“Machine translation is a very important technology for cross-border trade,” says eBay’s Hassan Sawaf.
eBay has developed its very own machine translation tools to help fuel expansion in Russia. The tools allow Russian buyers to get accurate translations of eBay seller listing and communications in real time, making working with eBay Russia much easier.
We recently asked Hassan Sawaf (pictured), eBay’s Senior Director of Machine Translation and Geo Expansion, to talk about the promise and challenges of machine translation. Sawaf is a data scientist with more than 20 years of experience and here’s what he had to say:
You can also read a TechCrunch story on the initiative here.
Rolling out in Russia. Last year, we began to use machine translation in the eBay environment to enable people across country boundaries to communicate and deal with each other. What we had before depended on third-party technology.
Now, in Russia, we are rolling out our own machine translation technology, which will eventually also reach other regions. We have real-time translation for queries, which can be written in English or Cyrillic. We can take Cyrillic input, translate it into English and instantly apply it to our huge global inventory, as expressed in English, for matches. This opens up new avenues for sellers around the world to reach Russian buyers.
Better technologies are enabling all of this. Better hardware, better databases and advancements in machine learning are helping us deliver more accurate machine translation. Often, machine translation is facilitated by capturing lots of parallel data on the Internet and building statistical models for translating words and phrases.
We’ve developed good ways to achieve translations with higher fidelity for the eBay user, learning from data we can find in the Internet, but also from the extensive data we have here at eBay.
For example, we can’t have a translation determining that something is “blue” when it is in fact “navy blue,” because navy might be the selling point for an item or product. We do and will continue to make extensive use of our own structured data to meet these challenges.
Educating users. We plan to educate users about how translation works and how it will help them. Users will want to know if there is a machine translating communications going back-and-forth between buyers and sellers, and education will help them have the highest level of trust in what we’re doing. Also, we will allow the users to grade the helpfulness of the translation, and learn to improve on the translation quality from this feedback.
Breaking down borders. We are going to do careful analytics as our machine translation strategy rolls out, and we think it will have a big market impact. In some countries, English is not a big barrier. For example, in Germany, where I was raised, the majority of people speak English as a second language.
In Russia, however, far fewer people speak English. As we’ve rolled out our machine translation capabilities there, and even before a lot of the education and outreach we plan to do, we’ve quickly increased the number of Russian users we see using these features by 50 percent. We’ve also seen a big increase in the number of relevant query results that people searching in Russian are receiving. There are a lot of people who need this technology.