Establishing a Question Answering System for Forensic Texts

Abstract In the field of criminal proceedings a large quantity of textual material is frequently confiscated or secured by criminologists for evaluating and conserving of evidential information or fulfilling any judicial investigation mandate. The search for specific information or finding of correlations between virtually countless documents is currently a time-consuming handcrafted work. The difficulties remain in the identification of evidential documents and valid relations between entities on the one hand and the adherence to time limits and data privacy-protection on the other. In this work, an integrated computational solution developed by the authors for supporting the evaluation process of forensic texts using computer linguistic technologies is outlined. The application framework under construction is designed towards a QA- system and especially being able to solve a specific criminal issue, and visualize issue-centred case-relevant relationships. For this purpose, several state-of-the-art techniques in the fields of text categorization and information/event extraction are analysed with respect to their suitability for the peculiarities of the considered domain. Subsequently, several approaches for solving domain- specific problems are introduced. The results of this study will form the basis for constituent parts of the currently developed framework.