Datasets with typos for testing LEA (https://arxiv.org/abs/2307.02912)

Datasets for testing the generalization capacity of LEA in the presence of typos. Link to the paper: https://arxiv.org/abs/2307.02912

Below is a list of raw public datasets and different versions of test splits to which automatically synthetically generated typos have been added by deleting and replacing characters.

Abt-Buy

Amazon-Google

WDC-Computers (small, medium, large and xlarge)

WDC-All (xlarge)

RTE

MRPC

Reference:

Almagro, M., Almazán, E., Ortego, D., & Jiménez, D. (2023, August). LEA: Improving Sentence Similarity Robustness to Typos Using Lexical Attention Bias. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (pp. 36-46).