staruhub/ClaudeSkills / skills/Geek-skills-wechat-article-writer
wechat-article-writer
Verified专业微信公众号文章助手,支持四个独立且可组合模式:article 写正文;image-prompts 从文章生成版本化、provider-neutral 的图片提示词 manifest 与稳定占位符,但不调用生图;layout 把文章和 manifest 确定性转换为微信安全的内联 HTML;full-pipeline 串联正文、提示词/占位符和排版。用户提到公众号、微信文章、推文、配图提示词、图片 manifest、公众号排版、内联 HTML 或完整图文流水线时触发。支持企业官号、个人技术博客、活动回顾和产品评测。不用于:带引用的研究报告(用 deep-research)、PPT/组图(用 deck-studio)、英文文章、仅润色一小段。永不自动发布,永不把提示词或占位符声称为已生成图片。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 14.0 KB |
| assets/emoji-library.txt | markdown | 8.3 KB |
| assets/template-structure.md | markdown | 8.3 KB |
| evals/routing-evals.json | config | 2.8 KB |
| references/PIPELINE-CONTRACT.md | markdown | 4.3 KB |
| references/linguistic-background.md | markdown | 3.2 KB |
| references/quality-checklist.md | markdown | 9.9 KB |
| references/style-guide.md | markdown | 10.1 KB |
| references/title-formulas.md | markdown | 8.3 KB |
| references/writing-techniques.md | markdown | 7.2 KB |
| schemas/image-manifest.schema.json | config | 3.1 KB |
| scripts/render_wechat_layout.py | script | 6.5 KB |
| scripts/validate_image_manifest.py | script | 8.1 KB |
references/linguistic-background.md
View on GitHub# 反翻译腔的语言学根因
本文不影响执行——agent 不需要读这个文件。它给维护者和用户提供"为什么要做翻译腔检查"的理论支撑。
---
## 1. 形合 vs 意合:两套不兼容的句法操作系统
**英语**是形合(hypotaxis)——靠显性连接词(which/that/because/although)焊接逻辑,每个分句必须有主语,关系严格标记。
**汉语**是意合(parataxis)——靠语序和语义自然流动。"山高月小,水落石出"没有连词,逻辑自明。
翻译腔的核心症状——被动语气泛滥、"的"字堆叠、抽象名词做主语、连词堆砌——**几乎每一条都是英语结构穿透到汉语骨架里**。
> 奈达:"就汉语和英语而言,也许在语言学上最重要的一个区别,就是形合和意合的对比。"(1983《Translating Meaning》)
## 2. LLM 翻译腔的三层机制
**预训练语料污染**:中文互联网、学术期刊、新闻报道、翻译著作——这些大规模语料本身已被翻译腔渗透。LLM 训练时见到的"自然中文",并不完全自然。
**SFT 监督微调偏差**:Li et al. (ACL 2025) 发现超过 34% 的 SFT 训练数据本身带翻译腔。模型学到的"标准答案"就是翻译腔。更反直觉的是:**base model 的中文比微调后的更自然**——SFT 把它教坏了。即便是 GPT-4 直译,翻译腔率超过 40%。
**English pivot**:多语言模型内部常以英语做推理轴心。输出中文时底层语义结构是英语的,最后一步只换了词。
## 3. Accuracy-Naturalness Tradeoff
2025 年有理论论文证明:**翻译不存在一个同时最大化准确性和自然度的最优点,这两者的权衡是数学必然**。
示例:德语 "Meine Lehrerin ist nett"(我的老师[女性]人很好)翻译到英语:
- 自然版:My teacher is nice(丢失性别信息)
- 准确版:My female teacher is nice(保留信息但不自然)
这就是 skill 里"准确性优先(保守模式)"的理论基础——不同场景需要在这条曲线上选不同的点。
## 4. 为什么 LLM 翻译腔和人译翻译腔不同
Kong & Macken (2025) 的英汉新闻语料研究发现:LLM 已经自己学会了规避余光中九病的大部分("的"、"被"、"进行"不再是首要问题),但产出了新一代翻译腔:
- 英文思维的物理隐喻(击穿论证、拆解问题)
- 英文学术风格的形容词起手式
- 英文技术写作的抽象名词骨架
- 段落内句长均匀、缺乏长短交错的节奏
这就是本 skill 的五类标记(§2.3)针对的病情。
## 参考文献
- Li et al. (2025). Lost in Literalism: How Supervised Training Shapes Translationese in LLMs. ACL 2025. arXiv:2503.04369
- Guo et al. (2024). Do LLMs Have an English "Accent"? arXiv:2410.15956
- Kong & Macken (2025). Decoding Machine Translationese in English-Chinese News. arXiv:2506.22050
- 余光中 (1987). 怎样改进英式中文?——论中文的常态与变态. 明报月刊
- 余光中 (1996). 论的的不休
- Nida & Taber (1964/1982). The Theory and Practice of Translation
- 连淑能 (2001). 英汉对比研究. 高等教育出版社
- 王力. 中国语法理论 / 汉语语法纲要
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |