As I point out elsewhere, machine learners and NLP researchers need this. This could offer tons of manually labeled training data.