Use of machine translation at JGU
The possibilities offered by machine translation have advanced significantly in recent years. Machine translation systems online today can produce fluent translations. However, machine translation should be used with caution, as it is not suitable for all types of text and often requires revision by a human translator. There are also privacy issues to consider.
How does machine translation work?
Machine translation is the process of translating text from one language to another without human intervention. Scientists have been working on this topic for a long time using various methods, but only neural machine translation substantially improved translation quality: with the help of self-learning algorithms, large amounts of data and huge computer capacities, machine translation systems are trained and learn. Two of the most well-known machine translators are DeepL and Google Translate.
What is the difference between CAT tools and machine translation?
CAT stands for computer-assisted translation. CAT tools serve to assist translators in the translation process, but do not translate. CAT tools allow you to create, edit, and manage translations of various file types. Translations performed in CAT tools are stored in a translation memory at the sentence level and are offered as suggestions for later translations of similiar passages. Additionally, a terminology database is integrated into the tool to store new terminology during the translation process. Terms stored in the terminology database are suggested during translation.
Additional material such as dictionaries or a machine translator can also be integrated into the tool. This allows the user to choose whether the text should be pre-translated or if the machine translator should be displayed for assistance purposes only.
The benefits of CAT tools:
- No sentence is translated twice
- Possiblity to integrate NMT (Neural Machine Translation)
- Saves time, documents are updated quickly
- Consistency is easily maintained through the use of an integrated termbase
- Previous translations can be used via the "alignment" function
What is post-editing?
The term post-editing refers to the revision and modification of a machine translation via a professional translator. The types of errors to look out for in this process differ from those that occur in human translations. Despite the fact that modern machine translations are almost always grammatically correct, a closer look often reveals reference errors. Other issues that arise are mixed registers, German syntax, and inconsistent use of terms.
When should machine translation be used?
Not every text should be translated with machine translation. When deciding whether or not to use machine translation, there are a number of factors to consider, including the subject matter, the target audience, the intended use, and the level of confidentiality.
Machine translation (MT) without post-editing is useful and appropriate for texts for personal use, for example, where the content can be roughly understood and is not confidential.
Machine translation is also used in product reviews and descriptions, as on Etsy or AirBnB, in forums, or in how-to guides, as Microsoft does. However, these types of translations often come with a disclaimer.
Machine translation with post-editing (MT-PE) is suitable for internal correspondence with tight deadlines, for example.
Human translation (HT) is appropriate for confidential documents, particularly challenging texts, or texts that are to meant to convey emotion and style to the reader.
The table below will help you decide whether machine translation (alone or in combination with post-editing) is the best option for your text.
Further recommendations on using machine translation
- The leading online machine translators, like DeepL and Google Translate, save texts for training purposes. Therefore, do not use them to translate text containing sensitive or personal information.
- To avoid incorrect translations or omission of information, you should know the language that you are translating into well enough to be able to avoid these mistakes.
- Consult the JGU Glossary or the Staff and Facilities Directory for the translations of official names and university-related terms.
Further reading:
Blog: Translation und Technologie (Hypotheses)
Ottmann, Angelika, and Carmen Canfora. "Risiken und Haftungsfragen bei neuronaler maschineller Übersetzung" Maschinelle Übersetzung für Übersetzungsprofis (2021): 171-184.
Brown et al.: "Language Models Are Few-Shot Learners" (2020).
Popel et al.: "Transforming Machine Translation: A Deep Learning System Reaches News Translation Quality Comparable to Human Professionals" (2020)