Open Translation Tools for Lezgi and Russian
A practical MT project built around real Lezgi texts. This hub brings together models, datasets, and a live Telegram bot to support bilingual communication and language technology for a low‑resource language.
What’s included
Bidirectional Lezgi↔Russian translation with curated parallel data and a deployment-ready model.
Project Overview
Dataset
Parallel corpora from Lezgi Wikipedia and Lezgi Gazet texts, with Russian translations.
View datasetWhy this matters
Low‑resource focus
Lezgi has limited digital resources. Open tools help communities, researchers, and educators.
Reproducible pipeline
Datasets, training notes, and model cards make the work transparent and reusable.
Practical access
A Telegram bot makes translation available without installing any software.
Bot Examples
Translator mode
Uses the trained NLLB model for sentence translation. You can also forward a message from another chat and the bot will translate it. In auto mode the bot detects source language, but very short texts can be misdetected.
Dictionary mode
Activates when you send a single word. It uses the M.M.Gadzhiev dictionary, with normalization and fuzzy search. For example, "Горы/Дагълар" can still match "Гора/Дагъ" after morphological normalization. For Lezgi, fuzzy suggestions are shown (e.g., "Пирем" -> "Перем", рубашка).
Get involved
Contribute data
Have Lezgi texts to share? Reach out to contribute or improve the dataset.
Evaluate quality
Report translation issues or domain gaps to help prioritize improvements.
Build on top
Use the model and dataset to create tools for education, media, and research.