topics: |  cognitive-linguistics

A classic work of cognitive linguistics proposing that grammatical
structures themselves carry meaning, e.g. change in perspective. 
Based on her Ph.D. at Berkeley, under George Lakoff. 

Excerpts

From intro

Work in Construction Grammar includes, for example, Fillmore, Kay and
O'Connor's analysis of the let alone and the more, the merrier constructions
(1988), Brugman's analysis of have constructions (1988), Kay 's work on even
(1990), the "What, me worry?" construction of Lambrecht (1990), and
Sweetser's analysis of modal verbs (1990). Construction Grammar is also
developed in Fillmore (1985b, 1987, 1988, 1990), Fillmore & Kay (1993), Filip
(1993), Jurafsky (1992), Koenig (1993), and Michaelis (1993). The present
work owes its greatest debts to Lakoff's in-depth study of there
constructions (1984) and to Fillmore (1987), who suggested that the meaning
of an expression is arrived at by the superimposition of the meanings of open
class words with the meanings of grammatical elements.

A central thesis of this work is lhat basic sentences of English are
instances of constructions -- form-meaning correspondences that exist
independently of particular verbs. That is, it is argued that constructions
themselves carry meaning, independently of the words in the sentence. p.1

The notion construction has a time-honored place in linguistics. Traditional
grammarians have inevitably found it useful to refer to properties of
particular constructions. The existence of constructions in the grammar was
taken to be a self-evident fact that required little comment. In the early
stages of transformational grammar (Chomsky 1957, 1965), constructions
retained their central role, construction-specific rules and constraints
being the norm. In the past two decades, however, the pretheoretical notion
of construction has come under attack. Syntactic constructions have been
claimed to be epiphenomenal, arising solely from the interaction of general
principles (Chomsky 1981, 1992); the rejection of constructions in favor of
such general principles is often assumed now to be the only way to capture
generalizations across patterns. p.1

It has long been recognized that differences in complement configuration are
often associated with differences in meaning. For example, the ditransitive
requires that its goal argument be animate, while the same is not true of
paraphrases with to:
	(1) a. I brought Pat a glass of water. (ditransitive)
	    b. I brought a glass of water to Pat.
	(2) a. *1 brought the table a glass of water. (ditransitive)
	    b. I brought a glass of water to the table. (Partee 1965: 60)
Fillmore (1968, fn. 49) noted that sentences such as the following differ in
meaning:
	(3) a. Bees are swarming in the garden.
	    b. The garden is swarming with bees.

Anderson (1971):
	    (4) a. I loaded the hay onto the truck.
	    	b. I loaded the truck with the hay.
While (4b) implies that the truck is entirely filled with hay (or at least
relevantly affected), no such implication exists in (4a). p.2

Wierzbicka (1988) contrasts (6a) and (6b):
	(6) a. 1 am afraid to cross the road.
	    b. 1 am afraid of crossing the road.
Only in (6a) is the speaker presumed to have some intention of crossing the
road. This difference in interpretation is argued to account for why (7a) is
infelicitous unless the falling is interpreted as somehow volitionally
intended:2
	(7) a. #1 am afraid to fall down.
	    b. 1 am afraid of falling down.

Argument structure constructions

I explore the idea that argument structure constructions are a special
subclass of constructions that provides the basic means of clausal expression
in a language.

1. Ditransitive			Subj V Obj Obj 2
  X CAUSES Y to RECEIVE Z	Pat faxed Bill the Jetter.

2. Caused Motion		Sub V Obj Obl
   X CAUSES Y to MOVE Z		Pat sneezed the napkin off the table.

3. Resultative	      		Subj V Obj Xcomp
   X CAUSES Y to BECOME Z	She kissed him unconscious.

4. Intrans. Motion		Subj V Obl
   X MOVES Y			The fly buzzed into the room.

5. Conative 			Subj V Oblat
   X DIRECTS ACTION at Y 	Sam kicked at Bill.

On a constructional approach to argument structure, systematic differences
in meaning between the same verb in different constructions are attributed
directly to the particular constructions. We will see that if we consider
various constructions on their own terms, interesting generalizations and
subtle semantic constraints emerge.

Brief introduction to construction grammar

The basic tenet of Construction Grammar as developed in Fillmore & Kay
1993, Fillmore, Kay & O'Connor 1988, Lakoff 1987, Brugman 1988, Lambrecht
1994, is that traditional constructions-i.e., form-meaning
correspondences-are the basic units of language.

In Construction Grammar, no strict division is assumed between the lexicon
and syntax. Lexical constructions and syntactic constructions differ in
internal complexity, and also in the extent to which phonological form is
specified, but both lexical and syntactic constructions are essentially the
same type of declaratively represented data structure: both pair form with
meaning.

Construction Grammar rejects the notion of a strict division between
semantics and pragmatics.

Construction Grammar is generative in the sense that it tries to account for
the infinite number of expressions that are allowed by the grammar while
attempting to account for the fact that an infinite number of other
expressions are ruled out or disallowed. 

Construction Grammar is not transformational. No underlying syntactic or
semantic forms are posited. Instead, Construction Grammar is a monostratal
theory of grammar like many other current theories, including Lexical
Functional Grammar (LFG) (Bresnan 1982), Role and Reference Grammar (Foley &
Van Valin 1984), GPSG (Gazdar et al. 1985), HPSG (Pollard & Sag 1987, 1994),
and Cognitive Grammar (Langacker 1987a, 1991). ... 

Contrast with lexicosemantic rule grammars

The recognition of subtle semantic differences between related syntactic
(subcategorization) frames has been growing, and there has also been
increasing focus on the fact that there appears to be a strong correlation
between the meanings of verbs and the syntactic frames they can occur in,
leading many researchers to speculate that in any given language the
syntactic subcategorization frames of a verb may be uniquely predictable from
the verb's lexical semantics (e.g., Levin 1985; Chomsky 1986; Carter 1988;
Levin & Rapoport 1988; Rappaport & Levin 1988; Pinker 1989; Gropen et
al 1989).

1.4 Advantages of the construction account

Implausible verb senses avoided

The constructional approach avoids the problem of positing implausible verb
senses to account for examples such as the following, in none of which does
the verb intuitively require the direct object complement:
	(8)  He sneezed the napkin off the table.

To account for (8), for example, a lexicosemantic theory would
have to say that sneeze, a parade example of an intransitive verb, actually
has a three-argument sense, 'X CAUSES Y to MOV Z by sneezing'.

	(9)  She baked him a cake. [3-arg bake: X intends to cause Y to HAVE Z]
	(10) Dan talked himself blue in the face.
	     	 	[sense of talk: 'X CAUSES Y to BECOME Z by talking']

1.4.2 Circularity Is Avoided

Another important advantage of the construction-based approach is that it
avoids a certain circularity of analysis resulting from the widespread claim
in current linguistic theories that syntax is a projection of lexical
requirements.

This claim is explicit in the Projection Principle of Government and Binding
Theory (GB) (Chomsky 1981), the Bijection Principle of Lexical Functional
Grammar (Bresnan 1982), and in all current accounts which attempt to predict
overt syntax from semantic roles or theta role arrays. In all of these
frameworks, it is the verb which is taken to be of central importance. That
is, it is assumed that the verb determines how many and which kinds of
complements will co-occur with it. In this way, the verb is analogized to the
predicate of formal logic, which has an inherent number of distinct
arguments. The verb is taken to be an n-place relation "waiting" for the
exactly correct type and number of arguments. But note, now, that an ordinary
verb such as kick can appear with at least eight distinct argument
structures:

	1. Pat kicked the wall.
	2. Pat kicked Bob black and blue.
	3. Pat kicked the football into the stadium.
	4. Pat kicked at the football.
	5. Pat kicked his foot against the chair.
	6. Pat kicked Bob the football.
	7. The horse kicks.
	8. Pat kicked his way out of the operating room.

Theories which assume that the verb directly determines particular complement
configurations are forced to claim that kick is a binary relation with agent
and patient arguments and therefore occurs with transitive syntax, except in
3 / 6.
circularity: That is, it is claimed that kick has an n-argument sense on the
basis of the fact that kick occurs with n complements; it is simultaneously
argued that kick occurs with n complements because it has an n-argument
sense.

[3, It is important to bear in mind that both semantic and pragmatic aspects
of grammatical form are relevant for determining synonymy, Only if two forms
have both the same semantics and the same pragmatics, they will be disallowed
by the Principle of No Synonymy of Grammatical Forms, This principle is
impossible to prove conclusively, since one would have to examine all f rms
in all languages to do so. 

6. "Meaning" is to be construed broadly enough to include contexts of use, as
well as traditional notions of semantics. That is, a construction is posited
when some aspect of the way it is conventionally used is not strictly
predictable. It would alternatively be possible to define constructions as
ordered triples of form, meaning, and context as is done by Zadrozny &
Manaster-Ramer 1993.]

Semantic parsimony

Levin (1985):
"there is evidence that when the verb slide is found in the double object
construction, ... its sense is not the purely physical transfer sense of
slide but rather a transfer of possession sense" (p. 35)...
"the goal argument of a change of possession verb must denote an entity
capable of ownership, but the goal argument of a change of location verb need
not," as illustrated by her examples:
      (17) a. She slid Susan/*the door the present.
      	   b. She slid the present to Susan/to the door.

Thus, 'slide-1', would constrain its goal to be animate, while the other,
'slide-2', would have no such constraint.  Would also need to stipulate that
'slide1' only occurs w the ditransitive.

Instead of positing both an additional sense of slide and a stipulation that
this sense can only occur in the ditransitive construction, we can attribute
the constraint that the goal must be animate directly to the construction.

I concur with Levin that the semantics of (and constraints on) the full
expressions are different whenever a verb occurs in a different construction.
But these differences need not be attributed to different verb senses; they
are more parsimoniously attributed to the constructions themselves.

Compositionality is preserved

Frege is generally acknowledged to have originally formulated the idea that
semantics need be compositional: the meaning of every expression in a
language must be a function of the meanings of its immediate constituents and
the syntactic rule used to combine them.

Montague stated the analogous condition that there must be a homomorphism
from syntax to semantics; that is, there must be a structure-preserving
mapping from syntax to semantics. Letting SEM be a function from syntax to
semantics, '+syn' a rule of syntactic composition, and '+sem' a rule of
semantic composition, the following is claimed to hold:

	(19) SEM(x +syn y) = SEM(x) +sem SEM(y)

The meaning of the expression is therefore taken to result from applying to
the meanings of the immediate constituents a semantic operation which
directly corresponds to the relevant syntactic operation.

Dowty (1979) observes that the claim is intended to imply that the relation
between syntactic expression and semantic representation is straightforward
and direct. That is, +syn, or syntactic composition, must be
straightforwardly related to +sem, or semantic composition.
Gazdar et al. (1985), also working within the Montague Grammar tradition: "We
assume that there exists a universal mapping from syntactic rules to semantic
translations .. . . We claim that the semantic type assigned to any lexical
item introduced in a rule .. . and the syntactic form of the rule itself are
sufficient to fully determine ... the form of the semantic translation rule"
(1985: 8-9).

[But] Because the rules of combination are so widely regarded as transparent,
it is easy to overlook the fact that there are any substantive rules at all.
Even Jackendoff, who in fact does recognize nonlexical meaning (cf. section
10.1.1), states in the introduction to his 1990 monograph Semantic
Structures: "It is widely assumed, and I will take for granted, that the
basic units out of which a sentential concept is constructed are the concepts
expressed by the words in the sentence, that is, lexical concepts"
(Jackendoff 1990a: 9).

[examples where purely lexical interpretations have trouble, e.g. in "Way
constructions: ]

(27) a. Pat fought her way into the room.
     b. Volcanic material blasted its way to the surface.
     c. The hikers clawed their way to the top.

By recognizing the existence of contentful constructions, we can save
compositionality in a weakened form: the meaning of an expression is the
result of integrating the meanings of the lexical items into the meanings of
constructions.12 In this way, we do not need to claim that the syntax and
semantics of the clause is projected exclusively from the specifications of
the main verb.

[FN12. However, extralinguistic knowledge is undoubtedly required as well in
order to arrive at a full interpretation of an expression in context;
cf. Lakoff 1977, Langacker 1987a.]

Psycholinguistic evidence: sentence processing

Certain psycholinguistic findings reported by Carlson and Tanenhaus (1988)
suggest that uses of the same "core meaning" of a verb in different syntactic
frames do not show the same processing effects that cases of real lexical
ambiguity do. For examp~e, notice that set truly has two different senses:
	(28) a. Bill set the alarm clock onto the shelf.
	     b. Bill set the alarm clock for six.

Load, on the other hand, although it can readily appear in the alternate
constructions in (29), according to Carlson and Tanenhaus's hypothesis (as
well as the current account) retains the same core lexical meaning in both
uses:
	(29) a. Bill loaded the truck onto the ship.
	     b. Bill loaded the truck with bricks.

Carlson and Tanenhaus reasoned that if a reader or hearer initially selects
an inappropriate sense of an ambiguous word like set, a garden path will
result, effecting an increased processing load. On the other hand, if an
inappropriate constructional use ("thematic assignment" on Carlson &
Tanenhaus's account) is selected, the reanalysis will be relatively cost free
since the sense of the verb remains constant and the verb's participant roles
("thematic roles" on Carlson and Tanenhaus's account) are already activated.

Sentences such as those in (28) and (29) were displayed on a CRT, and
subjects were asked to decide as quickly as possible whether a given sentence
"made sense." It was expected that subjects would anticipate an inappropriate
sense of set or an inappropriate use of load approximately half the time. A
theory which posits two distinct senses of load to account for the two uses
in (29), analogous to the situation with set in (28), would presumably expect
the two cases to work the same way. Carlson and Tanenhaus found, however,
that misinterpreted lexical ambiguity creates a more marked processing load
increase than misinterpreted uses of the same verb. The load increase was
witnessed by subjects' longer reaction time to decide whether sentences such
as (28) involving a true lexical ambiguity made sense, vis-a-vis sentences such
as (29), as well as by a marked increase in the number of "no" responses to
the question whether a given sentence made sense when a truly ambiguous verb
was involved.

1.4.6 Supportive Evidence from Child Language Acquisition
   By recognizing that the meanings of verbs do not necessarily change when
these verbs are used in different syntactic patterns - that the meaning of an
expression also depends on the inherent semantics of the argument structure
constructions - certain findings in language acquisition research can be made
sense of.

Landau and Gleitman (1985) note that children acquire verb meanings with
surprising ease, despite the fact that the situations in which verbs are used
only constrain possible meanings to a very limited degree (cf. also Quine
1960). For example, they note that their congenitally blind subject learned
the meanings of look and see without undue difficulty, despite the fact that
these meanings are nonphysical and, for this child, not directly
experientially based. They propose that children rely on syntactic cuing, or
syntactic bootstrapping, as they acquire verbal meaning. In particular, they
argue that children make use of the set of syntactic frames that a verb is
heard used with in order to infer the meaning of the verb. They argue that
this is possible because syntactic frames are surface reflexes of verbal
meanings: "The allowable subcategorization frames, taken together, often tell
a semantically quite transparent story, for they mark some of the logical
properties of the verb in question" (p. 140). Further, they assert that the
use of a verb in a particular syntactic frame indicates that the verb has a
particular component of meaning, one associated with that syntactic
frame. Certain experimental work by other researchers substantiates the idea
that syntactic frames aid in the acquisition of word meaning (see Brown 1957;
Katz, Baker & McNamara 1974; Naigles 1990; Fisher et al. 1991; Gleitman 1992;
Naigles et al. 1993).14

However, Pinker (1989) rightly criticizes Landau and Gleitman's formulation
of the claim. He notes that if different syntactic frames are assumed to
reffect different components of the meaning of verbs, as Landau and Gleitman
assume, then taking the union of these different components of meaning across
different syntactic frames will result in incorrect learning. For example, if
the appearance of an into-phrase in The ball floated into the cave is taken
to imply that float has a motion component to its meaning, then the child
will incorrectly infer that it will not be possible to float without moving
anywhere.

Pinker's criticism rules out the possibility that even adult speakers could
use the set of syntactic frames a verb is heard used with to determine the
verb's meaning. It does so because each distinct syntactic frame is taken to
reflect a different sense of the verb. This apparent paradox can be resolved
by recognizing that syntactic frames are directly associated with semantics,
independently of the verbs which may occur in them.

On this view, kick has the same sense in each of the eight argument
structures listed in section 1.4.2. The interpretations20 such as, 'X ACTS',
'X ACTS ON Y', 'X DIRECTS ACTION AT Y', 'X CAUSES Y to UNDERGO a CHANGE OF
STATE' -are associated directly with the particular constructions
involved. In this way, Landau and Gleitman's insight can be slightly
reinterpreted. What the child hypothesizes, upon hearing a verb in a
particular previously acquired construction, is not that the verb itself has
the component of meaning associated with the construction, but rather that
the verb falls into one of the verb clusters conventionally associated with
the construction (cf. chapter 5).

[FN 15]. Fisher et al. (1991) state this idea succinctly: "touch" is mapped
onto 'touch' because (a) the child can represent scenes observed as 'scenes
of touching' and (b) the wave form touch is likely to be heard when
touching is happening. That this has to be at least part of the truth about
word learning is so obvious as to be agreed upon by all theorists despite
their differences in every other regard (see e.g., Locke 1690 and Chomsky
1965 - and everybody in between who has commented on the topic). You can't
learn a language simply by listening to the radio" (1991 : 2).

Fisher, Cynthia, Geoffrey Hall, Susan Rakowitz, and Lila Gleitman. 1991. When
  it Is Better to Receive than to Give: Syntactic and Conceptual Constraints
  on Vocabulary Growth. IRCS Report 91-41. Philadelphia: University of
  Pennsylvania.]

(Fillmore 1968: 24).
"The case notions comprise a set of universal, presumably innate, concepts
which identify certain types of judgments human beings are capable of making
about the events that are going on around them, judgments about such matters
as who did it, who it happened to, and what got changed"

Langacker (1991) argues that language is structured around certain conceptual
archetypes: "Certain recurrent and sharply differentiated aspects of our
experience emerge as archetypes, which we normally use to structure our
conceptions insofar as possible. Since language is a means by which we
describe our experience. it is natural that such archetypes should be seized
upon as the prototypical values of basic linguistic
constructs. ... Extensions from the prototype occur ...  because of our
proclivity for interpreting the new or less familiar with reference to what
is already well established; and from the pressure of adapting ~. limited
inventory of conventional units to the unending, ever-varying parade of
situations requiring linguistic expression" (pp. 294-5).

The term case is used to identify 'the underlying syntactic-semantic
relationship'