Editor for this issue: <>
trl.oz.au>Some readers will not have failed to notice blatant contradictions between Trey Jones's answer and mine to Mark Durie and concluded "well, so there". There are no contradictions. We have each answered differently because we have read different meanings into the objection presented. Consider this exchange: Durie: >By Jacques Guy's method, even one of the daughter languages could be >a protolanguage, with 100% retention of vocabulary. Jones: >That is entirely correct. Me: >No. The daughter language is evidently not the protolanguage. Trey Jones interpreted Mark Durie's objection in mathematical terms: if two languages are lexicostatistically identical they are represented by the same point in the tree and they are the same. Correct, of course. I interpreted the same objection like the linguist I once was would have, for whom "daughter language" and "protolanguage" immediately conjure up notions of dates, and for whom Dick and his great-grandfather Harry are not the same person even though they are the spit image of each other, and, accordingly, I answered "No. The daughter language is evidently not the protolanguage; but they look the same." Durie: >Guy's 'proof' that the root can be placed in infinitely >many places only works on the assumption of infinitely >arbitrary variations in vocabulary replacement rates. Trey Jones has read "infinity" in its strict mathematical sense, in its context: graphs, arcs, hence the number of points on an arc; vocabulary replacement -- or retention -- expressed by a real number in the range 0 to 1, of which there are exactly as many as points on an arc. He has supplied the missing bits and ignored the irrelevant ones, thus reading something like: >Guy's 'proof' that the root can be placed in infinitely >many places only works on the assumption of infinitely >many different possible proportions of replaced/retained >vocabulary from 0 to 1. And, accordingly, he has answered, rightly, "Yes". Whereas I have puzzled over "infinitely arbitrary", which means... search me, and spotted "replacement rates" which do not enter in the proof at all, and, accordingly, and rightly too, I have answered "No". One valid comment, however, could have been that the root cannot be placed in infinitely many places because, lexicon being finite, sample wordlists can only consist of a finite number of items, so that there is only a finite number of distinct points in a lexicostatistical tree. But it changes nothing to the fact that the root can be any of those points, terminal nodes included. It is not the number of possible places where the protolanguage could be which is important, but that the protolanguage could be anywhere at all. Reminds me of the incident which led me to study statistics. It was around 1974, when I had just learnt programming and was cutting my ALGOL teeth on lexicostatistics. I had discovered a promising article by Wilhelm Milke in Lingua (1965) but could make neither head nor tail of it, not even the mathematical notation. So I did the thing any linguist would have done. With a photocopy of Milke's paper in my hot little hand, I went to our consultant statistician, and asked what it meant. She started reading and a smile lit up her face. As she read on the smile grew wider and wider. Until she burst out in a peal of laughter. Still flushed, she looked at me in a very peculiar way and asked: "Is that what you people believe?". Every year still brings its contingent of fresh nuts who have trisected the angle and squared the circle. Every year still brings its fresh harvest of factorial analyses, minimum-spanning trees, multidimensional scaling, linear programming and whatnot perorating on anything from Proto-Abelam to Proto-Zyryan, _without_the_slightest_regard_for_the_ properties_of_the_data_and_whether_those_methods_are_appropriate. (Has anyone tried fluid dynamics yet? No? Never mind, have chaos theory instead. Quantum cryptography is not bad, either. Neural nets are good, too. Petri nets, however, are no longer fashionable. Implement it all in LISP, or at the very least, in new, improved, C Triple++ with enzymes -- the beasties eat bugs) Some may turn to the extremes I cited, Icelandic and Muyuw, and say: "Oddballs." We do not know that. The number of languages for which we have dated documentary evidence is extremely small compared to those for which we have not, or which have disappeared without trace. Further, those languages for which we have evidence are all characterized by an old written tradition (encore une Lapalissade! If they were not, we would have no evidence, would we now?). Therefore they are not a representative sample. Short of Dr Who's time machine, whatever we may say about the past retention rates of the overwhelming majority of languages necessarily begs the question, and no amount of obfuscation will change that. The languages I used in my proof, at any rate (Alpha, Bravo, etc.), are witness enough. They are the very set of real languages of the New Hebrides (now Vanuatu) which led me to question the principles and methods of glottochronology some twenty years ago. I *knew*, by gut-feeling, by seat-of-the pants knowledge, that "Foxtrot", the subject of my Ph.D. thesis, learnt the hard way, in a monolingual situation, was an extremely close relative of the other languages of Espiritu Santo, even though lexicostatistics gave it as distant as the languages of New Caledonia, even though it was phonologically aberrant and syntactically non-Austronesian, with holophrastic tendencies worthy of Amerindian languages and internal noun inflections worthy of Semitic. Not to mention its seven degrees of deixis. Indeed, upon reading the transcription of my first tape, one of my supervisors had said: "That is Papuan". But back to Foxtrot: Alpha -830-----:-919-----:-972-----:-947-----: Bravo -770-----' | | | Charlie -----829-----------' | | Delta -----795-----------:-949-----' | Echo -----755-----------' | Foxtrot -----567-----------:-883-----:-895-----' Golf -----759-----------' | Hotel ----------772----------------' Wherever you put the protolanguage, you have widely divergent retentions and you can tell absolutely nothing about the dates of the splits, and, therefore, about the retention rates. If Foxtrot and Golf split 1000 years ago, then their respective retention rates were 56.7% and 75.9% per thousand years. Two thousand years ago, 75.3% and 87.1%. Five hundred years ago 32.1% and 57.6%. In fact, from toponymy and oral traditions, I would not be surprised at all if Foxtrot and Hotel had split only two or three hundred years ago! Grab your calculators: if Foxtrot split from Hotel 300 years ago, it has retained 0.567x0.883 of their common vocabulary over 0.3 millennia: 0.567x0.883=0.501, whence estimated retention rate for Foxtrot: 10% per thousand years. Do the same for Hotel: 42.2%. Why do I pick Foxtrot and Hotel rather than Foxtrot and Golf? I said "toponymy and oral traditions". They show that Foxtrot speakers lived originally right next door to Hotel. Golf, incidentally, is the language of Chief Buluk, once host to Jimmy Stephens' Nagriamel Movement, and is a variety of *that* language with the phonemic contrast between apico-alveolar s and lamino-alveolar s. (Yes, another weirdo. Those islands are full of them). Hard to believe but true, Golf shares no phonological innovations with Foxtrot. Yet I have not the slightest doubt that the "nutcracker proof" will make no difference whatsoever. First, from experience, it is most unlikely to get published. It has proved impossible to get the text of my 1983 Dunedin paper, which dealt with the mathematics and methodology of glottochronology, published. At the closest it came to publication, referees demanded emasculations, I carried them out, and then it fell into a black hole. Fashion, at any rate, demands length, padding, and thick bibliographies, not half-page proofs. Second, I doubt the observation and the proof are original, so simple they are. So they must have been discovered and presented before, and buried and forgotten. History repeating itself, they will be buried and forgotten once more. Third, their acceptance means writing off everything not backed up by dated historical or archaeological evidence, and eating humble pie from the hand of archaeologists, historians, botanists.... Gone would be the days of factorial analysis, multidimensional scaling, arcane manipulations and sleights of hand that dazzle the gaping audiences and cut a solid rut into that tenure track. The usual gravy train. You do not stand in the way of a train, whether coal or manure-powered. Good luck and a field day, then, to the Cavalli-Sforzas, Greenbergs, Marrs and other Dyen, Kruskal and Blacks. How long can linguistics ride on the coat-tails of mathematics in blissful ignorance? As long as astrology can on the coat-tails of astronomy. I would not hold my breath.Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue