Syntactic Query Models for Restatement Retrieval

We consider the problem of retrieving sentence level restatements. Formally, we define restatements as sentences that contain all or some subset of information present in a query sentence. Identifying restatements is useful for several applications such as multi-document summarization, document provenance, text reuse and novelty detection. Spurious partial matches and term dependence become important issues for restatement retrieval in these settings. To address these issues, we focus on query models that capture relative term importance and sequential term dependence. In this paper, we build query models using syntactic information such as subject-verb-objects and phrases. Our experimental results on two different collections show that syntactic query models are consistently more effective than purely statistical alternatives.