Your knowledge hub on nearshore software development | Pwrteams

AI Translation in e-Commerce – Best Cases

Written by Admin | August 16, 2019

You’ve probably heard tales of artificial intelligence taking over our jobs. And while some applications sound like a sci-fi fantasy, other cases prove that AI can work more efficiently than people. Take translation. The first computer translations were so mediocre that no one took them seriously. Now that neural networks are in the game, AI translations are almost as good as human translations. Businesses can now turn to AI to quickly translate massive volumes of text at a pace no human could compete with. Just look at how AI translation in e-commerce helped world-leading retailers boost their sales.

Why AI is so Valuable for e-Commerce

Data-driven programming, machine learning, and AI are more than just buzzwords for e-commerce. These technological advancements can select recommended goods for existing customers, target potential buyers, personalize, and enhance the chatbot experience.

AI learns patterns directly from the provided samples and doesn’t need human intervention to tell what products the customer will be most likely to buy next week. Security and fraud detection can also be enhanced thanks to AI. Identifying fake reviews, detecting forgery items, customizing CRMs – this is what artificial intelligence can do for your customers.

Using artificial intelligence for translating goods is a great way to make them available to a broader audience. In fact, 72.4% of people are more likely to purchase a product if the description is in their native language.  56.2% of internet users, seeing a product description in their language is more important than price. This is where AI translation comes to the rescue.

AI translation means using neural machine translation and data-driven algorithms to translate the source language into the target language. Today, AI translation is already enhancing many industries, but it’s especially important for large e-commerce retailers that have thousands of customers all over the globe. Language barriers are a critical factor of the global trade failure, so translation with the help of AI should be a necessary part of the modern approach to business operations. But why aren’t all retailers automating their translations with the help of AI yet?

Challenges of Implementing AI Translation in e-Commerce

Humans have background knowledge, intuition and flexibility no machine can afford. That poses challenges for AI translation in e-commerce human translators wouldn’t have to deal with. These are different title character limits for different retail channels. Others include determining if a word is a brand name (“apple”) that doesn’t require translation or a product name that has to be translated. Not to mention that the same brand can have different names in different countries (“Axe” in the USA and “Lynx” in Australia).

But the biggest concern is probably translation quality. Translation has to:

  • be accurate (especially the brand name, color and sizing)
  • be smooth and readable
  • block violent and obscene content
  • fit accurately into various interfaces and screen sizes

Alibaba, the world-renowned Asian e-commerce conglomerate, has faced many of these challenges, but they’ve found ways to handle them.

How Alibaba Tackles the Challenges of Machine Translation

Being one of the largest retailers and e-commerce companies in the world, Alibaba faced the need to translate more than 400 million goods. After all, 70% of their buyers speak or understand English, while 30% use other languages for communication. What’s more, nearly 96% of sellers don’t speak other languages at all.

To cope with the challenge, Alibaba employed several strategies and approaches to Machine Translation:

  • Rule-based Machine Translation (RBMT). It is excellent for translating numbers, dates, addresses, and commodity-related information when configured with simple rules and up-to-date dictionaries.
  • Statistical Machine Translation (SMT). It shows excellent results in translating product titles that consist of several independent phrases with no particular order or logic between them.
  • Neural Machine Translation (NMT). It uses the power of AI and Recurrent Neural Network to translate product descriptions, communication messages, offers, and comments of up to 30 words. The newest NN model called the Transformer is used alongside the older RNN seq-seq model to compare results. Usually showing better performance, Transformer still lacks smoothness and logic in translated sentences. Nevertheless, it’s the forefront of AI Machine Translation, where good performance with smooth translations is a matter of time.

To improve the quality of translations, Alibaba made an MT lab under the DAMO Academy to involve engineers and researchers in developing Machine Translation innovations.

Here are some of them:
  • Neural Inflection Prediction is crucial when translating from Chinese to English or Russian. Because Chinese characters don’t have singular or plural forms, they have to be predicted during translation.
  • Translation Intervention ensures that the MT system translates the key information to the target language accurately.
  • Distributed Training with Model Average is about the actual training of the model. Alibaba’s training corpora reached a billion sentences, so using multiple GPUs and averaging results is necessary to increase the speed of training.
  • Multi-Modal Translation is the forefront of Machine Translation. Being able to translate voice and text on images is necessary to unveil the real power of AI.

Alibaba has shown that employing machine learning and big data in translation can make more people around the world your clients. But eBay was the one to prove it.

Using AI Translation in e-Commerce: eBay’s Example

The best example of using AI translation in e-commerce is probably eBay’s case. The way eBay’s trade changed after the utilization of AI-driven translation techniques is astounding.

According to the eBay study conducted by researchers from MIT, NBER and Washington University, AI translation for non-English speaking users increased exports by 17.5% and revenue by 13.1%. Those were buyers from countries in Latin America, Asia and Europe. It was also estimated that the quality of translation with the help of AI increased by 10%.

The research claims that differentiated (categorized) products, cheaper goods and products that had more words in their title were responsible for increased exports. Also, AI translation enticed inexperienced buyers to purchase more goods.

eMT, eBay’s machine translation system, can translate from various languages as its algorithms were trained on eBay’s internal data. eBay Machine Translation (eMT) helped the retailer improve trade and economic activities by translating the product listing titles on eBay. This unique system allowed optimizing the process of translation and delivered first-class translations in milliseconds.

Conclusion

Bringing new automated translation approaches into business is an excellent way for companies to increase revenue and cut costs on manual translation. This is especially viable for e-commerce businesses that have thousands of customers all over the world.

AI translation can increase revenue by breaking down language barriers between foreign countries and prompting export between them. And while the cost of teaching a neural network automated translation might exceed some budgets, examples like eBay show that it’s a valuable investment.

Read also: