Learning to Learn to Disambiguate: Meta-Learning for Few-Shot Word Sense Disambiguation

The success of deep learning methods hinges on the availability of large\ntraining datasets annotated for the task of interest. In contrast to human\nintelligence, these methods lack versatility and struggle to learn and adapt\nquickly to new tasks, where labeled data is scarce. Meta-learning aims to solve\nthis problem by training a model on a large number of few-shot tasks, with an\nobjective to learn new tasks quickly from a small number of examples. In this\npaper, we propose a meta-learning framework for few-shot word sense\ndisambiguation (WSD), where the goal is to learn to disambiguate unseen words\nfrom only a few labeled instances. Meta-learning approaches have so far been\ntypically tested in an $N$-way, $K$-shot classification setting where each task\nhas $N$ classes with $K$ examples per class. Owing to its nature, WSD deviates\nfrom this controlled setup and requires the models to handle a large number of\nhighly unbalanced classes. We extend several popular meta-learning approaches\nto this scenario, and analyze their strengths and weaknesses in this new\nchallenging setting.\n