Machine translation (MT) is a fully automated process of translating content from one language into another by means of software. As part of computational linguistics, it is often directly connected with artificial intelligence (AI) engines to leverage human input for improving the automated translation output.
Main types of machine translation
Algorithms used by software for machine translation can be developed based on several principles – for example, linguistic rules – that demand a certain action for the output. This results in different approaches to obtain the required results for each machine translation type.
The main types of machine translation are:
- Rule-based machine translation,
- Dictionary-based machine translation,
- Statistical machine translation,
- Example-based machine translation,
- Hybrid machine translation,
- Neural machine translation.
The main differences between these types concern the foundation they are built upon: intelligent guessing, matching and analogies derived from already translated texts, linguistic rules, dictionary entries, or bilingual text corpora. The software, system, or network used can be trained through human input in the form of algorithm adjustments or deep learning methods.
The output obtained from different machine translation types vary greatly in quality. The availability for specific language pairs can be restricted due to a lack of sources to build on or human input for training. Only neural-based translation could so far reach a level of human parity, that is, output that can be compared to the output a human translator would have produced.
Quality differences in machine translation types
With technological advancements, such as neural networks, deep learning, and AI engines, machine translation reaches new levels of output accuracy and efficiency. These have a huge influence on the quality of the translation output.
There are specific factors within the source text that affect the output, independent from the type of machine translation used. And then there are some types of machine translation that are better suited for certain types of texts, whereas some can achieve good quality for any text.
The following factors in the input influence the machine translation output :
- Text format and length (i.e. lists, short sentences, divided into paragraphs),
- Text complexity (tripod consistent of text quantity, text quality, and reader-text-factors),
- Linguistic complexity (phonology, morphology, syntax, and semantic),
- Textual, contextual and interpersonal factors.
This means that a complex contract written in technical language for the legal field, adapted to the specific laws of the relevant region, requires a different processing approach than a list of items for grocery shopping or a short email update for a colleague abroad.
The scenario is very similar to a human translator who needs to have the necessary skills and subject matter expertise to be able to create a high-quality translation output. However, a human translator brings advanced cognitive language capabilities to the process that machine translation has yet to achieve, though its developers are constantly improving those abilities. Translation turns out to be an extremely complex computational challenge, especially when you consider the range of languages and cultural considerations involved.
How to leverage machine translation?
Machine translation can be used exclusively or in combination with human translation to render a higher-quality final translation product. Depending on the use-case and the complexity of the content, machine translation quality can be enough for the purpose of communicating something quickly and clearly without having to spend a fortune.
Here are some possible use cases for machine translation:
- Simple, unofficial emails,
- Price or product lists,
- High volumes of user-generated content (e.g. hotel or product reviews),
- Online chat,
- Product support answers to recurring questions,
- Status translations for community forums or social media.
Combining the power of machine translation with human translation is the safest way for any business to leverage the potential of MT without compromising on the quality. This means using previously translated content that is repeatedly used – via a translation memory (TM) – and automatically inserted during the process of machine translation. All other content that has no such matches in the translation memory will then be translated by a human translator.
This way, the human translator constantly refines the translation memory with good quality content. With each new translation project, the company leverages that continuously updated TM more and more. This results in cost savings due to elimination of redundant translation and shorter turnaround times, while improving quality.