AI Localization: What It Is, and How to Do It Right

By Pactera EDGE Editors
Diverse images

What exactly is AI localization, and what are the critical success factors for doing it right? These were among the topics that Pactera EDGE Chief Globalization Officer Jonas Ryberg discussed on November 18 in a webinar with research firm Nimdzi Insights. The webinar, “AI, Localization, and Lovable Experiences,” was based on a white paper published recently by Pactera EDGE and Nimdzi to identify the critical success factors for global organizations to create lovable experiences in every market. The wide-ranging webinar covered a lot of ground, including examples of companies applying AI localization. Here are some highlights of the points that Jonas made:

What Is AI Localization?

Jonas explained that AI localization is about localizing AI products, tools, and services to work in their respective markets. Localizing AI is about training AI with localized data – collecting and curating data sets that respond to cultures in different markets.

It’s important to note that AI localization goes beyond translation. Language translation is part of AI localization, but AI localization also includes localized experiences that people love no matter what country they live in – experiences that resonate based on their own cultures.

The Pactera EDGE/Nimdzi white paper defines AI localization even more fully: an ensemble of services centered around collecting and curating data in order to produce clean, and thoughtfully balanced AI training datasets that will reflect how locals think and interact with the world around them.

Ultimately, AI localization helps AI deliver memorable experiences. A memorable experience feels natural, personal, and made for you. This needs to happen on a global level, and this is where AI localization comes into play.

Why We Need AI Localization

Anyone who has traveled internationally knows that engaging with people in a culture different than yours often leads to misunderstandings based on cultural context, your heritage, or your background. Those misunderstandings go beyond language, too. For example, the meaning of a gesture can differ dramatically from one culture to the next.

The same holds true for AI-based products, services, and tools. We use AI to help us engage with people in different ways around the world, examples being voice recognition and computer vision. Those ways of engaging with people with AI change from culture to culture. So, we have to adapt AI accordingly.

AI Localization Is Essential for AI to Be Inclusive

There is a perception that AI is created by a small group of data scientists who represent a narrow set of markets. Therefore, AI is built for a small group of people. As Jonas noted, we really need to make sure that AI is accessible to as many people as possible through AI localization. And AI needs to be trained to address bias. We do that at the highest level by opening up AI to different languages. But it goes way beyond that. For example, a product that uses computer vision needs to be able to understand different types of characters in multiple markets. A search engine must adapt to the context of how people expect information to be presented to them in different markets.

A big part of adapting AI to be inclusive is curating and building data sets that address multiple types of users. Doing so is complicated, but it starts with addressing user personas who are more diverse than the generic user.

AI Localization in Action

Google is an example of a company that does AI localization well through its core search product.  If you use Google Search for the term football in the United States, Google produces results focused on the NFL.com, which makes sense for how Americans define football. But in the United Kingdom, the search brings results on World Cup Qualifiers. In Sweden, such a search results in a Wikipedia page with different types of football, which reflects a “Swedish-neutral way of doing things,” according to Jonas. This is an example of how AI localization is beyond language. Context needs to be localized as Google does with search results.

Spotify is another example of localizing an experience with AI. Spotify offers a global service by making itself available to 80 markets, and in some 36 new languages. Spotify uses AI to create customized music playlists based on your listening experience. Those playlists differ by global location. Spotify is another example of localizing content and an experience beyond translating language. If you want to present a list of products that will resonate with someone in Germany, that list should be different from a list you present to people in Sweden, India, or the United States. Localizing the product recommendation has nothing to do with language.

Success Factors for AI Localization

  • Design with people at the center. It is extremely important to design AI with humans at the center with a global perspective. Don’t let people be an afterthought. Don’t make AI localization an add-on.
  • Keep humans in the loop. It is extremely important to have humans in the loop through the entire design process for AI to be lovable. You need to have humans in the loop to provide the right data for each culture and each type of user.
  • Be inclusive. If you start with designing AI with people at the center, you have a good sense of who will be using the AI product. From there, you must ensure that the data sets you use to train AI also correspond to user groups. You cannot use a generic persona to meet the needs of those user groups.  You need data sets that do not have bias; that cover all ethnicities, genders, ages, communities, nationalities, and so on.  And that should be tracked and tested with metrics to ensure we do not introduce bias.
  • Unlock the power of partnerships. No single company can master AI localization by itself. This process requires an ecosystem of experts who possess the scale, reach, technology, and expertise – and, importantly, a partner who understands how to reduce and even eliminate bias in AI. For example, AI localization requires the right technology to collect and curate hundreds of thousands of data assets; and a framework to manage accuracy and quality, which is something language service providers have been doing for decades.

How to Test Data to Stop Bias

As noted, fighting bias through AI localization means, among other things:

  • Designing AI with humans at the center. Part of doing that means building personas that are more diverse and nuanced than generic personas businesses have become accustomed to using.
  • Relying on a globally diverse team of people who represent different cultures, languages, and domain expertise to train data that AI products use. The resources must possess domain expertise, too – for instance, specialty knowledge of industries such as healthcare.
  • Supporting people with the technology required to collect and curate data globally. For instance, it is far too expensive and impractical for people to collect hundreds of thousands of images around the world manually in order to train computer vision to be inclusive. People need a technology platform to support this process. Finding people who can collect hundreds of thousands of data assets with the help of technology is an enormous task. In many cases, data collection services need to be done properly respecting privacy.
  • Applying analytics to properly track the progress of your efforts.

These are among the reasons why businesses need an ecosystem of partners to do AI localization right.

The Business Case for AI Localization

Let’s say you are at a global organization, and you want to take your customer to the next level across multiple markets. How do you get started with a business case? Fortunately, there is a growing body of third-party data to help. For instance, McKinsey says that AI can create a $13 trillion uptick in the global economy by 2030. But for that value to be realized globally, AI needs to be localized to different markets.

As a practical matter, two ways to build a business case for the enterprise:

  • Focus on revenue generation – for instance, localizing product recommendations to each market means multiplying the revenue generated on a larger scale.
  • Lean into inclusivity. Businesses are under the microscope to be more accessible and inclusive. AI localization, when done right, makes a business more accessible and inclusive globally.

How Pactera EDGE Can Help

At Pactera EDGE, we provide AI localization services to clients around the world by combining our OneForma platform with globally crowdsourced resources who possess in-market subject matter expertise, mastery of 200+ languages, and insight into local forms of expressions, such as emoji on different social apps. We collect and localize billions of data units across a wide range of domains every year, enabling multilingual virtual personal assistants, locale-relevant search engines, enterprise-grade and machine translation engines, and automated image, text, and speech recognition solutions. 

We provide AI localization services to the world’s largest corporations. We understand our clients’ needs and expectations and offer mindful and secure custom AI data solutions at scale. 

To learn how we can help you, contact us. Download our white paper on AI localization here