Lecture 3: Lexical Semantics and Word Sense Disambiguation

Overview

• Focus on lexical semantics and relationships between word forms.
• Discussed WordNet and its hierarchical structure.
• Introduced the concept of Word Sense Disambiguation (WSD).

Key Concepts

Lexical Semantics

• Study of word meanings and relationships.
• Words can have multiple meanings (senses).

Word Sense Disambiguation (WSD)

• Problem of determining which sense of an ambiguous word is used in context.
• Example: The word "bass" can refer to a fish or a musical term depending on context.
• Important for natural language processing (NLP).

Challenges in WSD

• Ambiguity arises when a word has multiple meanings.
• Hard to apply disambiguation methods directly without context.

Approaches to WSD

Knowledge-Based Approaches

1. Overlap-Based Methods

   • Use machine-readable dictionaries (like WordNet) to identify overlaps between senses and context.
   • Construct sense bags (sense packs) and context bags for comparison.

2. Example of Sense Disambiguation Process

   • Given a sentence, identify all possible senses of a word.
   • Construct sense bags for each sense and a context bag from surrounding words.
   • Measure overlap and choose sense with the highest overlap.

Algorithms

• Lesk Algorithm: Measures overlap between context and sense definitions.
  • Steps:
    1. Create sense bags for each sense based on definitions.
    2. Create a context bag from the sentence.
    3. Calculate overlaps and select the sense with the highest overlap.

Graph-Based Methods

• Using PageRank-like algorithms to consider multiple words in a sentence together.
• Create a graph where nodes represent senses and edges represent similarities.
• Use PageRank to find the best sense connections among ambiguous words.

Machine Learning Approaches

1. Naive Bayes Classifier: A probabilistic model for classifying senses based on features.

   • Features might include grammatical and semantic characteristics of surrounding words.
   • Computes probabilities to find the most likely sense based on context.

2. Decision List Classifier: A list of collocations that associate specific senses with specific contexts.

   • Uses log likelihood ratios to decide which sense is used based on surrounding words.

Summary

• WSD is an important problem in NLP with various approaches including knowledge-based and machine learning methods.
• Different algorithms and techniques have been developed to tackle the challenges of disambiguating word senses based on context.
• Upcoming lectures will explore additional approaches and techniques in WSD.