I've kludged together a sentencizer for Norwegian (based on Text::Sentence) myself, so I know what you mean. It's not that bad, but the fact that some people are working at doing the same for audio streams is rather humbling.

Kleinbergs work seems interesting. I vaguely remember reading about it in the news, so I look forward to looking into it.

The negative/positive identifier is company work, so I'll probably be better of not elaborating on it for now.