Search Results (581)

Harmful language in conversational AI poses distinct safety and governance challenges, as Large Language Model (LLM) chatbots interact in private, one-to-one settings. Understanding the types of harm and their temporal concentration is crucial for responsible deployment and time-aware moderation. This study investigates the types and diurnal dynamics of harmful speech, comparing patterns between play-oriented chit-chat and task-oriented LLM services.We analyze two large-scale, real-world English corpora: a chit-chat service (SimSimi; 8.7 M utterances) and an LLM service (WildChat; 610 K utterances). Using the Perspective API for multi-label classification (Toxicity, Profanity, Insult, Identity Attack, Threat), we estimate the incidence of harm categories and compare their distribution across five dayparts. Our analysis shows that harmful speech is significantly more prevalent in the chit-chat context than in the LLM service. Across both platforms, Toxicity and Profanity are the dominant categories. Temporally, harmful speech concentrates most frequently during the dawn daypart. We contribute an empirical baseline on how harm varies by chatbot modality and time of day, offering practical guidance for designing dynamic, platform-specific moderation policies. Full article

This research explores the language features used by leading consumer brands with successful marketing in their promotional messages. Coca-Cola, McDonald’s, PepsiCo, Mondelez, and Unilever were selected because they appear in Effie’s Most Effective Marketers’ Index and are active on a range of media platforms. A group of 225 marketing texts, made up of social media posts, video advertisement transcripts, and website content, was examined using a corpus-based method based on Biber’s MDA framework. The goal was to find common lexicogrammatical patterns in top consumer brands on five different dimensions. Many advertisements included personal pronouns, commands, and words that suggest possibility or necessity. The findings also show that most social media posts provided information, yet had a moderate impact on persuasion. Abstract nouns, passive voice, and formal connectors were found to make the website and press release texts the most impersonal and explicit. The research discovered that Unilever’s language was more informational and abstract, but McDonald’s language was mixed-purpose and non-abstract. Overall, the results indicate that brands use vocabulary and grammar to fit each platform, but maintain their brand identity. Thus, successful consumer brands use different lexicogrammatical patterns in various media to achieve their objectives. Full article

Precision agriculture increasingly recognizes the transformative potential of unmanned aerial systems (UASs) for crop monitoring and field assessment, yet research consistently highlights significant usability barriers as the main constraints to widespread adoption. Complex mission planning processes, including detailed flight plan creation and way point management, pose substantial technical challenges that mainly affect non-expert operators. Farmers and their teams generally prefer user-friendly, straightforward tools, as evidenced by the rapid adoption of GPS guidance systems, which underscores the need for simpler mission planning in UAS operations. To enhance accessibility and safety in UAS control, especially for non-expert operators in agriculture and related fields, we propose a Secure UAS Control Framework (SAUCF): a comprehensive system for natural-language-driven UAS mission management with integrated dual-factor biometric authentication. The framework converts spoken user instructions into executable flight plans by leveraging a language-model-powered mission planner that interprets transcribed voice commands and generates context-aware operational directives, including takeoff, location monitoring, return-to-home, and landing operations. Mission orchestration is performed through a large language model (LLM) agent, coupled with a human-in-the-loop supervision mechanism that enables operators to review, adjust, or confirm mission plans before deployment. Additionally, SAUCF offers a manual override feature, allowing users to assume direct control or interrupt missions at any stage, ensuring safety and adaptability in dynamic environments. Proof-of-concept demonstrations on a UAS plat-form with on-board computing validated reliable speech-to-text transcription, biometric verification via voice matching and face authentication, and effective Sim2Real transfer of natural-language-driven mission plans from simulation environments to physical UAS operations. Initial evaluations showed that SAUCF reduced mission planning time, minimized command errors, and simplified complex multi-objective workflows compared to traditional waypoint-based tools, though comprehensive field validation remains necessary to confirm these preliminary findings. The integration of natural-language-based interaction, real-time identity verification, human-in-the-loop LLM orchestration, and manual override capabilities allows SAUCF to significantly lower the technical barrier to UAS operation while ensuring mission security, operational reliability, and operator agency in real-world conditions. These findings lay the groundwork for systematic field trials and suggest that prioritizing ease of operation in mission planning can drive broader deployment of UAS technologies. Full article

The Bunun are one of the Indigenous peoples of Taiwan, traditionally known for their mountain agriculture and communal cooperation. The cultural transmission of the Bunun people refers to the intergenerational process through which knowledge, values, and beliefs are passed down via language, rituals, music, hunting ethics, and daily practices. This system not only sustains ethnic identity but also demonstrates cultural resilience. However, historical colonization, forced relocation, assimilation in education, and modernization have disrupted these pathways. In recent years, elders, cultural health stations, community universities, and schools have collaboratively promoted cultural revitalization through curriculum design, ritual restoration, and language teaching. Full article

Large Language Models (LLMs) excel in translation, among other things, demonstrating competitive performance for many language pairs in zero- and few-shot settings. But unlike dedicated neural machine translation models, LLMs are not trained on any translation-related objective. What explains their remarkable translation abilities? Are these abilities grounded in “incidental bilingualism” in training data? Does instruction tuning contribute to it? Are LLMs capable of aligning and leveraging semantically identical or similar monolingual contents from different corners of the internet that are unlikely to fit in a single context window? I offer some reflections on this topic, informed by recent studies and growing user experience. My working hypothesis is that LLMs’ translation abilities originate in two different types of pre-training data that may be internalized by the models in different ways: Local and Global. “Local learning” makes use of bilingual signals present within a single training context window (e.g., an English sentence soon followed by its Chinese translation in the training data). “Global learning,” in contrast, capitalizes on mining semantically related monolingual contents that are spread out over the LLMs’ pre-training data. The key to explaining the origins of LLMs’ translation capabilities is a continuous iteration between Local and Global learning, which is a natural and helpful consequence of batch training. I discuss the prospects for testing the “duality hypothesis” empirically and its implications for reconceptualizing translation, human and machine, in the age of deep learning. Full article

Background: Cognitive Reserve (CR) is a theoretical construct developed to explain individual differences in resilience to age-related cognitive decline. Empirical evidence supports its positive role across multiple cognitive domains. However, behavioral research has primarily focused on areas either vulnerable to aging, such as memory, or relatively preserved, such as language. In contrast, the relationship between CR and task-specific performance in domains like visuospatial processing—a domain critical for everyday functioning—remains underexplored. This study investigates whether CR, as measured by the Cognitive Reserve Index Questionnaire (CRIq), predicts performance in mental rotation tasks in healthy older adults. Methods: Participants (age 55–85) completed two tasks: (1) a hand laterality task, requiring judgments about whether a rotated hand image (palm or back view) was left or right; and (2) a letter-congruency task, in which participants determined whether simultaneously presented rotated letters were identical or mirror-reversed. Results: Generalized and linear mixed-effects models revealed a protective effect of cognitive reserve, with higher CRIq scores significantly predicting greater accuracy in both tasks. Efficiency benefits (i.e., shorter reaction times) were evident mainly in the easiest conditions, suggesting that CR supports processing resources more effectively under moderate rather than maximal task demands. This pattern indicates that cognitive reserve does not uniformly enhance performance but instead modulates the allocation of cognitive resources in a context-dependent manner. Conclusions: To our knowledge, this is the first study to demonstrate a modulatory role of CR on visuospatial abilities in healthy older adults. These findings open new avenues for investigating how CR may differentially affect performance across a broader spectrum of cognitive functions, including attention, executive control, and spatial processing. A better understanding of these mechanisms could inform targeted cognitive interventions to strengthen resilience and promote successful aging. Full article

Large Language Models (LLMs) are increasingly used for rubric-based assessment, yet reliability is limited by instability, bias, and weak diagnostics. We present EvalCouncil, a committee-and-chief framework for rubric-guided grading with auditable traces and a human adjudication baseline. Our objectives are to (i) characterize domain structure in Human–LLM alignment, (ii) assess robustness to concordance tolerance and panel composition, and (iii) derive a domain-adaptive audit policy grounded in dispersion and chief–panel differences. Authentic student responses from two domains–Computer Networks (CNs) and Machine Learning (ML)–are graded by multiple heterogeneous LLM evaluators using identical rubric prompts. A designated chief arbitrator operates within a tolerance band and issues the final grade. We quantify within-panel dispersion via MPAD (mean pairwise absolute deviation), measure chief–panel concordance (e.g., absolute error and bias), and compute Human–LLM deviation. Robustness is examined by sweeping the tolerance and performing leave-one-out perturbations of panel composition. All outputs and reasoning traces are stored in a graph database for full provenance. Human–LLM alignment exhibits systematic domain dependence: ML shows tighter central tendency and shorter upper tails, whereas CN displays broader dispersion with heavier upper tails and larger extreme spreads. Disagreement increases with item difficulty as captured by MPAD, concentrating misalignment on a relatively small subset of items. These patterns are stable to tolerance variation and single-grader removals. The signals support a practical triage policy: accept low-dispersion, small-gap items; apply a brief check to borderline cases; and adjudicate high-dispersion or large-gap items with targeted rubric clarification. EvalCouncil instantiates a committee-and-chief, rubric-guided grading workflow with committee arbitration, a human adjudication baseline, and graph-based auditability in a real classroom deployment. By linking domain-aware dispersion (MPAD), a policy tolerance dial, and chief–panel discrepancy, the study shows how these elements can be combined into a replicable, auditable, and capacity-aware approach for organizing LLM-assisted grading and identifying instability and systematic misalignment, while maintaining pedagogical interpretability. Full article

This study explores the fluidity of multilingual and intercultural identities among Australian university students participating in a study abroad (SA) program in China. It pilots a structured analytical framework to analyse visual metaphors created by the participants, examining how Chinese language learners depict their evolving identities, thereby uncovering their fluid nature. An analysis of three case studies of Australian students—one of Chinese heritage, one Australian-born of Serbian heritage, and one Japanese-born who moved to Australia in childhood—illustrates their unique trajectories of identity formation. Multimodal data reveals that while the visual metaphors of all three students point to fluidity, with sub-themes of dynamism, blending, and layeredness, the SA experience triggers diverse individual processes of identity negotiation and transformation. This study contributes to the fields of multilingual and intercultural education and SA through its innovative use of a visual metaphor analysis approach, which effectively captures and decodes the complexities of intercultural development among language learners in an SA environment. The study advocates visual metaphor as a valuable tool for both researching and understanding how multilinguals, especially those whose first and home language is not English, reflect on their multilingual and intercultural identity and experiences. Full article

The study examines the complexities of the Aláròyé newspaper’s digital transformation. It seeks to understand how the organisation’s matrix, which includes circulation, readers, and advertising revenue, has been impacted by the shift to digital platforms while preserving its historical print business. Anchored on the uses and gratifications as well as technological determinism theory, the study demonstrates how digital technology influences the operational structures of media organisations. The study adopts a qualitative research design through interviews (with the newspaper’s publisher) and netnography with the newspaper’s social media community on Facebook. Data obtained are qualitatively analysed using the thematic analysis method. Findings from the study show that Aláròyé has made significant strides in embracing technology to enhance its operations, marking a notable shift from traditional practices to more modern, digital-driven strategies. The reception from the audience has been positive, with a growing number of readers engaging with the content across various digital platforms. The shift to digital media has allowed Aláròyé to expand its reach and foster a stronger connection with its audience, which is essential for the long-term success of the indigenous language newspaper. The study enhances the existing scholarship on indigenous language media by elucidating adaptive strategies and audience dynamics within African digital journalism. This establishes a framework for comprehending how indigenous language news outlets can sustain relevance in the digital era by preserving their cultural identity and social mission. Full article

Anorexia nervosa is among the most lethal psychiatric conditions. Online pro-anorexia (“pro-ana”) communities may frame starvation and restriction in moral or spiritual terms. This study explored how pro-ana discourse on X (formerly Twitter) encodes values, spirituality, and identity through language, with attention to clinical practice. A dataset of 2396 English-language tweets (2020–2025) was collected using dual criteria (pro-ana hashtags plus eating-disorder keywords). Only U.S.-based English tweets were included to maintain linguistic and cultural coherence with LIWC-22 norms and counseling frameworks developed in U.S. contexts. Tweets were separated into three corpora (full, hashtags, and tweet bodies) and analyzed using Linguistic Inquiry and Word Count 2022 (LIWC-22), supplemented with custom spirituality and pro-ana dictionaries, and keyword/keyness analysis against a 36-billion-token web reference corpus. Religious language appeared consistently higher in hashtags compared with tweets and Twitter norms. Tweets contained more authenticity and self-disclosure, while hashtags functioned as collective markers of identity and practice. Body and food terms were strongly elevated, and affiliation terms appeared comparatively suppressed. Keyness analysis identified distinctive items such as prayer fast, fasting prayer (Luke), OMAD fast, hunger hurt, and I’m punching, illustrating how sacred, cultural, and diet-related slogans were combined within pro-ana discourse. Pro-ana rhetoric may function as a sacralized identity frame that can provide existential meaning to disordered practices. These findings contribute to behavioral science by highlighting how online communities linguistically construct health-related identities and values. They also suggest that effective clinical interventions should address eating disorders not only at behavioral and cognitive levels but also at the level of values and spirituality. Full article

Background: Domain-specific large language models (LLMs) like Ortho GPT have potential advantages over general-purpose models in medical education, offering improved factual accuracy and contextual relevance. This study evaluates the performance of Ortho GPT against general LLMs and senior medical students on validated orthopedic examination questions. Methods: Six LLMs (Ortho GPT 4o, ChatGPT 4o, ChatGPT 3.5, Perplexity AI, DeepSeek-R1, and Llama 3.3-70B) were tested using multiple-choice items from final-year medical student orthopedic exams in German language. Each model answered identical questions under standardized zero-shot conditions; accuracy rates and item-level results were compared using McNemar’s test, Jaccard similarity, and point-biserial correlation with student difficulty ratings. Results: Ortho GPT achieved the highest accuracy across models. McNemar’s tests revealed the significant superiority of Ortho GPT over DeepSeek (p = 2.33 × 10⁻³⁵), Llama 3.3-70B (p = 1.11 × 10⁻³²), and Perplexity (p = 4.01 × 10⁻⁵). Differences between Ortho GPT and ChatGPT 4o were non-significant (p = 0.065), suggesting near-equivalent performance to the strongest general model. No LLM showed correlation with student item difficulty (|r| < 0.07, p > 0.05), indicating that models solved items independently of human-perceived difficulty. Jaccard indices suggested moderate overlap between Ortho GPT and ChatGPT 4o, but distinct response profiles compared with general LLMs. Conclusions: These findings illustrate the superiority of Ortho GPT in orthopedic exam accuracy and context relevance, attributed to its specialized training data. The domain-specific approach enables performance matching or exceeding top general LLMs in orthopedics, emphasizing the importance of domain specialization for reliable, curriculum-aligned support in medical education. Full article

Large language models (LLMs) are increasingly consulted by amateur gardeners who rely on them for diagnosing plant problems and selecting management strategies. This study evaluates whether such AI systems promote environmentally sustainable or chemically oriented practices. Fifteen real images of edible plants showing typical health issues were collected during 2024–2025, and four major models—ChatGPT 5.0, Gemini 2.5 Pro, Claude Sonnet 4.5, and Perplexity AI (standard version)—were queried in October 2025 using an identical user-style prompt. Each response was coded across four sustainability dimensions (ecological prevention, diagnostic reasoning, nutrient management, and chemical control) and aggregated into a composite Eco-Score (−1 to +1). Across cases, all models prioritized preventive and low-impact advice, emphasizing pruning, hygiene, compost, and organic sprays while recommending synthetic fungicides or pesticides only occasionally. The highest sustainability alignment was achieved by Perplexity AI (Eco-Score = 0.71) and Gemini 2.5 Pro (0.69), followed by ChatGPT 5.0 (0.57) and Claude Sonnet 4.5 (0.41). Although the models frequently converged in general reasoning, no case achieved full agreement in Eco-Score values across systems. These findings demonstrate that current LLMs generally reinforce sustainable reasoning but vary in interpretative reliability. While they can enhance ecological awareness and accessible plant care knowledge, their diagnostic uncertainty underscores the need for human oversight in AI-assisted amateur food production. Full article

The rapid integration of artificial intelligence (AI) into healthcare has reshaped medical education and clinical practice. While technological innovation is vital, soft skills are essential for preserving trust, ethical accountability, and humanistic care. This study explores the evolving role of soft skills in medical education in the AI era by examining definitional challenges, pedagogical strategies, and the integration of AI-related literacy. A narrative review methodology synthesized evidence across seven thematic domains, focusing on curricular integration, pedagogical strategies, and assessment approaches in medical education within AI-enabled learning environments. The findings demonstrated that soft skills improve patient adherence, satisfaction, safety, and trust; strengthen physicians’ professional identity, collaboration, and resilience; and enhance system-level outcomes, such as resilience, safety, and public trust. Experiential, reflective, and competency-based pedagogies remain the most effective instructional strategies, while AI-supported tools, including virtual patients, adaptive simulations, large language models (LLMs), and Retrieval-Augmented Generation systems (RAG), offer complementary benefits by enhancing doctor-patient communication, providing real-time personalized feedback, and strengthening clinical reasoning. Soft skills function as an interconnected and synergistic ecosystem that is reinforced by cognitive, affective, humanistic, and ethical mechanisms. Integrating these competencies with AI literacy promotes theoretical clarity, supports programmatic assessment, and fosters responsible innovation, ensuring that technological advancement enhances rather than diminishes the humanistic foundations of medicine. Full article

This article examines the development of Islamic education in Islamic primary schools in the Netherlands from 1988 to 2025. Since the early 1970s, the Muslim population in the Netherlands has grown significantly—initially due to labor migrants from Turkey and Morocco, and later as a result of asylum seekers from countries such as Somalia, Iraq, and Syria. The desire to practice and pass on their faith led to the establishment of mosques, educational centers, boarding schools, and eventually Islamic primary schools. In 1987, some of the founders of Islamic primary schools aspired to establish institutions similar to Madrasas, focusing heavily on Islamic instruction such as Qur’an recitation and Hadith studies. However, these ambitions could not be realized due to funding requirements. Others were inspired by the Imam Hatip schools in Turkey, which offer religious subjects such as Qur’an, Hadith, and Sira (the life of the Prophet Muhammad) alongside the national curriculum. Ultimately, a Dutch model of Islamic education emerged—partly influenced by the Imam Hatip concept, yet possessing a distinct identity. This study investigates how Islamic education has evolved in practice through semi-structured interviews, school observations, document analysis, and a national survey of religion teachers. The findings indicate that the desire to provide Islamic religious education was the primary motive behind the founding of the first Islamic primary school in 1988. Since then, this objective has remained central to school boards and parents alike. Religious education has progressed from fragmented teaching materials rooted in Arabic and Turkish contexts to coherent, Dutch-language curricula. By 2025, the teaching materials of Worden wie je bent (“Becoming Who You Are”) and the Amana have become dominant. Instruction encompasses not only religious knowledge and Qur’an recitation but also social-emotional development, citizenship, and sexuality education within an Islamic framework. Full article

This article presents the initial findings of the DIETALY project (Destination Italy in English Translation Over the Years), which explores the role of language and translation in shaping Italy’s international image as a tourist destination from the 1920s to the 1950s. Focusing on the national tourism agency ENIT, it analyses brochures, booklets and related materials produced for English-speaking markets during a period marked by Fascism, economic depression and post-war reconstruction. The study reveals that translation, localisation and adaptation were pivotal to ENIT’s communication strategy, facilitating cultural representation and adapting discourse in response to cultural, political and market changes. A case study of the Italy brochure series (1920–1937) illustrates the transition from literal translations to more adaptive, market-sensitive forms of linguistic mediation, reflecting growing awareness of audience expectations in Britain and the United States. Alongside this historical inquiry, the DIETALY project is developing a database that systematises the metadata of these dispersed materials. Although still in progress, this database is designed to support future qualitative and quantitative research, complementing the project’s demonstration of how ENIT’s multilingual discourse contributed to the construction of Italy’s identity as an attractive tourist destination for international audiences. Full article