A machine translation short guide for SMEs
Small and medium-sized companies require a maximum of 5000 pages of translation per language and year. Individualized MT solutions are too expensive for this volume and SMEs should make clever use of existing services:
- Do not use machine translation for confidential documents
- If possible, use DeepL as your machine translation engine
- Use MT without post-editing only for informative translation, support forums and online catalogs
- Use MT with post-editing for large amounts of text, short processing times
Machine translation, MT, post-editing – what is that?
Machine translation, or MT for short, is the automated translation of texts using computer software. MÜ and MT are used as equivalent abbreviations in the German-speaking area.
- Post-editing is the post-processing, checking and correction of machine-translated texts.
- MTPE is the combination of machine translation and post-editing.
The various methods of MT are discussed below.
Cleverly used machine translation can help you save costs. We will be happy to advise you on the use of MT for your translation tasks.
Dipl.-Kffr. Sanne Jerxsen
Machine translation at PRODOC
Use of MT in the CAT tool
We use memoQ’s MT plug-in to integrate MT into the translation process in a targeted and flexible manner. Machine translation is therefore part of the normal translation workflow with the following advantages:
- Flexibility with file formats
Any file format can be translated with MT without special preparation. InDesign files can, for example, be imported directly, translated via MT and output as InDesign files again.
- Combination of MT and Translation Memory
Existing translations come directly from the Translation Memory (TM), similar texts are also translated using the TM and machine translation is only used for completely new texts
- Consistent terminology
Use of company terminology can be controlled
- Quality assurance
Quality control is carried out with the usual tools and according to the same procedures as for other specialist translations
Post-editing is important for good quality
Today, machine translation is already very good, if not yet perfect. Therefore, PRODOC uses post-editing for optimal results. Post-editing is the checking and subsequent correction of machine-translated text by a technical translator. PRODOC’s technical translator adhere to the requirements of DIN ISO 18587 “Post-editing of machine-translations” for post-editing.
MT options with PRODOC
We generally offer 3 options for the use of MT:
Simplifies and accelerates your workflow for a large number of files or for data in formats that cannot be processed directly by the generic services – disadvantage: sometimes incorrect and/or grammatically/stylistically questionable translations.
This is recommended if not the quality but time or price have the highest priority
If you value certified service and high quality. Using DeepL as a translation engine generates results faster and cheaper than usual. Disadvantage: DeepL is only available in a few languages.
MT without post-editing for support articles – Example Microsoft
Microsoft uses machine translation on its support page to make the articles available in a variety of languages. Texts translated accordingly are marked with an icon. However, the company explicitly points out that the quality of the articles that come from machine translation is lower than that of professional translations. Here, too, the algorithms behind MT need to be improved further.
Workflow with MT in a CAT Tool
Languages for MT offered by PRODOC
We want to offer our customers added value with MT. This is only possible if we can integrate MT into the normal translation process using CAT tools. At the moment only DeepL is available as an engine for SMEs and this limits the number of languages in which we offer MT.
MT is available for the following languages:
Which translation service or engine is best suited for SMEs?
The state-of-the-art today is neural MT. Google, Microsoft, DeepL and many other major MT service providers use this technology – see below for more.
All of these services have the problems described above with inconsistent terminology because training data for the engines comes from many different sources.
Only full control over the training data can result in terminologically flawless MT results, which can then be published directly without post-editing for certain application purposes. However, this requires an extremely large amount of data. Daimler, Siemens, Bosch and similar industrial giants have sufficient data. SMEs, however, usually do not.
SMEs must therefore use Google, Microsoft or DeepL for machine translation.
We recommend (as of 2019) the use of DeepL – in our experience, this generates the least post-editing effort.
SMEs should only consider machine translation if they carefully select the texts to be translated. SMEs should also ensure that MT is integrated into a translation workflow with CAT tools and revision.
Large companies with high translation volumes can build their own machine translation engines and thus achieve terminologically consistent results. Smaller companies with a translation volume of less than 1000 pages per language and month have to use generic engines without terminology consistency. Post-editing is required for generic engines to achieve good results. Depending on the engine and deployment scenario, machine translation with generic engines can achieve good results if you set up the translation workflow accordingly.
Overview of MT technologies
Training data is the basis for machine translation. Training data are existing translations in the form of bilingual files that are fed into the machine translation engines. These existing translations are used by the systems in different ways to achieve more or less useful translation results. For good results, enormous amounts of data are required independent of the system: several million sentences per language.
This reinforces the fact that Google and Microsoft are leading in generic engines. They use their search engines to index all multilingual websites on the Internet and use their content to train their MT engines. DeepL was added at the end of 2017 and uses the data of the language portal linguee.com.
RMT – Rule-based MT
Based on morphological, syntactic and semantic rules, the training material is broken down into sentence components and a database is created which is then used for the creation of machine translations. The integration of dictionaries for terminological consistency is possible. The quality is particularly good if controlled language is also used for the German source text.
SMT – Statistical MT
Statistical analysis of a large bilingual corpus. Similar and frequently occurring sentence structures and grammatical structures are extracted in this way, which later serves as the basis for the translation. Here, translations are usually not very consistent. The results, however, are usually better than with RMT.
NMT – Neural MT
As with the other methods, NMT analyses training data and uses it as a basis for translation. In neural machine translation, however, the texts are “trained” by a neural network using deep learning algorithms. While the statistical approach of MT calculates all possible syntax options using word structures, the deep learning method starts a loop in which sentences are built until the best possible result is achieved. The use of artificial intelligence in machine translation has significantly improved the quality of the resulting translations.
Machine translation – Short history and future prospects
Global networking is in full swing. The speed at which the importance of digital connectivity has increased in recent years is hardly surprising for industry experts.
For many companies, the digital revolution was accompanied by operational restructuring and innovation.
The explosive spread of data and its processing can be exploited by companies through translation and localization to operate internationally and to open up new opportunities. The key to success here is to convey multilingual information and messages to potential customers in line with the target market.
The translation and localization industry also underwent a reorientation of its way of working. Increased data traffic and, in particular, the speed with which texts can be exchanged require fast and efficient localization of texts. Although the use of machine translation is not uncommon for fast content transfer, MT has become much more user-friendly in recent years due to the increased quality of the translated text.
MT is also used by PRODOC. Since machine translation is a complex subject, however, different aspects and requirements must be taken into account in order to achieve high-quality results. Starting with the nature and structure of the texts to be translated, the focus here is on adapted workflows and various types of machine translation.
With global connectivity and the speed needed to deliver new content, MT is hard to ignore these days. Despite all the progressive and positive developments, it should nevertheless be used with caution
Post-editing and a final review remain an absolute necessity in order to transfer texts into the foreign language in a way that is appropriate for the target group.
Are you interested in MT as well?
Use our quotation form and upload the files to be translated. Under “Remarks”, enter the MT service you are interested in:
- MT (without control, many files)
- MTPE (with corrections by a technical translator)
- MTPE certified (like MTPE with additional revision by a 2nd technical translator)
Dipl.-Ing. Stefan Weimar