Master Thesis Metadata Driven Language Resource Sharing

Language resources such as parallel-text messages and Multilingual Expression Tools (METs), which support searching and rewriting messages, are available in many multicultural fields; including schools where foreign students are studying. However, these METs present a common difficulty: the lack of parallel-texts. Thus we aimed to share language resources among METs. Service Oriented Architecture (SOA) has the potential to achieve this. On SOA, METs can search each others' resources which are wrapped as web services , in order to extend their contents. The problem is that, because each language resource has a different categorization criteria, METs cannot accurately pick up the parallel-text which their users need from the shared resources. To solve this problem, we adopt the metadata-driven approach: RDF meta-data are attached to parallel-texts, and a SPARQL queries are attached to each category in order to collect parallel-texts which suits the category. Within this approach, there are two issues to be solved. Architecture which preserves the merits of existing frameworks Previous studies on incorporating metadata technology to SOA spoil some merits of original SOA frameworks. They often standardize services' interfaces to achieve the interface-level interoperability, and implement access control mechanisms to protect the intellectual property rights of providers. We should design architecture that preserves these two achievements. Metadata design process which can reflect implicit categorization models of users To realize metadata-driven data sharing, we need me-tadata vocabularies which cover all essential aspects of parallel-texts. Guidelines for designing vocabularies are already proposed. However, previous guidelines assume that designers clearly know which aspects of resources are important for categorization. But in our case, they are not clear. Thus we should first find a way to capture implicit categorization model of users. In this paper, we propose an architecture and a metadata design process to solve those issues and enable sharing of language resources among METs. For the first issue, we propose an architecture which incorporates a metada-ta repository outside original services. This repository behaves not as a " broker " like previous studies on data integration with SOA, but rather as a " registry ". A " broker " not only executes query sent from MET, but access data sources on behalf of MET. This violates the access control mechanism. In our architecture, METs collect and integrate data by themselves, thus the access control mechanism is preserved. The architecture can also preserve the original interoperability. We add only one argument to the …