This site contains the accompanying supplementary materials for the paper "Analysis Methods in Neural Language Processing: A Survey", TACL 2019, available here.
- Table SM1: A categorization of work trying to find linguistic information in neural networks according to the neural network component investigated, the linguistic property sought, and the analysis method. See Section 2 in the paper.
- Table SM2: A categorization of challenge sets for evaluating neural networks according to the NLP task, the linguistic phenomena, the represented languages, the dataset size, and the construction method. See Section 4 in the paper.
- Table SM3: A categorization of methods for adversarial examples in NLP according to adversary's knowledge (white-box vs. black-box), attack specificity (targeted vs. non-targeted), the modified linguistic unit (words, characters, etc.), and the attacked task. See Section 5 in the paper.
The list of references is available here.
Miss your favorite neural analysis method? Great! Contributions to this site are welcome. Please open a pull request.
If you find this resource useful, please cite our paper:
@Article{belinkov:2019:tacl,
author = {Belinkov, Yonatan and Glass, James},
title = {Analysis Methods in Neural Language Processing: A Survey},
journal = {Transactions of the Association for Computational Linguistics (TACL)},
year = {2019},
volume = {7},
pages = {49--72},
doi = {10.1162/tacl\_a\_00254}
}