Automatic Language-Independent Induction of Gazetteer Lists
Diana Maynard, Kalina Bontcheva, Hamish Cunningham
University of Sheffield
Adaptation of existing Information Extraction (IE) systems to new languages and domains is the focus of much current research, but progress is often hindered by the lack of available resources to enable developers to get a new system up and running fast. It has previously been shown that a good set of gazetteer lists can have a vital role here, but creation of lists for a new language or domain can be time-consuming and laborious. In this paper we demonstrate a tool for inducing gazetteer lists from a small set of annotated corpora and creating a baseline IE system. We also describe an extension to this, using bootstrapping techniques in order to generate much larger volumes of noisy training texts. High quality results have been achieved in this way on Hindi, Chinese and Arabic.
lexical resources, language agility, information extraction, language engineering, bootstrapping
|Language(s)||English, Chinese, Arabic, Hindi|