What You Need to Know About Machine Translation: An Introduction

Warren Weaver — one of the pioneers of machine translation (MT) — said: “No reasonable person thinks that a machine translation can ever achieve elegance and style. Pushkin need not shudder.”

And you find yourself nodding along and agreeing with Weaver. After all, you have had multiple fails with Google Translate. Remember that embarrassing gaffe caused by a mistranslation of your proposal?

But, wait… that’s only one side of the story. What about the other side?

For starters, bear in mind that technology has advanced at an unprecedented pace since Weaver’s day. As for your very reasonable reservation about Google Translate, well, get a dose of this: there’s so much more to MT than Google Translate.

But, first, let’s back up and check out what exactly is MT?

What is Machine Translation?

MT is the process where a fully-automated software translates text or content from one language to another. Depending on the software, the translation may or may not involve human intervention.

How Does Machine Translation Work?

Over the years, three types of MT system emerged:

• Rule-based MT (RbMT)

With this system, you typically need an expert familiar with the source and the target language to develop various linguistic rules (including syntactic, semantic, and morphological) for the translation.

• Statistical MT (SMT)

Here, the system ‘learns’ to translate by analyzing a large amount of previously-translated text. This was the system that Microsoft Translator utilized before it launched a neural network.

• Neural MT (NMT)

A newer approach, the neural MT learns to translate by using a large neural network. Google Translate used this system before it advanced to a novel hybrid model architecture in June this year.

Free Machine Translation vs Paid Machine Translation

Now, there’s no doubt that free MT like Google Translate is extremely tempting. It’s free, it’s fast, and it’s easily accessible online.

But, a free MT really isn’t the best gauge of what MT can do. Here’s why:

• It often makes mistakes especially when it comes to grammar and context.
• There’s no editor, proofreader, or quality control in place.
• More often than not, there’s no way of reporting errors in the translations.
• The quality typically depends on the language pairing as most free MT’s database is built from existing online translations. So, less common language pairings are often less accurate.

This is where paid MT comes in:

• A good MT-provider should have procedures in place including post-translation edit by a human translator.
• With the editing done by a human translator (often a native speaker), the quality of the translation will not have to rely on the popularity of the language pairing.
• A proper provider will usually have strict security protocols in place for confidentiality and privacy reasons.

The Bottom Line

Now that you know about MT, we have no doubt that you’d want to give it a whirl. If you’re ready to experience the true magic of a quality machine translation, get in touch now!