paper-humanizer

name

paper-humanizer

description

Academic text polisher and humanizer for Chinese and English research writing. Removes AI-like patterns while preserving factual accuracy. Use when polishing academic text, reducing AI-generated patterns in papers/theses, improving naturalness of research writing while maintaining data integrity, or enhancing scholarly tone without fabricating facts. Supports both Chinese (中文) and English academic texts with configurable tone, field-specific terminology, and strict factuality controls.

Paper Humanizer

Polish and humanize academic writing to reduce AI-like patterns while preserving all factual content, data, and meaning.

Quick Start

Apply the system prompt from references/system_prompt.md and use the workflow below.

Output Format (Two-Stage)

Stage 1: Analysis & Ask

Produce the first 3 sections and then ask the user how to proceed. Do not output the full polished text in Stage 1.

1. 原文 AI 特征分析 (Original Text AI Pattern Analysis)
2. 核心优化策略 (Core Optimization Strategies)
3. 优化亮点说明 (Highlight Notes - optional)
4. 下一步行动 (Next Step): Ask the user to choose one of the following options:
   • 全文输出: Output the complete polished text at once.
   • 逐章修改: Output the polished text chapter by chapter, pausing for confirmation after each chapter.
   • 局部修改: Revise only the section(s) the user specifies.
   • Diff 模式: Show before/after snippets of key changes instead of the full text.

Stage 2: Execute on User Choice

Only after the user replies, deliver the polished content in their chosen form.

Core Constraints

Non-negotiable:

• Never fabricate facts, results, metrics, or references
• Never change numeric values, experimental settings, or data
• Preserve citation markers (e.g., [1], (Smith, 2023), \cite{...})
• Maintain all terminology and technical accuracy

Remove AI-isms:

• Chinese: "值得注意的是", "不难发现", "基于以上分析", "综上所述", "首先/其次/最后", "本文将"
• English: "It is worth noting that", "It can be seen that", "In summary", "Firstly, Secondly, Finally" (when overused)

See references/phrase_blacklist.md for complete list.

Parameters

Configure behavior using these parameters:

• language: auto (detect) | zh (Chinese) | en (English)
• tone: formal (default) | semi-formal | concise | persuasive
• strict_factuality: true (default) - preserve all data/facts
• keep_citations: true (default) - preserve citation markers
• blacklist_level: high (default) | medium | low

Workflow

1. Analyze: Identify 2-3 major AI pattern issues in the input text.
2. Strategy: Decide 3-5 targeted optimization strategies.
3. Highlight: (Optional) note 1-2 representative edits.
4. Ask: Present Stage 1 output and ask the user how they want to proceed.
5. Execute: Only after the user confirms, apply edits in their chosen mode (full text, chapter by chapter, partial, or diff).

Language Handling

• Auto-detect language from input text
• Maintain same language in output (unless user requests translation)
• Keep technical terms (GAN, WGAN-GP, NSL-KDD) as-is
• Preserve field-specific terminology

Advanced Features

For domain-specific terminology or special requirements, see:

• references/system_prompt.md - Complete editing principles and constraints
• references/phrase_blacklist.md - AI-pattern phrases to avoid
• references/user_template.md - Parameterized user prompt template

Script Usage (Optional)

Use the CLI wrapper for batch processing:

# Process a text file
python3 scripts/paper_humanizer.py --text-file input.txt --language auto

# Generate composed prompt for external use
python3 scripts/paper_humanizer.py --text-file input.txt --out prompt.md

The script combines system prompt and user template with specified parameters.