Logo Lanfrica

HappymoreMasoka/shona-spacy

Domaine:

natural language processing

Type de record:

software
Créateur:
Hap
HĂ´te:
Shona spaCy — Rule-Based NLP Pipeline for the Shona Language 🧩 Overview Shona spaCy is a lightweight, rule-based Natural Language Processing (NLP) pipeline for the Shona language built using the spaCy framework. It provides morphological analysis, noun class detection, and verb tense recognition by combining: a verified JSON lexicon (hand-annotated tokens), and a set of linguistic rules derived from Shona grammar. The system lets researchers and developers perform tokenization, POS tagging, and basic morphology analysis on Shona text — similar to what spaCy provides for English, French, or German. ⚙️ Features ✅ Token-level morphological analysis ✅ Automatic detection of noun classes (Mupanda) ✅ Verb tense and subject concord identification ✅ Closed-class recognition for pronouns, conjunctions, adverbs, etc. ✅ Extensible via JSON lexicon for domain-specific Shona terms ✅ Compatible with spaCy ≥ 3.7 🧠 Example Output Input: Mbudzi iri kumhanya mumunda. Output: Mbudzi | NOUN | Mupanda 9 | NounClass=9|Rule=True iri | VERB | | Rule=True|SC=i|Tense=None kumhanya | VERB | | Rule=True|SC=ku|Tense=None mumunda | NOUN | Mupanda 18 | NounClass=18|Prefix=mu-|Locative=True|Internal=True|Agricultural=True . | X | | Unknown 🧩 How It Works The pipeline analyzes each token in three stages: Lexicon Lookup (JSON-based) If the token exists in shona_lexicon.json, the analysis uses its manually verified fields: pos, lemma, category_detail, morph_features, gloss, comments. Closed-Class Matching Common function words (e.g., ini, iwe, kana, uye, mangwana) are classified directly from rule tables. Rule-Based Morphological Parsing For unknown words, the analyzer applies Shona morphological rules: Noun Class (Mupanda): Prefix-based detection (e.g., mu-, va-, chi-, zvi-). Verb Analysis: Detects subject concords (ndi-, u-, a-, ti-, mu-, va-), tense markers (no-, ka-, cha-, a-, na-), and derivational suffixes (-a, -e, - …