The pitfalls of MTPE in medical translation

The other day we were holding a team meeting to discuss future steps for MTEC, and at some point the use of MTPE (machine translation post-editing) in medical translation came up for discussion. We’ve all completed post-editing projects and we all ended up using the same strategy, which is to delete the machine translation and start translating from scratch instead. Now why would we do that? Simply because it’s faster and offers a higher level of consistency and quality.

The following is a list of some issues we’ve encountered with MTPE:

• No consistency: for example, MTPE for Dutch provides a happy mix of the version of Dutch that is spoken in the Netherlands and the version spoken in Belgium. So we see that ‘clinical trial’ is sometimes translated as ‘klinisch onderzoek’ (nlNL) and sometimes as ‘klinische studie’ (nlBE). The same applies for ‘study drug’, which can have all manner of translations while only ‘onderzoeksmiddel’ (nlNL) and ‘studiegeneesmiddel’ (nlBE) are the terms to be used. There is no call for hybrids such as ‘onderzoeksgeneesmiddel’ or ‘studiemiddel’. And it’s not just us having a preference for these terms - they are also mandatory in the respective ICF templates for nlNL and nlBE to get them approved.

  
  

• No localisation: this can sometimes lead to severe misunderstandings. For example, the word ‘prescription’ is often pre-translated as ‘recept’, which is perfectly fine in the Netherlands, but patients in Belgium will have no idea what it means because there a ‘recept’ is a recipe in a cookbook.

  
  

• Formal/informal tone indifference: in our experience with Dutch, MTPE sometimes uses the formal tone and sometimes the informal tone. The informal tone should only actually be used for children, so we only need it for assent forms or brochures targeted at paediatric patients. Changing every instance of informal tone to formal tone or vice versa can be a laborious affair.

  
  

• Tags: when you upload a source file into a CAT tool, you will almost inevitably have tags appearing in the text. They contain information regarding the layout of the text, but they derail machine translations because they end up cutting sentences (or words!) in the middle. The result is a completely useless machine translation.

  
  

• Overly literal translations: while the days of mocking Google Translate are certainly over, other machine translation tools do still adhere to the structure of the source language, which can make the target text hard to read or understand. At MTEC, we often translate documents that target patients participating in clinical trials, and we believe they deserve more than that. We take pride in making sure that the documents they receive are both accurate and easy to understand, so they don’t have to first waste time and energy trying to work out what a sentence actually means.

As you might understand from this post, we don’t recommend the use of MTPE just yet because it is still very much in its infancy. There is clearly still a whole lot of work to be done before it can meaningfully contribute to the quality of translations (rather than just increase the workload for the translator).