End of terms period
Please submit the brief report by email to the lecturer's email address. Use "NLP: Task 07" as the subject.
In this assignment your task is to become familiar with the UDPipe tool and the Universal Dependencies project. Download Slovak UD data, analyze its CoNLL-U format and use UDPipe to train a prediction model. Review the documentation on how to do this. While training models on the training data, tune the hyperparameters according to performance on development data. Finally, measure the best model's accuracy on test data.
The point of this task is to get acquainted with the tool and format widely used in the field of dependency parsing. As you should know how to implement a basic dependency parser using lecture material, you do not have to develop your own parser. Instead, you are encouraged to study the implementation of the various approaches used in UDPipe. Your model does not have to be significantly good. It is sufficient if you try a few hyperparameter configurations (at least using random hyperparameter search as described in the user's manual).
Briefly report your findings (as described in the previous section) in a document in a few paragraphs (2-4 paragraphs should be enough). Specify the method you've used to tune the parameters, any problems you've come across if any, the best few configurations and their performance on the train, dev, and test set.
The task should be simple enough, so the grading will most likely be binary (if you manage to train a model and write the brief report, you get 100%). Partial credit may be possible.