Uzbek MWE Tokenizer

This package identifies Multi-Word Expressions (MWEs) in Uzbek language texts using a transformer-based intelligent model (UzBERT).

Citation Requirement

If you use this model in your research or project, please cite the following paper:

Sharipov, M. S. (2026). Transformer-based intelligent model for identifying multi-word lexical units and reducing syntactic ambiguities in Uzbek language texts. Bulletin of TUIT: Management and Communication Technologies, 2(14), 103-107. DOI: 10.61663/262tuitmct14
Havola: https://uzjurnal.uz/2/2026/2/index?issue=14

Installation

pip install uzbek-mwe-tokenizer

Usage

You can use the model for both Latin (lot) and Cyrillic (cyr) texts.

Example in Cyrillic

from uzbek_mwe_tokenizer import UzbekMWETokenizer

# Initialize tokenizer in Cyrillic mode
tokenizer = UzbekMWETokenizer(mode="cyr")

sentences = [
    "Натижани кўриб, ҳамма ўқувчиларнинг бирданига тарвузи қўлтиғидан тушди.",
    "Қоронғи хонада унинг қаттиқ капалаги учди, гўё ерга кирса, қулоғидан тортиб чиқарадиган ҳолат эди.",
    "Рақибларимизни кўриб бизнинг асло тепа сочимиз тик бўлмади."
]

for gap in sentences:
    print("Gap:", gap)
    mwes = tokenizer.extract_mwe(gap)
    print("Topilgan MWE'lar:", mwes)
    print("-" * 50)

Kutilayotgan Natija (Expected Output):

Gap: Натижани кўриб, ҳамма ўқувчиларнинг бирданига тарвузи қўлтиғидан тушди.
Topilgan MWE'lar: [{'mwe': 'тарвузи қўлтиғидан тушди', 'confidence': 100.0}]
--------------------------------------------------
Gap: Қоронғи хонада унинг қаттиқ капалаги учди, гўё ерга кирса, қулоғидан тортиб чиқарадиган ҳолат эди.
Topilgan MWE'lar: [{'mwe': 'капалаги учди', 'confidence': 100.0}, {'mwe': 'ерга кирса', 'confidence': 76.0}, {'mwe': 'қулоғидан тортиб чиқарадиган', 'confidence': 99.0}]
--------------------------------------------------
Gap: Рақибларимизни кўриб бизнинг асло тепа сочимиз тик бўлмади.
Topilgan MWE'lar: [{'mwe': 'тепа сочимиз тик бўлмади', 'confidence': 100.0}]
--------------------------------------------------

Example in Latin

The package automatically transliterates text to Cyrillic for the model and translates the output back to Latin.

from uzbek_mwe_tokenizer import UzbekMWETokenizer

# Initialize tokenizer in Latin mode
tokenizer = UzbekMWETokenizer(mode="lot")

text = "Natijani ko'rib, hamma o'quvchilarning birdaniga tarvuzi qo'ltig'idan tushdi."
mwes = tokenizer.extract_mwe(text)
print(mwes)

Kutilayotgan Natija (Expected Output):

[{'mwe': "tarvuzi qo'ltig'idan tushdi", 'confidence': 100.0}]