Concordia University at the TREC 15 QA Track

In this paper, we describe the system we used for the TREC Question Answering Track. For factoid and list questions two different approaches were exploited: A redundancy-based approach using a modified version of aranea and a parse-tree based unifier. The modified version of aranea essentially uses Google snippets for extracting answers and then projects them to aquaint. The parse-tree based unifier is a linguistic-based approach that chunks candidate sentences syntactically and uses a heuristic measure to compute the similarity of each chunk in a candidate to its counterpart in the question. To answer other types of questions, our system extracts from Wikipedia articles a list of interest-marking terms related to the topic and uses them to extract and score sentences from the aquaint document collection using various interest-marking triggers. We submitted 3 runs using different variations of the system. In the factoid run, the average of our 3 runs is 0.202, for the list, we achieved an average of 0.084, and for the “Other”, we achieved an average F-score of 0.192.