This is implementation of a dependency parser of the language Bengali. This repository contains a dataset of around 7500 annotated tokens of Bengali Text. For the annotation decisions (and caveats) please refer the report.
Contents
Data/ - Contains the annotated dataset. All the files are in CoNLL format.
Data/Train - The training set (total of 5463 tokens annotated) Data/Train/train.txt.conll - the annotated train file
Data/Test/TestA/testA.conll - the annotated test file Data/Test/TestA/testB.conll - the annotated test file
To run the parser:
You can use your own training data or use the one shared my me. To run do the following. cd TurboParser-2.1.0.
Train script: ./run_train.sh
Test script : ./run_test.sh
It will run all the 3 (basic, standard and full models) of Turbo Parser. The labelled/unlabelled accuracy will be printed on the console after the test script is run.
Thanks!