Dialect Normalization using Large Language Models and Morphological Rules

TL;DR

This paper presents a novel dialect normalization approach combining rule-based methods and large language models with few-shot prompting, applied to Greek dialects without needing parallel data, and evaluates its impact on downstream tasks.

Contribution

It introduces a hybrid normalization technique leveraging linguistically informed rules and LLMs, applicable to low-resource dialects without parallel corpora.

Findings

Previous analyses relied on superficial linguistic features.

Semantic information persists despite normalization.

Normalization improves downstream task performance.

Abstract

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.