The 5 Pitfalls of Machine Translation

Machine translation (MT) has been around for decades, but in the last 10 years, global players such as Google, Microsoft, Baidu and SYSTRAN have been rushing to improve capabilities. Most recently, a new breakthrough in machine translation, called Neural MT, has been launched. While this platform shows promise in the long term, and CMI is constantly evaluating these new technologies, we believe MT still falls short of our clients’ requirements. For certain types of content, MT still presents too many pitfalls.

The pluses for MT are cost and quick turnaround time. The biggest negative is accuracy—still inadequate even with the latest platforms. Mistakes in accuracy can have significant adverse effects in business. This is especially true with creative content.

Yelena Makarczyk, our Director of Localization Services, has identified 5 pitfalls of machine translation.

  1. MT does not recognize shades of meaning or NUANCE. In some northern languages, there are multiple words for “snow”. MT will pick the literal one that may not reflect the proper ambiance.
  2. IRONY is critical in video or film translation. “Like hell I love her” may be picked up as “I love her very much”.
  3. Machines also don’t do CONTEXT. We came across a MT where “sit” was meant for a dog but was translated as “would you sit down, please”.
  4. Machines don’t RESEARCH. For example, an important scene might have amethysts, emeralds, etc. The machine will translate all these subtleties as precious stones. Our translators will identify the stones as intended by the filmmaker.
  5. TONALITY and GRAMMAR are critical to any content. MT will go to the most generic literal choice often not accounting for the integrity and syntax of the whole sentence. A comedy may be dull, a drama boring. When a translator who is an expert in these genres goes to work on 1,500 subtitles in a film, every line has texture, emotion and build. MT will never be able to pick up tonality.

In our experience, the best check against unintentional mistakes (because machines do not ‘intend”), is to have a translation retranslated back into the original and compare results. This is quite costly and in the end is easier addressed by not using MT to begin with...

In conclusion, we believe MT does have very legitimate applications, but these need to be carefully vetted in light of the criteria discussed in this blog post. So when would you consider using MT? When the translated version isn’t meant for a final audience. Internal documents? Intermediate communication? Personal use?

Please call us so we can help you with your dubbing, subtitling and captioning project today.

Learn how Hubspot Increased Engagement for its Global E-Learning Academy.