Merging Sentences using Shallow Semantic Analysis: A First Experiment

Multiple document summarisation involves the selection of important information in a collection of related documents, and the compilation of this information into one summary. This paper addresses a key problem in multiple document summarisation, that of merging two semantically similar sentences. Unlike the case of single document summarisation, when important sentences are extracted from multiple related documents, one is likely to find repetition of content within these extracted sentences. To avoid the resulting summary being repetitive and hence incoherent, there is a need for a merging process to handle such repetition. We present work in progress on an algorithm that takes annotated lexical dependency trees corresponding to an abstraction of the key semantics of the input sentences and produces a representation of their combined content that can be used to generate a new sentence.