Research on Chinese Sentence Compression for the Title Generation

Automatic Title Generation means generating a title which can show the central information of the original text via natural language processing technologies. One method is by extracting a sentence which represents the original text’s central information and then compressing it to a short sentence as the title, in which the core technology is the sentence compression. But the research of Chinese sentence compression has not carried out, it is mainly facing the following difficulties: lacking of the corpus, suffering from the poor performance of Chinese word segmentation and parsing, and having no unified automatic evaluation metric. This paper realizes a Chinese sentence compression method through simply shorting a sentence by deleting words or constituents which is main practice is by learning a subtree from the source parsing tree of a sentence, and then uses the manual and automatic evaluations to evaluate the sentence compression performance. The experimental results show that the method and evaluation metrics used in this paper are valid and effective.