Naver to announce research results on Korean-language AI model at EMNLP 2021

By Song Su-hyun
  • Published : Sept 3, 2021 - 11:22
  • Updated : Sept 3, 2021 - 11:22


Naver, the largest internet business in South Korea, will present seven research papers on natural language processing in artificial intelligence at a globally renowned academic conference, the company said Friday.

Set to be held in the Dominican Republic from Nov. 7 to 11, the Empirical Methods in Natural Language Processing 2021 is a global conference with 25 years of history.

At the upcoming conference, Naver will unveil its hyperscale Korean-language model Hyper Clova and introduce data used for its learning as well as the results of its research on capabilities of language models of various sizes.

Naver has proved Hyper Clova has outstanding capabilities for in-context learning even when it is given examples only, being able to complete a variety of tasks written in the Korean language, it said.

The company will also further elaborate on the level of discussions concerning prompt optimization and tokenization in language models, and give details on innovative changes to be brought by hyperscale AI in development methods for AI services.

“Naver has focused efforts on hyperscale AI since the second half of last year, and as a result, we have commercialized the country’s first Hyper Clova technology and achieved a feat of winning global recognition,” said Ha Jung-woo, director of Naver AI Lab.

“It is meaningful in that the global community recognizes the value of an AI model in the Korean language,” he said.