{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/can-bert-eat-rucola-topological-data-analysis","title":"Can BERT eat RuCoLA? Topological Data Analysis to Explain","arxiv_id":"2304.01680","date":"2023-04-04","proceeding":null,"authors":["Irina Proskurina","Irina Piontkovskaya","Ekaterina Artemova"],"abstract":"This paper investigates how Transformer language models (LMs) fine-tuned for acceptability classification capture linguistic features. Our approach uses the best practices of topological data analysis (TDA) in NLP: we construct directed attention graphs from attention matrices, derive topological features from them, and feed them to linear classifiers. We introduce two novel features, chordality, and the matching number, and show that TDA-based classifiers outperform fine-tuning baselines. We experiment with two datasets, CoLA and RuCoLA in English and Russian, typologically different languages. On top of that, we propose several black-box introspection techniques aimed at detecting changes in the attention mode of the LMs during fine-tuning, defining the LM's prediction confidences, and associating individual heads with fine-grained grammar phenomena. Our results contribute to understanding the behavior of monolingual LMs in the acceptability classification task, provide insights into the functional roles of attention heads, and highlight the advantages of TDA-based approaches for analyzing LMs. We release the code and the experimental results for further uptake.","url_abs":"https://arxiv.org/abs/2304.01680v1","url_pdf":"https://arxiv.org/pdf/2304.01680v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"can-bert-eat-rucola-topological-data-analysis","repo_url":"https://github.com/upunaprosk/la-tda","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"can-bert-eat-rucola-topological-data-analysis","repo_url":"https://github.com/upunaprosk/la-tda/blob/master/5_Head_importance.ipynb","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":null,"task_name":"CoLA"},{"task_slug":"linguistic-acceptability","task_name":"Linguistic Acceptability"},{"task_slug":"text-classification","task_name":"Text Classification"},{"task_slug":"topological-data-analysis","task_name":"Topological Data Analysis"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"cola","method_name":"COLA"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/linguistic-acceptability-on-cola","task":"Linguistic Acceptability","dataset":"CoLA","model":"BERT+TDA","rank_in_archive_order":2,"of":43,"metrics":{"Accuracy":"88.2%","MCC":"0.726"},"uses_additional_data":false},{"leaderboard":"/sota/linguistic-acceptability-on-cola","task":"Linguistic Acceptability","dataset":"CoLA","model":"RoBERTa+TDA","rank_in_archive_order":3,"of":43,"metrics":{"Accuracy":"87.3%","MCC":"0.695"},"uses_additional_data":false},{"leaderboard":"/sota/linguistic-acceptability-on-rucola","task":"Linguistic Acceptability","dataset":"RuCoLA","model":"Ru-RoBERTa+TDA","rank_in_archive_order":1,"of":9,"metrics":{"Accuracy":"85.7","MCC":"0.594"},"uses_additional_data":false},{"leaderboard":"/sota/linguistic-acceptability-on-rucola","task":"Linguistic Acceptability","dataset":"RuCoLA","model":"Ru-BERT+TDA","rank_in_archive_order":3,"of":9,"metrics":{"Accuracy":"80.1","MCC":"0.478"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}