A Curriculum-Based Approach to a QA Roadmap

The QA community is beginning to understand the core problems in the field, and they largely coincide with those of Natural Language Understanding. The difficulty of answering a question by a current QA system is a function of the match or lack of it between the question or its expression and the resources used to answer it, not how difficult it is for a human to answer it. A prominent factor in making a question hard now is not so much in finding an answer but in validating whether a candidate answer is correct. The problem in many ways parallels that of reading comprehension for children, which suggests a graduated approach to developing and evaluating the field. The difficulties faced by QA systems include long-standing issues in computational linguistics, such as anaphora resolution, metonymy etc.; logic-oriented issues such as scope and quantification as introduced by adverbs and articles; structural problems where the answer must be assembled from many sources, as well as reasoning about space, time and numbers. These problem areas are largely orthogonal, and can be introduced progressively with at each step accepted criteria for success.