book excerptise:   a book unexamined is wasted paper

Lexical Semantics Without Thematic Roles

Yael Ravin

Ravin, Yael;

Lexical Semantics Without Thematic Roles

Oxford University Press, 1990, 248 pages

ISBN 0198248318, 9780198248316

topics: |  language | semantics

Excerpts : History of Chomskyan approach

Chomsky, Syntactic Structures (1957):
    The notion "grammatical" cannot be identified with "meaningful" or
    "significant" in any semantic sense... Any search for a semantically
    based definition of "grammaticalness" will be futile. - p.15

Standard theory:
==> Syntactic component of generative grammar specifies rules for generating
    wff and their syntactic repr;  Phrase structure rules generate underlying
    structures, and transformational rules turn them into surface structures,
    or actual word strings.

Chomsky does not discuss the semantic component in syn structures, and
mentions it only briefly in Aspects of Syntax 65, stating that the sem
component specifies rules for the interpretation of the underlying (deep)
structure, and provides sem representations for sentences.  In the lexicon,
each item is provided with a descr of its sem properties.  Although there was
some corresp between syn structures and meaning, argues for the syntactic
component being autonomous.  Autonomous syntax became an imp underlying
assumpn of what came to be known as the Standard Theory. - 2

These Standard premises concerning the autonomy of syntax, and the existence
of two indep components (syntactic+semantic), each with its level of repr,
has been challenged a number of times; most importantly in the 1960s, by
several theories known as generative semantics.  The proponents of gen
semantics advocated deep structures that were both sem and syn in nature.
Diff paraphrases were the result of transformations yielding diff surface
forms. ("Do you beat your wife enthusiastically?" = "are you enthusiastic in
beating your w?")

Since syntax was derived from sem, there was no autonomy of syntax.  Gen Sem
eventually receded as it encountered increasing difficulties in accounting
for multiple paraphrases by means of transformations of a single deep
structure.  It also came under attack for failing to account for diff
syntactic generalizations across diff semantic representations (Chomsky 72)

Chomsky challenges autonomy of Syntax

[Autonomy of syntax has] now been challenged again, this time by Chomsky
himself.  In a drastic departure from his long-standing commitment to the
autonomy of syntax, the new version of Chomsky's theory, referred to as the
Government and Binding theory (GB), proposes semantics as the basis for
generating syntactic structures. p.3

Chomsky proposes theta-roles (semantic) as the basis for generating
syntactic structures.  Chomsky's position is different from generative
semantics - focus is on the semantic relations
between a head and its syntactic complements.  These relations, called
thematic roles, are stored in the lexican entries of the potential heads
(verbs, adj, and certain nouns).
 	     [But can't thematic roles be viewed as a property of syntax?]

Under the new view, thematic roles determine the syn structures in which
heads can appear.
	The introduction of thematic roles into the theory is a significant
	change of perspective from Chomsky's earlier versions of generative
	grammar.  First, the claim that thematic roles determine syntactic
	structure violates the previous principle of autonomy of syntax.
	Second, the claim that thematic roles are the only valid semantic
	entities in the meaning of predicates contradicts the previous
	assumption that there are semantic phenomena independent of syntax.
	All semantics information is now found in the lexicon or incorpated
	into syntactic structures. p.3

The idea of thematic roles or the claim that they determine syntactic
structures is not new.  Roles were first introduced by Fillmore 1968, then
Gruber, and more recently, Jackendoff.  Severely criticized by
Chomsky/Katz.

But Chomsky's GB theory explicitly refers to Fillmore's and Gruber's
theories.  Greeted enthusiastically:

  GB, GPSG, and LFG : Clause structure is largely predictable from the sem of
  the predicates.  The surprising thing (to linguists) has been how little
  needs to be stipulated beyond lexical meaning.  [Sells 1985]

2 Restrictive and Non-Restrictive theories


Restrictive: thematic roles assoc with a predicate determine the syntactic
     configurations in which the pred can grammatically occur.

Held by Fillmore, Chomsky, etc. Ravin shows that this faces problems in
handling events.

{Strongest form, Chomsky/Fillmore: Princ of Syntactic Relevance: Th Roles
are the only part of semantics that needs to be considered in Ling Theory,
since they are the only part that affects syntax.
	**** If not relev to syntax, it is not relev to language.)

Ravin: gives example of such an approach in phonology: the diff in
phonology between keep and postpone, which have the same syntactic
structure would be considered non-linguistic and wd not be represented,
whereas the difference between keep and kept is considered linguistic
and would be represented. 9

Non-Restrictive: Jackendoff, Ravin etc.  All sem econcepts found in
the meaning of ling expressions are to be accounted for, and syn structure
is autonom of sem.

Strong Syntactic Relevance view: If a them role is syntactically relevant,
then it must have been present in the  semantics of the l.u.; without this
aspect, semantic judgements are vague
and intuitive.

Ravin: This latter is not so for "All" sem structures - everyone agrees on
the sem of "I went to my office" or that "Tom is a female uncle" is
contradictory.  Some sem structures are intuitive - but this is true in
language of ALL RULES.

Chomsky 1957: 13-14: we may assume that certain sequences of phonemes are
   definitely sentences, and that certain other sequences are definitely not.
   In many intermiediate cases, we shall be prepared to have the grammar
   itself decide, when the grammar is set up in the simplest way so that it
   includes the clear sentences and excludes the clear non-sentences.

[***: Therefore finding a grammar is an exercise in induction / mc learning -
      and probabilistic theories work better.  If so, why the repugnance of
      data in later work? ]

Restrictive Theories: Syntactic categs of complements are predictable based
on thematic role.  e.g. Agent, Patient: NP.  Instrument: "headed by with", etc.
In Chomsky, thematic relns are one among several conditions constraining
surface struct derivation; for Fillmore they are the sole determinant.  In
MLP, they are derived from semantic principles.  In all, they determine the
syntactic configurations in which the pred can grammatically occur.

[IDEA: Fillmore: equiv to Panini; and much of this debate shadowed in the
critiques of K and Patanjali]

Chomsky 57:
    there is no regular correspondence between meaning and syn structure;
    e.g. Subj-Verb = actor-action - violated by "the fighting stopped" or
    "John received a letter"

[IDEA: The structure of grammatical rules.
Rule: decides Y/N : given an input, and a characterizn in terms of a finite
      set of discrete categories.
Problem: Also seeking simplification in rules.  Not happy if rules have a
huge set of parameters, and define the answer in terms of some complex
partitioning in this high-D space. ]

Correlation Principle

Correlation Principle: thematic structure specified by the meaning of a
lexical item determines the particular syntactic configuration in which the
lexical term occurs.  12

But such regular correspondences do not occur, as noted by Chomsky in SS:

    Such sentences as 'John received a letter' or 'the fighting stopped' show
    clearly the untenability of the assertion... that the grammatical
    relation subject-verb has the 'strutctural meaning' actor-action, if
    meaning is taken seriously as a concept indep of grammar.  Similarly the
    assignment... of any such structureal meaning as action-goal in the
    verb-object relation as such is precluded by such sentences as 'I will
    disregard his incompetence' or 'I missed the train'. 12

Thematic role can be present in verb without giving rise to the syntactic
structures supp associated with it; e.g. "John committed suicide", where the
thematic roles of Agent and Patient exist in the meaning of "commit suicide",
(as they do in "kill") but the syntactic structure is not as it is in "John
killed himself".  On the other hand, a thematic role may be absent from the
meaning of a certain verb and the syn structure of a sentence may not reflect
it, as in "He suffered a fatal injury" where only Patient is present, but the
sentence contains both a subj and an obj.
[I don't get this - this arg may be weak] 12

"John sliced the meat" conveys all the info of "John divided the meat into
slices, using an instrument with a sharp edge", except that by hiding the
implicit arguments, these do not shift the listerner's attention away from
the topic of the discourse.  Specifically mentioning arguments focuses on
their particular nature: "John sliced the meat into uneven pieces, using his
pocket-knife." 13

To maintain the Correlation Principle in spite of these problems, Restrictive
theories can:

a) broaden the sem content initially describeed as the "meaning" of the
   thematic roles, to the point of rendering them meaningless; or
   e.g. MLP: He broke the window vs The hammer broke the window ; broadens
   "agent: to accept latter.
   [NOTE: the chef cooked the meal vs the pot cooked the meal debate in
   Matilal]

b) if they keep the them roles restricted semantically, then they predict
   thematic ambiguity where tehre is no real sem ambiguity,
   In the exreme case [Chomsky GB's "theta roles"] thematic roles become
   simply another type of syntactic relation holding between syntactic
   constituents, corresponding to no semantic relation at all. [See Chap 3] 14

Fillmore restricts the meanings of his thematic roles.  "Find" does not
assign Agent to its subject.  But verbs like "break" become thematically
ambiguous - J broke the w, vs the hammer b t w.

In the absence of a semantic component, the Restrictive theories are not able
to determine what constitutes a genuine semantic concept. 15
This is why the lack of a thematic-semantic correspondence is not immediately
apparent in the Restrictive theories. 16

Jackendoff does not derive syn structures from thematic structures, and
therefore does not subscribe to the Principle of Syntactic Relevance.  His
semantics tries to capture all semantic concepts inherent in the meaning of
language expressions, whether expressing relations to other syn constituents or
not.

Also, he does not believe in the correlation principle - his sem repr and syn
repr are indep but linked through mapping conventions.

==> non-Restrictive.  However, he takes the them structure of motion verbs as
    protoypical, and adapts this to repr the thematic structure of all other
    verbs of events or states; but he has gradually moved away from thematic
    roles - his most recent model does not include thematic roles.


blurb: Ravin argues that thematic roles are not valid semantic entities, and that syntax and semantics are indeed autonomous and independent of one another. Suggesting a decompositional approach to lexical semantics in the spirit of Katz's semantic theory, the book considers such theoretical issues as indeterminacy and ambiguity, lexical configuration rules, and lexical projection, and analyzes the semantic content of event concepts such as causation, action, and change.

amitabha mukerjee (mukerjee [at-symbol] gmail) 2011 Oct 27