The code implements the BLLIP metric which is an automatic machine translation evaluation metric that uses dependency sets for comparing reference and candidate sentences.
This code uses the Hindi Dependency Parser designed by Siva Reddy to parse the candidate and reference sentences and get the dependency tags for them in two different output files (one for candidate and anither for reference sentences). Source: http://sivareddy.in/downloads
The reference paper used is: Pozar, Michael & Charniak, Eugene. (2018). Bllip: An Improved Evaluation Metric for Machine Translation.