I'm curruntly trying to improve the results of my dependency parser (arc-standard) when I found this article: that uses a dynamic oracle. https://www.aclweb.org/anthology/Q/Q13/Q13-1033.pdf The authors say that this technique can improve results by 2% but still I dont know how to integrate this to my neural network. I'm using Theanets ( 1 layer neural network) as inputs to the training I generated gold actions (shift, left-arc or right-arc) from conll-u data (many contextes[3 top words of the stack, 3 top words on the buffer and their pos tags ... etc]) and each context have a gold action that leads to the gold tree, as output I have the predicted action. Can anyone help me with this issue please.

  • This sounds like more of a computer science question than a linguistics question. Usually for programming questions it is necessary to show the code you are having problems with, and if this is the case it is off-topic here. You might have better luck on stackoverflow.com, although I think you'd need to flesh out your question some more. Also, keep in mind that there aren't likely to be that many experts in this particular field, so your question might remain unanswered for a long time. I would love to help you myself, but I don't have the background in neural nets. – CJ Dennis Apr 25 '16 at 3:16
  • @CJDennis It would be nice if you could explain to me how to calculate new gold actions after an error of the parser while parsing a sentence – HatemB Apr 28 '16 at 9:48
  • Unfortunately, while I find what you're trying to do fascinating, I have absolutely no experience with it. I have never used the tool you are working with. I would have to do a lot of studying before I could begin to answer your question! This is why I'm suggesting that it might be difficult to find an expert to help you. – CJ Dennis Apr 29 '16 at 5:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.