Improving Word Sense Disambiguation with Linguistic Knowledge from a Sense Annotated Treebank

Proceedings of Recent Advances in Natural Language Processing, pages 596–603, Hissar, Bulgaria, Sep 7–9 2015. Improving Word Sense Disambiguation with Linguistic Knowledge from a Sense Annotated Treebank Kiril Simov Alexander Popov Linguistic Modeling Department IICT-BAS {kivs|alex.popov|petya}@bultreebank.org Petya Osenova Abstract In this paper we present an approach for the enrichment of WSD knowledge bases with data-driven relations from a gold standard corpus (annotated with word senses, valency information, syntactic analyses, etc.). We focus on Bulgarian as a use case, but our approach is scalable to other languages as well. For the purpose of exploring such methods, the Personal- ized Page Rank algorithm was used. The reported results show that the addition of new knowledge improves the accuracy of WSD with approximately 10.5%. 1 Introduction Solutions to WSD-related tasks usually employ lexical databases, such as wordnets and ontolo- gies. However, lexical databases suffer from sparseness in the availability and density of rela- tions. One approach towards remedying this prob- lem is the BabelNet (Navigli and Ponzetto, 2012), which relates several lexical resources — Word- Net1, DBpedia, Wiktionary, etc. Although such a setting takes into consideration the role of lexical and world knowledge, it does not incorporate con- textual knowledge learned from actual texts (such as collocational patterns, for example). This hap- pens because the knowledge sources for WSD sys- tems usually capture only a fraction of the rela- tions between entities in the world. Many im- portant relations are not present in ontological re- sources but could be learned from texts. One possible approach to handling this sparse- ness issue is the incorporation of relations from sense annotated corpora into the lexical databases. We decided to focus on this line of research, by using the Bulgarian sense annotated treebank 1In this work we used version 3.0 of Princeton WordNet: https://wordnet.princeton.edu/. (Sensed BulTreeBank) in order to extract semantic relations and add them into the lexical resources. The hypothesis that this enrichment would lead to better WSD for Bulgarian was tested in the context of the Personalized PageRank algorithm. The structure of the papers is as follows: the next section discusses the related work on the topic. Section 3 presents the Bulgarian sense an- notated treebank. Section 4 focuses on the Bul- garian Syntactic and Lexical Resources. Section 5 introduces the WSD experiments and results. Sec- tion 6 concludes the paper. 2 Related Work Knowledge-based systems for WSD have proven to be a good alternative to supervised systems, which require large amounts of manually anno- tated training data. In contrast, knowledge-based systems require only a knowledge base and no additional corpus-dependent information. An es- pecially popular knowledge-based disambiguation approach has been the use of popular graph-based algorithms known under the name of ”Random Walk on Graph” (Agirre et al., 2014). Most ap- proaches exploit variants of the PageRank algo- rithm (Brin and Page, 2012). Agirre and Soroa (2009) apply a variant of the algorithm to Word Sense Disambiguation by translating WordNet into a graph in which the synsets are represented as vertices and the …

Content truncated to 500 words