A Knowledge-Based Approach to Syntactic Disambiguation of Biomedical Noun Compounds
Ramakanth Kavuluru, Daniel R. Harris · International Conference on Computational Linguistics · 2012
A Noun compounds (NCs) provide a convenient way of communicating complex biomedical concepts in natural language. NewNCs evolvewith scientific progress in various fields and are often not included in standard dictionaries. Thus, semantic analysis of NCs is an important task in applications including ontology alignment, semantic data integration, information extraction, and question answering. A first step in such analysis is the syntactic grouping or bracketing of the constituent nouns. The state-of-the-art in bracketing is mostly limited to compounds with three nouns using empirical studies involving corpora like the Web or Medline biomedical research article citations. Here, we present an alternative knowledge-based approach using the Unified Medical Language System (UMLS) concept labels and definitions for NCs with three or four tokens. Experiments indicate that our method offers comparable accuracy with those that use the Web or Medline for 3-token NCs. Preliminary evaluations with 4-token NCs also point to the potential of our approach to bracketing longer NCs.