User:Ruthenian8/GSOC 2021 progress report

From Apertium
Jump to navigation Jump to search
  • Title: Morphological analyzer for Bagvalal
  • Proposal: proposal
  • Abstract: Bagvalal is an endangered typologically rare Caucasian language from the Nakh-Daghestanian family. Its conservation and study are constrained by the lack of sufficient NLP-tools that can be used to process field data.
    My proposal is to develop an fst-powered morphological analyzer for Bagvalal using all the available grammatical and lexical information. In the future this project can allow Apertium to support morphological analysis for multiple Nakh-Daghestanian languages and develop corresponding language pairs.
  • GitHub repo: bagvalal
Week Intended changes Status
Week 1 Testing and refining the existing rules for the closed word classes (e. g. numerals, clitics and pronouns). Complete
Week 2 Writing documentation and tests. Complete
Week 3 & 4 Testing and refining the existing rules for the open word classes (e. g. verbs, nouns and adjectives).
Writing documentation and tests.
Complete
Week 5 Adding the missing adjectives and adverbs from the available dictionaries (see the Resources section above).
Testing the analysis results and the model performance.
Complete
Week 6 Adding the missing nouns.
Testing the analysis results and the model performance.
Complete
Week 7 & 8 Adding the missing verbs, participles, converbs and masdars.
Testing the analysis results and the model performance.
Complete
Week 9 & 10 Tokenizing the corpora.
Converting the existing annotations to an appropriate format
Creating word-analysis pairs.
Writing documentation.
In progress
Week 11 Expelling the false analyses from the model
Testing and debugging.
Finishing the work on the documentation
In progress
Week 12 Running all the tests and debugging In progress

Development log: 07.08.21 During the discussion with mentors it was determined that the Bagvalal-specific glosses which are present in the grammar and which have previously been incorporated in the analyzer should be left without modification. Taking into account that the analyzer is already in a ready state, it was decided to develop additional bilingual plug-ins for Russian and Avar.

Status: Naive coverage 82%, type coverage 76%.