Scholarlink

This study conducts a comparative analysis of four large language models-GPT—4o, Gemini 1.5 Flash, Claude 3.5 Sonnet, and LLama3—70B-in their ability to translate Moroccan Darija into English. By calculating the similarity percentage between the original text and the translated text using the Levenshtein Distance, the accuracy of each model was assessed. The results revealed that the Gemini model achieved the highest mean accuracy of 90.99% with an SDV of 10.98, followed closely by GPT-4 with a mean accuracy of 90.66% and an SDV of 11.93. Claude showed a mean accuracy of 80.13% and an SDV of 27.26, while LLama3 had the lowest mean accuracy of 77.34% with an SDV of 29.01. The normal distribution plots highlighted the concentration and spread of accuracy for each model, indicating that Gemini and GPT-4o provided more consistent and reliable translations compared to Claude and LLama3.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative Analysis of Large Language Models in Moroccan Darija Translation

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yassine El Kaneb,
  • Mohcine Kodad

摘要

This study conducts a comparative analysis of four large language models-GPT—4o, Gemini 1.5 Flash, Claude 3.5 Sonnet, and LLama3—70B-in their ability to translate Moroccan Darija into English. By calculating the similarity percentage between the original text and the translated text using the Levenshtein Distance, the accuracy of each model was assessed. The results revealed that the Gemini model achieved the highest mean accuracy of 90.99% with an SDV of 10.98, followed closely by GPT-4 with a mean accuracy of 90.66% and an SDV of 11.93. Claude showed a mean accuracy of 80.13% and an SDV of 27.26, while LLama3 had the lowest mean accuracy of 77.34% with an SDV of 29.01. The normal distribution plots highlighted the concentration and spread of accuracy for each model, indicating that Gemini and GPT-4o provided more consistent and reliable translations compared to Claude and LLama3.