Towards a Transformer-Based Reverse Dictionary Model for Quality Estimation of Definitions (Student Abstract)

In the last years, several variants of transformers have emerged. In this paper, we compare different transformerbased models for solving the reverse dictionary task and explore their use in the context of a serious game called The Dictionary Game. In its simplest form, a common language dictionary can be seen as an association between the meaning of a word and its definition. A task related to word sense disambiguation is called the reverse dictionary task. It focuses on the reverse association, i.e. guessing a word from its definition. A model capable of effectively solving this task should capture multiple semantic relationships such as synonymy, polysemy and homography. In order to solve the reverse dictionary task, some studies rely on information retrieval through lexical database (El-Kahlout and Oflazer 2004; Shaw et al. 2013) or on graph-based approaches (Thorat and Choudhari 2016). Recurrent networks have also been shown to be expressive enough to take them into account (