Recent studies show that Natural Language Processing (NLP) technologies propagate societal biases about demographic groups associated with attributes such as gender, race, and nationality. To create interventions and mitigate these biases and associated harms, it is vital to be able to detect and measure such biases. While existing works propose bias evaluation and mitigation methods for various tasks, there remains a need to cohesively understand the biases and the specific harms they measure, and how different measures compare with each other. To address this gap, this work presents a practical framework of harms and a series of questions that practitioners can answer to guide the development of bias measures. As a validation of our framework and documentation questions, we also present several case studies of how existing bias measures in NLP—both intrinsic measures of bias in representations and extrinsic measures of bias of downstream applications—can be aligned with different harms and how our proposed documentation questions facilitates more holistic understanding of what bias measures are measuring.
more »
« less
Socially Aware Bias Measurements for Hindi Language Representations
Language representations are an efficient tool used across NLP, but they are strife with encoded societal biases. These biases are studied extensively, but with a primary focus on English language representations and biases common in the context of Western society. In this work, we investigate the biases present in Hindi language representations such as caste and religion associated biases. We demonstrate how biases are unique to specific language representations based on the history and culture of the region they are widely spoken in, and also how the same societal bias (such as binary gender associated biases) when investigated across languages is encoded by different words and text spans. With this work, we emphasize on the necessity of social-awareness along with linguistic and grammatical artefacts when modeling language representations, in order to understand the biases encoded.
more »
« less
- Award ID(s):
- 1927554
- PAR ID:
- 10391939
- Date Published:
- Journal Name:
- Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies
- Page Range / eLocation ID:
- 1041 to 1052
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
As natural language processing methods are increasingly deployed in real-world scenarios such as healthcare, legal systems, and social science, it becomes necessary to recognize the role they potentially play in shaping social biases and stereotypes. Previous work has revealed the presence of social biases in widely used word embeddings involving gender, race, religion, and other social constructs. While some methods were proposed to debias these word-level embeddings, there is a need to perform debiasing at the sentence-level given the recent shift towards new contextualized sentence representations such as ELMo and BERT. In this paper, we investigate the presence of social biases in sentence-level representations and propose a new method, Sent-Debias, to reduce these biases. We show that Sent-Debias is effective in removing biases, and at the same time, preserves performance on sentence-level downstream tasks such as sentiment analysis, linguistic acceptability, and natural language understanding. We hope that our work will inspire future research on characterizing and removing social biases from widely adopted sentence representations for fairer NLP.more » « less
-
Large Language Models (LLMs) perpetuate social biases, reflecting prejudices in their training data and reinforcing societal stereotypes and inequalities. Our work explores the potential of the Contact Hypothesis, a concept from social psychology for debiasing LLMs. We simulate various forms of social contact through LLM prompting to measure their influence on the model’s biases, mirroring how intergroup interactions can reduce prejudices in social contexts. We create a dataset of 108,000 prompts following a principled approach replicating social contact to measure biases in three LLMs (LLaMA 2, Tulu, and NousHermes) across 13 social bias dimensions. We propose a unique debiasing technique, Social Contact Debiasing (SCD), that instruction-tunes these models with unbiased responses to prompts. Our research demonstrates that LLM responses exhibit social biases when subject to contact probing, but more importantly, these biases can be significantly reduced by up to 40% in 1 epoch of instruction tuning LLaMA 2 following our SCD strategy.more » « less
-
Perlman, Marcus (Ed.)Longstanding cross-linguistic work on event representations in spoken languages have argued for a robust mapping between an event’s underlying representation and its syntactic encoding, such that–for example–the agent of an event is most frequently mapped to subject position. In the same vein, sign languages have long been claimed to construct signs that visually represent their meaning, i.e., signs that are iconic. Experimental research on linguistic parameters such as plurality and aspect has recently shown some of them to be visually universal in sign, i.e. recognized by non-signers as well as signers, and have identified specific visual cues that achieve this mapping. However, little is known about what makes action representations in sign language iconic, or whether and how the mapping of underlying event representations to syntactic encoding is visually apparent in the form of a verb sign. To this end, we asked what visual cues non-signers may use in evaluating transitivity (i.e., the number of entities involved in an action). To do this, we correlated non-signer judgments about transitivity of verb signs from American Sign Language (ASL) with phonological characteristics of these signs. We found that non-signers did not accurately guess the transitivity of the signs, but that non-signer transitivity judgments can nevertheless be predicted from the signs’ visual characteristics. Further, non-signers cue in on just those features that code event representations across sign languages, despite interpreting them differently. This suggests the existence of visual biases that underlie detection of linguistic categories, such as transitivity, which may uncouple from underlying conceptual representations over time in mature sign languages due to lexicalization processes.more » « less
-
Linguistic communication is an intrinsically social activity that enables us to share thoughts across minds. Many complex social uses of language can be captured by domain-general representations of other minds (i.e., mentalistic representations) that externally modulate linguistic meaning through Gricean reasoning. However, here we show that representations of others’ attention are embedded within language itself. Across ten languages, we show that demonstratives—basic grammatical words (e.g., “this”/“that”) which are evolutionarily ancient, learned early in life, and documented in all known languages—are intrinsic attention tools. Beyond their spatial meanings, demonstratives encode both joint attention and the direction in which the listener must turn to establish it. Crucially, the frequency of the spatial and attentional uses of demonstratives varies across languages, suggesting that both spatial and mentalistic representations are part of their conventional meaning. Using computational modeling, we show that mentalistic representations of others’ attention are internally encoded in demonstratives, with their effect further boosted by Gricean reasoning. Yet, speakers are largely unaware of this, incorrectly reporting that they primarily capture spatial representations. Our findings show that representations of other people’s cognitive states (namely, their attention) are embedded in language and suggest that the most basic building blocks of the linguistic system crucially rely on social cognition.more » « less