Haozhe An

Haozhe An, Zongxia Li, Jieyu Zhao et al.

A common limitation of diagnostic tests for detecting social biases in NLP models is that they may only detect stereotypic associations that are pre-specified by the designer of the test. Since enumerating all possible problematic associations is infeasible, it is likely these tests fail to detect biases that are present in a model but not pre-specified by the designer. To address this limitation, we propose SODAPOP (SOcial bias Discovery from Answers about PeOPle) in social commonsense question-answering. Our pipeline generates modified instances from the Social IQa dataset (Sap et al., 2019) by (1) substituting names associated with different demographic groups, and (2) generating many distractor answers from a masked language model. By using a social commonsense model to score the generated distractors, we are able to uncover the model's stereotypic associations between demographic groups and an open set of words. We also test SODAPOP on debiased models and show the limitations of multiple state-of-the-art debiasing algorithms.

Shramay Palta, Haozhe An, Yifan Yang et al. · deepmind, microsoft-research

Retrieval based open-domain QA systems use retrieved documents and answer-span selection over retrieved documents to find best-answer candidates. We hypothesize that multilingual Question Answering (QA) systems are prone to information inconsistency when it comes to documents written in different languages, because these documents tend to provide a model with varying information about the same topic. To understand the effects of the biased availability of information and cultural influence, we analyze the behavior of multilingual open-domain question answering models with a focus on retrieval bias. We analyze if different retriever models present different passages given the same question in different languages on TyDi QA and XOR-TyDi QA, two multilingualQA datasets. We speculate that the content differences in documents across languages might reflect cultural divergences and/or social biases.

Christabel Acquaye, Haozhe An, Rachel Rudinger

Recent work has highlighted the culturally-contingent nature of commonsense knowledge. We introduce AMAMMER$ε$, a test set of 525 multiple-choice questions designed to evaluate the commonsense knowledge of English LLMs, relative to the cultural contexts of Ghana and the United States. To create AMAMMER$ε$, we select a set of multiple-choice questions (MCQs) from existing commonsense datasets and rewrite them in a multi-stage process involving surveys of Ghanaian and U.S. participants. In three rounds of surveys, participants from both pools are solicited to (1) write correct and incorrect answer choices, (2) rate individual answer choices on a 5-point Likert scale, and (3) select the best answer choice from the newly-constructed MCQ items, in a final validation step. By engaging participants at multiple stages, our procedure ensures that participant perspectives are incorporated both in the creation and validation of test items, resulting in high levels of agreement within each pool. We evaluate several off-the-shelf English LLMs on AMAMMER$ε$. Uniformly, models prefer answers choices that align with the preferences of U.S. annotators over Ghanaian annotators. Additionally, when test items specify a cultural context (Ghana or the U.S.), models exhibit some ability to adapt, but performance is consistently better in U.S. contexts than Ghanaian. As large resources are devoted to the advancement of English LLMs, our findings underscore the need for culturally adaptable models and evaluations to meet the needs of diverse English-speaking populations around the world.

Haozhe An, Christabel Acquaye, Colin Wang et al.

We examine whether large language models (LLMs) exhibit race- and gender-based name discrimination in hiring decisions, similar to classic findings in the social sciences (Bertrand and Mullainathan, 2004). We design a series of templatic prompts to LLMs to write an email to a named job applicant informing them of a hiring decision. By manipulating the applicant's first name, we measure the effect of perceived race, ethnicity, and gender on the probability that the LLM generates an acceptance or rejection email. We find that the hiring decisions of LLMs in many settings are more likely to favor White applicants over Hispanic applicants. In aggregate, the groups with the highest and lowest acceptance rates respectively are masculine White names and masculine Hispanic names. However, the comparative acceptance rates by group vary under different templatic settings, suggesting that LLMs' race- and gender-sensitivity may be idiosyncratic and prompt-sensitive.

Haozhe An, Rachel Rudinger

Through the use of first name substitution experiments, prior research has demonstrated the tendency of social commonsense reasoning models to systematically exhibit social biases along the dimensions of race, ethnicity, and gender (An et al., 2023). Demographic attributes of first names, however, are strongly correlated with corpus frequency and tokenization length, which may influence model behavior independent of or in addition to demographic factors. In this paper, we conduct a new series of first name substitution experiments that measures the influence of these factors while controlling for the others. We find that demographic attributes of a name (race, ethnicity, and gender) and name tokenization length are both factors that systematically affect the behavior of social commonsense reasoning models.

Haoran Liu, Haoyi Xiong, Yaqing Wang et al.

We find that different Deep Neural Networks (DNNs) trained with the same dataset share a common principal subspace in latent spaces, no matter in which architectures (e.g., Convolutional Neural Networks (CNNs), Multi-Layer Preceptors (MLPs) and Autoencoders (AEs)) the DNNs were built or even whether labels have been used in training (e.g., supervised, unsupervised, and self-supervised learning). Specifically, we design a new metric $\mathcal{P}$-vector to represent the principal subspace of deep features learned in a DNN, and propose to measure angles between the principal subspaces using $\mathcal{P}$-vectors. Small angles (with cosine close to $1.0$) have been found in the comparisons between any two DNNs trained with different algorithms/architectures. Furthermore, during the training procedure from random scratch, the angle decrease from a larger one ($70^\circ-80^\circ$ usually) to the small one, which coincides the progress of feature space learning from scratch to convergence. Then, we carry out case studies to measure the angle between the $\mathcal{P}$-vector and the principal subspace of training dataset, and connect such angle with generalization performance. Extensive experiments with practically-used Multi-Layer Perceptron (MLPs), AEs and CNNs for classification, image reconstruction, and self-supervised learning tasks on MNIST, CIFAR-10 and CIFAR-100 datasets have been done to support our claims with solid evidences. Interpretability of Deep Learning, Feature Learning, and Subspaces of Deep Features

Xingjian Li, Haoyi Xiong, Haozhe An et al.

Softening labels of training datasets with respect to data representations has been frequently used to improve the training of deep neural networks (DNNs). While such a practice has been studied as a way to leverage privileged information about the distribution of the data, a well-trained learner with soft classification outputs should be first obtained as a prior to generate such privileged information. To solve such chicken-egg problem, we propose COLAM framework that Co-Learns DNNs and soft labels through Alternating Minimization of two objectives - (a) the training loss subject to soft labels and (b) the objective to learn improved soft labels - in one end-to-end training procedure. We performed extensive experiments to compare our proposed method with a series of baselines. The experiment results show that COLAM achieves improved performance on many tasks with better testing classification accuracy. We also provide both qualitative and quantitative analyses that explain why COLAM works well.