I am doing a thesis on word embeddings for the Latvian language. I have made a bunch of embeddings and i want to evaluate them on an NLP task. I have chosen the POS tagging task. I am very basically familiar with neural networks, but i think at this point i dont have enough time to learn how to make a full architecture for the NLP task. I cant seem to find any easy solutions to just input my embeddings files into a pipeline and have it tag a text file. I have conllu format files for train, dev and test sets in my language to use for the POS tagger training and evaluating.

I have found UDPipe lib in python and it seems, that it is possible to do this with it, but the documentation and python bindings from c++ are inconsistent and dont line up. They are also just confusing, since some things are mentioned that can be done, but are not referenced anywhere else in the doc.

There is also an open source small project designed for a similar research paper in my language that does exactly what i want. It uses Tensorflow, but it is very outdated and i could not get it to work. Lots of different packages had different versions and language version requirements. Its a mess. I still need my environment to work with the stuff i've coded before, to generate the embeddings.

Preferably i would like to find a tool with a GUI even web or scriptable in python, as that is the fastest for me and deadlines are approaching. But i can also work with C# and node.js quite well.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.