Customized machine translation: The right option for your company?
When in 1999, as a translation student, I agreed to review a machine translation on HP servers, I was convinced that machine translation would never be able to come close to human translation.
Índice de contenidos
Index of contents
Index du contenu
For 3 years now, I can no longer say the same.
These two concepts are related and are undoubtedly the main characters of the greatest technological advances of this century.
This new model that makes possible the existence of autonomous vehicles or advanced medical diagnostics is the same one that allows us to say that Google Translator is much more smarter today than it was 5 years ago.
But can your company take advantage of this technological advance in the field of machine translation? Even if you're not a big multinational like Microsoft or Google?
The answer is clearly yes.
In this blog I would to introduce you what customized machine translation is. Also what requirements you must meet in order to take advantage of this new opportunity, lowering translation costs and even improving the quality of your translations.
During the last two decades, machine translation has been based on previous translations that serve as a model to generate new translations.
The number of previous translations, their quality and subject matter have a direct impact on the quality of the machine translations generated.
To get quality medical machine translations we should feed the machine translation system with many quality medical translations done by professional medical translators.
In the same way, we can improve machine translations about our company's products or services as long as we can customize the machine translation system with translations we already have.
The automatic translation "a la carte", "tailor-made" or "customized" is that machine translation that takes into account translations of a company or of a specific sector, in contrast to the generic automatic translation systems, such as Google Translator, Amazon or DeepL.
As you can read in the next section, the evolution of technology has made it more accessible and easier to have a customized machine translation system.
Machine translation has been with us since the 1950s. However, their fields of application have expanded significantly as technology has improved.
Early machine translation systems were based on grammar rules and dictionaries. The use of this type of machine translation was very limited. It was the emergence of what is known as statistical machine translation that allowed for a more widespread use of machine translation.
The main problem with statistical machine translation was that it required a large number of previous translations to start getting acceptable results. Around 2 million words of translations were needed for at least one language combination to start having a chance to get positive results.
This requirement meant that only large companies with huge volumes of translated content could make this approach feasible.
The new paradigm based on deep learning is known as neural machine translation.
This method has meant a significant qualitative advance in the quality of translations, but what you will be interested to know if you are not a multinational company translating millions of words a year, is that, with a translation volume of 100,000 words, you can already start to obtain decent results for machine translations.
Post-editing is the technical term we, translation companies, use to refer to the revision by professional translators of translations generated by automatic translation systems, whether generic or customized.
As I venture to say in my blog "Past, present and future of translation", post-editing will be the usual way of working for translators.
Post-editing improves productivity and, in some cases, helps to improve the quality of translations.
The quality of machine translation is directly related to post-editing productivity. The higher the quality of the machine translation, the greater the possibility of lowering our translation costs.
Neural machine translation, thanks to its better results, makes it feasible to use machine translation in scenarios where statistical machine translation did not work.
Despite the improvement in the last 3 years of machine translation, it still has limitations and there are scenarios in which it works better than others.
In general, we can say that the more creative the text to be translated is, the more unfeasible the use of machine translation will be.
Nowadays, it is not feasible to translate literary works or poetry with a machine translation software. Automatic translations generally lose the sense of the text when there are metaphors.
In a business translation context, it is not advisable to translate marketing materials by post-editing. Machine translation still tends to be literal and to use little idiomatic turns of phrase.
However, in the case of the translation of technical manuals, where the use of language is simpler and where the correct use of specialized terminology is crucial, an customized machine translation system can produce high quality translations.
As a rule of thumb, we can say that the productivity and feasibility of post-editing, i.e. proofreading machine translations, will increase as the creativity of the text and the importance of style decrease.
One of the techniques to achieve higher productivity rates is to write texts thinking that we will translate them through machine translation plus post-editing.
The three most important elements to consider when producing texts that produce quality translations are consistent use of terminology, grammatical simplicity of sentences and sentence length.
These elements are also important to facilitate the quality of the translation provided by professional translators, but have a greater impact in the case of post-editing.
If you are interested in this topic, you can read our blog entry: "What to do to improve the quality of machine translations".
To conclude, if your company has legacy translations with a volume of more than 100,000 words per language combinations and you need the translation of manuals or other types of technical documentation, such as financial reports or product databases, there is now the technology within your reach to study the creation of a customized machine translation engine that saves costs and guarantees the quality of the translations.
At AbroadLink Translations, like some other specialized translation companies, we now offer this service and technology to our clients. Do not hesitate to contact us so that we can study your case.
Josh Gambin holds a 5-year degree in Biology from the University of Valencia (Spain) and a 4-year degree in Translation and Interpreting from the University of Granada (Spain). He has worked as a freelance translator, in-house translator, desktop publisher and project manager. From 2002, he is a founding member of AbroadLlink and currently works as Marketing and Sales Manager.