transformers-ud-japanese-electra-base-ginza-510 Open-source Japanese Pre-trained Model - Accurately Handle Japanese Language Tasks

Transformers Ud Japanese Electra Base Ginza 510

Developed by megagonlabs

Japanese pretrained model based on ELECTRA architecture, pretrained on approximately 200 million Japanese sentences from the mC4 dataset and fine-tuned on UD_Japanese_BCCWJ corpus

Sequence Labeling

Transformers

JapaneseOpen Source License:MIT #Japanese Dependency Parsing #ELECTRA Architecture #spaCy Integration

Downloads 7,757

Release Time : 3/2/2022

Model Overview

This is a Japanese natural language processing model based on the ELECTRA architecture, specifically optimized for Japanese text and capable of recognizing Japanese bunsetsu structures.

Model Features

Trained on large-scale Japanese corpus

Pretrained on approximately 200 million Japanese sentences from the mC4 dataset

Domain-specific fine-tuning

Fine-tuned on UD_Japanese_BCCWJ corpus to optimize Japanese dependency parsing capabilities

GiNZA integration

Can be used with GiNZA v5 to provide comprehensive Japanese NLP processing capabilities

Model Capabilities

Japanese text analysis

Dependency parsing

Bunsetsu structure recognition

Use Cases

Natural Language Processing

Japanese text parsing

Analyzing the structure and dependency relationships of Japanese sentences

Accurately identifies Japanese bunsetsu structures

Japanese NLP application development

Serving as a base model for Japanese NLP applications

🚀 transformers-ud-japanese-electra-ginza-510 (sudachitra-wordpiece, mC4 Japanese)

This is an ELECTRA model pretrained on Japanese sentences from mC4 and finetuned on UD_Japanese_BCCWJ r2.8 using spaCy v3. It's distributed with GiNZA v5 to recognize Japanese bunsetu - phrase structures.

🚀 Quick Start

This is an ELECTRA model. It was pretrained on approximately 200M Japanese sentences extracted from the mC4 and then finetuned by spaCy v3 on UD_Japanese_BCCWJ r2.8.

The base pretrain model is megagonlabs/transformers-ud-japanese-electra-base-discrimininator.

The entire spaCy v3 model is distributed as a python package named ja_ginza_electra from PyPI along with GiNZA v5, which provides some custom pipeline components to recognize the Japanese bunsetu - phrase structures.

You can try running it as follows:

$ pip install ginza ja_ginza_electra
$ ginza

📦 Installation

$ pip install ginza ja_ginza_electra

📄 License

The models are distributed under the terms of the MIT License.

🙏 Acknowledgments

This model is permitted to be published under the MIT License under a joint research agreement between NINJAL (National Institute for Japanese Language and Linguistics) and Megagon Labs Tokyo.

📚 Citations

Contains information from mC4 which is made available under the ODC Attribution License.

@article{2019t5,
    author = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
    title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
    journal = {arXiv e-prints},
    year = {2019},
    archivePrefix = {arXiv},
    eprint = {1910.10683},
}

UD_Japanese_BCCWJ r2.8

Asahara, M., Kanayama, H., Tanaka, T., Miyao, Y., Uematsu, S., Mori, S.,
Matsumoto, Y., Omura, M., & Murawaki, Y. (2018).
Universal Dependencies Version 2 for Japanese.
In LREC-2018.

GSK2014-A(2019)

Property	Details
Model Type	ELECTRA
Training Data	mC4, UD_Japanese_BCCWJ r2.8, GSK2014 - A(2019)
Metrics	UAS, LAS, UPOS
License	MIT

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご