Yamagata Europe has been using machine translation technology since 2009. In this post, we want to give you a short overview of the most frequently used machine translation technologies today: Rule-Based Machine Translation (RBMT), Statistical Machine Translation (SMT) and Neural Machine Translation (NMT).

Today, Neural Machine Translation gets all the attention, but should we believe the hype?

MT in the translation process

The traditional translation flow – without MT – usually includes a human translator making use of a translation memory. Today, machine translation is added as an extra resource in the translation process.

Rule-based machine translation (RBMT)

RBMT generates translations based on morphological, syntactic and semantic analysis of the source and target language. This technology was first developed to translate Russian into English during the Cold War. The first commercial RBMT systems already became available in the sixties.

RBMT roughly works in five steps:

  1. A dictionary retrieves the basic information about each part of speech (verb, noun, adverb, etc.…).
  2. Syntactic information is retrieved from the source word (e.g. tense, gender, singular/plural, etc.).
  3. The source sentence is parsed.
  4. The source words are translated.
  5. The dictionary entries are mapped with the appropriate inflected forms.

RBMT usually generates consistent, stable translations, without having to rely on a large bilingual corpus. But there are also a number of drawbacks.

Let’s look at an example of an RBMT translation from German to English:

Starten Sie die Wiedergabe am angeschlossenen Gerät und stellen Sie eine moderate Lautstärke ein.
Start the rendition at the attached equipment and adjust you a moderate volume.

The problem with this sample is that it reveals itself as MT right away. The grammar is not correct and it doesn’t sound fluent. This might be OK when no other translation system is available, but the mechanical nature of RBMT is troublesome.

Statistical machine translation (SMT)

In the nineties, computational power and storage capacity of computers boomed. This gave way to the rise of statistical machine translation (SMT). This technology generates translations based on statistical models derived from large bilingual text corpora, which started to become available in the nineties.

SMT can practically handle any language and it sounds much more fluent compared to RBMT. Have a look at the SMT sample below (German to English): 

Starten Sie die Wiedergabe am angeschlossenen Gerät und stellen Sie eine moderate Lautstärke ein.
Playback starts from the connected device and set a moderate volume.

Still not a perfect translation, but much more fluent than the RBMT translation.

The drawback however for SMT is that you need a large, high-quality bilingual corpus. This quality is critical. Train your translation engine with low-quality data, and you’ll be in for quite a disappointment.

Neural machine translation (NMT)

Artificial Intelligence (AI) is gradually pervading all aspects of life and business. In the world of translation, Neural Machine Translation (NMT) is the new kid on the block. Neural MT makes use of neural network technology, a form of Artificial Intelligence (AI), and machine learning.

This is our sample translated by NMT:

Starten Sie die Wiedergabe am angeschlossenen Gerät und stellen Sie eine moderate Lautstärke ein.
Start playback on the connected device and set a moderate volume.

As you can see, NMT provides translations that are much more fluent and readable than RBMT or SMT.

Which machine translation technology is best?

The short answer to this question is that it depends on the volume of training data, the volume of data you want to translate and the type of content you want to translate.

The most recent technologies (SMT, NMT) have made huge progress in terms of fluency and contextual accuracy, in comparison to RBMT. However, this translation quality can only be offered when a large set of high-quality training data is available. SMT and NMT make most sense when you need to translate in high volumes, such as technical manuals.

RBMT is fine for the translation of small content volumes and will provide consistent translation quality for short sentences and for a fixed set of terminology data.

We can help!

Do you have:

  • A tight deadline?
  • Large translation volumes?
  • The need to keep long-term costs under control?

Then machine translation might just be the right solution for you. Yamagata Europe has the technology to do this and the right people to manage post-editing.

Let us know what you need. We guarantee a quotation within 24 hours.