This translates word by word

[wenchaoliu-93]

This would translate run by run (sometimes a run is a word), not sentence by sentence. I concatenated all the runs in a paragraph to make it translate better and faster. The drawback of my approach is that the format won't be preserved as well.
The below code starts from line 115.

            for index, paragraph_run in enumerate(paragraph.runs): # concatenate all the runs
                paraText += paragraph_run.text
            if len(paraText) > 0:
                try:
                    response = translate.translate_text(
                        Text=paraText,
                        SourceLanguageCode=source_language_code,
                        TargetLanguageCode=target_language_code,
                        TerminologyNames=terminology_names)
                    paragraph.clear()
                    run = paragraph.add_run()
                    run.text = response.get('TranslatedText')`

[wenchaoliu-93]

I fixed this in my fork: https://github.com/wenchaoliu-93/pptx-translator-bilingual/tree/master