Modelling Events through Memory-based, Open-IE Patterns for Abstractive Summarization

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 892–901, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Modelling Events through Memory-based, Open-IE Patterns for Abstractive Summarization Daniele Pighin Google Inc. biondo@google.com Marco Cornolti∗ University of Pisa, Italy cornolti@di.unipi.it Enrique Alfonseca Google Inc. ealfonseca@google.com Katja Filippova Google Inc. katjaf@google.com Abstract Abstractive text summarization of news requires a way of representing events, such as a collection of pattern clusters in which every cluster represents an event (e.g., marriage) and every pattern in the clus- ter is a way of expressing the event (e.g., X married Y, X and Y tied the knot). We compare three ways of extracting event patterns: heuristics-based, compression- based and memory-based. While the for- mer has been used previously in multi- document abstraction, the latter two have never been used for this task. Compared with the first two techniques, the memory- based method allows for generating sig- nificantly more grammatical and informa- tive sentences, at the cost of searching a vast space of hundreds of millions of parse trees of known grammatical utterances. To this end, we introduce a data structure and a search method that make it possible to efficiently extrapolate from every sentence the parse sub-trees that match against any of the stored utterances. 1 Introduction Text summarization beyond extraction requires a semantic representation that abstracts away from words and phrases and from which a summary can be generated (Mani, 2001; Sp¨arck-Jones, 2007). Following and extending recent work in semantic parsing, information extraction (IE), paraphrase generation and summarization (Titov and Klemen- tiev, 2011; Alfonseca et al., 2013; Zhang and Weld, 2013; Mehdad et al., 2013), the represen- tation we consider in this paper is a large collec- ∗Work done during an internship at Google Zurich. [John Smith] and [Mary Brown] wed in [Baltimore]... [Smith] tied the knot with [Brown] this Monday... #21: death #22: divorce #23: marriage PER married PER PER and PER wed PER tied the knot with PER PER has married PER John Smith married Mary Brown e1: J. Smith (PER) e2: M. Brown (PER) e3: Baltimore, MD (LOC) Figure 1: An example of abstracting from input sentences to an event representation and genera- tion from that representation. tion of clusters of event patterns. An abstractive summarization system relying on such a represen- tation proceeds by (1) detecting the most relevant event cluster for a given sentence or sentence col- lection, and (2) using the most representative pat- tern from the cluster to generate a concise sum- mary sentence. Figure 1 illustrates the summa- rization architecture we are assuming in this pa- per. Given input text(s) with resolved and typed entity mentions, event mentions and the most rele- vant event cluster are detected (first arrow). Then, a summary sentence is generated from the event and entity representations (second arrow). However, the utility of such a representation for summarization depends on the quality of pattern clusters. In particular, event patterns must cor- respond to grammatically correct sentences. …

Content truncated to 500 words