Interoperability between Service Composition and Processing Pipeline: Case Study on the Language Grid and UIMA

International Joint Conference on Natural Language Processing, pages 1052–1056, Nagoya, Japan, 14-18 October 2013. Interoperability between Service Composition and Processing Pipeline: Case Study on the Language Grid and UIMA Mai Xuan Trang, Yohei Murakami, Donghui Lin, and Toru Ishida Department of Social Informatics, Kyoto University Yoshida-Honmachi, Sakyo-Ku, Kyoto, 606-8501, Japan trangmx@ai.soc.i.kyoto-u.ac.jp, {yohei,lindh, ishida}@i.kyoto-u.ac.jp Abstract Integrating language resources is a critical part in building natural language process- ing applications. Processing pipeline and service composition are two approaches for sharing and combining language re- sources. However, each approach has its drawback. While the former lacks con- sideration about property rights of lan- guage resources, the later is not efficient to process and transfer huge amount of data through web services. In this paper we ad- dress the issue of interoperability between two approaches to mutually complement their disadvantages. We show an integra- tion of service composition and processing pipeline, and how the integration can be used to help developers seamlessly build NLP applications. We then present a case study that adopts the integration to inte- grate two representative frameworks: the Language Grid and UIMA. 1 Introduction The creation of language resources (LRs) remains a fundamental activity in the field of language technology. The number of language resources has been increasing year by year. Based on these resources, developers build advanced Natural Lan- guage Processing (NLP) applications (hereafter referred to as the applications) such as Watson and Siri by combining some of these resources. However, it is difficult for developers to collect and combine the most suitable set of language re- sources in order to achieve the developers’ goals. There are two types of language resource coor- dination frameworks supporting developers shar- ing and combining language resources and tools: Framework-based processing pipeline such as GATE (Cunningham et al. 2002) and UIMA (Fer- rucci el al., 2004) and framework-based service composition such as the Language Grid (Ishida, 2006). Interoperability between components in one framework is dealt by defining Common Data Exchange or standard interface for components. For example, UIMA defines Common Analysis Structure as data exchange between components, the Language Grid defines standard interfaces in a ontology for their language services (Hayashi, 2007). Interoperability among formats of two pro- cessing pipeline frameworks UIMA and GATE is explored in (Ide et al., 2009a). This paper ad- dresses the issue of how to bridge the gap between two data structures of common data exchange for- mat. In this work we focus on interoperability of two different types of frameworks. Therefore, this paper realizes interoperability between those two types of frameworks to mutu- ally complement their disadvantages. To this end, we address the following issues: • Integration between two types of frame- works: Service composition and processing pipeline. The integration provides ability to wrap components of one framework as com- ponents of another. This will lead to more language resources and tools becoming avail- able in both frameworks, facilitating the de- velopment process of NLP applications. • A case study of integration two representative frameworks: The Language …

Content truncated to 500 words