OpenAI Provider for Trados Studio documentation seems incorrect to me

Question

Please someone correct me if I am wrong, but I think the current documentation is misleading. If you follow it and create a custom batch task: 
 Pre-translate Files Pre-translate Files with OpenAI provider 
 &hellip;you won&rsquo;t get correct results. 
 Here&rsquo;s why: 
 In the first step, TM matches and OpenAI run together. The AI translations are cached immediately. By the time you run the second step, OpenAI mostly just reuses what was already cached. It doesn&rsquo;t re-translate based on TM matches. 
 The correct approach is: 
 Run Pre-translate Files with only the TM enabled, using a lower fuzzy match threshold to catch partial matches. Then run Pre-translate Files with the OpenAI provider. This way, OpenAI only translates segments the TM didn&rsquo;t cover. 
 This ensures TM fuzzy matches are fully leveraged before AI fills the gaps. Excerpt from documentation:

Patrick Andrew Hartnett · Accepted Answer

Important technical point: In a single pre-translate action, providers do not operate as if one provider can immediately consume another provider&rsquo;s output within that same pass. So if TM, DeepL, and OpenAI are all enabled together in one ordinary pre-translate step, OpenAI does not reliably get access to the translation produced by TM or DeepL in that same process in the way a &ldquo;review/update this translation&rdquo; prompt would require. That&rsquo;s why the OpenAI-specific batch task was added: to support the &ldquo;second-pass review/update&rdquo; scenario in one batch-task process. There is no other reason for that specific batch task to exist.

Patrick Andrew Hartnett · Answer

Hey Jesse Good , your observation is valid, and I agree the current documentation can be confusing. 
 For a normal pre-translation workflow , the best approach is typically not to use: 
 
 Pre-translate Files 
 followed by Pre-translate Files with OpenAI provider 
 
 as a standard two-step sequence. 
 The reason is that, in the first step, TM/MT/OpenAI can already run and OpenAI results may be cached. By the time the second OpenAI-specific step runs, it may mostly reuse cached output rather than meaningfully reprocessing segments based on TM matches. 
 So for everyday pre-translation, the better workflow is indeed: 
 
 run Pre-translate Files with TM first, so TM and fuzzy matches are leveraged, 
 then run OpenAI afterward only for content not already covered. 
 
 That said, the &ldquo;Pre-translate Files with OpenAI provider&rdquo; batch task was originally introduced for a more " specific scenario" : when OpenAI is being used to review or update an existing translation , rather than simply generate a first-pass translation. 
 For example: 
 
 first pre-translate with TM or another provider such as DeepL, 
 then run OpenAI with a prompt intended to review/improve/update the existing translation. 
 
 This is important because, within a single pre-translate action, providers run independently, so OpenAI does not necessarily have access to the translation output produced by TM/another provider in the way a &ldquo;review this translation&rdquo; prompt would require. 
 So in short: 
 
 for standard pre-translation , use Pre-translate Files with the OpenAI provider selected as needed; 
 the OpenAI-specific batch task is really intended for the review/update translation scenario. 
 
 I do agree the current naming/documentation is misleading, and we should improve that clarification.

Patrick Andrew Hartnett · Answer

Hey Angelo Caltagirone , 
 Yes, you can do this in one custom batch task . 
 Think of it as two phases inside the same custom batch task: 
 A. First phase: normal pre-translation Run Pre-translate Files using your TM and Lara. 
 Important: in this phase, OpenAI should not be used as a normal translation provider, otherwise it will also translate at the same time as Lara. 
 B. Second phase: OpenAI refinement Then add Pre-translate Files with OpenAI provider as the next step in the same custom batch task. 
 Here OpenAI is used specifically to review/refine/update the translations that already exist from the first phase. 
 C. Result Your custom batch task would look like this:

Pre-translate Files 
 
 TM enabled 
 Lara enabled 
 OpenAI not used for this normal pre-translation step

Pre-translate Files with OpenAI provider 
 
 OpenAI prompt used to refine the existing translations 
 for example: apply terminology, follow style rules, improve Lara output, etc.

So your PMs should not need to run two separate batch tasks manually or change settings between them, as long as the custom batch task is set up this way. 
 The key point is: 
 
 Use Lara/OpenAI together only if you want them both to generate translations independently. Use Lara first and OpenAI second if you want OpenAI to refine Lara&rsquo;s output. 
 
 In your case, because you want Lara to create the translation and OpenAI to improve it afterward, the combined custom batch task is the right approach.

RWS AppStore > 2. RWS AppStore Applications

OpenAI Provider for Trados Studio documentation seems incorrect to me

Top Replies