Multilingual Ontology Matching

[About] [Objective] [Example scenario] [Publications]

Ontology matching is a task that has attracted considerable attention in recent years. With very few exceptions, however, research in ontology matching has focused on the development of monolingual matching algorithms. As more and more resources become available in more than one language, novel algorithms are required which are capable of matching ontologies which share more than one language, or ontologies which are multilingual but do not share any languages. In this research we analyze several approaches to learning a matching function between two ontologies using a small set of manually aligned concepts, and evaluate them on different pairs of ontologies, showing that multilingual information can indeed improve the matching quality, even in cross-lingual scenarios.

In the ImREAL use cases we deal with data that comes from different countries across Europe. While the simulators are monolingual, our work is a first step towards the possibility of using multilingual training material or taking into account user experiences expressed in other languages. The grounding of the training material into international datasets can potentially enrich the simulators with knowledge contained in these external international sources.

Heterogeneity of data sources is a common problem that needs to be addressed in any research that requires data integration. Since different User Model services and applications produce information about users and store them in their proprietary formats (generic, XML, RDF or ontologies) often in different languages, data needs to be aggregated into a single format and/or single language.

Possible scenarios for overcoming the problem of heterogeneous user model data are:

• Using generic user model mediation framework with the goal of improving the quality of recommendations (translating the data between different models using inference and reasoning mechanisms).
• User model are standardized so data can be shared easily between applications
• Using top-level ontology and domain ontologies for user models
• All applications should use centralized user modeling system (used by all services and applications)

Services and applications that are used for user modeling work in the same domain but can use different domain ontologies (not just different in their structure, but written in different languages that represents an additional problem that needs to be addressed). In this project, ontology matching techniques are used for setting common understanding of the domain semantics. After the task of ontology matching is finished, services and applications should be able to align their domain models and mediate users information.

Let S be a source ontology and T a target ontology, and the S(l) and T(l) are sets of labels of S and T in a language l, and the LS and LT are sets of languages in S and T respectively.

Given a source ontology S with labels in English, German and Italian, monolingual ontology matching matches entities in S to entities in a target ontology T1 with English labels by comparing the English labels in S with those of T1. Multilingual ontology matching matches entities in S with entities in a target ontology T2 with English and German labels by considering the labels in English and German. Cross-lingual ontology matches entities in S to entities in a target ontology T3 with French labels either by translating the labels of S to French, by translating the labels of the T3 to one of the languages in LS, or by translating the labels of S and T to a third language.

1. Dennis Spohr, Laura Hollink, Philipp Cimiano. Multilingual and Cross-Lingual Ontology Matching and its Application to Financial Accounting Standards. In Proceedings of 10th International Semantic Web Conference (ISWC), Bonn, Germany, October 2011. [Talk and slides on videolectures.net]