An Environment for Named Entity Recognition and Translation

Proceedings of the 13th Annual Conference of the EAMT, pages 88–95, Barcelona, May 2009 An Environment for Named Entity Recognition and Translation Filip Graliński Adam Mickiewicz University filipg@amu.edu.pl Krzysztof Jassem Adam Mickiewicz University jassem@amu.edu.pl Michał Marcińczuk Wrocław University of Technology marcinczuk@gmail.com Abstract We present an environment for the recog- nition and translation of Named Entities (NEs). The environment consists of a new formalism for the Named Entity Recognition and Translation (NERT), a parsing mechanism that reads the rules, recognizes Named Entities in given texts and suggests their translation, as well as a set of tools for the evaluation. We sug- gest a method for the evaluation of (sets of) NERT rules that uses raw (not anno- tated) bilingual corpora. 1 Introduction The practical goal of our studies has been to de- velop a mechanism for correct processing of Named Entities in Machine Translation (MT) systems. Vilar et al (2006) claim that incorrect recognition of NE is responsible for approxi- mately 10% of errors made by MT programs. The authors’ experience in the field of MT (e.g. Jassem, 2004; Junczys-Dowmut and Graliński, 2007) tells that incorrect treatment of Named En- tities is responsible for most serious errors made by MT programs (i.e. errors that make output in- comprehensible to human readers). The research aims at improving the quality of Machine Trans- lation by finding a robust solution for processing NEs in MT systems. The paper is organized as follows: In Section 2 we describe the issue of Named Entity Recognition (NER). In Section 3 we show the importance of robust NER solutions for Ma- chine Translation. In Section 4 we present a for- malism for the description of NERT rules. In Section 5 we show some examples of rules com- patible with the grammar. In Section 6 we dis- cuss the problem of NERT evaluation and sug- gest an evaluation method that does not require a © 2009 European Association for Machine Translation. bilingual corpus to be annotated. We end with the reference to future work in Section 7. 2 Named Entity Recognition Named Entity Recognition consists in automatic determination of continuous fragments of texts (called Named Entities) which refer to informa- tion units such as persons, geographical loca- tions, names of organizations, dates, percentages, amounts of money, locations in texts. A NER module is usually expected to provide a markup on boundaries and types of included NEs. Here is an example of such a markup from Mikheev (1999), cited also in Nadeau (2007): On <Date>Jan 13th</Date>, <Person>John Briggs Jr</Person> contacted <Organization>Wonderful Stockbrockers Inc</Organization> in <Location>New York</Location> and instructed them to sell all his shares in <Organization>Acme</Organization>. NER is recognized as a field of Natural Lan- guage Processing since 1995. The sixth Message Understanding Conference (Grishman and Sund- heim, 1996) is usually considered a starting point of the NER history. First attempts in the field consisted in creation of handcrafted rules (Rau, 1991; Ravin and Wa- cholder, 1996). In recent years, this idea has been driven out by machine learning techniques. …

Content truncated to 500 words