Extracting and enriching workflows from text

Abstract

This paper is on a workflow extraction framework which allows to derive a formal representation based on workflows from textual descriptions of instructions, for instance, of aircraft repair procedures from a maintenance manual. The framework applies a pipes-and-filters architecture and uses NLP (Natural Language Processing) tools to perform information extraction steps automatically. In detail, the paper presents on the step of anaphora resolution to enrich the workflow extracted so far. We introduce a lexical approach and two further approaches based on a set of association rules which are created during a statistical analysis of a corpus of workflows. The results of the approaches are compared to each other. For the evaluation, we use 37 workflows which have been created by a human expert.