Towards Model Driven Architectures for Human Language Technologies

Abstract

Developing multi-purpose Human Language Technologies (HLT) pipelines and integrating them into the large scale software environments is a complex software engineering task. One needs to orchestrate a variety of new and legacy Natural Language Processing components, language models, linguistic and encyclopedic knowledge resources. This requires working with a variety of different APIs, data formats and knowledge models. In this paper, we propose to employ the Model Driven Development (MDD) approach to software engineering, which provides rich structural and behavioral modeling capabilities and solid software support for model transformation and code generation. These benefits help to increase development productivity and quality of HLT assets. We show how MDD techniques and tools facilitate working with different data formats, adapting to new languages and domains, managing UIMA type systems, and accessing the external knowledge bases.