What are the different types of machine translation?

Machine translation (MT) refers to the automatic translation of texts from one language into another language by a computer program. There are different types of machine translation, which differ in how they work.

Literal translation

The simplest and oldest variant is the direct method, in which the translation is carried out verbatim and the result is corrected afterwards according to the linguistic system of rules. However, this method is very error-prone and is now outdated.

Phrase-based translation

Example-based machine translators work with a translation memory in which entire phrases are stored. These phrases can then be recalled along with their equivalents, rather than being translated word for word.

Transfer and Sentence-Based Translation

For statistical machine translation, the program is first provided with a text corpus of source and target languages, which it can compare and from which it derives transmission rules. The more often words and phrases are matched in both languages, the more likely they are to be translated that way. The transfer method is the classic MT method with three steps: analysis, transfer, generation. First, the grammatical structure of the source text is analyzed, often in a tree structure. After that, the structures are transferred into the target language (=transferred). Finally, in the target language, sentences are generated from the structures with grammatical rules and thus the target text is generated (=generated). Since 2016, artificial neural networks, i.e. artificial intelligence, have been increasingly used for translation programs, which has led to a rapid increase in progress. Examples are DeepL, Google Translate, Yandex.Translate and Bing Translator, which achieved significantly better results from then on.

AI-based translation tools are based on deep learning. Instead of producing direct word-for-word translations, they rely on extensive databases, search for language patterns, and perform in-depth comparisons to provide better and more reliable translations. Artificial intelligence is completely redefining the language learning sector and the translation industry. Neural machine translation has proven to be a great improvement over rule-based translation technology, as it is able to understand nuances and complex language constructs such as idioms, expressions, and metaphors. This results in almost human-like translations, especially when paired with generative AI platforms.

How good is the quality of machine translation?

The quality and evaluation of machine translations is the key aspect for companies. So far, various mathematical algorithms, such as the Bleu score, have been used to measure the similarity of automatic translation to a human reference translation. These mathematical metrics correlate well with human evaluations and are particularly useful for evaluating large text documents with several thousand sentences. However, it is only the human evaluation of the translation quality by relevant stakeholders that increases the acceptance of translation quality in the company and is always useful for machine translation models specially trained for the company.

What role does artificial intelligence play in machine translation?

Artificial intelligence (AI) plays an important role in the machine translation of texts. Neural Machine Translation (NMT), in particular, has established itself as a reliable translation solution. AI-powered translations take the pressure off human translators by allowing them to access huge data sets and process relevant data incredibly quickly. And AI-supported translations are getting better and better – especially thanks to models specially trained for the company.

However, there are still requests for changes and linguistic innovations to a certain extent, so that revision and proofreading are still necessary. At Translate, these changes can be incorporated into the model through the continuous training of the translation AI. In this way, the ad-hoc intervention in translations is maintained, while the complexity of overcrowded glossaries is avoided. Regardless of the sector in which machine translation is used, companies with this technology no longer need to look for experienced translators when calculating lead time. Projects – such as website relaunches or international rollouts – can be completed much faster, while project milestones and deadlines are much easier to meet.

How can machine translation be improved?

Machine translation is widely used today and can help businesses process large volumes of documents, audio, and video material quickly. However, there are some factors that can affect the quality of machine translation.

1. Language pair (e.g. German – English) – some languages are easier to translate than others.
2. Domain of words – generic translation models do not know the technical terms and products in specific industries and B2B sectors.
3. Context – complex language constructs such as idioms and jargon and language jokes are difficult to implement by machine translators.

To improve the quality of machine translation, there are ways such as effective terminology management.
Glossaries can be created to include new, relevant terms and retire outdated terminology. Another option is to customize and train the machine translation engines.

The quality of machine translation depends heavily on how the models are trained. Both methods can help improve the quality of machine translation and reduce the need for post-editing. At Translate, we connect both glossaries to trained models and regularly empty the glossaries, using them as a basis for retraining the language models.

These technologies, together with stop-words (word equivalents that are the same in every language – e.g. product names) deliver the best possible machine translation result. Overall, machine translation has gone through several iterations and technological changes, which has also drastically improved the quality of machine translation. Companies can benefit from machine translation and streamline their internal content and translation processes to significantly reduce time-to-market for global content campaigns and product launches

[lc_the_author_full_name]

[lc_the_author_description]

learn more