Researchers have developed an AI tool that can detect online propaganda in Kinyarwanda, a Bantu language spoken by 350 million Africans. The dataset, called KinyaProp, provides examples of misinformation in Kinyarwanda for large language models to learn from and recognize.
Twenty NII papers were accepted at ACL 2026, including two that won the Best Theme Paper and Outstanding Paper awards. These achievements lay the groundwork for developing trustworthy AI and advancing research in explainability and transparent AI.
A new book by University of Auckland academics provides a detailed analysis of the Christchurch terror attack, including previously undiscovered online posts from the perpetrator. The authors used forensic linguistics to identify the terrorist and highlight gaps in the Royal Commission's inquiry.
Researchers created a tool that leverages large language models to build novel languages with unique grammatical structures and vocabularies. The ConlangCrafter tool generates diverse languages, which can be used for creative applications such as video games and movies, or aid researchers in studying poorly documented languages.
A large-scale study of an online patient portal shows that AI-generated responses can introduce errors and extraneous details, leading to increased editing time for physicians. Adapting AI to individual physician communication styles can improve accuracy by 33% and reduce editing by 26%.
A University of Washington-led research team created GovScape, an efficient search system for PDFs from the End of Term Web Archive. The system can currently search 10 million PDFs and uses highly efficient AI models to process documents, costing less than $1.5k.
Falk Huettig's new book delves into the effects of reading on memory, attention, language processing, and reasoning, as well as its impact on face recognition. The author argues that literacy continues to develop beyond decoding skills, training higher-level cognitive functions.
A study by the University of Sharjah found that fear-based messaging in CDC anti-smoking campaigns can be highly effective when combined with credible information and practical guidance. The research identified five core strategies: fear-induced narratives, health-hazard emphasis, statistics, expert endorsement, and urgency.
A new Drexel University study finds that most users see AI chatbots as supplements to human therapy, not substitutes. Users express concerns about emotional dependence, misinformation, and overreliance on the technology.
A new study reveals that human language is systematically biased toward safety, overturning a cornerstone assumption in psychology and linguistics. The researchers identified three independent dimensions: power, danger, and structure, which explain over 90% of the variance in meaning.
Researchers found that people follow a two-stage process, sampling behaviors at first and committing once enough evidence accumulates. The Tolerance Principle predicts when people have seen enough regularity to treat a pattern as a rule despite some exceptions.
Researchers found that men with a 'country' identity showed distinct vowel patterns in their pronunciations, different from those living in neighboring areas. Country identity was more robustly linked to language patterns than travel patterns.
New research uses transcripts from the Enron scandal to analyze linguistic moves made by energy traders to manipulate trust and encourage collusion. The study identified a 'playbook' of trust-related moves, including 'bond', 'build', 'confide', 'probe', and 'repair', which can inform practical interventions in combating fraud.
Researchers found that as systems grow, new functions are added at a slower rate, requiring infrastructure and existing functions to expand. Cities show an exception, where function diversity grows rapidly in smaller cities but slows down in larger ones.
The UJI's GRAPE group proposes a paradigm shift in language teaching with a computer tool that analyzes verbal and visual elements. The software, GRAPE-MARS, enables fast labeling of video files and incorporates AI tools for efficient analysis.
A new study by Doshisha University found that teachers' deliberate actions at the start of a lesson, such as responding to the school bell, can create social tightness or looseness. Teachers who enforced strict rules and verbal prompts maintained tightness, while those with more relaxed approaches created loose occasions for students.
The Centre for Computational Social Science and Humanities (CSSH) at NUS combines computational methods with social science and humanities to better understand complex social phenomena and develop solutions. CSSH aims to generate research that improves lives, strengthens institutions, and shapes more inclusive societies.
Researchers are working on a £3.5 million UK-Japan project to develop human-centred artificial intelligence for sign language translation, focusing on natural conversational Deaf data rather than interpreted signing.
A study by Iowa State University researchers found that anthropomorphizing language, such as using mental verbs like 'think' or 'know', is less common in news writing than expected. When used, these words often describe what AI requires to function, rather than attributing human-like qualities to the system. The research team identifie...
The verb kikoyu, originally meaning 'to hear', came to represent approximate understanding, then general understanding, and finally conviction. This process reflects cognitive simulation and mental representation transformation.
The German Research Foundation extends TRR 318 'Constructing Explainability' for a further three and a half years, focusing on context within explanations to develop more comprehensible AI systems. The collaboration involves over 60 researchers across seven disciplines, aiming to create social forms of explainable AI.
Researchers aim to understand and overcome the limitations of current large language models, which make mistakes despite extensive training. The Emmy Noether Research Group will focus on developing new architectures with predictable capabilities.
Large language models systematically rate speakers of German dialects less favorably than those using Standard German, associating dialects with negative traits. The bias grows when dialects are explicitly mentioned, and larger models display even stronger biases.
A new study by New York University researchers found that the brain processes multiple words at once in a hierarchical manner, similar to a subway system. The brain coordinates different levels of processing, such as sounds, words, and meanings, without interference, allowing for efficient interpretation of speech.
The DFG is establishing four new Research Units to address pressing issues in linguistics, security, precision medicine and environmental history. The Research Units will focus on topics like bilingual flexibility, security in catastrophic times, genome-based precision medicine and medieval coastal landscape development.
The new book examines language attrition in non-pathological settings, considering social, lexical, and grammatical constraints. Researchers aim to develop a multifaceted approach to combating language loss, improving pedagogy and promoting bilingualism.
A recent study found dramatic differences in how large language models classify hate speech, with significant variations for specific demographic groups. The analysis of seven models revealed that content moderation systems have inconsistent detection accuracy, which can erode trust and create perceptions of bias.
A new project aims to increase HPV vaccine uptake by harnessing AI to tailor health information to parents' specific concerns and communication styles. The AI-driven chatbot will simulate human conversation, addressing misconceptions and fears associated with the vaccine.
A new study argues that adolescents and young adults are the main drivers of language change, rather than young children. The research finds little evidence that early learning mistakes spread through communities and highlights the role of everyday language use among older speakers.
A study by University of Birmingham researchers found that corporate net-zero pledges often lack the substance needed to drive real change. Many companies use vague language and focus on direct emissions, omitting substantial indirect emissions generated by their products.
Researchers Marie Teich and Wilmer Leal develop a formal framework to analyze metaphors, confirming they are enduring linguistic and cognitive structures. The study reveals two significant metaphorical processes: mappings from concrete to abstract topics and the emergence of new mappings between domains.
A University of Washington study found that biased AI chatbots can influence users' political views, with those who interacted with the most biased systems shifting their views more significantly. The researchers suggest education about AI may help mitigate this effect.
A study analyzing over 7.9 million US Congressional speeches from 1873 to 2010 found that speakers of all ages contribute to linguistic changes, with older speakers adopting new words and meanings over time. The researchers used AI models to track word usage patterns and found shifts in meaning across decades.
A team of computer scientists created 2,300 original sudoku puzzles and asked AI tools like OpenAI's ChatGPT to solve them. The results showed that while some AI models could solve easy sudokus, most struggled to provide accurate explanations, raising questions about the trustworthiness of AI-generated information.
A new study by Waseda University researchers suggests that learners need to automatize word meanings in context to achieve fluent speech. By assessing automatized vocabulary knowledge (AVK) and declarative vocabulary knowledge (DVK), the team found AVK to be a stronger predictor of L2 speaking fluency.
A new study by psychologists, linguists, and policy experts reveals that policymakers and drinkers have different ways of framing alcohol consumption. The findings suggest that policies must reflect real-life behaviours and motivations to make a meaningful difference.
Researchers found that children's active exploration of their surroundings and use of multiple senses help them learn language quickly. This approach differs from AI systems' passive processing of data, highlighting the need for rethinking AI design.
Otago researchers have recovered ancient genomes from Papua New Guinea, shedding light on the genetic diversity of pre-colonial coastal communities. The study reveals that cultural groups did not intermarry for some time, suggesting a complex tapestry of different cultures and peoples.
A new study combines ancient DNA with dietary evidence and linguistics to paint a vivid picture of pre-colonial coastal communities in Papua New Guinea. The research highlights the influence of migrations and interactions on genetic makeup, supporting previous studies on the settlement of remote islands.
The Konstanz Cluster of Excellence 'The Politics of Inequality' will continue to investigate the political causes and consequences of growing inequality in society. Researchers aim to prevent self-reinforcing processes and reduce structural inequalities.
A Dartmouth team's AI model recognized Navajo with near-perfect accuracy, identifying related languages such as Apache and Native Alaskan languages. The study suggests that this technology could be a bridge to including smaller languages in online translation services.
Researchers at PolyU developed an AI-driven assessment system, LMLPA, to quantify LLM personality traits. The system analyzes linguistic patterns and style in LLM outputs to evaluate their personalities, with applications in education, manufacturing, business, and sustainable development.
A new study by Doshisha University demonstrates the feasibility and reliability of fully automated speaking tests for English language learners, enabling more frequent evaluation and larger-scale studies. The approach uses AI-powered speech recognition and computational metrics to assess language proficiency.
Researchers found bonobos create complex call combinations resembling human language, suggesting compositionality is evolutionary ancient. This challenges long-held assumptions and implies a shared trait among humans and their ancestors.
A new study found that AI-generated Arabic health messages are mostly positive but lack clarity. Human-created messages were rated as higher in terms of message quality and clarity. The study suggests that AI can augment public communication campaigns, particularly in multilingual populations.
A new genomic analysis suggests that human language capacity emerged at least 135,000 years ago, with regional groups spreading across the globe around 100,000 years after that. The study, which examined 15 genetic studies, provides increasingly converging evidence about the timing of geographic splits among early human populations.
A Cambridge study highlights the need for India to recognize its own sign language, Indian Sign Language (ISL), as a constitutional right. The study also calls for increased investment in education and training programs for deaf and hard-of-hearing students, including more schools and higher education institutions.
HKU achieves exceptional results in the 2025 QS World University Rankings by Subject, ranking 4th globally in Dentistry and Education. The university leads in 33 out of 50 subjects locally, accounting for two-thirds of the ranked subjects.
A recent study published in the Journal of Sociolinguistics suggests that people's judgments of accent are often culturally influenced rather than based on actual auditory differences. Participants rated speakers from different regions of Ohio, but surprisingly, those who believed certain regions had stronger accents did not necessaril...
The study compared the performance of seven AI models with that of 400 humans in comprehension tasks, revealing a significant difference in accuracy. Human accuracy reached 89%, while AI models struggled to achieve more than 70% correct answers.
Researchers found greater phonetic markers among Pacific Islander teens in a more diverse school, indicating prominent ethnic boundaries. The study suggests that language shifts serve as subtle signals to establish identity and connect with cultural communities.
Researchers found that humans still outperform ASR systems in noisy environments, but Whisper large-v3 matched human performance in all tested conditions except naturalistic pub noise. The system's ability to process acoustic properties and map them to the intended message was impressive.
Researchers found that culture plays a significant role in shaping interactions during casual conversations, fostering curiosity about diversity and group cohesion through humor. The study highlights the importance of cultural identity in bridging gaps between people from different backgrounds.
Professor Charles Schencking, a leading historian of modern Japan, has been elected as a Corresponding Fellow of the Australian Academy of Humanities. His expertise in digital humanities, Islamic Studies, Asian Studies, and linguistics will foster collaboration with international institutions.
A new study provides genomic evidence of early migration from New Guinea into the Wallacea archipelago, addressing major gaps in human genetic history. The research shows that Papuan ancestry is widespread across Wallacea, pointing to historical migrations from New Guinea.
The LiITA project creates an interoperable Knowledge Base for Italian language resources, enabling dialogue between resources and artificial intelligence model development. The project aims to foster innovation in linguistics and counteract declining interest in humanities studies.
A study published in Psychological Science found that 2-year-olds rely on grammatical information to understand the meaning of new words. The researchers used focus signals in sentences to manipulate children's responses, showing that young kids can absorb grammatical cues and leverage them to learn new words.
A new study found that speech sounds have texture and shape to them, with R sounds associated with roughness and L sounds with smoothness. The research showed strong tendencies for matching trilled R sounds with jagged lines and L sounds with straight lines across cultures.
Researchers at Tufts University have discovered that AI's poor performance in conversational skills stems from its limited exposure to transcribed spoken language. The AI is trained on vast datasets of written content but lacks the nuance of human conversation, making it struggle with timing and turn-taking.
A new study found that foreign accents can mitigate judgment for grammatical errors, except in cases where listeners are highly conscientious. The research suggests that English speakers are generally tolerant of grammatical errors when the speaker has a foreign accent.