Semantic textual similarity (STS) is a natural language processing (NLP) task that quantitatively assesses the semantic similarity between two text snippets. STS is a fundamental NLP task for many text-related applications, including text de-duplication, paraphrase detection, semantic searching, and question answering. Measuring STS is a machine learning (ML) problem, where an ML model predicts a value that represents the similarity of the two input texts.
This project catalogues the datasets annotated for semantic textual similarity. Then we perform a comparative evaluation of several deep learning based STS methods in these datasets. This site presents the datasets and pre-trained models evaluated in the Multilingual Semantic Textual Similarity project.
We would like to acknowledge following researchers who helped us with the error analysis.
- Claudia Schmitt Baeza
- Ana Isabel Cespedosa
- Maria Ferragud Ferragud
- Marie Escribe
- Marie Picot
- Rocío Caro Quintana
- Chrisse Amelie Soukai
If you use these resources, please cite (and read!) our paper:
Coming soon