ITRI - 01 - 10 Generative lexicon meets corpus data : the case of non - standard word uses

There are various ways to evaluate the Generative Lexicon (GL). One is to see to what extent it accounts for what we find in text corpora. This has not previously been done, and this chapter presents a first foray. The experiment looks at the “nonstandard” uses of words found in a sample of corpus data: “nonstandard” is defined as not matching a literal reading of any of the word’s dictionary definitions. For each nonstandard instance we asked whether it could be analysed using GL strategies. Most cases could not. The chapter discusses in detail a number of non-standard uses and presents a model for their interpretation which draws on large quantities of knowledge about how the word has been used in the past. The knowledge is frequently indeterminate between ‘lexical’ and ‘general’, and is usually triggered by collocations rather than a single word in isolation.