Distributional neural networks for automatic resolution of crossword puzzles

Abstract

Automatic resolution of Crossword Puz-zles (CPs) heavily depends on the qual-ity of the answer candidate lists produced by a retrieval system for each clue of the puzzle grid. Previous work has shown that such lists can be generated using In-formation Retrieval (IR) search algorithms applied to the databases containing previ-ously solved CPs and reranked with tree kernels (TKs) applied to a syntactic tree representation of the clues. In this pa-per, we create a labelled dataset of 2 mil-lion clues on which we apply an innovative Distributional Neural Network (DNN) for reranking clue pairs. Our DNN is com-putationally efficient and can thus take ad-vantage of such large datasets showing a large improvement over the TK approach, when the latter uses small training data. In contrast, when data is scarce, TKs outper-form DNNs.