Skip to main content

ConceptsConcepts lighthouse

To get you started on what our Machine Translation API is about, and its main functionalities, we have outlined some basic concepts that will help you understand the MT API flow and the underlying technology.

Machine translation (MT)​

Machine translation is the process of using artificial intelligence (AI), engines and algorithms to automatically translate content from one language into another without human input. Specifically Neural machine translation (NMT) is the next-generation of Machine Translation. This technology is exactly what underlines our MT API.

With an in-house dedicated AI experts, we keep our neural machine translation (NMT) engines on the cutting-edge, keeping them continuously well-trained with a large amount of data, including our users Translation Memories.

Translation Memories​

A Translation Memory is a database containing already translated words and sentences, this is usually registered in our system, and provided by the user so our translations are in line to the user´s terms. During the translation process, it detects identical sentences or fragments and it automatically suggests already translated texts.

You might assume using a Translation Memory (TM) would return similar outcome that of using our MT API. However, there are some differences between the two processes:

While a Translation Memory performs a lookup into a database of source and target (translated) strings, and looks for same or similar segments for the translation, during the Machine Translation process the target segment is completely created and transformed into "word vectors" using artificial neural networks.

Neural networks​

This is a more precise technology that takes the right terms for the users based on translation memories but also based on the learning done from the training data.

  • This means that dog is not only representing the forms d, o and g, but it can contain contextual information from the training data.

  • During the training phase, the NMT system tries to set the parameter weights of the neural network based on the reference values (source-target translation). Hence, words appearing in similar context will get similar word vectors.

  • The result is a neural network which can process source segments and transfer them into target segments. During translation, NMT is looking for a complete sentence, not just chunks (phrases) or isolated words.

Thanks to the neural approach, this is not about translating words but about transferring information and context, which returns in top level quality standards on your translations.

Imagine being able to translate more content, with a top-notch accuracy, quicker time-to-market and more cost-effective. This is now possible with our LanguageWire neural Machine Translation API!

Tip
  • Head to the Glossary to learn more about key terms and concepts used in this documentation.
  • See more details about our technology here.