In 2017, Amazon launched Amazon Translate, which can translate up to 71 languages and 4,970 language combinations – which is pretty impressive – but the big question on every Amazon Seller’s mind is if it can really translate a product listing well. Effective and accurate translations are crucial for truly connecting with a global audience, so it makes sense that lots of sellers have gotten curious enough about Amazon Translate to try it out themselves. In this week’s blog, we go deep into the translation engine of Amazon, and see if it’s a good solution for translating Amazon listings. We go into how it works, its pricing, and how well it does real-time translation work. Then we compare this neural machine translation service to human translations, and analyze which one would translate your Amazon product listing better.
Shall we get started?
If you’re interested in knowing more about our research into AI and translations, we touch on it a bit in our blog on YLT’s 5th birthday – Jana talks about predictions for the future, and AI most definitely is not exempt.
Introduction to Natural Language Processing, AI, and Amazon Translate
Amazon Translate is a cloud-based neural machine translation service that can deliver multilingual, natural-sounding translations. It’s powered by Amazon Web Services, and has more than 70 supported languages, including English, Spanish, French, German, Italian, Afrikaans, Korean, Catalan, Russian, Portuguese, and many more. The machine translation service uses deep learning, a type of machine learning based on neural networks. In other words, Amazon Translate is on a constant quest to improve its output to sound more natural. The design of artificial neural networks, like Amazon Translate, draws inspiration from the human brain’s learning process, using a natural language processing approach that reflects our personal experiences. Pretty soon, the output from Amazon Translate, and other machine translation services, is going to sound very similar to how a human being would communicate.
Mind you, “very similar” is key. Let’s not forget that human communication grows and changes; there are many words in different languages that we don’t use anymore because they’re outdated. Put it this way – next time you’re out with your friends and they start arguing about politics, religion, or other taboo topics, see what happens when you tell them to “cease their brabble.” They’ll probably look at you strangely. Language grows and changes with the times, and it remains to be seen if Amazon Translate – and other machine translation software – will roll with the times just as quickly.
Nevertheless, Amazon Translate can handle large volumes of translation requests. Amazon Translate allows for on-demand, real-time and batch translation. So, if you need to translate one massive block of data, Amazon Translate claims its deep learning models allows it to translate the encoded test efficiently, to emulate how your customer base would sound.
Furthermore, Amazon Translate can integrate the output into various applications using APIs by Amazon. It can process Word documents, PowerPoint presentations, Excel tables, and more. Amazon Translate also allows the user to create custom terminology bases with specific words that shouldn’t be translated, such as brand names and product names. Amazon Translate offers APIs in Java, JavaScript, PHP, .NET, Ruby, and Python.
How Much Does It Cost to Use Amazon Translate?
Google Translate is free to use – but Amazon Translate is not. Unless you only have 2 million characters to translate per month; then you can do that free of charge. After that quota, it charges $15 per million characters. Furthermore, Amazon Translate charges $30 per million characters for translation of .docx files, and $60 per million characters for active custom translation.
What does this mean for the Amazon Seller? Amazon Translate is good if you need a ton of data translated at the drop of a pin – but let’s not forget that there might be hidden costs. We’ll get into the limitations of Amazon Translate and other machine translation services later on in this blog, but suffice it to say that machine translations, no matter how smart, still can’t communicate the way human beings do, especially not when it comes to idiomatic expressions or humor.
Many people claim that Amazon Translate can localize text for you from one source language to another, smoothly translating common sayings, slang, and jokes. As of the moment, this is simply not true. Many machine translation apps have made big errors that resulted in bigger problems. In 2020, Facebook’s translation tools made pretty huge mistakes. In Burma, a post about Chinese President Xi Jinping’s visit was inappropriately translated; Xi Jinping’s name, on Aung San Suu Kyi’s official Facebook page, was mistranslated from Burmese to English – and the translation was derogatory. Facebook swears it was a technical glitch, since the system didn’t have President Xi’s name in its Burmese database, leading it to incorrectly guess the translation. Tests of similar words starting with “xi” and “shi” showed the same problem. Facebook made many similar gaffes – one that erroneously translated the Thai Public Broadcasting Service’s caption of a live stream about the King’s birthday candle-lighting ceremony. This error led to quite the strong reaction from the Thai public.
Amazon is also guilty of these errors. In 2020, Amazon Sweden became a laughing stock with many poorly translated product names and descriptions. For example, a hand-knit chicken was translated as “kuk” (another word for a rooster), and a die-cast cable was labeled “död” (deadly).
In short, whereas it looks quite affordable to use Amazon Translate to tailor a source text into many other languages, it simply can’t get the job 100% right – just yet. This is where those hidden costs come into play; you’d have to factor in a human translator’s time and effort spent to edit and proofread the output. Oftentimes, it takes even longer to proofread the machine translated output than it takes to translate and localize the data without the machines.
So, how much time do you really have?
How to Use the Amazon Translate App
Amazon Translate is a part of Amazon Web Services (AWS), a cloud computing platform established in 2006. AWS offers various IT resources like servers, storage, and databases via the cloud, allowing companies to rent these services without needing costly on-site hardware and software.
Amazon Translate enables users to translate unstructured text documents and is customizable to fit specific business needs. To access Amazon Translate, you need to have an AWS account. Although this might be a drawback for those not interested in using AWS, a demo version of Amazon Translate is available. Also, Amazon offers a comprehensive tutorial on the website, which can bring you through a step-by-step process in using Amazon Translate.
The process to use Amazon Translate involves navigating to the Amazon Translate console in the AWS Management Console, where users can translate text by choosing source and target languages, with the option for the service to auto-detect the source language. The console displays both the source text and the translated text, as well as JSON input and output for code debugging purposes when using AWS CLI or AWS SDK. Amazon Translate also offers a free tier, allowing customers to test the service before committing to it.
Sure, many features of Amazon Translate are handy, and will probably come into good use in the journey of an Amazon Seller, but if you really want to simplify your translation and localization journey, don’t rely on machine translations 100%.
What Are the Limitations of Amazon Translate and Other Machine Translation Services?
Machine translations like Amazon Translate have limitations. They often miss the subtleties of language, such as cultural nuances, idiomatic expressions, and industry-specific jargon, which are critical in the e-commerce space.
Machine translations, despite their advancements, have several limitations:
- Context Understanding: Machine translation often struggles with understanding context, especially in languages with high levels of ambiguity. It may not accurately grasp the meaning of words or phrases that depend heavily on context.
- Cultural Nuances: Capturing cultural nuances and idioms is challenging for machine translation. It might miss the cultural context or connotations, leading to translations that are technically correct but culturally inappropriate or awkward.
- Complex Syntax and Grammar: Languages with complex grammar and syntax can be problematic. Machine translation systems might struggle with long, complicated sentences or languages that have intricate grammatical rules.
- Limited Language Pairs: While machine translation has improved significantly for common language pairs (like English-Spanish), it’s less effective for less common language pairs or for languages with limited digital resources.
- Literal Translations: Machine translations often provide a literal translation, which can be problematic in languages where meaning is heavily dependent on tone, inflection, or context.
- Figurative Language and Idioms: Translating idioms, metaphors, and other figurative language can be difficult, as these often don’t have direct equivalents in other languages.
- Homonyms and Polysemy: Words with multiple meanings (homonyms) or words that have different meanings in different contexts (polysemy) can be challenging for machine translation to interpret accurately.
- Specialized Jargon and Terminology: Industry-specific jargon, technical terms, or specialized language can be problematic, as these terms may not be in the system’s database.
- Consistency Issues: In large projects, maintaining consistent terminology and style across the entire text can be difficult with machine translations.
- Subtle Language Elements: Subtleties like humor, irony, or sarcasm are often lost or misinterpreted in machine translations.
- Quality Variability: The quality of machine translation can vary greatly depending on the languages involved, the complexity of the text, and the specific machine translation tool used.
- Confidentiality and Data Security: Using online machine translation tools for sensitive or confidential information can pose data security risks.
All of these reasons are precisely why localization is much better than mere translation. Sure, you could feed a ton of data into machine translations, and you might get something that’s usable, but the fear is that you’ll publish something that will inadvertently offend your target audience – much like what happened in Burma, Thailand, and Sweden, as we illustrated above. Language is a living, breathing thing, and culture plays a big role in communications.
Even though machine learning is getting increasingly smarter with every month that passes, it still has a lot of limitations that restricts it to a tool, and not a complete solution.
Why Human Translations Will Always Trump Amazon Translate
This is where human translations shine. YLT Translations understands the subtleties of language and culture, ensuring that your product listings and marketing materials resonate with your target audience. Our translators are not just language experts; they are e-commerce specialists, familiar with the intricacies of Amazon selling.
At YLT, we don’t just translate listings. In fact, our translators are native speakers, who understand cultural nuances and complexities. Many times, customers have wondered why their sales in the US and UK are stellar, but their product is bombing in Germany or Japan. Our translators might have insight into these phenomena, since they represent your customer base in their respective countries.
Furthermore, YLT’s team is fully Amazon-trained. They understand what it takes to convert and engage on Amazon. We assure you, even though Amazon Translate is a neural machine translation service with a ton of beneficial uses, and it’s getting smarter and smarter every second, it still doesn’t understand the human aspect of what it means to appeal to a target audience. Oftentimes, copywriters will fine-tune their words to accurately illustrate an image. Sure, you could feed your infographic or A+ copy into Amazon Translate and hope it’ll churn out exact results, word-for-word, but do those words really illustrate the module or secondary image in each language? How you describe your barbecue gloves in US English is very different from how your Polish customer would describe them. Chances are, there aren’t very many backyard barbecues come Superbowl season in Warsaw, so that’s the level of contextual translation and localization you’d have to take into consideration.
Keep in mind that Amazon Translate doesn’t do keyword research, and keywords are different in every language. Think of our famous case of a customer who copy-pasted her US listing for baby diapers to her UK Seller Central – she was shocked when she started to rank for adult diapers for senior citizens. That’s because the Brits call them “nappies,” not “diapers.” It’s absolutely essential to conduct new keyword research per marketplace instead of just feeding the keywords through Amazon Translate, Google Translate, etc., and expecting the results to be the same.
Last but not least, to quote Jana: “A machine can’t log into a tool and pull out a list of keywords, and brainstorm what you’d want here or there,” Jana points out. “It’s still not good enough. It’s getting better, but it’s still not a replacement for the human brain.” It’s a machine. Until that machine starts to talk to you and helps you brainstorm how to present your product in Sweden, Brazil, Mexico, and Japan, you’ll want a human being with knowhow in localization and Amazon to bounce ideas with.
Alexa, sell my garlic press in Japan.
That’ll be the day.
Would It Be Practical to Get a Subscription to Machine Learning Like Amazon Translate?
Let’s say you have a bunch of data to translate, where minor errors are not a problem. Then go right ahead; grab that subscription, especially if you don’t need to bring the output through a human being after you generate the translation. Think minutes of meetings, analyzing social media and news feeds, or searching for specific information in many different languages. It’s a tool, after all.
But when it gets down to the nitty-gritty of a proper Amazon listing that you hope will convert and engage your audience, veer away from Amazon Translate.
The bottom line is, go ahead and assemble a decent tool kit for your Amazon selling experience. AI can help you in a variety of ways, and there are a ton of tools out there that will make your life easier. But AI and machines are tools. They’re meant to boost the human experience, not to replace it.
The day it does, is the day we’ve teleported into a sci-fi TV series, worthy of the biggest brightest imaginations in Hollywood.
Use Cases: We Compare Human Localization Support to Apps Similar to Amazon Translate
We’ve got use cases of machine vs. human translation to show you. That’s right; even YLT Translations does research on AI, machine translations, Amazon Translate – the whole shebang. Our team brought a product listing through machine translations, and translated it themselves, at the same time. Then the output was analyzed side-by-side to gauge accuracy and how engaging the result was.
To nobody’s surprise, the output wasn’t accurate at all. There were many sayings that didn’t make any sense, and wouldn’t convert an audience at all. We’ve got copies of this research to show you; email us at info@ylt-translations.com and we’ll gladly share a copy.
Folks, at the end of the day, AI is a great thing. Besides, it’s not like we can do away with it; AI is proving to be a valuable tool in a bunch of different industries. So, go ahead, and stuff your tool kit with AI tools. We’re moving into a time where working smarter, not harder, is the name of the game. But don’t forget – they’re tools. They’re not supposed to do all the work. You can mobilize them to do the heavy lifting, so you can go spend time doing the things that are more important, like attending to customers and finding new and better products (though there are AI apps for that too).
Amazon Translate and its peers, while wonderful, will never take away from the human experience. They can help, but the human experience will always be the better – and safer – option.
Now, isn’t that a relief?
Grab the study now while it’s hot – in other words, before AI develops yet again in the next 15 minutes and we have to redo the research to accommodate the change! Email info@ylt-translations.com – it’s a fascinating – and quick! – read.