to appear in Sign-based Construction Grammar, Boas, H.C. and Sag, I.A. (eds.

The FrameNet Constructicon

Charles J. Fillmore, Russell R. Lee-Goldman, and Russell Rhodes

1 Introduction
The Berkeley FrameNet Project1 has been engaged since 1997 in discovering and describing the semantic
and distributional properties of words in the general vocabulary of English.2 Notions from FRAME SEMAN -
TICS (see Fillmore and Baker 2009 and references therein) provide the basis of the semantic description of
the lexical units in the database, and sentences extracted from the FrameNet (FN) text corpora3 serve as
material for analysis and annotation. The goal is to describe the combinatorial properties of each word, both
semantically and syntactically, as these properties are revealed in the corpora.
The present chapter reviews some of the principles and displays some of the results of FrameNet’s
lexical research but focuses on an appended project for recognizing and cataloguing GRAMMATICAL CON -
STRUCTIONS in English. The registry of English constructions to be created by this secondary project—the
Constructicon—will describe the grammatical characteristics and semantic import of each construction, and
will link to each a collection of sample sentences that have been annotated to exhibit these characteristics,
using tools developed for the earlier lexical work.
While building a Constructicon has different goals from those of designing a construction-based gram-
mar of the language, the intention is that each construction will be represented in a way compatible with the
development of full grammar of the language of the sort presented elsewhere in this volume (see especially
Sag this volume). In some cases we offer precise proposals for the treatment of a construction as it would
appear in the grammar; in other cases the descriptions we present should be seen at least as organized ob-
servations about individual constructions, observations that need to be accounted for in a future complete
grammar. In all cases we expect that the constructicon will contain useful information for advanced language
pedagogy and that it will suggest new levels of expectation for parsing and other NLP activities.
The main body of the finished Constructicon will display the properties of constructional phenomena
in an abbreviated format, alongside of a representative sample of English sentences annotated to display
the properties claimed for each construction. The annotation procedure follows a method of identifying
and labeling phrases originally developed for FN’s lexicographic activities but adapted to indicate (1) the
stretches of language that count as instances of given constructions (e.g. the phrase bracketed in { } in They
1 FrameNet has been supported by a series of grants from the National Science Foundation: first under grant IRI #9618838,

March 1997–February 2000, ‘Tools for lexicon-building’; then under grant ITR/HCI #0086132, September 2000–August 2003, en-
titled ‘FrameNet++: An On-Line Lexical Semantic Resource and its Application to Speech and Language Technology.’ Since 2004
it has continued under a number of subcontracts, through NSF and other agencies of the U.S. government, among them a subcontract
from grant IIS-0325646 (Dan Jurafsky, PI) entitled Domain-Independent Semantic Interpretation for providing full-text FrameNet-style
annotation of a number of texts that had also been annotated in the PropBank project.
2 The authors are deeply indebted to Hans Boas and Ivan Sag, and three anonymous reviewers, for corrections and suggestions on

an earlier version of this chapter.

3 The main corpus used in the lexicographic work has been the British National Corpus. Information at http://www.natcorp.ox.ac.uk/.
hired me at {four dollars an hour}), (2) the segments of those stretches that are dictated by the requirements
of the construction (e.g. the phrases bracketed in [ ] in They hired me at {[four dollars] [an hour]}), and (3)
elements of the context whose properties require, or are selected by, the given construction instance (e.g. the
preposition, outside of the { } bracketing, in They hired me [at] {[four dollars] [an hour]}).
This last annotational practice – that of recognizing and categorizing linguistically relevant parts of
the governing context of an item under analysis – needs to be a feature of both lexical and constructional
annotations. Until now FN has been careful to identify the requirements lexical units might impose on
their dependents – constituting their VALENCE – but except for an attention to support verbs (Fillmore et al.
2003), by which, for example, advice, comfort, and disagreement appear in collocations like give advice,
take comfort, and have a disagreement, there has been no consistent way to indicate the kinds of features
seen in the preference for a ‘negative-ability’ context4 for verbs like stand in the meaning ‘tolerate’, as in
I can’t stand listening to this stuff, a context needed not only for simple lexical items, but also for some
complex or idiomatic expressions like I can’t quite put my finger on it.

1.1 FN’s Move from Words to Sentences

In their lexicographic work, FN researchers have concentrated on the semantic and syntactic properties of
lexical items, covering in particular their meanings, their syntactic and semantic combinatorial affordances,
and their relation to the semantic frames which inform their meanings.
Each lexical description in FN links a LEXICAL UNIT (LU) to the particular SEMANTIC FRAME that it
EVOKES in the mind of the user, roughly in the way that buy brings to mind a commerce scenario, eyebrow
evokes a face frame, hot finds its place in a temperature scale, and so on: the word FRAME covers
any kind of experience-based conceptual structure or relation that constitutes the experiential background
for understanding particular lexical meanings.
An LU is a pairing of a word and its meaning, usually a single word as it functions in a given frame. The
analyses are based on annotators’ examination of corpus instances of each word; sentences are selected to
exemplify the LU’s most typical COMBINATORY AFFORDANCES, as they have been observed in the corpus;
the example sentences are annotated – manually but with computer assistance – in a way that identifies the
frame being illustrated and displays the syntactic manner in which information about the frame is expressed
in the sentence. The database created in this way includes information about individual frames (a single
frame can provide the background for a large group of meaning-related LUs); each frame is associated with
descriptions of the roles of the entities that participate in them (individuals, props, setting features), known
as FRAME ELEMENTS (FEs); and each frame description includes a list of lexical units that belong to the
frame. A Commerce scenario includes alongside of buy a number of other words including purchase (noun
and verb), sell, pay, and a dozen or so others; the Face frame includes features and parts of faces beyond
eyebrow; the Temperature scale includes the full range of vocabulary relating to temperature – hot, cold,
warm, cool, tepid, freezing, and many others.
The FN annotations that have been created purely for lexicographic purposes illustrate one lexical unit
each, and contain no information about parts of the sentence outside the scope of the selected LU. The
purpose of such annotations is to show clear examples of the basic combinatorial possibilities for a single
targeted LU, and the selection of material to be annotated is to cover the main varieties of combinatorial pat-
terns observed in the corpus. The annotation is provided in a multi-layered, stand-off system with different
kinds of information, linked to particular words or phrases associated with the TARGET LU, shown on sepa-
4 Can+not is the most direct way to express the intended context here, but the ‘not’ part can be realized as any of the various ways of

creating a negative polarity context, the ‘can’ part can be expressed with paraphrases like be able (Are you able to stand living there?),
and a context with too can serve both meanings. That is, I’m too short to reach it = I’m so short that I can+not reach it.

2
rate dedicated layers, somewhat along the lines of a musical staff in which a chord extends across individual
phrases, and the individual notes in the chord represent (1) the SEMANTIC ROLE or frame element (FE) as-
signed to the phrase, (2) the GRAMMATICAL FUNCTION of the phrase, and (3) the lexico-syntactic form or
PHRASE TYPE . Reports based on such annotations, the FN LEXICAL ENTRIES , summarize the various ways
in which, say, information about BUYER, S ELLER, P RICE, and G OODS, are realized in the grammatical
neighborhood of words like buy, sell, cost, pay, and charge, in respect to such functional notions as subject,
object, oblique, etc., and in terms of the mandatory or optional choice of prepositions like to, from, and for,
and the like.
The kinds of grammatical structures needed to exemplify these basic structural properties are more or
less limited to relations of predication, modification, and complementation. Given the original purely lexico-
graphic purpose of the project, this limitation was not seen as a problem: FN’s goals were to characterize the
main distributional properties of verbs, nouns, adjectives, and contentful prepositions, following valence-
theoretic traditions,5 and for the most part we were able to achieve that. When the same words occur in
syntactically more complex contexts, their ability to do so usually depends on just those basic properties.
But there remain many sentences whose semantic and syntactic organization cannot be fully explained in
terms of the kinds of structures recognized in FN’s annotation database, or simple conjoinings or embeddings
of these, and that is where the new research on grammatical constructions comes in.
A simple example of a sentence that demands more than simple lexical analysis is (1):

(1) The skeptical are reluctant to believe that the unbelievable has happened.
This sentence contains two headless NPs, each containing only the definite determiner and an adjective.
These are instances of distinct constructions, recognizable by the fact that the first NP the skeptical shows
plural agreement (are reluctant) and stands for a generic class of human beings, while the second, the
unbelievable, shows singular agreement (has happened) and refers to an abstractly characterized situation
(See (73)). Neither phrase can be interpreted from its lexical material alone, and neither phrase can be
regarded as a conventional multiword expression.
Although a great many of the entries in the current FN database are monolexemic verbs, nouns, or
adjectives, in the course of this work the notion of LEXICAL UNIT as the target of analysis necessarily
expanded to include linguistic objects that contained more than one word – such as phrasal verbs, like take
off in the aeronautical sense, and preposition+noun patterns like under arrest, which semantically function as
predicate adjectives. We also recognize the existence of verb+noun patterns, like give advice, which belongs
more clearly to the concept of ‘advising’ than to ‘giving’. In these last two examples it is the noun (advice or
arrest) that evokes the frame, and the accompanying verb or preposition serves a secondary function: these
are the so-called SUPPORT WORDS.6
In addition the project has had to recognize IDIOMS and NOUN COMPOUNDS.7 While various makeshift
devices were developed early in the project’s history for including such structures within the limitations of
a lexicography project, it soon became clear that a very large number of linguistic objects, including noun
compounds, that function as units while at the same time having a describable internal structure could not
be formally recognized within the technical limitations of the project as it existed.
Since FN is capable of describing the semantic and combinatory properties of individual words one at
a time, the question came up of the possibility of analyzing entire sentences or texts by doing FN-style
5 Especially in the work that followed Tesnière 1959, such as Allerton 1982, Emons 1974, 1978, Helbig and Schenkel 1973, Herbst

1983, 1987, Herbst et al. 2004; cf. Fillmore 2007, Fillmore 2009.
6 Neither of these collocations is explicitly listed in the FN lexicon, but they are derivable from our annotation of give and under as

support words for advice and arrest respectively.

7 Lexical units that select PP complements with particular prepositions are simply treated as such, and not as phrasal in nature:

depend (on), defer (to); fond (of), complicit (in); appendage (to), member (of), etc.

3
annotations for each (frame-bearing8) word in it. Any attempt to do that, however, requires being prepared
for all contexts of words, not just exemplary ones, and recognizing a vast number of grammatical structures
that go beyond those needed for elucidating the basic properties of ordinary verbs, nouns, and adjectives.
It quickly became clear that a FN designed for valence-tracking lacked the requisite technology for many
aspects of full text analysis.
With the help of a Small Grant for Exploratory Research from the National Science Foundation9, FN
programmers introduced some changes to the structure of the database and created a number of adaptations
to the annotation software that made it possible to locate and annotate sentences in the FN corpora in terms
of the grammatical rules that licensed the structures found in them. These modifications to the FN toolbox
have enabled the recognition and description of many kinds of listable linguistic structures beyond the ones
that were earlier included in the FN lexicon.
The next two sections, which draw from discussions in Fillmore 2006, illustrate basic aspects (and limi-
tations) of a purely lexicographic or word-based analysis of a sentence, first with an invented simple sentence
in which all or most of the noteworthy characteristics are located in words and their syntactic/semantic va-
lences, and second with an attested passage for which a lexicographic analysis falls well short of a full
understanding of the sentence’s meaning.

1.2 Case 1: The Apparent Sufficiency of Lexical Analysis

The major unit of description for a frame-bearing lexical unit is its valence,10 beginning with a list of the
semantic roles (if any) of each of its syntactic dependents, identifying these as corresponding elements
of the frame the word evokes. Since the phrases that constitute the syntactic dependents of a predicator
can themselves include frame-bearing predicators, it seemed that the tools available in FN should make it
possible to characterize the basic semantic-syntactic structures of whole sentences by doing complete FN
analyses of each word in the sentence – technically by adding separate annotation layer sets for each LU.
The possibilities of clause embedding allowed for information from one frame to serve as an element of a
higher frame, and the possibilities of ARGUMENT SHARING (through ‘control’ relations) allowed detailed
cross-clause articulation of the syntactico-semantic structures in a sentence.11
To see how this works out, consider sentence (2):

(2) The Secretary ordered the Committee to consider selling its holdings to the members.

• The highest predicate in this sentence is the Communication12 verb order. By appropriately la-
beling the phrases that are in the grammatical construction with this word we can show who gave
the order (realized as the subject – the Secretary) and who received the order (the direct object – the
8 Not, strictly speaking, each word: many lexical items do not require the kind of treatment FN can provide. For example, in non-

specialist texts there is no need to provide for names of insect species, carbon compounds, Hindu gods, numerals, etc., and grammatical
markers and other function words appear only in the description of their lexical or grammatical hosts.
9 NSF #0739426 SGER: ‘Beyond the Core: A Pilot Project on Cataloging Grammatical Constructions and Multiword Expressions

in English,’ 2007–2008. Contributors to the work have been Dr. Collin F. Baker, Chris Oei, Jisup Hong, Michael Ellsworth, and the
present authors.
10 While this usage conforms with the use of ‘valence’ (or ‘valency’) in the tradition of valency dictionaries, it is narrower than

the use introduced in SBCG work, where VAL stands for the list of arguments a given linguistic entity ‘needs’ at a particular level of
representation.
11 This articulation is showable in the annotations only when there is a direct or indirect structural relation between predicates and

their arguments: relations of anaphora are not a part of FN annotation.

12 For this initial discussion the frames will be given very general names. FN makes more fine-grained distinctions than Communi-

cation, Cognition, and Commerce as presented in this paragraph. Throughout the chapter we use a fixed-width font for frame names.

4
 Committee), as well as the content of the order (expressed as the infinitive complement that begins
with to consider).
• By recognizing the constituents that are directly or indirectly grammatically linked to the Cognition
verb consider, the highest predicate of order’s complement, we show that it is the Committee that
is ordered to think about something (given the control requirements of order), namely selling the
Committee’s holdings.
• From the phrases that are in a grammatical relation with the Commerce verb sell we learn what is to
be sold (the direct object of selling, namely its holdings), and who the potential buyers are (represented
by the prepositional phrase at the end of the sentence, to the members). The sellers are identified as
the Committee due to the control properties specified by consider.
• The fact that in this context the words Secretary, Committee, and members all fail to be accompanied
by their potential of-complements indicates, by grammatical convention,13 that the organization to
which the secretary, the committee, and the members belong is known in the context of the ongoing
discourse. This sentence cannot be used in a conversation whose participants do not already have in
mind all of the information behind the speaker’s use of the definite article.

We thus approach an understanding of an entire sentence – that part of the ‘understanding’ of a sentence
that is explainable in terms of its linguistic structure alone – by integrating all of the parts into a single
representation. The basic semantic organization of this simple sentence has it that A gave an order to B;
B’s assignment is to consider performing the action C; C involves selling D to E; D is property owned by
B; A, B, and E belong to an organization that is not identified in the sentence but must be known to the
communicators.14

1.3 Case 2: The Inadequacy of Simple Lexical Analysis

Of course not every English sentence is so cooperative, with simple complement-taking verbs, simple noun
phrases, and orderly markers of grammatical structure. A sample of the kinds of linguistic observations that
require information that goes beyond single words and their contributions to sentence meaning can be found
in the first three sentences of a leader in the Economist magazine of May 10, 2007, on the resignation of
Tony Blair:

(3) For all the disappointments, posterity will look more kindly on Tony Blair than Britons do today. Few
Britons, it seems, will shed a tear when Tony Blair leaves the stage on June 17th after a decade as
prime minister, as he finally announced this week he would do. Opinion polls have long suggested
that he is unpopular.
Here are some observations about this sentence:
– for all the disappointments: for all X, unexplained through simple construals of for and all, is a
concessive structure with a meaning like ‘in spite of’. X is limited to either a definite NP with
or without of (*for all of several disasters in his first term, *for all events during his term) or
13 Actually, in the case of Secretary and Committee, the explanation involves the use of the definite article, since these words can

otherwise be used without the relatum being contextually given; for member, however, even if the word occurs without a definite article
– as in Are you a member? – the unmentioned club or organization is taken as pragmatically known.
14 Not explained by FN representations alone are the assumption that the its refers back to the mentioned committee and the parsi-

monious assumption that the unnamed background organization is the same for all three words Secretary, Committee, and members.

5
 a that-clause. In the latter case, the interpretation is ‘in spite of the fact that . . . ’. A related
construction involves a relative clause-like structure: for all he’s done, for all that has changed.
– look kindly on X: a verb-headed collocation meaning ‘judge positively’15
– posterity will look more kindly on Tony Blair than Britons do today: a comparison involving
different degrees of ‘looking kindly on Tony Blair’
– than Britons do today is a double-focus comparand: [Britons] [today], if each is accented, seems
to presuppose an analysis of posterity as something like [the world] [in the future]
– Britons do today: an elliptical clause in which the do is to be reconstructed as ‘look [degree]
kindly on Tony Blair’, where the nature of [degree] is unspecified here but it will be exceeded in
the case of ‘posterity’
– few Britons: unlike a few Britons, not a vague indication of cardinality, but a negator (= ‘not
many’), creating a negative polarity context in the predicate phrase
– it seems: epistemic parenthetical, applying to the surrounding clause, bearing no structural rela-
tion to anything else in the sentence
– shed a tear: a collocation (shed tears), but also a potential ‘minimizer’ expression fitting negative
polarity contexts, e.g. drink a drop, lift a finger, give a damn (Israel 2002, Horn 1989: 400)
– leave the stage: metaphor referring here to resigning from the PM-ship
– on June 17th: use of on with day references (compare with months, day parts, day-anchored
day parts, time points, etc.: in March, in the morning, on the morning of the next day, on the
weekend, at noon)
– June 17th: one of the several ways to associate a month-day number with a month name in
English (cf. June the 17th, the 17th of June)
– as he announced he would do: anaphoric or relativizing use of as (consider replacing as with
which).
– this week: an expression from a system with the three-way contrast this/ next/last for deictically-
anchored calendric expressions, occurring with week/month/year, etc., as well as weekday names
and seasons, but not day
– have long suggested: this use of long seems to be limited to the position between the auxiliary
have and the verbal participle, and the verbal meaning seems to be limited in ways yet to be
explored: compare I have long known that . . . with the unacceptable simple past tense *I
long knew that . . . and with inappropriate verb meaning *I have long lived in California; in
many other uses long as a time-duration adverb is limited to negative polarity contexts: I won’t
be long, etc.
This short passage illustrates the density of complex structures that can be found in ordinary texts. There
are binary patterns with specific kinds of fillers in two adjacent positions: this week, last year, next month,
etc.; June 17th, March 4th, etc. There are lexical items with very specific functions and positional possi-
bilities: as, long, on, do. There are phrasings with conventional collocations or non-predictable meanings:
for all, shed a tear, leave the stage. And there are words that participate in larger constructions that they
syntactically mark: few, more.
The Constructicon that is being produced is designed to handle most if not all of the linguistic features
relevant for the constructions identified or alluded to in the Economist passage.
15 Other fixed expressions with adverbial kindly include the negative polarity item (NPI) take kindly to and the usually parenthetical

phrase to put it kindly.

6
2 Annotation
Since FN annotations were presented in tiers or layers, one set of layers for each LU and the phrases that
count as its frame elements, it seemed that an adaptation of the tool for creating such annotations should
be possible for showing the corresponding kinds of information for grammatical structures. For each con-
struction instantiated in a sentence, it should be possible to assign a separate set of layers corresponding
to the construction: one layer to identify the expression licensed by the construction, and another layer to
identify the constituent components of that expression. Together these layers identify the CONSTRUCT de-
scribed by the construction (on which see below). Other layers may identify construct-external elements,
e.g., ‘co-textual’ requirements on uses of the construction.

2.1 FrameNet Annotation

FN, as described above, uses the notion of semantic frame as a basic unit of description: frames are schematic
understandings of relations among frame elements in a state of affairs. Frame elements are the participants,
props, situation determinants, etc. that are understood as present in any instance of a given frame. Partic-
ular lexical units are described as belonging to specific frames (polysemous words can belong to different
frames), and sentences containing them are annotated in a way that identifies the frame, marks the phrases
that are in a grammatical relation to the relevant LU. Those phrases are labeled with respect to their gram-
matical relation to the LU (GF, or grammatical function), to their syntactic form or phrase type (PT), and
the frame element which they express (FE). Since the resulting valence descriptions that make up the core
of each LEXICAL ENTRY are derived from the annotation, the annotators seek to include a representative
sample of sentences for each LU in each frame, to show the variety of lexical and phrasal realizations of the
FEs.
The annotations are presented in sets of layers, each layer dedicated to one kind of information (FE, GF,
the target LU, and so on). The FE is assigned manually by the annotator, who is generally working one
frame at a time; the GF and PT labels are assigned automatically but manually checked by the annotator.
The annotator uses a number of principles for determining the boundaries of the constituents blocked off
for labeling, as well as for dealing with LUs or FE-realizing phrases that are discontinuous. Since the
annotations are to show the linking between syntactic and semantic-role objects, the labeled phrases include
prepositions, complementizers, and other markers of syntactic role. Further, FEs that are interpreted in
‘higher’ predicates rather than in the structure immediately surrounding the target LU are also labeled, and
information about FEs that are core parts of the frame but which are unexpressed in the current sentence are
described within a taxonomy of implicit arguments.
An example of such annotation is given as Figure 1, which shows the annotation tool set up for what is
called the Compliance frame for an analysis of sentence (4):

(4) We complied fully with the instructions and resubmitted the paper in January 1992.
The bottom panel contains the labels available to the annotator for selecting the FEs.16 In the case of
this frame, the core FEs include the N ORM and the mutually exclusive trio P ROTAGONIST, ACT, and
S TATE _ OF _ AFFAIRS. (FE names are indicated with small capitals.) The list of peripheral FEs includes
D EGREE, which is illustrated in this sentence. The N ORM is the regulation, principle, doctrine, or law that
16 Other views of the workspace for the same sentence provide, in the bottom panel, the annotators’ choices for grammatical function

(GF) and phrase type (PT). Automatic chunking software supplies the labels on these levels, and the annotator uses the relevant
annotation panels only to make corrections.

7
 Figure 1: Lexicographic annotation with comply as the target

governs some kind of action. The P ROTAGONIST is a person whose behavior is judged in respect to com-
pliance with the N ORM, i.e. by honoring it or violating it, as in the subject of (5a); the ACT stands for an
act which is evaluated as appropriate or not by reference to the N ORM, as in the subject of (5b); and the FE
S TATE _ OF _ AFFAIRS is illustrated by the subject of (5c):

(5) a. You broke the law.

b. The removal of the child was in breach of my right of custody.
c. The wiring in this room is in violation of the county building code.
The N ORM can appear, depending on the lexical requirements, as direct object (broke the law), in an of-
phrase (violation of the law), in a with-phrase, as in Figure 1 complied with the instructions, and in a number
of other ways.
Thus, the FN database includes:
1. A list of frames, with informal frame descriptions, definitions of named frame elements, a partition
among the frame elements between those that are essential to an understanding of the frame (the CORE
frame elements) and those that are more generally applicable across frames (the PERIPHERAL frame
elements),17 and lists of lexical units that have (so far) been assigned to the frame. As noted, the
words that belong in a single frame do not have to be of the same part of speech: comply, compliant,
and compliance are in the same frame. And since many frames cover an area across which contrasts
and oppositions are defined (hot and cold are in the same frame, as are comply and flout, ascend and
17 Roughly, core elements are essential to the concept underlying a frame, and peripheral elements more or less correspond to what

valency theory writers refer to as ‘circumstantial’ (Fillmore 2003, Lyons 1977: 496–7, Tesnière 1959: 102f, 125f). Since one of our
annotational goals was to have something to say about each constituent that occurred within the grammatical scope of a target LU, we
needed a way of describing clausal elements that provided information not directly related to the current frame, and we referred to these
as EXTRATHEMATIC. Such elements often introduce frames of their own, possibly frames that take the content of the main clause as a
semantic component: for example, to describe the last PP in I set fire to my brother’s house in retaliation as extrathematic is to say that
the concept of ‘revenge’ is not a necessary component of the Ignition frame.

8
 descend, praise and criticize, etc.), the members of a single frame are not limited to synonyms or
near-synonyms.
2. A body of annotated sentences, with each sentence annotated for the individual LU it has been chosen
to illustrate. The annotators’ assignment is to survey the uses of the LU in the hope of offering samples
of each of its significant syntactic combinatory possibilities (e.g. both active and passive for passiviz-
able transitive verbs, varieties of path-indicating expressions for motion verbs, examples of words in
all parts of speech both with and without omissible arguments, and so on), and to prefer examples
with lexical material that reflects frame-relevant conditions.18 The annotator selects and labels the
phrases that express given FEs and then checks the results of the parsing-based automatic assignment
of grammatical functions and phrase types, selecting from a collection of categories deemed to be
‘lexicographically relevant’ (Atkins et al. 2003).
3. Lexical entries containing a dictionary-style definition of each LU, tables showing the syntactic real-
ization possibilities for each FE, and tables showing the range of valence possibilities, each of these
linked to sentences from the annotation set. These tables are generated automatically from the anno-
tations.
4. A network of semantic relations among the frames and FEs, covering various kinds of frame-to-
frame relations, including Inheritance, Subframe (a part-whole relation), Using, Causative_of, In-
choative_of, Perspective_on, Precedes, etc. Related frames have related FEs, but the realization of
these FEs may differ greatly across frames (indeed, within-frame differences in FE realization is not
uncommon, cf. Giving, with give, bestow, and noun donor).
5. The following kinds of information are derivable from the database, but not presented in it directly:
weak paraphrasability relations (complied with X paraphrases did not violate X); interpretation of
unexpressed core arguments (according to whether the omission (1) is dictated by a particular con-
struction, (2) is permitted under existential interpretation, or (3) is permitted under anaphoric interpre-
tation). In the existing FN entries it is possible, for nouns that participate as semantic heads in support
structures, to know which support verbs or support prepositions have been observed to occur with
them, and, by examining the assignment of FEs in annotations, to know what LEXICAL FUNCTIONS
they serve (in the sense of Mel’čuk 1996).
6. In addition to the material just described, which is limited to FN’s lexicographic research, FN has
accumulated a body of texts that have been annotated for (more or less) all of the frame-bearing words
they contain. In the layered representation (Figure 1) each frame-bearing LU token is given its own
set of annotation layers.

2.2 Constructional Annotation

Constructions are the rules that license ‘new’ linguistic signs based on other linguistic signs. The struc-
tures licensed by one or more constructions are called CONSTRUCTS, following the terminology of Sign-
based Construction Grammar. A construction can be described formally, in Attribute-Value Matrix form,
or informally in prose, but annotation must be of constructs: each annotation captures the properties of
a particular construct with respect to a particular construction that licenses it. To give a brief example, a
18 For these reasons FN annotation data cannot be relied on to show frequency information. In many cases the most frequent uses of

a verb will have only pronominal arguments: having representative examples of I didn’t like it or I think so, among the annotations for
like and think would not be useful.

9
Month-plus-Date construction licenses expressions like June 14th, May 2nd, and April 1st.19 If we are
momentarily interested in the Month-plus-Date construction, we identify June and 14th as the two nec-
essary parts of the phrase June 14th, and indicate their relation to the meaning of the phrase; annotation of
this phrase for Month-plus-Date requires identifying the categories month-name and ordinal-number
but ignores any other constructions which may be involved in the licensing of the complete phrase, such
as whatever apparatus may be needed for the creation of English ordinal numbers. By the same token, a
single stretch may be licensed by multiple constructions (e.g. a phrase which is simultaneously a VP and an
imperative), and so is annotated separately with respect to each construction.
Given these assumptions, two segmentations are needed for constructional annotation, one inside the
other: first, we need to block off the stretch of a passage that is accounted for by the construction, and
second, we need to identify the linguistic entities within that spread which represent its constituent elements.
Together these layers indicate the CONSTRUCT described by the construction. The entire stretch corresponds
to the construct’s MOTHER sign. The components are the DAUGHTERS; paralleling FN terminology we also
refer to these as CONSTRUCT ELEMENT (CEs). Mnemonic labels will be associated with each of these
segmentations, but their interpretation will be provided by the associated construction description.
The identification and labeling of constructs and CEs are only partial grammatical descriptions. The
construction description – prose, just as frame descriptions are – fills in other information: in the case of the
construct July 14th, the construction description will point to the relevant frames regarding dates, months,
years, etc., and will assign to this construct the information that it designates a ‘day’-size unit (rather than
week, month, year, etc.). This information is relevant in several structures, accounting for the selection of
the preposition on (rather than in or at). The description also specifies that the outputs of this construction
are assigned to the same calendric frame as those produced by the more complex construction that yields the
14th of July.
Two final layers (or sets of layers) may be needed in individual cases, one for showing the valence
realizations of valence-bearing constructs, and one for relevant elements of the governing context. For
constructs that are frame-evoking elements in their own right, the annotators are encouraged to demonstrate
this by showing the realization of the valents of such expressions. Thus, the construction that gives us make
one’s way, illustrated by (6), licenses the combination of a certain class of verbs (on which see Goldberg
1995) with a possessed way-headed NP to create what functions as a multiword verb evoking the Motion
frame:

(6) a. [Theme They] hacked their way [Source out] [Goal into the open].
b. [Theme We] sang our way [Path across Europe].

c. Hopefully [Theme he]’ll make his way [Goal to our location].

By assigning the products of this construction to Motion, our constructional annotation allows for noting
the Motion FEs that are present in any sentence containing the construction, in a manner entirely consistent
with FN annotation of regular verbs of motion: with labels indicating the spans of T HEME, S OURCE , G OAL,
etc., along with GF and PT information.20
For constructs of particular types, a layer may be needed to indicate needed or preferred GOVERNING
CONTEXTS . One such context is provided by support verbs or prepositions. The Rate.speed construc-
tion, which licenses such expressions as thirty miles an hour and five km per day, additionally evokes a
19 Construction names are given in italicized fixed-width font.
20 Fortemporary technical reasons we called these constituents ‘construct elements’ rather than frame elements in our annotation;
but conceptually these are frame element labels.

10
Speed frame which has an E NTITY FE (i.e. something must potentially move to have a speed). The con-
struction allows for expression of this FE, among other ways, by means of a support verb do: he’s doing
over a hundred miles an hour. Annotation of this proceeds exactly as it does for lexical annotation (with
some kinks, described below): he is labeled E NTITY, do is labeled Support, and the stretch a hundred miles
an hour is labeled as the construct which brings along this FE and support verb.
The parallels described above between LUs and constructions mean that the software needed for the one
should be adaptable to the other. Identification and labeling of FEs and support words is essentially identical.
Constructional annotation differs from FN lexicographic annotation primarily in that the frame-bearing units
are themselves complex. Annotation of constructs and their associated CEs is done in two parts. In place
of a TARGET LU (i.e. a frame-evoking element), we have a potentially multiword construct. The construct’s
span of text is delimited by a ConstructPT (CstrPT) label, whose name indicates the PT of the construct
(technically this is the phrase type of the sign that is the mother of the construct). Then, the construct-internal
constituents are identified as instantiating construct elements. The Rate.speed construction licenses an
NP construct (at [two miles per hour], a speed of [two miles per hour]) with two construct-internal CEs,
T IME and S PEED. The Month-plus-Date construction briefly outlined above specifies a construct of
type noun (or N/nominal), and two CEs – M ONTH and DATE – with semantic and syntactic constraints on
each. CE names are mostly mnemonic and do not, for instance, imply a hierarchy of syntactic or semantic
types (though they may correspond to such; if so, this is described in prose in the definition of these CEs or
the construction to which they belong). Like other CEs (and FEs), these construct-internal CEs are assigned
PT labels: both are NP, in the case of Rate.speed. No GF is necessarily assigned however, as within
the construct there is no necessary grammatical relation between the CEs: is thirty miles a dependent of an
hour? Or is it the head? In such cases the traditional categories are not always appropriate, or at least it
does not seem crucial to the annotation of the construction that a GF be chosen. The notion of GF is still
sometimes needed, as in make one’s way. A crucial property of this construction is that one’s way takes
the place of any object the verb might have had (whether it is itself a direct object depends on how such a
relation interacts with other parts of the grammar, e.g. passivization or case assignment).
An example of a constructional annotation illustrating most of these properties is given as Figure 2.
Moving from bottom to top: the construct comprises the stretch squeezed her way, and is a finite verb, as
indicated by the label on the CstrPT layer.21 The phrase her way is identified as the CEE, the Construction
Evoking Element, which indicates any lexically-limited material (if any). The two constituents within the
construct are identified as a verb and an NP. They are assigned no GF, as discussed above. Skipping to
the top line, we see that squeezed is labeled T RANSITIVE _ MEANS _ VERB – ‘transitive’ to indicate that
the verb normally has a valent (the thing squeezed) which is unexpressible in the construction (compare
the CE I NTRANSITIVE _ MEANS _ VERB).22 The way-NP is identified as the WAY _ EXPRESSION. Outside
the construct are found two FEs: T HEME and PATH, which are assigned GFs (External and Dependent,
respectively) and PTs (NP and PP). Finally note the T HEME label under her. This is located on a secondary
layer of CE annotation, available for additional indications of FE (or CE) fillers which do not strictly adhere
to the phrase structure of the sentence. On the assumption of FE-uniqueness per LU, this is an awkward
way of indicating that the possessive pronoun in this construction is identified with (and has to agree with)
the external argument. The verb-way construction is one of many which have such a subject-posessor FE
identification requirement, including expressions of body movement (crane one’s neck), and many other
multiword verbal expressions, e.g. change one’s mind, eat one’s fill.
21 Purely technical considerations led us to label this a finite VP rather than simply a verb.
22 The verb slot in this
construction is also a site for certain kinds of linguistic creativity: frequently words that are not verbs elsewhere
behave like verbs in one’s way constructs: examples in the database include nose, elbow, and acrobat.

11
 Figure 2: Annotation of an instance of the verb-way construction

3 Varieties of Constructions
We can classify constructions according to the kinds of constructs they create. The classes listed here are not
intended to be complete, nor are they mutually exclusive. In this section we survey a number of construc-
tional properties that need to be accounted for in a grammar and need to be explained in a constructicon.
Following this brief survey, we exemplify and analyze a number of constructions examined by the FN team,
bearing in mind this classification.

Frame-bearing constructions The expressions licensed by some constructions are frame-bearing entities,
analogous to words that evoke frames. The resulting sign specifies expected (obligatory or optional) frame
elements.
Several constructions have this property, including verb-way, which evokes the Motion frame and ap-
pears with FEs relevant for motion. The Rate construction evokes the Ratio frame with FEs N UMERATOR
and D ENOMINATOR. A subtler example is the construction that licenses brothers in He was brothers with the
class president. As we discuss in the following section, the construction evokes a frame of Reciprocity,
such that the relationship of brotherhood is understood as holding symmetrically even though that is not
lexically entailed by brother.

Valence-augmenting constructions Some constructions may be described as augmenting the valence of

an independently-licensed sign. One such construction is the comparative -er, which affixes to certain adjec-
tives or adverbs and adds to its base’s valence a phrase that indicates the standard against which some item
is compared: taller [than that one]. This phrase is optionally omitted with definite interpretation. Where the
modified adjective or adverb has complements of its own, the two sets are pooled and the valents may appear
in any order. Other such degree-modification constructions (so. . . that, for instance) require the added valent
to appear after all other arguments (see also Kay and Sag this volume, which offers a less heavy-handed way
of treating the ‘added’ FEs).

12
 (7) a. Her job is tougher than yours.
b. Her job is tougher.
c. The second layer of rock is richer in valuable minerals than the surrounding layers.
d. The second layer of rock is richer than surrounding layers in valuable minerals.

Constructions without Meanings While construction grammars originated in the recognition of con-
ventional pairings between specific formal patterns and the meanings they contribute to the expressions
that contain them – against a contrasting view that syntactic principles should be stated independently of
questions of meaning and use – there remains the question of whether all constructions should be seen
as meaning-bearing. The case for the necessity of semantics in every construction is made in Goldberg
2006: 166–182, in connection with SAI (Subject Auxiliary Inversion, here referred to as Aux-initial),
in response to several writers who had characterized it as a clear case of semantic-free syntax (Green 1985,
Newmeyer 2000: 46–9, Fillmore 1999). Goldberg examines a wide range of types of finite clauses contain-
ing the Aux-initial form, examines the separate functions of the types, and formulates a generalization
intended to cover all of them. While a ‘metagrammar’ of English might find some motivating concept that
underlies uses of this pattern, the actual work of building the FrameNet Constructicon is proceeding under
an assumption of the legitimacy of semantically null constructions.
There are three situations for which it is unnecessary to associate meanings with syntactic structures.
First, there are syntactic patterns with precise formal features whose interpretation depends on combining
information from their constituents in a completely regular way. Since construction grammars cover all of
the grammar, they include such basic structures as complementation (Head-Complement), modification
(Modifier-Head), and predication (Subject-Predicate). These constructions provide for realiza-
tion of the arguments of predicates and modifiers by specifying among other things the relative order of
head and dependent, and the syntactic category of the phrases produced (e.g. VP from V and NP). Aside
from phrase-building and argument linking, these constructions as such do not contribute meanings of their
own. Their function is to provide the contexts in which the syntactic and semantic expectations of their
constituents are satisfied. More specific constructions are necessary to accommodate the variety of semantic
and pragmatic meanings associated with instances of the abstract construction. For instance, specific modi-
fiers can assign different kinds of interpretations to their heads (compare the combinatory consequences of
confronting a noun with such modifiers as green, alleged, mere, former, so-called, economic, etc.), though
the means of syntactic combination and argument linking in general is constant. Similarly with the subject-
predicate relation. The simple idea that the subject establishes something to talk about and the predicate
establishes some property or relation of that thing (on the image that the left hand holds a thing and the right
hand decorates it) does not fit expletive-subject sentences like It’s raining or There’s a problem, or sentences
in which the subject’s relation to the verb is what provides the new information, as in A magnificent oak tree
developed out of that acorn that my father planted when he was a boy.
Second, there are constructions that determine syntactic patterns to which separate interpretations can be
given under differing variations. These include the abstract Aux-Initial construction and the Filler-Gap
construction. Aux-Initial creates signs that can serve as components of a large variety of specific con-
structions, treated in greater detail below, including the ability to produce expressions like those in (8), with
meanings like question, wish, condition, exclamation, etc. (Fillmore 1999):

(8) a. Has Dr. Padgett gone?

b. May my enemies live here in summer!

13
 c. Should that unthinkable thing happen, . . .
d. Was I mad!
Filler-Gap (Sag 2010) supports constructions that produce expressions like those in (9), with mean-
ings like interrogation, contrastive focus, conditionality, and so on (Kay and Fillmore (1999) called this
construction Left isolation):

(9) a. What are you writing?

b. That much, I can agree to.
Each of these more elaborated constructions specifies its own semantic and pragmatic features, but their
common syntax is succinctly captured by a general (abstract) construction. If there is a common meaning
discoverable in the many specific constructions, as suggested by Goldberg (2006) for Aux-Initial, this
may be associated with the general construction. However, because the specific subtypes of Aux-Initial
will need to fully specify their meaning in any case, the value of stating a potentially vague generalization at a
higher level is of dubious value unless independently motivated (psycholinguistically or sociolinguistically).
In section 4 we discuss three constructions which license an NP composed of the and an adjective phrase,
as in the very rich or the unexplainable. As with Aux-initial, it is in principle possible to posit an
abstract construction upon which the other three are elaborations.
Third, there are constructions that allow the omission of position-specific constituents that would other-
wise be repetitions. The interpretation of the constructs produced by such constructions involves cognitively
restoring and restructuring the missing elements into the remaining structure. These include Gapping,
Shared-Completion (a.k.a. Right-node Raising), and Stripping. Examples of such patterns can be
seen in (10):

(10) a. Some would take them to the top half of the dale, and others the bottom half.
b. He is clearly familiar with and fond of that cat.
c. I come to the office quite a bit, but not on Friday.
They have interpretation principles involving the reconstruction process, to be sure, but if they have ‘mean-
ings’ as such, they are at the level of information structure and focus.
There are, to be sure, patterns that are shared by numerous constructions to which we do indeed wish to
assign a general meaning. This is true, for example, of the Rate construction. That is, we feel it worthwhile
to assign a general meaning to the construction as a whole, captured in the concept of ‘rate’ or ‘ratio’, where
the first NP provides the numerator and the second the denominator. The constructions that are elaborations
of this construction can have special external properties and identify constraints on these two components:
measures of distance vs. time give speed; instances vs. time give frequency; dollars vs. hours give wages;
miles vs. time give mileage, and so on. In the case of the constructions we regard as non-meaning-bearing,
it is not possible to find a clear meaning structure in the general form of the construction that is exploited in
precise ways by the more specific constructions.

Contextually bound constructs Some constructions create expressions that have contextual requirements
that are not determined by their meaning. Certain measurement expressions using adjectives (tall, long,
high, wide, deep, old, and a few others) have variants restricted to use as predicates (six years old, 8 inches
thick) or as prenominal modifiers (six-year-old, 8-inch-thick). Special to the attribute version is the fact that

14
the measurement unit noun (here, year and inch) is singular even if the associated number is greater than
one: a six-year-old child, an 8-inch-thick wall, but not *a six-years-old child or *an 8-inches-thick wall. In
the predicate version the number value on the noun matches the number: The child is six years old, or the
wall is 8 inches thick.23 The same distinction characterizes classes of adjectives, discussed, e.g. in Bolinger
1967, such as attributive-only main or predicate-only asleep24 .
Less obvious examples of bound constructs are those licensed by the Uniqueness construction, e.g.
my first Russian, his favorite color, your first cigarette. The nominals created by such modifiers denote
relations, completed by some indication of a ‘possessor’. Without the modifier the possession relation
cannot be readily interpreted, or has a different meaning: my Russian, his color, your cigarette. Further
context dependencies can be seen when pronominally possessed expressions occur as arguments of a verb:
I’ve just met my first Russian; when did you smoke your first cigarette? where the interpretation is that I met
a Russian for the first time, and you smoked a cigarette for the first time.

Exocentric and headless Constructions A construction may license structures with syntactic properties
unpredictable from the properties of their parts. Such constructs may be described as ‘exocentric’ (the
category of the whole is not the category of the apparent head), or as not having a well-defined syntactic
head. Among such constructions are those licensing rate expressions (NPs from a juxtaposition of NPs),
expressions of generic humans with specific properties (NPs from the and an AP), the so-called ‘Big Mess’
construction (predeterminer modification of a noun by a degree-modified AP, dubbed ‘exceptional degree
markingâĂŹ by Zwicky (1995)), and time adverbials with last, this, or next and a calendric unit or subunit
(CU or CSU: next month, this Wednesday). These are briefly illustrated below, and some are given fuller
treatments in the following section (see Fillmore 2002 for details on calendric expressions).

(11) a. [NP/adverbial [NP thirty miles] [NP an hour]]

b. [NP the [AP very rich]]
c. [NP [AP how big] (of) [NP a box]]
d. [NP/adverbial [last/this/next] [C(S)U Tuesday]]

Pumping Constructions A number of constructions describe constructs with a single daughter sign, pos-
sibly a single word, and a mother sign with a form related or identical to the daughter, and a meaning which
incorporates the meaning of the daughter sign. The valence-modifying constructions discussed at length in
the literature on Argument Structure Constructions (Goldberg 1995, Boas 2003, Kay 2005) fall into this cat-
egory, as do general meaning-changing ‘zero derivation’ constructions that make use of the count-noncount
distinction in nouns (sometimes giving specific meanings and contexts to the result; see the discussion of the
meat-grinding construction in the following section). Several other phenomena, including those sometimes
23 Exceptions are attested in both directions: predicative she’s six foot tall, attributive a three-inches-thick wall, and even, with number
agreement reversed, our firm is today one years old. We describe the expressions that match the number-specifications stated here and
willingly leave the non-matching expressions unexplained.
24 The distinction is different from that between descriptive and relational adjectives; descriptive adjectives can occur in both posi-

tions, whereas relational adjectives tend to be attributive-only. The latter function in a way similar to that of the modifying noun in a
nominal compound, and are also known as pertainyms – having definitions of the type ‘of or pertaining to X’. They are exemplified
in criminal lawyer, nervous breakdown, favorite flower, economic policy, etc. Sometimes the same word may also occur as either a
descriptive or relational adjective (criminal lawyer is ambiguous), but with the relational meaning it is always prenominal. See Cop-
pock 2008: Chapter 5 for a recent discussion of the semantic and morphosyntactic features that distinguish these and other categories
of adjectives

15
given the name COERCION may also be understood as unary, pumping constructs (Michaelis 2005, forth-
coming, Pustejovsky 1995). Other constructions which could be handled as unary constructs (but which are
not, or not often, discussed in the coercion literature), are the English imperative (which pumps a VP to an
‘S’ (Ginzburg and Sag 2000: Chapter 2)), and ellipsis constructions such as sluicing and stripping, which
allow sub-clausal phrases to stand as clauses with clause-like interpretations (Ginzburg and Sag 2000: ch. 8).
In our SBCG analysis, constructions of this type are handled no differently than other constructions, except
that there is only one daughter sign called for by the construction rather than two or more. We avoid the name
‘coercion’ as it seems to imply a process by which some external distribution of the item in question forces
an interpretation not normally licensed. While this may be important for historical development of various
pumping constructions, or for online comprehension, the grammar treats the phenomenon no differently
from other sorts of constructions.
Among the other constructions in this category we can consider the use of plural nominals as predicates,
where they function as a familiar kind of reciprocal predicate (see Section 4.9), and the derivation of count-
noun ‘fractions’ from ordinal numbers. For example, the ordinal number third occurs as a pluralizable noun
in two thirds. The fraction vocabulary has one word that belongs independently to the fraction frame, half,
where the corresponding ordinal is second; and the ordinal first, of course, has no corresponding fraction.25

Clause-defining constructions Constructions account for constructs at all levels of grammatical descrip-
tion: for lexical units, maximal and non-maximal phrases, and clauses. Clausal constructs function as
declaratives, commands, questions, challenges, blessings, curses, etc, and these are realized linguistically
by a variety of patterns. Among these are Subject-Predicate, Imperative, and Aux-Initial
Question. Non-subject_wh-Questions are complexes made up of both Filler-Gap and Aux-Initial
components. Existential clauses (there’s a fire) are analyzed not a fixed clause type, but rather a special syn-
tactic valence possibility (especially with be) that can participate in Subject-Predicate, Aux-Initial,
Raising, and several other constructions.

4 Building the constructicon

This section introduces some notational conventions for presenting our analyses. In general, we will want
a way to indicate that a particular sign is built from one or more other signs: in other words, a construct.
A schematic representation of each construct will be given in the form of labeled bracketing, with outer
brackets ‘{ }’ enclosing the expressions produced by the construction, and inner brackets ‘[ ]’ enclosing the
individual construct elements. The constructs will be annotated in the same way, with mnemonic labels on
the brackets chosen for ease of recognition. For example, in the case of a construct with two daughters, the
general description will look like this:

(12) {M [D1 sign1 ][D2 sign2 ] }

In the simplest cases the phonological and morphological value of the mother is the juxtaposition of
the forms of sign1 and sign2 26 . Taking Rate as an example, we notate the whole (the Mother) and parts
(Daughters) as in (13b). Linguistic expressions annotated for a given construction will have wavy braces
25 Although the single words first and second are not usable as fractions, complex numerals containing them are – e.g. thirty first,

sixty second). There is probably little need to attribute this pattern to the synchronic grammar of English.
26 The non-simple cases will include the pacing and other prosodic indicators of segment boundaries in constructs that instantiate

Gapping or Shared_completion, as in section 4.2.

16
surrounding the whole construct, square brackets around the components, with labels the same as those
specified for the construction itself. In cases where blocks of examples are presented to illustrate the same
construction, we simply write, e.g. { [twenty dollars] [a day] }, without the labels, after the first example.
We also use this notation as in (13a) for abstractions of particular constructions:
(13) a. {Rate [Numerator sign1 ] [Denominator sign2 ] }
b. {Rate [Numerator twenty dollars] [Denominator [a day]] }
For each construction, the entry in the constructicon will include the following:

1. a bracketing formula offering mnemonic names for the mother and daughter constituents;

2. a mnemonic name for the construction;

3. an informal description of the properties of the mother consituent;
4. informal descriptions of each of the daughter constituents;
5. an interpretation of the manner in which properties of the daughters participate in the detailed proper-
ties of the resulting sign, in terms of syntax, semantics, pragmatics, and context.

For each construction examined, a formula of the above type is provided, along with example phrases or
sentences, an accompanying legend, and discussion. A simplified legend for Rate is shown in (14):

(14) {Rate [numerator ] [denominator ] }

Name Rate
M NP
D1 Numerator. A Quantified NP.
D2 Denominator. An indefinite singular NP.
Interpretation A ratio is built from numerator/denominator.

(15) a. They earn {M [D1 20 dollars] [D2 an hour]}.

b. The new hybrid gets {[sixty miles] [a gallon]}.

The legend states that: (1) the Mother (M) of the Rate construction will be an NP, (2) the first daughter
(D1), the Numerator, will be a quantified NP representing a quantity of units of one type, (3) the second
daughter (D2), the Denominator, will be an indefinite NP referring to a unit of another type, (4) the semantic
type represented by the mother will specify the concept created by a Ratio formed by the semantic repre-
sentations of the two constituent NPs. In labeled brackets we make reference either to D1, D2, etc, or to the
mnemonic given in the description of the daughter constituents, for ease of exposition.
Following this, we include, if relevant, further discussion of the details of the construction, including in
some cases a proposal for integrating the analysis into a complete SBCG grammar.

17
4.1 Lexical Idiom
Some constructions license idiomatic or partially idiomatic phrases that are assigned semantic roles not
predictable from their component parts. Take for instance the prepositional phrase in the distance ((16)–
(17)), which appear in descriptions of distant perception (and location), indicating where the percept is
located, namely, at a distance far from the person whose perception is being represented:

(16) {Loc_of _percept [D1 in][D2 the distance]}

Name In_the_distance
M PP, realizes the L OCATION _ OF _ PERCEPT FE in
distant perception frame: sight, hearing, and pos-
sibly smell.
D1 In. Takes the distance as complement.
D2 The distance. An NP with distance as the lexical
head and the as the marker.
Interpretation A phrase that indicates where a percept is located;
it may integrate into a P ERCEPTION frame (or a
subtype of P ERCEPTION) already evoked in the
context, or may evoke it itself.

(17) a. And then two human torsos appeared {M [D1 in][D2 the distance]}.
b. As I gazed out of the window I could see several groups of red deer {[in] [the distance]}, and in
the foreground the brown ferns with clumps of heather here and there; it was a wonderful sight.
c. Away {[in] [the distance]}, tucked between a fold in the surrounding hills, was home.

In FN terms, in the distance fills the FE L OCATION _ OF _ PERCEPT which occurs in perception-related
frames.
Frames of ‘distant perceptions’, i.e. vision and hearing, but not taste, touch or smell, allow the separation
in location phrases between (i) the location of the perceiver, (ii) the location of the whole event, and (iii) the
location of the thing perceived. In a sentence like Mrs. Grimshaw saw somebody on the neighbor’s roof it is
likely that the neighbor’s roof was not Mrs. Grimshaw’s location at the time of the event, nor the location of
the complete event; it is the location of what Mrs. Grimshaw saw. A phrase of the form in the distance, with
exactly these words (as opposed to freely generated phrases like from a considerable distance, at a distance
of several yards), always occurs as expressing the location of the object of perception.
We could say, then, that this phrase indicates the location of the perceived entity in expressions that
specifically evoke perception frames, but also that on its own, in a verbless sentence, it implies a perception
state-of-affairs, as in (17c) and (18):

(18) {[In] [the distance]} the rumble of tanks.

As noted above, the formulae and specifications for constructions in this format are built to be compatible
with a grammatical formalism such as that of SBCG. A full-fledged grammar will also make fine-grained
distinctions between degrees of fixed expression. In this case, in the distance is fixed except for a possible
modifier of distance. The AVM in Figure 3, a proposed SBCG analysis, shows this partially. The two

18
daughters are the word in and the phrase the distance.27 This is a slight simplification, as distance accepts
limited modification (e.g. in the far distance). It does not, however, license any complements: in the distance
from here to there is not described by this construction. The mother contributes the meaning that some entity
evoked by the phrase or clause it modifies is far away. Part of the contextual background of using this
construction is a scene of perception, in which the same entity which is located at a distance is understood as
perceived (indicated by coindexation of the fillers of the FAR _ ENTITY and L OCATION _ OF _ PERCEPT FEs).
We leave unresolved the question of whether there is a principled reason for why the distance cannot be
left-isolated (*it was the distance that we saw them in).
⎡ ⎡ ⎡ ⎡ ⎤⎤⎤⎤
prep
⎢ ⎢ ⎢ ⎡ ⎡ ⎥⎥⎥
⎢ ⎢ ⎢ ⎢
⎢   ⎤⎤⎥ ⎥⎥⎥⎥
⎢ ⎢ ⎢ CAT ⎢ ⎥⎥⎥⎥
⎢ ⎢SYN ⎢ entity_fr
,. . . ⎦⎦⎦⎥⎥⎥
⎢ ⎢ ⎢ ⎣ SEL ⎣SEM ⎣FRAMES ⎥⎥⎥
⎢ ⎢ ⎢ ENTITY i ⎥⎥⎥
⎢ ⎢ ⎣ ⎦⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ VAL   ⎥⎥
⎢ ⎢ ⎥⎥
⎢ MTR ⎢ ⎡   ⎤ ⎥⎥
in_the_dist-cxt ⇒⎢
⎢ ⎢ ⎥⎥
⎢ ⎢ ⎣ far_fr
⎦ ⎥⎥
⎢ ⎢SEM FRAMES ⎥⎥
⎢ ⎢ FAR _ ENTITY i ⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ ⎡   ⎤ ⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎣CNTXT ⎣ BCKGRND perception_fr
⎦ ⎦⎥
⎥
⎢ L OC _ OF _ PERCEPT i ⎥
⎣ ⎦
DTRS  [ FORM in], [ FORM the, distance] 

Figure 3: The in-the-distance construction

Other lexically-fixed or mostly-fixed idioms which would benefit from a constructional analysis include
in one’s own right, in which the only variable is the pronoun in the place of one’s, which covaries with the
subject of the phrase modified by in one’s own right (she was a renowned scholar in her own right). Other
candidate lexical idioms involve the noun word hurry as in in a hurry, what’s the hurry?, what’s your hurry?
(but not in my hurry, cf. in my haste).

4.2 Constructions with gaps

Several constructions are not associated with frames of their own, but rather specify arrangements of parts of
sentences in ways not normally licensed by the core clause-building constructions. Shared_completion
(commonly called right-node raising), Gapping, Aux-Initial clauses, and several others are of this
type. While many or most such constructions assign information-structural roles (topic, focus, etc) to their
constituent parts, such notions are distinct from ordinary semantic import.
We illustrate our coverage of this variety of construction with Gapping in (20d):

(19) {Gapping [Before ] [Middle ] [After ] ([conj ]) [Before ] [After ] }

27 The value of the FORM attribute is a list of morphological objects or formatives which are phonologically realized (Sag this

volume). It serves here to identify particular lexical units.

19
 Name Gapping
M A coordinate structure, whose non-initial con-
juncts are missing some linguistic material present
in the first conjunct.
D1 Before. In each conjunct, the material before the
Middle.
D2 Middle. A string in the first conjunct which con-
tains at least a verb, and which is omitted from
subsequent conjuncts.
D3 After. In each conjunct, the material after the Mid-
dle.
D4 Conjunction. A conjunction, if present. The con-
struction here relies on (or is a kind of) coordina-
tion construction. Must be nor if Middle contains
clausal negation.
Interpretation Each non-initial conjunct is missing some material
which is present in the initial conjunct, and each
conjunct is interpreted and parsed as though that
missing material were present (in the simple case,
D2 is interpolated between D1 and D3).

(20) a. {[Before He] [Middle adores] [After Mama], [conjunctionand] [Before she], [After him]}
b. Again, {[Targets] [were] [set], [times] [carefully recorded], [and] [fitness] [improved].}
c. {[A couple of bedrooms] [overlook] [Loch Ness], [and] [others] [the village and the Caledonian
Canal].}
d. {[He] [made no attempt to flirt] [with her] [nor] [she] [with him].}

Gapping licenses a coordination structure in which some verbal element in the first conjunct is omitted
from subsequent conjuncts, which are parsed and interpreted exactly as though the middle part of the first
conjunct were between its B EFORE and A FTER elements (with some modifications required for anaphora,
negation, etc., as in she had hardly enamoured herself with the locals, nor had they enamoured themselves
with her). We additionally note the conjunction as an optional CE. Though not strictly required by the
construction, when present it is a lexical indicator of the coordination struction that is required by Gapping.
The difference between our analysis of this variety of construction and other, frame-bearing construc-
tions, highlights the distinction between construct annotation and description, on the one hand, and writing
a full-fledged construction grammar on the other. A grammar must account for all aspects of every construc-
tion and their relations to each other. The goal in building a resource like the constructicon is to specify
the syntactic and semantic constraints on the components of the expressions that can realize each construc-
tion, to characterize the sort of object resulting from their assembly (the resulting sign), and to explain the
grammatical context in which that object finds itself.
This means that annotation of these semantically inert constructions is to a degree independent of any
particular constructional analysis. Shared completion (e.g. integral to and dependent on the system) may
be analyzed in a grammar as a complex sort of syntactic valence manipulation, or as phonological deletion.
To the constructicon builder and annotator, however, all that is available to analyze is what appears in the

20
sentence. The relevant units of the construct may be identified, and instructions as to how to interpret them
may be carried out regardless of the particular way the grammar might do so.
This contrasts with our lexical annotation, and also with annotation of semantically-valued constructions,
in that the labels applied to the various construct elements are not automatically informative as to how they
contribute to the meaning of a sentence. While Frame Elements identify event participants, are arranged in
a hierarchy, and may be connected to an external ontology, the labels applied to the subparts of a Gapping
construct must be looked up in a prose description which explains what to do with these constituents.
We include some level of detail for these constructions (e.g. the phrase type of the mother, and of the
subparts [if they are indeed constituents]) not because they are a crucial aspect of the construction, but in the
hopes that this sort of detail will aid the learning of these categories by an automatic parser or similar tool,
thus preventing misparsing of these often-occurring structures.
The ellipsis constructions illustrated in (21) are somewhat different in that the missing material is re-
trieved anaphorically or deictically:

(21) a. A: Do you swim? B: On the weekends.

b. I’m interested in astronomy, and my sister is too.
c. (while gesturing) A doggie!
Annotation of these constructions involves noting the segments which have material omitted and recognizing
that these represent phrases ‘larger’ than they might otherwise appear to (e.g. clauses, or ‘utterances’). As
before, the exact analysis of such constructions (as phonological deletion, valence manipulation, anaphora,
etc) is in large part independent of the analysis and annotation a constructicon can provide.

4.3 Clause-level constructions

In this section we consider two groups of constructions that license signs at the level of complete clauses:
Absolute constructions and Aux-initial clauses. The category also includes Conditionals, Comparative_
correlative, and many others.

4.3.1 Absolute constructions

There is a family of constructions consisting centrally of a subject NP and a nonfinite predicate, as in the
kids off to school. Such expressions are in many ways interpretable as the complements of one of the uses of
have (having the kids off to school) and it needs to be appended to a clause that provides the ‘subject’ of that
have. The first part of (22a) is basically synonymous with Naipaul had these events in mind. The paraphrase
is not always possible, however. The absolute clauses in (22b) are not (easily) paraphrasable with have (*he
had his hands shaking):
(22) a. These events in mind, Naipaul wrote a novel.
b. He sits there, his hands shaking, his movements robotic.
The construction also hosts a number of fixed expressions with bare nouns (arms akimbo, head held
high, hat in hand).

(23) {Secondary_predicate [D1 sign1 ] [D2 sign2 ] }

21
 Name Absolute:Plain
M Secondary Predicate, modifies a main
clause.
D1 a definite NP; external argument of D2.
D2 a nonfinite predicate (verbal or nonverbal).
Interpretation main clause’s subject ‘has’ D1 in state D2.
Related constructions Absolute:having;
Absolute:with;
Absolute:what_with

(24) a. He entered the room, {Secondary_predicate [D1 his head] [D2 held high]}
b. {[the team] [already up in arms]}
c. {[tax day] [approaching]}

A separate construction, Absolute:with, places with before D1 and D2, and yet another construction
(Absolute:what) licenses what before an Absolute:with phrase. The whole expression may also
have appended to it and all.
(25) a. {[with] [the house] [to herself]}
b. {[what] [with [everyone] [gone]]}
c. {[what] [with [everyone] [gone]] [and all]}
There is a distinction between depictive uses of these phrases (with his arms akimbo, with his hands in
his pockets) that merely describe some state of the subject of the main clause, and those uses that express
preconditions for the situation expressed in the main clause (what with the kids off to school, I have a lot
more time on my hands), which is what FN calls the C IRCUMSTANCES extrathematic FE (Ruppenhofer et al.
2006: Appendix A). Absolute clauses with only D1 and D2, or with initial with are compatible with both
uses, but what with can only hold the preconditions function:
(26) a. With his arms ready for a fight, he entered the room and looked around.
b.*What with his arms ready for a fight, he entered the room and looked around.
Given this we must consider Absolute:what_with to be a separate construction. We assumed
that the version with only with required a separate construction, but it may be possible to treat with as an
‘optional’ CE. We leave this question for further investigation. Another, related, difficulty is determining
the constituency of these constructs, especially when all the possible elements are present. The table above
analyzes with as taking two arguments, the first of which is specified as the external argument of the second
(similar to the ‘raising-to-object’ constructions described in Sag et al. 2003). The bracketing for versions
with what shows what as taking with plus its arguments as a single valent.

4.3.2 Aux-initial
Aux-initial constructions have a particular arrangement of subject, auxiliary, and predicate (Fillmore
1999). Like Subject-Predicate, Aux-initial does not introduce its own frame-semantic meaning,
and rather serves an organizational function. Aux-initial clauses consist of, in order, a finite auxiliary

22
verb, its subject, and a non-finite predicate complement of the auxiliary. This general Aux-initial con-
struction does not itself license any main clauses, but several more specific constructions instantiate the
general Aux-initial type while adding their own lexical or syntactic restrictions and semantic spec-
ifications. The general type may be understood as involved in larger main-clause constructions such as
non-subject wh-interrogatives:

(27) {Aux−initial [D1 sign1 ] [D2 sign2 ] [D3 sign3 ] }

One type of Aux-initial construction, with initial had, should, or were, is assigned a conditional
interpretation:

(28) {Aux−initial:cond [Aux sign1 ] [Subj sign2 ] [Pred sign3 ] }

Name Aux-initial:conditional
M Inverted finite clause.
D1 Auxiliary verb, either had, should, or were.
D2 NP, the subject of D1.
D3 Predicate (verbal or otherwise), selected by D1,
shares subject with D1.
Interpretation A conditional clause, with interpretation varying
with the identity of D1.

(29) a. {Aux−initial:cond [Aux had] [Subj I] [Pred known you were coming] }
b. {[had] [you] [arrived on time]} (past counterfactual)
c. {[should] [you] [encounter problems]} (present hypothetical)
d. {[were] [there] [any solution]} (present counterfactual)

Because the predicate, D3, is selected by D1, its form covaries with the identity of the auxiliary. That is,
with had, the predicate is a perfect participle, with should it is a bare infinitive, and with were any variety of
copular predicate including NP, PP, AP, and so on.
An Aux-initial clause with may as the auxiliary expresses a special pragmatic meaning, which one
may call ‘magic,’ or the expression of blessings, curses, wishes, and the like:

(30) {Aux−initial:magic [Aux may] [Subj sign2 ] [Pred sign3 ] }

(31) a. {Aux−initial:magic [Aux May] [Subj you] [Pred roast in hell]}!

b. {[May] [you] [live a long life]}!
c. {[May] [his hair] [fall out]}!

Aux-initial is the base construction upon which a number of elaborations are formed, among them
the construction that licenses polar (yes/no) interrogatives, as in did you see it? and won’t it be dangerous?
This construction places no restrictions on the auxiliary, which can be either positive or negative. The choice
of polarity has some pragmatic consequences with respect to presuppositions of the speaker and expected
responses to the question.

23
(32) {Aux−initial:polar [Aux Have] [Subj you] [Pred fed the cat?] }
Note that the polar interrogative construction is not implicated in non-subject wh-interrogatives (where
did you see it?), as those have quite different semantics, pragmatics, and intonation. Rather, the wh-
interrogative construction can simply be said to require of its non-wh-phrase daughter that it be an Aux-initial
clause of the most general type.
This does not exhaust the appearances of Aux-initial. As illustrated below, it also figures optionally
in the second part of the the x-er the y-er construction (Fillmore et al. 1988), and in comparative clauses.
Huddleston and Pullum (2002) argue that in comparatives, this is not quite the same construction, but rather
involves subject postposing. Note in particular that Aux-initial sentences in comparative clauses must
be accompanied by VP ellipsis:
(33) a. The faster the current flows in the channel, the more rapidly {[will] [the changes] [take place]}.
b. She interpreted it far more broadly than {[had] [her predecessors] [(*interpreted it)]}.

4.4 Co-text specifications

The simplest kind of lexical representation for complement-taking verbs regards the verb as the center and
the complements as annotated externally to the center as its semantic dependents. In the case of the verb
explain we find the associated argument structure of the verb specified in the verb’s lexical entry (Figure 4),
and annotated in-place in annotations of sentences containing the verb.
⎡ ⎤
FORM explain
⎢ARG - ST NPx , NPy , PP[to]z  ⎥
⎢ ⎡ ⎥
⎢
⎢ ⎡ ⎤ ⎤⎥
⎥
⎢ ⎢  explain-fr ⎥⎥
⎢ ⎢ ⎢S PEAKER ⎥ ⎥
⎢ ⎢ x⎥ ⎥⎥
⎢SEM ⎢FRAMES ⎢ ⎥ ⎥⎥
⎢ ⎢ ⎣T OPIC y⎦ ⎥⎥
⎣ ⎣ ⎦⎦
A DDRESSEE z

Figure 4: Lexical entry for explain

In the annotation of instances of this verb, the fillers of the argument structure requirements (i.e. the
frame elements) are indicated by bracketing independently of their syntactic position in the sentence. Thus,
for She explained her solution to her classmates, the frame elements S PEAKER, T OPIC, and A DDRESSEE
are indicated as follows:

(34) [Speaker She] EXPLAINED [Topic her solution] [Addressee to her classmates].

Most verbs, like explain, form with their valence dependents self-standing structures that impose no
requirements on their environment. But for certain verbs it is necessary to recognize both valence elements
(elements of the ‘governed’ context) and co-textual requirements (elements of the ‘governing’ context). The
verb bother, and the verb stand, in two of its uses, require the recognition of both complementation and
governing contexts: the latter is represented with italicization in the following annotations:
(35) a. [She] didn’t BOTHER [to tell me how to get out].

24
 b. [I] can’t STAND [to listen to his kind of music].
c. [I] won’t STAND [for this kind of nonsense].
The requirement for negation of some sort is well known (and not most accurately captured by the
term ‘negation’), but in the case of stand, a statement of governing modals (can, etc) is also necessary.
The best way to accomplish this sort of restriction in formal grammar is left open.28 In cases where a
particular construction specifies requirements of a governing context, we indicate those requirements with
an X element. While this may not always strictly correspond to a feature or bundle of features in an SBCG
analysis, it is a necessary part of understanding the distribution of particular classes of constructs and is at
least useful to gather such information under a single header. We do this for various of the measurement
constructions (section 4.13), for which the notion of support is relevant, and for a particular use of long
meaning a long time.
Phrases headed by long can occur as time-duration expressions in (at least) two separate contexts, mean-
ing something like ‘for a long time’. This is described not directly as a property of long itself but of long as
an adjectival or adverbial head. We treat long as an adjective here, rather than an adverb, but nothing crucial
rests on this decision.
The first context is in negative environments, as the argument of a head that requires a time-span (won’t
spend long or don’t take too long).

(36) {long.NPI [long] }

Name long.NPI
M NP, time duration expression.
D1 AP, headed by long.
X A ‘trigger’, the contextual element that licenses
long as an NPI, either lexical (e.g. negation) or
constructional (e.g. yes-no question).
Interpretation ‘for a long time’, plus any modification of long.
The construction describes an AP headed by the adjective long, which is interpreted as ‘a (degree)
long time’, has time-adverbial function, and requires a negative polarity context. The mother is understood
syntactically as an NP which may nonetheless be used in some slots usually reserved for adverbs (see also
Huddleston and Pullum 2002: 569). Note the parallel between stay very long and stay a month, and also stay
for very long, all of which indicate that long is treated as an NP. Examples of verbs which take long-phrases
or NPs as complements are be, spend, stay, take, and wait. The verb be in this list can be used without a
locative complement if the location has anaphoric zero interpretation: I won’t be (there) long.
Post-verbal long, as described by this construction, generally requires a negative polarity context, in-
dicated by the X constituent above. In our more informal constructional descriptions the X constituent
indicates some contextual feature called for by the construction, including negative and modal contexts. We
also use it to indicate possibilities for optional or required support words, e.g. be, stand, or measure for the
construct five feet tall. In the former case we also use the term trigger to designate the contextual element
that licenses the construct:
(37) a. Will you be {long.NPI [D1 long]}? (Aux-Initial question)
28 We follow the BNC tokenization, which separates negation from the modal in ‘contractions’ such as can’t, while recognizing that a

more linguistically sophisticated understanding of negated auxiliaries is possible, and compatible with our overall methods (e.g. Zwicky
and Pullum 1983).

25
 b. Have you been waiting {[long]}? (Aux-Initial question)
c. I ca [Trigger n’t] stay {[very long]}. (negation with not)
d. This wo[n’t] take {[long]}. (negation with not)
e. I hope we do[n’t] have to spend {[quite so long]} this time. (negation with not)
f. [If] you’re going to be {[long]}, maybe I’ll just go ahead. (conditional-clause)
Immediately before a verb, long also indicates ‘a long time’, as in (38):

(38) a. I have long known that this is the case.

b. They have long suspected there might be problems.
Long in this use prefers to accompany a VP with perfective have, though some other contexts, e.g. passive,
are well attested. Plain past tense seems rather awkward:

(39) She ?(has) long claimed that she was the rightful heir.
The collocation long since has a similar preference for perfective contexts, but indicates a past change-of-
state rather than a continuing state of affairs, paraphrasable as long ago:

(40) They have long since given up.

4.5 Degree modification

There are many ways of modifying the degree of a scalar adjective (or adverb; here we consider just adjec-
tives), including modifiers like very and quite, but there is a special set of modifiers that figure importantly
in other constructions. These include the words as, so, that, this, how, too, and enough. Following Kay
and Sag (this volume) we classify these words as belonging to a particular lexical class of degree modifiers
(accomplishable in HPSG/SBCG, by having the feature specification degree for the MARKING feature (Van
Eynde 2007)):

(41) {deg.mod.adj [deg.modifer sign1 ] [adjective sign2 ] }

Name Degree Modification

M AP with combined valences of D1 and D2.
D1 degree word, which may have its own valence (e.g.
so takes a that-clause).
D2 adjective without degree modification.
Interpretation A value on a scale is located with respect to a ref-
erence value, in ways specified by the particular
degree marker.
The constructs in question have two construct elements, the degree-marker and the adjective. In the
case of enough the degree marker follows what it modifies; elsewhere it precedes. The mother sign is a
complex adjective, and when the adjective itself has complements, its complements are combined with those
of the degree marker and must be distributed separately. Kay and Sag (this volume) provide a mechanism

26
for accomplishing this, by specifying the complement of a degree marker as being on an EXTRA list of
arguments which are potentially linearly ordered after any complements of the head adjective.
Some examples in which the degree-marker’s complement is alone are shown in (42):

(42) a. I was so angry that I could hardly speak.

b. I was too angry to speak.
c. She is more intelligent than you are.
d. She reacted so quickly that I was taken aback.
e. He acted meekly enough to avoid arousing suspicion.
f. I was so eager to win that I could hardly speak.

(43) a. I was too eager to win to speak.

b. This piece is easy enough to understand to assign to the class.
c. We were more interested in girls than in homework.
d.*We were more interested than in homework in girls.
e. You were more interested than I was in linguistics.
f. You were more interested in linguistics than I was.

(43a–f) are examples in which the degree-marker’s complements combine with the adjective’s own comple-
ments. Regarding ordering, in comparative cases (-er, more, less, and as) where the standard is compared
with the subject, the complement can occur before or after the adjective’s complement ((43c)–(43f)); in all
other cases the degree modifier’s complement must come at the end.
Degree-marked adjectives indicate a degree (or range of degrees) on a scale identified with the meaning
of the adjective, with respect to an implicit or explicit reference value. Comparatives (more, less, as) indicate
a degree higher than, lower than, or equal to a reference value that can be directly or indirectly expressed with
a than-clause or an as-clause. So and enough express sufficiency, and too expresses excess, a degree which
surpasses a level under which some event or situation is possible. Each is associated with specific patterns of
complementation. While enough and so both involve sufficiency, enough patterns syntactically with too in
licensing an infinitival complement and participating in control constructions.The complement indicates the
state-of-affairs that is made possible (in the case of enough) or impossible (too). The unexpressed subject
of the infinitival may be identified with the external argument of the AP (as in (44b)–(44c)), or may be
interpreted generically or from the speech or discourse context (as in (44a) and (44e)):
(44) a. It’s {deg.mod.adj[deg.mod too] [adj old]} to eat. (= It’s so old that one shouldn’t eat it.)
b. I’m {[too] [old]} to eat spicy foods. (= I’m so old that I shouldn’t eat spicy foods.)
c. He’s {[adj old] [deg.modenough]} to understand this. (= He can understand it)
d. He’s {[old] [enough]} to take to adult movies. (= Someone can take him)

27
 e. Is it {[warm] [enough]} to eat outside tonight?
f. It’s {[too] [hot]} outside for a picnic.
THIS NEEDS TO BE CORRECTED - THE CONNECTION BETWEEN THE PRECEDING EXAM-
PLES AND THE FOLLOWING PARAGRAPH.
These infinitivals are may also be ‘hollow,’ that is, missing a non-subject NP (cf. Huddleston and Pullum
2002: 1245) which is identified with the AP’s external argument or modified NP (as in (44a) and (44d)).
An alternate valence, with a PP[for], has no unexpressed subject, but nevertheless the expression indicates
directly or metonymically an event which is enabled or made impossible (see (44f)). The complement of too
(but not enough) is a negative polarity context (too tired to do anything) because exceeding the relevant limit
implies that the state-of-affairs is not possible: someone who is too tired to eat is so tired that he cannot eat.
The degree word so licenses a degree that-clause which expresses a state-of-affairs which holds by virtue
of reaching a particular threshold on the relevant scale:

(45) a. It was so hot that we couldn’t eat outdoors.

b. We were so tired that we decided to go straight home.
Compared adjectives constitute a subclass of degree-modified scalar adjectives. Several constructions
license constructs of this grammatical category: with the premodifiers more, less, and as), or by suffixation
of -er. The category is also assigned to two specific lexical items, better and worse. Any word or construct
of this category acquires the combinatory possibilities common to all expressions of the compared-adjective
type. The valence requirements of these phrases include:

– the primary entity of the comparison

– the ‘standard’ of comparison

The standard of comparison is introduced with the word as in the case of the Comparison-of-equality
version, but with than elsewhere. If compared adjectives have complement structures of their own, these are
combined with the valence of the comparative with the possibility of extraposition in a manner similar but
not identical to that for other degree expressions (on which see Kay and Sag this volume).
Some degree markers combine with their adjectives to create negative polarity items: that, so, too, as
seen in the following sentences (with the addition of an external T RIGGER constituent, the same sort of
element seen for long above):
(46) a. It’s [Trigger not] {[that] [expensive]}.
b. He’s [not] {[so] [smart]}.
c. He’s [not] {[too] [smart]}.
The NPI degree modifiers here have different meanings from non-NPI versions. That as an NPI in (46a)
is not understood as a response to a question about the exact cost of an item. So in (46b) is neither the
gushy modifier (It was so cute!) nor the modifier that expects a that-clause completion (45); as an NPI it is
unstressed). The too in (46c) similarly is not the modifier that expects an infinitival completion (It was too
hot to touch), and is again unstressed. Finally, we note that constructs produced by this construction are of a
type that permits their participation in the exceptional degree marking construction(see section 4.7, Zwicky
1995, Van Eynde 2007, and Kay and Sag this volume).

28
4.6 Degree qualifier realization
The ‘complements’ of complement-bearing degree markers, the infinitival VPs, that-clauses or than- or
as-phrases, require a separate construction for their realization, because the lexical constituent to which
the degree marker is formally appended is not necessarily the constituent to which the degree qualification
applies, or in other words, is not by itself the ‘scope’ of degree modification. The needed mechanism,
permitted by SBCG, allows the degree marker itself, together with the degree qualifier, to be ‘percolated’ to
a higher phrase, and provides a construction that juxtaposes the relevant degree-modified phrase with the
qualifier phrase. The constructional specification and examples can be seen below; the bracketing shows the
intended parse:

(47) {XP [scope sign1 ] [deg.qual sign2 ]}

Name Degree_qualifier_realization
M XP (identified with that of D1).
D1 XP containing a degree marker (more, less, as, so,
too, enough, etc.). Provides the scope of degree
modification introduced by the degree marker.
D2 Degree qualifier. With comparatives, a than-
phrase; with too and enough, an infinitival VP;
with so, a that-clause.
Interpretation The scope of the degree modification, provided by
Degree-qualifier, includes the entire meaning of
D1.

(48) a. He is {[scope heavier] [deg.qual than 200 pounds]}.

b. {[He is heavier] [than my mother]}.
c. a man {[heavier] [than 200 pounds]}
d. a man {[heavier] [than my mother]}
e.*a {[heavier man] [than 200 pounds]}
f. a {[heavier man] [than him]}
g.*a {[heavier man] [than my mother]}
h. He {[is so old] [that he remembers FDR’s inauguration]}.
i.*He {[looks so old] [that he remembers FDR’s inauguration]}.
j. She {[looks so young] [that the bartender refused to serve her]}.
k.*She looks as if she {[is so young] [that the bartender refused to serve her]}.
l. You {[make too many mistakes] [to deserve this job]}.
m. {[You need a bigger car] [than I do]}.

29
 n. Do you think I {[read big enough books] [to impress the librarian]}?
o. {[I have richer friends] [than you do]}.

The claim is that the formulation of the degree qualifier is sensitive to the contents of the scope. Compare
(48f) and (48g) where we see that gender must be consistent between the scope and the degree qualifier.
Similarly for (48e), which illustrates that if the scope indicates comparison of the weight of individuals, then
the degree qualifier cannot simply indicate a measurement expression.29 This requirement allows us to see
that in sentences like (48h) the scope is larger than the morphologically modified adjective – so old here.
The oddness of (48i) (contrasted with the lack of oddness of (48h)) shows that the scope must additionally
include the verb, since the choice of verb affects the acceptability of the sentence.

4.7 Exceptional degree markers

Degree-modified adjectives licensed by the constructions described in the previous two sections may not
directly precede nouns. In this they contrast with very hot, which appears wherever hot does. Rather, they
attributively modify only indefinite singular NPs, coming before the a. A variant of the construction inserts
of between the AP and the indefinite NP:

4.7.1 Exceptional degree markers.Plain

(49) {NP [deg.AP sign1 ] [indef.NP sign2 ]}

Name Attributive Degree Modification

M NP
D1 Degree-modified AP
D2 Indefinite singular NP, determiner: a
Interpretation An NP, with D1 modifying D2

(50) a. {NP [deg.APtoo big] [indef.NPa box]}

b. {[so old] [a building]}
c. {that big] [a problem]}
d. {[how serious] [a problem]}
e. {[more interesting] [a job]} than I thought

The degree-modified adjective includes those modified with so, too, that, as well as all comparatives.
This construction presupposes a type, namely degree-marked adjectives, and, following Kay and Sag this
volume, juxtaposes such adjectives before singular indefinite NPs. Adjectives with other manner of degree
modification, e.g. very serious, modify nouns NP-internally: a very serious problem.
29 The treatment suggested here matches Bresnan’s (1973) analysis of Chomsky’s (1965: 180) example with the entailments of more

successful lawyers than Bill, discussed in McCawley 1988: 694–5 as ‘circumnominal comparatives’: the N expresses not ‘a degree of
success’ but ‘a degree of being a successful lawyer’.

30
4.7.2 Exceptional degree markers with of
A version of the exceptional degree-marking construction has an intervening of before the indefinite NP.
We have not examined the possibilities carefully, but it does seem that some degree modifiers welcome of
more readily than others; the construction shows similarities with a pattern in which the NP’s modifier is
headed by much. Zwicky (1995) presents an analysis in which the headedness of the of phrase is shared
by the preposition and the noun, thus both prepositional features and nominal features appear at the ‘PP’
level. This allows the construction to select a PP with a particular preposition, which in turn selects an NP
with a particular determiner—i.e. a local account of what to all appearances is niece selection, a violation
of most notions of locality, accomplished by careful distribution of the relevant features across the syntactic
structure. The relevant structures are illustrated below.30
(51) a. {[too big] [of a problem]}
b. {[much bigger] [of a problem]} than we imagined
c. {[serious enough] [of a problem]}
d. [not] {[much] [of a problem]}
e. [not] {[that much] [of a problem]}
f. {[how much] [of a problem]}

4.8 Nominal Pumping

Some constructions do not license a mother sign from a combination of daughter signs, but rather from a
single daughter sign. These constructions license interpretations and contextual constraints not present in
the daughter sign but do not alter the morphophonology of the item in question. These are sometimes called
PUMPING constructions (for more details see section 3).31
By default nouns are either count or non-count, and this restricts their syntactic behavior (e.g. combi-
nation with determiners, pluralization) and construal (as denoting (a class of) individuals, or a substance,
etc.). Constructions exist which change nouns from one type to another, however, as in hand me a beer (= a
serving of beer) or give me some pillow (= some unbounded portion of the pillow).

4.8.1 Count-to-Mass: substance interpretation

(52) {N.mass,Substance [N.count,Entity ] }

30 One exhibits some puzzling behavior in this construction. Although a one is normally ruled out, it is possible with this construction:

The menu is as good a one as I have every seen, Use as large of a one as you can. It also seems marginally possible for the modified
NP to simply be one as long as of is present: It was a problem, but thankfully not so serious *(of) one that we had to stop everything.
We leave the analysis of how best to characterize this for future work.
31 It may turn out that some of the pumping constructions we describe here, including the well-studies cases of verb argument structure

constructions (caused motion, resultative, ditransitive, etc.) are best described not as fully-productive constructions but as patterns of
coining: a set of lexically-specified expressions which exhibit a pattern available for the creation of novel expressions (e.g. by analogy),
but the established members of which are not derivable from anything else in the language (Boas 2003, Kay 2005, to appear) Pumping
constructions seem to be central in the debate over productivity and the the status of ‘coercion’ in synchronic grammar (Ziegeler 2007),
but the construction/pattern of coining distinction is independent of the ‘size’ of the pattern. In any case, all these patterns, regardless
of their degree of productivity, are expressions which we believe can and should be included in a constructicon.

31
 Name Count-to-Mass.general
M Mass noun, interpreted as a substance.
D Count noun.
Interpretation The meaning of the noun construed as a substance,
or an unbounded region.

(53) a. Could you hand me some {NP.substance [N.count pillow]}?

b. Now there’s {[apple]} all over the floor.

4.8.2 Meat-grinding

(54) {N.mass,Meat [N.count,Animal ] }

Name Count-to-Mass.meat
M Mass noun, the meat of the animal.
D Count noun, animal name.
Interpretation For human consumption only.

(55) a. We always eat {[frog]} on Sundays.

b. Would anyone care for some {[mallard duck]}?

A particular variety of count-to-mass constructions concerns animals as food. Some meat names are not
the names of the animals from which the meat is taken: pork, beef, mutton, venison, etc. Other meat names
are derived from the names of the animals that provide the meat: chicken, lamb, turkey, fish, and so on. New
meat names can be generated from animal names, and of course animal names can be more than just single
nouns: beaver, pigeon, but also golden eagle or Welsh terrier. The nouns derived by this construction – as
well as the underived meat names (pork, etc.) – are specifically interpreted as dedicated for human cuisine.
This is shown by the fact that the pattern is not used (except playfully) for the food of carnivorous animals:

(56) a.?flies like venison

b.?the wolf who lives near our ranch likes lamb
Though this construction should in principle license phrases like cook cow, eat sheep, braise deer, the
existence of separate lexical items (beef, etc.) makes its use much rarer (or impossible).
Important to note is that as an appropriate construction is available to change the class of noun, the pro-
cess may be repeated. An animal name has the properties of a count noun, e.g. the possibility of occurrence
with the indefinite article a(n) or in plural number (a beaver, some beavers, while the same word as licensed
by the Count-to-Mass.meat construction can occur with quantifiers like much and some and is limited
to singular number. A lamb refers to an animal, some lamb refers to meat, but since both kinds of nouns can
occur in the singular with the definite article the lamb is ambiguous between the two possibilities.32
32 The output of the Count-to-Mass.meat construction may be the daughter of a non-count-to-count construction in the context

of food orders: they ordered two lambs, i.e. two dishes with lamb. For this and similar reasons, we prefer the constructional analysis of
these cases over ‘coercion,’ in which the noun class is determined based on the syntactic environment it finds itself in. Such an account
has difficulty with the ambiguity of the lamb, while an account with all the constructionally-specified constraints spelled out handily
captures the possibility.

32
4.8.3 Mass to Count: Portions and Varieties
Two constructions allow nouns that denote substances, such as beer, to denote instead (i) a portion of the
substance, or (ii) a particular variety of the substance. In each case, the size of the portion – cup, mug, pitcher
– or nature of variety – producer, brewing style, etc. – is determined by the linguistic and extralinguistic
context:
(57) a. {N.count,portion [N.mass,substance sign1 ] }
b. {N.count,variety [N.mass,substance sign1 ] }

(58) a. Would you care for another {N.count,portion [N.mass,substancebeer]}? (glass, mug, ..)
b. Here’s another {N.count,variety [N.mass,substancebeer]} you might like. (brew, . . . )

4.8.4 Proper to common

Proper nouns do not normally appear with determiners or pluralization, yet constructions exist which place
them in these contexts. Here we simply indicate the various interpretations of these constructions with
bracketing:
(59) a. There were three {[Michael]}s and a {[Robert]} in the class. (people with the name)
b. You’re no {[Julia Child]}. (person having salient properties of the individual bearing the name)
c. They are a modern-day {[Romeo and Juliet]}. (with modification)
Exactly how many constructions are necessary (we do not necessarily assume that even the above three
examples involve three separate constructions) is a topic we cannot address here, but the general idea is
similar to the constructions outlined above.

4.9 Reciprocal Relations

Not only single nouns, but nominal phrases may be the daughter sign of a pumping construction. One in-
teresting case of this is exemplified in (61). A plural nominal (i.e., a non-maximal noun-headed structure,
possibly consisting of only a noun) headed by a noun with a certain class of meanings can, by this construc-
tion, work like a reciprocal predicate, in which the two sides of a relation may be expressed jointly, as plural
or coordinated external argument, or separately, where one is foregrounded (as external argument) and one
is backgrounded (as a PP):

(60) {Reciprocal_pred [Plural nominal] }

Name Reciprocal_predicate_pumping
M Symmetric predicate with either distributed or collective
valence.
D Plural nominal, headed by a noun evoking a reciprocal
relation frame.

(61) a. I am {Recip.pred [Interpersonal.reln best friends]} with the president of Finland.

33
 b. The president of Finland and I are {[best friends]}.

Numerous lexical items have this pattern of valence alternation, including those below and also meet,
collide, and collaboate among many others:

(62) a. Diagrams A and B are similar. Diagram A is similar to diagram B.

b. Bill and Joe fought all day. Bill fought with Joe all day.
Nouns participating in this construction evoke frames inheriting from Personal_relations, either so-
cial or kin-related: friend is the most common (perhaps also historically prior), but also in the list are buddy,
pal, enemy, roommate, coworker, partner, neighbor, sibling, and cousin. Modification of the noun is also
possible (she is close friends with the producer), which indicates that the construct daughter is potentially
phrasal. Only modification which characterizes the nature of the relationship is allowed. Among the per-
mitted modifiers are close, best, old, bitter (e.g. rivals), and college, army; modifiers such as rich, tall, and
young are prohibited.
The mother sign additionally evokes Reciprocality, with the relation picked out by the daughter
specified as reciprocal (or symmetrical). This is often already the case, as with cousin, but need not be, as
with brother. The result is that a predicate like brothers incorporates the meaning of the lexical entry and
the construction: brother calls for two individuals such that (at least) one is the brother of the other, and the
construction guarantees that each is the brother of the other, while also constraining the syntactic means of
expressing who those individuals are.
The constructed sign is morphologically plural, but does not denote a (plural) individual in a personal
relationship. It is simply a predicate of that relationship. Because the resulting sign is neither a noun, nor
does it have (nominal) plural meaning, a singular external argument is possible (I am friends with them).
The mother sign does not itself have a valence specification, but its classification as a symmetric predicate
allows it to combine with either of two valence constructions compatible with symmetric predicates, by
which either one party of the relationship is the subject and the other is in a PP-with, or both parties are
realized by the subject.
In SBCG terms, the construction may be represented as in Figure 5. The type of the mother sign,
symmetrical-expression, is that class of predicates which may combine with valence constructions men-
tioned above, so it is unnecessary to state those valence possibilities in this construction. Unlike the earlier
examples of noun pumping, the mother sign in this case is not identified as a noun, or as any other part of
speech, but simply as a nonverbal predicate, which must be supported by be (or a separate construction that
accomplishes predication, e.g. Him, friends with the president?). Familiar constructions involving nouns,
such as determination, are correctly ruled out: *I am the best friends with the coach. The daughter of this
construction is specified as a plural nominal, but because the resulting predicate is not semantically ‘plural,’
that frame does not become part of the mother’s FRAMES list. The daughter designates an interpersonal
relation, while the mother sign adds the additional constraint that the relation be reciprocal.
The construction does not recognize the FE realization preferences of the daughter; independent of
this construction, the lexical head normally provides its own way of marking the second party of the rela-
tionship: friend of his, friend to the rich, her cousin. These are superseded by the valence constructions
which combine with symmetric predicates. Realizations not superceded by the construction include FEs
like S OURCE _ OF _ RELATIONSHIP (high school friends), which normally are pre-nominal modifiers. The
construction must also be able to articulate with the construction(s) licensing third cousin twice removed
(my barber is third cousins twice removed with my high school principal).

34
reciprocal_plural_predicate-cxt ⇒
⎡ ⎡ ⎤ ⎤
symmetrical-expression
⎢ ⎢    ⎥ ⎥
⎢ ⎢ ⎥ ⎥
⎢ ⎢SYN CAT
PRED +
⎥ ⎥
⎢ ⎥
⎢ MTR ⎢
⎢ NUM null ⎥
⎥ ⎥
⎢ ⎢ ⎡ ⎤⎥ ⎥
⎢ ⎢   ⎥ ⎥
⎢ ⎢ ⎥ ⎥
⎢ ⎣SEM ⎣FRAMES 1 ⊕ recip-fr
⎦⎦ ⎥
⎢ ⎥
⎢ RECIPROCAL _ SITUATION s ⎥
⎢ ⎥
⎢ ⎡ ⎤ ⎥
⎢ ⎡  ⎤ ⎥
⎢ ⎥
⎢ ⎢ ⎢
noun
⎥ ⎥ ⎥
⎢ ⎢ CAT
⎥ ⎥
⎢
⎢ ⎢SYN ⎢
⎢ NUM pl ⎥
⎥ ⎥ ⎥
⎢ ⎢ ⎢ ⎥ ⎥ ⎥
⎢ ⎢ ⎣ XPj  ⎦ ⎥ ⎥
⎥⎥
VAL
⎢ ⎢
⎢ ⎢ MRKG unmk ⎥ ⎥
⎥
⎢ DTRS ⎢⎢ ⎡ ⎤⎥
⎥ ⎥
⎢ ⎢ ⎡ ⎤ ⎥ ⎥
⎢ ⎢ interpers_reln_fr ⎥ ⎥
⎢ ⎢ ⎢    ⎥ ⎥
⎢
⎢ ⎢ ⎢ plural_fr ⎢
⎢ s ⎥
⎥ ⎥⎥
⎥ ⎥
⎥
⎢ ⎢
⎢SEM ⎢FRAMES  1 ⎢
SIT
⎥
⎥ , . . . ⎥⎥
⎢ ⎢ ⎣ ⎦ ⎥ ⎥
⎣ ⎣ ⎣ PL - ENTITY i SIDE _1 i ⎦⎦ ⎥
⎦
SIDE _2 j

Figure 5: Plural nominal as predicative reciprocal relation

4.10 Verb Pumping

Much work in recent construction grammatical literature has been devoted to ARGUMENT STRUCTURE
constructions, whereby a valence pattern associated with one class of verbs will welcome verbs with quite
different meanings; the descriptions have shown the ways in which the semantic requirements of the ‘guest’
word are bound to the semantic expectations of the ‘host’ context (Goldberg 1995, Boas 2003, see also:
Fillmore and Atkins 1992, Kay 2005).
The notion of a guest word in a host syntactic context is perhaps best illustrated by the valence patterns
associable with a particular set of verbs. There are some verbs that allow in their own right a valence of two
noun phrases indicating a transfer of possession or location: give, send, etc. This pattern serves as a host
in expressions such as I slid her the note or I kicked her the ball, it which the verb slide or kick indicates
the means (roughly) by which one gives someone something (Goldberg 1995). Similarly, a causative verb
like make takes in its normal complementation pattern an NP and a state-denoting predicate; placing other
verbs in place of make (wipe your face clean) results in one of a variety of so-called resultative constructions
(Goldberg 1995, Boas 2003).
A variant of this construction involves non-expression of the original verb’s direct object: shake the seeds
onto the ground, wring water from the towel, stir the milk into the batter. In this case the external arguments
of the original and ‘pumped’ verbs are coindexed, but the direct object is not obligatorily coidentified with
any valent of the original verb (the item which is shaken, wrung, or stirred, e.g. a container, a towel, a
combination of the milk and batter). This is what in FN would be called constructionally-licensed null
instantiation (CNI), and in particular CNI with indefinite interpretation. A sentence like I saw her shake
seeds onto the ground does not require a context in which the item being shaken has been established (and
it could simply be her hand). In some instances, the ‘original’ theme may be mentioned: shake the seeds

35
out of the container. We still analyze the downstairs verb’s direct object as unexpressed: so far as the
constructionally-licensed verb is concerned, no mention need be made of it.
Rough sketches of these two constructions are shown below. These constructions call for a constituent
that expresses the source, path, or goal of motion of the T HEME, indicated by the ‘SPG’ (Source/Path/Goal)
valent:

(63) {verb [verb sign1 ] }

Name Resultative_with_original_object
M a verb with valence: <Agent-NPi, Theme-NPj , SPG>.
The indicated action is performed on the Theme, resulting
in the Theme traveling as indicated by the SPG(s).
D1 a transitive verb, valence: < Agent-NPi , Theme-NPj >

(64) He {[mashed]} potatoes through a ricer.

(65) {verb [verb sign1 ] }

Name Resultative_without_original_object
M a verb with valence: <Agent-NPi, Theme-NPj , SPG>.
The indicated action is performed some item (NPk ),
which is unmentioned and has indefinite reference, result-
ing in the Theme traveling as indicated by the SPG(s).
D1 a transitive verb, valence: <Agent-NPi , Theme-NPk >

(66) She {[shook]} seeds onto the ground

4.11 The verb-way construction

The verb-way construction is similar to a pumping construction such as Caused-motion, but it operates on
units larger than single words. The construction is characterized by the presence of the noun phrase with a
possessive pronoun and the head way, which serves as direct object to a verb. The verb-plus-DO created is a
multiword verb of Motion. The selection of the type of verb distinguishes three kinds of constructions: the
most common has a verb that expresses an activity that enables the mover to make progress (squirm, push,
blast) as in She wriggled her way through the crack. Another option is for the verb to indicate an activity
roughly coterminous with motion (whistle, sing), as in He belched his way into the room. A separate option,
covered by a similar but distinct construction, is for the verb to serves mainly a support function with only
minimal (or no) semantics of its own, as in She made/wended her way to through the crowd. Nevertheless
the construction may play host to a number of verbs that are not canonically such, and often do not permit
easy classification into one of the three classes: George acrobated his way across involves a verb that may
well appear mostly or exclusively in this construction, at least for a period of time for a particular speaker.
As with other verb-pumping constructions, the guest LU loses its syntactic valence (except for the exter-
nal argument), but its core semantic valence is retained, with any missing arguments are understood indef-
initely: we know that you’re pushing against something if you push your way into a room, and if someone

36
elbowed her way to the front she pushed her elbow(s) into people. 33 Unlike inherently Motion-evoking
LUs, signs licensed by this construction require a S OUCE, PATH, or G OAL expression: she moved quickly is
fine, but *she elbowed her way is not.
The way-headed NP is peculiar in that it may be modified by an expression which semantically takes as
its argument either the Theme/Actor or the Path taken: walked his weary way back, weaved his precarious
way back. We state this in prose here, though the fact awaits a more precise desciption:

(67) {Motion verb [Verb ] [PossNP ] }

Name verb-way
M Verb, evokes the Motion frame. Requires at least
one S OURCE , PATH, or G OAL-related argument.
D1 A verb with at least an ACTOR argument; any
other arguments are suppressed and existentially
interpreted.
D2 An NP, headed by way and with a possessive pro-
noun coindexed to D1’s external argument; able
to be modified by ACTOR-modifying or PATH-
modifying expressions.
Interpretation the meaning of D1 (the verb) is incorporated into
the Motion frame as a M ANNER or M EANS of
motion. This is clear in many cases but the dis-
tinction is not always clear.
We find that such a structure as [poss + N], where the possessive pronoun is coreferential with the subject
of the clause containing the NP, is a component of a great many idioms and other constructions that make use
of (1) an anaphoric possessive pronoun with a (2) lexically specific head noun. These are parts of expressions
that are typically represented with one’s in idiom dictionaries: to have one’s way (’to be able to do what one
wants’), to blow one’s horn (’to advertise one’s virtues’), to take one’s time (’to be slow at accomplishing
something’), etc. Verbs such as blow and crane as in blow one’s nose and crane one’s neck also call for NPs
with a possessor coindexed with another of its arguments.
We can capture the properties common to all these idioms with an Anaphoric_possessed_noun
(APN) construction, which specifies an NP that consists of a head nominal with an anaphoric possessor,
making it available for participation in certain idioms, or as arguments of certain verbs. The construction
must also allow information about the required antecedent (person, number, gender) available at NP level
for coindexation with the appropriate argument. In most cases, the antecedent of the pronominal constituent
most is the subject of the predicate in which the NP occurs. For instance, the one in one’s nose must have
as antecedent the subject of [verb] one’s nose where [verb] may be blow, wriggle, and so on. There are
some idioms in which the antecedent is a non-subject, e.g. give him his due and put him in his place. The
identity of the head noun is determined by the context (construction, predicator) in which it finds itself.
Crane in crane one’s neck specifies that its first complement is licensed by APN and that the head noun is
neck; verb-way specifies way as the head.
33 The interpretation of the unmentioned direct object is very similar to the indefinite null instantiation (or ‘complementation,’) of

Fillmore 1986, which accounts for the interpretation of I saw someone reading on the train, in which there is an understood text which
is read, but it need never be mentioned. At the same time, it seems at best uncooperative to use a transitive verb like elbow or push in
the verb-way construction with no prior notion that there exist items to be pushed, elbowed, etc. Detailed corpus work should reveal
whether the anaphoric properties of verb-way line up with those of other INI lexical items or constructions, or if a separate category
is called for.

37
 Not all possessed NPs are licensed by this construction. The expressions in (68) involve only the familiar
possession and NP constructions, even when by coincidence the possessor is anaphoric to a core argument
of the verb:

(68) a. She found his book.

b. She found her book.
c. He put the book on its shelf.
There are many expressions (covered by the Experience_bodily_harm frame in FN) which are am-
biguous between canonical situations of possession and idiomatic uses of APN. (69) illustrates two sentences
which may be understood as non-intentional injury, in which case the body-part possessor must agree with
the subject:

(69) a. I broke my finger.

b. I bit my cheek.
Some idioms allow the noun itself to be modified (see (70a)) and the reflexive possessive to have the
emphatic extension in (70b):

(70) a. take your sweet time

b. toot his own horn
These may require specification of separate construct types which specifically allow or prohibit such flexi-
bility.
Individual constructions also specify the variety of modifiction permitted, as in (71):

(71) a. plodded his weary way

b. edged his precarious way
c.*took his weary time
See additional examples of constructions that make use of APN in (72):

(72) bear one’s cross, blink one’s eyes, put one’s mind to it, keep one’s fingers crossed, mind one’s own
business, have one’s way (sg.), have one’s ways (pl.)

4.12 Adjectives as Nominals

We recognize three different constructional processes by which NPs are created by joining the definite arti-
cle to an adjectival expression. We refer to these as Adjective-as-Nominal.Human [shown in (73a)],
Adjective-as-Nominal.Anaphoric [(73b)], and Adjective-as-Nominal.Abstract [(73c)]:

(73) a. Examine the plight of the very poor.

b. Their outfits range from the flamboyant to the functional.

38
 c. The unimaginable happened.
Each subconstruction has its own semantic and morphosyntactic specification. The Human version li-
censes a plural mother which denotes generically people with the given property. The Anaphoric version
semantically resembles ellipsis of a head noun, and nearly always appears in contrast/comparison contexts.
The Abstract version (73c) is morphologically singular and denotes an unspecified entity with the adjec-
tive’s property, or such entities in general (so far as we can ascertain, the interpretation as a specific entity is
largely limited to descriptions of event-occurrence, e.g., with happen, occur). This construction commonly
participates in constructions that juxtapose and contrast properties: we must separate the linguistic from the
paralinguistic, the practical includes the technological.

4.12.1 AdjectiveAsNom.Human

(74) {NP.plural [the ] [AP ] }

Name Adjective-as-nominal.Human
M NP, plural, generic reference.
D1 the word the.
D2 an AP describing a property of people.

(75) a. She is a friend to {NP [the the] [AP poor].

b. {[The] [hard of hearing]} are sure to appreciate this new device.

Although it is frequently said that the word poor in we need to provide housing for the poor is really
being ‘used as a noun’ the fact that the adjectives in such constructions may be themselves modified suggests
that the combination really needs to be analyzed as an NP with the form the followed by an AP (76). The
construction is affiliated with generic plural human NPs built from conjunctions of adjectives: rich and poor
(alike). For now we analyze only the construction with the. The construction specification is given below.

(76)
The elderly and {[the] [extremely young]} are most at risk.

The AP in each construct is licensed by the expected set of adjective-related constructions. The first
element is simply the definite determiner, semantically inert (which means that its only role is to be formally
a part of this construction: since it is an obligatory element, it has no way of contributing its own meaning).
There are uses of this word in generic expressions, but in those other uses, the resulting phrase could also be
an instance of an ordinary – singular or plural – function of definite determiners in NPs.
Adjectives denoting nationalities can participate in this construction, but only if the nationality adjective
ends in a sibilant consonant. That is, we cannot speak of members of the nationalities in general with
the phrases the German or the Italian (we have to say Germans or Italians), but the generic plural human
interpretation is freely given in the case of the French, the Chinese, the Irish, the Senegalese. This might be
one of the rare phenomena in grammar where a phonological feature constrains a grammatical pattern.
It is quite possible that this pattern is only one option in a more general construction that also describes
NPs composed of a genitive NP and an AP: England’s poor, the state’s persistently unemployed. Other

39
 the_AP_human-cxt ⇒
⎡ ⎡ ⎤⎤
FORM the, X
⎢ ⎢ ⎡  ⎤ ⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ noun ⎥⎥
⎢ ⎢ SYN ⎢CAT ⎥ ⎥⎥
⎢ ⎢ ⎣ NUM pl ⎦ ⎥⎥
⎢ ⎢ ⎥⎥
⎢ ⎢ ⎥⎥
⎢ MTR ⎢ MRKG det ⎥
⎢
⎢ ⎢ ⎡ ⎤⎥
⎥⎥
⎢ ⎢ INDEX i ⎥⎥
⎥
⎢ ⎢ ⎢    ⎥⎥⎥
⎢ ⎢ ⎢ ⎥
⎥⎥⎥
⎢ ⎢ SEM
⊕ 1 ⎦⎥
⎣FRAMES generic_fr human_fr ⎦
⎢ ⎣ , ⎥
⎢ GENERIC - OBJ i ENTITY i ⎥
⎢ ⎥
⎢ ⎡ ⎤ ⎥
⎢ FORM X
⎥
⎢   ⎥
⎢ ⎢ ⎥ ⎥
⎢ ⎢ ⎥ ⎥
⎢ ⎢ SYN CAT adj
⎥ ⎥
⎢ ⎢ VAL   ⎥ ⎥
⎢ DTRS ⎢ ⎥ ⎥
⎢ ⎢ ⎡   ⎤⎥ ⎥
⎢ ⎢ ⎥ ⎥
⎢ ⎢ property_fr ⎥ ⎥
⎢ ⎣ SEM ⎣FRAMES 1 list ⎦⎦ ⎥
⎣ ENTITY i
⎦

Figure 6: A the+AP construction for plural, generic, human-denoting NPs

definite determiners (this/these, that/those) are impossible in this pattern, so the more general construction
must find a common feature of both the and possession.

4.12.2 Adjective-as-nominal.Anaphoric
Noun-head deletion is familiar with number and superlative expressions – we pick the best, I’ll take two –
but it is also possible with the and a plain adjective phrase. This is a type of identity-of-sense anaphora: the
‘missing’ nominal is always recoverable in the preceding context (or rarely in the non-linguistic context),
and the AP is commonly in contrast with another AP modifying the antecedent noun:

(77) {NP [the ] [AP ] }

Name Adjective-as-nominal.Anaphoric
M NP, meaning is the result of D2 modifying an
available-from-context nominal, often in contrast with
the NP that provides that nominal. Singular/plural is
determined by the antecedent noun.
D1 the
D2 AP
Context Contains something construable as a binary operator
(commonly a conjunction) comparing two kinds of
things describable with the same noun.

(78) a. I prefer the short word to {NP [the the] [AP long]}.

40
 b. For all you can tell, the new situation is no better than {[the] [old]}.

Though we cannot explore the issue fully, this construction seems distinct from the types of nominal
deletion illustrated in (79). Though semantically close, the pattern with the is much narrower in use. It
prefers to find its antecedent very close by (and always from the linguistic, rather than extra-linguistic,
context), and is most comfortable in a non-colloquial register.

(79) a. I like Kim’s.

b. I’ll take three.

c. I bought the cheapest.

4.12.3 AdjAsNom.Abstract
The third subconstruction licenses NPs, composed of the plus an AP, which refer either to an unspecified
entity with the properties associated with the adjective, or to such entities in general: we have to face the
inevitable, distinguish the abstract and the concrete, they were witness to the supernatural.

(80) {NP [the ] [AP ] }

Name Adjective-as-nominal.Abstract
M NP, singular. Refers to an unspecified or generic entity
with the properties of D2.
D1 the
D2 AP

(81) a. {NP [the The] [AP seemingly impossible]} was somehow accomplished.
b. The artist shows us {[the] [familiar]} as if it were new.

4.13 Measurement Expressions

4.13.1 Measurement_with_adjective.Predicate
The combination of a measurement expression (a quantified unit, e.g. three feet) with a limited set of adjec-
tives may be used as a predicate to indicate the value of some item’s height, weight, etc. The adjectives that
participate in this pattern are tall, wide, high, thick, long, and deep, taking measurements of linear extent,
and old, taking measurements of age:

(82) {Predicate [meas.expr sign1 ] [adj sign2 ] }

41
 Name Measurement_with_adjective.Predicate
M AP, used predicatively (or in places where such con-
structs go, e.g. clause-initial modifier). Indicates a par-
ticular value (D1) on a scale given by D2. Possible
scales are linear measurement (and extensions to tem-
poral extent) and age.
D1 An expression built of a number and a unit, with appro-
priate number agreement.
D2 One of a restricted list of adjectives: high, tall, thick,
wide, deep, long, old.
X A supporting item. Be is possible for all adjectives.
Stand, measure, extend, etc. are specific to particular
scales. In the absence of a lexical supporter a support
construction is possible (e.g. Twenty miles long, it is the
longest such trail in the state).

(83) a. It was six feet long.

b. The structure is two hundred years old.
X indicates the construct-external contextual element that the construction calls for. In this case, the
construct appears with a support verb, or a support construction that specifies predication of the expression
over an NP without an overt lexical item like be, e.g. over seven feet tall, he stood well above his companions.

4.13.2 Measurement_with_adjective.Modifier
Examples with modified noun heads are found in (84). The construct in question is the measurement phrase.
Like the predicative version, it contains a measured unit and adjective, but shows different agreement prop-
erties (singular units only), and shows up differently in written text, namely with hyphenation:

(84) a. 62-year-old man

b. twelve-inch-thick wall
c. two-foot-long pole

(85) {Modifier [meas.expr sign1 ] [adj sign2 ] }

Name Measurement_with_adjective.Modifier
M AP, used attributively. Indicates a particular value
(D1) on a scale given by D2. Possible scales are lin-
ear measurement (and extensions to temporal extent)
and age. All word boundaries are generally indicated
with a hyphen.
D1 An expression built of a number and a singular unit.
D2 One of a list of adjectives: high, tall, thick, wide,
deep, long, old.
X A head noun which is modified by the measurement
construct.

42
(86) a. a six-foot-long rope
b. that two-year-old building
The fact that the unit in D1 is singular is probably best understood as an instance of the singularity of the
modifier noun in compounds (bookstore, pantleg, scissor-sharpener). A similar construction, which does
not include the scale-denoting adjective, is more properly considered a part of a compound: three-yard rope,
four-hour play. In this case the scale is left to interpretation based on the context or world knowledge. We
do not engage in a complete analysis, but point out that this adjective-less construction cannot simply be
the construction above with the adjective left out, as attested by the grammaticality of twenty-dollar haircut
and fifteen-pound turkey as opposed to *twenty-dollar-expensive haircut and *fifteen-pound-heavy turkey.
Conversely, although we can speak of a three-year-old child, there is no *three-year child.
Finally, we admit that these constructions present the most common case: a singular unit in attributive
position, and a plural unit in predicative position. Nevertheless, exceptions to this pattern are not hard to
find, even in the BNC:

(87) a. The first uses large, two metres tall puppets which can be easily seen by large audiences.
b. Within minutes of being trapped behind a 20-feet thick wall of coal, steel, and rubble completely
filling the 12-feet high and 16-feet wide tunnel, they were in contact with colleagues on the oppo-
site side using the undamaged Tannoy system which runs along the wall of the roadway.
c. He was about six foot tall and well-built.
d. Her replica was six foot high, made of spun meringue and sponge cake.
Given the above observations, some analysts might prefer to propose a single measurement-with-adjective
construction and add the comment that the choice of grammatical number exhibits certain tendencies distin-
guishing its use in attributive vs. predicative contexts. We prefer to claim that there are the two well-defined
and thoroughly productive constructions just described, while recognizing that speakers are able to express
their semantics by other means.

4.14 Rate phrases

The Rate construction and its several subconstructions provide means to express ratios. Here we consider
only those expressions with a(n) in the second NP:

(88) {Rate [numerator sign1 ] [denominator sign2 ] }

Name Rate
M NP
D1 The ‘numerator.’ Typically a quantified NP, head:
measurement unit; but also multiplicatives like
once, several times, etc.
D2 The ‘denominator.’ Typically an indefinite singu-
lar NP head: measurement unit; but also certain
dedicated words like each, apiece
Interpretation A ratio is built from num/denom; certain combi-
nations of units are recognized as their own type
of measure (e.g. frequency, mileage, cost, speed)

43
(89) a. {[numerator Two hundred pounds] [denominator a week]}, some of them can earn.
b. Milk was delivered {[twice] [a day]}, and bread daily.

This construction licenses a sign that denotes a ratio, with a particular internal syntax not seen elsewhere
in the grammar. The Mother of the Rate construction is an NP, as are both of the daughter signs. The first
daughter, the ‘Numerator’, is a quantified NP and the second daughter, the ‘Denominator’, is an indefinite
NP. Each NP refers to certain units, and the meaning of the mother is a Ratio formed by the semantic
representations of the two constituent NPs. In some cases these are already-recognized concepts: if the
units are recurrences vs. time, the interpretation of the whole is ‘frequency’; if they are distance in miles
vs. gallon of fuel, the mother’s category is ‘mileage’; if the two constituents represent monetary value vs.
unit of weight, the mother’s category is ‘cost per unit weight’. In this way it is possible to build up, from a
single abstract construction, interpretations of twenty times a day, twenty miles a gallon, and twenty dollars
an ounce. The construction also provides the basis for interpreting ad hoc ratios like four cookies a student.
Some expressions that convey a ratio concept are not instances of the construction as stated here: in I
earn twenty dollars every day (instead of a day) is a formation based on regular syntax, as shown by the fact
that the ‘denominator’ phrase would serve the same function if it had been every three or four days, and it
can be placed in clause-initial position (Every day I earn twenty dollars). The meaning of these expressions
differs from Rate-licensed ones: a car that traveled fifty miles every hour traveled for at least one hour, but
a car going fifty miles an hour may have only moved for a minute. These facts all lead to the conclusion that
formulations with { [twenty dollars] [a day] } need to be licensed separately.
It seems likely that the formation of the denominator with per rather than a is a separate but related
construction (90):

(90) a. Thus the athlete’s heart will beat 13 million fewer times per year.
b. That creature must have been travelling at 60 miles per hour.
Here, a few points of divergence may be mentioned. First, per is much more easily repeated than a, and
while the two variants may be combined, a must appears first, as in (91):

(91) a. one per customer per day

b. ??one a customer a day
c. one a customer per day
d. *one per customer a day
Second, a seems more suited to more familiar types of ratio, as in (92):

(92) After you hand out all the food, there should be three cookies per/??a child.
The same expression may fit both familiar and less familiar contexts. Both two dollars an inch and two
dollars per inch may appear in a context where the dollar amount is spent, as in the cost of printing a poster,
but the former is more awkward if the dollar amount represents additional costs incurred as a result of
parking one’s vehicle successively farther from the sidewalk (beyond one foot, tickets go up by an additional
two dollars per/??an inch). Third, constructs with per are syntactically more flexible, both in phrase-internal
syntax as well as external distribution. The denominator may be more easily modified (as in (93a)–(93b)),

44
and because per does not specify number, more complex ratios are permitted (as shown by (93d)). The
judgments on complex denominators depend on the precise nature of the complex NP, and seem to be quite
variable between speakers.

(93) a. two dollars per/??an inch purchased

b. twenty dollars per/?an hour of labor
c. twenty dollars per/?*a piece of furniture
d. twenty dollars per/*a three hours
Per-phrases are also potential nominal modifiers, not a possibility for the indefinite determiner, as illustrated
in (94).

(94)
I want to know the vehicle’s mileage per/*a gallon.

Further constructions are necessary to license expressions only used to express particular ratios, as with
to the for mileage and perhaps other sorts of ratios: My aeroplane, which also comes from the States, only
does eight miles to the gallon. It should also be noted that the characterization of the numerator and de-
nominator as expressions of some sort of measurement unit is not specified by the Rate construction, and
rather must be provided by the interpretation component of the constructions by which they themselves are
licensed.
An SBCG representation of the construction should capture at least the facts that the sign licensed is a
Ratio-denoting NP, made up of two NPs with particular shapes (quantified, with a). This is represented in
Figure 7. A Rate constructs two daughters correspond to the numerator and denominator. The first is a
quantified unit, and the second is an indefinite singular NP with a/an. The current formulation leaves open
the possibility that the two units are identical, e.g. twenty miles per mile.
A more complete formulation would include a way to understand that certain instances of ratios are
familiar (speed, frequency, and so on), or the semantics of the grammar should provide for recognition of
compositionally-built (constructional) meanings that coincide with already-established (often lexical) ones.

4.15 Magnitude Qualifying Predeterminers

In this section we compare two syntactic patterns, one of the form {[X] [of Y]}, the other {[X][Y]}. The
former is the Partitive Construction, not dealt with in the current Constructicon, and the latter is what
we are calling the Magnitude-Qualifying Predeterminer Construction (MQP). Of interest is the
way in which the sets of expressions that can occur in the X and Y positions in these two patterns overlap. In
some cases, in other words, the only difference is in the presence of the preposition of, heading the so-called
oblique partitive (in Huddleston and Pullum 2002: 333) in the case of the Partitive.
Examples of the former pattern are a third of the members, all of the beer, 50% of the population, one
tenth of his salary; examples of the latter are half the number, twice the size, two thirds the distance, one
tenth his salary. In the partitive expressions, the meaning of the whole is taken to be a portion (up to totality)
of the substance or set of things indicated in the Y phrase; in the magnitude case, the meaning of the whole
is a quantity computed from the quantity represented by the Y phrase. In all cases the NP heading the Y
expression is definite and explicitly or implicitly includes the understanding of something being quantified.

45
 rate-cxt ⇒

⎡ ⎡   ⎤ ⎤
CAT noun
⎢ ⎢SYN ⎥ ⎥
⎢ ⎢  ⎥ ⎥
⎢ ⎢ ⎡
VAL
⎥ ⎥
⎢ ⎢
⎢ ⎡ ⎤ ⎤⎥ ⎥
⎢ MTR
⎢ ⎢  ratio_fr ⎥⎥
⎥
⎥
⎢ ⎢ ⎢ ⎢ ⎥ ⎥⎥ ⎥
⎢SEM ⎢ FRAMES i⎦ ⎥
⎢
⎢ ⎣ ⎣ ⎣NUMERATOR ⎦⎥
⎦
⎥
⎥
⎢ j ⎥
⎢ DENOMINATOR
⎥
⎢ ⎡ ⎤ ⎥
⎢ ⎡ ⎤ ⎥
⎢ ⎥
⎢ ⎡ ⎡ ⎤ ⎤⎢ CAT noun
⎥ ⎥
⎢ ⎢ ⎥
⎢ ⎢SYN ⎣VAL  ⎦ ⎥ ⎥
⎢
CAT noun
⎥⎢ ⎥ ⎥
⎢ ⎢
⎢SYN ⎢
 ⎦
⎥ ⎥⎢ an ⎥⎥
⎥
⎣ VAL ⎥⎢ ⎤⎥
MRKG
⎢ ⎢ ⎡ ⎥ ⎥
⎢ ⎢ ⎥⎢ ⎥ ⎥
⎢ DTRS ⎢ MRKG num ,⎢ ⎥ ⎥
⎥
INDEX j
⎢ ⎢  ⎥⎢ ⎢ ⎡ ⎤ ⎥
⎢ ⎢ ⎥⎢ ⎢  indef_sg_fr   ⎥
⎥⎥ ⎥
⎢
⎢ ⎣SEM INDEX i ⎦⎢⎢SEM ⎢ ⎥
⎥⎥ ⎥
⎥
⎢ FRAMES  [unit_fr ], . . .  ⎢ ⎢FRAMES ⎢⎣ BOUND _ VAR
⎥ unit_fr
j⎦, ⎥
, . . . ⎦⎥ ⎥
⎣ ⎣ ⎣ UNIT j ⎦ ⎦
...

Figure 7: SBCG representation of Rate

Thus, 50% of the population is understood as referring to the population of some specific area or community,
explicitly expressible as, say, 50% of the population of Guam.
A number of pairings of X-expressions and Y-expressions fit only one of these constructions. In (95a),
for example, only the partitive interpretation is possible:
(95) a. 50% of the members resigned in protest. (*50% the members)
b. That’s twice the number we had the first time. (*twice of the number).
While ordinary fraction phrases can occur in the MQP construction, the equivalents expressed as percentages
cannot; and the plural noun members (as opposed to the noun membership) cannot designate the number of
members of an organization (cf. We now have twice the membership). In (95b), only the MQP construction
is represented: the word twice is a multiple, and therefore cannot represent a portion, and the word number
expresses a quantity but not directly the number of any set of measurables.
The partitive construction selects, in NPs designating the wholes, mass nouns like beer (half of the
beer), or plural nouns like members (two thirds of the members); the MQP construction, however, selects
NPs which indicate some measure or attribute that can be expressed quantitatively. Examples of nouns that
have this as their main sense are size, rate, frequency, height, age, cost, and many others.
Overlaps are possible because fraction phrases meet the conditions of both constructions (phrases ex-
pressing multiples obviously do not), and because some nouns incorporate both the concept of quantity and
an understanding of what it is that is being measured. Nouns that incorporate both the notion of magnitude
and information about the nature of the set or substance being measured include population (the number of
people living in a place), salary (the amount of money one earns per time unit), and others.
Similarly, the (a) examples in (96) and (97) refer to portions of some entity or set, while the (b) examples
refer to the quantities involved:

46
(96) a. Half of his salary goes to rent.
b. At the peak of my career I only earned half his salary.

(97) a. Two thirds of the population of Malawi live in poverty.

b. Hungary has two thirds the population of Malawi.

The correct understanding of these two constructions might not require the notion of selecting lexical
types in the Y position, but of imposing one or the other interpretation on it, in cases where the lexical classes
seem different from what is expected. Thus, instances of MQP sometimes seem to ‘coerce’ a magnitude
meaning onto a noun which, independently of the construction, refers to a substance. In (98a) the reference
is to an already determined supply of water, whereas in (98b) the reference is to a measurable quantity of
water:
(98) a. A third of the water leaked out.
b. My garden now needs only a third the water.
In the other direction, sometimes parameter-naming nouns can be coerced to refer to a substance, as in (99a)
and (99b):
(99) a. Half of the distance to the cottage is a muddy mess.
b. About a third of his weight is sheer muscle.
(100) displays the schematic representation and constructional specification of MQP:

(100) {Magnitude [Multiple sign1 ] [def.NP sign2 ] }

Name Magnitude-qualified entity

M A magnitude (as opposed to a portion).
D1 A multiplier (twice, four times) or fraction (two
thirds, half).
D2 A definite NP (superficially) denoting a measured
or measurable substance or (abstract) entity.
Interpretation A measured or measurable quantity of an entity.

5 Conclusion
This chapter has reported on the results of a year-long project to start a database of construction descrip-
tions and constructionally annotated sentences, built on the principles, and using the tools, of an ongoing
lexicography project, FrameNet. There were numerous reasons for trying to articulate a lexicon with a con-
structicon: serious work in lexical description was unable to escape the need to appeal to features of grammar
that go beyond the basic structures that define ordinary valence satisfaction; and the semantic notions that
were the central motivation for building a frame-based lexicon figure independently in accounting for the
meaning contributions of many of the constructions.

47
 The new project set out to analyze and exemplify a selection of English grammatical constructions in a
way that could contribute to the development of any full-fledged grammar of English, while having specifi-
cally in mind the demands and formalisms of SBCGȮur descriptions of the selected constructions and anno-
tations of expressions licensed by them, in this paper and on the FN website, represent this effort. In some
cases we have presented tentative SBCG analyses to highlight the compatibility with formalisms represented
elsewhere in this volume; but in many cases we felt that describing and illustrating the properties of individ-
ual constructions could be useful even when the level of analysis that would allow their incorporation into a
complete grammar was not yet achieved.
One difficulty any such project faces is the order in which to tackle the constructions of a language. Pro-
ceeding from the most frequent (or ‘importantâĂŹ in some sense) constructions is not practical. That would
mean figuring out in advance which constructions are most frequent or most important, and that also is not
easy. There is no preexisting list of constructions which could be ordered by frequency, and many construc-
tions are largely covert and cannot be easily recognized in running text or counted. Alternatively, choosing
a representative text and annotating the constructions in it as they appear would require devoting time both
to many constructions which are theoretically uninteresting (Head-complement, Modification, and
so on) in the sense that nothing the FN team could discover would be a genuine advance in grammatical
knowledge, and to constructions that are so rare that their properties could be difficult to explore, or so sub-
tle that their properties would be difficult to discover. If this effort was to be limited to selected interesting
(and findable) constructions, there is no advantage in staying with particular texts rather than taking on the
language as a whole.
Our cherry-picking approach had several advantages. The constructions we selected were by and large
relatively easy to locate in tagged, unparsed corpora, and that allowed us to focus less on search methods
and more on analytic and representational issues. We also selected several constructions treated prominently
in the literature.34 Furthermore, we hoped that our annotation methods would be able to capture the key
features of constructions without necessarily committing us to difficult decisions that would have to be
made in writing a full grammar. At the same time we feel it is flexible enough to handle constructions of any
level of complexity.
The move from lexical to constructional analysis, in the history of the FN project did not happen merely
because some members of the team were interested in both lexicon and grammar; it was necessitated by the
fact that we had begun to use FN analyses to represent the meaning structures of full texts, and it became
increasingly obvious that lexical analysis alone is not sufficient for such a task. The work of extending the
software tools originally designed for lexicon building to an analysis of syntactic constructions was facili-
tated by certain deep commonalities between lexical and grammatical structures in a theory of construction
grammar committed to a locality principle. Just as it had already become necessary in the course of lexical
analysis to look for the means of describing the inner structure of lexical units (e.g., with compounding and
other word-forming processes) in order to account for regularities between the components and the seman-
tic and combinatory properties of the resulting whole, so too has it been necessary to characterize both the
component parts and the semantics and combinatorics of given types of constructs. It is our hope that the
continuing development of the FN lexicon and constructicon will both inform and be informed by grammati-
cal and semantic theory, resulting in a unified resource of benefit to any endeavor to understand the meanings
of sentences and texts, and the linguistic structures that constitute them.
34 Some constructions, such as the calendar expressions detailed in Fillmore 2002 and let alone (Fillmore et al. 1988), we do not

mention here because it would be in large part a repetition of those works.

48
References
Allerton, D. J. 1982. Valency and the English Verb.
Atkins, S., Fillmore, C. J. and Johnson, C. R. 2003. Lexicographic relevance: selecting information from
corpus evidence. International Journal of Lexicography 16.3, 251–280.
Boas, H. C. 2003. A constructional approach to resultatives. Stanford: CSLI Publications.
Bolinger, D. 1967. Adjectives in English: Attribution and predication. Lingua 18, 1–34.
Bresnan, J. 1973. Syntax of the Comparative Clause Construction in English. Linguistic Inquiry IV(3), 275–
343.
Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Coppock, E. 2008. The Logical and Empirical Foundations of Baker’s Paradox. Ph. D.thesis, Stanford Uni-
versity.
Emons, R. 1974. Valenzen englischer Pradikatsverben. Tübingen: Niemeyer.
Emons, R. 1978. Valenzgrammatik für das Englische. Eine Einführung.. Tübingen: Niemeyer.
Fillmore, C. J. 1986. Pragmatically controlled zero anaphora. In Proceedings of the 12th annual meeting of
the Berkeley Linguistics Society.
Fillmore, C. J. 1999. Inversion and constructional inheritance. In G. Webelhuth, J.-P. Koenig and A. Kathol
(eds.), Lexical and constructional aspects of linguistic explanation, Stanford: CSLI.
Fillmore, C. J. 2002. Mini-grammars of some time-when expressions in English. In Complex Sentences
in Grammar and Discourse. Essays in Honor of Sandra A. Thompson, Amsterdam/Philadelphia: John
Benjamins Publishing Company.
Fillmore, C. J. 2003. Valency and Semantic Roles: the Concept of Deep Structure Case. In V. Agel, L. M.
Eichinger, H.-W. Eroms, P. Hellwig, H. J. Heringer and H. Lobin (eds.), Dependency and Valency: An
International Handbook of Contemporary Research, pp 457–75, Walter de Gruyter.
Fillmore, C. J. 2006. Articulating Lexicon and Constructicon. Presentation at the Fourth International Con-
ference on Construction Grammar, Tokyo, Japan.
Fillmore, C. J. 2007. Valency Issues in FrameNet. In T. Herbst and K. Götz-Votteler (eds.), Valency: Theo-
retical, Descriptive and Cognitive Issues, pp 129–160, Mouton de Gruyter, Berlin and New York.
Fillmore, C. J. 2009. A Valency Dictionary of English. International Journal of Lexicography 22(1), 55–85.
Fillmore, C. J. and Atkins, B. 1992. Towards a frame-based lexicon: The semantics of RISK and its neigh-
bors. In A. Lehrer and E. Kittay (eds.), Frames, Fields and Contrasts: New Essays in Semantics and
Lexical Organization, pp 75–102, Hillsdale: Lawrence Erlbaum Associates.
Fillmore, C. J. and Baker, C. F. 2009. A Frames Approach to Semantic Description. In B. Heine and H. Nar-
rog (eds.), The Oxford Handbook of Linguistic Analysis, Oxford: Oxford University Press.
Fillmore, C. J., Johnson, C. R. and Petruck, M. R. 2003. Background to FrameNet. International Journal of
Lexicography 16.3, 235–250.
Fillmore, C. J., Kay, P. and O’Connor, M. C. 1988. Regularity and Idiomaticity in Grammatical Construc-
tions: The Case of ‘Let Alone’. Language 64, 501–538.
Ginzburg, J. and Sag, I. A. 2000. Interrogative Investigations: the Form, Meaning, and Use of English
Interrogatives. Stanford: CSLI Publications.
Goldberg, A. E. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago:
Chicago University Press.
Goldberg, A. E. 2006. Constructions at Work: The Nature of Generalization in Language. Oxford: Oxford
University Press.
Green, G. M. 1985. The description of inversions in generalized phrase structure grammar. In Proceedings
of the 11th Annual Meeting of the Berkeley Linguistics Society.

49
Helbig, G. and Schenkel, W. 1973. Wörterbuch zur Valenz und Distribution deutscher Verben. Tübingen:
Max Niemeyer Verlag.
Herbst, T. 1983. Untersuchungen zur Valenz englischer Adjektive und ihrer Nominalisierungen. Tübingen:
Gunter Narr.
Herbst, T. 1987. A Proposal for a Valency Dictionary of English. In R. F. Ilson (ed.), A Spectrum of Lexicog-
raphy, Amsterdam and Philadelphia: Benjamins.
Herbst, T., Heath, D., Roe, I. F. and Götz, D. (eds.). 2004. A Valency Dictionary of English: A Corpus-
Based Anaysis of the Complementation Patterns of English Verbs, Nouns and Adjectives. Berlin/New
York: Mouton de Gruyter.
Horn, L. 1989. A Natural History of Negation. Chicago: Chicago University Press.
Huddleston, R. and Pullum, G. K. (eds.). 2002. The Cambridge Grammar of the English Language. Cam-
bridge: Cambridge University Press.
Israel, M. 2002. Minimizers, Maximizers, and the Rhetoric of Scalar Reasoning. Journal of Semantics 18(4),
297–331.
Kay, P. 2005. Argument Structure Constructions and the Argument-Adjunct Distinction. In M. Fried and
H. C. Boas (eds.), Grammatical Constructions: Back to the Roots, Amsterdam: John Benjamins Publish-
ing Company.
Kay, P. to appear. The Limits of Construction Grammar. In G. Trousdale and T. Hoffmann (eds.), The Oxford
Handbook of Construction Grammar, Oxford: Oxford University Press.
Kay, P. and Fillmore, C. J. 1999. Grammatical Constructions and Linguistic Generalizations: The What’s X
Doing Y? Construction. Language 75, 1–33.
Kay, P. and Sag, I. A. this volume. Discontinuous Dependencies and Complex Determiners. In Sign-based
Construction Grammar, Stanford: CSLI Publications.
Lyons, J. 1977. Semantics. Londong and New York: Cambridge University Press.
McCawley, J. D. 1988. The Syntactic Phenomena of English, vols I and II. Chicago: The University of
Chicago Press.
Mel’čuk, I. 1996. Lexical Functions in Lexicography and Natural Language Processing, Chapter Lexi-
cal Functions: A Tool for the Description of Lexical Relations in the Lexicon, pp 37–102. Amster-
dam/Philadelphia: Benjamins.
Michaelis, L. 2005. Entity and Event Coercion in a Symbolic Theory of Syntax. In J.-O. Östman and
M. Fried (eds.), Construction Grammars: Cognitive grounding and theoretical extensions, Chapter 2,
Amsterdam/Philadelphia: John Benjamins Publishing Comany.
Michaelis, L. A. forthcoming. Complementation by Construction. In The Proceedings of the Thirty-second
Annual Meeting of the Berkeley Linguistics Society: Parasession on Theoretical Approaches to Argument
Structure.
Newmeyer, F. J. 2000. Language Form and Language Function. Cambridge: MIT Press.
Pustejovsky, J. 1995. The Generative Lexicon. Cambridge, MA: The MIT Press.
Ruppenhofer, J. K., Ellsworth, M., Petruck, M. R. L. and Johnson, C. R. 2006. FrameNet: Theory and
Practice. Retrieved February 22th, 2006 from http://framenet.icsi.berkeley.edu/.
Sag, I. A. 2010. English Filler-Gap Constructions. Language 86.
Sag, I. A. this volume. Sign-Based Construction Grammar: An informal synopsis. In H. C. Boas and I. A.
Sag (eds.), Sign-based Construction Grammar, Stanford: CSLI Publications.
Sag, I. A., Wasow, T. and Bender, E. M. 2003. Syntactic Theory: a Formal Introduction. Stanford: CSLI
Publications, second edition.
Tesnière, L. 1959. Éléments de syntaxe structurale. Paris: Éditions Klincksieck.
Van Eynde, F. 2007. The Big Mess Construction. In S. Müller (ed.), Proceedings of the HPSG07 Conference.

50
Ziegeler, D. 2007. A word of caution on coercion. Journal of Pragmatics 39(5), 990–1028.
Zwicky, A. M. 1995. Exceptional degree markers: a puzzle in internal and external syntax. In OSU Working
Papers in Linguistics, pp 111–123.
Zwicky, A. M. and Pullum, G. K. 1983. Cliticization vs. Inflection: English N’T. Language 59(3), 502–512.

51