We built a dependency parser for Hindi with a web-interface and, further demonstratedits cross-lingual usage by applying on Marathi,a language that is low-resourced yet not very different from the former. For our parser,we have applied and tested various techniques ranging from transition-based parser that use SVM as a classifier to deep neural net-work based parsers which incorporate recent context-based word embeddings like BERT and FastText. We have demonstrated the immediate zero-shot application of such a parser on Marathi and also applied an existing word-embedding alignment method called MUSE toimprove the cross-lingual application performance.

This is a joint work with Abhishek and Mahesh under the guidance of Prof. Ashutosh Modi

The report of the work can be found here