I think of winemaking as a form of translation.
Product A (grapes) is transformed into Product B (wine). Very simple, really.
As you will know, traditionally, translation involves the conversion of a source text written in Language A into a target text written in Language B, meaning that the source text can now be understood and interpreted by readers of language B. It has become accessible through translation.
Of course, any translator will tell you that translation is much more complicated than this. Admittedly, large language models now mean anyone can translate anything into whichever language they choose in just a few seconds. Yet those who have actually used Google Translate, DeepL, or ChatGPT may have noticed that their offerings, while good, are less than perfect. Their translations are stilted and overly literal. A great deal is “lost” in translation, as the saying goes. More than that, these engines are frequently inaccurate; they misgender their subjects, make misleading grammatical errors, and perpetuate racial stereotypes.
To me, one of the greatest problems with these models is their inability to recognise the nuances of language, the subtext. If Google Translate were to make a wine, it would sit happily on the shelf next to Echo Falls, I Heart and Yellow Tail. It would be an affordable, serviceable product that communicates some of the flavours you might expect of a certain grape variety grown in a particular climate. But it would not have the subtlety, complexity and sense of place that you find
