Extending a Logic-Based Question Answering System for Administrative Texts

LogAnswer is a question answering (QA) system for German that uses machine learning for integrating logic-based and shallow (lexical) validation features. For ResPubliQA 2009, LogAnswer was adjusted to specifics of administrative texts, as found in the JRC Acquis corpus. Moreover, support for a broader class of questions relevant to the domain was added, including questions that ask for a purpose, reason, or procedure. Results confirm the success of these measures to prepare LogAnswer for ResPubliQA, and of the general consolidation of the system. According to the C@1/Best IR baseline metric that tries to abstract from the language factor, LogAnswer was the third best of eleven systems participating in ResPubliQA. The system was especially successful at detecting wrong answers, with 73% correct rejections.