A key challenge in Arabic NLP is the scarcity of dialectal data relative to Modern Standard Arabic (MSA), causing LLMs to overproduce MSA and struggle with dialectally accurate generation. From an interpretability perspective, this raises a fundamental question: where and how are dialectal features encoded within model internals, and can these representations be leveraged to improve dialect generation without fine-tuning? This study investigates two complementary inference-time approaches that serve simultaneously as interpretability probes and control mechanisms. First, we conduct a neuron-level analysis, identifying sparse neuron populations that encode dialect-specific features and showing that amplifying or suppressing these neurons can steer model outputs toward target dialects. Second, motivated by the entanglement of dialectal features at the single-neuron level, we apply a vector-steering approach that extracts dialect-specific activation directions and injects them during inference. Together, these methods illuminate the geometry of dialectal knowledge in Arabic LLMs and offer a principled, interpretability-grounded framework for dialect control without requiring dialect-specific fine-tuning.

There have been many attempts to model the morphological richness and complexity of Arabic, leading to numerous Part-of-Speech (POS) tagsets that differ in terms of (a) which morphological features they represent, (b) how they represent them, and (c) the degree of specification of said features. Tagset granularity plays an important role in determining how annotated data can be used and for what applications. Due to the diversity among existing tagsets, many annotated corpora for Arabic cannot be easily combined, which exacerbates the Arabic resource poverty situation. In this work, we propose an intermediate tagset designed to facilitate the conversion and unification of different tagsets used to annotate Arabic corpora. This new tagset acts as a bridge between different annotation schemes, simplifying the integration of annotated corpora and promoting collaboration across the projects using them.

We present the Camel Treebank (CAMELTB), a 188K word open-source dependency treebank of Modern Standard and Classical Arabic. CAMELTB 1.0 includes 13 sub-corpora comprising selections of texts from pre-Islamic poetry to social media online commentaries, and covering a range of genres from religious and philosophical texts to news, novels, and student essays. The texts are all publicly available (out of copyright, creative commons, or under open licenses). The texts were morphologically tokenized and syntactically parsed automatically, and then manually corrected by a team of trained annotators. The annotations follow the guidelines of the Columbia Arabic Treebank (CATiB) dependency representation. We discuss our annotation process and guideline extensions, and we present some initial observations on lexical and syntactic differences among the annotated sub-corpora. This corpus will be publicly available to support and encourage research on Arabic NLP in general and on new, previously unexplored genres that are of interest to a wider spectrum of researchers, from historical linguistics and digital humanities to computer-assisted language pedagogy.