Hindi Dependency Parser


Your input file should have the extension .input.txt e.g. hindi.input.txt To dependency tag your input file run "make <filename>.output" e.g.

make hindi.output

This will create a file named hindi.output

See hindi.input.txt and hindi.dependency.parser.out.txt for sample input and output files. You can change the entries in Makefile to work with any desired input.

Output format

The output format contains the following columns separated by tab space.

word id word lemma POS Tag parent id dependency label
8 वर्षों वर्ष NN 12 k7t

Dependency Tagset:

See dep-tagset.pdf for the dependency tagset details. The tagset is briefly described in [1].


We train and test Malt Dependency Parser [2] on Hindi ICON 2010 shared task data [3]. We use the features word, lemma and postag (for postposition tags we lexicalize the tag by appending the word to the tag). You can build a better parser using other features such as morphological information, which we do not do here.

Parser Accuracy: (Parser trained on features word, lemma and postag)

  Labelled   attachment score: 4948 / 6588 * 100 = 75.11 %
  Unlabelled attachment score: 5555 / 6588 * 100 = 84.32 %
  Label accuracy score:       5199 / 6588 * 100 = 78.92 %


Please cite wherever required.


Free to use for research purpose. You have to get a license from [LTRC IIIT Hyderabad] ( for commercial purposes. Please contact us for additional details.


Siva Reddy

Anil Krishna Eragani


Bharat Ram Ambati


[1] Begum, Rafiya, Samar Husain, Arun Dhwaj, Dipti Misra Sharma, Lakshmi Bai, and Rajeev Sangal. "Dependency Annotation Scheme for Indian Languages." In IJCNLP, pp. 721-726. 2008.

[2] Nivre, Joakim, Johan Hall, and Jens Nilsson. "Maltparser: A data-driven parser-generator for dependency parsing." Proceedings of LREC. Vol. 6. 2006.

[3] Husain, Samar, Prashanth Mannem, Bharat Ram Ambati, and Phani Gadde. "The ICON-2010 tools contest on Indian language dependency parsing." Proceedings of ICON-2010 Tools Contest on Indian Language Dependency Parsing, ICON 10 (2010): 1-8.