VerbCROcean: A Repository of Fine-Grained Semantic Verb Relations for Croatian

VERBCROCEAN: A Repository of Fine-Grained Semantic Verb Relations for Croatian Ivan Sekuli´c and Jan ˇSnajder University of Zagreb, Faculty of Electrical Engineering and Computing Text Analysis and Knowledge Engineering Lab Unska 3, 10000 Zagreb, Croatia {ivan.sekulic, jan.snajder}@fer.hr Abstract In this paper we describe VERBCROCEAN, a broad-coverage repository of fine-grained semantic relations between Croatian verbs. Adopting the methodology of Chklovski and Pantel (2004) used for acquiring the English VerbOcean, we first acquire semantically related verb pairs from a web corpus hrWaC by relying on distributional similarity of subject-verb-object paths in the dependency trees. We then classify the semantic relations between each pair of verbs as similarity, intensity, antonymy, or happens-before, using a number of manually-constructed lexico-syntatic patterns. We evaluate the quality of the resulting resource on a manually annotated sample of 1000 semantic verb relations. The evaluation revealed that the predictions are most accurate for the similarity relation, and least accurate for the intensity relation. We make available two variants of VERBCROCEAN: a coverage-oriented version, containing about 36k verb pairs at a precision of 41%, and a precision-oriented version containing about 5k verb pairs, at a precision of 56%. Keywords: verb relations, semantic relations, relation acquisition, Croatian language 1. Introduction Lexico-semantic resources such as WordNet (Miller, 1995) play an important role in many natural language process- ing (NLP) tasks. In particular, for applications that in- volve processing of events and states, resources that model verb semantics are of great importance. A number of such resources have been developed, most notably verb lex- ica FrameNet (Baker et al., 1998) and VerbNet (Kipper- Schuler, 2005), which model the predicate-argument rela- tions. More recently, the interest in textual entailment (Da- gan et al., 2013) has motivated the construction of large- scale verb entailment resources (Pekar, 2006; Szpektor and Dagan, 2008; Hashimoto et al., 2009). VerbOcean (Chklovski and Pantel, 2004) is a broad- coverage repository of fine-grained semantic relations be- tween English verbs. VerbOcean models five semantic relations between verbs: similarity, strength, antonymy, enablement, and happens-before. The resource has been acquired semi-automatically from the web using lexico- syntactic patterns, and covers about 3,500 verbs and 30,000 semantic relations. VerbOcean has been used for many NLP tasks, including event extraction (Mani et al., 2006), paraphrase detection, entailment recognition (Dagan et al., 2010; Mehdad et al., 2011) and contradiction detection (De Marneffe et al., 2008). In this paper we apply the VerbOcean methodology to ex- tract a broad-coverage repository of semantic relations be- tween Croatian verbs. We essentially adopt the procedure of Chklovski and Pantel (2004) and evaluate the quality of the resulting resource on a manually annotated sample of semantic verb relations. Based on our insights, we cre- ate VERBCROCEAN, a freely available repository of fine- grained semantic verb relations for Croatian. We make available two versions of the resource: a coverage-oriented version containing about 36k relations and a precision- oriented version containing about 5k relations. 2. Related Work Only a handful of semantic verb resources for Croatian are publicly available. The Croatian WordNet (CroWN) (Raf- …

Content truncated to 500 words