Disclaimer: I work on finding new ways to use machine translation in intercultural collaboration.
What happens most of the time with slang is as you say, a simple mapping. The most efficient way to deal with slang syntactically incorrect terms is to use a custom dictionary in the machine translator. For example, "U" is translated as "you". To make it more complete, you might want to use a complete translation memory, not taking single words.
To make it short, you just need a custom element in your translation, to adapt to your domain. That's why machine translation such as Google Translate can't work very well since it can't get the context of the content you want to translate.