Counterfactually Probing Language Identity in Multilingual Models. Srinivasan, A., Govindarajan, V. S., & Mahowald, K. October, 2023. arXiv:2310.18862 [cs]
Paper doi abstract bibtex Techniques in causal analysis of language models illuminate how linguistic information is organized in LLMs. We use one such technique, AlterRep, a method of counterfactual probing, to explore the internal structure of multilingual models (mBERT and XLM-R). We train a linear classifier on a binary language identity task, to classify tokens between Language X and Language Y. Applying a counterfactual probing procedure, we use the classifier weights to project the embeddings into the null space and push the resulting embeddings either in the direction of Language X or Language Y. Then we evaluate on a masked language modeling task. We find that, given a template in Language X, pushing towards Language Y systematically increases the probability of Language Y words, above and beyond a third-party control language. But it does not specifically push the model towards translation-equivalent words in Language Y. Pushing towards Language X (the same direction as the template) has a minimal effect, but somewhat degrades these models. Overall, we take these results as further evidence of the rich structure of massive multilingual language models, which include both a language-specific and language-general component. And we show that counterfactual probing can be fruitfully applied to multilingual models.
@misc{srinivasan_counterfactually_2023,
title = {Counterfactually {Probing} {Language} {Identity} in {Multilingual} {Models}},
url = {http://arxiv.org/abs/2310.18862},
doi = {10.48550/arXiv.2310.18862},
abstract = {Techniques in causal analysis of language models illuminate how linguistic information is organized in LLMs. We use one such technique, AlterRep, a method of counterfactual probing, to explore the internal structure of multilingual models (mBERT and XLM-R). We train a linear classifier on a binary language identity task, to classify tokens between Language X and Language Y. Applying a counterfactual probing procedure, we use the classifier weights to project the embeddings into the null space and push the resulting embeddings either in the direction of Language X or Language Y. Then we evaluate on a masked language modeling task. We find that, given a template in Language X, pushing towards Language Y systematically increases the probability of Language Y words, above and beyond a third-party control language. But it does not specifically push the model towards translation-equivalent words in Language Y. Pushing towards Language X (the same direction as the template) has a minimal effect, but somewhat degrades these models. Overall, we take these results as further evidence of the rich structure of massive multilingual language models, which include both a language-specific and language-general component. And we show that counterfactual probing can be fruitfully applied to multilingual models.},
urldate = {2025-01-17},
publisher = {arXiv},
author = {Srinivasan, Anirudh and Govindarajan, Venkata S. and Mahowald, Kyle},
month = oct,
year = {2023},
note = {arXiv:2310.18862 [cs]},
keywords = {Computer Science - Computation and Language},
}
Downloads: 0
{"_id":"XQF5KoQrRmXd6wm6X","bibbaseid":"srinivasan-govindarajan-mahowald-counterfactuallyprobinglanguageidentityinmultilingualmodels-2023","author_short":["Srinivasan, A.","Govindarajan, V. S.","Mahowald, K."],"bibdata":{"bibtype":"misc","type":"misc","title":"Counterfactually Probing Language Identity in Multilingual Models","url":"http://arxiv.org/abs/2310.18862","doi":"10.48550/arXiv.2310.18862","abstract":"Techniques in causal analysis of language models illuminate how linguistic information is organized in LLMs. We use one such technique, AlterRep, a method of counterfactual probing, to explore the internal structure of multilingual models (mBERT and XLM-R). We train a linear classifier on a binary language identity task, to classify tokens between Language X and Language Y. Applying a counterfactual probing procedure, we use the classifier weights to project the embeddings into the null space and push the resulting embeddings either in the direction of Language X or Language Y. Then we evaluate on a masked language modeling task. We find that, given a template in Language X, pushing towards Language Y systematically increases the probability of Language Y words, above and beyond a third-party control language. But it does not specifically push the model towards translation-equivalent words in Language Y. Pushing towards Language X (the same direction as the template) has a minimal effect, but somewhat degrades these models. Overall, we take these results as further evidence of the rich structure of massive multilingual language models, which include both a language-specific and language-general component. And we show that counterfactual probing can be fruitfully applied to multilingual models.","urldate":"2025-01-17","publisher":"arXiv","author":[{"propositions":[],"lastnames":["Srinivasan"],"firstnames":["Anirudh"],"suffixes":[]},{"propositions":[],"lastnames":["Govindarajan"],"firstnames":["Venkata","S."],"suffixes":[]},{"propositions":[],"lastnames":["Mahowald"],"firstnames":["Kyle"],"suffixes":[]}],"month":"October","year":"2023","note":"arXiv:2310.18862 [cs]","keywords":"Computer Science - Computation and Language","bibtex":"@misc{srinivasan_counterfactually_2023,\n\ttitle = {Counterfactually {Probing} {Language} {Identity} in {Multilingual} {Models}},\n\turl = {http://arxiv.org/abs/2310.18862},\n\tdoi = {10.48550/arXiv.2310.18862},\n\tabstract = {Techniques in causal analysis of language models illuminate how linguistic information is organized in LLMs. We use one such technique, AlterRep, a method of counterfactual probing, to explore the internal structure of multilingual models (mBERT and XLM-R). We train a linear classifier on a binary language identity task, to classify tokens between Language X and Language Y. Applying a counterfactual probing procedure, we use the classifier weights to project the embeddings into the null space and push the resulting embeddings either in the direction of Language X or Language Y. Then we evaluate on a masked language modeling task. We find that, given a template in Language X, pushing towards Language Y systematically increases the probability of Language Y words, above and beyond a third-party control language. But it does not specifically push the model towards translation-equivalent words in Language Y. Pushing towards Language X (the same direction as the template) has a minimal effect, but somewhat degrades these models. Overall, we take these results as further evidence of the rich structure of massive multilingual language models, which include both a language-specific and language-general component. And we show that counterfactual probing can be fruitfully applied to multilingual models.},\n\turldate = {2025-01-17},\n\tpublisher = {arXiv},\n\tauthor = {Srinivasan, Anirudh and Govindarajan, Venkata S. and Mahowald, Kyle},\n\tmonth = oct,\n\tyear = {2023},\n\tnote = {arXiv:2310.18862 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n\n\n\n\n","author_short":["Srinivasan, A.","Govindarajan, V. S.","Mahowald, K."],"key":"srinivasan_counterfactually_2023","id":"srinivasan_counterfactually_2023","bibbaseid":"srinivasan-govindarajan-mahowald-counterfactuallyprobinglanguageidentityinmultilingualmodels-2023","role":"author","urls":{"Paper":"http://arxiv.org/abs/2310.18862"},"keyword":["Computer Science - Computation and Language"],"metadata":{"authorlinks":{}}},"bibtype":"misc","biburl":"https://bibbase.org/zotero/abhishek-p","dataSources":["h7kKWXpJh2iaX92T5"],"keywords":["computer science - computation and language"],"search_terms":["counterfactually","probing","language","identity","multilingual","models","srinivasan","govindarajan","mahowald"],"title":"Counterfactually Probing Language Identity in Multilingual Models","year":2023}