Artifacts or Abduction: How Do LLMs Answer Multiple-Choice Questions Without the Question?

Multiple-choice question answering (MCQA) is often used to evaluate large language models (LLMs). To see if MCQA assesses LLMs as intended, we probe if LLMs can perform MCQA with choices-only prompts, where models must select the correct answer only from the choices. In three MCQA datasets and four LLMs, this prompt bests a majority baseline in 11/12 cases, with up to 0.33 accuracy gain. To help explain this behavior, we conduct an in-depth, black-box analysis on memorization, choice dynamics, and question inference. Our key findings are threefold. First, we find no evidence that the choices-only accuracy stems from memorization alone. Second, priors over individual choices do not fully explain choices-only accuracy, hinting that LLMs use the group dynamics of choices. Third, LLMs have some ability to infer a relevant question from choices, and surprisingly can sometimes even match the original question. We hope to motivate the use of stronger baselines in MCQA benchmarks, the design of robust MCQA datasets, and further efforts to explain LLM decision-making.

Multilingual large language models leak human stereotypes across language boundaries

Multilingual large language models have been increasingly popular for their proficiency in comprehending and generating text across various languages. Previous research has shown that the presence of stereotypes and biases in monolingual large language models can be attributed to the nature of their training data, which is collected from humans and reflects societal biases. Multilingual language models undergo the same training procedure as monolingual ones, albeit with training data sourced from various languages. This raises the question: do stereotypes present in one social context leak across languages within the model? In our work, we first define the term ``stereotype leakage'' and propose a framework for its measurement. With this framework, we investigate how stereotypical associations leak across four languages: English, Russian, Chinese, and Hindi. To quantify the stereotype leakage, we employ an approach from social psychology, measuring stereotypes via group-trait associations. We evaluate human stereotypes and stereotypical associations manifested in multilingual large language models such as mBERT, mT5, and ChatGPT. Our findings show a noticeable leakage of positive, negative, and non-polar associations across all languages. Notably, Hindi within multilingual models appears to be the most susceptible to influence from other languages, while Chinese is the least. Additionally, ChatGPT exhibits a better alignment with human scores than other models.

Towards Pragmatic Awareness in Question Answering: A Case Study in Maternal and Infant Health

Questions posed by information-seeking users often contain implicit false or potentially harmful assumptions. In a high-risk domain such as maternal and infant health, a question-answering system must recognize these pragmatic constraints and go beyond simply answering user questions, examining them in context to respond helpfully. To achieve this, we study pragmatic inferences made when mothers ask questions about pregnancy and infant care. Some of the inferences in these questions evade detection by existing methods, risking the possibility of QA systems failing to address them which can have dangerous health and policy implications. We explore the viability of detecting inferences from questions using large language models and illustrate that informing existing QA pipelines with pragmatic inferences produces responses that can mitigate the propagation of harmful beliefs.

It's Not Easy Being Wrong: Evaluating Process of Elimination Reasoning in Large Language Models

Chain-of-thought (COT) prompting can help large language models (LLMs) reason toward correct answers, but its efficacy in reasoning toward incorrect answers is unexplored. This strategy of process of elimination (PoE), when used with COT, has the potential to enhance interpretability in tasks like medical diagnoses of exclusion. Thus, we propose PoE with COT, a new task where LLMs must reason toward incorrect options on multiple-choice questions. We evaluate the ability of GPT-3.5, LLaMA-2, and Falcon to perform PoE with COT on 2-choice commonsense and scientific reasoning datasets. We show that PoE consistently underperforms directly choosing the correct answer. The agreement of these strategies is also lower than the self-consistency of each strategy. To study these issues further, we conduct an error analysis and give suggestions for future work.

Nichelle and Nancy: The Influence of Demographic Attributes and Tokenization Length on First Name Biases

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.

What to Read in a Contract? Party-Specific Summarization of Important Obligations, Entitlements, and Prohibitions in Legal Documents

Legal contracts, such as employment or lease agreements, are important documents as they govern the obligations and entitlements of the various contracting parties. However, these documents are typically long and written in legalese resulting in lots of manual hours spent in understanding them. In this paper, we address the task of summarizing legal contracts for each of the contracting parties, to enable faster reviewing and improved understanding of them. Specifically, we collect a dataset consisting of pairwise importance comparison annotations by legal experts for ~293K sentence pairs from lease agreements. We propose a novel extractive summarization system to automatically produce a summary consisting of the most important obligations, entitlements, and prohibitions in a contract. It consists of two modules: (1) a content categorize to identify sentences containing each of the categories (i.e., obligation, entitlement, and prohibition) for a party, and (2) an importance ranker to compare the importance among sentences of each category for a party to obtain a ranked list. The final summary is produced by selecting the most important sentences of a category for each of the parties. We demonstrate the effectiveness of our proposed system by comparing it against several text ranking baselines via automatic and human evaluation.

Agent-Specific Deontic Modality Detection in Legal Language

Legal documents are typically long and written in legalese, which makes it particularly difficult for laypeople to understand their rights and duties. While natural language understanding technologies can be valuable in supporting such understanding in the legal domain, the limited availability of datasets annotated for deontic modalities in the legal domain, due to the cost of hiring experts and privacy issues, is a bottleneck. To this end, we introduce, LEXDEMOD, a corpus of English contracts annotated with deontic modality expressed with respect to a contracting party or agent along with the modal triggers. We benchmark this dataset on two tasks: (i) agent-specific multi-label deontic modality classification, and (ii) agent-specific deontic modality and trigger span detection using Transformer-based (Vaswani et al., 2017) language models. Transfer learning experiments show that the linguistic diversity of modal expressions in LEXDEMOD generalizes reasonably from lease to employment and rental agreements. A small case study indicates that a model trained on LEXDEMOD can detect red flags with high recall. We believe our work offers a new research direction for deontic modality detection in the legal domain.

SODAPOP: Open-Ended Discovery of Social Biases in Social Commonsense Reasoning Models

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.

Theory-Grounded Measurement of U.S. Social Stereotypes in English Language Models

NLP models trained on text have been shown to reproduce human stereotypes, which can magnify harms to marginalized groups when systems are deployed at scale. We adapt the Agency-Belief-Communion (ABC) stereotype model of Koch et al. (2016) from social psychology as a framework for the systematic study and discovery of stereotypic group-trait associations in language models (LMs). We introduce the sensitivity test (SeT) for measuring stereotypical associations from language models. To evaluate SeT and other measures using the ABC model, we collect group-trait judgments from U.S.-based subjects to compare with English LM stereotypes. Finally, we extend this framework to measure LM stereotyping of intersectional identities.

Partial-input baselines show that NLI models can ignore context, but they don't

When strong partial-input baselines reveal artifacts in crowdsourced NLI datasets, the performance of full-input models trained on such datasets is often dismissed as reliance on spurious correlations. We investigate whether state-of-the-art NLI models are capable of overriding default inferences made by a partial-input baseline. We introduce an evaluation set of 600 examples consisting of perturbed premises to examine a RoBERTa model's sensitivity to edited contexts. Our results indicate that NLI models are still capable of learning to condition on context--a necessary component of inferential reasoning--despite being trained on artifact-ridden datasets.