Scalable Attentive Sentence-Pair Modeling via Distilled Sentence Embedding

Attention based models have become the new state-of-the-art in natural language understanding tasks such as question-answering and sentence similarity. Recent models, such as BERT and XLNet, score a pair of sentences (A and B) using multiple cross-attention operations - a process in which each word in sentence A attends to all words in sentence B and vice versa. As a result, computing the similarity between a query sentence and a set of candidate sentences, requires the propagation of all query-candidate sentence-pairs throughout a stack of cross-attention layers. This exhaustive process becomes computationally prohibitive when the number of candidate sentences is large. In contrast, sentence embedding techniques learn a sentence-to-vector mapping and compute the similarity between the sentence vectors via simple elementary operations such as dot product or cosine similarity. In this paper, we introduce a sentence embedding method that is based on knowledge distillation from cross-attentive models, focusing on sentence-pair tasks. The outline of the proposed method is as follows: Given a cross-attentive teacher model (e.g. a fine-tuned BERT), we train a sentence embedding based student model to reconstruct the sentence-pair scores obtained by the teacher model. We empirically demonstrate the effectiveness of our distillation method on five GLUE sentence-pair tasks. Our method significantly outperforms several ELMO variants and other sentence embedding methods, while accelerating computation of the query-candidate sentence-pairs similarities by several orders of magnitude, with an average relative degradation of 4.6% compared to BERT.