uzmorph-transformer: Uzbek Neural Morphological Analyzer (Transformer Encoder Architecture)

uzmorph-transformer is a state-of-the-art word-level morphological analyzer for the Uzbek language. Unlike sequential RNNs, it uses a Multi-Head Self-Attention (Transformer Encoder) mechanism to capture complex character dependencies.

Performance & Use Case

• Architecture: Transformer Encoder (Multi-Head Attention).
• Strength: Exceptional at analyzing very long or complex words (6+ suffixes) where sequential models might lose context. Handles long-range phonological dependencies with high precision.
• Accuracy: >95% (Scales significantly with complex data).
• Ideal For: Research-heavy NLP tasks and analyzing complex, highly-agglutinative Uzbek technical or literary texts.

Installation

pip install uzmorph-transformer

Quick Start (Usage Examples)

1. Simple Analysis (String Output)

from uzmorph_transformer.uzmorph_transformer import uzmorph_transformer

analyzer = uzmorph_transformer()
result = analyzer.analyze("kitoblarimizdagilar")
print(result)

# Output:
# Result: 'kitoblarimizdagilar' -> Stem: kitob | POS: NOUN | Tags: [plural=1, possession=1, cases=Locative, plural=1]

2. Structured Data Export

# To Dictionary
data = analyzer.analyze("yozayapmiz").to_dict()

# To JSON
json_out = analyzer.analyze("olma").to_json()

Supported Tags & Features

Part of Speech (POS)

• NOUN (Ot), VERB (Fe'l), ADJ (Sifat), ADV (Ravish), NUM (Son), PRN (Olmosh).

Grammatical Features

• Cases: Nominative, Ablative, Accusative, Dative, Genitive, Locative.
• Possession: 1, 2, 3.
• Tense: Past, Present, Future.
• Voice: Causative, Passive, Reciprocal, Reflexive.
• Mood: Conditional, Imperative, Progressive, Message, Proposal.

License

MIT