Sentence Meta-Embeddings for Unsupervised Semantic Textual Similarity
This work improves unsupervised STS for NLP applications, but it is incremental as it adapts existing meta-embedding methods to sentences.
The paper tackles unsupervised Semantic Textual Similarity by ensembling pre-trained sentence encoders into meta-embeddings, achieving a new unsupervised state-of-the-art with gains of 3.7% to 6.4% Pearson's r over single-source systems.
We address the task of unsupervised Semantic Textual Similarity (STS) by ensembling diverse pre-trained sentence encoders into sentence meta-embeddings. We apply, extend and evaluate different meta-embedding methods from the word embedding literature at the sentence level, including dimensionality reduction (Yin and Schütze, 2016), generalized Canonical Correlation Analysis (Rastogi et al., 2015) and cross-view auto-encoders (Bollegala and Bao, 2018). Our sentence meta-embeddings set a new unsupervised State of The Art (SoTA) on the STS Benchmark and on the STS12-STS16 datasets, with gains of between 3.7% and 6.4% Pearson's r over single-source systems.