Towards Logic-Based Question Answering under Time Constraints
Ingo Glöckner · 2008
Abstract—Using logic for question answering (QA) promises more accurate answers and increased flexibility of querying. However, the computational effort for a deep linguistic analysis of the involved natural language (NL) expressions and the subsequent logical inferencing makes it difficult to leverage this power for real-time QA. The paper attributes this difficulty in part to the popular answer validation paradigm, and it proposes a different solution which avoids parsing of answers at query time. The novel approach, which uses logic for assessing the relevance of extracted passages, allows better concurrency of QA subtasks since logic-based passage filtering can run in parallel with the answer extractors. Moreover the substitutions found in the proofs can guide a direct answer generation. A first implementation of logical passage filtering achieved 43 % F-measure for factual questions of CLEF07, with response times of only a few seconds. These results are especially interesting since neither aggregation (for better filtering) nor parallelization (for reducing latency) were used in the prototype yet.