Towards Few-shot Entity Recognition in Document Images: A Label-aware Sequence-to-Sequence Framework

Findings of the Association for Computational Linguistics: ACL 2022, pages 372 - 382 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Dynamically Refined Regularization for Improving Cross-corpora Hate Speech Detection Tulika Bose † Nikolaos Aletras ‡ Irina Illina † Dominique Fohr † † Universite de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France ‡University of Sheffield, United Kingdom {tulika.bose, illina, dominique.fohr}@loria.fr n.aletras@sheffield.ac.uk Abstract Warning: this paper contains content that may be offensive and distressing. Hate speech classifiers exhibit substantial performance degradation when evaluated on datasets different from the source. This is due to learning spurious correlations between words that are not necessarily relevant to hate- ful language, and hate speech labels from the training corpus. Previous work has attempted to mitigate this problem by regularizing spe- cific terms from pre-defined static dictionaries. While this has been demonstrated to improve the generalizability of classifiers, the coverage of such methods is limited and the dictionar- ies require regular manual updates from human experts. In this paper, we propose to automati- cally identify and reduce spurious correlations using attribution methods with dynamic refine- ment of the list of terms that need to be regular- ized during training. Our approach is flexible and improves the cross-corpora performance over previous work independently and in com- bination with pre-defined dictionaries.1 1 Introduction The relative sparsity of hateful content in the real world requires crawling of many of the standard hate speech corpora through keyword-based sam- pling (Poletto et al., 2021), rather than random sampling. Thus, hate speech classifiers (D’Sa et al., 2020; Mozafari et al., 2019; Badjatiya et al., 2017) often learn spurious correlations from the training corpus (Wiegand et al., 2019) leading to a substan- tial performance degradation when evaluated on a corpus with a different distribution (Yin and Zu- biaga, 2021; Bose et al., 2021; Florio et al., 2020; Arango et al., 2019; Swamy et al., 2019; Karan and Šnajder, 2018). Recent work has proposed regularization mecha- nisms to penalize spurious correlations by attempt- 1Code is available here: https://github.com/ tbose20/D-Ref Target corpus utterances Actual Predicted Genocide is never ok non-hate hate Women are goddesses non-hate hate Table 1: Spurious correlations learned by the source classifier between the shaded tokens and the hate label. ing to explain model predictions using feature at- tribution methods (Ross et al., 2017; Rieger et al., 2020; Adebayo et al., 2020). These methods assign importance scores to input tokens that contribute more towards a particular prediction (Lundberg and Lee, 2017). For instance, Liu and Avci (2019) penalize the attributions assigned to tokens con- tained in a manually curated dictionary consisting of group identifiers (e.g. women, jews) that are often known to be targets of hate. Kennedy et al. (2020) extract group identifiers manually from the top tokens indicated by a bag-of-words logistic re- gression model trained on the source corpus. How- ever, regularizing only group identifiers limits the coverage of such approaches, and may not capture other forms of corpus-specific correlations learned by the classifier limiting its performance on a new …

Content truncated to 500 words