Tim Atkin’s insightful post about terroir reminded me of one of my pet peeves in discussions about terroir—misguided expectations. He compares terroir to the practice of language translation, especially the generic translations coughed up by Large Language Models like ChatGPT
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 in a Mazis-Chambertin, a Cannubi, or an old-vine Barossa Shiraz.
Just as a good translator has to look beyond the literal meanings of words to capture the sub-text, the meaning beneath the words, in order to render the author’s intention, a winemaker making terroir-driven wines must translate soil, aspect to the sun, climate, and culture into a distinctive wine. But this is not a simple, straightforward process and the results are often surprising and uneven. Hence the complaints about terroir being “vague,” “subjective” or “a myth.”
Tim grudgingly gives into a bit of this skepticism before embracing the promise of uncertainty.
I agree with Jancis and Jamie; terroir is slippery. But for me, this slipperiness is precisely what makes it so interesting, so generative. Much will be lost in its translation, but attempts to do so are also ripe with possibility. Paying attention to terroir not only invites a more attentive, creative and thoughtful winemaking practice, but also recognises the importance of maintaining soil health, promoting biodiversity, and preserving culture and tradition. Terroir may not be scientifically proven, just as we cannot definitively prove what an author is saying “beneath their words”, but attempts to interpret it give way to the infinite possibilities of transformation and creation. These are worthy goals that, I think, are worth the risk of a little treachery.
But there is no treachery involved unless it’s the treachery of misguided expectations. We know that terroir is real because we know the flavors of grapes are influenced by weather, soil, and geography. What we can’t do is take that knowledge of the inputs to wine and generate rules for reliably producing output. In other words, winemaking cannot be reduced to a function or a formula. The inputs involve too many uncontrolled variables, too many feedback loops, non-linear causation, and interference effects to expect a clear “translation” of those inputs into reliably predictable outcomes.
However, the fact that we cannot use general knowledge to make accurate predictions about outcomes in particular cases does not mean that general knowledge is an illusion. We know that smoking causes lung cancer even though we can’t predict who will get it or when.
We timid humans always favor the certainty of black and white over the uncertainty of shades of grey. But in serious winemaking as well as translation its the nuances that do the work.
Exploring the Philosophy of Food and Wine 