Bias Identification and Attribution in NLP Models With Regression and Effect Sizes
In recent years, there has been an increasing awareness that many NLP systems incorporate biases of various types (e.g., regarding gender or race) which can have significant negative consequences. At the same time, the techniques used to statistically analyze such biases are still relatively simple...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Linköping University Electronic Press
2022-08-01
|
| Series: | Northern European Journal of Language Technology |
| Online Access: | https://nejlt.ep.liu.se/article/view/3505 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832591211360157696 |
|---|---|
| author | Erenay Dayanik Ngoc Thang Vu Sebastian Padó |
| author_facet | Erenay Dayanik Ngoc Thang Vu Sebastian Padó |
| author_sort | Erenay Dayanik |
| collection | DOAJ |
| description |
In recent years, there has been an increasing awareness that many NLP systems incorporate biases of various types (e.g., regarding gender or race) which can have significant negative consequences. At the same time, the techniques used to statistically analyze such biases are still relatively simple. Typically, studies test for the presence of a significant difference between two levels of a single bias variable (e.g., male vs. female) without attention to potential confounders, and do not quantify the importance of the bias variable. This article proposes to analyze bias in the output of NLP systems using multivariate regression models. They provide a robust and more informative alternative which (a) generalizes to multiple bias variables, (b) can take covariates into account, (c) can be combined with measures of effect size to quantify the size of bias. Jointly, these effects contribute to a more robust statistical analysis of bias that can be used to diagnose system behavior and extract informative examples. We demonstrate the benefits of our method by analyzing a range of current NLP models on one regression and one classification tasks (emotion intensity prediction and coreference resolution, respectively).
|
| format | Article |
| id | doaj-art-108c61e567694cb0ba97c2a966bec278 |
| institution | Kabale University |
| issn | 2000-1533 |
| language | English |
| publishDate | 2022-08-01 |
| publisher | Linköping University Electronic Press |
| record_format | Article |
| series | Northern European Journal of Language Technology |
| spelling | doaj-art-108c61e567694cb0ba97c2a966bec2782025-01-22T15:25:18ZengLinköping University Electronic PressNorthern European Journal of Language Technology2000-15332022-08-018110.3384/nejlt.2000-1533.2022.3505Bias Identification and Attribution in NLP Models With Regression and Effect SizesErenay Dayanik0Ngoc Thang Vu1Sebastian Padó2University of StuttgartUniversity of StuttgartUniversity of Stuttgart In recent years, there has been an increasing awareness that many NLP systems incorporate biases of various types (e.g., regarding gender or race) which can have significant negative consequences. At the same time, the techniques used to statistically analyze such biases are still relatively simple. Typically, studies test for the presence of a significant difference between two levels of a single bias variable (e.g., male vs. female) without attention to potential confounders, and do not quantify the importance of the bias variable. This article proposes to analyze bias in the output of NLP systems using multivariate regression models. They provide a robust and more informative alternative which (a) generalizes to multiple bias variables, (b) can take covariates into account, (c) can be combined with measures of effect size to quantify the size of bias. Jointly, these effects contribute to a more robust statistical analysis of bias that can be used to diagnose system behavior and extract informative examples. We demonstrate the benefits of our method by analyzing a range of current NLP models on one regression and one classification tasks (emotion intensity prediction and coreference resolution, respectively). https://nejlt.ep.liu.se/article/view/3505 |
| spellingShingle | Erenay Dayanik Ngoc Thang Vu Sebastian Padó Bias Identification and Attribution in NLP Models With Regression and Effect Sizes Northern European Journal of Language Technology |
| title | Bias Identification and Attribution in NLP Models With Regression and Effect Sizes |
| title_full | Bias Identification and Attribution in NLP Models With Regression and Effect Sizes |
| title_fullStr | Bias Identification and Attribution in NLP Models With Regression and Effect Sizes |
| title_full_unstemmed | Bias Identification and Attribution in NLP Models With Regression and Effect Sizes |
| title_short | Bias Identification and Attribution in NLP Models With Regression and Effect Sizes |
| title_sort | bias identification and attribution in nlp models with regression and effect sizes |
| url | https://nejlt.ep.liu.se/article/view/3505 |
| work_keys_str_mv | AT erenaydayanik biasidentificationandattributioninnlpmodelswithregressionandeffectsizes AT ngocthangvu biasidentificationandattributioninnlpmodelswithregressionandeffectsizes AT sebastianpado biasidentificationandattributioninnlpmodelswithregressionandeffectsizes |