Semantics

Going beyond syntax

1/27
 Semantics
• Relationship between surface form and
meaning
• What is meaning?
• Lexical semantics
• Syntax and semantics

2/27
 What is meaning?
• Reference to “worlds”
– Objects, relationships, events, characteristics
– Meaning as truth
• Understanding
– Inference, implication
– Modelling beliefs
• Meaning as action
– Understanding activates procedures

3/27
 Lexical semantics
• Meanings of individual words
– Sense and Reference
– What do we understand by the word lion ?
– Is a toy lion a lion? Is a toy gun a gun? Is a fake
gun a gun?
• Grammatical meaning
– What do we understand by the lion, lions, the
lions, … as in
The lion is a dangerous animal
The lion was about to attack
4/27
 Lexical relations
• Lexical meanings can be defined in
terms of other words
– Synonyms, antonyms, broader/narrower
terms
– synsets
– Part-whole relationships (often reflect real-
world relationships)
– Linguistic usage (style, register) also a
factor
5/27
 Semantic features
• Meanings can be defined (to a certain
extent) in terms of distinctive features
– e.g. man = adult, male, human
• Meanings can be defined (to a certain
extent) in terms of distinctive features

6/27
 Types of representation
1. Syntactic relations
The man shot an elephant with his gun

shot

subj obj adv

man elephant gun

det det mod

the an his
7/27
 Types of representation
2. Deep syntax
The man shot an elephant with his gun
An elephant was shot by the man with his gun
shot

dsubj dobj instr

man elephant gun

qtf qtf poss

the an his
8/27
 Types of representation
3. Semantic roles, deep cases
The man shot an elephant with his gun
An elephant was shot by the man with his gun
shot
The man used his gun to shoot an elephant
agent patient instr

man elephant gun

qtf qtf poss

the an his
9/27
 Types of representation
4. Event representation, semantic network
The man shot an elephant with his gun
An elephant was shot by the man with his gun
shooting
The man used his gun to shoot an elephant
shooter shot- instr
thing
man elephant gun

qtf qtf poss

the  man
10/27
 Types of representation
5. Predicate calculus
The man shot an elephant with his gun
An elephant was shot by the man with his gun
The man used his gun to shoot an elephant
The man owned the gun which he used to shoot an elephant
The man used the gun which he owned to shoot an elephant

event(e) & time(e,past) &

John punched Mary

12/27
 Types of representation
7. Semantic formulae (Wilks)

((THIS((PLANT STUFF)SOUR))
((((((THRU PART)OBJE) (NOTUSE *ANI))GOAL)
((MAN USE) (OBJE THING)
)))

door

13/27
 Uses for semantic
representations
• As a linguistic artefact (because it’s there)
• To capture the text  meaning relationship
• Identifying paraphrases, equivalences (e.g.
summarizing a text, searching a text for
information)
• Understanding and making inferences (e.g.
so as to understand a sequence of events)
• Interpreting questions (so as to find the
answer), commands (so as to carry them
out), statements (so as to update data)
• Translating 14/27
 Uses for semantic
representations
• Different levels of understanding/meaning
• Textual meaning may be little more than
disambiguating
– Attachment ambiguities
– Word-senses
– Anaphora (pronoun reference, coreference)
• Conceptual meaning may be much deeper
• Somewhere in between – a good example is
Wilks’ preference semantics: especially good
for metaphor
15/27
 Linguistic issues
• Words and Concepts
– Objects, properties, actions  n, adj, v
– Language allows us to be vague (e.g. toy gun)
• Semantic primitives – what are they?
• Meaning equivalence – when do two things
mean the same?
• Grammatical meaning
– Tense vs. time
– Topic and focus
– Quantifiers, plurals, etc.
16/27
 Linguistic issues
• There are many other similarly tricky
linguistic phenomena
– Modality (could, should, would, must, may)
– Aspect (completed, ongoing, resulting)
– Determination (the, a, some, all, none)
– Fuzzy sets (often, some, many, usually)

17/27
 Lexical semantics
• Lexical relations (familiar to linguists)
have an impact on NLP systems
– Homonymy –word-sense selection;
homophones in speech-based systems
– Polysemy – understanding narrow senses
– Synonymy – lexical equivalence
– Ontology – structure vocabulary, holds
much of the “knowledge” used by clever
systems
18/27
 WordNet
• Began as a psycholinguistic “theory” of how
the brain organizes its vocabulary (Miller)
• Organizes vocabulary into “synsets”,
hierarchically arranged together with other
relations (hyp[er|o]nymy, isa, member,
antonyms, entailments)
• Turns out to be very useful for many
applications
• Has been replicated for many languages
(sometimes just translated!)
19/27