Transcript Kotoe`s slides on Goldberg, Chapter 10 and Croft & Cruse, Chapter 10

Constructions at Work by Adele E. Goldberg
Chapter 10: Variations on a constructionist theme
▲ comparison with
• a subset of mainstream generative grammar approaches, which
are referred to as Syntactic Argument Structure theories (SAS)
• more closely related constructionist approaches
SAS vs. constructionist approaches
SAS
1. syntax
derivational
2. emphasis
rough paraphrases
3. constructions
underlying form
are pairings of
& coarse meaning
4. semantics
no reference
5. determined by Universal Grammar
6. language-internal
no
generalization
7. Minimalist
compatible
constructionist approaches
non-derivational
speaker construals
surface form
& detailed function
reference
use of languages ?
yes
not compatible
provide an alternative way
Hale and Keyser (1997)
▲ denominal verbs (e.g. dance, shelve, laugh, sneeze)
• possibly every verb has an internal argument
￭ syntactic incorporation
(1)
V*
V1
NP
Ni
dance
▲ meanings
• a noun ≠ a denominal verb
￭ a denominal verb has “an additional increment of meaning”
e.g. shelf & shelve
• shelve means that items must be individually placed upright
on something shelf-like rather than that the items are put on a
shelf
• “referential” component: come from the syntactic derivation
• “adverbial” component: captures the notion of the action invoked
from the meaning of the noun
▲ problem 1: the incorporation approach of denominal verbs
cannot account for their adverbial component
• it is not clear what role the referential component actually plays
• it seems that an adverbial component capture the referential
component
￭ syntactic derivation subverts the account for the adverbial
component
◘ the syntactic derivation
• be general across verbs
• no way to reference peculiar interpretations of specific words
◘ Therefore, an adverbial component
• cannot be attributed to the syntactic derivation
• not supposed to be listed in the lexicon
▲ problem 2
• the position from which incorporation takes place is expected
to be occupied by a trace, not by an over nominal
￭ the deletion of the original noun by incorporation
• the referential meaning is also eliminated
• the adverbial meaning is left
► However,
(3) a. She shelved her books on the windowsill.
b. I shelved the books on the highest shelf.
► against (3)
• the original noun is deleted
• a syntactic variable representing the argument is left through
lexical insertion
• the mysterious adverbial increment of meaning appears
• a real shelf is not involved in (3a)
• a real shelf is involved but is overtly specified in (3b)
◀ Goldberg: this is based on questionable empirical generalization
• Hale and Keyser’s incorporation analysis
motivated by Baker’s (1988) the syntactic generalization:
external argument cannot incorporate: only internal arguments
can
(4) a. A cow had a calf. internal
b. A cow calved.
c. *It cowed a calf.
external
► However,
• the generalization that all deverbal nouns correspond to
non-external argument is not accurate
• an external argument also can be incorporated into the verb
(6) a. The cook made dinner.
b. He cooked dinner.
▲ some denominal verbs are far more conventional than others
(9) a. A cow had a calf. A cow calved.
b. A woman had a baby. *A woman babied.
c. A kangaroo had a joey. *A kangaroo joeyed.
▲ constructionist approaches
• emphasize speaker construals of situations
• the difference in meaning between denominal verbs and
their corresponding nouns could not be ignored
Borer (2001)
• a “neo-constructionist” paradigm
• a derivational, autonomous view of syntax within the Minimalist
framework
• grammatical category information and the interpretation of
argument are derived from syntactic structure
• open-class words such as nouns and verbs, which are referred as
to “encyclopedic items” (Els), are stored in an “encycolopedia”
and do not contain any reference to grammatical categories or
argument structure
Q: how are category information and argument structure
properties in turn to be determined?
A: grammatical categories arise from allowing Els to merge with
grammatical features.
e.g dog + past tense → dogged
sink + a noun phrase (DP) → noun
sink + a verbal feature → verb
▲ Problem 1
► However, Borer also does not account for lexical meaning
• the noun dog and the verb dog
￭ words
• stored in the encyclopedia
• have only a category-neutral meaning
◀ Therefore, she does not discuss where these distinctions in
meaning between noun and verb come from
► Moreover,
• words cannot specify a number or type of obligatory arguments
e.g. dine, eat, devour
▲ Problem 2
￭ all external arguments must be interpreted as agents
• a multitude of counterexample
(10) a. The coma victim underwent the operation. --- experiencer
b. she received a package. --- recipient
▲ Goldberg
• meaning cannot be read off syntactic tree
• verbs and arguments make very real contributions to the most
basic of semantic interpretations
• Borer’s proposal greatly reduces the role of lexicon
Distributed Morphology (DM) (Marantz 1997)
• words are formed syntactically by combining roots with affixes
• roots are to be listed in an “Encyclopedia”, which involves real
world meaning and not linguistics knowledge
• The same root is resented by √WALK
• category information is not associated directly with roots and its
category is determined by its surrounding syntactic environment
► However, unlike the accounts discussed above, Marantz,
• attempts to account for the non-compositionality found
between the noun bake and the verb bake
• allows roots to make reference to a meaning associated with a
certain syntactic configuration within the Encyclopedia
• a second suffix must be compositionally related to the
corresponding word with a single affix
e.g. derivation & derivational
► however, exception
• an affix adds a non-compositional meaning → rare
• rare → infrequency
e.g. impress · ion · able is non-compositional
• able is normally attached to verbs
• able means “naïve” and not “able to make impression”
◀ therefore, Marantz’s Distributed Morphology cannot explain
this rare case
▲ Summary
￭ Three approaches:
• cannot account for idiosyncratic meaning of words such as the
noun shelf and verb shelve
￭ Constructionist approaches:
• each verb sense lexically specifies the number and semantic
type of arguments it has
• each argument structure constructions specifies its semantic
and information-structure properties
• the role of the lexicon is greatly expanded to include phrasal
patterns with their own idiosyncratic syntactic or semantic
properties
• the interaction of the argument structure of verb and
construction give rise to interpretation
10.2 A comparison of more closely related
constructionist approaches
▲ approaches that are much close to the constructionist approach
outlined in chapter 1.
• UCxG: Unification Construction Grammar (Kay, Fillmore, Sag
and Michaelis) → Construction Grammar (Croft: 2004)
• CG: Cognitive Grammar (Langacker)
• RCxG: Radical Construction Grammar (Croft)
• CCxG: Cognitive Construction Grammar (Lakoff, Goldberg, and
Bencini) → construction grammar (Croft: 2004)
▲ UCxG (Table 10 p.215)
•
•
•
•
does not uniformly adopt usage-based theory
puts heavy the heavy focus on unification-base formalism
does not adopt role of “motivation”
emphasizes formal explicitness; maximal generalization
10.3 Usage-Based or Maximal Generalization Only
◆ UCxG
• in line with generative framework
• aims to account for generalization without redundancy
• the frequencies of particular grammatical patterns are not
represented
• strict division between grammar and the use of the grammar
by Fillmore et al
◆ CG, CCxG, and RCxG
• usage-based frameworks
• aims to present grammatical knowledge in such a way that it can
interface with theories of processing, acquisition, and historical
change
• fully regular patterns may be stored
10.4 Formalism: Unification or Diagramatic or Other
▲ formalism
• for the sake of clarity and explicitness
• few of generative linguistic employ any systematic formalization
• the theories that have close ties to computational linguistics adopt
◆ UCxG
• close ties to HPSG and the FrameNet project of Fillmore
￭ HPSG: strong computational component, a unification-based
formalism
• adopts a unification-based formalism
▲ unification in UCxG
• a feature-based system, each constructions is presented by
features with values (feature structures), an Attribute-Value
Matrix (AVM)
• any pair of AVMs can be combined to license a particular
expression
• when two AVMs unify, they map onto a new AVM, which has
the union of attributes and values of the two original AVMs
￭ the verb phrase construction (e.g. found her bracelet)
(i) [cat v]
combination of a set of primitive
[role head]
atomic units → reductionist model
[lex +]
(Croft 2004: 266)
[role filler]
[loc +]
+
[gf –sbj]
▲ drawback to using the unification-based approach
• not sufficiently amenable to capturing detailed lexical
semantic properties since real meaning is not easily captured
by a fixed set of features
→ Fillmore (1975): need to recognize frame-based or
encyclopedic knowledge
• overemphasize syntactic elements insofar as the recurrent
features are most relevant to the unification mechanism
• cannot account for subtle differences in meaning between
different constructions since the features and categories of
languages are numerous
10.5 To Motivate or to Stipulate
◆ CCxG:
seeks to provide motivation for each construction in an effort to
constrain the theory and make it explanatorily adequate
• motivation is distinct from prediction
￭ motivation: could have been there
◘ aims to explain why this particular form-meaning
correspondence “make sense” and at best natural
◘ does not mean that the construction must exist since language
is contingent, not deterministic
◘ could not have had the opposite values of the properties
claimed to provide motivation
￭ prediction --- what is the aim of prediction?
◘
• functional and historical generalizations in languages are not
predictive but motivated by general forces since languages are
phylogenetic and ontogenetic
▲ The principle of Maximized Motivation (Goldberg 1995)
if construction A is related to construction B formally, then
construction A is motivated to the degree that it is related to
construction B semantically. Such motivation is maximized.
e.g. lower-trunk-wear constructions (pants, shorts, knickers)
→ plural structure
• motivated by the fact that the referents have bipartite structure
• motivated by a plural construction since the plural
construction and the lower-trunk-wear construction share the
same form and have related meaning.
▲ Japanese number terms: ichi (one), ni (two), …. jyu (ten)
• nijyu-ichi (21) is generated by the nijyu-n construction
• the nijyu-n construction is motivated by the niju (20)
construction and the jyu-n (10-n) construction
Jyu (10)
jyu-n
(10 + n, 0<n<10)
nijyu (20)
nijyu-n
(20 + n, 0<n<10)
◆ UCxG & CCxG
• be psychologically valid
• strive to be explicit and to capture relevant generalization
◆ CCxG
• psychological validity > being explicit or maximally general
◆ UCxG
• being explicit or maximally general > psychological validity
10.6 Cognitive Grammar
• Langacker (2003) provides 12 tenets that CxG, RCxG, and CG
all agree
(11) a. Constructions are the primary objects of description
b. The frameworks are non-derivational
c. Lexicon and grammar form a continuum of constructions
d. Constructions are form-meaning parings
e. Information structure is recognized as one face of
constructionist meanings
◆ CG:
• reductionist
• the finest-grained level of analysis is privileged
• grammar is however reducible to something more fundamental
• the function of a construction is believed to inhere in the
form
e.g. cross-linguistic differences in the conventional
expression of bodily sensations → Whorfian difference
◆ CCxG & RCxG
• non-reductionist
• there are interaction between parts that lead to emergent
properties that can only be described at the level of the whole
◆ CCxG
• each construction is understood as one potential option
among others
• the meaning of a construction is not necessarily inherent
but may be in part a pragmatic inference
￭ Goldberg: construction-specific properties of semantic role
◘ participant roles are specific to particular verbs
◆ RCxG
￭ Croft: construction-specific properties of grammatical
categories and relations
◘ syntactic roles must be defined construction-specifically
◘ “subject” and “object” do not define some fixed category
or syntactic structure, and they vary within and across
languages
◆ CG
￭ Langacker
◘ essentialist definitions
Subject – primary focal point
Object – secondary focal point
N – “thing”
V – “relation”
◘ a particular, schematic conceptual factor is part of every
such extension, constituting an invariant conceptual
characterization of each category
► CCxG & RCxG
￭ not essentialist definitions
• Croft: there is not consistent cross-linguistic distributional
pattern
• essentialist definitions for no-linguistic categories are
exception
• linguistic generalizations
formal patterns are extended and reused for related but
distinct functions
• cross-linguistic generalizations
recurrent archetypes across languages are based on
universal functional pressures
10.7 Construction Grammar and Radical Construction
Grammar (Croft 2001)
◆ RCxG
• extends work in construction grammar by investigating in
detail the cross-linguistic divergences, typology theory,
the usage-based model
• there are generalizations within or across languages
• the generalizations are determined by functional purpose
that each language’s constructions serve
• Croft points out from the cross-linguistic research
1. tense-mood-aspect inflection cannot be taken as criterial
for determining the category of Verb cross-linguistically
e.g. inflection
• Makah and a native American language:
verbs, nouns, adjectives, adverbs
• Vietnamese: no words
2. no syntactic test will pick out entities that one might wish to call
verbs, nouns, adjectives, subjects, objects, and so on across all
languages
3. even within a single language, a given criterion often only
applies to certain constructions
e.g. passivizability as the criterion for Direct Object
• its result tells some features of passivizability but it doesn’t
tell some global category Direct Object
▲ construction grammar: Croft (2004: chapter 10)
• general syntactic constructions have corresponding rule of
semantic interpretation → symbolic units (pairings of syntactic
structure (elements) and semantic structure (component))
→ symbolic links
• the symbolic link between form and meaning is internal to a
construction ↔ external in SAS (Fig. 10.2 & 10.3 p. 258-9)
• “meaning” means conventional meaning and includes
properties of the situation, the discourse, and the pragmatic
situation
▲ construction grammar: four questions
• has meronomic (part-whole) structure of grammatical units like
SAS (Fig. 10.4 p.268)
► however, the nature of categories of parts diverge among
theories
→ Q (i) What is the status of the categories of the syntactic
elements in construction grammar, given the existence of
constructions?
• internal structure of constructions
e.g. “subject” (2) p.262
• role to the whole construction (semantic relation)
• relation to another element on the construction (syntactic
relation)
→ Q (ii) What sorts of syntactic relations are posited?
• forms taxonomic network from more schematic representation
to idiosyncratic representation, which shows different kinds of
grammatical knowledge and syntax-lexicon continuum
(4)
[Verb Phrase]
[Verb Obj]
[kick Obj]
[kick [the habit]]
(Croft 2004: 263)
► however, constructions may be
liked by relations other than
taxonomic relations
→ Q (iii) What sort of
relations are found between
constructions?
• the taxonomic hierarchy appears to represent the same
information at different levels
e.g. the habit in kick the habit is the direct object at the idiom
construction and at more of the schematic level [Tr Verb Obj]
→ Q (iv) How is information stored in the construction
taxonomy?
(i) What is the status of the categories of the syntactic elements
in construction grammar, given the existence of constructions?
▲ Construction Grammar
• constructions: atomic primitive grammatical relations (e.g.
subject) and primitive syntactic categories (e.g. verb) →
reductionist
• specific constructions: contain syntactic and semantic information
that is not found in the units of the constructions that make up
▲ construction grammar
• a participant role is defined by the situation as a whole
→ complex events are primitive units of semantic representation
→ nonreductionist
• analysis of syntactic role and relation in argument structure is
reductionist like Construction Grammar
(i) What is the status of the categories of the syntactic elements
in construction grammar, given the existence of constructions?
▲ Cognitive Grammar
• have an essentially semantic basis
• syntactic categories (e.g. noun and subject): abstract semantic
construals of the conceptual content of their denotations
► However, the construal of specific experiences as belonging to
the semantic categories is language-specific
e.g. English sick & Russian bol(e)- verb + an adjectival
derivation suffix
(i) What is the status of the categories of the syntactic elements
in construction grammar, given the existence of constructions?
▲ Radical Construction Grammar
• nonreductionist
• constructions
￭ the primitive elements of syntactic representation
￭ defines categories in terms of the constructions that they
occur in
e.g. the elements of the Intransitive construction:
Intransitive Subject and Intransitive Verb
• allows for prototypes and extensions of constructions
(ii) What sorts of syntactic relations are posited?
▲ Construction Grammar: to assemble the parts of a
construction into a whole
• role: represents the role of the syntactic element in the whole
(modifier, filler, and head)
e.g. sing – head, Heather – filler in Heather sings.
• val: indicates the relation of the predicate to its argument by a
cross reference to the set of semantics
e.g. semantic feature structure for sings gives the argument A
to the singer argument
• rel: indicates the relation of each argument to its predicate, a
syntactic feature (grammatical function) and a semantic feature
(thematic role) e.g. Heather is “subject” and “agent”
◀ the answer to (ii)
• predicate-argument relation are syntactic and semantic
• However, predicate-argument relation are distinguished from
syntactic roles held by elements in the construction as a whole
e.g. The book is red and the red book
red – predicate, book – argument in both
be red – head, book – head
(ii) What sorts of syntactic relations are posited?
▲ construction grammar
Lakoff (1978)
• Syntactic elements (clause, noun phrase)
• Lexical elements (here, there)
• Syntactic conditions (liner order of elements)
• Phonological conditions (vowel length)
(Croft 2004: 273)
• allows for relations between syntactic elements as well as
relations between the elements and the constructions as a
whole
(ii) What sorts of syntactic relations are posited?
▲ Cognitive Grammar
• the concept of valance is symbolic like Construction Grammar,
but it is gradient unlike Construction Grammar
• the argument elaborates the relevant substructure of the
predicate
e.g. (34) I was reading this on the rain.
￭ a substructure of read elaborated by on the read → on the
read is less salient than I and this
￭ a substructure of on the read elaborated by read → highly
salient, more of an adjunct than a complement
• the roles represent a relation between the parts of constructions
and the whole, and they are defined semantically and
symbolically
(ii) What sorts of syntactic relations are posited?
▲ Radical Construction Grammar
• represents the role of a part of a construction in the whole
construction like Construction Grammar and Cognitive
Grammar
• defines relations between parts of a construction in purely
semantic terms unlike Construction Grammar and Cognitive
Grammar
(iii) What sorts of relations are found between constructions?
(iv) How is grammatical information stored in the
construction taxonomy?
▲ Construction Grammar
• a complete inheritance mode to avoid redundant representation
– represents information once in the construction taxonomy
at the highest level possible
• parts of a construction can inherit feature structures from
another construction (multiple inheritance)
e.g. nonsubject WH-question construction (why did she leave
him?)
￭ left-location construction and SAI construction
(iii) What sorts of relations are found between constructions?
▲ construction grammar
• the taxonomic links and the meronomic link (‘one construction is a
proper subpart of another construction’)
• the polysemy link (subtypes of a construction that inherit the
syntactic construction but are different in their semantics, a
metaphorical extension)
e.g. ditransitive constructions
(17) Suj causes Obj2 to receive Obj1. Joe gave Sally the bell.
(19) Suj enables Obj2 to receive Obj1. Joe permitted Chris an apple
(iii) What sorts of relations are found between constructions?
▲ Cognitive Grammar
• prototype-extension relations
• a schema subsuming both prototype and extension
▲ Radical Construction Grammar
• a taxonomic relationship, which must be linguistically
motivated
e.g. Verb superordinate to IntrV and TrV inflects with the
tense/agreement
• meronomic relation
• each unit of a construction is defined by its occurrence
(iv) How is grammatical information stored in the construction
taxonomy?
▲ construction grammar
• normal inheritance: inheritance can be blocked if it conflicts
with information in the more specific case
• a full entry model: the representation of information at all
levels in the taxonomic hierarchy of constructions
→ resolve a conflict in multiple inheritance
(iv) How is grammatical information stored in the construction
taxonomy?
▲ Cognitive Grammar
• a usage-based model, therefore the establishment of schematic
constructions is the result of language use
▲ Radical Construction Grammar
• allows for redundant representation of grammatical
information in accordance with the usage-based mode like
Cognitive Grammar and construction grammar
• semantic map model
￭ constructions are mapped onto a conceptual space according
to their function
￭ constructions can be related to one another by virtue of
having overlapping or neighboring functions in the
conceptual space