Cross-Thought for Sentence Encoder Pre-training
This addresses the need for reusable sequence embeddings in large-scale NLP tasks like question answering, representing an incremental improvement over existing pre-training methods.
The paper tackles the problem of pre-training sequence encoders for NLP tasks by introducing Cross-Thought, which trains on short sequences to select useful information for masked word prediction, resulting in outperforming state-of-the-art encoders on question answering and textual entailment and achieving new SOTA on HotpotQA.
In this paper, we propose Cross-Thought, a novel approach to pre-training sequence encoder, which is instrumental in building reusable sequence embeddings for large-scale NLP tasks such as question answering. Instead of using the original signals of full sentences, we train a Transformer-based sequence encoder over a large set of short sequences, which allows the model to automatically select the most useful information for predicting masked words. Experiments on question answering and textual entailment tasks demonstrate that our pre-trained encoder can outperform state-of-the-art encoders trained with continuous sentence signals as well as traditional masked language modeling baselines. Our proposed approach also achieves new state of the art on HotpotQA (full-wiki setting) by improving intermediate information retrieval performance.