Introduction to Semantic Processing (1)

In this lecture, we introduce fundamental issues in representing the meaning of natural language, and present three meaning representation strategies. We also review the steps that are taken in semantic analysis of utterances.

Issues

1. What are the representational requirements? ("Why?")
   • Question answering (e.g., LUNAR)
   • Database query (e.g, IDA)
   • Machine translation (e.g., KANT)
   • Expert knowledge (e.g., COMPASS)
   • General knowledge (e.g., CYC)
   • examples

2. What is the right grain size (level of semantic primitives)? ("How?")
   • Domain-specific (large grain-size)
   • General, exhaustive decomposition (small grain-size)
   • PROs and CONs

Meaning Representation

• Semantic networks (FrameKit)
  PRO: Extremely flexible and practical
  CON: No built-in theory (roll your own)

• Conceptual Dependency (Knight & Rich, Ch. 10)
  PRO: Well-defined set of universal semantic primitives
  CON: Grain size too small for many practical applications

• Scripts (from Knight & Rich, Ch. 10)
  PRO: Excellent for capturing situational semantics
  CON: Managing script change, abort can be tricky

Semantic Analysis

Knight & Rich, Ch. 15

• Lexical processing

• Sentence-level processing

• Semantic grammars

• Case grammars

• Conceptual Parsing