Add support to translate files

[LukasWallrich]

I am trying to translate some files with the Google Translate API. I don't think that is currently supported - but would be a great option for gl_translate, as I have not found any R code to do it and am a bit daunted by the API docs ... https://cloud.google.com/translate/docs/advanced/translate-documents ... might that be possible?

[MarkEdmondson1234]

Can you read the files into R at all? Chunk up the text and send in

[MarkEdmondson1234]

But reading your docs, it doesn't look like a big change, since there is an option to upload to Cloud Storage already.

[LukasWallrich]

If I just want the text, it's indeed not difficult - but I need to translate the full files, so that the formatting remains (somewhat) intact as they contain tables. I did not understand how to send the file with the request, so I ended up using reticulate and the google.cloud.translate package. With that, the Python function is straightforward, partly copied from the documentation. If this is not a common need, obviously feel free to close it - maybe this code helps others who face this issue.

import google.cloud.translate as translate
import os

os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "key.json"
os.environ["GOOGLE_PROJECT_ID"] = "XXXX"

def translate_pdf(file_path: str, destination: str, target_lang: str = 'en', source_lang: str = ''):

    if not os.path.isfile(file_path):
        raise ValueError("Error: The file does not exist or is not a regular file.")

    if not file_path.lower().endswith('.pdf'):
        raise ValueError("Error: The file is not a PDF file.")

    client = translate.TranslationServiceClient()

    location = "us-central1"

    parent = f"projects/{os.environ['GOOGLE_PROJECT_ID']}/locations/{location}"

    # Supported file types: https://cloud.google.com/translate/docs/supported-formats
    with open(file_path, "rb") as document:
        document_content = document.read()

    document_input_config = {
        "content": document_content,
        "mime_type": "application/pdf",
    }

    response = client.translate_document(
        request={
            "parent": parent,
            "target_language_code": target_lang,
            "source_language_code": source_lang,
            "document_input_config": document_input_config,
        }
    )

    # To output the translated document, uncomment the code below.
    f = open(destination, 'wb')
    f.write(response.document_translation.byte_stream_outputs[0])
    f.close()

[MarkEdmondson1234]

Thanks! This is helpful

[dietrichson]

@MarkEdmondson1234 I was trying to browse through the documentation at this link:
https://code.markedmondson.me/googleLanguageR/
but get a 404. Have you moved it?

[MarkEdmondson1234]

There were two mirrored websites so bit confused why but this one is still live: https://docs.ropensci.org/googleLanguageR/

[dietrichson]

@MarkEdmondson1234 I tried the following:

my_file1 <- readBin(my_out, "raw", n=10000)
my_file2 <- readBin(system.file(package = "googleLanguageR","test-doc-no.pdf"), "raw", n=10000)
expect_equal(my_file1, my_file2)

This stops working pretty quickly. I suspect that the PDF produced gets time-related metadata added, so they won't be 100% equivalent. Even the file-size is different by a couple of bytes. I'll try to rework the test using pdftools - although this will add a package dependency.