DialSeg-Ar: Arabic Dialectal Speech for Topic Segmentation

DialSeg-Ar is a multi-genre benchmark for linear semantic segmentation in Arabic, with a focus on dialectal conversational and transcribed speech. The dataset is designed to evaluate how well models can split a sequence of utterances into contiguous topic-coherent segments. It was created by MBZUAI and last updated on the platform in April 2026.

Use Cases

• Benchmarking topic segmentation models based on dialectal Arabic conversations.
• Training models for linear semantic segmentation based on topic-coherent segments.
• Evaluating NLP performance on code-switched Gulf Arabic–English podcast data.
• Analyzing model robustness on underrepresented Arabic genres, including emotional speech.

Strengths

• Focuses on underrepresented Arabic genres, including dialectal telephone conversations and code-switched podcasts.
• Designed specifically for evaluating linear semantic segmentation of topic-coherent segments.
• Created by the research institution MBZUAI.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

MBZUAI

Freshness

Last updated 2026-04-15 14:22:40; freshness should be verified.