This chapter addresses the application of graph-based algorithms to problems in the area of semantics. There has been growing interest in the automatic semantic analysis of text to support natural language processing applications, ranging from machine translation and information retrieval to question answering and knowledge acquisition. Significant research has been carried out in this area, including work on word-sense disambiguation, semantic-role labeling, textual entailment, lexical acquisition, and semantic relations.

The chapter describes synonym detection and automatic construction of semantic classes using measures of graph connectivity on graphs built from either raw text or user-contributed resources; measures of semantic distance on semantic networks, including simple path-length algorithms and more complex random-walk methods; textual entailment using graph-matching algorithms on syntactic or semantic graphs; word-sense disambiguation and name disambiguation, including random-walk algorithms and other structural approaches for knowledge-based word-sense disambiguation, as well as semi-supervised methods using label propagation on graphs; and sentiment classification using semi-supervised graph-based learning or prior subjectivity detection with min-cut/max-flow algorithms.

Semantic Classes

Some of the largest graph representations constructed to support a natural language processing task are perhaps those built from large text collections for unsupervised lexical acquisition (Widdows and Dorow 2002). One of the immediate applications of such large graphs is the construction of semantic classes by automatically extracting from raw corpora all of the elements belonging to a certain semantic category (e.g., “fruits” or “musical instruments.”)