Regularly, I have to reach for a dictionary to enjoy fully the rich use of language used by Janan Ganesh.
Last Saturday (“Mourning the Expat Havens”, FT Weekend, July 24) I’d made a mental note to research tabula rasa when, wonderfully, not only was the same phrase used in the directly-adjoining article but the interpretation (“a clean slate”) was included.
Might this become a standard service?
Jerry Blackett
Solihull, West Midlands, UK
Where does your enterprise stand on the AI adoption curve? Take our AI survey to find out.
Videoconferencing company Zoom today announced that it acquired Kites GmbH (Karlsruhe Information Technology Solutions), a startup developer of AI-powered real-time language translation technologies. Terms of the deal weren’t made public. Zoom said that Kites’ team of 12 research scientists will remain in Karlsruhe, Germany, helping the Zoom’s engineering team build translation capabilities for Zoom users.
Kites is among Zoom’s first acquisitions following the company’s $1.75 billion-plus share sale earlier this year. In a filing in January with the U.S. Security and Exchange Commission, Zoom said it could use part of the capital for merger and acquisition activity. As of March, the company had $4.2 billion in cash, which CFO Kelly Stackelberg said in an interview with Yahoo Finance Live would be put toward “opportunities for acquisitions to augment our talent and our technology,” among other efforts.
In its first acquisition in May 2020, Zoom bought Keybase, a security startup focused on encrypted communications, for an undisclosed sum.
Kites was founded in 2015 by Alex Waibel and Sebastian Stüker, faculty members at the Karlsruhe Institute of Technology. Waibel previously started the language technology group at Facebook, which became a part of the social network’s applied machine learning division. Waibel is also the founder of C-STAR, an international consortium for speech translation research, for which he served as chairman from 1998 to 2000.
Kites’ platform was originally designed as a tool to facilitate dialogue among international academic teams, but its focus was later broadened to become a general-purpose, AI-driven translation framework.
A number of third-party tools already allow Zoom users to engage in multilingual conversations. There’s Ligmo, which supports around 80 languages and 100 language pairs in real time. Another popular plugin, Worldly, can understand and translate between 16 different languages.
But Kites claims to leverage “state of the art” technology and predictive AI — built in-house and running on the cloud or on-premises — to deliver leading translation accuracy with low latency. Transcripts and translated text appear in real time, before speakers complete their sentences, and self-correct if a better interpretation is identified after additional context.
Kites says that when it comes to recognition, its system has an error rate of about 5%, with about only one second of delay behind a person’s speech.
“Kites emerged with the mission of breaking down language barriers and making seamless cross-language interaction a reality of everyday life, and we have long admired Zoom for its ability to easily connect people across the world,” Waibel and Stüker said in a statement. “We know Zoom is the best partner for Kites to help advance our mission and we are excited to see what comes next under Zoom’s incredible innovation engine.”
Following the acquisition, Waibel will become a Zoom research fellow, a role in which he’ll advise Zoom’s machine translation R&D. As for Zoom, it will explore opening an R&D center in Germany, according to Velchamy Sankarlingam, Zoom president of product and engineering.
“We are continuously looking for new ways to deliver happiness to our users and improve meeting productivity, and machine translation solutions will be key in enhancing our platform for Zoom customers across the globe,” Sankarlingam said in a statement. “With our aligned missions to make collaboration frictionless — regardless of language, geographic location, or other barriers – we are confident Kites’ impressive team will fit right in with Zoom.”
The administrator of your personal data will be Threatpost, Inc., 500 Unicorn Park, Woburn, MA 01801. Detailed information on the processing of personal data can be found in the privacy policy. In addition, you will find them in the message confirming the subscription to the newsletter.
In the wake of the pandemic, as profit margins ebb and flow, many businesses find themselves forced to reconsider the way they communicate and engage with their customers.
Take airlines. EasyJet bookings saw a 600 percent increase after Britain announced the lift of its lockdown. Yet business travel, the bread and butter of airline profits, isn’t expected to come back until 2025, according to The Global Business Travel Association. Low-paying consumers are flooding in, but profits continue to suffer. Moreover, new customers have new customer service needs. Older customers want information about updated safety protocols and services.
Other industries face similar uncertainties. Movie theaters were largely abandoned during COVID-19. It remains unclear whether consumers now prefer streaming and will continue to ignore theatres. Even handshakes might become obsolete.
All of this raises the question about the changing role of communication in customer service. How can organizations connect with customers, new and old, in an increasingly global world and as we ride the final waves of the pandemic?
While the temptation may be to slash prices to regain market share, there is something more powerful at play: trust.
Consumers who shy away from proximity and contact may well dictate the future of entire industries. Airlines, theaters, restaurants, retailers, and other impacted industries must think about recapturing market share with accurate language translations assisted by AI, to reflect a growing accessibility to customers around the world in their own preferred dialects.
Around the world, consumers increasingly expect a localized experience. This is as true of electronics providers like Panasonic as it is with massively multiplayer online gaming platforms (MMOGs) such as Wargaming. People could be visiting your online electronics shop from Portugal, Zambia, Brazil and Canada. If they don’t understand the language, or the localized version of that language doesn’t resonate with them, they have no reason to trust your credit-card submission form.
Likewise, players may be engaging with your MMOG from a similarly broad array of countries, but segregate their groups by language, because there is no good translation. Imagine the virtual economies to be built if anyone from anywhere could form alliances: the free players converting to paid players, the virtual goods changing hands. All of it is an exercise in trust, and trust comes with familiarity, and familiarity comes when we speak the same language.
The pandemic has changed assumptions. In my world of startups, for example, it used to be that fundraising had to happen in person. Nobody imagined that we’d be showcasing pitch decks via Zoom, but here we are. Zoom is widely accepted as being here to stay.
Many companies took the opportunity to focus solely on their product during the pandemic. Now that the products are dialed in, they find themselves needing to sell. Without services that communicate that product to people around the world, in a language they can understand, in a dialect with which they're familiar, all that product development will have limited applicability.
Companies have to make it easy for customers to trust them. Language is at the root. If you care about your brand, want to expose it to multiple markets, and provide a high level of customer service, you must translate. And that’s how language translation shines as the path to customer familiarity.
Among other things, trust is a function of repetition and dependability. What we see often grows familiar. Translation, long siloed in customer service, marketing, and other organizational teams, cannot provide the constancy needed to establish trust in many languages and dialects. Localization operates at human speed, and machine translation alone is too crude to hit the nuances that customers need to really familiarize and trust.
The entire technology stack should be harnessed for everyday communications across languages. This is evolving into an emerging discipline called Language Operations (LangOps). Data should be used to recommend the best language services for each need. A customer service app like Zendesk or Helpshift could “know” to address new customers with different services and options than older customers. Your MMOG players will receive communications in their own languages, be more likely to sign up for a paid version of the game, and build international teams of players. A Zambian visitor to your online electronics shop would receive the same levels of service as one from Canada, and be more likely to put in an order.
Such things were not possible before artificial intelligence evolved to its current level. With human refinement, organizations can create an entire translation layer for every market. As being global becomes more normal, language will be the tide that lifts all boats. It’s time to think of translation as less of a team, or group, and more as a technical operations layer that covers all things.
Vasco Pedro is a cofounder and chief executive officer of Unbabel, an AI-powered language operations platform that helps businesses deliver multilingual support at scale. A serial entrepreneur, Vasco has led Unbabel since 2013, taking it through Y Combinator and raising a total of $90 million in funding.
Fans of the life simulator (or vengeful deity simulator, depending on the player's mood) The Sims likely know about Simlish, the nonsensical-sounding language the people in-game speak. One voice actor, who worked on the second game as well as some other ones, spoke about his personal experience giving voices to these virtual people.
While the language of the Sims has a few classic phrases that originated in the first game and have stayed in the lexicon ever since (such as "Sul sul" for "Hello,") most of the language is entirely made up. Voice actor and singer Kid Beyond learned this fact when he signed up to help voice the people in The Sims 2 by working "hours + hours +hours" giving them voices for whatever mundane or shocking situations they may encounter. However, a key fact about Simlish is that it can't just be any old gibberish. It had to sound like American gibberish.
So. I was the male voice on The Sims 2, and a bunch of other Sims games.
I recorded hours + hours + hours of Simlish.
When I started the gig, they told me a lil’ secret:🤯 There is no Simlish. 🤯
🤖 — And no script. — 🤖The actors just make up gibberish.
But...(1/4) https://t.co/ag0nmw6ex8
— Kid Beyond — BLACK LIVES MATTER (@kidbeyond) June 28, 2021
So what was Kid Beyond's method for voicing Sims? He would take a magazine, flip it upside down, and read some "juicy" backwards words. Almost every language sounds bizarre when spoken backwards, and speaking English backwards provided that American gibberish feeling the studio was looking for. Kid Beyond jokes that after some of those four-hour sessions, it took a while to get his normal English back.
The fourth and current rendition of The Sims is available on PC, Xbox One, and PlayStation 4.
While there was an outpouring of grief and a whole book’s worth of reminisces by former Malacañang staff members in the days after the death of former Philippine President Benigno “Noynoy” Aquino III, the Mandaue City Public Information Office endured an embarrassing auto-translate gaffe on Facebook instead.
The original post, written in Cebuano, read: “Mandaue City nag half-mast agi ug pagbangutan sa kamatayon ni kanhi Presidente Noynoy Aquino.” (“Mandaue City [lowers the flag to] half-mast to mark the death of former President Noynoy Aquino.”)
Facebook’s auto-translate then gave this extremely unfortunate rendition: “Mandaue City is half-masturbating for the death of former President Noynoy Aquino.” (*cue shocked expression*)
(It doesn’t help that the still image from the accompanying video shows local government unit employees standing in front of a giant Pride banner, which has adorned the Mandaue City LGU building for Pride Month.)
Alerted by a commenter, the admin of the page changed the caption within an hour to avoid the mistranslation.
Since then, Facebook has issued an apology for the unfortunate gaffe.
_____________________
Also read: ‘YOLO’: Hong Kong police accidentally promotes drugs in anti-substance abuse campaign
_____________________
Facebook’s apology, which was also posted by the Mandaue City Public Information Office account, was addressed to Mandaue City information officer Atty. Eddu Ibañez, who had apparently written to complain about the mistranslation.
The Mandaue City PIO account went on to explain in the post that “the lapse was due to an isolated error by Facebook’s auto-translation technology. [The Facebook Head of Public Policy] added that they have made efforts to prevent the incident from happening again and has sent their deepest regrets on the unfortunate incident which happened on the occasion of the mourning of the death of a former president.”
The Mandaue PIO also said that they have forgiven the lapse. “Cognizant of the fact that technology remains imperfect, we chose to move on from this circumstance.”
This article, Facebook apologizes for ‘handy’ translation gaffe in Mandaue City post about PNoy’s death, originally appeared on Coconuts, Asia's leading alternative media company.
The Boyue P3 is a new digital pen that is designed for looking words up in the dictionary, by scanning text on an e-reader or print book. It also offers word search translation, voice translation, text translation, AI assistant, vocabulary book, textbook learning, text excerpting, listening practice, digital recording, history record and system settings. This device is currently only available in China and for Chinese text, but later this year it will support a multitude of new languages, including English.
On the front of the P3 is a 2.98-inch high-definition display and a button to engage the scanning feature. There is also 2 other buttons designed for volume and turning the device off. volume. There is a speaker on the back of it, so this is where everything will be read aloud. It has WIFI internet access and charging the device will be done via the USB-C port and there is a small 1050mAh battery for three weeks of use.
According to the only review posted for this, “For word search and translation, OCR text recognition technology is used, which can recognize 80 words in one minute, and the recognition rate is as high as 99%, which is 20 times faster than the query speed of paper dictionaries. In terms of the scanning speed, it was also fast, almost all scanning, the results were obtained immediately, and the performance was still excellent.”
The P3 should be released in the Fall, price is unknown.
Michael Kozlowski has been writing about audiobooks and e-readers for the past twelve years. His articles have been picked up by major and local news sources and websites such as the CBC, CNET, Engadget, Huffington Post and the New York Times. He Lives in Vancouver, British Columbia, Canada.
