Existing works examining Vision-Language Models (VLMs) for social biases predominantly focus on a limited set of documented bias associations, such as gender-profession or race-crime. This narrow scope often overlooks a vast range of unexamined implicit associations, restricting the identification and, hence, mitigation of such biases. We address this gap by probing VLMs to (1) uncover hidden, implicit associations across 9 bias dimensions. We systematically explore diverse input and output modalities and (2) demonstrate how biased associations vary in their negativity, toxicity, and extremity. Our work (3) identifies subtle and extreme biases that are typically not recognized by existing methodologies. We make the **D**ataset **o**f **r**etrieved **a**ssociations (**Dora**) publicly available.

Exploring the intersection of language and culture in Large Language Models (LLMs), this study critically examines their capability to encapsulate cultural nuances across diverse linguistic landscapes. Central to our investigation are three research questions: the efficacy of language-specific instruction tuning, the impact of pretraining on dominant language data, and the identification of optimal approaches to elicit accurate cultural knowledge from LLMs. Utilizing the GeoMLaMA benchmark for multilingual commonsense knowledge and an adapted CAMeL dataset (English-only) for evaluation of nuanced cultural aspects, our experiments span six different languages and cultural contexts, revealing the extent of LLMs’ cultural awareness. Our findings highlight a nuanced landscape: while language-specific tuning and bilingual pretraining enhance cultural understanding in certain contexts, they also uncover inconsistencies and biases, particularly in non-Western cultures. This work expands our understanding of LLMs’ cultural competence and emphasizes the importance of integrating diverse cultural perspectives in their development, aiming for a more globally representative and equitable approach in language modeling.

We study stereotypes embedded within one of the most popular text-to-image generators: Stable Diffusion. We answer the question: what stereotypes of gender and nationality/continental identity does Stable Diffusion display in the absence of such information i.e. what gender and nationality/continental identity is assigned to ‘a person,’ or to ‘a person from Asia.’ Using CLIP-cosine similarity for zero-shot classification of images generated by CLIP-based Stable Diffusion v2.1 verified by manual examination, we chronicle results from 136 prompts (50 results/prompt) of front-facing images of faces from 6 different continents, 27 countries and 3 genders. We observe how Stable Diffusion results of ‘a person’ without any additional gender/nationality information correspond closest to images of men (avg. similarity 0.64) and least with persons of nonbinary gender (avg. similarity 0.41), and to persons from Europe/North America (avg. similarities 0.71 and 0.68, respectively) over Africa/Asia (avg. similarities 0.43 and 0.41, respectively), pointing towards Stable Diffusion having a concerning representation of personhood to be a European/North American man. We also show continental stereotypes and resultant harms e.g. a person from Oceania is deemed to be Australian/New Zealander (avg. similarities 0.77 and 0.74, respectively) over Papua New Guinean (avg. similarity 0.31), pointing to the erasure of Indigenous Oceanic peoples, who form a majority over descendants of colonizers both in Papua New Guinea and in Oceania overall. Finally, we unexpectedly observe a pattern of sexualization of women, specifically Latin American, Mexican, Indian and Egyptian women, confirmed through an NSFW detector and verified by manual examination. This demonstrates how Stable Diffusion perpetuates Western fetishization of women of color through objectification in media, which if left unchecked will worsen this stereotypical representation. All code and relevant data will be made publicly available.

Previous work has established that a person’s demographics and speech style affect how well speech processing models perform for them. But where does this bias come from? In this work, we present the Speech Embedding Association Test (SpEAT), a method for detecting bias in one type of model used for many speech tasks: pre-trained models. The SpEAT is inspired by word embedding association tests in natural language processing, which quantify intrinsic bias in a model’s representations of different concepts, such as race or valence—something’s pleasantness or unpleasantness—and capture the extent to which a model trained on large-scale socio-cultural data has learned human-like biases. Using the SpEAT, we test for six types of bias in 16 English speech models (including 4 models also trained on multilingual data), which come from the wav2vec 2.0, HuBERT, WavLM, and Whisper model families. We find that 14 or more models reveal positive valence (pleasantness) associations with abled people over disabled people, with European-Americans over African-Americans, with females over males, with U.S. accented speakers over non-U.S. accented speakers, and with younger people over older people. Beyond establishing that pre-trained speech models contain these biases, we also show that they can have real world effects. We compare biases found in pre-trained models to biases in downstream models adapted to the task of Speech Emotion Recognition (SER) and find that in 66 of the 96 tests performed (69%), the group that is more associated with positive valence as indicated by the SpEAT also tends to be predicted as speaking with higher valence by the downstream model. Our work provides evidence that, like text and image-based models, pre-trained speech based-models frequently learn human-like biases when trained on large-scale socio-cultural datasets. Our work also shows that bias found in pre-trained models can propagate to the downstream task of SER.

We examine the effects of contrastive visual semantic pretraining by comparing the geometry and semantic properties of contextualized English language representations formed by GPT-2 and CLIP, a zero-shot multimodal image classifier which adapts the GPT-2 architecture to encode image captions. We find that contrastive visual semantic pretraining significantly mitigates the anisotropy found in contextualized word embeddings from GPT-2, such that the intra-layer self-similarity (mean pairwise cosine similarity) of CLIP word embeddings is under .25 in all layers, compared to greater than .95 in the top layer of GPT-2. CLIP word embeddings outperform GPT-2 on word-level semantic intrinsic evaluation tasks, and achieve a new corpus-based state of the art for the RG65 evaluation, at .88. CLIP also forms fine-grained semantic representations of sentences, and obtains Spearman’s 𝜌 = .73 on the SemEval-2017 Semantic Textual Similarity Benchmark with no fine-tuning, compared to no greater than 𝜌 = .45 in any layer of GPT-2. Finally, intra-layer self-similarity of CLIP sentence embeddings decreases as the layer index increases, finishing at .25 in the top layer, while the self-similarity of GPT-2 sentence embeddings formed using the EOS token increases layer-over-layer and never falls below .97. Our results indicate that high anisotropy is not an inevitable consequence of contextualization, and that visual semantic pretraining is beneficial not only for ordering visual representations, but also for encoding useful semantic representations of language, both on the word level and the sentence level.

We use a dataset of U.S. first names with labels based on predominant gender and racial group to examine the effect of training corpus frequency on tokenization, contextualization, similarity to initial representation, and bias in BERT, GPT-2, T5, and XLNet. We show that predominantly female and non-white names are less frequent in the training corpora of these four language models. We find that infrequent names are more self-similar across contexts, with Spearman’s rho between frequency and self-similarity as low as -.763. Infrequent names are also less similar to initial representation, with Spearman’s rho between frequency and linear centered kernel alignment (CKA) similarity to initial representation as high as .702. Moreover, we find Spearman’s rho between racial bias and name frequency in BERT of .492, indicating that lower-frequency minority group names are more associated with unpleasantness. Representations of infrequent names undergo more processing, but are more self-similar, indicating that models rely on less context-informed representations of uncommon and minority names which are overfit to a lower number of observed contexts.

Word embeddings learn implicit biases from linguistic regularities captured by word co-occurrence statistics. By extending methods that quantify human-like biases in word embeddings, we introduce ValNorm, a novel intrinsic evaluation task and method to quantify the valence dimension of affect in human-rated word sets from social psychology. We apply ValNorm on static word embeddings from seven languages (Chinese, English, German, Polish, Portuguese, Spanish, and Turkish) and from historical English text spanning 200 years. ValNorm achieves consistently high accuracy in quantifying the valence of non-discriminatory, non-social group word sets. Specifically, ValNorm achieves a Pearson correlation of r=0.88 for human judgment scores of valence for 399 words collected to establish pleasantness norms in English. In contrast, we measure gender stereotypes using the same set of word embeddings and find that social biases vary across languages. Our results indicate that valence associations of non-discriminatory, non-social group words represent widely-shared associations, in seven languages and over 200 years.