Comment Sigh not a breakthrough... (Score 0) 142
"using a vast database of annotated examples"
Oh Rlly? You have a massively perfect training set and things just magically work better? I'm shocked.
It reminds me of the paper I read about a "breakthrough" in text processing that mysteriously used the exact same algorithms already known in the field. When you went through the finer details of the paper you found that they had the grad-student slave manually go through all the training pairs and select the "correct" training data. Shockingly enough the hand-tuned results worked better when applied to an artificially narrowed test set....