The evolution of the cocktail

« previous post | next post »

A note in "Random Samples" in the July 9 Science relates how in graduate school, evolutionary biologist James Harriman

wondered whether [quirks of personal taste in drinks] evolve into popular cocktails much as mutations give rise to new species, through a sort of taste-based natural selection.

So Harriman, now a visiting scientist at Cornell University, fired up a computer program for generating phylogenetic trees. Instead of genes, he plugged in the ingredients of 100 cocktails, taking vodka as the tree's common ancestor. The program divided cocktails into several distinct families–drinks based on champagne or Irish cream , for example, or punch bowl drinks … A poster of the tree, which doubles as a mixology guide, is available online [for $20] from ThinkGeek.

Such programs do phylogenetic reconstruction based on the Darwinian assumption of descent with modification from a common ancestor. The trick is in the mathematics, of course, but otherwise this is the program of comparative reconstruction suggested to Darwin by the achievements of 19th-century historical linguistics (and ultimately traceable back to the reasoning used by philologists in studying manuscript descent), though in these other applications there is usually no stipulating the common ancestor (vodka in the cocktail case).

The poster as reproduced in Science is too small to be legible, and the one on the ThinkGeek site is, if anything, even less clear. Going to the piece on the sciencemag site doesn't help, and there are no further links there, so for the moment all I know is what I just told you.

I did find this statement by Jim Harriman on the Radaris site (presumably from LinkedIn), which tells you something about his research at Cornell, but nothing about the cocktail project:

I use bioinformatic algorithms to find the genes responsible for valuable traits in wine grapes. We can then test for these genes at each generation of a grape breeding program, so that the plants can be bred faster and more accurately. Ultimately, we will be able to grow new varieties that are resistant to disease, drought, or climate change. They will also be 100% organic, created only by conventional breeding.

I pursued these things only because of a (distantly) potential ambiguity in "the program divided cocktails into several distinct families". I'm sure the intended interpretation is that in the output of the program, cocktails were divided into several families, but it's also possible to read the clause as saying that the input of the program had the cocktails divided into families.

In any case, cultural artefacts (like cocktails, or for that matter lexica and manuscripts) are, in principle, just as eligible for comparative reconstruction (including subgrouping) as natural entities (like species as determined by "characters" or by genes) are — though I have no doubt that there are many difficulties and pitfalls in carrying over the procedures of reconstruction from one case to another. But none of this is an area of expertise for me; I'm just an interested outsider.


  1. Freddy Hill said,

    July 20, 2010 @ 1:09 am

    Hamilton or Harriman? I cursorily tried looking them up both at Science's site and at ThinkGeek and came up empty… Do you have a better link?

    [Ack. Harriman. A typo now corrected. I don't know why {Harriman} doesn't pull up anything — it didn't for me, either — but {cocktails} works at both sites.]

    You see, while you are an interested outsider when it comes to cocktail evolution, I'm definitely an interested drinker.

  2. HP said,

    July 20, 2010 @ 1:43 am

    There's actually a wealth of historical data that could be used to produce a detailed taxonomic tree of distilled alcoholic beverages stretching back to the invention of the alembic, if one had the time and inclination. It might be interesting to compare such a tree to one generated from existing cocktails, much as biologists compare phylogenetic to molecular trees.*

    One issue with a tree such as the one described here is that it would inevitably leave out entire classes of fossil cocktails, such as the egg-white-based Flips (of which only a small, relict population of eggless Tom and Jerries survive). I refer of course to the extinction event of 1920, which nearly decimated not only flips, but toddies, punches, and rye-based cocktails, all of which are severely threatened if not extinct in the wild.

  3. Dean said,

    July 20, 2010 @ 3:40 am

    There's a more legible copy of the poster here:

  4. Ginger Yellow said,

    July 20, 2010 @ 4:17 am

    I refer of course to the extinction event of 1920, which nearly decimated not only flips, but toddies, punches, and rye-based cocktails, all of which are severely threatened if not extinct in the wild.

    Ah yes, the Wilson-Harding boundary event.

  5. Ken Brown said,

    July 20, 2010 @ 4:55 am

    There are relict populations of egg-based drinks in the Netherlands and Belgium, with some history of occasional irruptions into the British Isles – most recently in the 1970s if I remember correctly.

    And saying "the Netherlands and Belgium" feels odd because I know "Netherlands" is, or was, "the low countries" in English and when our ancestors used to talk about the "The Low Countries" they included what is now Belgium, or most of it & some of France and Germany as well. So there is a built-in redundancy in the phrase.

  6. Mark Etherton said,

    July 20, 2010 @ 6:19 am

    The field work behind the table looks flawed: for example, the proportions in the Dry Martini recipe are clearly wrong and there is no brandy in the Champagne Cocktail. Nor are drinks with egg white in them anything like extinct, since the White Lady flourishes in Central London.

  7. John Cowan said,

    July 20, 2010 @ 7:51 am

    The group of algorithms in question produce a tree without specifying its root, much like one of those mobiles made of glass rods tied together which can be hung up from any point. Therefore, it's necessary to choose a root based on other considerations. The wonderful Ringe, Warnow, Nakhleh, and Harris tree of the Indo-European families, for example, is rooted between Anatolian and the rest of IE on known historical and geographical grounds, not as a consequence of the algorithm itself.

    The methodology of the IE tree is built on classical comparative data rather than by attempting an end-run around the comparative method. Thus although the Classical Armenian erku does not much resemble the other Indo-European words for 'two', it is known to be regularly derived from the same origin, and therefore is treated as the same character — indeed, all 24 languages in the database share that character, and so 'two' is not useful for tree reconstruction. Similarly, known occurrences of borrowing and parallel development have been removed from the input. The contrast to certain other studies much discussed here at the Log could hardly be more marked.

    The outputs of the study include not only the tree itself, with its mixture of familiar conjectures confirmed and previously unsuspected similarities, but also evidence for several episodes of hitherto undetected borrowing between families in the remote past.

  8. Danny Bloom said,

    July 20, 2010 @ 8:25 am

    Funny, but here in Taiwan, cocktail is called in Mandarin as "tail of a chicken" directly from the English….

  9. Mr Fnortner said,

    July 20, 2010 @ 8:40 am

    The entry in the tree for Miscellaneous is curious. In natural beings, it would be equivalent to placing the platypus, star-nosed mole, thorny devil lizard, and the narwhal in an all encompassing family because they are uniformly unusual.

  10. Mark Liberman said,

    July 20, 2010 @ 8:45 am

    John Cowan: "The group of algorithms in question produce a tree without specifying its root, much like one of those mobiles made of glass rods tied together which can be hung up from any point."

    This is true of some algorithms, and of some traits used as input, and not of others. Some changes are not intrinsically directional — a vocabulary item might either be lost or gained, for example. But others are pretty much a one-way street — palatalization of velars before /i/ and /e/ is a common change, for example, but it would be unexpected to find that change going in the opposite direction.

    Harriman's little "cocktail phylogeny" joke does seem to be one where the directionality of change needs to be specified externally. A deeper problem is that it appears to involve some obviously false assumptions about the underlying process of change. The idea that ingredients are evolutionary traits, for example, requires that brandy-based cocktails must have originated by deleting vodka from a recipe and then adding brandy to it, or carrying out a sequence of deletions, insertions and substitutions with the same effect.

    This is silly enough, and would be just as silly taken in the other direction. But the idea of taking vodka as the root — basically just alcohol and water — suggests an even sillier model of cocktail history, in which the chemical constituents of ingredients are taken as separate traits, so that brandy would have arisen by a process of adding grape flavorings to alcohol made from grain or other starches, and that wine must have been derived from brandy by adding additional water. This is clearly foolish, but it would be equally silly to insist that vodka was historically derived from brandy by the removal of grape-related impurities.

    This underlines the argument, made most strongly by Don Ringe, that such algorithms need to be grounded in a realistic understanding of the historical processes being modeled.

    Harriman's tree is not necessarily complete nonsense — as Russ Gray is fond of saying, adapting Picasso, "A model is a lie that leads us to the truth". But it would be wise to reserve judgment until we've had a look at some publication whose methods section gives more detail than we get from a 175-word note in Science and a ThinkGeek poster of the results.

  11. Doctor Science said,

    July 20, 2010 @ 9:28 am

    The problem that jumped out at me immediately is "taking vodka as the tree's common ancestor".

    Ah, youth. Harriman obviously doesn't realize that vodka is a comparatively recent player on the cocktail scene — the only common vodka-based cocktails as recently as the mid-1970s all had "Russian" in their names.

    Now that I see the legible version of the poster (thank you, Dean!) I am struck by how completely ahistorical the results are. Some extremely ancient drinks such as Egg Nog and Sangria have shallow roots; some modern drinks such as the Chocolate Martini have very deep roots.

    This is, in fact, *not* an evolutionary tree, it is a cladistic tree. It's *possible* that choosing a more reasonable root than vodka would give more historical results — personally I doubt it, but science demands that someone check. I think, though, that this tree could be a valuable textbook example of the difference between cladistic trees and evolutionary trees.

    It may also be a useful example for refuting "intelligent design". Harriman's tree exhibits the sort of pattern we expect with intelligent designers, such as bartenders. Evolutionary trees both exhibit and reflect history.

    [(myl) My own guess is that the "arbitrary root" problem is a small part of this, with the larger part of the problem being the assumption that the sole developmental process is adding, subtracting, or substituting ingredients (or chemical constituents of ingredients).

    You may be right that historical processes involving some measure of intelligence (and thus cultural history in general) pose special problems for phylogenetic algorithms. But there are lots of simple physical processes where a model based on the insertion or deletion of isolated "traits" (on some arbitrary definition of "trait") doesn't fit the facts very well. A lot of scientific research can be seen as a search for invariants that allow complex, entangled processes to be factored into separable pieces with tractable interactions.]

  12. Terry Collmann said,

    July 20, 2010 @ 9:40 am

    There are relict populations of egg-based drinks in the Netherlands and Belgium, with some history of occasional irruptions into the British Isles – most recently in the 1970s if I remember correctly.

    Ah yes, the Snowball – possibly so primitive a cocktail form it dates back to the time of Snowball Earth.

  13. JimG said,

    July 20, 2010 @ 10:23 am

    I second Mark Etherton's thought about poor fieldwork. Did author Harriman use a bartender's manual as a source? What were the effects of agriculture and trade? Sangría is a cocktail of alcoholic liquids and juices, as is a planter's punch. Tom and John Collins had a brother Ron. I intuit a closer connection than shown, for daiquiris, margaritas, caipirinhas and mojitos.

  14. Arnold Zwicky said,

    July 20, 2010 @ 10:43 am

    Several follow-up comments.

    First, yes, all the tree gives (or appears to give) is empirically-derived taxa. The labeling of these is a kind of interpretation (and results in the bizarre Miscellaneous taxon), and a historical account of the relationships is another massive act of interpretation — really, the relationships can't be interpreted without having some access to an independent source of evidence about history, as ML stresses.

    Second, I had appreciated that the project was light-hearted, but I hadn't entertained the possibility that some or all of it was simply a joke. Now I see how difficult it is to figure out what's intended seriously and what's simply invented; the "Random Samples" piece is absolutely no help here (maybe I should start assuming that everything on the page is a spoof, and just dismiss everything I read there; if so, it should be labeled Joke Page and not pretend to be reporting on science).

    In fact, I now wonder whether "James/Jim Harriman" is a real person or an invention. (See what happens when you begin to suspect that the well is poisoned?)

    The better version of the poster, which can be blown up, brought me to the Spaghetti Logic site, whose address can just barely be made out in the bottom right corner of the poster. There I found a posting — not attributed, but clearly to be understood as coming from "James Harriman" — "The Phylogeny of Cocktails", dated September 4th, 2009, which certainly should have been linked to in the "Random Samples" piece, if we're to take that even 10% seriously. It has the poster, the link to ThinkGeek and its order form, and this (linkless) text (in its entirety):

    This chart was generated by taking a list of cocktail drinks, representing each ingredient as a different gene, and then using a computer to generate a phylogenetic tree. Over Christmas 2008 I was thinking about how a lot of drink recipes have similar ingredients, and probably developed when someone modified an existing recipe. Take, for example, a Tom Collins and a John Collins, which are identical except that the John Collins uses bourbon instead of gin. I wondered if it was possible to reconstruct the phylogeny of the drinks. It turns out this is easier than one would think. The PHYLIP computer program can create family trees based on presence or absence of a trait (0 or 1), and it doesn’t matter what you consider a "trait". I collected about 90 recipes and used PHYLIP’s pars utility, with default options, to group them into families based on their 512 unique ingredients. The tree was generated with PHYLIP’s drawgram utility and cleaned up in Inkscape. The tree should technically be unrooted, but I rooted it on vodka since so many of the drinks contain vodka. The cool thing about grouping the drinks this way is that they form families with similar ingredients. So if you like martinis, you can look in the "gin family" to find other drinks that will taste similar.

    Some of the comments I’ve received:

    "An impressive feat of procrastination"

    "Phylogenetics + Liquor = Awesome"

    "Now this is the kind of phylogenetic research I can get behind."

    "I was skimming through it when I spotted the Salty Dog and wanted to try it."

  15. HECK said,

    July 20, 2010 @ 11:50 am

    Re: Doctor Science

    Your remark, "It may also be a useful example for refuting 'intelligent design'," displays what I think is a basic misunderstanding of science which actually plays into the hands of those who believe in intelligent design.

    Science can never refute the theory of intelligent design—all it can do is demonstrate the existence of an alternative theory. Those who hold to intelligent design can always argue that the designer hides his handiwork so that natural selection only appears to work. The same holds for the theory of the young Earth—fossils are placed there to make it appear old. I think a better name for it would be "insidious design."

    Of course this is why intelligent design is not a scientific theory—by definition a scientific theory must be open to the possibility of disproof. Trying to refute theories like intelligent design does a disservice to science by elevating them to the level of science. One should merely ask, "How might your theory be disproved?" And when no answer is forthcoming, simply dismiss the theory out of hand as not being in the realm of science.

  16. Mary Kuhner said,

    July 20, 2010 @ 1:03 pm

    I was once commissioned (by the offer of a nice lunch) to draw a phylogenetic tree of medieval weaponry, for a games designer. He gave me a list of weapons with various attributes, and I ran a standard parsimony algorithm (from PHYLIP, just as with the cocktails) to produce a tree. The tree made a fair amount of sense, grouping the polearms together, for example, and the daggers. He seemed happy with it. I suspect the intention was to make a "skill tree" where skill with guisearm would transfer at a certain level to skill with glaive, and a lesser level to skill with, say, longsword.

    Fine for the intended purpose, but as a practicing phylogeneticist I'd like to caution against the non-entertainment use of phylogenies for anything whose evolution is not predominantly descent-with-modification. It's a really bad model for, say, literary styles or TV shows. There are some descent-with-modification relationships among computer viruses, but that "tree" is really a series of trees with no common ancestor, and considerable crosstalk among them. In general, a lot of interesting stuff doesn't fit that model, and coercing the model has a bad history of simply reinforcing preconceptions.

  17. Doctor Science said,

    July 20, 2010 @ 1:18 pm


    Yes, I dithered over the word "refute", and couldn't come up with a good synonym. I meant that this sort of tree is IMHO just what one expects when things are being *designed*, and can be used as an educational tool to talk about how we can tell the difference between evolution and progressive design.


    the larger part of the problem being the assumption that the sole developmental process is adding, subtracting, or substituting ingredients (or chemical constituents of ingredients)

    It would be interesting to see how the tree would change as you include more chemical constituents. To me, the interesting thing to do with cocktail tree — or for students to consider — is to use it as a metaphor for problems in phylogeny, such that cocktails = living species (or extant texts). We know, from external sources, when and how each of the cocktails was invented — what data would we need to put into our tree program for it to generate anything like an evolutionary tree?

    Or do we conclude that evolution is the wrong metaphor for cocktails — and possibly for other cultural artifacts?

  18. chris said,

    July 20, 2010 @ 3:26 pm

    One problem with the evolution meme as applied to cocktails (or cultural artifacts in general) is that there is no barrier to hybridization between even arbitrarily widely separated parts of the "tree". A duck and a duckbilled platypus cannot interbreed (still less a duck and a duckweed), but there's nothing stopping someone from inventing a new drink that combines the best (in their opinion) characteristics of whatever two (or more) source drinks they want to start from, and if they can come up with a catchy enough name and a pool of people who enjoy drinking the result, then it's here to stay.

    This would tend to make the family tree look less like a tree and more like a net.

    [(myl) Indeed. What's at stake here is the general idea of historical development as "descent with modification", where "modification" is defined in terms of a limited set of possible independent and local innovations. The result of these assumptions is indeed a tree-structured history.

    There are aspects of both linguistic and genetic evolution that don't work this way, and therefore lead to less constrained sorts of historical networks. This includes most kinds of linguistic borrowing, for example, or in genetic history, genome incorporation (a la Lynn Margulis), sharing of genetic material among bacteria, or gene insertion in multi-cellular organisms via bacteria or viruses.

    But overall, it seems to be possible to limit the evidence used in phylogenetic reconstruction (whether or languages or of genomes) to traits for which the hypothesis of "descent with modification" (for well-defined and simple values of "modification") is valid.

    It's much less obvious (to me, at least) that it's possible even in principle to do this for cocktail recipes. Some recipes were indeed created by processes of insertion, deletion, substitution from earlier recipes. But others weren't. And I don't see how to tell the difference, without knowing the history that the algorithm is supposed to infer.

    I think that this is obvious enough to practitioners of the phylogenetic arts that the case in point must have been a sort of joke, intended to illustrate the nature of the techniques, rather than a serious attempt to infer the history of cocktail recipes.]

  19. Stephen Jones said,

    July 20, 2010 @ 4:07 pm

    John Cowan. Has anybody ever produced a plausible IE tree based on the Out of India theory? Koenraad Elst, a strong supporter of the theory has stated that supporters can't keep on denying that historical linguistics is a well established science, and that the linguistic evidence is not incompatible with the theory, but I have never seen a proposed history of IE that takes OIT as its basis.

  20. Peter Taylor said,

    July 20, 2010 @ 4:13 pm

    In addition to concurring that this can't be taken seriously as a reconstruction of cocktail evolution, I don't think I can take it seriously as a cocktail guide. Where's the crushed ice in the list of ingredients for a daiquirí? Without that it's just a sweet shot.

  21. chris said,

    July 21, 2010 @ 2:08 pm

    It's much less obvious (to me, at least) that it's possible even in principle to do this for cocktail recipes.

    It's not obvious to me that it's possible in principle to do it to languages, either. Or do you start your tree-building by ruling pidgins and creoles off-limits? (Are there any sizable families of languages that are descended from an ancient creole, and if there were, how would you know?)

RSS feed for comments on this post