hongcha1101/de-aigc-ch

De-aigc-ch is a toolkit for making Chinese text written by AI sound more like natural human writing. When AI language models write in Chinese, they tend to fall into recognizable patterns: certain opening phrases that go nowhere, repeated sentence structures, an unusually even length across paragraphs, and connecting words that feel mechanical. This repository collects rules, examples, and scripts for identifying and rewriting those patterns. The approach works in three layers. The first layer looks at word choice, flagging phrases like "worth noting" and "not only...but also..." that appear far more often in AI output than in human writing. The second layer looks at sentence structure, checking for things like overuse of colons and semicolons, stacked parallel phrases, and long fixed clauses that AI models tend to repeat. The third layer looks at how the whole piece is organized, such as sentences that are all the same length or a conclusion that simply restates the introduction. The repository includes a five-step editing workflow: scan for problems, categorize them, rewrite with variation, run a five-dimension self-evaluation, and do a second check. It also includes a six-dimension scoring rubric covering directness, sentence rhythm, cautious language, logical flow, punctuation naturalness, and authenticity markers. A passing score is 50 or above out of 100. A Python script handles automated pattern scanning. It checks text against 11 pattern categories and rates the result on a severity scale: PASS, LIGHT, MODERATE, or HEAVY. Reference material in the repository includes academic citations supporting each detection rule, a blacklist of specific phrases to delete, a blacklist of sentence structures to avoid, and 20 before-and-after rewrite examples. The project is aimed specifically at Chinese writing and the README is written in Chinese. The license is MIT.