Native English Tool

Stop writing like a robot. Start writing like a native speaker.

A tool that detects non-native English patterns — AI-overused words, overly formal phrasing, redundant expressions — and suggests natural American English replacements.

This is not an AI detection bypass tool. It's a genuine writing quality tool for anyone who wants their English to sound more natural.

The Problem

Non-native English speakers (and AI-generated text) share the same tells:

What they write	What a native would write
We must delve into the issue	We need to dig into the issue
Utilize our resources	Use our resources
In order to improve	To improve
It is worth noting that prices rose	Prices rose
A multifaceted and robust approach	A complex and solid approach
Due to the fact that it rained	Because it rained
This facilitates seamless integration	This helps with smooth integration

How It Works

Three-pass scanning engine:

1. Phrase scan — Catches multi-word patterns first ("in order to" → "to", "due to the fact that" → "because"). Regex-powered for flexible matching.
2. Word scan — Flags individual words from a 154-entry dictionary across 6 categories. Skips words already covered by phrase matches.
3. Structural analysis — Detects missing contractions, em-dash overuse, uniform paragraph lengths, repeated sentence starters, and excessive hedging.

Dictionary: 154 words + 89 phrases across 10 categories:

Category	Examples	Count
AI overused verbs	delve, leverage, foster, utilize, facilitate, harness, empower	36
Grandiose descriptors	meticulous, multifaceted, robust, seamless, unprecedented	31
Latinate verbs	investigate, eliminate, substantiate, exacerbate, mitigate	26
Formal verbs	commence, ascertain, procure, disseminate, juxtapose	26
Fancy nouns	tapestry, landscape, paradigm, nexus, zeitgeist	19
Stiff transitions	moreover, furthermore, consequently, henceforth, whereby	16
Filler phrases	"it is worth noting that", "in today's fast-paced world"	21
Redundant phrases	"in order to", "due to the fact that", "has the ability to"	41
AI clichés	"plays a significant role in", "serves as a testament to"	22
Structural rules	contractions, em dashes, paragraph uniformity, hedging	22

Quick Start

git clone https://github.com/wang2-lat/native-english-tool.git
cd native-english-tool
pip install -r requirements.txt
uvicorn app:app --reload --port 8000

Open http://localhost:8000 — paste text, hit Analyze.

Chrome Extension

Works standalone — no server needed. The full detection engine runs locally in your browser.

1. Open chrome://extensions/
2. Enable "Developer mode"
3. Click "Load unpacked"
4. Select the chrome-extension/ folder

Features:

• Popup: paste text and analyze instantly
• Right-click: analyze any selected text on a webpage
• Floating panel: see results right on the page

API Endpoints

Endpoint	Method	Description
/	GET	Web UI
/api/analyze	POST	Analyze text → issues + score
/api/fix-all	POST	Auto-fix all issues
/api/llm-rewrite	POST	AI rewrite (sentence-level, DeepSeek)

curl -X POST http://localhost:8000/api/analyze \
  -H "Content-Type: application/json" \
  -d '{"text": "We must leverage our robust capabilities.", "formality": "casual"}'

Nativeness Score

Each issue carries a severity weight:

Severity	Weight	Examples
High (red)	3	delve, utilize, tapestry, "it is worth noting that"
Medium (orange)	1.5	landscape, robust, nuanced, "in order to"
Low (yellow)	0.5	investigate, "when it comes to", missing contractions

Score = max(0, 100 - (penalty / word_count × 100) × 8)

A clean text scores 90-100. Heavy AI-speak scores below 30.

Security

• Input validation: max 50K chars, formality enum whitelist
• XSS protection: all DOM rendering via createElement/textContent, no innerHTML with user data
• Rate limiting: LLM endpoint capped at 10 req/min per IP
• CORS: localhost-only by default
• Security headers: X-Content-Type-Options, X-Frame-Options, X-XSS-Protection
• .gitignore protects .env with API keys
• 34 tests including ReDoS, adversarial input, null bytes, and unicode edge cases

For AI Rewrite (Optional)

The rule-based engine works without any API key. For sentence-level AI rewriting:

cp .env.example .env
# Add your DeepSeek API key to .env

Project Structure

native-english-tool/
├── app.py                  # FastAPI app + security middleware
├── engine/
│   ├── detector.py         # Three-pass scanning engine
│   ├── replacer.py         # Back-to-front replacement
│   └── llm_fallback.py     # DeepSeek API fallback
├── dictionary/
│   ├── words.json          # 154 word-level rules
│   ├── phrases.json        # 89 phrase-level rules
│   └── patterns.json       # Structural rules + contractions
├── chrome-extension/       # Standalone Chrome extension
│   ├── manifest.json
│   ├── popup.html
│   ├── engine.js           # Full JS port of detection engine
│   ├── content.js          # In-page highlighting
│   └── dictionary.js       # Auto-generated from Python dicts
├── templates/
│   └── index.html          # Web UI
└── tests/
    └── test_engine.py      # 34 tests (security included)

Who This Is For

• Non-native English speakers who want to sound more natural
• Content writers cleaning up AI-assisted drafts
• Students improving academic writing
• Developers writing documentation
• Anyone tired of reading (or writing) corporate buzzword soup

Contributing

PRs welcome. The easiest way to contribute:

1. Add dictionary entries — open dictionary/words.json or phrases.json, add entries following the existing format
2. Report false positives — open an issue with the text that was incorrectly flagged
3. Add language variants — British English, academic English, etc.

License

MIT