While neural machine translation (NMT) has achieved remarkable parity with human performance in general domains, handling multi-word expressions (MWEs) remains a persistent challenge, particularly in low-resource settings. This difficulty is twofold: models need not only decode the non-compositional meaning of MWEs but also navigate the stylistic choice between idiomatic ("MWE") and literal ("plain") outputs. In this study, we introduce a new 4-way parallel dataset for Greek and English, comprising of Greek and English MWE and plain sentences. We evaluate the style-transfer and style-switching capabilities of state-of-the-art Large Language Models (LLMs) across four tasks. Using traditional metrics, we demonstrate that LLMs often struggle with style-aware translation. Furthermore, we introduce StyleCOMET, a fine-tuned quality estimation metric to better evaluate stylistic nuances. Our results suggest that current limitations in MWE translation are a product of both model inability and metric-level insensitivity to idiomaticity.

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.

Multi-word expressions (MWEs) present unique challenges in natural language processing (NLP), particularly within the context of translation systems, due to their inherent scarcity, non-compositional nature, and other distinct lexical and morphosyntactic characteristics, issues that are exacerbated in low-resource settings.In this study, we elucidate and attempt to address these challenges by leveraging a substantial corpus of human-annotated Greek MWEs. To address the complexity of translating such phrases, we propose a novel method leveraging an available out-of-context lexicon.We assess the translation capabilities of current state-of-the-art systems on this task, employing both automated metrics and human evaluators.We find that by using our method when applicable, the performance of current systems can be significantly improved, however these models are still unable to produce translations comparable to those of a human speaker.