Frequently Asked Questions
What languages does Globalese support?
All of them.
Is your engine better than Google Translate?
First of all, let’s get this clear: Globalese is not an engine. It is a platform for building engines.
The quality of any engine depends on the input: the corpora. If you have the relevant corpora for your client’s project, you can expect better results than from an engine that is built on billions of words but lacks the specific data that only you have.
Will Globalese use my assets (TMs, files etc.) for building stock engines / engines for other users / selling it to Google/Microsoft/Facebook etc.?
Nope, we’re not the bad guys. Nothing that you upload is stored outside your own system or shared with anyone. We are not reusing your assets in any way.
Does Globalese provide stock engines?
No. We are very happy to help you find additional corpora if you need any, but we don’t think we know better what’s good for you than yourself.
What happens to inline tags in translated files?
Glad you asked. You will find them where you expect them to be — neatly reinserted in the proper positions in the target text. (Hey, we’re already beating Google!)
Can I stop paying all my human translators once I start using Globalese?
No. Machine Translation is not meant to replace human translators, but to assist them in their job to become more efficient. In other words: the same number of translators will be able to do more work given the same amount of time.
Also, keep in mind that even Neural MT is not the silver bullet. It is a great leap from Statistical MT, but you still have to have someone check and correct any mistakes the engine makes.
How does Globalese improve translation productivity?
Machine translation is not magic, but it can be a powerful productivity tool. By creating project-specific engines based on your translation memories and termbases, you get a kind of ‘intelligent translation memory’ which does not only show you already existing matches, but can generate good quality new translations based on low matches.
How do I measure my productivity gains?
Good question. We used to say you can compare the post-edited files to the raw MT output and look at the BLEU (F-measure, PER, TER etc.) scores. With neural MT output, things have changed and these automated metrics are less reliable than before. Probably the best way is to measure the time it takes to finish a project without/with MT.
What influences the quality of a Machine Translation engine?
The top influencers of the quality of an MT engine are the size, quality and relevance of the corpora.
What is the minimum Translation Memory size to create good quality MT engines?
One of the basic rules in Neural Machine Translation is that the quality of your engine depends on the volume of your corpora. On the other hand, volume is not everything: it is equally important to have good quality and relevant data. You can already achieve good results with a TM size of 100,000 segment pairs.
Who may benefit from using Globalese?
Organizations that wish to translate large volumes of texts in a short turnaround time, seeking data security, speed, accuracy and reliability – that is, Language Service Providers and Content Owners.
What are the advantages of Globalese compared to other NMT solutions?
Globalese can be fully integrated in the translation workflow, while your translation project managers and post-editors can work in their well-known CAT environment.
What types of documents should I use MT for?
Use MT in domains where the source text can be taken literally. That includes almost anything except marketing: manuals, software UI, technical documentation, travel websites, patents, you name it. Just don’t expect MT to translate War and Peace in style, or come up with an innovative translation of the next great catchphrase your client has just come up with. MT is not a literary translator or a copywriting tool.
User-generated content is a tough one, because very often the source text is loose, ambiguous, lacks punctuation or is full of typos. The more controlled the source text, the better.