Machine translation
Translation agency with AI experience
- Machine translation + post-editing
- Translations according to DIN ISO 17100 and DIN ISO 18587
- Optimization of training data for MT
Automatic translation of your texts
D.O.G. MT Solutions
In today’s world, cost-cutting measures are at the top of many managers’ priority lists. The interest in machine translation(MT) is greater than ever.
“How much can I save with machine translation?”
“What is the quality of MT and for which texts can I use it?”
These are some of the questions on the minds of many today. Here, we provide you with concrete answers and decision-making guidance. You can also download our whitepaper, which explains in detail how MT works and the usage scenarios.
Whether you are just looking for information, want to test MT or train an engine, contact us. We will be happy to help and advise you!
Machine translations have long been ridiculed. Since the successes of Google Translate and DeepL a few years ago, they are considered an important technology for the production of automatic translations. Expectations are high: reduction of translation costs with equivalent quality thanks to post-editing by trained post-editors. Is this possible? Yes, but not for all documentation and not for all goals.
D.O.G. began to actively address the topic of “machine translation” back in 2008. At that time, it was a joint research project with the French university ISIT: “Enhancing Machine Translation Quality with ErrorSpy”. This project resulted in functions in our ErrorSpy quality assurance software that can be used to better detect and correct translation machine errors.
Machine translation into all languages
We translate into all languages
Do you need an automatic translation? We will send you a quote within the shortest possible time. Send us your request using this quote form.
All-round solution for machine translation
What MT solutions does D.O.G. offer you?
Training progress: evaluation score
Training progress: gradient
- We advise you on the advantages and disadvantages of various alternatives.
- We work with you to develop specifications that define systems, workflows, integration requirements, and quality guidelines.
- We use an automatic translation system suitable for you, e.g. a neural machine translation (NMT) system, which we train with your data.
- You have a dedicated D.O.G. contact who coordinates the team of developers, post-editors and translators for your projects.
- We set up a quality management and deploy a team of post-editors who correct the machine-generated texts according to the agreed criteria. To do this, we use our quality assurance software and a terminology agreed with you, which we can maintain together with you in our LookUp terminology management system if you wish.
- We continuously maintain language resources such as translation memories and terminology, which are important for optimal training of translation engines.
Challenges of machine translation
We have the necessary knowledge
Machine translation yes, but not for all texts
Companies produce and use a whole lot of information. They serve different purposes (pure information and communication, advertising, operation of machines, etc.), are legally binding or not (like the instruction manuals specified by the European Machinery Directive 2006/42/EC), are formulated in a sophisticated way or are highly standardized.
Depending on how these characteristics and criteria turn out, machine translation is recommended or not.
In general, it is important to know that machine translation works better when:
- the subject and language combination has been previously learned by the translation engine. This is done in the machine learning phase, during which a relatively large amount of bilingual text (mainly segments from translation memories) was processed by the deep learning algorithm.
- a slightly lower quality compared to human translations are justifiable: a correct translation is not necessarily a good translation. In addition, the residual risk of mistranslation is greater with MT than with human translators.
- the quantities of texts are sufficient to operate the system economically.
- the text is written simply: short sentences, hardly any ambiguities, standardized terminology and syntax. Better yet, the author knows how MT works and writes texts that are optimized for machine translation.
- the text is for information purposes only. This is the case, for example, in internal communication, when you want to know "what's in the document/mail", when technical support needs to understand what kind of problem to solve.
- the time factor is in the foreground: MT is clearly faster, even with post-editing.
MT is not necessarily a competitor to translation memory systems. It is rather another tool with which translations can be produced.
Machines make other types of mistake
People make mistakes, machines do too, but others. For example, they translate proper names, do not understand many abbreviations, or add information that does not appear in the source text. They also sometimes omit words during translation. This is not always easy to detect, especially if the translated sentence otherwise sounds good.
That is why it is important to train post-editors, to inform them about the types of errors they need to look for. And the typology of these errors differs depending on the type of translation engine used: statistical machine translation (SMT) systems make different errors than neural machine translation (NMT) systems.
Since its foundation more than 20 years ago, D.O.G. has been working very intensively on the topic of translation quality and, since 2008, specifically on the quality of machine translation systems. We have developed our own tools and metrics to check this quality. Before machine translation starts, we identify the segments that are not suitable for the MT process and sort them out or mark them for further processing steps.
You can benefit from our experience.
Costs and benefits
Implementing a machine translation solution for your company is a project with two main components: setting up the system and getting value from the system.
Setting up a machine translation system: a model must be trained for each language combination. To do this, the system needs relatively large amounts of high-quality bilingual data. Training is done on special computers with GPU (Graphics Processor Unit) that allow massive parallel processing. Training can take several machine days and the training parameters must be optimized in several test runs. This is a cost factor.
Operating and using the system: Even in the deployment phase, the system must learn from errors and be trained with new topics and data. Finally, there is the cost of post-editing, which basically reflects the time that post-editors invest in correcting machine errors.
We charge separately for setting up the solution (a one-time cost) and for using the system. Thus, for the ongoing operation you pay a price per word that is lower than the cost of a human translation. After agreeing on the key data for your solution, we can provide you with a quote for this.
The cost-effectiveness of an MT solution ultimately depends on its deployment model. If you are already making big savings by using translation memory systems and want to machine translate the same documents in the future, then the savings might be rather small. However, if you are translating new texts and content that you have not translated before or have translated without translation technologies, then the profitability of a machine translation solution increases very quickly.
Your automatic translation in 9 steps
Workflow to your MT solution
Do you want to set up a translation portal
on your intranet that delivers fully automated translations? Do you want to support translators during their work with CAT tools (Computer-Aided-Translation Systems = Translation Memory Systems)? Or do you need post-edited documents that have a similar quality as human translations?
What do you want to achieve?
How do you plan to achieve it?
Training data for different languages
Linguistic data optimization
Structure of the base engine
Use of the engines
Quality assurance of translations
Learning from the MT errors
Engines update
Depending on the objective, different workflows and implementations are considered. Here, too, our team of specialists and software developers will help you implement the right solution.
MT for different industries
We translate for these industries
These are some of the industries in which we offer machine translation:
Language combinations
Automatic translation with post-editing in these languages
We offer a wide range of language pairs for machine translation. Here are some of the most requested language pairs:
- Machine translation German → English
- Machine translation German → French
- Machine translation German → Spanish
- Machine translation English → German
- Machine translation French → German
- Machine translation Spanish → German
- Machine translation German → Chinese
- Machine translation German → Japanese
- Machine translation German → Korean
- Machine translation German → Portuguese
- Machine translation Japanese → English
Machine translation + quality
Quality and time saving
The data and the training
Good data is crucial for machine learning. According to the motto “garbage in, garbage out”, the success of machine translation strongly depends on how extensive and how clean the training data is. First and foremost, translation memories (TMs) are the most suitable for training. But due to their creation history, TMs contain many segments that interfere with training: Empty segments, incorrectly assigned translations (e.g., in the case of incorrectly segmented sentences), or inconsistent terms and phrases.
Training is not a one-time affair. New texts and topics are added. Machine errors can be corrected through post-training. Our post editors give us regular feedback on typical recurring machine errors. Our developers use this information to retrain the translation engines on a regular basis. In this way, the engine adapts better and better to your linguistic wishes and specificities over time.
You can also benefit from our experience here. We help you optimize the training data to achieve a better result. We have developed methods and tools for this.
Machine translation and data security
An important reason for many companies to opt for machine translation is data security. No company management wants to be responsible for confidential information being tapped inadvertently because careless employees have it translated via a free translation service on the Internet.
The engines we train for you are either located on our own server in Germany, which is operated in accordance with EU law (GDPR, General Data Protection Regulation), or via secure cloud services in Germany.
You also have the option of installing the trained engine on your own server.
Post-editing according to ISO 18587
Many companies require machine translation solutions in conjunction with post-editing. Although the new standard on the post-editing of machine-produced translations (ISO 18587:2017) defines two post-editing levels Full and Light Post-Editing, in practice the criteria for acceptable quality depend heavily on the objectives pursued and engines used.
We help you to define the optimal quality targets for the work of post-editors for the solution you have chosen. This leads to specific guidelines for the work of the post editors.
Translation systems
Which engine?
If you want to travel to Rome, then you can choose between different means of transport. The same is true for machine translation. There is not one system and approach, but several alternatives depending on your objective, data volume and budget.
Thus, you can definitely choose between alternatives such as the following:
- Pre-trained neural systems (NMT) that have been trained with, for example, larger amounts of data in the field of engineering
- Neural machine translation systems (NMT) that are trained specifically on your data
- Statistical machine translation (SMT) systems, which are more suitable in certain situations due to better consideration of specialized terminology.
- We would be happy to advise you on further options!
Another SEO headline
Lorem ipsum dolor sit amet, consetetursadipscingelitr, sed diam nonumy eirmod tempor.
Another SEO headline
Loremipsumdolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor.
Another SEO headline
Lorem ipsum dolor sit amet, consetetursadipscingelitr, sed diam nonumy eirmod tempor.
Another SEO headline
Loremipsumdolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor.
FAQ
These questions are often asked in connection with MT
Yes. The risk of serious errors is greater with machine translation than with human translation. Some sentences read very well. However, some important information is missing or a statement has been misunderstood. This is not always recognized by inexperienced post-editors or post-editors who do not know the subject. With trained post-editors and quality assurance tools, as well as terminology optimized for MT, this risk can be significantly reduced. However, it can never be completely eliminated. A decision in favor of MT must therefore also take these aspects into account.
The cost of training an engine is not small. First of all, the hardware or infrastructure. Because of the very large amounts of data, you need powerful computers with very large memory. One can buy these computers or rent the infrastructure. Then, collecting and cleaning the training data costs time and money. Finally, the training itself, which requires a lot of know-how and some trial and error until the optimal engine configuration is found. Therefore, training a single engine (per language pair) is only worthwhile if the corresponding volumes of texts to be translated are available. Furthermore, in many cases the additional advantage of MT must be weighed against the existing advantages of translation memory systems.
The DIN ISO 18587 standard deals with post-editing and specifies the requirements for the qualification of post-editors. From a pragmatic point of view, post-editing is different from the revision of human translations, even though there are many similarities. The post-editor must be able to suppress the desire for perfection and limit himself to the necessary corrections, depending on the requirement for the quality level of the final result (full post-editing or light post-editing). He needs to know what specific types of mistakes MT machines make, because machines sometimes make mistakes that never or very rarely happen to humans (e.g., adding text). That’s why it is important to use service providers like D.O.G. GmbH who have experience with post-editing.
Technologies for successful machine translation
Technologies from D.O.G. to your advantage
The success of machine translation solutions stands and falls with the ability to detect translation errors and use the correct specialist terminology. This is where D.O.G. has very special technological advantages. Since the company was founded over 20 years ago, we have made it a priority to develop software products that support the quality of our translations. In particular, ErrorSpy and LookUp help us do just that.
LookUp is an intelligent terminology management system. In LookUp, relations between terms can be recorded, making contextual information available. Our software-based checks take these relations and your terminology into account. As a result, our post-editors are better able to detect machine errors.
ErrorSpy helps post editors detect and correct machine translation errors:
- ErrorSpy can detect missing translations
- ErrorSpy reports terminology errors
- ErrorSpy checks the punctuation, the numbers, the consistency of the translation
- ErrorSpy additionally works with regular expressions. Typical machine errors can be stored as regular expressions and automatically detected.
- ErrorSpy detects context-dependant translations and can report incorrect translations in a given context (e.g., when translating a word like "performance").
The AI know-how of D.O.G. developers and the ability to directly influence machine translation systems bring significant benefits already. You can benefit from this.
| German | Human | DOG Engine | DeepL | Kommentar |
|---|---|---|---|---|
|
Status Fehlteil setzen |
Set status reject part |
Set status reject par |
Set missing part status |
Terminologie wurde gelernt (Kein Sinnfehler) |
|
Das Handling wurde von der Maschine angefordert. |
Handling was requested by the machine. |
Handling was requested by the machine. |
The handling was requested by the machine. |
Stil wurde gelernt |
|
Cause: |
Cause: |
Cause: |
cause: |
Groß-/Kleinschreibung wurde gelernt |
|
Als erstes muß die Datei Einstellungen.cfg vom Hauptverzeichnis kopiert werden. |
File Einstellungen.cfg•must be copied beforehand from the main directory. |
First copy the file Einstellungen.cfg from the main directory. |
First, the Settings.cfg file must be copied from the root directory. |
Dateinamen (named entities) und Terminologie wurden gelernt |
Translation service
Automatic translation of your texts
These are some of the texts we machine translate. Of course, we also offer human translations.
- User manuals
- Website content
- Catalogs and e-commerce content
- Data sheets
- Hotel evaluation
- Reports or mails to customer service
- Presentations
- Tender documents
- Database contents
- Circulars and newsletters
- Training materials
Are you interested in machine translation?
Why not get in touch with us? In a non-binding conversation, we will discuss how MT can help you achieve your goals.
