Language Translation: Machines vs. Humans

This is my third article in a series on operating an ecommerce business in multiple countries. In “Language Translation Tools for Ecommerce Sites,” I reviewed various translation platforms and services.  Then, in “5 Keys for Building a Global Ecommerce Site,” I addressed localization variables, beyond language, to ensure a site fits within consumers’ shopping habits and modes of buying.

In this third installment, I will again address language translation. I’ll explain how so-called “machine” translations cannot be solely relied upon — they require a human review for accuracy and readability.

To demonstrate, I will use three free services — Google Translate, Bing Translator, and SDL Free Translation — and a sentence from a product description for a watch on Amazon. I will translate this sentence from English to Japanese and then use the Japanese output to translate back to English. Ideally, the end result in English should match the original English input.

The sentence for the product description is:

This light-up face, water resistance, rugged stainless steel case and digital chronograph watch is truly a companion timepiece.

Google Translate

Using Google Translate, we get this translated Japanese text.

Translating from English to Japanese, in Google Translate.

Then, translating the Japanese text back to English, we get this result.

This write -up , water resistance , robust stainless steel case and a digital chronograph watch is really a companion clock

Translating back to English, from Japanese, in Google Translate.

The English output does not match the input. It is close, but it would need a human to correct it.

Bing Translator

Now let’s try Bing Translator, for English to Japanese.

Using Bing Translator to translate from English to Japanese.

Now, translating back to English, we get this.

Watch the face of the light, waterproof, rugged stainless steel case, digital Chronograph watch fellow, really

The translation from Japanese back to English in Bing Translator is difficult to understand.

This output in Bing Translator is worse than the result in Google Translate. It will require a human translator to fix it.

SDL Free Translation

And lastly, let’s see how SDL Free Translation performs. Here’s the result for English to Japanese.

SDL Free Translation’s result from English to Japanese.

Translating the Japanese back to English, we receive this.

This light, the face of the water resistance and durable stainless steel case and a digital chronograph watch is a friend of in the true sense of the clock.

Translating back to English, from the Japanese translation.

This one is better than Bing Translator. But still it needs corrected.

Necessity of Human Translators

The three automated services are enormously helpful for quick translations. But they are not perfect. A retailer entering a new overseas market should not rely on them solely, as the user experience for consumers in those markets would suffer. The retailer needs human translators to review and correct the automated output.

Here are three options for locating a human translator:

• 100 companies for translation jobs;
• Translate By Humans;
• Fast Translator.

Human translators typically charge based on the number of words translated. Hence, to save money, translate as much as possible with free automated services and then engage a human to review.

Human translators, in my experience, offer lower per-translated-word rates with higher volume: If an entire site is being translated, the per-translated-word rates should be less. Another option is to involve a customer or a partner — perhaps via a monetary incentive — to review the automated translation.

See “Language Translation for Ecommerce: Good and Bad Examples,” the final article in Gagan Mehra’s 4-part series on language translation.