Auto-dubbing, which uses artificial intelligence to translate content into different languages, is a technology on which the global entertainment industry has increasingly come to rely in finding audiences among the planet’s 7.2 billion people, speaking more than 7,000 languages in roughly 200 countries. Companies like Flawless, Deepdub and Papercup use different approaches to offload to computers much of the labor required to fill that distribution pipeline. Another company, Spherex, emphasizes cultural awareness and the need for heightened sensitivity in pursuit of hits that travel across borders.

Though the Los Altos-based Spherex is a data and technology company, co-founder and CEO Teresa Phillips emphasizes that conscientious localization demands a nuanced degree of social understanding of which machines are not yet capable of fulfilling on their own when it comes to language translation.

“Artificial intelligence (AI) and machine learning (ML) are highly anticipated answers to this problem, but neither has reached the point of replacing the human localization component,” Phillips writes in TechCrunch. By way of example, Phillips cites two examples from South Korea: the Netflix drama “Squid Game,” which generated controversy over mistranslations, and the Oscar-winning film “Parasite,” whose director, Bong Joon-ho, “sent detailed notes to his translation team before they began work.”

“Literal translation is incapable of catching 100 percent of the story’s linguistic, cultural or contextual nuance included in the script, inflection or action,” Phillips explains in TechCrunch, noting “AI companies themselves admit to these limitations, commonly referring to machine-based translations as ‘more like dictionaries than translators,’ and remind us that computers are only capable of doing what we teach them while stating they lack understanding.”

The English title to the first episode of “Squid Game” is “Red Light, Green Light,” a reference to the children’s game that is part of the plot. But the original Korean title translates as “The Day the Mugunghwa Bloomed,” evoking a familiar Korean symbol for new beginnings.

“How can we expect to train machines to recognize these differences and apply them autonomously when humans don’t make the connection and apply them themselves?” asks Phillips. Even non-verbal cues, such as facial expressions, can color a translation, Phillips points out. In the end, it all comes down to “good training data.”

“AI is still a work in progress as it relates to explainability, interpretability and algorithmic bias,” she concludes. “The idea that machines will self-train themselves is far-fetched given where the industry stands concerning executing AI/ML. For a content-heavy, creative industry like media and entertainment, context is everything.”

Related:
Every Movie and TV Show Could Soon Be Dubbed into any Language You Want, The Washington Post, 10/14/21
AI Might Help Edit the Next Generation of Blockbusters, The Verge, 9/21/21
The Rise of the Robo-Voices, The Wall Street Journal, 10/7/21