Automatic Discrimination of Genres

The exponential growth in the size of information§resources has motivated research on selecting§information using non-topical criteria, such as§genre. This work assesses the feasibility of§automatically classifying documents by genre using§only adjectives and adverbs as discriminating§features, in particular: (a) trait adjectives, used§by psychologists to assess human personality traits,§and (b) speaker-oriented adverbs, studied by§linguists as markers of narrator attitude. It employs§human intelligence to select features a priori based§on insights from linguistics and psychology. Findings§show that one can build more parsimonious classifiers§whose power is equal to or greater than that of§classifiers which use ad hoc methods.§This work (a) presents a model for the systematic§study of a narrowly-defined set of features, (b)§demonstrates the principle of selecting features§motivated from other disciplines, and (c) uses a new§performance measure called accuracy gain. This work§will benefit researchers and practitioners in§information retrieval, computational linguistics,§communication, and anyone interested in identifying§style and other non-topical characteristics of documents.