Logo Lanfrica

Niger-Volta-LTI/iranlowo

Domaine:

natural language processing

Type de record:

software
Créateur:
Nig
Hôte:
Ìrànlọ́wọ́ is a utility library for analysis & (pre)processing of Yorùbá text → pypi.org # Ìrànlọ́wọ́ Ìrànlọ́wọ́ is a set of utilities to analyze & process Yorùbá text for NLP tasks. The focus is on *helping software developers* build large, clean text datasets for (further) diacritic restoration and machine translation tasks. ## Features ### ADR tools * [X] Strip all diacritics from word-types * [X] Verify that text is NFC or NFD * [X] Normalize a corpus (from MS Word or elsewhere) → NFC * [X] Split long sentences on certain characters like `;`,`:`, etc * [X] Automatically restore correct diacritics using a pre-trained model * [X] Find all variants of all word-type in a given corpus * [ ] Partially strip diacritics from word-types ### Ready to use webpage scrapers * [X] Bíbélì Mímọ́ (Biblica, Bible Society of Nigeria) * [ ] Yorùbá Blog * [ ] BBC Yorùbá ### Corpus analysis tools * [X] Dataset character distribution * [X] Dataset ambuiguity statistics → Lexdif, etc for a given corpus * [ ] Dataset scoring (proximity to correctly diacritized text, LM perplexity, KL divergence) ## Installation Obtainable from the Python Package Index (PyPI) → `pip install iranlowo` ## Example * Show computing environment and installation process * Diacritize a phrase ``` $ python Python 3.7.3 (default, Mar 27 2019, 16:54:48) [Clang 4.0.1 (tags/RELEASE_401/final)] :: Anaconda, Inc. on darwin Type "help", "copyright", "credits" or "license" for more information. >>> import iranlowo.adr as ránlọ >>> ránlọ.diacritize_text("lootoo ni pe ojo gbogbo ni ti ole") PRED AVG SCORE: -0.0037, PRED PPL: 1.0037 'lóòtóọ́ ni pé ọjọ́ gbogbo ni ti olè' ``` * Diacritize phrases, note we use `ipython` only because it renders nicer, easy-to-read text-colours in the terminal! ## Disclaimer This is beta software, if you pass the diacritizer out-of-domain text, English, pidgin or any other non-Yorùbá text, you will experience very marvelous, black-box results. Since this a work-in-progress and we are steadily improving, if you encounter any problems with …