v r i j d a G 2 3 s e p t e M b e r 2 0 1 1 Afscheidscollege prof. dr. C.H.M. Gussen H o v e n Of migrant men, shifting sounds and stagnant waters A farewell lecture by Carlos Gussenhoven Radboud Universiteit Nijmegen, 23 September 2011 Mijnheer de Rector Magnificus, dames en heren For over two decades, the second year Phonology course description has in- cluded the sentence Languages resemble each other in the way [their sounds] are struc- tured, but vary in the complexity of their sound systems. For instance, the smallest number of segments that has been found in any language is 11, the largest 144,... In this lecture, I would like to show, first, that counting sounds has led to interesting findings in linguistics, and second, argue that counting sounds has helped to redefine research questions in phonetics and phonology. 1 Variation in phonological complexity The data in that sentence came from Ian Maddieson’s 1984 survey of the phoneme inventories of 317 languages, a number that was later increased to 451, which represents 7% of the total number of languages spoken in the world (UPSID-II). Currently, these data have been incorporated in the phonological data base of World Atlas of Language Structures (WALS), which now includes phonological data for almost 600 languages. For instance, French has a more complex vowels system than Spanish, because it has 12 vowels against Spanish five. Complexity is spread throughout the phonology. Correlations between numbers of phonological elements, like vowels, consonants, tones, are largely positive, never negative. Having more vowels means having more tones, more tones means more consonants (Maddieson 2011). Languages with more phonemes also tend to have more complex syllable structures. For instance, Hawaiian, which has five vowels and eight consonants, only allows a single consonant before the vowel in the syllable and does not allow any consonant after it. As a result, all syllables have the form CV or just V. The only complexity is created by the fact that vowels may be long, giving 10 syllable rhymes and may combine with higher vowels to create a set of diphthongs, giving 1 another eight vowel types. The number of possible syllables therefore is 9 (eight consonants plus absence of a consonant) × (10 + 8), or 162. Following Maddieson (1984), Ryan Shosted (2006) did this kind calculation for a sample of 32 languages, to show the multiplicative effect of these structural options. 2 Migrant men: an important discovery Until 2007, it was always a matter of some fascination that no one had ever come up with any correlate of phonological complexity, other than the log-ical correlate of word length (Nettle 1998). But no external factors, like political organization or technological development, correlate with phonological complexity. Whatever the factors are that determine the phonological structure of languages, there are apparently many degrees of freedom. Languages with few phonemes are just as workable as those with many. While these observations remain true, a first indication that there was more behind phonological complexity came in 2007, when Jennifer Hay and Laurie Bauer (2007) showed that there was a weak, but robust positive correlation between population size and number of phonemes: smaller populations have smaller phoneme sets. They offered a plausible explanation, the only problem being that they would have had an equally plausible explanation if the situation had been the reverse. The next step occurred on 15 April of this year, when Quentin Atkinson published an article in Science based on 504 languages which showed that there was a fairly substantial correlation between the number of phonemes, measured as the mean of the normalized frequencies of the vowels, the consonants and the tones (‘Phoneme Diversity’), and the geographical distance from the southwest of Africa, as measured along the continental transit points known to have been used by modern man when colonizing the planet. The correlation explains 30% of the variance in phonological complexity against 14% by Hay and Bauer’s population size. Moreover, the variation due to population size is largely subsumed by that explained by distance from Africa. Spectacularly, the pattern of decreasing complexity would appear to reflect man’s first colonization of the planet, some 50 to 70 millenia ago (Fig. 1). Starting from south west Africa, we moved into the Eurasian landmass via the Middle East, moved along the southern edge of Asia and crossed over to Australia, some 40,000 years ago. About 14,000 years ago we trekked along the Bering Strait into North Amer-ica, making our way south over the next 6,000 years or so; lastly, about 3,000 years ago, we moved into Oceania, with New Zealand being among the most recently colonized areas, some 800 years ago. In addition to leaving a trail of diminishing phoneme sets, modern man also left a trail of ever smaller populations. 2 Exactly one month ago, Ian Maddieson discussed these findings at the Seven- teenth International Conference of Phonetic Sciences in Hong Kong. Tread- ing carefully, he suggested that it might be too early to draw conclusions from the Atkinson findings, since an observation made in 1996 about a cli- matic correlate of sound systems had not been taken into account: briefly, cold climates favour complex syllables. Complex syllables have many conso- nants, like English strictst (cf. a simple syllable like go). Munroe, Fought, & Macauley (2009) showed that it is the sonority of sounds that drives this correlation, since non-sonorant [i] and the sonorant consonants do not fall in with the broader generalisation. Incidentally, Munroe et al. (2009) explain the correlation by assuming that, other than in warmer climates, language use in colder climates largely takes place indoors. Possibly, it is rather the level of ambient noise that is the main explanatory variable, rain forests being particularly noisy places. An admittedly rhetorical example is Tamazight, the group of Berber language spoken in the Sahara. In some varieties, like Tarifit Berber, utterances can consist only of voiceless obstruents. A close example is (1), where [@] can be a voiceless release and [n] may thus be the only voiced sound (with thanks to Abder el Aissati). (1) [[email protected]@tQ.t@nt] /t tt ttQf d ten t/ 2SG IMPERF catch 2SG 3PL Fem ‘You catch them’ To see if this ecological factor has played a role the results presented by Atkinson, I first established that it can in fact be detected in his data. I de- fined it as the distance from the equator expressed as Absolute Latitude (see Fig. 2). Atkinson only used aggregate vowel, consonant and tone fequencies, which don’t allow a breakdown into more and less sonorant sounds. Still, there are negative correlations between Absolute Latitude and both vowel frequency (r=-0.19) and tone frequency (r=-0.26), and a positive correlation with consonant frequency (r=0.27), all p<.01), meaning that closer to the equator there are more tones and vowels and fewer consonants than in places further away, confirming the finding by Munroe and colleagues. I also con- structed a crude sonority index by dividing the mean normalized frequencies of vowels and tones by the normalized consonant frequency, after adding 2 to all values to avoid negative numbers. As predicted, there is a negative correlation with Absolute Latitude (r=-0.30, p<.01). There is no correlation with Distance from Origin. 3 Figure 1: Boxplot of phoneme diversity by global region showing highest diversity in Africa and lowest in Oceania and South America (top) and map of the world showing highest phoneme diversity in Africa with diminishing diversity in areas that were settled later, controlled for population size (bottom). From Atkinson (2011). 4 Figure 2: Map of the world indicating equator, Absolute Latitude and correlations between Absolute Latitude and normalized vowel, tone and consonant densities and a crude sonority index (see text). Adapted from Atkinson’s map with transit points.) I then ran a stepwise linear regression with Phoneme Diversity as the de- pendent variable and Distance from Origin, Log Population and Absolute Latitude as predictor variables. While unsurprisingly the results for the first two variables were as presented by Atkinson, no predictive contribution by 1 Absolute Latitude was found. We can conclude then that Ian Maddieson trod too softly, and that Atkinson’s data may suggest that language, as we know it today, existed when modern man left Africa and that all languages of the world can trace their origin to that source. Atkinson explains this finding as a linguistic version of the Serial Founder Effect, the loss of genetic variation that occurs when a new population is es- tablished by a small number of people moving away from a larger population. In the genetic domain, it explains the vulnerability of the American popula- tions to European diseases during the Columbian Exchange, which started in 1491 (Mann 2005), and the relative immunity of the European population to American diseases. While the Americans suffered near-extinctions as a result of European diseases, the only American disease to pose a problem to the 1There is in fact no correlation between Absolute Latitude and Distance from Origin. There is a weak correlation between Absolute Latitude and Phoneme Diversity (r=-.09, p<.05). I assume this is because Atkinson’s Phoneme Diversity measure contains two ‘sonorous’ componenents, vowels and tones, and only one ‘non-sonorous’ component, which biases the measure to sonority and thus closeness to the equator. If we take the mean of tones and consonants or the mean of vowels and consonants, the correlation disappears. 5 Europeans was syphilis. But language is not genetic.