Experiments on Term Extraction using Noun Phrase Subclassifications

Proceedings of Recent Advances in Natural Language Processing, pages 746–751, Hissar, Bulgaria, 12-14 September 2011. Experiments on Term Extraction using Noun Phrase Subclassifications Merley S. Conrado Walter Koza Universidad de S˜ao Paulo merleyc@icmc.usp.br kozawalter@opendeusto.es Josuka D´ıaz-Labrador Joseba Abaitua Solange O. Rezende Universidad Nacional de Rosario josuka@deusto.es joseba.abaitua@deusto.es solange@icmc.usp.br Thiago A. S. Pardo Zulema Solana Universidad de Deusto taspardo@icmc.usp.br zsolana@arnet.com.ar Abstract In this paper we describe and compare three approaches for the automatic extrac- tion of medical terms using noun phrases (NPs) previously recognized on medical text corpus in Spanish. In the first ap- proach, as baseline, we extracted all NPs, while for the second and third ones the extraction process is directed to “specific NPs” that are determined on the basis of the syntactic and positional criteria, among others. As contributions (i) we showed that it is possible to extract me- dical terms using “specific NPs”, (ii) new terms were added in the software dictio- nary, and (iii) terms that were not in the reference lists were extracted. For the third contribution, we used the SNOMED CT R⃝ terms lists, aiming at improving the IULA reference lists. 1 Introduction According to Moreno-Sandoval (2009), generally, noun phrases (NPs) correspond to specific terms of a particular domain. The terms can be formed by only a head or a head and complements. Then, the automatic term extraction task was mainly based on the recognition of this kind of phrases. In this paper, automatic extraction experiments for medical term extraction using noun phrases (NPs) previously recognized on medical text cor- pus in Spanish are described and compared. For this task, in a first stage, as baseline, all identified NPs are considered as term candidates, while in the other stages the extraction is directed to “spe- cific NPs” that are determined on the basis of syn- tactic and positional criteria, among others. The novelty of this work is that we are not using pure noun phrases, like many works utilize. In fact, we are using specific NPs, is to say, a subclassifica- tion of phrases. We use the IULA corpus (Bach et al., 1997) of medical texts in Spanish and results are compared with reference lists of unigrams, bi- grams and trigrams. According to the results, (i) we showed that it is possible to extract medical terms using “spe- cific NPs”, (ii) the software dictionary was im- proved with 2,445 new terms, and (iii) other terms that were not in the reference lists were extracted. For the third contribution we used the SNOMED CT R⃝ term lists aiming at improving the IULA re- ference lists. However, it should be mentioned that we detected other expressions that were nei- ther in the reference lists nor in SNOMED CT R⃝, although they could be considered medical terms. In this case, we have to say that new terms are added almost on a daily basis, and it is practically impossible to manually update the terms lists. 2 Term extraction in medicine There are different works about term extraction …

Content truncated to 500 words