Barbecued Opakapaka: Using Semantic Preferences for Ontology Population

Proceedings of Recent Advances in Natural Language Processing, pages 153–159, Hissar, Bulgaria, Sep 7–9 2015. Barbecued Opakapaka: Using Semantic Preferences for Ontology Population Ismail El Maarouf Georgiana Marsic Constantin Or˘asan University of Wolverhampton {i.el-maarouf, georgie, c.orasan}@wlv.ac.uk Abstract This paper investigates the use of seman- tic preferences for ontology population. It draws on a new resource, the Pattern Dic- tionary of English Verbs, which lists se- mantic categories expected in each syn- tactic slot of a verb pattern. Knowledge of semantic preferences is used to drive and control bootstrapped pattern extrac- tion techniques on the EnClueWeb09 cor- pus with the aim of identifying common nouns belonging to twelve semantic types. Evaluation reveals that syntactic patterns perform better than lexical and surface pat- terns, at the same time raising issues about assessing ontology population candidates out of context. 1 Introduction This paper investigates the use of weakly su- pervised techniques driven by semantic prefer- ences from the Pattern Dictionary of English Verbs (PDEV)1 on the task of ontology population. PDEV is the output of Corpus Pattern Anal- ysis (CPA; Hanks, 2004), a technique in corpus lexicography for mapping meaning onto words in text. PDEV (Hanks and Pustejovsky, 2005; Hanks, 2013; El Maarouf et al., 2014) is a new resource which organizes the description of a verb entry according to its main patterns of use. Its ma- jor features are (1) that it only accounts for uses found in a corpus in a bottom-up data-driven ap- proach, and (2) that the analysis focuses on the ac- curate description of word patterns, rather than on the analysis of word meanings in isolation. Ontology population is defined as the automatic identification of the nouns classed under a seman- tic category in the CPA ontology2. 1http://pdev.org.uk/ 2http://pdev.org.uk/#onto This paper describes ontology population tech- niques driven by PDEV semantic preferences ap- plied to a web-scale corpus. The next section de- scribes the resources used in this paper, section 3, the ontology population techniques, and section 4, the evaluation, before concluding in section 5. 2 Resources 2.1 The CPA Ontology PDEV aims to provide a well-founded corpus- driven account of verb meaning, using semantic types (STs) to stand as prototypes for collocational clusters occurring in each clause role. Current CPA practice has shown that the scientific con- cepts from WordNet (Fellbaum, 1998), the most widely used semantic repository in NLP, do not map well onto words as they are actually used. This is partly because folk concepts, and not sci- entific concepts, form the foundation of meaning in natural language (Wierzbicka, 1984). For this reason, the CPA Ontology has been developed for PDEV, and it contrasts with WordNet in the fol- lowing key aspects: (1) WordNet considers each synset (sense) as a node in the ontology while the CPA Ontology connects STs which cover multi- ple senses; thus WordNet synsets are either STs or word senses. (2) WordNet is intuition-based whereas the CPA Ontology is ‘corpus-driven’. The CPA ontology is inspired from the Brandeis Semantic Ontology (Pustejovsky et …

Content truncated to 500 words