Natural Language Processing
Phonetically-Informed Data Augmentation Improves Vietnamese Speech Translation by Addressing ASR Phonetic Errors
This paper addresses error propagation in cascaded speech translation (ST) systems, specifically for Vietnamese. The authors categorize Automatic Speech Recognition (ASR) errors, finding that phonetic confusions are the main cause of substitution errors, significantly degrading ST quality. They prop…
Hybrid Model Achieves State-of-the-Art Rumor Detection in Low-Resource Algerian Dialect
This paper introduces an end-to-end hybrid framework for rumor detection in the Algerian dialect, addressing challenges posed by informal language, code-switching, and limited annotated resources. The framework incorporates a novel dataset creation method combining real, synthetic, and existing corp…
Enhancing Dialogue Systems with Multilingual Emotional Validation: Datasets, Models, and LLM Evaluation
This work introduces M-EDESConv, a multilingual corpus, and M-TESC, a test set, to facilitate research in emotional validation for dialogue systems, addressing response identification, timing, and generation. A novel Multilingual Emotion-aware Gated Unit for Mutual Integration (MEGUMI) model demonst…
StanceNakba 2026: Transformer Models Excel in Conflict Discourse Stance Detection, but Cross-Topic Generalization Remains a Challenge
The StanceNakba 2026 shared task addressed actor-level and cross-topic stance detection in social media discourse related to the Palestinian-Israeli conflict. Utilizing a dataset of 2,606 posts, participating teams predominantly employed fine-tuned transformer models, achieving strong performance, p…
ArabiGEE: A Hierarchical Taxonomy for Structured Arabic Grammatical Error Explanation
ArabiGEE is a novel, hierarchical taxonomy for Arabic Grammatical Error Explanation (GEE), addressing the limitations of free-form text approaches. It systematically categorizes errors across orthographic, morphological, syntactic, and lexical dimensions, comprising 27 error types, 140 correction ty…