Thursday, 7 September 2017

Naming Wines: Let's Use Science!

So the dream of letting Artificial Intelligence design wine names has come a fraction closer: I have been in correspondence with the excellent Janelle Shane, wondering if her neural networks (which have had such success generating paint charts and craft beers) might work with wines. Her take on it? At this early stage, yes, all things are possible. There is some nervousness about stepping outside the world of, basically, Anglo-Saxon nomenclature; so Australia, South Africa, New Zealand and America are going to provide the basic structures - although some snippets of French may eventually creep in. Also, to work at all properly the neural networks need lots and lots of existing names to learn from, hundreds of the things. Back to me, and the next question: How to create such a list?

After an hour of persistent thought, I only have one idea, which is to physically comb through the lists of the big suppliers and supermarkets, churning through their Australia, New Zealand, South Africa and North America entries by hand, copying and pasting until my head swims. Oh, and some English wines, why on earth not? Well, I can already see why not, given that it seems utterly counter-intuitive to cobble together a list of brand names painstakingly by hand, like some piece worker from the nineteenth century, only to deliver the fruits of this drudging manual harvest to a cutting-edge twenty-first century machine and have it instantaneously translated into The Future. I might as well go out and pick the sloes from the hedges, it'd be just as time-consuming and tedious, but at least I'd have some sloes at the end, which I could then turn into a drink. Although on the debit side of that idea, most of the sloe gin I've ever made has tasted terrible, so perhaps the comparison is less watertight than I think. And, on the other hand, if I do go through the list, name-picking by hand, then maybe, just maybe, Ms. Shane's miracle program will create something so new and vibrant that the whole world will be enriched, just a tiny bit, and God knows we need something to brighten our days. My sacrifice, I start to tell myself, will be for the good of mankind.

After that, of course, I experience another small crisis when it occurs to me that I will have to taxonomise the wines as I go, but what taxonomy to use? I mean, I could just do all reds and all whites and leave it at that, a more or less random selection. But in her terrific craft beer re-stylings, Ms. Shane has already pinned the beers down to categories such as IPA, Stout and so on. A mere shopping-list of wines names isn't going to be enough to keep her happy, I can feel it. So how to divvy them up in anticipation? By principal grape? By price? By sub-region (although that sounds unnecessarily granular to me, all that Barossa Valley stuff)? By style - robust, medium-bodied, lightweight? By reference to topographical feature as mentioned on the label (valley, creek, ridge, hill, river) or colour (yellow, silver, ink, limestone, unless that's topography) or animal (dog, eagle, kangaroo, bird, whale, horse, I could go on) or even simply alcoholic strength? And what about all those characterful place names which to some extent already sound like fabrications - Kangarilla, Waimea, Oxney, Boschendal - how do I insinuate them into the Big List? Would the craft beer experiment have got off the ground if the neural networks had been given Bofferding as a building-block? How, come to think of it, could anything sound less plausible than Wirra Wirra Church Block, currently available at Tesco?

No, hold on, the thing is, given the amount of dumb toil ahead of me, I want as little mental involvement as possible. I don't want to have to check the alcohol count or find out if it's an easy-drinking medium white or even make great inroads into the grape variety. So it's going to be country and that's that. All right, country and colour.

I make a start. After about fifteen minutes with an on-line supermarket wine list, I have twenty-seven reds. I would have done it in ten minutes, but for the fact that the formatting on the web page kept invading my basic document and I had to scrub out images of bottles and presumptuous text alignments. Assuming I get better at it: twelve minutes for every twenty-five names; that's a hundred and twenty-five names in an hour, not too bad. Two, two and a half hours should give me most of what I need, assuming that there are two hundred and fifty-odd different Anglophone wines out there. Oh, and then I'll have to check for duplications. Still. Two and a half hours. That's less than it takes to watch Andrei Tarkovsky's Solaris from beginning to end, so really, how hard can it be?

CJ


2 comments:

  1. I tried asking Wine Searcher for temporary access to their API (https://www.wine-searcher.com/ws-api.lml) in order to pull out lots of names (and possibly their taxonomy as well). They wanted money. You may be able to use your celebrity status to get some cooperation. If you can negotiate the access, Janelle (or I) will have the techy nous to pull the data.

    ReplyDelete
    Replies
    1. Ah yes, our celebrity status…

      (I think we'll let CJ do the graft.)

      Delete