Large language models (LLMs) achieve strong performance on many tasks, but their progress remains uneven across languages and cultures, often reflecting values latent in English-centric training data. To enable practical cultural alignment, we propose a scalable approach that leverages national social studies curricula as a foundation for culture-aware supervision. We introduce CuCu, an automated multi-agent LLM framework that transforms national textbook curricula into open-ended, culture-specific question-answer pairs. Applying CuCu to the Korean national social studies curriculum, we construct KCaQA, comprising 34.1k open-ended QA pairs. Our quantitative and qualitative analyses suggest that KCaQA covers culture-specific topics and produces responses grounded in local sociocultural contexts.
Language Models (LMs) have been shown to exhibit a strong preference towards entities associated with Western culture when operating in non-Western languages. In this paper, we aim to uncover the origins of entity-related cultural biases in LMs by analyzing several contributing factors, including the representation of entities in pre-training data and the impact of variations in linguistic phenomena across languages. We introduce CAMeL-2, a parallel Arabic-English benchmark of 58,086 entities associated with Arab and Western cultures and 367 masked natural contexts for entities. Our evaluations using CAMeL-2 reveal reduced performance gaps between cultures by LMs when tested in English compared to Arabic. We find that LMs struggle in Arabic with entities that appear at high frequencies in pre-training, where entities can hold multiple word senses. This also extends to entities that exhibit high lexical overlap with languages that are not Arabic but use the Arabic script. Further, we show how frequency-based tokenization leads to this issue in LMs, which gets worse with larger Arabic vocabularies. We will make CAMeL-2 available at: https://github.com/tareknaous/camel2
Chen Cecilia Liu, Hiba Arnaout, Nils Kovačić
et al.
Large language models (LLMs) show promise in offering emotional support and generating empathetic responses for individuals in distress, but their ability to deliver culturally sensitive support remains underexplored due to a lack of resources. In this work, we introduce CultureCare, the first dataset designed for this task, spanning four cultures and including 1729 distress messages, 1523 cultural signals, and 1041 support strategies with fine-grained emotional and cultural annotations. Leveraging CultureCare, we (i) develop and test four adaptation strategies for guiding three state-of-the-art LLMs toward culturally sensitive responses; (ii) conduct comprehensive evaluations using LLM-as-a-Judge, in-culture human annotators, and clinical psychologists; (iii) show that adapted LLMs outperform anonymous online peer responses, and that simple cultural role-play is insufficient for cultural sensitivity; and (iv) explore the application of LLMs in clinical training, where experts highlight their potential in fostering cultural competence in novice therapists.
Large language models exhibit cultural biases and limited cross-cultural understanding capabilities, particularly when serving diverse global user populations. We propose MCEval, a novel multilingual evaluation framework that employs dynamic cultural question construction and enables causal analysis through Counterfactual Rephrasing and Confounder Rephrasing. Our comprehensive evaluation spans 13 cultures and 13 languages, systematically assessing both cultural awareness and cultural bias across different linguistic scenarios. The framework provides 39,897 cultural awareness instances and 17,940 cultural bias instances. Experimental results reveal performance disparities across different linguistic scenarios, demonstrating that optimal cultural performance is not only linked to training data distribution, but also is related to language-culture alignment. The evaluation results also expose the fairness issue, where approaches appearing successful in the English scenario create substantial disadvantages. MCEval represents the first comprehensive multilingual cultural evaluation framework that provides deeper insights into LLMs' cultural understanding.
Research on how humans perceive aesthetics in shapes, colours, and music has predominantly focused on Western populations, limiting our understanding of how cultural environments shape aesthetic preferences. We present a large-scale cross-cultural study examining aesthetic preferences across five distinct modalities extensively explored in the literature: shape, curvature, colour, musical harmony and melody. We gather 401,403 preference judgements from 4,835 participants across 10 countries, systematically sampling two-dimensional parameter spaces for each modality. The findings reveal both universal patterns and cultural variations. Preferences for shape and curvature cross-culturally demonstrate a consistent preference for symmetrical forms. While colour preferences are categorically consistent, ratio-like preferences vary across cultures. Musical harmony shows strong agreement in interval relationships despite differing regions of preference within the broad frequency spectrum, while melody shows the highest cross-cultural variation. These results suggest that aesthetic preferences emerge from an interplay between shared perceptual mechanisms and cultural learning.
Contrastive pre-training on large-scale image-text pair datasets has driven major advances in vision-language representation learning. Recent work shows that pretraining on global data followed by language or culture specific fine-tuning is effective for improving performance in target domains. With the availability of strong open-weight multilingual models such as SigLIP2, this paradigm has become increasingly practical. However, for Japanese, the scarcity of large-scale, high-quality image-text pair datasets tailored to Japanese language and cultural content remains a key limitation. To address this gap, we introduce WAON, the largest Japanese image-text pair dataset constructed from Japanese web content in Common Crawl, containing approximately 155 million examples. Our dataset construction pipeline employs filtering and deduplication to improve dataset quality. To improve the quality and reliability of evaluation on Japanese cultural tasks, we also construct WAON-Bench, a manually curated benchmark for Japanese cultural image classification comprising 374 classes, which addresses issues in the existing benchmark such as category imbalance and label-image mismatches. Our experiments demonstrate that fine-tuning on WAON improves model performance on Japanese cultural benchmarks more efficiently than existing datasets, achieving state-of-the-art results among publicly available models of comparable architecture. We release our dataset, model, and code.
Michelle J. Cummings-Koether, Franziska Durner, Theophile Shyiramunda
et al.
The swift diffusion of artificial intelligence (AI) raises critical questions about how cultural contexts shape adoption patterns and their consequences for human daily life. This study investigates the cultural dimensions of AI adoption and their influence on cognitive strategies across nine national contexts in Europe, Africa, Asia, and South America. Drawing on survey data from a diverse pilot sample (n = 21) and guided by cross-cultural psychology, digital ethics, and sociotechnical systems theory, we examine how demographic variables (age, gender, professional role) and cultural orientations (language, values, and institutional exposure) mediate perceptions of trust, ethical acceptability, and reliance on AI. Results reveal two key findings: First, cultural factors, particularly language and age, significantly affect AI adoption and perceptions of reliability with older participants reporting higher engagement with AI for educational purposes. Second, ethical judgment about AI use varied across domains, with professional contexts normalizing its role as a pragmatic collaborator while academic settings emphasized risks of plagiarism. These findings extend prior research on culture and technology adoption by demonstrating that AI use is neither universal nor neutral but culturally contingent, domain-specific, and ethically situated. The study highlights implications for AI use in education, professional practice, and global technology policy, pointing at actions that enable usage of AI in a way that is both culturally adaptive and ethically robust.
We examine how production and the development of property rights interact with cultural transmission to shape the treatment of the elderly across societies. Our model posits that respect for the elderly arises endogenously: parents invest in cultivating cultural values in their children, who later reciprocate in proportion to this investment. We show that this model is functionally equivalent to one in which cultural goods are transferred by the elderly. We focus on the distinct roles of property rights, finding that while insecure output rights may promote elderly welfare, secure rights over productive resources can have comparable benefits. The model reveals a nonlinear relationship between cultural sophistication, property rights, and economic factors such as the capital and land intensity of production, driving variations in elderly well-being across societies. Finally, we consider how the model suggests demographic, technological, and policy changes influence elderly well-being across the spectrum of development.
Much work on the cultural awareness of large language models (LLMs) focuses on the models' sensitivity to geo-cultural diversity. However, in addition to cross-cultural differences, there also exists common ground across cultures. For instance, a bridal veil in the United States plays a similar cultural-relevant role as a honggaitou in China. In this study, we introduce a benchmark dataset CUNIT for evaluating decoder-only LLMs in understanding the cultural unity of concepts. Specifically, CUNIT consists of 1,425 evaluation examples building upon 285 traditional cultural-specific concepts across 10 countries. Based on a systematic manual annotation of cultural-relevant features per concept, we calculate the cultural association between any pair of cross-cultural concepts. Built upon this dataset, we design a contrastive matching task to evaluate the LLMs' capability to identify highly associated cross-cultural concept pairs. We evaluate 3 strong LLMs, using 3 popular prompting strategies, under the settings of either giving all extracted concept features or no features at all on CUNIT Interestingly, we find that cultural associations across countries regarding clothing concepts largely differ from food. Our analysis shows that LLMs are still limited to capturing cross-cultural associations between concepts compared to humans. Moreover, geo-cultural proximity shows a weak influence on model performance in capturing cross-cultural associations.
Pretrained large language models have revolutionized many applications but still face challenges related to cultural bias and a lack of cultural commonsense knowledge crucial for guiding cross-culture communication and interactions. Recognizing the shortcomings of existing methods in capturing the diverse and rich cultures across the world, this paper introduces a novel approach for massively multicultural knowledge acquisition. Specifically, our method strategically navigates from densely informative Wikipedia documents on cultural topics to an extensive network of linked pages. Leveraging this valuable source of data collection, we construct the CultureAtlas dataset, which covers a wide range of sub-country level geographical regions and ethnolinguistic groups, with data cleaning and preprocessing to ensure textual assertion sentence self-containment, as well as fine-grained cultural profile information extraction. Our dataset not only facilitates the evaluation of language model performance in culturally diverse contexts but also serves as a foundational tool for the development of culturally sensitive and aware language models. Our work marks an important step towards deeper understanding and bridging the gaps of cultural disparities in AI, to promote a more inclusive and balanced representation of global cultures in the digital domain.
Toxicity and abuse are common in online peer-production communities. The social structure of peer-production communities that aim to produce accurate and trustworthy information require some conflict and gate-keeping to spur content production and curation. However, conflict and gate-keeping often devolve into hierarchical power structures which punish newcomers and lock out marginalized groups through entrenched cultural norms. Community administrators often focus on content quality, rather than consideration for all user safety, to promote community growth and survival. Once toxic cultural norms dominate a peer-production community, it is very difficult for community administrators to stop these behaviors from undermining inclusive peer-production. We propose developing a "handbook of intelligent system design" that attempts to frame design protocols to better read user-community culture and accurately distinguish toxic negative interactions from beneficial conflict.
In this paper, we address the modeling issues of cell movement and division with a special focus on the phenomenon of volume exclusion in a lattice-based, exact stochastic simulation framework. We propose a new exact method, called Reduced Rate Method -- RRM, that is substantially quicker than the previously used exclusion method, for large number of cells. In addition, we introduce three novel reaction types: the contact-inhibited, the contact-promoted, and the spontaneous reactions. To the best of our knowledge, these reaction types have not been taken into account in lattice-based stochastic simulations of cell cultures. These new types of events may be easily applied to complicated systems, enabling the generation of biologically feasible stochastic cell culture simulations. Furthermore, we show that the exclusion algorithm and our RRM algorithm are mathematically equivalent in the sense that the next reaction to be realized and the corresponding sojourn time both belong to the same reaction and time distributions in the two approaches -- even with the newly introduced reaction types. Exact, agent-based, stochastic methods of cell culture simulations seem to be undervalued and are mostly used as benchmarking tools to validate deterministic approximations of the corresponding stochastic models. Our proposed methods are exact, they are easy to implement, have a high predictive value, and can be conveniently extended with new features. Therefore, these approaches promise a great potential.
James M. Borg, Andrew Buskell, Rohan Kapitany
et al.
The goal of Artificial Life research, as articulated by Chris Langton, is "to contribute to theoretical biology by locating life-as-we-know-it within the larger picture of life-as-it-could-be" (1989, p.1). The study and pursuit of open-ended evolution in artificial evolutionary systems exemplifies this goal. However, open-ended evolution research is hampered by two fundamental issues; the struggle to replicate open-endedness in an artificial evolutionary system, and the fact that we only have one system (genetic evolution) from which to draw inspiration. Here we argue that cultural evolution should be seen not only as another real-world example of an open-ended evolutionary system, but that the unique qualities seen in cultural evolution provide us with a new perspective from which we can assess the fundamental properties of, and ask new questions about, open-ended evolutionary systems, especially in regard to evolved open-endedness and transitions from bounded to unbounded evolution. Here we provide an overview of culture as an evolutionary system, highlight the interesting case of human cultural evolution as an open-ended evolutionary system, and contextualise cultural evolution under the framework of (evolved) open-ended evolution. We go on to provide a set of new questions that can be asked once we consider cultural evolution within the framework of open-ended evolution, and introduce new insights that we may be able to gain about evolved open-endedness as a result of asking these questions.
This study aims: 1) How to Planning, Organizing, Monitoring and Implementing Change Management 2) What are the steps of the Principal in developing organizational culture 3) What are the supporting and inhibiting factors of the effectiveness of change management in developing organizational culture at SMP IT Darul Azhar Southeast Aceh. This type of research is qualitative. The approach in this study uses a descriptive approach. As informants in this study were the principal, teachers, and students. Research data collection techniques using observation, interviews and documentation. Data analysis techniques used data reduction, data presentation, and drawing conclusions. Techniques to guarantee data validity are credibility, transferability, dependability, and conformability. The results of this study indicate that there are three most important points in the effectiveness of change management in developing the organizational culture of SMP IT Darul Azhar, namely: In change management, the principal is planning according to the needs of students and then by organizing by dividing tasks to teachers to carry out planning, then supervision, namely to know the progress of this change management. In developing the school principal's organizational culture, namely the implementation of morning apples, dhuha prayers, and tadarus al-qur'an after the midday prayers. The supporting factor is the willingness of teachers and students to be better at applying religious values. The inhibiting factor is the lack of competent human resources (HR), loyal to change.
This study aims to determine the effect of Organizational Culture Transformation and Talent Management on Organizational Effectiveness mediated by Work. This research is based on organizational effectiveness in companies that have changed strategy due to encouragement from external industry factors so that organizational culture and talent management need to be carried out which are expected to increase organizational effectiveness. organization. The sample in this study was taken using a data collection method called purposive sampling. The sample used was 80 respondents. Data analysis using PLS (Partial Least Square) analysis technique through SmartPLS software. In addition, to test the hypothesis and the seventh, a trial was conducted to test the mediating/intervention variable in the hypothesis. The results showed that organizational culture transformation had a positive but not significant effect on organizational effectiveness. Furthermore, talent management has a positive and significant effect on organizational effectiveness. The results also show that the transformation of organizational culture and talent management has a positive but not significant effect on organizational effectiveness through job satisfaction owned by employees of PT. XYZ.
The term glass ceiling is applied to the well-established phenomenon in which women and people of color are consistently blocked from reaching the upper-most levels of the corporate hierarchy. Focusing on gender, we present an agent-based model that explores how empirically established mechanisms of interpersonal discrimination coevolve with social norms at both the organizational (meso) and societal (macro) levels to produce this glass ceiling effect for women. Our model extends our understanding of how the glass ceiling arises, and why it can be resistant to change. We do so by synthesizing existing psychological and structural theories of discrimination into a mathematical model that quantifies explicitly how complex organizational systems can produce and maintain inequality. We discuss implications of our findings for both intervention and future empirical analyses, and provide open-source code for those wishing to adapt or extend our work.
Many cultural traits characterizing intelligent behaviors are now thought to be transmitted through statistical learning, motivating us to study its effects on cultural evolution. We conduct a large-scale music data analysis and observe that various statistical parameters of musical products approximately follow the beta distribution and other conjugate distributions. We construct a simple model of cultural evolution incorporating statistical learning and analytically show that conjugate distributions emerge at equilibrium in the presence of oblique transmission. The results demonstrate that the distribution of a cultural trait within a population depends on the individual's model for cultural production (the conjugate distribution law), and reveal interesting possibilities for theoretical and experimental studies on cultural evolution and social learning.
Pablo Andres Contreras Kallens, Rick Dale, Paul E. Smaldino
Categorization is a fundamental function of minds, with wide ranging implications for the rest of the cognitive system. In humans, categories are shared and communicated between minds, thus requiring explanations at the population level. In this paper, we discuss the current state of research on the cultural evolution of categorization. We begin by delineating key properties of categories in need of evolutionary explanation. We then review computational modeling and laboratory studies of category evolution, including their major insights and limitations. Finally, we discuss remaining challenges for understanding the cultural evolution of categorization.
Victor Braberman, Nicolas D'Ippolito, Jeff Kramer
et al.
An architectural approach to self-adaptive systems involves runtime change of system configuration (i.e., the system's components, their bindings and operational parameters) and behaviour update (i.e., component orchestration). Thus, dynamic reconfiguration and discrete event control theory are at the heart of architectural adaptation. Although controlling configuration and behaviour at runtime has been discussed and applied to architectural adaptation, architectures for self-adaptive systems often compound these two aspects reducing the potential for adaptability. In this paper we propose a reference architecture that allows for coordinated yet transparent and independent adaptation of system configuration and behaviour.