What is Machine Translation (MT)?

Machine Translation is computer generated translation, based on specific algorithm sets. Usually fast and simple to use, MT engines represent a quick and easy, although not always the best solution for translation.

Machine Translation can be rule based, statistical or neural - or even a hybrid of several systems.

Machine Translation basics

Machine Translation (MT) is an automated translation of text performed by a computer. It provides text translations based on computer algorithms without human involvement. With Machine Translation, source text is easily and quickly translated into one or more target languages. Maybe the most well-known Machine Translation Engine is Google Translate, and LingoHub put its trust as well in DeepL and Amazon Translate for machine translation purposes.

Machine Translation is not to be confused with Computer Assisted Translation (CAT). While CAT includes the use of different machine translation engines, their role is usually supportive to the human translator. On the other hand, Machine Translation is the sole product of the computer, although human reviewers might be included for quality assurance.

Things to consider when opting for Machine Translation:

• Accelerated translation workflow: translations are done much faster
• Type of content that is being translated: not all content is suitable for MT. However, content which is frequently changed or updated would demand high costs of human translation and MT might be the perfect solution.
• Not all languages provide the same quality of results: MT still provides the best results when translating from and to English, with several languages close behind: Spanish, French, German, Portuguese.
• Human factor: no matter the quality of MT, human professional translators are still the most reliable option. Are you translating generic phrases or specialized or niche texts? Before choosing MT, consider including a human reviewer in your translation process.

How is Machine Translation done?

The most basic function of Machine Translation is the word-by-word translation where a word in source language is replaced with the corresponding word in another language. This system has been a source of much confusion and general distrust in MT engines. However, since its humble beginnings, MT has developed significantly and the quality and precision of translation is setting higher bars.

We can distinguish several types of Machine Translation:

1. Rule-based Machine Translation (RBMT)

Data: Relies on linguistic rules and bilingual dictionaries for every language pair.

Translation: uses rule sets to transfer the grammatical structure of source content into the target language

Costs: initial and ongoing investment to increase translation quality steadily

How it works: Basically, a dictionary for the source language is used to select appropriate words in the target language. Syntax and grammar rules of both the source and target locale are observed, and the words taken from the dictionary are adapted appropriately (gender, grammatical number, word order etc).

2. Statistical Machine Translation (SMT)

Data: uses statistical translation models from analyzing training data

Translation: is selected from training data with algorithms to select the most commonly appearing words.

Costs: extensive hardware configuration is necessary to run MT.

How it works: By learning and comparing to training documents, desired source text is translated based on probability of occurrence in the target language. Works best for language pairs with similar word order.

3. Neural Machine Translation (NMT)

Data: uses deep learning system to teach itself and constantly improve

Translation: created by using representation learning and deep learning.

Costs: training models are quite expensive

How it works: NMT functions similar to a human brain, by using neural network models to create statistical translation models.