COOL, a Context Outlooker, and its Application to Question Answering and other Natural Language Processing Tasks

Fangyi Zhu, See-Kiong Ng, Stéphane Bressan

Published 2022-04-01Version 1

Vision outlookers improve the performance of vision transformers, which implement a self-attention mechanism by adding outlook attention, a form of local attention. In natural language processing, as has been the case in computer vision and other domains, transformer-based models constitute the state-of-the-art for most processing tasks. In this domain, too, many authors have argued and demonstrated the importance of local context. We present and evaluate an outlook attention mechanism, COOL, for natural language processing. COOL adds, on top of the self-attention layers of a transformer-based model, outlook attention layers that encode local syntactic context considering word proximity and consider more pair-wise constraints than dynamic convolution operations used by existing approaches. A comparative empirical performance evaluation of an implementation of COOL with different transformer-based approaches confirms the opportunity of improvement over a baseline using the neural language models alone for various natural language processing tasks, including question answering. The proposed approach is competitive with state-of-the-art methods.