1

I am wondering how, if possible at all, one might train a new SyntaxNet model that uses the training data from the original, out-of-the-box "ready to parse" model included on the github page. What I want to do is add new training data to make a new model, but I don't want to make an entirely new and therefore entirely distinct model from the original Parsey McParseFace. So my new model would be trained on the data that the included model was trained on (Penn Treebank, OntoNotes, English Web Treebank), plus my new data. I don't have the money to buy from the LDC the treebanks the original model is trained on. Has anyone attempted this? Thanks very much.

Luke Strom
  • 11
  • 1
  • There are some pretty detailed instructions relating to your question [here](https://github.com/tensorflow/models/tree/master/syntaxnet). They are for python. What programming language do you want? – Jason Aug 08 '16 at 18:19
  • 1
    Where exactly in the training section does the readme mention how I can achieve my goal? When you train a new model, your training data is all that it has for features. I don't see a way to extend the Parsey McParseface model's training data as I would like to do. – Luke Strom Aug 11 '16 at 20:52

0 Answers0