KETI Unveils KE-T5 Model Based on Google T5
Available in both Korean and English
AI supports minority learning to increase efficiency


On the 28th, the Korea Electronics Technology Institute (KETI) unveiled the cutting-edge language processing model, 'KE-T5'. KE-T5 is a dual language processing model that can be used for both Korean and English processing, using self-supervised learning, built based on Google's language model algorithm, T5 (Text-to-Text-Transformer).

KE-T5 is the first Korean data-centric T5 series model and a general-purpose language model that includes both semantic recognition and expression characteristics of language. It supports AI few-shot learning, which builds learning models with a small amount of learning data, and shows high performance in various language processing, so it has high usability compared to the construction cost.

It is capable of processing Korean and English simultaneously, and is suitable for not only language comprehension such as document summarization, English-Korean and Korean-English translation, and conversation, but also high-level language processing that requires linked learning of expressions.

KETI has distributed 16 models on GitHub free of charge according to the type of text and the scale of learning, so that developers can select and use them according to their development environment and engine characteristics. In addition, 24 Korean-English summary and translation models that can be used immediately have been released. They can be used and distributed freely according to the Apache 2.0 license.

This technology was developed through the projects 'Technology to restore invisible parts in images using visual common sense through self-supervised learning' (2021-0-00537), 'Development of AI-based empathy service technology for emotional stability' (S0316-21-1002), and 'AI source technology that can infer facts and logical basis for each issue by learning non-standard text' (2021-0-00354) supported by the Ministry of Science and ICT, the Institute of IT Planning and Evaluation (IITP), and the National IT Industry Promotion Agency (NIPA).