Fall Research Intern

In my ongoing internship, I am working on multilinguality, intepretability and model analysis of neural networks in machine translation. The following summarizes the work I have been doing:

• Understanding the translationese phenomenon in human translation by exploring feature learning and feature engineering approaches in translationese classification. The BERT-based model outperforms all other models in both monolingual and multi-lingual settings.
• Investigating if representations learned by classifiers, the classifiers’ architecture, or both contribute to the performance difference.
• Utilizing deep learning explainability techniques like integrated gradients and probing to better understand performance differences.