On Thursday, January 26th we discussed the failings of machine translation and put Google Translate, https://translate.google.com DeepL https://www.deepl.com/en/translator and ChatGPT https://chat.openai.com through a series of challenges.
Translation fails pre-date machine translation of course. The following articles summarise some amusing (and expensive) mistakes:
Worst Translation Fails & Mistranslations: 20+ Hilarious Examples: https://translatepress.com/translation-fails-mistranslations-examples/
When a Global Journey Goes South: 10 Examples of Bad Translation: https://phrase.com/blog/posts/5-examples-localization-gone-tragically-awry/
Chris explained how today’s discussion was inspired by an e-mail she had received regarding the renewal of her digital certificate which Google Translate had automatically translated into English.
“We inform you that your digital citizenship certificate (NIF/NIE: ******8Y) issued on 02-12-2021 is in the renewal period. The definitive expiration will occur on 02-12-2024.
Note the expiry date – 2nd December 2024 Plenty of time to renew then … no hurry.
However, upon reading the original (in Valenciano and Castillian) she read this:
“Le informamos que su certificado digital de ciudadano (NIF/NIE: ******8Y) emitido en fecha 12-02-2021 se encuentra en período de renovación. La caducidad definitiva se producirá el día 12-02-2024.
Note the actual expiry date: 12th February 2024 (!) Whoops, better get it done soon!
The problem is that Google Translate has translated the date from European format (Day/Month/Year ) into US English format (Month / Day/ Year) . There is no option in Google Translate for UK English. Scotty remarked that he had noticed this with regard to a medical appointment.
We challenged Google Translate (a US product) Deepl (founded in Germany) and ChatGPT (AI) to translate a number of texts:
1. “It was a sad and disappointing day when I discovered that my universal remote control did not, in fact, control the universe, not even remotely.” Translate this into Spanish. This caused some confusion. Was “remote control” a mystical power (control remoto) or a gadget to control your TV? (mando a distancia) -In the former case the joke would translate. sort of. But not in the latter.
2. Time flies like an arrow, fruit flies like a banana. None of them got this …the translation involved flying bananas . The problem in both cases is the English!
3. We played around with this fake letter to the town hall translating it into Spanish then back translating.: “Dear Councillor, I am writing to ask you about the lighting in our urbanisation. The lights are not working and the residents are very angry.” Well, to be polite you have to know if the councillor is male (concejal) or female (concejala). Should “angry” be enfadados ( “angry” as suggested by Google and Deepl) ? or given the context, the less extreme molestados (“upset” – ChatGPT ) . “Residents” translated as “vecinos”, which translated back as “neighbors” – aaagh American English. although with DeepL you have the option to use English English ! “Urbanisation” becomes a “development” with back translation in Google.
3 We tried some Vietnamese in Google and ChatGPT: “Giao dịch dự kiến thực hiện từ ngày 29/1 đến ngày 27/2 theo phương thức khớp lệnh hoặc thỏa thuận.”
Google said: The transaction is expected to be carried out from January 29 to February 27 by order matching or agreement method.
ChatGPT said: The transaction is expected to take place from January 29 to February 27 through matching orders or negotiation.
We noted that the date format seems to have translated correctly in this case, and that ChatGPT read more fluently. (Deepl can’t handle Vietnamese yet)
We looked through the following article which compares different translation engines for accuracy: Research vs Practice: How Accurate Is Google Translate? https://phrase.com/blog/posts/is-google-translate-accurate/ Don’t even attempt to use them for Armenian! We noted that the article mentioned Systran https://www.systransoft.com/translate/ This is one of the early translation tools and is probably worth looking at again.
We noted that Samsung’s latest Galaxy phones offer live translation over phone calls, texts https://techcrunch.com/2024/01/17/samsungs-latest-galaxy-phones-offer-live-translation-over-phone-calls-texts/ Chris remarked that there are two opportunities for bloopers to be created and multiply..: The phone mis-interpreting the voice or text input, and then mis-translating it! We have to remember that Garbage in leads to Garbage out. Badly written text will be translated badly and a message isn’t only made of words it also contains context, intention and non-verbal components. Far from what AI techies would like you to believe, the job of Copy-editor is not going to go away soon!
Chris Betterton-Jones – Knowledge junkie