Use syntax information

[alexanderkoller]

One thing we could do that might help us across the board is to use syntax information from the companion data. The easiest thing to do would be to pass the label of the incoming dependency edge of each word to the neural network.

@namednil, how much work would this be? I suppose we would simply add another column to amconll and pass the syntax edge label there?

[namednil]

Sure, we could do that, or, even use the relative POS-based encoding.

That's not much work, an hour I guess (spent in am-tools and in the parser), and we don't even have to re-decompose everything.

[alexanderkoller]

Sounds great! Let's try it!