/CherokeeLemmatizer

Utility to guess some affix splits on Cherokee texts. Developed to use with the Moses Machine Translation software.

Primary LanguageJavaApache License 2.0Apache-2.0

CherokeeLemmatizer

Utility to analyze Cherokee language texts.

  • Generate likely lemmas.
  • Show likely affixes.

Developed for use with the Moses Machine Translation software.

build:

./gradlew clean build fatjar

run: cat text.in | java -jar build/libs/CherokeeLemmatizer.jar > text.out

Where text.in is assumed to be pre-tokenized text. And text.out is output in "factored" format for use by translation systems like Moses-SMT.

The output format looks like:

ᏃᏊᏍ|ᎣᏊᏍ|Ꮒ-| ᏗᏥᏲᏟ|ᏗᏥᏲᏟ|| ,|,|| ᏛᏨᏃᏎᎵ|ᎥᏨᏃᏒᎢ|Ꮥ-|-ᎡᎵ ᏥᏔᎦ|ᏔᎦ|Ꮵ-| ᏄᏛᏂᏗ|ᎤᏛᏂᏗ|Ꮒ-| ᎦᏅᏅᎢ|ᎦᏅᏅᎢ|| ᏧᏂᏗᏫᏍᏗᎢ|ᏧᏂᏗᏫᏍᏗᎢ|| .|.||

Where each entry on a line corresponds to a word in the source text. Each entry consists of:

surface-form|lemma|prefixes|suffixes

Where:

  • surface-form = original word in text
  • lemma = guessed lemma for the original word.
  • prefixes = guessed sequence of prefixes applied to the lemma
  • suffixes = guessed sequence of suffixes applied to the lemma

Note:

  • The program uses heuristics via regex pattern matches to extract the likely prefixes and suffixes. The program will generated incorrect lemma|prefix|suffix results for some words which match the lookup patterns, but are already in their "lemma" form.

See examples for some example input/output files.