by Jan Pedersen
Metaphors make up one of the classic translation problems, maybe even the most important one, according to the late translation scholar Peter Newmark. Still, they are severely under-researched in audiovisual translation (AVT). This struck me as odd, and I decided to do something about it.
Metaphors and AVT have a lot in common, and that complicates things for AVT, and particularly for subtitling, more than for other kinds of translation. Metaphors are verbal images, and images arguably make up the most central part of audiovisual texts, such as a movie. In a movie, you can visualize metaphors, for humorous or other purposes. And what do you do if that metaphor is not used in the target language? For instance, in Swedish, there is a metaphor-based idiom, “det är ingen ko på isen”, which expresses the notion that there is no need to panic. In direct translation, it would be “there is no cow on the ice”. Clearly, this is total nonsense in English (hence the above AI-generated image), and the standard first strategy would then be to replace it with a different metaphor with the same meaning, like “we’re sitting pretty”. But what are you to do, if there is an on-screen cow on the ice?
So, what did I do to rectify the scarcity of research on metaphors in AVT? Well, I did a lot of research, obviously, over a number of years. The first article I published on these issues is freely available online here. I then continued my research and last year, I published a full-length book on the matter. It is frightfully expensive, but if you can find someone to buy it for you, it is available here. That’s it for the ads, now let me tell you what I found out.
I investigated the Swedish subtitles of Yes, Prime Minister, The Great British Bake-Off and both the subtitles and the dubbed translation of two SpongeBob SquarePants movies. The reason for choosing this odd combination of material is that they all have a multitude of metaphors and come from different genres.
The general results were fairly unsurprising. If the same metaphors were used in the source and target languages, they tended to be translated directly. If not, a variety of strategies were used. This is similar to previous findings from other forms of translation. Overall, the metaphors tended to be translated directly, and if that didn’t work for some reason, a different metaphor or reduction to sense – i.e. replacing the metaphors with a non-metaphor, as in replacing “no cow on the ice” with “don’t panic” – was used, and on rare occasions metaphors were omitted.
I found an interesting difference between subbing and dubbing. Replacing a metaphor with another one was more common in dubbing than in subtitling, where metaphors were more often reduced to sense. I think this is due to differences in media constraints. In dubbing, the original dialogue is gone, so the viewer doesn’t know which metaphor has been used, whereas in subtitling, the original metaphor is still audible (and, for many viewers in the English-to-Swedish language combination, understandable) and then using a different metaphor would draw more attention to the subtitles than just saying what the speaker actually means.
What makes AVT different from other forms of translation is when the metaphor is visualized or present in some other way. That is when the subtitlers really come into their own and produce solutions that are amazing feats of creativity. In my material, the subtitlers were willing to go to great lengths, twisting and turning the words of the speakers and even language itself, to make sure that the subtitles don’t clash with the on-screen action. For instance, when the subtitler of Yes, Prime Minister replaced the English mixed metaphor “take the bull by the horns and grasp the nettle”; they not only tweaked the dialogue in the subtitles to introduce Swedish metaphors that matched the sense of the original, but the translation also matched the movements of the character, Bernard, who illustrated how difficult it would be to grasp a nettle while taking a bull by the horns.
In my material, there was just one case where the subtitler was stumped by the combination of dialogue and image. This was in The Great British Bake-off, where everybody started laughing when a female contestant with a beaver-shaped cake was greeted by the judges with the words: “Tell us about your beaver”. Since the “beaver” was clearly visible on-screen, the only solution was to render the non-metaphorical sense, and hope that the viewers would get the implied metaphorical sense from the reactions on screen. (Incidentally, a clip of this metaphor-based double entendre went viral and can be seen e.g. here.) I would say that cases like this are extremely rare, but cause nigh on unsolvable translation problems, and I will be talking about them at Languages and the Media in London on November 5, 2026, if you happen to be there. If not, I would like to point out that a successful way of rendering a metaphor is to ascertain whether it is used in both source and target language or if it is an original metaphor, in which case it can normally be translated directly. If not, I would suggest considering using a different metaphor with the same sense, if one is available and it doesn’t clash with the image. If that fails, one can normally reduce it to sense and render it using non-metaphorical language.
I was thus overall very impressed by the problem-solving skills displayed by the subtitlers when it comes to rendering metaphors in a functionally equivalent, effective, idiomatic, and easily accessible way. But this is 2026, and it is basically illegal to write a blog post about translation without discussing AI, so let’s do that. I carried out a small experiment in which I took an example from Yes, Prime Minister, where an old and barmy banker, Sir Desmond Glazebrook, gives his view on how the City would react to the appointment of a new manager of the Bank of England:
If you spill the beans, you open up a whole can of worms. How can you let sleeping dogs lie, if you let the cat out of the bag? Bring in a new broom and if you’re not careful, you’ll find you’ve thrown the baby out with the bath water. If you change horses in the middle of a stream, the next thing you know, you’re up the creek without a paddle.
The Swedish subtitler handled this tirade very well indeed, replacing it with a string of Swedish metaphor-based clichés or translating them directly when they work in Swedish This means that the translation was functional (i.e. funny) as well as idiomatic.
I asked four AI tools (Google Translate, Gemini, ChatGPT and DeepL) to do the same. The results were a bit mixed, but the overall impression is that AI cannot really handle metaphors. The first metaphor was reduced to sense by all of them (and thus not funny) and almost all the rest were translated directly, making the result very unidiomatic and strange indeed. The weirdest part was that ChatGPT lied to my face by claiming that its direct translations were all natural Swedish idioms, which most clearly are not. I’m sure the results would have been better by more precise prompting, but to do that would mean that AI created more, rather than less, work for the translator.
So, if you’re worried that AI is skilled enough in metaphor translation to take over your job, I can only say: there’s no cow on the ice!
Professor Jan Pedersen received his Ph.D. in 2007 with the dissertation Scandinavian Subtitles. He is co-editor of Benjamins Translation Library (BTL) and Journal of Audiovisual Translation, which he co-founded. His publications include the 2011 monograph Subtitling Norms for Television, and many articles on subtitling, translation and linguistics. His recent monograph Metaphors in Audiovisual Translation is available from BTL. He is former President of ESIST and former Director of The Institute of Interpreting and Translation Studies at Stockholm University, where he still teaches and researches audiovisual translation. He also worked as a subtitler for many years, and in 2023, he received FBO’s award Teksterprisen.
