How Artificial Intelligence revolutionizes ecommerce translation
If you want to organize your ecommerce translations yourself and/or manually insert product descriptions in different languages, it’s absolutely fine … if you have a lot of patience and time on your hands.
However, depending on your inventory and how often it changes, you may be in for quite a lot of work, dealing with translations and retranslations on your own.
And here’s the deal: today you really don’t have to.
Good business solutions are often automated and translation tools for ecommerce purposes are no exception. Especially now that the development of Artificial Intelligence (AI) makes machines much more human than they used to be.
Ecommerce & machine translation
Artificial Intelligence denotes the simulation of intelligent behaviour in machines so that more and more complex tasks can be performed by machines with less human supervision.
Machine learning is a subset of AI: algorithms are used to break down data, learn from it and make predictions. This way, machines can be trained to perform specific tasks.
Let’s focus on international ecommerce. At present sellers have to do a great deal of manual work, matching product categories, attributes, sizes and so forth.
However, thanks to a big data analysis and AI, most of this product tagging can be carried out on a larger scale. A simple raw description will be enough for AI to accurately localize the whole content, map the right category, extract attributes or convert sizes.
At Webinterpret we’re focused on machine learning and machine translation. We’ve learned from experience their benefits. Previously we’d use humans to convert sizes from one country to another. This resulted in a lot of critical errors due to human mistakes despite the fact that we hire professional, well-trained translators.
After we taught the machine to convert sizes based on brands, we no longer had errors. Size conversion accuracy and quality grew by leaps and bounds.
According to Marek Kostykowski, the leader of the Product Department at Webinterpret:
Mixing human expertise with continuously teaching the machine how to accurately localize product listings is the key to delivering consistent quality.
Say cheers to AI & automation!
Automation is a big benefit for retailers who want to spend less time on managing their catalogue and more time on developing their business strategy and delivering great buyer experience.
Ecommerce automation takes brains
Automatic solutions may seem simple and work … well … automatically, but creating a good, reliable automatic solution is an extremely complex process.
Patrick Smarzynski, CEO at Webinterpret, has worked with ecommerce technology for over 15 years. In his interview on ecommerce trends, Patrick stresses that automation isn’t a straightforward task. Specific rules and exceptions must be recognized by machines.
For example, machines shouldn’t translate brand names. Otherwise, the translation process may produce hilarious effects when, for instance, the brand Apple gets translated into the name of a fruit in a given language.
The role of AI is to teach the tool to recognize this and prevent faulty translation.
To account for such specificities, we need a human brain that will teach the machine to perform the task correctly in an efficient, correct and accurate way. Then we can benefit from the machine’s speed and consistency.
So to have it all for optimal ecommerce translation, we need both machines AND humans.
According to Patrick,
It is humans that have knowledge and do the thinking, but it is machines that are fast and more consistent. Technology combined with the human touch can produce amazing results.
Further, the machine must identify the right word/s to produce an optimal listing translation. To do so, it’s not enough to simply recognize words and phrases and combine them.
At least not if the expected end result should include keywords that will sell the product.
Consider this. We can very often use more than one word to describe the same item, e.g. trousers and pants. And here’s the deal…
The word you decide to use may make a huge difference when it comes to your conversion rates.
Simply because in some contexts online buyers, searching for the same product, are more likely to use the word ‘trousers’. In other contexts, ‘pants’ may be the winner.
Ecommerce data analysis helps to establish which keywords will maximize the possibility of selling a given product.
Of utmost importance are titles: these should include searchable terms on a given platform. We use big data techniques to examine millions of titles to check how words/synonyms correlate with revenue.
According to our findings,
Selecting the right words can double the number of transactions.
So to recap…
If you think that literary descriptions of your product will guarantee successful sales, think again. Some buyers may flinch at lengthy, full-sentenced descriptions without bothering to read them! It truly is product attributes that are most essential.
Buyers are not looking for great product descriptions, but great product offers. Clearly listed attributes and right keywords are the core of optimal ecommerce translation.
Webinterpret’s AI-based translation solution
Webinterpret uses an AI based, natural language processing solution and autonomous machine-learning to deliver affordable, automated human-quality translation.
We fully embrace AI technology to minimize operational time and cost while providing our clients with a high-quality, domain-specialized localization process.
Using a plugin, we connect to your domestic store or marketplace account. We then extract your product categories, titles, descriptions and attributes.
The localization process starts with:
- analyzing the product to be localized
- understanding its structure
- extracting its localizable components
- inferring important attributes, such as its brand.
Then, the different components are localized according to the specific requirements posed by:
- the marketplace, e.g. length limitations
- the category of the product, e.g. accounting for shoe size conversion between EU and UK or specific vocabulary in electronics)
- sellers’ preferences.
This translation process, tailored to the specific requirements of the product, can be fully automated with Thot, our in-house translation software tool. Thot has pioneered the implementation of adaptive machine translation.
This innovative technology allows the translation system to learn from human feedback in a continuous, self-improving loop. Thanks to this solution, the correctness of our translations has got closer to 100% with no need for human intervention.
To deliver the most accurate translations, we summarize all product descriptions, extract all attributes and store product descriptions as structured content. We use standardized descriptions optimized for search engines
Further, we analyze your product data to establish context and understand your products in a way that standard machine translation cannot. For example, the way we’ll translate ‘face mask’ will depend on whether it belongs to sports accessories or health & beauty.
Our ecommerce optimized machine translation is different from:
- Standard machine translation: it’s built and trained specifically for ecommerce products and it understands the notion of products, attributes, categories and context in a way that standard machine translation cannot.
- Human translation: that is slow and expensive.
A big advantage of automatic translation is speed and efficiency. For instance, our machine translation system takes seconds to translate a huge amount of listings.
If your store offers 1,000 items, manual translation of all of your inventory can take approximately a month. To do this kind of job, Webinterpret will need … 1 day.
If you sell on international markets and need to change your inventory regularly, slow translation carried out by humans can cost you a loss in sales potential.
Finally, at Webinterpret we translate over 100,000 products every week. With every new translation our system becomes better at translating and optimizing your products.
Our ecommerce translation solution in a nutshell:
- It’s automated and integrated and plugs straight into your existing store.
- Thanks to ecommerce and category optimized libraries, our translations are consistent and standardized.
- We use a statistical machine translation engine, which is specifically optimized and trained for ecommerce products, attributes and categories. As a result, your products are searchable and indexed correctly.
- Our solution is perfectly integrated with the top ecommerce platforms, such as Magento, Shopify, Bigcommerce or Woocommerce.
- Our automated process enables us to offer our translation services for free as part of one of our free or paid plans.
In a nutshell
You may think that automatic translation isn’t a very sophisticated and reliable method compared to translation carried out by humans.
Yet if we take automatic translation solutions for ecommerce purposes, creating them requires a very specific ecommerce context. This is something that standard machine translation and most human translators don’t take into account.
The result of successful ecommerce translation are keywords and attributes that will help to find your product and sell it.
Thus, international ecommerce experts, analysts and developers are striving to create machine translation solutions that will help you multiply your revenue in the most efficient and hassle-free way.
A cross-border trade partner can help you optimize your translation processes and costs, thus helping you get the best return-on-investment.
Interested in finding out more? Book a FREE demo!
Further reading & other sources
New ecommerce trends 2017 & onwards: interview with P. Smarzynski
3 myths about automated translation for international ecommerce sellers
7 must-know facts about the automatic translation of listings for ecommerce sellers
How to turn knowledge into profits: data analysis in ecommerce