Purpose: To benchmark open-source or commercial medical image-specific VLMs against real-world radiologist-written reports. Methods: This retrospective study included adult patients who presented to the emergency department between January 2022 and April 2025 and underwent same-day CXR and CT for febrile or respiratory symptoms. Reports from five VLMs (AIRead, Lingshu, MAIRA-2, MedGemma, and MedVersa) and radiologist-written reports were randomly presented and blindly evaluated by three thoracic radiologists using four criteria: RADPEER, clinical acceptability, hallucination, and language clarity. Comparative performance was assessed using generalized linear mixed models, with radiologist-written reports treated as the reference. Finding-level analyses were also performed with CT as the reference. Results: A total of 478 patients (median age, 67 years [interquartile range, 50-78]; 282 men [59.0%]) were included. AIRead demonstrated the lowest RADPEER 3b rate (5.3% [76/1434] vs. radiologists 13.9% [200/1434]; P<.001), whereas other VLMs showed higher disagreement rates (16.8-43.0%; P<.05). Clinical acceptability was the highest with AIRead (84.5% [1212/1434] vs. radiologists 74.3% [1065/1434]; P<.001), while other VLMs performed worse (41.1-71.4%; P<.05). Hallucinations were rare with AIRead, comparable to radiologists (0.3% [4/1425]) vs. 0.1% [1/1425]; P=.21), but frequent with other models (5.4-17.4%; P<.05). Language clarity was higher with AIRead (82.9% [1189/1434]), Lingshu (88.0% [1262/1434]), and MedVersa (88.4% [1268/1434]) compared with radiologists (78.1% [1120/1434]; P<.05). Sensitivity varied substantially across VLMs for the common findings: AIRead, 15.5-86.7%; Lingshu, 2.4-86.7%; MAIRA-2, 6.0-72.0%; MedGemma, 4.8-76.7%; and MedVersa, 20.2-69.3%. Conclusion: Medical VLMs for CXR report generation exhibited variable performance in report quality and diagnostic measures.

Like visual processing, language processing is susceptible to illusions in which people systematically misperceive stimuli. In one such case--the comparative illusion (CI), e.g., More students have been to Russia than I have--comprehenders tend to judge the sentence as acceptable despite its underlying nonsensical comparison. Prior research has argued that this phenomenon can be explained as Bayesian inference over a noisy channel: the posterior probability of an interpretation of a sentence is proportional to both the prior probability of that interpretation and the likelihood of corruption into the observed (CI) sentence. Initial behavioral work has supported this claim by evaluating a narrow set of alternative interpretations of CI sentences and showing that comprehenders favor interpretations that are more likely to have been corrupted into the illusory sentence. In this study, we replicate and go substantially beyond this earlier work by directly predicting the strength of illusion with a quantitative model of the posterior probability of plausible interpretations, which we derive through a novel synthesis of statistical language models with human behavioral data. Our model explains not only the fine gradations in the strength of CI effects, but also a previously unexplained effect caused by pronominal vs. full noun phrase than-clause subjects. These findings support a noisy-channel theory of sentence comprehension by demonstrating that the theory makes novel predictions about the comparative illusion that bear out empirically. This outcome joins related evidence of noisy channel processing in both illusory and non-illusory contexts to support noisy channel inference as a unified computational-level theory of diverse language processing phenomena.

Large language models (LLMs) are increasingly deployed in culturally diverse environments, yet existing evaluations of cultural competence remain limited. Existing methods focus on de-contextualized correctness or forced-choice judgments, overlooking the need for cultural understanding and reasoning required for appropriate responses. To address this gap, we introduce a set of benchmarks that, instead of directly probing abstract norms or isolated statements, present models with realistic situational contexts that require culturally grounded reasoning. In addition to the standard Exact Match metric, we introduce four complementary metrics (Coverage, Specificity, Connotation, and Coherence) to capture different dimensions of model's response quality. Empirical analysis across frontier models reveals that thin evaluation systematically overestimates cultural competence and produces unstable assessments with high variance. In contrast, thick evaluation exposes differences in reasoning depth, reduces variance, and provides more stable, interpretable signals of cultural understanding.

Modern preference alignment techniques, such as Best-of-N (BoN) sampling, rely on reward models trained with pairwise comparison data. While effective at learning relative preferences, this paradigm fails to capture a signal of response acceptability, leaving systems vulnerable to selecting the least bad of many unacceptable options. This is particularly problematic for hard prompts, where the risk of such false acceptances increases with the number of samples. In this paper, we address this critical reliability gap by introducing a new data collection and modeling framework. By augmenting preference data with an outside option, inspired by discrete choice models, we train a reward model that can distinguish not just what is better, but what is good enough. We leverage this capability to create an adaptive inference strategy, best of mini-N in-loop, which partitions the generation budget into sequential loops with a calibrated, early-exit condition. Our experiments show that when tuned as an alignment guardrail, it reduces reliability failures by 70%, and when tuned as an inference accelerator, it improves average inference speed by over 22% in IMDB-sentiment setting. We thus provide a principled and flexible framework for practitioners to explicitly manage the trade-off between reliability and computational efficiency.

Conditional acceptability refers to how plausible a conditional statement is perceived to be. It plays an important role in communication and reasoning, as it influences how individuals interpret implications, assess arguments, and make decisions based on hypothetical scenarios. When humans evaluate how acceptable a conditional "If A, then B" is, their judgments are influenced by two main factors: the $\textit{conditional probability}$ of $B$ given $A$, and the $\textit{semantic relevance}$ of the antecedent $A$ given the consequent $B$ (i.e., whether $A$ meaningfully supports $B$). While prior work has examined how large language models (LLMs) draw inferences about conditional statements, it remains unclear how these models judge the $\textit{acceptability}$ of such statements. To address this gap, we present a comprehensive study of LLMs' conditional acceptability judgments across different model families, sizes, and prompting strategies. Using linear mixed-effects models and ANOVA tests, we find that models are sensitive to both conditional probability and semantic relevance-though to varying degrees depending on architecture and prompting style. A comparison with human data reveals that while LLMs incorporate probabilistic and semantic cues, they do so less consistently than humans. Notably, larger models do not necessarily align more closely with human judgments.

As artificial intelligence rapidly transforms society, developers and policymakers struggle to anticipate which applications will face public moral resistance. We propose that these judgments are not idiosyncratic but systematic and predictable. In a large, preregistered study (N = 587, U.S. representative sample), we used a comprehensive taxonomy of 100 AI applications spanning personal and organizational contexts-including both functional uses and the moral treatment of AI itself. In participants' collective judgment, applications ranged from highly unacceptable to fully acceptable. We found this variation was strongly predictable: five core moral qualities-perceived risk, benefit, dishonesty, unnaturalness, and reduced accountability-collectively explained over 90% of the variance in acceptability ratings. The framework demonstrated strong predictive power across all domains and successfully predicted individual-level judgments for held-out applications. These findings reveal that a structured moral psychology underlies public evaluation of new technologies, offering a powerful tool for anticipating public resistance and guiding responsible innovation in AI.

Abstract--Large Language Models (LLMs) are increasingly integrated into software engineering (SE) tools for tasks that extend beyond code synthesis, including judgment under uncertainty and reasoning in ethically significant contexts. We present a fully automated framework for assessing ethical reasoning capabilities across 16 LLMs in a zero-shot setting, using 30 real-world ethically charged scenarios. Each model is prompted to identify the most applicable ethical theory to an action, assess its moral acceptability, and explain the reasoning behind their choice. Responses are compared against expert ethicists' choices using inter-model agreement metrics. Our results show that LLMs achieve an average Theory Consistency Rate (TCR) of 73.3% and Binary Agreement Rate (BAR) on moral acceptability of 86.7%, with interpretable divergences concentrated in ethically ambiguous cases. A qualitative analysis of free-text explanations reveals strong conceptual convergence across models despite surface-level lexical diversity. These findings support the potential viability of LLMs as ethical inference engines within SE pipelines, enabling scalable, auditable, and adaptive integration of user-aligned ethical reasoning. Our focus is the Ethical Interpreter component of a broader profiling pipeline: we evaluate whether current LLMs exhibit sufficient interpretive stability and theory-consistent reasoning to support automated profiling. Autonomous systems are increasingly becoming an integral part of our daily lives across diverse domains [1], [2]. These systems can operate independently without any human intervention and make decisions acting on behalf of their users [3]-[6]. Their rapid growth brings both opportunities and challenges. From a software engineering perspective, as these systems become pervasive, a key challenge is designing systems that, beyond meeting technical requirements, also account for ethical considerations [7]-[11]. Recently, various studies have focused on the ethical implications of these software-intensive systems on individuals and society [10], [12]-[15]. Software engineering ethics encompasses principles and rules that guide engineers' decisions throughout the design and development process [16]. V arious approaches have also been introduced that ensure that systems align with broad ethical values like fairness, transparency, and safety [17]-[22].

Artificial intelligence (AI)-based computer perception (CP) technologies use mobile sensors to collect behavioral and physiological data for clinical decision-making. These tools can reshape how clinical knowledge is generated and interpreted. However, effective integration of these tools into clinical workflows depends on how developers balance clinical utility with user acceptability and trustworthiness. Our study presents findings from 20 in-depth interviews with developers of AI-based CP tools. Interviews were transcribed and inductive, thematic analysis was performed to identify 4 key design priorities: 1) to account for context and ensure explainability for both patients and clinicians; 2) align tools with existing clinical workflows; 3) appropriately customize to relevant stakeholders for usability and acceptability; and 4) push the boundaries of innovation while aligning with established paradigms. Our findings highlight that developers view themselves as not merely technical architects but also ethical stewards, designing tools that are both acceptable by users and epistemically responsible (prioritizing objectivity and pushing clinical knowledge forward). We offer the following suggestions to help achieve this balance: documenting how design choices around customization are made, defining limits for customization choices, transparently conveying information about outputs, and investing in user training. Achieving these goals will require interdisciplinary collaboration between developers, clinicians, and ethicists.

This paper presents an investigation into the impact of adding adjustment features to an existing sign language (SL) avatar on a Microsoft Hololens 2 device. Through a detailed analysis of interactions of expert German Sign Language (DGS) users with both adjustable and non-adjustable avatars in a specific use case, this study identifies the key factors influencing the comprehensibility, the user experience (UX), and the acceptability of such a system. Despite user preference for adjustable settings, no significant improvements in UX or comprehensibility were observed, which remained at low levels, amid missing SL elements (mouthings and facial expressions) and implementation issues (indistinct hand shapes, lack of feedback and menu positioning). Hedonic quality was rated higher than pragmatic quality, indicating that users found the system more emotionally or aesthetically pleasing than functionally useful. Stress levels were higher for the adjustable avatar, reflecting lower performance, greater effort and more frustration. Additionally, concerns were raised about whether the Hololens adjustment gestures are intuitive and easy to familiarise oneself with. While acceptability of the concept of adjustability was generally positive, it was strongly dependent on usability and animation quality. This study highlights that personalisation alone is insufficient, and that SL avatars must be comprehensible by default. Key recommendations include enhancing mouthing and facial animation, improving interaction interfaces, and applying participatory design.

There is a lack of an evaluation methodology that estimates the extent to which large language models (LLMs) use code-switching (CS) in the same way as bilinguals. Existing methods do not have wide language coverage, fail to account for the diverse range of CS phenomena, or do not scale. We propose an intervention based on minimal pairs of CS. Each minimal pair contains one naturally occurring CS sentence and one minimally manipulated variant. We collect up to 1,000 such pairs each for 11 language pairs. Our human experiments show that, for every language pair, bilinguals consistently prefer the naturally occurring CS sentence. Meanwhile our experiments with current LLMs show that the larger the model, the more consistently it assigns higher probability to the naturally occurring CS sentence than to the variant. In accordance with theoretical claims, the largest probability differences arise in those pairs where the manipulated material consisted of closed-class words.