A fairly robust LeetSpeak to English converter using probability of word usage
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
data
words
.gitignore
Dictionary.py
LeetSpeak.py
README.md
Spelling.py
leet.py
phonetics.py

README.md

leetspeak

Years ago I started working on Garot Bot 2.0. Well, that project was enormous and never completed. However, as part of the project I wanted to have a LeetSpeak to English translator. This multi-threaded Python 3 converter/translator deciphers your messages using probability of word usage taken from the Project Gutenberg from all the possibilities it can generate. I have yet to see any translator that works in as many cases as this one.

Using

  1. Install python3.
  2. Run: python3 leet.py --demo to test or python3 leet.py [-e] "Your Message" to decode (by default) or encode a message.

Notes

  • It will run slower the first time while it generates the .pickle files.
  • This program decodes better than encodes since encoding just randomly choses characters, which does not look at all like normal leet.
  • The program ignores all punctuation. How does it know if that period is part of the leet or is a period?
  • I'm not fond of the coding style I used back then. This was one of my first major Python projects.

Example

Some things work.

 [garrett leetspeak]$ python3 leet.py -e "How are you doing today mother?"
 )-(ω\_1_/ /\|2ə \j()/_/ [)¤ai/|/gee -l-¤l]/-\-/ /|/|{}-l-[-]€[z?

 [garrett leetspeak]$ python3 leet.py ")-(ω\_1_/ /\|2ə \j()/_/ [)¤ai/|/gee -l-¤l]/-\-/ /|/|{}-l-[-]€[z?"
 how are you doing today mother

And some things don't. Note that Github was not a common word a few hundred years ago.

 [garrett leetspeak]$ python3 leet.py -e "This is a test for Github."
 '][']-[eyeš 3y35 /-\ +e$-1- |=0® (_+l†}-{yuu|3.

 [garrett leetspeak]$ python3 leet.py "'][']-[eyeš 3y35 /-\ +e$-1- |=0® (_+l†}-{yuu|3."
 this is a tehiubhig for

License

The words lists are from /usr/share/dict, which I believe is from aspell, under GPL2. I would say where stopwords_en.txt came from, but I think that's a combination of many sites and my own thoughts. As for slang, it's from No Slang, and I don't see a license. The gutenberg_small.txt file is obviously from Gutenberg; although, I have since lost the script to generated it (hence why I just provide the file here). The phonetics.py file is from AdvaS Advanced Search and is GPL. All of my code, as always, is under the ISC license.