Search Results (253)

Data scarcity is a significant barrier to developing high-performing AI models for medical text classification. To improve stroke prediction from electrocardiogram (ECG) interpretations reports where training data are scarce, we propose a novel data augmentation technique, Multi-Expert Perspective Augmentation. We use the LLM Phi-4 to rewrite original machine-generated ECG reports from the simulated perspectives of five different medical specialists. This prompt-based approach generates text with diverse clinical viewpoints while preserving the original meaning. We trained a BiomedBERT-based Gradient Boosting model to classify for stroke, comparing a baseline model trained only on the original data against a model trained with our augmented data. The model trained solely on original data showed poor performance (F1-score: 0.6698) with a severe precision–recall imbalance. In contrast, the augmented model achieved a significantly improved and balanced performance, with an F1-score of 0.8421, an accuracy of 0.8427, a precision of 0.8571, and a recall of 0.8276. Our method also outperformed other LLM-based augmentation techniques. Our findings demonstrate that rewriting text from multiple simulated expert perspectives is an effective strategy for data augmentation, enhancing the linguistic and contextual diversity of training data and leading to a more balanced and accurate classification model. Full article

In the rapidly evolving landscape of digital marketing and electronic commerce, short-form content—particularly on platforms like Twitter (now X)—has become pivotal for real-time branding, community engagement, and product promotion. The rise of Non-Fungible Tokens (NFTs) and Web3 ecosystems further underscores the need for domain-specific, engagement-oriented social media content. However, automating the generation of such content while balancing linguistic quality, semantic relevance, and audience engagement remains a substantial challenge. To address this, we propose RL-TweetGen, a socio-technical framework that integrates instruction-tuned large language models (LLMs) with reinforcement learning (RL) to generate concise, impactful, and engagement-optimized tweets. The framework incorporates a structured pipeline comprising domain-specific data curation, semantic classification, and intent-aware prompt engineering, and leverages Parameter-Efficient Fine-Tuning (PEFT) with LoRA for scalable model adaptation. We fine-tuned and evaluated three LLMs—LLaMA-3.1-8B, Mistral-7B Instruct, and DeepSeek 7B Chat—guided by a hybrid reward function that blends XGBoost-predicted engagement scores with expert-in-the-loop feedback. To enhance lexical diversity and contextual alignment, we implemented advanced decoding strategies, including Tailored Beam Search, Enhanced Top-p Sampling, and Contextual Temperature Scaling. A case study focused on NFT-related tweet generation demonstrated the practical effectiveness of RL-TweetGen. Experimental results showed that Mistral-7B achieved the highest lexical fluency (BLEU: 0.2285), LLaMA-3.1 exhibited superior semantic precision (BERT-F1: 0.8155), while DeepSeek 7B provided balanced performance. Overall, RL-TweetGen presents a scalable and adaptive solution for marketers, content strategists, and Web3 platforms seeking to automate and optimize social media engagement. The framework advances the role of generative AI in digital commerce by aligning content generation with platform dynamics, user preferences, and marketing goals. Full article

Background/Objectives: Children with Autism Spectrum Disorder (ASD) often display heightened sensitivity to simple auditory stimuli, but have difficulty discriminating and integrating multiple phonological features (segmental: consonants and vowels; suprasegmental: lexical tones) at the syllable level, which negatively impacts their communication. This study aims to investigate the neural basis of segmental, suprasegmental and combinatorial speech processing challenges in Mandarin-speaking children with ASD compared with typically developing (TD) peers. Methods: Thirty children with ASD and thirty TD peers will complete a multi-feature oddball paradigm to elicit auditory ERP during passive listening. Stimuli include syllables with single (e.g., vowel only), dual (e.g., vowel + tone), and triple (consonant + vowel + tone) phonological deviations. Neural responses will be analyzed using temporal principal component analysis (t-PCA) to isolate overlapping ERP components (early/late MMN), and representational similarity analysis (RSA) to assess group differences in neural representational structure across feature conditions. Expected Outcomes: We adopt a dual-framework approach to hypothesis generation. First, from a theory-driven perspective, we integrate three complementary models, Enhanced Perceptual Functioning (EPF), Weak Central Coherence (WCC), and the Neural Complexity Hypothesis (NCH), to account for auditory processing in ASD. Specifically, we hypothesize that ASD children will show enhanced or intact neural discriminatory responses to isolated segmental deviations (e.g., vowel), but attenuated or delayed responses to suprasegmental (e.g., tone) and multi-feature deviants, with the most severe disruptions occurring in complex, multi-feature conditions. Second, from an empirically grounded, data-driven perspective, we derive our central hypothesis directly from the mismatch negativity (MMN) literature, which suggests reduced MMN amplitudes (with the exception of vowel deviants) and prolonged latencies accompanied by a diminished left-hemisphere advantage across all speech feature types in ASD, with the most pronounced effects in complex, multi-feature conditions. Significance: By testing alternative hypotheses and predictions, this exploratory study will clarify the extent to which speech processing differences in ASD reflect cognitive biases (local vs. global, per EPF/WCC/NCH) versus speech-specific neurophysiological disruptions. Findings will advance our understanding of the sensory and integrative mechanisms underlying communication difficulties in ASD, particularly in tonal language contexts, and may inform the development of linguistically tailored interventions. Full article

Languages vary in their availability of productive endocentric bare-stem compounds (e.g., flower hat) and a range of complex predicates (separable verb-particles, double object datives, adjectival resultatives, put-locatives, make-causatives, and perceptual reports). To account for these cross-linguistic variations, two parameters have been proposed: the Compounding Parameter (TCP), which governs the formation of bare-stem compounds, separable verb-particles, and adjectival resultatives, and the Small Clause Parameter (SCP), which determines whether a verb can take a small clause complement. These parameters make testable predictions about children’s acquisition. If TCP and SCP are on the right track, we would expect correlations in the acquisition of structures governed by each parameter. This study examines these predictions by analyzing longitudinal corpora from 23 English-speaking children, assessing both the correlation between the acquisition of different structures and their acquisitional ordering. Our findings support both TCP and SCP, confirming that the acquisition of bare-stem compounds is closely associated with that of separable verb-particles, while the acquisition of (some) complex predicates is related. In addition, our results offer new insights into the potential triggers that children use to set each parameter. These findings contribute to our understanding of language variation and the role of parameter setting in first language acquisition. Full article

This study investigates how second language (L2) learners of Spanish identify the regional origin of native Spanish speakers and whether specific phonetic cues predict dialect identification accuracy across proficiency levels. Situated within a growing body of work on sociolinguistic competence, this research addresses the development of learners’ ability to use linguistic forms not only for communication but also for social interpretation. A dialect identification task was administered to 111 American English-speaking learners of Spanish and 19 native Spanish speakers. Participants heard sentence-length stimuli targeting regional phonetic features and selected the speaker’s country of origin. While L2 learners were able to identify regional dialects above chance, accuracy was low and significantly below that of native speakers. Higher-proficiency learners demonstrated improved identification, especially for speakers from Spain and Argentina, and relied more on salient phonetic cues (e.g., [θ], [ʃ]). No significant development was found for identification of Mexican or Puerto Rican varieties. Unlike native speakers, L2 learners did not show sensitivity to broader macrodialect groupings; instead, they frequently defaulted to high-exposure varieties (e.g., Spain, Mexico) regardless of the phonetic cues present. Findings suggest that sociophonetic perception in L2 Spanish develops gradually and unevenly, shaped by cue salience and exposure. Full article

The rapid growth of scientific literature necessitates effective information extraction. However, existing methods face significant challenges, particularly when applied to multi-domain documents and low-resource languages. For Kazakh and Russian, there is a notable lack of annotated corpora and dedicated tools for scientific information extraction. To address this gap, we introduce SciMDIX (Scientific Multi-Domain Information extraction), a novel multi-domain dataset of scientific documents in Russian and Kazakh, annotated with entities and relations. Our study includes a comprehensive evaluation of entity recognition performance, comparing state-of-the-art models, such as BERT, LLaMA, GLiNER, and spaCy across four diverse domains (IT, Linguistics, Medicine, and Psychology) in both languages. The findings highlight the promise of spaCy and GLiNER for practical deployment in under-resourced language settings. Furthermore, we propose a new zero-shot relation extraction model that leverages a multimodal representation by integrating sentence context, entity mentions, and textual definitions of relation classes. Our model can predict semantic relations between entities in new documents, even for a language encountered during training. This capability is especially valuable for low-resource language scenarios. Full article

Background: Older adults often struggle to comprehend speech in noisy environments, a challenge influenced by declines in both auditory processing and cognitive functions. This study aimed to investigate how differences in speech-in-noise perception among individual with clinically normal hearing thresholds (ranging from normal to mild hearing loss in older adults) are related to neural speech tracking and cognitive function, particularly working memory. Method: Specifically, we examined delta (1–4 Hz) and theta (4–8 Hz) EEG oscillations during speech recognition tasks to determine their association with cognitive performance in older adults. EEG data were collected from 23 young adults (20–35 years) and 23 older adults (65–80 years). Cognitive assessments were administered to older adults, and both groups completed an EEG task involving speech recognition in Speech-Shaped Noise (SSN) at individualized noise levels based on their Sentence Recognition Scores (SRS). Results: The results showed that age significantly impacted hit rates and reaction times in noisy speech recognition tasks. Theta-band neural tracking was notably stronger in older adults, while delta-band tracking showed no age-related difference. Pearson’s correlations indicated significant associations between age-related cognitive decline, reduced hearing sensitivity, and Mini-Mental State Examination (MMSE) scores. Regression analyses showed that theta-band neural tracking at specific SRS levels significantly predicted word list recognition in the higher SRT group, while constructional recall was strongly predicted in the lower SRT group. Conclusions: These findings suggest that older adults may rely on theta-band neural tracking as a compensatory mechanism. However, regression results alone were not sufficient to fully explain how working memory affects neural tracking, and additional cognitive and linguistic factors should be considered in future studies. Furthermore, cognitive assessments were administered only to older adults, which limits the ability to determine whether group differences are driven by age, hearing, or cognitive status—a major limitation that should be addressed in future research. Full article

Human language comprehension relies on predictive processing; however, the computational mechanisms underlying this phenomenon remain unclear. This study investigates these mechanisms using large language models (LLMs), specifically GPT-3.5-turbo and GPT-4. We conducted a comparison of LLM and human performance on a phrase-completion task under varying levels of contextual cues (high, medium, and low) as defined using human performance, thereby enabling direct AI–human comparisons. Our findings indicate that LLMs significantly outperform humans, particularly in medium- and low-context conditions. While success in medium-context scenarios reflects the efficient utilization of contextual information, performance in low-context situations—where LLMs achieved approximately 25% accuracy compared to just 1% for humans—suggests that the models harness deep linguistic structures beyond mere surface context. This discovery implies that LLMs may elucidate previously unknown aspects of language architecture. The ability of LLMs to exploit deep structural regularities and statistical patterns in medium- and low-predictability contexts offers a novel perspective on the computational architecture of the human language system. Full article

The understanding of hidden emotions—situations in which individuals deliberately express an emotion different from what they genuinely feel—is a key skill in theory of mind (ToM) development. This ability allows children to reason about discrepancies between internal emotional states and external expressions and is closely tied to linguistic development, particularly vocabulary related to mental states, which supports complex emotional reasoning. Children who are deaf or hard of hearing (DHH), especially those born to hearing non-signing families and raised in oral language environments, may face challenges in early language exposure. This can impact the development of social and emotional skills, including the ability to understand hidden emotions. This study compares the understanding of hidden emotions in hearing children (n = 59) and DHH children (n = 44) aged 7–12 years. All children were educated in spoken language environments; none of the DHH participants had native exposure to sign language. Participants completed a hidden emotions task involving illustrated stories where a character showed a certain emotion in front of two observers, only one of whom was aware of the character’s true emotional state. The task assessed children’s understanding of the character’s emotional state as well as their ability to reason about the impact of hiding emotions on the beliefs of the observers. The results showed that the hearing children outperformed their DHH peers in understanding hidden emotions. This difference was not attributed to hearing status per se but to language use. Specifically, children’s spontaneous use of cognitive verbs (e.g., think or know) in their explanations predicted task performance across the groups, emphasizing the role of mental state language in emotional reasoning. These findings underscore the importance of early and accessible language exposure in supporting the emotional and social cognitive development of DHH children. Full article

This empirical pilot study explored the use of wearable eye-tracking technology to gain objective insights into interpersonal interactions, particularly in healthcare provider training. Traditional methods of understanding these interactions rely on subjective observations, but wearable tech offers a more precise, multimodal approach. This multidisciplinary study integrated counseling perspectives on therapeutic alliance with an empirically motivated wearable framework informed by prior research in clinical psychology. The aims of the study were to describe the complex data that can be achieved with wearable technology and to test our primary hypothesis that the therapeutic alliance in clinical training interactions is associated with certain behaviors consistent with stronger interpersonal engagement. One key finding was that a single multimodal feature predicted discrepancies in client versus therapist working alliance ratings (b = −4.29, 95% CI [−8.12, −0.38]), suggesting clients may have perceived highly structured interactions as less personal than therapists did. Multimodal features were more strongly associated with therapist rated working alliance, whereas linguistic analysis better captured client rated working alliance. The preliminary findings support the utility of multimodal approaches to capture clinical interactions. This technology provides valuable context for developing actionable insights without burdening instructors or learners. Findings from this study will motivate data-driven methods for providing actionable feedback to clinical trainees. Full article

Customers heavily depend on reviews for product information. Fake reviews may influence the perception of product quality, making online reviews less effective. ChatGPT’s (GPT-3.5 and GPT-4) ability to generate human-like reviews and responses to inquiries across several disciplines has increased recently. This leads to an increase in the number of reviewers and applications using ChatGPT to create fake reviews. Consequently, the detection of fake reviews generated or rephrased by ChatGPT has become essential. This paper proposes a new approach that distinguishes ChatGPT-rephrased reviews, considered fake, from real ones, utilizing a balanced dataset to analyze the sentiment and linguistic patterns that characterize both reviews. The proposed model further leverages Explainable Artificial Intelligence (XAI) techniques, including Local Interpretable Model-agnostic Explanations (LIME) and Shapley Additive Explanations (SHAP) for deeper insights into the model’s predictions and the classification logic. The proposed model performs a pre-processing phase that includes part-of-speech (POS) tagging, word lemmatization, tokenization, and then fine-tuned Transformer-based Machine Learning (ML) model DistilBERT for predictions. The obtained experimental results indicate that the proposed fine-tuned DistilBERT, utilizing the constructed balanced dataset along with a pre-processing phase, outperforms other state-of-the-art methods for detecting ChatGPT-rephrased reviews, achieving an accuracy of 97.25% and F1-score of 97.56%. The use of LIME and SHAP techniques not only enhanced the model’s interpretability, but also offered valuable insights into the key factors that affect the differentiation of genuine reviews from ChatGPT-rephrased ones. According to XAI, ChatGPT’s writing style is polite, uses grammatical structure, lacks specific descriptions and information in reviews, uses fancy words, is impersonal, and has deficiencies in emotional expression. These findings emphasize the effectiveness and reliability of the proposed approach. Full article

Fake news generation and propagation occurs in large volumes, at high speed, in diverse formats, while also being short-lived to evade detection and counteraction. Despite its role as an enabler, Artificial Intelligence (AI) has been effective at fake news detection and prediction through diverse techniques of both supervised and unsupervised machine learning. In this article, we propose a novel Artificial Intelligence (AI) approach that addresses the underexplored attribution of information asymmetry in fake news detection. This approach demonstrates how fine-tuned language models and emotion embeddings can be used to detect information asymmetry in intent, emotional framing, and linguistic complexity between content creators and content consumers. The intensity and temperature of emotion, selection of words, and the structure and relationship between words contribute to detecting this asymmetry. An empirical evaluation conducted on five benchmark datasets demonstrates the generalizability and real-time detection capabilities of the proposed AI approach. Full article

Adversarial attacks in Natural Language Processing (NLP) present a critical challenge, particularly in sentiment analysis, where subtle input modifications can significantly alter model predictions. In search of more robust defenses against adversarial attacks on sentimental analysis, this research work introduces two novel defense mechanisms: the Lexicon-Based Random Substitute Model (LRSM) and the Word-Variant Voting Model (WVVM). LRSM employs randomized substitutions from a dataset-specific lexicon to generate diverse input variations, disrupting adversarial strategies by introducing unpredictability. Unlike traditional defenses requiring synonym dictionaries or precomputed semantic relationships, LRSM directly substitutes words with random lexicon alternatives, reducing overhead while maintaining robustness. Notably, LRSM not only neutralizes adversarial perturbations but occasionally surpasses the original accuracy by correcting inherent model misclassifications. Building on LRSM, WVVM integrates LRSM, Frequency-Guided Word Substitution (FGWS), and Synonym Random Substitution and Voting (RS&V) in an ensemble framework that adaptively combines their outputs. Logistic Regression (LR) emerged as the optimal ensemble configuration, leveraging its regularization parameters to balance the contributions of individual defenses. WVVM consistently outperformed standalone defenses, demonstrating superior restored accuracy and F1 scores across adversarial scenarios. The proposed defenses were evaluated on two well-known sentiment analysis benchmarks: the IMDB Sentiment Dataset and the Yelp Polarity Dataset. The IMDB dataset, comprising 50,000 labeled movie reviews, and the Yelp Polarity dataset, containing labeled business reviews, provided diverse linguistic challenges for assessing adversarial robustness. Both datasets were tested using 4000 adversarial examples generated by established attacks, including Probability Weighted Word Saliency, TextFooler, and BERT-based Adversarial Examples. WVVM and LRSM demonstrated superior performance in restoring accuracy and F1 scores across both datasets, with WVVM excelling through its ensemble learning framework. LRSM improved restored accuracy from 75.66% to 83.7% when compared to the second-best individual model, RS&V, while the Support Vector Classifier WVVM variation further improved restored accuracy to 93.17%. Logistic Regression WVVM achieved an F1 score of 86.26% compared to 76.80% for RS&V. These findings establish LRSM and WVVM as robust frameworks for defending against adversarial text attacks in sentiment analysis. Full article

This paper creates a global similarity network between city-level dialects of English in order to determine whether external factors like the amount of population contact or language contact influence dialect similarity. While previous computational work has focused on external influences that contribute to phonological or lexical similarity, this paper focuses on grammatical variation as operationalized in computational construction grammar. Social media data was used to create comparable English corpora from 256 cities across 13 countries. Each sample is represented using the type frequency of various constructions. These frequency representations are then used to calculate pairwise similarities between city-level dialects; a prediction-based evaluation shows that these similarity values are highly accurate. Linguistic similarity is then compared with four external factors: (i) the amount of air travel between cities, a proxy for population contact, (ii) the difference in the linguistic landscapes of each city, a proxy for language contact, (iii) the geographic distance between cities, and (iv) the presence of political boundaries separating cities. The results show that, while all these factors are significant, the best model relies on language contact and geographic distance. Full article

Trust in Traditional Thai Medicine (TTM) among Thai people has been reduced due to a lack of objective standards and the susceptibility of the general population to false information. The emergence of generative artificial intelligence (Gen AI) has significantly impacted various industries, including traditional medicine. However, previous Gen AI models have primarily focused on prescription generation based on Traditional Chinese Medicine (TCM), leaving TTM unexplored. To address this gap, we propose a novel fast-learning fine-tuned language model fortified with TTM knowledge. We utilized textual data from two TTM textbooks, Wat Ratcha-orasaram Ratchaworawihan (WRO), and Tamra Osot Phra Narai (NR), to fine-tune Unsloth’s Gemma-2 with 9 billion parameters. We developed two specialized TTM tasks: treatment prediction (TrP) and herbal recipe generation (HRG). The TrP and HRG models achieved precision, recall, and F1 scores of 26.54%, 28.14%, and 24.00%, and 32.51%, 24.42%, and 24.84%, respectively. Performance evaluation against TCM-based generative models showed comparable precision, recall, and F1 results with a smaller knowledge corpus. We further addressed the challenges of utilizing Thai, a low-resource and linguistically complex language. Unlike English or Chinese, Thai lacks explicit sentence boundary markers and employs an abugida writing system without spaces between words, complicating text segmentation and generation. These characteristics pose significant difficulties for machine understanding and limit model accuracy. Despite these obstacles, our work establishes a foundation for further development of AI-assisted TTM applications and highlights both the opportunities and challenges in applying language models to traditional medicine knowledge systems in Thai language contexts. Full article