Incorrect translation with xml tags

[twarkie]

Hello! I'm seeing some strange translations when using the DeepL API with xml tags. The tags are just used to map the result back to the original text.

Here's a simple Swedish string that can be used to reproduce the problem: <x key="0">Köpesumma</x>. It should translate to Purchase amount.

$this->client = new \DeepL\Translator(config('services.deepl.api.key'));

$translated = collect(
    $this->client->translateText(
        ['<x key="0">Köpesumma</x>'],
        'sv',
        'en-US',
        [
            TranslateTextOptions::TAG_HANDLING => 'xml',
            TranslateTextOptions::PRESERVE_FORMATTING => false,
            TranslateTextOptions::FORMALITY => 'prefer_less',
        ]
    )
)

After looking through the code I found that the following form data was generated for cURL: target_lang=en-US&source_lang=sv&formality=prefer_less&text=%3Cx+key%3D%220%22%3EKopesumma%3C%2Fx%3E&preserve_formatting=0&tag_handling=xml

And the following response was returned via the library:

{
    "translations": [
        {
            "detected_source_language": "SV",
            "text": "<x+key=\"0\">Copayment</x>"
        }
    ]
}

...which is incorrect! If I manually call the API and use JSON to post the data, it works as expected:

Input:

{
  "text": [
    "<x key=\"0\">Köpesumma</x>"
  ],
  "target_lang": "en-US",
    "source_lang": "sv",
    "formality": "prefer_less",
    "preserve_formatting": false,
    "tag_handling": "xml"
}

Response:

{
    "translations": [
        {
            "detected_source_language": "SV",
            "text": "<x key=\"0\">Purchase price</x>"
        }
    ]
}

...which is correct! I'm not sure if there is a problem with the library or if I'm using it incorrectly?

[JanEbbing]

Hm, if I var_dump the output of your code I get the correct result:

array(1) {
  [0]=>
  object(DeepL\TextResult)#19 (2) {
    ["text"]=>
    string(29) "<x key="0">Purchase price</x>"
    ["detectedSourceLang"]=>
    string(2) "sv"
  }
}

Looking at what you posted, the issue seems to be in your curl client: target_lang=en-US&source_lang=sv&formality=prefer_less&text=%3Cx+key%3D%220%22%3EKopesumma%3C%2Fx%3E&preserve_formatting=0&tag_handling=xml is lacking the ö in Köpesumma, and instead has an o. This should be correctly URL-encoded in HttpClientWrapper::urlEncodeWithRepeatedParams, but I'm not sure what would cause this to differ on your machine (Translating Kopesumma gives me Copayment as well). Can you install a different HTTP Client (eg Guzzle) and configure the translator to use that instead and try with that? It is described in the README.

[twarkie]

Very nice catch @JanEbbing! I will take a look to see why that happens. I do get different results on different machines but now I know where to look. Thanks.