Parsing and Comprehending with Word Experts (A Theory and its Realization)
Steve Small Chuck Rieger University of Maryland
In Rieger ( 1977, 1978), Rieger and Small ( 1979), and Small ( 1980), we advanced a model of natural-language understanding called the Word Expert Parser (WEP). In this model, we treat each word of language as a complex procedural knowledge source -- a "word expert" -- which contains decision pathways reflecting the range of linguistic and other world knowledge about the word necessary to understand the word in a broad variety of contexts. This model's structure rests on our hypotheses that: (1) human knowledge about language is organized primarily as knowledge about words rather than as knowledge about rules; and (2) language understanding is largely the coordination of information exchange among word experts as each diagnoses its own involvement in its linguistic and conceptual environment. Rather than rule application or matching, the WEP's main concern is, therefore, one of mediating the passing of linguistic signals and concept fragments among words experts as each goes about its own comprehension activities.
This model structure derives from our theory that parsing knowledge is inseparably entwined with complex, frequently idiosyncratic word-based world knowledge. It is our belief that language is not suited for description by rules that describe general principles. It is our belief that humans do not understand language by the orderly application of rules, but rather by the controlled interchange of semaphores and concepts among internally complex "experts," one per linguistic unit (that is, word or morpheme).
The Word Expert Parser has reached a level of development that we can now characterize as a good first approximation to the general theory of language we