Dialect Normalization using Large Language Models and Morphological Rules
This addresses the problem of low-resource dialect processing for NLP systems, though it is incremental as it applies existing techniques to a specific language context.
The authors tackled dialect-to-standard normalization for Greek dialects by combining rule-based morphological transformations with LLM few-shot prompting, achieving results evaluated by human annotators on a proverb dataset and revealing that previous analyses relied on superficial linguistic artifacts.
Natural language understanding systems struggle with low-resource languages, including many dialects of high-resource ones. Dialect-to-standard normalization attempts to tackle this issue by transforming dialectal text so that it can be used by standard-language tools downstream. In this study, we tackle this task by introducing a new normalization method that combines rule-based linguistically informed transformations and large language models (LLMs) with targeted few-shot prompting, without requiring any parallel data. We implement our method for Greek dialects and apply it on a dataset of regional proverbs, evaluating the outputs using human annotators. We then use this dataset to conduct downstream experiments, finding that previous results regarding these proverbs relied solely on superficial linguistic information, including orthographic artifacts, while new observations can still be made through the remaining semantics.