Vietnamese startup's AI translation is better than Google and ChatGPT

Doctranslate.io is an AI (artificial intelligence) translation platform, founded by Dr. Tran Vu Anh. The startup was founded in 2023 and was selected by Google to participate in the Google Accelerator for Startups 2024 program. Recently, Doctranslate.io continued to be in the top 5 most potential startups at the National Innovation and Startup Festival (TECHFEST 2024).

AI translates better than Google, ChatGPT

After many years of living and working abroad, Vu Anh realized that language is a big barrier to bringing knowledge closer to Vietnamese people. While big language models like Google Translate and ChatGPT do a great job of translating in communication and conversation, Vu Anh realized that there is still a "niche market" with a lot of room that the big guys have not "reached" yet: text translation.

"If you have a 20-page document or scientific research paper, you can send it to ChatGPT to translate. But the chatbot will only return the translations in written form. Very important parts such as charts, images, data tables, etc. will almost lose all formatting. If you have to sit down and redesign those tables, it will take a lot of time and effort, and there may even be errors. Although Google and OpenAI's large language models are powerful, they still cannot completely handle the problem. That is why Doctranslate.io was born, focusing on translating text, images, audio, and video," said Vu Anh.

Original image file ( left ) and translation file from Doctranslate.io

The special feature of the platform is that users can input a text file, many formats with charts, images. Then AI not only translates paragraphs but also translates images, keeping the original format, helping users to compare directly and use immediately, without spending more time redesigning. With a content-rich image file, Doctranslate.io even allows exporting to PowerPoint files so that users can customize every detail and use it for many different purposes.

In a scientific study published at the IEEE-RIVF (International Conference on Computing and Communication Technologies) conference and IEEE - the world's largest professional technical organization), Doctranslate.io's ALMA-Gemma-7B-IT-ST model scored higher than Google Translate, OpenAI's GPT-3.5-Turbo... in the English to Vietnamese translation test.

Specifically, while GPT-3.5 Turbo scored 39.3 points and 34.1 points for English-Vietnamese and Vietnamese-English translations, Doctranslate.io's AI outperformed with 56.21 points and 57.32 points. Google Translate scored 39.86 points and 35.76 points.

Vietnamese startup's own formula

In addition to providing niche products that have not been conquered by large AI models, one of the reasons why Doctranslate.io is superior to ChatGPT, Google Translate is its ability to deeply understand Vietnamese.

According to Vu Anh, Vietnamese has many characteristics compared to English and other languages. If you do not understand them clearly, you will not be able to improve AI's translation ability. One of the simplest examples is the first person. "We have many nouns and pronouns to indicate relationships and nuances in different contexts. For example, when translating stories, saying 'Mr. Chi Pheo' and 'that Chi Pheo' are very different. This is something that large AI models, if they do not have a deep understanding of Vietnamese and are not fine-tuned, will not be able to improve their accuracy," said the founder of Doctranslate.io.

Doctranslate.io translation platform interface

During the implementation process, Vu Anh's team also realized that the demand for translation of documents in the medical industry, documents of multinational companies, banks, translation of meeting content, translation of stories... is very large. However, these organizations are not always ready to use large free language models such as those of Google, OpenAI in the system. Some institutions need "offline" models that can be easily integrated into the system, bringing good work efficiency but more importantly, ensuring data privacy.

Understanding that need, Doctranslate.io has developed both a translation platform for ordinary users and a customized version that can be easily integrated into the systems of digital transformation enterprises that require privacy and security. Currently, users can use the free trial version of Doctranslate.io to translate text, images, videos, and conversations. Meanwhile, businesses can use the cloud version or API of Doctranslate.io. More advanced, Doctranslate.io also provides an on-premises version, which is highly customizable and the data is processed right on the company's system.

After more than a year on the market, Doctranslate.io has more than 200,000 active users every month and is integrated into many business systems. The Vietnamese AI startup plans to continue expanding its market to countries in the Southeast Asian region, then globally. "With Doctranslate.io, we are gradually realizing the goal of bringing knowledge closer to everyone and affirming the position of Vietnamese intelligence on the global technology map," Tran Vu Anh shared.

Source: https://thanhnien.vn/ai-dich-thuat-cua-startup-viet-tot-hon-ca-google-chatgpt-185250115142214607.htm