Higher-order Lexical Semantic Models for Non-factoid Answer Reranking

Lexical semantic models provide robust performance for question answering, but, in general, can only capitalize on direct ev-idence seen during training. For example, monolingual alignment models acquire term alignment probabilities from semi-structured data such as question-answer pairs; neural network language models learn term embeddings from unstructured text. All this knowledge is then used to estimate the semantic similarity between question and answer candidates. We in-troduce a higher-order formalism that al-lows all these lexical semantic models to chain direct evidence to construct indirect associations between question and answer texts, by casting the task as the traversal of graphs that encode direct term associa-tions. Using a corpus of 10,000 questions from Yahoo! Answers, we experimentally demonstrate that higher-order methods are broadly applicable to alignment and lan-guage models, across both word and syn-tactic representations. We show that an important criterion for success is control-ling for the semantic drift that accumu-lates during graph traversal. All in all, the proposed higher-order approach improves five out of the six lexical semantic mod-els investigated, with relative gains of up to +13 % over their first-order variants.