Shallow Semantics for Automatic Text Summarization


Most automatic text summarization systems operate by selecting sentences from the source documents. Consequently, the task of improving summaries generated in such a way requires developing methods to make better choices of what sentences should be included in the final summary.  Another stream of text summarization utilizes some form of query to request particular information be present in the summary. In such a case, the selection of sentences would have to take into account the words in the query.  In this presentation, I will examine the degree in which the use of semantic resources can assist a sentence selection task in query based summarization.  Semantic information could provide details pertaining to the function of particular sentences as well as the type of information contained within it.  This work attacks this problem from two angles.  The first is to compare semantic properties of sentences included in human written summaries with those written by a variety of automatic systems.  The second is to add heuristics based on semantic information to an existing summarization system to test for improvement using at least two automated evaluation measures.