Enriching Frame Representations with Distributionally Induced Senses

Enriching Frame Representations with Distributionally Induced Senses Stefano Faralli†, Alexander Panchenko‡, Chris Biemann‡, Simone Paolo Ponzetto† †Data and Web Science Group, School of Business Informatics and Mathematics, Universit¨at Mannheim, Germany ‡ Language Technology Group, Department of Informatics, Universit¨at Hamburg, Germany {stefano,simone}@informatik.uni-mannheim.de {panchenko,biemann}@informatik.uni-hamburg.de Abstract We introduce a new lexical resource that enriches the Framester knowledge graph, which links Framnet, WordNet, VerbNet and other resources, with semantic features from text corpora. These features are extracted from distributionally induced sense inventories and subsequently linked to the manually-constructed frame representations to boost the performance of frame disambiguation in context. Since Framester is a frame-based knowledge graph, which enables full-fledged OWL querying and reasoning, our resource paves the way for the development of novel, deeper semantic-aware applications that could benefit from the combination of knowledge from text and complex symbolic representations of events and participants. Together with the resource we also provide the software we developed for the evaluation in the task of Word Frame Disambiguation (WFD). Keywords: distributional semantics, linked open data, frame semantics. 1. Introduction Recent years have witnessed an impressive amount of work on the automatic construction of wide-coverage knowl- edge resources. Web-scale information extraction sys- tems like NELL (Carlson et al., 2010) or Knowledge Vault (Dong et al., 2014) can acquire massive amounts of machine-readable knowledge from the Web, whereas projects like DBpedia (Bizer et al., 2009), YAGO (Rebele et al., 2016) or BabelNet (Navigli and Ponzetto, 2012) have turned collaboratively-generated content into large knowl- edge bases. However, all of these resources are entity- centric in that they are primarily built around the notion of entities, as either provided by an external resource (e.g., Wikipedia pages) or automatically discovered from text (e.g., by clustering entity mentions). The entities are most commonly represented by nouns, noun phrases, and named entities. Entities lie at the heart of the Semantic Web and are central to enable semantic technologies on a large scale (Dietz et al., 2017). Besides, they also provide the founda- tion for more complex semantic representations like event templates or frames (Fillmore, 1968), which are in the fo- cus of our interest in this paper. Recent work looked at ways to populate the Linked Open Data (LOD) cloud with wide-coverage information about semantic frames on the basis of Framester (Gangemi et al., 2016a), a frame-centric resource that is meant to act as a hub between several other linguistic resources that are al- ready part of the LOD cloud. In parallel, researchers looked at ways to combine knowledge graphs of this kind with distributional semantics to include human-readable mean- ing representations based on semantic vector spaces within the Semantic Web ecosystem (Faralli et al., 2016). Con- sequently, in this paper we bring these two lines of re- search together and present a new resource that combines a wide-coverage symbolic repository of frames with human- readable distributional semantic representations from text. In Table 3, we show an excerpt from a distributional-based sense inventory, a proto-conceptualization (PCZ) proposed by Faralli et al. (2016) and further …

Content truncated to 500 words