Linear Semantic Segmentation for Low-Resource Spoken Dialects

TL;DR

This paper introduces a new benchmark and segmentation model tailored for low-resource spoken dialects of Arabic, addressing challenges like informal syntax and code-switching.

Contribution

It provides a multi-genre benchmark for dialectal Arabic and proposes a segmentation model that improves robustness and local coherence in low-resource spoken dialects.

Findings

Segmentation models trained on MSA news genres perform poorly on dialectal speech.

The proposed model outperforms strong baselines on dialectal non-news genres.

The benchmark and approach are applicable to other low-resource spoken languages.

Abstract

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. In this paper, we introduce a new multi-genre benchmark (more than 1000 samples) for semantic segmentation in conversational Arabic, focusing on dialectal discourse. The benchmark covers transcribed casual telephone conversations, code-switched podcasts, broadcast news, and expressive dialogue from novels, and was annotated and validated by native Arabic annotators. Using this benchmark, we show that segmentation models performing well on MSA news genres degrade on dialectal transcribed speech. We further propose a segmentation model that targets local semantic coherence and robustness to discourse discontinuities, consistently outperforming strong baselines on dialectal non-news genres. The benchmark and approach generalize to other low-resource spoken languages.