languages come to be as they are.
Perceived problems included the circularity of the historical principles in-
voked, and the kinds of psychological claims that had been made. Sapir (1929,
p. 208) wrote that the psychological interpretation of language change was
“desirable and even necessary,” but that the existing psychological explana-
tions were unhelpful and “do not immediately tie up with what we actually
know about the historical behavior of language.” Bloomfield (1933, p. 17)
complained about the circularity of psychologizing, saying that there was no
independent evidence for the mental processes other than the linguistic pro-
cesses they were supposed to explain. The historicist paradigm was not really
1.3 Linguistics as the study of E-language 13
refuted or shown to be wrong: rather, it was abandoned as yielding diminish-

ing returns and not asking the truly fundamental questions about the nature
of language. Work on language change continued to flourish, but subsequent
generations did not, by and large, appeal to historicist explanations.
In chapter 8 below, we will ask what accounts of language history we can
give if we take a different, more contingent approach. We shall shift away from
a study of the results of behavior and toward a study of the states and properties
of the mind/brain that give rise to those products, and we will see that a rather
deeper (if quite different) understanding of change results. But at this point, let
us continue to trace the evolution of thought about what linguistics ought to
concern itself with.
Saussure’s concern to re-orient linguistics from the study of parole to that of
langue gave rise to structural linguistics, or the study of linguistic systems: the
sounds and grammatical structures of, e.g., French, as opposed to the history of
how Latin originals became the individual words of Modern French. Saussure
himself argued that the right basis for such an understanding was to construe
the linguistic system as something with a reality that was primarily social: a
set of conventions among speakers of a given language to do things in a certain
way. For him, langue was a system existing outside of any individual, located
rather in the speech community that employs a particular code.
The nature of that code was that of a structured inventory of signs, where each
sign consists of the unity of some expression with the semantic content which it
signifies (thus, “[kæt] means ‘Felis catus’ ”). The structural value of any unit in
the system derives from the principle that each sign is distinct from every other
sign: thus, the value of cat is provided by the facts that in terms of expression,
[kæt] is different from [ræt], [kɔt], [kæn], etc.; while in terms of content, a cat is
distinct from a lion, tiger, dog, etc. The system of signs (langue) is what makes
communication possible within a speech community; it constitutes the common
knowledge of that community, and is therefore independent of the particular
properties of any individual member of the community, or of any individual
utterance that may be produced on a particular occasion by a speaker of the
language (an act of parole).
While Saussure’s conception of langue as a primarily social entity was not
widely adopted, his replacement of diachronic change by synchronic system as
the object of study in the field was. Structural linguistics evolved in various ways
through the first half of the century, and a consideration of those developments
is well beyond the scope of the present book (see Anderson 1985, among many
other sources, for further discussion).
With the rise of strongly empiricist, verificationist views in the philosophy
of science during this period, the study of language eventually became mired
down in the assumptions of behaviorist (or as Leonard Bloomfield, a major
influence in this development, preferred to put it, “mechanist”) psychology. On
that view, scientific investigation can only concern itself with external observ-
ables. Properties of mind (if indeed there are such) are intrinsically beyond the
reach of science. Bloomfield in particular (contrary to his subsequent reputa-
tion in some circles) did not deny the existence of the mind, but maintained that
its properties were so far beyond the reach of (current) science as to make its
investigation pointless.
What was consistently true through this period was that the primary object
of inquiry in the field was not something historical, but something present to
contemporary speakers of a given language. This was always something ex-
ternal, and the basic question of the field had become: what are the properties
of the sets of sounds, words, sentences, etc. recorded in attested acts of speak-
ing? Whether thought of as deriving from social conventions or as the external
responses corresponding to particular stimuli, linguistic objects were consis-
tently seen in this external mode. The commitment to an E-language view of
what linguistics studies was thus axiomatic and complete.
1.4 Linguistics as the study of I-language

The study of E-language entails the study of languages as sets of externally
observable elements – sounds, morphological units, words, phrases, sentences,
texts, etc. But there is no reason to believe that such sets are actually coherent
or well-defined objects of inquiry. That is, there is no precisely definable notion
of a “language,” such that one can show in some non-circular way that a given
sentence is “a sentence of English” (see Chomsky 1986, Lightfoot 1999, and
others). A given sentence might be used by a particular speaker in Arkansas, but
not by another speaker in Yorkshire. Does that make it a sentence of English?
Apparently the answer is “yes” for at least some Arkansas speakers, but “no”
for at least some Yorkshire speakers.
There is no general algorithm for characterizing the sentences of English,
and there is no reason to expect to find one. If different speakers of the “same”
language speak differently, there can be no general algorithm. Languages in
this extensional sense are conglomerations of the outputs of various grammars,
where each grammar is represented in the mind/brain of (at least) one individual
speaker. It is not, of course, necessary that people who consider themselves to
speak the “same” language should have different grammars, different language
organs, but it is certainly not necessary that their grammars be the same, either,
and empirical evidence suggests considerable variation within particular speech
communities. Some of this variation, as it is reflected in differences in output of
the grammars concerned, may be identified by speakers as reflecting different
“languages,” but this is a sociological notion, not a linguistic one. For example,
speakers across a large part of the former Yugoslavia used to consider that they
spoke “Serbo-Croatian,” but as a result of changes in political identity, some of
these came to be speakers of “Serbian,” others of “Croatian,” “Bosnian,” etc. –

all of this in the absence of any significant change in the grammars of the
speakers involved.
The E-language set of sentences in use in a speech community, then, is not a
coherent object of study because it corresponds to the outputs of a diverse range
of (possibly very different) grammars. Indeed, even the E-language notion of
the set of sentences produced by a given speaker cannot serve this purpose.
Ignoring the fact that an individual speaker will typically control a number
of (possibly very different) styles, corresponding to different grammars, the
actual set of sentences produced and/or understood by a speaker is a completely
accidental consequence of the range of things s/he has had occasion to say
or hear. Insofar as there are systematic properties of this collection, they are
systematicities either of non-linguistic experience or of the underlying grammar,
but not of the set of sentences itself. “Languages” in the E-language sense are not
well-defined entities. They do not “exist” independent of contingent, concrete
acts of speaking; they do not “descend” from one another; and there is no
reason to believe that their study will reveal any interesting properties. They
are convenient fictions, like the rising and the setting of the sun. It follows that
if we want to develop a genuine science of language, this must be concerned
instead with the I-language notion of grammars, the properties of a person’s
language organ.
In the 1950s, the intellectual coherence and scientific status of American
structuralist theories of (E-)language were not seriously in doubt. When an
alternative to the distribution-based theories of linguistic structure then in
vogue5 developed around the work of Noam Chomsky, this did not initially
challenge the notion that the goal of linguistics was to provide rigorous and
principled descriptions of the set of possible utterances that constitute a par-
ticular language. As Chomsky worked out his ideas, however, it became clear
that the conceptual underpinnings of the field were in principle incapable of
providing a substantive account of language.
The central difficulty for existing theories was the problem of how lan-
guage could be acquired. Chomsky (1959) presented a devastating critique of
Skinnerian behaviorism that centered on this issue, an assessment that eventu-
ally had far-reaching effects in psychology, but whose immediate consequence
was a shift of seismic proportions in the expectations scholars had for linguistic
theory. He observed that when we consider language use (“verbal behavior” in
Skinner’s terms), we must recognize that
[t]he child who learns a language has in some sense constructed the grammar for him-
self on the basis of his observation of sentences and non-sentences (i.e. corrections by
the verbal community). Study of the actual observed ability of a speaker to distinguish
5 Harris 1951 was perhaps the epitome of this approach.

sentences from non-sentences, detect ambiguities, etc., apparently forces us to the con-
clusion that this grammar is of an extremely complex and abstract character, and that the
young child has succeeded in carrying out what from the formal point of view, at least,
seems to be a remarkable type of theory construction. Furthermore, this task is accom-
plished in an astonishingly short time, to a large extent independently of intelligence,
and in a comparable way by all children. Any theory of learning must cope with these
facts. (Chomsky 1959, p. 57)
These remarks had a number of profound consequences. For one, they made
it clear that the generalized mechanisms of behaviorist learning theory, based on
attributing as little structure as possible to the organism in particular domains,
was quite incapable of dealing with the acquisition of human language. For
another, they brought the problem of learning to the forefront in the study
of language: where previously linguists had generally been content to char-
acterize the language itself, Chomsky made it clear that an understanding of
language cannot proceed without asking how a speaker’s ability to use it arises
in development.
But for linguists, the most profound effect of these arguments was a shift
in our conception of the object of study in the field. Chomsky stressed that
the basic problem is not one of characterizing what people do: it is rather one
of characterizing what they know. The central reality of language is the fact
that we call someone a speaker of, say, Japanese, because of a certain kind of
knowledge that he or she has. If that is the case, linguists need to find a way
to study the structure of this knowledge, and while the things people say and
do can constitute important evidence, that is not all there is, or even the most
important thing. This knowledge is what we are calling a person’s language
organ.
In this focus on the nature of language as a form of knowledge, an aspect of
the structure of the mind, linguists have thus returned to a conception much like
Sapir’s of the centrality of the individual to an understanding of linguistic and
cultural phenomena. The decline of behaviorist assumptions in the last several
decades of the twentieth century necessarily led to a much broader consensus
about the need to understand the mind in its own terms: if Bloomfield was
indeed right that the mind was beyond the reach of current science, the thing to
do was to develop the relevant science, not study something else.
Much of academic psychology still finds itself preoccupied with externalist
issues, and for one reason or another rejects the validity or utility of conceiving
its object in terms of the minds of individuals described at some appropriate level
of abstraction. The result has been the rise of cognitive science as a discipline
whose goal is precisely a science of the mind. Combining ideas from linguistics,
computer science, philosophy, anthropology, and cognitive psychology, this
emerging field focuses squarely on the nature of mental and cognitive life.
Linguists and linguistics have been important in these developments, in view

of the fact that it is in this sphere that the central role of the individual can be
studied particularly concretely and productively. Cognitive science thus plays
for the contemporary linguist the kind of role Sapir foresaw for psychiatry: one
that makes it possible to study phenomena that emerge from the individual by
providing methods for investigating the content of the individual’s mind and
the structure of mental life. If the insight of Sapir’s 1938 paper now appears in
somewhat different form, its validity has only been reinforced by subsequent
developments in the study of language.
The result of the developments surveyed above has been the rise of a con-
sensus that linguistics really is (a branch of) cognitive science: the object of
study, I-language, is an aspect of the structure of the mind/brain, the language
organ, defined at some level of abstraction, which must be studied as such and
whose study can lead to further insights about the architecture of cognition.
This view is certainly not a universal one: some linguists, for a variety of rea-
sons, continue to focus on external objects and events rather than on the form
of linguistic knowledge. It is also important to note that many linguists who
would assent in principle to the view we have presented nonetheless continue
to pursue in practice the study of particular linguistic phenomena in ways that
remain unchanged from their origins within an E-language conception of the
discipline. In what follows, we will attempt to draw out the differences in ap-
proach that follow from this shift of attention, and the relation between such
abstract, general considerations and the everyday practice of linguistics.
Here we have sketched some aspects of the path by which the field of lin-
guistics has arrived in its own way at what were essentially Sapir’s conclusions.
In subsequent chapters, we will outline some of the basic properties and sub-
structures of language organs, considering as we proceed how we have been
able to discover them.
2 Language as a mental organ
In this chapter, we pursue an important source of evidence for the claim that
human language has a specialized basis in human biology: the relation between
what a speaker of a language can be said to “know” and the evidence that is
available to serve as the basis of this knowledge. The apparently common-sense
notion that an adult speaker’s knowledge of his/her language arises by simple
“learning,” that is, as a direct generalization of experience, turns out to pose
a logical paradox. We begin with two brief examples that illustrate this point,
and then explore the consequences of this for the mechanisms that must in fact
underlie the development of language organs in normal human speakers.
2.1 We know more than we learn

A striking property of language acquisition is that children attain knowledge
which, quite literally, infinitely surpasses their actual experience. On the basis
of very limited experience, a productive system, a grammar, arises in each
speaker which not only encompasses the actual facts to which they have been
exposed, but also permits the production and comprehension of an unlimited
range of novel utterances in the language. There must, therefore, be much more
to language acquisition than mimicking what is heard in childhood; and there
is more to it than the simple transmission of a set of words and sentences from
one generation of speakers to the next.
2.1.1 Two grammatical puzzles

Consider some subtleties that people are usually not consciously aware of.
The verb is may be used in its full form or its reduced form: English speakers
can say either Kim is the professor here or Kim’s the professor here, He is
happy or He’s happy. However, certain instances of is never reduce: for ex-
ample, the underlined items in Kim is happier than Tim is or I wonder who
the professor is here. Most speakers are not aware of this, but we all know,
subconsciously as it were, that we can not use the reduced form in such cases.
How did we come to know this? As children, we were not instructed to avoid
18
the reduced form in certain places. Yet, all children typically attain the ability
to use the forms in the adult fashion, and this ability is quite independent of
intelligence level or educational background. Children attain it early in their
linguistic development. More significantly, children do not try out the non-
occurring forms as if testing a hypothesis, in the way that they “experiment”
by using forms like goed and taked. The ability emerges perfectly and as if by
magic.
Another example. Pronouns like she, her, he, him, his sometimes may refer
back to a noun previously mentioned in a sentence (2.1a–c). However, one can
only understand (2.1d) as referring to two men, Jay and somebody else; here
the pronoun may not refer to Jay, unlike (2.1a–c).
(2.1) a. Jay hurt his nose.
b. Jay’s brother hurt him.
c. Jay said he hurt Ray.
d. Jay hurt him.
To extend this point, consider some more complex examples, as in (2.2):
(2.2) a. When Jay entered the room, he was wearing a yellow shirt.
b. Jay was wearing a yellow shirt when he entered the room.
c. When he entered the room, Jay was wearing a yellow shirt.
d. He was wearing a yellow shirt when Jay entered the room.
e. His brother was wearing a yellow shirt when Jay entered the
room.
In all of the sentences in (2.2) the pronoun (he or his) may refer to some
other individual, not mentioned. It may also refer to Jay – in all cases, that
is, except (2.2d), where the wearer of the yellow shirt can only be understood
to be someone other than Jay. Again, all speakers understand these sentences
in the same way, but we may legitimately be puzzled at the source of this
commonality. It is quite unlikely to have come from any explicit instruction: as
far as we know, these points about the interpretation of pronouns had not been
systematically noted, even by grammarians, prior to the late 1960s (Ross 1967,
Langacker 1969, Reinhart 1976, McCawley 1999).
As adults we generalize that a pronoun may refer to another noun within the
same sentence except under very precise conditions (as in (2.1d) or (2.2d)). But
then, how did we all acquire the right generalization, particularly knowledge of
the exceptions? In the case of (2.2d), we might be tempted to say that it is only
natural that a pronoun should not be able to refer to an individual mentioned only
later in the sentence, but the evidence of (2.2c,e) shows that such “backwards
anaphora” is in fact possible under some circumstances. Furthermore, we will
see in chapter 9 that even very young children know when backwards anaphora
is possible and when it is not.
20 2 Language as a mental organ
2.1.2 Where does this knowledge come from?

In approaching both of these problems, recall the nature of our childhood expe-
rience: we were exposed to a haphazard set of linguistic expressions. We heard
various sentences containing both the full verb is (am, are) and the reduced
form ’s (’m, ’re). We also heard sentences containing pronouns, in some of
which the pronoun referred to another noun in the same sentence, and in others
to a person not mentioned there. The problem is that, because we were not
informed about what cannot occur, our childhood experience provided no evi-
dence for the “except” clause(s), the cases in which the reduced form is impos-
sible or where a pronoun and a noun in the same sentence may not corefer. That
is, we had evidence for generalizations like “is may be pronounced [z]”1 and
“pronouns may refer to an individual named by a noun in the same sentence,”
but no evidence for where these generalizations break down.
As children, we came to know the generalizations and their exceptions, and
we came to this knowledge quickly and uniformly. Yet our linguistic experience
was not rich enough to determine the limits to the generalizations. We call this
the problem of the p o verty of the s tim ulus. Children have no data
which show them that is may not be reduced in some contexts and they have
no data showing that him may not refer to Jay in (2.1d). These two small
illustrations are examples of the form that the poverty-of-stimulus problem
takes in language. It may look as if children are behaving magically, but there
is no magician and magic is no answer.
There are two “easy” solutions to the poverty-of-stimulus problem, but nei-
ther is adequate. One is to say that children do not overgeneralize, because they
are reliable imitators. That is, children do not produce the reduced ’s in the
wrong place or use a pronoun in (2.1d) or (2.2d) wrongly to refer to Jay, be-
cause they never hear language being used in this way. In other words, children
acquire their native language simply by imitating the speech of their elders. We
know this cannot be literally true, because everybody constantly says things that
they have never heard. We express thoughts with no conscious or subconscious
consideration of whether we are imitating somebody else’s use of language.
This is true of the most trivial speech: in saying I want to catch the 3:25 pm
bus, which leaves from outside Borders bookstore, one is using a sentence that
one has almost certainly not heard.
The alternative of saying that we form new sentences “by analogy” with
specific sentences we have heard before simply conceals the problem, because
it does not account for the fact that some possible “analogies” are good and
others are not. Why does the existence of the contracted ’s in Tim’s happy not
provide an analogical foundation for a similar reduced form in Kim’s happier
1 We follow the convention in linguistics of enclosing phonetic representations in square brackets.
than Tim is? Why do the sentences (2.2a–c,e) not provide an analogical basis for
coreference between Jay and he in (2.2d)? The point is that language learners
arrive at certain very specific generalizations, and fail to arrive at certain other
logically possible ones, in ways that cannot be founded on any independent
general notion of induction or analogy.
A variant on this “solution” is to claim that children learn not to say the deviant
forms because they are corrected by their elders. Alas, this view offers no better
insight for several reasons. First, it would take an acute observer to detect and
correct the error. Second, where linguistic correction is offered, young children
are highly resistant and often ignore or explicitly reject the correction. Third,
in the examples discussed, children do not overgeneralize and therefore parents
have nothing to correct; this will become clearer when we discuss experimental
work on young children later in this chapter and in chapter 9.
So the first “easy” solution to the poverty-of-stimulus problem is to deny
that it exists, to hold that the environment is rich enough to provide evidence
for where the generalizations break down. But the problem is real, and this
“solution” does not address it.
The second “easy” answer would be to deny that there is a problem on the
grounds that a person’s language is fully determined by genetic properties. In
that case, there would be nothing to be learned. Yet this answer also cannot
be right, because people speak differently, and many of the differences are
environmentally induced. There is nothing about a person’s genetic inheritance
that makes her a speaker of English; if she had been raised in a Dutch home,
she would have become a speaker of Dutch.
The two “easy” answers either attribute everything to the environment or
everything to the genetic inheritance, and we can see that neither position is
tenable. Instead, language emerges through an interaction between our genetic
inheritance and the linguistic environment to which we happen to be exposed.
English-speaking children learn from their environment that the verb is may
be pronounced [z] or [z], and native principles prevent the reduced form from
occurring in the wrong places. Likewise, children learn from their environment
that he, his, etc are pronouns, while native principles entail that pronouns may
not refer to a preceding noun under specific circumstances. The interaction of
the environmental information and the native principles accounts for how the
relevant properties emerge in an English-speaking child.
We will sketch some relevant principles below. It is worth pointing out that
we are doing a kind of Mendelian genetics here, in the most literal sense. In
the mid-nineteenth century, Mendel postulated genetic “factors” to explain the
variable characteristics of his pea plants, without the slightest idea of how these
factors might be biologically instantiated. Similarly, linguists seek to identify
information which must be available independently of experience, in order for
a grammar to emerge in a child. We have no idea whether this information is

encoded directly in the genome or whether it results from epigenetic, devel-
opmental properties of the organism; it is, in any case, native. As a shorthand
device for these native properties, we shall write of the linguistic genotype,
that part of our genetic endowment which is relevant for our linguistic develop-
ment. Each individual’s genotype determines the potential range of functional
adaptations to the environment (Dobzhansky 1970, p. 36), and we assume that
the linguistic genotype (what linguists call Universal Grammar or UG) is uni-
form across the species (apart from fairly severe and specific pathology). That
is, linguistically we all have the same potential for functional adaptations and
any of us may grow up to be a speaker of Catalan or Hungarian, depend-
ing entirely on our circumstances and not at all on variation in our genetic
make-up.
It is important to understand that UG in this sense is not to be confused
with the grammar of any particular language: to say that would be close to the
second fallacious approach to the problem of the poverty of the stimulus which
we discussed above. Rather, UG can be seen as the set of principles by which the
child can infer, on the basis of the limited data available in the environment, the
full grammatical capacity which constitutes a mature speaker’s (unconscious)2
knowledge of a language.
Since children are capable of acquiring any language to which they happen
to be exposed between infancy and puberty, the same set of genetic principles
which account for the emergence of English (using genetic now in the ex-
tended sense we have indicated) must also account for the emergence of Dutch,
Vietnamese, Hopi, or any other of the thousands of languages spoken by hu-
man beings. This plasticity imposes a strong empirical demand on hypotheses
about the linguistic genotype; the principles postulated must be open enough
to account for the variation among the world’s languages. The fact that people
develop different linguistic capacities depending on whether they are brought
up in Togo, Tokyo, or Toronto provides a delicate tool to refine claims about
the nature of the native component.
We conclude that there is a biological entity, a finite mental organ, which
develops in children along one of a number of paths. The range of possible paths
of language growth is determined in advance of any childhood experience. The
language organ that emerges, the grammar, is represented in the brain and plays
a central role in the person’s use of language. We have gained some insight
into the nature of people’s language organs by considering a wide range of
phenomena: the developmental stages that young children go through, the way
language breaks down in the event of brain damage, the manner in which people
2 Needless to say, not everything we want to say a person “knows” is accessible to direct reflection
and articulation. This is a point which has been made at great length in the literature of philosophy,
psychology, and diverse other fields, and which we will not attempt to justify here.
2.2 The nature of grammars 23
analyze incoming speech signals, and more. At the center is the biological notion
of a language organ, a grammar.
2.2 The nature of grammars

Children acquire a productive system, a grammar, in accordance with the re-
quirements of the genotype. If asked to say quite generally what is now known
about the linguistic genotype, we would say that it yields finite grammars, be-
cause they are represented in the finite space of the brain, but that they range
over an infinity of possible sentences. Finite grammars consist of a set of oper-
ations which allow for infinite variation in the expressions which are generated.
The genotype is plastic, consistent with speaking Japanese or Quechua. It is
modular, and uniquely computational.
By modular we mean that the genotype consists of separate subcomponents
each of which has its own distinctive properties, which interact to yield the
properties of the whole. These modules are, in many cases, specific to lan-
guage. The grammar is one such module. Another may be a distinct “parser”
which interacts with the grammar to assign structures and meanings to incom-
ing speech signals, and thereby captures our capacity to understand spoken
language in real time. Within the grammar, one module of innate linguistic
capacity contains abstract structures which are compositional (consisting of
units made up of smaller units) and which fit a narrow range of possibilities.
These structures reflect computational operations which encompass the ability
to relate one position to another by movement (or what we treat as copying and
deletion below, p. 24). Another module is the mental lexicon, a list of word-
forms and their crucial properties. Research has undermined the notion that the
mind possesses only general principles of “intelligence” which hold of all kinds
of mental activity.
These modules may or may not be separately represented in neural tissue:
for example, Grodzinsky (2000) has recently argued that movement relations –
and not other aspects of syntactic form – are computed by specific tissue within
the classical Broca’s area. We return to these matters in chapter 10: the claim of
modularity does not in any sense rest on such physical separation, however. It
refers, rather, to the fact that various aspects of linguistic knowledge are logically
and functionally independent of one another, yielding the full complexity of
human language through the interaction of individually rather simple systems.
To see the kind of compositionality involved, consider how words combine.
Words are members of categories like noun (N), verb (V), preposition (P),
adjective/adverb (A). If two words combine, then the grammatical properties of
the resulting phrase are determined by one of the two words, which we call the
head: we say that the head projects the phrase. If we combine the verb visit with
the noun Chicago, the resulting phrase visit Chicago has verbal and not nominal
properties. It occurs where verbs occur and not where nouns occur: I want to
visit Chicago, but not *the visit Chicago3 nor *we discussed visit Chicago. So
the expression visit Chicago is a verb phrase (VP), where the V visit is the head
projecting the VP. This can be represented as a labeled bracketing (2.3a) or
as a tree diagram (2.3b). The verb is the head of the VP and the noun is the
complement.
(2.3) a. [VP V visit N Chicago]
b. VP
V N
visit Chicago
In general, two categories “merge” to form a new category. So an

“inflectional” (I) element like will might combine with the VP visit Chicago, to
yield the more complex expression will visit Chicago, with a structure [IP I will
[VP visit Chicago]]. The auxiliary, inflectional will, heads the new phrase and
projects to a phrasal category IP. This means that visit Chicago is a unit (VP),
which acts as the complement of will (it completes the meaning of will and
is structurally the sister of will), but will visit is not a unit; that is, there is no
single node which dominates will visit (and nothing else) in this example. The
unit consisting of will and its complement can, in turn, be the basis of yet more
structure in merging with a subject phrase (e.g., [DP D the N student]) to form a
full IP: The student will visit Chicago.
The units defined by these trees are the items which the computational opera-
tions manipulate; they are the items which are copied and deleted, and to which
reference (including coreference) can be assigned. Non-units are not available
to these operations.
One of the computational operations involved is that of copying an ele-
ment already present and merging it with the tree. This accounts for the fact
that in the surface forms of many sentences, elements occur in positions other
than those with which their syntactic function is naturally assigned, i.e., dis-
placed. For example, an expression like What city will the student visit? (where
what city is understood as the complement of visit) can be described by a
structure along the lines of (2.4). Will is copied in a position labeled C (for
“complementizer”), which in such question constructions precedes the rest
of the IP, and what city is copied as a specifier to that head. The copies are
deleted, which we indicate with “e” (for empty) and an index. We take phrases
like what city and the student to be determiner phrases (DP), where what and
the are determiners projecting as DPs; more on this in chapter 3 (see note 1
there).
3 Following a standard convention in linguistics, we indicate phrases or sentences which are not
well-formed in English with a preceding ∗ .
(2.4) CP
DP CP
D NP C IP
what cityj willi
DP IP
D NP I VP
the student [ei]
V DP
visit [ej]
The syntactic component of a speaker’s language organ, then, can be repre-

sented by a system of rules that describes (or “generates”) the set of structures
(similar to that in (2.4)) occurring in sentences of the language; it characterizes
the range of possible structural relations between displaced elements and their
functional positions, etc. Other aspects of a full grammar provide explicit ac-
counts of the relation between this kind of syntactic organization and the way
the words of sentences are constructed and pronounced. We will have more to
say about these various areas of grammar in later chapters. The set of possible
grammars of this type is narrowly limited by the principles of UG, which re-
quire that systems within this range (and no others) are in principle attainable
by the specific principles available to the child, with differences among them
corresponding to differences in the child’s experience.
2.3 Back to the puzzles

Let us return now to the poverty-of-stimulus problems raised above in section
2.1, beginning with that of ’s, the reduced form of is. Recall that this reduced
form is not available in some cases, though it is possible in other, superficially
quite similar, examples.
(2.5) a. I wonder who the professor is/*’s here.
b. I wonder who the professor is/’s talking about now.
Poets make linguistic jokes from these phenomena. The Gershwins were
famous for contraction jokes: in Girl Crazy (1930) a chorus begins I’m bidin’
my time / ’Cause that’s the kind of guy I’m, which is not normal English.
What differentiates the two cases in (2.5), such that the reduced form ’s
is possible in the one instance but not in the other? One potential difference
emerges if we consider the structures of the sentences in (2.5) in more de-
tail. In particular, if we take into account the notion that the element who
has been displaced from its natural position in the sentence, we see that the
position from which this movement takes place differs in the two cases:4
,
(2.6) a. I wonder whoi the professor is/* s ei here.
,
b. I wonder whoi the professor is/ s talking about ei now.
We can notice that in example (2.6a) the verb is/’s immediately precedes
the trace of the displaced element who, while in (2.6b) it does not. The correct
generalization seems to be that reduced forms of auxiliary elements cannot
immediately precede empty elements, such as the trace of displacement in
these cases.
This does indeed describe many of the relevant circumstances under which
reduced auxiliaries are impossible, but it is too narrow to be satisfying. There
is no direct evidence for the situation in which reduced forms are excluded,
so some native principle must be involved, but it seems unlikely that such a
principle as “reduced auxiliary verbs cannot appear before an empty element
such as a trace” forms a distinct part of the human genetic endowment for
language; or rather, to the extent that it does, it must surely be as a by-product
of something more general and principled.
We might attempt to improve on this hypothesis as follows. Note that, in a
sentence like Kim’s happy, the auxiliary element ’s is grammatically the head
of the IP, taking the adjective phrase (AdjP) happy as its complement. In pro-
nunciation, however, it forms part of a single unbroken unit with the preceding
word Kim, as the apostrophe in the conventional spelling ’s suggests, despite
the fact that the sequence Kim+’s does not constitute a phrase syntactically.
In the case that interests us, reduced forms of auxiliaries such as ’s (as well as
’ve, ’re, and the reduced forms of am, will, would, shall, and should) do not have
enough phonological “substance” to be words on their own, and thus necessarily
combine with a word on their left to make up a single phonological word in the
pronunciation of sentences in which they occur. In terms of pronunciation, that
is, Kim’s in Kim’s happy is just as indissoluble a unit as birds in Birds fly.
4 As above, we indicate the source of a copied element with an empty symbol “e,” generally
referred to in the syntactic literature as a t r a c e. The identical subscripts attached to the copied
element and its trace indicate that both have the same reference.
Not every element that functions as a unit from the point of view of the
syntax corresponds to a whole word in pronunciation. Syntactic units that do
not contain enough phonetic material to make up a whole word by themselves,
such as the reduced auxiliaries in English, are referred to as (simple) cli ti cs.
Clitics are little words which occur in many, perhaps all languages, and have
the property of not being able to stand alone.
In some languages, these elements attach systematically to the word to their
left; in others, to the right, and in others the direction of attachment depends
on details of the syntactic and/or phonological structure. This is a property of
particular languages that is plausibly learned from overt data: the child need
only determine that some element is a clitic (in the sense above), and then
see whether it is pronounced indissolubly with the word on its left or on its
right. What is consistently the case, however, is that syntactic elements that
do not constitute words in their own right must attach to some other word as
clitics in order to be pronounced at all. This much plausibly reflects properties
of UG.
The phonological evidence clearly supports the claim that in English, clitics
attach to their left, not their right. This direction of association is supported
by the fact that the pronunciation of the clitic varies from [s] to [z] to [əz] as
a function of the final sound of the preceding word (2.7a); and is completely
insensitive to the shape of the word on its right. This variation in shape is exactly
the same as that which we can observe in the shape of the regular plural ending
(spelled (e)s but pronounced in the same three ways as ’s), again as a function of
the final sound of the preceding word (2.7b). Similarly, while the third person
singular present ending of verbs is always spelled -s, it shows the same variation
in pronunciation (2.7c), as does the ending of possessive forms (2.7d).
(2.7) a. Pat’s ([s]) leaving, Kim’s ([z]) coming in, and Chris’s ([əz])
replacing Jan.
b. packs ([s]), pals ([z]), passes ([əz])
c. infects ([s]), cleans ([z]), induces ([əz])
d. Pat’s ([s]), Kim’s ([z]) (or) Chris’s ([əz]) corkscrew
In Kim’s happy, the element ’s is a clitic in terms of its pronunciation. This
does not alter the fact that in syntactic terms, it is a verb which serves as the head
of the phrase [IP ’s happy], a simple structure which involves no displacement
or deletion. Compare this with the case of the underlined is in Kim’s happier
than Tim is, though. This latter element is also a verb that serves as the head of
a phrase, though what it is followed by within that phrase is something which
is understood but not pronounced. Syntactically, the phrase has the structure
[IP is e], where the unpronounced element e is understood as happy.
Now recall that our problem is to account for the fact that not only are
sentences like (2.8a) impossible in English, but language learners know this
without ever having been explicitly told not to use reduced auxiliary forms in
such positions.
(2.8) a. *Tim’s happier than Kim’s.
b. Tim’s happier than Kim is.
What differentiates the bad sentence (2.8a) from its well-formed near-twin
(2.8b)? The only difference is between the final phrase [IP ’s e], in (2.8a), vs. the
phrase [IP is e], in (2.8b). While reflecting no distinction of syntactic structure or
meaning, this difference does have one consequence: in each of the sentences in
(2.8) the final phrase contains only a single element, but in (2.8a) that element
is a clitic (’s), while in (2.8b) it is the word is.
We already know that a clitic cannot (by definition) constitute a word by
itself. A rather natural part of the theory of sound structure is the prosodi c
h ie r a r c h y (see Nespor and Vogel 1986 for basic discussion of this notion),
according to which utterances are composed of phrases, phrases of words,
words of smaller units called feet, which are themselves composed of syllables.5
It stands to reason that since a clitic by itself cannot constitute a word, a phrase
consisting only of a clitic would contain no word, and would thus be ill-formed.
We might, then, attribute the ill-formedness of (2.8a) in English to the fact that
it contains the ill-formed phrase [IP ’s e].
It may appear that we have played a sort of shell game here, trading on the
fact that the word “phrase” has more than one sense: on the one hand, it refers
to a unit of pronunciation, associated with particular intonational contours,
possibilities for pause, etc. and composed of smaller units of pronunciation,
(phonological) words. On the other hand, however, a “phrase” can be interpreted
as a unit of syntactic organization, composed of units of grammar and meaning
whose actual pronunciation is, for syntactic purposes, quite irrelevant. Thus the
final syntactic phrase in both sentences in (2.8) consists of two elements: a verb
(’s or is) and an adjective (e, interpreted as “happy”). The explanation we are
proposing requires us to say that because [IP ’s e] is syntactically a phrase, its
pronunciation as a phonological phrase would be ill-formed because it would
not contain enough phonological material to make up even a single word. But
what justifies us in the apparent equation of syntactic and phonological notions
of “phrase”?
We must note that syntactic phrasing is not always faithfully reflected in
phonological phrasing. In a sentence like This is the cat, that chased the rat,
that ate the cheese, . . . the boundaries of phonological phrases coincide roughly
with the commas. As a result, that chased the rat is a phonological phrase here,
but not a syntactic phrase: syntactically the phrasing is something like [DP the
cat [IP that chased [DP the rat [IP that ate the cheese . . .]]]].
5 Various authors have proposed slightly different inventories of prosodic constituent types, but
without affecting the overall point here in any material way.
While the correspondence between syntactic and phonological phrase struc-

ture is not a simple isomorphism, neither is it arbitrary and unconstrained. As
we will see below in chapter 9, children are sensitive at a very early age (around
eight to ten months) to the phonological organization of utterances into phrases,
and there is no doubt that they use this structure as a crucial key to discovering
the syntactic organization of sentences and phrases. In order to do this, however,
they must assume that in general a syntactic phrase will correspond to a phono-
logical phrase. If this is the case, things that are true of phonological phrases
will tend also to be true of syntactic ones; and we have seen that a phonological
phrase consisting of some phonetic material, but not enough to make up even a
single well-formed word, is itself ill-formed. From this property of the Prosodic
Hierarchy, together with an assumption that syntactic and phonological phrases
will in general correspond, we can arrive at the following principle:
(2.9) Every syntactic phrase which corresponds to any pronounced
material must contain enough material to make up at least one
phonological word.
The principle in (2.9) is not by itself an independent part of UG, but it
is plausible to suggest that it follows as a theorem from things that are: the
Prosodic Hierarchy and the assumption of a correspondence between syntactic
and phonological constituency.
Armed with (2.9), we can now see the basis for our earlier observation that
reduced forms of the auxiliaries cannot immediately precede empty elements,
as in the ill-formed sentence (2.8a). The problem is not with the empty ele-
ment itself, but rather with the fact that where a phrase consists of an auxiliary
verb together with a following element such as an adjective, and that follow-
ing element is unpronounced, the use of a full (rather than clitic) form of the
auxiliary is necessary to provide the phrase with enough phonetic substance to
satisfy (2.9).
Sentence-internal understood elements can have similar effects, as in I
wonder wherex the concert [IP [IP is ex ] on Wednesday]. Here wherex has moved
from the position indicated as ex , but is still understood there. The verb is
(whether reduced to its clitic form ’s or not) has to be combined with a word
that stands in a certain kind of structural relationship to it: it forms a phrase
together with its complement. This phrase is subject to (2.9), and so if the com-
plement does not provide a word (because it has been deleted), the auxiliary
must: ’s itself would not suffice to supply a well-formed phonological corre-
spondent to a syntactic phrase, and so does not tolerate a deletion site to its
immediate right that would exhaust the phrase of which ’s is the head.
On Wednesday is not the complement of is in this example: rather, it is the
deleted wherex . In chapter 3, we argue that this kind of deletion results from
incorporation of a copied word into an adjacent word. The non-occurrence of *I
wonder who the professor’s here, *I wonder where the concert’s on Wednesday,
*Kim’s happier than Tim’s, etc., suggest that clitics like ’s cannot host a deleted
(understood) item in this way.
How do these facts interact with the observation we made above that clitic
elements are in fact attached to material on their left, forming part of a phrase
with that material in actual pronunciation? Such “migration” of clitic material
from one phrase to another does not actually alter the force of (2.9). In (2.10a),
for example, happy is the complement of is and therefore reduced ’s may at-
tach to the preceding word without leaving the phrase it heads phonologically
deficient. The same is true for the first is of (2.10b). However, Tim is not the
complement of the underlined is in (2.10b); in this case, the subject Tim and
the copula verb is have permuted. As a result, the underlined is is the only overt
representative of its phrase, and cannot be reduced.
(2.10) a. Kim’s happy.
b. Kim is (/’s) happier than is(/*’s) Tim.
But while the principle in (2.9) is plausibly derivable from general require-
ments that are part of UG, and a consequence of (2.9) is that a reduced is may
not be the only phonological material representing the syntactic phrase of which
it is the head, this is not the whole story. Consider the following examples:
(2.11) a. John, my dear’s, a bastard.
b. *John’s, my dear, a bastard.
c. He is/*’s too going to fix it.
d. Fred is tired of Spinoza, just as Mary is/*’s of Schopenhauer.
e. She’s a better scientist than he is/*’s an engineer.
In all of these cases, the reduced auxiliary is impossible in circumstances
where the syntactic phrase it heads contains other material that ought to satisfy
(2.9). There is, however, a property that characterizes the sentences in (2.11):
the fact that each involves a construction with special intonation. In each of the
bad cases, the material immediately following the auxiliary is/’s is set off as
(part of ) a separate (phonological) phrase in pronunciation.
At this point, we know enough to provide an account of the circumstances
under which reduced auxiliaries are impossible. Suppose that we start with the
syntactic structure of the sentence, and then identify the corresponding pro-
nunciation, in terms of phonological phrases. Simplifying somewhat, we can
say that the most natural phrasing is one that mirrors the syntax: syntactic
phrases correspond to phonological phrases. Some constructions, however, are
exceptions to this in that they enforce a phonological phrase boundary in a
place where one might not be motivated syntactically. These include the vari-
ous parenthetical insertions, emphases, and ellipses illustrated in (2.11). Once
phonological phrases have been delimited, we can say that any such phrase that
contains some phonological material, but not enough to constitute at least one
phonological word, is ipso facto ill-formed. Since clitics do not qualify as full
phonological words, any phrase consisting of a clitic alone will suffer such a
fate. In actual pronunciation, finally, prosodically deficient elements (simple
clitics) are attached phonologically to the word on their left.
Though fairly intricate, this account provides an answer to the problem
sketched at the outset: how to account for the fact that the child comes to know
when the reduced forms of the auxiliary elements are or are not possible, in the
absence of direct evidence. In order to achieve this, the child needs to learn that
in English (a) the elements am, is, has, etc. have simple clitic alternants ’m, ’s,
etc.; (b) clitics attach to their left; and (c) certain constructions (parenthetical
insertions, emphasis, ellipses) get phrased in a special way phonologically. All
of these facts are directly attested in the linguistic data available to the child,
and can thus be learned without difficulty.
But these observable phenomena interact with principles of UG, which are
part of the child’s initial linguistic endowment, and thus do not have to be
learned, principles which are not language-particular but rather apply to all
languages: (a) notions of the phonological word and phrase, together with the
requirement that phonological phrases must be made up of at least one phono-
logical word; and (b) the preference for parallel structure in syntax and in
prosodic organization. Jointly these will entail that a syntactic phrase with
a clitic element such as ’s as its only phonological content will be excluded,
without requiring the child to have overt evidence for the badness of such ex-
amples, as we have seen above.
Part of what a child growing a grammar needs to do is to determine the clitics
in his or her linguistic environment, knowing in advance of any experience that
“clitics” are small, unstressed items attached phonologically to an adjacent word
in ways that may be contrary to the syntactic relations they bear to surrounding
material. This predetermined knowledge – the nature of clitics and the fact that
they cannot by themselves satisfy the requirement that phrases be represented
by at least one phonological word – is contributed by the linguistic genotype
and is part of what the child brings to language acquisition. The environment
provides examples such as Pat’s happy, Bob’s happy, and Alice’s happy too.
The child can observe that the three instances of ’s in these cases vary in their
pronunciation ([s] after Pat, [z] after Bob, and [əz] after Alice). This variation
is quite systematic, and, as noted, follows the same principles as those that
determine the form of the plural ending in cats, knobs, palaces among other
endings in English. These facts confirm that ’s must be a clitic, and must attach
phonologically to its left.
Under this approach, the child is faced with a chaotic environment and in
scanning it, identifies clitics . . . among many other things, of course (Lightfoot
1999). This is the answer that we provide to our initial problem, and it is an
answer of the right shape. It makes a general claim at the genetic level (“clitic”
is a predefined category) and postulates that the child arrives at a plausible
analysis on exposure to a few simple expressions. The analysis that the child
arrives at predicts no reduction for the underlined is in Kim is happier than Tim
is, I wonder who the professor is here, and countless other cases, and the child
needs no correction in arriving at this system. The very fact that ’s is a clitic,
a notion defined in advance of any experience, dictates that it may not occur
in certain contexts. It is for this reason that the generalization that is may be
pronounced as ’s breaks down at certain points and does not hold across the
board.
Consider now the second problem, the reference of pronouns. An initial
definition might propose that pronouns refer to a preceding noun, but the data
of (2.1) and (2.2) showed that this is both too strong and too weak. It is too strong
because, as we saw, in (2.1d) him may not refer to Jay; in (2.1b) him may refer
to Jay but not to Jay’s brother. The best account of this complex phenomenon
seems to be to invoke a native principle which says, approximately (2.12).
(2.12) Pronouns may not refer back to a higher nominal element
contained in the same clause or in the same DP.
In (2.13) we give the relevant structure for the corresponding sentences of
(2.1). In (2.13b) the DP Jay’s brother is contained in the same clause as him
and so him may not refer back to that DP: we express this by indexing them
differently. On the other hand, Jay is contained inside the DP and is not “higher”
than him, so those two nouns do not need to be indexed differently – they may
refer to the same person and they may thus be coindexed. Again we see the
constituent structure illustrated earlier playing a central role in the way in which
the indexing computations are carried out. In (2.13d) Jay is in the same clause
as him and so the two elements may not be coindexed; they may not refer to the
same person. In (2.13c) Jay is not contained in the same clause as he: Jay and
he may thus refer either to the same person or to different people. In (2.13a) his
is contained inside a DP and may not be coindexed with anything else within
that DP; what happens outside the DP is not systematic; so his and Jay may
corefer and do not need to be indexed differently.
(2.13) a. [IP Jayi hurt [DP hisi/j nose]]
b. [IP [DP Jayi ’s brother]k hurt himi/j/∗k ]
c. [IP Jayi said [IP hei/j hurt Ray]]
d. [IP Jayi hurt himj/∗i ]
The idea that pronouns refer to a preceding noun is shown to be too weak
because sometimes, as in (2.2c,e), the pronoun refers to a following noun. In
this case, the relevant principle seems to be that such “backwards anaphora” is
not possible if the pronoun not only precedes the noun, but is also “higher” (in
a precise sense whose details are not relevant to our present concerns) in the
syntactic structure than the noun which is to serve as its antecedent.6 In (2.2c),
the pronoun precedes Jay, but this is acceptable because the pronoun appears
within a subordinate clause, and thus is not relevantly “higher.” In (2.2e), the
pronoun is subordinated by virtue of appearing as a possessor within a larger
DP. In (2.2d), however, the pronoun appears as subject of the main clause, and
is thus (in the relevant structural sense) syntactically higher than the following
noun, which therefore cannot serve as its antecedent.
We could have illustrated these points equally well with data from French or
from Dutch, or from many other languages, because the principles apply quite
generally, to pronouns in all languages. If we assume a native principle, available
to the child independently of any actual experience, language acquisition is
greatly simplified. Now the child does not need to “learn” why the pronoun
may refer to Jay in (2.13a) or (2.13b,c) but not in (2.13d); in (2.2a–c,e) but
not in (2.2d), etc. Rather, the child raised in an English-speaking setting has
only to learn that he, his, him are pronouns, i.e. elements subject to Principle B
(see note 6). This can be learned by exposure to a simple sentence like (2.1c)
(structurally (2.13c)), uttered in a context where he refers to Jay; that suffices to
show that he is neither an anaphor nor a name – the other possible noun types,
according to most current views – and hence must be a pronoun.7
One way of thinking of the contribution of the linguistic genotype is to view it
as providing invariant principles and option-points or p a rameters. There are
invariant principles, such as that clitics attach phonologically to adjacent words
by virtue of their prosodically “weak” character; that phonological phrases are
based on words, which are based on smaller prosodic units; that phonological
and syntactic phrases are generally related in a particular way; that pronouns
cannot be locally coindexed and that names may not be coindexed with a higher
DP, etc. Taken together, these have consequences, such as the principle in (2.9)
which requires phrases that are pronounced at all to contain at least one full
phonological word. Meanwhile, there are also options: direct objects may pre-
cede the verb in some grammars (German, Japanese) and may follow it in others
(English, French); some constructions may have special intonation associated
with them; clitics in some grammars attach to the right and in others to the
left, etc. These are parameters of variation and the child sets these parameters
one way or another on exposure to particular linguistic experience. As a re-
sult a grammar emerges in the child – a language organ, part of the linguistic
6 The principle alluded to in the last paragraph, that pronouns may not be locally coindexed, is
Principle B of the binding theory; (2.12) is only an informal (and partly inaccurate) rendering of
Principle B. Here we allude to Principle C, that names may not be coindexed with a higher DP
anywhere. In chapter 9 we discuss the acquisition of Principle C.
7 Anaphors are elements that are locally coindexed, according to Principle A of the binding theory,
while names, by Principle C, are never coindexed with a higher DP.
phenotype. The child has learned that ’s is a clitic and that he is a pronoun;
the genotype ensures that ’s cannot be the only phonological material within a
syntactic phrase and that he is never used in a structurally inappropriate context.
2.4 The acquisition problem

In the preceding sections we have looked at some specific acquisition problems
and considered what ingredients are needed for their solution. Now let us stand
back and think about these matters more abstractly.
The child acquires a finite system, a grammar, which generates structures
which correspond more or less to utterances of various kinds. Some structural
principle prevents forms like *Kim’s happier than Tim’s from occurring in
the speech of English speakers, as we have seen. Children are not exposed to
pseudo-sentences like this or informed systematically that they are not to be
produced. Speakers come to know subconsciously that they cannot be said and
this knowledge emerges somehow, even though it is not part of the environ-
mental input to the child’s development. It is not enough to say that people do
not utter such forms because they never hear them: people say many things
that they have not heard, as we have noted. Language is not learned simply by
imitating or repeating what has been heard.
2.4.1 The poverty of the stimulus

This poverty-of-stimulus problem defines our approach to language acquisi-
tion. Since the late 1950s, much of the linguistic literature has focused on areas
where the best description cannot be derived directly from the data to which the
child has access, or is underdetermined by those data, as in the examples with
the clitic ’s and the pronouns discussed above. If the child’s linguistic experi-
ence does not provide the basis for establishing a particular aspect of linguistic
knowledge, another source must exist for that knowledge.
This is not to say that imitation plays no role, but only that it does not provide
a sufficient explanation. This is worth emphasizing, because antagonists some-
times caricature this approach to language acquisition as “denying the existence
of learning,” when in fact we merely deny that learning is the whole story, a very
different matter. The quoted phrase comes directly from a remarkable article
in Science (Bates and Elman 1996), in which the authors assert that “Noam
Chomsky, the founder of generative linguistics, has argued for 40 years that
language is unlearnable” and that they, on the other hand, have “rediscovered”
learning!
Caricatures of this type show up in the writing of people who claim that all in-
formation is derived from the environment and that there is no domain-specific
genetic component to language acquisition. These people deny the poverty-of-
stimulus problems, claiming that children may derive all relevant information
from their linguistic environment. Bates and Elman provide a recent and partic-
ularly striking instance of this line, claiming that artificial neural networks can
learn linguistic regularities from imperfect but “huge computerized corpora of
written and spoken language.”
Nobody denies that the child must extract information from the environment;
it is no revelation that there is “learning” in that technical sense. After all,
children learn to speak one way when surrounded by speakers of French and
another when surrounded by speakers of Italian. Our point is that there is more
to language acquisition than this. Children react to experience in accordance
with specific principles.
The problem demanding explanation is compounded by other factors. Despite
variation in background and intelligence, people’s mature linguistic capacity
emerges in fairly uniform fashion, in just a few years, without much apparent
effort, conscious thought, or difficulty; and it develops with only a narrow
range of the logically possible “errors.” Children do not test random hypotheses,
gradually discarding those leading to “incorrect” results and provoking parental
correction. In each language community the non-adult sentences formed by very
young children seem to be few in number and quite uniform from one child
to another, which falls well short of random (see chapter 9). Normal children
attain a fairly rich system of linguistic knowledge by five or six years of age and
a mature system by puberty. In this regard, language is no different from, say,
vision, except that vision is taken for granted and ordinary people give more
conscious thought to language.
These, then, are the salient facts about language acquisition, or more properly,
language growth. The child masters a rich system of knowledge without sig-
nificant instruction and despite an impoverished stimulus; the process involves
only a narrow range of “errors” and takes place rapidly, even explosively be-
tween two and three years of age. The main question is how children acquire
so much more than they experience and how the growth takes place.
A grammar, the language organ, represents what a speaker comes to know,
subconsciously for the most part, about his or her native language. It represents
the fully developed linguistic capacity, and is therefore part of an individual’s
phenotype. It is one expression of the potential defined by the genotype. Speak-
ers know what an infinite number of sentences mean and the various ways in
which they can be pronounced and rephrased. Most of this largely subconscious
knowledge is represented in a person’s grammar. The grammar may be used for
various purposes, from everyday events like expressing ideas, communicating,
or listening to other people, to more contrived functions like writing elegant
prose or lyric poetry, or compiling and solving crossword puzzles, or writing a
book about the language organ.
We do not want to give the impression that all linguists adopt this view
of things. In fact, people have studied language with quite different goals in
mind, ranging from the highly specific (to describe Dutch in such a way that it
can be learned easily by speakers of Indonesian), to more general goals, such

as showing how a language may differ from one historical stage to another
(comparing, for example, Chaucerian and present-day English). However, the
research paradigm we sketch, which construes a grammar as a biological object,
the language organ, has been the focus of much activity since the late 1950s
and it seems to us to provide the most satisfying approach to the attempt to
understand the fundamental nature of language.
2.4.2 The analytical triplet

A grammar, for us, is a psychological entity, part of the psychological state of
somebody who knows a language. For any aspect of linguistic knowledge, three
intimately related items are included in a full account of this state. First, there is
a formal and explicit characterization of what a mature speaker knows; this
is the grammar, which is part of that speaker’s phenotype. Since the grammar
is represented in the mind/brain, it must be a finite system, which can relate
sound and meaning for an infinite number of sentences.
Second, also specified are the relevant principles and parameters common
to the species and part of the initial state of the organism; these principles and
parameters make up part of the theory of grammar or UG, and they belong to
the genotype.
The third item is the trigger experience, which varies from person to person
and consists of an unorganized and fairly haphazard set of utterances, of the
kind that any child hears.8 The universal theory of grammar and the variable
trigger together form the basis for attaining a grammar.
In (2.14) we give the explanatory schema, using general biological terminol-
ogy in (2.14a) and the corresponding linguistic terms in (2.14b). The triggering
experience causes the genotype to develop into a phenotype; exposure to a
range of utterances from, say, English allows the UG capacity to develop into a
particular mature grammar. One may think of the theory of grammar as making
available a set of choices; the choices are taken in the light of the trigger expe-
rience or the primary linguistic data (PLD), and a grammar emerges when the
relevant options are resolved. A child develops a grammar by setting the open
parameters of UG in the light of her particular experience.
(2.14) a. linguistic triggering experience (genotype −→ phenotype)
b. primary linguistic data (Universal Grammar −→ grammar)
Each of the items in the triplet – trigger, UG, and grammar – must meet
various demands. The trigger or PLD must consist only of the kinds of things that
children routinely experience and includes only simple structures (see chapter 9
8 The notion of a trigger is from ethologists’ work on the emergence of behavioral patterns in
young animals.
for discussion). The theory of grammar or UG is the one constant and must
hold universally such that any person’s grammar can be attained on the basis
of naturally available trigger experiences. The mature grammar must define an
infinite number of expressions as well-formed, and for each of these it must
specify at least the sound and the meaning. A description always involves these
three items and they are closely related; changing a claim about one of the items
usually involves changing claims about the other two.
The grammar is one subcomponent of the mind, a mental organ which inter-
acts with other cognitive capacities or organs. Like the grammar, each of the
other organs is likely to develop in time and to have distinct initial and mature
states. So the visual system recognizes triangles, circles, and squares through
the structure of the circuits that filter and recompose the retinal image (Hubel
and Wiesel 1962). Certain nerve cells respond only to a straight line sloping
downward from left to right within a specific, narrow range of orientations;
other nerve cells to lines sloped in different directions. The range of angles that
an individual neuron can register is set by the genetic program, but experience
is needed to fix the precise orientation specificity (Sperry 1968).
In the mid-1960s David Hubel, Torsten Wiesel, and their colleagues devised
an ingenious technique to identify how individual neurons in an animal’s vi-
sual system react to specific patterns in the visual field (including horizontal
and vertical lines, moving spots, and sharp angles). They found that particular
nerve cells were set within a few hours of birth to react only to certain visual
stimuli, and, furthermore, that if a nerve cell is not stimulated within a few
hours, it becomes totally inert in later life. In several experiments on newborn
kittens, it was shown that if a kitten spent its first few days in a deprived optical
environment (a tall cylinder painted only with vertical stripes), only the neurons
stimulated by that environment remained active; all other optical neurons be-
came inactive because the relevant synapses degenerated, and the kitten never
learned to see horizontal lines or moving spots in the normal way.
We see learning as a similarly selective process: parameters are provided by
the genetic equipment, and relevant experience fixes those parameters (Piattelli-
Palmarini 1986, 1989). A certain mature cognitive structure emerges at the ex-
pense of other possible structures which are lost irretrievably as the inactive
synapses degenerate. The view that there is a narrowing down of possible con-
nections out of an overabundance of initially possible ones is now receiving
more attention in the light of Hubel and Wiesel’s Nobel Prize winning success.
On the evidence available, this seems to be a more likely means of fine tuning
the nervous system as “learning” takes place than the earlier view that there is
an increase in the connections among nerve cells.
So human cognitive capacity is made up of identifiable properties that are
genetically prescribed, each developing along one of various preestablished
routes, depending on the particular experience encountered during the individ-
ual’s early life. These genetic prescriptions may be extremely specialized, as
Hubel and Wiesel showed for the visual system. They assign some order to our
experience. Experience elicits or triggers certain kinds of specific responses but
it does not determine the basic form of the response.
This kind of modularity is very different from the view that the cognitive fac-
ulties are homogeneous and undifferentiated, that the faculties develop through
general problem-solving techniques. In physical domains, nobody would sug-
gest that the visual system and the system governing the circulation of the blood
are determined by the same genetic regulatory mechanisms. Of course, the pos-
sibility should not be excluded that the linguistic principles postulated here may
eventually turn out to be special instances of principles holding over domains
other than language, but before that can be established, much more must be
known about what kinds of principles are needed for language acquisition to
take place under normal conditions and about how other cognitive capacities
work. The same is of course true for other aspects of cognitive development.
Only on such a basis can meaningful analogies be detected. Meanwhile,
we are led to expect that each region of the central nervous system has its own special
problems that require different solutions. In vision we are concerned with contours and
directions and depth. With the auditory system, on the other hand, we can anticipate a
galaxy of problems relating to temporal interactions of sounds of different frequencies,
and it is difficult to imagine that the same neural apparatus deals with all of these
phenomena . . . for the major aspects of the brain’s operation no master solution is likely.
(Hubel 1978, p. 28)
2.4.3 Real-time acquisition of grammars

In the domain of language, there is good evidence that the sophisticated dis-
tinctions that we discussed at the beginning of this chapter do not result from
learning and that the hypothesized genetic (or epigenetic: recall our earlier
extension of this term) constraints seem to be at work from the outset. Exper-
imenters have constructed situations in which the over-riding temptation for
children would be to violate the relevant constraints. The fact that children con-
form to the hypothesized constraints, resisting preferences they show in other
contexts, shows that the constraints under investigation are active for them,
and that this is true at the earliest stage at which they might be manifested
(Crain 1991).
Stephen Crain and Rosalind Thornton (1998) developed an elicitation task
that encouraged children to ask questions like *Do you know what that’s up
there, if these were compatible with their grammars. They hypothesized that
children would generally show a preference for the reduced ’s form whenever
this was consistent with their grammars. This preference would be revealed in
a frequency count of legitimate forms, like Do you know what that’s doing up
there? Comparing the frequency of the reduced forms in the contexts where
adults find reduced forms unacceptable with that of non-adult reduced forms
more generally would indicate whether or not children’s grammars contained
the hypothetical genetic constraint. If the genetic constraint is at work, there
should be a significant difference in frequency; otherwise, not.
The target productions were evoked by the following protocols, in which
Thornton and Crain provided children with a context designed to elicit
questions.
(2.15) Protocols for cliticization:
a. Experimenter: Ask Ratty if he knows what that is doing
up there.
Child: Do you know what that’s doing up there?
Rat: It seems to be sleeping.
b. Experimenter: Ask Ratty if he knows what that is up there.
Child: Do you know what that is up there?
Rat: A monkey.
In (2.15a), the child is invited to produce a sentence where what is understood

as the object of doing: Do you know whatx that is doing ex up there? Since
the resulting phrase of which is is head, [IP is [VP doing ex ]], contains at least
one phonological word in addition to is itself, is can be replaced with the clitic
form ’s without resulting in an ill-formed correspondence between syntactic
and phonological structure. However, in (2.15b), the child produces a sentence
where what is understood as the complement of is, but is not pronounced in
that position: Do you know whatx that is ex up there? (cf. That is a bottle up
there). Here a deletion site is the only other component of the phrase headed
by is. As a result, that phrase ([IP is ex ]) only corresponds to a phonological
word to the extent that is itself is a word – and not merely a clitic. This fact
prevents the is from cliticizing in adult speech; no adult would use the reduced
form to produce *Do you know what that’s up there? (cf. That’s a bottle up
there).
Crain and Thornton found that young children behaved just like adults, mani-
festing the hypothetical genetic constraint. The children tested ranged in age
from 2 years, 11 months, to 4 years, 5 months, with an average age of 3 years,
8 months. In the elicited questions there was not a single instance of the reduced
form where it is impossible in adult speech. Children produced elaborate forms
like those of (2.16), but never with ’s, the reduced form of is.
(2.16) a. Do you know what that black thing on the flower is?
(4 years, 3 months)
b. Squeaky, what do think that is? (3 years, 11 months)
c. Do you know what that is on the flower? (4 years, 5 months)
d. Do you know what that is, Squeaky? (3 years, 2 months)
There is, of course, much more to be said about grammars and their ac-
quisition, and we will return to this topic in chapter 9 below. There is also
an enormous technical literature, but here we have briefly illustrated the kind
of issue that work on real-time acquisition can address under our I-language
approach.
2.5 Conclusion
Recent theoretical developments have brought an explosive growth in what
we know about human languages. Linguists can now formulate interesting hy-
potheses and account for broad ranges of facts in many languages with elegant
abstract principles, as we shall see. They understand certain aspects of lan-
guage acquisition in young children and can model some aspects of speech
comprehension.
Work on human grammars has paralleled work on the visual system and has
reached similar conclusions, particularly with regard to the existence of highly
specific computational mechanisms. In fact, language and vision are the areas of
cognition that we know most about. Much remains to be done, but we can show
how children attain certain elements of their language organs by exposure to
only an unorganized and haphazard set of simple utterances; for these elements
we have a theory which meets basic requirements. Eventually, the growth of
language in a child will be viewed as similar to the growth of hair: just as hair
emerges at a particular point in development with a certain level of light, air, and
protein, so, too, a biologically regulated language organ necessarily emerges
under exposure to a random speech community.
From the perspective sketched here, our focus is on grammars, not on the
properties of a particular language or even of general properties of many
or all languages. A language (in the sense of a collection of things people
within a given speech community can say and understand) is on this view an
epiphenomenon, a derivative concept, the output of certain people’s grammars
(perhaps modified by other mental processes). A grammar is of clearer status:
the finite system that characterizes an individual’s linguistic capacity and that
is represented in the individual’s mind/ brain, the language organ. No doubt the
grammars of two individuals whom we regard as speakers of the same language
will have much in common, but there is no reason to worry about defining “much
in common,” or about specifying precise conditions under which the outputs of
two grammars could be said to constitute one language. Just as it is unimportant
for most work in molecular biology whether two creatures are members of the
same species (as emphasized, for example, by Dawkins 1976), so too the notion
of a language is not likely to have much importance if our biological perspec-
tive is taken and if we explore individual language organs, as in the research
program we have sketched here and which we elaborate in later chapters.
3 Syntax
3.1 The emergence of syntax within linguistics

Before the development of generative grammar in the late 1950s, linguists
focused almost entirely on the smallest units of language: sounds, minimal
meaningful elements (“morphemes” like ed, ful, con – see chapter 7 below
for more on this notion), and words, where the model of the Saussurian sign
has most plausibility. “Syntax” was largely a promissory note to the effect
that such sign-based analysis would eventually encompass the larger units of
phrases, sentences, etc. Meanwhile, what went by that name was largely a kind
of applied morphology: some instructions for what to do with the various kinds
of words (inflected and otherwise).
For example, drawing from our bookshelves more or less at random, we find
that Morris Jones’ (1913) comprehensive grammar of Welsh is divided into two
sections, phonology and accidence (inflectional properties), and has nothing
under the rubric of syntax. Arthur MacDonnell’s (1916) grammar of Vedic
Sanskrit has two chapters on sounds, four chapters on inflections, and a final
chapter entitled “Syntax”. There he has some observations about word order and
agreement phenomena, and then a discussion of the uses of cases, tenses, and
moods. He notes that the subjunctive mood has a fundamental sense of “will”
and lists the uses of the subjunctive mood in main clauses, relative clauses, and
with “relative conjunctions.” Joseph Wright’s grammar of the Gothic language
(1910) also has one final chapter on syntax, in which he illustrates the use
of the various cases, the agreement properties of adjectives, some uses of the
pronouns, verb voices, tenses, and moods. These works are entirely typical.
One could think of languages as made up from a small inventory of sounds,
a larger inventory of morphemes, and a lexicon containing the words. Those
inventories might be transmitted from one generation of speakers to another
and language acquisition could be seen as a process of children acquiring the
relevant inventories. That is not implausible for these small units, although
the view turns out to be inadequate for reasons that we shall discuss in the
next few chapters; and there was no productive way to think of syntax or
of larger units in that fashion. Certainly children do not just acquire a set of
41
42 3 Syntax
sentences as they become mature speakers. Perhaps they acquire some set of
basic structures and construction-types, which may then be elaborated into some
open-ended set, and there were efforts to build models along those lines. But that
involved positing some kind of system, an open-ended, recursive system, and the
available models did not have that capacity. Certain “typological” approaches,
inspired by Greenberg (1963), adopted E-language formulations in identifying
harmonic properties of languages which have “predominant subject–object–
verb” word order, for example, asking to what extent such languages would
show, say, noun–adjective order.
When the productive mechanisms of syntactic structure came under serious
scrutiny, with the development of generative grammar, it became apparent that
even though a description of the words of a language is certainly necessary, it is
not sufficient: languages cannot be viewed simply as inventories of signs. The
resulting insights had profound effects in all areas of the field. There has been
a tremendous amount of work over the last forty years, yielding discoveries in
language after language and theories which apply productively to wide ranges
of phenomena in many languages. Since pregenerative work on syntax was so
scant, there is little to be learned from a comparison. Rather, in this chapter we
shall take two features of current models – deletion and Case theory – and we
shall show some of the ramifications of these ideas, how they are shaped by
the cognitive, I-language nature of our analyses, how they capture details about
language structure, remembering that God is in the details, and the devil, too.
The details are fascinating in themselves; they represent distinctions which
are not reported in standard language textbooks, they are not taught to second-
language learners, and, indeed, for the most part they were not known until
rather recently, when they were discovered by theoreticians. There is no way
that these distinctions could be communicated directly to young children as
they develop their language capacity. However, the distinctions we shall dis-
cuss are not the object of our inquiry, but data which provide evidence about
the inner mechanisms of the mind. Our problem is to pick one or two more
or less self-contained illustrations of those mechanisms, parts of the internal
systems represented in individual brains and acquired under normal childhood
conditions.
The earliest attempts to carry out the program of generative grammar quickly revealed
that even in the best studied languages, elementary properties had passed unrecognized,
that the most comprehensive traditional grammars and dictionaries only skim the sur-
face. The basic properties of languages are presupposed throughout, unrecognized and
unexpressed. This is quite appropriate if the goal is to help people to learn a second
language, to find the conventional meaning and pronunciation of words, or to have some
general idea of how languages differ. But if our goal is to understand the language
faculty and the states it can assume, we cannot tacitly presuppose “the intelligence of
the reader.” Rather, this is the object of inquiry. (Chomsky 2000, p. 6)
3.2 Successive merger and deletion

We begin by returning to the basic mechanisms for building syntactic structure
that we began to discuss in chapter 2. One way to think of this (developed in
recent “Minimalist” work) is that the mental lexicon contains words belonging
to various categories. They may be drawn from the lexicon and “merged” into
a syntactic structure, which is built bottom-up. To adjust our earlier example:
the verb visit might merge with a determiner phrase1 (DP) what city to yield a
verb phrase (VP) containing a head, the verb visit, and its complement [DP what
city]: [VP V visit [DP what city]]. The inflectional element (I) will now may be
merged with the VP, yielding the inflection phrase (IP) in (3.1a). Subsequently
the determiner phrase the student may be merged with the resulting IP, to yield
another IP in which the DP the student is the specifier (Spec) of the IP already
formed (3.1b); this yields the sentence The student will visit what city? with the
structure of (3.1b–c). The structures may be represented as a labeled bracketing
(3.1b) or as an equivalent tree (3.1c); the linear brackets of (3.1b) and the
tree diagram of (3.1c) carry precisely the same information about constituent
structure. The structures define the units of expressions: what city is a unit but
student will is not; will visit what city is a unit, but not will visit. The units are
the items manipulated by the computational mechanisms; they are the items
which may be copied, deleted or indexed in ways that we shall explore in a
moment.
(3.1) a. [IP I will [VP V visit [DP what city]]]
b. [IP [DP the student][IP I will [VP V visit [DP what city]]]]
c. IP
DP IP
I VP
V DP
the student will visit what city
This kind of merger may continue indefinitely, building structures of arbitrary

size through simple recursive devices. All languages may draw on three kinds of
recursive devices which enable one to take any sentence and make it longer: co-
ordination (Reuben and Phil and Fred and . . . went fishing and caught lobsters
1 DPs are headed by a determiner like the or what, which in turn is followed by a noun phrase. This
yields structures like [DP D the [NP N student [PP P of physics]]], [DP D what [NP city]], [DP [SpecDP
Kim] [DP D ’s [NP tall mother]]], or [DP D we [NP linguists]].
44 3 Syntax
and . . .), complementation (Reuben said that Phil said that Fred thought
that . . .), and relativization (This is the cow that kicked the dog that chased the
cat that caught the mouse that ate the cheese that . . .).2 It is always possible to
construct a more complex sentence.
Instead of considering how the language capacity deals with the (literal)
infinity of possible structures, let us explore the property of “displacement”:
words and phrases may occur in displaced positions, positions other than those
corresponding to their interpretation. In an expression What did you see in
Berlin?, what is understood as the direct object of see (cf. We saw the Reichstag
in Berlin), but it has been displaced. It is pronounced at the front of the sentence
but understood in the position of a direct object, to the right of the verb. English-
speaking children hear sentences like these and consequently the displacement
operation is learnable; children learn from experience that a wh-phrase typically
occurs at the front of its clause, even though it is understood elsewhere. Chinese
children have no such experiences and their grammars have no comparable
displacement. They hear questions like “You saw what in Berlin?” in Chinese,
where (the equivalent of) “what” is pronounced in the same position in which
it is understood.
Displacement can be viewed as a special case of merger: an item already
in the structure is copied and merged, with subsequent deletion of the copied
element. There are other ways of viewing the displacement effect, of course,
but let us pursue this account. For an example, let us continue with the structure
of (3.1b). The I element will, which is already present in the structure, might
now be copied and merged with (3.1b) to yield the more complex (3.2a), with
a complementizer phrase (CP) headed by the copied will; the lower will is
subsequently deleted to yield (3.2b). Now another step: the DP what city, the
direct object of visit, may be copied and merged, yielding (3.2c), (where what
city is the specifier of the CP), with subsequent deletion to yield (3.2d), the
structure of the sentence What city will the student visit?
(3.2) a. [CP C will [IP [DP the student] [IP I will [VP visit what city]]]]
b. [CP C will [IP [DP the student] [VP visit what city]]]
c. [CP [SpecCP what city] [CP C will [IP [DP the student][VP visit
what city]]]]
d. [CP [SpecCP what city][CP C will [IP [DP the student][VP visit]]]]
We shall focus here on a condition for deleting elements after they have
been copied: the manner in which elements may be deleted limits the way that
2 It has been suggested by Hale 1976 that in some (Australian aboriginal) languages, the functions
of relative clauses are actually filled by a kind of “adjoined” structure which is more like a form
of coordination than it is like the embedded relative clauses of, e.g., English. This does not
compromise the point that in these languages, too, sentences can be arbitrarily extended by the
introduction of the functional equivalents of English relative clauses.
elements may be displaced, as we shall see. We shall postulate a Universal

Grammar (UG) condition on deletion, which will interact with what children
learn about English to distinguish the kinds of sentences that one might hear
from those which do not occur.
Let us be clear on the logic of the enterprise here. English-speaking children,
unlike their Chinese-speaking counterparts, learn from their environment that
“wh-phrases” (as we will call question phrases like who, when, what city, etc. –
even though Chinese “wh-phrases” do not involve the sound wh) are displaced
to the front of their clause. We want an analysis which entails that that is
all they need to learn from the environment in this regard. In fact, there are
many interesting distinctions involving wh-expressions, as we are about to see.
We seek an analysis, however, in which those distinctions are not themselves
learned directly from the environment, since much of what we will discuss is not
explicitly present in the child’s experience and thus could not be learned in this
way. Rather, these aspects of the knowledge that develops – the language organ –
follow from such relatively simple, directly attested facts as the clause-initial
position of English wh-expressions in combination with what is contributed by
general genotypic properties, or UG.
3.2.1 Delete that

First, observe a very simple instance of deletion: English complementizers or
“sentence introducers” like that may be deleted in structures like (3.3).3
(3.3) a. It was apparent [CP that/e Kay had left].
b. The book [CP that/e Kay wrote] arrived.
c. It was obvious [CP that/e Kay had left].
This operation does not apply in languages like Dutch and French, where the
complementizers dat and que are invariably present. Nonetheless, the English-
specific operation is learnable, because children typically hear sentences in
both forms, sometimes with the complementizer present, sometimes not: It was
apparent that Kay had left/It was apparent Kay had left. Therefore an operation
“Delete that” meets the basic requirements for inclusion in the grammars of
English speakers: children have evidence for it. (3.3a) is the structure for the
sentence It was apparent (that) Kay had left, (3.3b) for The book (that) Kay
wrote, and (3.3c) for It was obvious (that) Kay had left.
However, as with virtually every grammatical operation, we find that some
aspects are not determined by normal childhood experience. In particular,
generalizations typically break down and the point at which they break down
cannot be determined by the data available to children; we shall see many
3 As in chapter 2, “e” indicates an empty element, in these structures a deleted that.

46 3 Syntax
examples over the next few pages. This matters if grammars are elements of
individual cognition, emerging in people’s brains on exposure to some linguis-
tic experience: we need to tease apart what can plausibly be learned from the
environment and what cannot. Our current generalization, “Delete that,” breaks
down in that certain instances of that may not be deleted. Consider (3.4), where
only the structure with that occurs in speech: *It was apparent to us yesterday
Kay had left, *The book arrived Kay wrote, *Kay had left was obvious to all of
us, *Fay believes, but Ray doesn’t, Kay had left (a boldface “*e” indicates an
illicit empty item, a place where that may not be deleted).
(3.4) a. It was apparent to us yesterday[CP that/*e Kay had left]

b. The book arrived [CP that/*e Kay wrote].
c. [CP that/*e Kay had left] was obvious to all of us.
d. Fay believes, but Ray doesn’t, [CP that/*e Kay had left].
This is a standard poverty-of-stimulus problem. Since children are not told

that a sentence like *Kay had left was obvious to all of us does not occur
(corresponding to (3.4c), with no that), principles of UG are implicated. The
fact that that is deletable can be derived directly from experience, since children
hear sentences in both forms with and without a sentence-introducing that, but
the fact that that may not be deleted in the structures of (3.4) has no comparable
basis in experience, since children are exposed only to utterances which actually
occur and they are not supplied with lists of utterances which do not occur, as
we discussed in chapter 2. Therefore knowledge of the limits to a generalization
like “Delete that,” of the exceptions where it does not apply, must come from
elsewhere; this knowledge is not derivable from the external world and it must
reflect internal properties.
All of this represents subconscious knowledge that every speaker of English
has. To know English is to know where that may, may not, or must appear: to
know the distribution of that. No one taught us these things, and the distinctions
we have just noted will come as news to nearly all of our readers who have not
studied syntax. Some of this knowledge is learnable in the sense that it can be
derived from normal childhood experience, but much is simply not accessible
in this way.
The difference between (3.3) and (3.4) is that in (3.3) that is the top-most
element in a clause (CP) which is the complement of an adjacent, overt word. A
complement is something which completes the meaning of a word, and the CP
is the complement of apparent in (3.3a), book in (3.3b), and obvious in (3.3c),
and in these contexts that may be deleted. In (3.4) the embedded clause [CP that
Kay had left/wrote] is the complement of apparent in (3.4a), book in (3.4b),
nothing in (3.4c), and believes in (3.4d). In no case is the CP the complement
of the adjacent word in (3.4) and therefore that may not be deleted. Evidently,
that may be deleted only if its clause is the complement of an adjacent, overt
word.
Somehow this limitation must be derived from intrinsic, native properties; it
cannot be a direct consequence of environmental information, because children
are not informed about what does not occur. This is crucial. “Delete that”
is a very simple generalization, easily learnable by children on the basis of
normal experience, but we see that a UG principle is implicated; we need some
sort of internal principle to prevent the system from over-generating, yielding
structures like those of (3.4) with that deleted, which do not occur in normal
speech. Let us formulate the principle in question as (3.5).
(3.5) UG condition: an element may be deleted if it is the top-most

item in a complement of an overt, adjacent word.
3.2.2 Deleting copies

Returning to the deletion of copied elements in the structure-building opera-
tions, we now see that the identical condition holds for deleted copies, marked
henceforth as an indexed “e.” In (3.6a), who is deleted in the position marked as
ei , where it is the complement of see. In (3.6b), who is copied first at the top of
its local CP and then at the top of the higher CP;4 in each case, the lower element
may be deleted, because it is the complement of an adjacent, overt word saw
or the top-most item in the complement of say. These are the structures for the
well-formed sentences Who did Jay see? and Who did Jay say that Fay saw?
(3.6) a. Whoi did Jay see ei ?

b. Whoi did Jay say [CP ei that Fay saw ei ]?
Our UG condition on deletion (3.5) makes lots of nice distinctions, which

we shall illustrate, but let us again be clear on the logic. The idea is that there
are simple generalizations that children can learn from their environment, for
example, “Delete that” or “Delete a copied wh-phrase.” However, these gen-
eralizations break down, and the points at which they break down cannot be a
function of external, environmental information, because there is none. There-
fore we need to formulate an appropriate internal condition which interacts
with the learned generalizations so that the system does not over-generate to
produce the non-occurring (3.4), but yields only the kinds of sentences which
actually occur. Condition (3.5) suffices for (3.4), and more.
4 It is crucial that movement is local: a wh-phrase moves first to the front of its own clause. From
there it may move on to the front of the next clause up.
48 3 Syntax
Consider (3.7a), which illustrates a common-or-garden conjoined sentence.

Here the second verb may be unpronounced, as we see from (3.7b). In (3.7b),
we say that there is a “gapped” verb, which is understood to be present in the
position indicated but not pronounced.
(3.7) a. Jay introduced Kay to Ray and Jim introduced Kim to Tim.
b. Jay introduced Kay to Ray and Jim gap Kim to Tim.
c. *Which mani did Jay introduce ei to Ray and which womanj
(did) Jim gap *ei to Tim?
d. *Jay wondered whati Kay gave ei to Ray and whatj Jim (did)
gap *ej to Tim.
e. *Jay admired ei greatly [DP his uncle from Paramus]i but
Jim (did) gap *ej only moderately [DP his uncle from New
York]j .
f. Jay gave his favorite racket to Ray and Jim gap his favorite
plant to Tim.
g. *Jay gave ei to Ray [DP his favorite racket]i and Jim (did)
gap *ej to Tim [DP his favorite plant]j .
The fact that verbs may be gapped in this way in all (as far as we know) forms
of English is readily learnable: children hear sentences like Jay went to Rome
and Fay to Paris, Jay introduced Kay to Ray and Jim Kim to Tim, and so on,
and they can infer from their experience that verbs may be omitted in conjoined
sentences. This is not a universal fact and we are told that such gapping does
not occur in Chinese; children in Beijing do not have the requisite experiences.
The system as we have it now predicts (as a result of (3.5)) that, where
gapping occurs, a wh-phrase cannot be deleted to the right of the gap (see
3.7c,d) if the wh-phrase is the complement of the gapped verb but the verb is
not overt. Sentences corresponding to these structures fail to occur: *Which
man did Jay introduce to Ray and which woman (did) Jim to Tim? (3.7c), *Jay
wondered what Kay gave to Ray and what Jim to Tim (3.7d). Similarly with
the non-wh-words displaced in (3.7e,g); (3.7e,g) illustrate a generalization that
we shall return to below, to the effect that a “large” DP may be displaced to
the far right of its clause. This principle is learnable on exposure to sentences
like Jay admired greatly his uncle from Paramus, Jay introduced to Fay all
the students from Los Angeles. Our condition on deletion prevents this learned
generalization from over-generating to yield (3.7e,g), and we seem to have the
right kind of interaction between intrinsic and acquired properties.5
5 There are various kinds of conjoined structures, and there is more to be said – in fact, much
more – about ellipsis operations. For example, while the sentence corresponding to (3.7c) is
ill-formed, one encounters sentences like Which man did Jay introduce to Ray and Jim to Kim?
and Which man did Jay introduce to Ray and which woman to Tim? We have no intention of
providing a comprehensive account of ellipsis operations here. Our goal is rather to illustrate
The same point holds for a deleted that to the right of a gapped verb (3.8b)
and a deletion at the front of an embedded clause (3.8c). (3.8a) is a well-formed
structure yielding the normal, everyday sentence Jay thought Kay hit Ray and
Jim that Kim hit Tim. However, (3.8b,c) are ill-formed; the deleted that in
(3.8b) and the deletion at the front of the indicated CP in (3.8c) fail to meet
our condition, not being in a complement of an adjacent, overt word; so the
structures do not occur, nor do the corresponding sentences *Jay thought Kay
hit Ray and Jim Kim hit Tim (3.8b), *Who did Jay think Kay hit and who (did)
Jim (that) Kim hit? (3.8c).
(3.8) a. Jay thought Kay hit Ray and Jim gap [CP that Kim hit Tim].
b. *Jay thought Kay hit Ray and Jim gap [CP *e Kim hit Tim].
c. *Whoi did Jay think Kay hit ei and whoj (did) Jim gap [CP
*ej (that) [CP Kim hit ej ]]?
Our UG condition (3.5) captures these distinctions with a plausible bifur-
cation between internal properties and information which is learned from the
environment. Children learn from their environment that a sentence-introducing
that may be deleted, that a copied wh-phrase may be deleted, that verbs may
be omitted in conjoined sentences, that large DPs may be displaced to the far
right of their clause. Our internal UG condition (3.5) guarantees that these
learned operations do not over-generate to yield non-occurring structures and
their corresponding sentences.
There are more subtleties holding of the speech of every mature speaker of
English, which follow from this particular condition of UG, making deletion
possible only in certain structures. Again, these are distinctions known sub-
consciously to every speaker of English: to know English means having these
distinctions, even though they were not taught to us and they could not be
derived entirely from our direct experience.
A simple possessive phrase like Jay’s picture is three ways ambiguous. Jay
might be the owner or the painter of the picture, or the person portrayed, i.e.
the object. Traditional grammarians label the last of these readings an “objec-
tive genitive.” Following the traditional intuition, modern linguists say that the
structure for the reading in which Jay is the object, the person portrayed, is
(3.9): Jay is copied from the complement position, where it is understood, to
the possessive position, where it is pronounced. Its copy is deleted in the usual
fashion, where it is adjacent to and the complement of the noun picture.
(3.9) a. [DP Jayi ’s [NP picture ei ]]
But now consider The picture of Jay’s. Here the ambiguity is different and
the phrase is only two ways ambiguous. It means that Jay is the owner or the
some poverty-of-stimulus problems and the kind of reasoning that is involved in solutions to
them.
50 3 Syntax
painter of the picture, but not the object: the expression cannot refer to a picture
in which Jay is portrayed, somewhat surprisingly. Similarly in A picture of the
Queen’s, the Queen can be the owner or the painter but not the person portrayed.
This is something that most adults are not aware of: indeed, it was not observed
until recently and certainly is not something we impart explicitly to children.
Again, a condition of UG must be involved and it is our condition on deletion
(3.5). Possessive elements like Jay’s only occur introducing DPs and therefore
must be followed by an NP even if nothing is pronounced (see note 1). If
nothing is pronounced, then the noun heading the NP must be null (e). So The
picture of Jay’s, where Jay is the owner or the painter, would have the well-
formed structure of (3.10a), where the indicated NP is empty and understood as
“picture.” If The picture of Jay’s were to have the impossible reading where Jay
is the person portrayed, the “objective genitive,” then the structure would be
(3.10b) and the copied element to be deleted (ei ) is the complement of another
empty element, the same empty noun understood as “picture.” The deletion of
the copied Jay is illicit because there is no adjacent overt word (cf. 3.9). Similarly
for The picture is Jay’s (3.10c) and The picture which is Jay’s (3.10d), which
also lack the objective genitive reading for Jay and whose structures involve an
illicit deletion.
(3.10) a. The picture of [DP Jayi ’s [NP e]]

b. *The picture of [DP Jayi ’s [NP e *ei ]]
c. *The picture is [DP Jayi ’s [NP e *ei ]]
d. *The picture which is [DP Jayi ’s [NP e *ei ]]
Data like these are not the input to language acquisition; rather, they emerge
from the properties of the system which is triggered by simple, accessible
data available in children’s everyday experience and now we understand how.
Children hear expressions like Jay’s picture, understanding it to mean “picture of
Jay,” and thereby learn that English, unlike many other languages, has displaced
objective genitives; the UG condition then blocks structures like (3.10b,c,d).
For a further illustration of the utility of our UG condition (3.5), the deviant
(3.11c) has an illicit deletion right-adjacent to the gapped verb. Because the
verb is a gap, deletion of he i there (the top-most element of the complement)
is illicit, as is now familiar. That is the only relevant difference from (3.11a),
which is perfectly comprehensible and straightforward, involving no ill-formed
deletion. (3.11b) is also well-formed: here there is no gapped verb and known
licenses the deleted copy.
(3.11) a. It is known that Jay left but it isn’t gap that he went to the
movies.
b. Jay is known [ei to have left] but hei isn’t known [ei to have
gone to the movies].
c. *Jayi is known [ei to have left] but hei isn’t gap [*ei to have
gone to the movies].
Before we change gear, consider one last distinction. The sentence corres-
ponding to structure (3.12a), The crowd is too angry to organize a meeting, is
ambiguous: the understood subject of the embedded clause, written here as “e,”
may refer back to the crowd (the “anaphoric” reading, indicated by coindexing),
or it may be arbitrary in reference, meaning “anybody” (indicated by the “arb”
index): either the crowd is going to organize the meeting or somebody else,
unspecified.
(3.12) a. The crowdi is too angry [CP ei/arb to organize a meeting].

b. Whati is the crowdj too angry [CP ei [ej to organize ei ]]?
The ambiguity corresponds to an ambiguity of structure: the anaphoric reading

occurs if the CP is the complement of angry; while if the CP is not a complement,
but some kind of adjunct (where the CP is a sister not to angry but to some
higher phrasal category), then the embedded subject is taken to be arbitrary in
reference. Take this on faith (we cannot argue it here) and see what follows.
The interrogative sentence corresponding to (3.12b), What is the crowd too
angry to organize?, is not ambiguous in the same way and allows only the
anaphoric reading, in which the understood subject is the crowd; it cannot be
understood to mean that some other, unspecified person is to organize some-
thing. This is because what originates as the complement of organize and moves
first to the front of the embedded CP, and then to its final position. Two instances
of what must be deleted, in the positions indicated. The fact that what is deleted
at the front of the embedded clause indicates that the clause is the complement
of angry and not an adjunct: if it were not the top-most item in the complement
of angry, it would not be deletable (as a result of (3.5)). As a result the under-
stood subject must be anaphoric, ej . It is inconceivable that children learn from
experience that structures like (3.12b) do not have the ambiguity of (3.12a).
This can only be a function of the general properties of the language organ, as
on our analysis.
Our UG condition on deletion does a lot of work; it enables us to distinguish
many well-formed and deviant structures, and to do so in such a way that we can
offer plausible stories about what is learned and what is not learned by children
who will become speakers of some form of English. Children learn that there
is deletion of sentence-introducing thats, displacement of wh-phrases, copied
objective genitives, and right-moved large DPs; our UG condition on deletion
(3.5) then makes the distinctions we have discussed. Notice, by the way, that we
have been focusing on a property of our model, deletion, and not on phenomena
from some language; in fact, the deletion cuts across many sentence types and
constructions. The mathematical model we are working with defines topics of
52 3 Syntax
research, as is usual in scientific work, and phenomena of the kind that we have
discussed constitute evidence for the properties of the model.
3.2.3 Incorporating deleted items

Let us now take things a step further. One way of thinking of deletion in the
light of our discussion so far would be to take the structural position left empty
by a deleted item to be incorporated, clitic-like, into an adjacent phonological
host of which it is the complement. So in (3.12b) the lowest ei incorporates into
the verb organize; the higher ei incorporates into angry; however, in (3.11c) the
boldface *ei cannot incorporate into an adjacent overt word, because there is not
an overt word in an adjacent position. In (3.7c), ei incorporates into introduce,
but *ej has no adjacent overt word to incorporate into. And so on. We will
pursue this in a moment.
This perspective generalizes to some of the phenomena involving the reduced
’s, ’re, ’ve, etc., discussed in chapter 2. If these reduced forms are cliticized to
their left, as we suggested, then we can understand why the second is in Tim’s
happier than Kim is may not be reduced (*Tim’s happier than Kim’s): at some
level the structure for such a sentence contains Tim is happier than Kim is happy
and happy deletes only if it has a full phonological host to its left. If reduced
is (’s) is itself a clitic, then it cannot host the deleted happy. Similarly for
I wonder whati that is ei up there, where the lower what could not be deleted if
there were no host to its left and therefore is must occur in its full, unreduced
form (compare the non-existent *I wonder what that’s up there).
We can now unify these phenomena by saying that the reduced is is not
an appropriate host for a deletion site: since it is itself a clitic, it is not a full
phonological word, as we argued in chapter 2, and not an appropriate host for
a deleted item. Similarly, one finds the simple The concert’s on Wednesday,
which is well formed because there is no deleted element requiring a host, but
not *I wonder where the concert’s on Wednesday, since the structure would
be . . . wherei the concert’s ei on Wednesday and there would be no host for
ei .6 The facts discussed in chapter 2 concerning the distribution of reduced
forms of is, am, have, etc. thus interact with another story, reflexes of our UG
condition (3.5).
Note that the total picture involving the reduced auxiliaries is not provided
by either part of our account alone. Thus, at least some examples of the types
6 Here is a nice distinction: Max is dancing in London and Mary is in New York is ambiguous: Mary
might be dancing in New York or perhaps running a restaurant. However, the same sentence with
a reduced is is unambiguous: Max is dancing in London and Mary’s in New York means only
that Mary is in New York, not necessarily dancing there. If Mary were dancing in New York,
the structure would be . . . and Mary is [e] in New York and the empty verb would require a full
phonological word as its host, for which the reduced ’s would not suffice.
represented by (2.11) do not involve deletion, and so (3.5) would have nothing
to say about them. On the other hand, some examples involving displacements
(and thus deletion) would not fall under (2.9) either:
(3.13) Tiredi as he [IP is/*’s [AdjP ei of his job at the car-wash]], Fred
won’t go looking for something better.
In many examples involving deletion affecting the entire complement of an
auxiliary, the two conditions converge to rule out the reduction of auxiliaries,
but each has a separate role to play in other examples.
If we adopt this view, then the notion of a host for a deletion site illumi-
nates more distinctions, which have to do with the extractability of subjects of
embedded clauses. Many years ago Joan Bresnan postulated a Fixed Subject
Constraint to capture the observation that subjects of finite embedded clauses
seemed to be fixed, unmovable. Now we can dispense with a distinct Fixed
Subject Constraint and relate the cases it was intended to account for to what
we have discussed in this section, explaining them through our UG condition
on deletion.
We noted that English embedded clauses are introduced by a complementizer
which may or may not be pronounced; so, sentences corresponding to (3.14a)
occur with and without that. This is also true if a wh-item is copied from
the embedded object position (3.14b): that may or may not be present. The
deleted complementizer in (3.14a) incorporates into the adjacent verb thought,
as indicated. Similarly the deleted wh-word at the front of the (embedded)
clause in (3.14b) incorporates into the adjacent think, whether or not that is
present. The same goes for the deleted wh-word which is the complement of
saw in (3.14b). In each case, the deleted element is the top-most item of the
host’s complement.
(3.14) a. I thought [CP that/e [IP Ray saw Fay]].

b. Whoi did you think [CP ei that/e [IP Ray saw ei ]]?
c. *Whoi did you think [CP *ei that [IP *ei saw Fay]]?
However, a who in an embedded subject position may not incorporate into an
adjacent that (the boldface *ei in 3.14c): *Who do you think that saw Fay? That
cannot host the deleted item: incorporation is not possible and the deletion is
illicit (hence the boldface).7
The same is true of indirect questions introduced by a word like how: a direct
object incorporates into the adjacent verb solved in (3.15b) (Which problem do
you wonder how John solved?), but a subject wh-word may not incorporate into
7 The higher element, at the front of the clause, poses no problems, incorporating into the higher
verb, think in (3.14c), wonder in (3.15a,b), etc.
54 3 Syntax
how (3.15a), the ill-formed structure for the non-occurring sentence *Who do
you wonder how solved the problem? It cannot be the case that complementizers
like that and how are not appropriate hosts for deletion sites because they are not
full, phonological words in some sense, because (apart from the phonologically
unmotivated nature of this move) the same is also true for expressions like what
time in indirect questions like (3.15c), *Who were you wondering what time
finished the exam?
(3.15) a.*Whoi do you wonder [CP ei how [IP *ei solved

the problem]]?
b. Which problemi do you wonder [CP ei how [IP John
solved ei ]]?
c.*Whoi were you wondering [CP ei what time [IP *ei finished
the exam]]?
d. Which exami were you wondering [CP ei what time
[IP Kim finished ei ]]?
In these cases, what blocks the incorporation of the deletion site into the
preceding word is the requirement in (3.5) that only the top-most element in
the complement of a word can incorporate into that word. This suggests either
that *ei in (3.15a,c) is not the top-most element in the embedded finite clause
(perhaps the finite verb, the bearer of the clause’s inflectional marking, is its
structural head) or else that the clauses [IP *ei solved the problem] (3.15a) and
[IP *ei finished the exam] (3.15c) are not properly the complements of how and
what time, respectively. We leave this open here.
A wh-word also may not be deleted in subject position if its clause (an
“indirect question” in traditional terms) is part of a relative clause. In (3.16b)
the wh-word has been copied from an underlying object position (where the
deleted who is the complement of and incorporated into bought) and the result
is a more or less normal sentence: This is the sweater which I wonder who
bought. However, (3.16a) is impossible, where who cannot be deleted in subject
position; [IP ei bought ej ] is not the complement of the adjacent what. The
sentence corresponding to (3.16a) would be the impossible *This is the student
who I wonder what bought.
(3.16) a. *This is the student whoi I wonder [CP whatj [IP *ei bought ej]].
b. This is the sweater whichi I wonder [CP who bought ej ].
Now consider another limitation on the deletion of subjects, but here in a

simple, unembedded clause. As we mentioned in connection with (3.7), English
has an operation whereby a “large” DP may occur in a displaced position at

the far right of its clause. In (3.17a) the moved element, All the students from
LA, is understood as the complement of the verb introduced and it may be
deleted in its original position, incorporated into the adjacent verb. In (3.17b)
the moved element is in a subject position, but the subject of a non-tensed or
“infinitival” clause; it is the top-most item in the complement of the verb expect
and may be incorporated. Crucially, unlike the case of finite embedded clauses
(which, like other finite clauses, must have subjects), a subject associated with
an embedded non-finite verbal expression is not structurally subordinate and
can incorporate. However, the generalization breaks down at a certain point:
the sentence corresponding to (3.17c) sounds very un-English (*Are unhappy
all the students from LA), and the problem is that the deletion site is neither a
complement nor the top-most element in any complement.
(3.17) a. I introduced ei to Mary [DP all the students from LA]i

b. I expect [CP ei to be at the party] [DP all the students
from LA]i
c. *[ *ei are unhappy] [DP all the students from LA]i
So if we take deletion to involve incorporation into an appropriate host word,

this move has the effect of blocking it where such incorporation is impos-
sible, thus preventing the copying of wh-phrases from the subject of tensed
clauses. This blocks the generation of ill-formed structures like those of (3.14c),
(3.15a,c), (3.16a), and (3.17c), while permitting the well-formed structures. This
enables us to understand a wide range of data, as we have sketched.
3.2.4 Understanding some diversity

As the next part of our story, we will see that our condition on deletion (3.5)
also enables us to understand some remarkable diversity among languages.8
The restriction on the deletion of the subjects of tensed clauses apparently
conflicts with a desire/need to ask questions about subjects of tensed clauses,
just as one may ask questions about entities in other structural positions. The
evidence for this claim is that strategies are adopted to circumvent the effects
of this UG principle in certain contexts, and these strategies are manifested
quite differently in individual languages. Because they vary so much, these
individual strategies obviously do not reflect genotypical principles directly, but
they do exist and they take on such diverse forms because of our genotypical
principles.
8 As more languages have been examined with carefully worked out models, it has become clear
that earlier work had underestimated not only their complexity and the extent to which they are
determined by initial conditions, but also their diversity (see Chomsky 2000, p. 7).
56 3 Syntax
We have argued for our UG condition (3.5) here entirely on the basis of
poverty-of-stimulus facts from English, but the condition holds at the level of
UG, and therefore of all grammars; and we expect to be able to argue for the
condition from the perspective of any language showing displacement prop-
erties. Indeed, we can and probably should make those arguments, but to do
so would exhaust the tolerance of our publisher if not our readers, so we will
demonstrate some consequences for linguistic diversity more succinctly. Rather
than give comparable arguments from several languages, we shall illustrate one
effect as it is manifested in various languages.
Luigi Rizzi (1990) identified three kinds of strategies used in different lan-
guages to circumvent the UG ban on extracting the subject of a tensed clause,
here subsumed under our condition on deletion (3.5). Each strategy employs an
ad hoc, learned device which licenses extraction from a subject position. The
particular devices are quite different from language to language, and our UG
condition on deletion helps us understand that diversity.
(3.18) Three strategies to license an extracted subject:
a. Adjust the complementizer so as to license the extraction
b. Use a resumptive pronoun in the extraction site
c. Move the subject first to a non-subject position and then
extract
English exploits strategy (3.18a) and permits extraction of a subject if the
complementizer that is adjusted – in fact, not present, as in Who do you think
saw Fay?, which has the structure (3.19). Recall that who originates as the
subject of saw. Because it is a wh-expression, it is copied at the front of its
local clause, and then at the front of the upstairs clause. So two copies need
to be deleted in the positions indicated. The lowest deletion (the subject) is
licensed by the higher coindexed (agreeing) position at the front of the CP (and
is incorporated into the coindexed position),9 and the higher deletion at the front
of the CP is licensed by and incorporated into the verb think (being the top-
most item in think’s complement). In the comparable (3.14c) and (3.15a), there
was no host for the deleted item. In other words, subjects of tensed clauses in
English are movable only if the CP contains only an empty, unpronounced, co-
indexed or “agreeing” item: that permits a subject wh-word to be incorporated
9 Example (3.16a) illustrates the importance of indexing, a complex matter. Indexed elements are
effectively overt. This can be seen in a language where verbs are displaced. French Qui voyez-
vous? “Who do you see?” has the structure (i), where ej is the deletion site of voyez and hosts
ei , the deleted qui.
(i) [CP quii voyezj [IP vous ej ei ]]

By contrast, the English “gapped” verbs in (3.7) and (3.8) do not result from displacement
and, therefore, behave differently.
appropriately.10 That and how, on the other hand, in these positions are not
appropriate hosts, as we saw in connection with (3.14c) and (3.15a).
(3.19) Whoi did you think [CP ei [CP ei saw Fay]]?

French also exploits strategy (3.18a), but somewhat differently. The com-
plementizer que is never deleted (3.20a). Again we see that objects may be
extracted freely: Qui crois-tu que Marie a vu?Qui originates as the direct ob-
ject of vu, is copied at the front of its local clause, and then at the front of the
upstairs clause (3.20b). The trace of the lowest qui is incorporated into vu and
the higher one is incorporated into crois, as indicated. But a subject is not ex-
tractable in a comparable way and que, like English that, cannot host a deleted
subject: *Qui crois-tu que a vu Jean? (3.20c). However, French speakers can
adjust the complementizer que to the “agreeing” form qui, if and only if it is
followed by a deletion: Qui crois-tu qui a vu Jean? Effectively the agreeing
form hosts the deleted element (3.20d).
(3.20) a. Je crois [CP que [IP Marie a vu Jean]].

I think that Marie has seen Jean
I think that Marie has seen Jean.
b. Quii crois-tu [CP ei que [IP Marie a vu ei ]]?
who think you that Marie has seen
Who do you think Marie has seen?
c. *Quii crois-tu [CP ei que [IP *ei a vu Jean]]?
d. Quii crois-tu [CP ei qui [IP ei a vu Jean]]?
Here we see a very specific, ad hoc device, in this case an operation changing
que to qui, whose sole motivation is to permit deletion of a subject DP. In French
the agreeing complementizer is an overt qui, while in English the comparable
form is the deleted, unpronounced complementizer.
Rizzi identified similar devices in a variety of languages, which host deleted
subjects. West Flemish (a language like German and Japanese and unlike
English and French, where direct objects precede their verb; hence the right-
ward cliticization of the direct object in (3.21a)) behaves similarly to French:
the usual form of the complementizer is da (3.21a) but a special “agreeing”
form die occurs where a deleted subject needs to be hosted (3.21b).
10 In chapter 9 we discuss the course of language acquisition by children who use “medial”
wh-items, as in What do you think what Cookie Monster eats? (Thornton 1995). Such chil-
dren retain medial wh-items longest in contexts where the wh-word is extracted from a sub-
ject position and where it acts as a kind of “agreeing” complementizer: Who do you think
who’s under there?
58 3 Syntax
(3.21) a. Den vent dai Pol peinst [CP ei da Marie ei getrokken heet]
the man that Pol thinks that Marie photographed has
The man that Pol thinks that Marie has photographed
b. Den vent dai Pol peinsti [ CP ei die ei gekommen ist]
the man that Pol thinks that come is
The man that Pol thinks has come
Norwegian is similar to West Flemish: it shows a special complementizer

som only in embedded questions with a subject deleted (3.22a); its function
seems to be to host a deleted subject which otherwise would violate our UG
condition. It never occurs with a deleted object (3.22b).
(3.22) a. Vi vet [CP hvemi som/*e [CP ei snakker med Marit]].

we know who that talks with Mary
We know who talks with Mary.
b. Vi vet [CP hvemi (*som) [CP Marit snakker med ei ]].
we know who that Mary talks with
We know who Mary talks with.
In these four languages we see copied subjects being deleted if the comple-
mentizer is adjusted in some fashion: deleted in English, amended to an agree-
ing form in French, or with a special form in West Flemish and Norwegian.
In all cases there is a kind of agreement. Hebrew is a bit different. Hebrew also
does not allow deletion of a subject DP (3.23a), although objects extract freely
(3.23b), as is now familiar. Subjects are extractable under special circumstances.
A special device adjusts the complementizer, in this case cliticizing the com-
plementizer še onto an adjacent head (3.23c). In (3.23c) the complementizer
cliticizes rightward onto lo, vacating the complementizer position and permit-
ting the subject to be incorporated leftwards, as in the analysis of (3.23d) (see
Shlonsky 1988). Because the complementizer position has been vacated, the
subject is extractable; this is reminiscent of the English device of emptying the
complementizer position not by cliticization but by deletion.
(3.23) a.*Mii ein ex joda’at [CP ’im *ei mešaret bamilu’im]?

who not you know whether serves in reserves?
Who do you not know whether (s/he) serves in the
reserves?
b. Et mii ein ex joda’at [CP ’im hamilu’im me’aifim ei ]?
acc+ who not you know whether the reserves tire
Who do you not know whether the reserves tire him/her?
c. Mi at ma’mina [CP šelo ohev salat xacilim]?

who you believe that not likes salad eggplants
Who do you believe doesn’t like eggplant salad?
d. Mii at ma’mina [CP e[IP ei še+lo ohev salat xacilim]]?
The second general strategy is to use a “resumptive” pronoun in positions

where incorporation is not possible (3.18b). Swedish exploits this strategy:
(3.24a) shows that a complementizer does not host a deleted subject, as is now
familiar; instead, the resumptive pronoun det appears and so there is no dele-
tion, no empty element to be licensed. On the other hand, if no complementizer
is present, the resumptive pronoun does not appear and the deleted item incor-
porates first into the empty complementizer and then into the verb of which its
clause is the complement (3.24b), more or less along the lines of English and
Hebrew, exploiting strategy (3.18a) in this instance (Engdahl 1985, p. 40).
(3.24) a. Vilket ordi visste ingen [CP hur det /*ei stavas]?
which word knew no one how it/e is spelled
Which word did no one know how it is spelled?
b. Kalle i kan jag sla vad om [CP ei /*han kommer att klara sig].
Kalle can I bet about e/he is going to succeed
Kalle, I can bet (*he) is going to succeed.
The West African language Vata adopts the same strategy, but here even for
local movement in a simple, unembedded clause. Again we see the familiar
subject–object asymmetry: an extracted subject has a resumptive pronoun in its
underlying position, never a deletion (3.25a), while the opposite is true for an
extracted object (3.25b). To express English Who ate rice?, one says “Who did
he eat rice?,” with a resumptive pronoun in the subject position, and not “Who
ate rice?”; to express English What did Kofi eat?, interrogating the direct object,
one says “What Kofi ate?,” with no resumptive pronoun (the lower “what” is
incorporated into its verb le), and not “What Kofi ate it?” The resumptive
pronoun is used only where a wh-word may not be incorporated.
(3.25) a. Álói *(òi ) le saká la?
who (he) eat rice wh
Who ate rice?
b. Yii Kòfı́ le (*mı́i ) la?
what Kofi eat (it) wh
What did Kofi eat?
Italian, on the other hand, manifests a third strategy: moving the subject
first to a non-subject position (3.18c). Subjects may occur to the right of the
60 3 Syntax
VP (3.26a): Credo che abbia telefonato Gianni “I think that Gianni has tele-
phoned.” Here wh-words may be incorporated leftward into an adjacent verb,
and that is the position from which they are copied; so (3.26b) is the structure
for a sentence like Chi credi che abbia telefonato? “Who do-you-think has
telephoned?”
(3.26) a. Credo [CP che [VP abbia telefonato Gianni]].

I-think that has telephoned Gianni
I think that Gianni has telephoned.
b. Chii credi [CP che [VP abbia telefonato ei ]]?
who do-you-think that has telephoned
Who do you think has telephoned?
The Arabic dialect of Banni-Hassan employs a similar device. Here subjects

may occur in front of the verb or after it, as in Italian. Indeed, there is a morpho-
logical distinction between a post-verbal subject miin “who” and its preverbal
counterpart min. If the complementizer innu “that” occurs (3.27a), then the
post-verbal subject form is required. In other words, if the complementizer is
present, it cannot host a deleted element (just like English that, French que,
West Flemish da, Hebrew še, Swedish hur, Italian che, etc.) and consequently
the element must move from the post-verbal position, showing the appropriate
morphology; the structure is (3.27a). On the other hand, if the complemen-
tizer is absent, then the subject may be incorporated leftwards, hosted by the
higher verb gaal, as illustrated by English above (3.19), and the preverbal sub-
ject is a possible extraction site; the pronoun shows the appropriate preverbal
morphology (3.27b).
(3.27) a. Miin/*mini Fariid gaal [CP innu [VP kisar ei albeeda]]

who Fariid said that broke the egg
Who did Fariid say broke the egg?
b. Min/*miini Fariid gaal [ei [VP kisar albeeda]]?
We have discussed a bewildering range of facts, but a comprehensible pattern

emerges from the data. UG blocks extraction of subjects, because unlike overt
verbs, nouns, and indexed null elements, complementizers cannot host deleted
items. However, for reasons of expressibility, speakers “need” to extract sub-
jects; that is what the evidence from specific languages shows. Because of the
UG constraint, they are forced to adopt ad hoc strategies which eliminate illicit
deletions (Swedish, Vata), provide a post-verbal alternative to them (Italian,
Banni-Hassan Arabic), or adjust the complementizer in some way so as to
3.3 Case 61
license them through an agreeing form (English, French, West Flemish, Hebrew,
Norwegian).
The UG constraint explains the need for ad hoc, language-specific devices.
Each of the devices we have examined is learnable, assuming children are
prohibited genetically from extracting embedded subjects in the normal case.
That is, children are exposed to positive, accessible data which demonstrate the
language-specific operation that adults use: the deletability of that in English,
the operation changing que to qui in French, the need for a resumptive pronoun
only in subject positions in Swedish and Vata, etc. The conditions under which
these apply follow from our UG condition (3.5). We therefore have accounts for
the specific languages, which meet our basic requirements. We also see that the
consequences of a condition of the linguistic genotype may be circumvented
sometimes in the interest of expressivity, and we understand why there is such
diversity in these cases.
Our UG principle on deletion (3.5), requiring that deleted elements are in-
corporated in a clitic-like fashion into an adjacent, overt head, was postulated
on the basis of poverty-of-stimulus arguments. It provided a way to tease apart
generalizations that a child might induce from her environment and the limits to
those generalizations, preventing them from over-generating and yielding non-
occurring structures. We know that the principle has similar effects in other
languages, but we have not illustrated that here. Instead, we have illustrated
some effects in a variety of languages relating to the extractability of subjects.
That suffices to show that our principle is operative in other languages, al-
though it certainly has many more consequences than we have illustrated here
for analyses of French, West Flemish, and so on.
3.3 Case
Let us turn now to a second aspect of UG, which connects with matters central
for traditional grammarians, the existence of cases. As it turns out, this notion –
familiar to anyone who has studied German, Russian, Latin, Greek, etc. – is
both closely similar to and subtly different from a more abstract relation on
syntactic structure which plays a central role in the theory of UG.
3.3.1 Morphological case

It is a striking property of natural languages that a noun often shows a different
form depending on its function. Modern English manifests these “case” alterna-
tions only in the pronoun system. A pronoun shows one form if it is the subject
of a finite verb (“nominative,” or “nom”), another if it is the complement of a
62 3 Syntax
verb or of a preposition (“accusative,” or “acc”), and another if it acts as the

specifier of a DP (“genitive,” or “gen”) (3.28).
(3.28) a. He saw him in [DP his car].

b. They saw them at [DP their house].
Other languages have much richer morphological systems. Finnish has,

strictly speaking, fifteen different cases, of which nine are local, expressing
location, destination, and source, each of which may appear alone or with
markers for interior and exterior. Source is thus indicated by the suffix tta;
if the phrase indicates “from inside,” the noun may have the elative case
marker sta, and if the phrase indicates “from outside,” the noun may have the
ablative case suffix lta. Other languages have a comitative case indicat-
ing accompaniment, an abessive or privative case to mean “not having”
(Finnish rahtatta is [money+abessive] “without money, moneyless”); some
Australian languages encode the opposite idea in a proprietive case indicat-
ing “having” (Kalkatungu putuyan [stomach+proprietive], means “pregnant”).
Languages may display quite an array of different inflectional endings on nouns.
If one adds all combinations of orientation markers and regular case markers,
one can identify forty or more markings in some Finno-Ugric and Northeast
Caucasian languages. Barry Blake’s (1994) book Case takes his readers on a
wonderful trip through a spectacular range of case systems.
Old English had a morphological case system, with essentially four cases –
quite simple when compared to Finnish. Figure 3.1 gives four sample para-
digms, for different types of nouns. The demonstrative (sē, etc.) has somewhat
richer inflection than the nouns.
“that stone” “that ship” “that tale” “that name”
Singular
Nominative sē stān æt scip sēo talu sē nama
Accusative one stān æt scip ā tale one naman
Genitive æs stānes æs scipes ǣre tale æs naman
Dative ǣm stāne ǣm scipe ǣre tale ǣm naman
Plural
Nominative ā stānas ā scipu ā tala ā naman
Accusative ā stānas ā scipu ā tala ā naman
Genitive āra stāna āra scipu āra tala āra namena
Dative ǣm stānum ǣm scipum ǣm talum ǣm namum
Figure 3.1 Old English (nominal) case inflection
Morphological cases are not just ornamental decorations but they interact
with core syntactic operations. Polish, like Russian, shows an accusative mark-
ing on direct objects (3.29a), but the marking is genitive if the verb is negated
(3.29b).
3.3 Case 63
(3.29) a. Janek przyniósl´ kwiaty.

John bought flowers [acc]
John bought flowers.
b. Janek nie przyniósl´ kwitów.
John not bought flowers[gen]
John did not buy flowers.
3.3.2 Case theory

Surprisingly, the intertwining of case with core syntactic operations shows up
even in languages with no overt, morphological case system. This indicates
that there is more to case than meets the eye: case must represent a significant
abstraction, regardless of overt markings, and “case theory” has become part of
modern models. The idea is that all grammars have some kind of case system,
but only some have morphological case; the rest have abstract Case. From now
on we shall use Case to refer to that abstract notion of case, and lower case case
(sorry) to refer to overt, morphological markings. The distinction is crucial for
our discussion.
Any lexical head (noun, verb, adjective, preposition) may have a complement
DP. However, DPs are pronounced only in certain positions, and these are
positions where they have Case (by analogy with a traditional understanding of
the distribution of case). The abstract Case system of modern English is simple.
Verbs and prepositions generally assign (accusative) Case to their complement
DPs, as is overt in the form of pronouns: Kay V saw him P with them. The
subject of a finite verb receives (nominative) Case (he, they left) and a DP in the
specifier of a larger DP is Case-marked by the clitic ’s and has (genitive) Case:
Kay’s, his, their book. These are the positions which receive Case and these are
the positions in which a DP may be pronounced.
DPs may be understood in other positions, but unless that position is Case-
marked, the DP may not be pronounced there; it must be copied into a position
in which it is Case-marked in order to be pronounced. This, in fact, is the
motivation for certain movement: DPs move in order to be Case-marked. All
overt DPs must be Case-marked in order to be pronounced.11 That is the essence
of Case theory.
So, if a DP originates in a position which is not Case-marked, somehow it
must acquire Case. In (3.30) DPs move to another DP position. Kay is under-
stood as the complement of arrested and picture, and as the subject of like, but
11 One position in which a DP is not Case-marked and may not be pronounced is the complement
to an adjective: *Kay is proud Ray. Such structures are salvaged not through movement but
through the insertion of the meaningless preposition of, whose sole function is to assign Case
to the DP. The preposition for plays a similar role in salvaging the subject of an infinitival verb:
*him to learn Greek would be difficult vs. for him to learn Greek would be difficult.
64 3 Syntax
these are not Case-marked positions. Again, the Case-marked positions are the
complement of verbs and prepositions and the specifier of a DP. In (3.30) Kay
does not originate in such a position; in (3.30a) Kay originates as the comple-
ment not of a verb arrest but of a participle arrested, and passive participles
(unlike verbs) are not Case assigners. In (3.30b) Kay originates as the comple-
ment of a noun, not a Case assigner, and in (3.30c) the original position isn’t
the complement of anything nor the specifier of a DP nor the subject of a finite
clause. Consequently Kay may not be pronounced in those positions and must
move to a position where it does receive Case: one does not find anything like
*Ray was arrested Kay, *Ray’s picture Kay, *Ray seems Kay to like Jay, or *it
seems Kay to like Jay, because, in each example, Kay is in a position which is
not Case-marked.
(3.30) a. Kayi was participle arrested ei .
b. Kayi ’s N picture ei . (meaning “picture of Kay”)
c. Kayi seems [ ei to like Ray].
If a DP originates in a Case-marked position, on the other hand, then it does
not move. Kay originates as the complement of a verb in (3.31a) and may not
move (3.31aii), as the complement of a preposition in (3.31b) and may not
move (3.31bii), and as the subject of a tensed verb in (3.31c) and may not move
(3.31cii). Compare the analogous structures of (3.30), which differ crucially in
that Kay originates there in non-Case-marked positions and must move.
(3.31) a. i. Somebody V arrested Kay.
ii. *Kayi V arrested ei . (intended to mean “Kay arrested her-
self ”)
b. i. Picture P of Kay.
ii. *Kayi ’s picture P of ei .
c. i. It seems Kay likes Ray.
ii. *Kayi seems [ ei likes Ray].
If we are going to distinguish the positions from which movement takes place
in modern English in terms of case ((3.30) vs. (3.31)), we need an abstract notion
of Case, defined independently of morphological endings, because morpholog-
ical case does not exist in the language outside the pronoun system. Abstract
Case is what is at work in the distinctions of (3.30)–(3.33).
In (3.30a), the deleted Kay (ei ) is not the complement of a verb or preposition
or any other Case assigner, but of the participle arrested, and it has no Case. Sim-
ilarly in (3.30b), ei is the complement of the noun picture and has no Case. And
in (3.30c) ei is the subject of an infinitive verb and is Caseless. As a result, the
deletion sites in (3.30) are positions from which the DP Kay moves to another DP
position and receives Case there. In (3.31a), on the other hand, the deletion site
is the complement of a transitive verb, of a preposition in (3.31b), and in (3.31c)
it is the subject of a tensed verb. All these positions are Case-marked, DPs may
3.3 Case 65
be pronounced in these positions, and these are not positions from which they
must move to other DP positions; indeed, they may not move. We draw the
relevant distinctions in terms of an abstract notion of Case.
Conversely, wh-movement shows the mirror image: a wh-phrase moves to
the specifier of CP, not to another DP position, and it may not move there from
Caseless positions (3.32), but only from Case-marked positions (3.33). (3.32)
and (3.33) correspond to (3.30) and (3.31), respectively. In (3.33a) the deletion
site is the complement of the verb (as in (3.31a)), in (3.33b) the complement of
the preposition of (as in (3.31b)), and in (3.33c) the deletion site is the subject
of a tensed verb (as in 3.31c).
(3.32) a. *Whoi was Kay participle arrested ei ?
b. *Whoi did you see a N picture ei ?
c. *Whoi did it seem [ ei to like Ray]?
(3.33) a. Whoi did Kay V arrest ei ?

b. Whoi did Kay see a picture P of ei ?
c. Whoi did it seem [ ei likes Ray]?
3.3.3 Links between morphological case and abstract Case

This shows that Case is a fundamental item in grammars, even when it is not
morphologically marked. Current work within the Minimalist Program is in-
vestigating the intertwinement of Case and syntactic operations, with intriguing
results. One way of thinking of the relationship between the morphological and
abstract systems is to claim that abstract Case occurs in all grammars and must
be realized in some fashion; morphological case is then one means of realizing
abstract Case. We shall return to the relationship between morphological case
and abstract Case in chapter 8, when we examine some interesting syntactic
effects of the loss of morphological case in the history of English.
Meanwhile we have illustrated another principle of UG, which characterizes
the positions in which DPs may be pronounced. That principle forces certain
DPs to move, to be copied elsewhere. In that event, the copied element has to be
deleted; in section 3.2, we examined some aspects of the way in which deletion
takes place. All the deletions illustrated in this section meet the requirements
discussed in section 3.2.12 That is true, for example, of the deleted items in
(3.30); ei is the complement of the adjacent participle arrested in (3.30a), of
12 We have assumed but not demonstrated that copied elements must be deleted. Therefore, if they
cannot be deleted, not meeting the requirements for incorporation into an appropriate host, the
resulting structure is ill-formed. We adopt the analysis of Nunes 1995: copied elements must
be deleted because if they weren’t, there would be two instances of, say, what and those two
instances are non-distinct. If they are non-distinct, they cannot be linearized in the phonology.
This would take us further into grammatical theory than we want to go here, but interested
readers can follow the details in Nunes 1995.
66 3 Syntax
the adjacent noun picture in (3.30b). In (3.30c), the trace ei , it is the top-most
element of the complement of seems, and thus these are all legitimate deletion
sites. That is also true of the deletion sites in (3.31): ei is the complement of
the adjacent verb arrested in (3.31a), the adjacent of in (3.31b) and it is the
top-most element of the complement of seems in (3.31c), all legitimate deletion
sites. The problem with the structures of (3.31) is that they violate, not the
condition on deletion (3.5), but Case theory: Kay receives Case in its original
position and therefore may not move. Case theory interacts with our principle of
deletion (3.5) and together they characterize the possibilities for moving DPs,
i.e. copying and deleting them. We will leave it as an exercise for the reader
to determine that the deletions in (3.32) and (3.33) also meet the requirements
discussed in section 3.2.
3.4 Conclusion
Here we have put a searchlight on two aspects of current syntactic theory,
deletion and Case, and shown how they help to capture distinctions typical
of English speakers, and how they distinguish what a child learns from the
environment from what she knows independently of experience. It is these
detailed distinctions which make up the subconscious knowledge that people
have when they are speakers of some form of some language. That knowledge is
characterized by the kinds of grammars that people have, by their cognitive
systems. It is too gross to say merely that structures are made up of subunits or
that languages with case systems tend to have freer word order than languages
without a rich morphology. Modern work takes us beyond E-language bromides
like this.
4 Sound patterns in language
In this chapter and the two following ones, we turn from issues of syntactic
organization in natural language to the systematicities of sound structure. There
is a conventional division between phonetics, or the study of sounds in speech,
and phonology, the study of sound patterns within particular languages. As
we will see, there is a reasonably clear conceptual distinction here, and we will
follow it in devoting most of this chapter and the next to the more obviously
linguistic domain of phonology while postponing substantive discussion of
the nature of phonetics until chapter 6, after some necessary preliminaries in
section 4.1. We will attempt to tease apart these notions, but that process will
reveal that questions of sound structure, seemingly concrete and physical in
their nature, are actually abstract matters of cognitive organization – aspects of
I-language and not measurable external events.
Ideally, we should broaden our scope a bit: signed languages also have a
“phonology” (and a “phonetics”) despite the fact that this is not based on sound,
although we cannot go into the implications of that within the scope of this
book.1 In recent years, the study of signed languages has revealed the fact that
their systems of expression are governed by principles essentially homologous
with those relevant to spoken language phonology and phonetics. This close
parallelism reinforces the conclusion that we are dealing here with the structure
of the mind, and not simply sound, the vocal tract, and the ears (or the hands
and the eyes).
A rough way to distinguish phonetics from phonology is as follows: phonetics
provides us with a framework for describing the capacities of the organism – the
range of articulatory activities humans use (or can use) in speech, the properties
of the sounds that result, and the way the peripheral auditory system deals with
those sounds. A learner still has to determine how these capacities are deployed
in the language of the environment, but the capacities themselves and their
relation to physical events develop in the organism independent of particular
1 A basic survey of the formational system of American Sign Language is provided in Klima and
Bellugi 1979. Diane Brentari (1995) discusses the relation of this system to the phonological
systems of spoken languages; some more technical papers on this topic will be found in Coulter
1993.
67
68 4 Sound patterns in language
languages, in ways we will examine briefly in chapter 9. Phonology, in contrast,

provides us with a framework for describing the knowledge of sound structure
that is acquired as part of learning a specific language.
The limitation of phonetic concerns to properties that are actually employed
in speech is significant. Phoneticians do not, in principle, consider non-linguistic
sounds – burps, coughs, whistles, etc. There are purposes for which a represen-
tation that goes well beyond anything that might be linguistically significant
can be motivated: consider the interests of the psychotherapist, for example,
who might attend closely to much more than the purely linguistic content of a
patient’s vocalizations. Phoneticians and other linguists, however, have gener-
ally agreed that a representation of just those properties that might distinguish
linguistic utterances in some language is the appropriate starting point for such
a notion.
A phonetic representation, then, specifies the sound properties of an
utterance that differentiate it from all other possible human linguistic utter-
ances, in terms of all the dimensions that function in human languages. Such a
representation can be interpreted as a language-independent set of instructions
for pronouncing the utterance.
Phonological (or “phonemic”) representations, in contrast, spec-
ify the sound-related properties of linguistic elements (morphological units,
words, whole utterances) that distinguish these from other elements of the same
“granularity” within the same language. This involves the way differences in
sound are (and are not) used to differentiate one linguistic element from an-
other within the systems of particular languages; the way sounds are modified
(or replaced by other sounds) in particular environments, in ways particular to
individual languages; and other matters. A theory of sound structure, on this
picture, must specify at least the following things:
(4.1) a. What information belongs in the phonetic representation;
b. What a phonological representation looks like, and how it
may differ from a phonetic representation; and
c. What principles relate one to the other, and how these are
organized.
4.1 Phonetics as theory

Before we explore the goals and content of phonology, it is necessary to say
a bit about the nature of phonetics, and the relation between the acoustic sig-
nal and our perception of it. Phonetics is the place where linguistic data seem
most clearly observable. That is, sound seems to be a measurable physical phe-
nomenon which you can actually get your hands on (or at least your ears). In
contrast, phonological structure, syntax, semantics, and so on are all something
we infer from judgments and intuitions: sentences do not come with little trees
on them (representing their syntactic organization) which we could study di-
rectly, for instance. Because it deals with observables, phonetics seems to many
people not to be much of a “theory” at all: just a sort of neutral observation
language in which we describe utterances. On that view, a phonetic description
might be more or less accurate, but there is no other sensible way to evaluate it.
In fact, in the nineteenth and early twentieth century, phoneticians proceeded
along just this line: they attempted to refine their techniques for measuring as
many dimensions as possible of speech with maximal precision. As equipment
got better, and researchers’ observations could be more and more fine-grained
and accurate, there was an explosion of data – a result that had the paradoxical
effect of convincing most students of language that they were on the wrong
track.
Much of what was being turned up, for instance, followed from the observa-
tion that speech is continuous: what is taking place at any particular moment
is at least a little bit different from what is going on just before or just after
that moment. As a result, if the phonetician attempts to measure everything
at every moment where there are distinct values to be recorded, there is no
limit in principle to the amount of measuring to be done. It is clear that a full
characterization of an utterance as a physical event requires us to recognize an
unlimited number of points in time – but it is also clear that our understanding
of the utterance in linguistic terms is not thereby improved.
In fact, we usually represent the phonetic structure of an utterance in a much
simpler way: as if it were a sequence of separate units, like beads on a string,
each one a “sound” representing about the same amount of phonetic material as
a letter in common writing systems. A phonetic representation usually has the
form of a sequence of segments, where each of these is characterized as a point
in a multi-dimensional space. The dimensions of this space are the phonetically
relevant properties of sounds. These include articulatory details such as the
location of the highest point of the tongue body, presence (vs. absence) of
vibration of the vocal cords, rounding of the lips, etc. This is rather like a
sequence of snapshots, one per segment, where each one shows just the detail
about the articulatory apparatus that we deem relevant.
This characterization may seem to be simply a direct observation statement
about the utterance – after all, if the representation is really phonetic, there
surely couldn’t be anything too abstract about it, right? And isn’t it the case that
an utterance of the English word pit “really” consists of the sound p, followed
by the sound i, followed by the sound t?
But in fact while a phonetic representation has a certain claim to “objec-
tivity,” it is not just a mechanical recording of utterances. Rather (like any
serious record of an observation), it is a theory of what it is about the physical
form of this utterance that is of linguistic significance: it deals in abstractions.
Observers who are completely in agreement on the measurable, physical facts

can nonetheless disagree strongly on their interpretation, and thus on the cor-
rect phonetic transcription of a given utterance. A physicist, on the other hand,
would be quite unhappy with a phonetic representation as a characterization of
acoustic or biomechanical reality. We will have much more to say about the
actual content of a theory of phonetic representations in chapter 6, but at this
point it is still worth establishing that there is something substantial here to
have a theory of.
Compare the characteristics of a tape recording of an utterance of the word
pit with a transcription of that same utterance: [ph it]. There are (at least) three
important ways in which the transcription is an idealized abstraction from the
actual signal:
(4.2) Segmentation pretends the flow of speech can be divided into
a sequence of discrete intervals (here, three such segments);
Segmental independence pretends each of these intervals
contains all and only the information about a single phonetic
segment (thus ignoring co-articulation); and
A system of features chooses some aspects of speech pro-
duction to talk about and ignores others.
However intuitive it may seem, there is no physical warrant for representing
the continuous event as a sequence of (some small number of) discrete points.
Even the spaces in writing between words do not correspond to anything that is
(necessarily) present in the speech signal. Our division of the speech signal into
a sequence of discrete words is actually a theory of its structure. Sometimes,
more than one theory can be consistent with the same signal, which is the source
of a class of unintended distortions called mondegreens. The word comes from
the misconstrual of a line from a seventeenth century Scottish ballad, where
“they hae slain the Earl of Murray, and hae laid him on the green” was heard
as “ . . . and the Lady Mondegreen.” Jackendoff (1994) cites the fact that the
English sentences We needed a cantor and We need a decanter sound the same
without context or a particularly careful pronunciation, a relation described by
saying that the two sentences are oronyms. In such a case, a listener needs other
evidence to choose between multiple theories of the same utterance, and often
(particularly in the case of songs) such evidence is simply lacking.
Just as the division of utterances into words is a matter of linguistic structure,
and not physics, so also is the division of words into letter-sized units. This is the
abstraction involved in segmentation. Both from an articulatory and from
an acoustic point of view, that is, utterances give us no justification for cutting
them up as we do. For example, X-ray movies or other ways of picturing the
speech apparatus at work show continuous fluid movement of the articulators:
there are no (appropriate) discontinuities to tell us where to cut. This is not to
say that the acoustic facts do not have discontinuities: they often do, but usually
not where we want them. For instance, the acoustic structure of a stop consonant
like the initial [ph ] at the beginning of pit looks like three separate parts, not
one – but where the last one is integrated with the properties of the following
vowel in such a way as to make it impossible to separate them.
Secondly, there is the abstraction of segmental independence. The seg-
mental picture suggests that all of the properties of a segment are located at the
same point, and that segments do not overlap. But in fact, this is not at all the
way speech works. Instead, there is an enormous amount of co-articulation, or
overlapping of articulatory gestures. This is both anticipatory (right-to-left) and
perseverative (left-to-right), and it smears the segments together so as to make
it impossible to define their identity except relative to their particular context.
Another consequence of coarticulation is the fact that what counts as the
“same” sound may be heavily context-dependent. For instance, the [g] sounds
in ghoul and geek are quite different, as a function of differences in the vowel
that follows. This kind of intersegmental interaction is quite typical of speech
articulation: at any given point in time, what is going on is likely to be the
product not just of one but of several of the “segments” in the utterance. Even if
we could find discontinuities, then, we would not in general be able to present
a snapshot that is physically real and consists exclusively of the (invariant)
properties of a single segment we wanted to characterize.
So why do we do this (segmentation)? Because we have learned that this
idealization is actually more appropriate and useful than the literal truth. The
segmental representation is really the only basis for finding the regularities that
obtain in languages with respect to the forms shown by individual words. Of
course, it is only adequate once we specify the relation between such a picture
and the vastly more complex reality that we could (in principle) measure, and
phoneticians therefore take on (at least implicitly) the responsibility of describ-
ing the ways in which the representations they work with are implemented in
all of their messy detail.
The phoneticians of the nineteenth century were concerned to say how transi-
tions between segments fill in the intervening values of a continuous function – a
function which we only specify at some small finite number of points (corre-
sponding to the distinct segments we recognize in the utterance). Having done
that, though, the segmented result is more enlightening for further linguistic
study than a full reconstruction of the actual continuous nature of the facts. It
is an appropriate level of abstraction from those facts because it organizes the
data in a way that is more coherent: a way that allows us to see their structure,
otherwise obscured by detail.
There is a third idealization involved in phonetic representations: these are
also abstract in that they choose some things to characterize at the expense of
others. We might describe the sound [i] (roughly the vowel of Pete), for instance,
as a “high front unrounded oral vowel.” This tells us about some things involved
in the production of [i]: the position of the highest point of the tongue body, the
configuration of the lips, and the position of the velum. It says nothing about
what the rest of the tongue is doing (except that it is “lower” than the point we
described), or what the epiglottis is doing, or how wide the nose itself is opened,
what the facial muscles are doing, how loudly the speaker is talking, what his
(or her) precise vocal range is, etc. These omissions are not because the values
of these other properties cannot be determined: rather, they represent an implicit
claim that these and other aspects of the physical event are not linguistically
significant (or else that they can be predicted from the other things that we have
described).
No language, for example, ever seems to contrast utterances on the basis
of degree of spreading of the nostrils, or of loudness of the voice. Nor do
languages differ systematically in this respect (such that, e.g., language A is
spoken louder than language B, and to speak language A softly gives you a
foreign accent). Linguistically relevant transcriptions, then (as opposed to what
might be of interest to a psychiatrist or other specialist in non-verbal behavior)
do not need to record the width of the nasal opening, or the loudness of the
voice. In fact, we can go further, and say that a fortiori, such a transcription
should not record these properties – at least not if “phonetics” refers to the
linguistically relevant dimensions of speech. The choice of a set of parameters
(or distinctive features) is (part of) a phonetic theory: the data alone
do not establish their own meaning, or impose a particular choice of a set of
features as a matter of physical necessity.
What is required is a representation that will describe all and only those
aspects of sound production that can play a part in the system of language.
This is linguistic phonetics, after all, and not physics, or physiology. Any
degree of freedom in the representation should correspond to something that
at least potentially could be under linguistic control. The segmented phonetic
representation actually stands at a level of abstraction some distance away from
the physical reality, and there must be an explicit statement of the relation
between the two, a matter which will occupy us further in chapter 6. Theory
kicks in, even at the level of representing the sounds of language.
Phonetics connects with claims about innateness and Universal Grammar
(UG), etc.: linguistic phonetics is an attempt to delineate exactly the aspects
of sound structure that are available for use in natural languages – the things
that one has to be able to control (potentially) as part of learning a particular
language, matters which fall within the domain of the language organ. There
are indeed a number of independent dimensions with this property, but certainly
not an infinite number – and more to the point, not everything that one could
measure is a candidate. And if a discrete, segmented representation is close to
what is linguistically significant, the dimensions of control in speech are rather
abstract and removed from the continuous, concrete physical reality.
What is part of UG, then, and thus innate, is not the actual sound of any words,
but rather a set of abstractly organized possibilities. The child approaching the
task of learning a language knows in advance what to pay attention to, and what
range of possibilities might be important. In fact, as we will discuss in chapter 9,
children attend to these matters in astonishingly structured detail from the very
beginning: according to some research, indeed, even before birth. If the terms
of analysis are given in advance, then we can begin to understand how humans
cope with the infinite variability of experience in a common fashion, in such a
way that we arrive at shared representations and analyses of speech events.
4.2 Phonology: language-particular structure

In contrast to the “general purpose” nature of phonetics, phonology studies
the ways in which sounds are organized and related internal to individual
languages. While phonetics is concerned to provide a universal framework for
the linguistic sound properties of human speech, phonology is concerned with
questions such as those of (4.3).
(4.3) Inventory: Which of the possible sound types made available
by phonetic theory actually occur in utterances in language X?
Contrast: Which of the phonetic differences that occur in lan-
guage X can serve to distinguish utterances (words, sentences)
from one another in this language?
Contextual limitation: Even though some property P oc-
curs in language X, are there environments (characterized in
terms of other properties) from which P is excluded, or where
the value of P is fixed? Where P is a property of some element,
and might be expected to occur in a position from which it is
excluded, what – if anything – appears in its place?
Alternation: When the “same” linguistic element (mor-
pheme, word, sentence) appears in different forms in different
environments, what systematic differences occur in its pronun-
ciation? What conditions govern the range of phonetically dif-
ferent elements that count as the “same” morpheme, word, etc.?
Of course phonological theory is concerned to specify the range of answers
to these questions that are possible in general. As such, it is another universal
property of the human language faculty, since we presume that the range of
answers (and indeed the set of appropriate questions) is independent of any
particular language. Just as the child brings a set of phonetic categories to
the task of language learning, there is also a limited range of properties that
characterize the systems of individual languages in comparison with one an-
other. Phonological theory attempts to describe the dimensions of this aspect of
I-language.
4.2.1 Phonological representations

In this section, we discuss the basic nature of phonological representations,
showing the kinds of abstraction that are involved. We shall see that there are
actually a wide range of quite different approaches that all correspond in one way
or another to the fundamental insight underlying the concept of “phonological
representation,” implying distinct notions of the phonology–phonetics relation.
Within the linguistics of the mid-twentieth century, certain a priori notions about
the nature of this relation implied that some significant facts about linguistic
sound structure had to be relegated to some other aspect of the grammar, for
which yet a third kind of representation was proposed: morphophonemics.
Phonetic representations are an idealization from the complex facts of speech
to a sequence of discrete, internally homogeneous segments. Each segment is
characterized with respect to a set of features on a limited, universal list. The
point of such a theory is to provide a principled account of what we include and
what we leave out in giving a language-independent characterization of par-
ticular utterances: importantly, we leave out the irrelevant and the predictable.
But if phonetic theory provided our only account of the sound properties of
languages, we would leave unexpressed the fact that many things which are not
predictable in general are nonetheless quite predictable by principles specific
to a given language. When we know the language of a given utterance, many
of its properties that are significant from the perspective of general phonetics
can be seen to be determined by the system of that language, rather than being
distinguishing properties of just that utterance.
For instance, the aspirated character of the stop in English pat (phonetically
[ph æt]) must be indicated phonetically, since the languages of the world contain
many instances of unaspirated voiceless stops in essentially the same position
(e.g., French [pat] patte “paw”), and in some cases the two possibilities even
contrast (e.g. Punjabi [pət] “honor” vs. [ph ət.] “split”). And yet somehow this
aspiration seems less relevant from the point of view of English: indeed, most
English speakers are quite unaware of it.
This is not to say that a phonetic representation of utterances is wrong,
but only that there are facts of another order to be described as well, another
(distinct) level of analysis that must be available for the sound structure of
utterances. We want not only a characterization in the language-independent
terms of phonetic theory, but also one that takes into account the sound pattern
of a particular language, in terms of which we can identify the ways in which
sound properties function in that language.
Following a line originally stressed by Ferdinand de Saussure (1974, an
edition drawn from notes on his courses first published in 1916), we want this
phonological representation to be systematically constrained so that differences
between the representations of two utterances correspond to differences in their
linguistic content, not simply their physical properties. For reasons that will
become apparent, we will often refer to the phonological representation as the
“underlying form” of an utterance. And to further confuse matters, the history of
this kind of representation results in a tendency to call its elements “phonemes”
and thus to speak of a “phonemic” representation. There are subtle distinctions
to be made here, but by and large “phonological,” “underlying” and “phonemic”
representations are all ways of talking about the same thing.
The essence of what this representation is all about is expressed by the
following principle:
(4.4) Two phonological representations should differ only insofar as

they correspond to different messages within the language in
question.
Thus, in English, phonetic representations contain both [ph ] (as in pit, phonet-
ically [ph ιt]) and [p] (as in spit, phonetically [spιt]); but we do not want to allow
two phonological representations to differ only in that one has a [ph ] where
the other has a [p], since it is never the case that this difference alone serves to
distinguish meanings in English. Simplifying slightly, aspirated [ph ] appears
always and only at the very beginning of stressed syllables, while unaspirated
[p] appears elsewhere (after a syllable-initial s or under conditions of reduced
stress). A similar generalization applies to other voiceless stops (t, k) in English.
The point is that a speaker’s knowledge of English, her language organ,
includes some characterization of a set of principles defining the notion of
“phonological representation in English” that exclude two such representations
from both being possible, even though the definition of “phonological represen-
tation in Punjabi” (for instance), which consitutes a part of a Punjabi speaker’s
language organ, does allow two representations that differ only in this way.
Just as the notion of “phonetic representation” provides us with an implicit
definition of “possible human linguistic utterance” or the like, a phonological
representation is the basis of an account of what the possible linguistically dis-
tinct utterances are in a given language. The phonological principles that are
thus inherently particular to a specific language form a part of I-language, and
it is part of the task of learning a language to determine the appropriate set of
such principles from among those made available by phonological theory.
We have formulated the principle in (4.4) in extremely general terms, in
order to try to accommodate the many different ways in which linguists have
thought of phonological representations. We can distinguish at least three variant
conceptions of how to satisfy the requirement of the phonemic principle (4.4).
One approach, which we can call a theory of incompletely specified
representations, proposes that where some property is predictable within a
given language, we omit it altogether from phonological representations within
that language. It is then a function of the rules mapping phonological onto

phonetic form to fill in the unspecified features. On this picture, phonological
representations in a given language are arrived at by quite literally abstracting
away from the predictable properties of utterances in the language in question.
Many people assume that the incompletely specified view is the only pos-
sible (or coherent) way to think of phonological representations, but this is
not the case. Some scholars, such as Edward Sapir, have maintained a fully
specified basic variant view of phonological form. This notion also fo-
cuses on the idea that only a limited number of the sound types that are pho-
netically possible can be the basis of contrasts in a given language, but instead
of effectively omitting all of the non-contrastive properties, it treats a group of
(phonetically distinct) sounds that do not contrast with one another as a sort of
equivalence class, represented by a kind of “prototype” sound.
On this picture, we say that English has only one kind of voiceless stop in
phonological representations, but this is nonetheless a fully specified phonetic
segment. Suppose (for concreteness’ sake) that we choose to have underlying
unaspirated voiceless stops like [p]. Then the phonological representations of
pit and spit are /pιt/ and /spιt/, respectively,2 and one of the things that must be
specified about the relation of phonological form to phonetic form is that some
/p/s are realized phonetically as [ph ]s.
These two views are different, but they both satisfy the phonemic princi-
ple. On the fully specified basic variant view, no two English phonological
representations could differ only in that one has a ph where the other has a p
because English phonological representations only have /p/s, and not /ph /s.
On the incompletely specified view, such a difference could not arise because
the property of aspiration is not present at all in phonological representations
(in English).
There is a third way to satisfy principle (4.4) which has been maintained by
some other phonologists, including on one reading (Anderson 1985), Saussure
himself. We can call this a fully specified surface variant view. On
this picture, we let phonological representations look just like phonetic ones:
pit= /ph ιt/, spit= /spιt/. But we also say that a phonological representation in
English is only well-formed if it conforms to a set of defining regularities.
For instance, a representation in which an aspirated stop follows an /s/ within
a syllable is ill-formed, and on the other hand a representation in which an
unaspirated stop appears at the beginning of a stressed syllable is also ill-formed.
Now if we get the restrictions right (and tight enough), we have satisfied the
phonemic principle even though our representations do not look any different
from phonetic ones! That is because if we take two representations that differ
2 Note that the usual convention is to put phonological representations within /s and phonetic
representations within square brackets.
only in that one has a ph where the other has a p, one or the other of these is
going to be ill-formed because it violates some constraint, and is thus ruled out
as a phonological representation in English.
If all of these approaches to phonological representation can satisfy the basic
desiderata for such a notion, as expressed in (4.4), does it make any difference
which one we choose? In fact here, as elsewhere, the issue of an appropriate
level of abstraction arises, and the determination of the properties of that level
is ultimately an empirical question. If all three of the varieties of phonological
representation just considered can be made to fare equally with respect to the
basic principle in (4.4), there may still be other ways in which they are distinct in
their ability to facilitate an insightful account of the complete reality constituted
by sound structure in I-language. When we consider not simply the nature of
phonological form itself, but also the systematic formulation of its relation to
other aspects of language (including phonetic form, word structure, and other
matters), differences will emerge.
4.2.2 Relating phonological and phonetic form

The differences among possible conceptions of phonological representation
considered in the previous section are connected with different conceptions of
how phonological form is to be related to phonetic form. On the incompletely
specified view, what has to be done to get from phonological to phonetic form
is to “fill in the blanks”: i.e., to specify the values of redundant parameters. On
the fully specified basic variant view, on the other hand, the relation consists not
in filling in the blanks, but rather in changing a “prototypical” value of a given
parameter in a context-dependent fashion. Finally, on the fully specified surface
variant view, there is no alteration of the representations at all, but rather a system
of constraints serves to define directly the range of potentially contrasting forms
within a given language. In choosing among these possibilities, the phonological
theorist asks which leads to a more enlightening and comprehensive view of
the nature of I-language.
Independent to some extent of the mechanics of the phonology–phonetics
relation, another issue is just how “abstract” that relation can be. In particular,
during the 1940s and 1950s, many linguists came to agree on a condition that
should be imposed on the relation between the two representations, radically
limiting the permissible degree of abstraction in phonological form by requiring
that phonological form be directly and unambiguously derivable from phonetic
representation. This is often called the biuniqueness condition:3
(4.5) The phonetic and the phonological representation can each be
uniquely recovered from the other.
3 See Anderson 1985, pp. 312ff. for some discussion of these matters.
With regard to the path from phonological to phonetic representation, every-

one would accept this (at least up to the point of free variation): it just means
that the principles of the phonology have to operate so as to assure a determinate
result.
In the other direction, though, the requirement that a phonological represen-
tation can be uniquely determined on the basis of phonetic form alone is not so
obvious. In particular, (4.5) means that two phonological representations cannot
be “neutralized” in the sense that both map onto the same phonetic form, for
if they were, there would be no way to resolve the ambiguity without appeal-
ing to something outside of phonetic representation. But many cases in many
languages seem to have that character, prima facie.
For instance, obstruents in German are always [–Voice] in final position.
This has the consequence that both Bund “association” and bunt “varicolored”
are pronounced [bυnt], although the difference reappears when the stops cease
to be final (cf. Bunde [bυndə] “associations,” bunte [bυntə] “varicolored (fem
nom sg)”). To maintain biuniqueness, we would have to say that both Bund and
bunt have the same phonological representation (e.g., /bυnt/), because there is
no way to tell for a given instance of phonetic [bυnt] which word it represents.
But this conclusion is quite counter-intuitive.
Why did scholars believe in the necessity of a condition like biuniqueness? In
fact, the notion that phonological form should be biuniquely related to phonetic
substance grows fairly directly out of the notion that both are characteriza-
tions of utterances as they exist apart from speakers themselves: that is, of the
E-language conception of what is being described in linguistics, with phono-
logical representations hugging the (supposedly objective) phonetic ground and
minimizing their abstractness. As we will see, the abandonment of condition
(4.5) is intimately tied up with the replacement of that conception of the field
by one that takes I-language, the speaker’s knowledge of the language, as the
proper object of linguistic inquiry.
There are at least three sorts of reason that seem to characterize the motiva-
tions for earlier scholars’ assumption of the condition in (4.5), all deriving from
the E-language orientation of linguists at the time. First, speech perception was
assumed to work in a purely passive, bottom-up mode, starting from the acoustic
input (alone) and yielding a linguistic percept. If phonological representations
correspond to this perceptual reality, they would then have to be recoverable on
the basis of data in the signal alone.
Secondly, a bias toward procedural accounts of linguistic structure dominates
the linguistic literature (at least in North America) from the 1930s through the
early 1950s. Linguists were not supposed to have any biases or preconceived
notions of what languages were like, since it was assumed they could differ with-
out limit (i.e., that there are no universals). Further, since meaningful elements
are constructed out of sound units, it seemed logical to linguists of the time
that one had to have a complete account of the latter before trying to identify
the former. But in that case, the phonetic data are all the analyst has to go on
when constructing an analysis, so the analysis has to be the sort of thing that
can be attained on that basis. The conceptual mistake that follows is to identify
what the language “must be” with a particular path by which the linguist can
discover its structure. This has the (presumably unintended) consequence that
linguistic theory winds up describing the behavior and knowledge of linguists,
rather than that of speakers.
Finally, many felt that what the phonemic principle “really” required was
that phonological representations encode linguistic contrast. If we take contrast
to be something that can be operationally determined by asking speakers
“do the utterances U1 and U2 contrast or not?” (or any one of a thousand
variants of this), the principle again follows, since all and only those differences
that can be unambiguously recovered from phonetic presentation alone will
correspond to contrasts in this sense.
As this view became more explicit, however, it became clear that it failed to do
justice to the facts of language. First, students of the nature of perception came
to entertain the possibility that rather than simply registering its input, the mind
actively constructs hypotheses about the world, and compares them with the
incoming data to validate a particular interpretation. Perhaps the purest form of
this picture is the notion of “analysis by synthesis,” according to which percepts
are quite generally constructed by the perceptual system itself (the “synthesis”)
and then confirmed (as the “analysis”) to the extent that the available data do
not contradict them.4 The kind of “motor theory” of speech perception which
we will presume in chapter 6 below generally makes somewhat similar assump-
tions. These and other active views of perception made it clear that the perceptual
system might well involve (at least in part) “top-down” generation and testing of
hypotheses, and not only simple, bottom-up identification of the acoustic signal.
If that is the case, though, more than one such hypothesis might be consistent
with the same surface facts, and thus more than one phonological representation
might represent a speaker’s interpretation of the same phonetic form.
Secondly, when we recognize that there are in fact substantive universals
of language, it becomes clear that linguists are not limited in their grammar-
writing to simple induction over collections of surface data, and also that the
analysis of different parts of the grammar (e.g., sound structure and word
structure) can proceed hand-in-hand. Thus, there is no reason to believe that
the surface facts alone directly constrain the range of possible grammars.
Finally, it turns out that “contrast” is a lot harder to define operationally than
it initially appears. True, we do need to be able to define this notion (insofar
4 See Stevens and Halle 1967 for the specific application of this view to the problem of speech
perception.
as it is a coherent one), but it does not follow that contrast ought to be the
only thing determining what a phonological representation should be like. A
coherent view of this relation, in fact, along with the other matters discussed
just above, emerges only when we realize that the basic object of inquiry in the
study of sound structure (as with the rest of language) is the language organ, a
form of knowledge, rather than a direct characterization of external events.
4.2.3 Morphophonemic representations

Let us explore the consequences of accepting the biuniqueness condition (4.5)
for our ability to provide an adequate description of the content of I-language
knowledge of sound structure. We will see that adherence to this condition
actually prevents us from articulating some regularities that clearly constitute
part of what speakers know about their language.
Consider an example that is similar to that of the final devoicing we saw above
in German, but with a bit more structure. In Korean we find the following set
of obstruent consonants in initial position and between vowels:
(4.6) ph th h kh
p’ t’ ’ k’
p/b5 t/d / k/g
s
s’
In final and preconsonantal position, on the other hand, we find only [p,t,k],
sounds in which the consonantal closure is not immediately released into a
following vowel or brief aspiration. In different forms of the same basic word,
final or preconsonantal [p] will correspond to any one of the labials, [t] to any
of the coronals (the second and third columns in (4.6)), and [k] to any of the
velars.
In medial position, any of the segments in the chart in (4.6) can occur. What
is obviously the same stem, however, changes shape when a medial consonant
comes to be in final or preconsonantal position. For example, an object case
suffix /- l/ appears in Korean in (object case) forms like those in (4.7).
(4.7) a. nad l “grain”
b. nath l “piece”
c. nas l “sickle”
d. načh l “face”
e. na l “day”
5 For the sounds in this row, the first (p, t, , k) occurs initially and the second (b, d, , g) occurs
between vowels. Since the members of these pairs are in complementary distribution, we treat
them as units.
In isolation, though (as for instance in the citation forms of the same words),
all of these stems are pronounced in the same way: as [nat].
The matter of how to analyze the object case forms in (4.7) phonologically is
straightforward, since all of the segments in the chart are in contrast, and we can
set up phonological representations essentially equivalent to the phonetic ones
(though we still need to abstract away from the environmentally conditioned
variation in voicing in p/b, t/d, etc.). But what about the phonological analysis
of [nat]? We seem to have two choices.
On the one hand, we might say that [nat] always has the same phonological
representation (e.g., /nat/), one which satisfies biuniqueness (4.5), because it
can always be uniquely, if uninformatively, recovered from the phonetic form.
The alternative is to say that there are at least five (and potentially eight) different
phonological forms that all correspond to the surface form [nat], such that it
is not possible to tell in isolation which phonological representation should be
associated with any given phonetic one (hence violating biuniqueness).
Most linguists are tempted to posit phonological representations /nat/, /nath /,
/nas/, /načh /, /nač/, etc., and to say that there is a principle by which any final or
preconsonantal coronal obstruent (in phonological representation) is replaced
by (phonetic) [t] (with corresponding rules for labials and for velars). This
violates biuniqueness, but it seems to express something real about the language.
What is the basis of this feeling?
In essence, the situation is the following. For any given lexical element of the
language, the prevocalic variant in which it occurs is idiosyncratic, but constant:
e.g., the word for “day” appears with the same consonant in [naei] “in the
daytime” as in [na l] “day (obj),” as well as in any other form where the stem-
final consonant is followed by a vowel. In contrast, forms of, e.g., “sickle” in
which the stem is immediately followed by a vowel always show a stem-final [s].
Given the form that occurs before a vowel, the final or preconsonantal form is
predictable, but not vice versa. That is, we can establish as a general principle
of the language that any segment which is [+Obstruent,+Coronal] in medial
position (i.e. [t, th , s, č] etc.) will correspond to [t] in final position. From the
non-medial form, however, we cannot predict the medial one in a unique way.
Given a biunique notion of the relation between phonological representations
and phonetic form, /nas/, /nac/, etc. cannot be the phonological representations
of words all pronounced [nat], since there would be no way to recover one
of these (as opposed to any of the others) from a given instance of [nat]. Any
phonological representation is a representation of the sound properties of a
message, but biuniquess further limits the sound properties that can potentially
differentiate one message from another to ones that are overtly realized in the
phonetic form of utterances expressing that message – an E-language notion.
But what else could “sound properties” possibly mean? In fact, what dif-
ferentiates e.g. /nas/ from /nac/ as the phonological representation for [nat] is
something a native speaker of Korean knows about the form the item in question
takes in general and not just in this utterance. As a part of speakers’ knowledge
of the language, these relations clearly belong in a full account of the I-language
Korean, part of a Korean speaker’s language organ. Any theory of phonologi-
cal form which prevents their expression is thereby seen to be inadequate as a
general account of the structure of the language faculty.
Note that the proposal to distinguish /nas/, /nat, /nač/, etc. (all realized pho-
netically as [nat]) is still a matter of the “sound properties differentiating one
(potential) message from another,” however. We have not, for instance, pro-
posed a way to give different representations for, e.g., pair and pear in English,
since these words never differ in their sound properties. While the differences
among Korean /nas/, /nat/, /nač/, etc. are manifested in some (though not all)
phonological environments, there are no such differences among English pare,
pear, pair.
When American structuralists in the 1940s and 1950s came up against this
sort of example, they could not incorporate it into a (biunique) phonemic rep-
resentation. Since their focus was on an E-language conception of the object of
their study, and for other reasons sketched above, they considered biuniqueness
a valuable requirement a priori on phonological form, and were not particularly
troubled by this conclusion. On the other hand, examples such as the Korean
one here (and the widely discussed case of final de-voicing in German, Russian,
and a variety of other languages) made it clear that even if these facts did not
belong in the “phonology,” there was still more to be said to include them in an
overall description.
To allow for this, they constructed a new kind of object: a “morphophonemic”
representation. This was assumed to represent a more abstract characterization
of linguistic elements, related in a non-biunique way to phonemic representa-
tions as in (4.8).
(4.8) Morphophonemic form Phonemic form Phonetic form

|nas |
ˆ
|nac| /nat/ [nat ]
ˆ
|nath|
These morphophonemic representations were not taken too seriously as real-

ities of language by most phonologists at the time, precisely because they were
not recoverable in a biunique fashion from the phonetic data: they were regarded
simply as linguists’ conveniences for representing a lot of data in a compact
fashion. This makes perfect (ideological) sense: the need for a phonological rep-
resentation of the “morphophonemic” variety does not spring from E-language
considerations, which were the focus of attention at the time, but rather from
the need to express the content of speakers’ knowledge of the language. From
our perspective, however, it can readily be seen that these representations actu-
ally articulate something important that a speaker knows, hence properties of
that individual’s language organ: they (and not the biunique phonemic forms)
uniquely determine the full range of shapes a given linguistic element will dis-
play across environments, which is surely part of the characterization of the
“sound properties” that oppose that element to other elements). An account
of sound structure as I-language must thus include these matters; while the
necessity of the biunique phonemic representation, in contrast, remains to be
established.
4.3 Morphophonemics and I-language

Biuniqueness has been abandoned as a requirement by most phonologists for
many years, and in practice the phonological representations of the 1960s and
later have looked much more like the “morphophonemic” representations of the
1950s than like the phonemic representations of that era. Given the intellectual
investment of the field in a biunique, surface oriented, E-language interpretation
of the phonemic principle at the time, how did this change in outlook come
about?
There is a standard version of the history of phonology – a kind of creation
myth that has as its background the picture of structuralists in America (and also
in Europe) concentrating on the discovery of (biunique) phonemes as minimal
units of surface contrast through the 1930s, 1940s, and 1950s. Then at the end
of the 1950s (according to this interpretation), Morris Halle presented some
data from Russian, facts concerning voicing assimilation in that language, from
which it was immediately clear that the whole notion of phonemes as they were
conceived at the time was indefensible and led inevitably to loss of generality.
As a result (with allowances for entrenched prejudices and the time necessary to
retool), phonologists reoriented their attention toward the previously marginal-
ized domain of morphophonemics, and phonemics was replaced by generative
phonology.
To what extent does that picture correspond to reality? Let us look at the facts,
historical and linguistic, which can teach us something about shifts in scientific
paradigms, among other things. Halle’s argument was first presented at a 1957
meeting (Halle 1957), and appears in print two years later in The Sound Pattern
of Russian:
In Russian, voicing is distinctive for all obstruents except /c/, /č/, and /x/, which do
not possess voiced cognates. These three obstruents are voiceless unless followed by
a voiced obstruent, in which case they are voiced. At the end of a word, however, this
is true of all Russian obstruents: they are voiceless, unless the following word begins
with a voiced obstruent, in which case they are voiced. E.g., [m’ok l,i] “was (he) getting
wet?,” but [m’og b ] “were (he) getting wet”; [žeč l,i] “should one burn?,” but [že
ˇ b ]
“were one to burn.”
In a phonological representation which satisfies both condition (3) [phonemic→

phonetic determinacy] and (3a) [phonetic→phonemic determinacy], the quoted utter-
ances would be symbolized as follows: /m’ok l,i/, /m’og bi/, /žeč l,i/, /žeč bi/. Moreover, a
rule would be required stating that obstruents lacking voiced cognates – i.e., /c/, /č/, and
/x/ – are voiced in position before voiced obstruents. Since this, however, is true of all
obstruents, the net effect of the attempt to meet both condition (3) and (3a) would be a
splitting up of the obstruents into two classes and the addition of a special rule. If con-
dition (3a) is dropped, the four utterances would be symbolized as follows: {m’ok l,i},
{m’ok bi}, {žeč l,i}, {žeč bi}, and the above rule could be generalized to cover all
obstruents, instead of only {č}, {c} and {x}. It is evident that condition (3a) involves a
significant increase in the complexity of the representation.6 (Halle 1959, 22f.)
Halle’s argument is standardly adduced as having persuaded phonologists to
abandon the phonemics of the time, and take up the rather different pursuits
and abstractions of generative phonology. The logic of the argument is that a
description interposing a level of representation meeting the biuniqueness con-
dition (4.5 above) between the morphophonemic and phonetic representations,
as in (4.8), leads necessarily to a situation in which a regularity of the language
(here, the assimilation of voicing in clusters of obstruents) cannot be stated in
a unitary way.
As we will argue below, the force of this argument rests in its focus on the need
for a grammar not only to assign the correct representations to linguistic forms
but also to give an account of the rules of the language, rules which correspond
directly to components of a speaker’s linguistic knowledge. By “rules” here,
we do not of course refer to the pronouncements of prescriptive grammarians,
but rather to the collection of regularities that can be shown to characterize an
important component of the way individuals produce and understand sentences
of their language – the language organ, as represented by a “grammar.” As we
shall see in chapter 5, these regularities can be formulated in a variety of ways.
What matters is that the replacement of biunique phonemic representations
by “morphophonemic” ones must have followed from the replacement of
E-language conceptions of language by those of I-language. While quite plau-
sible in retrospect, however, the extent to which that logic was actually apparent
to the linguists of the time who heard Halle’s argument is not obvious.
It is reasonably clear that the force of Halle’s argument cannot have come
from its novelty, since similar arguments had been offered before. Halle, indeed,
cites one (though curiously, not one of the better known cases noted below):
Analogous examples can be cited from many languages. An especially interesting ex-
ample is discussed by G. H. Matthews, “A Phonemic Analysis of a Dakota Dialect”
[IJAL 21. 56–59 (1955)], who shows that the labial and dental nasal consonants are
automatic alternants of the cognate stops as well as of /m/ and /n/, while the velar nasal
is an alternant of the velar stop exclusively. (Ibid. p. 22, note 12)
6 Halle uses braces – {} – to indicate a morphophonemic representation, as opposed to biunique

phonemic representations in /s and phonetic representations in square brackets.
In fact, the example Halle refers to is not analogous in the relevant details to
the Russian case. Matthews shows that /b,t,k/ are replaced by [m,n,ŋ] in syllable-
final position after a nasal vowel in this form of Dakota. But he also makes it
clear (Matthews 1955, p. 57, note 3) that “[m, n] do not otherwise occur in this
position.” This is thus a case of “partial overlapping,” rather than neutralization,
and the example does not necessitate the loss of a generalization in order to
maintain a biunique analysis. Since the nasalization rule for stops following a
nasal vowel does not involve a mixture of neutralization and allophonic effects,
but can be formulated entirely as a relation between phonemic and phonetic
forms, its unitary nature is not compromised by the principle requiring phonemic
forms to be uniquely recoverable from the surface phonetics. Other examples
that do have this character, however, were known at the time, as we will note.
There were actually several examples that had been discussed in the literature
before Halle’s paper that involve facts whose logic is entirely parallel to that
of Russian voicing assimilation. It is instructive to look at the treatment they
received, because it shows us something about the extent to which linguists of
the period held to the principles of their theory.
One way of dealing with such facts is illustrated by Bloomfield’s discussion
of Menomini. Here, as in Russian, we have an apparent generalization which
(when applied to morphophonemic forms) involves a mixture of phonemic and
subphonemic effects. But instead of concluding that this showed the inadvis-
ability of phonemic representations, Bloomfield interprets the facts as showing
that the allophonic variation is probably phonemic too, after all. “If it looks
like a phoneme, walks like a phoneme, quacks like a phoneme, it must be a
phoneme” (with apologies to Walter Reuther).
If postconsonantal y, w, or any one of the high vowels, i, ı̄, u, ū, follows anywhere in
the word, the vowels ē and ō are raised to ı̄ and ū, and the vowel o in the first syllable
of a glottal word is raised to u: mayı̄čekwaʔ that which they eat, cf. mayēček that which
he eats; ātεʔnūhkuwεw he tells him a sacred story, cf. ātεʔhnōkεw . . . Since ū occurs
only in this alternation, it is not a full phoneme. (Bloomfield 1939, §35)
“Not a full phoneme.” What does that mean? In the inventory of “the actual
Menomini phonemes,” the element ū appears in parentheses, and is identified
as a “semi-phoneme” (Bloomfield 1939, §5). Bloomfield must have been some-
what uncomfortable with this analytic result, because in a later treatment, his
posthumously published grammar of Menomini (edited by Charles Hockett),
he gives some rather marginal arguments that ū is a (“full”) phoneme after all.
Since the occurrence of u is normally confined to the forms in which it replaces o
under the regular alternation of 1.8 [referring to the rule above], it might be viewed as
a mere positional variant of o . In this alternation, however, the difference of o and
u is parallel with that of e and i , two sounds which unmistakably figure as separate
phonemes. Moreover, the difference of o and u is maintained by persons in whose
speech this alternation has lost its regularity. Also, the sound of u (and never of o ) is
used in a few interjections: capu q “splash!,” ku h “stop it!” A contrast of o and u

appears in the foreign words co h “Joe,” cu h “Jew.” (Bloomfield 1962, §1.16)
Fairly clearly, the invocation of such marginal evidence (the speech of in-
dividuals who do not really control the phonology of the language and the
pronunciation of a synchronically non-Menomini word) stretches the intuitive
notion of what phonology is about in order to maintain consistency.
A somewhat different response, and a real triumph of honesty in dealing
with such facts, is illustrated by Bernard Bloch. Discussing (in Bloch 1941) an
example from English that is logically just like Halle’s, he notices exactly the
same point Halle makes: an apparently unitary rule must be broken in two as a
result of the requirements for a phonemic representation. But does he conclude
that (biunique) phonemes should be discarded? Not at all. Instead, he concludes
that science has saved us from a seductive but ultimately false generalization.
In essence, he denies the intuitively obvious analysis of the facts on the basis
of a priori theoretical considerations.
These reactions are among the more principled. In fact, when we look at the
examples that began to accumulate by the 1950s which suggested that phone-
mic representations had properties that led to incorrect or otherwise deficient
analyses, we see that linguists of the time found various ways to preserve their
principles in the face of the apparent facts. On an issue other than biuniqueness,
this can be illustrated from reactions to the famous example of writer/rider,
where the surface contrast is in the “wrong” place as illustrated in (4.9). For
many speakers, the pronunciations differ in terms of the length of the first vowel
and not in terms of the the medial stop, which is pronounced in the same way
(typically as a sort of “flap”) in both words.
?
(4.9) [rajD] “writer” vs. [ra jD] “rider” = /rajtr/ vs. /rājtr/ or /rajtr/
vs. /rajdr/
One possible way to deal with such a situation is to force the theory to provide
the correct result. When the principles lead to absurdity, adapt the principles
so that they will yield what you know intuitively to be correct. An example of
this approach is provided by Harris’ (1951) procedures of “rephonemicization,”
which allow the linguist to massage the analysis in a variety of ways so as to
arrive at a satisfying analysis even though the basic premises of the theory do
not naturally provide one.
An alternative is to follow the principles consistently, and if they lead to
absurdity, then deny the facts. With respect to the specific facts in (4.9), this is
illustrated by Householder’s (1965, p. 29) conviction that “I can tell you from
experience that you will, if the words are in fact consistently distinguished,
invariably find one or more of several other differences [between the flaps of
writer and rider].” That is, even though all of the apparent evidence suggests
that the difference between writer and rider (in the relevant dialect) is a matter
of the quantity or quality of the stressed vowel, a sufficiently assiduous search
for phonetic detail will uncover some basis for assigning the difference to the
medial consonant (where it intuitively “belongs”) and treating the patent vowel
difference as allophonic.
The difficulties that emerged for the phonemic theory of this period follow
directly from the fact that it was a theory of E-language. The biuniqueness con-
dition (4.5) and the approach to language that motivated it forced the terms of
the theory to limit themselves to descriptions of the external observables sup-
posedly provided by phonetics. As a result, facts indicating that speakers’ know-
ledge of a language is not limited in this way had to be dealt with in uncomfortable
ways or not at all.
A phonological representation is, by its very nature, a characterization of
the sound properties that distinguish linguistic objects for a speaker of a
given language. In order to translate this notion into a description of external
reality, however, phonemicists found it necessary to rebuild it on the basis of
observable properties and operational tests, ideas that turn out to be quite prob-
lematic in practice and to lead to a host of difficulties that in fact have nothing
important to do with phonology itself. As we will see in chapter 6, even the
notion of a purely E-language approach to phonetic description turns out to be
inadequate.
When we ask why Halle’s argument should have been so earth-shaking, it is
hard to say. Not only did it not involve a completely novel complex of facts,
it is not even the case that it shows biunique phonemic analyses in general to
lead to loss of generalization. This is a point that several authors have made,
with respect to earlier theories such as that of Trubetzkoy (1939), one of the
founders of the European variety of phonological theory to which generative
phonology often traces its roots.7
Principled discussion in the 1940s and 1950s of facts that were embarrassing
for phonemic theory did not in general consider, as Halle did, the possibility
that the appropriate conclusion to be drawn was that the basic premises of
structuralist phonology were misconceived. On the other hand, Halle’s argument
when it was presented in 1957/1959 was of a sort that had been offered in
substance before; and in any event, it did not really suffice to prove its point
in a fully general form. So why, then, did it have such major consequences,
while other similar cases had little or no effect? It appears that the special force
of Halle’s argument came from the fact that it was embedded in a theory that
was not limited to representations and the alphabets of elements that compose
them:
7 For discussion of alternatives to Halle’s analysis within Trubetzkoy’s framework – involving the
notion of the “Archi-phoneme” – and also within current generative phonology, see Anderson
2000, from which the present section is derived.
[T]he effectiveness of Halle’s argument . . . lay in the emphasis it put on the centrality of
rules in a phonological description. Note that the entire argument rests on the observation
that, in certain situations, a level meeting the conditions of bi-uniqueness requires some
unitary regularity of the language (here, voicing assimilation) to be split up into two ef-
fectively unrelated rules. Now in a theory (such as American structuralist phonemics) in
which only the representations of forms have “real” status, such an argument is nonsen-
sical or at best irrelevant: the principles relating one representation to another (the rules)
are simply parts of the definitions of individual elements of representations, and have
no independent status whatsoever in the grammar. If they can be formulated in a simple
and concise way, so much the better: but the notion that the elements of representations
themselves should be chosen for the convenience of the rules was inconceivable.
The immediate consequence of Halle’s discussion was a change in phonology in
the direction of much more abstract representations than those permitted within a the-
ory which concentrated on biunique phonemics. But it must be emphasized that this
move was, in an important sense, an ancillary consequence of a more fundamental re-
orientation in phonological research: a shift from a concentration on the properties of
phonological representations and their elements to a much greater stress on the rules
of a grammar. Naturally, concern with questions of representations and their nature did
not disappear overnight. Nonetheless, the recognition was dawning that rules as well
had to be taken seriously as part of a grammar if language was to be examined as a
complex cognitive system rather than an inventory of phonemes, morphemes, words,
and constructions. Since the study of rules, their properties, and their organization into
linguistic systems was virtually unexplored territory, this reorientation had a much
more important effect on the nature of phonological research than the mere fact that
generative underlying representations are more abstract than biunique phonemic ones.
(Anderson 1985, pp. 321f.)
Halle’s innovation, on this view, was the focus he put on the need to get
the rules right in the statement of a language’s phonology, and not simply
to provide the right representations. These rules, as part of the content of a
speaker’s language organ, are intrinsically an aspect of I-language. So long as
linguistic theory remained focused on the (E-language) issue of how to represent
utterances in a principled alphabet, though, an argument based on the need to do
justice to the rules could have no real force, since the content of the statements
that relate phonology to phonetics had no independent external (E-language)
status of the sort the utterances themselves have.
Ultimately, the shift of attention from alphabets (inventories of basic repre-
sentational elements) and representations based on them to rules is significant
because it reflects a more profound shift in the object of inquiry, from the study
of the properties of observable linguistic events, the forms, to the study of
the knowledge speakers have of their language that underlies their production
and perception of such events. Rules are preeminently a characterization of
speakers’ knowledge, while the representations are in some sense primarily a
characterization of the forms. The change is thus a shift from the study of lan-
guage as an external, physical or social reality to the study of the structure and
organization of an aspect of human cognition: from E-language to I-language.
Now during the heyday of American structuralism, it was pretty much out
of bounds to study internalized knowledge: all there was to study was the
observable external form. But by the 1950s the world was gradually coming to
be more receptive to talk about minds, and so such a shift was at least logically
possible. The link between rules and individual cognition is quite explicit, at
least by the time of Chomsky and Halle’s fundamental statement in The Sound
Pattern of English (Chomsky and Halle 1968, pp. 3f):
The person who has acquired knowledge of a language has internalized a system of rules
that determines sound–meaning connections for indefinitely many sentences . . . [W]e
use the term “grammar” to refer both to the system of rules represented in the mind
of the speaker-hearer . . . and to the theory that the linguist constructs as a hypothesis
concerning the actual internalized grammar of the speakerhearer.
As late as 1965, when Fred Householder provided Chomsky and Halle with
a debating platform for use in going through the bases of alternative approaches
to phonology, it is clear that at least a significant fraction of the field did not (and
perhaps could not) understand the notion that linguistics might have speakers’
knowledge, rather than the properties of linguistic forms, as its proper object.
Householder was certainly a very intelligent man, and an experienced linguist,
but the very idea of linguistics as the study of an aspect of the mind was
quite incomprehensible to him. In discussing the claim of Chomsky (1964) that
“A grammar that aims for descriptive adequacy is concerned to give a correct
account of the linguistic intuition of the native speaker,” Householder (1965,
p. 14) finds that “[o]nly . . . ‘observational adequacy’ is intelligible (at least to
me) . . . it is sheer braggadocio to talk about descriptive adequacy, even if one
knew how to discover what a ‘correct account of the linguistic intuition of the
native speaker’ is.”
By the mid to late 1960s, as new generations of students appeared whose
training originated in the work of Chomsky, Halle, and their colleagues at
Massachusetts Institute of Technology, the basic point about the central im-
portance of rules – the need to get those right because they are really what the
study of language is all about – came to be more generally appreciated. But
recall that the persuasiveness of Halle’s original argument rests crucially on
one’s willingness to take seriously this need to get the rules right. And in fact
it took ten years or so after Halle’s original presentation for this to become a
generally accepted notion,8 so it is clear that whatever was responsible for the
rise of generative phonology, it probably was not simply the logic of Halle’s
conclusion about the obstructive role of phonemes in a descriptively adequate
account of Russian voicing assimilation.
So what in fact did happen to change the direction of phonologizing in
the early 1960s? A part of the responsibility undoubtedly should be laid to a
8 See Anderson 2000 for a more detailed account of the relevant history.
principle that “plus c’est la m ême chose, plus ça change.” That is, by the end of
the 1950s, phonemic theory had increasingly become a settled discipline within
which only quite minor adjustments seemed necessary (or possible). With little
left to do, new generations of students inevitably looked for new challenges –
and new approaches that would provide them. While the fundamentally distinct
scientific premises of the new theory of generative grammar may have been
apparent to its originators, students did not have to appreciate these differences
to see that something quite new and different was going on, and that they could
make real contributions to it.
It is important to understand the content of our creation myths, since they tell
us something about the structure we actually give to our world. On the other
hand, it is also important not to confuse them with explanations of how the
world actually came to be the way we find it. In the end, Halle’s argument about
Russian voicing assimilation probably did not in itself persuade the linguists
of the time to drop their externalist presumptions, their phonemes and their
exclusive focus on representations, so as to become mentalists focusing on
rules as the expression of internalized knowledge. But on the other hand, it is
exactly in the context of that development that we still have to see the logical
force of the original argument. We really only come to appreciate the sense
of this important argument after the shift in point of view that it supposedly
produced has been achieved.
It is not particularly satisfying to discover that a field can change its character
fairly rapidly for reasons that are primarily pragmatic, and not purely principled.
But on the other hand, this case is instructive, not just in its own right, but
because it suggests that the same kind of influence may have been responsible,
on a smaller scale, for a number of the changes we have seen since then (and
probably many times before).
For example, phonological theory in the period immediately before and af-
ter the publication of Chomsky and Halle 1968 was intensely occupied with
highly formal concerns, issues such as rule ordering and the role of formally
defined notational conventions in producing an explanatorily adequate theory of
grammar.9 Within a rather short period in the late 1970s, these were almost com-
pletely abandoned in favor of the study of “auto-segmental representations” –
a notion of the organization of phonological (and phonetic) representations
in terms of the synchronization of properties with respect to the time-course
of speaking. This considerable shift of attention did not come about because
auto-segmentalists solved the earlier problems, however, or even showed that
they were misconceived. Rather, it happened because auto-segmental work
9 That is, a theory which has enough internal deductive structure to ensure that for any given
set of empirical facts, exactly one grammar will be provided – and that grammar will be the
“descriptively adequate” one, in the sense introduced by Chomsky above of providing a “correct
account of the linguistic intuition of the native speaker.”
developed an impressive array of fascinating results within a rather short time,

especially in an area that had previously been rather intractable (the study of
tone). Only much later was it shown that some classical chestnuts of earlier
work, like the analysis of length, could really be solved in auto-segmental
terms, in ways that had just not been available within the framework of stan-
dard segmental phonology (see for instance the discussion in Kenstowicz 1994,
pp. 410ff. and references there).
In this chapter, we have considered the way in which the shift from a focus on
E-language to speaker’s knowledge or I-language went along with changes in
linguists’ views of the nature of phonological representation, and of the scope a
description must have if it is to be adequate to its object. We have said little, as
yet, about the form such a description should take. In the next chapter, we turn
to those matters, where we will see that linguists’ views of the nature of the
field have changed again, as consequences of the need to focus on I-language
have gradually been absorbed.
5 Describing linguistic knowledge
In the previous chapter, we traced the path of linguists’ interests in sound

structure as these evolved from an E-language-based focus on representations
alone to an I-language approach. Over time, it has come to be appreciated
that knowledge of language includes not only (representational) questions of
what speakers of a language know about the sound properties of its words,
etc., but also the characterization of what they know about overall regularities
that transcend particular items (see Anderson 1985). In the domain of sound
structure, the description of these regularities originated in important respects
from the study of what had been previously thought of as “morphophonemics”
(see section 4.2.3 above). It inherited from that work a descriptive framework
going back to one of the oldest grammatical traditions about which we have
evidence, that of ancient Indian grammarians such as Pān.ini (c. 500 BC). In
those terms, regularities are formulated as a system of rules, each of which
performs some limited, local modification of a representation. Collectively, and
in the context of a theory of the way they interact with one another, these rules
describe a mapping between phonological representation and overt phonetic
form.
Until relatively recently, linguists assumed that the description of a speaker’s
knowledge of overall regularities, of the general principles that are not part
of any individual word or other linguistic form, was essentially equivalent to
such a system of rules. Around the early 1990s, however, a challenge arose
to this position, in the form of proposals (associated with the framework of
Optimality Theory (OT)) claiming that the rule system could and should
be replaced with a hierarchy of ranked, violable constraints on surface form.
Descriptions in terms of rules and in terms of constraints are, on the face of
it, rather different, but both have as their goal a description of the linguistic
knowledge of a speaker – that is, of the language organ. We attempt below
to give some of the flavor of the issues involved, and of the similarities and
differences between the two approaches.
The discussion here is limited to phonological matters, but the consequences
of these considerations are potentially much broader. In the concluding section,
we return briefly to this matter.
92
5.1 Phonological knowledge as it appears in borrowing

In asking what a speaker of a language knows about its sound structure, most
of the relevant evidence comes from a close study of the regularities that seem
to characterize forms, their pronunciation and systematic relations among them
within the language, under normal circumstances of use. Sometimes, however,
this knowledge is put to use in other ways, whose study provides another source
of evidence about what speakers know. In phonology, one example is the way
words of one language are adapted when borrowed by another language.
We consider here some of the conclusions we can draw from one such case, the
way French words are adapted when borrowed into the West Atlantic language
Fula.1
French and Fula have rather different inventories of sounds, as well as differ-
ent structures within which those sounds appear. The sounds of standard French
are given in figure 5.1.
Consonants
labial coronal dorsal

+anterior −anterior
stops p/b t/d k/g
fricatives f/v s/z /
nasals m n
glides /j w
liquids r/l
Vowels [i, y, u, e, ø, o, , , œ, , a; ã, ˜, œ̃, ˜ ]
Figure 5.1 The sounds of French
These segmental units appear within syllables that may have zero, one or
two consonants in their onset, possibly followed by a glide, preceding an oral
or nasal vowel, followed optionally by one or two additional consonants as a
coda. Monosyllabic words illustrating some of these various possibilities are in
(5.1), where “V” represents the vowel of the syllable and “C” and “G” represent
a consonant and a glide (or semi-vowel) respectively.
(5.1) V eau [o] “water”
GV oui [wi] “yes”
CV peaux [po] “skin”
CGV moi [mwa] “me”
1 Fula is a language spoken in various forms by some 12–15 million people across West Africa.
In much of this region, the colonial language was French, and use of French in a variety of
circumstances continues in many countries. The discussion here is based on the work of Carole
Paradis and her colleagues (e.g., Paradis and LaCharité 1997); where dialect differences are
relevant, the forms are those of a Futa Toro speaker.
94 5 Describing linguistic knowledge
CCṼ plein [plε̃] “full”

CCGV croix [krwa] “cross”
CCGVC truite [trɥit] “trout”
CVC lac [lak] “lake”
VCC ours [urs] “bear”
CVCC parc [park] “park”
CCGVCC croı̂otre [krwatr] “to grow”
These can be reduced to a formula for the structure of the French syllable,
given as (5.2).
(5.2) French syllable structures: (C(C))(G)V(C(C))
The sounds of Fula are rather different from those of French. These are given
in figure 5.2.
Consonants
labial coronal dorsal laryngeal

+anterior −anterior
stops p/b t/d č/̌ k/g
m n n
prenasal stops b d j g
implosives
fricatives f s h
nasals m n
glides w j (w) 2
liquids r/l
Vowels i, u, , , a (long and short); and have closed variants [e] and
[o] in the vicinity of a high vowel or another raised mid vowel.
Figure 5.2 The sounds of Fula
The structure of syllables in Fula also differs from French. As summarized

in (5.3), Fula allows at most a single consonant as a syllable onset, followed by
either a short or a long vowel (but not a sequence of glide plus vowel or other
diphthong), followed by at most a single consonant as an offset.
(5.3) Fula syllable structures: (C)V(:)(C)
When French words are borrowed into Fula, their forms are generally changed
so as to adapt to the possibilities offered by the language. For instance, Fula
2 The segment [w] appears in both the labial and the dorsal columns, because it involves both
lip and tongue body in its articulation. The duality has consequences for the way the segment
behaves with respect to regularities of the language, as shown in Anderson 1976.
does not have the sound [

] (similar to the consonant at the end of the English
word garage), and when French words containing this sound are borrowed, it
is replaced by [s] as in the examples in (5.4).
(5.4) French Fula gloss

barrage [bara
] baras dam
collège [kɔlε
] kɔlε:s school
journal [
urnal] su:rnal newspaper
Similarly, Fula does not have the sound [v]. When French words with [v] are
borrowed, this sound is replaced (unpredictably) with one of [w,b,f] as in (5.5).

avocat [avɔka] awɔka lawyer
civil [sivil] siwil civil
verre [vεr] wε:r glass
avion [avjɔ̃] abijɔn airplane
livre [livr] li:ba:r book
vinaigre [vinεgr] binε:gara vinegar
élève [elεv] εlεf student
mouvement [muvmã] mufmaŋ movement
télévision [televizjɔ̃] tεlεfisjɔŋ television
Since Fula does not have nasal vowels, French words with those sounds
are borrowed with a sequence of oral vowel followed by nasal consonant. The
nasal consonant is articulated at the same point of articulation as a following
consonant; or, if there is no following consonant, the velar nasal [ŋ] appears.

bandit [bãdi] bann di gangster
canton [kãt ɔ̃] kantɔŋ canton
marin [marε̃] marεŋ sailor
changer [ʃã
e] sans-u-dε (to) change
gendarme [
ãdarm] sann darma policeman
ventilateur [vãtilatœr] wantilatɔr fan
French consonant clusters at the beginning of a syllable are broken up by a

vowel in words borrowed into Fula. The quality of this vowel is the same as
that of the following vowel; or, if the second member of the cluster is a glide
([w] or [ j]), a similar vowel ([u] or [i], respectively) is inserted.

briquet [brikε] birikεt lighter
classe [klas] kala:s class
drapeau [drapo] darapɔ flag
boisson [bwasɔ̃] buwasɔŋ drink
coiffer [kwafe] kuwa:f-a:-dε coif (one’s hair)
lieutenant [ljøt(ə)nã] lijεtinaŋ lieutenant
A word-initial cluster of [s] plus consonant is avoided by the insertion of a
preceding [i], as in Fula istati “statue,” from French statue ([staty]).
Syllable-final clusters in French words are also broken up by inserting a
vowel. This vowel appears either between the consonants of the cluster or
following the entire cluster, depending on which consonant of the cluster is a
liquid ([l] or [r]).

contre [kɔ̃tr] kɔntɔr against
filtre [filtr] filtir filter
mètre [mεtr] mε:tεr meter
table [tabl] ta:bal table
carde [kard] karda card (comb)
force [fɔrs] fɔrsɔ force
course [kurs] kursi course
These diverse ways of coping with French sounds and sound combinations
that do not occur in Fula may seem complex, but a little reflection will show that
they are quite conservative: in each case, as much of the content and organization
of the French word is preserved as possible, within the framework of Fula sound
patterns.
5.2 Can rules express phonological knowledge?

What is crucial for our purposes is the fact that these various systematic patterns
of adaptation provide clear evidence that Fula speakers know the sound pattern
of their language quite well. Since this knowledge governs the way foreign
words are borrowed, there is no serious doubt that it forms part of the I-language
system of speakers. It therefore behooves us to characterize that knowledge in
a way that can account for the role(s) it plays.
But if our description of the phonological knowledge of Fula speakers takes
the form of a system of rules relating underlying phonological forms to surface
phonetic forms, it is not at all obvious that that requirement will be satisfied.
For instance, French [v] has to become [f], [b], or [w] in Fula, because there are
5.2 Can rules express phonological knowledge? 97
no voiced fricatives in the language. The independently motivated phonology

of Fula itself, however, would not contain any rule turning /v/ into something
else. For one thing, phonological representations of Fula words do not contain
/v/, and so there is no need for a rule to convert it into something else; and there
is no reason to believe that any other rule of the language would have this effect.
The same is true for most of the other adaptations we have seen above.
The conclusion from this observation is that a system of “morphophonemic”
rules does not properly represent the character of I-language in the phonological
domain. The regularities of sound structure seem to have a reality whether or
not the language can be argued to contain rules that modify its own forms so as
to make them conform.
In some instances, such rules may exist: in particular, the combination of
independently motivated morphological material may give rise to configurations
that are at variance with the language’s notion of what is “well-formed.” For
instance, clusters of consonants that are not otherwise possible can be formed
by affixation; and when that happens, a rule inserts a vowel to break up the
ill-formed sequence, as in (5.9).
(5.9) a. /ɔjj-/ “cough” + /n/ “causative” −→ ɔjj-i-n “make cough”
b. /utt/ “(be) fat” + --“inchoative” + -t “repetitive” −→
utt-i--i-t “become fat again !”
Clearly, the existence of rules such as that inserting [i] in the forms in (5.9) is
motivated by (the need to conform as closely as possible to) the sound pattern
of the language. We cannot, however, say that a speaker’s knowledge of that
pattern is comprehensively expressed by the set of rules that correct violations
of it, since the range and existence of such rules is a largely adventitious side-
effect of the range of morphological and other processes in the language that
might give rise to otherwise ill-formed configurations.
Similar reservations about the adequacy of the rule-based form of description
can be motivated in other ways, in a variety of languages. English, for example,
is subject to a number of restrictions on consonant sequences, including those
in (5.10).
(5.10) a. Sequences of obstruent consonants must have the same
value for voicing (e.g., the sequences ts or zd can occur
within a word, but tz or sd cannot).3
b. Clusters of “similar” coronal consonants within a single
syllable (e.g., two sibilants, such as šs, or two coronal stops,
such as dd) are prohibited.
3 A very small number of exceptions to this generalization – words like Aztec – exist within
morphologically simple forms, but there are no exceptions in word plus affix combinations. We
disregard the exceptions here. In fact, they lose their exceptional character within a more detailed
analysis than we have space to develop here.
These limitations on clusters can be related to the surface form taken by

several productive morphological elements in English. For example, if we dis-
regard “irregular” plurals such as those of (5.11a), the regular formation of the
plural forms of nouns displays three phonetically distinct shapes, as in (5.11b).
(5.11) a. ox/oxen, radius/radii, woman/women, etc.
b. cats ([s]), dogs ([z]), horses ([əz])
It is natural to assume (see Anderson 1974, ch. 4, among many other dis-
cussions) that the English regular plural has a unitary basic shape /-z/. When
suffixed to words ending in a voiceless obstruent, however, such as cat, or to a
word ending in a sibilant such as horse, the resulting combination runs afoul
of the restrictions in (5.10). To remedy this, the rules of (5.12) are generally
assumed to form part of speakers’ knowledge of English.
(5.12) Epenthesis: Insert the vowel [ə] between two adjacent sibi-
lants in the same word.
Devoicing: A voiced obstruent becomes voiceless after a
voiceless obstruent.
These rules can be seen to yield the correct phonetic forms by way of the
derivations in (5.13).
(5.13) underlying: /kæt#z/ dɔg#z/ /hɔrs#z/
epenthesis: — — /hɔrs#əz
devoicing: kæt#s — —
surface: [kh æts] [dɔgz] [hɔɹsəz]
These same rules can be seen to govern variation in shape in several other ele-
ments (as we have already noted above, in chapter 2), especially if the epenthesis
rule in (5.12) is generalized so as to apply between two stops as well as between
two sibilants.
(5.14) 3rd sg present of verbs: eat/eats, ride/rides, raise/raises
Possessives: Rick’s [book], Fred’s [book], Alice’s [book]
Contractions: Rick’s, Fred’s, Alice’s [leaving now, left
already]
Past tense of verbs: pick/picked ([t]), raise/raised ([d]),
wait/waited ([əd])
In these and many other cases, we can formulate a set of regularities of the
language, and we see that the rules in our description serve to “enforce” those
regularities when configurations arise that violate them.
But there are clearly problems in the claim that the knowledge involved
is exhaustively characterized by these rules. For example, consider English
leak/leaked vs. drug/drugged, as in the last set of forms in (5.14). We assume
the basic form of the regular past ending is /-d/, as suggested by vowel-final
verbs such as delay/delayed. The devoicing of the past tense ending in picked,
leaked, etc. is clearly due to the regularity that English final clusters of obstruents
must agree in voicing, as enforced by the devoicing rule in (5.12).
Now consider leave/left, lose/lost, and other similar verbs. These involve
an alternative form of the past ending, one which we also find in mean/meant,
deal/dealt, etc., and which we might represent as basic /-t/. But then in left (from
/lijv#t /), lost (from /luwz#t/), we seem to have quite a different rule applying
from the devoicing rule in (5.12): one that devoices the end of the stem, not
the suffix consonant. Despite this difference in effect, though, the two clearly
both enforce the same generalization: that of (5.10a). Somehow the grammar
containing these rules is not actually capturing this overall regularity. This is
a variation on the same kind of insufficiency we saw before in the ability of a
language’s rule system to account for patterns of adaptation in borrowing.
5.3 Constraint-based theories of phonological knowledge

Observations like those of the preceding subsection led phonologists to consider
the possibility that a set of constraints on phonological form, rather than a set
of rules converting one representation into another, would better represent the
basic knowledge speakers have about sound structure. And in the early 1990s,
this led to the development of a concrete theory known as Optimality Theory
whose first full programmatic formulation was that of Prince and Smolensky
1993.
5.3.1 Constraints vs. rules: patterns vs. effects

A basic rationale for OT might be formulated as the claim that cross-
linguistically valid regularities of phonological form are to be found in output
configurations – that is, in the shapes of linguistic forms as they are pronounced –
rather than in input (lexical) configurations or in the details of rules mapping
one onto the other. The importance in particular of “cross-linguistically valid
regularities” is of course that these are the most likely to derive from aspects
of Universal Grammar (UG), and thus to be in some sense definitional of the
language organ.
The approach followed by generative phonologists up to that point had been
to proceed from an interpretation of the phonemic principle (4.4) to construct
a system of underlying phonological forms. These have two related desiderata,
as in (5.15).
(5.15) a. Abstract away from surface differences that are due to reg-
ularities of the language; and
b. To the extent possible (i.e., except for cases of suppletion),

attribute the same phonological shape to a given linguistic
element (word, morpheme) in all environments in which it
appears.
The rules of the grammar then, exist (a) to state the regularities of sound
structure in the language; and (b) to relate the abstract phonological forms of
linguistic elements to their phonetic realizations in various contexts. The idea
pursued by Prince and Smolensky and which has driven OT is that this approach
is ultimately unproductive in satisfying the main goal of phonological theory:
to provide a substantive definition of what constitutes a possible phonologi-
cal system for a natural language. While it is obviously important to provide
accurate descriptions of individual languages, the task of understanding UG
requires us to provide a more general account of the content and organization
of I-language.
Traditional generative phonology thinks of a grammar as a collection of
rules, each of the form A → B/C D. Such a rule looks for input sequences
of the form CAD and performs an operation of the form A → B (“A takes on
property B”) on them. But “[f]or this format to be worth pursuing, there must
be an interesting theory which defines the class of possible predicates CAD
(Structural Descriptions) and another theory which defines the class of possible
operations (Structural Changes).” These theories have proven to be “loose and
uninformative,” and thus we should conclude that “the locus of explanatory
action is elsewhere” (Prince and Smolensky 1993, p. 3).
The point here is that the rules themselves do not really seem to be very
useful in arriving at generalizations about universal properties of phonologi-
cal form. We can try to establish generalizations about what sorts of things
rules can do, but all such theories seem to allow for the formulation of lots of
things we “know” to be impossible. This suggests we should look elsewhere
for explanations.
Furthermore, theories of rules have been limited to theories of individual
rules. Even the best theory of the Structural Descriptions and Structural Changes
of particular rules misses the overall nature of phonologies: that sets of rules
have a coherence that cannot be seen in the individual rules themselves.
As illustrated by the two rules for the pronunciation of consonant clusters in
inflectional forms proposed above, devoicing can affect either the stem-final
consonant or the ending itself depending on which form the ending has. These
two rules are in principle quite independent of one another, but together clearly
express a single regularity.
Now in the 1970s and 1980s there were a variety of proposals made to the
effect that the basis of phonological rules was to be sought in their effects:
that is, that there were various regularities of surface pattern that provided the
motivation for the differences between underlying and surface form that are
expressed by individual rules. A language does not have an epenthesis rule
because it likes epenthesis, but rather because as a result of this rule, it will
avoid ill-formed clusters (as in English inflection, discussed above, or the rule
of Spanish that avoids initial /sC/ by inserting an initial [e] in words like España).
A language has assimilation rules not because of the way they work, but because
as a result, all clusters will be homogeneous in some property.
Often it seems to be the case that a language has multiple rules, each of
which by itself is only part of the picture, but which taken together have the
effect that some pattern exists on the surface. Thus English assimilates voice
progressively in inflection (/kæt + z/→[kh æts]), but regressively in some other
cases of inflection (lose/lost), as we have seen just above. The two formally
quite different rules have one surface effect, a matter which we will take up in
more detail below.
These “conspiracies” (a term introduced by Kisseberth 1970a, 1970b) seem
to have a natural formulation as ways to satisfy some constraint on surface
representations. Suppose we take this effect as the locus of explanation in
phonology. Then we can attempt to develop a theory of how representational
well-formedness determines the assignment of phonological structure: a theory
of constraints and their interaction, as opposed to a theory of rules.
The nature of these constraints has been the subject of intense investigation
in recent years. An important basic notion is that constraints are instantiations
of universal aspects of sound structure – hence, they are the stuff of UG. Con-
straints address representational well-formedness (rather than the mechanics
of converting one representation into another), and it is presumed that most of
the content of this notion is due to the structure of the human language faculty,
rather than to arbitrary interlinguistic variation.
A conceptually important difference between OT and related theories lies in
the claim that constraints can be operative in a language even when they are
not necessarily true (or satisfied) in every form. The members of a given set of
constraints are typically in conflict, and not mutually consistent: satisfying one
constraint may require the violation of another. The way a particular language
resolves these conflicts is what characterizes its particular phonology as opposed
to those of other languages.
We can make this concrete by suggesting that there are two fundamentally
conflicting demands in sound structure:
(5.16) Markedness: the tendency for phonetic forms to be pro-

nounced in a simple, natural way (as determined in part by
the nature of speech articulation, acoustics, and audition, and
in part perhaps by more abstract cognitive factors – all aspects
of the human language faculty).
Faithfulness: the tendency for properties that distinguish

lexical elements from one another within a given language to
be preserved in the phonetic realizations of those elements.
The notion of markedness certainly has a long series of antecedents. Attempts

to incorporate it into generative phonology, however, had taken a form that
effectively allowed the “unmarked” status of an element or configuration to
figure in a grammar only to the extent it was true or completely satisfied.
Relative markedness basically has no expression in systems such as that of
Chomsky and Halle 1968: rules could be evaluated as “cheaper” to the extent
they contributed to a universally unmarked set of possibilities, but this was
essentially an all-or-nothing effect. OT, in contrast, allows for the possibility
that a markedness effect can contribute to the sound pattern of a language
even though some (perhaps even most) of the language’s actual forms
violate it.
Consider these issues with respect to an example. In terms of the places of
articulation found in the world’s languages, labials, dentals, and velars are all
quite generally found, but labialized velars (e.g., [kw ]) are much less common.
We could formalize this as a (markedness) constraint to the effect that labialized
velars are prohibited:4
(5.17) *kw
In a language like English, where there are no labialized velars, the effects
of this constraint are absolute. We could express this by saying that “even if
the lexical representation of an English word had a lexical labialized velar, it
would be pronounced without velarization.” Such a markedness constraint in-
evitably comes into conflict with the basic faithfulness property, expressed by
a constraint to the effect that lexical values must be preserved. A language like
English can then be described by saying that in such a language, the marked-
ness constraint (5.17) takes precedence over (or “dominates”) the faithfulness
constraint (5.18).5
(5.18) IdentIO (Round)
In the Native American language Kwakw’ala (often called “Kwakiutl”),

however, labialized velar stops are found in words like [bəgw anəm] “man”
or [kw ag u ĺ] “Kwakiutl.” We can account for this by saying that the same two
4 The content of this statement is that a surface representation is disfavored to the extent it contains
instances of rounded velar consonants.
5 The content of this constraint is that a representation is disfavored to the extent underlying or
input values of the property [Round] are not preserved in the output.
constraints apply to both languages, but that in Kwakw’ala faithfulness (5.18)

dominates markedness (5.17) in this instance.
Actually, the facts about labialized velars are a bit more complicated than
that. It is true that they are not common; but under some circumstances, they
are actually preferred. In particular, when an adjacent vowel is round, the velar
often is too. We might say, in fact, that it is dis-preferred to have a sequence
of a vowel plus a consonant that differ in rounding. An approximation to this
constraint might be (5.19):
(5.19) *[+syll,+round][−syll,−round]
Consider a language like Menomini (Bloomfield 1962), in which k following
a round vowel is itself round (as in [okw i:? san] “his son”), but following a
non-round vowel is not round (as in [meki:? s] “my son”). We could describe
this by saying that in general, the markedness of labialized velars prevents their
occurrence; but in the specific circumstance where they follow a rounded vowel,
these occur to the exclusion of unrounded ones. We get this result by saying
that the contextual markedness condition dominates the context-free one, and
both dominate faithfulness.
Another possibility is illustrated by Kwakw’ala. In this language, we have
both plain and labialized velars (k’asa “beat soft” vs. xwasa “dance”). After
the vowel [u], however, only [kw ] appears. The element /-kas/ “really” appears
without rounding in [gæ ĺa-kas] “really long” as opposed to rounded (after [u], a
round vowel) in [əwu-kw as] “really big.” Here faithfulness dominates context-
free markedness, but both are dominated by contextual markedness.
In English, Menomini, and Kwakw’ala, we have languages whose sound pat-
terns differ in systematic ways with respect to the appearance of labialized
velar consonants. Those differences in sound pattern are not easily expressed
as differences in the content of the system of morphophonemic rules applicable
to each language. We can, however, characterize the ways in which the know-
ledge of English, Menomini, and Kwakw’ala speakers differ in terms of the
relative dominance of members of essentially the same system of constraints.
The system of UG provides us with the relevant constraints, at least in some
form. What differentiates one language from another (and thus, what individual
speakers know about their language in the domain in question) is the way these
constraints relate to one another. In other words, a specific language organ is
described not in terms of a collection of rules, but rather in terms of the relative
ranking of the constraints made available by UG.
Note that unrounding [u] would be another way for a language to avoid
violating (5.19): to the extent this solution is not chosen, it must be prevented
by an even higher ranking constraint (IdentIO(V)), requiring that the input
features of vowels be preserved. If this outranks (5.17), then the violation of
(5.19) will have to be repaired by rounding both (rather than unrounding both).
Constraints thus come in several flavors:
(5.20) Markedness constraints (the tendency for phonetic forms to

be pronounced in a simple, natural way):
Context-free markedness constraints describe overall
preferences for certain feature configurations over others
within a single segment; and
Context-sensitive markedness constraints describe
preferences for certain feature configurations over others in
combinations of segments.
Faithfulness constraints (the tendency for properties that
distinguish lexical elements from one another to be preserved
in their phonetic realizations):
MaxIO constraints require that every element in the input
have a correspondent in the output;
DepIO constraints require that every element in the output
be the correspondent of some element in the input; and
IdentIO constraints require particular properties of output
segments to be identical to those of the corresponding input
segments.
We have as yet said nothing about just how a system of constraints (such
as (5.17), (5.18), etc.) allows us to compute a surface form corresponding to a
given input. In essence, this process consists in a comparison among all of the
formally possible surface forms that might correspond to that input (a set of
“candidates”), resulting in the selection of that candidate that best conforms to
the system of ranked constraints. The grammar thus consists of two components,
called Gen and Eval. Gen operates on input representations to produce a
set of candidates; these, in turn, are assessed by Eval. The candidate with the
highest degree of harmony (i.e., the one which violates highly ranked constraints
to the smallest possible degree) is (by definition) optimal, and is thus chosen
as the output.
When we investigate languages from this perspective, what we find is that the
same set of constraints can describe a number of different systems, depending
on their relation to one another. In any given language, the constraints are
organized in a hierarchy, and then contribute to the determination of correct
surface forms via principles of “harmony” that include those of (5.21).
(5.21) Optimality: An output is “optimal” when it incurs the

smallest possible violation of a language-specific hierarchy of
constraints.
Strict Domination: Any candidate which violates the
highest-ranking constraint is eliminated, regardless of its
evaluation by lower-ranking constraints, if candidates are avail-

able which do not violate the higher-ranking constraint.
How does Gen do its work? This is a matter of some dispute among practi-
tioners of OT. One view is that Gen produces, for any input, the full range of
possible well-formed expressions over the alphabet of phonological represen-
tations. This requires Eval to be able to screen out vast masses of irrelevant
candidates with particular focus on the few that are legitimate possibilities.
While it has never really been shown how this is computationally possible with
finite resources, something along these lines is commonly assumed in theoret-
ical discussion.
Another possibility (one implicit in the notion that Gen “operates on input
representations”) is that Gen is smart enough to know, for a given input form,
what forms could conceivably be at least possible output candidates. But that in
turn requires that Gen incorporate some intelligence, mechanisms that sound
suspiciously like a set of phonological rules. And of course if the same old set
of rules turns out to be necessary as a (covert) part of the grammar, it is not
obvious how different this system would be from that of classical generative
phonology. These issues remain to be clarified, especially if OT is to be taken
seriously as providing a model of how speakers bring what they know to bear
on what they do, but at least the conceptual outlines of such a constraint-based
theory are relatively clear.
Of course, apart from the constraint system provided by UG (whose inter-
nal ranking characterizes an individual language), a speaker’s knowledge also
includes information about individual linguistic forms: the lexicon, a set of
representations including phonological, morphological, syntactic, and seman-
tic information. We will discuss the organization and content of the lexicon
in chapter 7, but one strictly phonological issue in the structure of the lexicon
forms an important aspect of OT. The goal of this theory is to characterize the
phonological knowledge a speaker has of his or her language entirely in terms
of a system of (ranked) constraints applying to output forms. It follows, there-
fore, that no constraints of this sort should crucially hold at the level of input
forms. This notion (known as the hypothesis of the richness of the base)
has quite interesting consequences for the form of our description.
For instance, in English, a language with no labiovelars, the absence of such
segments follows entirely from the constraint ranking in (5.22).
(5.22) *kw (“dominates”) IdentIO(Round)
We do not need to say explicitly, in addition, that there are no underlying

labiovelars: as we noted above, it would make no difference if such segments oc-
cured in inputs, since they would never survive into output forms. This suggests
a general condition of Lexicon optimization: where a number of distinct
inputs would all yield the same output, pick (as the lexical representation) the
one whose output incurs the fewest violations. What that means in practice is
that the input which is closest in its properties to the output excludes any of the
others from the lexicon.
5.3.2 An example: voice assimilation

Throughout this discussion, we have presented as a basic motivation for
constraint-based account of phonological knowledge the argument that the ele-
mental units of that knowledge are sound patterns in language, not the specifics
of the relation between underlying and surface forms. For a concrete example,
let us return to the facts about English consonant clusters in inflected forms
that we have already touched on several times above, and where the unity of
formally distinct effects is perhaps clearer.
It is well known that clusters of obstruents differing in voicing are quite rare,
and that most languages have ways of dealing with these when they would
be expected to arise. Let us assume that voicing is specified by a property (or
distinctive feature) ([Voice]) in obstruents. Voiced obstruents like [b,d,g,z] etc.
will possess this property, while voiceless obstruents like [p,t,k,s] etc. will lack
it. Then we can say that there is a general markedness constraint requiring
clusters of obstruents to be uniform with respect to this feature.
(5.23) agree(voice): If one member of a sequence of adjacent

obstruent consonants has the property [Voice], every other
member of the sequence should have that property as well.
This constraint, a part of UG, dis-prefers structures like those of (5.24), since
in each case the property [Voice] is associated with only one member of the
sequence to the exclusion of the other.
(5.24) a. [Voice]
[+Obstruent] [+Obstruent]
b. [Voice]
[+Obstruent] [+Obstruent]
Let us see how we might invoke this apparatus to derive a familiar pattern of
phonological variation (known as voicing assimilation) found in English
inflections such as the plural or past tense, as discussed above. For input
/kæt + z/, /lajk + d/ we should prefer the output forms [kæts] and [lajkt]. The fact
that the lexical value of voicing is not generally preserved here suggests that a
faithfulness constraint must be being violated. Let us call it IdentIO(Laryngeal).
(5.25) identio(laryngeal): Consonants should be faithful to input

laryngeal specifications.6
A constraint like (5.25) actually stands for two subcases: on the one hand,
input specifications should be preserved; and on the other, output specifications
should reflect input. Jointly, these two aspects of (5.25) have the effect that
input specifications of [Voice] should correspond exactly with those in the
output. Clearly, since input specifications of [Voice] are not in fact always
reflected faithfully in the output (e.g., the final voiceless segments of cats, liked
correspond to segments that are voiced in the input), some other constraint
must outrank (5.25) in the English constraint hierarchy. The relevant constraint
which has this effect, of course, is Agree(Voice), (5.23).
But that does not determine a unique output form. Corresponding to inputs
such as /lajk + d/, for example, both [laikt] and [lajgd] would satisfy (5.23),
each incurring a single violation of (5.25) and otherwise the same. What aspect
of the grammar of English, then, results in the fact that we prefer [laikt] over
[lajgd] as the realization of /lajk+ d/?
The answer to this is probably to be sought in more general considerations
in UG. In the languages of the world, voiceless obstruents are much commoner
than voiced ones, and while it is quite rare for a language to make use of voiced
obstruents without matching voiceless ones, the opposite is not at all unusual.
This suggests that overall, the property of voicing is dis-favored in obstruents,
a fact that we could represent as (5.26).7
(5.26) *voiced obstruent: avoid voiced obstruents.
Since English obviously has voiced obstruents, this constraint must be out-
ranked by others, such as (5.25) which requires the preservation of input values
if possible. But in comparing [lajgd] and [lajkt] as possible outputs correspond-
ing to /lajk+ d/, we saw that each incurs exactly one violation of (5.25) – the
stem-final consonant in one case, and the affix in the other, but each with the
same overall value, a violation forced by the higher ranked (5.23). Since these
6 Including, among others, [Voice], a (possibly complex) property characterizing the activity of
the vocal folds.
7 This is presumably related to the articulatory fact that airflow conditions during the production
of obstruent consonants are such as to inhibit vocal fold vibration, and some additional gesture
is necessary if voicing is to be maintained under these circumstances. Vowels and sonorant
consonants, in contrast, are articulated in such a way that vocal fold vibration is a normally
expected concomitant, and indeed some additional gesture is necessary in these sounds if voicing
is to be inhibited. Here part of the content of UG can be grounded in – though probably not
mechanically deduced from – properties of the speech production system.
two are thus tied with respect to their faithfulness to the input, the choice be-
tween them falls to (5.26), which tells us to prefer [lajkt] over [lajgd] since the
former contains no voiced obstruents.
If we were to formulate this effect in terms of a rule, it would look something
like (5.27).
(5.27) [+Obstruent] → [−Voice]/[−Voice]+ —
Such a rule would quite adequately express the fact that underlyingly voiced
affixes are devoiced when added to a stem ending in a voiceless segment. We
can show, however, that it would not suffice to describe the knowledge about
voicing assimilation which English speakers bring to bear in determining the
forms of their language.
As we mentioned above, English regular verbs form their past tense (and
past participle) by adding a uniform ending whose basic shape appears to be
/-d/ (although it also appears as [t] or [ d] when required by the shape of
the stem). Among the classes of irregular verbs, however, we find a number
(e.g., learn/learnt, burn/burnt, mean/meant, deal/dealt) which involve a similar
but distinct ending whose basic shape appears to be /-t/. This ending shows a
different pattern of assimilation from the “regular” /-d/: instead of altering
the shape of the ending to assure compliance with a requirement that voicing
be uniform in clusters, in this case it is the stem that changes, in verbs like
leave/left, lose/lost. Notice this direction of assimilation is also what we need
for derived forms such as describe/descriptive (cf. retain/retentive for the form
of the ending), absorb/absorption, five/fifth, etc.
The rule in (5.27) is not adequate to describe this situation, although we can
easily formulate a rule that is.
(5.28) [+Obstruent]→[−Voice]/ — +[−Voice]
(5.28) is clearly a different rule from (5.27), formally quite distinct despite the
fact that both are basically ways of enforcing a single regularity. Such dupli-
cation, where we need multiple formally unrelated rules to express a single
generalization, seems quite unsatisfactory if our aim is to describe the know-
ledge a speaker of English has about the language.
But now notice that this redundancy or duplication can be eliminated if we
base our account on constraints rather than rules. To deal with outputs such as
[lεft] corresponding to input /lijv + t/,8 we do not in fact have to add anything
to the constraint system as we have elaborated it above. As we saw, this system
will prefer outputs with uniform voicing to ones that are strictly faithful to their
8 In accounting for these verbs, we must obviously also include an explanation of the alternation
in the stem vowel ([ij]↔[ε]), but that issue is orthogonal to the one involving voicing which we
are discussing here.
5.4 The extension of constraint-based description 109
input values; it will also prefer devoicing (either regressively, right-to-left, or

progressively, from left-to-right) and not voicing as the optimal way to achieve
this, given the general markedness constraint (5.26).
In fact, the same constraint system also describes yet another way of enforcing
the same overall regularity. In Swedish, we find bidirectional spreading of
voicelessness, as in the examples of (5.29).
(5.29) a. hög “high”; högtid [hœkti:d] “festival”
b. viga “to marry”; vigsel [viksəl] “marriage”
c. dag “day”; tisdag [tista] “Tuesday”
d. skog “forest”; skogsbrand [skoksprand] “forest fire”
e. äga “to own”; ägde [ægdə] “owned”
In this case, the assimilation may go in either direction, but we see that
whichever segment is altered, the result is to prefer a (uniformly) voiceless
cluster to a (uniformly) voiced one. We derive this result again by saying that
agreement in voicing (5.23) outranks laryngeal faithfulness (5.25), while (5.26)
tells us that (other things being equal) it is better to devoice both than to voice
both as a way of ensuring homogeneity of voicing in a cluster.
The constraint-based approach thus allows us to describe three rather different
patterns of voicing assimilation (progressive, regressive, or bi-directional) as
instances of the application of the same underlying knowledge, knowledge
which we can attribute in substance to constraints provided by UG (and ranked
within the particular grammars of individual languages). Notice further that even
if English did not have any actual alternations in voicing at all, this knowledge
could still play a role in the adaptation of borrowed words: we predict that
where such words are borrowed into a language with such a constraint hierarchy,
clusters that differed in voicing in the original language should be replaced by
uniformly voiceless ones as a part of the nativizing process.
5.4 The extension of constraint-based description

In the sections above, we have explored some reasons to believe that a system of
constraints implemented in the general way proposed by OT may provide a more
accurate description of the content of a speaker’s linguistic knowledge than a
system of rules that specify the mechanics of converting input to output forms.
The replacement of rule-based theories by constraint-based ones is certainly
not complete (even in phonology), and many other details of the architecture
of the language faculty remain to be explored in order fully to understand the
nature and role of such constraints. Nonetheless, it is fair to say that the bulk
of research in phonology at present is being conducted within this general
framework; and furthermore, that the replacement of rules by constraints has
followed from a greater appreciation of the demands of a theory that purports
to offer a description of speakers’ knowledge, not just of the forms speakers

employ.
It is natural, then, to ask whether this result is confined to the study of phon-
ology. In fact, recent years have seen a number of efforts to apply the analytic
methods of OT as practiced by phonologists directly to syntactic problems: see
papers in Barbosa et al. 1998 and Dekkers, van der Leeuw, and van de Weijer
2000, among others. But regardless of whether the specifics of theoretical
apparatus transfer literally from one domain of linguistic structure to another in
this way, there is a more general trend of which OT is a part, which is certainly
relevant to areas other than phonology.
Just as generative phonology was originally focused on the notion of rules
as they describe the specifics of how one representation is related to another, so
generative syntax originally concentrated on rules (transformations) describing
specific constructions in specific languages (the passive in English, causatives
in Turkish, etc.). In the 1980s, however, talk of such specific rules came to
be replaced by talk of more general principles and parameters of grammatical
structure, notions not specific to any particular construction but rather char-
acteristic of syntactic form overall.
Along lines like those pursued in chapter 3, exploration of the properties of
Binding Theory, Case Theory, etc. led to greater explanatory depth in syntactic
theory as a general account of the human language capacity. Most current
work in syntax seeks general properties of linguistic structure from which the
specifics of individual constructions follow, rather than the other way around.
Though much of the concrete theoretical apparatus of the resulting theories
differs signficantly from that of OT, the two can be seen as intellectually similar
responses to the challenge of characterizing the human language faculty, and
both are opposed to earlier E-language-oriented attempts to characterize directly
the external phenomena of language.
There is no obvious reason to believe a priori that sound structure and syn-
tactic structure are governed by exactly the same sorts of principle, but it is
relatively clear by now that knowledge of language is better characterized by
systems of principles, parameters, and constraints than by construction-specific
(or configuration-specific) rules of transformation. In that sense, OT is simply
one instance of a broad class of theories that have arisen as a result of the shift
from E-language to I-language as the object of inquiry in linguistics.
6 Phonetics and the I-linguistics of speech
Our knowledge of a language is determined by the language organ we develop

as a child on the basis of exposure to utterances in that language, and includes
what we know about contrasts, relations, and regularities within the set of
linguistic objects. Obviously, though, it also includes what we know about the
objects themselves. The structure of that knowledge is described by a theory of
representations of the various sorts of object that form parts of our language.
Seeing the foundation of these representations as an aspect of our knowledge
(an I-language point of view) has somewhat different consequences from seeing
them as based purely on externally determined properties, part of E-language.
There may be much formal similarity between the actual representations that
result from these two differing perspectives, but the conceptual content is still
quite distinct.
In this chapter, we address the nature of the representations that seem to
be most obviously and irreducibly based on observable, physically measurable
properties: phonetic representations. We argue that when phonetics is seen as
genuinely part of language, rather than a subpart of physics or physiology,
the resulting conception of “phonetic representation” (while still recognizable)
differs in a number of important ways from what is often taught (or more
accurately, assumed) in standard textbooks.
6.1 Representations and the study of sound structure

Most linguists assume, as we argued in chapter 4, that the principles of sound
structure in a given language mediate between a p honologi cal repre-
se n t at i on that indicates all and only the properties of an utterance in terms
of which it contrasts with other utterances in that language, and a phoneti c
r e pr e s e n t at i on that provides a language-independent characterization of
its pronunciation.
The nature of phonological representations has occasioned a great deal of
discussion and a vast literature. Some of the most radical proposals within OT
have argued that representations of any sort other than surface phonetic ones
are redundant, but the linguistic significance of some sort of phonological
111
112 6 Phonetics and the I-linguistics of speech
representation for linguistic items and utterances has generally not been in
doubt. Many readers may be surprised to learn, though, that the status of
phonetic representations themselves in linguistic theory has not always been
quite so clear.
To be sure, there has often been some question about the extent to which pho-
netics is properly part of linguistics at all. If this kind of investigation of the artic-
ulatory, acoustic, and perceptual properties of concrete acts of speaking is essen-
tially a matter of more and more precise measurement of physiological, physical,
and neurological events, it seems to have little to do with linguistic structure
per se, especially if we construe the latter as primarily cognitive in its basis.
Phonetics would have the status of an auxiliary discipline – overlapping with,
but properly included within, physics, physiology, and the neurophysiology of
the auditory system – that simply described the externally observable proper-
ties of the abstract objects with which linguistics is concerned. As Trubetzkoy
(1939) put it, “phonetics is to phonology as numismatics is to economics.”
We argued in section 4.1 of chapter 4 that the kind of representation
generally called “phonetic” is a significant abstraction from the raw physical
facts. Nonetheless, few would question the premise that acts of speaking do
have some observable properties, and that the business of phonetics is to settle
the facts of the matter as to what these are, as far as the language system
is concerned. Relevant results of such observations can then be presented
in some appropriate form, and who could question that such a “phonetic
representation” describes the things phonologists have to account for?
Leonard Bloomfield, in contrast, argued that there is no linguistic signifi-
cance to phonetic representations (cf. Anderson 1985, pp. 262ff.). His point was
that insofar as these deviate in any way from a full physical record of the speech
event (such as might be provided by a tape recording, supplemented with
cineradiographic and other records of the articulatory details), they represent an
arbitrary selection of some properties to the exclusion of others and cannot be
said to be based on theoretically interesting principles. As such, they serve more
as a record of biographical details about the phonetician (what properties he or
she has learned to record and what to omit) than as a theoretically significant
record of a linguistic event. Somewhat similar objections have been revived
(on different grounds) in an updated form by Pierrehumbert (1990).
Bloomfield’s objection is a very serious one, and one to which linguists have
not always devoted enough attention – if only to be clear in their own minds
about why they reject it. Why, after all, should we attribute to some particular
subset of the physical properties of utterances the status of a fundamental
characterization of language? The essential nature of language is that of a
system of tacit knowledge as represented by the language organ, an aspect
of the organization of the mind and the brain. In that light, the selection of
some external properties of linguistic utterances as systematic to the potential
exclusion of others requires at least some justification.
6.1 Representations and the study of sound structure 113
In this chapter, we will defend the claim that there is a significant notion
of “phonetic representation,” one that is distinct both from a phonological
representation and from a complete physical record of a speech event. This is
part of our I-language system, and thus it merits the attention of linguists. The
kind of representation to which we wish to attribute this status, however, is
at least at first blush rather different from the sort of thing linguists typically
teach their students to produce in a beginning phonetics course.1
Let us begin by asking about the factors that contribute to determining the
physical properties of an act of speaking. A talker, let us assume, has something
to say and initiates a sequence of gestures of the vocal organs which affect the
surrounding air and are thus conveyed, perhaps, to the ears of potential listeners.
Any particular speech event of this sort can be regarded as resulting from the
interaction of a number of logically distinguishable aspects of the system that
implements that intention:
a. The talker’s intention to produce a specific utterance (i.e., the properties that
characterize the particular linguistic items – words, etc. – that compose it);
b. The fact that the utterance is produced by a speaker of a particular language
(i.e., patterns of neuromuscular activity characteristic of the sound pattern
of the particular language being spoken);
c. The fact that the utterance is a speech event (i.e., that its production invokes
neurophysiological and motor control mechanisms that are brought into play
in speech in general, as opposed, for instance, to the control regimes that are
relevant to swallowing, breathing, etc.); and
d. The physical and physiological properties of the speech apparatus, the acous-
tics of such systems, etc.
Since we can decompose speech into its articulatory, acoustic, and perceptual
aspects, we might envision (at least) three separate representations, one in each
domain. Alternatively, we might seek a single representation that unifies all of
these sorts of property in terms of one set of independent variables. Without
going into the matter in more detail here, we should make it clear that we are
quite persuaded by the arguments of advocates of a Motor Theory of Speech Per-
ception (see Liberman and Mattingly 1985, Mattingly and Studdert-Kennedy
1991 for discussion) to the effect that the primes of phonetic specification lie in
the articulatory domain, and not (directly) in the acoustics or in perception. This
decision goes against much work in automated speech recognition, for example,
which tends to be resolutely grounded in a “bottom up” approach to recovering
linguistic structure on the basis of the structure of the acoustic signal alone.
Even granting the apparent difficulties that arise in the effort to specify an
architecture for perception that implements a motor theory, we think that the
1 Another recent proposal involving a notable expansion of the notion of “phonetics” beyond the
traditional is that of Kingston and Diehl 1994. The analyses and proposals of these authors are
in some ways similar to the point of view presented below, though they differ in other technical
and substantive respects that we do not go into here.
problem for those who would model perception is to find a way to implement
such an architecture. The tendency in much technologically oriented work on
speech perception and recognition is rather to retreat into the notion that some-
how the invariants must be out there in the acoustics if we will only keep
looking, because perception would be more straightforward if they were. This
is simply another example, we feel, of the drunk who persists in looking for his
keys under the streetlight because the light is best there. We could refine our
view of just what a Motor Theory is, but for present purposes that is not really
necessary. Our reason for bringing the matter up at all is simply to be explicit
about the assumption that it is articulatory activity that we wish to characterize.
6.2 A linguistic basis for phonetic representation

Returning to the problem of justifying the attribution of linguistic significance
to a phonetic representation, suppose that we could arrive at a representation
of the speech event that met the conditions in (6.1), one which characterizes all
and only those aspects of it that are under the control of the linguistic system
(as opposed to those that are language-independent consequences of other
factors). We submit that this could then be said to be genuinely linguistic, rather
than accidental and external, and to furnish an appropriate notion of “phonetic
representation” for inclusion in linguistic theory. Such a representation would
clearly describe speech events from the point of view of the language organ.
(6.1) A phonetic representation characterizes all and only those as-
pects of a speech event that are under linguistic control, in the
sense of being managed by the system of linguistic knowledge
that the talker brings to bear in performance.
That means that we certainly want to include indications of properties that
distinguish utterances from one another within a language. We also want to
include indications of non-contrastive properties of an utterance that are under
the control of the specific linguistic system within which it is produced – an
indication, for example, of whether the occlusion of utterance-final stop con-
sonants, or of stops in clusters, is released or not (recall the discussion of un-
released stops in Korean in chapter 4). This property seems not to be used
contrastively in any language. That is, there is apparently no language in
which, for instance, [tap] and [tap] might be distinct words. Despite this,
it is still the case that languages can differ from one another in terms of it
(cf. Anderson 1974), which means it is manipulated by the systems of individual
languages.
On the other hand, we want to omit many things which we could, in principle,
measure and record. By and large, for example, we can neglect the activity of the
epiglottis in speech – not because it is completely inactive (it is not) or because
its position has no auditory consequences (also untrue), but rather because, as
far as we can tell, epiglottal activity is not something we manipulate per se, a
dimension which is controlled differently in some languages than in others.
This is perhaps a controversial thing to assert, but let us be clear on why:
some phoneticians have reported that in some languages epiglottal activity is
manipulated for linguistic purposes. But in the absence of clear support for such
claims, we would want to exclude the epiglottis from phonetic representation
in spite of its physiological and acoustic importance unless and until it can be
shown that epiglottal activity is independent of other articulatory events, events
which correspond to dimensions of irreducible linguistic importance.
What we seek, then, is a representation of all and only those aspects of a
speech event that are under linguistic control, in the sense of being managed by
the language organ: the system of linguistic knowledge that the talker brings
to bear in performance. Another way to put this is to say that we want to
characterize everything in the talker’s linguistic intention, as opposed to aspects
of the event that follow from the physical, neurophysiological, and other extra-
linguistic properties of the apparatus that is employed in talking. Providing
such a representation would respond substantively to Bloomfield’s objection,
by grounding the properties attributed to it in their role in the cognitive system
that constitutes our linguistic knowledge. It would certainly be distinct from a
full physical record of the speech event, since it would explicitly abandon the
attempt to describe everything that is true of this event in favor of a description
of everything that is linguistically determined about it.
One particular view of language in which the place of such a representation
is clear is that of Chomsky’s recent m inim alis t analyses (Chomsky 1995).
In that approach, there are only three significant kinds of representation. Post-
poning one of these, the nature of lexical items, for chapter 7, the two key
representations are the interfaces to sound, on the one hand (phonologi cal
fo r m, or PF), and to meaning, on the other (logical form, or LF). The first
of these is intended to characterize all and only the aspects of an utterance’s
form that are managed by the linguistic computational system, and it must be
(exactly) adequate to serve as an interface to the language-independent systems
of articulation and perception. PF in these terms is exactly the representation
we seek.
6.2.1 Measurable but “unintended” effects

It is, however, far from simple to construct such a representation. Disentangling
the properties of the speech event that are due to linguistically formed intention
from those that are due to the physics and physiology of the vocal apparatus, or
to general motor control mechanisms that are operative in speech, is far from
trivial. Particular properties certainly do not come labeled as to their source, and
there are few generally applicable diagnostics in this area that can be applied
directly to observations of physical events of speaking.
Pitch microcontours We have said that the range of distinctions that

separate linguistic elements from one another in a given language is a matter
that should be resolved in the phonological representations characteristic of that
language. However, the way in which particular phonological distinctions are
realized phonetically is also a part of the phonology of the language. Phonetic
representation must therefore provide us with a range of descriptive properties
adequate to differentiate languages from one another in these terms. As a special
case, we must also be able to discriminate between, on the one hand, phonetic
differences that correspond to different ways in which the vocal apparatus is
controlled in distinct languages, and on the other, differences that correspond
to the fact that some dimension which is specifically controlled in one language
is left to the unguided control of the speech apparatus itself in another. In the
first case, the grammars of the two languages in question differ in the way they
specify the phonetic realization of the property in question. In the second case,
the grammars differ in that one provides specific instructions for this realization,
while the other leaves it unspecified, under the control of the speech apparatus.
Presystematically, here is the sort of distinction we wish to make: particular
phonetic phenomena may be represented in the grammars of some languages
but not others. Consider the fact that in general, vowels following syllable-initial
voiced obstruents begin at a somewhat lower pitch than their target value, and
rise to that value over a relatively short period of time. Following voiceless ob-
struents, vowels begin on a slightly elevated pitch and fall to their target values.2
The basis of this effect might be the following: in order to overcome the fact
that the aerodynamic conditions for vocal fold vibration are less than optimal
when an obstruent constriction exists in the supra-laryngeal vocal tract, a com-
pensatory internal adjustment of vocal fold elasticity or stiffness is a common
accompaniment of voiced obstruents. Correspondingly, voiceless consonants
are often accompanied by a stiffening that ensures the lack of vibration during
their closure period. We do not need to understand the precise basis of the effect,
however, to be sure that there is one: the testimony of the world’s languages is
quite adequate on that score.
It is not hard to show that the vocal fold stiffness adjustments accompanying
consonants are logically and phonologically independent of the dimensions
of specification of vowel pitch. Nonetheless, such gestures associated with
consonants have consequences for the rate of vibration, when, at some point
shortly after the release of the occlusion, it occurs. This is a matter of the physics
and physiology of the structures brought into play in speech production. On the
other hand, we might well see the actual time course of the pitch contour between
the consonant release and the following vowel as characteristic of the baseline
control regime of the vocal organs as this applies in speech. The stiffness or
2 See Lehiste 1970 for a review of the classic literature establishing these facts.
slackness of the folds disappears with some particular latency and elasticity,
and it seems reasonable to attribute the particular contour of pitch that results
in a language like, say, English to general mechanisms of speech motor control,
not to the grammar of English.
In general, we find essentially the same pitch contours in post-consonantal
vowels across many languages. In some languages, however, such as Yoruba
and Thai, where tone is linguistically significant, we find that the pitch contours
of post-obstruent vowels are much sharper. We might suggest that in these cases,
the language specifies an independent compensatory gesture that has the effect
of bringing the vowel to its intended (and significant) pitch value more rapidly
than in English. Compare the contours in figure 6.1. In English, as represented
by the graphs in figure 6.1a, we see that there is still a difference in pitch in
the vowel following a voiced consonant (e.g., [b]) as opposed to a voiceless
one ([p]) more than 100 millseconds after the stop release. In Yoruba, on the
other hand, there are three contrastive tonal levels. We see in figure 6.1b that
a vowel following voiced vs. voiceless consonants displays the same (high,
mid, or low) pitch value within roughly 50 milliseconds after the stop release
(slightly longer in the case of a high tone, but still sooner than in the non-tonal
language English). Hombert (1976, p. 44) argues “that there is a tendency in tone
languages (which does not exist in non-tonal languages) to actively minimize
the intrinsic effect of prevocalic consonants.”
Fo Fo
k
140Hz
180Hz
k k k k k
k g
g
p
130
k
k
150 k
b g
g
k
g
k
g
120
g
120
time time
20 100 msec
20 100 200 msec
a. English b. Yoruba
Figure 6.1 Post-obstruent pitch contours in tonal vs. non-tonal languages
(from Hombert 1976)
We suggest that Yoruba phonetic representations thus should specify the

presence of this property, which we can distinguish from matters of general
physics and physiology (including the fact that the vowel pitch deviates at all
from its target initially) as well as from matters of general speech motor control
(such as the time course of pitch change that occurs, other things being equal).
What matters is that in Yoruba, other things are not in fact equal. Phonetic
representations in this language have to indicate a language-specific compen-
satory gesture that is part of its system, part of what the language organ of
Yoruba speakers determines. Phonetic representations in English, on the other
hand, ought not to say anything about the post-consonantal pitch contours: the
contours that we can observe physically are a matter simply of the vocal organs,
not of the language organ.
Vowel length effects In the case of English post-consonantal pitch

contours, we can see how we want the answer to turn out, but in the general
case there are many questions whose answers are not self-evident about what
does and does not belong in a phonetic representation. A first problem, noted
in Fowler 1990, results from the fact that not all of what is systematically
carried out in speaking is due to linguistic intention. We might hope that any
pattern which occurs in some languages but not in others must surely be under
linguistic control, and thus suitable for inclusion in our phonetic representation.
This is, in fact, a very standard form of argument in the linguistic phonetic
literature.
Fowler (1990) points out, though, that this apparently obvious diagnostic
may fail us. She notes that one of the most often cited phonetic regularities in
the literature is the fact that, in most languages, vowels are shortened by ca. 20–
30 milliseconds before voiceless obstruents as opposed to the duration they show
in other environments. In a few languages (cf. Anderson 1981 and references
cited there) this effect is suppressed, and vowels display the same durations
regardless of the nature of a following consonant. In English, on the other hand,
the difference in length between vowels before voiced and voiceless consonants
is considerably exaggerated, and (in contrast to the case in most others) well
above the threshold of perceptibility.
In languages like Saudi Arabic where the expected contextual vowel length
difference is apparently suppressed, it is necessary to indicate this as a fact
about the linguistically governed intentions of speakers. This is so despite the
fact that the result is no difference at all in vowel duration regardless of context.
And, of course, it is also necessary to indicate in a representation of English
utterances that vowels are considerably longer in voiced environments than
would otherwise be expected. But what about the languages that simply show
the expected small differences? Fowler argues that in fact these differences arise
as a mechanical consequence of the different dynamics of closure gestures in
voiceless as opposed to voiced obstruents: the talker “intends” the same thing
with respect to the vowels in both cases, but the implementation of that intention
leads to an unintended consequence as a result of the coordination of the vowel
and consonant gestures. In this case, then, the phonetic representations of vowels
ought to be uniform, despite the fact that acoustically the vowel portions of
utterances differ quite systematically by 20–30 milliseconds, depending on
the environment. Paradoxically, vowels of equal duration may thus come to be
specified differently (the Saudi Arabic case) while ones of different duration will
not differ in their phonetic representation where the differences are not intended
per se. On the other hand, where the durational differences considerably exceed
what can be attributed to the mechanical effects (as in the English case), we
must again indicate the presence of some specific phonetic intention responsible
for this.
Laryngeal specifications in Korean The differences we observe

across languages in the duration of vowels suggest the possibility that some
of the observable properties of phonetic elements may result not from our ac-
tive articulatory intentions, governed by the language organ as properties of
a particular language, but rather as consequences of the necessities associated
with other articulations. This notion has been discussed in the phonetic literature
by Patricia Keating (1988) as a matter of unders p ec i fi cati on.
Keating begins from the premise that parameters of a segment that are pre-
dictable should be unspecified in a phonological representation indicating con-
trasts, i.e. that phonological representations, at least, are underspecified in the
sense that not all segments have specified values for all possible properties.
This is thus a version of the “incompletely specified” view of phonological
representations discussed earlier in section 4.2.1. She goes further, however, in
suggesting that under some circumstances such underspecification may persist
into the phonetics. It may be the case, that is, that the values assumed along some
dimension during some period of an utterance result not from any linguistically
governed property of that part of the utterance, but simply from interpolation,
as a result of the exigencies of getting from where you were to where you have
to be. The spirit of such phonetic underspecification is entirely consistent with
the view being developed here.
For example, Korean (cf. Kagaya 1971) contains three series of stops:
“voiceless aspirated” stops, followed by a puff of breath as in for the [ph ] in
English pit, as opposed to “unaspirated” ones like the [p] of spit; and “voiceless
glottalized” ones, associated with a simultaneous constriction of the vocal folds.
Two of these, the voiceless aspirated and the voiceless glottalized, clearly have
characteristic properties: in the one case, a spread glottis gesture (which is what
produces the puff of breath) and in the other, a constriction of the glottis. The
remaining series, however, is what interests us.
These sounds are described as voiceless unaspirated initially, but voiced in-
tervocalically. Measurement of the time course of opening of the glottis (Kagaya
1971) during their production suggests that the voicelessness in initial position
is actually a gradual transition from the open glottis position associated with
the neutral position for speech in that environment to the approximated position
required for the following vowel; and that intervocalic voicing represents merely
the maintenance of the position appropriate to the consonant’s surroundings.
The “unaspirated” stops, then, have no value of their own for the position of the
larynx: they are phonetically underspecified for this property.
Vowel quality differences Identifying real instances of phonetic un-

derspecification may in fact be quite difficult, since the mere fact of apparent
predictability does not suffice to distinguish our linguistic intentions from the
baseline behavior of the systems employed in speech. Sometimes properties that
may seem thoroughly predictable nonetheless look as if they have linguistically
enforced target values. Boyce, Krakow, and Bell-Berti (1991) argue that for
some speakers, at least, there is apparently an articulatory target for the prop-
erty of lip rounding, under the control of the language organ even for segments
like [t] where this would seem to be particularly irrelevant and thus most likely
to follow from interpolation.
Fowler (1990) suggests that in some cases, what appears to be a gestural
target may in fact follow from the assumption that the speech apparatus tends
to return to the neutral position for speech insofar as it is otherwise unoccupied,
and some apparent deviations from straightforward interpolation may result
from this effect rather than from an active articulatory target. Sorting out these
hypotheses is not easy. Nonetheless, even though many apparently predictable
properties may turn out to be actively determined after all, it seems eminently
plausible that a number of examples will remain in which some articulatory
phenomenon results not from phonetically controlled gestural intentions but
rather as consequences of interpolation between other, surrounding gestures.
An interesting case in which we can ask how much of what is articulated
is actually a part of the (I-language) grammar concerns the vowel systems of
languages with very few contrasting segments.
Anderson (1978) takes up a proposal originally due to Aert Kuipers (1960)
concerning the Circassian languages, particularly Kabardian. This language has
a full range of vowel types (including [i,e,a,o,u], etc.), and a listener with no
knowledge of the language would notice nothing unusual about its vowels. On
closer inspection, however, it turns out that in Kabardian, the actual quality of a
vowel in any given position in a word is almost completely predictable from its
surroundings. That is, both [i]s and [u]s occur in Kabardian words, for example,
but [i] only occurs where adjacent to certain consonants, and [u] only adjacent
to some others. The two are in com p lem entary d i stri buti on in that
there are no environments in which we might potentially find either sound, with
the difference between them serving as the basis of a contrast between words.
Furthermore, the consonants surrounding [i] vs. [u] are not arbitrary: those
favoring [u] are ones whose articulation is similar to that of an [u] in involving
lip rounding and a relatively retracted tongue position, while those favoring [i]
involve no rounding, but a relatively high, front tongue body position. Similar
observations can be made for the environments in which the other phonetic
vowels are found.
These facts suggest that in Kabardian, the wide phonetic variety of vowels
corresponds to little or no difference in linguistic intention: essentially, the
vowels consist of a vocalic transition from the position of a preceding consonant
to that of a following one. In any given environment, there are in general only
two possibly distinct qualities that can occur, which we can differentiate by
saying that the transition is made with or without an accompanying deflection
of the tongue body downward. The presence or absence of this deflection is
thus the only way in which the articulation of one vowel differs from that of
another per se.
This kind of analysis was examined more deeply by John Choi (1992). Choi
explored a variety of possibilities quite closely, and examined articulatory and
acoustic data on languages of this type. One such language which he analyzed
in detail is Marshallese. Choi concluded that the front–back dimension in the
realization of Marshallese vowels does not correspond to anything that is under
independent linguistic control: no gesture of fronting or backing intervenes
in the relatively smooth transition from preceding to following tongue body
position as determined by the surrounding consonants, although tongue height
is independently controlled. This state of affairs is thus intermediate between the
almost completely underspecified vowels of Kabardian and a fully determined
system.
When is a “schwa” not a [ə]? Even in less “exotic” languages, we

may find support for the conclusion that certain vowels are phonetically un-
derspecified despite the fact that they are pronounced in a particular way. In
particular, reduced vowels (e.g., those of unstressed syllables in English, where
there are many fewer possible vowel qualities than in stressed syllables) may
have some surprises in store for us as far as their phonetic content is con-
cerned. Browman and Goldstein (1992c) examined the production of schwa
(the vowel in the first syllable of tomato) in English, with the expectation that
this would prove to have no distinct articulatory target associated with it. In fact,
the acoustic character of the vowels we transcribe as [ə] varies considerably as a
function of the vowels in surrounding syllables (and to some extent, of adjacent
consonants). English schwa is generally the result of a principle of reduction
which neutralizes other differences in vowel quality under given conditions,
especially absence of stress: compare the final syllables of relate, compare,

precede, where there is a stress, with the corresponding syllables in relative,
comparable, precedence, where that stress does not appear. Since the thing that
seems to distinguish schwa from the other vowels of English is precisely its
lack of potentially contrastive properties, it might be logical to expect it to have
few or no positive properties of its own.
Nonetheless, Browman and Goldstein found evidence that schwas produced
by English speakers in certain environments do indeed have a target associated
with them: [ə] in /pVpəpVp/ (e.g., [pipəpip]) has an articulatory target in the
middle of the English vowel space (Browman and Goldstein 1992c), a sort of av-
erage of all of the other vowels, but they observe a distinct gesture deflecting the
tongue body toward this position in the transition from a preceding vowel’s ar-
ticulation to that of a following one. The [ə] target frequently has little or
no direct acoustic consequence (beyond what we would expect of a smooth,
targetless transition), but it is apparently present nonetheless in the speaker’s
intention.
Interestingly, though, a later study (Browman and Goldstein 1992b) showed
that while schwas in syllables within lexical items display this kind of articula-
tory target, schwas that appear epenthetically before the productive inflectional
endings of the language may well be targetless after all. Thus, in “If Nita’d even
k n o w n . . .” the underlined schwa shows a distinct lowering of the tongue
dorsum, but in “It’s needed even now . . .” there is little or no such lowering.
This suggests that the “schwas” in such endings, though often acoustically
indistinguishable from other schwas, nonetheless correspond not to the
speaker’s intention to produce a specific vowel, but rather to a dynamic ef-
fect in articulation whereby two similar gestures are sequenced in time so as to
separate them, resulting in a brief interval whose only relevant property is the
duration of the gestural offset. Nita’d thus contains two real, phonetic vowels,
while needed contains only one, plus an additional vocalic interval that arises
as a consequence of the need to separate the two ds.
In all of these cases, we have portions of the speech event that bear properties
which we could perfectly well measure, and record, and which may correspond
closely to properties that are clearly linguistic in other cases; but for which
a case can be made that in these instances, they correspond to nothing spe-
cific in the speaker’s intention. There is thus a clear difference in the extent
to which the same E-language facts correspond to something present in the
I-language of a given speaker, depending on the language. In a principled phon-
etic representation of the sort we are interested in here, then, we ought to
omit them in the case where they are not in fact governed by the language
organs of speakers of the relevant languages; but it does not appear that we
can always determine that fact directly from examination of the physical facts
alone.
6.2.2 Intended effects hidden from observation

We must, furthermore, cope with the fact that not all of what we intend in speech
is actually realized physically (articulatorily and/or acoustically). Lindblom’s
“H&H” theory (Lindblom 1990), for example, suggests that the phonetic char-
acterization of phonetic elements might be in terms of a “hyper-articulated”
form whose actual implementation within a larger context may be (indeed,
generally is) incompletely achieved by virtue of other properties of the context.
This notion is further supported by Johnson, Flemming, and Wright (1993),
who provide evidence that listeners identify vowels in terms of extreme, hyper-
articulated targets rather than by their actual values. When asked to select the
simple vowel that best corresponds to a vowel in a word in running speech, that
is, they choose a “pure” quality rather than one closer to the actual physical
signal they have just heard. Here we have a variant of the traditional notion of
“undershooting the articulatory target,” the point being that there is good reason
to believe that the talker’s actual phonetic intention, as well as the listener’s
interpretation of the event, is characterized by an ideal target and not by its
undershot (or “hypo-articulated”) realization.
The most direct sort of argument for this general conclusion is provided by
facts of casual speech cited by Browman and Goldstein (1990b). They observed
that in relatively casual pronunciations of sequences such as must be, ground
pressure, “phonetically” [msbi], [graumprεš], there is apparent “loss” of
the segments [t] and [d]. X-ray microbeam studies, however, reveal that an
alveolar stop closure is actually made in each of these cases. It has no acoustic
consequences, though, since it is (more or less) completely concealed by its
overlap with the labial closure of the following [b] or [p]. The presence of the
stop in the speaker’s I-language representation is confirmed by the articulatory
data, even though this has no reflex in the actual signal, and thus would not be
present in a traditional (E-language) phonetic representation.
In many forms of English, final stop consonants are lost in certain environ-
ments in fluent speech: thus, mist may be pronounced as [ms]. It is natural to
ask whether the same sort of gestural “hiding” is responsible for this. The phe-
nomenon of variable final stop deletion in English is clearly very complex and
related to a number of conditions.3 In some instances, such as the loss of final
stops in morphologically simple (“monomorphemic”) forms before pause (e.g.,
the pronunciation of mist just cited), it is clear that gestural reduction rather than
hiding must be at work, since there is no other gesture in the environment which
could be said to overlap with the final [t] so as to obscure it. We must conclude
that for such speakers, the reduction of the [t] in this case must be due to an
I-language principle of phonetic realization.
3 Guy (1991a, 1991b) provides a summary of what is known about this variation, and Kiparsky
(1993) suggests an account in terms of OT of the range of systems that have been found.
Potentially, there is further empirical evidence to be found for Browman

and Goldstein’s proposal: the gestural overlap account suggests that varying
degrees of overlap ought to be found in a wide enough sample of speech, while
the traditional (deletion) story is only up to describing variable probabilities
for a categorical process deleting (or retaining) the stop. These considerations
should not obscure the point of the example for our purposes, however: in at
least some instances a gesture is actively intended, and indeed accomplished,
despite the fact that other conditions interact with it so as to deprive it of any
audible effect.
Such a gesture is obviously part of our intention, and surely appears in
the instructions to the vocal apparatus (i.e., in PF, the I-language phonetic
representation), despite the fact that it is quite undetectable in the acoustic event.
Obviously, then, such a phonetic representation is not at all a neutral observa-
tion language for speech, simply reflecting physically measurable properties of
the sounds produced. Instead it is a rather abstract and theoretical articulation
of our understanding of what is orderly and coherent about speech events from
the point of view of the dimensions controlled by the system of language, the
language organ.
6.2.3 Discreteness of phonetic dimensions

Let us return to the issue of whether phonetic transcription is in fact a coherent
notion that ought to play a role in linguistic theory. A recent argument that it is
not is due to Pierrehumbert (1990), whose objections are based on the fact that
traditional phonetic transcriptions are discrete and categorial (e.g. symbols of
the In t e r n at i on al Phonetic A lp habet, or IPA), but there are several
ways in which this fails to reflect linguistic reality:
a. effacement of segments in reduction (and epenthesis, such as the oral stop in
nasal+fricative clusters) is a matter of “less and less” (or “more and more”),
not “something vs. nothing”;
b. language-specific variation may involve continuous detail, so there is no
(finite) list of possible sounds;
c. gradient (but systematic) phenomena such as intonation and laryngeal
lenition (Pierrehumbert and Talkin 1992) require attention to hierarchical
organization, not just fine transcription;
d. phonetic details, as opposed to (categorical) phonological contrasts, are out
of awareness, and thus not “cognitive.”
There is some cogency to these complaints about phonetic representations
of the sort phoneticians teach beginning students to produce, but their force
is to show that this needs to be revised, not abandoned. We adopt here the
gestural perspective of Articulatory Phonology (Browman and Goldstein 1990a,
1990b, 1992a, and elsewhere), on which the properties we specify correspond
to dynamic gestures of the vocal organs, aspects of what a speaker intends to do

in speaking (as opposed to frameworks based on describing static articulatory
configurations or targets). Recall from chapter 4 that the representation of speech
as a sequence of discrete, static segments is after all a theoretical idealization,
warranted essentially by the insight it can bring to the study of sound in language.
To the extent that that picture actually interferes with a coherent view of the
relation between I-language structure and the implementation of speech, it needs
to be replaced with a view that is more appropriate to its subject matter.
The gestural view provides us with a language for describing some of
these phenomena, and for responding to the substance of the objections raised
by Pierrehumbert. Browman and Goldstein’s model is not the only one that
might serve for this purpose: the “C/D” model proposed by Fujimura (1994),
for instance, has somewhat similar goals but interestingly different dynamic
properties. This diversity suggests that there are quite genuine empirical issues
involved in the determination of the properties of an appropriate phonetic rep-
resentation, but that these issues are independent of the logical basis of phonetic
representation as a characterization of the talker’s intentions. This is surely a
desirable result (if not a very surprising one).
In responding to Pierrehumbert’s points, important advantages derive from
the fact that the gestural representations of Browman and Goldstein do not
have the exclusively categorial and segmentalized character of traditional tran-
scriptions. Instead, gestures have magnitudes and temporal extent which are
(at least in part) subject to specification and manipulation by the principles of
individual linguistic systems. The representational vocabulary of I-language,
on this account, thus includes some aspects of speech dynamics, as is apparently
required, and not simply the description of a sequence of static configurations.
Furthermore, the overall organizing principles of gestural synchronization and
sequencing, and the description of the time course involved in the transition
between distinct gestures, are also (at least in principle) parameterizable as a
property of particular linguistic systems, thus determined as a property of in-
dividual language organs. This sort of specification of the overall system of
gestural realization in a given language can be understood as something like a
formalization (in part) of the traditional impressionistic notion of the “base of
articulation” of the language (including statements like “French is spoken in
the front of the mouth,” etc.).
6.2.4 Timing effects

On the basis of what we have seen, let us assume that the language organ deals
with phonetic specifications in terms of intended gestures, rather than (as in
much traditional terminology) semi-snapshots of ideal articulatory configura-
tions or an acoustically based parameterization of speech events. An important
component of gestural representations is the fact that they bring a non-discrete

notion of time into consideration, in ways that traditional phonetic representa-
tions based on sequences of individual independent phonetic segments do not.
Once non-discrete phenomena based on the detailed time course of speech are
taken into account, the problems of isolating those dimensions of control that
are under the control of linguistic intention (that is, that are part of I-language)
and of identifying the range of specifiable variation along those dimensions
becomes even more difficult. It must be stressed that phonologists – the people
most concerned about language-particular characterization – have no serious
descriptive language for low-level temporal dynamics, let alone an articulated
theory of the independent variables in this domain.
In fact, phonologists have generally had little to say about the details of speech
timing. Phoneticians know a lot about the time course of speech events, but
phonologists have until quite recently worked hard to develop models in which
there is literally no place for the details of time course: phonological elements
are points in time that can succeed one another but that do not have duration,
magnitude, elasticity, and so on. A widespread (but completely unsubstantiated)
assumption among phonologists is that it is precisely in this domain that we
can attribute everything we observe to the (language-independent) mechanics
of speech articulation.
But of course the increased information content of non-discrete gestural
representations potentially comes at a high price: it provides us with too much
freedom in its raw form. In order to isolate the linguistically significant degrees
of freedom in the domain of the time course of speech, we must separate the
dimensions of rhythm and timing that are under the control of the language
organ from the background dynamic properties of the articulatory system.
It seems clear that there are some important language-particular components
to the determination of speech rhythm and timing: that is, there are properties
of this sort that differentiate languages from one another, as anyone can testify
who has vaguely heard a conversation across a crowded room and recognized
immediately that the speakers involved must be, say, Dutch, though without
actually picking up a single word. See Beckman, Edwards, and Fletcher (1992)
for some discussion. And if these factors have a language-particular component,
they must be reflected in phonetic representations of the sort we seek. But until
we have phonological models that take time course seriously, we have quite
literally no principled way to talk about what might be language-particular
about speech dynamics. Such models are in fact beginning to emerge,4 but the
problem has not, perhaps, received the attention it deserves from phonologists.
4 See Byrd and Saltzman 1998 and Byrd et al. 2000 for one approach to a phonological model of
speech dynamics.
6.3 Speech microprosody: a research program 127
And of course, until we know what aspects of speech organization are subject
to language-particular determination, we do not know what we need to have a
principled theory of in order to describe this aspect of the language organ.
Introduction of an appropriate specification of rhythmic and temporal effects
into the description of the aspects of speech that fall under linguistic control
is thus an important refinement of the range of problems to which phonolo-
gists should attend. We have assumed above that the form this might take is
simply a specification of appropriate aspects of the time course of particular
gestures in speech, but there is another aspect to the problem. Not only can the
implementation of particular gestures over time fall under linguistic control, but
independent of this, their synchronization and relative organization can provide
yet another dimension of possible variability. Browman and Goldstein and their
colleagues (Browman and Goldstein 1998, Browman et al. 1998) develop this
point in suggesting that otherwise comparable gestures in English and Tashlhiyt
Berber, for example, differ in their “bonding” or relative synchronization
more than in their individual component timings. These ways of describing
timing relations need further refinement and application to a variety of language
types.
6.3 Speech microprosody: a research program

A greater understanding of the ways in which languages can differ in their over-
all patterns of speech rhythm and timing requires much more study of speech
microprosody: the low-level dynamics of pitch, intensity, and a variety of mea-
sures of the temporal control of articulation within and below the level of
syllables. The problem in this investigation is to hold physical and phonetic
factors (roughly) constant, while varying control of the prosodic system
(in a rather general sense of “prosody”) within an individual language.
One possibility which can be exploited is that of comparing ordinary native
speakers of (e.g.) English with speakers who have a good command of the
language, but a “foreign accent.” A plausible hypothesis is that prosody is
the hardest thing to learn, partly because it cannot be taught (since we do not
know how to articulate our knowledge of it) and partly because in general
we are not even aware of it. As we will note below in chapter 9, prosody
is apparently controlled by the language organ at a very early stage of the
language acquisition process – even before the appearance of the first words
in the child’s production. A similar point is argued at length by Locke 1993.
Although this is surely not all of the story, one large component of a “foreign
accent” is likely to be prosodic anomaly. On the other hand, these speakers are
still perfectly normal with respect to the basic phonetic mechanisms underlying
articulation; and so we can assume that ways in which they deviate from the
patterns evidenced by ordinary native speakers will reveal differences in the

cognitive system controlling the dynamics of articulation, and thus aspects of
the dimensionality of that system.
But there are several other populations of interest where we can hope to
find anomalies vis à vis normal prosody. One of these is a class of clinically
brain damaged patients who display what is called Acquired Foreign Accent
Disorder.5 These are people who (sometimes only for a relatively short period)
during recovery from cerebral insult (strokes and other sources of brain damage)
are perceived to have acquired an accent – typically that of a language to which
they have never been exposed. A notable example from the literature (Monrad-
Kröhn 1947) is a Norwegian woman during the Second World War who was
perceived to have a German accent following head trauma (sustained, ironi-
cally, in a German bombing attack), and who was subsequently ostracized in
her village on this account, when in fact she had never had any exposure to
spoken German. Some rather similar symptoms can occasionally be induced
by electrical stimulation of areas of the cortex in epileptic patients undergoing
exploratory investigation of anatomical brain function prior to surgery intended
to relieve intractable seizures.
Examination of the phonetic details of the speech of such patients suggests
that in general, whatever they are doing, it bears little or no specific resemblance
to the speech of genuine foreign accented (normal) speakers. Furthermore,
attempts to characterize what these patients are doing wrong in terms of seg-
mental articulations seem to fail (or at least to be quite insufficient). Their sim-
ilarity to foreign accented normal speakers lies in the fact that while they get
(most of) the segmental phonology right, their prosody seems to be disordered.
Our hypothesis here is that the cognitive deficit underlying “acquired” for-
eign accents is – or at least often involves – impaired control of rhythm and
timing, the prosodic organization of the native language. When prosody sounds
wrong, but the speech is still basically segmentally correct and intelligible,
other speakers perceive this as a “foreign accent.” This is thus a circumstance
in which the language organ’s control of speech prosody is disrupted, and com-
parison of these patients with normal speakers provides another potential basis
for exploring the properties of the system itself.
Studying both natural and acquired foreign accents in comparison to the
prosody of normal speech gives us two somewhat different windows on the
question of what aspects of prosodic organization are part of the linguistic sys-
tem, such that control of them could be lost (or fail to have been correctly
5 Some references on this topic include Graff-Radford et al. 1986, Blumstein et al. 1987, Kurowski,
Blumstein, and Alexander 1996, and Carbary, Patterson, and Snyder 2000. The interpretation of
this disorder as one of a prosodic, rather than segmental, nature, as suggested below, has come to
be fairly widely accepted, though details vary considerably among both patients and researchers.
A particularly useful review with respect to this issue is provided by Berthier et al. 1991.
6.4 Conclusion 129
acquired). There is also another kind of population that would be of interest:

subjects in which a disorder exists that affects the system of articulation in gen-
eral, rather than the cognitive systems that provide the instructions for control
of articulation. The reason to look at this as well is to disentangle systematic-
ities that are, as it were, “post-PF” (i.e., implementational matters external to
the I-language organization of speech) from those that represent dimensions of
variation in a cognitively significant phonetic representation.
At least two such populations are potentially available for study, both in-
volving subcortical damage that results in speech disorders. In cerebellar lesion
patients with ataxic dysarthria, we find a perceived speech rhythm impairment,
without (in general) corresponding impairment to higher level language func-
tions. Somewhat similar symptoms are displayed by some early AIDS patients
in whom subcortical degenerative damage has occurred. Our hypothesis is that
the deficit in these patients is in the system responsible for general control and
organization of rhythmic speech movements (cf. Keller 1987). Theirs is a prob-
lem in the general phonetics of prosody, not the language-specific cognitive
system.
These populations can be studied with respect to standard acoustic measures
of timing, the time course of acoustic intensity and pitch, and also with articu-
latory measures (e.g., the time course of jaw movement, ultrasound studies of
tongue movement, and other articulatory variables) as possible, depending on
the conditions of access to individual subjects. The hope is that comparing these
talkers along these dimensions will give us insight into what aspects of normal
performance are attributable to what aspects of the overall speech and language
system. Such investigations are still in early stages of development 6 – there are
major problems of principle and experimental design yet to be resolved, and
few concrete results. The purpose of bringing up this work at all is simply to
propose a methodology for gathering the sort of information we will need in
order to arrive at an improved understanding of those dimensions of speech
rhythm and timing that are part of the system of I-language, and which thus
ought to be recognized as components of a linguistically significant phonetic
representation.
6.4 Conclusion
If we take the lessons of the sections above to heart, the representation we
arrive at for “PF” – the interface of the language organ with the mechanics
of articulation7 – is quite distant from the sort of fine phonetic transcription
6 See Boatman et al. 1994.

7 The same representation is also appropriate as the interface between the langauge system and
perception, if something like the Motor Theory referred to above is correct, as we presume.
using symbols of the IPA (or some other, similar system) which most current
linguists learned as students (and generally continue to teach). An accurate PF
representation:
a. fails to indicate some properties that are perfectly true, measurable, and sys-
tematic about utterances, insofar as these are the consequences, not specifi-
cally intended, of other aspects of speech;
b. indicates gestures that are intended even under circumstances where those
gestures do not have their intended auditory (or perhaps even articulatory)
effect;
c. specifies gestures with associated temporal dynamics, and not just timeless
points, and which are related in hierarchical ways;
d. indicates gradient temporal relations rather than mere succession;
e. indicates gradient gestural magnitudes and not only categorical presence vs.
absence; and
f. indicates the overall pattern of articulatory dynamics within which the ges-
tural intentions of the talker are realized.
A representation with these characteristics obviously cannot be taken to be the
kind of physical observation language phonetics is often presumed to provide.
It is rather a description of the cognitively real (though largely unconscious)
representations that underlie speech motor control and that are in some sense
recovered in perception. Despite its non-traditional character, it does seem to
serve the larger goal of characterizing what is linguistically significant about
the facts of speech. It is neither a full physical record of speech events, nor
a restricted characterization of the minimal distinctive core that distinguishes
higher-level linguistic elements from one another. As such, it serves as the output
(or perhaps better, as the implicature) of the sound-structural regularities of a
language. We claim that these are the things that ought to be demanded of a
“phonetic representation” by those for whom such a notion finds its significance
in the theory of language and the mind, rather than in physics or physiology.
That is, it is an appropriate way of characterizing PF, the connection between
the language organ and the organs of speech and hearing.
7 Morphology
If you ask a naive person-in-the-street – the kind of person the British call “the
man on the Clapham omnibus” – what the central thing is that has to be learned in
order to “know” a language, the chances are that a major part of the answer will
be “the words of the language.” This notion that the words of a language are the
essence of its identity is reinforced by standard language courses, which devote
great attention to the systematic presentation of vocabulary. Indeed, much of
what passes for “grammar” in many language courses is actually a subpart of
the theory of words: what has to be learned about things like conjugation and
inflection is first and foremost how to form inflected words. Compared with the
effort usually devoted to drilling vocabulary and word formation, the amount of
attention devoted to exemplifying the uses of the various forms and providing
usage notes is usually quite limited, and the space given to fundamental matters
of syntactic structure virtually none at all.
So if the set of words is such an important property of, say, English, how do
we determine what that set is? A standard answer is provided by a dictionary
(though that, of course, simply puts the problem off by one step: how did the
dictionary makers know what to include?). Most speakers behave as if the
question “Is [such and such] a word of English?” has a determinate answer,
but if so, the dictionary probably does not provide that answer, at least in the
general case. For instance, overlook “disregard” is clearly a word of English,
and is recognized as such in dictionaries, but what about overlookable “subject
to being disregarded”? Surely this is also a word of English, though it will not be
found in any of the standard dictionaries, even the Oxford English Dictionary.
And this ignores the fact that new words enter the language all of the time: if a
word like frobnicate “to manipulate or adjust, to tweak” did not exist prior to
the second half of the twentieth century, how did that change? And why, as soon
as we recognize frobnicate, do we have no trouble at all accepting prefrobnicate
“to manipulate or adjust something prior to performing some other operation
on it”? Clearly a dictionary of a language reflects the language’s words, but
equally clearly no dictionary can be taken as a definition of what constitutes a
word of the language.
131
132 7 Morphology
Linguists generally think of the lexicon of a language as a component of

its grammar separate from its syntax and its sound system, the component that
provides an account of the words whose organization into phrases, clauses, etc. is
given by the syntax and whose pronunciation is specified by the phonology and
the phonetics. Beyond this, however, there tends to be rather a lot of vagueness
about just what sort of system the “lexicon” is.
Our argument in this chapter is that the standard construal of the lexicon as
primarily an inventory derives from the conception of language as E-language.
The deficiencies and lacunae in that view suggest the need to replace it with an
I-language idea, that of the lexicon not as a list but as a form of knowledge – a
component of the language organ.
7.1 The lexicon

Mark Aronoff (1988) points out that there are a number of distinct senses that
can be (and have been) assigned to the notions “lexicon” and “lexical,” including
on the one hand that of the locus of idiosyncrasy in language, and on the other
that of a collection of all of the items that belong to “open” or “major” word
classes (typically nouns, verbs, and adjectives).
The first of these conceptions is usually associated with Leonard Bloomfield,
who said that
A complete description of a language will list every form whose function is not de-
termined either by structure or by a marker [a “function word,” or closed-class item
characteristic of some grammatical structure]; it will include, accordingly, a lexicon, or
list of morphemes, which indicates the form-class of each morpheme, as well as lists of
all complex forms whose function is in any way irregular. (Bloomfield 1933, p. 269)
The “morphemes” of a language are supposed to be the smallest discrete pieces

into which words can be broken that constitute irreducibly arbitrary associations
of form and meaning. Thus, overlook consists of two morphemes over and look
because each of these parts has a more or less distinguishable meaning, and
the meaning of the word as a whole is built on those partial meanings, but
there is no separable meaning to smaller parts (e.g., o, ver, loo, etc.). Larger
constructions containing several morphemes appear in Bloomfield’s lexicon to
the extent their properties are not entirely those that would be projected from
those of their components together with the rules of the grammar.
If this notion is taken literally, the lexicon in this sense
is incredibly boring by its very nature. It contains objects of no single specifiable type
(words, VPs, morphemes, perhaps intonation patterns, and so on), and those objects that
it does contain are there because they fail to conform to interesting laws. The lexicon
is like a prison – it contains only the lawless, and the only thing its inmates have in
common is lawlessness. (Di Sciullo and Williams 1987, p. 3)
7.1 The lexicon 133
In contrast to this notion, Aronoff shows that the notion of the lexicon as
an inventory of “members of a major lexical category” is presumed in much
writing on generative grammar. Evidently these two conceptions are not coex-
tensive: on the one hand, some idiosyncratic items (including both members
of the set of grammatical items like determiners, pre- or post-positions, etc.;
and idiomatic phrasal constructions) are not members of open word classes;
and on the other, many words that are members of the classes noun, verb, and
adjective (or whatever the correct set might be) will, in many languages, be
completely compositional formations composed of more than one morpheme,
such as overlookability. Ignoring these differences, the a priori interest of the
lexicon defined in either way is not self-evident.
What these two ways of construing the notion of “lexicon” have in common
is that they are both kinds of list. Perhaps by analogy with dictionaries in the
real world, it seems often to be taken for granted that the lexicon is a kind of
set or database.
Now of course no one would take seriously the notion that the “syntax” of a
language is a list of its sentences (whether the unpredictable ones or all those
of some given type), or that the “phonology” is a list of sound combinations.
Both of these aspects of grammar are generally construed as kinds of knowledge
speakers have about their language: in the one case, knowledge of the patterns by
which words can be organized into larger structures, and in the other, knowledge
of how sound units of various sorts combine and the modifications they undergo
within such larger combinations. These are all components of our language
organ, which develop in particular ways on the basis of our exposure to data
characteristic of a given language. It seems productive to interpret the “lexicon”
of a language in a similar sense as, roughly, “the knowledge a speaker has of
how words can instantiate positions in a syntactic structure.”
When construed in this I-language fashion, the lexicon is not just a list of
items. Of course, much of what we know about the words of our language
does have the character of individual and rather local stipulations, like “cat is
a noun pronounced /kæt/ and meaning ‘a carnivorous mammal (Felis catus)
long domesticated and kept by man as a pet or for catching rats and mice.’”
But in addition, a speaker’s lexical capacity must also include a system of
rules or principles by which words are related to one another, insofar as these
relations are (at least partially) systematic and thus part of what we know
about words, qua speakers of a particular language. Such regularities are also
implicated in describing the formation of new words not hitherto part of a
speaker’s explicit knowledge of the language but implicit in its principles of
word formation. Even when a language has a rich set of general principles for
forming new words, it is typically the case that the question of whether or not
a given possible word is also an actual one remains a lexical issue – not, say, a
syntactic one.
134 7 Morphology
It is not immediately clear that a notion of “lexicon of language L” in the

E-language sense of “the set of words of L” is conceptually well formed, or that it
would be particularly interesting even if it were. Given potential sound/meaning
correspondences, speakers can assign interpretations to these, such as “a word
of my language that I have heard before,” “a word that I can’t recall having
heard, but which probably means [such and such],” “a word I’ve heard before,
and that has something to do with a part of a boat, but I don’t really know what
part,” “a word I don’t know, that sounds like it has something to do with [such
and such], but I can’t tell exactly what,” “surely not a word of my language,” etc.
These are all aspects of the set of principles a speaker controls, and it is the study
of those principles that ought to be primary in the study of lexical knowledge.
It seems quite unlikely that any particular list of specific forms, however
annotated, can be formulated so as to express the same information. The fact
that much of our knowledge of the lexicon is rather specific to individual words
may seem to give it such a list-like character overall, but this is probably illusory.
Much the same problems arise in studying the lexicon from the point of view
of its extension (i.e., as a list of words) that arise in studying other aspects of
language extensionally rather than as systems of knowledge.
The knowledge that speakers have of a language’s lexicon probably applies
over a superset of both notions discussed by Aronoff. Idiosyncratic things that
speakers know about particular words are certainly “lexical,” as are the proper-
ties of idiomatic units larger than the word. Similarly, the lexicon comprehends
our knowledge of the words that can fill major lexical categories, whether these
are related in completely (or only partially) compositional ways to “simpler”
words or are themselves irreducible lexical primes. But while both of these
kinds of knowledge can be presented in the form of lists, no particular list can
be taken to exhaust the content of the knowledge that is the lexicon. For instance,
speakers can certainly apply their knowledge to provide analyses of parts of un-
familiar potential words, analyses that may be insufficient to assign a structure
to the entire word ( just as, in syntax, speakers may be able to assign structural
descriptions to only some parts of a sentence that is not fully grammatical).
Central to our knowledge of language in this area, as in others, is the notion
of rule-governed regularities. In the discussion below, we will speak of these as
Word Formation Rules, but we should bear in mind the lesson of chapter 5 that
there may be quite different ways to express such things. Only further investi-
gation can show whether, for instance, a system of hierarchically organized
constraints might not provide a more satisfactory account of this component of
the language organ.
7.2 Words and “morphemes”

We would like, then, to replace “externalist” points of view that construe the
lexicon of a language as a collection of individual lexical items with a conception
of the lexicon as a particular form of knowledge of the language – knowledge

about its words and their relations with one another, part of a person’s grammar
and properties of the language organ. Let us start from the nature of words:
we can agree that what we know about a particular word in and of itself can
be represented as an association among several distinct sorts of information.
A lexical item, then, is an association of sound (phonological form) with
syntax (morphosyntactic properties) and meaning (semantics).
This conception of words as associations is, of course, a version of the Saus-
surean notion of words as signs: the minimal units in language that connect
form (sound, or in the case of signed languages, gesture) with content (meaning).
A minimal sign, like “cat”={/kæt/, noun, cat}, displays an irreducible con-
nection among these properties. As far as the syntax is concerned, these can be
taken to be the atoms of linguistic structure.
Now if all words were like cat there would be very little left to do but list them,
but it quickly becomes apparent that this is not the case. And apart from the
question of whether all words are signs in just this sense, we can also ask the
converse: are all signs words? That is, do parts of words – “morphemes” in
traditional terms – have the logical structure of signs?
This question comes up because of the fact that not all lexical items are
equally arbitrary: some are partially motivated (de Saussure 1974) in that
they are related to other lexical items. We saw that above, in considering the
word overlook: some proper subparts of the meaning of this word are associated
with proper subparts of its form (over, look) in a way that would contradict the
notion of signs as irreducible if signs and words were the same thing. The
relation of overlook to over and to look is not just a connection among
three items that happen (independently) to be words of English: there are
things shared among these words of a sort that do not apply to other sets (e.g., cat,
cake, hat) that show some overlap in form.
How are we to explicate this notion of partial motivation, and in general the
idea that some words are related to one another? The traditional notion of the
morpheme is driven in large part by this question. Within structural linguistics,
the morpheme was proposed as the “real” locus of the irreducible sign relation.
Some words of course consist of only a single morpheme on this view (cat, look),
while others are composed of more than one (overlook, which contains the same
morpheme as look plus the same morpheme as that occurring alone in the word
over). In some languages, the number of such pieces within a word can get
really very large. Some examples from languages that are particularly noted for
this are given in (7.1).1
1 “Eskimo” is a name to be avoided, since (like the names of many other groups around the world) it
was assigned to the people involved by speakers of another language, and is considered offensive.
In this case, the word “Eskimo” was apparently supplied by speakers of an Algonquian language
who served as guides to early French and English speaking explorers. It means roughly “people
who eat really disgusting stuff.”
136 7 Morphology
(7.1) West Greenlandic (“Eskimo”)

tusaanngitsuusaartuaannarsiinnaanngivipputit
you simply cannot pretend not to be hearing all the time
Siberian Yupik
avelghaghtesnaanghisaghqaqsaghanka . . .
I am supposed to prevent them from missing it (school) in
the future, but . . .
German
Lebensversicherungsgesellschaftsangestellter
life insurance company employee
Some morphemes (traditionally called “free”) can appear alone as words,
while others (traditionally called “bound,” such as pre- in prefrobnicate) are
always part of larger constructions, where their status as morpheme units is
based on the fact that they make (roughly) the same contribution to form and to
meaning in each of the more complex words in which they appear. In these terms,
we could define the notion of Morphological Relatedness as follows:
words are related insofar as they share a common morpheme (or morphemes).
A word W1 is derived from another word W2 if (and only if) W1 = W2 + µ for
some morpheme µ (disregarding order).
This leads to a formulation like the following of the relation between words
that share morphological material:
   
(7.2)   
 /əbl/ 
 
 /inflejtəbl/ 

 /inflejt/     
Adj Adj
[V NP ] + ⇒
  
 able to be  
 able to be 
i n f l a te    
[V ]ed [i nflat]ed
“inflate” + “-able” ⇒ “inflatable”
Here we have a complex word (inflatable) representing a collection of properties
including sound (the phonological representation, /inflejtəbl/), the morphosyn-
tactic property of being an adjective, and (what we frankly admit is little more
than a stand-in for) meaning.2 The diagram indicates that the properties of this
complex word can be regarded as composed from corresponding properties of
its component parts (inflate and -able) in a straightforwardly additive fashion.
If this picture is viewed seriously as a claim about how words are constructed
in a speaker’s mental grammar, by his language organ, and not simply as a
2 We have essentially nothing to say in this book about semantics, the study of meaning, but that
certainly does mean we think this is unimportant, or not related to the nature and structure of the
language organ. This omission simply reflects the limitations of our own research competence.
The work of Ray Jackendoff, for instance (including Jackendoff 1983, 1990 among others)
sheds important light on the way the language organ associates linguistic form with meaning.
Representations in small capitals are intended to refer to elements of meaning: while these are
not quite as arbitrary as they may appear, we will not attempt to defend claims about them here.
kind of metaphor, it has some rather strong consequences. Some of these are
the following. (a) Since the relation between a (complex) word and another
from which it is derived consists exactly in the addition of another minimal
sound–meaning complex, sign composition must always be strictly monotonic
(additive, concatenative). (b) Since the basis of the sign is the indissoluble
unity of sound and meaning (and (morpho)syntax), there should be a one-
to-one association between form elements and content elements. That is, every
chunk of form ought to correspond to exactly one chunk of meaning, and vice
versa. Exceptions to this are unsystematic instances of accidental homophony
or synonymy. (c) Derivational relations are directional, in the sense that (all of
the) properties of the base form are presupposed by a derived form that involves
an additional marker. We will examine in turn each of these implicit empirical
claims of the morpheme-based theory of lexical structure, and conclude that
none of them are consistent with the facts of natural language in general. This
will lead us to the conclusion that the lexical component of the language organ
should be viewed as a system of rule-like relations among words, rather than
just an inventory of minimal signs (the morphemes).
It should be stressed that the notion that words are exhaustively composed
of morphemes, and that the theory of word structure is essentially a kind of
“syntax of morphemes,” has a long history in linguistics, and a good deal of
initial appeal. Nonetheless, we suggest that this is essentially an E-language
notion that ought to be replaced by a somewhat different conception if the
structure of the language organ is to be properly understood.3
7.2.1 Monotonicity in word formation

The first consequence of the morpheme-based view that we will consider is the
claim of monotonicity: the idea that a derived word consists of its base with
some additional material added. If complex words are just simpler stems with
the addition of (one or more) sound–meaning complexes, we should find that
derived words always contain their bases as a proper subset, in terms of form,
meaning, and morphosyntax. We will see, however, that the relations we find
in natural languages are actually more complicated than that.
Problems of morpheme form As far as word shape is concerned, it

has been clear at least since such work as Hockett 1947 that not all morpho-
logy is monotonic (i.e., some morphology is “non-concatenative”). Following
the analytic tradition initiated by Hockett’s paper, the consequences of this
observation are frequently obscured by giving descriptive names to the kinds
3 Again, the work of Jackendoff, especially Jackendoff 1975, furnishes important precedents for
the ideas developed here.
138 7 Morphology
of case that pose problems, providing the illusion that these have somehow
been solved. In fact, though, they present a solid core set of cases in which a
morphological relation between two forms is not signalled by the addition of
new material, as the morphemic view would require.
One large group of examples of this sort can be collected as instances
of apophony, or relation by replacement. English pairs such as sell/sale,
sing/song, blood/bleed, food/feed, etc., as well as the familiar examples of
man/men, woman/women, mouse/mice, etc., indicate relations such as those
between a verb and a corresponding noun or between singular and plural by
changing the main stressed vowel, not by adding an affix of some sort. In some
languages, such as the Semitic languages Arabic and Hebrew, replacive oper-
ations of this sort (e.g., Modern Hebrew semel “symbol,” simel “symbolize”;
xašav “think,” xišev “calculate,” etc.) can be treated as the association of a basic
set of consonants with one word pattern or another (see McCarthy 1981), but
this analysis has no appeal for English examples like those above.
Other apophonic relations involve consonants, rather than vowels, and
are sometimes called instances of mutation (as opposed to “Ablaut” or
“Umlaut” where vowel changes are concerned). Consider English pairs
such as believe/belief, prove/proof, speak/speech, bath/bathe, breath/breathe,
glass/glaze “provide with glass.” Again, the relations involved are signalled (in
the modern language, at least) not by the addition of some marker, but rather
by a change from one consonant to another.
Some languages (though not English) go so far as to indicate a class of derived
words not by adding material to a base, but rather by subtraction. An interesting
class of nouns is derived from verbs in Icelandic by deleting the final -a that
marks the infinitive: thus, hamr [hamr] “hammering” from hamra [hamra] “to
hammer”; pukr [pü:kr] “concealment” from pukra [pü:kra] “make a secret of.”
One might think that the infinitive is here derived from the (shorter) noun form
by adding the ending -a, but we can see that that is not the case from the vowel
length in forms like pukr [pü:kr]. The fact that this vowel is long appears to
contradict the rules for vowel length in Icelandic; but it makes sense in terms
of the form of the untruncated infinitive ([pü:kra]). Forms derived from this
infinitive preserve its vowel length, but this explanation requires us to assume
that pukr is derived from pukra – by deletion of the final -a.4
Problems of morpheme content The cases above, and similar types as

discussed in Anderson 1992, make it clear that morphological complexity is
not always represented in an additive way as far as word form is concerned.
As already noted, such examples have been well known for some time, and
4 See Orešnik and Pétursson 1977, Kiparsky 1984 for details and discussion.
a number of efforts have been made to accommodate them within theories of

word structure. What is rather less familiar, however, is the fact that the same
conclusion holds for the semantic and morphosyntactic dimensions of word
formation. Here too it is not hard to see that word formation cannot in general
be monotonic, as required by the morpheme model. That is, the syntax and/or
the semantics of a complex form may not be just an additive function of its
pieces. Actually, we can see something like that even in the case of inflatable:
the syntactic properties of this word show little or no reflex of those of the
underlying verb. We might suggest that they are somehow still there, but are
just covered up by the over-riding syntax of being an adjective. Other examples
do not seem to be as simple, though.
Subtractive semantics. An example of semantically non-monotonic morphol-

ogy is provided by a category of verbs in Icelandic traditionally known as
“middle voice” forms (Anderson 1990). These are built by adding a suffix -st to
a basic verb, and as far as their form is concerned, they are quite unexceptional.
The suffix itself represents a fossilized form of what was originally (that is,
earlier in the history of the language) a reflexive pronoun attached as an enclitic
to the verb, but that connection has now been lost. There is no reason to believe
that in contemporary Icelandic the -st is in any syntactic sense a pronoun rather
than simply a suffix appearing in a large number of verbs. One might think this
is a perfectly ordinary “morpheme,” but the problem is that the relation be-
tween the basic and derived verbs is subtractive: typically, one of the arguments
of the basic (transitive) verb, together with the semantic relation associated
with it, is simply absent from the derived verb. Consider the pairs of verbs
in (7.3).
(7.3) gleðjast “rejoice” gleðja “gladden (trans)”

kveljast “suffer” kvelja “torture (trans)”
lýjast “get tired, worn out” lýja “tire (someone) out”
hefjast “begin (intrans)” hefja “begin (trans)”
opnast “open (intrans)” opna “open’ (trans)”
finnast “exist, be to be found” finna “find”
heyrast “be audible” heyra “hear”
týnast “be, get lost” týna “lose”
ágoggast “(of a fish) be hooked” gogga “catch with a hook”
The morphology here is clearly additive in form. Could we think of it as

additive in content too? In logical terms, we might think of the semantics as
adding an operator that binds the Agent argument in semantic representation,
along the lines of (7.4).
140 7 Morphology
 
(7.4)  /kvelja/   

 
  /st/ 
[V [NP ]]
+ V

 (Caus e x, 
  
  ∃x (Cause x, P)
(Be c om e tortured, y))
“torture” + “-st”
 

 /kveljast/ 

 
Adj
⇒

 ∃x (Caus e x, 

 
(B ecom e tortured, y))
“suffer”
In (7.4), the “meaning” of the element -st is that where the meaning of the
base includes the notion that some participant causes some state of affairs,
the causer is some indefinite agent that need not be explicitly mentioned. Thus,
when the subject of the verb kveljast “suffers,” what that means is that this
individual comes to be tortured by some person or persons unknown.
But saying this makes an incorrect prediction: it predicts that in the inter-
pretation of the derived (-st) verb, there is an implied generic or other agent
responsible for the patient’s state, and this is simply false. Verbs of this sort in
Icelandic do not necessarily have any such interpretation. When an Icelander
suffers, s/he doesn’t need to feel there is someone or something to blame it on.
Compare this with the rather similar English passive form in (7.5a).
(7.5) a. During their captivity, Kim and Sandy were tortured (by
thoughts of home and family).
b. When called on in class, Terry always looks tortured (by
insecurity).
When no explicit agent is given (for instance, in a by phrase), a sentence like
(7.5a) still carries the implication that there was some cause for the suffering,
even if only mental; and this cause can be made explicit. Even in the adjectival
form (7.5b), the source of the suffering, though not as strongly implied as in the
verbal form, can be made explicit. In Icelandic, however, there is no possibility
of adding an af phrase (the equivalent of an English by phrase, used with true
passives to indicate an agent), and there is no necessary implication that the
suffering has a determinate cause. Any such source can only be mentioned in
the form of an adverbial adjunct phrase.
(7.6) a. *Jón kveljast af tannpı́nu
John is-tortured by toothache
John is tortured by toothache
b. Jón kveljast (ı́ tannpı́nu)
John suffers from toothache
John suffers (from toothache)
English passives thus have a place for reference to a causer – a place that
may be represented explicitly (with a by phrase), or implicitly, through the in-
troduction of a non-specific or indefinite semantic operator. Icelandic -st verbs,
in contrast, do not involve such reference (even though ordinary passive forms
in Icelandic do, like English). The right semantic analysis of the relation be-
tween the basic and the -st verbs in (7.3) must therefore involve not the binding
(by an abstract or impersonal operator) of an agentive argument, but rather the
omission of the entire layer of semantic structure associated with agentivity.
Semantically, the verbs that serve as bases for this formation have the schematic
form in (7.7).
(7.7) (cause x, (become (P y)))
That is, “(subject) causes (object) to become tired, miserable, started, open,
etc.” The addition of the ending -st has the effect of deleting the highest predi-
cate (cause x,y) from this structure. This means the corresponding argument
disappears from the syntax, so that a basic transitive verb becomes intransitive;
and also that the role played by this argument is no longer present in the se-
mantics of the derived verb. In modern Icelandic, “suffering” is not a matter of
being tortured by someone/something, even though such a meaning may well
provide the etymology of the word.
What does this mean? Simply that there is no way to describe the verbs in
(7.3) as being related by the addition of a “morpheme” -st, because there is no
way to characterize this morpheme’s contribution to meaning as the addition
of something to the meaning of the base verb. The relation between the two
columns of (7.3) is relatively straightforward: speakers know that when a given
verb involves causation of this sort, a related verb ending with -st may describe
the same resulting state, but without reference to its causation. This I-language
characterization, however, has no clear correspondent in terms of the E-language
conception of words as built by combining morphemes.
This example is actually typical of a large class of cases. In many languages,
we find some morphological element that converts transitive verbs to intran-
sitives, often with no remaining trace (in the form of “understood” material)
of the missing argument(s). The subtractive nature of this operation is usually
concealed by grammar writers, who give the element in question a superficially
additive interpretation “detrans” or the like. But we must not lose sight of the
fact that the actual content of “detrans” may be precisely the suppression of
some part of the base verb’s content – an operation just like that of phonologi-
cal subtraction (truncation), when Icelandic pukra “make a secret of ” becomes
pukr “concealment.”
Subtractive morphosyntax. We turn next to a case that also involves

“subtractive” operations as the concomitant of a derivational relation, but where
142 7 Morphology
the subtractive operation in question alters the syntactic information in the lex-
ical items affected.
English has a large array of nominals formed with the ending -er, many of
which are derived from related verbs. An interesting point about these concerns
the extent to which the nominal does or does not appear associated with the same
arguments as the basic verb. Hovav and Levin 1992 distinguish two classes of
-er nominals: some (typically with an agentive interpretation) inherit the argu-
ment structure of the base verb, while others (typically with an instrumental
interpretation) do not. Consider, for example, two senses of the noun wrapper
in relation to the syntax of the base verb wrap, as illustrated in (7.8). Here the
difference in prepositions reflects a difference in argument structures, a vari-
ation that is associated with the difference between agentive and instrumental
readings of the nominal.
(7.8) a. The best job for Fred would take advantage of his experi-
ence as a wrapper {of/*for} presents in fancy gold paper at
Tiffany’s.
b. The best use I can think of for The New York Times is as a
wrapper {for/*of} fish that didn’t keep well overnight.
Note that, in association with nominals, of marks complement arguments (as
in (7.8a) – see similar examples in chapter 3), while various other prepositions
(like for in (7.8b)) mark adjuncts.
Hovav and Levin draw an analogy with two types of derived nominals, dis-
tinguished by Grimshaw (1990) among others:
(7.9) a. The examination (of/*for the graduating seniors) lasted three
hours.
b. The examination/exam (for/*of prospective students) was
three pages long.
“Event”-derived nominals such as (7.9a) refer to events, and they can take
complements, as the possibility of (7.9a) with of shows. Non-event-derived
nominals such as (7.9b) refer to objects, results of actions, etc. rather than
events, and these do not take complements, as the impossibility of (7.9b) with
of rather than (the adjunct-marking) for shows.
Hovav and Levin propose to unify the differences found in (7.8) and (7.9)
in the following way. They suggest that the syntax of basic verbs like wrap,
examine includes a reference to an event (of wrapping, examining).5 Both in
the -er cases (7.8b) and the others (7.9b), the derived word involves a change
in the “event” position, as a correlate of the derivational relations illustrated
5 We do not attempt to justify this analysis here, but refer the reader to Hovav and Levin’s work
for the motivation and details.
by (7.8b) and (7.9b). The relations illustrated by (7.8a) and (7.9a), in contrast,
involve the same formal marker, but quite different semantics; and here the argu-
ment structure of the basic verb, including its “event” argument, are preserved
in the derived noun.
If this account is correct in its essence, it provides an example of “subtrac-
tive” morphosyntax in association with “additive” phonological marking: the
formation of instrumental and non-event-derived nominals involves the dele-
tion of syntactic argument structure with respect to the properties of the base
verb. Again, this makes sense if we think of the derivation of one lexical item
from another as based on a relation which is part of the I-language that
develops as the language organ of speakers of (in this case) English. It is much
harder to understand if we attempt to represent this part of the language by an
E-language-based inventory of morphemes.
Derivation involving semantic or syntactic change. In some cases of appar-

ent subtraction, it is at least logically possible that the relation is actually mono-
tonic (additive), and we have simply gotten the direction of derivation wrong:
perhaps the shorter form is actually the base, with the longer one derived from it
(by addition) rather than vice versa. We now consider more complex examples,
where the derived form is changed, rather than being added to or truncated. In
these cases, playing with the question of which form is the base is not really an
option, since the relation in question cannot be construed as additive in either
direction.
Derivation with morphosyntactic (and/or semantic) Abwechslung (rather than
monotonic addition of material) is illustrated by a quite productive relation in
the Australian language Warlpiri (Nash 1986, Laughren 1988). In this language,
transitive verbs can be associated with either of two case-marking patterns on
their subject and object arguments, with a difference in meaning.
Warlpiri is an ergative language. This refers to a pattern of case marking
on the arguments of verbs which differs systematically from that found in, say,
German or Latin, where subjects appear in the nominative (nom) case and direct
objects (usually) in the accusative (acc). In an ergative language, the subject6 of
a transitive verb is marked with a special case, called the ergative (erg),
while the subject of an intransitive verb appears in a different form, called the
absolutive (abs). In the basic clause pattern of Warlpiri, as in most ergative
languages, the (direct) object is marked with the absolutive case (the same form
that appears on the subjects of intransitive verbs).
6 There is an extensive literature on the structure of “ergative” languages, and a major point of
contention is the nature of grammatical relations within basic clauses in such a language. Since
the resolution of those issues does not concern us here, our use of “subject” and “object” in
referring to Warlpiri can be be taken as presystematically referring to the arguments that would
correspond to the subject and object of a literal translation of the sentence into English.
144 7 Morphology
Most transitive sentences, however, have an alternative form in which the

object is in the dative (dat) case. When the object takes this form, there is also
a special interpretation: the object is interpreted as less completely or directly
affected by the action than in a corresponding sentence with an absolutive
object. This replacement of erg+abs marking by erg+dat is thus associated with
a semantic distinction which is also found in many other languages, though the
way in which it is formally indicated may be somewhat different. For instance,
in English many transitive verbs can appear either with a direct object, or
with an object marked with a preposition, as in the sentence pairs in (7.10).
The second sentence of each pair implies a less comprehensive effect on the
object.
(7.10) a. i. Every night, I read War and Peace to my wife.
ii. Every night, I read from War and Peace to my wife.
b. i. I can’t believe my own cousin shot me!
ii. I can’t believe my own cousin shot at me!
Without attempting to give a comprehensive account of these differences,7
for present purposes we need only attend to the fact that in Warlpiri, two distinct
morphosyntactic patterns are associated with two semantically distinct inter-
pretations of sentences involving what appears to be the same basic verb. In
the pairs of Warlpiri sentences below, we have an alternation between the two
case frames (erg+abs vs. erg+dat) that is associated with a shift from “basic”
to “incompletive” or “partitive” semantics.
(7.11) a. i. Yarla-rna pangu-rnu ngajulu-rlu
yam(abs)-1sg dig-past 1sg-erg
I dug (up) yams
ii. Yarla-ku-rna-rla pangu-rnu ngajulu-rlu
yam-dat-1sg-3dat dig-past 1sg-erg
I dug for yams
b. i. Walya-rna pangu-rnu ngajulu-rlu
earth(abs)-1sg dig-past 1sg-erg
I dug the earth
ii. Walya-ku-rna-rla-jinta pangu-rnu ngajulu-rlu
earth-dat-1sg-3dat-dd dig-past 1sg-erg
I dug at the earth
Although there is no relevant overt affix present on the verb in (either alternant
of ) these forms, they are nonetheless quite typical of the sort of relation we see
in many languages between semantically similar verbs, appearing with two
distinct case-marking patterns, as in the English examples we saw above in
7 See Anderson 1988 among much other literature for some discussion.
(7.10). The English expressions that gloss the examples in (7.11) – dig yams,
dig the ground vs. dig for yams, dig at the ground, etc. – illustrate the same kind
of contrast.
It happens that the (quite systematic, familiar and productive) relation be-
tween these two patterns is not marked inWarlpiri by an overt verbal affix, but
in various other languages (including one to which we will turn shortly, West
Greenlandic) entirely comparable relations are overtly marked. We assume that
the relation between these two patterns in Warlpiri, as in languages where there
is an overt affix, is appropriately established in the lexicon, since there are two
distinct morphosyntactic behaviors correlated with a difference in meaning.
There have been a number of proposals in the literature as to how to express
the difference between the members of such pairs. Mary Laughren (1988) makes
an interesting suggestion about the Warlpiri case. She suggests that the affected
verbs have a semantic interpretation involving more than one layer of structure,
and that the morphosyntactic relation is correlated with a reorganization of
those layers.
(7.12) a. i. [V [erg][abs]], “I got yam s by I di g”
ii. [V [erg][dat]], “I dug in order to I get yams”
b. i. [V [erg][abs]], “I broke up e arth by I di g”
ii. [V [erg][double-dat]], “I dug in order to I break
up earth”
If Laughren’s account is more or less on target, we have an example of what
we might call “semantic metathesis”8 as the concomitant of a lexical relation.
That is, the relation between two lexical variants of the same verb involves no
phonological change, but only the morphosyntactic replacement of one case
frame by another, and the reorganization of subordination relations within the
logical conceptual structure that represents the verb’s semantics.
A somewhat different take on a very similar construction is provided in an
interesting paper by Maria Bittner (1987). In West Greenlandic (an “Eskimo”
language), the “object-demoting” or anti-passive construction is overtly
marked – there are at least five different suffixes, in fact, that have this ef-
fect. Bittner argues, contrary to previous accounts, that these affixes are in fact
independent. But what interests us is the change in the morphosyntax of case
marking that accompanies each of the affixes.
We illustrate first with some simple cases, to show the form of the alternation.
The first member of each pair in (7.13) is formally transitive, while the second
is intransitive and “anti-passive” (AP) with a complement in the instrumental
case (inst).
8 metathesis is a change whereby two elements switch position, without otherwise changing their
form. An example is provided by English dialects in which the verb ask is pronounced [æks].
146 7 Morphology
(7.13) a. i. Jaakup ujarak tiguaa.

Jacob-erg stone-abs he-took-it
Jacob took a/the stone.
ii. Jaaku ujaqqamik tigusivuq.
Jacob-abs stone-Inst he-took-AP
Jacob took a/the stone.
b. i. Jaakup illu taanna sanavaa.
Jacob-erg house-abs this-abs he-builds-it
Jacob built/was/is building this house (and may have
finished it).
ii. Jaaku illumik taassuminnga sanavug.
Jacob-abs house-inst this-inst he-builds-AP
Jacob was/is building this house (and hasn’t finished).
More interesting differences show up when the sentence contains various

logical operators. When that happens, Bittner suggests that a semantic difference
appears which corresponds to the scope in a conceptual representation of some
logical variable.
(7.14) a. i. Atuartut ilaat ikiurtaraiqarpara.

of-students one-abs I-must-help-him/her
I must help one of the students (namely, Nanuq).
≡ ∃x[x is one of the students and I must help x]
ii. Atuartut ilaannik ikiuisariaqarpunga.
of-students one-inst I-must-help-AP
I must help one of the students (but it doesn’t matter which one).
≡ It is necessary that (∃x[x is one of the students and I help x])
b. i. Jaakup siumukkurmiuq ajugaassasuraa.
Jacob-erg member-of-“Siumut”-abs he-believes-he-will-win
Jacob believes a member of “Siumut” will win.
≡ ∃x[x is a member of “Siumut” and Jacob believes (x will win)]
ii. Jaaku siumukkurmiumik ajugaassasurinnippuq.
Jacob-abs member-of-“Siumut”-inst he-believes-will-win-AP
Jacob believes a member of “Siumut” will win.
≡ Jacob believes (∃x[x is a member of “Siumut” and x will win])
The difference between members of the pairs in (7.14) is comparable to a

well-known ambiguity that appears in “intentional” contexts, as illustrated by
the two senses of John wants to marry a Norwegian: either (a) there is some
particular Norwegian, namely Sigrid, that John wants to marry; or else (b) his
interest is motivated by his desire to get a Norwegian passport, so he will settle
for any marriageable Norwegian who might be available. Similar phenomena
have been pointed out in a number of languages, including Finnish, Basque,

Polish, and others. A Finnish example is provided in (7.15).
(7.15) (Finnish)
i. Matkustajat ovat jo laivassa.
passengers-nom are already on-ship
The passengers are already on the ship.
≡ ∃x[x a group of passengers and already (x is on the ship)]
ii. Laivassa on jo matkustajia.
on-ship is already passengers-Part
There are already passengers on the ship.
≡ already (∃x[x a group of passengers and x is on the ship])
To accommodate these examples, Bittner (1987, p. 230, paraphrased) pro-
poses what she calls the “Scope Generalization”: where the basic casemarking
pattern of a language involves interpretations with certain logical operators tak-
ing wide scope, an alternative case-marking pattern may be associated with
interpretations where those operators may (or must) take narrow scope.
Of course, in order to make the descriptions of Warlpiri, West Greenlandic,
etc. suggested above into fully formal accounts, we would need to have a much
more explicit theory of logical conceptual structure. But assuming the gen-
eral plausibility of Laughren’s and Bittner’s analyses, it seems that the lexical
relation between basic and anti-passivized verbs in Warlpiri and West Green-
landic involves an operation of rearrangement on these structures, rather than
the simple addition of new material.
The reader who has struggled this far through the exotica of the last few pages
may well be wondering what these data have to tell us about the nature of the
language organ. The answer is that, from the variety of considerations adduced
above, we conclude that derivational relations are not in general monotonic in
character. They may involve subtractive or replacive operations on the mor-
phosyntactic and/or semantic content of an item, as well as on its phonology.
The importance of that fact is that it shows us that our knowledge of words
cannot be reduced to a list or inventory of words or morphemes: rather, the
language organ must deal in more general relations among words that are not
reducible simply to the presence vs. the absence of meaningful elements. This
is perfectly natural on the I-language view of our subject matter, of course, but
it also reinforces that picture.
7.2.2 The solidarity of form and content

We turn now to the question of whether recurring elements of word form are
consistently and uniformly associated with recurring elements of word content,
and vice versa. That is, do word classes defined by form correspond one-to-one
148 7 Morphology
with those defined by meaning (or function)? 9 When we look seriously at

this issue, we see that a single formal marker typically goes with a particular
derivational content, and vice versa, but this is not true in general, so we must
separate the two kinds of relation. This point has been made in great detail by
Robert Beard (1995); see also work by Arnold Zwicky and Richard Janda, and
much earlier, by Franz Boas.
The categories of word form and word content may fail to be isomorphic
in both directions. On the one hand, we often find a single function which
is represented by several forms. Consider the formation of English “action
nominals” (destruction, settlement, arrival, laughter, etc.) as an example. Most
verbs have such a nominal, and only one: if there is more than one, there is
semantic specialization, as in the difference between recital and recitation. The
class of action nominals is functionally unitary, but the forms are quite diverse.
However, the fact that these multiple potential forms are mutually exclusive
shows the significance and underlying unity of the class.
Another example is furnished by the binyan system of Semitic languages such
as Arabic and Hebrew. There is a limited set of fixed word structural patterns
(made up of syllable structure together with associated vowels) in such a lan-
guage, and the same lexical material (represented by a set of – typically three –
consonants) can be associated with various of these. Each pattern has a typical
semantic correlate, but in some cases very similar semantics can be associ-
ated with more than one pattern. An example is furnished by the two Arabic10
causative forms ʕallam “teach” and ʔaʕlam “inform.” Both are causatives of
the basic verb ʕalim “know,” and thus provide an example in which the same
function can be associated with more than one formal characteristic.
On the other hand, a single form can also be used to represent several distinct
functions. For instance, in the formation of nominals in English, the ending -er is
associated with agentives (trainer, writer) and instruments (wringer, mixer) de-
rived from verbs, as well as other forms derived from nouns meaning “person
connected with (N),” such as roomer, second-grader, honeymooner, carpet-
bagger, footballer, Londoner, Pittsburgher; nouns referring to non-humans,
somewhat parallel, such as ten pounder, freighter, tanker, Nor’easter, etc.;
nouns from numbers, like fiver, forty-niner, (seventy-)sixer, etc. The formally
unitary class of nouns in -er is thus functionally quite diverse.
9 The assumption that they do underpins what Bazell (1952), in discussing American Structura-
list theory of the 1950s, called the “Yale morpheme,” one of his canonical instances of the
“Correspondence Fallacy” (the assumption that structure in one domain will be isomorphic to
structure in another). The reference here is to the fact that linguists at Yale such as Bernard
Bloch, Rulon Wells, and others were particularly prominent in the development of a picture of
morphological structure closely analogous to the then-prevalent view of phonology based on
biunique phonemes, as sketched in chapter 4.
10 In the transcription of Arabic forms here, [ʕ] is a fricative consonant produced with a constriction
in the pharyngeal region, while [ʔ] is the somewhat more familiar glottal stop.
Much the same can be said, again, for the formal patterns of Semitic mor-
phology. Arabic verbs in the C1 aC2 C2 aC3 pattern, for example, include both
intensives (e.g. qattal “massacre”; cf. qatal “kill”), and causatives (e.g. ʕallam
“teach”; cf. ʕalim “know”). While the pattern presents a single category of
word form, that category is associated with quite different content in different
instances.
These facts suggest that within the lexicon of a language, we need to recognize
a collection of formal classes (like -er nouns), and a collection of content-based
classes (like “derived nominals”), independently of one another. Typically a
given form-based class will be largely coextensive with a particular content-
based class (or at least with a small number of these), and vice versa; but since
the two have independent bases, there is no need for the structure of the two
domains to map onto one another in any simple way.
Let us take a look at one such content-based class, to see how it is related
to formal structure. English contains a great many “agent nouns” derived from
verbs, such as [N [V bake]-r], [N [V preach]-er], etc. Some other agent nouns
appear to end in -er, but not an -er which has been added to a more basic verb:
rather, if there is a related verb, it is homophonous with the noun, and should
probably be regarded as derived from it, as in the cases of [V [N butcher]], [V
[N huckster]], and (perhaps) [V [N minister]]. In yet other cases, the noun ends
(phonetically, at least) in the same way as other agent nouns, but there is no
related verb at all: [N carpenter], [N soldier], [N janitor], [N bursar]. We also have
agent nouns that apparently display the same ending, but where the related word
that serves as the base is another noun, rather than a verb: messenger, adulterer,
lawyer, astronomer, furrier, clothier, hatter, etc.
The sources of agent nouns in -er can thus be quite diverse, but that is not at
all the end of the story. Other nouns seem to be just as validly “agent nouns”
in terms of their content, but do not display -er at all: poet, musician, artist,
linguist. Many such nouns are related to verbs in the language, but where such
a non-er agent noun exists, the expected regular formation with -er added to
the verb is not well-formed: cf. cook (*cooker in the sense “one who cooks”:
cf. Kiparsky 1982), judge (*judger), student (*studier), representative (*repre-
senter), correspondent (*corresponder).
Notice that these regular forms are not blocked per se as possible words: in
fact, cooker exists, at least in British English, but as an instrumental (meaning
“oven”), rather than an agentive. The regularity is rather that once a verb is
associated with one action noun in the lexicon, we cannot create another syn-
onymous action noun for that verb by applying the regular process. This is quite
parallel to the fact that a single verb has a single associated action nominal, as
we observed above, though this may come from any one of a wide range of
formal types. The same form can be employed for two different functions – e.g.
wrapper, which can be either an agent or an instrument – but not vice versa.
150 7 Morphology
This complementarity establishes the significance of the content-based classes

within the lexicon of a language, and also shows us that the nature and content
of these classes cannot in general be equated with any particular formal marker.
Again we see that a speaker’s knowledge of the lexicon of a language cannot
be equated with a list of sign-like items.
7.2.3 Directionality in word formation

We usually think of derivation as a process that leads from base to derived
form, and thus has an inherent directionality. That is suggested naturally by
the notion of derivation as adding a morpheme to a base. But once we take
the classes that are involved in derivational relations seriously, we see that that
is not necessary: relations between lexically significant classes might have no
relevant directionality at all.
An example of this state of affairs is presented by the verbal system of
West Circassian (Dumézil 1975). In these languages,11 transitive verbs with
the canonical shape /C(ə)/ are commonly related to intransitives with the shape
/Ce/, where the consonants are the same in such semantically related pairs, and
only the vowel differs. In some cases (e.g., 7.16a,b), it seems plausible that the
transitive meaning is the more “basic” one, while in others (e.g., 7.16c,d) it
would seem to be that of the intransitive that has this status.
(7.16) a. se-tχe “I write (intrans)”; zə-tχəʎ se-tχə “I write a letter”
b. ṡ◦ əzə-r me-th.ač ’e “the woman washes, does the dishes
(intrans)”;
ṡ◦ əzə-m šəq◦ ə-r ye-th.ač ’ə “the woman washes (trans) the
dishes”
c. me-laž e “he works”; yə-χate ye-lež ə “he works his garden”
d. ma-ko ’e “he goes”; mə-γ ◦ eg◦ ə-r ye-k◦ ’ə “he goes this road”
In such a system, there is no independent motivation for thinking of either
the transitive or the intransitive form as “derived from” the other, though there
is a clear, systematic and productive relation between the two.
Another similar example is provided by the gender system of the African
language Maasai (Payne 1998). In this language, every noun belongs to one or
the other of two genders, masculine or feminine.12 Each gender has an associated
prefix. There are systematic connections between the two, but the relation is
such that where there exist pairs of words differing only in gender, it can go
11 In terms of the number of distinctions they show among consonants, the West Circassian lan-
guages are among the most elaborate in the world. The reader for whom the transcriptions
below seem like gibberish should not be concerned, since these details of pronunciation are not
material to the point at issue.
12 We ignore here a separate “gender” which applies only to one or two locative words.
either way, in the sense that either the masculine form or the feminine might be
taken as basic depending on the example.
The simplest cases are those in which the relation is perfectly symmetric;
and those in which only one gender is possible.
(7.17) a. εnk-apυtánı̀ “wife’s mother”; ɔl-apυtánı̀ “wife’s father”
b. ε-mɔ́dáı́ “female fool”; ɔl-módáı́ “male fool”
c. εnk-áı́ “God”; *ɔlk-áı́
d. *ε-mεná; ɔl-mεná “contempt”
In other cases, however, it appears that the feminine is basic, and the mascu-
line derived.
(7.18) a. en-kı́né “goat; female goat”; ol-kı́né “male goat”
b. εn-k έráı́ “child (either gender)”; ɔl-kέráı́ “large male child”
c. εnk-anáshὲ “sister”; ɔlk-anáshὲ “very large sister (pejorative)”
d. en-tı́t!o “girl”; ol-tı́t!o “large, shapeless hulk of a woman
(pejorative)”
e. en-kitók “woman”; ol-kitók “very respected man”
In a number of other cases, there is a relation of relatively small/relatively
large between the two forms, and in these instances the feminine seems sec-
ondary with respect to the masculine.
(7.19) a. εn-dóı́nyó “hill”; ol-dóı́nyó “mountain”
b. εnk-ál έm “knife”; ɔl-ál έm “sword”
c. εnk-aláshὲ “weak brother (pejorative)”; ɔl-aláshὲ “brother”
d. εnk-abáánı̀ “female or small doctor, quack”; ɔl-abáánı̀
“(male) doctor, healer”
e. εn-dεk έt “ineffectual curse”; ɔl-dεk έt “curse”
f. ε-l έε “man (pejorative)”; ɔ-l έε “man”
It seems that in this instance, we should recognize a basically symmetric
relation between the two genders:
   
(7.20)  feminine   masculine 
(female, relatively small) ⇔ (male, relatively large)
   
/e(n)-/ /o(l)-/
In some cases, basic items have the same status (but different sense) in either
class. In other cases, the lexical item has a “basic” gender, and a shift in either
direction may imply a pejorative value. There is still no sense in which one of
the genders is in general derived from the other, however.
The theme being developed here should be familiar by now. E-language-
based views have assumed that the lexicon of a language can be characterized
by a set of items (signs) whose properties of sound, meaning, and morphosyntax
152 7 Morphology
can be identified in the utterances of speakers. In contrast, we see, the “lexicon”

of a language is based on relations among items that are not reducible to the
mere presence or absence of individual meaningful signs. It is the systematicities
of these relations that the language organ must accommodate. Eventually, of
course, we must ask how much of this structure is due to properties of Universal
Grammar (UG) and how much variation exists among individual languages;
and how, where such variation exists, the properties of the language faculty
allow it to be derived from the data available to the child learning a specific
language. Before such questions can even be contemplated, however, we must
have an adequate view of what the basic content of lexical knowledge must
consist in.
7.3 Productivity
Let us summarize what seems to be true about our knowledge of words from
what we have seen thus far. First, it is quite impossible to maintain that that
knowledge takes the form of a list of full words we have learned individually.
In fact, we could not possibly have just memorized a list of all the complex
words – certainly not in languages like West Greenlandic or Siberian Yupik,
where much of the expressive power of the language is built into the system
that forms new words, so the creative aspect of language is not limited to the
syntax. Indeed, this is also the case for German or English.
A list would not represent our knowledge for several reasons. First, we can
clearly make new compounds (e.g., PlayStation, ThinkPad, earbud [part of
a hands-free cell phone]) and derived words (prefrobnicate, overlookable),
and at least the first time we use them, such forms would not be on the list.
Secondly, in languages with extensive inflectional paradigms (such as Finnish,
Hungarian, and many others), every lexical item may have thousands of inflected
forms, many of which a given speaker might never have encountered for a
particular word, but which can all be produced and recognized when required.
We recognize new inflected forms of words, and may not even know whether we
have ever seen them before. We can even provide such forms for nonsense words
(as in Jabberwocky). Most generally, such a list does not take advantage of or
express our knowledge of systematic regularities that may be more complicated
than what can be expressed merely as the addition of a “morpheme”; and there
are a great many such regular relations to take advantage of among the words
we know. All of this suggests that our knowledge of words is better represented
by something more like a system of rules than (just) a list. Of course, we have
to have a list of the irreducible bits – the fact that [kh æt] is a noun that means
“cat,” for instance – but we also have to have a system of rules or the like to
represent our knowledge that expresses – and indeed goes well beyond – the
ways in which these bits are combined.
7.3 Productivity 153
A standard, if incomplete, response to this is to treat word structure as analog-

ous to syntax. We could say that the list is a list of word structure components, or
morphemes, and that complex words are formed by combining and recombining
these elements. That approach allows us to express the possibility of forming
new words (from known elements), but it also makes a number of concrete
predictions about the kinds of relation we ought to find among words, and
as we have seen in the preceding sections, those predictions (monotonicity,
solidarity of form and content, directionality) are often not satisfied.
An alternative to the strictly morpheme-based theory is to treat our know-
ledge of words as knowledge of systematic classes of lexical items, together
with the systematic relations among them. For instance, we know that there is
a large class of adjectives in English ending in -able (sometimes spelled -ible),
each of which has a meaning involving roughly “able to be (something)ed”
(inflatable, navigable, potable, etc.). Many of these are related to other words,
verbs whose meaning is more or less the “something” in that formula, and
whose shape is similar (but not necessarily identical) to the rest of the -able
word (inflate, navigate – but not *pote). Word knowledge includes both know-
ledge of the regularities of form and meaning that characterize such classes and
knowledge of the relations among them.
Derivational relations can be expressed by rules that operate over the lexicon,
relating one class of lexical items to another: for instance, “given a transitive
verb, the same shape extended by /əbl/ corresponds to an adjective with the
meaning able to be verbed.” The relations involved (often) have the char-
acter that they can be used to support the positing (or creation) of new lexical
items: given an appropriate verb (e.g., frobnicate), we can construct prefrob-
nicate, and given that, we can further produce an adjective prefrobnicatable
“able to be manipulated or adjusted prior to the performance of some other
operation.” Each form follows from our knowledge of the relationships among
lexical classes.
On the other hand, some instances of derivational relations are only partial,
serving as a reconstruction at a finer granularity of Saussure’s notion of the
“partially motivated sign.” For instance, we recognize that potable, ineluctable,
eligible, possible, probable, credible, etc. are also members of the class of -able
words, because they have the right form and sense even though there is no more
“basic” word from which they could be derived by adding the suffix -able.
We have also seen that the important class of agent-nominals in English
includes not only words ending in -er, some of which (e.g., baker) are related to
simple verbs and some of which (e.g., butcher) are not, but also others that do
not have that shape (e.g., cook). The membership of these latter in the relevant
class is confirmed by the fact that their presence obviates the formation of
new words that might seem to be perfectly regular (e.g., the impossible *cooker
“one who cooks”).
154 7 Morphology
We can clearly see that our knowledge of lexical relations can be more or
less exhaustive. This is known in the literature on morphology as the issue of
“productivity”: the formation of new adjectives in -able from verbs is essentially
completely productive, but the relation between spit and spittle is essentially
limited to this pair (and perhaps prick/prickle) in modern English, since there
is no reason to assume any relation at all between such superficially similar
pairs as cod and coddle. Many other connections fall somewhere in between,
applying in a number of instances but not really providing a warrant for novel
creations.
At least a lot of what goes by the name of productivity in the study of
morphology is probably a reflection of the extent to which what we know
about lexical relations determines the properties of new instances. That is, the
properties of a class may sometimes be sufficient to allow us to predict all of
the characteristics of a potential new member, but in other instances the internal
diversity of the class might leave much underdetermined.
A case where our knowledge is clearly only partial, and where the relation
in question is only marginally productive, is that of English adjectives in a-:
e.g., ablaze, abroad, aground, afoot, afield, ajar, alight, asleep, askew, astride,
aspread, awash, aghast, etc. These are derived from verbs (e.g., ablaze), nouns
(e.g., afield), or other adjectives (abroad). Some do not correspond to more
“basic” stems from any class: ajar, aghast (but ghastly?). The class of these
adjectives displays a sort of family resemblance semantically, but no consistent
semantics: Marchand (1969) suggests that they mean “in a state or position
of . . . ,” but that is not interestingly distinct from what we could say about the
class of adjectives as a whole. And of course the conclusion we can draw is that
our inability to make up new adjectives of this class – the lack of productivity of
“a- adjectives” – follows from the fact that the existing relations do not provide
enough limitations to determine what such a word would mean: if we were to
coin the word awalk, we would not know what use to put it to.
Much derivational morphology is lexically isolated in this way. Aronoff
(1976) suggests that productivity might be reduced, when properly under-
stood, to the transparency of a derivational relation. Relations are more or less
productive depending on how completely they determine the properties of a
potential word as a function of those of an existing base, an account which
clearly rests on a conception of word structure as an elaborate and articu-
lated kind of knowledge, rather than just a set of basic elements together with
rules for combining them. Surely some account of the differences among word
formation patterns in terms of their productivity is necessary as part of a de-
scription of what a speaker knows about a language, and it seems inevitable
that any such account will need to be based on the I-language notion of word
structure relations, rather than on the E-language notion of an inventory of
signs.
7.4 Conclusions about lexical organization 155
7.4 Conclusions about lexical organization

A number of consequences follow from the discussion in this chapter for our
understanding of lexical relatedness. First, of course, we must avoid thinking
of the “lexicon” as a static list – either of full words or of morphemes. Second,
we must recognize that while similarities of form are quite typically associated
with similarities of meaning, the two are not as solidary as the picture of the
Saussurean sign (or the structuralist morpheme) might lead us to hope. The need
to analyze the two separately, while treating the connections between them as
another complex dimension of word structure, goes back at least to the proposal
of Franz Boas (1911) to distinguish “grammatical processes” from the “ideas
expressed by grammatical processes.” The “processes” Boas had in mind corre-
spond to formal characteristics, including both affixal and non-affixal markers;
the “ideas” correspond to the semantic content of morphological classes. The
two are not necessarily associated one-to-one, and a coherent analysis of word
form must not assume that they are.
A modernized version of Boas’ picture distinguishes “derivation” in the
abstract sense of relations among lexical subclasses, each characterized by
various substantive properties, from the formal expression of these relations.
Perhaps the most explicit form of this position is the “Separation Hypothesis” of
Robert Beard (1995), but a number of variations have appeared in the linguistic
literature of recent years. The central point of these is the need to recognize a
class of relations among existing lexical items, rather than just rules for building
new items (a notion introduced to the generative literature by Jackendoff 1975).
Such a move also requires us to give up on what Bazell called “Yale
morphemes” as minimal signs (see note 9 above). Instead of thinking of deriva-
tion as the addition of a morpheme, we should think of it as a bundle of phono-
logical, semantic, and morphosyntactic mappings among lexical classes. Often,
of course, these will be coherent in the sense that a given phonological effect
will be primarily associated with unique, specific semantic and morphosyntactic
effects. But not always.
When we look at the full range of relations established in word formation,
we get a very different picture of what “lexical relatedness” can consist in from
the one that has dominated the field for many years. The resulting picture is
quite different from what theories based on morpheme combination – at least
in any serious sense – would suggest. If we think of language, and especially
the lexicon, in E-language terms as an inventory or collection of signs that
can be combined, the morphemic view seems self-evident. But if we think of
the lexicon as a kind of knowledge, it seems equally obvious that it might be
partitioned in the various ways we have seen: knowledge of form, of meaning,
of grammatical properties, etc. – and that the connections among these might
be only partial. Thus, the shift from a morpheme-based view of derivation to a
156 7 Morphology
more complex relational one goes along with a shift in the nature of linguistics
from a focus on items and the properties of a collection of linguistic objects to
a focus on language as a cognitive system, and a kind of knowledge.
We are really only at the beginning of an understanding of how the language
organ accommodates and organizes knowledge of words and their relations
with one another, and of how principles of UG both limit and enable the kinds
of things that can occur in particular languages. We have not even begun to
address issues such as those raised in chapter 5, concerning whether a system
of rewriting rules is the most appropriate way to formulate a description of the
systematicities of I-language in this domain.
It seems self-evident that there is much more variation among languages in
their lexicons than elsewhere in grammar, but perhaps, once the contribution of
UG to lexical structure is properly understood, that will turn out to be illusory,
just as it has turned out that languages are much more similar in their syntactic
organization than was once thought. Any approach to such problems, however,
must start from a clearer picture of just what lexical structure involves than
is provided if we concentrate simply on the identification of a collection of
irreducible basic signs: a “lexicon” in the sense of a “dictionary.”
8 Language change
In chapter 1, we saw how nineteenth-century linguists promoted the rise of

linguistics as a distinct discipline, thinking of texts as the essential reality and
taking languages to be entities “out there,” existing in their own right, waiting to
be acquired by speakers. For them, languages were external objects and changed
in systematic ways according to “laws” and general notions of directionality.
They focused on the products of human behavior rather than on the internal
processes that underlie the behavior, dealing with E-language rather than
I-language. By the end of the nineteenth century, the data of linguistics con-
sisted of an inventory of sound changes but there were no general principles:
the changes occurred for no good reason and tended in no particular direction.
The historical approach had not brought a scientific, Newtonian-style analysis
of language, of the kind that had been hoped for, and there was no predictability
to the changes – see section 1.2. The historicist paradigm – the notion that there
are principles of history to be discovered – was largely abandoned in the 1920s,
because it was not getting anywhere.
In sections 8.3 and 8.4 we shall ask what kinds of accounts of language
history we can give if we take a more contingent, I-language-based approach.
Following our general theme, we shall shift away from a study of the products
of behavior toward a study of the states and properties of the mind/brain that
give rise to those products. This will enable us to link language change closely
to childhood acquisition and to offer different kinds of explanation, which are
more satisfactory. Before we do that, however, we shall consider more recent,
twentieth-century work, which perpetuates aspects of nineteenth-century think-
ing and pursues E-language approaches to explaining how languages change.
8.1 Long-term directionality in the twentieth century

Nineteenth-century linguists viewed languages changing as objects floating
smoothly through time and space, and that image continued to be adopted
throughout the twentieth century. Despite the move away from historicism in
the 1920s, linguists resumed the search for historical principles in the latter
decades of the twentieth century. In the 1970s much work recast the notion of
157
158 8 Language change
“drift,” originally due to Sapir (1921, ch. 7). Sapir dealt with long-term change
by postulating drifts. A drift represented the unconscious selection of those
individual variations that are cumulative in some direction. So he attributed the
replacement of English whom by who to three drifts:
a. the leveling of the subject/object distinction;
b. the tendency to fixed word order;
c. the development of the invariant word.
Sapir was concerned that in positing a “canalizing” of such “forces” one
might be imputing a certain mystical quality to this history. Certainly the modern
work confirms that fear. Robin Lakoff (1972), for example, examined changes
in various Indo-European languages which yield a more “analytic” surface
syntax, and she sought to combine Sapir’s three drifts into one. The phenomenon
cannot be described, she pointed out, by talking about individual changes in
transformational rules or other aspects of a grammar:
Rather, it must be described as a meta-condition on the way the grammar of a language

as a whole will change . . . Speaking metaphorically, it instructs the language to segmen-
talize where possible . . . It is not at all clear where this meta-condition exists: neither
as part of a grammar nor as a universal condition on the form of grammars. It is not
clear how a constraint on change within one language family, a constraint which is not
absolute but which is nevertheless influential, is to be thought of. But there is no other
way to think of these things: either there is such a meta-condition, whatever it is, or all the
Indo-European languages have been subject to an overwhelming series of coincidences.
(Lakoff 1972, pp. 178, 192)
If the explanation is admittedly incoherent, then maybe the fallacy is in re-

quiring a principled explanation for such a large-scale change taking place over
such a long period. What is wrong with a series of independent events? Why
should we believe that a meta-condition is needed to explain these changes?
Only slightly less mystical is the approach to drifts based on Greenberg’s
(1963) word order typologies. This distinguishes “transitional” and “pure” lang-
uage types, which are defined universally in terms of hierarchically ranked
word-order phenomena. Languages change from one pure type to another by
losing/acquiring the relevant orders in the sequence specified by the hierarchies.
Again, the unit of change is taken to be a language, an object “out there” which
can be described independently of its speakers.
A pure subject–verb–object language, for example, has verb–object order,
auxiliary–verb, noun–adjective, and preposition–NP/DP; and these orders are
ranked in some hierarchy. A subject–object–verb language is essentially the mir-
ror image and has the opposite orders: object–verb, verb–auxiliary, adjective–
noun, and NP/DP–postposition, etc. If a language changes from the object–verb
type to the verb–object type, it acquires all of the new orders in the sequence pre-
scribed by the hierarchy: first verb–object, then auxiliary–verb, and so on. The
8.1 Long-term directionality in the twentieth century 159
hierarchy is the substance of a historical law which stipulates how a language

of one type changes to a language of a different type.
This raises the question of how a child attains a language which is exactly
halfway between the subject–verb–object and subject–object–verb types; how
does she know whether this is a subject–verb–object language changing to
subject–object–verb or vice versa? How does she know that her generation
must push the language a little towards, say, the subject–verb–object type? It
seems that the only conceivable answer is to postulate a racial memory of some
kind, such that the child knows that this is a subject–verb–object language
changing towards subject–object–verb. This is presumably what Robin Lakoff
had in mind in postulating a “meta-condition on the way the grammar of a
language as a whole will change.”
Sapir stressed that whether or not an individual change is part of a larger drift,
its cause must be found locally, thereby avoiding the mystical quality that he
warned against or the need for racial memories. The replacement of whom by
who is part of the general drift towards invariable words and the loss of the case
system, but Sapir nonetheless isolated four reasons for this particular change:
first, whom was felt not to belong to the set of personal pronouns, which have
distinct subject/object forms, but to a set of interrogative and relative pronouns,
which show no such distinction; second, the emphatic nature of interrogative
pronouns militates in favor of their invariableness; third, an objective form rarely
occurs in initial position; and fourth, [hu:m] was alleged to be phonetically
clumsy before the alveolar stops of do and did. We are not interested here in
the validity of this account, but rather in the form that Sapir felt an explanation
should take: local reasons for the change suffice to account for it. This is very
different from the ideas of the 1970s typologists, who argued that notions like
the subject–object–verb to subject–verb–object continua constituted diachronic
explanations (Vennemann 1975); for them, the drift was the explanatory force,
rather than being something which required explanation, and no local causes
were needed.
The typologists remain faithful to the methods of the nineteenth century. They
retain the same kind of historical determinism, believing that a language of one
type changes inexorably to a language of another type; and they deal with the
products of the language capacity rather than with the capacity itself, like their
nineteenth-century predecessors. The goal remains one of finding “straightline
explanations for language change” (Lass 1980), generalizations which would
hold of history. And they are no more successful.
A recent variant on this approach offers historicist accounts in a biologi-
cal, evolutionary guise. Bauer (1995) argued that Latin was a thorough-going
left-branching language, which changed into a thoroughgoing right-branching
system in French. Where Latin has a head to the right of its complement
(exercitum duxit “he led the army,” deorum munus “gift of the gods,” etc.),
French has its head to the left (il conduisit l’armée “he led the army,” le don
des dieux “the gift of the gods”). She explains the change through “an evolu-
tionary concept of language change: . . . languages evolve in the direction of
features that are acquired early” (Bauer 1995, p. 170). She says that “Latin must
have been a difficult language to master, and one understands why this type of
language represents a temporary stage in linguistic development” (Bauer 1995,
p. 188), but she gives no reasons to believe this and she gives no reason why
early languages should have exhibited structures which are hard to acquire.1
If a diachronic change is “adaptive,” one needs to show how the environment
has changed in such a way that the new phenomenon is adaptive in a way that
it wasn’t before. However, proponents of this kind of evolutionary explana-
tion do not do this; instead, they set up universal “tendencies” by which any
change is “adaptive,” such as a tendency for left-branching languages to become
right-branching, and, like the typologists, they postulate inexorable, historical
tendencies as explanatory forces.
Another line of work, again focusing on how languages change in some
global fashion, has similarly emphasized the alleged unidirectionality of change.
Accounts of “Grammaticalization” also treat languages as external objects “out
there,” subject to change in certain inevitable ways. Grammaticalization, a
notion first introduced by Antoine Meillet in the 1930s, is taken to be a semantic
tendency for an item with a full lexical meaning to be bleached over time and
to come to be used to mark a grammatical function. Such changes are said to
be quite general and unidirectional; one does not find changes proceeding in
the reverse direction, so it is said.
We shall discuss an instance of grammaticalization in section 8.3 and there are
many examples that have been described this way in the literature (for a survey,
see Hopper and Traugott 1993). One which is often cited concerns negative
markers in French. In Old French, the negative particle ne was often reinforced
with an appropriate noun. With motion verbs, ne was reinforced with the noun
pas “step.” Over time, pas began to occur even where there was no motion, and
eventually some reinforcing noun became effectively obligatory. As a result,
the reinforcing nouns like pas (others were point “point,” mie “crumb,” gote
“drop”) underwent grammaticalization from noun to negative particle.
Grammaticalization is a real phenomenon, but it is quite a different matter to
claim that it is general and unidirectional, or an explanatory force. If there were a
universal tendency to grammaticalize, there would be no counter-developments,
by which bound forms become independent lexical items (affixes becoming clit-
ics or independent words – we mention an example of this later in this chapter,
1 The same logic, another throwback to nineteenth-century thinking, shows up in the evolutionary
explanations of Haspelmath 1999b; see the commentary on this paper by Dresher and Idsardi
1999.
8.2 Grammars and time 161
when we point to genitive endings in -es in Middle English being reanalyzed

as his, yielding genitives like Christ his sake, Mrs Sands his maid). In our
view, when grammaticalization takes place, it is explained when one points to
local factors which promoted the new grammar; these would be new trigger-
ing experiences, changes in cues, or what Kiparsky (1996) calls the “enabling
causes.”
Grammaticalization, while interesting as a phenomenon, is not an explana-
tory force. We have no well-founded basis for claiming that languages or
grammars change in one direction but not in another, no basis for postulating
historical algorithms mapping one kind of grammar into another kind
(Gx →Gy ). The fact that we observe locative case endings coming to be used with
partitive force in some language does not mean that it cannot be otherwise. Van
Gelderen 1997, Janda 2001, Joseph 2001, Newmeyer 1998, ch. 5, and others
offer careful studies showing changes which run counter to grammaticalization,
“degrammaticalization,” where affixes or minor categories become full lexical
items.2
If we observe a lexical verb being reanalyzed as a functional category in a
given language, we need to seek a local cause, rather than invoke principles
of history as explanatory forces.3 Those that have been invoked do not work
and, if they did, they would themselves need deeper explanation . . . unless one
believes that Clio, the muse of history, or some other unseen hand is directing
human events. The work we have surveyed in chapter 1 and in this section
sees languages following historical laws in order to change from one type
to another. The search for principles which require that a particular type of
system change historically into another type was no more successful in the late
twentieth century than it was in the nineteenth, not surprisingly; and all such
accounts view languages as objects “out there,” external to individual speakers,
E-languages.
8.2 Grammars and time

If we switch our perspective from language change to grammar change, from
E-language to I-language, from the products of the language system to the
2 Janda 2001 offers many references. He also has good critical discussion of how fundamental the
issue of unidirectionality is for grammaticalizationists and how cavalier some of them have been
in dismissing changes which appear to run counter to their predispositions. Imperious blanket
denials that such changes occur, as in the writings of Haspelmath (1999a, 1999c), do not remove
them from history. Newmeyer (1998, ch. 5) provides an excellent general discussion of grammat-
icalization and examines studies which use reconstructions as evidence for “grammaticalization
theory,” despite the fact that it was assumed in the very reconstruction.
3 Nor is it appropriate to explain the change by invoking some principle of UG which favors the
new grammar ( pace Roberts 1993; see Lightfoot 1999, section 8.3 for discussion).
system itself, we explain grammatical change through the nature of the acquisi-
tion process, as we indicated in chapter 2. A grammar grows in a child from some
initial state (Universal Grammar, or UG), when she is exposed to primary ling-
uistic data (PLD) (schematically, as in (8.1)). So the only way a different gram-
mar may grow in a different child is when that child is exposed to significantly
different primary data.
(8.1) primary linguistic data (Universal Grammar → grammar)
There is more to language change, a phenomenon of social groups, than just

grammar change, a phenomenon of individuals. Grammar change is nonethe-
less a central aspect of language change, and it is (naturally enough) intimately
related to other aspects of language change. The explanatory model is essen-
tially synchronic and there will be a local cause for the emergence of any new
grammar: namely, a different set of primary linguistic data. Time plays no role
and there are no principles which hold of history.
St. Augustine held that time comes from the future, which does not exist;
the present has no duration and moves on to the past, which no longer exists.
Therefore there is no time, only eternity. Physicists take time to be “quantum
foam” and the orderly flow of events may really be as illusory as the flickering
frames of a movie. Julian Barbour (2000) has argued that even the apparent
sequence of the flickers is an illusion and that time is nothing more than a sort
of cosmic parlor trick. So perhaps linguists are better off without time, without
attributing intrinsic properties to history.
Let us consider the kinds of explanations that are available if we view gram-
mars as elements of cognition, language organs in individual brains, and then
ask whether we can or should be more ambitious. In the next section we shall
illustrate a different strategy in considering an instance of grammaticalization,
a set of changes affecting modal auxiliaries in English, offering local causes for
the various changes and not invoking any general tendency to grammaticalize
as an explanatory force.
8.3 English auxiliary verbs

English modal auxiliaries like can, could, may, might, will, would, shall, should,
and must differ from ordinary verbs in their distribution. A modal auxiliary is
fronted in a question, but a verb like understand is not (8.2a); a modal occurs
to the left of a negative particle, unlike a verb (8.2b); a modal does not occur
with a perfective (8.2c) or present participle (8.2d), unlike a verb; a modal
does not occur in the infinitival complement to another verb (8.2e), nor as the
complement of another modal (8.2f), unlike a verb like try; and no modal may
occur with a complement DP, whereas some verbs may (8.2g).
(8.2) a. i. Can he understand chapter 4?

ii. *Understands he chapter 4?
b. i. He cannot understand chapter 4.
ii. *He understands not chapter 4.
c. i. *He has could understand chapter 4.
ii. He has understood chapter 4.
d. i. *Canning understand chapter 4 . . .
ii. Understanding chapter 4 . . .
e. i. *He wanted to can understand.
ii. He wanted to try to understand.
f. i. *He will can understand.
ii. He will try to understand.
g. i. *He can music.
ii. He understands music.
The distribution of these modal auxiliaries is peculiar to modern English. For
example, the French verb pouvoir “can” behaves the same way as a regular verb
like comprendre “understand” with respect to movement in a question (8.3a)
and negation (8.3b). Unlike can, pouvoir may occur as a complement to another
verb (8.3c), even to another modal verb (8.3d), and may take a clitic direct object
(8.3e), and to that extent it behaves like ordinary, common-or-garden verbs in
French. In French grammars, the words which translate the English modals,
pouvoir, devoir, etc., walk like verbs, talk like verbs, and are verbs, just like
comprendre.
(8.3) a. i. Peut-il comprendre le chapitre?
can-he understand the chapter
Can he understand the chapter?
ii. Comprend-il le chapitre?
understands-he the chapter
Does he understand the chapter?
b. i. Il ne peut pas comprendre le chapitre.
he neg1 can neg2 understand the chapter
He cannot understand the chapter.
ii. il ne comprend pas le chapitre.
he neg1 understands neg2 the chapter
He doesn’t understand the chapter.
c. Il a voulu pouvoir comprendre le chapitre.
he has wanted to be able to understand the chapter
He wanted to be able to understand the chapter.
d. Il doit pouvoir comprendre le chapitre.
he must to be able to understand the chapter
He must be able to understand the chapter.
e. Il le peut.
he it can
He can (do) it [e.g., understand the chapter].
Furthermore, not only may languages differ in this regard, but also different
stages of one language. Sentences along the lines of the non-existent utterances
of (8.2) were well-formed in earlier English. If the differences between Old and
modern English were a function of separate features with no unifying factor
(Ross 1969), we would expect these features to come into the language at dif-
ferent times and in different ways. On the other hand, if the differences between
Old and modern English reflect a single property, a categorical distinction, then
we would expect the trajectory of the change to be very different. And that is
what we find. If the differences between can and understand were a function
of the single fact that understand is a verb while can is a member of a different
category, inflection (I), then we are not surprised to find that (8.2ci), (8.2di),
(8.2ei), (8.2fi), and (8.2gi) dropped out of people’s language in parallel, at the
same time.
In Middle English Kim can understand the chapter had the structure (8.4a)
and in present-day English (8.4b). If in present-day English can is an I element,
as in (8.4b), then one predicts that it cannot occur to the left of a perfective or
present participle, as in (8.2ci), (8.2di) (those participial markers are generated
in Spec VP), that it is mutually exclusive with the infinitival marker to (which
also occurs in I) (8.2eii), that there may only be one modal per VP (8.2fi),
and that a modal may not be followed by a complement DP (8.2gi). Simply
postulating the structure of (8.4b) accounts for the data of (8.2c–g) in present-
day English. Earlier English had structures like (8.4a), where can is a verb and
behaves like understand.
(8.4) a. Middle English IP
Spec IP
I VP
V VP
Kim can understand the chapter

b. Present-day English IP
Spec IP
I VP
Kim can understand the chapter
If we attend just to changing phenomena, the historical change between

Middle and present-day English consists in the loss of various forms, not in
the development of new forms; people ceased to say some things which had
been said in earlier times. Before the change, all of the utterances in (8.2c–g)
(we return to (8.2a–b) in a moment) might have occurred in a person’s speech,
but later only those forms not marked with an asterisk. That fact alone sug-
gests that there was a change in some abstract system. People might start to use
some new expression because of the social demands of fashion or because of
the influence of speakers from a different community, but people do not cease
to say things for that sort of reason. There might be an indirect relationship,
of course: people might introduce new expressions into their speech for exter-
nal, social reasons, and those new expressions (e.g. analogues to (8.2aii) and
(8.2bii) with the periphrastic do: Did he understand chapter 4?) might entail
the loss of old forms, but one needs an abstract system in order to make such
a link. Changes involving only the loss and obsolescence of forms need to be
explained as a consequence of some change in an abstract, cognitive system.
This methodological point is fundamental.
If one focuses on the final disappearance of the relevant forms, one sees that
they were lost at the same time. The most conservative writer in this regard was
Sir Thomas More, writing in the early sixteenth century. He used many of the
starred forms in (8.2c–g) and had the last attested uses of several constructions.
His grammar treated can, etc. as verbs in the old fashion (8.4a), and the fact
that he used all the relevant forms and his heirs none, suggests that his grammar
differed from theirs in one way and not that the new grammars accumulated
unrelated features. The uniformity of the change suggests uniformity in the
analysis, and that tells us about people’s language organs. There was initially
a single change, a change in category membership: can, etc., formerly verbs
which moved to I in the course of a derivation, came to be analyzed as I elements
(8.4b). The fact that there was a single change in grammars accounts for the
bumpiness: several phenomena changed simultaneously.
The change in category membership of the English modals explains the

catastrophic nature of the change, not in the sense that the change spread through
the population rapidly, but that phenomena changed together. The notion of
change in a grammar is a way of unifying disparate phenomena, taking them
to be various surface manifestations of a single change at the abstract level.
Anthony Kroch and his associates (see Kroch 1989, for example) have done
interesting statistical work on the spread of such changes through populations
of speakers, showing that it is the grammars which spread: competing grammars
may coexist in individual speakers for periods of time. They have shown that
the variation observed represents oscillation between two fixed points, two
grammars, and not more random oscillation in which the phenomena vary
independently of each other. In addition, we now have computer models that
simulate the spread of grammars through populations of speakers, exploiting
the methods of population genetics (Niyogi and Berwick 1997, Briscoe 2000,
Niyogi 2002, Yang 2002).
So much for the nature of the change, a change in people’s grammars. Now let
us ask why this change took place. Under the cognitive view of grammars, we
do not appeal to historical forces as explanations. The cause of the grammatical
change can only be earlier changes in PLD; there must have been differences in
what people – in particular, children – experienced. In this instance, there were
two such changes.
First, the modal auxiliaries became distinct morphologically, the sole sur-
viving members of the preterite-present class of verbs. Here is how that hap-
pened. In early English, there were many verb classes, and the antecedents of
the modern modals were preterite-presents. The preterite-presents (so-called
because their present tense forms had past tense or “preterite” morphology)
were distinct in that they never had any inflection for the third person singular,
although they were inflected elsewhere: þ u cannst “you(sg) can,” we cunnan
“we can,” we cuð on “we could.” Nonetheless, they were just another class of
verbs, one of many, and the forms that were to become modal auxiliaries be-
longed to this class, along with a number of other verbs which either dropped out
of the language altogether or were assimilated to another more regular class of
verbs. For example, unnan “grant” was lost from the language and witan “know”
simply dropped out of the preterite-present class, coming to be treated like non-
preterite-presents. After the simplification of verb morphology, verb classes
collapsed and the only inflectional property of present tense verbs to survive
was the -s ending for the third person singular, and the preterite-present verbs
had always lacked that property. The preterite-presents did not change in this
regard, but a great mass of inflectional distinctions had disappeared, and now
the preterite-presents were isolated; they looked different from all other verbs
in lacking their one morphological feature, that -s ending. And the surviv-
ing preterite-presents were the elements which would be recategorized as I
items.
The morphological distinctiveness of the surviving preterite-presents, the

new modals, was complemented by a new opacity in their past tense forms.
The past tense forms of the preterite-present verbs were phonetically identical
in many instances to the subjunctive forms and, when the subjunctive forms
were lost, past tense forms survived with subjunctive-type meanings rather
than indicating past time reference. While loved is related to love pretty much
exclusively in terms of time reference in present-day English, the relationship
between can and could is sometimes one of time (8.5ai) and sometimes has
nothing to do with time (8.5aii). And might is never related to may in terms of
time in present-day English (8.5bi,bii); in earlier times, might did indicate past
time (8.5biii) but the thought of (8.5biii) would need to be expressed as might
not have intended in present-day English. So might, could, should, etc. came
to take on new meanings which had nothing to do with past time, residues of
the old subjunctive uses; that is what we mean when we say that the past tense
forms became semantically opaque.
(8.5) a. i. Kim could understand the book, until she reached page
56.
ii. Kim could be here tomorrow.
b. i. *Kim might read the book yesterday.
ii. Kim may/might read the book tomorrow.
iii. These two respectable writers might not intend the mis-
chief they were doing. (1762 Bp Richard Hurd, Letters
on Chivalry and Romance, 85)
As a result of these two changes, the preterite-present verbs came to look
different from all other verbs in the language: they were morphologically dis-
tinct, and also their past tense forms did not have the meanings associated with
the past tense of real verbs. UG provides a small inventory of grammatical
categories and elements are assigned to a category on the basis of their mor-
phological and distributional properties. Consequently, morphological changes
entail new primary linguistic data: after the morphological changes occurred,
children heard different things, and these new PLD may trigger new category
distinctions. In this case, we know that, following the morphological changes,
the surviving verbs of the preterite-present class were assigned to a new gram-
matical category, and that change was complete by the early sixteenth century.
The evidence for the new category membership is the simultaneous loss of the
phenomena we discussed in (8.2c–g).
There were two stages to the history of English modal auxiliaries (Lightfoot
1999, ch. 6). First, a change in category membership, whereby can, etc. ceased to
be treated as verbs and came to be taken as manifestations of the I category; this
change affected some verbs before others, but it was complete by the sixteenth
century. Consequently, for a sentence like Kim can sing, early grammars had
structures like (8.4a), where can is an ordinary verb which sometimes moves to
I, but later grammars had structures like (8.4b), where can is an I item, drawn
from the lexicon and merged into a structure as an instance of I. As a result,
sentences like (8.2ci–gi) dropped out of the language and no longer occurred
in texts.
The second stage was that the grammars of English speakers lost the operation
moving verbs to a higher I position (e.g., in 8.4a). This change was completed
only in the eighteenth century, later than is generally supposed (Warner 1997).
At this point, sentences with a finite verb moved to some initial position (8.2aii)
or to the left of a negative (8.2bii) became obsolete and were replaced by equiv-
alent forms with the periphrastic do: Does Kim understand this chapter? Kim
does not understand this chapter, etc. Also sentences with an adverb between
the finite verb and its complement became obsolete: Kim reads always the
newspapers. This change has been discussed extensively and Lightfoot (1999,
section 6.3) argues that it was caused by prior changes in PLD, most notably the
recategorization of the modal verbs just discussed and the rise of periphrastic do
forms (above). These changes had the effect of greatly reducing the availability
of the relevant cue, [I V], i.e. a verb occurring in an I position.
The two changes are, presumably, related in ways that we do not entirely un-
derstand: first, the Inflection position was appropriated by a subclass of verbs,
the modal auxiliaries and do, and the V-to-I operation no longer applied gener-
ally to all tensed clauses. Somewhat later, the V-to-I movement operation was
lost for all verbs other than the exceptional be and have (see below) and I was
no longer a position to which verbs might move. We pass over the details of
this change here, in order to discuss something else.
An intriguing paper by Anthony Warner (1995) shows that there is a third
stage to the history of English auxiliaries, involving changes taking place quite
recently affecting the copula be, and this turns out to be of current theoretical
interest. It has often been observed that VP ellipsis is generally insensitive to
morphology. So one finds ellipses where the understood form of the missing
verb differs from the form of the antecedent (8.6).
(8.6) a. Kim slept well, and Jim will [sc. sleep well] too.
b. Kim seems well behaved today, and she often has [sc. seemed
well behaved] in the past, too.
c. Although Kim went to the store, Jim didn’t [sc. go to the
store].
There is a kind of sloppy identity at work here. One way of thinking of this is
that in (8.6a) slept is analyzed as [past+V sleep] and the understood verb of the
second conjunct accesses the verb sleep, ignoring the tense element. However,
Warner noticed that the verb be works differently. Be may occur in elliptical
constructions, but only under conditions of strict identity with the antecedent
form (8.7). In (8.7a,b) the understood form is identical to the antecedent, but
not in the non-occurring (8.7c,d,e).
(8.7) a. Kim will be here, and Jim will [sc. be here] too.
b. Kim has been here, and Jim has [sc. been here] too.
c. *Kim was here, and Jim will [sc. be here] too.
d. *If Kim is well behaved today, then Jim probably will [sc.be
well behaved] too.
e. *Kim was here yesterday, and Jim has [sc. been here] today.
This suggests that was is not analyzed as [past + V be], analogously to slept,
and be may be used as an understood form only where there is precisely a be
available as an antecedent; not was or is, but just be, as in (8.7a). Similarly for
been; compare (8.7b) and (8.7e). And similarly for am, is, are, was, were.
Warner goes on to note that the ellipsis facts of modern English were not
always so and one finds forms like (8.7c,d,e) in earlier times. Jane Austen was
one of the last writers to use such forms and she used them in her letters and in
speech in her novels, but she did not use them in narrative prose (8.8a,b). These
forms also occur in the work of eighteenth century writers (8.8c), and earlier,
when verbs still moved to I (8.8d).
(8.8) a. I wish our opinions were the same. But in time they will [sc. be
the same]. (1816 Jane Austen, Emma, ed. by R.W. Chapman,
London: OUP, 1933. 471)
b. And Lady Middleton, is she angry? I cannot suppose it pos-
sible that she should [sc. be angry]. (1811 Jane Austen, Sense
and Sensibility, ed. by C. Lamont, London: OUP, 1970. 237)
c. I think, added he, all the Charges attending it, and the Trouble
you had, were defray’d by my Attorney: I ordered that
they should [sc. be defrayed]. (1740–1 Samuel Richardson,
Pamela, London. 3rd edition 1741. Vol. II, 129)
d. That bettre loved is noon, ne never schal. (c. 1370 Chaucer,
A Complaint to his Lady, 80. “So that no one is better loved,
or ever shall [sc. be].”)
These forms may be understood if were in (8.8a) was analyzed as subjunctive
+be and the be was accessed by the understood be. In other words, up until the
early nineteenth century, the finite forms of be were decomposable, just like
ordinary verbs in present-day English. This is what the ellipsis facts suggest.
Warner then points to other differences between present-day English and
the English of the early nineteenth century. Present-day English shows quite
idiosyncratic restrictions on particular forms of the verb be, which did not exist
before the late eighteenth century. For example, only the finite forms of be may
be followed by to+infinitive with a modal sense of obligation (8.9a); only been
may occur with a directional preposition phrase, effectively meaning “gone”
(8.9b); and being, unlike any other form of be, has a special restriction that it
does not permit an ing complement (8.9c).
(8.9) a. i. Kim was to go to Paris.

ii. *Kim will be to go to Paris.
b. i. Kim has been to Paris.
ii. *Kim was to Paris.
c. i. I regretted Kim reading that chapter.
ii. I regretted that Kim was reading that chapter.
iii. *I regretted Kim being reading that chapter.
Restrictions of this type are stated in the lexicon and these idiosyncrasies show
clearly that been, being, etc. must be listed as individual lexical entries, because
they have individual peculiarities. However, these restrictions are fairly new in
the language and we find forms corresponding to the non-occurring sentences
of (8.9a–c) through the eighteenth century:
(8.10) a. You will be to visit me in prison with a basket of pro-
visions; . . . (1814 Jane Austen, Mansfield Park, ed. by
J. Lucas, Oxford: Oxford University Press, 1970. 122)
b. I was this morning to buy silk. (1762 Oliver Goldsmith, Cit
W: 158 (meaning “I went to . . . ,” not “I had to . . . ”))
c. Two large wax candles were also set on another table, the
ladies being going to cards. (1726 Daniel Defoe, The Polit-
ical History of the Devil, Oxford: Talboys, 1840. 336)
d. . . . he being now going to end all with the Queene . . . (1661
Samuel Pepys, Diary II 129.1(30 June))
e. One day being discoursing with her upon the extremities
they suffered . . . (1791 Daniel Defoe, Robinson Crusoe.
Vol. II, 218)
f. . . . and exclaimed quite as much as was necessary, (or,
being acting a part, perhaps rather more,) at the conduct
of the Churchills, in keeping him away. (1816 Jane Austen,
Emma. 145)
g. Their being going to be married. (1811 Jane Austen, Sense
and Sensibility, ed. by R.W. Chapman, Oxford: Oxford
University Press, 1923. 182)
h. The younger Miss Thorpes being also dancing, Catherine
was left to the mercy of Mrs. Thorpe. (1818 Jane Austen,
Northanger Abbey, ed. by R.W. Chapman, Oxford: Oxford
University Press, 1923. 52)
Warner concludes that after the change
was continues to carry the information that it is third singular past finite indicative. The
point is that it does not do so as a consequence of inflection or inflectional relationships,
but is essentially monomorphemic. The relationship was:be becomes fundamentally
different not only from regular verbs loved:love, etc. but also from that of irregular or
suppletive verbs (slew:slay, went:go), which are in some sense essentially compositional,
as the contrast of behavior in ellipsis shows. (Warner 1995, p. 538)
Whether the change affects be alone or also have, it is clearly restricted to a

very narrow class of lexical items. The analysis, however, reflects quite general
properties and the change relates to the matter of category membership. Forms
of be cease to be analyzed as a verb with the usual morphological structure and
came to be listed in the lexicon atomically with no internal structure.
Words are stored in different ways in the mental lexicon. Irregular words are
stored individually, while words formed by a regular, productive process are not.
The various forms of be are listed individually in the lexicon, like mice and men,
and they are not composed by productive lexical processes. Our conclusion,
then, is that was, been, etc. have no internal structure and occur in the lexicon
as unanalyzed primitives. In that case, computational operations like those
involved in ellipsis may access atomic forms like was or been but not any
internal structure along the lines of [past+be], because they do not exist.
Grammars have quite small sets of categories and individual lexical items
are assigned to one or other of these categories. Evidence suggests that items
are assigned to grammatical categories on the basis of their morphological
properties. If a given item becomes morphologically distinct from other items
in its category, then it is liable to be assigned to a new category. This is what
was involved in the earlier changes involving the modal auxiliaries, formerly
preterite-present verbs. As a result of changes affecting the preterite-present
class of verbs and the subjunctive mood, these items became morphologically
and semantically distinct from other verbs and were assigned to a new category,
I(nflection), as we just saw. Similarly with the changes affecting be in the
eighteenth century.
If we ask why this third change took place, we get some insight into the
grammatical property involved. Again, the only explanation available to us
must derive from earlier changes in PLD, in people’s linguistic experience. The
change in the internal structure of be forms was preceded by two other changes.
First, the operation moving verbs to an I position, V-to-I, was finally lost
in the eighteenth century, as manifested by the obsolescence of sentences like
(8.2aii) and (8.2bii). The loss of V-to-I movement further distinguished the
modal auxiliaries from verbs, with the modals continuing to occur in the old
I positions, clause-initially and to the left of a negative. Unlike modals in other
respects, be patterns with them in this regard and not with verbs: Is she happy?
and She is not happy.
Second, the pronoun thou and the accompanying inflectional forms in (e)st
were lost from informal spoken English in the eighteenth century. Thou had the
use of French tu, directed to children, intimates, and inferiors. It was lost for
social reasons, as speakers abandoned a linguistic manifestation of these social

distinctions. The obsolescence of Thou shalt, Thou shouldest, etc. removed the
last inflectional property of modal verbs which was shared with ordinary verbs.
Modals had become very different from verbs in the course of changes that we
have linked to their change in category, completed by the sixteenth century. They
ceased to have non-finite forms, ceased to have mood and tense forms, ceased
to have complements, etc; all of which showed that they were recategorized as
I elements. The loss of the second person singular forms constituted a further
step in this bifurcation: the last remaining inflectional commonality shared by
modals and ordinary verbs was lost. This last change also affected the verb be,
which had shown no symptoms of the category change affecting the modals: be
did not lose its non-finite forms in the way that can did, nor did it lose its mood
and tense forms, nor its ability to be followed by a complement. Now, however,
forms of be became distinct from ordinary verbs in that they continued to move
to I and on to C, while main verbs lost this ability, and they did not look like
verbs any more in terms of their finite inflectional properties.
The first of these changes affected all verbs – they ceased to move to I – and
the second change affected all verbs and (modal) elements generated in the
I position. Neither change affected be in particular, but their effect was to single
out be and make it less like a verb. As a result of these changes, be differed from
verbs in that it could occur in (structurally) high functional positions like I and
C, and it lacked the only morphological properties characteristic of verbs, the
universal third person singular marker in s and the usual past tense ed marker.
It is plausible that these were the changes which led children to treat be
differently. Be was once a verb, but now it ceased to look like a verb and
ceased to have the internal structure of a verb. Nor was it an I element: be did
not undergo the category change that the modals underwent by the sixteenth
century and it was never associated intrinsically with the (finite) I position in
the way that modals are, and continues to occur in non-finite contexts (I want
to be happy, Being happy is a good thing, Be happy!, etc). This shows that the
finite forms of be move to I and are not base-generated there like the modals.
In the eighteenth century verbs ceased to move to I but the finite forms of be
continued to do so. As a result, be no longer walked like a verb and no longer
talked like a verb. The evidence suggests that, in addition, in the nineteenth
century be ceased to show the paradigmatic properties of verbs. Be was no
longer categorized as a verb; instead, its inflectional paradigm was individually
specified as a series of lexical items, stored individually in the mental lexicon.
This was the grammatical change.
Forms of be (and similarly verbal forms in French) are fully formed by lexical
operations but verbs in English are bare, acquiring affixes via the syntax, when
syntactic features are realized on the verb. Before that takes place, the abstract
sleep of slept is identifiable as a distinct unit and therefore is an appropriate
antecedent for ellipsis in (8.6a) Kim slept well, and Jim will [sc. sleep well]
too. This reveals how elements are stored in the mental lexicon: is is stored
in just that form while slept is stored as V sleep with the form slept created
morphologically by the attachment of an affix. If all verbs were treated the same
way, as in Chomsky 1995, there would be no obvious way to make the distinction
between those which may be antecedents for ellipsis under conditions of sloppy
identity (sleep, etc.), and those which may not (is, are, and other forms of be).
Lasnik (1999, ch. 5) drew a similar distinction between “affixal” and
“featural” verbs and keyed the distinction to whether the verb moves: if a
verb moves in the syntax (e.g., be forms and all finite verbs in French), then it
already has its inflectional features attached when it is merged into the syntactic
structure and is “featural,” but if a verb does not move to a higher inflectional
position, then it is “affixal” and has affixes lowered onto it in the syntax. How-
ever, this correlation is not general and there is more to the story than this.
Modal elements are featural and are generated in I, not moving there. Finite
be, on the other hand, clearly moves to I, because be may also occur in other,
non-finite positions if I is filled with a modal (8.11).
(8.11) Kim might still be reading that chapter.
So forms of be (and have) move to I; they are and always have been featural.
They have always moved to I at all stages of their history but it was only in the
late eighteenth century that they came to be stored atomically and developed the
odd properties discussed here. We conclude that if a verb is featural, it moves
to I. However, a featural item may be base-generated in I (modern modals) and
may or may not be stored atomically: was is not a verb and it is stored atomically
in modern grammars.
What is important about this story is that, while the changes we have discussed
only involve the verb be, they have the hallmarks of grammatical change. There
are several surface changes, all involving be, which can be attributed to one
analytical notion. The changes reflect quite general properties of the grammar.
One can identify the structural property which is relevant and we can tell a
plausible and rather elegant story about why and how the grammatical change
might have come about. We distinguish how items are stored in the lexicon.
We see, again, that morphology has syntactic effects. It is particularly impor-
tant in defining category membership; children assign items to categories on the
basis of their morphology. We have explained the third change by pointing to
changes in the trigger experience which led to the new morphological structure
of be forms. Those changes in the trigger are a function of prior grammatical
shifts, relating to the change in category membership of the modal auxiliaries
and the loss of V-to-I movement; there are links among the three changes and
we have another domino effect. Again we have local causes and we do not need
to appeal to internal motivating factors.
While morphology clearly influences category membership, one finds a

stronger claim in the literature. It is sometimes argued that richly inflected
languages differ in a fundamental, global way from poorly inflected languages
like English, Swedish, and Chinese. In this context, it might be argued that
grammars with rich inflectional systems, like those of French speakers, list
forms individually in the lexicon and do not form them by general operations.
On the analysis offered here, this would entail that there would be no bare forms
available to syntactic operations and one would find various lexical restrictions
on particular forms, as we saw for be in present-day English in (8.10) above.
In grammars like those of present-day English speakers, involving fewer in-
flectional distinctions, there are fewer individual listings. Plural nouns in s and
third person singular forms of verbs in s are composed in the syntax through
the affixal realization of featural content, and they are not listed in the lexicon
fully formed. As a result, the bare form of the noun and of the verb, shorn of
its affixal ending, is available to various grammatical operations and it is not
possible to state lexical restrictions on particular third person singular forms.
However, the material of this section shows that this is not correct and it sug-
gests that the way items are stored in the lexicon is neither a function of move-
ment nor a simple function of morphological richness. Verbs and be could occur
in ellipses without strict identity with their antecedent up until the nineteenth
century. This suggests strongly that their forms were not stored atomically, even
though they were richly inflected in the early stages of the language; instead,
they were formed by operations applying internal to the lexicon. Consequently,
we conclude that the way elements are stored in the lexicon has many syntactic
consequences, but that distinction represents variation which is independent of
whether an item moves to an Inflection position and independent of whether an
item is richly inflected.
In affecting a narrow class of words, the third change described here is small
scale. It can be understood in terms of prior changes, including the highly
contingent loss of thou forms, and it reflects ways in which items may be stored
in the mental lexicon. It is a function of earlier changes.
In this section we have tracked some changes affecting the English modal
auxiliaries, changes which might be labeled “grammaticalization.” We have
shown local causes for each of the three changes in grammars (the new cate-
gory membership of the modal auxiliaries, the loss of V-to-I movement, and
the new atomic lexical forms was, etc.), taking grammars to be individual, in-
ternal systems existing in individual brains. There was nothing inevitable about
these changes: the equivalent words in French and Swedish did not undergo
parallel changes, because there were no parallel local causes. Grammaticaliza-
tion theory has nothing to say about small-scale changes like these, nor does it
tell us why Swedish and French have not behaved like English. The problem
with postulating general historical tendencies is that they are too “Cyclopean”
(to adopt a useful term from Calvert Watkins’ (1976) critique of typological
analyses) and too gross to be enlightening, and they predict that languages
should undergo parallel historical changes.
If changes in category membership are relatively common (whatever that
means), they still need local causes. Identifying local causes enables us to
understand the details of the change, as we have illustrated here. This case
study suggests that category changes may result from morphological changes.
Not many of the world’s languages have a richly recorded history, but many
that do have undergone morphological simplification, sometimes with category
changes. If our historical records included languages with increasing morpho-
logical complexity, we would be in a stronger position to relate morphological
and categorial changes. However, given the records that we have, we can see
the precariousness and uselessness of seeking to explain categorial changes by
general historical tendencies.
8.4 Syntactic effects of the loss of case

In chapter 3 we sketched a theory of abstract Case, which determines the
positions in which DPs may be pronounced, extending the traditional notion
of morphological case. DPs occur as the subject of a finite clause (they left),
as a specifier of a larger DP (their book), or as a complement to a verb or a
preposition (read them, look at them). Our abstract Case seems to be related
to the overt, morphological case studied by earlier grammarians, and one can
probe that relationship by examining what happens syntactically when a mor-
phological case system is eroded and eventually lost.
In this section we shall examine some curious syntactic effects resulting
from the disappearance of the morphological case system in English. The loss
of morphological case will enable us to understand to a significant degree the
hitherto mysterious emergence of new “split genitives” in Middle English. What
strikes us is the tightness of the explanation, the way in which one element of
Case theory explains the details of the development. We shall see that if one
part of a child’s linguistic experience changes, namely the transparency of the
case morphology, then other things must also change in the grammars which
emerge.4
The Case theory we adopt links Case to thematic roles. Thematic roles define
the semantic relationship of DPs to a head. In (8.12a) the subject DP is an Agent,
in (8.12b) a Theme, in (8.12c) a Location, and in (8.12d) an Instrument. These
4 The loss of morphological case is discussed more fully in Lightfoot 1999, ch. 5, from which this
study is drawn, with some revisions.
thematic roles are a function of the meaning of the verb and are “assigned”
by the verb, so the DPs are thematically linked to the verbs. In a sentence like
Kay drove to New York, New York is thematically linked to the preposition to
and not to the verb drove; in a phrase John’s mother’s house, the DP John’s
mother is thematically related to house but the smaller DP John is thematically
related only to mother.
(8.12) a. [DP The striker] kicked the defender.

b. [DP The striker] received the award.
c. [DP The Indian Ocean] surrounds Madagascar.
d. [DP The wind] blew the door open.
If UG stipulates that heads may assign Case to the left or to the right in
accordance with the head-order parameter, as we indicated in chapter 3, one is
not surprised to find Old English nouns assigning Case to the left and to the
right. There is good reason to believe that the head-order parameter was shifting
in late Old English: one finds verbs preceding and following their complement,
object–verb order alternating with verb–object. There is independent evidence
that OE nouns assigned genitive Case not only to the left (8.13a) but also
to the right (8.14b). One finds possessive–head order alternating with head–
possessive. Old English has a very simple analysis. It is more or less a direct
manifestation of this UG theory of Case: nouns assigned Case to the left and
to the right, and only to DPs with which they were thematically related, as we
shall see. Case was assigned in that fashion and then was realized on both sides
of the noun with the morphological, genitive suffix. Lof assigns a thematic role
to god in (8.13ai) and lufu to god and mann in (8.13bi).
(8.13) a. i. Godes lof “praise of God” (Ælfric)

ii. Cristes læwa “betrayer of Christ”
b. i. Lufu godes and manna “love of God and of men”
(Ælfric, Catholic Homilies ii, 602.12)
ii. Ormæte stream wæteres “huge stream of water”
(Ælfric, Catholic Homilies ii, 196.5)
If Old English nouns assigned Case to the left and to the right, and if in both
positions it was realized as a morphological genitive, then one is not surprised
to find that Old English also manifested “split genitives” (the term is Eilert
Ekwall’s (1943)). These were split in that a single genitive phrase occurred
on both sides of the head noun. In (8.14) we see an example where the split
element occurring to the right of the noun was a conjunct. Jespersen (1909,
p. 300) notes that with conjuncts, splitting represents the usual word order in
Old English.
(8.14) a. Inwæres broþur ond Healfdenes

Inwær’s brother and Healfden’s
Inwær’s and Healfden’s brother (AS Chron., 878)
b. Sodoma lande 7 gomorra
Sodom’s land and Gomorra
The Sodomites’ and the Gomorreans’ land (West Saxon
Gospels (Ms A), Matt 10.15)
In addition, appositional elements, where two DPs are in parallel, were usu-
ally split: the two elements occurred on either side of the head noun (8.15a–c),
although (8.15d) was also possible, where Ælfredes cyninges is not split.
(8.15) a. Ælfredes godsune cyninges.

King Alfred’s godson. (AS Chron., 890; Laud (Peterbor-
ough) [E] 1122))
b. þæs cyninges dagum herodes.
In the days of Herod the king. (West Saxon Gospels (Ms A),
Matt 2.1)
c. Iohannes dagum fulwihteres.
From the days of John the Baptist. (West Saxon Gospels
(Ms A), Matt 11.12)
d. Ælfredes cyninges godsunu.(AS Chron.,890(Parker c.900))
e. *The book’s cover about Rio (= The book about Rio’s cover)
f. *þæs cyninges godsune Frances
the of king godson of France
The king of France’s godson
Splitting within DPs was restricted to conjuncts (8.14) and to appositional

elements (8.15a–c). In particular, Old English did not show split constructions
with a preposition phrase (PP), along the lines of (8.15e). So there was no
general rule “extraposing” a PP. Nor does one find anything like (8.15f), where
Frances has no thematic relation to godsune.
Split genitives in Old English had a structure along the lines of (8.16).
Ælfredes was in the specifier of DP. Godsune assigned a thematic role and
Case to the left and to the right.5
(8.16) [DP Spec Ælfredes D [NP godsune [cyninges]]]
These grammars had an overt genitive case on the right or on the left of the
head noun; and they had split genitives, where the head noun assigned the same
5 Cynthia Allen (2002) argues that cyninges is an adjunct to godsune rather than a complement.
This raises interesting questions which we shall not discuss here.
thematic role and Case in both directions. So much for splitting in Old English
grammars.
Now for the mysterious changes. Middle and early Modern English also
manifested split genitives but they included forms which are very different
from the split genitives of Old English, as the examples of (8.17) show.
(8.17) a. The clerkes tale of Oxenford. (Chaucer, Clerk’s Tale,
Prologue)
b. The Wive’s Tale of Bath. (Chaucer, Wife of Bath’s Tale,
Prologue)
c. Kyng Priamus sone of Troy. (Chaucer, Troilus & Cressida,
I, 2)
d. This kynges sone of Troie. (Chaucer, Troilus & Cressida,
III,1715)
e. The Archbishop’s Grace of York. (Shakespeare, 1 Henry IV,
III.ii.119)
The meaning is “The clerk of Oxford’s tale,” “King Priam of Troy’s son,”
etc, and the genitive is split in the same sense as in Old English grammars: the
rightmost part of the genitive phrase (italicized) occurs to the right of the head
noun which the genitive phrase modifies. Mustanoja (1960, p. 78) notes that
“the split genitive is common all through ME [Middle English]” and is more
common than the modern “group genitive,” The clerk of Oxford’s tale. Jespersen
(1909, p. 293), exaggerating a little, calls this splitting “the universal practice
up to the end of the fifteenth century.” However, these Middle English split
forms are different from those of Old English grammars, because the rightmost
element is neither a conjunct nor appositional, and it has no thematic relation
with the head noun, tale, son, Grace, but rather with the item to the left: clerk,
wife, etc. How did these new split forms emerge and become so general?
We can understand the development of the new Middle English split genitives
in light of the loss of the overt morphological case system and the theory of
Case related to thematic role. Culicover (1997, pp. 37f.) discusses the “thematic
case thesis,” under which abstract Case realizes thematic-role assignment quite
generally. This is where we seek to connect work on abstract Case with the
morphological properties discussed by earlier grammarians.
Old English had four cases (nominative, accusative, genitive, and dative) and
a vestigial instrumental, but they disappear in the period of the tenth to thirteenth
century, the loss spreading through the population from the north to the south
probably under the influence of the Scandinavian settlements (O’Neil 1978). In
early Middle English, grammars emerged which lacked the morphological case
properties of the earlier systems, in particular lacking a morphological genitive.
Put yourself now in the position of a child with this new, caseless grammar;
your grammar has developed without morphological case. You are living in
the thirteenth century; you would hear forms such as (8.15a) Ælfredes godsune
cyninges, but the case endings do not register: that is what it means not to have
morphological case in one’s grammar. You are not an infant and you are old
enough to have a partial analysis, which identifies three words. Ælfredes was
construed as a “possessive” noun in the specifier of DP.
The modern “possessive” is not simply a reflex of the old genitive case.
Morphological case generally is a property of nouns. On the other hand,
“possessive” in modern English is a property of the DP and not of nouns: in
(8.18a) My uncle from Cornwall’s cat the possessor is the whole DP My uncle
from Cornwall. Allen (1997) shows that the ’s is a clitic attached to the preced-
ing element and that the group genitive, where the clitic is attached to a full DP,
is a late Middle English innovation.
(8.18) a. [DP [DP my uncle from Cornwall]’s cat]
b. Poines his brother. (Shakespeare, 2 Henry IV, 2.4.308)
c. For Jesus Christ his sake. (1662 Book of Common Prayer)
d. Mrs. Sands his maid. (OED, 1607)
e. Job’s patience, Moses his meekness, and Abraham’s faith.
(OED, 1568)
As the case system was lost, the genitive ending es was reanalyzed as
something else, a Case-marking clitic. If ’s comes to be a clitic in Middle
English, which Case-marks DPs, this would explain why “group genitives” be-
gin to appear only at that time, as Allen argued. Allen’s analysis also predicts
Jespersen’s observation that splitting was the universal practice until the clitic
became available.
It is likely that there was another parallel reanalysis of the genitive es ending,
yielding the his-genitives which were current in the sixteenth and seventeenth
centuries (8.18b,c) for “Poines’ brother,” “Christ’s sake,” etc. The genitive
ending in ’s was sometimes spelled his, and this form occurs even with females
(8.18d), and occurs alongside possessive clitics (8.18e).
UG dictates that every phonetic DP has Case, as we sketched in chapter 3.
The new caseless children reanalyzed the old morphological genitive suffix
es as a clitic, which was recruited as a Case-marker. The clitic ’s Case-marks
the element in the specifier of the containing DP. So Ælfred has Case and the
Case is realized through the ’s marker (usually analyzed as the head D, as in
the structure given for (8.19a); see also chapter 3). In short, the Ælfredes of the
parents is reanalyzed as Ælfred’s, although orthographic forms like Ælfredes
occur in texts when mental grammars surely yielded Ælfred’s. Orthographic ’s
is a recent innovation. So far, so good.
What about cyninges in (8.15a)? The evidence suggests that the phrase
became (8.19a) Ælfred’s godsune king. One finds phrases of just this form
in (8.19b,c), where the post-nominal noun is not overtly Case-marked, and
Jespersen (1909, pp. 283f.) notes that these forms are common in Middle
English.
(8.19) a. Ælfred’s godsune king
[DP [DP Ælfred] D ’s [NP godsune [king]]]
b. The kynges metynge Pharao
Pharaoh the king’s dream (Chaucer, Book of the Duchess,
282)
c. The Grekes hors Synoun
Sinon the Greek’s horse (Chaucer, Squire’s Tale, 209)
The forms of (8.19), where the rightmost element is appositional, are direct
reflexes of OE split genitives like (8.15), corresponding exactly, except that the
split element, Pharao, Synoun, has no overt Case. Despite the absence (for us
new, caseless children – remember our thought experiment) of an overt, mor-
phological genitive case, UG prescribes that the post-nominal DP must carry
some abstract Case. After the loss of the morphological case system, it can no
longer be realized as a genitive case ending. That means that there must be
another way of marking/realizing the abstract Case in (8.19). Perhaps Pharao
receives its Case by coindexing with the Case-marked kynges; the two forms
are in apposition and therefore are coindexed and share the same thematic role.
This is what one would expect if there is a one-to-one relationship between
Case and thematic role, the key element of our theory of Case. In that event, no
independent Case-marker is needed for Pharao.
There is another option for realizing Case on the rightmost element. The
dummy preposition of could be used as a Case-marker, as it is in (8.17)
(see chapter 3, note 11). This is not possible in Ælfred’s godsune king or the
phrases of (8.19), because if of were to Case-mark the DP, one would expect
it also to assign a thematic role (given a one-to-one relation between Case and
thematic role) and in that event the DP could not be interpreted as an apposi-
tional element. The sentences of (8.17), on the other hand, are not like those of
(8.19) and have different meanings. In (8.17b), for example, Wive and Bath are
not appositional, not coindexed, and therefore an independent Case-marker and
thematic-role assigner is needed; this is the function of of. 6 Under this view, the
emergence in Middle English of the new N of DP forms (8.17) is an automatic
consequence of the loss of the morphological case system: of was introduced
in order to Case-mark a DP which would not otherwise be Casemarked. In
particular, the DP could not be Case-marked like the rightmost item in (8.19),
which carries the same Case as Ælfred’s because it has the same thematic role.
Of assigns Case to a DP only if it has an independent thematic role.
6 Nunnally (1985, p. 21) finds no genitival of phrases in his study of the OE translation of
St. Matthew’s Gospel (of was used frequently to show origin or agency, best translated by
modern from or by).
With the introduction of the of Case-marker in these contexts, there is a

further change and the split genitive construction is extended, as we have noted
(8.17). In Old English, the post-nominal genitive always had a thematic relation
with the head noun; one does not find expressions such as (8.15f) þ æs cyninges
son Frances “The king of France’s son,” where Frances is neither a conjunct
nor appositional and is thematically related to “king” (Nunnally 1985, p. 148,
Cynthia Allen, Willem Koopman, personal communication). In such a phrase,
Frances could not be Case-marked by any adjacent element; in particular, it
could not receive Case from son because it has no thematic relation with son.
In Middle English, one does find post-nominal, split DPs even where there is
no thematic relation with the head noun, and the post-nominal items are Case-
marked by of. So, in (8.17a) Oxenford is construed with clerkes and not with
tale, and it is Case-marked by of. It is crucial to note that the Middle English split
expressions only involve of phrases: one does not find The book’s cover about
Rio for “the book about Rio’s cover,” mirroring the situation in Old English and
showing that there is no general operation “extraposing” PPs in Middle English,
any more than there was in Old English. Additionally, – and this is crucial – the
post-nominal noun in (8.17) always has a thematic role of Locative/Source.
We know of no claim to this effect in the literature but it has been pointed out
to us by Cynthia Allen, Olga Fischer, and Willem Koopman in independent
personal communications and it seems to be right. So, for example, one does
not find forms like (8.20), where the post-nominal noun is a Theme (8.20a) or
a Possessor (8.20b).
(8.20) a. *The portrait’s painter of Saskia (= the portrait of Saskia’s
painter)
b. *The wife’s tale of Jim (= the wife of Jim’s tale)
The fact that the of phrase is associated with a unique thematic role makes
sense if UG prescribes a link between Case and thematic-role assignment. As
we have noted, in Old English (8.15a) Ælfredes godsune cyninges, godsune
assigns the same Case to the right and to the left, realized in both instances as
an overt, morphological genitive case; it also assigns the same thematic role to
the two DPs to which it assigns Case. That is what it means for the two DPs
to be appositional (same Case, same thematic role), and all of this is easy to
understand if Case and thematic-role assignment are linked at the level of UG.
The same applies to conjuncts (8.14). Consequently, the extension of these split
genitives in Middle English (to the new forms with of, (8.17)) is not surprising
under a theory which allows nouns to assign Case and which links Case to
thematic-role assignment.
This much we can understand under the theory of Case. The properties of
the new grammar must have emerged in the way that they did, if children (a)
heard expressions like Ælfredes godsune cyninges (8.15a), (b) did not have
the morphological case system of their parents, and (c) were subject to a Case
theory requiring all DPs to have Case (assigned and realized) and linking Case
with the assignment of thematic roles. We have a tight explanation for the new
properties of Middle English grammars. In particular, we explain the distinction
between (8.17) and (8.19), with of occurring where there is no thematic relation
with the head noun (8.17), but not where there is such a relation (8.19). We
see that change is bumpy; if one element of a grammar changes, there may be
many new phenomena (8.17). Children do not just match what they hear and
they may produce innovative forms, as required by UG. UG defines the terrain,
the hypothesis space, and a change in initial conditions (loss of morphological
case) may have syntactic effects.7
This is an explanation for the form of the split genitives of (8.17) in Middle
English. They were around for four centuries and then dropped out of the
language. This was probably a function of the newly available clitic ’s which
made possible group genitives like The clerk of Oxford’s tale; these became
possible only when ’s was construed as a clitic, which Case-marked DPs, and
that in turn was a function of the loss of morphological cases, including the
genitive in es.
Here we have taken a notion (“case”) from traditional grammar, and con-
strued Case as an element in cognitive grammars, in people’s language organs.
Phonetic DPs, DPs which are pronounced, have an abstract Case which must
be realized somehow. This is required by UG, and abstract Cases are often
realized as morphological cases. Children scan their linguistic environment for
morphological cases and, if they find them, they serve to realize abstract Cases.
If children do not find morphological cases, then different grammars emerge. In
that event, a P or V (or other categories) may Case-mark a complement DP. We
have examined here what happens when everything else remains constant. There
came a point in the history of English when children ceased to find morpholog-
ical cases. Those children were exposed to much the same linguistic experience
as their parents, but the transparency of overt case endings had dropped below a
threshold such that they were no longer attained. Given a highly restrictive the-
ory of UG, particularly one linking Case-assignment by nouns to thematic-role
assignment and requiring Cases to be realized on phonetic DPs, other things
then had to change.
In this way our abstract theory of Case enables us to understand how some
of the details of Middle English grammars were shaped, why things changed
as they did and why Middle English grammars had their odd split genitives.
7 Our account leaves open the question of why these extended split genitives (8.17) should have
arisen. Lightfoot (1999) appeals to the reanalysis of one special type of Old English split geni-
tive, those involving double names like Thomasprest Doreward (= priest of Thomas Doreward)
and, crucially, those where the second part begins with of : Rogereswarenner of Beauchamp
(= warrener of Roger of Beauchamp), which may have triggered the new split genitives.
8.5 Chaos 183
8.5 Chaos
From the Greeks to Newton, people have believed in a predictable universe.
Where unpredictable behavior was observed, for example in weather, the un-
predictability was attributed to lack of knowledge: if we just knew more, we
would have better weather forecasts. Pierre Simon Laplace said that he could
specify all future states if he could know the position and motion of all particles
in the cosmos at any moment. Recently, however, scientists in various fields
have found that many systems are unpredictable despite the fact that they follow
courses prescribed by deterministic principles. The key to understanding how
systems may be both determinate and unpredictable – an oxymoron from the
point of view of classical science – lies in the notion of sensitive dependence
on initial conditions.
Predicting final outcomes – or indeed anything beyond the very short-term –
becomes impossible for many types of system. Chaos incorporates elements of
chance, but it is not random disorder. Rather, chaos theory tries to understand
the behavior of systems that do not unfold over time in a linearly predictable
manner. When viewed as a whole, these systems manifest definite patterns and
structures. However, because the evolution of a chaotic system is so hugely
complex and so prone to perturbation by contingent factors, it is impossible to
discern its underlying pattern – its attractor – by looking at a single small event
at a single point in time. At no single point can future directions be predicted
from past history.
So it is with the emergence of a new species in evolutionary change, with
changes in the political and social domain, and in grammar change. Change
is not random, but we are dealing with contingent systems and we offer
retrospective explanations, not predictions. Grammatical change is highly con-
tingent, sensitive to initial conditions, chaotic in a technical sense. Linguists
can offer satisfying explanations of change in some instances, but there is no
reason to expect to find a predictive theory of change, offering long-term, linear
predictions.
The emergence of a grammar in a child is sensitive to the initial conditions of
the primary linguistic data. If those data shift a little, there may be significant
consequences for the abstract system. A new system may be triggered, which
generates a very different set of sentences and structures. There is nothing
principled to be said about why the data should shift a little; those shifts often
represent chance, contingent factors. Contingent changes in the distribution of
the data (more accurately, changes in the “cues”: Lightfoot 1999) may trigger a
grammar which generates significantly different sentences and structures, and
that may have some domino effects, as we have seen.
Changes in languages often take place in clusters: apparently unrelated
superficial changes may occur simultaneously or in rapid sequence. Such clusters
manifest a single theoretical choice which has been taken divergently. The
singularity of the change can be explained by the appropriately defined
theoretical choice. The principles of UG and the definition of the cues
constitute the laws which guide change in grammars, defining the available
terrain. Any given phenomenal change is explained if we show, first, that the
linguistic environment has changed in such a way that some theoretical choice
has been taken differently (say, a change in the inflectional properties of verbs),
and, second, that the new phenomenon (may, must, etc. being categorized as I
elements, for example) must be the way that it is because of some principle of
the theory and the new inflectional system.
Sometimes we can explain domino effects of this type. Linguists have
argued that a changing stress pattern may leave word-final inflection mark-
ings vulnerable to neutralization and loss. Loss of inflectional markings may
have consequences for category membership and changes in category member-
ship may have consequences for computational operations moving verbs to an
I position. In that event, one establishes a link between a change in stress pat-
terns and changes in the positions of finite verbs. Benjamin Franklin would
understand: “For want of a nail, the shoe was lost; for want of a shoe the horse
was lost; for want of a horse, the rider was lost.” However, to say that there
may be domino effects is not to say that there is a general directionality of
the kind sought by nineteenth-century linguists and by modern typologists and
grammaticalizationists.
What we cannot explain, in general, is why the linguistic environment should
have changed in the first place (as emphasized by Lass 1997 and others).
Environmental changes are often due to what we have called chance factors,
effects of borrowing, changes in the frequency of forms, stylistic innovations,
which spread through a community and, where we are lucky, can sometimes be
documented by variation studies. Changes of this type need not reflect changes
in grammars. But with a theory of language acquisition which defines the range
of theoretical choices available to the child and specifies how the child may take
those choices, one can predict that a child will converge on a certain grammar
when exposed to certain environmental elements. This is where prediction is
possible, in principle. We thus have a determinist theory of language acquisition,
but not a determinist theory of history or of language change.
We have an interplay of chance and necessity, and appropriately so: changes
are due to chance in the sense that contingent factors influence a child’s PLD and
make the triggering experience somewhat different from what the child’s parent
was exposed to. Necessity factors, the principles of UG and the cues, define the
range of available options for the new grammar. We take a synchronic approach
to history. Historical change is a kind of finite-state Markov process, where
each state is influenced only by the immediately preceding state: changes have
8.5 Chaos 185
only local causes and, if there is no local cause, there is no change, regardless
of the state of the grammar or the language at some previous time.
In that way, the emergence of a grammar in an individual child is sensitive
to the initial conditions, to the details of the child’s experience. So language
change is chaotic, in a technical sense, in the same way that weather patterns
are chaotic. The historian’s explanations are based on available acquisition
theories, and in some cases our explanations are quite tight and satisfying.
Structural changes are interesting precisely because they have local causes.
Identifying structural changes and the conditions under which they took place
informs us about the conditions of language acquisition; we have indeed learned
things about properties of UG and about the nature of acquisition by the careful
examination of diachronic changes. Under this synchronic approach to change,
there are no principles of history; history is an epiphenomenon and time is
immaterial.
9 “Growing” a language
We have argued throughout this book that the cognitive system underlying a
person’s language capacity has intrinsic properties which are there by biolog-
ical endowment. Those properties interact with contingencies resulting from
exposure to a particular linguistic environment and the interaction yields a final
state in which the person may communicate, perhaps some form of French.
In that case, the person, Brigitte, will have incorporated from her environment
the contingent lexical properties that livre is a word to refer to the novel she is
reading and cooccurs with forms like le and bon (being “masculine”), père may
refer to her father. She has also incorporated contingent structural properties:
interrogative phrases like quel livre may be displaced to utterance-initial posi-
tion, verbs raise to a higher functional position, and so on. We have described
ways in which linguists have teased apart the intrinsic properties common to
the species and the contingent properties resulting from individual experience.
That work has been guided by the kind of poverty-of-stimulus arguments that
we have discussed, by theoretical notions of economy and elegance, and by the
specific phenomena manifested by the mature grammar under investigation.
Viewing a person’s language capacity in this way and focusing on what we
have called I-language leads one to ask novel questions about children and their
linguistic development. The perspective we have sketched has already led to
productive research and we have learned a great deal about the linguistic minds
of young children. In many ways, results concerning the attainment of syntactic
knowledge present the most dramatic examples of the bearing of this research
on our overall thesis, and we focus on this area for much of the present chapter.
Work in recent years has also shown remarkable things about the path by which
children attain knowledge of the expression system of their language, however,
and we turn briefly to those matters in section 9.5.
9.1 Principles of Universal Grammar: active early

If one postulates a principle of Universal Grammar (UG), one can ask whether
it is available to children from birth or whether it is subject to a biological
clock, becoming available only at a certain age, in the way that puberty or
186
baldness or death are genetically determined and subject to timing constraints.

In chapter 2 we discussed an experiment which showed that children manifest
the hypothesized constraints about clitic nature of the reduced ’s and that this
is true at the earliest stage where they can be tested.
The same holds for another reduction process, whereby want to is often
pronounced wanna: I want to be happy or I wanna be happy. Here, the to may
be attached to the preceding want and then reduced to give the wanna form;
but an intervening understood element blocks the reduction. For example, Who
do you want to see? has the structure of (9.1a) (corresponding to the statement
You want to see Jill) and it may be pronounced with the wanna form. Who is
understood as the direct object (complement) of the verb see, indicated here
by x.
(9.1) a. Whox do you want [IP I to see x]?
b. Whox do you want [IP x I to go]?
However, when an understood element intervenes between want and to,
wanna does not occur in the speech of most people. So an expression Who
do you want to go? has the structure of (9.1b), where who is understood in the
position x as the subject of the verb go (corresponding to the statement I want
Jill to go). Structures like (9.1b) do not yield a pronunciation where want to is
reduced to wanna.
The locality restriction discussed in chapter 3 also holds for the reduced
wanna form: to attaches to want to yield the reduced form only if it is adjacent
to and the top-most element in the complement of want. This is so in (9.1a) and
therefore want to may be reduced. However, it is not true in the sentences of
(9.2) and want to may not be reduced to wanna. We give some partial structure
in (9.2aii,bii), enough to show that the to clause is not the complement of want.
In (9.2aii) the IP to win games is not the complement of want; it acts as the
subject of the next IP, which, in turn, is the complement of want. Similarly, to
vote for the incumbent is not the complement of want in (9.2bii).
(9.2) a. i. They don’t want to win games to be their only goal.
ii. . . . want [IP [IP I to win games] I to be their only goal]
b. i. They expect people who continue to want to vote for the
incumbent.
ii. They expect [IP [NP people who continue to want] I to vote
for the incumbent]
A productive approach is to treat reduced to as a clitic, like the reduced is
discussed in chapter 2, and it may attach to a higher want. Part of what a child
growing a grammar needs to do is to determine the clitics in his or her linguistic
environment, knowing in advance of any experience that these are small, un-
stressed items left- or right-attached to an adjacent element in an appropriate
188 9 “Growing” a language
structural relationship, with no other (phonetic or “understood”) element inter-

vening. This predetermined knowledge is contributed by the linguistic genotype
and is what the child brings to language acquisition.
In this case too, experimenters have shown that the distinctions we have dis-
cussed do not result from learning and that the hypothesized genetic constraints
are operative from the outset. They constructed situations in which children
would be tempted to violate the relevant constraints. The fact that children con-
form to the hypothesized constraints, resisting the preferences for reduction
in some contexts, shows that they have the constraints under investigation and
they have them at the earliest stage where they might be manifested.
Stephen Crain and Rosalind Thornton developed an elicitation task that in-
vited children to ask questions like (9.1b), *Who do you wanna go?, if these
were compatible with their grammars. They guessed that children would gen-
erally prefer the reduced wanna form whenever this was consistent with their
grammars. This preference would be revealed in a frequency count of legiti-
mate forms, like (9.1a) Who do you wanna see? Comparing the frequency of
the reduced forms in these contexts with that of non-adult reduced forms would
indicate whether or not children’s grammars contained the hypothetical genetic
constraint. If the genetic constraint is operative, there should be a significant
difference in frequency; otherwise, not.
In the simplest case, an experimenter asked for a child’s help in finding
information about rats. Help was sought because the rat (a puppet) was too
timid to talk to grown-ups. The experimenter said “The rat looks hungry. I bet
he wants to eat something. Ask Ratty what he wants.” And the children, who
ranged in age from two years, ten months, to five years, five months, typically
would ask “What do you wanna/want to eat?” In this example, the wh-word is
understood as the complement of eat and the reduced form could occur freely,
as in (9.1a). In fact, the reduced form occurred 59 percent of the time in these
examples. So children do show a preference for the reduced form and that is
the extent of it.
Something very different happened when the wh-word had to be understood
in the subject position, as in (9.1b), Who do you want to go? The protocol for
this experiment was that the experimenter would ask: “There are three guys
in this story: Cookie Monster, a dog, and this baby. One of them gets to take
a walk, one gets to take a nap, and one gets to eat a cookie. And the rat gets
to choose who does each thing. So, one gets to take a walk, right? Ask Ratty
who he wants.” And the child would typically proceed: “Who do you want to
take a walk?” Here who is understood as the subject of take, i.e. between want
and to: whox you want [x to take a walk]? In these contexts the frequency of
the reduced form was quite different: the reduced forms occurred 4 percent
of the time. In fact, one child accounted for all three actual occurrences of
the reduced form, which suggests that this particular child had some other
analysis of wanna forms. So children prefer to use the reduced form in asking
questions like (9.1a), but correctly resist this preference when it conflicts with
UG principles. They use the reduced form in asking questions like (9.1a) but
not in questions like (9.1b), so they manifest the hypothetical genetic constraint
at a stage when their spontaneous production manifests very few instances of
long-distance wh-movement. The ingenuity of the experiment shows that even
at this stage the relevant principles are operating (Crain 1991).
The experiments we have described deal with elicited production, but com-
prehension studies also show that hypothetical genetic constraints are in effect
in very young children, at the earliest stage where they can be tested. Thornton
(1994) reported children’s comprehension of yes/no questions containing nega-
tion, such as (9.3). The difference between the two forms lies in the structural
position of the negative: in (9.3ai) the negative is inside the IP (partial structure
given in (9.3aii)) but in (9.3bi) it has formed a word with did and moved out of
the IP to C (9.3bii).
(9.3) a. i. Did any of the turtles not buy an apple?
ii. [CP did [IP any of the turtles not buy an apple]]
b. i. Didn’t any of the turtles buy an apple?
ii. [CP didn’t [IP any of the turtles buy an apple]]
The position of the negative corresponds to two distinct interpretations. That
correspondence between meaning and structural position follows from prin-
ciples of UG, which we need not go into here; essentially, a negative has an
effect on any element within its complement; logicians say that negatives have
sc o pe over certain elements. The phenomenon is clear. Suppose that turtles A
and B bought an apple but turtle C did not. Then if somebody asked question
(9.3ai), an appropriate answer would be that turtle C did not. If somebody asked
(9.3bi), then the appropriate answer would be very different: turtles A and B
did. So children’s responses to questions like (9.3ai,bi) reveal how they interpret
negatives. In particular, responses to (9.3bi) show whether children interpret the
negative in the higher structural position. This is worth testing because Thornton
found that all her children produced non-adult negative questions. Most dou-
bled the auxiliary verb (What do you don’t like?) and one failed to move the
auxiliary to the position of C: What you don’t like?
In testing comprehension, Thornton found that the children had no difficulty
interpreting negative questions in the adult fashion; significantly, all children
were able to access interpretations like (9.3bi), where the negative needs to be
interpreted in the position of C. She tested children between the ages of three
and a half and four and a half. The comprehension test used a modified form
of the Truth Value Judgement task (Crain 1991). A story was acted out by one
experimenter and watched by the child and a second experimenter, who was
playing the role of a puppet, in this case “Snail.” At the end of each story,
the experimenter asked Snail a targeted question. Snail had difficulty with the
question (“That’s a hard one . . .”), and requested help from the child. If the
child was cooperative, she answered the question for Snail.1 The scenarios used
to test children’s comprehension of questions like (9.3ai) and (9.3bi) were de-
signed so that either (9.3ai) or (9.3bi) could be asked appropriately; children’s
answers, however, indicate their analysis of the structural position of the neg-
ative. Thornton found that, while these children made production errors with
expressions like adult What don’t you like?, their comprehension was adult-like
and manifested the UG principles which determine the scope of negatives.
So there is a clear production/comprehension asymmetry, which, of course, is
no surprise under the modular view of mind that we have articulated. Whatever
it is that causes the delay in producing the adult forms, the fact that children
interpret the negative questions in adult fashion shows that they have access
to whatever principles of UG assign scope relations. The difficulty evidently
lies with the behavior of the element n’t: children produce non-adult questions
which retain the n’t in the IP until they figure out that n’t may form part of a
word with did 2 and move with it outside the IP to C.
Evidence suggests, then, that at least some principles of UG are available
and operative at the earliest stage where they might be tested. The same might
turn out to be true of all UG principles. But perhaps not. Some principles of UG
may turn out to be subject to a puberty-like clock, emerging only at a certain
age in a kind of “maturation.” There is nothing implausible about that view and
we know that it holds of some genetic properties in the physiological domain;
so why not also in the cognitive domain?
In fact, the case that grammars mature in this fashion has been made. Some
have argued that very young children have “semantically based” grammars and
that they graduate to “syntactically based” grammars at a certain age just as
tadpoles develop into frogs (Bowerman 1973; for discussion, see deVilliers and
deVilliers 1985 and Gleitman and Wanner 1982). Others have argued that prin-
ciples of the binding theory are not operative until they mature in the organism
(Borer and Wexler 1987, Manzini and Wexler 1987). A little later we shall con-
sider a claim that the principles of the binding theory are based on notions of
linear precedence in very young children, later becoming structurally based. We
do not find these maturation claims convincing and will not discuss them here,
but there is nothing implausible in the general idea of maturation of genetic
properties.
1 The child was not asked the question directly, in order to alleviate any feeling she might have
of being tested; in this setup, Snail is being quizzed, not the child. Here we are giving only the
briefest description of the experiments, but the experimental techniques used for this kind of work
require great ingenuity and are of enormous interest in themselves. For excellent discussion, see
Crain and Thornton 1998.
2 The most common account of this is to say that n’t is a clitic, and attaches to did by cliticization.
Zwicky and Pullum 1983, however, show that words like didn’t actually represent inflected forms
which exist only for a limited set of verbs, rather than cliticization. The difference between these
two analyses does not bear on the point under consideration here, though.
9.2 New phenomena

The last experiment (dealing with the interpretation of (9.3)) illustrates the ob-
vious fact that young children are not entirely similar to adults in their syntactic
behavior, even if they seem to manifest at the earliest stage possible the hypo-
thesized genetic constraints that we have discussed. In another study, Thornton
analyzed an intriguing type of non-adult speech. She observed long-distance
medial wh-questions like (9.4) in some three- and four-year-old children, where
a copy of the wh-word shows up at the front of the clause where it is understood;
such sentences do not occur in adult English, but analogues to them do occur
in dialects of German and in other languages (Thornton 1995).
(9.4) a. What do you think what pigs eat?
b. Who do you think who eats trash?
Such phenomena had not been previously observed in the literature on child
language acquisition but Thornton was struck by them in light of theoretical
ideas about the displacement of wh-phrases. Recall that in chapter 3, where we
discussed the derivation of sentences like What city will the student visit?, we
noted that wh-phrases move to the front of an utterance. We noted (footnote 4
in chapter 3) that in a complex expression like Who did Jay say that Fay saw?
(3.6b), movement is always local and the wh-phrase is copied first at the front
of its own clause, then at the front of the next clause, and so on, as in (9.5).
(9.5) Whoi did Jay say [CP ei that Fay saw ei ]?
We know that movement is possible in an embedded clause (I know who Fay

saw, which has the structure I know [CP whox Fay saw x]) and postulating that
the movement is always local enables us to distinguish the well-formed Who
did you believe that Fay claimed that Ray saw? from the ill-formed *What
did you believe Fay’s claim that Ray saw?3 Theorists postulated that move-
ment is always local and successively cyclic, and Thornton observed children
using overt wh-phrases in positions where they are always deleted in adult
speech (9.4).
Not only did Thornton introduce expressions like (9.4) into the acquisition
literature but she also noted further distinctions. After children stop using medial
wh-words corresponding to a deleted complement (9.4a), they persist with them
where they correspond to a deleted subject (9.4b). That developmental sequence
illustrates principles of UG: the extra wh-word in subject extraction questions is
3 The distinction is a function of a Subjacency condition which blocks movement across more than
one bounding node (typically IP and DP in English grammars) and the claim that a wh-phrase
moves to the specifier of CP (see chapter 3).
an overt manifestation of the special “agreement” properties needed to license

an empty subject, as we discussed in chapter 3. Recall that a copied wh-phrase
may not be deleted in the subject position of an embedded clause, unless it is
licensed in some special, language-specific way (see the discussion following
(3.18)); no such inhibitions constrain the deletion of a complement. That means
that more has to be acquired from the environment in order for a person to delete
a wh-phrase which is the subject of an embedded clause, and that explains why
it takes a little longer.
There is a more primitive point to be made in this connection, primitive
but of crucial importance: some children systematically produce things like
(9.4), even though they never hear such sentences uttered by adult speakers
of English. Children do not simply imitate models; they develop a system, a
grammar, which at certain stages of development yields things which no child
hears from English-speaking adults. When we study the development of these
grammars (language organs), often we can understand the properties they have,
even when they do not reflect what children have heard.
The sentences which children produce (e.g., (9.4)) make sense. They reflect
properties of their language organs as we understand them, but they do not
reflect what they hear from adults. In fact, the idea that children imitate adult
speech would provide very little understanding of the stages that children go
through and the kinds of things they say. And that idea cannot survive close
observation of the things children actually do say, in comparison with the speech
of the models available to them.
Another illustration of how grammatical theory has served as a diagnostic
tool for acquisitionists, yielding discoveries about children’s capacities, comes
from the domain of so-called optional infinitives (Wexler 1994). We noted in
chapter 8 that modal auxiliaries like can, may, etc. are generated as instances
of Inflection (I) (9.6a) and may be displaced, moving to a higher functional
position, as in (9.6b).
(9.6) a. Kim [I can] [VP understand the chapter].
b. [I Can] [IP Kim understand the chapter]?
In most European languages other than (modern) English, ordinary verbs may
also move to the I position and then still further to a higher functional position.
So a French sentence like Elle lit les journaux has the structure of (9.7a), where
the verb lit has moved to the I position, as is clear in the corresponding negative
(9.7b), where the finite lit has moved across the intervening pas (cf. Ne pas
lire les journaux . . . , where the non-finite lire stays in its original position to
the right of pas). Lit-elle les journaux? has the structure of (9.7c). Verb-second
languages like Dutch and German are often analyzed with the same movement:
Met Hans bezoekt zij Amsterdam “she is visiting Amsterdam with Hans” would
have the structure of (9.8), where the finite verb bezoekt moves first to I and
then to a higher position outside IP (Dutch has a head-final system, in which

verbs and inflection elements occur at the right edge of their phrases).
(9.7) a. Elle [I liti ] [VP ei les journaux].

b. Elle ne [I liti ] pas [VP ei les journaux].
c. Liti [IP elle ei [VP ei les journaux]].
(9.8) a. Met Hans bezoekti [IP zij [VP Amsterdam ei ] ei ].
In general, as noted in chapter 3, movement takes place in order to check
features. So lit and bezoekt have finite tense markers, which are checked in the
I position. They may also have a feature which requires checking in a higher
position, as in (9.7c) and (9.8).
David Poeppel and Ken Wexler (1993) noted that children acquiring gram-
mars like these seem to go through an “optional infinitive stage”: not only do
they use adult-style finite verbs but they also use non-finite verbs in unem-
bedded clauses, unlike adults. However, in a language like German or Dutch,
the alternation is not random: the form of the verb correlates with its posi-
tion in the clause. Finite verbs systematically appear in second position, as in
(9.8), and non-finite verbs systematically occur clause-finally, in their original,
base-generated position, showing up in infinitive forms (9.9).
(9.9) a. Thorsten Caesar haben.

Thorsten C (=doll have
Thorsten has (the doll) Caesar.
b. Du das haben.
you that have
You have that.
Mieke Weverink (1989) supplied comparable examples from child Dutch

(9.10), although no adult would produce such forms.
(9.10) a. Pappa schoenen wassen.

Daddy shoes wash
Daddy washes shoes.
b. Pappa nieuwe scooter kopen.
Daddy new scooter buy
Daddy buys a new scooter.
c. Ik ook lezen.
I also read
I’m reading too.
d. Pappa kranten weg doen
Daddy newspapers away do
Daddy gets rid of the newspapers.
At this optional infinitive stage, children know that finite verbs may not oc-
cur (more technically, may not be “checked”) in clause-final position in matrix
clauses, because they do not produce clause-final inflected verbs: *Pappa schoe-
nen wast, *Pappa nieuwe scooter koopt. And they know that non-finite verbs
may not occur in an inflectional position and do not produce forms like *Pappa
wassen schoenen or *Ik lezen ook.
Comparable phenomena are found in young French children (Pierce 1992).
They alternate between the non-finite, non-adult forms of (9.11), where the verb
stays in its first, base-generated position inside the VP and to the right of the
negative marker pas, and the finite forms of (9.12), where the verb is finite and
therefore occurs in the I position where its finite features are checked, to the
left of pas.
(9.11) a. Pas [VP manger la poupée].

Not eat the doll.
b. Pas [VP tomber bébé].
Not fall baby.
c. Pas [VP attraper une fleur].
Not catch a flower.
(9.12) a. Patsy est pas là-bas.
Patsy is not down there.
b. Marche pas.
Walks not.
c. Est pas mort.
Is not dead.
d. Trouve pas.
Finds not.
There is now an extensive literature on optional infinitives, including in chil-

dren with Specific Language Impairment (SLI) (Wexler, Rice, and Cleave 1995),
drawn from several languages. Much has been discovered about child language.
These discoveries show that while the children produce non-adult forms, which
they have not heard from their parents or older siblings, they nonetheless know
some very abstract principles of grammar, including head movement and the
properties of inflection which require head movement.
9.3 Experimental technique

Not only have we learned much about child language and found that chil-
dren are subject to abstract principles of UG from the earliest testable stages,
but we have also discovered a lot about how to analyze children’s capacities,
about experimental technique. In chapter 2, we noted that sometimes a pronoun
may precede its antecedent and sometimes not. This is the phenomenon of
“backwards anaphora.” So in (9.13a) (repeated from chapter 2), he may refer
to Jay, but not in (9.13b). In both instances, he precedes Jay, but in (9.13b) he is
also structurally higher than Jay in a sense that we shall not make precise here.
It is Principle C of the binding theory which prevents Jay being coindexed with
he in (9.13b), allowing it in (9.13a).4
(9.13) a. When he entered the room, Jay was wearing a yellow shirt.
b. He was wearing a yellow shirt, when Jay entered the room.
Given what we said in the last section, we would expect children to conform
to Principle C and to produce and understand sentences like (9.13) in the adult
fashion. However, that is not what the early literature suggests. Larry Solan
(1983) and Susan Tavakolian (1978) discussed experiments with children acting
out sentences like (9.13), when provided with suitable toys.5 The children were
three- to eight-year-olds; not so young. For adults, (9.13a) may be interpreted as
referring to two men, Jay and somebody unnamed; (9.13b) must be interpreted
that way, involving two men. That interpretation is also open to children and
the act-out studies found that children interpreted both types of sentences that
way most commonly: two thirds of the time, in fact. This led to the conclusion
that backwards anaphora does not exist in young children, that the conditions
on coreference are purely linear and not structural in the early stages. Put
differently, children do not permit the interpretation allowed by Principle C
and are not subject to Principle C but rather only to a linear condition that a
pronoun may never precede a noun to which it refers. However, the conclusion
was premature.
First, one third of the responses, in fact, permitted backwards anaphora,
with a pronoun referring to a noun to its right. Second, even if all responses
disallowed backwards anaphora, that would not show that children were not
subject to Principle C: perhaps they are only displaying a strong preference to
have the pronoun refer to some second person unnamed in the sentence. To test
adherence to Principle C one needs an experiment which shows that children
sometimes allow backwards anaphora and that they reject it in the appropriate
circumstances. Or not.
4 The binding theory, one of the more resilient aspects of grammatical theory, divides nouns into
three types: anaphors, pronouns, and everything else (sometimes called “names”). Anaphors are
subject to Principle A and must be locally coindexed with a “higher” element (the technical
notion is a “c-commanding” item). A pronoun must not be locally coindexed with a higher
element, by Principle B. And Principle C requires that names not be coindexed with a higher
element anywhere. See Chomsky 1981. In (9.13b) he is higher than Jay in the relevant sense, and
therefore they may not be coindexed (Principle C); in (9.13a), on the other hand, he is contained
in a subordinate clause and is not higher than Jay and therefore Principle C is irrelevant.
5 The actual sentences used included For him to kiss the lion would make the duck happy, That he
kissed the lion made the duck happy.
Crain and McKee (1986) constructed a Truth Value Judgement task, of the
kind we saw earlier. Children were exposed to sentences like (9.14) (analogous
to (9.13)), in situations where the pronoun was backwards anaphoric and in
situations where the pronoun had another referent not mentioned in the sentence.
They judged the truth value of the sentences.
(9.14) a. While he was dancing, the Ninja Turtle ate pizza.

b. He was dancing while the Ninja Turtle ate pizza.
Sentence (9.14a) was presented twice, once where the Ninja Turtle was danc-
ing and eating pizza and once where somebody else was dancing while the
Ninja Turtle was eating pizza. Similar conditions attended the presentation of
(9.14b), although no adult would use such a sentence in a situation in which the
Ninja Turtle was dancing and eating pizza. For each scene, Kermit the Frog said
what he thought happened in that trial, using sentences like (9.14a) or (9.14b). If
Kermit said something appropriate, children could feed him something he liked;
if Kermit said the wrong thing, children could get Kermit to eat something
“yucky,” like a rag or a cockroach, or to do push-ups. There is much to be said
about the design of the experiment, but the results clearly showed that children
correctly accepted the backwards anaphoric reading in sentences like (9.14a)
about two thirds of the time. In addition, 90 percent of the time sentences like
(9.14b) were correctly judged to be wrong in contexts displaying coreference.
Thus even two- and three-year-olds allow backwards anaphora and reject it
when structural conditions dictate that they should. Kermit ate some pretty
unpleasant things as a consequence of the fact that these children behaved in
accordance with Principle C.
The construction of a good experiment is far from trivial. We cannot sim-
ply observe children’s spontaneous expressions, because it might be several
years before a child is confronted with a situation which draws on her know-
ledge of Principle C in a way that can be measured. Crain and Thornton (1998,
chs. 27–30) provide an excellent discussion of the design features of experi-
ments relating to Principle C.
The literature is full of experiments showing that children do not know such-
and-such at some early age, only coming to acquire some principle of UG
at a later age. Since children’s language differs from that of adults, one can
easily allow extraneous factors to distort findings, as in the early experiments
on Principle C. Confounding factors have led many researchers to conclude too
hastily that children do not know certain principles of UG. In fact, we know of
no good demonstration to this effect.
Consider another example. Hornstein and Lightfoot 1981 argued that children
know in advance of any experience that phrase structure is hierarchical and not
flat. So the structure of an expression the second striped ball would be that of
(9.15a) and the flat structure of (9.15b) would not be a candidate, precluded
by UG.6
(9.15) a. DP
D NP
A NP
A NP
the second striped ball
b. DP
D A A N
the second striped ball
There were various reasons for this analysis and for the claim that flat struc-
tures are generally unavailable. In particular Hornstein and Lightfoot argued that
if children allowed a flat structure they would never be exposed to positive data
enabling them to acquire the correct hierarchical structure of adult language.
Nonetheless, it was argued, for example by Matthei (1982), that children must
operate with flat structures, because they have difficulty interpreting phrases
like the second striped ball. Confronted with an array like that in figure 9.1
and asked to identify “the second striped ball,” adults invariably identify the
third ball but children often identify the second ball, which is also striped.
Figure 9.1 An array of objects
6 Hornstein and Lightfoot stated their claims in terms of then-current notions of phrase structure,
in which NP included determiners and intermediate N elements. Twenty years later, we translate
their claims into present-day frameworks using DP, etc.
Hamburger and Crain 1984 then showed that the difficulty that children have
in identifying “the second striped ball” in an array does not relate to their
grammars; in fact, it cannot. They hypothesized that the difficulty might lie in
the pragmatic complexity of identifying “the second striped ball” rather than
in syntactic complexity. A dramatic improvement in children’s responses was
effected by two changes. First, children were given a pre-test session where they
handled and counted sets of striped and unstriped balls. Second, if children were
first asked to identify the first striped ball, forcing them to plan and execute part
of what is involved in identifying the second striped ball, they then performed
much better when asked to identify the second striped ball. As Crain put it,
“these simplifying maneuvers made it possible for children to reveal mastery
of the syntax and semantics of such expressions” (1991, p. 609).
Furthermore, Hornstein and Lightfoot had shown that the pronoun one refers
to an NP and not a noun head (see note 6). Hamburger and Crain noted that if
children employ such hierarchical structures, then, once they know the meaning
of ordinals, they will behave in adult fashion when confronted with an array like
that of figure 9.2 and asked “Point to the first striped ball; point to the second
one,” using the pronoun one. In fact, children pointed consistently to the fifth
object in the array. Using one in this way indicates hierarchical structures like
(9.15a) and is quite incompatible with the flat structure hypothesis ((9.15b)), if,
as is generally assumed, pronouns corefer with syntactic constituents (because
second striped ball is not a constituent in (9.15b); consequently children would
be expected to identify the second ball, not the second striped ball).
Figure 9.2 Another array of objects
Here we have shown more instances where we can see that the language of
young children manifests properties of UG at the earliest stages that we can
test . . . if we do the experiments properly and are careful to tease out their syn-
tactic capacities, designing the experiment to exclude extraneous confounding
factors. As we learn more about experimental technique, so we shall learn more
about children’s linguistic capacities, and vice versa.
9.4 Nature of the trigger

In chapter 2, our analytical triplet had children endowed with UG, which de-
velops into a mature grammar on exposure to primary linguistic data (9.16).
Let us turn now to children’s trigger experience and ask what it consists in,
and we shall find that I-language notions are needed in order to define even
children’s initial experience, the primary linguistic data (PLD) that trigger the
development of a mature grammar.
(9.16) primary linguistic data (Universal Grammar → grammar)
The primary linguistic data are positive and robust. That is, they consist of
actual expressions which occur frequently enough for any child to hear them.
As we observed in chapter 2, a rough-and-ready formulation would say that the
PLD do not include negative data, information about what does not occur, nor
exotic expressions, nor expressions used in exotic circumstances. Nor do they
include paraphrase relations or indications of the scope of quantifiers. These
are legitimate data, of course, part of what a grammar should characterize, part
of the output of the emerging system, but plausibly not part of the input to
language acquisition and so not “primary.” These seem to be plausible assump-
tions about the input, but the proof will come if they support successful models
of the form of (9.16), yielding optimal claims about UG and about mature
grammars.
One might also argue that the PLD are structurally simple, that children do not
need to hear complex expressions in order to develop a normal, mature grammar.
Children do hear complex expressions, of course, and they may understand
them, but they may not be a necessary part of experience. In that event, the
question arises what one means by “simple.”
Lightfoot (1991, 1994) has argued that children need exposure to simple,
unembedded clauses and the front of an embedded clause, but not to any-
thing more complex than that. This is degree-0 learnability, the idea that
grammars are learnable by exposure only to unembedded structures. The rel-
evant “unembedded” structures are defined in terms of the I-language, gram-
matical notion of a binding domain. So the PLD are drawn from unembedded
binding domains.
Children need to learn, for example, which verbs are transitive and which are
intransitive, and therefore need access to VPs which may or may not include
a complement. Not everything, however, can be learned from simple, unem-
bedded clauses. English speakers learn that the complementizer that may be
omitted from the front of an embedded finite clause, unlike French que or Dutch
dat, and that must require exposure to the front of an embedded clause, where
sometimes that occurs and sometimes it does not. Similarly some verbs require
that their complement clause be topped by a wh-item (I wonder who she saw)
and others do not: I believe that she saw Reuben. This can only be learned,
it would seem, by exposure to the front of an embedded clause. Furthermore,
in English, some verbs allow a lexical subject in their infinitival complement,
while their counterparts in other languages do not: I expect/want her to see
Reuben. This fact about English verbs also can only be learned if children have
access to embedded clauses.
Languages differ in the tenses of reported speech, the so-called sequence-

of-tenses phenomenon. For example, if somebody says Yesterday Fay said she
was happy, Fay’s words might have been “I am happy” or “I was happy.” In
other languages the tense of the embedded clause manifests directly the tense
of the direct speech. In these languages, if Fay said “I am happy,” that would
be reported as Fay said that she is happy. Again, it is hard to imagine how
differences like this could be learned unless children have access to embedded
clauses.
This suggests that children need access to simple, unembedded clauses and
to the top of an embedded clause and to the highest inflectional position, but
that may be enough. These are the positions for which the binding domain
is not the immediate CP containing them, but the next CP up, which may be
unembedded.7 Lightfoot argued that children do not need access to structures
more complex than this. Nothing about the mature grammar is to be learned
from the fact that an embedded verb has a complement DP or that a relative
clause has a certain tense.
Restricting the trigger in this way is a strong claim which runs against analyses
in the literature. We will sketch three examples briefly. First, it was once argued
that the verb serve selected a complement containing a transitive verb ((9.17a)
vs. (9.17b)). If that were the right generalization, then children would need
access to embedded verbs in order to learn it.
(9.17) a. *The ice served to melt.
b. The ice served to chill the beer.
c. *Edison served to invent the light bulb.
However, that is not the right generalization, as can be seen by the ill-
formedness of (9.17c), where the embedded verb is transitive. (9.17c) is ill-
formed because it violates what is, in fact, the correct generalization, which
is that the subject of serve must be an instrument, hence inanimate. Not only
is the subject of serve an instrument, but it also acts as the understood subject
of the embedded verb (melt, chill, invent). The correct generalization, that the
subject of serve is an instrument, is learnable by exposure just to simple, un-
embedded clauses. All verbs with instrumental subjects are transitive (hence
the ill-formedness of (9.17a)), but that is a deep fact which does not need to be
learned, holding quite generally.
Second, several linguists (beginning with Rizzi 1982) had argued that the
well-formedness of (9.18) in French and Italian indicates that IP is not a
7 In note 4, we observed that an anaphor must be locally coindexed with a higher element. Com-
parably, pronouns are subject to the same local requirement. “Locally” means within its binding
domain. An element’s binding domain is usually the clause (CP) or DP which contains it. If the
element is in a high structural position (namely in C, Spec of CP, or the subject of a non-finite IP),
then its binding domain is the next CP up. See Chomsky 1981 for the details.
bounding node in those grammars (see note 3), unlike in the grammars of
English speakers. Furthermore, this fact could be learned on exposure to com-
plex sentences like those of (9.18), where there are two levels of embedding.
(9.18) a. Tuo fratello [CP a cui mi domando [CP che storie abbiano
raccontato]] era molto preoccupato.
Your brother, to whom I wonder which stories they have
told, is very troubled.
b. C’est à mon cousin [CP que je sais [CP lequel offrir]].
It is to my cousin that I know which to offer.
However, Lightfoot argued that the relevant conclusion (that IP is not a

bounding node) could be drawn from simpler structures like those of (9.19).
There combien (9.19a) and ne (9.19b) move across a DP and IP, as indicated
(they are understood in the position marked ei ), and therefore (assuming that
DP is always a bounding node) IP cannot be a bounding node in grammars
allowing structures like these and the well-formedness of (9.18) follows.
(9.19) a. Combieni as [IP tu vu [DP ei de personnes]]?

How many have you seen people?
b. Nei ho visti [IP [DP molti ei ] corregli incontro].
Of-them I saw many run towards him.
Third, degree-0 learnability would preclude a generalization that a certain

verb is deleted only in embedded clauses, as Andersson and Dahl 1974 argued
for the Swedish auxiliary verb ha. Ha can be deleted in (9.20c) in an embedded
clause, but it appears not to be deletable in a main clause (9.20a,b).
(9.20) a. Han hade sett henne.

He had seen her.
b. *Han e sett henne.
c. . . . att han (hade) sett henne.
. . . that he had seen her.
d. Allan kanske redan (har) skrivit sin bok.
Allan perhaps already has written his book.
Lightfoot showed that Andersson and Dahl’s generalization was incorrect and
that ha may be deleted in a matrix clause if the second position to which it
would ordinarily move is already filled by an adverb like kanske “perhaps,” as
in (9.20d). This suggests that the correct generalization is that ha may be deleted
quite generally, but in fact it fails to be deleted when moved to C (Swedish is a
verb-second language, like Dutch and German, in which finite verbs typically
move to C in matrix clauses). The restriction that it may not be deleted in C
may then be understood in terms of the binding theory: if ha were deleted in
that position, its deleted copy (trace) would not be bound – it would lack an
antecedent. So ha may be deleted only in its original, base-generated position,
where it is not needed as an antecedent for a deleted copy.
This would also explain the non-deletability of a moved do or modal auxiliary
in English, as in (9.21); compare the non-moved can in (9.21c), which may be
deleted. Under this analysis, the Swedish facts are not as peculiar as one might
have thought and there are no special conditions to be learned, and nothing that
requires children to learn anything from embedded binding domains.
(9.21) a. *Who did Jay greet and who Ray treat?
b. *Who can Jay visit and who Ray eat with?
c. Jay can visit Fay and Ray eat with Kay.
More productive syntax of this type followed, where better analyses were
found by following through on the assumption that learning is based only on
structures from unembedded binding domains. What is relevant here is that the
notion of a binding domain is itself a grammatical, I-language notion, and that
is what we need to define the limits to children’s trigger experience. I-language
notions are implicated in analyses of language acquisition from the outset.
In fact, there is a still more fundamental point: not only must the PLD be drawn
from simple structures, but they are abstract structures themselves, not unana-
lyzed sentences. This is so-called cue-based acquisition, which we shall discuss
below, but first let us consider other, more E-language-based approaches.
Chomsky’s Aspects of the theory of syntax (1965), now a classic, viewed
children as endowed with a metric evaluating the grammars which can generate
the primary data to which they are exposed, along with appropriate structural
descriptions for those data. The evaluation metric picks the grammar which
conforms to the principles of UG and is most successful in generating those
data and those structural descriptions. The child selects a grammar whose output
matches her input as closely as possible.
The same point holds for more recent models. Gibson and Wexler (1994)
posit a Triggering Learning Algorithm (TLA), under which the child–learner
uses grammars to analyze incoming sentences and eventually converges on the
correct grammar. If the child–learner cannot analyze a given sentence with the
current grammar, then she follows a procedure to change one of the current
parameter settings and then tries to reprocess the sentence using the new para-
meter values. If analysis is now possible, then the new parameter value is
adopted, at least for a while. So the TLA is error-driven and permits the child
to reset a parameter when the current grammar does not give the right results.
This model has the child seeking grammars which permit analysis of incoming
data, where the data consist of more or less unanalyzed sentences.
Clark (1992) offers a similar kind of model, but one which differs from that
of Gibson and Wexler in that the child does not revise particular parameter
settings. Clark posits a Darwinian competition between grammars needed to
parse sets of sentences. All grammars allowed by UG are available to each child,
and some grammars are used more than others in parsing what the child hears.
A “genetic algorithm” picks those grammars whose elements are activated most
often. A Fitness Metric compares how well each grammar fares, and the fittest
grammars go on to reproduce in the next generation, while the least fit die out.
Eventually the candidate grammars are narrowed to the most fit, and the child
converges on the correct grammar.
What these models have in common is that learners eventually match their
input, in the sense that they select grammars which generate the sentences of
the input. It is only accurate grammars of this type which are submitted to
Chomsky’s (1965) evaluation metric, and Gibson and Wexler’s error-driven
children react to inaccurate grammars by seeking new parameter settings until
a sufficient degree of accuracy is achieved. The child converges on a gram-
mar which analyzes the input successfully, where the input consists of sets of
sentences.
There are problems with these input-matching, E-language-based ap-
proaches. First, the models will need great elaboration to deal with the fact,
observed several times in this chapter, that children produce non-adult forms.
That is, they operate with inaccurate grammars which do not match the input.
The models will need to explain why certain inaccuracies are tolerated and
others not.
Second, the models require extensive appeals to memory, because children
resetting a parameter need to know the full set of sentences which required
earlier resettings, lest they now be lost by picking the wrong parameter to reset.
Third, it is hard to see how these input-matching models can succeed when
children are exposed to unusual amounts of artificial or degenerate data, which
in fact are not matched. In particular, it is hard to see how they could ac-
count for the early development of creole languages, as described by Bickerton
(1984, 1999) and others. In these descriptions, early creole speakers are not
matching their input, which typically consists to a large degree of pidgin data.
Pidgins are primitive communication systems, cobbled together from fragments
of two or more languages. They are not themselves natural languages, and they
tend not to last long, before giving way to a creole with all the hallmarks of a
natural grammar. The first speakers of creoles go far beyond their input in some
ways, and in other ways fail to reproduce what they heard from their models,
arriving at grammars which generate sentences and structural descriptions quite
different from those of their input.
Nowadays we can observe these effects in the development of deaf children
acquiring various kinds of signing grammar. 90 percent of deaf children are
born into hearing homes and are exposed initially to degenerate pidgin-like data,
as their parents and older siblings learn an artificial gestural system in order
to communicate in primitive fashion. Nonetheless, like early creole speakers,
these deaf children go far beyond their models and develop natural systems.
Goldin-Meadow and Mylander (1990) show how deaf children go beyond their
models in such circumstances and “naturalize” the system, altering the code and
inventing new forms which are more consistent with what one finds in natural
languages.8
Fourth, severe feasibility problems arise if acquisition proceeds by evaluating
the capacity of grammars to generate sets of sentences. If UG provides binary
parameters of variation, then 11 parameters permit 4,096 grammars, and 32
allow 8-9 billion grammars, and so on. The child eliminates grammars which
fail to generate the sentences encountered. If we assume thirty to forty structural
parameters and if children converge on a grammar by, say, age seven, then they
eliminate grammars at a fantastic rate, several in each waking second on average.
As if that problem is not enough, it is by no means clear that parameters can be
kept down to between thirty and forty. If there are only thirty to forty structural
parameters, then they must look very different from present proposals.
Cue-based acquisition offers a very different, more I-language-based ap-
proach. Under this view, advocated by Dresher 1999, Fodor 1998, Lightfoot
1999, and others, children do not evaluate grammars against sets of sentences.
Rather, UG specifies a set of “cues” and children scan their linguistic environ-
ment for these cues and converge on a grammar accordingly. A cue is some
kind of structure, an element of grammar, which is derived from the input. The
cues are to be found in the mental representations which result from hearing,
understanding, and “parsing” utterances (“parsing” means assigning structure
and meaning to incoming speech signals). As a person understands an utterance,
even partially, he or she has some kind of mental representation of the utterance.
Similarly for children, but a child may only have a partial understanding of what
is heard, hence a partial parse. The child scans those mental representations,
derived from the input, and seeks the designated cues.
The child scans the linguistic environment for cues only in simple syntactic
domains (this is the degree-0 learnability just discussed). Learners do not try
to match the input; rather, they seek certain abstract structures derived from
the input, and this shapes the emerging grammar without regard to the final
result. That is, a child seeks cues and may or may not find them, regardless
of the sentences that the emerging grammar can generate; the output of the
grammar is entirely a by-product of the cues the child finds, and the success of
the grammar is in no way evaluated on the basis of the set of sentences that it
generates, unlike in the input-matching models.
So, for example, a child scans her environment for nouns and determiners.
She would find the nouns livre, idée, and vin and the determiners le, la, and mon,
if she lives in Paris; she finds book, idea, and wine, if she lives in Boston, and the
determiners the, that, and my. Our Bostonian would also find a determiner ’s,
8 Newport (1999) extends these ideas and Kegl, Senghas, and Coppola (1999) report on the spec-
tacular emergence of Nicaraguan Sign Language over the last twenty years.
which has no counterpart in French. She would find that this determiner
also assigns Case to a preceding DP; she would discover this on exposure
to an expression the player’s hat, analyzed partially as [DP [DP the player] D ’s
[NP hat]]. This partial analysis is possible, of course, only after the child has
identified player and hat as separate words, both nouns projecting an NP, etc. In
this way, the order in which cues are identified, the “learning path” (Lightfoot
1989), follows from dependencies among cues and follows from their internal
architecture.
Our Parisian would also find the cue I V, that is, instances of verbs occurring
in an inflectional position (cf. chapter 8). She would find this on exposure to
sentences like Elle lit toujours les journaux and understanding that les journaux
is the complement of the finite verb lit. In that case, since verbs are first generated
adjacent to their complement, there must be a partial analysis of (9.22a), which
represents the movement of lit out of the verb phrase, across the adverb toujours,
to the higher inflectional position. She would also find the cue when confronted
with and partially understanding an expression Lit-elle les journaux?, which
requires the partial analysis of (9.22b).
(9.22) a. Elle [I V liti ] [VP toujours V ei les journaux].
b. V Liti [IP elle I ei [VP V ei les journaux]]?
Our Bostonian, on the other hand, would not be confronted with any such
expressions and would never be exposed to sentences which required postulating
a verb raised to a higher inflectional position. Since she never finds the cue I V,
her grammar would never have such a verb-raising operation.
So children scan the environment for instances of I V. This presupposes prior
analysis: children may scan for this cue only after they have identified a class of
verbs and when their grammars have a distinct inflectional position, I. The cue
must be represented robustly in the PLD. The approach is entirely I-language
based and children do not test or evaluate grammars against sets of sentences; in
fact, the set of sentences generated by the emerging grammar is quite irrelevant –
the chips fall where they fall.
Cue-based acquisition finesses the feasibility problems which arise for input-
matching models. We are free to postulate 100 or 200 cues, if that is what analy-
ses of different grammars require. That does not raise comparable feasibility
problems for the child learner. Our child would not be evaluating quadrillions
of grammars against sets of sentences, rejecting hundreds every waking second.
Rather, the child would be scanning her environment for the 100 or 200 cues,
much in the way that she scans her environment and identifies irregular past-
tense verbs like took, went, fit, spoke, and so on. That task may raise difficulties
that we do not now understand, but it does not raise the particular, devastating
feasibility problems of input-matching parametric systems.
The cue-based approach has been productive for phonologists concerned
with the parameters of stress systems (Dresher 1999) and it comports well with
work on the visual system, which develops as organisms are exposed to very
specific visual stimuli, such as horizontal lines (Hubel 1978, Hubel and Wiesel
1962, Sperry 1968). Current theories of the immune system are similar; specific
antigens amplify preexisting antibodies. In fact, this kind of thing is typical of
selective learning quite generally.
9.5 Acquiring sound patterns

Our discussion thus far in this chapter has been limited to the development of
syntax, but of course the child also comes to control the sound pattern of the
native language. Just as recent work on the acquisition of syntax makes it clear
that, once we do the experiments correctly, children can be seen to have sub-
stantial elements of mature grammars earlier than had been thought, so current
work makes it clear that the phonological system of the language organ arises
rather earlier in the child’s life than had long been thought, and in fact pro-
vides essential prerequisites for the development of syntactic knowledge. The
sequence of events in the course of the acquisition of phonology is quite regular
and shows every evidence of being driven by the organism’s biological nature.
This impression is confirmed by the parallels with another, quite undeniably
biological developmental sequence: the emergence of song in oscine songbirds
(one of the twenty-seven orders of living birds). While syntactic organization is
apparently uniquely human, phonological form and its development show simi-
larities – limited, but quite striking – to this aspect of the cognitive organization
in very different species.
9.5.1 Children’s early knowledge

The development of children’s control of the speech patterns of the language
they will speak is a subject that merits a book of its own – indeed, several books
have been devoted to exactly this, as the experimental results of recent decades
have shown how rich the child’s early experience is in this area. Well before
specific words or the principles of their combination are available, sophisticated
knowledge of the patterns of their realization in sound (or gesture, in the case of
deaf infants raised in the presence of adult signers) is already present. We sum-
marize below what is generally accepted about a universal, culture-independent
progression in the development of speech, following the very useful and ac-
cessible presentation by Bénédicte de Boysson-Bardies (1999).9 We make no
serious attempt to document these results independently or to give details of
the experiments on which the conclusions are based, for which the reader is
encouraged to consult the work just cited.
9 See also Jusczyk 1997 and Kuhl 1999 for further amplification and references.
It has been known for many years that infants follow a regular progression
in their sound productions during the first year or so of life. Between one
and about five months, their vocalizations are often referred to as cooing: at
this stage, they produce sounds resembling vowels and begin to control the
process of phonation. Around seven months, they begin the stage known as
babbling: early on, this consists of repetitive productions that rhythmically
alternate consonants and vowels (“babababa,” “mamamama,” etc.). By around
eight months, the vowels produced in babbling begin to approach those specific
to the language spoken by their parents and others around them, and may
alternate different syllables rather than repeating the same one. The intonation
contours of babbling also begin to resemble those of their (soon to be) native
language, and by the age of about ten months, it is possible to differentiate
children on the basis of the linguistic environment in which they have developed.
This development is a crucial part of the infant’s attaining control of language.
We can see this in part from the behavior of congenitally deaf babies: up until
the onset of babbling, their vocalizations are entirely comparable to those of
hearing babies, but deaf babies do not babble. Around the age of seven months
or so, their vocalizations diminish, and do not reemerge until later, at a point
where they are dominated by syllables with labial consonants that the baby can
see how to pronounce.
The absence of vocal babbling in deaf infants, however, does not mean that
this stage of language development is absent. For deaf children raised among
signing adults, there is indeed a linguistic environment: it is simply in another
modality from that of hearing children. And it has been shown (Petitto and
Marentette 1991) that these babies do indeed engage in manual “babbling” that
evokes the constituent elements of signed language, a kind of manual activity
that is qualitatively different from that of other children and that serves the same
sort of function of attunement to a linguistic world as oral babbling does for the
hearing child.
By around 10–15 months, children have arrived at a selection of vowel and
consonant types appropriate to their native language. While babbling may per-
sist, in the production of nonsense repetitions (generally with appropriate sen-
tential intonation), the first stable words begin to appear by the end of the first
year, and infants now come to use a consistent phonetic form to refer to an
object. Around 20–24 months, when most (though by no means all) children
have a vocabulary of roughly 250 to 300 words, they begin to combine words
in meaningful ways and produce their first sentences.
A child’s early productions are of course available for observation, and the
path of development sketched above has been documented for some time. It is
naturally rather harder to study the sequence of events in the development of
perception, since we cannot directly observe what is going on in the mind of
the prelinguistic child. Recent years have seen the emergence and refinement
of a number of ingenious experimental techniques for drawing conclusions

about a child’s perceptual abilities, however, and we do know quite a bit about
the sequence in which linguistic perceptual ability arises.
Even before birth, it has been demonstrated that the fetus responds to auditory
input, recognizing changes in sounds and reacting preferentially to the sound
of the mother’s voice. At birth (or as soon thereafter as it is practical to do
experiments), infants can discriminate sounds along all of the dimensions used
by the world’s languages. They also can detect changes in intonation patterns,
while recognizing the sameness of a speech sound under variations in intonation.
At (or shortly after) birth, infants show a preference for the prevailing inton-
ation patterns of their “mother tongue.” Up to the age of 2 months, infants show
no right-ear advantage for speech (though they do show a left-ear advantage
for musical contrasts). By 3 or 4 months, though, a right-ear advantage for
speech is present. The importance of this lies in the fact that (as we will note in
chapter 10) the primary language areas in the brains of most humans are in the
left hemisphere, to which the right ear provides the most direct access.
By the age of 5 months, infants can make some connections between visual
and auditory information. Around 6 months, infants show a preference for
speech containing the vowel sounds of the language spoken around them. At
about the same time, they are able to detect the prosodic cues for the boundaries
of clauses in different languages, despite the fact that it will be some time
before they actually produce utterances that could be said to be structured
into “clauses.” Around 8–10 months, their sensitivity to prosodic organization
increases still further, and they can be shown to be sensitive to phrase boundaries
within clauses – obviously an important ability if they are to be in a position to
impose a syntactic organization on their linguistic input, as we have assumed at
a number of points in this book. Although they do not yet produce stable words
at this point, they can still be shown to prefer word forms that respect the stress
patterns of their native language, and the constraints on sequences of sounds
that are characteristic of it.
Already, around the age of 10 months, infants’ ability to discriminate sound
contrasts that are not used in the language spoken around them is greatly re-
duced. This degradation of the phonetic perceptual virtuosity they are born
with will continue as they acquire their native language, and adults have con-
siderable difficulty hearing sound contrasts not present in their native language
(Japanese speakers distinguishing [r] from [l], English speakers hearing French
or Spanish [p] vs. [b], etc.), contrasts which they could discriminate as babies.
When the same physical acoustic dimensions are presented in a non-speech
signal, however, adults perceive them with roughly the same accuracy regard-
less of native language, showing that it is specifically linguistic perception that
becomes preferentially tuned to the dimensions of contrast utilized in a partic-
ular language.
These observations suggest something like the following course of devel-

opment. At birth, the child’s perceptual system is capable of making a range
of auditory discriminations. Both speech and general auditory processing have
roughly the same capacities. During the first months of life, the quality of the
speech around the child results in a “tuning” of the speech perception system
that begins to focus on the kinds of sound found in the surrounding language. At
around six months, the child begins to experiment with his own vocal apparatus,
discovering the connections between articulation and sound. As the perceptual
system becomes tuned to a specific language, and integrated with the motor
system, it comes to disregard distinctions that do not occur in this language.
This has no effect on non-speech perception, however.
This process of progressive refinement seems to proceed as a product of
the availability of evidence. Notice that the child in the womb has access to
some limited information about speech: the rhythm and intonation pattern of
the mother’s speech, but little or nothing else. And, in fact, we can demonstrate
some specialization for those patterns right after birth.
It has long been considered that this sort of development takes place during
a specific stage in life: the “critical” or “sensitive” period. For some aspects of
language learning, we have actual evidence that if the right kind of stimulus is
not present during the right age (roughly, up to puberty), learning does not take
place.10 In the domain of speech, there is little relevant information: the few
known studies of children deprived of linguistic input during their early years
did not apparently include the kind of perceptual tests that would tell us how
normal their speech perception systems were.
We can show, however, that in “late” bilinguals (people who learn a second
language well, but after childhood), as opposed to “early” bilinguals, there
are processing differences: imaging studies seem to show that the latter use
overlapping brain regions to process the two languages, while the former use
distinct regions. This suggests that whatever we learn after the period of normal
first-language acquisition, we learn in a different way.
Why should this be true? One possibility is that it is just a part of development:
certain mechanisms cut in at a particular point, and cut out at a particular point.
A slightly less metaphysical point of view, however is the following. In the
early years of life, the brain develops vast numbers of new synapses, peaking
at about 2 years of age. Over the next several years, neurons die and synapses
wither, resulting in a specifically limited organization. Neuron death levels out
at around age 7, while overall synaptic organization is determined by around
the age of puberty. If we assume that the role of experience is to shape and mold
this specific organization, then whatever is in place by puberty is what we are
“stuck with.”
10 For a survey of the classic cases relevant to this issue, and their interpretation, see Curtiss 1988.
9.5.2 Birds’ development of song

Somewhat surprisingly, perhaps, much the same story seems to be characteristic
of song learning in oscine birds (suborder Passeres) – but not of the communica-
tion systems of other animals, or even other communication systems in the same
animals.11 In some birds, song is learned, in the sense that some interaction with
the environment is required for normal development. In other birds, however,
the song is completely innate: a young phoebe, for instance, will sing nearly
correctly at the age of about 2 weeks, and the course of the bird’s improvement
depends solely on improved control of the vocal apparatus with physical mat-
uration. Crucially, these birds will sing correctly if (a) raised in isolation from
other phoebes; or (b) deafened shortly after hatching. This is not true for birds
that learn: these birds need experience and models, or else their vocalizations
will resemble pathetic squawks rather than real species-typical song.
It is conventional to distinguish true song from other avian vocalizations,
referred to as “calls.” Bird song is opposed to bird calls in its complexity, in its
ontogeny, and also in the underlying neural structures. To cite Marler (1999,
pp. 295f.):
Most calls seem to be innate . . . In most birds the song is also innate. In only three of the
27 orders of birds is there unequivocal evidence of vocal learning. These are the parrots,
the hummingbirds, and above all . . . the oscine songbirds . . . Oscines [have] uniquely
complex vocal apparatus, and [a] specialized network of brain nuclei that constitutes the
“song system,” lacking from the brains of suboscines, and as far as we know, from the
brains of all birds with innate songs.
On the other hand, “parrots seem to have achieved a similar end by their own
neuroanatomically distinct neural system, apparently independently evolved”
(Marler 1999, p. 296). The neural organization in hummingbirds shows some
initial similarities to that of oscines, but little evidence has been collected on
these species.
Song may range from a simple series of a few more or less identical notes
through long arias that may last 10 seconds or more. The difference between
songs and calls is only in part a categorial one. Marler (1999, pp. 295f.) describes
songs (as opposed to calls) as
especially loud, longer in duration than calls, often highly patterned, with a variety of
acoustically distinct notes . . . often a male prerogative, with many functions, the most
obvious of which are signaling occupation of a territory and maintenance of sexual
bonds. Songs are sometimes seasonal and sometimes given year round . . . Some learned
11 There is a vast literature on song and other acoustic communication in birds, including the
development of this ability as a function of differences among species. Our presentation here
relies especially on work by Peter Marler (1970, 1991, 1999, and elsewhere), along with other
research represented by chapters in Kroodsma and Miller 1996. Marler has explored the parallels
(and differences) between the development of song in birds and of language in human infants
in great detail, and his work has given rise to a great deal of further exploration of these issues.
birdsongs are relatively simple, on a par with those that are innate. Other learned songs
are extraordinarily complex, with individual repertoires numbering in the tens, hundreds,
and in a few cases, even in the thousands.
Songs are quite distinctive from one species to another, of course. The song
itself is made up of a number of separate notes, of different types. These occur
in a particular sequence, and the sequence is essentially the same across repeti-
tions. These matters are important: female song sparrows, for instance, respond
preferentially to songs that are (a) composed of “song sparrow” notes, and
(b) follow “song sparrow” patterns. Experiments show that female receptive-
ness is sensitive to both of these dimensions.
In fact, as Marler observes, the same bird will typically have a repertoire of
several different songs (two to ten, or even many more in other species), gen-
erally similar but distinct. All of these songs serve the same purpose, however:
to claim territory. Different songs do not convey different messages. Females
appreciate the variety, though: diversity of the male’s repertoire helps attract
mates, even though it cannot be shown to be linked to any other objectively
valuable genetic characteristic.
Songs often – maybe usually – display differences of dialect. That is, there
may be local variations that characterize the song. These are not genetic: if you
move a baby bird into a different area, he will learn the local dialect. Females
may even prefer locally appropriate variation, providing evidence that although
in most species they do not sing, females do some song learning too.
The role of learning is quite diverse across species. In some species (e.g.
cuckoo), even a rather complicated song is clearly innate, since it is sung without
learning. This is adaptive for this bird, since cuckoos typically lay their eggs in
other birds’ nests, with the result that the babies would not have other cuckoos
around to serve as models for song learning. In other species, though, learning is
definitely involved. This may consist in identifying some particular conspecific
individual’s song and copying it, perhaps picking up the right pattern from
within a range of possible choices (chaffinch), perhaps relatively free learning
(bullfinch).
The course of learning seems to involve four periods.12 First is the acquisition
of a song model from experience during days 15–35 after birth. The bird does
not sing yet, but it is during this time that he is listening – and song models heard
at this time are internalized and saved. This is followed by a period in which
the bird produces subsong, from roughly 25–40 days, consisting of relative
soft, broad band, unstructured sounds. This stage is thought to be the process by
which a young bird calibrates his vocal instrument, akin to babbling in human
infants. At about 35 to 80 days, the bird begins to produce plastic song, and
12 The timings cited here for these periods are for white crowned sparrows, the bird Marler studied
first in detail. Other species will vary somewhat from these precise ages, while following the
same overall sequence of development.
this period is marked by the gradual approximation of the young male’s song
output to the stored model(s). Finally, at about 90 days, the adult song is fixed
(or “crystallized”) in its permanent form. In some species, several different
songs are sung during plastic song and all but one are dropped at the stage of
crystallization.
Much of this is actually highly parallel to human language acquisition, given
the fact that the bird only has to learn to speak, as it were, and not to say anything:
he has to develop a command of phonology, but not syntax. The stages of song
learning, and the role of a critical (or “sensitive”) period (during which input
has to be available, or else song will not develop normally) are just like what
we find in human infants. Quite clearly, a given bird has a specific range of
systems that it is capable of learning, providing an obvious parallel with the
role of UG in determining the range of human languages that are accessible to
the child. The bird does not have to learn how song works: it only has to learn
which song to sing, within a narrowly constrained range.
Similarly, we can note that at least some birds, during the “plastic song”
phase, produce song elements that are not the same as any they have actually
heard, but which still fall within the range of possible song for that species.
This is comparable to the fact that children’s “errors” during language learning
correspond to different possible grammatical systems, at a stage where they are
still working out just what system the language they are learning instantiates.
In birds, these “errors” may well persist in adult song as creative innovations
(a form of “sound change”?).
This sequence shows us how a particular kind of learning takes place. It
seems quite incontrovertible that this learning sequence is determined by the
bird’s biology: vary the species, and the range of systems that can be acquired
changes, regardless of the nature of input. Song sparrows cannot learn to be
bluebirds, although they can learn to sing somewhat like swamp sparrows
(a closely related species with a rather different song) if that is the only model
available.
While the localization of human language functions in the brain is known
only sketchily, as we will see in chapter 10, the corresponding issues in the
neuroanatomy of birds are understood in much greater detail. The control of
song is centered in connections among several specific, neuroanatomically
distinct areas (especially those known as HVc, MAN, and RA). In song
birds, there are basically four functions that are subserved by this specialized
apparatus:
(9.23) a. Initiating non-song calls

b. Initiating and controlling song
c. Perceiving song
d. Learning song
One of the major findings in human neurophysiology is hemispheric

asymmetry: the fact that although the two sides of the brain are roughly simi-
lar anatomically, they are specialized in somewhat different ways for particular
functions. For instance, as we will discuss in chapter 10, most structural lan-
guage functions are concentrated (in most adult right-handed males) in the left
hemisphere, while non-structural, emotional, paralinguistic function is mostly
in the right. Since birds have two hemispheres too, we might ask whether their
brains are lateralized as well.
The answer seems to be: yes and no. Lesioning different sides of the bird’s
brain has slightly different results on the song. Actually, that is somewhat puz-
zling, since birds do not have a corpus callosum connecting the two hemi-
spheres, which are largely independent of one another (though there are a few
connections). Each hemisphere controls motor activity on the corresponding
side of the animal. A bird’s syrinx has two independently controllable parts,
and it might be thought that the right syrinx would be a right brain function and
the left syrinx a matter for the left brain.13 But in fact physical coupling makes
it more complex than this: what the bird does with the muscles on one side also
affects the configuration of the other. It seems that different, but interrelated,
parts of the song can be controlled separately by the two hemispheres. This is
not the same kind of lateralization we find in humans, as far as we can tell.
It is particularly interesting to note that the motor control areas that are
involved in song production are also involved in song perception. Williams
and Nottebohm (1985) have shown that zebra finch motor areas, right down to
the nucleus innervating the syrinx, respond when the bird is hearing roughly
conspecific song, but much less when the bird hears stimuli with auditory
properties that could not be song; not even when it hears its own song backwards.
This is obvious support for the motor theory of speech perception (mentioned
above in chapter 6), at least for birds, since the system whose basic function is
to control production also plays a role in perception. This does not mean that the
bird is literally singing along (though birds probably come closer to this than
people): it just means that the production system is involved, and hence that its
dimensions of control can be expected to play a part in perception.
Other interesting results from neurological investigations concern the mech-
anisms involved in learning. For one thing, a comparison of species in which
song is learned (e.g., zebra finches) and those in which it is unlearned
(e.g., phoebes) shows that brain areas that are essential to song production
and learning in the former (e.g., HVc) are simply absent in the latter. Birds that
learn have not one but two pathways that are involved in song production, one of
which seems to be specially involved in learning, because it becomes redundant
13 Birds do not have the kind of contralateral organization that connects the right side of the human
body to the left hemisphere, and vice versa.
once song is crystallized. This shows that specialized brain physiology is inti-
mately connected with the learning process, which is of course related to the
notion that song learning (though not the specific song) is innate.
While human infants display a relatively uniform pattern of linguistic devel-
opment, different species of birds behave somewhat differently with respect to
learning. In most species (e.g., zebra finch), song is learned once, in the first
year, and stays constant through the rest of life. In other species, though, new
songs are learned each year: this is the case with the canary. It turns out that
when we look at the birth and death of neurons in the song-relevant parts of
the bird brain, cell birth and death are associated with song learning and song
forgetting, respectively. Interestingly, when we compare canaries with zebra
finches, we find that neurogenesis occurs in close-ended learners (that is, those
who learn their song(s) once and for all), but in contrast to open-ended learners,
this process is generally arrested after the first year of life. These observations
provide a neurological basis for the observation that learning is associated with
critical or sensitive periods, and that the timing of these is a consequence of
physical changes that play themselves out in the stages of a bird’s maturation.
No one would seriously doubt that the control of birdsong is largely organized
as a function of the bird’s neuroanatomy, and thus of its biology. In some birds,
the whole development is innate, since the bird can come to sing its song
with little or no environmental input. In the birds we have been discussing,
though, the song control system develops on the basis of an interaction with
data provided by the environment. The productivity of this interaction is rather
precisely dependent on physical changes in the bird’s neuroanatomy, changes
that are clearly controlled by its specific genetic program.
All of this is strongly consistent with the picture of human language learning
as similarly driven by human biology: normal language acquisition in humans,
also, takes place preferentially during a specific stage of maturation. But just as
what a bird is capable of learning is a function of its species-specific biology,
so also what we are capable of learning as a first language is undoubtedly
determined by our genetic program. Our language organ is distinctly biological
in nature.
9.6 Conclusion
There is, of course, much more to be said about grammars and their acquisition,
as well as the development of phonology (including more detailed aspects than
we have attended to above); and there is an enormous technical literature.
Here we have tried to show how the biological view of grammars, focusing on
the internal representations occurring in individual brains, influences the way
one studies the acquisition of linguistic knowledge in young children. In this
connection, we have outlined a rigorously I-language-based approach.
9.6 Conclusion 215
Meanwhile, we have an approach to the riot of differences that we find in the

languages of the world and even within languages: there is a universal language
faculty and it is this UG which makes us accessible to one another, defining the
ways in which we analyze our linguistic experience. As children, our linguistic
experience varies tremendously; no two children experience the same set of
sounds and sentences. Nonetheless, the approach we have sketched enables
us to understand the universality of our development, why we categorize the
linguistic world so similarly and can talk to each other despite the enormous
variation in our childhood experience.
10 The organic basis of language
Our ability to speak and understand a natural language results from – and is made
possible by – a richly structured and biologically determined capacity specific
both to our species and to this domain. In this chapter we review arguments
that show that the language faculty is a part of human biology, tied up with the
architecture of the human brain, and distinct at least in significant part from
other cognitive faculties. We also discuss some of the work that has tried to link
the language organ with specific brain tissue and its activity.
Previous chapters have explored the structure of various components of our
language organ, and some aspects of the course by which that structure arises.
Some component of the mind must be devoted to language, and in its original
state (determined by Universal Grammar (UG)), prior to any actual linguistic
experience, it seems predisposed to infer certain quite specific sorts of system
on the basis of limited and somewhat degenerate data. This is what we mean
when we say that our language organ can be described by a grammar, and
the shape of particular grammars is determined by the system of UG as this
interprets the primary linguistic data available during the period of growth of
the language organ.
Thus far, our description is largely an abstract or functional one: that is, it
does not depend on the specific properties of the physical system that realizes
it. For a parallel, consider the nature of multiplication. We can characterize the
function of multiplication over the natural numbers in terms of some general
properties (commutativity, associativity, etc.), together with some specific re-
sults. Any computation that produces those results, consistent with the general
properties of multiplication, counts as multiplication: repeated addition, binary
shift-and-add strategies, the kind of logarithmic addition that used to be imple-
mented on slide rules, etc. Multiplication remains the same function, regardless
of the algorithm by which we compute it.1
Suppose we take a specific algorithm, for concreteness’ sake – perhaps the
standard one we learn in grade school, by which we carry out multi-digit
1 The discussion here is based on the kind of analysis of cognitive functions proposed by David
Marr (1982).
216
10 The organic basis of language 217
multiplications one place at a time, with carrying, etc. That algorithm seems
quite clearly specified, but in fact it can be implemented in various ways: with
the paper and pencil technique we learned in school (at least prior to the ubiquity
of calculators), on mechanical adding machines, old-fashioned moving-wheel
adding machines, an abacus, digital computers, etc. The inner workings of all
of these devices differ in various ways – even specifying the algorithm does not
tell us exactly how the system does it, at the level of implementation.
Much of what we have seen in previous chapters concerning language remains
at the functional level. We can determine properties of languages, often very
abstract and surprising ones; and we can establish the properties which UG
must have in order to allow the learner to establish a particular instance of the
“language function” on the basis of the kind of data available, but there are
myriad ways in which this function could be “computed” and (at least as far
as we know) few limits in principle on the kind of mechanism with which that
computation could be carried out.2 Silicon-based forms of artificial life (such
as contemporary digital computers) cannot at present fully replicate human
knowledge of language, let alone the acquisition of that knowledge, but we
have no particular reason to believe that the existing limitations reflect some
special property of the stuff of which we (as opposed to our workstations) are
made.
Nonetheless, when we ask how our knowledge of language is implemented,
there is some fact of the matter: that is, there is some physical aspect of human
beings that realizes the knowledge in question, and something that renders
this knowledge accessible in the observed way in the cases of real, physical
human beings. We maintain that there is no serious alternative to the notion of a
language organ: that is, a highly specialized aspect of our cognitive organization
which is common to all of us, particular to humans, specialized for the tasks it
performs (i.e., specific to the domain of language), and determined by human
biology – specifically including (though not limited to) the organic structure of
the human brain.
A commonly posed alternative to this position is the notion that language
is a product not of human biology but of human culture. On that view, lang-
uage might be seen as something that develops within every human society,
for reasons having to do with the imperatives of social interaction, but without
further determination beyond the constraints imposed on effective communi-
cation in naturally occurring human societies. Every known human society has
2 In the terms above, the study of the “algorithms” by which humans carry this out is referred to
as the study of language or speech processing, and forms a part of the discipline of psycho-
linguistics. We have said virtually nothing about processing in this book, which is not at all meant
to deny its significance. There are many different views of language processing that could all
correspond to the same picture of the properties of the language organ in the sense we have been
pursuing; and furthermore, for any particular theory of processing, there are in principle many
ways in which it could be carried out by actual physical systems.
218 10 The organic basis of language
conventions of dress, of table manners, etc., but there is no particular reason to

believe that this is due to a faculty for fashion, etiquette, etc., which is embodied
somehow in the human genome. If language were nothing but another form of
social convention, there would be no reason for it to be treated differently.
But the proposition that language is only one more socially determined arti-
fact, with the differences among the world’s languages mere cultural accidents,
is belied by the deep and detailed similarities that evidently exist among them,
as the research of the last forty years or so has made clear.3 A particularly
striking instance of the commonality of the human language faculty is supplied
by the manual languages which develop in Deaf communities independently
of one another or of the language of the surrounding hearing community, and
which share fully in the fundamental properties we have explored in the case of
spoken languages. We must conclude that the profound structural similarities
between signed and spoken languages,4 including not only the basic princi-
ples of their organization but the specific path of their development, the brain
regions associated with their control, and many other features, can be neither
the result of simple shared history nor necessary consequences of the articula-
tory/acoustic/auditory modality of spoken language. They must, rather, derive
from shared biology.
The notion that human language acquisition is primarily a matter of cul-
tural transmission, rather than biologically driven maturation in the presence
of relevant experience, is also undermined by instances of the development
of structurally deficient pidgins into the essentially normal linguistic systems
found in creoles. The deep reorganization of pidgins into creoles which takes
place as an essentially automatic result of transmission through the natural
language learning process in new generations of children, provides additional
support for the richness of the genotypic system involved in linguistic devel-
opment (Lefebvre 1998, Bickerton 1999). This kind of reorganization can be
observed on a daily basis in the emergence of the gestural languages of the Deaf.
Ninety percent of Deaf children are born into hearing homes, and often such
children are exposed initially to a kind of pidgin, as their parents struggle to
learn some kind of manual system of communication. Nonetheless the children
reorganize their experience and attain more or less natural, creole-like systems
(Lightfoot 1999, pp. 170–173).
The language faculty has properties typical of a bodily organ, a specialized
structure which carries out a particular function. Some organs, like the blood
and the skin, interact with the rest of the body across a widespread, complex
interface, and all organs are integrated into a complex whole. Often the limits
to an organ are unclear, and anatomists do not worry about whether the hand is
3 It is increasingly clear that the superficial diversity of the world’s languages conceals striking
commonalities among them. For an introduction to the evidence for this, see Baker 2001, who
considers a stunningly wide range of languages in a highly accessible way.
4 See Newport 1999, Supalla 1990, Anderson 1993 among other sources for some discussion.
10.1 Only humans have language organs 219
an organ or whether this designation should be reserved for one of its fingers. If
we think of an organ as a distinguishable aspect of an organism’s structure, in-
trinsic to and determined by its biological nature and implementing a particular
function, then the human capacity for language can be viewed in those terms.
10.1 Only humans have language organs

It appears that the capacity for language in the sense we have been developing is
specific to humans, and lacks significant parallels even in those species closest
to us, including the higher apes. This conclusion follows from the failures of
a half century of intensive efforts to teach human languages to individuals of
other species, especially chimpanzees and other primates.
The failure of attempts to teach spoken languages to non-human primates was
initially attributed to deficiencies in these animals’ vocal apparatus. To over-
come this, attention shifted in the 1960s to studies based on manual languages
such as American Sign Language (ASL; Gardner and Gardner 1969). Research
has demonstrated that ASL and other natural signed languages have the struc-
tural properties of spoken languages,5 but obviously without requiring a human-
like vocal tract. While some relatively minor differences in the structure and
control of the hands exist between humans and other primates, it is reasonably
clear that the basic articulatory demands of, e.g., ASL are well within the limits
of dexterity of chimpanzees and other primates.
Despite this, however, the animals in these experiments have never been
shown to acquire even the rudiments of the syntactic, lexical, or phonological
systems of natural languages.6 This has been repeatedly challenged by members
of the ape language research community, especially in connection with more
recent experiments involving bonobos (“pygmy chimpanzees” – pan paniscus)
(Savage-Rumbaugh et al. 1993, 1998), but the fact remains that nothing resem-
bling the natural human capacity for free, recursive syntactic combination, the
systematicities of relations among lexical items, or the organization of sound
(or sign) structure based on systematic, patterned combinations of a limited
number of basic components has been shown in any serious scientific work to
date.
5 See Klima and Bellugi 1979, Perlmutter 1991 and a rich technical literature. Note that there
is considerable controversy about the extent to which the input to which the apes in these
experiments were exposed actually constituted a natural language like ASL, as opposed to a
system that lacked essential properties of ASL.
6 For discussion of this work during the period when it was particularly prominent, see Petitto and
Seidenberg 1979, Terrace et al. 1979, Wallman 1992. Unfortunately, the negative conclusions
that arose from careful examination of these experiments held little appeal for the general public,
and the impression has gradually grown that some higher apes have been successfully taught
to use a human natural language. Despite the fascinating abilities that these apes have in fact
displayed, such a conclusion is simply not correct. See Anderson (forthcoming) for review of
these and other issues from the more general perspective of animal communication systems.
In contrast to the failure of these attempts to instill in primates a syntactic,

lexical, or phonological ability comparable to that which appears naturally and
spontaneously in every normal human child, a capacity for arbitrary symbolic
reference has been elicited in some higher apes (Premack 1978, 1980, 1990,
Savage-Rumbaugh 1986, Savage-Rumbaugh et al. 1986, Savage-Rumbaugh
1987, Seidenberg and Petitto 1987, Savage-Rumbaugh, Shanker, and Taylor
1998) and perhaps even in other animals, such as parrots (Pepperberg 2000).
This is extremely interesting, since such use of arbitrary symbols had not seemed
to occur in nature in non-human species. The demonstration that in some cases it
is nonetheless within their cognitive capacities is quite surprising and important.
It does not, however, compromise the conclusion that the systematic syntactic,
morphological, and phonological properties of human language are provided
by our language organ as a consequence of our specific genotype, and as such
are outside the capacity of non-humans.
This should hardly come as a great surprise, since every species has some
unique systems and capacities (and lacks others) as determined by its specific
genotype – inaccessible in the absence of the appropriate biology. It is not far-
fetched to compare the situation regarding language in other animals with the
fact that humans, even with intensive training, are incapable of free flight. So
birds fly, and humans talk.
The functional properties of our language organ develop along a regular
maturational path, such that it seems appropriate to see our linguistic knowledge
as “growing” rather than being “learned.” As with the visual system, much of
the detailed structure we find is “wired in,” though triggering experience is
necessary to set the system in operation and to determine some of its specific
properties.
In this respect, human language shows fascinating and detailed analogies (as
well, of course, as significant dis-analogies) with the development of song in
birds (Marler 1970, 1991), a system we explored in section 9.5.2 and which
is quite uncontroversially to be attributed to properties of the animal’s specific
biology rather than to some system of generalized learning or the like.
We marvel at the exotic structure and precise elaboration of systems such
as the bat’s echo-location facility, or the entirely different sonar systems of
dolphins, the electric field sensitivity (and communication based on it) of certain
fish, etc. (Hughes 1999). No serious scientist would question that these are
faculties quite specific in their structure and development to the individual
species concerned, and that no amount of structured training will enable even
the brightest or most sensitive human child to find insects by emitting short
pulses of high frequency sound and listening for their reflection. Why, then,
should we doubt that the intricacies of the human language faculty would be
similarly unique to our own species, especially given the repeated failure of
attempts to prove otherwise?
One might, of course, reject the validity of this analogy, and say that humans
lack certain skills of bats or electric fish because they lack the appropriate
sensory organs. But what, after all, is an “appropriate sensory organ”? Precisely
some bit of tissue which is specialized, as a result of the animal’s biological
organization, to be sensitive to certain environmental events. But that is exactly
what we claim for the language organ: it is a biologically determined aspect of
certain tissue (primarily in the brain), rendering it uniquely sensitive to linguistic
events in the environment, and allowing the development of a highly specialized
capacity as a consequence of that environmental interaction.
The importance of quite species-specific determination in these matters can-
not be underestimated: we must not imagine that an animal’s capacities develop
simply through the application of very generally defined cognitive systems to
the problems posed by its Lebenswelt. Many animals have noses, and olfac-
tory cortex, but no matter how much a man may desire truffles, he cannot find
them by their scent without the aid of his dog or pig. Similarly, non-human
primates certainly have ears to hear with and (with more qualification) a vocal
tract capable of producing sound; or hands and eyes capable of producing and
identifying signs; but this does not appear to endow even the cleverest of them
with the capacity for human language.
Of course, if one defines “language” generally enough, perhaps simply as
“structured communication” or the like, there is every reason to think that a
vast range of organisms display something of the sort, but we have seen in
earlier chapters that the human language organ has a much more specific and
substantive content than that – a content which in the final analysis has no
significant parallels in any other species.
10.2 Language is a function of the brain

What can we say about the physical implementation of language – about the
“hardware” that the language program runs on? The nature and capacity of the
system seem clearly to be specified (at least in part) as an aspect of the biological
structure of the organism. We can show that from the fact that disruptions in the
system seem to be heritable, as we will note in the following section. It is also
clear that language ability is not simply another aspect of general intelligence
or overall cognitive capacity. It looks as if there must be some specific part of
the organism where a distinctive language organ resides.
So where should we expect to find it, in anatomical terms? That may seem
obvious: in the brain. Interestingly enough, however, the relation of the brain to
capacities such as language has not always appeared self-evident. Historically,
we find mention in Babylonian and Assyrian cuneiform tablets from c. 2000 BC
of a connection between disorders of intelligence and brain disease. In an
Egyptian record from c. 1700 BC, we find mention of a connection between a
brain injury and the fact that the affected individual suffered a loss of speech.
In later theorizing, though, the relation of the brain to cognition (including lan-
guage) became somewhat more obscure: Aristotle, for instance, taught that the
brain was really a large sponge, whose function was to serve as a radiator to
cool the blood.
Only fairly recently (in historical terms), with the development of good
microscopes – and then later, with the development of more elaborate imag-
ing facilities – did it become possible to see enough of the structure of the
brain as an organ to develop any kind of coherent picture of its organiza-
tion and functions. The notion that the brain, and the nervous system more
generally, is a collection of cells that communicate with one another at spec-
ific connection points (the neuron doctrine) dates only from the nineteenth
century.
On a larger scale, there have been two distinct views of the brain that have
been in contention for much of recent history. One of these has its modern
origins in the theories of Franz Gall. Gall believed that the brain is composed
of a large number of very specific faculties, each specialized for a very limited
sort of function. He offered specific charts of this kind of organization (see
figure 10.1).
Gall also held the view that exercising a particular mental faculty caused
the corresponding part of the brain to enlarge, pressing against the skull and
producing corresponding bulges (or, in the case of underutilized faculties, de-
pressions). The result was supposed to be palpable irregularities in the surface
of the skull, which an appropriately trained interpreter could use as indices of
personality traits. That gave rise to the pseudo-science of phrenology, which
tended to give “faculty psychology” a bad name from which it has still not
completely recovered. There is no logical necessity, however, to the connection
between this sort of thing and the notion that the mind (as represented in the
brain) has a number of distinct components, modules, or faculties . . . among
them, the language organ.
At the beginning of the nineteenth century, the neurologist Marie-Jean-Pierre
Flourens (1824) performed experiments on animals in which he excised parts
of their brains and then looked to see what specific abilities were affected. By
and large, what he found was that all of his experimental subjects were affected
globally: that is, rather than losing just one particular function, Flourens’ ex-
perimental subjects were all turned into vegetables. From this he concluded
(Flourens 1846), in explicit reaction against Gall’s views, that cognitive capac-
ities are not localized in the brain, but rather distributed globally, so that injury
to any part results in a general degradation. This view that the brain represents
a single very general faculty, rather than a lot of individual and specific ones, is
sometimes called the “aggregate field” view of the brain: the view caricatured
by Pinker (1994) as that of the “brain as meatloaf.”
1. Amativeness 2. Philoprogenitiveness 3. Concentrativeness

4. Adhesiveness 5.Combativeness 6. Destructiveness
6a. Alimentiveness 7. Secretiveness 8. Acquisitiveness
9. Constructiveness 10. Self-esteem 11. Love of approbation
12. Cautiousness 13. Benevolence 14. Veneration
15. Conscientiousness 16. Firmness 17. Hope
18. Wonder 19. Ideality 20. Wit
21. Imitation 22. Individuality 23. Form
24. Size 25. Weight 26. Colour
27. Locality 28. Number 29. Order
30. Eventuality 31. Time 32. Tune
33. Language 34. Comparison 35. Causality
Figure 10.1 The personality organs of the human brain, according to Gall
and his followers Johann Spurzheim and George Combe
While the aggregate field view seems somewhat implausible in many ways,
it is quite attractive on one particular notion of the nature of human beings. If
we view most higher mental faculties as actually properties of an incorporeal
soul, or something of the sort, it seems natural to say that the brain is just a
large, functionally amorphous structure that serves as the locus for the effects
of the soul on the body (and perhaps vice versa), but nothing more precise
than that. If, on the other hand, we hold that specific faculties are anatomically
localized, this increases the extent to which we see those faculties as grounded
in the properties of specific material (brain tissue) in the physical world; and
as a result, the materialist view of the basis of the mind becomes much more
congenial. Indeed, in 1802, Gall was forced by conservative Catholic authorities
and the Austro-Hungarian emperor to stop lecturing in Vienna (and soon after, to
leave Austria) precisely because he was seen to be “championing materialism,
atheism, and fatalism bordering on heresy” (Finger 2000, p. 124).
In the late nineteenth and early twentieth century, however, an accumulation
of results suggested that the localist view is correct, at least in part. As people
looked into the question in more detail, it became clear that very specific in-
juries can result in rather specific functional impairments, rather than general
degradation (as in the case of Flourens’ animal subjects). And in this connection,
the study of language has played a central role.
10.2.1 Language localization in the brain

Nowadays, it is a generally accepted research assumption that brain functions
are localized. For instance, evidence from a number of sources makes it clear
that core aspects of language function are located predominantly in the left
hemisphere in normal, right-handed, adult males.7 This evidence includes
(a) brain damage: left but usually not right hemisphere stroke often produces
disorder or breakdown in language functions; (b) the Wada test: sodium amytal
is injected in the major artery supplying one or the other hemisphere, which
allows us to effectively suppress its activity. When this happens, we find that lan-
guage function is affected when the left hemisphere is anesthetized in this way;
(c) commissurotomy: severing the connections between the hemispheres allows
us to explore their functions separately; (d) imaging studies, which show clearly
that particular left hemisphere areas are metabolically and electro-chemically
active during language-related tasks; and (e) stimulation studies, such as those
made possible in presurgical examination of epileptic patients through the im-
plantation of subdural grids.
We should note that when we investigate these questions of localization,
we find that the organization of language in the brain is entirely comparable
regardless of the modality of the language: that is, signers seem to use the
7 Some (but not all) left-handed individuals have an opposite lateralization of language functions.
Lateralization is less prominent in young children, and develops in part as an aspect of maturation.
And some recent studies (e.g., Shaywitz et al. 1995) suggest that the asymmetries of language
function between hemispheres are less pronounced in women than in men, a result which is
consistent with the evidence that anatomical asymmetries between Broca’s area and the analogous
brain region in the right hemisphere are less pronounced in women than in men (Uylings et al.
1999).
same parts of their brains for language as speakers. Sign aphasias are entirely
comparable to conditions affecting spoken language, and we can find evidence
that the linguistic use of visual information is the responsibility of different parts
of the brain from its non-linguistic use (Poizner, Klima, and Bellugi 1987). ASL
is produced and perceived in the visual spatial modality, as opposed to spoken
languages, and right hemisphere lesions often produce marked visuospatial
deficits. Nonetheless, signers (like speakers) with left hemisphere strokes often
display significant problems with language, while those with right hemisphere
strokes display essentially normal (signed or spoken) language in the presence
of such spatial deficits as left hemi-neglect (essentially ignoring the existence
of the left half of their visual field). We can thus see that it is language and not
( just) speech which is primarily based in the left cerebral hemisphere.
However, this overall localization of language function in the left hemisphere
has been seen in recent years to be a significant oversimplification (Kosslyn
et al. 1999). Rather more language function can be identified in the right hemi-
sphere (by the kinds of arguments suggested above) than was once thought, and
there is considerable individual variation in these matters.
10.2.2 Studying the functional anatomy of language

Of course, it would be nice to be able to say something much more specific
than just that language functions are (generally) located in the left hemisphere.
Actually, though, there is little reason to expect detailed maps of language areas
in the brain to emerge soon. In an exploration of the specifics of localization,
George Ojemann and his colleagues (1989) studied over a hundred preopera-
tive epilepsy patients, using direct cortical stimulation. They discovered that
stimulation in precise locations could interrupt naming ability – but apart from
the fact that by and large, the relevant sites were in the perisylvian region, the
exact locations varied tremendously from one individual to the next.8 Still, we
would like to have as much precise information as possible. What techniques
are there for exploring these matters, and what does the evidence suggest?9
For most areas of cognition (other than language) that have been explored
in detail, researchers have an enormous advantage that is not available to those
studying language: animal models. The human visual and auditory systems, as
well as systems underlying other sensory modalities, motor control, and much
else, have close analogs in other animals, and we permit ourselves to study these
in ways that are completely excluded in the case of other humans. Much has been
8 These differences are important in determining appropriate sites for surgery to ameliorate epilep-
tic conditions.
9 We certainly cannot cover all of this emerging field here, nor is it necessary to do so to make
our overall point about the biological nature of the language organ. The papers in Brown and
Hagoort 1999 provide a summary of recent views and results, in much greater depth.
learned about, say, the visual system by studying cats and monkeys: removing
selective bits of brain tissue and observing the resultant deficits, implanting
electrodes directly into individual neurons in the intact brain and measuring
its activity, disturbing neural development in various ways and observing the
results, etc. We have learned much about these systems in animals evolutionarily
close to us, and our understanding of the role of various cortical areas in visual
processing is relatively precise. These animal models, in turn, have generally
turned out to be valid in their essentials for the human visual system as well.
The corresponding studies have not been carried out directly on humans, for
obvious moral and ethical reasons, but it appears we can largely dispense with
them, given what we know about other animals.
When we come to language, however, this path to knowledge is closed to us,
for the precise reason that no other animal besides homo sapiens possesses a
language organ, so far as we know. To understand the precise role of the brain in
language, we must start from the very beginning – and our options are relatively
limited, for the same reasons we cannot study the human visual system in the
kind of detail the monkey visual system has been explored. With few excep-
tions, we are limited to the evidence provided by non-intrusive experimental
methods, or to interpreting what we might call the “experiments of nature”: the
consequences of localized brain damage, as caused by stroke, trauma, or other
accidents.
Evidence from language deficits The primary source of evidence to

date concerning the precise localization of language functions has come from
the study of language deficits, or aphasia. The logic of this is to attempt to
infer the distinctness of the mechanisms underlying particular functions from
the evidence of patients who have suffered brain damage, and in whom these
functions are differentially affected. When we find a patient who can write
to dictation, but cannot repeat what we say, we presume something like the
following: auditory processing is intact, as is the patient’s access to the lexicon,
but there is a problem going from the lexicon to a spoken output. This implies
the potential distinctness of brain regions serving this last function. We then
look to the location of the lesions in our patient, and infer that the function(s)
impaired must have been located in (or at least essentially dependent on) the
area(s) destroyed.
The original, basic step in this study was made by Paul Broca, who in 1861
described a patient with good comprehension of language but severely impaired
production. At the time, the only way to find out where the brain was in fact
damaged in such cases was to wait for a post-mortem autopsy; and Broca’s
patient had a lesion specifically in the left posterior frontal lobe. Broca stud-
ied eight patients with similar language deficits, and found that in each case
subsequent autopsies revealed lesions in the left hemisphere (generally in the
same overall area, though it appears the actual lesions were rather diverse). This
led Broca to announce that “Nous parlons avec l’hémisphère gauche!” More
specifically, a particular kind of language deficit could now be associated with
damage to a relatively specific part of the brain, now referred to as “Broca’s
area” (though see Uylings et al. 1999 for some of the difficulties in defining
that notion more precisely in anatomical terms).
This result provided important impetus for the development of brain science,
and in the 1870s others discovered that electrical stimulation of the brain could
result in movements of very specific parts of an animal’s anatomy. In fact,
it was possible to correlate particular locations in the cortex with particular
motor activities, resulting in a topographic map of the primary motor strip that
shows an obvious resemblance to the body – with interesting differences, largely
related to the fact that some small body parts (the vocal organs, e.g.) require
disproportionately large amounts of controlling tissue, while other body parts
(the trunk and abdomen, e.g.) require less than their relative size would suggest.
A familiar and amusing exercise is to relate the brain tissue subserving
primary motor control with the parts of the body, resulting in a sort of
“homunculus” analogy (cf. figure 10.2).
More recent research suggests there are actually several of these “body maps”
in the brain, rather more diffused, and with some variation in structure in differ-
ent instances. Similar maps link areas of brain tissue bit by bit with the visual
field, with the frequency response range of the cochlea, etc. Neuroscientific re-
search is usually based on the view that activities are rather precisely localized.
And of course, once the plausibility of this position was well established, many
other researchers got busy looking for the specific localization of a variety of
functions – such as language.
In the late 1870s, Carl Wernicke proposed another rather specific relation
between brain lesion and functional deficit. This involved a class of patients
whose production is relatively fluent but who have a deficit in comprehension.
This pattern was associated with left posterior temporal lobe lesions, affecting
the area now known as Wernicke’s area. Wernicke noted that Broca’s area is
quite near the motor areas responsible for control of speech articulators, and
“Broca’s aphasics” have trouble producing fluent speech. “Wernicke’s area,”
on the the other hand, is close to areas of the brain responsible for auditory
comprehension, and “Wernicke’s aphasics” have comprehension diffculties.
Wernicke therefore suggested that language production is controlled by Broca’s
area, and comprehension by Wernicke’s area.
But Wernicke’s main contribution was the suggestion that only some of
the brain basis of cognitive structures such as language is to be found in the
localization of specific functions. Much of the way cognitive systems work,
according to him, involves interconnections among these specific functions.
Thus, on this view both the localists and the aggregate field people were partly
Figure 10.2 Topographic map of the body musculature in the primary motor
cortex (from Purves et al. 1997, p. 316)
right: some functions are local, but mental functioning involves making overall
patterns of connections among these functions.
On this basis, Wernicke predicted another kind of aphasia, one that would
be based not on a lesion in Broca’s area or Wernicke’s area, but rather on
impaired connections between the two. Such a patient ought to have relatively
spared production and comprehension, but to be very bad at tasks that involved
connecting the two, such as repetition. Patients displaying just this sort of
deficit (“Conduction aphasia”) were later actually found, providing one of those
wonderful instances that are all too rare in science of the confirmation of an
empirical prediction made on theoretical grounds.
How much can we refine and extend the information provided by cases
of language pathology in order to derive a more precise picture? As one
might expect, clinicians have gone far beyond the classification of “Broca’s,”
“Wernicke’s,” “Conduction” aphasics, etc. There are many clinical
“syndromes” that have been identified, and one might expect it to be possi-
ble to look for the locus of brain damage in each case to get a sense of what
goes on where.
Unfortunately, as more and more patients are documented, it becomes clearer
that each is an individual, and the deficits patients show are subtly specific.
In part, the interpretive difficulties arise from the fact that lesions are rarely
precise, and a variety of functions are inevitably implicated when a large area
of brain tissue is affected. But some neuropsychologists argue that so much
individual variation exists among brain-damaged individuals that the only way
to learn about cognition is to study each one as an individual, sui generis.
Lumping patients with similar but distinct lesions together and averaging over
them – e.g., treating a population of “Broca’s aphasics” as being comparable –
necessarily loses important detail on this view. Categories like “Broca’s” or
“agrammatic” aphasia may conceal as much as they reveal, by averaging out
the differences among distinct patients.
Indeed, aphasic symptoms can be exquisitely detailed. This is particularly
obvious in the anomias: patients can lose the ability to name quite precise
categories, such as animate objects, abstract objects, plants . . . one of us recalls
once hearing the neuropsychologist Frieda Newcombe describe a patient who
had lost the ability to name green vegetables, but nothing else.
Given the diffuse and contradictory evidence from attempting to correlate
anomias with lesion sites, it is vanishingly unlikely that this will ever lead to
the discovery of “the green vegetable center” in the brain; but even so, we learn
something more general from the cumulative effect of such cases. They suggest
that our knowledge of words is organized in such a way that semantically similar
items share some brain mechanisms.
Another bit of evidence in the same direction is provided by Tourette’s Syn-
drome patients. These people have a variety of involuntary seizures, including
occasional episodes in which they suddenly start to shout out strings of blas-
phemies and obscenities. Now since there seems to be no reason to assume
that these have any specific semantic content, it would appear that somehow
the patient’s brain has obscenities organized together as well as items similar
in substance, at least if we take Tourette’s Syndrome to be a localized brain
disorder of some sort.
A major problem in studying the brain through the deficits of aphasic patients
is that the ability to localize is rather poor, since most aphasias result from
rather comprehensive lesions. Also, the system as we study it in an aphasic is
an abnormal one: perhaps just the normal with something missing (as the logic
of most research seems to assume), but more likely with something missing and
something else added in the mind/brain’s attempt to make up for the deficits. It
would clearly be much better if we could study people without brain damage,
and in ways that let us see finer spatial detail.
In summary, evidence from brain lesions suggests that a number of important
language functions are largely based in certain areas in the left hemisphere,
especially around the perisylvian fissure. There is good reason to believe that
subcortical structures have considerable importance as well, but these have been
much less accessible to study and so are much less well understood. There seems
little hope that we can go much further than this toward localizing components
of the language organ on the basis of deficit studies, however.
Evidence from non-invasive imaging So how can we go about exam-

ining what goes on in the rather well protected tissue of the brain so as to explore
the neurological underpinnings of cognitive phenomena (including language)?
X-ray is one way to study the inside of a normal head, but even forgetting
about radiation, this is not useful for our purposes, because it does not show
soft tissue. Rather better is Magnetic Resonance Imaging (MRI), a technique
that shows the soft tissue in considerable detail. From a series of MRI scans
at regular spatial intervals, it is possible to reconstruct the three dimensional
structure of the brain tissue with some precision. But of course a picture of the
tissue itself, no matter how detailed, fails to reveal what the brain is doing. For
this, we need an indication of the activity taking place in the brain’s individual
neurons and groups of neurons.
Even at the level of the individual neuron, this activity is incredibly compli-
cated. It involves large numbers of tiny channels through which chemical ions of
various sorts can (or cannot) pass, resulting in changes in the electrical charge at
various points in the neuron. When certain such electro-chemical events occur,
they result in the release of a variety of other substances (neurotransmitters)
that effect the state of other neurons with which a given neuron is in contact.
A full picture of what is going on in even the simplest process with cognitive
content is surely bewilderingly complex – and, of course, we have to arrive
at our picture without actually penetrating the skull or introducing observing
instruments into the brain itself.
Two general approaches to the problem of imaging the brain’s activity have
evolved in recent years. One of these is based on the fact that neural activity
requires energy, and when a region of the brain becomes more active, it requires
an increase in local blood flow. If we can identify such regions of greater
blood flow, we can infer from their associated metabolic activity that something
interesting might be going on there. The other approach relies on the fact that
neural activity implies the flow of electrical current. While incredibly slight, this
current flow can under some circumstances be detected externally. In addition,
when current flows it produces a magnetic field, and this also can be detected
under some conditions. We discuss the potential results of these two sorts of
study in turn.10
Hemodynamic methods. One way of imaging metabolic activity (blood flow)

in the brain from outside is the technique of Positron Emission Tomography
(PET), which has in fact produced the great majority of the technicolor pict-
ures of working brains that appear in the popular literature. This technique
involves injecting subjects with radioactively “tagged” water, whose presence
in particular areas of brain tissue reflects blood flow (and thus, by inference,
metabolic levels). The location from which radiation is emitted can be detected
with considerable accuracy from outside the head, and areas where more of this
radioactivity can be measured can be assumed to be more active than others.
Despite the immediate appeal of the pictures that derive from PET experi-
ments, the technique is quite limited in its application to the study of cognitive
phenomena that take place rapidly, as most language processing surely does. It
tells us where activity is taking place, but temporal resolution is quite poor: on
the order of 30 seconds for a single picture. A longer period is required between
images, to allow for the decay of the radioactive isotope in the tagged water
from the earlier dose; and the radioactivity involved means that only a dozen
or so images can be collected from a given subject. These physical limitations
mean that any experiment inevitably involves a number of subjects whose im-
ages under parallel conditions must be averaged to obtain significance. Also,
since it is relative increase in activity that is of interest, not just absolute levels
of blood flow, these experiments generally involve “subtractive” designs, in
which two conditions (in one of which the function of interest is involved, but
not the other) are compared. One image is then “subtracted” from the other to
obtain a picture of the brain regions whose activity is presumably correlated
with the function under study.
These experimental details are by no means insignificant, because they make
the results extremely difficult to interpret in terms of cognitive functioning. For
one thing, pairs of conditions that have an appropriately “subtractive” character
have proven difficult if not impossible to construct. Although good spatial
resolution is possible with this technique, that advantage is partially lost as
a result of the need to average across the different brains of several subjects.
There is also the fact that increased blood flow to a region implies that overall,
it has greater metabolic needs; but if some of a group of neurons increase their
activity while others decrease, no such net increase in flow will be observed.
Despite these difficulties, PET has been used to explore some more inter-
esting issues about how language is represented in the brain. Differences in
processing between words and other stimuli can be detected, but only in an
10 See Rugg 1999 for a good survey of these methods, their possibilities and their limitations.
overall fashion that averages over long periods of time and many events. Jaeger
and her colleagues (1996) studied the processing of regular vs. irregular verbs
in English using PET, and concluded that these are stored in the brain in ways
that imply different cognitive mechanisms. Such studies are still a matter of
static pictures, though: the experiment involves having several subjects each do
the same thing over and over, which yields a single picture that represents the
whole period. The possible inferences to the functions of the language organ are
limited, at best. Poeppel (1996) discusses a series of PET experiments that pur-
port to identify brain regions active in phonological processing, and concludes
that a combination of weak experimental designs and intrinsic limitations of
the method result in essentially no significant result.
Another way of imaging blood flow is by the use of f(unctional)MRI: the
same method noted above, but with much more powerful magnets than those
normally used for static imaging of tissue. Somewhat paradoxically, the blood
draining from very active tissue is richer in oxygen than that draining from less
active tissue. The extent to which hemoglobin is affected by a strong external
magnetic field depends on its oxygen content, so these differences in metabolic
activity can result in an effect that can be measured from outside the head in
the presence of a powerful magnet.
The images that are possible with fMRI have even better spatial resolution
than PET images, and since they can be acquired within about 100 milliseconds
for a single “slice,” their temporal resolution is better as well. In addition, since
they involve no radiation exposure, there is no resulting limit in the number of
images that can be collected from a single subject, thus avoiding the problems
of intersubject normalization and averaging.
Despite these advantages, fMRI is still some distance from providing a
satisfactory approach to cognitive events, because its temporal resolution is
still somewhat too coarse to allow us to capture the dynamics of what brain
tissue is doing at an appropriate level. It also presents some problems of its
own (extreme sensitivity to slight movements, non-uniform imaging of tissue
in different regions, high noise level of the apparatus) which limit its usefulness.
Nonetheless, as experimental designs are developed which allow the study of
more sophisticated cognitive functions (“event-related fMRI”), its virtues, es-
pecially in terms of spatial resolution, will make contributions to studies of the
localization of language functioning in the brain.
One important problem with the majority of studies conducted using these
hemodynamic techniques is that they are designed by researchers who have
little or no familiarity with the concepts, methods, results, and research issues
of those working on language from a linguistic point of view. Most are based on
studies of individual words, rather than active use of language; they typically
involve visual presentation of written material, rather than the linguistically
more natural mode of aural presentation; and in general they make almost
no contact with the literature on grammars. Linguists certainly cannot expect

to be able to pose their questions directly and find answers in brain images;
but if those designing the experiments have no real sense of what is known
about the structure of the language organ, it is unlikely that they will find
interesting answers, either. One of our hopes in writing this book is to improve
on this situation. In any event, linguists and neuroscientists are increasingly
cooperating in this work, especially as the greater benefits of fMRI are realized
by both communities.
Electro-chemical methods. As mentioned above, as an alternative to the imag-

ing of blood flow as a clue to neural activity, we can attempt to measure the
consequences of electrical current flow in the neuron. This can be done either
on the basis of scalp measurements of current flow (electroencephalogram,
or EEG) or by using special sensors to record the patterns of magnetic field
variation that result from that flow. Each of these has its strengths and its dis-
advantages, as summarized well by Kutas, Federmeier, and Sereno 1999. The
overwhelming advantage of both methods over hemodynamic imaging is its
vastly improved temporal resolution: changes on the order of a millisecond can
be recorded, making it possible to study language processing on the temporal
scale on which it occurs.
The primary technique presently in use to study dynamics of processing is
the measurement of Event Related Potentials (ERP) – low level changes in the
scalp EEG signal that can be coordinated with linguistically interesting events.
This has been widely employed in studying syntactic processing (Hagoort,
Brown, and Osterhaut 1999), and a number of characteristics of the ERP signal
have been identified that appear to correlate well with language functions.
ERP imaging has been applied in a large number of studies of the dynamics of
linguistic processing. Interestingly, and unlike the experimental literature based
on hemodynamic methods, much more of the ERP literature on language has
been driven by questions asked by linguists (rather than neuroscientists). An
important reason for this is the fact that the equipment involved is vastly less
expensive: even some departments of linguistics can afford to develop an ERP
facility, while PET and fMRI remain more or less the exclusive province of
medical schools and their immediate affiliates.
The basis of ERP studies is the measurement of electrical potentials that can
be detected on the scalp – but that has some serious disadvantages. One of
these is the fact that although it is possible to record the signal with millisecond
accuracy, the related issue of connecting the signal to a particular area of brain
tissue has been shown to be unsolvable in principle: it is effectively impossible
to reconstruct the three-dimensional location of the subdural brain activity that
gave rise to the measured signal without additional information. A second
issue is the fact that only when a substantial number of neurons are active in
a simultaneous, synchronized fashion will there be a detectable signal. Much

brain activity fails to meet this threshold, and accordingly cannot be recorded
this way.
A different technique called Magneto-EncephaloGraphy (MEG) shows great
promise of being able to image actual electro-chemical events of very short
duration (like ERP), while retaining high spatial accuracy (like fMRI). This is
based on the detection of the magnetic fields associated with neural activity: like
the hemodynamic methods, this requires elaborate and expensive apparatus, but
its development has been driven more by questions of the study of cognitive
activity than by medical diagnostic considerations, and as a result, linguists
have been active participants in organizing the research agendas that have been
explored (see the papers in Marantz, Miyashita, and O’Neil 2000 for some
examples).
This technique too has technical problems which limit the extent to which
it can be considered the holy grail in the study of cognitive processes (Kutas,
Federmeier, and Sereno 1999). Like ERP, it is only sensitive to the activity of
substantial numbers of neurons acting together: indeed, these neurons must all
be aligned in the same way, and this must be on a tangent to the surface of the
skull in the relevant region. Some neural events related to language meet these
criteria: as a result, MEG has made progress in identifying interesting aspects of
neural functioning (see, e.g., Poeppel and Marantz 2000), but it is obviously far
from providing a way of looking at language-related brain activity in general.
Can we image the language organ? The hemodynamic methods of investi-

gating brain function thus provide us with observational techniques that have
quite good spatial resolution, but very poor temporal resolution; while electro-
physiological studies give excellent temporal resolution but (at least in the case
of ERP) very little (or quite ambiguous) spatial information. Is it possible to
improve on these approaches? Current research pursues this goal from sev-
eral perspectives. For example, fMRI studies conducted in concert with ERP
recordings can provide the additional spatial information needed to improve
the interpretation of the ERP signal. This kind of work is difficult to organize,
and still in its infancy, but it shows some promise of reducing the limitations of
each technique.
Does this mean that we can realistically expect to be able to present a fully
objective physical picture of the activity of the language organ, once the exper-
imental techniques available are suitably refined? We can certainly anticipate
an increase in our understanding of the physical underpinnings of that activity.
For example, in a series of papers (Poeppel 1996, Hickok and Poeppel 2000,
Poeppel and Marantz 2000), David Poeppel and his colleagues have examined
the neurological basis of speech perception, whereby physical signals in the
auditory system are transformed into symbolic mental representations that enter
into further computation. One might imagine that it would be fairly straightfor-
ward to identify the functional neuroanatomy of speech perception: it is tightly
linked to a single sensory modality,11 unlike higher-level lexical or syntactic
processes. However, it has proved surprisingly difficult to characterize. Poeppel
argues that the difficulty stems in part from the fact that the neural systems
supporting “speech processing” vary as a function of the task: the neural sys-
tems involved in discriminating syllables under laboratory conditions overlap
only partially with those involved in natural, on-line comprehension of spoken
language. Progress can be made in disentangling these results on the basis of
a variety of imaging techniques, but only in the presence of well-worked-out
theories of the cognitive functions involved. The images do not, as it were, bear
their interpretations on their sleeves. Similarly, Phillips (2001) discusses the
complexity of relating acoustic signals to higher-level lexical representations,
using electro-physiological results to refine theories of the abstract, functional
representations.
On a more cautionary note, let us not forget that none of these techniques
give us a way to examine what the neuron is actually doing: although we
talk about neurons “firing,” current flow is not the same thing as the complex
electro-chemistry of real neurons, where a variety of neurotransmitters having
diverse effects are in play. There is accordingly no reason to believe that, if the
functioning of the language organ is somehow reducible to (or representable
in terms of ) the activity of neurons, blood or current flow is the right level of
analysis. And, so far, we have no way of getting information from outside the
head about these more detailed matters.
A more fundamental point here is that the representations provided by non-
invasive imaging techniques are not simply the unvarnished reality of brain
activity, but a significant abstraction from it: they represent some features of
these events (metabolic or electromagnetic correlates of the coordinated ac-
tivity of a large number of spatially associated neurons), but omit much else.
There is no reason a priori to believe that the specific abstractions provided by
these technologies will correspond in a particularly close way to the kinds of
abstraction we need to understand our functional descriptions of the language
organ. We have already seen in chapter 6, for example, that the dimensions of
a linguistically relevant (phonetic) representation of speech correspond only in
a rather indirect way to what is measurable in the acoustic signal. There is no
particular reason to expect that the fit between brain imaging data and other
11 This is true to the extent we disregard work suggesting that non-auditory evidence for the activity
of the speech apparatus – visual observation of the talker’s face, for instance, as in the classical
“McGurk effect” (McGurk and MacDonald 1976) – can also enter into our perception of speech.
And of course the kind of “analysis-by-synthesis” view of perception discussed in chapter 4
further complicates the extent to which we can expect speech perception to be simply a matter
of transforming the information provided by the peripheral auditory system alone.
aspects of language function need be any closer than this. We may, of course,
be able to make inferences about functional organization from evidence of this
sort – at least, if we have sufficiently articulated theories in both domains to
make it possible to confront each with the other. Even adopting a thoroughly
materialist view of the language organ as a cognitive function arising from the
activity of the brain and nervous tissue, though, we are hardly committed to the
proposition that the kinds of physical phenomena we can measure are exactly
and exhaustively coextensive with that function.
The conclusion to be drawn from the remarks above is that while we can be
quite sure that the language organ is a biologically determined aspect of human
physiology, and that it arises primarily as an aspect of the electro-chemical
activity of the brain (more specifically, in most individuals, of the left cerebral
hemisphere), we do not yet have techniques that would allow us to relate specific
cognitive events to specific neurophysiological (electro-chemical) events in
particular assemblages of neurons with any certainty. What evidence we have
suggests that the relevant events are rather complex and implicate substantial
regions of cortical and subcortical tissue, at least some of which subserves other
functions as well as language. And in any event, the cognitive interpretation of
that activity can only be offered in the presence of a fully articulated picture of
its functional nature: of the language organ, as we have interpreted that notion
in this book.
10.3 Language is a particular faculty

The language organ is thus not comparable to, say, the kidney, in having a
clear and specifiable anatomical localization (at least in the present state of
our knowledge). Our understanding of the localization of cognitive function
in brain tissue is currently too fragmentary and rudimentary to allow for clear
claims of this sort. Certain areas of the brain (both cortical and subcortical) can
be shown to subserve functions essential to language, in the sense that lesions
in these areas disrupt language functioning, sometimes in remarkably specific
ways (Mayeux and Kandel 1991), but inferences from this evidence to claims
that, for example, “language is located in Broca’s (and/or Wernicke’s) area” are
quite unwarranted. Indeed, even the overall localization of language function
in the left cortical hemisphere has been seen in recent years to be a significant
oversimplification, as we noted above.
But in fact, even if it were to become clear that there is no real segregation
between language-related and non-language-related brain tissue, it would still
be useful and important to treat the language capacity as a discrete and speci-
fiable human biological system in functional if not anatomical terms, on the
basis of arguments of the sort we have adduced.
The domain-specificity of the language faculty is supported by the extensive
literature documenting dissociations between control of language structure and
10.3 Language is a particular faculty 237
of other aspects of cognition. Where a system operates and is subject to discrete

impairment independently of other systems, it is a candidate for modular status.
Thus in the domain of the senses, one can be deaf without being blind, and vice
versa, which supports (though it does not by itself require) the claim that hearing
and sight are the products of distinct systems. Neil Smith (in press) provides an
excellent discussion of this point. He discusses the case of a linguistic “savant”
Christopher, whose hand–eye coordination is severely impaired and whose
psychological profile shows “moderate to severe disability in performance tasks,
but results close to normal in verbal tasks.” Despite low general intelligence,
not only is his language essentially unimpaired, but in fact he has an astonishing
capacity to pick up languages; see Smith and Tsimpli 1995 for more extensive
documentation and analysis.
In contrast, the phenomenon known as Specific Language Impairment (SLI;
for an overview, see Joanisse and Seidenberg 1998) represents an apparently
genetically determined condition in which language ability is impaired in fairly
precise ways in the presence of otherwise normal abilities in other domains:
most SLI children are cognitively normal but fail to develop age-appropriate
linguistic capacities (Bishop 1997). The homogeneity of the cases which have
been grouped together under this diagnosis is quite controversial, but in sup-
port of the biological nature of the faculty in question, the distribution of SLI
in some well-studied populations has been shown in both epidemiological and
genetic studies (Tomblin 1997) to be that of a relatively simple Mendelian trait
(Gopnik 1990, Gopnik and Crago 1991), perhaps even with a specific, identifi-
able chromosomal location. Researchers have postulated a range of grammatical
deficits associated with this genetic abnormality (Clahsen, Bartke, and Gollner
1997, Gopnik 1997, van der Lely 1996); see Levy and Kavé 1999 for a useful
overview.
Smith (in press) points to other dissociations:
Just as intelligence and language are dissociable, so also is it possible to separate linguis-
tic ability and Theory of Mind, with autistic subjects lacking in the latter but (potentially,
especially in the case of Asperger’s Syndrome – see Frith 1991) language being retained
within normal limits. Some Down’s Syndrome children provide a contrary scenario,
with their Theory of Mind being intact, but their linguistic ability moderately to severely
degraded.
Similarly we find “submodular” dissociations within the language organ,

suggesting that grammars have their own internal modules. Smith points to
dissociations between the lexicon and the computational system. Christopher’s
talent for learning second languages “is restricted largely to mastery of the
morphology and the lexicon, whilst his syntactic ability rapidly peaks and then
stagnates . . . [A] reverse dissociation [is] found in the case of children with
Spinal Muscular Atrophy, who seem to develop a proficient syntactic rule sys-
tem but have correspondingly greater difficulty with lexical development” (see
Sieratzki and Woll in press). Edwards and Bastiaanse (1998) address this issue
for some aphasic speakers, seeking to distinguish deficits in the computational
system from deficits in the mental lexicon.
We also know that focal brain lesions can result in quite specific language
impairments in the presence of normal cognitive abilities, and vice versa
(Caplan 1987). Friedmann and Grodzinsky (1997) argue that agrammatic apha-
sics may be unable to compute certain abstract structural elements (“functional
categories”), while Grodzinsky (2000) identifies much of agrammatism with a
disorder specifically impairing the computation of movement relations, local-
ized in the classical “Broca’s area.” Ingham (1998) describes a young child in
similar terms, arguing that she lacked one particular functional category.
This modular view runs contrary to a long tradition, often associated with
Jean Piaget, which claims that language is dependent on prior general cogni-
tive capacities and is not autonomous and modular (Piaget and Inhelder 1968,
Piattelli-Palmarini 1980). Such a claim is undermined by the kinds of dissoci-
ations that have been observed, however. Bellugi et al. (1993) have shown, for
example, that Williams Syndrome children consistently fail to pass seriation
and conservation tests but nonetheless use syntactic constructions whose ac-
quisition is supposedly dependent on those cognitive capacities. Clahsen and
Almazan (1998) demonstrate that Williams Syndrome children have good con-
trol of the rule-governed aspects of syntax and word formation, but are severely
impaired in certain irregular, memory-based functions, while SLI children dis-
play an essentially symmetrical pattern of affected and spared abilities. More
generally, language and other cognitive abilities dissociate in development just
as they do in acquired pathology (Curtiss 1988).
10.4 Conclusions
The psychologist Eleanor Rosch once said that she wanted her new field to be
empirical but not barbarically so. The key to successful scientific work is to
find a level of abstraction at which one can state interesting generalizations.
The point of our book has been to demonstrate that generative grammarians
have found an appropriate level of abstraction in the notion of I-language,
which has enabled them to understand the language faculty better. The details
of that abstraction make up the focus of empirical work of the kind that we have
described.
Does the appropriate level of abstraction incorporate the deletion operations
that we discussed in chapter 3? Perhaps yes, if what we sketched endures, as
ideas about constituent structure and the binding theory have endured now for
long periods. Or perhaps not. In fact, most likely not. As work progresses, our
ideas about deletion will probably come to be seen in a different light, related
to other ideas, whose connections we do not now see, and the technology
will change. That is what it means to work at a certain level of abstraction,

constantly refining and revising hypotheses. It is striking that generative theories
of grammar have been reformulated fairly radically a number of times over
the last several decades in all the domains that we have sketched in this book.
Each of those reformulations has involved rethinking analyses across the board,
but grammarians have always dealt with primitives which are abstract and
formal, involving categories like N, V, [Coronal], structure-building and lexical
operations, deletion, indices, and the like. These are elements of the I-language
abstraction and they are enduring elements.
Rosch’s “barbarism” comes if one tries to work at inappropriate levels of
abstraction. At several points earlier in the book we have contrasted the work
we have described with so-called E-language approaches, where the primitives
are drawn from the external world of sounds, words, and sentences: low-level
abstractions. We have shown that a person’s language capacity can hardly begin
to be described at that level. The kinds of distinctions between well-formed and
ill-formed sentences discussed in chapter 3 require a higher level of abstraction
and cannot be captured in the E-language approaches which dominated earlier
periods of the field and continue to be pursued in certain quarters.
One cannot understand major facts about the bumpy way in which languages
change from generation to generation, if one limits oneself to describing lan-
guages in terms of low-level units like sounds, morphemes, words, and sen-
tences. Those E-language approaches lead to Rosch’s “barbarism,” we would
say. When confronted with the kinds of distinctions we have indicated, they
invoke vague analogical notions, or Cyclopean ideas that certain kinds of lan-
guages tend to change to other kinds of languages, sliding down some mystical,
diachronic slope (cf. chapter 8), or global pronouncements that children slowly
learn to match their input (cf. chapter 9). The knives just are not sharp enough
to make the right distinctions, to capture the details, and these approaches are
not adequate to analyze language as a cognitive phenomenon. They might per-
haps be adequate to analyze languages from other points of view, in their social
dimensions or as literary tools, though we doubt that, as well.
At the other end of the spectrum, someone might argue that the language
faculty is brain based, and that therefore one should work at still higher levels
of abstraction and describe it in terms of brain operations, analyzing cortical
matter and synapses. We have indicated in this chapter that interesting work is
emerging which seeks to correlate what we know about language with what we
know about the brain. At this stage of our knowledge, however, observations
of the brain are unlikely to reveal the kinds of distinctions discussed in earl-
ier chapters, and there is no reason to stipulate that work on the brain should
be privileged. If syntacticians, working with their formal abstractions, can dis-
cover the distinctions among the sentences of English discussed in chapter 3 and
then tease apart information which is plausibly native, intrinsic to the species,
distinguishing it from crucial information which English speakers derive from

their external experience, and if that enables them to understand aspects of ling-
uistic diversity, distinctions between Italian, Swedish, and Banni-Hassan, then
they have a productive research program. That program will, surely, enable us
to understand better the kinds of distinctions which occur in brain-damaged,
aphasic patients or how Williams Syndrome children differ from others in their
language acquisition. It will also help those conducting brain imaging studies
to ask more substantive questions about the functions that neural activity might
correspond to. If the program doesn’t help brain scientists, then something is
wrong and ideas need to be adjusted, either those of the grammarian or those
of the brain scientist, or more likely, both. If the program does help, then the
productivity of the program is increased. That is why it is important that gen-
erative grammarians attend to the literature on language acquisition, aphasia,
and brain processes, and why acquisitionists, aphasiologists, and neurophysi-
ologists should attend to the grammatical literature. It is the importance of that
interaction that drove us to write this book.
Humanists often complain about scientific work becoming narrower and
narrower, but Sir Peter Medawar (1967, p. 115) argued that the opposite is
actually the case in the sciences:
One of the distinguishing marks of modern science is the disappearance of sectarian

loyalties. Newly graduated biologists have wider sympathies today than they had in
my day, just as ours were wider than our predecessors’. At the turn of the century
an embryologist could still peer down a microscope into a little world of his own.
Today he cannot hope to make head or tail of development unless he draws evidence
from bacteriology, protozoology, and microbiology generally; he must know the gist of
modern theories of protein synthesis and be pretty well up in genetics.
This kind of breadth is what we aspire to in the cognitive sciences, particularly

in work on language organs.
To study brain processes with only minimal knowledge of work on grammars
is to restrict oneself to another kind of barbarism in understanding language.
One may claim that “human linguistic and cognitive ability can be traced back
to learned motor responses” (Lieberman 2000, p. 2), but one restricts oneself
to barbarism by failing to connect that idea to the details of what we know
about human linguistic and cognitive function. We are beginning to discover
interesting things about brain operations, and of course we want to connect
them to what we know about language. That means that people who work on
imaging the electro-chemical activity of the brain or on aphasia should know
about grammatical analyses conducted at the functional level we have described,
and that grammarians should know something of what we are beginning to learn
about brain processes. In this way, each group benefits in refining its approach
to learning more about people’s language organs.
In fact, people work at levels where they believe that they can make significant
progress. Strikingly, over the last forty years work on syntax and phonology has
blossomed and led to countless discoveries. A good case can be made that studies
of children’s language have blossomed more recently, say over the last ten years,
and that we are learning a great deal along the lines of what we described in
chapter 9. On the other hand, it is less clear that comparable productivity has
been achieved in studies of aphasia over recent years. All this can change and
aphasiologists might discover some productive tools which will enable them,
over the next decade, perhaps, to make discoveries in a way comparable to
what has happened in syntax and language acquisition. We would bet that
aphasiologists might discover useful tools in the grammatical and acquisitional
literature – or elsewhere, of course. Hunches about future progress and the
discovery of useful tools guide people in figuring out what they want to study
in graduate school, as they prepare to undertake careers in the cognitive sciences.
So there are different levels at which one might work. E-language levels do
not look at all promising for work on the language faculty as an element of
human cognition. I-language approaches of the kind we have described have
been very productive, we contend. Brain-based approaches so far have not led
to much understanding about the language faculty, but there has been some
and, now that we have better tools, it is likely that we shall learn much more in
this domain in the near future, hoping for some correspondence between what
we find about brain processes and the nature of I-language structures and their
acquisition.
What we seek quite broadly is a theory of the mind, seen as an element of
nature. Chomsky (2000, ch. 4) has construed the kind of work we have described
as a development of the rational psychology of the seventeenth century: there
are “principles or notions implanted in the mind [that] we bring to objects from
ourselves [as] a direct gift of Nature, a precept of natural instinct . . . common
notions [and] intellectual truths [that are] imprinted on the soul by the dictates
of Nature itself, [which, though] stimulated by objects [are not] conveyed” by
them (Herbert of Cherbury 1624/1937, p. 133). We try to discover the “principles
or notions implanted in the mind” that are a “direct gift” of nature. We begin
with common-sense formulations. If we observe that Juan knows Spanish, we
focus on a state of the world, including a state of Juan’s brain. He knows
how to interpret certain linguistic signals, certain expressions. We might try to
characterize that knowledge and to ask how his brain reached this state, how his
language organ developed in the interplay of nature and nurture. Inquiry leads to
empirical hypotheses about biological endowment, information derived from
the environment, the nature of the state attained, how it interacts with other
systems of the mind, and so on.
One conducts that inquiry as best one can, typically invoking deeper and
deeper abstractions as one moves beyond common-sense formulations. There
is no a priori reason why analyses must be stated at one particular level of

abstraction. One uses those levels that lead to productive insights. In particular,
there is no reason why everything must be stated in terms of brain physiology.
If somebody claimed that there were such reasons, we would want to know
whether they should be stated in terms of what we know of the chemistry
of the brain or at a still higher level of abstraction, in terms of some kind of
particle physics. Nor is there any reason to require that grammatical analyses be
reducible in some transparent way to what is now known about brain physiology.
We have very little idea how the nervous system computes simple, arithmetical
operations, but that does not mean that algorithmic accounts of arithmetic should
be discarded as an account of human computational capacity because they have
no demonstrable basis in what we now understand of brain processes.
Working at different levels of abstraction is normal practice. One wants to
unify various theories about the world by discovering correspondences between
them, but that does not necessarily entail the kind of reduction which took
place in the mid-twentieth century, when much of biology was reduced to
biochemistry. That was a rare event in the history of science. More usually,
one or other of the sciences has to be reformulated, sometimes radically, before
unification can proceed. Chomsky gives the example of nineteenth-century
chemistry: its accounts of molecules, interactions, properties of elements, etc.
was in no way continuous with or harmonious with physics, as then understood.
“By the 1930s, physics had radically changed, and the accounts (themselves
modified) were ‘continuous’ and ‘harmonious’ with the new quantum physics”
(Chomsky 2000, p. 82).
One picks research strategies which are opportune at any given time. We have
tried to show here that an approach to the language faculty which sees it as part of
our cognitive make-up can be fruitful, if one proceeds at the levels of abstraction
that we have illustrated, teasing apart intrinsic and extrinsic properties and
postulating abstract grammars. We believe that these abstract grammars are
represented in the brain somehow and we are beginning to learn something
of how that might be, hoping to learn more along those lines over the near,
foreseeable future.
It is no news that things mental, of the mind, are emergent properties of the
brain. We are not much better at saying how they emerge than our predecessors
were when such ideas were expressed in the seventeenth century, but perhaps
we do have better ideas about the nature of things mental, most strikingly in the
domains of language and vision. Those ideas should be pursued, while neurosci-
entists, informed by grammatical work, discover how grammatical operations
are implemented at the cellular level; and grammarians should attend to work in
neuroscience, which may suggest that they have cut the empirical cake wrongly.
We already have some convergences, for example in work on aphasia showing
that inflectional morphology is quite distinct from other linguistic processes
(see remarks by Alfonso Caramazza in Gazzaniga 1996, pp. 131–151). That is

how we shall elaborate a cognitive physiology.
Perhaps one day progress in these domains will lead to some understanding of
consciousness, but that seems more remote, harder to foresee. It is more useful,
we think, to pursue for now the kinds of abstractions and idealizations where
we have some understanding, notably regarding vision and language organs,
in the hope and expectation that they will provide better tools for tackling
more remote problems, working towards an understanding of consciousness,
rationality, morality, emotions, and other attributes of the natural world, the
world of human beings and their cognitive capacities.
References
Allen, C. 1997. Investigating the origins of the “group genitive” in English. Transactions
of the Philological Society 95. 111–131.
2002. Case and Middle English genitive noun phrases. In Lightfoot 2002. 57–80.
Anderson, S. R. 1974. The organization of phonology. New York: Academic Press.
1976. On the description of consonant gradation in Fula. Studies in African Linguistics
7. 93–136.
1978. Syllables, segments and the northwest Caucasian languages. In Syllables and
segments, ed. by A. Bell and J. B. Hooper. 47–58. Amsterdam: North-Holland.
1981. Why phonology isn’t “natural.” Linguistic Inquiry 12. 493–539.
1985. Phonology in the twentieth century. Chicago: University of Chicago Press.
1988. Objects (direct and not so direct) in English and other languages. In On lan-
guage: a Festschrift for Robert Stockwell, ed. by C. Duncan-Rose, T. Vennemann
and J. Fisiak. 279–306. Beckenham, Kent: Croom-Helm.
1990. The grammar of Icelandic verbs in -st. In Icelandic syntax, ed. by J. Maling and
A. Zaenen. Syntax & Semantics 24. 235–273. New York: Academic Press.
1992. A-morphous morphology. Cambridge: Cambridge University Press.
1993. Linguistic expression and its relation to modality. In Current issues in ASL
phonology, ed. by G. R. Coulter. Phonetics and Phonology 3. San Diego: Academic
Press.
2000. Reflections on “on the phonetic rules of Russian.” Folia Linguistica 34. 1–17.
forthcoming. Doctor Dolittle’s delusion: animal communication and the nature of
human language.
Anderson, S. R. and D. W. Lightfoot. 1999. The human language faculty as an organ.
Annual Review of Physiology 62. 697–722.
Andersson, A-B. and O. Dahl. 1974. Against the penthouse principle. Linguistic Inquiry
5. 451–454.
Aronoff, M. 1976. Word formation in generative grammar. Cambridge, MA: MIT Press.
1988. Two senses of lexical. Proceedings of the Eastern States Conference on
Linguistics 5. 13–23.
Baker, M. 2001. The atoms of language. New York: Basic Books.
Barbosa, P., D. Fox, P. Hagstrom, M. McGinnis and D. Pesetsky, eds. 1998. Is the best
good enough? Cambridge, MA: MIT Press.
Barbour, J. 2000. The end of time. Oxford: Oxford University Press.
Bates, E. A. and J. L. Elman. 1996. Learning rediscovered. Science 274. 1849–1850.
Bauer, L. 1995. The emergence and development of SVO patterning in Latin and French.
Oxford: Oxford University Press.
244
References 245
Bazell, C. E. 1952. The correspondence fallacy in structural linguistics. Studies by

Members of the English Department, Istanbul University 3. 1–41.
Beard, R. 1995. Lexeme-morpheme base morphology: a general theory of inflection and
word formation. Albany: SUNY Press.
Beckman, M., J. Edwards and J. Fletcher. 1992. Prosodic structure and tempo in a
sonority model of articulatory dynamics. In Docherty and Ladd 1992. 68–86.
Bellugi, U., S. Marks, A. Bihrle and H. Sabo. 1993. Dissociation between language and
cognitive functions in Williams Syndrome. In Language development in exceptional
circumstances, ed. by D. Bishop and K. Mogford. Hillsdale, NJ: Lawrence Erlbaum.
Berthier, M. L., A. Rutz, M. I. Massone, S. E. Starkstein and R. C. Leiguarda. 1991.
Foreign accent syndrome: behavioral and anatomical findings in recovered and
non-recovered patients. Aphasiology 5. 129–147.
Bickerton, D. 1984. The language bioprogram hypothesis. Behavioral and Brain
Sciences 7. 173–221.
1999. How to acquire language without positive evidence: what acquisitionists can
learn from creoles. In DeGraff 1999. 49–74.
Bishop, D. V. M. 1997. Uncommon understanding: development and disorders of
language comprehension in children. London: Psychology.
Bittner, M. 1987. On the semantics of the Greenlandic antipassive and related construc-
tions. International Journal of American Linguistics 53. 194–231.
Blake, B. 1994. Case. Cambridge: Cambridge University Press.
Bloch, B. 1941. Phonemic overlapping. American Speech 16. 278–284.
Bloomfield, L. 1933. Language. New York: Holt.
1939. Menomini morphophonemics. Travaux du cercle linguistique de Prague 8.
105–115.
1962. The Menomini language. New Haven: Yale University Press. [Charles F.
Hockett, ed.].
Blumstein, S. E., M. P. Alexander, J. H. Ryalls, W. Katz and B. Dworetzky. 1987. On
the nature of the foreign accent syndrome: a case study. Brain and Language 31.
215–244.
Boas, F., ed. 1911. Handbook of American Indian languages. Bulletin 40, part I. Bureau
of American Ethnology.
Boatman, D., B. Gordon, M. Stone and S. Anderson. 1994. Studies of articulatory timing
in normal and foreign accent syndrome speech. Brain and Language 47. 549–552.
Borer, H. and K. Wexler. 1987. The maturation of syntax. In Parameter setting, ed. by
T. Roeper and E. Williams. 123–172. Dordrecht: Reidel.
Bowerman, M. F. 1973. Structural relationships in children’s utterances: syntactic
or semantic? In Cognitive development and the acquisition of language, ed. by
T. E. Moore. New York: Academic Press.
Boyce, S. E., R. A. Krakow and F. Bell-Berti. 1991. Phonological underspecification
and speech motor organization. Phonology 8. 219–236.
Brentari, D. 1995. Sign language phonology: ASL. In The handbook of phonological
theory, ed. by J. A. Goldsmith. 615–639. Cambridge, MA: Blackwell.
Briscoe, E. 2000. Grammatical acquisition: innovative bias and coevolution of language
and the language acquisition device. Language 76. 245–296.
Broe, M. B. and J. B. Pierrehumbert. 2000. Papers in laboratory phonology vol. V:
Acquisition and the lexicon. Cambridge: Cambridge University Press.
246 References
Browman, C. and L. Goldstein. 1990a. Gestural specification using dynamically-defined

articulatory structures. Journal of Phonetics 18. 299–320.
1990b. Tiers in articulatory phonology, with some implications for casual speech. In
Kingston and Beckman 1990. 341–346.
1992a. Articulatory phonology: an overview. Phonetica 49. 155–180.
1992b. Targeted and targetless schwas. Paper presented at International Phonologie-
tagung, Krems, Austria.
1992c. “Targetless” schwa: an articulatory analysis. In Docherty and Ladd 1992.
26–56.
1998. Competing constraints on intergestural coordination and self-organization of
phonological structures. Read at Current Trends in Phonology, II, Royaumont,
France.
Browman, C. P., L. M. Goldstein, D. N. Honoroff, A. Jebbour and E. Selkirk. 1998.
Gestural organizations underlying syllable structure in Tashlhiyt Berber. Read at
Current Trends in Phonology, II, Royaumont, France.
Brown, C. M. and P. Hagoort, eds. 1999. The neurocognition of language. Oxford:
Oxford University Press.
Byrd, D., A. Kaun, S. Narayanan and E. Saltzman. 2000. Phrasal signatures in articula-
tion. In Broe and Pierrehumbert. 70–87.
Byrd, D. and E. Saltzman. 1998. Intragestural dynamics of multiple prosodic boundaries.
Journal of Phonetics 26. 173–199.
Caplan, D. 1987. Neurolinguistics and linguistic aphasiology. Cambridge: Cambridge
University Press.
Carbary, T. J., J. P. Patterson and P. J. Snyder. 2000. Foreign accent syndrome following
a catastrophic second injury: MRI correlates, linguistic and voice pattern analyses.
Brain and Cognition 43. 78–85.
Choi, J. D. 1992. Phonetic underspecification and target interpolation: an acoustic study
of Marshallese vowel allophony. Working Papers in Phonetics 82. Los Angeles:
Phonetics Laboratory, UCLA.
Chomsky, N. 1959. Review of B. F. Skinner, Verbal Behavior. Language 35. 547–578.
1964. The logical basis of linguistic theory. Proceedings of the ninth international
congress of linguists, Cambridge, MA, August 27–31, 1962, ed. by H. G. Lunt.
914–978. The Hague: Mouton.
1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press.
1975. Reflections on language. New York: Pantheon.
1981. Lectures on government and binding. Dordrecht: Foris.
1986. Knowledge of language. New York: Praeger.
1995. The minimalist program. Cambridge, MA: MIT Press.
2000. New horizons in the study of language and mind. Cambridge: Cambridge Uni-
versity Press.
Chomsky, N. and M. Halle. 1968. The sound pattern of English. New York: Harper &
Row.
Clahsen, H. and M. Almazan. 1998. Syntax and morphology in Williams Syndrome.
Cognition 68. 167–198.
Clahsen, H., S. Bartke and S. Gollner. 1997. Formal features in impaired grammars:
a comparison of English and German SLI children. Essex Research Reports in
References 247
Clark, R. 1992. The selection of syntactic knowledge. Language Acquisition 2. 83–149.

Coulter, G. R. ed. 1993. Current issues in ASL Phonology. Phonetics and Phonology 3.
San Diego, CA: Academic Press.
Crain, S. 1991. Language acquisition in the absence of experience. Behavioral and Brain
Sciences 14. 597–650.
Crain, S. and C. McKee. 1986. The acquisition of structural restrictions on anaphora. In
Proceedings of North Eastern Linguistics Society 16, ed. by I. S. Berman, J. Choe
and J. McDonough. 94–110. Amherst, MA: University of Massachusetts, GLSA.
Crain, S. and R. Thornton. 1998. Investigations in Universal Grammar: a guide to
experiments on the acquisition of syntax and semantics. Cambridge, MA: MIT
Press.
Culicover, P. 1997. Principles and parameters: an introduction to syntactic theory.
Oxford: Oxford University Press.
Curtiss, S. 1988. Abnormal language acquisition and the modularity of language. In
Linguistics: the Cambridge survey, ed. by F. J. Newmeyer. Vol. II. 96–116.
Cambridge: Cambridge University Press.
Darnell, R. 1990. Edward Sapir. Berkeley: University of California Press.
Darwin, C. 1874. The descent of man. New York: Appleton. [Second edition, 1889.]
Dawkins, R. 1976. The selfish gene. Oxford: Oxford University Press.
de Boysson-Bardies, B. 1999. How language comes to children. Cambridge, MA: MIT
Press.
de Saussure, F. 1879. Mémoire sur le système primitif des voyelles dans les langues
indo-européennes. Leipzig: Tübner.
1974. Cours de linguistique générale. critical edn. Paris: Payot. [Prepared by Tullio
de Mauro.]
DeGraff, M., ed. 1999. Language creation and change: creolization, diachrony and
development. Cambridge, MA: MIT Press.
Dekkers, J., F. van der Leeuw and J. van deWeijer, eds. 2000. Optimality theory: phonol-
ogy, syntax and acquisition. Oxford: Oxford University Press.
deVilliers, J. and P. deVilliers. 1985. The acquisition of English. In The crosslinguistic
study of language acquisition, ed. by D. Slobin. 27–139. Hillside, NJ: Lawrence
Erlbaum.
Di Sciullo, A-M. and E. Williams. 1987. On the definition of word. Cambridge, MA:
MIT Press.
Dobzhansky, T. 1970. Genetics of the evolutionary process. New York: Columbia Uni-
versity Press.
Docherty, G. J. and D. R. Ladd, eds. 1992. Papers in laboratory phonology, Vol. II:
Gesture, segment, prosody. Cambridge: Cambridge University Press.
Dresher, B. E. 1999. Charting the learning path: cues to parameter setting. Linguistic
Inquiry 30. 27–67.
Dresher, B. E. and W. Idsardi. 1999. Prerequisites for a theory of diachronic adaptation.
Zeitschrift für Sprachwissenschaft 18. 212–215.
Dumézil, G. 1975. Le verbe oubykh: études descriptives et comparatives. Mémoires de
l’Académie des Inscriptions et Belles-lettres 1. Paris: Imprimerie Nationale.
Edwards, S. and R. Bastiaanse. 1998. Diversity in the lexical and syntactic abilities of
fluent aphasic speakers. Aphasiology 12(2). 99–117.
Ekwall, E. 1943. Studies on the genitive of groups in English. Lund: Gleerup.
248 References
Engdahl, E. 1985. Parasitic gaps, resumptive pronouns, and subject extractions.

Finger, S. 2000. Minds behind the brain. Oxford: Oxford University Press.
Flourens, M.-J.-P. 1824. Récherche expérimentale sur les propriétés et les fonctions du
système nerveux dans les animaux vertébrés. Paris: Ballières.
1846. Phrenology examined. Philadelphia: Hogan & Thompson. Translated from the
2nd French edition (1842) by C. de L. Meigs.
Fodor, J. D. 1998. Unambiguous triggers. Linguistic Inquiry 29. 1–36.
Fowler, C. A. 1990. Some regularities of speech are not consequences of formal rules:
Comments on Keating’s paper. In Kingston and Beckman 1990. 476–489.
Friedmann, N. and Y. Grodzinsky. 1997. Tense and agreement in agrammatic production:
pruning the syntactic tree. Brain and Language 56. 397–425.
Frith, U. 1991. Autism and Asperger Syndrome. Cambridge: Cambridge University
Press.
Fujimura, O. 1994. C/D model: a computational model of phonetic implementation.
In Language computations, ed. by E. S. Ristad. 1–20. American Mathematical
Society.
Gardner, R. A. and B. T. Gardner. 1969. Teaching sign language to a chimpanzee. Science
165. 664–672.
Gazzaniga, M. S. 1996. Conversations in the cognitive neurosciences. Cambridge, MA:
MIT Press.
Gibson, E. and K. Wexler. 1994. Triggers. Linguistic Inquiry 25. 407–454.
Gleitman, L. and E. Wanner. 1982. Introduction. In Language acquisition: the state of
the art, ed. by E. Wanner and L. Gleitman. 3–48. Cambridge: Cambridge University
Press.
Goldin-Meadow, S. and C. Mylander. 1990. Beyond the input given: the child’s role in
the acquisition of language. Language 66. 323–355.
Gopnik, M. 1990. Feature blindness: a case study. Language Acquisition 1(2). 139–164.
1997. Language deficits and genetic factors. Trends in Cognitive Sciences 1. 5–9.
Gopnik, M. and M. Crago. 1991. Familial aggregation of a developmental language
disorder. Cognition 39. 1–50.
Graff-Radford, N. R., W. E. Cooper, P. L. Colsher and A. R. Damasio. 1986. An unlearned
foreign “accent” in a patient with aphasia. Brain and Language 28. 86–94.
Greenberg, J. H. 1963. Some universals of grammar with particular reference to the
order of meaningful elements. In Universals of language, ed. by J. H. Greenberg.
73–112. Cambridge, MA: MIT Press.
Grimm, J. 1848. Geschichte der deutschen Sprache. Vol. I. Leipzig: Weidmannsche
Buchhandlung.
Grimshaw, J. 1990. Argument structure. Cambridge, MA: MIT Press.
Grodzinsky, Y. 2000. The neurology of syntax: language use without Broca’s area.
Behavioral and Brain Sciences 23. 1–21.
Guy, G. R. 1991a. Contextual conditioning in variable lexical phonology. Language
Variation and Change 3. 223–239.
1991b. Explanation in variable phonology: an exponential model of morphological
constraints. Language Variation and Change 3. 1–22.
Hagoort, P., C. M. Brown and L. Osterhaut. 1999. The neurocognition of syntactic
processing. In Brown and Hagoort 1999. 273–316.
References 249
Hale, K. 1976. The adjoined relative clause in Australia. In Grammatical categories

in Australian languages, ed. by R. M. W. Dixon. 78–105. Canberra: Australian
Institute of Aboriginal Studies.
Halle, M. 1957. On the phonetic rules of Russian. Unpublished paper read to the Annual
Meeting of the Linguistic Society of America, Chicago.
1959. The sound pattern of Russian. The Hague: Mouton.
Hamburger, H. and S. Crain. 1984. Acquisition of cognitive compiling. Cognition 17.
85–136.
Harris, Z. 1951. Methods in structural linguistics. Chicago: University of Chicago Press.
Haspelmath, M. 1999a. Are there principles of grammatical change? Journal of
1999b. Optimality and diachronic adaptation. Zeitschrift für Sprachwissenschaft 18.
180–205.
1999c. Why is grammaticalization irreversible? Linguistics 37. 1043–1068.
Hauser, M. D. and M. Konishi, eds. 1999. The design of animal communication.
Cambridge, MA: MIT Press.
Herbert of Cherbury. 1624/1937. De veritate. University of Bristol Studies Number 6.
Translated by M. H. Carré.
Hickok, G. and D. Poeppel. 2000. Towards a functional neuroanatomy of speech per-
ception. Trends in Cognitive Sciences 4. 131–138.
Hockett, C. F. 1947. Problems of morphemic analysis. Language 23. 321–43.
Hoenigswald, H. 1963. On the history of the comparative method. Anthropological
Hombert, J.-M. 1976. Consonant types, vowel height and tone in Yoruba. UCLA Working
Papers in Phonetics 33. 40–54.
Hopper, P. J. and E. Traugott. 1993. Grammaticalization. Cambridge: Cambridge
University Press.
Hornstein, N. and D. Lightfoot. 1981. Introduction. In Explanation in linguistics: the
logical problem of language acquisition, ed. by N. Hornstein and D. Lightfoot.
9–31. London: Longman.
Householder, F. W. 1965. On some recent claims in phonological theory. Journal of
Hovav, M. R. and B. Levin. 1992. -er nominals: implications for the theory of argument
structure. In Syntax and the lexicon, ed. by T. Stowell and E. Wehrli. Syntax &
Semantics 26. 127–153. San Diego: Academic Press.
Hubel, D. 1978. Vision and the brain. Bulletin of the American Academy of Arts and
Sciences 31(7). 28.
Hubel, D. and T. Wiesel. 1962. Receptive fields, binocular interaction and functional
architecture in the cat’s visual cortex. Journal of Physiology 160. 106–154.
Hughes, H. C. 1999. Sensory exotica. Cambridge, MA: MIT Press.
Ingham, R. 1998. Tense without agreement in early clause structure. Language Acqui-
sition 7. 51–81.
Jackendoff, R. S. 1975. Morphological and semantic regularities in the lexicon.
Language 51. 639–71.
1983. Semantics and cognition. Cambridge, MA: MIT Press.
1990. Semantic structures. Cambridge, MA: MIT Press.
1994. Patterns in the mind. New York: Basic Books.
250 References
Jaeger, J. J., A. H. Lockwood, D. L. Kemmerer, R. D. Van Valin, B. W. Murphy and

H. G. Khalak. 1996. A positron emission tomography study of regular and irregular
verb morphology in English. Language 72. 451–497.
Janda, R. D. 2001. Beyond “pathways” and “unidirectionality”: On the discontinuity
of language transmission and the counterability of grammaticalization. Language
Sciences 23. 265–340.
Jespersen, O. 1909. Progress in language. London: Swan Sonnenschein. [Second
edition.]
Joanisse, M. and M. Seidenberg. 1998. Specific language impairment: a deficit in gram-
mar or processing? Trends in Cognitive Sciences 2. 240–247.
Johnson, K., E. Flemming and R. Wright. 1993. The hyperspace effect: phonetic targets
are hyperarticulated. Language 69. 505–528.
Joseph, B. 2001. Is there such a thing as grammaticalization? Language Sciences 23.
163–186.
Jusczyk, P. 1997. The discovery of spoken language. Cambridge: MIT Press.
Kagaya, R. 1971. Laryngeal gestures in Korean stop consonants. Annual Bulletin,
Research Institute of Logopedics and Phoniatrics, University of Tokyo 5. 15–23.
Keating, P. 1988. Underspecification in phonetics. Phonology 5. 275–292.
Kegl, J. A., A. Senghas and M. Coppola. 1999. Creation through contact: sign language
emergence and sign language change in Nicaragua. In DeGraff 1999. 179–237.
Keller, E. 1987. The cortical representation of motor processes of speech. In Motor and
sensory processes of language, ed. by E. Keller and M. Gopnik. 125–162. Hillsdale,
NJ: Lawrence Erlbaum.
Kenstowicz, M. 1994. Phonology in generative grammar. Cambridge: Blackwell.
Kingston, J. and M. Beckman, eds. 1990. Papers in laboratory phonology, Vol. I:
Between the grammar and the physics of speech. Cambridge: Cambridge
University Press.
Kingston, J. and R. L. Diehl. 1994. Phonetic knowledge. Language 70. 419–454.
Kiparsky, P. 1982. Word formation and the lexicon. In Proceedings of the 1982 Mid-
America linguistics conference, ed. by F. Ingemann. 3–29. Lawrence: University
of Kansas.
1984. On the lexical phonology of Icelandic. In Nordic prosody, Vol. III, ed. by
C.-C. Elert, I. Johansson and E. Strangert. 135–164. Stockholm: University of
Umeå (Almqvist and Wiksell).
1993. Variable rules. Unpublished paper read at Rutgers Optimality Workshop.
1996. The shift to head-initial VP in Germanic. In Studies in comparative Germanic
syntax, ed. by H. Thráinsson, S. J. Epstein and S. Peters. 140–179. Dordrecht:
Kluwer.
Kisseberth, C. 1970a. On the functional unity of phonological rules. Linguistic Inquiry
1. 291–306.
1970b. Vowel elision in Tonkawa and derivational constraints. In Studies presented to
Robert B. Lees by his students, ed. by J. Sadock and A. Vanek. 109–138. Edmonton:
Linguistic Research.
Klima, E. and U. Bellugi. 1979. The signs of language. Cambridge, MA: Harvard
University Press.
Kosslyn, S. M., M. S. Gazzaniga, A. M. Galaburda and C. Rabin. 1999. Hemispheric
specialization. In Fundamental neuroscience, ed. by M. J. Zigmond, F. E. Bloom,
S. C. Landis, J. L. Roberts and L. R. Squire. 1521–1542. San Diego: Academic
Press.
References 251
Kroch, A. 1989. Reflexes of grammar in patterns of language change. Language Varia-

tion and Change 1. 199–244.
Kroodsma, D. E. and E. H. Miller, eds. 1996. Ecology and evolution of acoustic com-
munication in birds. Ithaca, NY: Cornell University Press.
Kuhl, P. K. 1999. Speech, language and the brain: innate preparation for learning. In
Hauser and Konishi 1999. 419–450.
Kuipers, A. H. 1960. Phoneme and morpheme in Kabardian. The Hague: Mouton.
Kurowski, K. M., S. E. Blumstein and M. Alexander. 1996. The foreign accent syndrome:
a reconsideration. Brain and Language 54. 1–25.
Kutas, M., K. D. Federmeier and M. I. Sereno. 1999. Current approaches to mapping
language in electromagnetic space. In Brown and Hagoort 1999. 359–392.
Lakoff, R. T. 1972. Another look at drift. In Linguistic change and generative theory, ed.
by R. P. Stockwell and R. Macauley. 172–198. Bloomington: Indiana University
Press.
Langacker, R. W. 1969. Pronominalization and the chain of command. In Modern stud-
ies in English, ed. by D. Reibel and S. Schane. 160–186. Englewood Cliffs, NJ:
Prentice-Hall.
Lasnik, H. 1999. Minimalist analysis. Oxford: Blackwell.
Lass, R. 1980. On explaining linguistic change. Cambridge: Cambridge University
Press.
1997. Historical linguistics and language change. Cambridge: Cambridge University
Press.
Laughren, M. 1988. Toward a lexical representation of Warlpiri verbs. In Thematic rela-
tions, ed. by W. Wilkins. Syntax and Semantics 21. 215–242. San Diego: Academic
Press.
Lefebvre, C. 1998. Creole genesis and the acquisition of grammar. Cambridge:
Cambridge University Press.
Lehiste, I. 1970. Suprasegmentals. Cambridge, MA: MIT Press.
Levy, Y. and G. Kavé. 1999. Language breakdown and linguistic theory: a tutorial
overview. Lingua 107. 95–113.
Liberman, A. M. and I. G. Mattingly. 1985. The motor theory of speech perception
revised. Cognition 21. 1–36.
Lieberman, P. 2000. Human language and our reptilian brain: the subcortical bases of
speech, syntax and thought. Cambridge, MA: Harvard University Press.
Lightfoot, D. W. 1989. The child’s trigger experience: Degree-0 learnability. Behavioral
and Brain Sciences 12. 321–334.
1991. How to set parameters: arguments from language change. Cambridge, MA:
MIT Press.
1994. Degree-0 learnability. In Syntactic theory and first language acquisition:
crosslinguistic perspectives, ed. by B. Lust, G. Hermon and J. Kornfilt. Vol. II.
453–171. Hillsdale, NJ: Lawrence Erlbaum.
1999. The development of language: acquisition, change and evolution. Oxford:
Blackwell.
ed. 2002. Syntactic effects of morphological change. Oxford: Oxford University Press.
Lindblom, B. 1990. Explaining phonetic variation: a sketch of the H and H theory. In
Speech production and speech modelling, ed. by W. J. Hardcastle and A. Marchal.
403–439. Dordrecht: Kluwer Academic.
Locke, J. 1993. The child’s path to spoken language. Cambridge, MA: Harvard Univer-
sity Press.
252 References
MacDonnell, A. 1916. A Vedic grammar for students. Oxford: Oxford University Press.
Manzini, R. and K. Wexler. 1987. Parameters, binding theory, and learnability. Linguistic
Inquiry 18. 413–444.
Marantz, A., Y. Miyashita and W. O’Neil, eds. 2000. Image, language, brain. Cambridge,
MA: MIT Press.
Marchand, H. 1969. The categories and types of present-day English word formation.
München: Beck. [Second edition.]
Marler, P. 1970. Birdsong and human speech: could there be parallels? American
Scientist 58. 669–674.
1991. Song-learning behavior: the interface with neuroethology. Trends in Neuro-
sciences 14. 199–206.
1999. On innateness: are sparrow songs “learned” or “innate”? In Hauser and Konishi
1999. 293–318.
Marr, D. 1982. Vision. San Francisco: Freeman.
Matthei, E. 1982. The acquisition of prenominal modifier sequences. Cognition 11.
201–332.
Matthews, G. H. 1955. A phonemic analysis of a Dakota dialect. International Journal
of American Linguistics 21. 56–59.
Mattingly, I. G. and M. Studdert-Kennedy, eds. 1991. Modularity and the motor theory
of speech perception. Hillsdale, NJ: Lawrence Earlbaum.
Mayeux, R. and E. R. Kandel. 1991. Disorders of language: the aphasias. In Principles
of neural science, ed. by E. Kandel, J. H. Schwartz and T. M. Jessel. 839–851.
New York: Elsevier.
McCarthy, J. J. 1981. A prosodic theory of non-concatenative morphology. Linguistic
Inquiry 12. 373–418.
McCawley, J. D. 1999. Why surface syntactic structure reflects logical structure as much
as it does, but only that much. Language 75. 34–62.
McGurk, H. and J. MacDonald. 1976. Hearing lips and seeing voices. Nature 264.
746–748.
Medawar, P. 1967. The art of the soluble. London: Methuen.
Monrad-Kröhn, G. H. 1947. Dysprosody or altered “melody of language”. Brain 70.
405–415.
Morris Jones, J. 1913. A Welsh grammar. Oxford: Clarendon.
Mustanoja, T. 1960. A Middle English syntax. Helsinki: Société Néophilologique.
Nash, D. 1986. Topics in Warlpiri grammar. New York: Garland.
Nespor, M. and I. Vogel. 1986. Prosodic phonology. Dordrecht: Foris.
Newmeyer, F. J. 1998. Language form and language function. Cambridge, MA: MIT
Press.
Newport, E. L. 1999. Reduced input in the acquisition of signed languages: contributions
to the study of creolization. In DeGraff 1999. 161–178.
Niyogi, P. 2002. The computational study of diachronic linguistics. In Lightfoot 2002.
351–65.
Niyogi, P. and R. Berwick. 1997. A dynamical systems model for language change.
Complex Systems 11. 161–204.
Nunes, J. 1995. Linearization of chains and sideward movement. PhD thesis. University
of Maryland. College Park, MD.
Nunnally, T. 1985. The syntax of the genitive in Old, Middle, and Early Modern English.
PhD thesis. University of Georgia.
References 253
Ojemann, G., J. Ojemann, E. Lettich and M. Berger. 1989. Cortical language organization
in left, dominant hemisphere. Journal of Neurosurgery 71. 316–326.
O’Neil, W. 1978. The evolution of the Germanic inflectional systems: a study in the
causes of language change. Orbis 27. 248–285.
Orešnik, J. and M. Pétursson. 1977. Quantity in modern Icelandic. Arkiv för Nordisk
Filologi 92. 155–171.
Paradis, C. and D. LaCharité. 1997. Preservation and minimality in loanword adaptation.
Journal of Linguistics 33. 379–430.
Paul, H. 1880. Prinzipien der Sprachgeschichte. Tübingen: Niemeyer.
Payne, D. L. 1998. Maasai gender in typological and applied perspective. Read at 29th
Annual Conference on African Linguistics, Yale University.
Pepperberg, I. M. 2000. The Alex studies: cognitive and communicative abilities of grey
parrots. Cambridge, MA: Harvard University Press.
Perlmutter, D. M. 1991. The language of the Deaf. New York Review of Books 38(6).
65–72. [Review of Sacks 1989.]
Petitto, L. A. and P. F. Marentette. 1991. Babbling in the manual mode: evidence for the
ontogeny of language. Science. 251. 1493–1496.
Petitto, L. and M. Seidenberg. 1979. On the evidence for linguistic abilities in signing
apes. Brain and Language 8. 162–183.
Phillips, C. 2001. Levels of representation in the electrophysiology of speech perception.
Cognitive Science 25. 711–731.
Piaget, J. and B. Inhelder. 1968. The psychology of the child. London: Routledge.
Piattelli-Palmarini, M., ed. 1980. Language and learning: the debate between Jean
Piaget and Noam Chomsky. London: Routledge and Kegan Paul.
1986. The rise of selective theories: a case study and some lessons from immunol-
ogy. In Language learning and concept acquisition: foundational issues, ed. by
W. Demopoulos and A. Marras. 117–130. Norwood, NJ: Ablex.
1989. Evolution, selection, and cognition: from “learning” to parameter setting in
biology and the study of language. Cognition 31. 1–44.
Pierce, A. 1992. Language acquisition and syntactic theory: a comparative analysis of
French and English child grammars. Dordrecht: Kluwer.
Pierrehumbert, J. 1990. Phonological and phonetic representation. Journal of Phonetics
18. 375–394.
Pierrehumbert, J. and D. Talkin. 1992. Lenition of /h/ and glottal stop. In Docherty and
Ladd 1992. 90–117.
Pinker, S. 1994. The language instinct. New York: William Morrow.
Poeppel, D. 1996. A critical review of PET studies of phonological processing. Brain
and Language 55. 317–351.
Poeppel, D. and A. Marantz. 2000. Cognitive neuroscience of speech processing. In
Marantz et al. 2000. 29–50.
Poeppel, D. and K. Wexler. 1993. The full competence hypothesis of clause structure in
early German. Language 69. 1–33.
Poizner, H., E. S. Klima and U. Bellugi. 1987. What the hands reveal about the brain.
Cambridge, MA: MIT Press.
Premack, D. 1978. Chimpanzee problem-solving: a test for comprehension. Science
202. 532–535.
1980. Representational capacity and accessibility of knowledge: the case of chim-
panzees. In Piattelli-Palmarini 1980. 205–221.
254 References
1990. Words: What are they, and do animals have them? Cognition 37. 197–212.
Prince, A. and P. Smolensky. 1993. Optimality theory: constraint interaction in generative
grammar. Manuscript, Rutgers University and University of Colorado.
Purves, D., G. J. Augustine, D. Fitzpatrick, L. C. Katz, A.-S. LaMantia and J. O.
McNamara, eds. 1997. Neuroscience. Sunderland, MA: Sinauer.
Rask, R. K. 1818. Undersøgelse om det gamle Nordisk eller Islandske sprogs oprindelse.
Copenhagen: Gyldendals.
Reinhart, T. 1976. The syntactic domain of anaphora. PhD thesis. Massachusetts Institute
of Technology.
Rizzi, L. 1982. Violations of the wh-island constraint and the subjacency condition. In
Issues in Italian syntax, ed. by L. Rizzi. 49–76. Dordrecht: Foris.
1990. Relativized minimality. Cambridge, MA: MIT Press.
Roberts, I. 1993. A formal account of grammaticalization in the history of Romance
futures. Folia Linguistica Historica 13. 219–258.
Ross, J. R. 1967. On the cyclic nature of English pronominalization. To honor Roman
Jakobson. The Hague: Mouton.
1969. Auxiliaries as main verbs. In Studies in philosophical linguistics, ed. by
W. Todd. Vol. I. Evanston: Great Expectations.
Rugg, M. D. 1999. Functional neuroimaging in cognitive neuroscience. In Brown and
Hagoort 1999. 15–36.
Sacks, O. 1989. Seeing voices. Berkeley: University of California Press.
Sapir, Edward. 1921. Language. New York: Harcourt, Brace & World.
1925. Sound patterns in language. Language 1. 37–51.
1929. The status of linguistics as a science. Language 5. 207–214.
1938. Why cultural anthropology needs the psychiatrist. Psychiatry 1. 7–12.
1994. The psychology of culture. Berlin: Mouton de Gruyter. Reconstructed and edited
by Judith T. Irvine.
Savage-Rumbaugh, E. S. 1986. Ape language: from conditioned response to symbol.
New York: Columbia University Press.
1987. Communication, symbolic communication, and language: reply to Seidenberg
& Petitto. Journal of Experimental Psychology: General 116. 288–292.
Savage-Rumbaugh, E. S., K. McDonald, R. A. Sevcick, W. D. Hopkins and E. Rupert.
1986. Spontaneous symbol acquisition and and communicative use by pygmy
chimpanzees (Pan Paniscus). Journal of Experimental Psychology: General 115.
211–235.
1993. Language comprehension in ape and child. Monographs of the Society for
Research in Child Development 58. 1–221.
Savage-Rumbaugh, E. S., S. G. Shanker and T. J. Taylor. 1998. Apes, language, and the
human mind. New York: Oxford University Press.
Schleicher, A. 1848. Über die Bedeutung der Sprache für die Naturgeschichte des
Menschen. Weimar: Hermann-Böhlau.
1861–62. Compendium der vergleichenden Grammatik der indogermanischen
Sprachen. Weimar: Hermann-Böhlau.
1863. Die Darwinische Theorie und die Sprachwissenschaft. Weimar: Hermann-
Böhlau.
Seidenberg, M. and L. Petitto. 1987. Communication, symbolic communication, and
language: comment on Savage-Rumbaugh et al. Journal of Experimental Psychol-
ogy: General 116. 279–287.
References 255
Shaywitz, B., S. Shaywitz, K. Pugh, T. Constable, P. Skudlasrski, R. Fulbright, R. Bronen,

J. Fletcher, D. Shankweiler, L. Katz and J. Gore. 1995. Sex differences in the
functional organization for language. Nature 373. 607–609.
Shlonsky, U. 1988. Complementizer cliticization in Hebrew and the ECP. Natural
Language and Linguistic Theory 6. 191–206.
Sieratzki, J. and B. Woll. in press. Toddling into language: precocious language devel-
opment in motor-impaired children with spinal muscular atrophy. Lingua.
Sievers, E. 1881. Grundzüge der Phonetik. Leipzig: Breitkopf & Hartel.
Smith, N. V. in press. Dissociation and modularity. In Mind, brain and language:
multidisciplinary perspectives, ed. by M. T. Banich and M. Mack. Hillsdale, NJ:
Lawrence Erlbaum.
Smith, N. V. and I. M. Tsimpli. 1995. The mind of a savant: language-learning and
modularity. Oxford: Blackwell.
Solan, L. 1983. Pronominal reference: child language and the theory of grammar.
Dordrecht: Reidel.
Sperry, R. 1968. Plasticity of neural maturation. Developmental Biology Supplement 2.
306–27.
Stevens, K. and M. Halle. 1967. Remarks on analysis by synthesis and distinctive fea-
tures. In Models for the perception of speech and visual form, ed. by W. Whaten-
Dunn. 88–102. Cambridge, MA: MIT Press.
Supalla, S. 1990. Segmentation of manually coded English: problems in the mapping of
English in the visual/gestural mode. PhD thesis. University of Illinois.
Tavakolian, S. 1978. Children’s comprehension of pronominal subjects and missing
subjects in complicated sentences. University of Massachusetts occasional papers
4: Papers in the structure and development of child language, ed. by H.Goodluck
and L. Solan. 37–83. Amherst, MA: University of Massachusetts Graduate Student
Linguistic Association.
Terrace, H. S., L. A. Pettito, R. J. Sanders and T. G. Bever. 1979. Can an ape create a
sentence? Science 206. 891–902.
Thornton, R. 1994. Children’s negative questions: a production/comprehension asym-
metry. In Proceedings of Eastern States Conference on Linguistics 1994, ed. by
J. Fuller, H. Han and D. Parkinson. 306–317. Ithaca: Cornell University Depart-
ment of Linguistics.
1995. Referentiality and wh-movement in child English: juvenile dlinkuency.
Language Acquisition 4. 139–175.
Tomblin, J. B. 1997. Epidemiology of specific language impairment. In The inheri-
tance and innateness of grammars, ed. by M. Gopnik. 91–110. New York: Oxford
University Press.
Trubetzkoy, N. S. 1939. Grundzüge der Phonologie. Travaux du cercle linguistique de
Prague.
Uylings, H. B. M., L. I. Malofeeva, I. N. Bogolepova, K. Amunts and K. Zilles. 1999.
Broca’s language area from a neuroanatomical and developmental perspective. In
Brown & Hagoort 1999. 319–336.
van der Lely, H. K. J. 1996. Specifically language impaired and normally developing
children: verbal passive vs. adjectival passive sentence interpretation. Lingua 98.
243–272.
van Gelderen, E. 1997. Verbal agreement and the grammar behind its “breakdown”:
minimalist feature checking. Tübingen: Max Niemeyer.
256 References
Vennemann, T. 1975. An explanation of drift. In Word order and word order change,
ed. by C. N. Li. 269–302. Austin: University of Texas Press.
Wallman, J. 1992. Aping language. Cambridge: Cambridge University Press.
Warner, A. R. 1995. Predicting the progressive passive: parametric change within a
lexicalist framework. Language 71. 533–557.
1997. The structure of parametric change and V movement in the history of English.
In Parameters of morphosyntactic change, ed. by A. van Kemenade and N. Vincent.
380–393. Cambridge: Cambridge University Press.
Watkins, C. 1976. Toward Proto-Indo-European syntax: problems and pseudo-problems.
In Diachronic syntax, ed. by S. Steever, C. Walker and S. Mufwene. 305–326.
Chicago: Chicago Linguistic Society.
Weverink, M. 1989. The subject in relation to inflection in child language. Master’s
thesis. Utrecht University.
Wexler, K. 1994. Optional infinitives, head movement, and the economy of derivations.
In Verb movement, ed. by D. W. Lightfoot and N. Hornstein. 305–350. Cambridge:
Cambridge University Press.
Wexler, K., M. Rice and P. Cleave. 1995. Specific language impairment as a period of
extended optional infinitives. Journal of Speech and Hearing Research 38. 850–
863.
Williams, H. and F. Nottebohm. 1985. Auditory responses in avian vocal motor neurons:
a motor theory for song perception in birds. Science 229. 279–282.
Wright, J. 1910. Grammar of the Gothic language. Oxford: Clarendon.
Yang, C. D. 2002. Grammar competition and language change. In Lightfoot 2002.
367–380.
Zwicky, A. M. and G. K. Pullum. 1983. Cliticization vs. inflection: English n’t. Language
59. 502–13.
Index
a- adjectives 154 birdsong 210–14, 220

Acquired Foreign Accent Disorder Bishop, D. V. M. 237
128–9, 128 n5 Bittner, M. 145–7
agent nouns 149, 153 biuniqueness condition 77–8, 80–1, 83
Allen, C. 177 n5, 179, 181 Blake, B. 62
Almazan, M. 238 Bloch, B. 86, 148 n9
anaphora 195–6, 195 n4 Bloomfield, L. 12, 13–14, 85–6, 112, 132
Anderson, S. R. 76, 88, 112, 118, 121, Boas, F. 4, 12, 148, 155
218 n4, 219 n6 Boatman, D. et al. 129 n6
Andersson, A.-B. 201 Borer, H. 190
apes 219–20, 219 n5–6 borrowing 93–6, 109
aphasia see language deficits Bowerman, M. F. 190
apophony 138 Boyce, S. E. 120
Arabic 138, 148, 148 n10, 149 brain
Banni-Hassan 60 aggregate field view 222, 223–4
Saudi 118 birds 212, 213–14, 213 n13
Aronoff, M. 132, 133, 134, 154 body maps 227, 228f
Articulatory Phonology 124–5 Broca’s area 227, 238
Asperger’s Syndrome 237 and cognition 221–2
autism 237 electroencephalograms (EEG) 233
auxiliary verbs see modal auxiliaries Event Related Potentials (ERP)
233–4
babbling 207 functional anatomy 225–36, ix–x
Baker, M. 218 functional MRI (fMRI) 232–3, 234
Banni-Hassan Arabic 60 language localization 208, 213, 224–5,
Barbour, J. 162 224 n7, 236
Bastiaanse, R. 238 Magnetic Resonance Imaging (MRI)
Bates, E. A. 34, 35 230
Bauer, L. 159–60 Magneto-EncelphaloGraphy (MEG) 234
Bazell, C. E. 148 n9, 155 materialist view 222, 223f, 224
be 168–71, 172, 173 neural activity 230–1, 235, 240–1
see also is/’s non-invasive imaging 230–6
Beard, R. 148, 155 Positron Emission Tomography (PET)
Beckman, M. 126 231–2, 233
behaviorism 13–14, 15, 16 Wernicke’s area 227
Bell-Berti, F. 120 see also language deficits
Bellugi, U. 219 n5, 225, 238 Bresnan, J. 53
Berthier, M. L. et al. 128 n5 Broca, P. 226–7
Bickerton, D. 203 Broca’s area 227, 238
bilingualism 209 Browman, C. 121, 123, 124–5, 127
binding theory 33 n6, 33 n7, 190, 195–6, Brown, C. M. 225 n9, 233
195 n4, 200–2, 200 n7 Byrd, D. 126 n4
257
258 Index
Caplan, D. 238 and I-language 238–9

Caramazza, A. 243 incorporation 52–5
case merger 43–5
of 180–1, 180 n6, 182 that deletion 45–7
Case theory 63–5 derived nominals 142–3, 148, 149
ergative languages 143, 143 n6 determiner phrases (DPs) 43 n1, 63–5,
Middle English 175, 178, 179, 180, 181–2 175–6, 177
morphological/abstract links 65–6 deVilliers, J. and P. 190
morphological case 61–3 devoicing 98–9
Old English 62, 176–8, 180 n6, 181, Di Sciullo, A.-M. 132
182 n7 Diehl, R. L. 113 n1
possessive 179 displacement 44
scope generalization 147 Down’s Syndrome 237
syntactic effects of loss 175–82 DPs see determiner phrases
thematic roles 175–6, 178, 180, 181 Dresher, B. E. 204, 205
see also genitives drift 158–9
Choi, J. D. 121 Dumézil, G. 150
Chomsky, N. 15–16, 34, 42, 89, 102, 115, Dutch 192–4
202, 203, 241, 242, vi, xii, xv
Circassian languages 120, 150, 150 n11 E-language 12–15, 239, 241, xv
cladograms 7–8 Edwards, J. 126
Clahsen, H. 238 Edwards, S. 238
Clark, R. 202–3 Ekwall, E. 176
Cleave, P. 194 electroencephalograms (EEG) 233
clitics 27–8, 30–1, 187–8 Elman, J. L. 34, 35
cognition 37–8, 221–2, xiii–xiv epenthesis 98–9
cognitive science 16–17, 241–3 -er nominals 142–3, 148, 149, 153
Combe, G. 223f ergative languages 143, 143 n6
consonant clusters 97–9, 101, 106–9 “Eskimo” 135 n1
constraints 101 see also West Greenlandic
Eval 104, 105 Event Related Potentials (ERP) 233–4
faithfulness 102–4
Fixed Subject 53 faithfulness 102–4
Gen 104, 105 Federmeier, K. D. 233, 234
harmony principles 104–5 Finger, S. 224
markedness 101, 102–4 Finnish 62, 147
richness of the base 105 Fischer, O. 181
Coppola, M. 204 n8 Fixed Subject Constraint 53
copy theory of movement 43–5 Flemish, West 57–8
Correspondence Fallacy 148 n9 Flemming, E. 123
Crago, M. 237 Fletcher, J. 126
Crain, S. 38–9, 188–9, 196, 198 Flourens, M.-J.-P. 222
creole languages 203, 218 fMRI (functional Magnetic Resonance
Culicover, P. 178 Imaging) 232–3, 234
Curtiss, S. 209 n10, 238 Fodor, J. D. 204
Fowler, C. A. 118, 119, 120
Dahl, O. 201 French
Dakota 84–5 binding theory 200–1
Darwin, C. 11 borrowing into Fula 93–7, 93 n1
de Boysson-Bardies, B. 206 branching system 159–60
de Saussure, F. 12, 13, 74, 76, 135 deletion 56 n9, 57
deafness see signing and signers negative markers 160
degree-0 learnability 199–202, 204 pouvoir 163–4
deletion verb movement 192, 194, 205
deleting copies 47–52 Friedmann, N. 238
diversity 55–61 Frith, U. 237
Index 259
Fujimura, O. 125 I-language 15–17, 75, 83–91, 238–9, 241,

Fula 93–7, 93 n1 xii, xv
Icelandic 138, 139–41
Gall, F. 222, 223f, 224 inflection 174
Gazzaniga, M. S. 243 Ingham, R. 238
gender 150–1 Inhelder, B. 238
genitives International Phonetic Alphabet (IPA) 124, 130
endings 160–1 is/’s 18–19, 25–32, 38–9
group genitive 179, 182 Italian 59–60, 200–1
split 175, 176–8, 181, 182, 182 n7
German 78, 136, 193 Jackendoff, R. S. 70, 136 n2, 137 n3, 155
Gibson, E. 202, 203 Jaeger, J. J. et al. 232
Gleitman, L. 190 Janda, R. D. 161, 161 n2
Goldin-Meadow, S. 204 Jespersen, O. 11–12, 176, 178, 180
Goldstein, L. 121, 123, 124–5, 127 Joanisse, M. 237
Gopnik, M. 237 Johnson, K. 123
grammars Jones, M. 41
change 161–2 Joseph, B. 161
generative theories 238–9, 240
historical view 8–9 Kabardian 120
as language organ 35–7, 40 Kagaya, R. 119
maturation 190 Kavé, G. 237
nature of 23–5 Keating, P. 119
real-time acquisition 38–40 Kegl, J. A. 204 n8
rules 84, 89, 90 Keller, E. 129
see also Universal Grammar (UG) Kingston, J. 113 n1
grammaticalization 160–1 Kiparsky, P. 123 n3, 161
English auxiliary verbs 162–75 Klima, E. 219 n5
Great Vowel Shift 9–10 Klima, E. S. 225
Greenberg, J. H. 42, 158 Koopman, W. 181
Greenlandic, West 136, 145–6, 147 Korean 80–2, 119–20
Grimm, J. 11 Krakow, R. A. 120
Grimm’s Law 9–10, 11 Kroch, A. 166
Grimshaw, J. 142 Kroodsma, D. E. 210 n11
Grodzinsky, Y. 23, 238 Kuipers, A. H. 120
Guy, G. R. 123 n3 Kutas, M. 233, 234
Kwakw’ala (Kwakiutl) 102–3
Hackel, E. 8, 129 n6
Hagoort, P. 225 n9, 233 labialized velars 102–3
Hale, K. 44 n2 Lakoff, R. T. 158, 159
Halle, M. 79 n4, 83–5, 87–8, 89, 90, language
102, xvii as brain function 221–36, 239–40
Hamburger, H. 198 and culture 217–18
Harris, Z. 86 diversity 55–61, 55 n8
Haspelmath, M. 160 n1, 161 n2 functional anatomy 225–36
have 173 identity 14–15
Hebrew 58–9, 138, 148 as knowledge 16, 92, 111
Herbert of Cherbury 241 localization in the brain 208, 213,
historical linguistics 5–12 224–5, 224 n7, 236
Hockett, C. F. 85, 137 organic basis 216–19, 221
Hombert, J.-M. 117, 117f and other species 219–21, 219 n5–6
Hornstein, N. 196–7, 197 n6, 198 as particular faculty 236–8
Householder, F. W. 86, 89 processing 216–17, 217 n2
Hovav, M. R. 142 language acquisition 1–3
Hubel, D. 37, 38 analytical triplet 36–8
Hughes, H. C. 220 binding theory 195–6, 200–2
260 Index
language acquisition (cont.) lexicon 105, 132–5

cue-based acquisition 204–6 derivational relations 153–4, 155–6
degree-0 learnability 199–202, 204 mental lexicon 171, 172–3, 174
evaluation metric 202, 203 optimization 105–6
experimental technique 194–8 organization 155–6
Fitness Metric 203 productivity 152–4
learning path 205, 207 word classes 153
linguistic genotype 15–16, 22–3, 31, 33–4, words and “morphemes” 134–52
36–7 LF (logical form) 115
optional infinitives 192–4 Liberman, A. M. 113
phrase structure 196–8 Liebermann, P. 240
poverty of the stimulus 18–21, 25–33, Lightfoot, D. W. 161 n3, 167, 168, 175 n4,
34–6 182 n7, 196–7, 197 n6, 198, 199,
primary linguistic data (PLD) 162, 167, 168, 201–2, 204, 205, 218
199 Lindblom, B. 123
real-time acquisition of grammars 38–40 linguistic genotype see Universal Grammar
sequence-of-tenses phenomenon 200 (UG)
sound patterns 206–9 linguistic knowledge 16, 92, 111
speech perception 207–9 see also phonological knowledge
syntax 186–206 linguistic relativity 4
trigger 2, 36–7, 198–206 linguistic savants 237–8
Triggering Learning Algorithm (TLA) 202, linguistics
203 as discipline x–xi, xiv
Truth Value Judgement tasks 189, 196 historical development xi–xiii
UG principles 186–90 as history 5–12
wh-phrases 45, 191–2 and mind/brain 3–5
language change as study of E-language 12–15
analogical change 9 as study of I-language 15–17
case loss 175–82 Locke, J. 127
chaos 183–5 logical form (LF) 115
drift 158–9
English auxiliary verbs 162–75 Maasai 150–1
evolutionary explanation 159–60 MacDonald, J. 235 n11
grammar change 161–2 MacDonnell, A. 41
grammaticalization 160–1 Magnetic Resonance Imaging (MRI) 230
long-term directionality 157–61 functional (fMRI) 232–3, 234
sound change 6, 7, 9–11 Magneto-EncephaloGraphy (MEG) 234
spread through populations 166 Manzini, R. 190
time 162 Marantz, A. 234
word order typologies 158–9 Marchand, H. 154
language deficits 2, 3, 128–9, 222, 224–5, Marentette, P. F. 207
226–30, 237–8, 241 markedness 101, 102–4
see also Specific Language Impairment Marler, P. 210–11, 210 n11, 211 n12, 220
(SLI) Marr, D. 216 n1
language organ 3, 16, 221, ix Marshallese 121
langue 12, 13 Matthei, E. 197
Laplace, P. S. 183 Matthews, G. H. 84
Lasnik, H. 173 Mattingley, I. G. 113
Lass, R. 159, 184 McGurk, H. 235 n11
Latin 159–60 McKee, C. 196
Laughren, M. 143, 145 meaning 136, 136 n2, 137
Lehiste, I. 116 n2 Medawar, P. 240
Levin, B. 142 MEG (Magneto-EncephaloGraphy) 234
Levy, Y. 237 Meillet, A. 160
lexical items 135 Menomini 85–6
Index 261
mental lexicon 171, 172–3, 174 of 180–1, 180 n6, 182

metathesis 145, 145 n8 Ojemann, G. et al. 89 225
Middle English 164, 175, 178, 179, 180, Old English 62, 164, 176–8, 180 n6, 181,
181–2 182 n7
Miller, E. H. 210 n11 O’Neil, W. 178, 234
Miyashita, Y. 234 Optimality Theory 99–106, 110
modal auxiliaries 162–75 optional infinitives 192–4
be 168–71, 172, 173 oronyms 70
category membership 164, 166, 167–8, 171, Osterhaut, L. 233
173
distribution 162–4 Paradis, C. 93 n1
morphological distinction 166 parole 12, 13
nature of change 164–6 Paul, H. 6
past tense forms 167 Payne, D. L. 150
and thou 171–2 Pepperberg, I. M. 220
V-to-I operation 168, 171 perception
mondegreens 70 analysis by synthesis 79
Monrad-Kröhn, G. H. 128 in birdsong 213
More, Sir Thomas 165 in child development 207–9
morphemes 132, 135–52 motor theory 113–14, 213
content 138–47 neurological basis 234–5
derivation and change 143–7 Perlmutter, D. M. 219 n5
subtractive morphosyntax 141–3 PET see Positron Emission Tomography
subtractive semantics 139–41 Petitto, L. 219 n6, 220
directionality in word formation 150–2 Petitto, L. A. 207
form 137–8 PF see phonological form
solidarity of form and content 147–50 Phillips, C. 235
morphological relatedness 136 philology 5–6
morphology phonemes 75
case 61–3 phonemics see phonology
and category membership 173–4 phonetic representation 68, 69–70, 76 n2
productivity 152–4 biuniqueness condition 77–8, 80–1, 83
see also morphemes discreteness of phonetic dimensions 124–5
morphophonemics distinctive features 70, 72
and I-language 83–91 hidden intended effects 123–4
representations 80–3 linguistic basis for 112–13, 114–27
MRI see Magnetic Resonance Imaging measurable “unintended” effects 115–22
Mustanoja, T. 178 laryngeal specifications in Korean
mutation 138 119–20
Mylander, C. 204 pitch microcontours 116–18, 117f
schwa 121–2
Nash, D. 143 vowel length effects 118–19
negation 160, 189–90, 190 n2 vowel quality differences 120–1
Newcombe, F. 229 and phonological form 77–80, 115,
Newmeyer, F. J. 161, 161 n2 129–30
Newport, E. L. 204 n8, 218 n4 segmental independence 70, 71
Norwegian 58 segmentation 70–1
Nottebohm, F. 213 timing effects 126–7
nouns phonetics 10–11, 67–8
agent nouns 149, 153 faithfulness 102–4
binding theory 195 n4 International Phonetic Alphabet 124, 130
derived nominals 142–3, 148, 149 markedness 101, 102–4
-er nominals 142–3, 148, 149, 153 theory 68–73
Nunes, J. 65 n12 see also phonetic representation
Nunnally, T. 180 n6, 181 phonological form (PF) 77–80, 115, 130
262 Index
phonological knowledge Rizzi, L. 56, 57, 200

borrowing 93–6, 109 Roberts, I. 161 n3
constraints vs. rules 99–110 Rosch, E. 238, 239
rules 96–9 Ross, J. R. 164
phonological representations 68, 74–7, Rugg, M. D. 231 n10
76 n2 rules 84, 88–9, 90, 92, 96–9, 100–1, 152
biuniqueness condition 77–8, 80–1, 83 Russian 83–4
fully specified basic variant 76
fully specified surface variant 76 Saltzman, E. 126 n4
incompletely specified 75–6 Sapir, Edward 4–5, 12, 17, 76, 158, 159
morphophonemic representations 80–3 Sapir-Whorf Hypothesis 4
nature of 111–12 Saudi Arabic 118
and phonetic form 77–80, 115, 130 Savage-Rumbaugh, E. S. et al. 219, 220
underspecification 119–20 Schleicher, A. 7–8, 11
phonology 67, 68, 73 schwa 121–2
articulatory 124–5 segmental independence 70, 71
auto-segmental representations 90–1 segmentation 70–1
generative 83, 84, 89, 100 Seidenberg, M. 219 n6, 220, 237
history 83–91 Senghas, A. 204 n8
morphophonemic representations 80–3 Separation Hypothesis 155
sound change 6, 7, 9–11 Sereno, M. I. 233, 234
sound pattern acquisition 206–14 serve 200
theory 73, 87 Shanker, S. G. 220
see also phonological form; phonological Siberian Yupik 136
knowledge; phonological Sieratzki, J. 238
representations Sievers, E. 11
phrase structure 28–31, 196–8 signing and signers
physiology ix–x apes 219–20, 219 n5–6
Piaget, J. 238 aphasia 225
Piattelli-Palmarini, Massimo 238 brain function 224–5
pidgins 203, 218 language acquisition 203–4, 204 n8,
Pierce, A. 194 206, 207
Pierrehumbert, J. 112, 124, 125 language faculty 2, 67, 218
Pinker, S. 222, xvi signs 13, 135, 153
pitch microcontours 116–18, 117f SLI see Specific Language Impairment
PLD see primary linguistic data Smith, N. V. 237
Poeppel, D. 193, 232, 234–5 Smolensky, P. 99, 100
Poizner, H. 225 sociolinguistics 3–4, xi
Polish 62, 63 Solan, L. 195
Positron Emission Tomography (PET) sound change 6, 7, 9–11
231–2, 233 sound patterns
poverty-of-stimulus problem 18–21, 25–33, babbling 207
34–6 birdsong development 210–14
Premack, D. 220 children’s acquisition 206–9
preterite-presents 166–7 cooing 207
primary linguistic data (PLD) 162, 167, perception 207–9
168, 199 see also phonetics; phonology
Prince, A. 99, 100 Specific Language Impairment (SLI) 194,
pronouns 19, 32–3, 61–2, 195 n4, 200 n7 237, 238
prosodic hierarchy 28 speech
psychiatry 4–5 microprosody 126, 127–9
Pullum, G. K. 190 n2 perception 79, 113–14, 207–9, 234–5
physical properties 113
Rask, R. K. 11 split genitives 175, 176–8, 181, 182, 182 n7
recursive devices 43–4 Spurzheim, J. 223f
Rice, M. 194 Stevens, K. 79 n4
rider/writer 86–7 structural linguistics 13
Index 263
Studdert-Kennedy, M. 113 movement 192–3, 205

Supalla, S. 218 n4 preterite-presents 166–7
Swedish 59, 109, 201–2 serve 200
syllables 28 subjunctive forms 167
syntax want to/wanna 187–9
children’s acquisition 186–206 West Circassian 150
emergence within linguistics 41–2 see also modal auxiliaries
generative 110 visual system 37, 206, 226
merger and deletion 43–5 voice assimilation 106–9
deleting copies 47–52 vowels
diversity 55–61 Great Vowel Shift 9–10
incorporation 52–5 length effects 118–19
that deletion 45–7 quality differences 120–1
see also case schwa 121–2
Talkin, D. 125 Wallman, J. 219 n6

Tavakolian, S. 195 Wanner, E. 190
Taylor, T. J. 220 want to/wanna 187–9
tense 200 Warlpiri 143–5, 147
Terrace, H. S. et al. 219 n6 Warner, A. R. 168–71
that deletion 45–7 Watkins, C. 175
thematic roles 175–6, 178, 180, 181 Wells, R. 148 n9
Thornton, R. 38–9, 188–90, 191–2, 196 Wernicke, C. 227, 228
thou 171–2 Wernicke’s area 227
TLA see Triggering Learning Algorithm West Circassian 150, 150 n11
Tomblin, J. B. 237 West Flemish 57–8
tone languages 117–18 West Greenlandic 136, 145–6, 147
Tourette’s Syndrome 229 Weverink, M. 193–4
trace 26 n4 Wexler, K. 190, 192, 193, 194, 202, 203
tree diagrams 7–8 wh-phrases 45, 191–2
trigger 2, 36–7, 173, 198–206 who/whom 159
Triggering Learning Algorithm (TLA) 202, Wiesel, T. 37
203 Williams, E. 132
Trubetzkoy, N. S. 87, 112 Williams, H. 213
Tsimpli, I. M. 237 Williams Syndrome 238
Woll, B. 238
underspecification 119–20 words 28, 131–2
Universal Grammar (UG) 15–16, 22, 36–7 classes 153
Case 176, 179, 180, 182 derivational relations 153–4, 155–6
constraints 101, 103, 109 directionality in formation 150–2
deletion principle 47, 61, 65 lexical items 135
grammatical categories 167 lexical organization 155–6
principles 31, 33–4, 186–90 lexicon 132–5
production/comprehension asymmetry monotonicity in formation 137–47
189–90 and “morphemes” 135–52
utterances 28 productivity 152–4
Uylings, H. B. et al. 224 n7, 227 solidarity of form and content 147–50
Wright, J. 41
van Gelderen, E. 161 Wright, R. 123
Vata 59 writer/rider 86–7
verbs
affixal 172–3 Yale morpheme 148 n9, 155
be 168–71, 172, 173 Yoruba 117–18, 117f
featural 173 Yupik 136
have 173
is/’s 18–19, 25–32, 38–9 Zwicky, A. M. 148, 190 n2

Anderson, Lightfoot 2002 The Language Organ

Uploaded by

Copyright:

Available Formats

Anderson, Lightfoot 2002 The Language Organ

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anderson, Lightfoot 2002 The Language Organ

Uploaded by

Copyright:

Available Formats

The Language Organ

Linguistics as Cognitive Physiology

stephen r. anderson is Professor of Linguistics and Cognitive Science

david w. lightfoot is Dean of the Graduate School, Georgetown

  

© Stephen R. Anderson and David W. Lightfoot 2004

First published in printed format 2002

ISBN 0-511-04210-8 eBook (netLibrary)

1 Studying the human language faculty 1

2 Language as a mental organ 18

4 Sound patterns in language 67

5 Describing linguistic knowledge 92

6 Phonetics and the I-linguistics of speech 111

8 Language change 157

9 “Growing” a language 186

10 The organic basis of language 216

One of the great success stories of post-Second-World-War intellectual inquiry

As a result, linguistics departments sometimes have a bit of a hard time mak-

In truth, our understanding of the overall architecture of cognition is still very

with so speciﬁc an understanding of higher-level abstract representations of

harder reading. Like restaurateurs, we aim to cater to a range of different

Chapter 6 deals with phonetics, often thought of not as a theoretical aspect of

1.1 Linguistics and the mind/brain

1.2 Linguistics as history

reconstruct an ancestral language (e.g., Proto-Indo-European) from which a

Aryan Italo-Celtic Balto-Slavic

Indic Iranian Greek Albanian Italic Celtic Slavic Baltic Germanic

As has often been pointed out, the Schleicher-style cladograms, familiar

These cladograms assume that languages are like species, as articulated

(1.2) a. Grimm’s Law

b. The Great English Vowel Shift

phonetics of speech using instrumental techniques that were precise and

1.3 Linguistics as the study of E-language

refuted or shown to be wrong: rather, it was abandoned as yielding diminish-

1.4 Linguistics as the study of I-language

these came to be speakers of “Serbian,” others of “Croatian,” “Bosnian,” etc. –

5 Harris 1951 was perhaps the epitome of this approach.

Linguists and linguistics have been important in these developments, in view

2.1 We know more than we learn

2.1.1 Two grammatical puzzles

2.1.2 Where does this knowledge come from?

a grammar to emerge in a child. We have no idea whether this information is

2.2 The nature of grammars

In general, two categories “merge” to form a new category. So an

The syntactic component of a speaker’s language organ, then, can be repre-

2.3 Back to the puzzles

While the correspondence between syntactic and phonological phrase struc-

2.4 The acquisition problem

2.4.1 The poverty of the stimulus

can be learned easily by speakers of Indonesian), to more general goals, such

2.4.2 The analytical triplet

2.4.3 Real-time acquisition of grammars

In (2.15a), the child is invited to produce a sentence where what is understood

3.1 The emergence of syntax within linguistics

3.2 Successive merger and deletion

the student will visit what city

This kind of merger may continue indeﬁnitely, building structures of arbitrary

elements may be displaced, as we shall see. We shall postulate a Universal

3.2.1 Delete that

(3.15) a.Whoi do you wonder [CP ei how [IP ei solved

(3.23) a.Mii ein ex joda’at [CP ’im ei mešaret bamilu’im]?