1996 SALS-SIG Seminars
SALS-SIG Research Seminar
Linguistics in an Age of Engineering
Department of Linguistics, University of Sydney
When: Friday, 22nd November 1996
Where: Room E6A357, Macquarie University
Within Human Language Technology research, the 1990s have seen an enormous swing away from (hand-crafted) knowledge-based systems towards the use of Statistical and Corpus-based Natural Language Processing methods. In this talk, I will motivate this approach, showing how concerns with ambiguity, real text, and the knowledge acquisition bottleneck motivate a Statistical NLP approach. But beyond that, I want to discuss the role of linguistics in such a brave new world, and the questionable theoretical bases of various prior work, both within and outside StatNLP. Among the topics I want to touch on are: (I) If you're writing a StatNLP parser, you no longer need a linguist to write the grammar and lexicon. Do you need one at all? What kind of knowledge of language should be being included and how? (II) Part of speech tagging systems are now widespread, but do they have a coherent linguistic (as opposed to statistical) basis? (III) Consideration of data from language variation and change shows that non-categorical approaches and sensitivity to frequency of use are necessary within linguistics for purely scientific reasons. What should a StatNLP-informed linguistics look like?
|Last modified: July 1997|