Our discussion topic for Thursday, March 26 was: Translating and writing with AI – we’ll all be sounding like robots soon
AI large language models like ChatGPT are getting better at translating as well as writing essays and summaries. But what are the downsides? ..we will play around with some text using various AI models, compare their outputs and discuss how AI English differs from Human English and what the implications are.
Translating with AI:
Internet articles:
We looked at the English version of Jávea.com. This site is clearly machine translated, without editing. It is clunky and sometimes weird to read. https://en.javea.com/la-historia-oculta-del-faro-del-cabo-de-san-antonio-de-xabia/?u We copied some of its Spanish text and asked ChatGPT to translate it into English. The results were similar to the site’s own translation, so there was nothing remarkable there. As a side note, members said that the specialist AI translation app, DeepL, https://www.deepl.com/en/translator seems to give particularly good results. Its free for short texts and there’s a browser plugin.
Literature:
Chris compared the official English translaton of the modern Spanish novel, La Sombra del Viento, by Carlos Ruiz Zafón, with translations generated by ChatGPT and Claude. Here is the opening paragraph to compare:
>Original Spanish
Todavía recuerdo aquel amanecer en que mi padre me llevó por primera vez a visitar el Cementerio de los Libros Olvidados. Desgranaban los primeros días del verano de 1945 y caminábamos por las calles de una Barcelona atrapada bajo cielos de ceniza y un sol de vapor que se derrama sobre la Rambla de Santa Mónica en una guirnalda de cobre líquido.
>Chat GPT
I still remember that dawn when my father took me for the first time to visit the Cemetery of Forgotten Books. It was the early days of the summer of 1945, and we walked through the streets of a Barcelona trapped beneath ashen skies and a vaporous sun that spilled over the Rambla de Santa Monica like a garland of liquid copper.
>Claude
I still remember that dawn when my father took me for the first time to visit the Cemetery of Forgotten Books. The first days of the summer of 1945 were slipping away, and we walked through the streets of a Barcelona trapped beneath ashen skies and a vaporous sun that spilled over the Rambla de Santa Monica in a garland of liquid copper.
>Human Translator: Lucia Graves
I still remember the day my father took me to the Cemetery of Forgotten Books for the first time. It was the early summer of 1945, and we walked through the streets of a Barcelona trapped beneath ashen skies as dawn poured over Rambla de Santa Monica in a wreath of liquid copper.
Although the AIs translations were technically correct, they lacked the visual poetry of the human translation. Translating literature is an art, since the translator has to understand the emotional content of the source material and then re-write it in the target language, rather than translating word by word or phrase by phrase.
Cockney slang:
We then had a bit of fun asking AI models to translate Cockney rhyming slang and Christmas cracker jokes. Here are some of our results!
Prompt: Translate this into Spanish: The mother said to her son, “Up the apples and the pears” The boy said, “I want to take a butchers at my phone first”.
>Chat GPT
La madre le dijo a su hijo: «Sube por las escaleras».
El chico respondió: «Quiero echarle un vistazo a mi móvil primero».
It understood the slang and translated the meaning into Spanish.
>Deep Seek
La madre le dijo a su hijo: “Arriba las manzanas y las peras”.
El niño dijo: “Primero quiero echar un vistazo a mi móvil”.
It translated “apples and pears” literally, but understood that “butchers” meant “look”
>Claude
Here is the translation, with an important note:
La madre le dijo a su hijo: “Sube las escaleras.”
El chico dijo: “Primero quiero echar un vistazo a mi teléfono.”
A note on the translation: Both lines contain Cockney rhyming slang, which has no direct Spanish equivalent and cannot be translated literally:
“Apples and pears” = stairs
“Butcher’s (hook)” = look
Since Cockney rhyming slang is a purely English cultural phenomenon, I translated the meaning rather than the slang itself. If you wanted to capture the flavour of colourful, working-class vernacular in Spanish, you could use informal or regional slang — for example, “echar un ojo” or “pegarle un vistazo” have a casual, everyday feel — but there is no true Spanish equivalent of the Cockney device itself.
Claude was brilliant! It then asked “Would you like a version that tries to mimic the playful, coded feel using Spanish slang?“
Christmas Cracker Joke:
Prompt: Translate this Joke: How much does Santa have to pay to park his sleigh?
Nothing. It’s on the house!
> Chat GPT
¿Cuánto tiene que pagar Papá Noel por aparcar su trineo?
¡Nada, es gratis!
>Deep Seek
Here’s the translation of the joke into Spanish:
¿Cuánto tiene que pagar Santa para estacionar su trineo?
Nada. ¡Cortesía de la casa!
Note: The pun “on the house” (free, paid for by the establishment) is preserved in Spanish with “cortesía de la casa,” which conveys the same double meaning: it’s free parking because it’s “on the house” (literally, courtesy of the house/establishment).
>Claude:
Here is the translation:
¿Cuánto tiene que pagar Papá Noel para aparcar su trineo?
¡Nada! ¡Está en el tejado!
This joke actually works beautifully in Spanish — perhaps even better than in English. In English, “on the house” is an idiom meaning free of charge, but it also literally means on the house, where Santa’s sleigh lands. In Spanish, “está en el tejado” (it’s on the roof) delivers the visual punchline directly and naturally, and Spanish speakers will get the joke immediately since everyone knows Santa lands on the roof.
The only small trade-off is that Spanish lacks a perfect equivalent of “on the house” as an idiom for something free — the usual expression would be “invita la casa” — so the joke relies purely on the literal image rather than a double meaning. But it lands cleanly all the same!
Claude wins again!
Writing with AI:
There is growing evidence that using AI’s Autocomplete to finish sentences in e-mails and documents erases your personal style. https://theconversation.com/with-ai-finishing-your-sentences-what-will-happen-to-your-unique-voice-on-the-page-276036 : Quote:
“Because generative AI composes and suggests text in highly standardized, predictable patterns, its outputs can read as if they’re dressed-up versions of what linguists call “phatic expression.” These are the overly common phrases that function as social glue more than as conveyors of sentiment: “How are you?”, “Have a good day” or “See you soon.” But this glue can lose its hold if the technology is used in the wrong situations. Using artificial intelligence to compose a social media post in the wake of a tragedy, or using it to write a fan letter to an Olympic athlete, comes off as insincere. People are starting to catch on to generative AI’s prose, not because it’s clunky or poorly written, but because it all sounds the same. That’s because large language models are trained on gigantic masses of examples of human writing, and they predict text based on probabilities and commonalities.“
AI English has its own characteristics which which differ from the English written by human beings: https://theconversation.com/how-ai-english-and-human-english-differ-and-how-to-decide-when-to-use-artificial-language-277455 Quote:
“Human English contains persistent, if subtle, linguistic patterns of variation and readability. By contrast, AI uses what (one might) call exam English – a rather formal, dense English that is favored in academic tests and papers. It is less varied and less readable. People perceive it as robotic, but they also perceive it as smart... “AI English favours what exam English has always rewarded: homogeneity and information density. And thus far, instructional tuning – training AI models to follow human instruction – only makes AI English less like human English. It doesn’t help that AI writing is part of what AI bots train on. At its best, AI English is a language database driven by statistics. It’s big, but it’s canned. History tells us that a full range of global human English gives people the greatest possibilities for expression and connection.
When AI is used to write novels and stories, it seems to overdo literary tropes and devices, e.g as described in this FanFiction Reddit :https://www.reddit.com/r/FanfictionNet/s/9QnTCfjTHl Here are some quotes:
THE AMPLIFICATION ECHO
You wrote a thing. Then you wrote it again, but harder.
“He did not hurry. He never did.” “She was not afraid. She had never been afraid.” “It did not work on him. Very little did.”
The test is simple: does the second sentence contain any information not already in the first? If the answer is no, you’ve been amplified.
SILENCE AS PUNCTUATION
“Silence settled over the room.” “No one spoke.” “The room fell silent.” “A long silence followed.”
Count these in your text. If you have more than two per scene, your characters are spending more time not talking than talking. AI uses silence as a transition the way bad PowerPoints use fade-to-black. The reader doesn’t need to be told the room is quiet after every significant statement. If the statement is strong, the silence is implied.
THE MASTER TEST
Read your text aloud. If every paragraph sounds like it was written by the same person in the same mood, something has gone wrong. A grief scene should not have the same rhythm as a comedy scene. A fight should not have the same rhythm as a political negotiation. If your text has been AI-edited, the most likely symptom is rhythmic monotony — every scene given the same weight, the same fragment patterns, the same solemn pauses, the same trailing wit.
Here’s a prediction: Generative AI will never win the Booker prize for literature! ( nor me neither)
Christine Betterton-Jones – Knowledge Junkie