Tag
This paper presents a 17k-sentence corpus with annotations for argumentative passages across three German political arenas during COVID-19, and a pilot study on automatically identifying such passages, finding that boundaries are hard to pin down and models exhibit confirmation bias.
This paper studies authorship verification on transcribed German-language videos, comparing traditional n-gram methods with transformer-based approaches across three self-compiled corpora. Traditional character and token n-gram methods outperformed modern transformers, achieving up to 88% accuracy and 90% AUC.
A two-dialect finite-state morphological analyzer for the Dungan language is presented, with a multi-genre evaluation measuring inflection, ambiguity, and lexical coverage.
This paper investigates whether LLMs can accurately predict item difficulty levels in large-scale reading and writing tests, finding that GPT-4.1 achieves moderate accuracy but is outperformed by ConvBERT, and that LLMs tend to underestimate difficulty for hard items.
This paper empirically evaluates how well LLMs align with human emotional perception of news framing, using a YouGov survey of 3,011 UK adults and seven LLMs assessing sympathy in headlines. It finds that alignment varies across models and demographic subgroups, highlighting the importance of differential alignment for AI development.
This Data Descriptor presents a large-scale corpus of transcribed religious radio broadcasts captured from live webstreams over one month in July 2025, comprising over 700,000 recordings and 60 million transcript lines, annotated using LLMs for program format and topic. It enables descriptive study of religious broadcasting and analysis of social/political issues in religious media.
This paper assesses the adversarial robustness of five Arabic language models under character, word, and sentence-level attacks, showing that diacritic insertion can reduce accuracy by 92% and adversarial training improves resilience but has limitations.
Proposes a deep neural model combining multi-layer temporal convolutional networks with label-wise attention for medical coding, achieving significant improvements in F1 and recall scores over previous state-of-the-art.
This paper evaluates LLMs' ability to recognize unspoken beliefs (implicatures) and their updates through implicature cancellation, introducing the expert-annotated ImplicatureX dataset. Results show LLMs lag behind humans, especially in natural scenarios.
This study evaluates four proprietary LLMs (GPT-4o, GPT-5.2, Claude Sonnet 4.5, DeepSeek) for specialized terminology translation from English to French across two domains, comparing prompting strategies. Results show Claude Sonnet 4.5 performs best, but LLMs cannot yet replace specialized corpora.
This paper surveys mechanisms for calculating word embeddings, investigates popular toolkits and embedding matrices, and experiments with selected implementations to understand their properties.
This paper proposes a diagnosis and optimization framework for cross-domain and cross-lingual generalization in offensive language detection, decomposing performance degradation into dataset and language effects, and quantifying the trade-off between multilingual capability and source-task performance.
This paper presents the IHLC submission to the LT-EDI 2026 shared task, using LoRA fine-tuning for gender-neutral rewriting (Rank 3) and activation steering for counter-narrative generation (Rank 6), highlighting both promise and limitations.
This paper presents a formative study using a two-stage LLM pipeline (Gemini 2.5 Pro and Flash) to detect internal documentation inconsistencies in electronic health records, analyzing 3,000 discharge summaries and proposing a graded ontology for categorizing inconsistencies.
This study analyzes how different types of reviewer guidelines (official conference guidelines vs. reviewer-imitating ones) affect LLM-based automated peer review, finding that official guidelines produce more human-consistent results while strict rubric-style scoring degrades performance.
This paper introduces GAND, a benchmarking resource of gender-ambiguous natural English sentences for analyzing gender bias in machine translation, and presents an interpretability analysis using contrastive translations to reveal source words influencing gender assignment.
Cleanlist AI is a product for natural-language prospecting that helps find, enrich, and sync leads.
This paper investigates the use of large language models (LLMs) and supervised classifiers for depression detection from social media text, proposing a prompt-based embedding method that enhances interpretability. Experiments on multiple datasets show that zero-shot LLMs perform well for binary classification but struggle with fine-grained severity, while supervised models on LLM summary embeddings achieve more consistent performance across multi-class and ordinal tasks.
This paper presents a method for multilingual grievance labeling using contextual models to improve upon word-level dictionaries, addressing circular evaluation and achieving better performance by reading surrounding context.
This paper explores whether valence features can reflect morality in natural language by analyzing human annotations of moral scenarios, finding significant correlations and achieving a Matthew's correlation coefficient of 0.764 for binary morality classification.