I run your code both on PTB and CTB dataset, but in CTB data LAS could not achieve the performence in paper

[Woooooody]

Could you offer the LAS result of PTB and CTB. thx

[jcyk]

Sorry, but I don't have the results. Could you report your results here? Just the LAS of CTB is not good?

[Woooooody]

Thank you for your response. At present, I got USA: 89.86, LAS: 84.09 on CTB. But in paper, this number is 89.30 and 88.23, respectively. I got more good result than paper on UAS, but LAS is not good enough. By the way, I got the same result on PTB3.03 both of UAS and LAS. All in all, thanks for your share.

[baoy-nlp]

in the lib/data.py line 25 and line 29 ,it assumes that the root rel must be “root”.This is not necessary in some data sets, It may be "ROOT".

[jcyk]

@baoy-nlp wow, thank you!