On inference theories and code theories: Corpus evidence for semantic schemas

This article illustrates a method of studying the evaluative connotations of words and phrases, by studying their most frequent collocates in large corpora. The semantic schemas which can be identi®ed in this way are clusters of preferred lexis and syntax, which often have conventional pragmatic connotations. Examples of individual schemas are given. A broader research project is then proposed to study the connotations of words and phrases in a signi®cant area of social meaning: the many category terms in English for talking about groups of people in terms of the human life cycle (e.g., infant, baby, child, adolescent, teenager, youth, adult, senior citizen). Such analyses contribute to the long-standing debate between `inference theories' and `code theories' of language comprehension, and suggest that more is conventionally encoded in language structure than has often been suggested in recent work. In addition, they explore a central issue of corpus semantics: the relation between stability and variation in textual units.