Introduction to Lexical Semantics

1
Introduction to Lexical Semantics

• Vasileios Hatzivassiloglou
• University of Texas at Dallas

2
What this course is about

• Recent advances in NLP
• Advances in the area of lexical semantics
• Semantics meaning
• Lexical related to words

3
Language Constraints

• Several mechanisms operate to control allowable
  messages in a language and their meaning
• Basic block a letter / grapheme
• Letters combine to form morphemes (e.g., re-) and
  words

4
Types of constraints

• Men dogs walks (syntax)
• Colorless green ideas sleep furiously (semantics)
• The stock market made a gain (lexical
  preferences)
• Discourse/pragmatics
• inference, missing information, implicature,
  appropriateness

5
Word meaning

• Partly compositional (derivations)
• Mostly arbitrary
• Also not unique, in many ways
• How to represent a words meaning?

6
Meaning representation

• Logical form
• Attributes / properties
• Relationships with other words
• Specialization
• Synonymy
• Opposition
• Meronymy

7
Polysemy

• Multiple meanings for a word
• A central issue for interpreting/understanding
  text

8
Contrastive Polysemy

• Weinreich (1964)
• a. The bank of the river
• b. The richest bank in the city
• (2) a. The defendant approached the bar
• b. The defendant was in the pub at the bar
• 25 senses of bar

9
Complementary Polysemy

• The bank raised interest rates yesterday.
• The store is next to the new bank.
• (2) Mary painted the door.
• Mary walked through the door.
• (3) Sam enjoyed the lamb.
• The lamb is running on the field.

10
Metaphor and Metonymy

• All the world's a stage,
• And all the men and women merely players
• They have their exits and their entrances
• The White House said ...
• The pen is mightier than the sword

11
Synecdoche, Allegory, Hyperbole

• Synecdoche
• Part for whole
• head for cattle
• Whole for part
• the police, the Pentagon
• Species for genus
• kleenex
• Genus for species
• PC

12
Main Questions

• How can we model lexical semantics?
• Discuss properties or attributes relating to word
  meaning, constraints on word use
• How can we learn those properties and
  constraints?
• What can we use them for?
• Focus on applications in bioinformatics

13
Dictionaries

• Representing meaning via definitions, examples
• Core vocabulary
• The problem of circular reference
• Automated construction

14
Ontologies

• Representing word meaning via inheritance/speciali
  zation
• Manual and automated construction
• Domain vs. general ontologies
• Specific ontologies (PenMan, SENSUS)

15
Lexical Databases

• Representing meaning via intersections of
  concepts and links (semantic nets)
• WordNet, manual construction and verification
• Automating lexical relationship extraction
• Multiple languages

16
Context as a means for determining lexical
relationships

• A word is known by the company it keeps
• Statistical tests for word use, compositional
  preferences
• Measures for coincidence, estimation issues

17
Disambiguation

• Selecting among multiple meanings
• Dictionary and corpus-based approaches
• Training and avoiding training data
• Evaluations (SENSEVAL)
• Role of domain and discourse
• Multiple levels

18
Non-compositional preferences

• Collocations
• Non-compositional (kick the bucket)
• Non-substitutable (white wine)
• Non-transformable
• Types of collocations
• How to find them
• Domain specialization, translation

19
Lexical properties

• Lexical relationships (specialization, synonymy,
  antonymy, meronymy)
• Orientation
• Markedness
• Domain/register applicability

20
Semantic Similarity

• Used for classification, organization, clustering
• Vector representations of context
• Similarity based on vector comparison,
  probabilistic models, LSI
• Robustness and bias
• Clustering and content-based smoothing

21
Orientation and Ordering

• Semantic orientation or polarity
• Lexical vs. document level (review)
• Semantic strength
• Linguistic scales and implicature

22
Text mining

• Using large quantities of unnanotated text for
  learning lexical properties
• The web as corpus

23
Mapping across languages

• Static mapping (bilingual dictionaries)
• Dynamic mapping in MT
• Interlingua representations
• Statistical transfer

24
Evaluation Issues

• Suitable reference standards
• Agreement between evaluators
• Avoiding bias

25
Selectional constraints

• Preposition/Article selection
• Text generation
• Lexical cohesion (for rewriting, but also for
  selecting words)
• math/statistics vs. math/food

26
Terminology

• Deciding what is a term
• Terminological databases
• Issues of consistency, reference concepts,
  currency, coverage
• Automatic detection of terms
• Constraining and classifying terms
• Definitions for terms

27
Bioinformatics

• Emerging field
• Meaning of technical terms
• Disambiguation (e.g., protein/gene)
• Classification
• Functional roles
• Abbreviations

28
List of topics

• Dictionaries, ontologies, databases
• Measures for word coincidence, similarity
• Disambiguation
• Collocations
• Word categorization and clustering
• Orientation and ordering
• Text mining, the web as corpus
• Evaluation
• Multilingual issues
• Selectional constraints and cohesion
• Terminology
• Bioinformatics