Lexical Semantics Chapter 16

1
Lexical SemanticsChapter 16

• Lindsay Butler
• Ling 538
• 5 December 2006

2
What is lexical semantics?

• Systematic meaning-related structure
• Lexeme pairing of an orthographic or
  phonological representation with a meaning
  (Saussurian sign)
• Lexicon a list of lexemes (finite)
• Sense the meaning component of a lexeme
• Lexemes are not analyzable units. They have
  internal structure that determines how they
  combine with other elements in the sentence
• The lexicon is not simply a finite listing, but
  rather a creative generator of infinite meanings

3
Senses of lexemes

• Homophony words that have the same form but
  different meaning
• bank financial institution
• bank sloping mound
• Homograph (orthographic)
• bass type of fish
• bass musical instrument
• versus homophone (phonological)
• would auxiliary verb
• wood hard, fibrous substance

4
Senses of lexemes

• Polysemy a single lexeme with multiple related
  meanings
• bank (financial institution) and bank (sloping
  mound) are not related (etymologically)
• but, bank (financial institution) and
• blood bank (not financial, but same concept of
  holding a deposit, just of blood)
• You can bank on Mans (not financial, but it has
  the sense of security)
• Finding the right meaning is the task of word
  sense disambiguation

5
Senses of lexemes

• Synonymy different lexemes with the same
  meaning
• Test of substitutability
• Example big and large

6
Senses of lexemes

• Hyponymy a class of synonymy pairings of
  lexemes where one denotes a subclass of the other
• Hyponym the more general of the pair
• Car is a hyponym of vehicle
• Hypernym the more specific of the pair
• Set of hyponyms have proved useful approximations
  of ontologies, taxonomies, and object structures

7
WordNet

• A database of lexical relations for English
• http//wordnet.princeton.edu
• Three databases for nouns, verbs, adjectives and
  adverbs
• Based on the concept of a synset
• synonymy chump, fish, fool, gull, mark, patsy,
  fall guy, sucker, schlemiel, shlemiel, soft
  touch, mug a person who is gullible and easy to
  take advantage of

8
Structure of lexemes

• Thematic roles a set of categories that
  characterize certain arguments of verbs into a
  shallow semantic language
• Jon climbed the wall
• Shannon washed his hands
• Deep roles are specific to the event climb, wash
• Shallow roles reveal a commonality between
  climbing and washing They have animate
  volitional actors that are causers of the event.
  Thus, they demonstrate the thematic role of agent

9
More thematic roles

• Theme participant most directly affected
• Experiencer simply, the experiencer
• Force non-volitional causer
• Instrument simply, an instrument used
• Beneficiary simply, the beneficiary
• Source origin of the object of a transfer
• Goal destination of the object of a transfer

10
FrameNet

• Lexical resource for English thematic roles
  (Baker et al., 1998 Lowe et al., 1997)
• http//framenet.icsi.berkeley.edu
• More than 625 semantic frames.

11
Structure of lexemes

• Selectional restrictions
• Lexemes have restrictions on which concepts can
  perform certain thematic roles
• Example
• I wanna eat someplace thats close to campus
• eat is intransitive and doesnt select an object
  (or theme)
• You dont want to eat the someplace thats close
  to campus
• I wanna eat some really good Chinese food today
• eat is transitive and does select an object (or
  theme) some really good Chinese food

12
Representing selectional restrictions

• Using event-oriented semantics to capture
  selctional restrictions
• Hyponomy relations in WordNet Evidence that
  hamburgers are edible

13
Structure of lexemes

• Primitive decomposition
• Example (motivated by McCawley (1968)
• Andrew killed his evil twin
• Andrew caused his evil twin to become not alive
• Though kill and cause to become not alive are not
  synonyms, they have the same meaning
• Decomposing a predicate into a more complex set
  of predicates DO, CAUSE, BECOME, NOT, ALIVE
• Conceptual Dependency (Schank, 1972) (more
  decomposition) is the most widely used in NLP
• 11 primitives such as ATRANS (the abstract
  transfer of possession or control from one entity
  to another), PROPEL (the application of physical
  force to move an object)

14
Structure of lexemes

• Semantic field
• set of words from a single domain may be captured
  by a more integrated or holistic relationship
  among them
• The semantic domains that FrameNet employs, such
  as HEALTH CARE, CHANCE, PERCEPTION,
  COMMUNICATION, TRANSACTION, TIME, SPACE, BODY,
  MOTION, etc., can be used to represent a semantic
  field

15
Creativity in the lexicon

• Metaphor We have in mind a certain concept or
  situation, but we use words and phrases that are
  relevant to totally different kinds of concepts
• Conventional metaphor (one type)
• Such as CORPORATION AS PERSON
• Fuqua Industries, Inc. said Triton Group, Ltd., a
  company it helped resuscitate, has begun
  acquiring Fuqua shares
• And Ford was hemorrhaging its losses would hit
  1.54 billion in 1980.

16
Creativity in the lexicon

• Metonymy We denote a concept by using a closely
  related concept
• Example PLACE FOR INSTITUTION
• The White House had no comment
• Example AUTHOR FOR AUTHORS WORKS
• He likes Shakespeare

17
Computational approaches

• For metaphor and metonymy
• Convention-based
• apply language specific knowledge
• Reasoning-based
• not specifically language related but rather a
  general reasoning ability

18
Conclusions

• Lexical semantics deals with the vast meaning and
  structure of words/lexemes
• Words cannot be analyzed in isolation
• They can have multiple meanings, selectional
  restrictions on what can co-occur with them, and
  can be decomposed
• Databases to help deal with the complexity of
  sense and structure WordNet and FrameNet
• The lexicon, though a finite list of lexemes, has
  infinite generative power (creativity of
  language)
• How do we deal with the vastness and creativity
  of language computationally? Decomposition