Comparison and Combination of Sentence Embeddings Derived from Different Supervision Signals

TL;DR

This paper compares sentence embeddings from models fine-tuned on natural language inference and word prediction tasks, analyzing their properties and demonstrating that combining these methods improves performance on semantic similarity and downstream tasks.

Contribution

It provides a comparative analysis of sentence embeddings based on different supervision signals and shows that their combination enhances performance.

Findings

Combining the two embedding methods improves unsupervised STS performance.

Different supervision signals capture distinct properties of sentence semantics.

The combined approach outperforms individual methods on downstream tasks.

Abstract

There have been many successful applications of sentence embedding methods. However, it has not been well understood what properties are captured in the resulting sentence embeddings depending on the supervision signals. In this paper, we focus on two types of sentence embedding methods with similar architectures and tasks: one fine-tunes pre-trained language models on the natural language inference task, and the other fine-tunes pre-trained language models on word prediction task from its definition sentence, and investigate their properties. Specifically, we compare their performances on semantic textual similarity (STS) tasks using STS datasets partitioned from two perspectives: 1) sentence source and 2) superficial similarity of the sentence pairs, and compare their performances on the downstream and probing tasks. Furthermore, we attempt to combine the two methods and demonstrate that combining the two methods yields substantially better performance than the respective methods on unsupervised STS tasks and downstream tasks.