…with just a hint of Naive Bayes in the nose

« previous post | next post »

Coco Krumme, "Velvety Chocolate With a Silky Ruby Finish. Pair With Shellfish.", Slate 2/23/2011:

Using descriptions of 3,000 bottles, ranging from \$5 to \$200 in price from an online aggregator of reviews, I first derived a weight for every word, based on the frequency with which it appeared on cheap versus expensive bottles. I then looked at the combination of words used for each bottle, and calculated the probability that the wine would fall into a given price range. The result was, essentially, a Bayesian classifier for wine. In the same way that a spam filter considers the combination of words in an e-mail to predict the legitimacy of the message, the classifier estimates the price of a bottle using its descriptors.

The analysis revealed, first off, that "cheap" and "expensive" words are used differently. Cheap words are more likely to be recycled, while words correlated with expensive wines tend to be in the tail of the distribution. That is, reviewers are more likely to create new vocabulary for top-end wines. The classifier also showed that it's possible to guess the price range of a wine based on the words in the review.

Winetalk was a LL theme for a while a few years ago:

"The legal treatment of quantifiers", 1/11/2004
"Just a trace of the obligatory rubber", 4/9/2004
"Editor impresses", 5/4/2004
"Ritual verbal enthusiasm for food", 5/11/2004
"Modification as social anxiety", 5/16/2004
"More winetalk imports into coffee lingo", 5/24/2004
"Grand Cru smackdown", 6/2/2004
"More on winetalk culture", 6/2/2004
"Apologia pro risu suo", 6/2/2004
"What do wine tasting notes communicate?", 6/5/2004

And again, for a bit, last year:

"Two brews", 2/6/2010
"X forward", 3/12/2010

But it didn't occur to me that an online compendium of wine-tasting notes with prices and ratings would be a lovely subject for an after-hours Breakfast Experiment™.


  1. army1987 said,

    February 23, 2011 @ 8:04 pm

    I've read lots of nonsense in wine reviews, but “graphite”? For f***'s sake.

    I've heard that no-one seems to be able to tell expensive wines and average-priced wines in blind tasting tests, can they?

  2. Jangari said,

    February 23, 2011 @ 8:19 pm

    I enjoyed that article. Some of my friends are wannabe wine connoisseurs who rabbit on about their wine with such descriptors as Coco points out. Some of them redefine ludicrous – I heard once a friend genuinely describe a shiraz as 'melancholy'.
    Some descriptors sound totally fair enough; particular fruits and berries, comments on tannins (velvety, grippy, etc.) and body (light, rich, etc.) but I think there's a lot of herd mentality out there, a well as blind concordance. I'll admit my palate is less finely tuned than others', but I notice when sharing a bottle of wine, if someone says 'yeah, pears and boysenberries', it's likely to trigger that connection in my head and I'll start to look for (and eventually find) those flavours – even if they're not there.
    I once saw an experiment, where white wine was coloured various shades of red, and I think red wine was decoloured somehow, and the tasters (all chefs and foodies) all gave classic 'red' descriptions to the red-coloured white wines (and vice versa). So much of our flavour experiences are confabulated from context and suggestion. Just eat, drink and be merry.

    [(myl) Pointers to some of the experimental literature on this topic are here and here (also above in the list of links).]

  3. WillSteed said,

    February 23, 2011 @ 8:58 pm

    It seems to me a deliberate bit of marketing. The reviewers know that they're reviewing an expensive bottle of wine. The people who are going to be buying the more expensive bottles of wine want to think that they're experiencing something special, so they use words that make it sound more special.

  4. Peter said,

    February 23, 2011 @ 10:44 pm

    Upon tasting this bottle of Night Train Express, vintage 2011, there were strong notes of gasoline, varnish and embalming fluid, with undertones of ammonia, antifreeze and Liquid Plumber.

  5. Charles Gaulke said,

    February 23, 2011 @ 11:38 pm

    Hm… I think WillSteed is onto something. We ought to make people review wines "blind", and then revisit these word distributions in a year or two.

    Is there some formally described linguistic phenomenon that covers something like the adoption of "graphite" as a flavour descriptor? I've never been able to find an actual description or definition of what it's actually supposed to mean, but it seems widely accepted and, looking at some wine books, it's been around for quite a while.

    [(myl) If you're interested in this, you could start by checking out Venetia L. Joscelyne et al., "Partial Shading of Cabernet Sauvignon and Shiraz Vines Altered Wine Color and Mouthfeel Attributes, but Increased Exposure Had Little Impact", Journal of Agricultural and Food Chemistry 55(26) 2007, which used a panel that "underwent 5 weeks of high-level training in aroma, taste, and trigeminal (mouthfeel) sensation detection and evaluation":

    Six 1 h sessions were held over 5 weeks to train the panel. During training, panelists were presented with a selected replicate of each experimental wine treatment in coded, covered, XL5 (ISO standard), 215 mL tasting glasses and asked to individually generate and then reach panel consensus on appropriate descriptive terms for each variety. Descriptive terms were narrowed to three color, eight aroma, four flavor (where flavor is defined as aromas by mouth), two taste, and three mouthfeel attributes for Shiraz and two color, eight aroma, four flavor, three taste, and three mouthfeel descriptive terms for Cabernet Sauvignon.

    Their "Aroma and Palate Attribute List with Agreed Definitions" includes "pencil shavings", defined as "shavings of a gray-lead or graphite pencil", and a footnote indicates that "Reference aromas that match the definition were provided".

    From their principal components analysis of the experimental results, you can see that the placing of "pencil shavings" on the first dimensions of the PC space (which accounted for ~90% of the variance) was not very different from the placing of (say) "plum".

    And in their tasting results, you can see that the distribution of the "pencil shavings" attribute differed very significantly among tasters (meaning that some of them used it more than others) but not among wines (meaning that it had no intersubjective value as a descriptor). It's not clear whether this is because the tasters had different chemical receptors in this case (possible), or because they were using the term in an idiosyncratically evocative way without any particular physiological referent, or perhaps because the wines tested didn't really differ in this attribute.]

  6. rone said,

    February 24, 2011 @ 4:02 am

    Eric Asimov, Wine in Two Words: http://mobile.nytimes.com/article?a=748958&f=32

  7. richard howland-bolton said,

    February 24, 2011 @ 7:33 am

    à propos army1987's doubt. I have no opinion about claims that oenophiles can tell which side of the mountain the grapes were grown or if Pierre-the-Grape-Stomper's foot rot is better, but I do have a personal non-evidential anecdote.
    Many years ago I was a food chemist, working for a company that wanted to learn a lot about the chemistry of wine in a short time, part of my responsibility was checking the sugar levels as fermentation proceeded by analysis and by a 'quick check method' that used diabetic's blood sugar strips (CliniTest, if I remember) I would also taste the sample (Yes! Taste: a sip not a swig! And I'm sticking to that story!!) and after a couple of months I found that I was as good as the quick test. Of course I had almost immediate feedback to help me, but it does make me wonder about the sensitivity of the human palate.

    [(myl) Evaluation of sugar content by taste is a highly evolved (because highly valuable) ability — see e.g. Mattias Laska et al., "Taste Difference Thresholds for Sucrose in Two Species of Nonhuman Primates", Am. J. Primat. 1999. Evaluation of the physiochemical basis of descriptors like "pencil shavings" is less, shall we say, ecologically essential.]

  8. Neal Goldfarb said,

    February 24, 2011 @ 8:50 am

    And let's not forget this:

    A lot of people in this country pooh-pooh Australian table wines. This is a pity as many fine Australian wines appeal not only to the Australian palate but also to the cognoscenti of Great Britain.

    Black Stump Bordeaux is rightly praised as a peppermint flavoured Burgundy, whilst a good Sydney Syrup can rank with any of the world's best sugary wines.

    Château Blue, too, has won many prizes; not least for its taste, and its lingering afterburn.

    Old Smokey 1968 has been compared favourably to a Welsh claret, whilst the Australian Wino Society thoroughly recommends a 1970 Coq du Rod Laver, which, believe me, has a kick on it like a mule: eight bottles of this and you're really finished. At the opening of the Sydney Bridge Club, they were fishing them out of the main sewers every half an hour.

    Of the sparkling wines, the most famous is Perth Pink. This is a bottle with a message in, and the message is 'beware'. This is not a wine for drinking, this is a wine for laying down and avoiding.

    Another good fighting wine is Melbourne Old-and-Yellow, which is particularly heavy and should be used only for hand-to-hand combat.

    Quite the reverse is true of Château Chunder, which is an appellation contrôlée, specially grown for those keen on regurgitation; a fine wine which really opens up the sluices at both ends.

    Real emetic fans will also go for a Hobart Muddy, and a prize winning Cuivre Reserve Château Bottled Nuit San Wogga Wogga, which has a bouquet like an aborigine's armpit.

  9. What’s the Point, you Ask? | You Had Me At Merlot said,

    February 24, 2011 @ 9:45 am

    […] Interesting article (Click Here) […]

  10. Sybil said,

    February 24, 2011 @ 4:20 pm

    How do I love the title of this post? Let me count the ways.

    Whereas I was so sad that you did not entitle this one "What does 'even' even mean?":

    I see that "even" is making a reappearance in the newest post, so off to read…

    [(myl) I've adopted your suggestion for the title of the cited post.]

  11. maidhc said,

    February 25, 2011 @ 3:52 am

    I imagine Sir Les Patterson may have had some further suggestions, although the only one I remember is "Australian Wine-Tasting … BYOB!"

  12. Dick Leed said,

    February 26, 2011 @ 9:15 pm

    Flavor names are just tags. Tasters could just as well use numbers. It doesn't matter whether or not the flavor name for wine matches the flavor of the analogous non-wine substance. I started raising black currants a few years ago and gave some to my son, a wine buyer. He said, "Aha! so that's what black currants taste like!" Although he had never tasted them before, he had been using and reading the word as applied to wine for years. He found it interesting that the flavors matched, but it was unimportant. What was important was that one wine be similar to another with respect to a particular flavor, what ever the name for it might connote in the outside world.

  13. drinksnob said,

    February 27, 2011 @ 9:58 pm

    I don't think the science cited in this article actually backs up her conclusions, and, in fact, is a dreadful under-representation of the literature on the subject. The majority of which (lazy search) in fact supports clearly the possibility of spectrometric differentiation of varietals. In addition, since the current understanding of how odorants combine synergistically to form flavor in a food matrix (that is, when not fed individually into a detector from a gas chromatograph) is primitive at best, the sensory evaluation literature (lazier search) is probably more relevant.

    I'm surprised no-one is calling out this article for its use of capital-S Science for an appeal-to-authority argument.

    PS – Related.

  14. Robert Howells said,

    February 28, 2011 @ 3:23 pm

    army1987 said,
    February 23, 2011 @ 8:04 pm

    I've read lots of nonsense in wine reviews, but “graphite”? For f***'s sake.

    I've heard that no-one seems to be able to tell expensive wines and average-priced wines in blind tasting tests, can they?

    I have experienced "pencil shavings" in a wine that had a high percentage of Cab Franc in the blend.

  15. Dave T said,

    June 18, 2011 @ 11:44 pm

    As Robert suggests, 'graphite' is just a pretentious way of saying 'pencil lead'. It's a flavor many people are familiar with.

    (And I believe Perth Pink is a wine for laying down and voiding, not 'avoiding'…)

RSS feed for comments on this post