LocWorld41 localizes for Silicon Valley

LocWorld41, held November 6-8,2019, took the theme of “Go Global, Be Global” to Silicon Valley.

In the keynote on November 7, local entrepreneur Vitaly M. Golomb offered a presentation looking inside the “Corporate Black Box.” Golomb was born in the Soviet Union and immigrated to the US at eight years old, and subsequently found himself “torn between a couple of different cultures” and speaking Russian at home.

His current company bridges “the world to Silicon Valley,” said Golomb, and his presentation focused on startups and the emerging world they’re addressing. Golomb noted that in 1983, Isaac Asimov predicted a future where “untrained millions find themselves helpless to do the jobs that most need doing,” an accurate prediction considering the amount of automation present in the current workforce. Golomb stated that nationalism is a reaction to this phenomenon; misplaced outrage over losing traditional jobs, not to waves of immigration but to tech.

Golomb discussed what startups actually are: “an experiment in search of a business model” that scales up quickly. Golomb broke down the “S curves” of funding and profitability, and explained that you need the right collaborators to scale startups into successful ventures. You need venture capitalists and other funding sources, savvy management, professionals such as lawyers who can put together a venture capital deal, universities from which to draw employees and so on.

LocWorld hosted an open house for the first time, drawing 124 Silicon Valley professionals to the exhibit hall and a few afternoon sessions. This cross-section of the conference focused among other things on content strategy, machine translation evaluation, human-centered design and going global as a startup localization manager.

A session the morning of November 8 covered one of Silicon Valley’s most prescient projects: “Robot Voices in Multimedia.” Gilbert Segura of Compass Languages discussed the challenges of text to speech. He mentioned Amazon Polly, Microsoft Azure which is “closely related to Speech Translator” and Google Compute Cloud. Google, he said, has “gotten all the people, they’ve gotten all the money,” and is currently offering 187 voices. With its neural network, Google found a way to consider various learning data — speech and vocal patterns, including accent.

Engineering text to speech and more requires some basic phonetic knowledge. For example, said Segura, “speech is more than just sounds put together,” it’s about how the sounds are put together.

Meaning isn’t just encapsulated in text; it’s also in the way you say something, so there needs to be a way to figure out what the meaning is. Thus, said Segura, “when you’re doing speech to text,” you need to encode a lot of information.