Amilcare : adaptive information extraction for document annotation

Abstract

The traditional process of document annotation for knowledge identification and extraction in Knowledge Management (KM) is complex and time consuming, as it requires manual annotation by domain experts. In the typical scenario a domain expert:(1) builds an ontology describing the application domain: (2) annotates a number of texts in order to identify instances of elements (i.e., concepts and relations) in the texts. There is a strong interest in Text Mining technologies (and in particular in Human Language- based Technologies), for reducing the burden of text annotation for KM [1]. In this paper we show how Information Extraction from texts (IE) can provide support for document enrichment and make the text annotation process more effective and efficient. The main challenge to be addressed by IE researchers in this framework is portability of IE systems to new applications with no knowledge of HLT. As a matter of fact, in terms of IE each annotation task (e.g. tagging texts about failures in cars in order to identify faults and involved car parts) requires to port the IE system to a new application domain. IE is just one of the many technologies required by complex KM environments: wide use of IE tools will come only when the definition of such application domain will not require any specific IE skill apart from notions of KM. Moreover there is the need to port across different text types without major recoding of system resources, as documents in KM can range from free texts (technical reports, newspaper-like texts) to structured marked up texts (e.g. highly structured XML/HTML documents) and even a mixture of them [2].