To ask "What's in a word?" presupposes a great deal, including the assumption that things such as "words" exist as psychologically real mental entities. This seems like a reasonable assumption; it would appear trivially obvious that to know a language, we must know the words of that language. However, in this talk, I want to suggest a different way of thinking about this kind of knowledge.
Virtually every theory of language assumes the existence of a mental lexicon, within which word-like representations are stored. The content of the representations varies from theory to theory, as does the form (dictionary-like entries in many symbolic theories; nodes and weights in most connectionist models). But despite these differences, most theorists subscribe to long-standing and honorable division of labor between a type of knowledge that is implemented via processes (read, rules or grammar) and the kind of knowledge that is implemented via lists (read, the lexicon).
In recent years, the information that is assumed to be packed into the lexicon has grown significantly. The role of context in modulating the interpretation of words has also become increasingly apparent. Indeed, there exists now an embarrassment of riches which threatens the representational capacity of the lexicon. In this talk I will review some of these results, including recent experimental work from adult psycholinguistics and child language acquisition, and suggest that the concept of a lexicon may be stretched to the point where it is useful to consider alternative ways of capturing the knowledge that language users have of words. I will propose that rather than thinking of words as static representations that are subject to mental processing --operands, in other words --they might be better understood as operators, entities that operate directly on mental states in a lawful and predictable manner. This shift from words as operands to words as operators offers insights into a number of phenomena that I will discuss at the end of the talk.