Abstract

Main verb identification is a fundamental task in natural language processing to automatically identify the predicate-verb in an input sentence. It is particularly useful for many applications in Chinese language processing. Most previous studies focused on developing a computational model for identifying the main verbs. However, the definition of main verb should not be overlooked. In our studies to work out a sound specification design, we have found many complicated issues that require more effort than we thought before for a thorough study. We first work on the specification for annotating the main verbs, with special attention to handle those complicated cases, and then we propose an approach to identifying the main verbs by virtue of utilizing chunk information. This approach gives better results than the approach utilizing only part-of-speech information. Finally, the local contextual features are integrated into a Support Vector Machine (SVM) model. This model achieves a promising F score of 92.8%, and can be applied to improve the performance of a Chinese Sentence Breaker by 2.4%.