Recently, machine translation technology based on artificial intelligence is a field that is attracting attention around the world. As such, many leading companies and researchers, such as Google and MS, are intensively researching, and actually meaningful research results continue to be published. Thanks to this, the performance and quality of machine translation (Machine Translation) has evolved rapidly day by day, and the use of machine translation is expanding in various fields in the translation industry and field.
However, it is also undeniable that the quality of machine translation still has many shortcomings compared to that of professional translators. However, considering the situation where the use of machine translation is bound to increase due to realistic reasons such as cost and time savings, translation companies and translators in the user's position are bound to use it based on an accurate understanding of machine translation quality issues.
In this blog post, I'll talk about the “BLEU (Bilingual Evaluation Understudy) Score (BLEU) Score),” an indicator that measures actual machine translation performance, to help you understand this issue.
What is a BLEU (bilingual evaluation understudy) score?
Accurate evaluation and confirmation of translation quality is essential in order to make a good machine translator or to use the translator properly. Also, among the various methods for evaluating the performance and quality of actual translators, the “BLEU score” is the most widely used.
Simply put, the BLEU score is a quantitative index for measuring translation quality, which is measured by comparing the accuracy between machine-translated sentences and correct answers. namely The more similar the sentence translated by the machine translator to the correct sentence determined by the human, the higher the BLEU scoreYou will do it.
Actual measurement examples of BLEU scores
In general, the performance of machine translation Determined by the size of the training data and the level of the algorithm modelIt is said that it will be In particular, the most important thing is to secure a large amount of high-quality data for learning translators, and apparently, leading companies that have secured many users through long-term services have an advantage in this area.
Therefore, in order for Gicon Studio, a latecomer, to catch up with them, they had no choice but to differentiate themselves with a different approach. It is an improvement method through the algorithm mentioned earlier, and provides a professional translation solution with a translation model optimized for specific topics and domains. In fact, Translation Dictionary (Translation Dictionary), a customized database of technical terms and expressions in the field, played a decisive role in improving the professional translation quality of Gicon Studio services.
Zicon Studio better than Google
Also, in 2019, the Gicon Studio team underwent a verification and verification test (Verification & Validation) by the Korea Information and Communication Technology Association (TTA) Software Testing and Certification Institute to receive an objective evaluation of professional translation performance and quality. At the time, we compared the translation quality with Google Translator in 5 fields (law, finance, machinery, chemistry, and medicine) and scored higher scores than Google Translator in 4 fields (law, finance, machinery, and medicine). In particular, it recorded an excellent BLEU score (Google 0.2464, Gicon Studio 0.3277) in legal tests and was recognized for its technical prowess.
The meaning and use of BLEU scores from the user's point of view
At first glance, it's easy to think that the BLEU score is an important indicator for developers or researchers who need evaluation or feedback on research and development results. However, it can also be a good reference for translation companies and translators that use machine translation solutions in actual translation projects to determine which one is most suitable by comparing various options.
While most translators are designed for general use and are easy to use regardless of the subject, there are also models optimized for specific topics or fields. In other words, an accurate translation is possible when using the optimal one for the purpose, considering that even the same word or sentence may need to be translated with a completely different meaning depending on the field of expertise used.
Therefore, for high-quality translation results, it is necessary to use tailor-made solutions for the relevant field. Since there are various fields that require translation, such as patents, law, medicine, science, and technology, it is necessary to be wise to select and use the optimal one by referring to evaluation indicators such as BLEU scores in order to accurately translate documents in these fields.
*This content has been transferred from Gicon Studio to Letterworks.
