A transformational-based learner for dependency grammars in discharge summaries

A Transformational-based Learner for Dependency Grammars in Discharge Summaries David A. Campbell M. Phil., Stephen B. Johnson Ph.D. Department of Medical Informatics, Columbia University Abstract NLP systems will be more portable among medical domains if acquisition of semantic lexicons can be facilitated. We are pursuing lexical acquisition through the syntactic relationships of words in medical corpora. Therefore we require a syntactic parser which is flexible, portable, captures head-modifier pairs and does not require a large training set. We have designed a dependency grammar parser that learns through a transformational-based algorithm. We propose a novel design for templates and transformations which capitalize on the dependency structure directly and produces human-readable rules. Our parser achieved a 77% accurate parse training on only 830 sentences. Further work will evaluate the usefulness of this parse for lexical acquisition. 1 Introduction Natural Language is a vital medium in medicine. Health care providers rely on medical narratives for recording, representing and sharing complex medical information such as the description of images, explanation of test results, or the summary of a patient’s hospital visit. Natural Language Processing (NLP) tools have been applied to medical narrative for a variety of applications, such as triggering clinical alerts (Friedman, 1997) and document classification (Wilcox, 2000). The effort required to create and maintain NLP systems in the medical setting can be prohibitive. Most language processors require a domain-specific semantic lexicon to function and, so far, these lexica have been created manually. The time and cost involved in creating these knowledge structures put limits on the extensibility and portability of NLP systems (Hripcsak, 1998). One solution to this bottleneck is to use machine learning to assist in categorizing lexemes into semantic classes. Such a tool could reduce the difficulty in porting NLP systems from one domain to another. 2 Dependency Grammars One approach to semantic categorization is the use of syntactic features (Kokkinakis, 2001). This is based on the assumption that lexemes that share similar syntactic relations to other lexemes in the corpus will be semantically similar (Dorr, 2000). The idea of clustering words based on syntactic features has been well investigated in general language (Pereira, 1993; Li, 1998) However, (Harris, 1991) states that the syntactic relationships are more well-defined and have less variation in scientific languages (sublanguages), such as the ones used in medical texts. Identifying word classes using syntactic relationships should be simpler and potentially more useful in these types of languages. Dependency grammars (Hudson, 1991) generate parses where words in a sentence are related directly to the word which is its syntactic head. Each word, except for the root has exactly one head, and the structure is a tree. The analysis does not generate any intermediate syntactic structures. Figure 1 shows an example of a sentence with a dependency grammar parse. There has been interest in learning dependency grammars from corpora. Collins (Collins, 1996) used dependencies as the backbone for his probabilistic parser and there has been work on learning both probabilistic (Carroll, 1992; Lee, 1999; Paskin, 2001) and …

Content truncated to 500 words