Towards understanding Dutch automatically

Automatic language analysis is a first step towards the understanding
of natural language by computer. In this presentation I show why
language analysis is required, what problems it faces, and how some of
these problems have been solved. In our work, we focus on written Dutch.

In the first part of the presentation, I will motivate why automated syntactic analysis
(which words belong together, and which relations hold among groups of words)
is required if your goal is to understand natural language by computer.
A difficult problem in automated syntactic analysis is constituted by the
inherent ambiguity of natural language. Although this ambiguity is hardly
ever a problem for humans, for computers this problem is rather fundamental.

We present Alpino: an automated syntactic analyzer. In Alpino, ambiguities
are resolved by means of a statistical model (based on Maximum-Entropy).
I will focus on some aspects of the ambiguity resolution model, with special
emphasis on the use of a set of features which characterize the likelihood
that two words are combined in a particular syntactic relationship. Such
features are based on very large automatically analysed corpora (more than
one billion words). We also present formal evaluation results.

About the speaker:

Gertjan van Noord (1961) was born in Culemborg. After obtaining a masters degree (cum laude) in General Linguistics at the university of Leiden he worked on a PhD project with the subject of ‘reversibilty in natural language processing’ at the University of Utrecht. Since 1992 he is affiliated to department of Humanities Computing at the University of Groningen. He currenly holds a professorship at this department. He is on the Executive Board of the Association of Computational Linguistics (currently vice-president elect) and is co-founder of Computational Linguistics in the Netherlands working-group. He served in a number of similar positions (theme leader for the NWO and many other chairs, editorial boards, etc).