May you provide some example about conllu format ?

[hecongqing]

No description provided.

[Oneplus]

Please refer the readme for an example.

Then, prepare your input file in the conllu format, like
1 Sue Sue _ _ _ _ _ _ _
2 likes like _ _ _ _ _ _ _
3 coffee coffee _ _ _ _ _ _ _
4 and and _ _ _ _ _ _ _
5 Bill Bill _ _ _ _ _ _ _
6 tea tea _ _ _ _ _ _ _

[Oneplus]

I will close this issue. If there is a further question, please reopen it.

[frankier]

You might find this helpful: https://github.com/EmilStenstrom/conllu

[jbrry]

Hi Yijia @Oneplus,

Could you give an example of running elmoformanylangs to compute ELMo representations for a conllu file and loading the vectors in a python file?

I have tried something similar to #9 but I cannot "get" the vector for a specific sentence. Using the allennlp module, I am able to get a sentence based on the sentence index, like so:

vecs_file = os.path.join(args.elmo_output_dir, 'en_lines-sentences.hdf5')   
h5py_file = h5py.File(vecs_file, 'r')  
embedding = h5py_file.get("0")

However, when I use elmoformanylangs to compute the ELMo representations, when I try get the vector it always returns a None value. I have even tried providing the full tab-separated sentence but I can never find the sentence in my program, e.g.

embedding = h5py_file.get("He and Ron went down to breakfast to find Mr and Mrs Weasley and Ginny already sitting at the kitchen table .") or embedding = h5py_file.get("0") always returns None.

The command I am running is:

python -m elmoformanylangs test --input_format conll --input ~/en_conllu/en_lines-ud-dev.conllu --model ~/ELMo/en.model/ --output_prefix sample --output_format hdf5 --output_layer -1

The output of such a command gives rather unusual looking output. E.g.

tail sample.ly-1.hdf5 looks like:
https://imgur.com/a/JWKeV4v

So my question is, does something look erroneous with my output/command or how do I load a particular sentence embedding when using a conllu file?

Thanks!