Dialect-Aware NLP

My PhD project: developing NLP solutions tailored to Kurdish dialectal variation, with the long-term goal of extending the methodology to other dialect-rich language families.

Dialect-Aware NLP is the working title for my PhD research at Saarland University. The core question is: how do we build NLP systems that do not merely tolerate dialectal variation but actively account for it? The project takes Kurdish as its primary testbed - a language family with extensive, well-documented dialectal diversity and very limited digital resources - and develops modular, transferable solutions intended to generalise to other dialect-rich, low-resource language situations.

Kurdish is an ideal research language for dialect-aware NLP for several reasons. It has a large number of speakers across multiple countries, yet is severely underrepresented in NLP benchmarks and tools. Its dialects - Kurmanji, Sorani, Zazaki, Gorani, and many sub-varieties within each - are sometimes mutually unintelligible, yet existing NLP resources conflate them into a single 'Kurdish' label. This makes Kurdish a challenging and scientifically interesting case study, and any improvements directly benefit a real community of speakers.

The methods developed for Kurdish are designed to be language-agnostic wherever possible. The long-term goal is a transferable methodology for dialect-aware NLP that can be applied to other dialect-rich, low-resource language families - particularly those where dialect variation is linguistically significant but computationally ignored.

This is an active PhD project. Results from several of the research threads have already been presented at international conferences (ICKL-6, ICKL-7). Further publications and updates will be added here as the work progresses. [TODO: add links to papers and talks as they are published]