Abstract The lack of reasoning capabilities in Vision-Language Models (VLMs) has remained at the forefront of research discourse. We posit that this behavior stems from a reporting bias in their training data. That is, how people communicate about visual content by default omits tacit information needed to supervise some types of reasoning; e.g., “at the game today!” is a more likely caption than “a photo of 37 people standing behind a field”. We investigate the data underlying the popular VLMs OpenCLIP, LLaVA-1.5 and Molmo through the lens of theories from pragmatics, and find that reporting bias results in insufficient representation of four reasoning skills (spatial, temporal, negation, and counting), despite the corpora being of web-scale, and/or synthetically generated. With a set of curated benchmarks, we demonstrate that: (i) VLMs perform poorly on the aforementioned types of reasoning suppressed in the training data by reporting bias; (ii) contrary to popular belief, scaling data size, model size, and to multiple languages does not result in emergence of these skills by default; but, promisingly, (iii) incorporating annotations specifically collected to obtain tacit information is effective. Our findings highlight the need for more intentional training data curation methods, rather than counting on scale for emergence of reasoning capabilities.

Most vision-language models today are primarily trained on English image-text pairs, with non-English pairs often filtered out. Evidence from cross-cultural psychology suggests that this approach will bias models against perceptual modes exhibited by people who speak other (non-English) languages. We investigate semantic and expressive variation in image captions across different languages; we analyze both human-annotated datasets and model-produced captions. By analyzing captions across seven languages (English, French, German, Russian, Chinese, Japanese, Korean) in high-quality image captioning datasets (Crossmodal and Visual Genome), we find that multilingual caption sets tend to provide richer visual descriptions than monolingual (including English-only) ones; multilingual sets contain 46.0% more objects, 66.1% more relationships, and 66.8% more attributes. We observe the same results with multilingual captions produced by LLaVA and the Google Vertex API: for example, compared to monolingual captions, they cover 21.9% more objects, 18.8% more relations, and 20.1% more attributes. These suggest that, across a large number of samples, different languages bias people and models to focus on different visual concepts. Finally, we show that models trained on image-text data in one language perform distinctly better on that language’s test set. Our work points towards the potential value of training vision models on multilingual data sources to widen the range/variation of descriptive information those models are exposed to.

Because large language models are expensive to pretrain on different datasets, using smaller-scale experiments to decide on data is crucial for reducing costs. Which benchmarks and methods of making decisions from observed performance at small scale most accurately predict the datasets that yield the best large models? To empower open exploration of this question, we release models, data, and evaluations in DataDecide—the most extensive open suite of models over differences in data and scale. We conduct controlled pretraining experiments across 25 corpora with differing sources, deduplication, and filtering up to 100B tokens, model sizes up to 1B parameters, and 3 random seeds. We find that the ranking of models at a single, small size (e.g., 150M parameters) is a strong baseline for predicting best models at our larger target scale (1B) ($\tilde$ 80% of comparisons correct). No scaling law methods among 8 baselines exceed the compute-decision frontier of single-scale predictions, but DataDecide can measure improvement in future scaling laws. We also identify that using continuous likelihood metrics as proxies in small experiments makes benchmarks including MMLU, ARC, HellaSwag, MBPP, and HumanEval $>$ 80% predictable at the target 1B scale with just 0.01% of the compute.

We examine diverging preferences in human-labeled preference datasets. We develop a taxonomy of disagreement sources spanning ten categories across four high-level classes and find that the majority of disagreements are due to factors such as task underspecification or response style. Our findings challenge a standard assumption in reward modeling methods that annotator disagreements can be attributed to simple noise. We then explore how these findings impact two areas of LLM development: reward modeling training and evaluation. In our experiments, we demonstrate how standard reward modeling (e.g., Bradley-Terry) and LLM-as-Judge evaluation methods fail to account for divergence between annotators. These findings highlight challenges in LLM evaluations, which are greatly influenced by divisive features like response style, and in developing pluralistically aligned LLMs. To address these issues, we develop methods for identifying diverging preferences to mitigate their influence in evaluations and during LLM training.

Kaitlyn Zhou, Jena D. Hwang, Xiang Ren, Nouha Dziri, Dan Jurafsky, Maarten Sap. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Human values are crucial to human decision-making. Value pluralism is the view that multiple correct values may be held in tension with one another (e.g., when considering lying to a friend to protect their feelings, how does one balance honesty with friendship?). As statistical learners, AI systems fit to averages by default, washing out these potentially irreducible value conflicts. To improve AI systems to better reflect value pluralism, the first-order challenge is to explore the extent to which AI systems can model pluralistic human values, rights, and duties as well as their interaction. We introduce ValuePrism, a large-scale dataset of 218k values, rights, and duties connected to 31k human-written situations. ValuePrism’s contextualized values are generated by GPT-4 and deemed high-quality by human annotators 91% of the time. We conduct a large-scale study with annotators across diverse social and demographic backgrounds to try to understand whose values are represented. With ValuePrism, we build Value Kaleidoscope (or Kaleido), an open, light-weight, and structured language-based multi-task model that generates, explains, and assesses the relevance and valence (i.e., support or oppose) of human values, rights, and duties within a specific context. Humans prefer the sets of values output by our system over the teacher GPT- 4, finding them more accurate and with broader coverage. In addition, we demonstrate that Kaleido can help explain variability in human decision-making by outputting contrasting values. Finally, we show that Kaleido’s representations transfer to other philosophical frameworks and datasets, confirming the benefit of an explicit, modular, and interpretable approach to value pluralism. We hope that our work will serve as a step to making more explicit the implicit values behind human decision-making and to steering AI systems to make decisions that are more in accordance with them.

As natural language becomes the default interface for human-AI interaction, there is a need for LMs to appropriately communicate uncertainties in downstream applications.In this work, we investigate how LMs incorporate confidence in responses via natural language and how downstream users behave in response to LM-articulated uncertainties.We examine publicly deployed models and find that LMs are reluctant to express uncertainties when answering questions even when they produce incorrect responses.LMs can be explicitly prompted to express confidences, but tend to be overconfident, resulting in high error rates (an average of 47%) among confident responses.We test the risks of LM overconfidence by conducting human experiments and show that users rely heavily on LM generations, whether or not they are marked by certainty.Lastly, we investigate the preference-annotated datasets used in post training alignment and find that humans are biased against texts with uncertainty.Our work highlights new safety harms facing human-LM interactions and proposes design recommendations and mitigating strategies moving forward.

Procedural planning, which entails decomposing a high-level goal into a sequence of temporally ordered steps, is an important yet intricate task for machines. It involves integrating common-sense knowledge to reason about complex and often contextualized situations, e.g. ``scheduling a doctor's appointment without a phone''. While current approaches show encouraging results using large language models (LLMs), they are hindered by drawbacks such as costly API calls and reproducibility issues. In this paper, we advocate planning using smaller language models. We present PlaSma, a novel two-pronged approach to endow small language models with procedural knowledge and (constrained) language-based planning capabilities. More concretely, we develop *symbolic procedural knowledge distillation* to enhance the commonsense knowledge in small language models and an *inference-time algorithm* to facilitate more structured and accurate reasoning. In addition, we introduce a new related task, *Replanning*, that requires a revision of a plan to cope with a constrained situation. In both the planning and replanning settings, we show that orders-of-magnitude smaller models (770M-11B parameters) can compete and often surpass their larger teacher models' capabilities. Finally, we showcase successful application of PlaSma in an embodied environment, VirtualHome.

The recent wave of generative AI has sparked unprecedented global attention, with both excitement and concern over potentially superhuman levels of artificial intelligence: models now take only seconds to produce outputs that would challenge or exceed the capabilities even of expert humans. At the same time, models still show basic errors in understanding that would not be expected even in non-expert humans. This presents us with an apparent paradox: how do we reconcile seemingly superhuman capabilities with the persistence of errors that few humans would make? In this work, we posit that this tension reflects a divergence in the configuration of intelligence in today's generative models relative to intelligence in humans. Specifically, we propose and test the **Generative AI Paradox** hypothesis: generative models, having been trained directly to reproduce expert-like outputs, acquire generative capabilities that are not contingent upon---and can therefore exceed---their ability to understand those same types of outputs. This contrasts with humans, for whom basic understanding almost always precedes the ability to generate expert-level outputs. We test this hypothesis through controlled experiments analyzing generation vs.~understanding in generative models, across both language and image modalities. Our results show that although models can outperform humans in generation, they consistently fall short of human capabilities in measures of understanding, as well as weaker correlation between generation and understanding performance, and more brittleness to adversarial inputs. Our findings support the hypothesis that models' generative capability may not be contingent upon understanding capability, and call for caution in interpreting artificial intelligence by analogy to human intelligence.

Wenting Zhao, Justin T. Chiu, Jena Hwang, Faeze Brahman, Jack Hessel, Sanjiban Choudhury, Yejin Choi, Xiang Lorraine Li, Alane Suhr. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Abstract Large language models (LLMs) have garnered a great deal of attention for their exceptional generative performance on commonsense and reasoning tasks. In this work, we investigate LLMs’ capabilities for generalization using a particularly challenging type of statement: generics. Generics express generalizations (e.g., birds can fly) but do so without explicit quantification. They are notable because they generalize over their instantiations (e.g., sparrows can fly) yet hold true even in the presence of exceptions (e.g., penguins do not). For humans, these generic generalizations play a fundamental role in cognition, concept acquisition, and intuitive reasoning. We investigate how LLMs respond to and reason about generics. To this end, we first propose a framework grounded in pragmatics to automatically generate both exceptions and instantiations – collectively exemplars. We make use of focus—a pragmatic phenomenon that highlights meaning-bearing elements in a sentence—to capture the full range of interpretations of generics across different contexts of use. This allows us to derive precise logical definitions for exemplars and operationalize them to automatically generate exemplars from LLMs. Using our system, we generate a dataset of ∼370kexemplars across ∼17k generics and conduct a human validation of a sample of the generated data. We use our final generated dataset to investigate how LLMs reason about generics. Humans have a documented tendency to conflate universally quantified statements (e.g., all birds can fly) with generics. Therefore, we probe whether LLMs exhibit similar overgeneralization behavior in terms of quantification and in property inheritance. We find that LLMs do show evidence of overgeneralization, although they sometimes struggle to reason about exceptions. Furthermore, we find that LLMs may exhibit similar non-logical behavior to humans when considering property inheritance from generics.

Chandra Bhagavatula, Jena D. Hwang, Doug Downey, Ronan Le Bras, Ximing Lu, Lianhui Qin, Keisuke Sakaguchi, Swabha Swayamdipta, Peter West, Yejin Choi. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Jack Hessel, Ana Marasovic, Jena D. Hwang, Lillian Lee, Jeff Da, Rowan Zellers, Robert Mankoff, Yejin Choi. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Valentina Pyatkin, Jena D. Hwang, Vivek Srikumar, Ximing Lu, Liwei Jiang, Yejin Choi, Chandra Bhagavatula. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Emily Allaway, Jena D. Hwang, Chandra Bhagavatula, Kathleen McKeown, Doug Downey, Yejin Choi. Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics. 2023.

Transformer large language models (LLMs) have sparked admiration for their exceptional performance on tasks that demand intricate multi-step reasoning. Yet, these models simultaneously show failures on surprisingly trivial problems. This begs the question: Are these errors incidental, or do they signal more substantial limitations? In an attempt to demystify transformer LLMs, we investigate the limits of these models across three representative compositional tasks---multi-digit multiplication, logic grid puzzles, and a classic dynamic programming problem. These tasks require breaking problems down into sub-steps and synthesizing these steps into a precise answer. We formulate compositional tasks as computation graphs to systematically quantify the level of complexity, and break down reasoning steps into intermediate sub-procedures. Our empirical findings suggest that transformer LLMs solve compositional tasks by reducing multi-step compositional reasoning into linearized subgraph matching, without necessarily developing systematic problem-solving skills. To round off our empirical study, we provide theoretical arguments on abstract multi-step reasoning problems that highlight how autoregressive generations' performance can rapidly decay with increased task complexity.

Peter West, Chandra Bhagavatula, Jack Hessel, Jena Hwang, Liwei Jiang, Ronan Le Bras, Ximing Lu, Sean Welleck, Yejin Choi. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Recent years have brought about a renewed interest in commonsense representation and reasoning in the field of natural language understanding. The development of new commonsense knowledge graphs (CSKG) has been central to these advances as their diverse facts can be used and referenced by machine learning models for tackling new and challenging tasks. At the same time, there remain questions about the quality and coverage of these resources due to the massive scale required to comprehensively encompass general commonsense knowledge. In this work, we posit that manually constructed CSKGs will never achieve the coverage necessary to be applicable in all situations encountered by NLP agents. Therefore, we propose a new evaluation framework for testing the utility of KGs based on how effectively implicit knowledge representations can be learned from them. With this new goal, we propose Atomic 2020, a new CSKG of general-purpose commonsense knowledge containing knowledge that is not readily available in pretrained language models. We evaluate its properties in comparison with other leading CSKGs, performing the first large-scale pairwise study of commonsense knowledge resources. Next, we show that Atomic 2020 is better suited for training knowledge models that can generate accurate, representative knowledge for new, unseen entities and events. Finally, through human evaluation, we show that the few-shot performance of GPT-3 (175B parameters), while impressive, remains ~12 absolute points lower than a BART-based knowledge model trained on Atomic 2020 despite using over 430x fewer parameters.

Jeff Da, Maxwell Forbes, Rowan Zellers, Anthony Zheng, Jena D. Hwang, Antoine Bosselut, Yejin Choi. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Peter West, Ximing Lu, Ari Holtzman, Chandra Bhagavatula, Jena D. Hwang, Yejin Choi. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

In social settings, much of human behavior is governed by unspoken rules of conduct rooted in societal norms.For artificial systems to be fully integrated into social environments, adherence to such norms is a central prerequisite.To investigate whether language generation models can serve as behavioral priors for systems deployed in social settings, we evaluate their ability to generate action descriptions that achieve predefined goals under normative constraints.Moreover, we examine if models can anticipate likely consequences of actions that either observe or violate known norms, or explain why certain actions are preferable by generating relevant norm hypotheses.For this purpose, we introduce Moral Stories, a crowd-sourced dataset of structured, branching narratives for the study of grounded, goaloriented social reasoning.Finally, we propose decoding strategies that combine multiple expert models to significantly improve the quality of generated actions, consequences, and norms compared to strong baselines.1

Social norms-the unspoken commonsense rules about acceptable social behavior-are crucial in understanding the underlying causes and intents of people's actions in narratives.For example, underlying an action such as "wanting to call cops on my neighbor" are social norms that inform our conduct, such as "It is expected that you report crimes."We present SOCIAL CHEMISTRY, a new conceptual formalism to study people's everyday social norms and moral judgments over a rich spectrum of real life situations described in natural language.We introduce SOCIAL-CHEM-101, a large-scale corpus that catalogs 292k rules-of-thumb such as "It is rude to run a blender at 5am" as the basic conceptual units.Each rule-of-thumb is further broken down with 12 different dimensions of people's judgments, including social judgments of good and bad, moral foundations, expected cultural pressure, and assumed legality, which together amount to over 4.5 million annotations of categorical labels and free-text descriptions.Comprehensive empirical results based on state-of-the-art neural models demonstrate that computational modeling of social norms is a promising research direction.Our model framework, NEURAL NORM TRANSFORMER, learns and generalizes SOCIAL-CHEM-101 to successfully reason about previously unseen situations, generating relevant (and potentially novel) attribute-aware social rules-of-thumb.Punching a friend who stole from me.RoT 1: It is unacceptable to injure a person.RoT 2: People should not steal from others.RoT 3: It is bad to betray a friend.RoT 4: It is OK to want to take revenge.

Lianhui Qin, Vered Shwartz, Peter West, Chandra Bhagavatula, Jena D. Hwang, Ronan Le Bras, Antoine Bosselut, Yejin Choi. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP). 2020.

Nathan Schneider, Jena D. Hwang, Vivek Srikumar, Jakob Prange, Austin Blodgett, Sarah R. Moeller, Aviram Stern, Adi Bitan, Omri Abend. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2018.

This paper describes a new approach to detection and tracking of potential cyberattacks from analyzing large quantities of cyber-related webpage text, using ontological knowledge about such attacks combined with composable causal models represented in Probabilistic Soft Logic. The stages of a cyberattack kill chain are viewed as a sequence of both observed and unobserved events (e.g., reconnaissance, weaponize, exploit, install) and explicit mentions of, or related to, such events are examined as potential signals for a future attack. Using a suite of natural language processing techniques, sentences from input news texts are automatically classified according to the described cyberattack event, then enriched with named entity recognition for the rapid detection of key elements that might be associated with potential cyberattacks. We present our work as a framework for rapid and flexible predictive analysis of the ever-increasing amount of cyber-related text data, with initial experiments indicating that event detection using parsing and named entity recognition combined with statistical relational learning show promise in time-series prediction from news text.

OBJECTIVE: To create annotated clinical narratives with layers of syntactic and semantic labels to facilitate advances in clinical natural language processing (NLP). To develop NLP algorithms and open source components. METHODS: Manual annotation of a clinical narrative corpus of 127 606 tokens following the Treebank schema for syntactic information, PropBank schema for predicate-argument structures, and the Unified Medical Language System (UMLS) schema for semantic information. NLP components were developed. RESULTS: The final corpus consists of 13 091 sentences containing 1772 distinct predicate lemmas. Of the 766 newly created PropBank frames, 74 are verbs. There are 28 539 named entity (NE) annotations spread over 15 UMLS semantic groups, one UMLS semantic type, and the Person semantic category. The most frequent annotations belong to the UMLS semantic groups of Procedures (15.71%), Disorders (14.74%), Concepts and Ideas (15.10%), Anatomy (12.80%), Chemicals and Drugs (7.49%), and the UMLS semantic type of Sign or Symptom (12.46%). Inter-annotator agreement results: Treebank (0.926), PropBank (0.891-0.931), NE (0.697-0.750). The part-of-speech tagger, constituency parser, dependency parser, and semantic role labeler are built from the corpus and released open source. A significant limitation uncovered by this project is the need for the NLP community to develop a widely agreed-upon schema for the annotation of clinical concepts and their relations. CONCLUSIONS: This project takes a foundational step towards bringing the field of clinical NLP up to par with NLP in the general domain. The corpus creation and NLP components provide a resource for research and application development that would have been previously impossible.