Book Reviews: Natural Language Processing: A Paninian Perspective

This book is an elementary introduction to natural language processing, but it is a very unusual one for two reasons. First, its subject languages are those of India; second, although the book sets out and discusses the better-known linguistic grammars that are commonly used in natural language processing, the grammar that the book actually uses is one based on the Astfidhy~y~ (pronotmced 'ashtad'yayee'), humankind's earliest extant grammar. To begin with, some background with regard to this grammar is in order. The Astfidhy~y~ is a grammar of the Sanskrit language, as it was spoken by the descendants of the Indo-Aryan tribe living on the Gangetic plain around the sixth century B.C. The grammar, written or compiled by P~inini, contains a little over 4,000 rules. Sanskrit, like its European cousin Latin, is a language in which the most salient grammatical device is inflection, while the least salient one is word order, and Panini's grammar reflects this. The grammar is designed so that, together with a Sanskrit lexicon and a canonical specification of a situation, it is supposed to generate all and only the correct Sanskrit sentences expressing the situation. The assumption made is that each situation to be expressed can be framed as an action or event (kriyd) with associated factors (h/raka), roughly valences. Once a one-to-one correspondence has been established between the elements of a situation canonically framed, on the one hand, and underived root morphemes from the lexicon, on the other, the grammar operates nondeterministically to yield a set of morphologically well-formed words that together express the situation. Up until the latter half of the twentieth century, Panini's grammar was the most rigorous and comprehensive ever written for any language. Moreover, many, if not most, of the key ideas in modern linguistic theory have their origin in the ideas of his grammar (e.g., sandhi) or were anticipated by it (e.g., the theta criterion). The tradition that the grammar initiated made Sanskrit, until this century, the most thoroughly studied human language. Though the languages of the Indian subcontinent belong to two distinct language families, Indo-European (those descended from Sanskrit) and Dravidian, they have much in common: in particular, they tend to be highly inflected. Hence, the treatment of inflection, and not word order, must play the most important role in the processing of such languages. English is a language in which inflection has only a marginal role, while word order has …