January 2020 – March 2020
Coursework for COMP0087: Statistical Natural Language Processing,
UCL,
London, United Kingdom
In this paper we explore various NLP-inspired methods for configuring automated crossword solvers. We consider techniques including word embeddings and recurrent neural networks, and define several models applicable to the problem. To quantitatively compare performances, we then apply our models to solving crossword puzzles from The Guardian. The main finding of our paper is that simple pooling methods (e.g., bag-of-words) are more effective than RNN architectures for crossword solving, due primarily to the typically short lengths of crossword clues.