MCSE: Multimodal Contrastive Learning of Sentence Embeddings
This work addresses the challenge of sentence embedding quality for natural language processing applications, offering an incremental improvement over existing methods.
The paper tackled the problem of learning semantically meaningful sentence embeddings by proposing a multimodal contrastive learning approach that uses both visual and textual information, resulting in a 1.7% improvement in state-of-the-art average Spearman's correlation across semantic textual similarity tasks.
Learning semantically meaningful sentence embeddings is an open problem in natural language processing. In this work, we propose a sentence embedding learning approach that exploits both visual and textual information via a multimodal contrastive objective. Through experiments on a variety of semantic textual similarity tasks, we demonstrate that our approach consistently improves the performance across various datasets and pre-trained encoders. In particular, combining a small amount of multimodal data with a large text-only corpus, we improve the state-of-the-art average Spearman's correlation by 1.7%. By analyzing the properties of the textual embedding space, we show that our model excels in aligning semantically similar sentences, providing an explanation for its improved performance.