Semantics and Complexity of Question AnsweringSystems: Towardsa Moore's Law for Natural LanguageEngineering

Abstract

There’s a proliferation of QA systems and NL chat systems on the Weband in intranets (e.g. www.neuromedia.com or www.ask.com). The question arises: how far are these systems from "real" NLU? After all, they seem to follow Eliza. Do they? How do we settle such questions? QAsystems typically comprise of an NLunderstanding systems and a knowledge base/database. Given a user’s query, what is the complexity of retrieving the information from the database? How do we approachthis problem? What does it mean that one query is more complex than the other? NLP is in demand. For this demand to be sustained we need a "Moore’s Law for natural language engineering", so that the business community could understand what is possible when and at what price.