We introduce a method for learning to provide feedback on a given English sentence with a correction.
In our approach, the sentences and corrections are analyzed to identify the error type and a problem word, aimed at customizing explanations according to the context of the error type and the problem word.
The method involves identifying an error type and a problem word of the given sentence, generating common feedback patterns of each error type, and extracting grammar patterns along with collocations and example sentences for explanation.
At run-time, a sentence with a correction is classified, and the problem word and template are identified to provide detailed explanations.
We present a prototype feedback system, \textit{TellMeWhy}, that applies the method to an annotated corpus and an underlying grammatical error correction system.
Preliminary evaluation on a set of representative sentences with common errors shows that the method has potential to outperform existing commercial writing services.