Awesome

VnMarMoT for Vietnamese POS tagging

VnMarMoT is a pre-trained MarMoT model for Vietnamese Part-of-Speech (POS) tagging. VnMarMoT obtains a state-of-the-art POS tagging accuracy at 95.88% on the benchmark Vietnamese treebank, with a tagging speed at 25K words/second computed on a personal computer of Intel Core i7 2.2 GHz. See more details in our paper:

Dat Quoc Nguyen, Thanh Vu, Dai Quoc Nguyen, Mark Dras and Mark Johnson. 2017. From Word Segmentation to POS Tagging for Vietnamese. In Proceedings of the 15th Annual Workshop of the Australasian Language Technology Association, ALTA 2017, pages 108-113. [.bib]

Please cite our ALTA 2017 paper when VnMarMoT is used to produce published results or incorporated into other software.

VnMarMoT has also been incorporated into our Java NLP annotation pipeline VnCoreNLP for Vietnamese. VnCoreNLP provides rich linguistic annotations through key NLP components of word segmentation, POS tagging, named entity recognition and dependency parsing.

Usage

// Convert a word-segmented corpus into column-based representation
$ python Utility.py test.txt test.col.txt
// Perform POS tagging using VnMarMoT
$ java -cp marmot.jar marmot.morph.cmd.Annotator --model-file vn.marmot --test-file form-index=<WORD-FORM-COLUMN-INDEX>,<INPUT-COLUMN-FORMATTED-FILE> --pred-file <OUTPUT-FILE>
// Example:
$ java -cp marmot.jar marmot.morph.cmd.Annotator --model-file vn.marmot --test-file form-index=0,test.col.txt --pred-file test.pred.txt