Is there a dataset somewhere with sentences/paragraphs labeled with cognitive biases? The idea of a machine learning program auto labeling college essays with faulty deductive logic is enticing. Then run the system backward to generate logically sound arguments like the yahoo Nsfw detector post recently.