AugCSE: Contrastive Sentence Embedding with Diverse Augmentations
This work addresses the challenge of developing robust sentence embeddings for NLP applications, though it is incremental as it builds upon existing models.
The authors tackled the problem of creating a general-purpose sentence embedding model by introducing AugCSE, a framework that uses diverse data augmentations with an antagonistic discriminator to handle conflicting contrastive signals, achieving state-of-the-art results on downstream transfer tasks and competitive performance on semantic textual similarity tasks using only unsupervised data.
Data augmentation techniques have been proven useful in many applications in NLP fields. Most augmentations are task-specific, and cannot be used as a general-purpose tool. In our work, we present AugCSE, a unified framework to utilize diverse sets of data augmentations to achieve a better, general purpose, sentence embedding model. Building upon the latest sentence embedding models, our approach uses a simple antagonistic discriminator that differentiates the augmentation types. With the finetuning objective borrowed from domain adaptation, we show that diverse augmentations, which often lead to conflicting contrastive signals, can be tamed to produce a better and more robust sentence representation. Our methods achieve state-of-the-art results on downstream transfer tasks and perform competitively on semantic textual similarity tasks, using only unsupervised data.