Search Results (505)

The advancing fluency of large language models (LLMs) raises important questions about their ability to emulate complex human traits, including emotional expression and personality, across diverse linguistic and cultural contexts. This study investigates whether state-of-the-art LLMs can convincingly mimic emotional nuance in English and personality markers in Arabic, a critical under-resourced language with unique linguistic and cultural characteristics. We conduct two tasks across six models: Jais, Mistral, LLaMA, GPT-4o, Gemini, and DeepSeek. First, we evaluate whether machine classifiers can reliably distinguish between human-authored and AI-generated texts. Second, we assess the extent to which LLM-generated texts exhibit emotional or personality traits comparable to those of humans. Our results demonstrate that AI-generated texts are distinguishable from human-authored ones (F1 > 0.95), though classification performance deteriorates on paraphrased samples, indicating reliance on superficial stylistic cues. Emotion and personality classification experiments reveal significant generalization gaps: classifiers trained on human data perform poorly on AI-generated texts and vice versa, suggesting LLMs encode affective signals differently from humans. Importantly, augmenting training with AI-generated data enhances performance in the Arabic personality classification task, highlighting the potential of synthetic data to address challenges in under-resourced languages. Model-specific analyses show that GPT-4o and Gemini exhibit superior affective coherence, while LLaMA performs worse. Linguistic and psycholinguistic analyses reveal measurable divergences in tone, authenticity, and textual complexity between human and AI texts. These findings have significant implications for affective computing, authorship attribution, and responsible AI deployment, particularly within under-resourced language contexts where generative AI detection and alignment pose unique challenges. Full article

Urban ethnic enclaves are historically layered cultural landscapes whose public signage encodes community vitality, power relations, and cultural identity in ways that conventional land-use inventories cannot capture. Addressing the absence of scalable, longitudinal computational methods for monitoring such at-risk landscapes, this study develops a reproducible digital-mapping pipeline that operationalises linguistic-landscape analysis as a cultural-heritage monitoring tool for heritage-sensitive land-use planning. Taking Daerim-dong—Seoul’s primary Joseonjok (Korean Chinese) enclave—as a case, we process 38,640 Kakao Map Road View images across 17 annual cross-sections (2008–2024). The pipeline integrates four methodological components: a bounded Spatial Weighting Correction that adjusts for uneven historical coverage; zero-shot semantic sign-function classification using the Qwen2-7B-Instruct model; an exploratory Difference-in-Differences design probing the 2016–2017 THAAD geopolitical disruption; and a Boundary Permeability Ratio (BPR) for tracking enclave edge dynamics. The results document a three-phase trajectory—rapid bilingual expansion (2008–2016), stabilisation (2016–2019), and a COVID-period contraction (2019–2024)—and show that raw sign-count metrics can systematically overstate minority-language decline during economic crises once crisis-period signage is isolated. The BPR is presented as a candidate leading indicator of enclave contraction whose operational thresholds remain to be calibrated through multi-enclave validation. As a methodological proof-of-concept, the study illustrates how computational street-view analysis can support cultural-landscape governance, offering urban planners and heritage managers an actionable, transparent baseline for monitoring at-risk multicultural urban landscapes. Full article

This paper presents a hybrid model that combines linguistic data summarization techniques, algorithms for constructing probabilistic trees, and various generative artificial intelligence models for learning and generating linguistic summaries to aid decision-making. The proposal is validated using methodological triangulation techniques that demonstrate high consistency in the knowledge discovered. The proposal also compares different generative artificial intelligence models; among the evaluated models, Gemini achieved the best performance. However, it is evident that, in certain contexts and tasks, small language models can be effective, yielding results comparable to large language models (LLMs) at a lower computational cost. This study applies the algorithms in a case study analyzing oceanographic data from Northern Chile. In the validation scenario, the combination of linguistic data summarization methods with unsupervised learning techniques effectively models human tolerance for imprecision when processing complex data and generated linguistic summaries easily interpretable by human decision-makers with high levels of confidence. Studies of energy capacities in the studied region and their behavior in both winter and summer are presented. Full article

Serious games represent a promising pedagogical approach by combining therapeutic and educational goals through play. This study examined the potential of a culturally adapted version of Érase una vez, a narrative-based card game originally developed in Brazil, to foster and assess narrative competence in school-aged children. Twenty typically developing children (ages 8–9 and 11–12) participated in both individual written and group oral storytelling tasks. Usability perceptions were assessed through questionnaires and the System Usability Scale, while linguistic performance was analyzed using quantitative and qualitative approaches. Statistical methods and computational linguistic tools were applied to measure lexical, syntactic, and pragmatic features. Results indicated high engagement and satisfaction, with over 95% of participants reporting positive experiences. Narrative productions revealed significant age-related differences: older children generated longer and syntactically more complex stories, while younger groups produced simpler structures with lower lexical variety. Group narratives reflected classroom-level effects, with sixth graders achieving greater cohesion and creativity. Findings support the integration of Érase una vez as both a pedagogical and therapeutic tool. Its playful and flexible format promotes motivation, reduces performance anxiety, and elicits rich, naturalistic language samples. Despite limitations of sample size and design, results encourage further exploration of serious games in educational and clinical contexts. Full article

Recently, computational sentiment analysis has become an essential tool for detecting evaluative language in large text collections. However, its application to many low-resource language families and historical corpora remains largely unexplored. This paper reviews the evolution of sentiment analysis methods in the Turkic language family, with a particular focus on Chagatai, the classical predecessor of several modern Turkic languages. We outline the methods that have evolved since the advent of lexicon-based and rule-based approaches up to the present day with large language models, addressing longstanding problems in agglutinative morphology, data scarcity, orthographic instability, and multilingual lexical mixing. To examine the available options, we conducted a pilot experiment using multilingual models in a zero-shot setting on a curated Chagatai corpus. In the absence of ground-truth annotations, prediction stability was validated with ensemble consistency and inter-model agreement. The results show real promise but also distinct limitations when adapting traditional NLP technologies for historically remote, low-resource languages. Progress in the field will require cross-disciplinary work, systematic diachronic dataset deployment, and a nuanced adaptation of multilingual representation learning to handle linguistically rich, low-resource settings. Full article

Writing is both a cognitive–linguistic activity and an emotional academic experience. For students with Specific Learning Difficulties (SLDs), written production may involve tension, anticipated failure, reduced confidence, and frustration, especially when handwriting demands compete with idea generation, spelling control, and text organization. These emotional responses are educationally important because they may influence persistence, written productivity, revision behavior, and academic participation. This study examined emotional responses to handwritten and digitally supported computer-based writing among 60 secondary education students, including 40 students with formally diagnosed SLD and 20 students without learning difficulties. Each participant completed two writing tasks, one handwritten and one digitally supported computer-based, in a counterbalanced order. The computer-based condition functioned as a digitally supported writing environment, with spell-checking and grammar-checking enabled. After each condition, students completed the Writing Emotional Response Scale (WERS), a 12-item study-specific instrument assessing writing anxiety, writing confidence, writing frustration, and perceived ease. The WERS was developed as a preliminary measure of immediate, task-specific emotional responses. Students with SLD reported lower anxiety, lower frustration, higher confidence, and higher perceived ease during digitally supported writing. The study contributes to educational psychology by linking writing modality, SLD, and emotional accessibility. Full article

Ethnic hate speech is a form of intersectional violence that affects Indigenous groups in México. Despite the severity of this social phenomenon, there is a lack of computational resources, specifically labeled datasets, to enable the development of automated tools for its detection. This paper presents a methodology for calculating benchmarks of the EthnoHate dataset, designed for the classification of ethnic hate speech in México, and focuses on establishing baselines using machine learning algorithms. Moreover, different levels of linguistic representation are evaluated. The results reveal that hate in the dataset is predominantly explicit, with a strong lexical component, enabling models such as BoW, and TF-IDF to achieve an F1-macro competitive. However, semantic techniques like BERT with a contextual classifier achieves the best overall performance in F1-macro, demonstrating that there is a significant proportion of implicit cases that require deep semantic understanding. Analysis of algorithms reveals that ethnic hate speech in México manifests through a recurring vocabulary with complex combinations, and that, while lexical approaches are highly effective, contextual models are necessary to capture the subtlety and diversity of hate expressions. This work establishes the feasibility of the task, validates the quality of the EthnoHate dataset, and lays the groundwork for future research employing more complex architectures. Full article

Speech emotion recognition (SER) has attracted increasing attention in human–computer interaction, mental health monitoring, and multimedia retrieval. However, many existing multimodal SER systems exhibit a strong bias toward the text modality: because utterance-level labels are often easily inferred from lexical content, models tend to under-utilize non-verbal acoustic cues, which can lead to erroneous predictions when crucial emotional information is predominantly conveyed by prosodic and spectral features. To alleviate this imbalance, we propose an audio-sensitive SER framework that explicitly enhances the contribution of the audio modality through a two-step strategy. First, we construct an Audio Sensitive Network (ASN) by pretraining on the parallel Emotional Speech Dataset (ESD), in which identical linguistic content is spoken with different emotions. This setting allows the ASN to learn speech content-independent emotional representations that emphasize paralinguistic information. Second, we introduce a threshold fusion scheme that integrates the ASN with existing SER classifiers. Specifically, we employ the Tree-structured Parzen Estimator (TPE) to optimize label-wise decision thresholds, enabling flexible calibration of the joint prediction space across modalities and models. We conduct experiments on both the IEMOCAP and ESD corpora, comparing multiple baseline classifiers with and without the proposed audio-sensitive enhancement. The results show consistent, albeit moderate, improvements in emotion recognition performance (e.g., up to +11.7% absolute accuracy on angry for MMAN on IEMOCAP), particularly for emotions that rely heavily on prosodic and spectral cues, thereby demonstrating the effectiveness of the proposed framework in boosting audio sensitivity within multimodal SER systems. Full article

Objectives: The rise in the frequency of mental health concerns (depression and suicide) expressed on social media calls for reliable, explainable, and efficient computational methods for mental health surveillance. In this paper, we propose an interpretable framework for text-based detection of post- and community-level mental health risk on social media. Methods: The framework combines (i) Secretary Bird Optimization (SBO) for feature selection of informative linguistic and psychological features, (ii) a BERT (Bidirectional Encoder Representations from Transformers)—CNN (Convolutional Neural Network) model for post-level reasoning, and (iii) a Graph Neural Network (GraphSAGE) for community-level reasoning. The graph is estimated based on semantic similarity between posts and author relations, instead of social interactions (e.g., mentions, replies) between authors. We use SHAP and LIME for model interpretability, uncertainty, and calibration analysis to evaluate the trustworthiness of predictions. Results: The model delivers 93.1% accuracy, 0.91 F1-score, and 0.944 ROC-AUC on the eRisk and CLPsych datasets using a strict user-disjoint validation strategy. SBO lowers the number of features by about 38%, leading to better generalization. The graph-based model enables improved learning of post and user representations by capturing relational dependencies. Conclusions: Our approach offers an explainable and robust means of detecting mental health risk from text. Graph-based representations of semantic and authorship interactions enable community-level analyses, while interpretability and uncertainty estimation facilitate possible human-in-the-loop decision-making. This research does not explicitly consider a human-in-the-loop experiment. Full article

Large Language Models (LLMs) are increasingly explored for educational tasks such as grading, yet their alignment with human evaluation in real classrooms remains underexamined. In this study, we investigate the feasibility of using OpenAI GPT-4o to evaluate short-answer quizzes and project reports in an undergraduate Computational Linguistics course. We collect responses from approximately 50 students across five quizzes and receive project reports from 14 teams. LLM-generated scores are compared against human evaluations conducted independently by the course teaching assistants (TAs). Our results show that GPT-4o achieves strong correlation with human graders (up to 0.98) and exact score agreement in 55% of quiz cases. For project reports, it also shows strong overall alignment with human grading, while exhibiting some variability in scoring technical, open-ended responses. We release all code and sample data to support further research on LLMs in educational assessment. This work highlights both the potential and limitations of LLM-based grading systems and contributes to advancing automated grading in real-world academic settings. Full article

Imagined speech decoding has attracted growing interest in brain–computer interface (BCI) research, as it may enable language-related information to be recovered from non-overt neural activity. Current studies in this area are often treated as a single, unified research problem, despite substantial differences in decoding target, output constraints, and system output forms. This review examines recent imagined speech decoding research from a task-oriented perspective, with a focus on how different neural decoding tasks are defined, constrained by their output spaces, and expressed through different output pathways. The included studies are organized into four main task levels: semantic/intent, phoneme/syllable, word, and sentence/language decoding. They are further compared along two auxiliary dimensions: output-space property and output pathway, with particular attention to closed-set and open-vocabulary settings. The review shows that current studies span markedly different linguistic granularities and communication objectives, from low-bandwidth intent recognition to text or speech reconstruction. Finally, it concludes that imagined speech should not be treated as a single homogeneous decoding problem, and that a task-oriented framework provides a clearer basis for comparing heterogeneous studies and guiding future communication-oriented BCI research. Full article

Depression is a critical global public health challenge, and the demand for accurate automated detection methods has generated considerable research interest in Transformer-based models. Despite their substantial promise, a comprehensive investigation into their architectural efficacy, intrinsic mechanisms, and barriers to practical implementation remains lacking. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, this systematic review was conducted across six databases (IEEE Xplore, Elsevier, Springer, MDPI, PubMed, and arXiv). The final search was performed in October 2025, covering English-language empirical studies published between 2020 and 2025 that employed Transformer-based architectures for depression detection. Risk of bias and methodological quality were independently appraised by two authors using a six-dimension structured rubric, with disagreements resolved by a third author. Findings were narratively synthesized given substantial cross-study heterogeneity. This systematic review analyzed 46 studies and provided the first comprehensive, mechanism-level, architecturally stratified comparison of encoder-only, decoder-only, hybrid, and multimodal fusion paradigms, examining self-attention dynamics and transfer learning strategies. Since 2019, these frameworks have evolved from text-centric approaches to advanced multimodal systems. Encoder-only models show consistently strong results in high-throughput text-based screening, decoder-only models demonstrate stronger few-shot learning capabilities, hybrid architectures show the highest observed median performance in clinical interview settings across the reviewed studies, and multimodal fusion systems offer complementary advantages when heterogeneous signal integration is critical. These trends are task-contextualized and should not be interpreted as unconditional rankings, given heterogeneity in evaluation metrics and tasks across studies. Nonetheless, four principal challenges hinder clinical translation: overreliance on self-reported data, cross-linguistic bias, absence of uncertainty quantification, and substantial computational overhead. Future efforts should shift from incremental benchmark improvements toward clinical utility through standardized psychiatric validation, uncertainty-aware architectures, fairness-enforced training across diverse populations, and the integration of Transformer-based models with wearable and mobile health data to improve detection stability and reduce translational risk. This systematic review was registered on the Open Science Framework (OSF; DOI: 10.17605/OSF.IO/SYF9N). This research was funded by the Faculty of Information Science and Technology and by Universiti Kebangsaan Malaysia under Grant TAP-K014364. Full article

Adaptive cruise control (ACC) is a key component of advanced driver assistance systems, as it maintains a safe distance from preceding vehicles by regulating speed and spacing. However, vehicle dynamics, measurement uncertainty, and traffic variability pose significant challenges for conventional control methods. In this context, fuzzy logic (FL) has been widely explored for its ability to handle uncertainty and incorporate expert knowledge via linguistic rules. This article presents a systematic literature review on the application of FL in ACC systems, proposing a functional taxonomy based on the role of the fuzzy system within the control architecture. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology, 103 initial records were identified, of which 87 studies were included in the final analysis. Four main categories are defined: Direct Fuzzy Control/Learning-Based, Fuzzy Supervisory Decision Control, Fuzzy Adaptive Robust Control, and Fuzzy Model-Based Control. Results indicate that Direct Fuzzy Control/Learning-Based and Fuzzy Supervisory Decision Control dominate the literature, accounting for 35.6% and 28%, respectively, while Fuzzy Adaptive Robust Control and Fuzzy Model-Based Control represent 20.7% and 14.9%. Mamdani-type systems predominate (78.16%), followed by Takagi-Sugeno (T–S) systems (17.24%), while type-2 fuzzy systems remain limited (4.60%) due to higher computational complexity. Recent trends highlight growing interest in adaptive and robust FL-based strategies. Full article

Small language models (SLMs) are valued for their computational efficiency and suitability for edge deployment, but often underperform in localized linguistic and cultural contexts due to their limited parameter size. This study explores integrating real-time web search into Qolda, a 4B-parameter Kazakh-centric SLM, to close the performance gap with larger models. We assess two strategies: Naïve Retrieval-Augmented Generation (RAG), which uses raw benchmark questions as search queries, and Query-Refined RAG, which applies various refiner models, including a supervised distillation-tuned Qolda, to optimize queries. On the KazCulture and KazMMLU benchmarks, the Naïve RAG approach in reasoning-enabled mode achieved an average accuracy of 76.00%, improving on the Zero-Shot evaluation result of 60.37%, and, in this system-level comparison, exceeding the Zero-Shot accuracy of larger open-source models such as Qwen3-32B (64.72%) and Gemma-3-27b-it (60.24%), which were evaluated without retrieval augmentation. Query refinement improved the accuracy by about 3%, from 76.00% to 79.46%, but nearly doubled the computational cost. Inference time analysis shows that Naïve RAG adds approximately 1 s of retrieval latency per question. Query refiners introduce up to 4 s of additional overhead. However, the retrieved context reduces the time required for model reasoning in think mode. The most notable gains were observed in localized cultural knowledge, where web search integration correctly answered 32.9% of KazCulture questions that the Zero-Shot baseline failed on, while losing only 16.9% in return. These results suggest that retrieval-augmented SLMs can offer a cost-effective and competitive alternative to larger models for tasks in the domains of Kazakh language and Kazakh culture. Full article

Augmented Reality (AR) has been found to produce significant effects on individual learning outcomes but its impact on collaborative applications remains moderate. Existing AR frameworks emphasize individual instructional design, whereas frameworks for collaborative learning rarely engage with the spatial and device-mediated affordances of mobile AR. In response to this inadequacy in the literature, we introduce the Mobile Augmented-Reality Storytelling for Vocational Education and Training (MARS-VET) framework, a four-dimensional conceptual architecture that integrates Computer-Supported Collaborative Learning (CSCL) scripting principles with mobile AR affordances for collaborative English as a Foreign Language (EFL) writing in Vocational Education and Training (VET) settings. MARS-VET synthesizes theoretical perspectives across four dimensions: contextual anchoring, which embeds activities within authentic workplace scenarios; collaborative orchestration, which structures group interaction through macro- and micro-level scripts; competency cultivation, which sequences writing progression from model-based reproduction toward autonomous professional text production; and capacity building, which addresses the professional-development requirements of implementing educators. Content validity was established through expert panel evaluation involving international specialists (N = 11) who rated the framework against 36 items using a four-point relevance scale and provided additional qualitative feedback. The Scale-level Content Validity Index (S-CVI/Ave = 0.91) exceeded established thresholds, with all four dimensions achieving satisfactory item-level indices. Experts reached unanimous agreement on items addressing workplace scenario identification and co-located access to linguistic resources. Qualitative feedback led to terminology refinements and clarification of orchestration mechanisms. The framework offers VET institutions and educators a reference for the design and evaluation of collaborative AR experiences in an area where integrative frameworks have so far been lacking. Full article