Mapping Ontologies Using Ontologies: Cross-lingual Semantic Role Information Transfer

Mapping Ontologies Using Ontologies: Cross-lingual Semantic Role Information Transfer Bal´azs Indig1,2, M´arton Mih´altz2, Andr´as Simonyi2,3 1P´azm´any P´eter Catholic University, Faculty of Information Technology and Bionics 2MTA-PPKE Hungarian Language Technology Research Group Pr´ater u. 50/a, 1083 Budapest, Hungary 3Analogy Zrt. Logodi u. 44, 1012 Budapest, Hungary {indig.balazs, mihaltz.marton}@itk.ppke.hu, andras.simonyi@analogy.co Abstract This paper presents the process of enriching the verb frame database of a Hungarian natural language parser to enable the assignment of semantic roles. We accomplished this by linking the parser’s verb frame database to existing linguistic resources such as VerbNet and WordNet, and automatically transferring back semantic knowledge. We developed OWL ontologies that map the various constraint description formalisms of the linked resources and employed a logical reasoning device to facilitate the linking procedure. We present results and discuss the challenges and pitfalls that arose from this undertaking. Keywords: linked resources, ontology, verb argument frames 1. Introduction Semantic role labeling (SRL) – identification of the seman- tic arguments of verbs and their classification into seman- tic (thematic) roles – is an important step in making sense of natural language sentences, enabling further applications such as semantic search, question answering etc. (Palmer et al., 2010). This paper details the process of enriching the verb frame database of a novel, psycholinguistically mo- tivated Hungarian natural language parser (detailed in Sec- tion 1.1.) with semantic role information. We accomplished this by linking the verb frame database to available external linguistic resources such as VerbNet (Schuler, 2014) and WordNet (Fellbaum and Miller, 2006) in order to transfer as much semantic role information as possible. The link- ing was achieved by mapping the different constraint de- scription formalisms of the source and target resources us- ing two OWL ontologies and by employing the Racer OWL reasoner (Haarslev et al., 2012). 1.1. The parser Our parser is based on ANAGRAMMA, a computa- tional text understanding approach which does not fol- low the parsing algorithms that are well-established in lan- guage technology, but uses some of the principles of hu- man sentence processing set forth in (Pl´eh, 1989). This performance-based algorithm processes linguistic input in a parallel way from left to right. Each parsing step uses a trigram window where the first of the three tokens is pro- cessed, sometimes with the help of the two following to- kens. The basic unit of processing is a (written) word. We may also think of the series of input words as a clock signal co- ordinating the work of the processing threads. The first thread is morphological analysis, but for the sake of sim- plicity – and because our focus is on syntactic processing – we conceptualize this process as a monolithic one, which provides the lemma of the input and those linguistic and non-linguistic features which serve as the basis for further processing. Some of these features belong to the so-called “demand” class, which create threads that look for suit- able features that may satisfy them, while others belong to the so-called “supply” class, which may satisfy the …

Content truncated to 500 words