SHSK0118/HFTransformersT5Translation_jp-en-ru

---
task_categories:
- translation
language:
- ja
- en
- ru
---

<h1>HFTransformersT5Translation_jp-en-ru</h1>

<p>
Multilingual Neural Machine Translation using T5 (Hugging Face Transformers)
for Japanese, English, and Russian.
</p>

<hr>

<h2>Model Description</h2>

<p>
This repository provides a multilingual neural machine translation system
built using T5 from
<a href="https://huggingface.co/docs/transformers/index">Hugging Face Transformers</a>.
</p>

<p>
The system supports translation between:
</p>

<ul>
<li>Japanese (ja)</li>
<li>English (en)</li>
<li>Russian (ru)</li>
</ul>

<hr>

<h2>Supported Language Pairs</h2>

<ul>
<li>ja ↔ en</li>
<li>en ↔ ru</li>
<li>ja ↔ ru</li>
</ul>

<hr>

<h2>Training Data</h2>

<p>
The training corpus is publicly available at:
</p>

<p>
<a href="https://huggingface.co/datasets/SHSK0118/HFTransformersT5Translation_jp-en-ru">
HFTransformersT5Translation_jp-en-ru Dataset
</a>
</p>

<p>
Each corpus is split into:
</p>

<ul>
<li>Train (90%)</li>
<li>Validation (5%)</li>
<li>Test (5%)</li>
</ul>

<hr>

<h2>Training Procedure</h2>

<ul>
<li>Model: T5 (Transformers)</li>
<li>Tokenizer: T5Tokenizer</li>
<li>Loss: Cross-entropy</li>
<li>Evaluation Metrics: BLEU, RIBES</li>
</ul>

<hr>

<h2>Evaluation</h2>

<p>
Evaluation is performed using corpus-level:
</p>

<ul>
<li>BLEU</li>
<li>RIBES</li>
</ul>

<hr>

<h2>Usage</h2>

<pre>
python src/train.py
python src/t5_test.py
python src/translate.py
</pre>

<hr>

<h2>Limitations</h2>

<ul>
<li>Domain-specific corpus</li>
<li>Performance varies across language pairs</li>
<li>Japanese–Russian pair contains fewer training samples</li>
</ul>

<hr>

<h2>Intended Use</h2>

<p>
This implementation is intended for:
</p>

<ul>
<li>Educational purposes</li>
<li>Research prototyping</li>
<li>Multilingual translation experiments</li>
</ul>

<hr>

<h2>Author</h2>

<p>
Independent implementation by Shota Tokunaga.
</p>