Comparison and Combination of Sentence Embeddings Derived from Different Supervision Signals
This work addresses the problem of understanding and improving sentence embeddings for researchers and practitioners in NLP, but it is incremental as it builds on existing methods.
The paper investigates how different supervision signals affect sentence embeddings by comparing two fine-tuning methods on natural language inference and word prediction tasks, and finds that combining them yields substantially better performance on semantic textual similarity and downstream tasks.
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.