CHAPTER 29

LANGUAGE PRODUCTION
GRAMMATICAL ENCODING

KATHRYN BOCK AND WILLEM LEVELT

I. INTRODUCTION

The processes of language production can be divided into those that create the
skeleton of an utterance and those that flesh the skeleton out. In this chapter
we are concerned chiefly with the former, a set of processes which we term
GRAMMATICAL ENCODING (Levelt, 1989). Grammatical encoding comprises
both the selection of appropriate lexical concepts (entries in the speaker's
vocabulary) and the assembly of a syntactic framework. It contrasts with PHO-
NOLOGICAL ENCODING, which comprises the assembly of sound forms and the
generation of intonation. The product of these processes is not speech itself,
but a specification of an utterance that is adequate for controlling the processes
of articulation or speech production.
The components of grammatical encoding are no more accessible to con-
scious experience than the corresponding components of comprehension. Just
as in comprehension, we typically become aware only of disruptions. But
unlike disruptions of comprehension, many disruptions of production are public
events: A speaker who intends to say meals on wheels and instead says wheels
on meals usually knows that something has gone wrong, as does anyone within
earshot. Because of their ready availability, speech errors are a rich source of
clues to how language production works (Cutler, 1988).
Deciphering these clues has been the focus of several pioneering studies
(Dell & Reich, 1981; Fromkin, 1971; Garrett, 1975; Meringer & Meyer, 1895/
1978). The details of the analyses diverge in important ways (some of which
we touch on later), but there is reasonable agreement on the broad outline of
production processes that is sketched in Figure 1. This outline roughly follows
proposals by Garrett (1980, 1982, 1988) and, although it is motivated primarily
by analyses of speech errors, it is intended to provide an account of normal
production. The bridge from errors to normal production is built largely on the
existence of strong constraints on the forms of speech errors, which are taken
to point to relatively immutable components of the production process.
COPYRIGHT © 1994 BY ACADEMIC PRESS. INC.
HANDBOOK OF PSYCHOLINGUISTICS 945 ALL RIGHTS OF REPRODUCTION IN ANY FORM RESERVED.
946 KATHRYN BOCK AND WILLEM LEVELT

MESSAGE

FUNCTIONAL

Lexical Function
Selection Assignment

PROCESSING

POSITIONAL

Constituent Inflection
Assembly

PROCESSING

PHONOLOGICAL
ENCODING

to output systems
FIG. 1 An overview of language production processes.

We use the model in Figure 1 to organize and introduce the main topics of
this chapter. It shows four levels of processing, the message level, the functional
level, the positional level, and the phonological level. The message captures
features of the speaker's intended meaning and provides the raw material for
the processes of grammatical encoding. These processes are grouped into two
sets, functional and positional. The primary subcomponents of functional pro-
cessing are lexical selection (which involves the identification of lexical concepts
that are suitable for conveying the speaker's meaning) and function assignment
(which involves the assignment of grammatical roles or syntactic functions).
Positional processing involves the creation of an ordered set of word slots
(constituent assembly) and morphological slots (inflection). Finally, phonologi-
cal encoding involves spelling out the phonological structure of the utterance,
in terms of both the phonological segments of word forms and the prosody of
larger units.
The processes of grammatical encoding can be more concretely specified
by going through the steps involved in generating a simple utterance and con-
CHAPTER 29 GRAMMATICAL ENCODING 947

structing errors that might arise at each step. We number these steps for exposi-
tory convenience, but the numbers are not intended to denote a strict ordering
of implementation. As the target utterance we use She was handing him some
broccoli. The message behind this utterance presumably includes notions about
a past progressive event in which a female action-agent transfers by hand a
nonspecific object from a certain class of vegetables to a male action-recipient.
The first step, lexical selection, involves identifying the lexical concepts
and LEMMAS suitable for conveying the message. Lemmas carry the grammatical
information associated with individual lexical concepts, such as their form class
(noun, verb, etc.). For conveying the broccoli message, appropriate lemmas
include masculine and feminine pronominal indices, a noun {broccoli), and a
verb {hand) that relates the elements or ARGUMENTS of events involving an
agent, a recipient, and a theme. 1 A common type of speech error that appears
to reflect a problem of lexical selection is a SEMANTIC SUBSTITUTION, which
would occur if our hypothetical speaker said She was handing him some cauli-
flower. These substitutions preserve general features of the meaning of the
intended word (Hotopf, 1980) and are nearly always members of the same
grammatical form class (noun, verb, adjective, adverb, or preposition). In
Stemberger's error corpus (1985), 99.7% of all lexical substitutions represented
the same form class as the target.
The second step is function assignment. This involves assigning syntactic
relations or grammatical functions (e.g., subject-nominative, object-dative).
During the formulation of She was handing him some broccoli, the feminine
pronoun lemma should be linked to the nominative (subject) function, the
masculine to what we will call the dative function,2 the argument represented
by broccoli to the accusative function, and hand to the main verb function.
Errors of function assignment arise when elements are assigned to the wrong
functions. For example, if the feminine and masculine pronoun lemmas were
linked to the dative and nominative functions respectively, the resulting utter-
ance would most likely be He was handing her some broccoli. These EXCHANGE
errors, like other types of exchanges, involve constituents of the same type
(both are noun phrases). They are not simple exchanges of word forms, as our
example illustrates: The error is not Him was handing she some broccoli.
The next two steps constitute positional processing, so called because it
fixes the order of the elements in an utterance. As this implies, the order may
not be imposed during functional processing. One indication comes from a
contrast in scope between the features of different types of errors (Garrett,
1980). Exchanges of whole words occurred within the same phrases only 19%
of the time in Garrett's corpus (1980), implying that adjacency is not a strong
conditioning factor. In contrast, when sounds are exchanged (as in sot holdering
iron), they originated in the same phrase 87% of the time.
We consider constituent assembly first. This is the creation of a control
hierarchy for phrasal constituents that manages the order of word production

1
In event-role terminology, the theme is the object in the event that undergoes movement.
This sense of theme should not be confused with the unrelated sense of discourse theme.
2
The dative is roughly the same as the traditional indirect object.
948 KATHRYN BOCK AND WILLEM LEVELT

and captures dependencies among syntactic functions. For She was handing
him some broccoli, the hierarchy can be depicted in this way:

The basic features of such hierarchies are largely predictable from the types
of syntactic functions that have to be represented and from the syntactic features
of the selected lemmas.
The last of the grammatical encoding processes, inflection, involves the
generation of fine-grained details at the lowest levels of this structure. In English,
many of these details involve elements that carry information about number,
tense, and aspect but are bound to other words. So, the expression of the
progressive feature on the verb handing requires elaboration of one node of
the tree, as shown below:

The generation of these details is in no strict sense distinguishable from the

rest of constituent assembly, but we discuss it separately in order to showcase
a debate over whether the elements dominated by the "twigs" of the structural
tree behave uniquely.
One type of error that is identified with inflection is known as STRANDING.
Stranding is illustrated in the utterance of a speaker who intended to say You
ended up ordering some fish dish and instead said You ordered up ending some
fish dish (Garrett, in press). In such errors, the bound suffixes (-ed, -ing) show
up in their proper locations in the utterance but affixed to the wrong words,
arguing that the inflections are positioned separately from their word stems.
Another type of error that may arise during inflection is called a SHIFT (Garrett,
1975) and consists of the mislocation of an affix. Such an error could lead to
the utterance of She was hand himming some broccoli by our hypothetical
speaker. The elements involved in such errors are much more likely to be
involved in errors than the final syllables of word stems, such as the -id in
CHAPTER 29 GRAMMATICAL ENCODING 949

morbid (Stemberger, 1985), implying that strandings and shifts are not simple
mislocations of syllables but mislocations of pieces of grammatical structure.
With all this done, it still remains necessary to spell out the phonological
content of the utterance. That is the province of phonological encoding, which
we will not treat here (see Gerken, this volume).
In the remainder of this chapter, we fill out the picture of grammatical
encoding by critically examining each of its hypothesized subcomponents and
marshalling evidence about them from different sources, including computer
modeling and experimental research on production. The experimental work
serves at least three essential purposes. First, it serves to test hypotheses
derived from error observations under better controlled circumstances, making
it possible to rule out alternative explanations of production processes. Second,
it permits examination of features of language production that errors cannot
illuminate, if only because those features are seldom or never involved in errors.
Even the most familiar types of speech error are surprisingly rare events (Deese,
1984; Garnham, Shillcock, Brown, Mill, & Cutler, 1982; Heeschen, in press).
And finally, experimental work makes it possible to explore whether the features
of production that are postulated on the basis of error analyses hold equally
under the circumstances that lead to normal, error-free production. Errors, by
definition, reflect unusual circumstances that cannot straightforwardly be taken
to represent the norm. So, any hypothesis that attributes a certain property to
the production system in order to account for a particular sort of error is
vulnerable to the objection that the property is in fact aberrant. 3
At the outset, we adopt a very strong position about the nature of these
processing systems. It is that each one is influenced only by information repre-
sented at the level directly above it. For example, we assume that the processes
of lexical selection and function assignment are under the control of information
in the message and are unaffected by the sounds or phonological features of
words. This is neither a majority view nor an obviously correct one, and there
are compelling reasons to subject it to careful scrutiny (Dell, 1986; Stemberger,
1985). However, it is an assumption that is a testable and (perhaps all too easily)
disconfirmable, so that its flaws can be readily corrected as further evidence
about these processes accumulates.
We also assume that language production is incremental (Kempen & Hoen-
kamp, 1987; Levelt, 1989), so that variations in the order in which information
is delivered from one component to the next can readily affect the order in
which elements appear in speech (Bock, 1982). When higher level processing
components drive lower level ones, incremental production implies that the
higher levels need not complete their work on an utterance before the next
level begins. This is illustrated in Figure 1 in terms of hypothetical temporal
connections between the processing levels. The implementation of incremen-
tality requires the formulation, at every level, of piecemeal units relevant to
the form and content of the developing utterance, so our review touches on
the information partitionings within each processing component.

3
It is for this reason that the most persuasive hypotheses that emerge from error analyses
are based on what stays right in an utterance when something else goes wrong.
950 KATHRYN BOCK AND WILLEM LEVELT

II. LEXICAL SELECTION

In fluent speech we normally produce two to three words per second (Maclay
& Osgood, 1959), but there are occasional bursts (ANACRUSES) of up to seven
words per second (Deese, 1984). Even at these rates, we retrieve the appropriate
items from our mental lexicons. This is a surprising skill, given that we know
tens of thousands of words (Oldfield, 1963) and that errors of lexical selection
are rare: estimates of selection-error rates per thousand words of speech range
from 0.25 (Deese, 1984) through 0.41 (Garnham et al., 1982) up to 2.3 (Shallice
& Butterworth, 1977).
Empirical research in lexical selection relies on three sources of evidence.
First, though selection errors are rare, they have been carefully collected and
analyzed. Second, word finding can be particularly troublesome in aphasic
patients; the ways in which they err can reveal processes of retrieval that are
deeply hidden in normal speech. Third, lexical selection has increasingly come
to be studied experimentally. The experiments often involve picture naming,
with naming latencies measured under various conditions. In the following we
address only the first and last sources of evidence (for studies in aphasia, see
Garrett, 1992 and chapters in this volume by Caplan and by Zurif and Swinney)
in terms of a theoretical framework developed by Levelt (1989; Levelt et al.,
1991a) and Roelofs (1992). That framework is presented first.

A. A Network Model of Lexical Access

Our mental store of words and basic information about them is called the mental
lexicon. It is obviously not the case that all possible words of our language are
stored somewhere in our minds, because there is an infinity of possible words.
Take the numerals. They form an infinite set and a corresponding infinite set
of words, including compounds such as twenty-three thousand two hundred
seventy-nine. This is unlikely to be an entry in the mental lexicon. Rather, such
words are constructed when needed. Languages differ greatly in the use their
speakers make of this ability: Speakers of Turkish, for instance, produce new
words in almost every sentence (cf. Hankamer, 1989), whereas speakers of
English rarely do so. When we talk about lexical access here, we sidestep this
productive lexical encoding to focus on the retrieval of stored words from the
mental lexicon.
Our knowledge of words involves three types of information. First, we
know a word's meaning. We know that a sheep is a kind of domestic animal,
that it has a wool pelt, that it produces milk, etc. These are all properties of
our concept SHEEP.
Second, a word has syntactic properties. The word sheep is a noun. In
French mouton is also a noun, but in addition it has male syntactic gender, in
contrast to chèvre 'goat', which has female gender. A word's syntactic proper-
ties can be fairly complex. Verbs, in particular, are specified for the optional
or obligatory arguments they command. For example, the verb hit typically
takes a subject and a direct object (i.e., it is a transitive verb), and because
this is something that a speaker knows about the verb hit, it is part of the
mental lexicon. This type of information is called the verb's SUBCATEGORIZA-
TION FRAME. The verb hand, from our earlier example, has two subcategoriza-
CHAPTER 29 GRAMMATICAL ENCODING 951

tion frames. The first one, the prepositional frame, includes a direct object
position and an oblique (prepositional) object position (as in She was handing
some broccoli to him), and the second one, the double object frame, maps the
dative to the direct object position and the accusative to a so-called second
object position (as in She was handing him some broccoli). The word as a
syntactic entity is technically called a LEMMA.
Lemmas contrast with LEXEMES, which capture the word's form properties.
These constitute its morphological and phonological shape. The word sheep is
monomorphemic and consists of three phonological segments, / /, /i/, and /p/.
The word handing consists of two morphemes, a stem and a suffix, and six
phonological segments, /h/, / /, /n/, /d/, / /, and / /.
In the network model, these different types of information correspond to
nodes within three levels of representation, the conceptual level, the lemma
level, and the lexeme level. A part of this lexical network is shown in Figure
2. It depicts some of the knowledge we have about the words sheep and goat.
At the conceptual level, the nodes represent concepts. They are linked by
labeled arcs that represent the nature of relationships. Since a sheep is an
animal, this is represented by an ISA connection between the nodes SHEEP

visual form

CONCEPTUAL
LEVEL

LEMMA
LEVEL

LEXEME
OR
SOUND
LEVEL

FlG. 2 A part of the lexical network. Note that the arrows represent types of connections within
the network, not the flow of information during production or comprehension.
952 KATHRYN BOCK AND WILLEM LEVELT

and ANIMAL. A word's meaning as a whole is represented by such a network

of relations (as introduced by Collins & Loftus, 1975, and Collins & Quillian,
1969), although individual lexical concepts themselves are represented by uni-
tary nodes. In this respect, the model departs from a compositional representa-
tion of word meaning. We will not go into this perennial issue in lexical represen-
tation (for further discussion, see Bierwisch & Schreuder, 1992; Fodor, Garrett,
Walker, & Parkes, 1980; Levelt, 1989; McNamara & Miller, 1989).
Some conceptual nodes have direct connections to nodes at the second,
lemma level. This subset of conceptual nodes represents lexical concepts. Not
all concepts are lexical: DEAD TREE is a perfectly well formed concept, but
one without a lexical concept. Yet English has a lexical concept for dead body
(CORPSE).
The nodes at the lemma level represent syntactic properties. The lemma
sheep has a category link to the noun node; in French the lemma mouton has
a gender link to the male node, and so on. At the lexeme level, the network
represents the word's form properties. The lexeme node / ip/ thus has labeled
links to its constituent phonological segments, / /, /i/, /p/.
Lexical access in this model is represented by activation spreading from
the conceptual level to the lemma level to the lexeme level (note that Fig. 2
does not depict the activation trajectories; the arrows in the figure characterize
permanent relationships rather than processing dynamics). We will not consider
how a speaker first conceives of the notions to be expressed (see Levelt,
1989, and for a very different view, Dennett, 1991), but beyond that, the first
requirement for lexical selection in normal speech is the existence of an active
lexical concept. A concept node can become activated in myriad ways. One
simple procedure to induce this is to present a picture for naming. In an experi-
ment, a subject can be given a picture (e.g., one of a sheep, as shown in Fig.
2) and asked to name it as fast as possible. The assumption is that the picture
activates the concept.
An active lexical concept spreads its activation to all connected concept
nodes. So if the SHEEP node is active, the GOAT node will receive some
activation as well (either directly, or via mediating nodes such as ANIMAL or
MILK). In addition, activation will spread from the lexical concept node to
the corresponding lemma node. In this framework, lexical selection is selection
of the appropriate lemma node. So, if SHEEP is the active lexical concept,
the lemma sheep should be retrieved. It would be an error of selection if goat
were retrieved. There is nonetheless a small chance for such a mishap, because
some activation spreads from SHEEP to GOAT and from there to the lemma
goat.
In Roelofs' (1992) implementation of this model, the probability that any
given lemma will be selected during a specified time interval is the ratio of its
activation to the total activation of all lemmas in an experimental set (i.e., the
Luce ratio; Luce, 1959). This makes it possible to predict the time course of
lexical selection under various experimental conditions (see below). Some of
those conditions are designed to directly activate lemma nodes through the
presentation of spoken or written words (see Fig. 2), creating competitors for
other lemmas activated from the conceptual level.
The model as it is depicted deals only with lemmas for lexical concepts.
CHAPTER 29 GRAMMATICAL ENCODING 953

But not all words in fluent speech correspond to lexical concepts. In listen to
the radio, to does not represent a concept. Rather, the lemma for the transitive
verb listen requires the preposition to, so the lemma to must be activated via
an indirect route at the lemma level. We refer to this as INDIRECT ELECTION.
The major joint in the model is between the lemma and lexeme levels of
representation. Between lexical concepts and lemmas, there are systematic
relations. So, a verb's meaning is regularly related to its subcategorization frame
(Fisher, Gleitman, & Gleitman, 1991; Keenan, 1976). But between lemmas
and lexemes, the relation is highly arbitrary (de Saussure, 1916/1955). There
is no systematic reason why a SHEEP should be called sheep (or moutori).
Still, there are some statistical relations between the syntactic and phonological
properties of words (Kelly, 1992). Nouns, for instance, tend to contain more
syllables than verbs; they also contain front vowels more often than verbs
(Sereno & Jongman, 1990). Kelly (1992) argues that language learners and
users may sometimes rely on such statistical relations in parsing and speech
production.
The most dramatic reflection of the rift between the lemma and lexeme
levels is the so-called tip-of-the-tongue (TOT) phenomenon. It was described
by William James in 1890 in one of the most frequently quoted passages in
cognitive psychology:
Suppose we try to recall a forgotten name. The state of our consciousness is
peculiar. There is a gap therein: but no mere gap. It is a gap that is intensely
active. A sort of wraith of the name is in it, beckoning us in a given direction,
making us at moments tingle with the sense of our closeness, and then letting
us sink back without the longed-for term. If wrong names are proposed to us,
this singularly definite gap acts immediately so as to negate them. They do not
fit into its mould. And the gap of one word does not feel like the gap of another,
all empty of content as both might seem necessarily to be when described as
gaps. . . . The rhythm of a lost word may be there without a sound to clothe
it; or the evanescent sense of something which is the initial vowel or consonant
may mock us fitfully, without growing more distinct (1890/1950, pp. 251-252).
The TOT phenomenon was later discussed by Woodworth (1938) and systemati-
cally studied for the first time by R. Brown and McNeill (1966). R. Brown and
McNeill presented the definitions of infrequent words such as sextant and asked
subjects to produce the defined word. Whenever subjects entered a tip-of-the-
tongue state, they reported whatever came to mind about the target word. In
many cases the subjects knew the initial consonant or vowel, the number of
syllables, and the stress pattern. Related words might come to mind that shared
these properties (such as secant for sextant). These findings have been con-
firmed and elaborated in many subsequent studies (see A. S. Brown, 1991, and
Levelt, 1989, for comprehensive reviews). Most of these studies deal with TOT
states in normal speakers, but there are also clinical conditions that persistently
arouse TOT states. These are called anomias (see Butterworth, 1992, and Gar-
rett, 1992, for further discussion).
In terms of the network model, the TOT phenomenon is a failure to access
the lexeme from the lemma. The speaker knows the meaning to be expressed
(i.e., the concept) and the word's syntax (that it is a plural noun, a transitive
verb or whatever; i.e., the lemma). Only the word form is blocked. Some
954 KATHRYN BOCK AND WILLEM LEVELT

aspects of the form may surface, revealing something about the process of
phonological encoding (see Levelt, 1989, and Gerken, this volume, for reviews).
Because TOTs appear to arise subsequent to lemma activation, they are not
problems of lexical selection, but of lexeme activation.

B. Errors of Lexical Selection

There are three major types of lexical selection errors, called substitutions,
blends, and exchanges. In all three cases a nontarget lemma is activated and
an incorrect word form is produced. But there are different ways in which this
derailing activation can come about. Consider examples (l)-(6) of substitutions.
(1) . . . carrying a bag of cherries. I mean grapes (Stemberger, 1985)
(2) He's a high-low grader (Fromkin, 1973)
(3) Get out of the clark [intended: car] (Harley, 1984)
(4) A branch falling on the tree [intended: roof] (Fromkin, 1973)
(5) He's the kind of soldier a man . . . wants to emanate [intended:
emulate] (Bock, 1987)
(6) / urgently request you to release the hostages unarmed-unharmed
(Fromkin, 1973)
One potential cause of a substitution error is that an alternative lexical
concept is activated along with the target. In (1) the speaker intended to express
the notion GRAPE, but CHERRY was activated at the same time. This may
result from activation spreading at the conceptual level. Because GRAPE and
CHERRY are semantically related (both are small round fruits), there is some
linkage between them in the conceptual network. If both lexical concepts then
activate their lemmas, there is a chance for the unintended one (cherry) to be
accidentally selected (given a probabilistic selection rule like that of Roelofs,
1992).
Example (2) also involves a semantic relation: high and low are antonyms.
Antonyms and other semantic oppositions in fact form the most frequent type
of word substitution. Their causation may be similar to the above case, but
there is an additional feature. High and low are strong associates (stronger than
grape and cherry). It is not clear where word association should be represented
in a network model such as the one in Figure 2. It may be a special form of
conceptual relation, but it might also involve direct lemma-to-lemma connec-
tions.
Example (3) has a different etiology. The speaker intended to say Get out
of the car to someone but at that moment glanced up at a storefront with the
word Clark's printed on it. Then clark intruded, creating an environmental
contamination (Garrett, 1980). There was no conceptual spreading of activation
from CAR to CLARK. Rather, the printed word Clark seems to have activated
the corresponding lemma.
Example (4) has a still different cause. It appears that branch may have
activated its associate tree, allowing the lemma tree to be selected instead of
the target lemma roof. Again it is unclear whether activation spread at the
conceptual level (from BRANCH to TREE), at the lemma level (from roof to
tree), or both. Was the speaker really thinking of a tree when the error occurred?
We will never know.
CHAPTER 29 GRAMMATICAL ENCODING 955

Example (5), in which the target word was replaced by a sound-related

word, is due neither to conceptual- nor to lemma-level priming. In fact, it is
not strictly an error of lexical selection under our present definition because,
in terms of the model, the error need not have involved the activation of a
nontarget lemma (i.e., the lemma emanate). Fay and Cutler (1977) called this
type of error a malapropism and argued that such errors arise during lexeme
processing. At that level the lexicon is organized in terms of form, not meaning.
And indeed, malapropisms show no systematic meaning relation to the corre-
sponding targets (Garrett, 1980), as testified by such cases as sympathy for
symphony, bodies for bottles, and garlic for gargle. Revealingly, there is a
strong similarity to the "wrong names" that occur during TOT states, which
also seem to arise during lexeme processing.
The final example, (6), is a mixed error: The error unarmed and the target
unharmed have both a semantic and a phonological connection. Mixed errors
such as dictionary for directory and oyster for lobster are controversial. In
their corpus of naturally observed errors, Dell and Reich (1981) found that the
probability of a mixed error was higher than would be predicted if semantic
and phonological errors have independent sources. They concluded that phono-
logical similarity increases the probability of a semantic substitution. This con-
clusion has been supported in other research (Harley, 1984; Martin, Weisberg,
& Saffran, 1989; Stemberger, 1983; but see del Viso, Igoa, & García-Albea,
1991).
In a network model such as the one developed by Dell (1986) or the one
depicted in Figure 2, this can be handled by postulating feedback from the
lexeme to higher levies, crossing the lexeme/lemma rift (rather than the purely
top-down flow that we have thus far assumed). However, that move may
be unnecessary. First, there is experimental evidence from error-free speech
(Levelt et al., 1991a) that is inconsistent with this option. We return to this
below. Second, mixed errors may be predominantly environmental contamina-
tions, as suggested by Garrett (in press). If so, it may be that their origin is
special, not the consequence of general feedback between lexemes and lemmas.
Third, the overrepresentation of mixed errors could result from a mechanism
of self-monitoring. According to Levelt (1989), self-monitoring can begin as
soon as there is a phonetic plan for the word, and so before articulation is
initiated. If unarmed is internally but erroneously planned, a cohort of sound-
related words will be activated in the speaker's comprehension system, among
them unharmed. Its meaning can be activated via this phonological route, and
in addition via a semantic route, allowing the activated notion UNARMED to
prime the related notion UNHARMED even further. Since that is the intended
meaning, the monitor may pass the (erroneous) item.
Let us now turn to a second type of lexical selection error, blends (7)-(8).

(7) The competition is a little stougher [stiffer/tougher] (Fromkin, 1973)

(8) The sky is shining [The sky is blue/The sun is shining] (Harley, 1984)

Most blends are of type (7), the fusion of two words that are near-synonyms
in the context of conversation. Whereas substitutions reveal a predilection for
antonyms and close associates, blends of antonyms are exceptional (Hotopf,
1980; Levelt, 1989). Instead, it is quasi-identity of meaning that characterizes
956 KATHRYN BOCK AND WILLEM LEVELT

the blending components. The source of blends may therefore be earlier than
substitutions.
This makes blends something of a puzzle (Garrett, 1980). Their antecedents
are early, but the errors themselves—the phonological merging of two word
forms—are late. The merging is phonologically systematic, respecting the sylla-
ble constituency of both components (MacKay, 1972; Wells, 1951). It is possible
that this late merging is the result of the parallel encoding of two different
utterances (Butterworth, 1982; Garrett, 1980; Harley, 1984) triggered by the
speaker's conceptual indecision. This possibility is reinforced by the existence
of sentence blends such as (8), which likewise appear to result from the parallel
encoding of two related notions.
The third type of selection error includes exchanges such as (9)—(12).
(9) Seymour sliced the knife with a salami (Fromkin, 1973)
(10) I got into this guy with a discussion (Garrett, 1980)
(11) a hole full of floors (Fromkin, 1973)
(12) threw the window through the clock (Fromkin, 1973)
In (9), knife slipped into the noun slot in the direct object noun phrase in place
of salami (perhaps because it was at that moment more activated). So far, this
is simply a sort of word subsitution. But then, because knife was no longer
available for the next noun slot, salami was inserted in its stead to create a
second error. Because insertion in the wrong syntactic slot is possible only if
the syntactic category of the word is the same, word exchanges usually occur
between words of the same form class (over 80% of the time; Garrett, 1980;
Stemberger, 1985).
In (9), the exchange involved words, not whole phrases. A whole-phrase
exchange would have yielded Seymour sliced a knife with the salami, in which
the articles accompany their respective nouns. Phrase exchanges do occur,
however, as example (10) shows. Here the phrases this guy and a discussion
were exchanged. The existence of such exchanges complicates the picture,
because it can be difficult to tell a word exchange from a phrase exchange. A
clear case is shown in example (11) which, like (9), is a word exchange. The
target was a floor full of holes, and when the nouns floor and hole exchanged,
hole left its inflectional marking behind. This is characteristic of unambiguous
word exchanges: they strand other parts of their phrases, including adjectives
and closed-class 4 elements (see Berg, 1987, for further discussion). However,
when all the phrasal elements that accompany the exchanging words are the
same, as they are in (12), it is impossible to tell whether the exchange is lexical
or phrasal.
Such ambiguities are problematic because the exchange straddles the
boundary between lexical and syntactic processing. Genuine phrase exchanges
may have a different etiology than genuine word exchanges, one more similar
to the one sketched for pronoun exchanges in the introduction. Some support
for this conjecture comes from an informal survey of the word exchanges in
Fromkin's (1973) Appendix, which showed that unambiguous word exchanges
(e.g., takes plant in the place) are more likely to exhibit sound similarities and
less likely to exhibit meaning similarities than exchanges which could be phrase
exchanges (e.g., used the door to open the key).

4
The members of the closed class include function words and inflectional morphemes.
CHAPTER 29 GRAMMATICAL ENCODING 957

C. Experimental Studies of Lexical Selection

There is a long tradition of experimental research in lexical selection that falls

under the heading of "object naming" and dates back to Cattell (1885). Cattell
found that subjects were slower in naming pictures than in reading words. This
result finds a natural explanation in the network model of Figure 2. Written
words have direct access to lemmas, whereas picture information has to be
relayed via concepts.
Cattell's result has been extensively studied. One of its offshoots is research
on interference between words and pictures, as embodied in what is now known
as the picture interference paradigm (Lupker, 1979; see Glaser, 1992, for an
excellent review of this literature since Cattell). This is a double stimulation
paradigm. The primary stimulus is a picture, which the subject is instructed
to name as fast as possible. The secondary stimulus is a printed or spoken
distractor word, which the subject is instructed to ignore. Subjects are
rarely completely successful in carrying out this latter instruction, however:
The picture-naming latencies are normally affected by the presence of the
distractor.
There are usually two variables in such an experiment. The first is the
relation between the distractor and target word (the picture's name). When the
picture is one of a sheep, the distractor may be a superordinate {animal), the
identical word (sheep), a subordinate (ram), a cohyponym (goat), a sound-
related word (sheet), or an unrelated word (house). The second variable is the
stimulus onset asynchrony or SOA. This is the interval between picture onset
and distractor onset. If it is negative, the distractor precedes the picture onset;
if it is positive, the distractor follows the picture onset.
A classic picture-interference study is Glaser and Düngelhoff's (1984). In
one of their experiments they used an SOA range from — 400 to + 400 ms in
steps of 100 ms. A printed distractor word was either semantically related to
the target (a cohyponym) or unrelated (there was also an identity condition
which we ignore here). AH distractors were names of pictures in the response
set. The REAL line plotted in Figure 3 shows the difference between the related
and unrelated conditions in naming latencies over the whole range of SOAs.
Clearly, the naming response was sometimes delayed when semantically
related distractors were presented, compared to when the distractors were
unrelated to the target word. This is called SEMANTIC INHIBITION, and it can
be understood in terms of the network model (see Fig. 2). When the picture
depicts a sheep, activation spreads to the concept SHEEP, and thence to the
lemma sheep. An unrelated distractor word such as house directly activates
the corresponding lemma house. Because there are now two active lemmas,
and both are possible responses in the experiment (i.e., house is sometimes a
target), the probability of selecting sheep at any one moment is smaller than if
there were no distractor (because the Luce ratio is smaller). If a related distractor
is presented (e.g., goat), the delay should be even greater. This is because
activation from the concept SHEEP spreads to the concept GOAT and down
to the lemma goat. The latter will therefore be more activated than house is
in the unrelated condition. The results of Roelofs' (1992) simulation of Glaser
and Düngelhoff's experiment is shown in Figure 3 as the SIM line. The fit is
statistically perfect. Roelofs' own experiments produced further support for
this model.
958 KATHRYN BOCK AND WILLEM LEVELT

FlG. 3 Picture naming in the picture-word interference paradigm at nine stimulus-onset asychro-
nies (SOAs). The REAL data {filled squares) are from Glaser and Düngelhoff (1984); the SIM data
(open squares) are a simulation reported by Roelofs (1992). In the related (RED condition the
interfering probe word was semantically related to the picture target, and in the unrelated (UNR)
condition it was not.

In a very similar study, but with spoken distractor words, Schriefers,

Meyer, and Levelt (1990) found a comparable semantic inhibition effect. In
addition, they showed that semantic inhibition disappeared when the subject's
task was not picture naming but picture recognition. In the recognition task,
the subject was first shown the pictures (as in the picture naming experiment).
Then the pictures were presented among a set of new ones. The subject's task
was to push a yes button when the picture was an old one and the no button
when it was new. The pictures and distractors were the same ones used in the
naming experiment, but now there was no trace of semantic inhibition. This
implies that the effect is lexical, not conceptual. Because semantic inhibition
cannot be merely a word form effect, this finding points to the involvement
of lemma representations even in picture naming, when the subject is not
constructing sentences. A similar disappearance of semantic inhibition in a
recognition task was observed by Levelt et al. (1991a).
Schriefers (1990) was also able to separate conceptually and lexically in-
duced latency effects in an experimental setting. The subjects viewed two
geometrical shapes of different sizes (e.g., a large and a small triangle), one of
which was marked by a cross. The task was to say bigger when the marked
shape was the bigger one and smaller when it was the smaller one of the two.
When both shapes were rather large the bigger response was facilitated, and
when both were rather small the smaller response was facilitated. Schriefers
argued that this congruency effect is of conceptual, nonlexical origin, because
it was also found when the response was nonverbal (made by push buttons),
when no lexical access was required. The situation was quite different for
another effect, the markedness effect: Bigger responses were usually faster
CHAPTER 29 GRAMMATICAL ENCODING 959

than smaller responses (and similarly for other marked versus unmarked adjec-
tives; see Bierwisch, 1969, for an analysis of this distinction), but this
markedness difference disappeared when subjects responded nonverbally. On
the basis of these and other findings, Schriefers suggested that markedness is
a property of adjective lemmas.
Earlier we argued that the major rift in lexical access is between the lemma
and the lexeme levels of processing. In their picture naming study, Schriefers
et al. (1990; see above) used not only semantic distractor words, but also
phonological ones in the picture-word interference task. The semantic dis-
tractors caused inhibition at an SOA of - 150 ms (i.e., the onset of the spoken
word preceded the picture by 150 ms). Phonological distractors (e.g., sheet
when the picture was one of a sheep) produced a facilitatory effect at SOAs
of 0 and + 150 ms (in agreement with findings by Meyer, 1990, 1991). But there
was no trace of phonological facilitation at the SOA of - 150 ms. The implication
is that phonological encoding strictly follows lexical selection.
The two-stage theory that is suggested by this result was reconfirmed by
Levelt et al. (1991a). A different type of dual stimulation task was used but,
as in the previous experiments, the subjects' main task was picture naming.
On about one third of the trials, a spoken probe (a word or nonword) was
presented at one of three SOAs. The subjects' secondary task was lexical
decision: They pushed a yes button when the probe was a word and a no button
when it was a nonword. The dependent variable was the latency of this response.
Among the probes were semantically related ones (e.g., goat when the picture
was one of a sheep) and phonologically related ones (e.g., sheet). Assuming
that the processes of lexical selection affected latencies for semantic probes
and that lexeme encoding affected latencies for phonological probes, Levelt et
al. were able to examine whether the data fit their two-stage model better than
a connectionist network model which allows for feedback from the lexeme to
the lemma level (Dell, 1986). That turned out to be true (see Dell & O'Seaghdha,
1991, 1992; Levelt, 1992, and Levelt et al., 1991b, for detailed discussion of
these controversial issues). Further findings from this research indicated that
a lemma spreads activation to its lexeme only after it has become selected;
lemmas that are merely active do not spread activation to the lexeme level.
This contradicts predictions from both connectionist (Dell, 1986; Mac Kay,
1987) and cascade-type models (Humphreys, Riddoch, & Quinlan, 1988).
The conclusion was that the lexical access system for production has a
highly modular organization. Lexical selection strictly precedes and is unaf-
fected by phonological encoding. And that makes good sense. Lexical selection
and phonological encoding are dramatically different: Lexical selection involves
a semantically driven search through a huge lexicon, whereas phonological
encoding involves the creation of a pronounceable phonetic pattern for each
individual word. Interactions between such processes pose the threat of mutual
disruption, yet lexical access is remarkably fast and accurate. Modularity may
be nature's protection against error.
A final question about lexical selection concerns word frequency. Fre-
quency seems to have reliable effects on production, as reflected in picture
naming times. Oldfield and Wingfield (1965; also see Lachman, Shaffer, &
Hennrikus, 1974) found a high correlation between the latency to name a pic-
tured object and the frequency of the object's name in the language. So, the
960 KATHRYN BOCK AND WILLEM LEVELT

average speech onset latency was 640 ms for high-frequency basket, compared
to 1080 ms for low-frequency syringe. What is the locus of this effect in the
network model? Is it the concept, the lemma, the lexeme, or all three? It could
even be a very late phenomenon, having to do with the initiation of articulation.
Wingfield (1968) excluded the first alternative by measuring recognition latencies
for the pictures—a conceptual process—and found no effect of frequency.
Going from picture to concept therefore does not create the frequency effect;
lexical access is apparently essential.
At the other extreme, articulatory initiation, the chief evidence for a word
frequency effect comes from Balota and Chumbley (1985). They asked subjects
to read a word, but to utter it only after a go signal which appeared at SOAs
ranging from 150 to 1400 ms. Under these conditions one probably measures
articulatory initiation rather than selection or phonological encoding, but there
was a frequency effect of 26 ms averaged across SOAs in two experiments (for
further discussion, see Balota & Chumbley, 1990; Monsell, 1990; Monsell,
Doyle, & Haggard, 1989). Clearly this is not the full word frequency effect, as
measured by Oldfield and Wingfield. And perhaps it is not a word frequency
effect at all, but a syllable frequency effect. Levelt and Wheeldon (in press)
found that word and syllable frequency contribute independently and additively
to production onset latencies. It may be, then, that most or all of the "real" word
frequency effect has its origin somewhere between conception and articulation.
How, then, to distinguish between the lemma and lexeme levels as sources
of word frequency effects? Jescheniak & Levelt (in press) assessed the contribu-
tion of lemmas with a gender decision task using Dutch words which, like
French words, come in one of two grammatical genders. In this task, Dutch-
speaking subjects saw pictures and indicated the gender of the word that named
the depicted object. They did this by pressing one of two buttons, thereby
judging the gender of the target noun without actually uttering it. In another
task they simply named the pictures. Each picture appeared three times under
both task conditions, and in both there was an initial frequency effect. But in
the gender decision task the frequency effect dissipated, disappearing entirely
on the third trial. In naming, however, the frequency effect remained undimin-
ished over trials. From these and other experiments, Jescheniak & Levelt
concluded that the persistent frequency effect is a lexeme effect. The ephemeral
effect of frequency on gender judgment may have its origin in the connection
between the lemma and gender nodes (see Fig. 2) and is perhaps only a recency
effect. After a lemma's gender is accessed, that information may be readily
available for reuse.
In conclusion, the lexeme may be the primary locus of the frequency effect.
This conclusion is consistent with findings on prelexical hesitations in spontane-
ous speech (Butterworth, 1980; Garrett, 1975; Levelt, 1983).

III. FUNCTION ASSIGNMENT

As message elements are mapped onto concepts and lemmas, they must also
be assigned to syntactic functions. The primary problem of function assignment
is to specify which elements will serve as the subject of the incipient utterance
and which, if any, will serve as objects of various kinds. It is obviously necessary
CHAPTER 29 GRAMMATICAL ENCODING 961

to separate this problem from lexical selection, since the same words may serve
different functions in different sentences (e.g., Girls like boys versus Boys like
girls) and even in the same sentence (e.g., People need people). It is also useful
to treat this problem as one of grammatical encoding rather than one of message
formulation, because very similar messages may be expressed in ways that
differ only in the assignments of grammatical functions (e.g., She was handing
him some broccoli vs. She was handing some broccoli to him). But just as the
selection of lemmas is heavily influenced by the content of a message, so is
the process of function assignment.
Function assignment should also be separated from constituent ordering
for reasons that can be difficult to appreciate for speakers of English. English
observes a relatively rigid ordering of the constituents that play different roles,
but in languages with more flexible constituent orders, constituents can appear
in different positions serving the same grammatical functions (often signaled
by differences in case). Even in English, there are deviations from canonical
word order which point to a function assignment process that is different from
the ordering process. For example, a speaker can emphasize an object by
"fronting" it, as in Him I can't stand, and if, as in this example, the fronted
constituent is a pronoun (the only type of English element that reliably marks
grammatical function), it will retain its objective case.
The problems that have to be addressed by a theory of function assignment
have to do with the nature of the functions that are assigned, the kinds of
information that control the assignment, the nature of the elements that the
functions are assigned to, and the organization of the processes that carry out
these operations. The first three of these problems are matters of intense debate
in linguistics, and the last, more obviously the province of psycholinguistic
research, has received little systematic attention. We briefly examine the first
two and the last one in turn, from the perspective of the kinds of psycholinguistic
data that have been brought to bear on them (the third problem remains unad-
dressed in the psycholinguistic literature). Again, the data come from experi-
ments on normal speech and from observations of speech errors. However,
because speech errors that are unambiguously attributable to syntactic problems
are woefully scarce (Fay, 1980; Garrett, 1975; Stemberger, 1983, 1985), we rely
heavily on experimental data.

A. What Functions Are Assigned?

The most familiar candidate functions are those known as the SUBJECT and
DIRECT OBJECT (and, less familiarly, INDIRECT and OBLIQUE objects). The famil-
iarity of these labels disguises enormous linguistic problems of specification
and definition that we cannot begin to address (for discussion, see Bresnan,
1982; Marantz, 1984; Perlmutter, 1982; Williams, 1981), but we assume that an
adequate account of grammatical functions will highlight something close to
the traditional set, and that they are marked morphologically in case languages
and structurally in configurational languages (such as English). To simplify the
discussion, we use traditional case terminology to refer to the grammatical
functions that are assigned (e.g., nominative, accusative, dative, genitive), and
traditional grammatical relations terminology (subject, direct object, etc.) to
refer to where the elements that are assigned these functions actually appear
962 KATHRYN BOCK AND WILLEM LEVELT

in English sentences (since most of the work that we consider is on English).5

So, in English, the element that is assigned the nominative function appears in
subject position.
This apparently innocuous statement disguises a substantive theoretical
claim about the process of function assignment. The claim is that, within gram-
matical encoding, there is no level of processing at which the element that
serves as the subject of the sentence plays a role that can be realized as a
different grammatical relation. On this argument, there is no point at which
(for example) the direct object of an active sentence (e.g., the bone in A dog
carried the bone) has the same representation as the subject of its passive
paraphrase (e.g., The bone was carried by a dog).
This claim runs counter to a traditional conception of deep structure in
psycholinguistics (see Foss & Hakes, 1978, for a review), according to which
"underlying" objects may be realized as subjects. The problem with this con-
ception is that there is no evidence that function assignments normally undergo
changes during grammatical encoding, and some evidence that they do not
(Bock, Loebell, & Morey, 1992; Tannenbaum & Williams, 1968). Because
relation changing operations (such as transformations) are likely to introduce
considerable processing complexity (see Bresnan & Kaplan, 1984, for discus-
sion), it is more parsimonious to assume that the relations are assigned just
once and maintained throughout the grammatical encoding process.
This does not deny the existence of a level at which—to return to the
example above—there is some uniform representation of the bone in A dog
carried the bone and The bone was carried by a dog. However, we would
prefer to locate this uniformity within nonlinguistic cognition (cf. Bransford,
Barclay, & Franks, 1972), either in the conception of the event itself or in the
components of the message. The referent of the phrase the bone may play the
same part in a mental model of the event, regardless of how the event is
described.
Likewise, the rejection of relation changing operations does not mean that
there can be no underlying grammatical representation for utterances. It implies
only that there is a one-to-one correspondence between the underlying and
surface roles. In our minimalist conception, the underlying roles are the ones
assigned during functional processing and the surface roles are the ones assigned
during positional processing. Figure 4 sketches this arrangement.
Phrase exchanges (e.g., I went to the mechanical mouse for an economy
five and dime instead of/ went to the economy five and dime for a mechanical
mouse; Garrett, 1980) represent a type of error that may arise from missteps
of function assignment. They have two properties which point to something
other than a simple misordering of words. The first, noted briefly in the introduc-
tion, is restricted to errors in which the inverted phrases are made up of pro-
nouns (e.g., you must be too tight for them instead of they must be too tight
for you; Stemberger, 1982), because only pronouns exhibit their function assign-
ments. The distinctive feature of pronoun errors is that the pronouns bear the
appropriate case for the position in which they erroneously appear, rather than

5
Since nonconfigurational languages lack strict isomorphisms between functions and positions,
it is important to remember that the relations here refer to positions only in English sentence
structure.
CHAPTER 29 GRAMMATICAL ENCODING 963

Accusative
Dative

Nominative

AFTER
FUNCTION
ASSIGNMENT

AFTER
POSITION
ASSIGNMENT

FlG. 4 The relationship between grammatical functions (after function assignment) and grammati-
cal relations (after position assignment).

the case for the position in which they should have appeared. According to
Stemberger (1982), this is the norm for such errors in English, and Berg (1987)
reports a similar trend for German.
The second property of phrase exchange errors that appears to favor func-
tion assignment as the source of the problem is that the verbs in the error-
bearing utterances tend to agree with the subject that is actually produced
rather than with the subject that was intended (e.g., that's supposed to hang
onto you instead of you're supposed to hang onto that and most cities are true
of that instead of that's true of most cities; Stemberger, 1982). Stemberger
(1985) reports that this occurred in 6 of the 7 relevant errors in his corpus. It
suggests that the element that appears in subject position in the error also bears
the role of subject during the formulation of the agreeing verb, compatible with
the hypothesis that a function assignment error is the source of the exchange.
The element's appearance in an incorrect position is only a secondary conse-
quence of a deeper malfunction.
Experimental evidence consistent with the separation between functional
and positional assignments comes from Bock et al. (1992). They used a sentence
structure priming paradigm in which speakers first produced a priming sentence
in one of two different syntactic structures and then saw a conceptually unre-
lated event which they described with a single sentence. The event was designed
to be describable in either of the two primed structures (see Bock, 1990, for a
more complete description of the paradigm). The results revealed separate,
independent effects of the primed structure itself and of the conceptual features
of the elements that served different grammatical functions. In the present
scheme, these separate effects may be traced to positional and functional pro-
cessing, respectively.
964 KATHRYN BOCK AND WILLEM LEVELT

B. What Information Controls Functional Assignments?

Most discussions of the controllers of function assignments have focused on
subject assignments (e.g., Bates & MacWhinney, 1982; Bock, 1982), although
object assignments are receiving increasing attention (e.g., Bock & Warren,
1985; Gropen, Pinker, Hollander, & Goldberg, 1991; Gropen, Pinker, Hollander,
Goldberg, & Wilson, 1989). The sets of controllers that dominate these discus-
sions are (a) thematic or EVENT roles coupled with the primitive conceptual
features that may help to individuate these roles and (b) discourse or ATTEN-
TIONAL roles. We assume that these kinds of information are represented in
the message, and that their effects on the process of function assignment are
in part mediated by the structural and semantic conventions of the speaker's
language, importantly including the subcategorization conventions or argument
structures of lemmas represented in the lexicon.

1. Event Roles
The sets of event roles proposed in the literature vary widely, with little
agreement about appropriate criteria for individuating them. Most of the sets
include something corresponding to an agent (the instigator of an event), a
patient or theme (a person or object that is affected, moved, or located), a
recipient or goal (a beneficiary or moved-to location), an experiencer or instru-
ment (the vehicle of an event or action), as well as other roles such as time
and source.
There is a seductive and well-known correspondence between event roles
and grammatical relations. Agents are often subjects, patients are often direct
objects, and recipients are often indirect objects. However., there are both
systematic and idiosyncratic violations of these correspondences: Agents some-
times appear as the oblique (by) objects of passive verbs and patients, recipients
sometimes serve as subjects of certain active verbs (e.g., undergo, receive),
and the same participants standing in roughly the same conceptual relationship
sometimes appear in different grammatical relations (e.g., Many people fear
snakes; Snakes frighten many people). Because the mapping between event
roles and functional roles seems to be heavily influenced by the specific require-
ments of different verbs and verb forms (Grimshaw, 1990), one of the most
important factors in the control of functional role assignment is the choice of
the verb during lexical selection. This has so far received little attention in
production research (but see Jarvella & Sinnott, 1972; Gropen et al., 1989,
1991).
The difficulty of specifying a uniform set of event roles has led to various
linguistic proposals for reducing them to more primitive meaning relations
(Bierwisch, 1986; Jackendoff, 1987). An array of psycholinguistic evidence
suggests that these relations are in some way bound up with such substantive
notions as animacy (see Bock et al., 1992, for review) and concreteness (Bock
& Warren, 1985; Clark & Begun, 1971; C. T. James, Thompson, & Baldwin,
1973). In general, this work suggests that the more animate or concrete the
participant in an event, the more likely it is to appear in the subject relation in
an utterance.
The simplest interpretation of many of these results is that animate or
concrete elements are more likely to appear early in a string of words. However,
CHAPTER 29 GRAMMATICAL ENCODING 965

there is evidence that implicates functional role assignment rather than serial
positioning in these effects. It comes from experiments in which the effects
of animacy and concreteness on word order in conjunctions (where the role
assignments are the same but the positions of the words differ; e.g., the farmer
and the refrigerator vs. the refrigerator and the farmer) were contrasted with
their effects on the order of constituents in sentences (where the roles as
well as the positions differ; e.g., The farmer bought the refrigerator vs. The
refrigerator was bought by the farmer). The results show that the simple word
ordering impact of these factors is weak to nonexistent, whereas the ordering
variations that follow from changes in grammatical role assignment are robust
(Bock & Warren, 1985; McDonald, Bock, & Kelly, 1993).

2. Attentional Roles
Event roles and attentional roles are intimately related, insofar as different
event roles naturally vary in attentional values. For example, Osgood (1980)
emphasized the natural perceptual prominence of agents that derives from their
movements, and studies of visual attention have confirmed such a tendency in
young children (Robertson & Suci, 1980). Still, relative attentional values may
vary with changes in the relative prominence of participants, with corresponding
consequences for functional role assignments.
This is a natural expectation that has surprisingly weak confirmation in
studies of event or scene descriptions (Bates & Devescovi, 1989; Flores
d'Arcais, 1987; Osgood, 1971; Sridhar, 1989). The general finding from such
studies is that when the elements' event roles and animacy (for example) are
equated, variations in the prominence of elements within events have only
weak effects on function assignments.
A much more powerful influence is found when prominence is manipulated
by discourse or conversational means. Perhaps the most potent device is a
question. Imagine that a person observes a scene in which a girl chases a boy
and then is asked What was going on with the girl? or What was going on with
the boy? Many studies have shown that in these or similar circumstances, the
questioned entity tends to be assigned the subject role in the answer (Bates
& Devescovi, 1989; Bock, 1977; Carroll, 1958). Pictures of individual event
participants or single words referring to them, presented as cues for the descrip-
tion of previously or subsequently apprehended events, seem to have similar
effects (Bock, 1986a; Bock & Irwin, 1980; Flores d'Arcais, 1975; Perfetti &
Goldman, 1975; Prentice, 1967; Turner & Rommetveit, 1968).
It is a short step to the information structure of sentences in discourse. By
information structure, we mean the distribution of given (or topical) and new
information (Clark & Haviland, 1977: Halliday, 1970). The linguistic marking
of given information differs from that of new information in a variety of ways,
including prosody (Cutler & Isard, 1980; Fowler & Housum. 1987; Needham,
1990; but see Eefting, 1991) and positioning within sentences (MacWhinney &
Bates, 1978; Smith, 1971). Linked to given information's general tendency to
appear early in sentences is its affinity for the subject relation (Tomlin, 1986).
It seems likely that the sentence-level effects of topicalization are attribut-
able to forces similar to those responsible for the effects of concreteness of
individual entities. Both may be regarded as increasing the definiteness or
relative mental prominence of participants in the events that sentences describe.
966 KATHRYN BOCK AND WILLEM LEVELT

Bock and Warren (1985) termed this mental prominence CONCEPTUAL ACCESSI-
BILITY.
Although grammatical functions could in principle be assigned in any order
or even all at once, there are reasons to suspect that, at least in English, there
is a preference for combinations of elements that permit the nominative function
to be assigned first. Elements that are accessible (in the senses described above)
tend to appear as subjects more often than as objects (see Bock, 1987, for
review), particularly when accessibility arises from the message or the meaning
rather than from factors that primarily affect the word form (such as frequency or
phonological simplicity; see Bock, 1986a; Levelt & Maassen, 1981; McDonald et
al., 1993; Streim & Chapman, 1987). This tendency finds a reflection in
proposals about hierarchies of grammatical functions or relations (Keenan &
Comrie, 1977), in which subjects dominate all other functions. From a pro-
cessing standpoint, the advantage of such an arrangement is clear: Things that
present themselves more prominently or more readily are given a function that
allows them to lead in the utterance itself.

C. What Is the Nature and Organization of the Processes That Carry

Out Function Assignments?
Woven into the preceding discussion was a claim about the organization of
function assignment that we now consider explicitly. It is that verbs somehow
control function assignment.
A verb's specification of its normally expressed arguments may serve to
organize function assignment around a unit that is roughly equivalent to the
clause. A simple one-clause sentence such as She was handing him some
broccoli consists of a single main verb and its arguments. The verb hand
requires three arguments, an agent, a recipient, and a theme. During functional
processing, the element corresponding to the agent should be assigned the
nominative function, the one corresponding to the recipient should be assigned
the dative function, and the one corresponding to the theme should be assigned
the accusative function. The realization of these as the subject, first object (the
object that immediately follows the verb), and second object creates a full or
simple clause.
One of the implications of this view of the organization of production was
examined by Bock and Cutting (1992). Their method involved the elicitation
of a type of verb agreement error called an ATTRACTION ERROR. Such errors
occur when the head of the subject noun phrase is separated from the verb, as
are generalization and are in the observed error The only generalization I would
dare to make about our customers are that they're pierced. Bock and Cutting's
speakers were asked to convert complex subject phrases into full sentences by
completing them. The phrases contained a head noun (e.g., The claim) followed
either by a phrase postmodifier of the head (as in The claim about the newborn
baby . . .) or a clause postmodifier of the head (as in The claim that wolves
had raised the baby . . .). Although these subject phrases differed in structural
complexity, they were equated in length (in terms of numbers of syllables).
The critical fragments ended in a plural noun (babies) intended to elicit verb
agreement errors in the completions (cf. Bock & Miller, 1991). The question
was whether the clause postmodifier would promote or retard this tendency
relative to the phrase postmodifier.
CHAPTER 29 GRAMMATICAL ENCODING 967

A simple sequential view of production suggests that the clause imposes a

processing load analogous to the problems created by clauses in comprehension
(Caplan, 1972; Jarvella, 1971), predicting an increase in errors after clauses.
Alternatively, if production is hierarchically organized, guided by the require-
ments of verbs, the prediction is that clause postmodifiers will actually reduce
the number of errors. Consider a fragment completion along the lines of The
claim that wolves had raised the babies was rejected. Here, agreement in the
outer clause (e.g., The claim was rejected) may be partially protected from the
material in the inner clause (wolves had raised the babies). Because the error-
eliciting word babies occurs within a different clause (bound to a different verb)
in The claim that wolves had raised the babies was rejected than in The claim
about the newborn babies was rejected, the agreement operation may be pro-
tected from the irrelevant plural. The results from three experiments supported
the "protection" hypothesis: Errors were more likely to occur after phrase
than after clause postmodifiers. This points to clauses as important organizing
forces in functional processing.
The centrality of the verb to this organization becomes even clearer when
the straightforward equation between verbs and clauses breaks down. Not all
clauses are full, simple ones, and these divergences offer a better glimpse of
the role that the verb may play.
Ford and Holmes (1978; also see Ford, 1982; and Holmes, 1988) examined
such cases in a study in which subjects spoke extemporaneously on a prescribed
topic while at the same time monitoring for a randomly presented auditory
tone. The reaction times to the tones were then analyzed as a function of
their location in the subject's speech stream. The critical locations were the
beginnings and ends of functional verb units that did and did not straightfor-
wardly correspond to the beginnings and ends of simple clauses. For example,
I began the book is a simple clause with only one functional verb unit, whereas
I began working a lot harder contains two functional verb units, one for the
finite (tensed) verb began and a second for the nonfinite verb working. The
results revealed a reliable increase in tone detection latencies at the ends of
functional verb units, regardless of whether those units corresponded to simple
clauses.
Other results consistent with verb-centered control of function assignment
comes from evidence about the minimum scope of advance preparation in
production (Lindsley, 1975), which seems to require at least some planning of
the verb. Evidence about the maximum scope comes primarily from contextual
speech errors, errors in which the source seems to be interference from another
element of the intended utterance. The wide majority of such errors originate
from material in the same clause (Garrett, 1980). However, word exchange
errors originate in adjoining clauses 20% of the time, leading Garrett (1980) to
the suggestion that no more than two clauses may be planned at once. Holmes
(1988) discusses whether such two-clause errors typically involve verbs that
take clausal arguments (and so require the formulation of two clauses at once),
but the question remains open.
Finally, there is an intriguing (but inconclusive) asymmetry between verbs
and the other major grammatical categories (nouns, adjectives, and adverbs) in
their susceptibility to semantic substitution. Hotopf (1980) reported error data
from both English and German which suggests that the tendency for verbs to
undergo semantic substitution is vastly lower, both in actual incidence and as
968 KATHRYN BOCK AND WILLEM LEVELT

a percentage of opportunities. This resistance to substitution could stem from

the centrality of the verb to higher level production processes. But it could
also be because the lexical organization of verbs is different from that of nouns
(Huttenlocher & Lui, 1979) or because the nature of meaning relationships
among verbs makes the diagnosis of a substitution more difficult (Gentner &
France, 1988).

D. Summary of Functional Processing

Functional processing, as we have described it, yields an activated set of lemmas
and a set of syntactic functions, linked together via the argument structures of
the lemmas (notably that of the verb). This is illustrated in Figure 5. Beyond this,
there must be a specification of individual elements, such as the indefiniteness of
the "broccoli" argument indicated by some, the past progressive nature of the
action, and the singularity of the verb. We show some of these specifications
as annotations on the argument structure in Figure 5 but postpone their discus-
sion until we get to the topic of inflection below.

IV. CONSTITUENT ASSEMBLY

The partial functional structure in Figure 5 consists of temporary (and therefore

labile) linkages among stored elements and carries no intrinsic order. To convert
this into an utterance, something has to impose a sequence on the elements.
There is a great deal of evidence that in order to do this, speakers follow

MESSAGE
THEME RECIPIENT
AGENT
ACTION

noun
broccoli
she

verb
him

pronoun
hand

hand DATIVE
NOM PAST (=recipient)
(=agent) PROGRESSIVE
SINGULAR
ACCUSATIVE
(=theme)

FlG. 5 The products of functional processing.

CHAPTER 29 GRAMMATICAL ENCODING 969

something like the scheme specified in a hierarchical constituent structure. The

evidence comes from formal analysis, from pauses in speech, and from errors
in sentence recall.
Formal linguistic analysis provides the traditional arguments for hierarchi-
cal structure. Without such a notion it is difficult to explain structural ambiguity
(as found in the alternative readings of The old [men and women] were left
behind in the village vs. The [old men] and [women] were left behind in the
village), or sentence segmentation (why a sentence such as The girl that kissed
the boy blushed is not understood to assert that a boy blushed, despite the fact
that it contains the sequence the boy blushed), or verb agreement (verbs agree
not with what immediately precedes them, in a positional sense, but with a
particular constituent structure category, roughly, the highest noun phrase in
the same clause; compare The boy who watched the clowns was amused and
The boys who watched the clown were amused).
Data from language performance indicate that such structures somehow
characterize the products of speech production processes. Normal prosodic
patterns (Cooper, Paccia, & Lapointe, 1978; Grosjean, Grosjean, & Lane, 1979)
and hesitations (Boomer, 1965; Butterworth, 1980; Butterworth & Beattie, 1978;
Maclay & Osgood, 1959; see Garrett, 1982, for review) have been argued to
reflect structures that are larger than individual words but smaller than full
clauses. Although pause patterns are multiply determined, reflecting forces
other than syntactic structure (Gee & Grosjean, 1983; Levelt, 1989; Selkirk,
1984), they appear to be heavily influenced by phrase structure. Likewise, the
products of sentence recall (which are also products of language production)
indicate that speakers organize sentences in terms of phrasal constituents (John-
son, 1965, 1966a, 1966b).
Such things help to establish that speakers create utterances that have
hierarchically organized phrase groupings, or frames. However, they say noth-
ing about the information that is encoded or elaborated in frames or about the
processes that create them. The next two sections review those questions.

A. What's in a Frame?
The structure of a sentence could in principle reflect the structure of any of
several different sorts of information, including event role information, syntactic
function information, and prosodic information. Since phrase structure often
confounds these possibilities, it is difficult to disentangle them by observation
alone. Bock and Loebell (1990) employed an experimental approach to this
issue that relies on a tendency among speakers to use the same form repeatedly,
sometimes with different words (Bock, 1986b; Levelt & Kelter, 1982; Schenk-
ein, 1980; Tannen, 1987; Weiner & Labov, 1983). Bock and Loebell examined
whether the form repetition tendency changed when the repeated structures
represented different event roles or different prosodic patterns. They found no
effects of these variations, although the form repetition tendency itself was
clearly in evidence. Together with the findings of Bock et al. (1992), the appear-
ance is that the structure is formed under the control of information that is not
readily interpretable as conceptual, semantic, or prosodic.
The obvious alternative candidates are the syntactic functions and the
grammatical categories of the lemmas that realize them. For example, subjects
970 KATHRYN BOCK AND WILLEM LEVELT

are typically configured in one way within a sentence structure and direct
objects in another. Nouns occur as the heads of noun phrases, verbs as the
heads of verb phrases, prepositions as the heads of prepositional phrases, and
so on. So, given the nominative function and a noun lemma to fill it, adequate
information is available to create or retrieve the rudiments of a subject noun
phrase in the proper position in an utterance frame.

B. The Processes of Constitutent Assembly

An influential theory of phrase structure elaboration was proposed by Yngve
(1960). According to the model, production processes generate a phrase struc-
ture tree from top to bottom and left to right, so the first part of the sentence
to be elaborated is the leftmost daughter. As the processor traverses this branch,
it stores information about commitments to rightward branches that have yet
to be elaborated. For example, to produce the simple structure shown below
(which would be appropriate for an uninflected version of a sentence such as
Our dog chases squirrels).

the generator starts at the top, where a commitment must be made to one
rightward elaboration (the verb phrase) and proceeds leftward past the first
node (the noun phrase), where it incurs another commitment to a rightward
elaboration (the head noun) and terminates at the determiner branch. The
number of commitments at this point, two, is shown in parentheses. The genera-
tor then returns to elaborate the noun phrase by creating the noun branch.
Finally, it returns to the top and proceeds to elaborate the verb phrase.
Yngve's theory makes a very concrete prediction about the effects of
branching on the difficulty of producing a sentence. Since commitments to right
branches are stored while the generator elaborates other branches to the left,
the cost of storage may appear as an impairment to fluency, perhaps as a slowing
of speech rate or as an increase in the probability of error. The number of such
commitments grows as a function of the depth in the tree of the left branches,
with a corresponding increase in the storage cost. Storage cost is typically
assessed by counting the number of left branches dominating each terminal
element (word) of the sentence (which yields the number of right-branching
commitments) and dividing by the total number of words (which yields a mea-
sure of mean depth). For the structure above, the mean depth is 1.0.
This model has been examined by a number of investigators, including
Johnson (1966a, 1966b), Martin and Roberts (1966, 1967; Martin, Roberts, &
Collins, 1968), and Perfetti (1969a, 1969b; Perfetti & Goodman, 1971). Little
consistent support has been found for the detailed predictions of the depth
hypothesis (see Fodor, Bever, & Garrett, 1974, Chaps. 5 and 6, and Frazier,
CHAPTER 29 GRAMMATICAL ENCODING 971

1985, for review and further discussion), perhaps because mean depth is insuffi-
ciently sensitive to structure assembly. The measure is a global one, whereas
disruptions of surface syntactic elaboration may be local. Likewise, most tests
of the depth hypothesis have employed methods (such as sentence recall) that
are not suited to the detection of a speaker's transient encoding problems.
Some support for a broader implication of Yngve's model came from experi-
ments by Forster (1966, 1967, 1968a, 1968b). Forster looked at the ease and
speed of completing sentences that had words deleted at the beginning or end.
He found that it was more difficult to create the beginnings of sentences, as
would be expected if the existence of rightward commitments burdens the
generation of sentences. Evidence that this was not exclusively the result of
practice in generating sentences from left to right came from comparisons of
sentence completion performance across languages that differed in the degree
to which their sentences characteristically branch to the left.
Still, there is something highly artificial about the task that Forster em-
ployed, although it is an artificiality that is built into Yngve's model. Speakers
may rarely know exactly how their sentences will end before they begin them,
but depth calculations cannot be made precisely unless they do. So, while
Forster's experiments generally supported the original theory, they involved a
task that diverges from ordinary production in just the way that the theory does,
making it unclear whether the results can be generalized to normal formulation
processes.
A related form of support for Yngve's theory comes from the tendency for
"heavier" or more complex constituents to appear later in sentences, which
reduces their depth. Thus, the sentence The clerk showed the woman a book
with a picture on its cover of Nancy Reagan glaring at Raisa Gorbachev sounds
much more natural than The clerk showed a book with a picture on its cover
of Nancy Reagan glaring at Raisa Gorbachev to the woman. There is no
comparable disparity between the formally similar sentences The clerk showed
the woman a book and The clerk showed a book to the woman. A related
phenomenon occurs in language acquisition, where subject-elaborated noun
phrases have been found to appear later in the course of development than
object-elaborated noun phrases (Pinker, 1984). However, as Frazier (1985)
pointed out, these facts are compatible with any approach which predicts that
complex constituents tend to appear at points of low complexity within a sen-
tence.
A computational model that avoids the pitfalls of Yngve's approach has been
proposed by de Smedt (1990; also see Kempen & Hoenkamp, 1987; Lapointe &
Dell, 1989). What distinguishes the model is that it permits incremental produc-
tion. It does this by building pieces of phrase structure as the lemmas and
function assignments that demand particular phrasal fragments become avail-
able, and fitting the fragments together according to constraints on possible
unifications (see Kay, 1985, for a discussion of unification procedures). The
phrase structure is thereby assembled in a piecemeal and heuristic fashion
under the control of lemmas and their functions, rather than by means of an
algorithm that generates a tree into which words must be inserted. The predic-
tions of the model seem most likely to concern problems that might arise
during unification attempts among incompatible fragments, but these predictions
remain to be worked out and tested.
972 KATHRYN BOCK AND WILLEM LEVELT

V. INFLECTION

In order to examine a heated controversy in the production literature, we

consider under the heading of inflection not only inflection proper, but also the
formulation of the function words that are often associated with grammatical
phrases of different types (e.g., determiners for noun phrases, auxilliaries for
verb phrases, and prepositions for prepositional phrases). Function words and
inflectional affixes (together with derivational affixes,6 which we will largely
ignore) constitute the elements of the CLOSED CLASS, SO called because its
inventory (both in the language and in the vocabulary of individual adult speak-
ers) undergoes change much more slowly than the inventory of the open class
(nouns, verbs, adjectives, and adverbs).
One source of the controversy is a relatively undisputed fact about speech
errors: The elements of the closed class are less likely to be involved than
elements of the open class. So, the words in blends, in semantic and phonological
substitutions, and in exchanges tend to be members of the open class. Even
sound errors, which are much more likely to be indiscriminate about syntactic
classifications than word errors, seem to be constrained by open and closed
class membership, occurring principally within open class words in spontaneous
speech.
At issue is how to account for this regularity. Open and closed class words
differ in many other ways, and these differences suggest alternative accounts
for their behavior. So, open class words by and large occur less frequently
than closed class words, they are learned later in language acquisition, they
are longer, and they are more likely to bear stress. Such factors, alone or
together, could create a predisposition to error that has nothing to do with word
class per se.
In the following sections, we present two alternative accounts of how the
elements of the closed class receive their places within a sentence structure.
Along with these accounts we consider some of the other evidence that has
been brought to bear on the issue.

A. Inflections as Frame Features

Beyond the general features of the behavior of closed class elements in errors,
Garrett (1982) has called on another disparity between them and open class
elements in arguing that the closed class is a special word class or separate
vocabulary. Among some aphasics, the closed class is disproportionately absent
from speech (Saffran, Schwartz, & Marin, 1980), despite the general rule that
high-frequency words are more likely to be preserved in aphasic speech. Garrett
has also presented an analysis of normal speech errors which suggests that
they are more likely to occur among open class words even when frequency
is controlled (Garrett, 1990).
6
The distinction between derivational and inflectional affixes is based in part upon whether
they change the grammatical category of the word to which they apply. By this criterion, the plural
affix for nouns and number, tense, and aspect affixes for verbs are inflectional, whereas derivational
affixes change verbs into nouns (e.g., -tion, as in creation), nouns into verbs (e.g., -ate, as in
pulsate), nouns into adjectives (e.g., -ly, as in princely), and so on. However, not all derivational
affixes change form class (e.g., un-, mis-).
CHAPTER 29 GRAMMATICAL ENCODING 973

To account for such evidence, Garrett (1982) argued that the elements of
the closed class are intrinsic features of the grammatical frame. Unlike open
class words, which have to be linked to the frame in some fashion (e.g., by
being assigned a grammatical function or a position), the closed class elements
in an important sense are the frame, serving to define as well as mark the
functions and grammatical features (e.g., definite, plural, past tense, and so
on) of the open class words.
For this to happen, we might imagine that during functional assignment,
each function is tagged with additional specifications appropriate to its realiza-
tion. For example, if the subject is specified as definite and plural, the frame
generated for the subject noun phrase should include, in addition to a branch
for the head noun itself, a definite determiner branch and a branch for the plural
inflection, along the lines of

Since the additional branches are nothing more than the corresponding closed
class elements themselves, they may be directly encoded after a few minor
morphophonological adjustments.

B. A Mixed Model
The challenges to the frame view come from at least two directions. One focuses
on an alternative explanation for the disparate behaviors of open and closed
class elements in speech errors, seeking to attribute them to the differences in
frequency of the class members. These differences are enormous: In the CELEX
database (which includes almost 18 million words of English text; Burnage,
1990), the 70 most frequent English words are function words (though some of
them, like can, have content usages) and range from 60,958 occurrences per
million (the) to 1,798 per million (now); the first unambiguously open class word
(time) has a frequency of 1,791 per million. Examinations of the relationship
between word frequency and error proclivity have shown that infrequent forms
are more likely to participate in errors than frequent forms (Dell, 1990; Stem-
berger & MacWhinney, 1986), consistent with a frequency hypothesis.
A second challenge is directed at the claim that closed class words define
the frame. To test this, Bock (1989) examined whether structural repetition
(the tendency to repeat similar phrase structures across successive sentences)
is dependent on identity of closed class elements. She found equally strong
structural repetition when the closed class members of sentences were different
or the same, suggesting that the phrasal configurations of sentences are con-
trolled by forces that are not fully equatable with their closed class elements.
974 KATHRYN BOCK AND WILLEM LEVELT

To accommodate such challenges to Garrett's view of the inflection process,

Lapointe and Dell (1989) offered a modified account that distinguishes between
the free standing elements of the closed class (such as determiners and auxillia-
ries) and inflectional affixes (such as the plural and past tense). In their model,
only the affixes are given directly in the frame. The freestanding function words
have to be inserted by an additional operation, so that the frame for the noun
phrase shown above would appear as something closer to the following.

The motivation for this treatment was an analysis by Lapointe (1985) of

simplification errors in the speech of English- and Italian-speaking aphasics.
Within the errors, Lapointe noted a difference in the behavior of freestanding
function words and affixes. Whereas function words tended to be omitted,
affixes tended to be replaced with other affixes, suggesting that affixes are in
some sense more intrinsic to the frame than function words.
The model's treatment of function words nonetheless distinguishes them
from content words as well as from affixes. They differ from content words in
the mechanism by which their phonological or morphosyntactic representation
is linked to the frame, prior to phonological encoding, maintaining something
of the spirit of the separate class view.7 Specifically, the assumption is that for
each designated function word slot there is only one filler, so that there is
no competition among candidates. However, during phonological encoding,
function and content words undergo the same operations, suggesting that, other
things equal, open and closed class words should be equally prone to sound
errors. This prediction received support from experimental studies of error
elicitation reported by Dell (1990).

C. The Generation of Bound Inflections

Spontaneous speech errors strongly suggest that bound inflected forms are
accessed separately from stem forms during generation. Most of the evidence
for this comes from stranding errors such as the one cited in the introduction
(You ordered up ending some fish dish; Garrett, in press). Stemberger (1985)
found that inflectional affixes were stranded in 88.9% of the errors in which it
was possible in his corpus. Both the frame model and the mixed model imply
that stranding is a consequence of normal frame generation coupled with some
failure of lexical access, and not a frame generation problem, in agreement
with Stemberger (1985). The question to be addressed here, then, is how the
hierarchical framework comes to have the appropriate configuration to control
the appearance of the bound elements of the closed class.

7
However, this separate class does not constitute a separately stored vocabulary, but a class
of words whose use is heavily constrained by syntactic features.
CHAPTER 29 GRAMMATICAL ENCODING 975

For many elements of the closed class, including the freestanding ones,
the notion of indirect election can be called on to explain how such elements
become part of the frame. As discussed in the section on lexical selection,
certain lemmas carry specifications about the closed class elements that can
or must accompany them. These specifications may be represented in a way
that can be directly incorporated into a structural frame, so that the choice of
a lemma that carries such information guarantees the compilation of the element
into the developing utterance. For example, if the plural form of goat (goats)
is selected, the lemma should mandate the construction of a noun phrase in
which the stem of the head noun is affixed with the plural /-s/. However, if the
choice were the plural form of sheep (sheep), no affix would be called for.
A related but more difficult question has to do with the circumstances that
lead to the selection of lemmas that require closed class elements. In some
cases these selections may be under direct control of message elements, as
when a verb is specified for past tense. But in others the connection to message
features can be less straightforward, as when there are syntactic dependencies
among inflectional features. So, why do speakers say She was handing him some
broccoli rather than She were handing him some broccoli? In such dependent
relationships, two (or more) constituents of a sentence reflect a value of some
feature that triggers an inflectional variation. These constituents need not be
adjacent: In subject-verb agreement, for example, agreement can cross an
indeterminate amount of intervening material. What is necessary is that the
agreeing constituents stand in appropriate structural or syntactic-functional
relationships (so in English, agreement operates between the head of the subject
noun phrase and the finite verb).
Indirect evidence about the workings of the agreement operation in produc-
tion comes from studies of errors of attraction. As noted above, attraction
errors have the property that the number of the verb agrees with the number
of some (usually plural) constitutent of the sentence other than the head noun
phrase. Assuming that such errors are constrained by the factors that control
normal agreement (an assumption that obviously may be wrong), Bock and
Miller (1991) and Bock and Eberhard (1993) explored how various number
characteristics affect the incidence of attraction errors in speech. These charac-
teristics included the "multipleness" of the referent of the subject (as it might be
represented in the message; cf. Pollard & Sag, 1988), the semantic multipleness
versus grammatical plurality of the attracting noun phrase, the regularity of
plural marking, and spurious surface features of plurality (plural-like pronounci-
ation, as in the word rose). The only factor that reliably created attraction
errors was grammatical plurality (i.e., subcategorized plurality) of the attracting
noun. Because grammatical plurality is a property of lemmas rather than of
nonlinguistic concepts or messages, lemmas may be the principal source of
number agreement features in English utterances.
The obvious place to state this dependency in the general architecture we
have set out is within functional processing, since it is there that the relevant
relationships are represented. In functional processing terms, the creation of
the dependency requires that the finite (tense and number carrying) verb and
the noun lemma linked to the nominative function have the same number. For
this to happen, the verb must inherit the subject's number feature, or the subject
must inherit the verb's number feature, or both must inherit the same value of
976 KATHRYN BOCK AND WILLEM LEVELT

a feature that is stated elsewhere (one possible locus is in the message, on

a linguistic argument persuasively developed by Pollard & Sag, 1988). The
distribution of attraction errors, both in spontaneous and elicited speech, points
toward the first of these alternatives.
Linked to the question of the origins of the frame features that control the
appearance of inflections is a current controversy over the representation of
regularly and irregularly inflected forms in the lexicon (Kim, Pinker, Prince,
& Prasada, 1991; Pinker, 1991; Rumelhart & McClelland, 1987). One position
in this debate is that regular and irregular forms are represented in the same
way, 8 so that there is no explicit sense in which inflected regular forms consist
of a stem and an affix. As Stemberger (1985) observed, this position is challenged
by the evidence that inflected forms in complete utterances tend to be cre-
ated—or to fall apart—in a piecemeal way, along morphological rather than
phonological lines. Such evidence adds weight to the alternative, a rule-based
origin for the production of inflected forms in connected speech.

VI. CONCLUSION

By way of summary. Figure 6 sketches the products of each set of grammatical

encoding processes that we have discussed, taking the (by now hackneyed)
example She was handing him some broccoli as the target utterance. Functional
processing serves to integrate a set of lexical specifications with a set of syntactic
functions, which in turn guide the creation of a framework for the positioning
of words. This framework controls positional processing, the output of which
is an ordered set of word forms and their inflections.
We have reviewed several types of evidence in developing this picture,
among them the constraints that have been observed on errors in spontaneous
and elicited speech. In closing we should point out one notable absence from
this discussion. Missing is Freud's (1917/1976) account of errors such as the
parliamentarian's Gentlemen, I take notice that a full quorum of members is
present and herewith declare the sitting closed. Because Freud's account has
become part of the fabric of popular culture, it is important to consider its
bearing on an explanation of how people talk. The main drawback of Freud's
analysis, as Freud himself acknowledged, is that few speech errors have discern-
ible psychodynamic content. Yet most speech errors, whether or not they carry
clues to a speaker's unconscious impulses, display an impressively regular set
of linguistic restrictions (note that the parliamentarian's slip is a thoroughly
ordinary semantic substitution). It follows that errors of speech may carry fewer
clues to the mysteries of unconscious motivation than to the mundane and
relatively mechanical underpinnings of speech.
Unfortunately, the clues about these underpinnings are sometimes ambigu-
ous or conflicting and are always open to alternative interpretations. For such

8
In a certain sense, they may not be represented at all as lemmas or lexical forms (Seidenberg
& McClelland. 1989).
CHAPTER 29 GRAMMATICAL ENCODING 977

FUNCTIONAL
PROCESSING
him
she hand DATIVE
NOMINATIVE PAST
PROGRESSIVE broccoli indefinite
SINGULAR
ACCUSATIVE

AUXILIARY
SINGULAR
PAST
stem -ing

POSITIONAL
PROCESSING
FlG. 6 An illustration of the events of grammatical encoding.

reasons, computational and experimental approaches have assumed increasing

importance in the study of language production. Computational models like
those of Dell (1986; Dell, Juliano, & Govindjee, 1993), de Smedt (1990),
Kempen and Hoenkamp (1987), and Roelofs (1992) offer concrete proposals
about the organization of processing and, in the best cases, generate specific
predictions about the consequences of the mechanisms they embody. Because
the overt characteristics of language production are readily observable and
quantifiable, the predictions made by these models are amenable to testing
across a wide array of data. For such reasons, the computational approach to
production offers great promise.
Systematic, controlled empirical testing is a necessary complement to com-
putational models, and here too, there have been promising developments. As
the present review suggests, there is now an array of experimental methods
that strategically target the underlying dynamics of production, most of them
relying on techniques (like interference and priming) that transiently sideswipe
or enhance specific subcomponents of formulation between messages and artic-
ulation. These developments are nonetheless fairly new and narrowly spread
978 KATHRYN BOCK AND WILLEM LEVELT

over the range of issues in production, in part because of the challenge of

manipulating the language production process without disrupting the fundamen-
tal features of the underlying communicative intention. 9 Critical observations
are therefore sparse at many points, making the research we have reviewed
little more than a preliminary step toward the understanding of grammatical
encoding.

ACKNOWLEDGMENTS

During the preparation of this chapter, Kathryn Bock was supported in part by a fellowship from
the Max Planck Society and by grants from the Fulbright Program and the National Science
Foundation (BNS 90-09611).

REFERENCES

Balota, D. A., & Chumbley, J. I. (1985). The locus of word-frequency effects in the pronunciation
task: Lexical access and/or production? Journal of Memory and Language, 24, 89-106.
Balota. D. A., & Chumbley, J. I. (1990). Where are the effects of frequency in visual word
recognition tasks? Right where we said they were! Comment on Monsell et al. (1989). Journal
of Experimental Psychology: General, 119, 231-237.
Bates, E., & Devescovi, A. (1989). Crosslinguistic studies of sentence production. In B. MacWhin-
ney & E. Bates (Eds.), The crosslinguistic study of sentence processing (pp. 225-253). Cam-
bridge: Cambridge University Press.
Bates, E., & MacWhinney, B. (1982). Functionalist approaches to grammar. In E. Wanner &
L. R. Gleitman (Eds.), Language acquisition: The state of the art (pp. 173-218). Cambridge:
Cambridge University Press.
Berg, T. (1987). The case against accommodation: Evidence from German speech error data.
Journal of Memory and Language, 26, 277-299.
Bierwisch, M. (1969). On certain problems of semantic representations. Foundations of Language,
5, 153-184.
Bierwisch, M. (1986). On the nature of semantic form in natural language. In F. Klix & H. Hagendorf
(Eds.), Human memory and cognitive capabilities: Mechanisms and performances (Part B,
pp. 765-784). Amsterdam: North-Holland.
Bierwisch, M., & Schreuder, R. (1992). From concepts to lexical items. Cognition, 42, 23-60.
Bock, J. K. (1977). The effect of a pragmatic presupposition on syntactic structure in question
answering. Journal of Verbal Learning and Verbal Behavior, 16, 723-734.
Bock, J. K. (1982). Toward a cognitive psychology of syntax: Information processing contributions
to sentence formulation. Psychological Review, 89, 1-47.
Bock, J. K. (1986a). Meaning, sound, and syntax: Lexical priming in sentence production. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 12, 575-586.
Bock, J. K. (1986b). Syntactic persistence in language production. Cognitive Psychology, 18,
355-387.
Bock, J. K. (1987). Coordinating words and syntax in speech plans. In A. Ellis (Ed.), Progress in
the psychology of language (pp. 337-390). London: Erlbaum.
Bock, J. K. (1989). Closed-class immanence in sentence production. Cognition, 31, 163-186.
Bock, J. K. (1990). Structure in language: Creating form in talk. American Psychologist, 45,
1221-1236.
Bock, J. K., & Cutting, J. C. (1992). Regulating mental energy: Performance units in language
production. Journal of Memory and Language, 31, 99-127.

9
The assumption that this is possible is itself a point of controversy (Butterworth & Hadar,
1989; McNeill, 1985, 1987, 1989).
CHAPTER 29 GRAMMATICAL ENCODING 979

Bock, J. K., & Eberhard, K. M. (1993). Meaning, sound, and syntax in English number agreement.
Language and Cognitive Processes, 8, 57-99.
Bock, J. K., & Irwin, D. E. (1980). Syntactic effects of information availability in sentence produc-
tion. Journal of Verbal Learning and Verbal Behavior, 19, 467-484.
Bock, J. K., & Loebell, H. (1990). Framing sentences. Cognition, 35, 1-39.
Bock, J. K., Loebell, H., & Morey, R. (1992). From conceptual roles to structural relations:
Bridging the syntactic cleft. Psychological Review, 99, 150-171.
Bock, J. K., & Miller, C. A. (1991). Broken agreement. Cognitive Psychology, 23, 45-93.
Bock, J. K., & Warren, R. K. (1985). Conceptual accessibility and syntactic structure in sentence
formulation. Cognition, 21, 47-67.
Boomer, D. S. (1965). Hesitation and grammatical encoding. Language and Speech, 8, 148-
158.
Bransford, J. D., Barclay, J. R., & Franks, J. J. (1972). Sentence memory: A constructive versus
interpretive approach. Cognitive Psychology, 3, 193-209.
Bresnan, J. (Ed.). (1982). The mental representation of grammatical relations. Cambridge, MA:
MIT Press.
Bresnan, J., & Kaplan, R. M. (1984). Grammars as mental representations of language. In W.
Kintsch, J. R. Miller, & P. G. Poison (Eds.), Method and tactics in cognitive science (pp.
103-135). Hillsdale, NJ: Erlbaum.
Brown, A. S. (1991). A review of the tip-of-the-tongue experience. Psychological Bulletin, 109,
204-223.
Brown, R., & McNeill, D. (1966). The "tip of the tongue" phenomenon. Journal of Verbal Learning
and Verbal Behavior, 5, 325-337.
Burnage, A. G. (1990). Celex: A guide for users. Nijmegen, The Netherlands: University of Nij-
megen.
Butterworth, B. (1980). Evidence from pauses in speech. In B. Butterworth (Ed.), Language
production (Vol. 1, pp. 155-176). London: Academic Press.
Butterworth, B. (1982). Speech errors: Old data in search of new theories. In A. Cutler (Ed.).
Slips of the tongue and language production (pp. 73-108). Berlin: Mouton.
Butterworth, B. (1992). Disorders of phonological encoding. Cognition, 42, 261-286.
Butterworth, B. L., & Beattie, G. W. (1978). Gesture and silence as indicators of planning in
speech. In R. N. Campbell, & P. T. Smith (Eds.), Recent advances in the psychology of
language: Formal and experimental approaches. New York: Plenum.
Butterworth, B., & Hadar, U. (1989). Gesture, speech, and computational stages: A reply to
McNeill. Psychological Review, 96, 168-174.
Caplan, D. (1972). Clause boundaries and recognition latencies for words in sentences. Perception
& Psychophysics, 12, 73-76.
Carroll, J. B. (1958). Process and content in psycholinguistics. In Current trends in the description
and analysis of behavior (pp. 175-200). Pittsburgh: University of Pittsburgh Press.
Cattell, J. M. (1885). Über die Zeit der Erkennung und Benennung von Schriftzeichen, Bildern
und Farben [The time to recognize and name letters, pictures, and colors]. Philosophische
Studien, 2, 635-650.
Clark, H. H., & Begun, J. S. (1971). The semantics of sentence subjects. Language and Speech.
14, 34-46.
Clark, H. H., & Haviland, S. E. (1977). Comprehension and the given-new contract. In R. O.
Freedle (Ed.), Discourse production and comprehension (pp. 1-40). Norwood, NJ: Ablex.
Collins, A. M., & Loftus, E. F. (1975). A spreading-activation theory of semantic processing.
Psychological Review, 82, 407-428.
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal
Learning and Verbal Behavior, 8, 240-247.
Cooper, W. E., Paccia, J. M., & Lapointe, S. G. (1978). Hierarchical coding in speech timing.
Cognitive Psychology, 10, 154-177.
Cutler, A. (1988). The perfect speech error. In L. M. Hyman & C. S. Li (Eds.), Language, speech
and mind: Studies in honour of Victoria A. Fromkin (pp. 209-223). London: Routledge.
Cutler, A., & Isard, S. D. (1980). The production of prosody. In B. Butterworth (Ed.), Language
production (Vol. 1, pp. 245-269). London: Academic Press.
Deese, J. (1984). Thought into speech: The psychology of a language. Englewood Cliffs, NJ:
Prentice-Hall.
980 KATHRYN BOCK AND WILLEM LEVELT

Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological

Review, 93, 283-321.
Dell, G. S. (1990). Effects of frequency and vocabulary type on phonological speech errors.
Language and Cognitive Processes, 5, 313-349.
Dell, G. S., Juliano, C, & Govindjee, A. (1993). Structure and content in language production:
A theory of frame constraints in phonological speech errors. Cognitive Science, 17, 149-195.
Dell, G. S., & O'Seaghdha, P. G. (1991). Mediated and convergent lexical priming in language
production: A comment on Levelt et al. (1991). Psychological Review, 98, 604-614.
Dell, G. S., & O'Seaghdha, P. G. (1992). Stages of lexical access in language production. Cognition,
42, 287-314.
Dell, G. S., & Reich, P. A. (1981). Stages in sentence production: An analysis of speech error
data. Journal of Verbal Learning and Verbal Behavior, 20, 611-629.
del Viso, S., Igoa, J. M., & García-Albea, J. E. (1991). On the autonomy of phonological encoding:
Evidence from slips of the tongue in Spanish. Journal of Psycholinguistic Research, 20,
161-185.
Dennett, D. (1991). Consciousness explained. Boston: Little, Brown,
de Saussure, F. (1955). Cours de linguistique générale. [Course in general linguistics]. (5th ed.).
Paris: Payot (Original work published in 1916).
de Smedt, K. J. M. J. (1990). Incremental sentence generation. Doctoral dissertation, Katholieke
Universiteit Nijmegen, Nijmegen, The Netherlands.
Eefting, W. (1991). The effect of "information value" and "accentuation" on the duration of Dutch
words, syllables, and segments. Journal of the Acoustical Society of America, 89, 412-424.
Fay, D. (1980). Transformational errors. In V. A. Fromkin (Ed.), Errors in linguistic performance:
Slips of the tongue, ear, pen, and hand (pp. 111-122). New York: Academic Press.
Fay, D., & Cutler, A. (1977). Malapropisms and the structure of the mental lexicon. Linguistic
Inquiry, 8, 505-520.
Fisher, C, Gleitman, H., & Gleitman, L. R. (1991). On the semantic content of subcategorization
frames. Cognitive Psychology, 23, 1-62.
Flores d'Arcais, G. B. (1975). Some perceptual determinants of sentence construction. In G. B.
Flores d'Arcais (Ed.), Studies in perception: Festschrift for Fabio Metelli (pp. 344-373).
Milan, Italy: Martello-Giunti.
Flores d'Arcais, G. B. (1987). Perceptual factors and word order in event descriptions. In G.
Kempen (Ed.), Natural language generation (pp. 441-451). Dordrecht: Martinus Nijhoff.
Fodor, J. A., Bever, T. G., & Garrett, M. F. (1974). The psychology of language. New York:
McGraw-Hill.
Fodor, J. A., Garrett, M. F., Walker, E. C. T., & Parkes, C. H. (1980). Against definitions.
Cognition, 8, 263-367.
Ford, M. (1982). Sentence planning units: Implications for the speaker's representation of meaning-
ful relations underlying sentences. In J. Bresnan (Ed.), The mental representation of grammati-
cal relations (pp. 797-827). Cambridge, MA: MIT Press.
Ford, M., & Holmes, V. M. (1978). Planning units and syntax in sentence production. Cognition,
6, 35-53.
Forster, K. I. (1966). Left-to-right processes in the construction of sentences. Journal of Verbal
Learning and Verbal Behavior, 5, 285-291.
Forster, K. I. (1967). Sentence completion latencies as a function of constituent structure. Journal
of Verbal Learning and Verbal Behavior, 6, 878-883.
Forster, K. I. (1968a). The effect of removal of length constraint on sentence completion times.
Journal of Verbal Learning and Verbal Behavior, 7, 253-254.
Forster, K. I. (1968b). Sentence completion in left- and right-branching languages. Journal of
Verbal Learning and Verbal Behavior, 7, 296-299.
Foss, D. J., & Hakes, D. T. (1978). Psycholinguistics: An introduction to the psychology of
language. Englewood Cliffs, NJ: Prentice-Hall.
Fowler, C. A., & Housum, J. (1987). Talkers' signaling of " n e w " and "old" words in speech and
listeners' perception and use of the distinction. Journal of Memory and Language, 26,489-504.
Frazier, L. (1985). Syntactic complexity. In D. R. Dowty, L. Karttunen, & A. M. Zwicky (Eds.),
Natural language parsing: Psychological, computational, and theoretical perspectives (pp.
129-189). Cambridge: Cambridge University Press.
Freud, S. (1976). Introductory lectures on psychoanalysis. In J. Strachey (Ed.), The complete
psychological works (Vol. 15). New York: Norton. (Original work published 1917).
CHAPTER 29 GRAMMATICAL ENCODING 981

Fromkin, V. A. (1971). The non-anomalous nature of anomalous utterances. Language, 47,

27-52.
Fromkin, V. A. (Ed.). (1973). Speech errors as linguistic evidence. The Hague: Mouton.
Garnham, A., Shillcock, R. C, Brown, G. D. A., Mill, A. I. D., & Cutler, A. (1982). Slips of the
tongue in the London-Lund corpus of spontaneous conversation. In A. Cutler (Ed.), Slips
of the tongue and language production (pp. 251-263). Berlin: Mouton.
Garrett, M. F. (1975). The analysis of sentence production. In G. H. Bower (Ed.), The psychology
of learning and motivation (pp. 133-177). New York: Academic Press.
Garrett, M. F. (1980). Levels of processing in sentence production. In B. Butterworth (Ed.),
Language production (Vol. 1, pp. 177-220). London: Academic Press.
Garrett, M. F. (1982). Production of speech: Observations from normal and pathological language
use. In A. Ellis (Ed.), Normality and pathology in cognitive functions (pp. 19-76). London:
Academic Press.
Garrett, M. F. (1988). Processes in language production. In F. J. Newmeyer (Ed.), Linguistics:
The Cambridge survey: III. Language: Psychological and biological aspects (pp. 69-96).
Cambridge: Cambridge University Press.
Garrett, M. F. (1990, March). Processing vocabularies in language production. Paper presented
at the CUNY Conference on Human Sentence Processing, New York.
Garrett, M. (1992). Disorders of lexical selection. Cognition, 42, 143-180.
Garrett, M. F. (in press). Errors and their relevance for models of language production. In G.
Blanken, J. Dittman, H. Grim, J. Marshall, & C. Wallesch (Eds.), Linguistic disorders and
pathologies. Berlin: de Gruyter.
Gee, J. P., & Grosjean, F. (1983). Performance structures: A psycholinguistic and linguistic ap-
praisal. Cognitive Psychology, 15, 411-458.
Gentner, D., & France, I. M. (1988). The verb mutability effect: Studies of the combinatorial
semantics of nouns and verbs. In S. I. Small, G. W. Cottrell, & M. K. Tanenhaus (Eds.),
Lexical ambiguity resolution (pp. 343-382). San Mateo, CA: Morgan Kaufman.
Glaser, W. R. (1992). Picture naming. Cognition, 42, 61-105.
Glaser, W. R., & Düngelhoff, F.-J. (1984). The time course of picture-word interference. Journal
of Experimental Psychology: Human Perception and Performance, 10, 640-654.
Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press.
Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Syntax and semantics in the acquisi-
tion of locative verbs. Journal of Child Language, 18, 115-151.
Gropen, J., Pinker, S., Hollander, M., Goldberg, R., & Wilson, R. (1989). The learnability and
acquisition of the dative alternation in English. Language, 65, 203-257.
Grosjean, F., Grosjean, L., & Lane, H. (1979). The patterns of silence: Performance structures
in sentence production. Cognitive Psychology, 11, 58-81.
Halliday, M. A. K. (1970). Language structure and language function. In J. Lyons (Ed.), New
horizons in linguistics (pp. 140-165). Baltimore, MD: Penguin.
Hankamer, J. (1989). Morphological parsing and the lexicon. In W. Marslen-Wilson (Ed.), Lexical
representation and process (pp. 392-408). Cambridge, MA: MIT Press.
Harley, T. A. (1984). A critique of top-down independent levels models of speech production:
Evidence from non-plan-internal speech errors. Cognitive Science, 8, 191-219.
Heeschen, C. (in press). Morphosyntactic characteristics of spoken language. In G. Blanken, J.
Dittman, H. Grim, J. Marshall, & C. Wallesch (Eds.), Linguistic disorders and pathologies.
Berlin: de Gruyter.
Holmes, V. M. (1988). Hesitations and sentence planning. Language and Cognitive Processes, 3,
323-361.
Hotopf, W. H. N. (1980). Semantic similarity as a factor in whole-word slips of the tongue. In
V. A. Fromkin (Ed.), Errors in linguistic performance: Slips of the tongue, ear, pen, and
hand (pp. 97-109). New York: Academic Press.
Humphreys, G. W., Riddoch, M. J., & Quinlan, P. T. (1988). Cascade processes in picture identifi-
cation. Cognitive Neuropsychology, 5, 67-103.
Huttenlocher, J., & Lui, F. (1979). The semantic organization of some simple nouns and verbs.
Journal of Verbal Learning and Verbal Behavior, 18, 141-162.
Jackendoff, R. (1987). The status of thematic relations in linguistic theory. Linguistic Inquiry, 18,
369-411.
James, C. T., Thompson, J. G., & Baldwin, J. M. (1973). The reconstructive process in sentence
memory. Journal of Verbal Learning and Verbal Behavior, 12, 51-63.
982 KATHRYN BOCK AND WILLEM LEVELT

James, W. (1950). The principles of psychology (Vol. 1). New York: Dover. (Original work published
1890).
Jarvella, R. J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and
Verbal Behavior, 10, 409-416.
Jarvella, R. J., & Sinnott, J. (1972). Contextual constraints on noun distributions to some English
verbs by children and adults. Journal of Verbal Learning and Verbal Behavior, 11, 47-53.
Jescheniak, J. D., & Levelt, W. J. M. (in press). Word frequency effects in speech production:
Retrieval of syntactic information and of phonological form. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition.
Johnson, N. F. (1965). The psychological reality of phrase-structure rules. Journal of Verbal
Learning and Verbal Behavior, 4, 469-475.
Johnson, N. F. (1966a). The influence of associations between elements of structured verbal
responses. Journal of Verbal Learning and Verbal Behavior, 5, 369-374.
Johnson, N. F. (1966b). On the relationship between sentence structure and the latency in generating
the sentence. Journal of Verbal Learning and Verbal Behavior, 5, 375-380.
Kay, M. (1985). Parsing in functional unification grammar. In D. R. Dowty, L. Karttunen, &
A. M. Zwicky (Eds.), Natural language parsing: Psychological, computation, and theoretical
perspectives (pp. 251-278). Cambridge: Cambridge University Press.
Keenan, E. L. (1976). Towards a universal definition of "subject." In C. N. Li (Ed.), Subject and
topic (pp. 303-333). New York: Academic Press.
Keenan, E. L., & Comrie, B. (1977). Noun phrase accessibility and universal grammar. Linguistic
Inquiry, 8, 63-99.
Kelly, M. H. (1992). Using sound to solve syntactic problems: The role of phonology in grammatical
category assignments. Psychological Review, 99, 349-364.
Kempen, G.. & Hoenkamp, E. (1987). An incremental procedural grammar for sentence formula-
tion. Cognitive Science, 11, 201-258.
Kim, J. J., Pinker, S.. Prince, A., & Prasada, S. (1991). Why no mere mortal has ever flown out
to center field. Cognitive Science, 15, 173-218.
Lachman, R., Shaffer, J. P., & Hennrikus, D. (1974). Language and cognition: Effects of stimulus
codability, name-word frequency, and age of acquisition on lexical reaction time. Journal of
Verbal Learning and Verbal Behavior, 13, 613-625.
Lapointe, S. G. (1985). A theory of verb form use in the speech of agrammatic aphasics. Brain
and Language, 24, 100-155.
Lapointe, S. G., & Dell, G. S. (1989). A synthesis of some recent work in sentence production.
In G. N. Carlson & M. K. Tanenhaus (Eds.), Linguistic structure in language processing
(pp. 107-156). Dordrecht: Kluwer.
Levelt, W. J. M. (1983). Monitoring and self-repair in speech. Cognition, 14, 41-104.
Levelt, W. J. M. (1989). Speaking: From intention to articulation. Cambridge, MA: MIT Press. '
Levelt, W. J. M. (1992). Accessing words in speech production: Stages, processes and representa-
tions. Cognition, 42, 1-22.
Levelt, W. J. M., & Kelter, S. (1982). Surface form and memory in question answering. Cognitive
Psychology, 14, 78-106.
Levelt, W. J. M., & Maassen, B. (1981). Lexical search and order of mention in sentence production.
In W. Klein & W. Levelt (Eds.), Crossing the boundaries in linguistics (pp. 221-252). Dor-
drecht: Reidel.
Levelt, W. J. M., Schriefers, H., Vorberg, D., Meyer. A. S., Pechmann, T., & Havinga, J.
(1991a). The time course of lexical access in speech production: A study of picture naming.
Psychological Review, 98, 122-142.
Levelt, W. J. M., Schriefers, H., Vorberg, D., Meyer, A. S., Pechmann, T., & Havinga, J. (1991b).
Normal and deviant lexical processing: Reply to Dell and O'Seaghdha (1991). Psychological
Review, 98, 615-618.
Levelt, W. J. M., & Wheeidon, L. (in press). Do speakers have access to a mental syllabary?
Cognition.
Lindsley, J. R. (1975). Producing simple utterances: How far ahead do we plan? Cognitive Psychol-
ogy, 7, 1-19.
Luce, R. D. (1959). Individual choice behavior. New York: Wiley.
Lupker, S. J. (1979). The semantic nature of response competition in the picture-word interference
task. Memory & Cognition, 7, 485-495.
MacKay, D. G. (1972). The structure of words and syllables: Evidence from errors in speech.
Cognitive Psychology, 3, 210-227.
CHAPTER 29 GRAMMATICAL ENCODING 983

MacKay, D. G. (1987). The organization of perception and action: A theory for language and
other cognitive skills. New York: Springer-Verlag.
Maclay, H., & Osgood, C. E. (1959). Hesitation phenomena in spontaneous English speech. Word,
15, 19-44.
MacWhinney, B., & Bates, E. (1978). Sentential devices for conveying givenness and newness:
A cross-cultural developmental study. Journal of Verbal Learning and Verbal Behavior, 17,
539-558.
Marantz, A. (1984). On the nature of grammatical relations. Cambridge, MA: MIT Press.
Martin, E., & Roberts, K. H. (1966). Grammatical factors in sentence retention. Journal of Verbal
Learning and Verbal Behavior, 5, 211-218.
Martin, E., & Roberts, K. H. (1967). Sentence length and sentence retention in the free learning
situation. Psychonomic Science, 8, 535.
Martin, E., Roberts, K. H., & Collins, A. M. (1968). Short-term memory for sentences. Journal
of Verbal Learning and Verbal Behavior, 7, 560-566.
Martin, N., Weisberg, R. W., & Saffran, E. M. (1989). Variables influencing the occurrence of
naming errors: Implications for models of lexical retrieval. Journal of Memory and Language,
28, 462-485.
McDonald, J. L., Bock, J. K., & Kelly, M. H. (1993). Word and world order: Semantic, phonologi-
cal, and metrical determinants of serial position. Cognitive Psychology, 25, 188-230.
McNamara, T. P., & Miller, D. L. (1989). Attributes of theories of meaning. Psychological Bulletin,
106, 355-376.
McNeill, D. (1985). So you think gestures are nonverbal? Psychological Review, 92, 350-371.
McNeill, D. (1987). Psycholinguistics: A new approach. New York: Harper & Row.
McNeill, D. (1989). A straight path—to where? Reply to Butterworth and Hadar. Psychological
Review, 96, 175-179.
Meringer, R., & Meyer, K. (1978). Versprechen und Verlesen. Amsterdam: Benjamins. (Original
work published 1895)
Meyer, A. S. (1990). The time course of phonological encoding in language production: The en-
coding of successive syllables of a word. Journal of Memory and Language, 29, 524-
545.
Meyer, A. S. (1991). The time course of phonological encoding in language production: Phonological
encoding inside a syllable. Journal of Memory and Language, 30, 69-89.
Monsell, S. (1990). Frequency effects in lexical tasks: Reply to Balota and Chumbley. Journal of
Experimental Psychology: General, 119, 335-339.
Monsell, S., Doyle, M. C, & Haggard, P. N. (1989). Effects of frequency on visual word recognition
tasks: Where are they? Journal of Experimental Psychology: General, 118, 43-71.
Needham, W. P. (1990). Semantic structure, information structure, and intonation in discourse
production. Journal of Memory and Language, 29, 455-468.
Oldfield, R. C. (1963). Individual vocabulary and semantic currency: A preliminary study. British
Journal of Social and Clinical Psychology, 2, 122-130.
Oldfield, R. C, & Wingfield, A. (1965). Response latencies in naming objects. Quarterly Journal
of Experimental Psychology, 17, 273-281.
Osgood, C. E. (1971). Where do sentences come from? In D. D. Steinberg & L. A. Jakobovits
(Eds.), Semantics: An interdisciplinary reader in philosophy, linguistics and psychology (pp.
497-529). Cambridge: Cambridge University Press.
Osgood, C. E. (1980). Lectures on language performance. New York: Springer-Verlag.
Perfetti, C. A. (1969a). Sentence retention and the depth hypothesis. Journal of Verbal Learning
and Verbal Behavior, 8, 101-104.
Perfetti, C. A. (1969b). Lexical density and phrase structure depth as variables in sentence retention.
Journal of Verbal Learning and Verbal Behavior, 8, 719-724.
Perfetti, C. A., & Goldman, S. R. (1975). Discourse functions of thematization and topicalization.
Journal of Psycholinguistic Research, 4, 257-271.
Perfetti, C. A., & Goodman, D. (1971). Memory for sentences and noun phrases of extreme depth.
Quarterly Journal of Experimental Psychology, 23, 22-23.
Perlmutter, D. M. (1982). Syntactic representation, syntactic levels, and the notion of subject. In
P. Jacobson & G. K. Pullum (Eds.), The nature of syntactic representation (pp. 283-340).
Dordrecht: Reidel.
Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard
University Press.
Pinker, S. (1991). Rules of language. Science, 253, 530-535.
984 KATHRYN BOCK AND WILLEM LEVELT

Pollard, C, & Sag, I. A. (1988). An information-based theory of agreement. In D. Brentari, G.

Larson, & L. MacLeod (Eds.), Papers from the 24th annual regional meeting of the Chicago
Linguistic Society: Part Two. Parasession on agreement in grammatical theory (pp. 236-257).
Chicago: Chicago Linguistic Society.
Prentice, J. L. (1967). Effects of cuing actor vs. cuing object on word order in sentence production.
Psychonomic Science, 8, 163-164.
Robertson, S. S., & Suci, G. J. (1980). Event perception by children in the early stages of language
production. Child Development, 51, 89-96.
Roelofs, A. (1992). A spreading activation theory of lemma retrieval in speaking. Cognition, 42,
107-142.
Rumelhart, D. E., & McClelland, J. L. (1987). Learning the past tenses of English verbs: Implicit
rules or parallel distributed processing? In B. MacWhinney (Ed.), Mechanisms of language
acquisition (pp. 195-248). Hillsdale, NJ: Erlbaum.
Saffran, E. M., Schwartz, M. F., & Marin, O. S. M. (1980). Evidence from aphasia: Isolating the
components of a production model. In B. Butterworth (Ed.), Language production (Vol. 1,
pp. 221-241). London: Academic Press.
Schenkein, J. (1980). A taxonomy for repeating action sequences in natural conversation. In B.
Butterworth (Ed.), Language production (Vol. 1, pp. 21-47). London: Academic Press.
Schriefers, H. (1990). Lexical and conceptual factors in the naming of relations. Cognitive Psychol-
ogy, 22, 111-142.
Schriefers, H., Meyer, A. S., & Levelt, W. J. M. (1990). Exploring the time course of lexical
access in language production: Picture-word interference studies. Journal of Memory and
Language, 29, 86-102.
Seidenberg, M. S., & McClelland, J. L. (1989). A distributed, developmental model of word
recognition and naming. Psychological Review, 96, 523-568.
Selkirk, E. O. (1984). Phonology and syntax. Cambridge, MA: MIT Press.
Sereno, J. A., & Jongman, A. (1990). Phonological and form class relations in the lexicon. Journal
of Psycholinguistic Research, 19, 387-404.
Shallice, T., & Butterworth, B. (1977). Short-term memory impairment and spontaneous speech.
Neuropsychologia, 15, 729-735.
Smith, C. (1971). Sentences in discourse. Journal of Linguistics, 7, 213-235.
Sridhar, S. N. (1989). Cognitive structures in language production: A crosslinguistic study. In B.
MacWhinney & E. Bates (Eds.), The cross-linguistic study of sentence processing, (pp.
209-224). Cambridge: Cambridge University Press.
Stemberger, J. P. (1982). Syntactic errors in speech. Journal of Psycholinguistic Research, 11,
313-345.
Stemberger, J. P. (1983). Speech errors and theoretical phonology: A review. Bloomington: Indiana
University Linguistics Club.
Stemberger, J. P. (1985). An interactive activation model of language production. In A. Ellis (Ed.),
Progress in the psychology of language (pp. 143-186). London: Erlbaum.
Stemberger, J. P., & MacWhinney, B. (1986). Frequency and the lexical storage of regularly
inflected forms. Memory & Cognition, 14, 17-26.
Streim, N., & Chapman, R. S. (1987). The effects of discourse support on the organization and
production of children's utterances. Applied Psycholinguistics, 8, 55-66.
Tannen, D. (1987). Repetition in conversation: Toward a poetics of talk. Language, 63, 574-605.
Tannenbaum, P. H., & Williams, F. (1968). Generation of active and passive sentences as a function
of subject or object focus. Journal of Verbal Learning and Verbal Behavior, 7, 246-250.
Tomlin, R. S. (1986). Basic word order: Functional principles. London: Croom Helm.
Turner, E. A., & Rommetveit, R. (1968). Focus of attention in recall of active and passive sentences.
Journal of Verbal Learning and Verbal Behavior, 7, 543-548.
Weiner, E. J., & Labov, W. (1983). Constraints on the agentless passive. Journal of Linguistics,
19, 29-58.
Wells, R. (1951). Predicting slips of the tongue. Yale Scientific Magazine, 26, 9-30.
Williams, E. (1981). Argument structure and morphology. Linguistic Review, 1, 81-114.
Wingfield, A. (1968). Effects of frequency on identification and naming of objects. American Journal
of Psychology, 81, 226-234.
Woodworth, R. S. (1938). Experimental psychology. New York: Holt.
Yngve, V. H. (1960). A model and an hypothesis for language structure. Proceedings of the
American Philosophical Society, 104, 444-466.