A dependency parser for code-switched Telugu-English

This study addresses a notable gap in the availability of Code Switched Telugu-English dependency parsers. Conducting research into Code Switched data, especially Telugu-English could be very important, with the prevalence of Code Switching in native Telugu speakers. This is due to the fact that English is widely understood in Telugu-Speaking areas, and since code switching is a prevalent phenomenon in those areas. Through an informal blog dataset, the first Code Switched dependency treebank for Telugu-English was annotated. This treebank is limited to 300 sentences, to serve as an experiment to understand the feasibility of developing more Telugu-English Code Switched technologies. An error analysis was also performed, exploring the common issues faced by dependency parsing models and their possible resolutions. Using the UDify parser, a universal part of speech accuracy of 56.17%, an unlabeled attachment score of 30.61%, and a labeled attachment score of 11.79% was achieved.