We demonstrate the importance of nominalizations for prepositional phrase attachment for biomedical journal articles. We outline several significant features of the GENIA corpus data and compare them to Wall Street Journal Data. We evaluate a heuristics-based approach to PP attachment based on shallow chunking and domain dependent resources. We conclude that the heuristics based approach performs well, is appropriate for shallow levels of text analysis, and can easily be adapted to or used with other techniques, such as a filter after a statistical parse, or as features in a more complex machine learning environment.
展开▼
