JS/TS Library for Transliterating Between 50 Most Popular Languages (2450 combinations)

说明与用途

基于 ICU RBT 的 JS/TS 库实现了 50 种语言（2450 组合）的准确、上下文感知的转写，取代昂贵的外部 API 和 AI 生成规则，采用可复现的规则生成流程。

问题

Existing transliteration libraries mainly convert non‑Latin scripts to Latin and lack accurate, context‑aware transliteration from Latin scripts to many other languages, producing unreadable or incorrect results.

结果

The library achieves accurate, context‑aware transliteration across 50 languages, supporting 2450 combinations, without external API calls, and with a reproducible rule generation workflow.

工作流

• 1. Define transliteration pipeline: choose IPA or Standardized Latin as pivot based on language pair.
• 2. Generate ICU RBT rules using AI, then manually review for correctness.
• 3. Apply rules to a set of 60 examples (cities, people, companies) for each language pair.
• 4. Evaluate results (0‑100) and iterate: generate fixes, review, reapply until evaluation ≥100 or >95 after 20‑30 cycles.
• 5. Integrate rules into JS/TS library exposing a simple API for transliteration.
• 6. Publish library and documentation for developers to use locally.

证据

"We could also use Google Translate API, but there are some caveats: ... It is not free."

来源

"Human review of all AI suggestions is a must…"

来源

"The process I’m following to generate valid set of ICU rules…"

来源