1989 Reports

Lexical Co-occurrence: The Missing Link

Smadja, Frank A.

Aside from syntax, linguistic knowledge can be separated into two distinct parts, encyclopedic knowledge and dictionary knowledge. Encyclopedic knowledge describes the world whereas the dictionary describes individual word features, thus capturing lexical knowledge. Among the various types of lexical knowledge, one has generally been overlooked and should bring new results in computational linguistics: co-occurrence knowledge. Co-occurrence knowledge stands for the extent to which an item is specified by its environment independently of syntactic or semantic reasons. The basic concept is that of a lexical relation due to Saussure [49]. A lexical relation between two units of language stands for a correlation of common appearance of the two units in the utterances of the language.