Generate Compressed Sentences with Stanford Typed Dependencies towards Abstractive Summarization

In this paper, we implement sentence generation process towards generate abstractive summarization which is proposed by (Genest and Lapalme, 2010). We simply use Stanford Typed Dependencies1 to extract information items and generate multiple compressed sentences via Natural Language Generation engine. Then we follow LexRank based sentence ranking combined with greedy sentence selection to build final summary. Although the quantitative evaluation based on Rouge metric demonstrates poor performances, we believe that this sentence generation process make important role towards generate abstractive summarization.