lee@uhccux.uhcc.hawaii.edu (Greg Lee) (09/08/89)
From article <1933@ncrsecp.Copenhagen.NCR.dk>, by larsbc@ncrsecp.Copenhagen.NCR.dk (Lars Ballieu Christensen): )... is it possible, maybe on a )longer term basis, to provide better NW systems by developing )grammars, which are build with higher degree of respect to the )theoretical linguistic description? I don't think it's possible, unless the long term basis is long enough to permit adequate theoretical linguistic descriptions to be discovered. Say a century. One might reasonably expect linguists' descriptions now to provide a good account of the facts of a language, or a fact in a group of languages, and to give a convenient terminology for describing facts. But that's not `theoretical' in the usual sense. )Generally when developing automatic NL system, the goal seems to )be building parsers, which form the most correct analysis of the )language structures on each level of the specific system. But who knows what "the most correct analysis" is? Should verbs in English go with their objects in a `verb phrase', or with their subjects, or with neither? Intonation sometimes suggests the second choice. Tradition and some not altogether conclusive arguments about possible idioms suggest the first. My colleague Stan Starosta has a well-developed theory that makes the third choice. In this and other instances, it seems to me that fashion or convenience more than theory dictates what turns up in current descriptions. Some syntacticians are enthralled with binary branching trees, but for any principled reason? Not so far as I know. )This strategy causes, at least for grammars for danish language, )that the NW grammars, which primarily have been build/specified )by theoretical language scientists, that the theoretical basis of )the language often seems to be neglected. Instead, ad hoc )grammars are build in order to meet the demands for correctness, But that's what linguists are doing, too. Greg, lee@uhccux.uhcc.hawaii.edu