Feature-Based Selection of Dependency Paths in Ad Hoc Information Retrieval

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 507–516, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Feature-Based Selection of Dependency Paths in Ad Hoc Information Retrieval K. Tamsin Maxwell School of Informatics University of Edinburgh Edinburgh EH8 9AB, UK t.maxwell@ed.ac.uk Jon Oberlander School of Informatics University of Edinburgh Edinburgh EH8 9AB, UK j.oberlander@ed.ac.uk W. Bruce Croft Dept. of Computer Science University of Massachusetts Amherst, MA 01003, USA croft@cs.umass.edu Abstract Techniques that compare short text seg- ments using dependency paths (or simply, paths) appear in a wide range of automated language processing applications including question answering (QA). However, few models in ad hoc information retrieval (IR) use paths for document ranking due to the prohibitive cost of parsing a retrieval collection. In this paper, we introduce a flexible notion of paths that describe chains of words on a dependency path. These chains, or catenae, are readily applied in standard IR models. Informative catenae are selected using supervised machine learning with linguistically informed fea- tures and compared to both non-linguistic terms and catenae selected heuristically with filters derived from work on paths. Automatically selected catenae of 1-2 words deliver significant performance gains on three TREC collections. 1 Introduction In the past decade, an increasing number of techniques have used complex and effective syntactic and semantic features to determine the similarity, entailment or alignment between short texts. These approaches are motivated by the idea that sentence meaning can be flexibly captured by the syntactic and semantic relations between words, and encoded in dependency parse tree fragments. Dependency paths (or simply, paths) are compared using techniques such as tree edit distance (Pun- yakanok et al., 2004; Heilman and Smith, 2010), relation probability (Gao et al., 2004) and parse tree alignment (Wang et al., 2007; Park et al., 2011). Much work on sentence similarity using dependency paths focuses on question answering (QA) where textual inference requires attention to linguistic detail. Dependency-based techniques can also be highly effective for ad hoc information retrieval (IR) (Park et al., 2011). However, few path-based methods have been explored for ad hoc IR, largely because parsing large document collections is computationally prohibitive. In this paper, we explore a flexible application of dependency paths that overcomes this difficulty. We reduce paths to chains of words called catenae (Osborne and Groß, 2012) that capture salient semantic content in an underspecified manner. Catenae can be used as lexical units in a reformu- lated query to explicitly indicate important word relationships while retaining efficient and flexible proximity matching. Crucially, this does not require parsing documents. Moreover, catenae are compatible with a variety of existing IR models. We hypothesize that catenae identify most units of salient knowledge in text. This is because they are a condition for ellipsis, in which salient knowledge can be successfully omitted from text (Osborne and Groß, 2012). To our knowledge, this paper is the first time that catenae are proposed as a means for term selection in IR, and where ellipsis is …

Content truncated to 500 words