He'd invented a basic evolving text-analysis algorithm.
Employing statistical method using a cloud of text, he could produce an intelligible version of almost any sample, a sort of a translation of ones' speech into ones' own speech as clearly elocuted.
One evening, he (or she) decided to see what it could do. To his or her surprise, the intention of the writer or speaker became clear.
Ecstatic with a latenight manic buzz of curiosity, he couldn't stop trying to figure out what the program would be able to do.
Adjusting parameters and experimenting on various written material, he discovered that most of what people say is just a bunch of bullshit.