D2CSE: Difference-aware Deep continuous prompts for Contrastive Sentence Embeddings

TL;DR

D2CSE introduces a novel, memory-efficient method for learning highly distinguishable sentence embeddings using continuous prompts, outperforming existing models on semantic similarity benchmarks.

Contribution

It proposes a difference-aware deep continuous prompt approach that avoids multiple PLMs, reduces training parameters, and enhances sentence embedding quality.

Findings

Outperforms state-of-the-art on seven STS benchmarks.

Reduces training parameters to about 1% of existing methods.

Improves embedding space quality measured by multiple metrics.

Abstract

This paper describes Difference-aware Deep continuous prompt for Contrastive Sentence Embeddings (D2CSE) that learns sentence embeddings. Compared to state-of-the-art approaches, D2CSE computes sentence vectors that are exceptional to distinguish a subtle difference in similar sentences by employing a simple neural architecture for continuous prompts. Unlike existing architectures that require multiple pretrained language models (PLMs) to process a pair of the original and corrupted (subtly modified) sentences, D2CSE avoids cumbersome fine-tuning of multiple PLMs by only optimizing continuous prompts by performing multiple tasks -- i.e., contrastive learning and conditional replaced token detection all done in a self-guided manner. D2CSE overloads a single PLM on continuous prompts and greatly saves memory consumption as a result. The number of training parameters in D2CSE is reduced to about 1\% of existing approaches while substantially improving the quality of sentence embeddings. We evaluate D2CSE on seven Semantic Textual Similarity (STS) benchmarks, using three different metrics, namely, Spearman's rank correlation, recall@K for a retrieval task, and the anisotropy of an embedding space measured in alignment and uniformity. Our empirical results suggest that shallow (not too meticulously devised) continuous prompts can be honed effectively for multiple NLP tasks and lead to improvements upon existing state-of-the-art approaches.