Our long-time readers know very well how ASTW has always a special attention towards machine translation. In fact, we have often dedicated a wide space to the matter within our bi-weekly articles (viewable at the link in orange). Therefore, we could not overlook Translated’s publication of a very interesting report concerning the singularity point of machine translation. A potential hub for the future of our industry.

Translated is an Italian language services company, based in Rome, founded by entrepreneur and IT expert Marco Trombetti. In 2014, Translated began developing its own machine translation software (CAT Tool) called Matecat. Benefiting from funding provided by the European Union and collaborating with important entities such as the Fondazione Bruno Kessler, the University of Edinburgh and the Université du Maine.

It is precisely using the computer-aided translation tool (Matecat) that the team of Translated came up with a whole series of data and variables to quantify the progress made by machine translation, measuring the speed at which we approach the singularity.


The results of this study were presented at the Association for Machine Translation in the Americas 2022. In order to carry out the research, the team first determined which variables would best determine quantity and quality of machine translation.

Over the past few years, the most widely established practice has been the use of the BLEU (Bilingual Evaluation Understudy) index, which, however, as mentioned in one of our previous articles:

“[…] gives a qualitative score to translation outputs. Although the benchmark allows comparison of the progress of different machine translation models, it does not offer an absolute measure of the software’s ability to achieve human quality.”

The Translated research group shares the same view, which believes that this benchmark “would not place the proper value on the translator’s cognitive effort.” For this purpose, a variable was introduced that could best represent the quality level of MT: Time to Edit (TTE).

The variable TTE refers to total time applied by the translator while controlling and editing pre-translated texts, divided by total word counting. Basically, it is measured the time a linguist takes to perform a translation in order to tangibly evaluate the necessary effort needed for a good-quality translation. 

Not for the first time, in order to assess the quality of a translation, the parameters of the so-called “human evaluation” have been used; in fact, such an initiative was proposed and undertaken in mid-2022 by Meta.


“We all realize that we are approaching the singularity in AI. For the first time, we have been able to quantify how fast we are approaching it.”

The CEO of Translated commented, with these words, about the chart concerning breakthrough on machine translation singularity based on artificial intelligence. That is, the time when machine translation will equal the output achieved by human translators.

Going further with Trombetti’s statement, “The TTE data show a surprisingly linear trend. If this trend continues to decline at the same rate as in previous years, we expect to reach a point in the near future, where MT will provide what could be called a perfect translation”.

Understanding “a perfect translation” as a machine translation output that takes the same amount of time for a linguist to check a translation done by their colleagues.

Nevertheless, this should not frighten human translators.

The singularity of machine translation should not be interpreted as the point at which the human translator will have to step aside. Instead, it is seen as a milestone to aim for in order to be able to increase one’s productivity, as well as the quality of the final product. We should remember that any translation will always need the intelligence, creativity and choices of human beings.

What do you think? Do you believe it is possible to reach the singularity in the near future? Let us know in the comments of our social channels!

Image by Serpstat from Pexels