Machine learning models are often trained with US-based data. Artificial intelligence localization is an effort to create data sets to train models for the many other markets in the world.

Jessica Davis, Senior Editor

October 4, 2021

5 Min Read
Erick Nguyen via Alamy

Enterprise customers of companies like Microsoft and Google have come to expect their office productivity suites -- Office 365 and Google Docs -- to incorporate localization for the various markets where employees work and where they do business with customers. For instance, that means that Office 365 could be customized in preferences to use certain languages and currencies as a default, depending on where the user works. In the US it's dollars and in the UK it’s pounds.

That kind of localization technology has been around for decades and extends to many different enterprise applications and services.

But emerging technologies generally focus on a single market when they are developed. That means they aren’t initially equipped with this kind of localization technology. For instance, if a US-based company has launched an AI program, the AI models may do a terrific job of reflecting how things work in the US, but may very well fall short in markets abroad. Localization for AI has not yet arrived.

That’s not hugely surprising. Just a few years ago many enterprises were finding it challenging to implement AI into production at scale. The pandemic has accelerated adoption, however, and now some organizations may find themselves at the point of refining these systems.

More specifically, now that AI is finally taking hold in enterprises, these organizations may be looking to add localization to their models.

Such enhancements could be extremely important for particular industry verticals such as retailers, according to Bradley Shimmin, Omdia chief analyst for AI platforms, analytics, and data management.

“If I’m doing endcaps in a bookstore, what I place on that endcap is going to look radically different depending on whether that store is in Tennessee or Vermont,” he said. “A nationwide retailer would know those differences and put them into practice.”

International localization presents more of a challenge in terms of both language and culture.

Retail and other industry verticals may have been using language localization in the form of translation software for some time now. Shimmin notes that AI language libraries such as GTP-3 (generative pre-trained transformer 3) have provided developers with an “embarrassment of riches” when it comes to language tools.

However, localization is about more than just language. It’s about culture, too. Jonas Ryberg, chief globalization officer for Pactera EDGE, tells InformationWeek that his company is working on helping AI products to work in multiple markets.

“The majority of AI products are made for the US market,” he says. Pactera EDGE has been working with some tech giants to help them make their AI relevant for non-US markets, too. In the past couple years, the company has seen demand from the next tier of companies that have added AI to their technology stack. These organizations may lack the data sets they need to train models for new markets or markets where they are newly applying AI. Ryberg says that his company is seeing interest across the board in several industries, although retail is one of the first focused on by Pactera EDGE.

For instance, in the US if a customer puts a lamp in their cart, they will get a next recommendation based on American preferences. But the preferences of customers in the UK or European countries may be very different than American preferences.

“We would create data sets for those markets and companies,” Ryberg says. Creating those data sets may be more challenging depending on the regulations for each country, particularly around personal identifiable data. In certain jurisdictions it may be necessary to operate a local data center as well. Pactera EDGE is working to build those data sets. For instance, for a retail customer it may go out to a store and take photos of products on shelves to feed a computer vision model.

AI localization has come into play for AI-powered contract lifecycle management company Agiloft, too, according to Andy Wishart, the company’s chief product officer. Data sets to train the company’s models frequently come from existing contracts loaded into the system. By nature, these data sets will be specific for local languages and the laws and regulations of local jurisdictions.

“We can create custom models based on the training sets of a particular language and jurisdiction,” Wishart says.

Yet even though there are rumblings of this new trend in the market, it’s just a start, according to Shimmin, who compares the effort to create a formal understanding of many different cultures that anyone can use to the introduction of the CIA Fact Book, a publicly available book of facts on cultures around the world.

“This isn’t about how to overthrow regimes,” he says. “It’s a very detailed manual on how to understand other countries, the differences between them, what is happening in those countries at different points in time. AI localization is in the spirit of that Fact Book and how that information can be put into practice.”

Adding the AI element is the new part that is poised for growth in the years to come.

“I feel like we are just at the outset of this,” Shimmin says.

What to Read Next:

Data Engineers in High Demand, Winning High Salaries

Data Fabrics: What CIOs Need to Know

Gaining Control Over Data Decay

What CIOs Need to Know About Graph Database Technology

About the Author(s)

Jessica Davis

Senior Editor

Jessica Davis is a Senior Editor at InformationWeek. She covers enterprise IT leadership, careers, artificial intelligence, data and analytics, and enterprise software. She has spent a career covering the intersection of business and technology. Follow her on twitter: @jessicadavis.

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like

More Insights