Neural machine translation (NMT) uses machine learning and an artificial neural network to perform language translation. It predicts the likelihood of a sequence of words, typically modeling entire sentences in a single integrated model.
Let's deep dive into NMT and how it can streamline translation processes.

How neural machine translation works

Neural machine translation (NMT) automatically converts source text in one language to target text in another language.

Unlike traditional statistical machine translation (SMT) models, NMT only requires a fraction of the memory. Furthermore, unlike conventional translation systems, all parts of the neural translation model are trained jointly (end-to-end) to maximize the translation performance.

With the power of Neural Machine Translation (NMT) and available public translation platforms, users can generate instant translations with little to no customization. To increase accuracy the terminology can be tailored based on the context, category, style, and target audiences.

Neural machine translation vs. professional human translation

We believe that NMT is an unconditional part of today's translation process. NMT combined with translation memories is the most helpful tool for professional translators to increase efficiency and output. It's basically impossible to scale the translation business without NMT nowadays.

Professional translators are tasked with NMT post-editing to make sure the translation is natural and fits the context accurately for target audiences. To control translation quality effectively, translators can rely on mandatory QA checks and QA operators. Quality output is achieved through this collaboration of artificial intelligence, language professionals, and QA procedures, resulting in a hybrid translation process.

Neural machine translation for everybody?

NMT is broadly available to the public using cloud services on platforms, servers, or via software integration using an API. Users can utilize independent or open-source machine translation systems to build their very own NMT system. If a corpus of source and target texts in two languages are provided a neural language model can be established.

In combination with CAT tools users can provide live translation suggestions to professional translators while improving suggestion quality learning from the sentences previously chosen.

Data is key. It is essential to create an effective NMT network.

Public machine translation platforms and why NOT use them

We strongly advise our clients to use an on-site machine translation engine. Public machine translation (MT) platforms are often open and shared, and the translations are not always kept confidential.

For example, the NMT platform of idioma® runs inside our corporate network with no external access. Our cloud solution uses data encryption to avoid data leakage and we can guarantee the data never reach the public.

Currently, NMT is the most advanced translation solution. It can produce adequate translations fast and, with the help of professional translators, it can generate quality output. Free public MT engines are practical but have their limits. We believe that to use public NMT platforms in a satisfactory manner, the next level of development, adaptation, and security is necessary.

Neural machine translation solution at idioma®

Our IT developers built a flexible NMT service based on idioma®'s 40 years of experience in the translation industry.

We provide many options integrated with our NMT solution:

• Data selection: manually or automatically choose appropriate data to feed the solution from your corpora or our corpora to best fit your translation projects.
• Data cleaning: edit or remove data to increase the quality and efficiency of your NMT
• Neural domain adaptation: tuning neural networks to individually fit specific translation content
• Terminology management: terminology, DNT, and tag control.
  

Neural Machine Translation (NMT) is moving faster than ever before. And we aim to provide our clients and partners with the latest techniques and solutions available.

Do you have any questions on NMT or would like a translation quote?
Please contact us at info@idioma.com.