The role of taxonomy in language engineering

Abstract

To achieve systems that successfully process real–life written and spoken language, a prime need is for taxonomic work that introduces the kind of order and well–definedness into the diversity of linguistic forms and structures that the Linnaean system introduced into the biological realm. The current emphasis on statistically based techniques brings this need into particularly sharp focus. To date, the emphasis within the discipline has tended to be different from this. Computational linguistics has focused more on developing parsing software, for instance, than on defining the target analyses that a parser hits or misses. Enterprises that have discussed standards for language engineering have tended to list sets of approved categories, rather than to specify precise boundaries between categories. There is an analogy with the history of software engineering. Creating software is a relatively glamorous activity, but good software is achieved by postponing the glamour in favour of the plodding work of refining specifications.