Semantic-oriented cross-lingual ontology mapping

Abstract

Ontologies support knowledge discovery, sharing and reuse among people and enable semantic interoperability between computer-based systems. To establish correspondences between knowledge concepts represented in ontologies, ontology mapping is at the heart of dealing with heterogeneity on the semantic web. A great deal of effort has focused on the matching of ontologies that are written in the same natural language and various tools have been developed to facilitate this monolingual ontology matching process. However, as knowledge and knowledge representations are not restricted to the usage of a single natural language, to make use of knowledge bases across natural language barriers, matching tools and techniques must be able to work with ontologies that are written in heterogeneous natural languages. This research identifies key challenges, tools and techniques to support the process of cross-lingual ontology mapping between independent ontologies that are written in diverse natural languages. One approach to cross-lingual ontology mapping (CLOM): the translationbased approach, is to use translation techniques to convert a cross-lingual mapping problem into a monolingual mapping problem which can then be solved via existing monolingual matching tools. However, noise can be introduced during the translation process which leads to poor mapping quality in the subsequent monolingual matching step. This thesis aims to address this challenge faced by translation-based approach to cross-lingual ontology mapping by proposing the concept of appropriate ontology label translation (AOLT). Appropriate translations in the context of cross-lingual ontology mapping are those translations that are most likely to maximise the success of the subsequent monolingual ontology matching step. In particular, this thesis presents two realisations of the AOLT concept, which have been integrated in two SemanticOriented Cross-lingual Ontology Mapping systems: SOCOM and SOCOM++. It is shown through the evaluations of SOCOM and SOCOM++ that the proposed AOLT concept is effective at improving CLOM quality compared to the baseline system. A major contribution of this thesis is the AOLT concept, its demonstration and evaluation. The proposed AOLT concept distinguishes translations that take place for the purpose of cross-lingual ontology mapping and those that take place for the purpose of localisation. This AOLT concept is the first attempt that aims to improve mapping quality in translation-based cross-lingual ontology mapping systems.