Linguistic Structure

Productivity occurs at both lexical and sentence level; possible to generate an infinite number of words/sentences from a finite set of phonemes/words.

How is this possible? A rule structure (or grammar) is proposed (there are alternatives; e.g., connectionism) that allows for linguistic productivity.

Criteria for evaluating a grammar (from Chomsky):

1. Observational adequacy; must specify what counts as acceptable strings (evidence = intuition)

2. Descriptive adequacy; must specify relationship between strings similar in some manner (e.g.,active - passive)

3. Explanatory adequacy; language acquisition based on universal principles.

4? Psychological reality (i.e., representational level)

Pre-Chomsky grammars

Behaviorist (Skinner): Markov finite state models; each word serves as a stimulus for next word

Sequential (left to right) model. Transitional probabilities based on actual use

e.g.: The boy hit the ball

Probability of "boy" following "The" > probability of "hit" following "The"

Acquisition based on reinforcement

Problems:

1. Acquisition (explanatory adequacy); can construct sentences never heard (and hence not reinforced)

2. Transitional probabilities can't explain acceptability (observational adequacy):

e.g.: Colorless green ideas sleep furiously

Transitional probabilities = 0, yet grammatical

e.g.: Goes down here is not large feet are happy

Transitional probabilities are high but not grammatical

3. Sentence ambiguity; can't explain sentences with multiple interpretations

e.g., They are visiting fireman

4. Can't explain sentence embedding (dependencies need not be adjacent

e.g., If...then/ either S1 or S2 constructions.

e.g.: The team that decided to show up won the game. ("won" related to "team" not "won")

Phrase Structure Grammars

Proposed by neobehaviorists (e.g., Jenkins & Palermo, 1964) and a component of early Chomsky grammar.

Sentences are grouped into constituents that are organized hierarchically (rather than sequentially)

Phrase structure grammar consists of rules for rewriting a constituent into a set of one or more constituents

e.g.: The boy hit the ball

Rewrite rules:

S -> NP VP S
NP -> det (adj) N NP VP

VP -> V NP det N V NP

N -> boy, ball The boy hit det N

V -> hit the ball

det -> the

Advantages:

 

Handles embedded elements (hierarchical)

 

Allows for recursion (NP -> NP S); can generate infinite number of sentences

 

e.g., The boy who plays baseball hit the ball

 

S

NP VP

NP S V NP

det NP NP VP hit det N

The boy who V NP the ball

(the boy)

plays baseball


Problems:

Descriptive adequacy:


1. Can't explain similarity of sentences with different surface structures (e.g., active-passive)


2. Can't explain underlying differences between sentences with similar surface structures (e.g., eager/easy)


3. Can't explain sentence ambiguity; how to parse sentences with multiple meanings:


E.g.: They are visiting fireman


Transformational Grammar


Chomsky's late 1950s and 1960s work - revolutionized linguistics and cognitive psychology

Assume existence of two levels:

Deep structure - underlying structure of sentences most related to its meaning (but not identical to meaning). Deep structure generated based on phrase structure rules

Surface structure - actual arrangement of constituents when written or spoken. Derived from deep structure via rewrite rules (deep structure transformed into surface structure.

Examples (rules are simplified):


1. Active - passive transformation:

NP1 V NP2 -> NP2 Aux V by NP1

The boy (NP1) hit (V) the ball (NP2) ->

The ball (NP2) was (Aux) hit (V) by the boy (NP1)


2. Particle movement transformation:

NP1 V part NP2 -> NP1 V NP2 part

He called up his bookie -> He called his bookie up

Restriction: NP2 = pronoun (*He called up him)


3. Dative (transfer verbs) movement transformation:

NP1 V NP2 NP3 -> NP1 V NP3 NP2

He gave the money to his bookie -> He gave to his bookie the money.

4. Declarative - interrogative relationship:

e.g.: He hit the ball. What did he hit?

He hit the ball

He hit what (substitution)

What he hit (displacement)

What he did hit (implicit auxiliary made explicit)

What did he hit? (Permutation; transpose subject and auxiliary)


Lexical Constraints

Transformational grammar is concerned with syntax. But Chomsky (e.g., 1965; Aspects of syntax) did attempt to deal with meaning (in a limited way).

Subcategorization rules: verbs can be categorized in terms of the types of constituents with which they can co-occur.

Dative (e.g., give) give [V+ NP NP] (2 object noun phrases required)

Intransitive (e.g., die) die [V - ] (no object noun phrases)


Selectional restrictions: restrictions on selection of NPs specified by subcategorization rules

E.g.: Verbs eliciting emotion (e.g., scare, frighten, disgust, etc.) are transitive and so require an object NP (subcategorization rule). But NP must be something that can experience an emotion


E.g., scare [V + NP[animate]]

*I scared the wall.


Note: semantics more ambiguous than syntax. E.g.: talk - animate; *?I talked to my computer. (Role of world knowledge)

Evaluation


Meets (with varying degrees of success) the criteria of descriptive, observational, and explanatory adequacy.

Problems:

1. Parsimony. Not an elegant theory. Numerous rules that children must learn and learn quickly. Explanatory adequacy weak

2. Psychological reality (representational level). Little empirical evidence for actual occurrence of transformations during processing.

Derivational theory of complexity (Miller); more transformations should take longer to process. Little support.

Plus, transformations may not be independent of meaning.

E.g., Slobin (1966); Present picture and ask Ss to verify

Reversible The dog chased the cat.

The cat was chased by the dog.

Nonreversible: The boy raked the leaves

The leaves were raked by the boy.

Passive > active for reversible but not for nonreversible (semantics short-circuits need for syntactic work; leaves can't rake boys.



3. Role of semantics underdeveloped

 

Government and Binding Theory


Summary of changes:

1. Units can be intermediate in size between noun and verb phrases and emphasis is no most important word in constituent.

X (X-bar) -> N (N-bar) V (V-bar) I (I-bar)

X = head (most important word/phrase) + arguments+spec

Arguments = role players (essential)

adjuncts (modifiers)

Spec = specifier (optional); subject

Universal structure, but language variability in terms of order:

English - head first

Japanese - head last



Sample rule in English: role players (essential) must be closer to head than adjuncts (invariant over phrase type)

N:

The president (Head) of Ball State University (RP) with the great retirement package (adjunct) will be leaving soon (V).

The president with the great retirement package of Ball State University will be leaving soon.


V:

He (NP) paid (head) his bookie (RP) in the park (adjunct).

He paid in the park his bookie.


I:

Ball State (NP) will (I-head) make (V-head) the NCAA tournament (NP) this year (V adjunct).


2. Lexicon plays greater role (less need for transformations); (Recall selection restrictions)

Verbs carry enough information so that transformations become less important

E.g., congratulate [V Agent (NP) Patient (NP)]


If S contains "congratulate", it will require a NP subject (agent) and NP patient (subject)

E.g., Bill congratulated Al;

*Bill congratulated

*congratulated Al


Note: similarity to lexical functionalist grammar (Breslin); Lexical information preempts need for transformations to generate structure.


give - subject, object, to object

give [V, agent, object, patient]


3. Fewer transformations; deep structure (d-structure) less important.

Transformations now handled largely by movement.


Passive:

IP

NP I

The ball aux V NP PP

was hit (the ball) by him

trace

hit [V object] - object moved to NP, trace remains.