|
We call a `domain theory' a collection of generalisations characterising what does and does not typically happen in some application domain: a type of `commonsense understanding'. We use domain theories to carry out disambiguation in natural language processing by characterising some interpretations of a sentence as unlikely or impossible. Building domain theories by hand is impractical except for tiny domains: we describe a method of automatically inducing such theories from disambiguated corpora using a combination of statistical and inductive logic programming techniques. The talk briefly describes earlier work with the ATIS corpus, using the resulting theory for syntactic disambiguation, and in more detail, work in progress with the Penn Treebank aimed at reference resolution.
|
Informatics Forum, 10 Crichton Street, Edinburgh, EH8 9AB, Scotland, UK
Tel: +44 131 651 5661, Fax: +44 131 651 1426, E-mail: school-office@inf.ed.ac.uk Please contact our webadmin with any comments or corrections. Unless explicitly stated otherwise, all material is copyright © The University of Edinburgh |
