Efficient Open Domain Question Answering With Delayed Attention in Transformer-Based Models

Open Domain Question Answering (ODQA) on a large-scale corpus of documents (e.g. Wikipedia) is a key challenge in computer science. Although Transformer-based language models such as Bert have shown an ability to outperform humans to extract answers from small pre-selected passages of text, they suffer from their high complexity if the search space is much larger. The most common way to deal with this problem is to add a preliminary information retrieval step to strongly filter the corpus and keep only the relevant passages. In this article, the authors consider a more direct and complementary solution which consists in restricting the attention mechanism in Transformer-based models to allow a more efficient management of computations. The resulting variants are competitive with the original models on the extractive task and allow, in the ODQA setting, a significant acceleration of predictions and sometimes even an improvement in the quality of response.