Graphene: Semantically-Linked Propositions in Open Information Extraction

Proceedings of the 27th International Conference on Computational Linguistics, pages 2300–2311 Santa Fe, New Mexico, USA, August 20-26, 2018. 2300 Graphene: Semantically-Linked Propositions in Open Information Extraction Matthias Cetto1, Christina Niklaus1, Andr´e Freitas2, and Siegfried Handschuh1 1 Faculty of Computer Science and Mathematics, University of Passau {matthias.cetto, christina.niklaus, siegfried.handschuh}@uni-passau.de 2 School of Computer Science, University of Manchester andre.freitas@manchester.ac.uk Abstract We present an Open Information Extraction (IE) approach that uses a two-layered transforma- tion stage consisting of a clausal disembedding layer and a phrasal disembedding layer, together with rhetorical relation identification. In that way, we convert sentences that present a complex linguistic structure into simplified, syntactically sound sentences, from which we can extract propositions that are represented in a two-layered hierarchy in the form of core relational tuples and accompanying contextual information which are semantically linked via rhetorical relations. In a comparative evaluation, we demonstrate that our reference implementation Graphene outper- forms state-of-the-art Open IE systems in the construction of correct n-ary predicate-argument structures. Moreover, we show that existing Open IE approaches can benefit from the transfor- mation process of our framework. 1 Introduction Information extraction (IE) turns the unstructured information expressed in natural language (NL) text into a structured representation (Jurafsky and Martin, 2009) in the form of relational tuples that consist of a set of arguments and a phrase denoting a semantic relation between them, e.g. ⟨Barack Obama; served as; the 44th President of the US⟩. Traditional IE systems have concentrated on identifying and extracting relations of interest by taking as input the target relations, along with hand-crafted extraction patterns or patterns learned from hand-labeled training examples. As this manual labor scales linearly with the number of target relations, such a supervised approach does not scale to large, heterogeneous corpora which are likely to contain a variety of unanticipated relations. To tackle this issue, Banko et al. (2007) introduced Open IE as a new extraction paradigm that allows for a domain-independent discovery of relations in large amounts of text by not depending on any relation-specific human input. Instead, detecting the relations is part of the problem. State-of-the-art Open IE systems (see Section 2) identify relationships between entities in a sentence by matching patterns over either shallow syntactic features in terms of part-of-speech (POS) tags and noun phrase (NP) chunks or dependency tree structures. However, particularly long and syntactically complex sentences pose a challenge for current Open IE approaches. By analyzing the output of such systems (see Figure 1), we observed three common shortcomings. First, relations often span over long nested structures or are presented in a non-canonical form that cannot be easily captured by a small set of extraction patterns. Therefore, such relations are commonly missed by state-of-the-art approaches. Consider for example the first sentence in Figure 1 which asserts that ⟨Sonia Sotomayor; became; the first Supreme Court Justice of Hispanic descent⟩. This information is encoded in a complex participial construction that is omitted by both reference Open IE systems, OLLIE (Mausam et al., 2012) and ClausIE (Del Corro …

Content truncated to 500 words