Linear Semantic Segmentation for Low-Resource Spoken Dialects

Semantic segmentation is a core component of discourse analysis, yet existing models are primarily developed and evaluated on high-resource written text, limiting their effectiveness on low-resource spoken varieties. In particular, dialectal Arabic exhibits informal syntax, code-switching, and weakly marked discourse structure that challenge standard segmentation approaches.

We introduce a new multi-genre benchmark for semantic segmentation in spoken Arabic, focusing on dialectal discourse. The benchmark covers casual telephone conversations, code-switched podcasts, expressive dialogue, poetry, and broadcast news, and is annotated by dialect-competent annotators. Using this benchmark, we show that segmentation models performing well on news degrade sharply on dialectal speech. We further propose a segmentation model that targets local semantic coherence and robustness to discourse discontinuities, consistently outperforming strong baselines on non-news genres. The benchmark and approach generalize to other low-resource spoken languages.