At eBay, machine translation technology has been steadily breaking down borders. In many regions around the world, from Russia to Mexico, eBay leverages machine translation on the eBay platform to enable people across language and country boundaries to connect with each other. While eBay has used third-party technologies to boost this effort in the past, the majority of their current tools were created in-house. In addition, they host research workshops focused on how machine translation can improve customer experiences.
Evgeny Matusov is eBay’s Senior Manager of Machine Translation Science. He notes that eBay’s machine translation tools benefit buyers and sellers alike.
“Machine translation at eBay is key in promoting cross-border trade,” he said. “Our technology helps overcome language barriers and lets buyers order items from foreign countries. Our buyers can search the site in their native language, but see inventory from far away. If the item they are looking for is sold in another country, they will see the item title and/or description in their native language. Obviously, this benefits sellers as well as buyers, as sellers get new customers from other countries.”
Enabling machine translation on eBay in countries like Russia, Mexico, Argentina, Chile, Brazil has resulted in significant increases in items sold from regions such as the U.S. and Europe. Within Europe, machine translation seamlessly allows buyers and sellers from Germany, France, Italy, Spain, and other countries to effectively speak the same language.
One of the challenges in Europe, of course, is to cover the multitude of countries and language pairs. Most of the research on machine translation for Europe so far has focused on language pairs involving English as a source or target language. But what if you want to translate between French and Dutch, or German and Polish? Most of the free on-line translation tools do this using two systems: translating to English first, and then translating from English to the target language. Machine translation errors are multiplied in this two-step process. At eBay, they train dedicated translation systems for such language pairs, producing high-quality results.
As they introduce new product pages and new browsing experiences on eBay, machine translation will also be applied to product reviews, making it easier for their customers to judge whether a certain product is a good choice for them based on input from people residing in other countries. Product descriptions, buying guides, and other information will all appear in the user’s native language.
eBay’s machine translation systems rely on state-of-the-art algorithms. They use statistical translation models, which they’ve specifically developed for the ecommerce domain, and eBay employs several researchers who are well-known in the machine translation community. eBay’s models are trained to deal with challenging, noisy user-generated content. They are intelligent about brand names and alphanumeric product specifications, and they even know how to translate eBay jargon, such as NIB (new in box) as well as other acronyms. In addition, these systems are very fast, and they have to be: eBay has to handle more than eight billion translation requests per day.
“As we work on new releases of our machine translation systems, we actively experiment with emerging deep learning algorithms,” said Matusov. “For instance, neural network architectures for machine translation allow us to make use of wider context and eBay’s structured metadata in the most effective way.”
All of this benefits eBay’s customers, Matusov emphasized, and that is the most important goal. “Machine translation can connect global customers, enabling on-demand translation of messages and other communications between sellers and buyers,” he said. “It helps them solve problems and have the best possible experiences on eBay.”
Also published on Medium.