All rights reserved 1998 - 2023 ABZ ONLINE

NEWS

Neural Machine Translation (NMT) - When can Neural Machine Translation be considered as an alternative?

This question is often asked, and customers usually have their doubts. Here is a short list of situations where you should consider discussing Neural Machine Translation (NMT) with our project manager.

Internal communication

Texts intended for communication within the company, such as internal emails, minutes, or intranet news, are well-suited for NMT. A light post-editing is sufficient for internal purposes since the main focus is on text comprehension.

User-generated content in e-commerce, support and portals

Texts created by users or customers, such as product reviews, feedback, or comment sections under blog posts, are ideal for machine translation. They only require light post-editing to serve their purpose.

However, if newsletters, blog posts, or social media posts are part of an advertising campaign, they fall under marketing texts. In such cases, NMT should be avoided.

Product information

Large volumes of frequently repeated text can be handled efficiently with machine translation. However, technical specifications and product information, such as ingredient lists or instructions for use, should still undergo extensive post-editing. This is because they provide important sales and decision-making information.

Product information should not be confused with product descriptions, as the latter are marketing texts that focus on different aspects not considered by machines.

User manuals

Machine translation is also suitable for user manuals with purely informative content. As with all the other examples, the machine must first be trained using deep learning. This ensures factual accuracy. In addition, extensive post-editing ensures error-free printing.

When should neural machine translation (NMT) be avoided?

Marketing texts
 

Creative texts, especially slogans, are often rich in puns, idioms, irony, and other rhetorical devices. They are intended to appeal emotionally to the target audience. Neural machine translation (NMT), at its current state of the art, is unable to faithfully reproduce such linguistic devices with the same meaning.

Websites and Apps
 

A website must be intuitively usable from the home page to the last subpage. The same goes for apps on devices. Textual references must be created and a consistent, coherent user experience must be provided. Devices lack overview, so be careful. Many website operators also make a distinction between high-quality and standard-quality text, with the latter sometimes being more suitable for machine translation.

The term "transcreation" is often used to refer to the adaptation of web content for a target market. It goes beyond translation and involves interpreting the text for a foreign culture's target audience, including product and service names and color schemes. Again, translation engines would be completely overwhelmed in this scenario.

Speeches and presentations

The transcription of spoken speeches, such as for company events or business presentations, relies on rhetorical phrasing, allusions, irony, or creative wordplay. These aspects surpass a computer's capabilities, and machine translation can yield grotesque results in such cases.

Legal and other specialized texts, such as patents

Nested texts with long sentences, where every word is crucial, often do not lend themselves well to machine translation. Machines may lose context and mix up individual concepts or themes. Proofreading such translations is often more of a detriment than a benefit, as mistranslated terms or even sentences can cause confusion rather than help.

Literary texts

It may go without saying, but for the sake of completeness, we strongly advise against the temptation of having literary texts translated by a machine.