The dream of being able to automatically translate from one language to another has come true. But just how effective is machine translation? What are the requirements for adopting Neural Machine Translation?
In this post, we aim to de-mystify machine translation by providing you with the information you need to decide whether machine translation is the solution you are looking for.
What is Machine Translation?
Let’s start with some basic definitions. Machine translation, also referred to as automated translation, is the process of automatically transferring text from one language to another using one or a combination of the following methods:
- Rule-based machine translation (RBMT) – Uses language and grammar rules combined with specific dictionaries to generate translations.
- Statistical machine translation (SMT) – Generates translations using statistical models created from the analysis of large databases of multilingual content (bilingual text corpora)
- Neural machine translation (NMT) – This relative newcomer gained traction in 2017 and uses deep learning based on neural networks to learn linguistic rules from statistical models, resulting in faster and better translations.
If you want to know more about the evolution of machine translation, take a look at this informative article from TextMaster.
Check out the infographic we have prepared for you here below. If you like it, copy and paste this code to embed it on your site:
<blockquote class="embedly-card"><h4><a href="https://blog.acolad.com/hs-fs/hubfs/Final%20-%20social%20(003).png?width=1650&name=Final%20-%20social%20(003).png">null</a></h4><p>null</p></blockquote><script async src="//cdn.embedly.com/widgets/platform.js" charset="UTF-8"></script>
What level of quality can Neural Machine Translation achieve?
The quality of machine translation output will depend on various factors:
- the configuration of the machine translation engine
- the quality of the source text
- the domain of the source text
- the language combination
- the amount of previously translated material available
Because of these limitations, it is practically impossible for raw machine translation output to be used for any purpose that goes beyond that of “gisting” (trying to grasp the overall meaning of a text).
As a result, in recent years, the language service industry has adapted with the introduction of integrated processes for machine translation that combine automatic translation with human post-editing of machine translated text. In fact, this skill is becoming increasingly requested.
Neural Machine Translation is not the same as Google Translate
A very common misconception regarding machine translation solutions is to think that they are the same thing as Google Translate.
The main difference is that professional companies providing translation services, such as Acolad, provide customized translation engines that are trained and optimized to meet the specific requirements of our customers and their projects.
How does Neural Machine Translation post-editing work?
Post-editing consists of carrying out an in-depth proofread of a text that has been translated by a machine translation engine. This technique is used in addition to the glossary and translation memory. The automatic translation output is followed by a Machine Translation Post-Editing that ensures the accuracy of the text.
This post-editing process is different from the revision process, in that precise rules need to be established in order to achieve the desired output. However, post-editing requires new skills for translators and we have therefore been providing training and guidance to our translators who are interested in specializing in this new discipline. A machine sees a sentence as merely a series of words. Unlike a human, it cannot recognize a cultural reference or a historic date. During the proofread, the post-editor pays great attention to these elements to guarantee the quality and accuracy of translated content.
Generally, we adopt hybrid solutions for our customers that combine Machine Translation with Computer-Assisted Translation (i.e. Translation Memory Technology). In this scenario, results from the Translation Memory have priority and will be proposed to the translator when available, and the Machine Translation engine is used for new content.
Below, you can see a general Neural Machine Translation project workflow:
This way, the translator/post-editor can work within a familiar environment and reap the advantages of the combined technologies.
When it comes to offering machine translation with post-editing to our customers, we generally propose the following service levels:
Raw Neural Machine Translation:
Neural machine translated text is delivered as is without any human intervention. We recommend that you only use this option when you need to obtain the overall gist of a text for internal use, very quickly. It is the least expensive alternative as it requires less initial set up and resources.
Neural Machine Translation with Light Post-editing:
Human translators make minor edits to the machine output text. This option is recommended for obtaining the gist of large volumes of content quickly. Even if you can expect a grammatically correct and understandable translation, we only recommend it for content intended for internal use.
Neural Machine Translation with Full Post-editing:
This option is ideal for content that needs to be published or widely distributed. Human translators perform complete corrections of the machine translation output to make sure that it is adapted to the target audience, the tone and style are appropriate, and it matches the content’s messages in the source language. The final output is comparable to that of a regular human translation. We recommend this option for:
- Technical documentation
- Annual reports
You can find more information about this here.
What are the requirements for adopting Neural Machine Translation?
For a long time, adopting an integrated machine translation process only made sense for certain types of content and for limited language pairs. Continuous technological improvements now allow us to better translate more language pairs and all type of contents in terms of output quality.
What can you do to improve Neural Machine Translation quality?
The quality, style and characteristics of the source text will determine your overall savings and the output quality of machine translation. At Acolad, our project managers analyze your documents carefully to decide whether they are suited to an automatic pre-translation and give you recommendations. If you follow a few simple guidelines, you should be able to optimize the quality and cost of your machine translation:
- Keep your sentences short and use controlled language
- Ensure you are using standard terminology
How much can you save with Neural Machine Translation?
How much can you save with Neural Machine Translation? Contrary to what many people think, using Neural Machine Translation will not completely reduce your translation budget (unless you just concentrate on using raw machine output for gisting purposes), but depending on your specific situation, you may be able to save up to 50% on your translation costs.
Are you ready to take the next step?
If you think that your translation processes could benefit from Neural Machine Translation solutions, contact us for a free consultation. Our experts will be able to advise you on the suitability of your content and provide suggestions for the best translation workflow.