The Battle of Translator vs. AI

The Battle of Translator vs. AI

How Does Honyaku Plus Stack Up?

The consultant looked concerned. “Is Google Translate a threat to Honyaku Plus?” I laughed out loud. I couldn’t help it. “Have you ever used Google Translate?” I asked.

It was 2012. We were looking over Honyaku Plus’s SWOT analysis. Google Translate had been released six years earlier. Yet, it was still the butt of many jokes among the general public, and even more in the translation industry. The output was gobbledygook. So, no, Google Translate was not a threat. But the consultant asked, “If not now, how about in the future?” It was an ominous question that would haunt me for years to come.

AI meets translation

Fast forward to 2019. The AI boom is gaining steam, and one of the most interesting developments is Neural Machine Translation, also called NMT. This type of machine translation is supposed to be a game changer. I see multiple advertisements for an NMT system in Japan. XYZ Corporation (not their real name) in Osaka is offering this service, and quite cheaply I might add. Could this be the end of Honyaku Plus?

I signed up for a trial with XYZ Corporation, and I arranged a test. For this test, I used documents that Honyaku Plus had previously translated, so I had a human translation to compare with the machine translation.

It may not surprise you to know that the Japanese and English language pair is one of the most challenging in the translation world. For example, the proper use of “a” and “the” is often difficult even for human translators.

Imagine my shock when I discovered that the machine had in fact correctly used “a” and “the.” Now I was worried.

Putting NMT to the test

After getting over my initial shock, I dug deeper into the machine translation results and found many errors. For example, the words “he” and “she” alternated even though the subject had not changed: “He researched the topic for three years, then she published a report.” I also noticed that the longer the sentence, the more chance of translation errors. In some cases, the subject and object were reversed, producing sentences like “The fish ate man,” rather than “The man ate fish.”

The same test on other machine translation systems produced different vocabulary and sentence structures, but the same low level of quality. Trying other fields did not change the outcomes. I had the machine translate a document in the field of law and another in the field of machine tools, but they produced similar results.

In the end, all the machine translations required human intervention to make the translation useable. A professional translator needed to read the original Japanese text and analyze the English translation to determine what needed to be corrected and then make the correction. That is a lot of work.

A machine that turns out errors is not just broken, it is dangerous.

The verdict is in

In any kind of study, the methods used must be examined, and the results should be verified or validated by someone else. Fortunately, I found a report that did just that. It was a report on the use of NMT produced by CSA Research. The results were revealing.

• Only 50% of the translation companies included in the report (the bigger ones) have tried NMT

• Less than 30% actually use it for real work

If machine translation is so great, why are so few companies using it? The report also neatly answered that question:

Most [language service] providers don’t trust it sufficiently, feel it hurts more than it helps on jobs, . . . [and clients] don’t want it used on their projects.”

Things happen for a reason

One might ask, “Is there any real use for NMT?” The answer is “yes,” but it is complicated. Machine translation is the result of two forces: One is the desire of scientists to make machines that are smarter than people, and the other is the desire to translate mountains of text that no one wants to pay for.

Scientists have been dreaming about AI for at least 100 years, and the concept of machines doing all our work goes back to ancient times. No surprise there.

More recently however, we have a vast collection of text that needs to be translated. An example is the voluminous end user agreements that change almost weekly, and another is the highly repetitive content in online shopping sites. Companies like Google, Facebook, and Amazon are the main culprits, and the main drivers of this trend.

Honyaku Plus beats The Machine

We are translation experts that save you time and money by doing it right the first time. We deliver the best quality possible to maintain the professional standards of our clients. We use reliable, modern translation tools to achieve the best results, but at the moment, human translation is the gold standard for important translation projects.

Contact us with all your translation questions, even those about machine translation.

To read the full report, see “Are AI Deployments All They Are Cracked Up to Be?”

https://csa-reshttps://csa-research.com/Blog/ArticleID/549/LSP-technology-adoptionearch.com/Blog/ArticleID/549/LSP-technology-adoption