DiaLLM: An Investigation into the Robustness-Generation Gap in English Dialect Adaptation
For NLP researchers working on dialectal adaptation, this paper reveals a fundamental gap between robustness and generation, highlighting the need for richer reward designs and dialectal resources.
Large language models understand dialectal English but fail to generate it. DiaLLM shows that robustness and generation are dissociated: alignment reshapes generation in ways benchmarks miss, and no single alignment method dominates, with a reward-quality gap confirmed by human evaluation.
Large language models increasingly \emph{understand} dialectal English, yet still \emph{produce} only standard, US-leaning English, leaving dialectal generation, the harder half of the problem, largely unaddressed. We introduce \textbf{DiaLLM}, which continually pretrains three open-weight language model families on the International Corpus of English and applies implicit and explicit post-training paradigms, each combined with three model alignment strategies, giving the first controlled comparison of these components across Australian, Indian, and Northern British English. Our results reveal that dialectal robustness and generation are \emph{dissociated}: benchmarks are shaped by continual pretraining and SFT, while alignment visibly reshapes generation in ways benchmarks do not capture. Explicit variety-targeted adaptation produces output reliably recognised as dialectal and preferred over broad alignment, yet the method that most aggressively optimises the dialectal reward is not preferred by human evaluators. Independent linguistic analysis corroborates this reward-quality gap, most clearly on two of the three families. No single alignment method dominates, and closing the gap will require richer reward designs and continued investment in dialectal resources. We release all code, checkpoints, and preference datasets.