Improving Statistical Machine Translation in the Medical Domain using the Unified Medical Language system

Improving Statistical Machine Translation in the Medical Domain using the Unified Medical Language System Matthias Eck matteck@cs.cmu.edu Stephan Vogel Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213, USA vogel+@cs.cmu.edu Alex Waibel ahw@cs.cmu.edu Abstract Texts from the medical domain are an important task for natural language processing. This paper investigates the usefulness of a large medical database (the Unified Medical Language System) for the translation of dialogues between doctors and patients using a statistical machine translation system. We are able to show that the extraction of a large dictionary and the usage of semantic type information to generalize the training data significantly improves the translation performance. 1 Introduction Hospitals in the United States have to deal with an increasing number of patients who have no knowledge of the English language. It is not surprising that in this area translation errors can lead to severe problems (Neergard, 2003; Flores et al. 2003). This is one of the main reasons why the medical domain plays an important role in many of the current projects involving natural language processing. Especially many text or speech translation projects include tasks to translate texts or dialogues with medical topics. The goal of this research was the improvement of translation quality in the medical domain using a statistical machine translation system. A statistical machine translation system deduces translation rules from large amounts of parallel texts in the source and target language. The general approach to gather as much training data as possible is usually complicated and expensive. So it is necessary to make use of already available data and databases and it is reasonable to hope that some ideas and special methods could actually improve the performance in limited domains, like the medical domain. The Internet and especially the WWW offers a lot of data related to medical topics. Especially interesting and promising for us was the Unified Medical Language System® (UMLS, 1986-2004) available from the US National Library of Medicine. It provides a vast amount of information concerning medical terms and we extracted information from this database to improve an existent translation system. The paper will first give an introduction into the Unified Medical Language system. We will then point out which parts could be useful for statistical machine translation and later show how the baseline system was actually significantly improved using this data. 2 The Unified Medical Language System 2.1 Introduction The Unified Medical Language System (UMLS, 1986-2004) project was initiated in 1986 by the U.S. National Library of Medicine. It integrates different knowledge sources into one database (e.g. biomedical vocabularies, dictionaries). The goal is to help health professionals and researchers to use biomedical information from these different sources. It is usually updated about 3 or 4 times per year. It consists of three main knowledge repositories, the UMLS Metathesaurus, the UMLS Semantic Network and the SPECIALIST lexicon. Interesting facts about the UMLS, related work and further information can be found in (Lindbergh, 1990; Kashyap, 2003; Brown et al., 2003; Friedman et al., 2001; Zweigenbaum …

Content truncated to 500 words