HeLI-OTS 1.3

HeLI off-the-shelf language identifier with language models for 200 languages.

Usage:
java -jar HeLI.jar -r <infile> -w <outfile>

The program will read the <infile> and classify the language of each line as one of the 200 languages it knows
and writes the results, one ISO 639-3 code per line, into file <outfile>.

You can use the -c option to make the program print a confidence score for the identification after each language code.

Usage:
java -jar HeLI.jar -c -r <infile> -w <outfile>

You can give the list of comma-separated ISO 639-3 identifiers for relevant languages after -l option.

Usage:
java -jar HeLI.jar -r <infile> -w <outfile> -l fin,swe,eng

You can give the number of top-scored languages to print after the -t option. (overrides confidence)

Usage:
java -jar HeLI.jar -r <infile> -w <outfile> -l fin,swe,eng -t 2

If you omit both of the filenames, the program will read the standard input one line at a time and write the result to standard output.

It can identify c. 3000 sentences per second using one core on a 2021 laptop and around 3 gigabytes of memory.

If you use this program in producing scientific publications, please refer to: 
 @inproceedings{jauhiainen-etal-2017-evaluation,
     title = "Evaluation of language identification methods using 285 languages",
     author = "Jauhiainen, Tommi  and
       Lind{\'e}n, Krister  and
       Jauhiainen, Heidi",
     booktitle = "Proceedings of the 21st Nordic Conference on Computational Linguistics",
     month = may,
     year = "2017",
     address = "Gothenburg, Sweden",
     publisher = "Association for Computational Linguistics",
     url = "https://www.aclweb.org/anthology/W17-0221",
     pages = "183--191",
 }

Producing and publishing this software has been partly supported by The Finnish Research Impact Foundation Tandem Industry Academia -funding in cooperation with Lingsoft.