Braille Act at 10: How Is Braille Made?

Kook Hyun-woo, managing director at Acens, examined the significance of the Braille Act in his op-ed, "Braille Act at 10: How Is Braille Made?" published in IT Daily. Enacted in 2016 and implemented the following year, the act defined Braille as the script used in South Korea for people with visual disabilities and placed it on par with Hangul. Since then, it has been amended several times.

During that process, the system has been revised to further strengthen Braille's legal recognition and the obligation to provide it. Public institutions and others became obligated to provide Braille documents containing the same content when requested by people with visual disabilities, and they were also required to disclose the status of such requests and their fulfillment. In this way, institutional improvements have continued.

However, even with this legal recognition and the obligation to provide Braille in place, the op-ed said the change in the actual experience of receiving Braille has been minimal. In particular, the issue of producing Braille textbooks was presented as a long-accumulated challenge that needed to be institutionalized separately. Earlier this year, revisions to the Elementary and Secondary Education Act and a public-private agreement led to separate institutionalization of "timely distribution before the start of the semester."

In this regard, the National Assembly Research Service pointed out that simply shortening the deadline for publishers to submit original files has limitations. It also proposed that the entire process, from ordering and supply through production and distribution, needs to be improved at the same time. The discussion was also presented as moving beyond the timing of obtaining original files and toward reforming the entire process together.

In the end, the op-ed organized the task into two tracks. One is to secure the original text quickly, and the other is to efficiently turn the secured original text into Braille. Even as institutional improvements have continued over the 10 years since the Braille Act took effect, the context is that comprehensive process reform, including these two tasks, must work together to change the actual distribution experience.

The structure for providing Braille documents in public institutions is one in which an obligation to provide them arises when a person with a visual disability makes a request. The staff member receiving the request typically commissions an outside Braille transcription agency and waits for the result. But information has a shelf life. In cases such as closed recruitment notices or welfare programs whose application period has ended, access to meaningful information can become difficult when Braille is delivered weeks later.

The cause of this problem cannot be reduced to staff indifference alone. The current system imposes obligations, but does not provide the production tools needed to make compliance easy. For staff who do not know Braille, the only options are to request outside help and wait. As a result of that structure, requests decline and demand becomes invisible. When demand becomes invisible, it becomes difficult to build budgets and infrastructure.

The Disability Rights Guarantee Act was enacted in May 2026, and its implementation date is May 2028. The act explicitly defines access to knowledge and information as a basic right for people with disabilities. However, its limitation is that while it can create an obligation to provide, it cannot automatically create production capacity. Provision presupposes production, and the past 10 years have also shown that an obligation to provide alone is not enough. Accordingly, the task needs to shift from whether Braille needs to be provided to how Braille should be produced.

The Braille production process remains fragmented. The Braille production procedure consists of obtaining the original text → proofreading Braille grammar → converting it through Braille transcription software → human proofreading, but these steps are not linked in a single environment. The main steps also depend on individual workers' PCs and installed software.

In this structure, there is also a problem with original-file management. Sensitive materials such as prepublication manuscripts and nonpublic documents may be copied to multiple workers' PCs during production. This makes it difficult to identify who holds the original file, and it is also hard to verify whether the original has been deleted after the work is finished.

This becomes a factor that makes publishers and institutions reluctant to provide the original file. As a result, Braille production often requires additional scanning and OCR even without the original. The extra scanning and OCR increase the production burden.

Problem 2 is the disconnect between work and data. It is difficult to grasp the stage of work progress, it is hard for multiple people to divide the same material for processing, and it is also difficult to check whether another institution has already produced the same material, leading to duplicate production. In addition, edits made by transcribers and proofreaders, along with the reasons for those edits, are lost after the work ends, showing that in an environment where work processes and data are separated, progress sharing, collaboration, duplication prevention, and preservation of revision history are not being handled properly.

Problem 3 is that continuous quality improvement is difficult. When Braille rules are revised, installed software must be updated individually on each PC, and because there are many cases in which previously produced materials do not have the applied engine or rules recorded, it is difficult to determine what should be re-produced when standards change. If expert-proofread automated Braille data remains on a personal PC, it fails to become an asset for improving the accuracy of future transcription, so the same fragmented structure blocks responses to rule changes, management of existing materials, and reuse of proofreading data.

The common cause of these three problems is presented as the fragmentation of the production environment and data. Accordingly, the Braille production environment is said to need a server- and web-centered redesign.

A web-based Braille production platform is a method that combines original-file storage, work execution, and standards application into a single system in a web environment. This platform makes it possible to integrate the production environment, and in a web-based production setting, originals, work processes, and transcription standards can be managed in an integrated way.

The original files are managed on the server, and workers view them through a browser. This makes it possible to record file checks and revision history, and to reduce the problem of files being left indiscriminately on individual PCs.

This platform also makes collaborative work possible. Multiple transcribers and proofreaders can divide up the same material, and because progress can be checked, personnel can be concentrated on urgent materials.

Before starting new production, it is also possible to search for previously produced materials. In addition, data on production volume, processing time, and user institutions are accumulated, and the accumulated data becomes a basis for future policy and budget design.

The transcription engine is also centrally managed. When new Braille rules are reflected, all users can work with the same standards.

It is also possible to record the applied engine and rules for each output. This makes it possible to identify the scope of impact when rules are changed in the future.

Rather than processing materials through the same procedure, the discussion suggests that responses can be divided between those requiring immediacy and those requiring accuracy. Examples of materials where immediacy matters include notices and learning materials needed today, which are handled by providing the results of automatic Braille transcription immediately. By contrast, examples of materials where accuracy matters include textbooks and test materials, which are handled by providing them after review by professional transcribers and proofreaders. Previously, both needs were placed in the same production queue, but a digital platform makes it possible to separate them by purpose.

The school setting is cited as a particularly meaningful area for this approach. As inclusive education expands, students with visual disabilities are increasingly learning together in regular classrooms, but teachers in those classrooms rarely know Braille, making reliance on outside transcription support unavoidable for now. In particular, beyond government-produced textbooks, learning materials are also created frequently during class, making it difficult to provide those materials in Braille at the needed moment. This platform would allow teachers to prepare lessons without knowing Braille, and the issue they need to judge would not be the Braille rules but the material's urgency and need for accuracy.

To implement this environment, three technologies need to be in place together: electronic document processing, electronic Braille, and vision AI.

The first foundational technology is electronic document processing. Target formats include HWP, HWPX, DOCX, PDF, and EPUB, and the text and structure of different document formats must be extracted accurately. In addition, the original text and Braille output must be comparable in the browser, and workers must be able to compare them without downloading the original file.

The second technology is electronic Braille. A transcription engine that complies with the latest Korean Braille rules is needed, as is file generation for Braille note-takers. In addition, web-based Braille editing and web-based Braille proofreading functions are needed.

The transcription area has a characteristic in which it requires stable results that comply with the rules, rather than plausible results in the style of generative AI. For that reason, the parts that require human judgment need follow-up review by specialists.

The third technology is vision AI. The target materials are scanned books and image PDFs, and the scope of restoration includes text, tables, equations, and reading order. As OCR and vision-language models advance, practical applicability has also been rising rapidly.

Document processing, transcription, and artificial intelligence are different specialties, so combining the three technologies into a single service is highly difficult. But if these technologies are linked, the entire process from reading the original text to generating and proofreading Braille can be connected within a single web environment. This is not a concept that remains at the idea stage; it is explained in the context that public authorities will first verify it and then decide how to spread it.

The author's company has developed each of the three areas separately. In electronic documents, it developed document conversion technology, server-based streaming viewer technology, and file verification technology, and these electronic document technologies have been applied in the public and administrative sectors. In electronic Braille, it developed web-based electronic Braille generation and editing technology and a server-based transcription engine.

It also participated in projects such as the National Institute of the Korean Language's parallel corpus construction project for print and Braille. In that process, it conducted transcription and proofreader verification on automatic Braille transcription results totaling 135,000 sentences. The review found an accuracy rate of more than 96%.

The related technology has already been applied to internet banking services at financial institutions and various government and public institution systems, and recently research and development using OCR and vision-language models has also continued. The R&D work is focused on structuring the text, tables, equations, and reading order of scanned documents.

However, possessing the technology and proving it works in the field are separate matters. Before public funds are invested in a new approach, evidence is needed regarding how quickly users receive materials, the level of automatic transcription quality, the degree to which expert review burdens are reduced, and changes in institutional operating costs.

Article 50 of the Disability Rights Guarantee Act specifies the state's responsibility for developing and disseminating digital technology, and Article 51 of the same act stipulates that pilot projects must be carried out and their results reflected when introducing new systems. Accordingly, the author's company has prepared a plan to pre-implement a cloud-based service for the production environment and move forward with field verification. The goal is field verification, with the aim of checking data on processing speed, quality, usage volume, and operating burden through real-world use of materials by schools, institutions, and users.

As evidence accumulates, the public sector can choose an introduction method based on results already confirmed. In that case, the private sector's advance role is to prove the technology and service's feasibility, while the public sector's follow-up choice determines how to expand private-sector achievements. Either way, it reduces the budget and time burden involved in the initial verification of technological feasibility.

There are two options. One is shared use in the form of a verified service's SaaS model, which has the effect of eliminating the need for each institution to build its own separate system and allows service use only in the amount needed. The other is separate construction of a public Braille production platform based on verified technology and operational data.

The Disability Rights Guarantee Act takes effect on May 27, 2028. By that time, Braille provision results must not remain marked as "none," and production infrastructure must be prepared in advance to ensure the actual fulfillment of rights under the law.

Source: IT DAILY · Kook Hyun-woo
Original: https://www.itdaily.kr/news/articleView.html?idxno=241308

References

• https://www.itdaily.kr/news/articleView.html?idxno=241308