Interpreting Pretrained Contextualized Representations via Reductions to Static Embeddings
Open accessThe DOI matches its Crossref record; 31 of the 31 references it cites were found in a registry or an index.
Abstract
Contextualized representations (e.g.ELMo, BERT) have become the default pretrained representations for downstream NLP applications.In some settings, this transition has rendered their static embedding predecessors (e.g.Word2Vec, GloVe) obsolete.As a side-effect, we observe that older interpretability methods for static embeddings -while more mature than those available for their dynamic counterparts -are underutilized in studying newer contextualized representations.Consequently, we introduce simple and fully general methods for converting from contextualized representations to static lookup-table embeddings which we apply to 5 popular pretrained models and 9 sets of pretrained weights.Our analysis of the resulting static embeddings notably reveals that pooling over many contexts significantly improves representational quality under intrinsic evaluation.Complementary to analyzing representational quality, we consider social biases encoded in pretrained representations with respect to gender, race/ethnicity, and religion and find that bias is encoded disparately across pretrained models and internal layers even for models that share the same training data.Concerningly, we find dramatic inconsistencies between social bias estimators for word embeddings.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1423 · Cited by 54,325
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
Deep Contextualized Word Representations
Peters, M., Neumann, M., Iyyer, M., Gardner, M., Clark, C., Lee, K., & Zettlemoyer, L. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers) · doi:10.18653/v1/n18-1202 · Cited by 11,983
Enriching Word Vectors with Subword Information
Bojanowski, P., Grave, E., Joulin, A., & Mikolov, T. (2017) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00051 · Cited by 9,645 · cited 1×
“…from integrating additional information such as syntax (Levy and Goldberg, 2014a; Li et al., 2017), morphology (Cot-terell and Sch¨utze, 2015), subwords (Bojanowski et al., 2017), subcharacters (Stratos, 2017; Yu et al., 2017) and, most recently, context (Peters et al., 2018; Devlin et al., 2019).”
BackgroundXLNet: Generalized Autoregressive Pretraining for Language Understanding
Yang, Z., Dai, Z., Yang, Y., Carbonell, J., Salakhutdinov, R., & Le, Q. V. (2019) · arXiv · doi:10.48550/arxiv.1906.08237 · Cited by 6,687
A unified architecture for natural language processing: deep neural networks with multitask learning
Collobert, R., & Weston, J. (2008) · Proceedings of the 25th international conference on Machine learning - ICML '08 · doi:10.1145/1390156.1390177 · Cited by 5,058 · cited 1×
“Word embeddings (Bengio et al., 2003; Collobert and Weston, 2008; Collobert et al., 2011) have been a hallmark of modern natural language processing (NLP) for many years.”
BackgroundDistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019) · arXiv · doi:10.48550/arxiv.1910.01108 · Cited by 4,653
Energy and Policy Considerations for Deep Learning in NLP
Strubell, E., Ganesh, A., & McCallum, A. (2019) · arXiv · doi:10.48550/arxiv.1906.02243 · Cited by 3,735
Energy and Policy Considerations for Deep Learning in NLP
Strubell, E., Ganesh, A., & McCallum, A. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1355 · Cited by 3,684 · cited 9×
“Bias is a complex and highly relevant topic in developing representations and models in NLP and ML.”
BackgroundInfluentialSemantics derived automatically from language corpora contain human-like biases
Caliskan, A., Bryson, J. J., & Narayanan, A. (2017) · Science · doi:10.1126/science.aal4230 · Cited by 2,887
BERTScore: Evaluating Text Generation with BERT
Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., & Artzi, Y. (2019) · arXiv · doi:10.48550/arxiv.1904.09675 · Cited by 2,865
Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016) · arXiv · doi:10.48550/arxiv.1607.06520 · Cited by 1,615 · cited 10×
“…certainty regarding the extent to which embeddings are biased, it is impossible to properly interpret the meaningfulness of debiasing procedures (Bolukbasi et al., 2016; Zhao et al., 2018a,b; Sun et al., 2019) as we cannot reliably estimate the bias in the embeddings both before and after the…”
BackgroundMethodInfluentialA primer in BERTology: What we know about how BERT works
Rogers, A., Kovaleva, O., & Rumshisky, A. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00349 · Cited by 1,552
BERT Rediscovers the Classical NLP Pipeline
Tenney, I., Das, D., & Pavlick, E. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1452 · Cited by 1,307 · cited 3×
“…et al., 2018) and BERT (Devlin et al., 2019) has served as the impetus for exciting and diverse interpretability research: Liu et al. (2019a); Tenney et al. (2019a) study what is learned across the layers of these models, Tenney et al. (2019b); Ethayarajh (2019) consider what is learned from…”
BackgroundResultWord embeddings quantify 100 years of gender and ethnic stereotypes
Garg, N., Schiebinger, L., Jurafsky, D., & Zou, J. (2018) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1720347115 · Cited by 1,065 · cited 4×
“Social bias in NLP has been primarily evaluated in three ways: (a) using geometric similarity between embeddings (Bolukbasi et al., 2016; Garg et al., 2018; Manzini et al., 2019), (b) adapting psychological association tests (Caliskan et al., 2017; May et al., 2019), and (c) considering down-stream…”
BackgroundMethodInfluentialAn Analysis of Deep Neural Network Models for Practical Applications
Canziani, A., Paszke, A., & Culurciello, E. (2016) · arXiv · doi:10.48550/arxiv.1605.07678 · Cited by 969 · cited 1×
“Future research may consider downstream gains as improved static embeddings are critical for resource-constrained settings and may help address environmental concerns in NLP (Strubell et al., 2019), machine learning (Canziani et al., 2016), and the broader AI community (Schwartz et al., 2019).”
BackgroundEthayarajh, K. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1006 · Cited by 841
Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraints
Zhao, J., Wang, T., Yatskar, M., Ordonez, V., & Chang, K.-W. (2017) · Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d17-1323 · Cited by 829 · cited 1×
“…et al., 2019), (b) adapting psychological association tests (Caliskan et al., 2017; May et al., 2019), and (c) considering downstream behavior (Zhao et al., 2017, 2018a, 2019a; Stanovsky et al., 2019).8 Our bias evaluation is in the style of (a) and we consider multi-class social bias in the…”
BackgroundGender bias in coreference resolution: Evaluation and debiasing methods
Zhao, J., Wang, T., Yatskar, M., Ordonez, V., & Chang, K.-W. (2018) · Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers) · doi:10.18653/v1/n18-2003 · Cited by 684 · cited 2×
“…regarding the extent to which embeddings are biased, it is impossible to properly interpret the meaningfulness of debiasing procedures (Bolukbasi et al., 2016; Zhao et al., 2018a,b; Sun et al., 2019) as we cannot reliably estimate the bias in the embeddings both before and after the procedure.”
BackgroundUniversal Adversarial Triggers for Attacking and Analyzing NLP
Wallace, E., Feng, S., Kandpal, N., Gardner, M., & Singh, S. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1221 · Cited by 642
(2019) · doi:10.18653/v1/n19-1419 · Cited by 601
What BERT Is Not: Lessons from a New Suite of Psycholinguistic Diagnostics for Language Models
Ettinger, A. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00298 · Cited by 511 · cited 1×
“…(2019) consider what is learned from context, Clark et al. (2019); Michel et al. (2019) look at specific attention heads, Hewitt and Manning (2019); Ettinger (2020) address linguistic understanding such as syntax and negation, and Wallace et al. (2019); Tan and Celis (2019) address ethical…”
BackgroundMitigating Gender Bias in Natural Language Processing: Literature Review
Sun, T., Gaut, A., Tang, S., Huang, Y., ElSherief, M., Zhao, J., Mirza, D., Belding, E., Chang, K.-W., & Wang, W. Y. (2019) · Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/p19-1159 · Cited by 492 · cited 1×
“…regarding the extent to which embeddings are biased, it is impossible to properly interpret the meaningfulness of debiasing procedures (Bolukbasi et al., 2016; Zhao et al., 2018a,b; Sun et al., 2019) as we cannot reliably estimate the bias in the embeddings both before and after the procedure.”
BackgroundMoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distance
Zhao, W., Peyrard, M., Liu, F., Gao, Y., Meyer, C. M., & Eger, S. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1053 · Cited by 490 · cited 5×
“is not uniformly distributed throughout model layers and practitioners often do not use the last layer of deep Transformer models (Liu et al., 2019a; Zhang et al., 2020; Zhao et al., 2019b).”
BackgroundMethodResultInfluentialOn Measuring Social Biases in Sentence Encoders
May, C., Wang, A., Bordia, S., Bowman, S. R., & Rudinger, R. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1063 · Cited by 461 · cited 1×
“…geometric similarity between embeddings (Bolukbasi et al., 2016; Garg et al., 2018; Manzini et al., 2019), (b) adapting psychological association tests (Caliskan et al., 2017; May et al., 2019), and (c) considering down-stream behavior (Zhao et al., 2017, 2018a, 2019a; Stanovsky et al., 2019).”
BackgroundGonen, H., & Goldberg, Y. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1061 · Cited by 454 · cited 1×
“This is further compounded with the existing evidence that current intrinsic measures of social bias may not handle geometric behavior such as clustering (Gonen and Goldberg, 2019).”
BackgroundAre Sixteen Heads Really Better than One?
Michel, P., Levy, O., & Neubig, G. (2019) · arXiv · doi:10.48550/arxiv.1905.10650 · Cited by 404
Designing and Interpreting Probes with Control Tasks
Hewitt, J., & Liang, P. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1275 · Cited by 393 · cited 1×
“Inspired by concerns with probing methods/diagnostic clas-sifiers (Liu et al., 2019a; Hewitt and Liang, 2019) regarding whether learning can be attributed to the classifier and not the underlying representation, we employ an exceptionally simple parameter-free method for converting from…”
MethodMeasuring Bias in Contextualized Word Representations
Kurita, K., Vyas, N., Pareek, A., Black, A. W., & Tsvetkov, Y. (2019) · Proceedings of the First Workshop on Gender Bias in Natural Language Processing · doi:10.18653/v1/w19-3823 · Cited by 348 · cited 3×
“This is the only layer studied in Kurita et al. (2019).”
BackgroundMethodEvaluating the Underlying Gender Bias in Contextualized Word Embeddings
Basta, C., Costa-jussà, M. R., & Casas, N. (2019) · Proceedings of the First Workshop on Gender Bias in Natural Language Processing · doi:10.18653/v1/w19-3805 · Cited by 189 · cited 2×
“Similarly, Basta et al. (2019) considers intrinsic gender bias in ELMo via gender-swapped 7The authors also identified several empirical concerns that draw the meaningfulness of this method into question.”
Background
Cited by
On the Opportunities and Risks of Foundation Models
Bommasani, R., Hudson, D. A., Adeli, E., Altman, R., Arora, S., von Arx, S., Bernstein, M. S., Bohg, J., Bosselut, A., Brunskill, E., et al. (2021) · arXiv · doi:10.48550/arxiv.2108.07258 · Cited by 2,676
Self-citationA primer in BERTology: What we know about how BERT works
Rogers, A., Kovaleva, O., & Rumshisky, A. (2020) · Transactions of the Association for Computational Linguistics · doi:10.1162/tacl_a_00349 · Cited by 1,552
Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color
Abdou, M., Kulmizev, A., Hershcovich, D., Frank, S., Pavlick, E., & Søgaard, A. (2021) · arXiv · doi:10.48550/arxiv.2109.06129 · Cited by 70
Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color
Abdou, M., Kulmizev, A., Hershcovich, D., Frank, S., Pavlick, E., & Søgaard, A. (2021) · Proceedings of the 25th Conference on Computational Natural Language Learning · doi:10.18653/v1/2021.conll-1.9 · Cited by 64
Mitigating Language-Dependent Ethnic Bias in BERT
Ahn, J., & Oh, A. (2021) · Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2021.emnlp-main.42 · Cited by 64
“The first branch of methods is biased space subtraction (Bolukbasi et al., 2016; Manzini et al., 2019; Liang et al., 2020a; Wang et al., 2020; Bommasani et al., 2020).”
Influential
Metadata
- Authors
- Bommasani, R., Davis, K., & Cardie, C.
- Venue
- Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
- Published
- 2020
- Pages
- 4758-4781
- OpenAlex
- W3035102548
- Cited by
- 139
- References
- 31
- FWCI
- 10.78
- Type
- conference paper
- Language
- en
- Open access
- hybrid
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.