autoINcorrect 🤖💥

"A spell-checker-in-reverse"

Ever stared at your phone, absolutely convinced you typed "meeting," only for it to send "meating"? We've all been victims of autocorrect. But it got me thinking: what if we turned the tables?

This is the autoINcorrect system. It's a "spell-checker-in-reverse" that, instead of fixing mistakes, artfully creates them.

The Idea

This isn't just random chaos. It's a sophisticated system built on a simple, but (if I do say so myself) pretty smart idea: what if, to test how good a "fixer" is, we first built the perfect "breaker"?

By simulating how humans make real, plausible errors, we can build the ultimate stress-test for any real autocorrect model out there. It's an entire NLP pipeline, built in reverse.

This system understands:

Physical Errors: Simulating "fat-finger" slips from an actual keyboard layout.

Cognitive Errors: Simulating "brain-fart" moments like homophone swaps (your vs. you're).

Semantic Errors: Simulating "near-miss" malapropisms like banking -> banker using NLP (NLTK) and word embeddings (GloVe).

Formatting Errors: Simulating real-world punctuation mistakes and random case-flipping.

📦 First-Time Setup

Before you run the script, you'll need the necessary NLP models. Your pipeline.py file includes a helper function to do this automatically.

The first time you import or run the package, it will:

Check for the necessary nltk data (WordNet, Punkt, Tagger).

Download any missing nltk packages.

The first time you use the "Malapropism" module, it will download the glove-wiki-gigaword-100 model (this may take a minute).

🚀 Quick Start & Usage

This package is designed to be simple to use and easy to chain. The main wrapper function is auto_incorrect().

--- Import the main functions ---

from autoINcorrect import auto_incorrect, word_error, corrupt_format

my_text = "I am running a test on my banking application for its creative standards. It's a beautiful day to see if your function works correctly."

--- Option 1: Run the full chain with one function ---

(This uses all default settings)

print("--- Full Corruption (Default) ---")

This is the main wrapper function you'll usually use

corrupted_text = auto_incorrect(my_text) print(corrupted_text)

--- Option 2: Manually chain functions for more control ---

print("\n--- Manual Chaining (Custom Settings) ---")

1. Apply word-based errors

(Let's make errors common, with 70% typos and 30% smart errors)

custom_distribution = { 'fat_finger': 0.7, 'brain_fart': 0.3, 'malapropism': 0.0 } text_with_word_errors = word_error( my_text, error_rate=0.3, # Corrupt 30% of words distribution=custom_distribution )

2. Apply punctuation and case errors to that result

(Let's be aggressive: remove all punctuation and flip 80% of caps)

final_corrupted_text = corrupt_format( text_with_word_errors, add_rate=0.0, replace_rate=0.0, remove_rate=1.0, # 100% chance to remove punctuation case_corruption_prob=0.8 # 80% chance to flip a CAPITAL )

print(f"\nOriginal:\n{my_text}\n") print(f"Corrupted:\n{final_corrupted_text}")

🛠️ Key Modules

Your package is split into two main corruption functions, which are combined in the auto_incorrect() wrapper.

1. word_error()

This function handles all word-based corruption. It contains three sub-modules:

⌨️ "Fat Finger" Module: Simulates physical keyboard mistakes.

adjacent_swap: standard -> stamdard (m near n)

transpose: standard -> standrad

delete_char: standard -> standad

insert_char: standard -> standfard

🧠 "Brain Fart" Module: Simulates cognitive "word choice" mistakes.

replace_homophone: your -> you're

replace_morphological: decision -> decide (uses NLTK/WordNet)

🎭 "Malapropism" Module: The smartest module. Creates "near-miss" semantic errors.

replace_semantic_neighbor: banking -> banker (uses GloVe embeddings to find related words with the same root).

2. corrupt_format()

This function handles all punctuation and case-based corruption.

Punctuation: Randomly adds (word,), removes (word. -> word), or replaces (word? -> word;) punctuation marks.

Case: Randomly flips CAPITAL letters to lowercase based on a probability.

📦 Installation

pip install autoINcorrect

📜 License

This project is licensed under the MIT License.