For more information please see our paper Hierarchical Character-Word Models for Language Identification.

To train a model run the command

python langid.py path/to/outdir

For some simple visualization of a trained model run

python langid.py --mode=debug path/to/outdir

To evaluate a trained model run

python langid.py --mode=eval path/to/outdir

The data directory holds an example input file created from Wikipedia sentence fragments. The file is saved in tab separated format. We partition the data according to the last digit of the id number in the data file. Separate lines are used for training, validation, and testing.