Mention Detection Crossing the Language Barrier

Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing, pages 600–609, Honolulu, October 2008. c⃝2008 Association for Computational Linguistics Mention Detection Crossing the Language Barrier Imed Zitouni and Radu Florian IBM T.J. Watson Research Center 1101 Kitchawan Rd, Yorktown Heights, NY 10598 {izitouni, raduf}@us.ibm.com Abstract While significant effort has been put into an- notating linguistic resources for several lan- guages, there are still many left that have only small amounts of such resources. This paper investigates a method of propagat- ing information (specifically mention detec- tion information) into such low resource languages from richer ones. Experiments run on three language pairs (Arabic-English, Chinese-English, and Spanish-English) show that one can achieve relatively decent perfor- mance by propagating information from a lan- guage with richer resources such as English into a foreign language alone (no resources or models in the foreign language). Fur- thermore, while examining the performance using various degrees of linguistic informa- tion in a statistical framework, results show that propagated features from English help improve the source-language system perfor- mance even when used in conjunction with all feature types built from the source language. The experiments also show that using propa- gated features in conjunction with lexically- derived features only (as can be obtained di- rectly from a mention annotated corpus) yields similar performance to using feature types de- rived from many linguistic resources. 1 Introduction Information extraction is a crucial step toward un- derstanding a text, as it identifies the important con- ceptual objects and relations between them in a dis- course. It includes classification, filtering, and se- lection based on the language content of the source data, i.e., based on the meaning conveyed by the data. It is a crucial step for several applications, such as summarization, information retrieval, data mining, question answering, language understand- ing, etc. This paper addresses an important and basic task of information extraction: mention detection1: the identification and classification of textual refer- ences to objects/abstractions mentions, which can be either named (e.g. John Smith), nominal (the presi- dent) or pronominal (e.g. he, she). For instance, in the sentence President John Smith said he has no comments. there are three mentions: President, John Smith and he. This is similar to the named entity recognition (NER) task with the additional twist of also identi- fying nominal and pronominal mentions. A few languages have received a lot of attention in terms of natural language resources that were cre- ated – for instance, in English one has access to la- beled part-of-speech data, word sense information, parse tree structure, discourse, semantic role labeles, named entity data, to name just a few (our apologies if we missed your favorite resource). There are a few other languages that also have annotated resources (such as Arabic, Chinese, German, French, Spanish, etc), but also a very large number of languages with few resources. It would be very useful if one could make use of the resources in the former languages to help bootstrapping (or just the projection) …

Content truncated to 500 words