LLM 輔助翻譯與 pollm 工具設計

December 16, 2024

沒什麼組織、一些倏忽即逝有趣或不有趣的想法

我們是否應該或不應該將 LLM 導入翻譯
- 為什麼需要 LLM 來翻譯 (接續 [[2024-12-14]])
  - 翻譯的本質是資訊搬運，僅僅是將一種語言的資訊轉換為另一種語言的資訊，LLM 相當擅長於理解文本的內容並且生成相對應的文本，此外加上使用者大部分對於文件的需求是不大的，只有在有情境需要時才會查看文件，因此 LLM 可以幫助使用者快速的獲取到所需的資訊節省搬運的力氣，人甚至不會去查看文件，在此 LLM 最大的幫助即是有效率的搬運資訊
- 人類可以做什麼
  - 從一個語系轉移至另一個語系，需要考慮到語言的語法、語義、語境、文化背景、缺乏的詞彙等因素，這些因素容易讓 LLM 產生誤導性或不一致的回答，我們仍然需要人為介入調整即定義細節的內容，讓產出更能夠符合目標語言的語言特性
pollm Design
- Why pollm
  - pollm is a tool that helps the translator to translate the Python documentation in the python-doc-zh-tw repository more efficiently and effectively. (理由如上項)
- Functionalities
  - pollm is a CLI tool based on Aider that helps you traverse the messages in the Po files and provide the suggestions for the translation.
  - since pollm is based on Aider, the translator can use the Aider-like CLI interface to interact with the Po files and get the suggestions for the translation, and provide the feedback to the Aider, and Aider will store the feedback and improve the suggestions in the future.
  - Customizable prompt templates for the translator to provide the suggestions for the translation.
- Future
  - Integrate the pollm with the pogrep, powrap, etc. to provide the full translation experience for the translator.
  - CI integration to provide the automatic translation suggestions for the translator and the reviewer once new messages are added to the Po files.