The Limitations of Machine Translation
Getting started
Machine translation has continued toadvance exponentially in line with the progress of AI, but its limitationsstill persist. What are some of these limitations? First of all, it can be saidthat interpreting omitted information is difficult. Unconsciously, we oftendon't articulate every component of a sentence perfectly when speaking.Sometimes we miss the subject, other times the object, and quite frequently, wefail to properly handle the end of a sentence. Shall we illustrate this with anexample of a conversation?
Drama <Marry My Husband> Episode 10
Yang Joo-ran: How could you let such anaccident happen? No, even if it was an accident, how could you cover it up?
Park Min-hwan: Huh? Uh, what...?
Yang Joo-ran: I heard there was an uproarat Jo Mart due to lack of allergy notices. How could you let it slide?
Park Min-hwan: How could...
Yang Joo-ran: Jo Mart didn't evenapologized... A mere employee's mistake?
These are the lines from the drama<Marry My Husband>. As you can see, there are sentences that do not endperfectly. Also, you can notice instances where the mentioned content isomitted. If we were to write these sentences as complete sentences, the contentwould look like this.
Yang Joo-ran: How could you let such anaccident happen? No, even if it was an accident, how could you cover it up?
Park Min-hwan: Huh? I'm not sure whataccident you're referring to.
Yang Joo-ran: I heard there was an uproarat Jo Mart due to lack of allergy notices. Why did you not report the accident?
Park Min-hwan: How did you know there wasan accident due to the missing allergy notice?
Yang Joo-ran: They didn't even apologize atJo Mart, they even raised their voice and said it was their employee's mistake.
두 번째로 쓴 대사가 문법적으로 옳은 문장이지만 이걸 영상으로 본다고 생각하니 오히려 더 어색하게 느껴지실 거 같아요. 완벽한 문장 성분을 갖춘 문장을 만들었을 때 오히려 더 어색하게 느껴지기도 한다는 게 ‘말'의 특징이에요. 우리는 실생활에서 이렇게 제대로 말하지 않기 때문이죠. 바로 이러한 부분에서 문서 번역과 영상 번역에는 차이가 생기게 되는데 이에 대해 조금 자세히 적어볼까 합니다.
The Difference Between Document Translationand Video Translation
In document translation, sentences oftenhave all their components perfectly structured. Therefore, there isn't as muchconcern about the limitations of machine translation as there used to be.Through ongoing research and development, machine translation systems canproduce fairly accurate interpretations for sentences with perfect structuralelements. However, the issue arises when translating videos, as the limitationsbecome apparent. In cases where sentence components are omitted, as in theexample provided earlier, there are instances where these omissions aren'ttranslated. In cases where the mentioned content is often omitted later,machine translation may fail to accurately translate this part.
How is LETR WORKS address this issue?
Subtitles character count settings areadjustable. Before uploading a project to LETR WORKS, you can set the charactercount in advance, allowing you to establish maximum character limits for easierediting during the subsequent process. This feature helps streamline theediting process, saving both time and effort.
The TB glossary function enables fastertranslation. By unifying repetitive terms in videos, subtitle quality isimproved, leading to more accurate translations.
Utilize Translation Memory (TM) andsentence recommendation features. As you may know, AI improves with learning,yielding better results over time. By learning from past translation projectsat, LETR WORKS, the system can recommend sentences tailored to specificsituations, enabling faster translation.
With increasing interest in the localization ofcontent these days, various fields such as webtoons and videos are attractingattention to translation. This suggests that the importance of colloquialtranslation is expected to grow. At LETR WORKS, we are constantly pondering andproviding solutions for the inevitable limitations of machine translation, soplease continue to show interest and keep an eye on us in the future.
Editor / Lee inhye
inhye.lee@twigfarm.net
