Machine Translation Evaluation using ROSE Metric
Objective
To improve accuracy of automatic evaluation of Machine Translated sentences.
What I did?
- Implemented the ROSE Machine Translation metric in Python.
- Tested multiple classifiers, such as, Pairwise Ranking SVM with RBF kernel for classification.
- Implemented a 3-layer fully connected Neural Net architecture using Tensorflow to use as a classifier for evaluation.
- Improved the accuracy of the classifier by using semantic features (extracted using spacy package) at later layers of neural net.
Results
- Improved the test accuracy from baseline of 52%(Simple METEOR) to 56%
Technologies Used:
- Python
- TensorFlow
- scikit-learn