Between Computational Linguistics and Computation for Linguistics

Khalil Sima'an (University of Amsterdam)


The bulk of research in Computational Linguistics (CL) until the 1990's aimed at providing the computational means for increasingly more complex linguistic theories and at establishing a link with language technology. Most work at that time concentrated on devising formal and computational representations, grammars and algorithms for these linguistic theories, e.g. work on LFG, HPSG. Over the past 2 decades, CL witnessed a radical change in the kind of methodology it employs and in the goals that it aims at. Probabilistic modelling, statistical inference and an (alarmingly) increasing diversity of machine learning methods underly the bulk of current work. For an outsider it might seem that this field is falling pray to methodological frenzy. Inspite of this, we claim that the current situation has a reason and an aim. Furthermore, we believe that the new CL is complementary to linguistics in aims, and possibly even in methodology. In this talk we will dwell on these issues. Specifically we will wonder about the main concept (``the gist") that underlies the current research in CL, on whether the different ``ground elements" of the current CL constitute pieces that fit together as in a puzzle, and on where knowledge of language (and linguistic work) could fit in as the missing piece.

Back to ISCOL'05 homepage