Generating Pattern-Based Entailment Graphs for Relation Extraction

Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017), pages 220–229, Vancouver, Canada, August 3-4, 2017. c⃝2017 Association for Computational Linguistics Generating Pattern-Based Entailment Graphs for Relation Extraction Kathrin Eichler, Feiyu Xu, Hans Uszkoreit, Sebastian Krause German Research Center for Artificial Intelligence (DFKI) Berlin, Germany firstname.lastname@dfki.de Abstract Relation extraction is the task of recogniz- ing and extracting relations between enti- ties or concepts in texts. A common ap- proach is to exploit existing knowledge to learn linguistic patterns expressing the target relation and use these patterns for extracting new relation mentions. De- riving relation patterns automatically usu- ally results in large numbers of candi- dates, which need to be filtered to de- rive a subset of patterns that reliably ex- tract correct relation mentions. We ad- dress the pattern selection task by exploit- ing the knowledge represented by entail- ment graphs, which capture semantic rela- tionships holding among the learned pat- tern candidates. This is motivated by the fact that a pattern may not express the target relation explicitly, but still be use- ful for extracting instances for which the relation holds, because its meaning en- tails the meaning of the target relation. We evaluate the usage of both automati- cally generated and gold-standard entail- ment graphs in a relation extraction sce- nario and present favorable experimental results, exhibiting the benefits of structur- ing and selecting patterns based on entail- ment graphs. 1 Introduction The task of relation extraction (RE) is to recog- nize and extract relations among entities or con- cepts mentioned in texts. One common approach to RE is to learn and exploit extraction patterns (e.g., based on syntactic dependency trees), which express the targeted semantic relations. In order to circumvent the manual creation of patterns, nu- merous approaches have been investigated to de- rive patterns automatically. Automatic methods generally induce large numbers of unique candi- date patterns, which only potentially express the target relation and need to be filtered in order to derive a subset of high-quality patterns for the re- lation extraction task. The task of filtering or se- lecting patterns can be tackled in various ways, e.g., based on frequency information, or by apply- ing syntactic or semantic criteria. For many RE applications, such as knowledge base population, patterns are not only relevant if they express the target relation explicitly, but also if they extract facts from which the target relation can be inferred. For example, all patterns below can be utilized for extracting pairs of people who are or were involved in a marriage relation: P1: PERSON1 <marry> PERSON21 P2: PERSON1 <be widow of> PERSON2 P3: PERSON1 <divorce from> PERSON2 However, only pattern P1 expresses the target relation explicitly. Patterns P2 and P3 are seman- tically different from P1, but express a fact that entails the marriage relation2. As being aware of these semantic relationships holding among pat- terns can be of help in the pattern selection pro- cess, we propose to capture and exploit these re- lationships using pattern-based …

Content truncated to 500 words