Transcript November 20, 2003 Chapter 16 Lexical Semantics

November 20, 2003
Chapter 16
Lexical Semantics
Words have structured meanings
• Lexeme – a pairing of a form with a sense
• Orthographic form – the way the lexeme looks on the page
• Phonological form – the way the lexeme sounds
• Lexicon – finite list of lexemes
eats
eaten
eat
ate
Lexeme
eat
Lexical Relations
• Homonymy
• Polysemy
• Synonymy
• Hyponymy
Homonymy
A relation that holds between two lexemes that have the same
form with unrelated meanings
• Homophones
• Homographs
• “found”
Lexeme
sloping
mound
bank
Lexeme
Financial
institution
Homonymy causes problems
• Spelling correction
– Confusables – your vs. you’re
• Speech recognition
– Homophones and pure homonyms
• Text-to-speech
– Homographs – conduct
• Information retrieval
– Homographs and pure homonyms
Polysemy
The phenomenon where a single lexeme has multiple
related meanings
Biological
repository
bank
Lexeme
Financial
institution
Polysemy
• How many senses does a word have?
– Zeugma: Does Midwest Express serve breakfast and
Philadelphia
– Kim has an uncle and so does Sandy
– Kim has a bat and so does Sandy
• How are they related?
• How can you tell which sense should be attributed to a given
word?
Synonymy
A relation that holds between two lexemes with the
same sense
older
big
lexeme
Positive
size
large
lexeme
hyponymy
A relation that hold between two lexemes where one denotes a
subclass of the other
vehicle
car
• ontology
• taxonomy
• object hierarchy
hypernym
vehicle
hyponym
car
WordNet
• A large electronic database of lexical relations
• A web-based interface
• Sets of lexical entries corresponding to unique
orthographic forms, accompanied by sets of senses
associated with each form
Category
Unique Forms
Number of Senses
Noun
94474
116317
Verb
10319
22066
Adjective
20170
29881
Adverb
4546
5677
WordNet synset
• Synonymy is organized around the notion of a synset
• {chump, fish, fool, gull, mark, patsy, fall guy, sucker,
schlemiel, shlemiel, soft touch, mug}
• The synset is the sense associated with the WordNet entry.
• The semantic relations are relations between sysnets
Internal Structure of Words
• Thematic roles
• Selectional restrictions
• Primitive decomposition
• Semantic fields
Thematic Roles
• Deep roles:
Houston’s Billy Hatcher broke a bat
e, x, y Isa(e, Breaking)  Breaker(e, BillyHatcher) 
BrokenThing(e, y)  Isa(y, BaseballBat)
He opened a door
e, x, y Isa(e, Opening)  Opener(e, he)  OpenedThing(e, y) 
Isa(y, Door)
Thematic Roles
• Breaker and Opener are agents
• BrokenThing and OpenedThing are themes
• Some other commonly used thematic roles
– Experiencer
– Force
– Result
– Content
– Instrument
– Beneficiary
– Source
– Goal
Thematic Roles
• Thematic roles can be used as a shallow semantic language
• Can be used to determine surface realization:
AGENT > INSTRUMENT > THEME
The highest thing in the hierarchy will typically be the subject;
the lowest thing will be the object.
• Linking theory looks at the mapping between conceptual
structure and grammatical function.
Thematic Roles
• Doesn’t work for all verbs: donate, return, transfer
• It only helps for NP and PP arguments of verbs
• Differing perspective
– Amie bought the sandwich from Benson for three dollars
– Benson sold Amie the sandwich for three dollars
– Amie paid Benson three dollars for the sandwich
Selectional Restrictions
The senses of lexemes enforce selectional restrictions on their
arguments
• Which airlines serve Denver
– The ServedThing is a geographical location
• Which airlines serve breakfast
– The ServedThing is a meal
This helps tell which sense of a lexeme is intended in a given
context
Selectional Restrictions
Selectional restrictions occur at varying levels of specificity
In rehearsal, I often ask the musicians to imagine a tennis
game.
They tell of jumping over beds they can’t imagine clearing
while awake.
I cannot even imagine what this lady does all day.
Atlantis lifted Galileo from the launch pad.
Mr. Kruger lifted the fish from the water.
To diagonalize a matrix is to find its eigenvalues.
Representing Selectional Restrictions
The semantic contribution of a verb like eat
e, x, y Eating(e)  Agent(e, x)  Theme(e, y)
 Isa(y, EdibleThing)
The phrase ate a hamburger would get something like
e, x, y Eating(e)  Agent(e, x)  Theme(e, y)
 Isa(y, EdibleThing)  Isa(y, hamburger)
Instead of using logical concepts, we can just use WordNet
synsets: { food, nutrient }
{ hamburger, beefburger } is a hyponym of this.
Creativity and the Lexicon
We can use more word meanings that can be explicitly listed in
the lexicon.
There are productive processes for creating new senses from
those explicitly listed, including
• Metaphor
• Metonymy
Metaphor
Using metaphor, we refer to, and reason about, concepts using
terminology appropriate to completely different kinds of
concepts.
CORPORATION AS PERSON
• That doesn’t scare Digital, which has grown to be the worlds
second-largest…
• Triton Group Ltd., a company it helped resuscitate, has begun
acquiring Fuqua shares
• But if it changed its mind, however, it would do so for
investment reasons, the filing said.
Metonymy
The use of one concept to refer to another concept closely related
to it.
PRODUCT FOR PROCESS
• GM killed the Fiero because it had dedicated a full-scale
factory to building the plastic bodied car…
AUTHOR FOR WORKS
• He likes Shakespeare
PLACE FOR INSTITUTION
• The White House had no comment
Approaches to Metaphor and Metonymy
• Convention-based approaches hard-wire metaphors like
CORPORATION AS PERSON and metonyms like
PRODUCT FOR PROCESS
• Reasoning-based approaches treat this as a problem for
general, not necessarily linguistic, reasoning, such as
analogical reasoning.
Summary
• Lexeme
• Lexical Relations
– Homonymy
– Polysemy
– Synonymy
– Hyponymy
• WordNet
• Thematic Roles
• Selectional Restrictions
• Metaphor & Metonymy