This AI web-based tool provides real-time AI translations in 82 languages and 40 dialects.

With Verbum, OneMeta AI aims to make language barriers a thing of the past

This AI web-based tool provides real-time AI translations in 82 languages and 40 dialects.

For years, artificial intelligence and technology researchers have been trying to solve one of the world’s most complicated problems: translations fast and accurate enough that they can make conversations possible between any two people in the world.

In January, OneMeta AI announced Verbuma web-based tool that provides real-time translations in 82 languages and 40 dialects. The announcement declared Verbum was “designed to revolutionize communication” and “solves millennia of cross-language confusion.”

Verbum can provide verbal translation regardless of what language the speaker is using, as well as closed captions and automatically translated chats. Verbum is built for in-person events as well as for phone calls, meaning it can show instantaneous translation on the devices of audience members. The app was announced at CES in Las Vegas with a key emphasis on why the program is much faster than anything else on the market.

“No digital tool or person-based service exists today that can come close to matching Verbum’s translation and/or transcription capabilities, especially when taken in concert,” said Saul Leal, the CEO of OneMeta AI, in a statement during the announcement. “In fact, it is not unusual for transcription service providers to take weeks to produce a foreign language transcription of a recorded conversation or discussion. By contrast, we do so in roughly a second.”

During a video call, Leal demonstrated the software to me in Spanish, English and Arabic. Verbum was able to translate in real-time in all of the languages with high accuracy.

For Leal, the ramifications of this technology are incredibly important. He thinks about its potential implications in courtrooms, where translators can cost large amounts of time and money for defendants or migrants trying to make their case to stay in the U.S. He thinks about parent-teacher conferences that might be improved for parents that don’t speak the same language as their teachers or hospitals that have to treat patients through a language barrier.

“We talk about inclusivity,” Leal says. “The number one factor of exclusivity is actually language.”

The culture of technology companies attempting to solve social and cultural problems runs deep within the U.S. business market. While most companies frame themselves as making a positive impact on the world, tech companies in particular have a history of positioning their marketing around tackling major problems and bringing people together—attempting to bring internet to remote areas, provide innovative health care or tackle climate change through big-swing ideas, for example. All of this has given rise to the idea of “social entrepreneurship” in the past few decades.

OneMeta AI’s origins for Verbum have a similarly social motivation. A case study into the potential application for Verbum within U.S. communities found that schools would be an important area of focus, Leal says. The team at OneMeta AI has talked to 43 schools and districts, he says, and out of those, 40 are extremely interested in the software.

Many international students, undocumented or not, might be fearful to provide their name if they get hurt and require emergency medical services, Leal points out. Some of these students don’t speak English, prompting hospitals and schools to scramble to find people who can help.

Another area where major problems can arise is in the court system, Leal says, compounding the financial challenges facing defendants who don’t speak English well.

“You have someone who is paying $150 for a ticket,” Leal says, pointing out that translators are only available if booked in advance. “And if there’s not a schedule stacked correctly, they’re paying another $600 for translation.”

Attempts have been made for a long time to address this problem.

Google Translate, for example, is an incredibly common tool—and it even allows speech-to-text translation that happens more or less in real time. The difference is that, while it’s an incredibly useful tool for any two people who speak different languages, it can only translate small chunks of text at a time. This dramatically slows down any conversation.

Verbum, by contrast, addresses this by providing a “transcript” of sorts for the listener to read.

"We talk about inclusivity. The number one factor of exclusivity is actually language."

The origin for Verbum actually came from OneMeta AI working on a similar program to Google Translate, Leal says. He and his team were working on a different project focused on language that had a latency period of 3 to 5 seconds. When they focused on speeding that period up through AI, he says, “everything changed.” The latency period became one-eighth of a second.

This works because the application is trained by feeding a huge amount of content into its machine learning program, Leal continues. Often, this means finding libraries and phonetic dictionaries of words and languages and then working with linguists to perfect the results.

Technologies like Verbum will likely open up the whole world for business, travel and general communication. And while it seems that might disincentivize people from learning a new language, linguists hope it will do the opposite.

Writing for The Conversation in 2017, linguist Michael Haugh argued that the nuances in translations would make it even more important that language interpretation lives on. While translations can help connect the world, understanding comes from people who can bridge the gaps in what the subtler uses of words are.

“While such technologies are an increasingly useful tool, they can no more replace the deep cross-cultural knowledge that comes with learning languages than the advent of calculators meant we no longer needed to learn math,” Haugh wrote.

Another tricky part of training Verbum is honing in on accents and dialects. In Spanish, for example, there are many variations within South America alone—not to mention the major differences from the Spanish spoken in Spain. In Argentina, some people pronounce the “ll” sound like a “ch.”

Verbum also learns by drawing from radio stations all over the world. These types of improvements can ensure accuracy, Leal says. The South Indian language of Tamil, for example, only had an accuracy of 85 percent through Verbum, so the team used the radio tactic to improve it. Within 24 hours, the accuracy was listed as between 90 and 95 percent.

As a result of these kinds of real-time results, Leal and his team can build things much faster than large companies like Google or Amazon. The company has even attracted Microsoft, which reached out to secure a spot on OneMeta AI’s board.

“We’re definitely ahead of the game in terms of accuracy and time,” Leal says. “It’s kind of genius.”

But the fleet of other programs attempting to address similar problems—projects by MIT, Microsoft and Taia, to name a few—ensures that the market will be a complicated one to navigate. All of these programs focus on a slightly different structure and niche, and Verbum is one of the first to really embrace real-time translations within conversations like video calls or in-person events.

​​“In terms of the competition, it will be very tough as we move forward. It will be a very crowded field,” Leal says. “Right now, we have the lead.”

Jack Dodson is a reporter and documentary filmmaker most recently based in Palestine-Israel from 2018-2022. He has reported for Vice, BBC, The Intercept, Middle East Eye, among many others. He has a master’s in investigative journalism and documentary from Columbia Journalism School and a bachelor’s degree from Elon University in rhetoric.