Relation- and Phrase-level Linking of FrameNet with Sar-graphs

Relation- and Phrase-level Linking of FrameNet with Sar-graphs Aleksandra Gabryszak, Sebastian Krause, Leonhard Hennig, Feiyu Xu, Hans Uszkoreit LT Lab @ German Research Center for Artificial Intelligence Alt-Moabit 91c, Berlin, Germany {alga02,skrause,lehe02,feiyu,uszkoreit}@dfki.de Abstract Recent research shows the importance of linking linguistic knowledge resources for the creation of large-scale linguistic data. We describe our approach for combining two English resources, FrameNet and sar-graphs, and illustrate the benefits of the linked data in a relation extraction setting. While FrameNet consists of schematic representations of situations, linked to lexemes and their valency patterns, sar-graphs are knowledge resources that connect semantic relations from factual knowledge graphs to the linguistic phrases used to express instances of these relations. We analyze the conceptual similarities and differences of both resources and propose to link sar-graphs and FrameNet on the levels of relations/frames as well as phrases. The former alignment involves a manual ontology mapping step, which allows us to extend sar-graphs with new phrase patterns from FrameNet. The phrase-level linking, on the other hand, is fully automatic. We investigate the quality of the automatically constructed links and identify two main classes of errors. Keywords: Linking linguistic resources, knowledge graphs, relation extraction 1. Introduction Linguistic linked open data (Chiarcos et al., 2013) is the idea and movement of publishing linguistic resources according to the linked data principles (Bizer et al., 2009). A pre- requisite for releasing such data is to identify semantically corresponding elements in distinct datasets. We describe our work on linking two English linguistic resources, one from the area of lexical semantics (FrameNet) and one being a repository of linguistic expressions for knowledge-graph relations (sar-graphs)1. A sar-graph is a graph containing linguistic knowledge at syntactic and lexical semantic levels for a given language and target relation. Sar-graphs (Uszkoreit and Xu, 2013) are a semi-automatically created resource which explicitly links the semantic relations of knowledge graphs (Dong et al., 2014; Lehmann et al., 2015) to the linguistic patterns which can express these relations in natural-language text. The current version of sar-graphs contains syntactic depen- dency relations between content words, word senses, and semantic arguments. The linguistic patterns in sar-graphs are automatically acquired via a pattern discovery method based on distant supervision (Mintz et al., 2009; Krause et al., 2012). Thus sar-graphs can be directly applied to free texts for relation extraction. Early work on lexical-semantics resources has focused on gathering information about individual words and their dif- ferent meanings in varying contexts, the famous example being WordNet (Fellbaum, 1998). Linguistic knowledge resources that go beyond the level of lexical items are scarce and of limited coverage due to significant investment of human effort and expertise required for their construction. FrameNet (Baker et al., 1998) is such a resource and pro- vides fine-grained semantic relations of predicates and their arguments. However, FrameNet does not provide an explicit link to real-world fact types. There is increasing research in automatically creating large- scale linguistic resources, often these have been built on 1http://sargraph.dfki.de top of existing resources. For example, BabelNet (Navigli …

Content truncated to 500 words