AI Translation Firm Unveils ‘world-first’ Timeline toSingularity

A new approach to gauging AI development has been unveiled by an Italian business: examining advancements in machine translation.

The approach was used by Translated, a company that offers translation services, to forecast when we will reach singularity, a vague concept that is frequently described as the point at which machines surpass human intelligence.

When AI offers “a perfect translation,” the Rome-based company reaches this milestone. This occurs when machine translation (MT) is superior to the best human translations, according to the new research.

“[It will be] within this decade, at least for the top 10 languages in a context of average complexity,”

“The reality is that in some specific domains and in a few languages this has already happened. For some rare languages and domains it may never come.” Marco Trombetti, the company’s CEO, tells TNW

The estimates provided by Translated are based on information obtained from Matecat, a CAT tool.

The platform got its start in 2011 as a research project funded by the EU. The system was made available as open-source software three years later, and experts now use it to enhance their translations.

Matecat is a freemium product available from Translated. Users give the business data that is used to enhance its models in exchange.

Translated monitored the amount of time users spent reviewing and revising 2 billion MT suggestions in order to map the road to singularity. Over the course of Matecat’s 12 years of operation, approximately 136,000 professionals from all over the world had made these edits. The translations covered a wide range of topics, including both technical and literary topics. They also included industries like speech transcription, where MT is still having difficulty.

“Singularity is really close.

The evidence suggests that AI is advancing quickly. Leading translators in the world checked and corrected MT suggestions on average once every 3.5 seconds in 2015. Today, that time per word is 2 seconds.

The time will reach one second at the current rate in about five years. When that happened, MT would offer the historic “perfect translation.” The translations made by a machine will then be easier to edit than those made by a top expert.

Trombetti asserts that any task requiring comprehension, listening, communication, and knowledge sharing can be made multilingual with little effort.

The precise time we will arrive at the singularity point may vary, but the trend is undeniable: we are getting very close, he claims.

When plotted graphically, Translated's TTE data shows a surprisingly linear trend
The “Time to Edit” metric assigns the quality evaluation to professional translators. Credit: Translated

“All our customers who are deploying machine translation on a large scale are also spending more on human translation,” says Trombetti.

“Machine translation is an enabler in that it creates more interactions between markets and users that were not in contact before. This generates business, and business generates higher-quality content that requires professionals.”

Trombetti also expect new roles to emerge for elite translators.

“To get the best quality out of machine translation you need it to be trained by the best linguists. A significant volume of translations is required to train language models and fix errors in them, so I guess it’s likely that we’ll witness huge competition for the best translators in the upcoming years.”

“MT is a good predictor of what’s next in AI.

The new study, according to Translated, is the first to ever quantify the rate at which the singularity is drawing near. Although not all skeptics will be won over, MT is a compelling barometer for AI development.

Machines have a notoriously difficult time learning human languages. Computers may struggle to understand the linguistic subjectivity, the dynamic nature of conventions, and the subtleties of cultural allusions, wordplay, and tone.

These complexities need to be modeled and connected in two languages during translation. As a result, the field frequently sets the standard for algorithmic research, data gathering, and model sizes. For instance, the Transformer model was used in MT for many years prior to being implemented in OpenAI’s GPT systems.

“MT is simply a good predictor of what is coming next in AI.” says Trombetti.



We will be happy to hear your thoughts

      Leave a reply

      Shopping cart