Improved Representation Learning for Question Answer Matching

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 464–473, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Improved Representation Learning for Question Answer Matching Ming Tan, Cicero dos Santos, Bing Xiang & Bowen Zhou IBM Watson Core Technologies Yorktown Heights, NY, USA {mingtan,cicerons,bingxia,zhou}@us.ibm.com Abstract Passage-level question answer matching is a challenging task since it requires effec- tive representations that capture the com- plex semantic relations between questions and answers. In this work, we propose a series of deep learning models to address passage answer selection. To match pas- sage answers to questions accommodat- ing their complex semantic relations, un- like most previous work that utilizes a sin- gle deep learning structure, we develop hybrid models that process the text us- ing both convolutional and recurrent neu- ral networks, combining the merits on ex- tracting linguistic information from both structures. Additionally, we also develop a simple but effective attention mechanism for the purpose of constructing better an- swer representations according to the in- put question, which is imperative for bet- ter modeling long answer sequences. The results on two public benchmark datasets, InsuranceQA and TREC-QA, show that our proposed models outperform a variety of strong baselines. 1 Introduction Passage-level answer selection is one of the es- sential components in typical question answering (QA) systems. It can be defined as follows: Given a question and a pool of candidate passages, se- lect the passages that contain the correct answer. The performance of the passage selection task is not only crucial to non-factoid QA systems, where a question is expected to be answered with a se- quence of descriptive text (e.g. the question in Ta- ble 1), but also very important to factoid QA sys- tems, where the answer passage selection step is Question: Does Medicare cover my spouse? Ground-truth answer: If your spouse has worked and paid Medicare taxes for the entire required 40 quarters, or is eligible for Medicare by virtue of being disabled or some other reason, your spouse can receive his/her own medicare benefits. If your spouse has not met those qualifications, if you have met them, and if your spouse is age 65, he/she can receive Medicare based on your eligibility. Another candidate answer: If you were married to a Medicare eligible spouse for at least 10 years, you may qualify for Medicare. If you are widowed, and have not remarried, and you were married to your spouse at least 9 months before your spouse’s death, you may be eligible for Medicare benefits under a spouse provision. Table 1: An example of a question with the ground-truth answer and a negative answer ex- tracted from the InsuranceQA dataset. also known as passage scoring. In factoid QA, if the sentences selected by the passage scorer mod- ule do not contain the answer, it will definitely lead to an incorrect response from the QA system. One central challenge of this task lies in the complex and versatile semantic relations observed between questions and …

Content truncated to 500 words