Modeling UIMA type system using web ontology language: towards interoperability among UIMA-based NLP tools

Abstract

With the recent development and adoption of NLP framework architectures, NLP modules/tools developed independently in the research community can be adopted as integrated applications. Development of wrappers and interfaces required to adopt NLP modules/tools, however, still requires huge amount of efforts. In this paper, we focus on one NLP framework architecture, UIMA (Unstructured Information Management Architecture), which defines annotations as types described in a type system and can achieve direct interoperability if a common type system is used. We explore the use of ontology to model UIMA types and argue existing ontology development or reasoning tools can be utilized to understand types (we use types and annotations interchangeably) from existing NLP systems developed under UIMA, define equivalent annotations in different NLP systems, and apply the practice in the ontology community to draw agreements on the definition of common NLP types, thereby achieving better interoperability among NLP modules/tools.