The Balancing ACT: Combining Symbolic and Statistical Approaches to Language

This volume is worth it for the opening paper, Steve Abney's brilliant "Statistical methods and linguistics". Reading this paper in Chris Manning's statistical NLP class at CMU changed the way I think about the field. I believe it should be required reading for *linguists*. In my experience, most computational linguists either don't care about mainstream theoretical linguistics, have evaluated it and dismissed it as useless, or have taken on Abney's arguments; the remainder use logic and formal language theory without statistics. I considered myself in that remainder before reading this paper. Abney argues that Chomsky's original motive for setting up the paradigm of generative grammar though intuited grammaticality judgements was a simple expedient that allowed him to apply the mathematics of his day (automata and formal language theory). Abney dispatches "classical" generative grammar with the finesse that Chomsky showed in dismissing behaviorism. Today, information theory (read Cover and Thomas's excellent book) as applied to natural language (read Manning and Schuetze's excellent book) is the paradigm of choice for the mathematically and computationally savvy linguist. All in all, anyone who is interested in a scientific approach to linguistics should read Abney's paper.