Grammars, Inference, Parsing and Transduction

Abstract

At the very beginning of my own work on TTS synthesis, I was looking for a short but accurate introduction to grammars, parsers, and lexicons; an introduction in which the related terminology would be comprehensively explained, with not too many details, but not being too cursory either; an introduction in which the goals and, better still, the coverage limits of grammar models and the efficiency limits of parsers could be perceived.¹ I now know that this research was bound to fail, for even though grammars are often based on simple facts, their implications are particularly hard to appreciate. What is more, linguistic theories often insist on their peculiarities, rather than on their common properties. I thus grasped pieces of information throughout my research and painfully tried to form a coherent view. It was not, however, until I read the inescapable (Gazdar and Mellish, 1989), the didactic (Charniak, 1993) and (Winograd, 1983), the very complete (Sabah, 1989), the rigorous (Miclet, 1984) and the brilliant (and concise) (Gibbon, 1991), and after I had attended the comprehensive introduction to natural language processing courses of Prof. Martin Kay at the ELSNET summer school on prosody (University College of London, 1993) that I began to have deeper insights in this intricate area. The following pages are merely a concise presentation of some basic facts extracted from these references. It does not claim to be complete nor truly accurate, for it inevitably oversimplifies some aspects of grammars.

“Speak English!” said the Eaglet. “I don’t know the meaning of half those long word, and, what’s more, I don’t believe you do either!” Lewis Carroll, Alice’s Adventures in Wonderland