MiNLP

Why we included this project

MiNLP is a set of Chinese-language processing components that have been used across Xiaomi's own business, so the shipped pieces are tested on real traffic rather than just benchmark data. The most finished part is MiNLP-Tokenizer, a Chinese word segmentation library available as a Python package that was refined in production before being open-sourced in November 2020. The repo also carries duckling-fork-chinese, a JVM fork of Facebook's Duckling that turns natural-language mentions of numbers, dates, and times into structured objects, the kind of extraction you need for voice assistants, scheduling, or form-filling. Teams building Chinese NLP pipelines, especially ones parsing user input for entities, can use both components directly. Just be aware that the rest of the platform, POS tagging, NER, and dependency parsing, is still roadmap rather than shipped code, so the tokenizer and structured parser are the concrete deliverables today.