MoonPinyin

MoonPinyin is a MoonBit library for adaptive polyphone-to-pinyin conversion.

Public repository: tyytyy68/moonpinyin-osc2026

Competition repos:

• GitHub: liuminyuan/MoonPinyin
• GitLink: lmylmy78/moonpinyin

It is designed as a small but extensible text-processing core for:

• Chinese search normalization
• TTS and speech pipeline preprocessing
• educational annotation and accessibility tooling
• pinyin initials, tone-marked pinyin, and rule tracing

What it does

MoonPinyin combines two layers:

1. Phrase-level matching
2. Single-character fallback readings

That keeps common polyphones like 重, 行, 长, 乐, 还, 处, and 朝 readable in practical text.

The current implementation is intentionally compact and readable so it can serve as a competition submission, a demo package, and a base for later extension.

Outputs

• pinyin_with_tone(text) returns tone-number pinyin
• pinyin(text) returns plain pinyin without tone marks
• initials(text) returns pinyin initials
• trace(text) returns a rule trace for debugging and presentation

Example

let s = "重来一次，重要的是银行行为要看上下文。"
println(pinyin_with_tone(s))
println(pinyin(s))
println(initials(s))
println(trace(s))

Project layout

• moonpinyin.mbt core library
• moonpinyin_test.mbt black-box tests
• cmd/main/main.mbt runnable demo
• LICENSE Apache-2.0
• SOURCE.md source provenance note
• docs/proposal.md one-page proposal draft

Competition notes

This repository is written to fit the MoonBit Open Source Ecosystem Competition:

• MoonBit is the primary implementation language
• the repo is open and testable
• the code is structured as a reusable package rather than a one-off script
• the README is kept concise and reviewer-friendly
• the repository keeps a single maintainer identity and avoids mixed authorship
• the source note explains the hand-curated rule set and intended open-data extension path

Source notes

The phrase rules and single-character defaults are hand-curated for this demo submission. They are not copied from any proprietary dataset. The structure is designed so a larger open dictionary or a trained model can be added later without changing the public API.