Dialectal Filtering: Synthesize Text Corpora for Kurdish Varieties by Enriching Datasets with Multi-Labels

Kurdish, a low-resource language with extensive dialectal variation, faces significant challenges in NLP due to noisy, mixed-dialect text corpora. This research introduces a dialect-aware text filtering framework to preprocess and enrich Kurdish datasets, synthesizing variety-specific sub-corpora for underrepresented dialects (see Figure 1). By leveraging lexicographic features, this approach generates over 100 sub-corpora from Northern Kurdish data, offering resources for dialect-specific NLP tasks but has currently limitations in precision.