A few weeks ago, we hosted our very first webinar on how MT and post-editing can be used to translate various types of content. We think we owe you a follow up with a more detailed description of three approaches to machine translation and human review and various options that companies have at their disposal thanks to them.
Obviously, there is no out-of-the-box solution that will work equally well for all companies. This is why we always point out that we should discuss specific company needs, before jumping to solutions. We will talk about approach too!
#1. Training a machine translation engine model and post-editing
Yes, it is possible to train a machine translation engine model. First, you have to define whether the model will be company-specific (using the company’s content, terminology for training), or whether you will be training a more generic, vertical model, which would work within a certain industry.
Let’s say that you have trained your own company machine translation model.
- What do you do now, and who will be checking the translated content?
- Will they be doing classical review/proofreading tasks, or should this be a different task?
- If you are doing this internally, will you be allocating your colleagues, or will you be managing a team of freelancers?
- If you are outsourcing to freelancers, how do you explain the task to them, and how much should you pay them?
Suddenly, all of these questions, and many more, become relevant. The main about this approach is that you have to take into account that this is a trained model. This means that probably, there will be specific mistakes it will be making (like terminology).
Once you trained the model and you know how it should behave, it should be easier for you to see the mistakes and to tell the people doing the post-editing what they should pay attention to. If you prepare the guidelines properly, you can improve the turnaround times, and spare everyone a lot of stress.
If you combine this approach with the use of translation memories (collections of human translations), you can further improve the quality and save all the hard work your team does.
#2. Generic machine translation engine model and post-editing internally
Most companies won’t immediately start training their own machine translation engines. Many will not have either content nor people needed to do this.
Using this approach to machine translation and human review you have to bear in mind that you won’t be familiar with the mistakes the MT engine will be making. In effect, this might take you longer to prepare guidelines that the people within your company might use when checking the machine-translated content.
In any case, it would be very useful to use a TMS with an integrated machine translation engine. Usually, they are paid versions, which means that you get privacy (always check this), and better quality.
#3. Generic machine translation engine model and outsourcing post-editing
For many companies, this might actually be the best approach to machine translation and human review because it’s the easiest one. They don’t have to train their machine translation engine, they don’t have to provide high-quality content to be used for machine learning, and they don’t put additional stress on their employees, asking them to check the machine-translated content.
In such cases, the best option for you would be to use a company which has both the system to support you and the services under the same hood. In other words, the review should not then be sub-outsourced to other companies, as this creates additional responsibility issues in case of faulty translations, and your review and approval process will drag out unnecessarily.
Our approach to machine translation and human review
At Text United, we work on an integrated machine translation engine and we cooperate with freelancers. Because of that, we are responsible for the work they do.
If we outsourced to other companies, we would forward you to the company who actually did the review, and since they never had actual contact with you, it might be difficult to define responsibilities, as they could shift it back to the company that contracted them.
The implications of introducing new technologies and innovative workflows need to be discussed before actually implementing the solution. Since we have been working with our clients on such workflows for several years, we can provide you with advice on how to approach the above questions.
If you want to discuss your specific case, simply reach out to us! As always, there are no strings attached, and we would love to hear more about your projects and challenges you are trying to overcome.