elmo-bilstm-cnn-crf icon indicating copy to clipboard operation
elmo-bilstm-cnn-crf copied to clipboard

Published 20 hours ago • verified publisher icon UKPLab

do this code also suitable for Chinese?

Open EvaluationResearch opened this issue 6 years ago • 4 comments

EvaluationResearch avatar Aug 03 '18 09:08 EvaluationResearch

Not really. As far as I know there are no ELMo embeddings for chinese so far. Further, you would need to provide word boundaries in order that the code is runnable. Otherwise it would read the complete sentence as a single token.

nreimers avatar Aug 03 '18 09:08 nreimers

@zhaoyueplc check here for a pretrained model for many language https://github.com/HIT-SCIR/ELMoForManyLangs but it is trained with customer setting, further modifier would be need according the issue reported there. I am planning to use the German pretrained model, would be nice if anyone can share that if it works. Thanks. John

Jorigorn avatar Sep 19 '18 11:09 Jorigorn

It appears that ELMoForManyLangs is sadly not compatible with the original AllenNLP architecture (and hence sadly cannot be used with this framework): https://github.com/HIT-SCIR/ELMoForManyLangs/issues/1

Maybe I find a way how to support their ELMo weights for AllenNLP

nreimers avatar Oct 23 '18 13:10 nreimers

Anyone have any success with this? I tried running their model on a conllu file but I am unable to "get" the representations like you would with the AllenNLP version.

If anyone knows the best way to do this please let me know.

https://github.com/HIT-SCIR/ELMoForManyLangs/issues/7

jbrry avatar Dec 06 '18 14:12 jbrry