ETRI Commercializes Korean AI Technology Exobrain
Successful development of deep question answering technology and securing versatility
Hancom Office 2020 applied to legal inquiry service


Domestic researchers have succeeded in commercializing Korean AI technology.

The Electronics and Telecommunications Research Institute (ETRI) announced on October 31 that it has successfully developed and commercialized 'Exobrain,' a natural language deep question-answering technology that receives keywords and questions written in natural language and finds the correct answer.

This technology enables not only web search functions to locate documents and provide short-answer responses, but also high-difficulty descriptive question-and-answer capabilities. By utilizing this, the 'General Knowledge In-depth Q&A Technology' and 'Legal Knowledge In-depth Q&A Technology' services were developed, which ultimately led to commercialization.

The general knowledge deep question-answering technology developed by the research team analyzes Wikipedia to find relevant answers. It works by having a machine identify the question type and then applying a solution optimized for that type to find the correct answer.

ETRI developed the language model KorBERT last June. ETRI developed this Exobrain deep question answering technology by advancing the neural search and machine reading comprehension (MRC) technologies applied to Covert and ensuring their versatility.

This technology incorporates Korean language question analysis technology, semantic knowledge extraction technology, Wikipedia-based short-answer and descriptive question-answering technology, and a distributed question-answer processing platform technology.

Hancom has incorporated this technology into the 'Knowledge Search' function of 'Hancom Office 2020,' which was released on the 10th of last month. Users can use this feature through the cloud service after signing up for 'OfficeTalk'.

If you select OfficeTalk from the tools in Hangul, press the # key, and ask a question, you can obtain an immediate answer on the right side of the screen without needing to search through a separate portal.

Hancom announced that when comparing Exobrain with Google Knowledge Graph search on general knowledge questions, Exobrain showed performance up to 10% higher.

Kim Man-soo, Head of the Future Technology Research Division at Hancom, stated, “Previously, users had to search for information on portals when working with Hangul,” adding, “With Exobrain integrated into the latest version of Hangul, users can now find the information they need conveniently and quickly, thereby increasing document creation productivity and efficiency.”

The research team also succeeded in developing deep question-and-answer technology capable of providing not only short-answer responses but also descriptive answers. It is at a level where it can be utilized in the legal field, which is rich in technical terms and Sino-Korean words.

In the case of legal Q&A, a specialized deep learning language model was constructed for legal documents to understand the context of specialized terminology and sentence order within the legal documents. In addition, the service was implemented by applying machine reading comprehension technology capable of solving problems optimized for both short-answer and essay question types.

The National Assembly Library and the National Science and Technology Council (NST) plan to utilize Exobrain as software for their AI-based legal services starting next year. The NST plans to use Exobrain to respond to simple legal questions and assist experts in the search and response process, replacing the services currently provided by hired lawyers.

Dr. Kim Hyun-ki of ETRI’s Language Intelligence Lab stated, “Exobrain’s deep question-answering technology will contribute to the commercialization of AI in Korea,” adding, “Going forward, we plan to research and develop technologies related to ‘AI knowledge avatars’ that enable question-answering not only through text but also through voice, allowing for closer communication with users.”

Meanwhile, Exobrain's machine reading technology currently ranks first with 95.02 points in KorQuAD 1.0, a Korean machine reading competition.