The inability to do this before was the lack of self-correcting sentences in the training data. Presumably new training corpuses add many more examples of self-

The inability to do this before was the lack of self-correcting sentences in the training data. Presumably new training corpuses add many more examples of self-correcting sentences / paragraphs?