-
Notifications
You must be signed in to change notification settings - Fork 0
Description
Hello,
I am using the DeepL translate API as well as as the write API with the Java client library. I noticed that the "rephrase" endpoint of the write API regularly produces a wrong source language. It is significantly less precise in recognizing the language than the translate API. Since I need the source language in my use case, this is forcing me to do two calls: One to translate, only to get the correct source language and another one to write to get the rephrased text.
Example:
Dutch example text: "Met deze zin wil ik laten zien dat herformulering niet altijd werkt."
deepLClient.translateText("Met deze zin wil ik laten zien dat herformulering niet altijd werkt.", null, "en-US")
returns
"text":"With this sentence, I want to show that restatement does not always work.",
"detectedSourceLanguage":"nl",
"billedCharacters":68
while
deepLClient.rephraseText("Met deze zin wil ik laten zien dat herformulering niet altijd werkt.", "en-US", null)
returns
"text":"In this case, I will later see that the reformulation does not always work.",
"detectedSourceLanguage":"en",
"targetLanguage":"en-US"
The API here wrongly detects 'en' as source language. In this case the rephrased sentence also does not fully reflect the meaning of the original sentence. This might also be due to a wrongly detected source language.
I'd highly appreciate a fix here since having to make two requests does not only take more time but is also billed twice.