Professional Documents
Culture Documents
1. Introduction
Atkinson’s intriguing article in Science suggests that there is a signal to be
found in the global distribution of the richness of phonological contrasts across
a large sample of languages which reflects the process of the spread of anatom-
ically modern humans across the habitable areas of our planet from an origin
in Africa. In our commentary we will focus on three principal issues. The first
has to do with what the signal is based on and whether it seems to be reliable.
The second has to do with whether the concept of a serial founder effect is a
persuasive explanation for the global patterns found. The third has to do with
considering whether there are alternative explanations which might account for
the patterns found.1
1. A few details in this commentary were changed after the author of the target paper had seen
it. Specifically, a note was added about which languages create the major peaks and valleys
seen in Figure 2, and an improved method of calculating east-west distances in the Americas
was used. These changes do not affect any of the major points made in this commentary and
are not of a nature to require a separate response from the target author.
(the central bin contains the largest number of languages and the numbers de-
crease away from the center). The third is highly skewed in that about 60 % of
the languages fall in the lowest bin, having no tone contrasts.
As we will show below, there seems to be a real statistical relationship be-
tween Atkinson’s phoneme diversity variable, its components, and the mea-
sures that underlie these components and distance from Africa calculated
through certain travel nodes. However, it is worth noting that the label phoneme
diversity is potentially misleading in two ways. The full set of phoneme and
prosodic contrasts is not taken into account, and the internal diversity of the
phoneme set is not considered. Notably, the WALS variable of “basic vowel in-
ventory” strips away contrasts among matching vowels which differ in length,
nasalization, voice quality, and other properties. For example, Navajo has 4 ba-
sic vowel qualities /i, e, a, o/, but each of these can be long or short, and oral or
nasalized. So this North American language has 16 vowel phonemes, but falls
into the “small” bin in terms of its basic vowel inventory. On the other hand,
the West African language Igbo has 8 basic vowel qualities which equals its
total number of vowel phonemes, so it falls into the large bin, even though it
has only half the number of vowel phonemes as Navajo.
Adding tone to the phoneme diversity calculation incorporates one way in
which the syllable inventory of a language can be enriched, but much greater
phonological variation between languages is created by differences in permit-
ted phonotactic combinations (Maddieson 1984, Shosted 2006). A language
that has 10 consonants and 5 vowels but allows only CV syllables can con-
struct only 50 syllables. But if the same inventory can be freely deployed in a
(C)V(C) template the number of possible syllables is 605.
The phoneme diversity variable Atkinson calculates is thus neither a true
measure of the size of the phoneme inventory, nor of the phonotactic possi-
bilities in the languages concerned, but a hybrid touching on some aspects of
these properties. Further, it concerns only the number of distinctions, whereas
the term diversity might more readily be understood to refer to the content of
an inventory, rather than its size. The set of consonants in (1a) has less diversity
than the set in (1b), as the same few properties are re-combined in (1a) whereas
most of the consonants in (1b) have unique properties. In terms of the number
of phonetic traits exploited, the set in (1c) is equally diverse as that in (1a),
even though it has fewer members.
(1) a. p t k b. p ts q c. t k
b d g á nd k’ b
m n N m R ì n
However, although we consider the measure Atkinson uses to be rather an
odd construct and would have preferred that it not be labeled “diversity”, the
signal he has discovered seems to be valid. The most important part of this
Table 1. Fit between phonological variables and best-origin distance (simple regression
on means by language family, n = 49)
signal comes from the size of the consonant inventory.2 We have conducted
several examinations of the data that Atkinson provides and also of the un-
derlying data that forms the basis for the material in the WALS chapters that
he used. Table 1 shows the individual correlations of the three constituents of
his “diversity” measure, and their combination, with the best-origin distance
from Africa using data taken directly from Atkinson’s Table S2. This table
lists the means by language family for all families recognized which have two
or more members in the sample (using the classification given in WALS). One
language family listed in the table, Yanomam, has been omitted as there is only
one Yanomami language in the dataset – the Arawakan language Shiriana (ISO
639-3 code xir) was misidentified and wrongly counted as Yanomami. We did
not correct for another error, the inclusion of the Tai-Kadai language Po-Ai in
the Austronesian family. Since there are 41 Austronesian languages this error
will have little impact on the means.
Taken in isolation, the mean size of the basic vowel inventory does not corre-
late significantly with mean distance by family, and the role of tone is marginal.
When the three variables are simultaneously considered in a multiple regres-
sion the role of tone is negligible, and that of basic vowel inventory is marginal.
Results are shown in Table 2. In both analyses, the influence of the consonant
inventory size is paramount.
To further examine this issue we calculated the correlation between the best-
fit distance for individual languages and normalized values of the raw data
which underlies the binned values for consonant and vowel inventory that are
reported in WALS and were used by Atkinson. In addition, normalized val-
2. Largely because of the attention drawn by Atkinson’s article it came to light that a number
of errors had crept into the WALS data on consonant inventory size due to misalignment of
multiple data columns in a single spreadsheet. These errors affect 63 of the 504 languages
(12.5 %) used in Atkinson’s analysis, and have marginal impact on the results. We have re-
peated the analysis of the total diversity measure using corrected values and find the relation-
ship between this and the best-fit distance remains robust both across the set of languages
individually (F(1,502) = 255.9, p < .0001; R2 = .338), and across language families (using
those with three or more members – F(1,30) = 22.1, p < .0001; R2 = .424).
Table 2. Fit between phonological variables and best-origin distance (multiple regres-
sion on means by language family)
ues of the total vowel inventory and of a four-step tone index and an 8-step
index of syllable complexity were analyzed. The tone index distinguishes lan-
guages with simple (2 tones), moderately complex (3 tones), and complex (4
or more tones) tone systems from those which are not tonal (58 %). The syl-
lable complexity index is the sum of values assigned to the onset (range 0–3),
nucleus (range 1–2), and coda (range 0–3). A language which allows nothing
more complex than CV syllables and has no complex nuclei earns a score of 1.
A language which allows elaborate onsets and codas and has complex nuclei
(long vowels and/or diphthongs), such as English, earns a score of 8.
We were able to identify data for 495 of the 504 languages that Atkinson
lists in his Table S2. Not all data points are available for all the languages,
and values for some are undoubtedly disputable. As Table 3 shows, consonant
and vowel inventories and tone system complexity all individually correlate
to a highly significant degree with Atkinson’s best-fit distance from Africa,
although the R2 values are not especially high. The highest R2 value is with
the tone index, a reflection of the fact that 109 of the 123 languages in families
of Africa in this data are tonal (89 %), whereas in all other continent-sized
areas tonal languages are a minority. In a stepwise regression analysis, the tone
variable is the first to enter, but consonant inventory, basic vowel inventory,
and total vowel inventory all make significant independent contributions (in
diminishing order of significance) to the model, which yields a cumulative R2
value of .331 (n = 426). Strikingly, the syllable index shows no relationship
to distance from the best-fit origin in Africa. In other words, what is arguably
the greatest contributor to phonological diversity among languages patterns
geographically quite unlike the variables that relate to paradigmatic distinctions
among segments and tones.
The relative importance of consonant inventory and tone system complex-
ity differs notably between the analysis of the binned data at language family
level, and the analysis of the individual language data (using the raw conso-
nant inventory size and a 4-level tone index). The low significance of the tone
system variable in the family-level analysis may be due to the fact that Africa
reduces to five families (the familiar four – Niger-Congo, Nilo-Saharan, Afro-
Asiatic, and the disputed Khoisan, plus Kadugli which WALS treats as separate
Table 3. Fit between phonological variables and best-origin distance (simple regression
on normalized individual language values)
Verb (VOS, OVS, OSV) are predominently found far from Africa (Feature 81A
in WALS; Dryer 2011). In the WALS chapter on this issue all examples of the
VOS order are outside Africa (including Madagascar, counted as part of South-
East Asia by Atkinson). None of these less common orders occur anywhere in
mainland Eurasia. Most cases of the OVS order are found in South America,
where all six possible orders occur. If linguistic complexity in general declines
with distance from Africa, this distribution would be unexpected.
That there is no obvious signal of an African origin in the traits that con-
ventional historical linguists have been studying for more than a century and
a half could be due to large-scale language replacement having occurred in
Africa, resulting in loss of much of its earlier linguistic diversity (see Blench
2006 for discussion, also Nettle 1999). For example, the vast geographical
range of the Niger-Congo family in West and Central Africa suggests such re-
placement was in progress well before the more recent “Bantu expansion”, and
the distribution of Semitic languages in Africa is due to comparatively recent
back-migrations from Arabia. Given an assumption of replacement, the small
number of language families currently found in Africa would fail to reflect its
earlier linguistic richness. Under these circumstances, phonological richness in
surviving individual language could certainly be a trace of an earlier situation,
maintained through both continuity and cross-language contact. However, for
phonological richness – specifically phoneme inventory size – to decline with
distance from Africa, two things must be true. Processes that reduce the size of
a phoneme inventory must be more frequent overall than those that increase it.
And a mechanism must exist to increase loss that has some relation to distance.
Both internal processes and contact can lead to increases or decreases in the
number of distinct segments (or tones) a language uses. For example, Stan-
dard French in the last generation or so has lost a vowel through the merger
of /œ̃/ with /Ẽ/ and has gained a consonant /N/ through nativization of the En-
glish suffix -ing. There seems nothing in the linguistic processes themselves
that would lead loss to be more frequent than gain. Atkinson, relying inter alia
on Hay & Bauer 2007, argues that the imbalance between loss and gain is me-
diated through an indirect effect of community size. On the leading edge of
modern humans’ spread through new territories each new founder community
is assumed, reasonably enough, to be of small size. Small community size is
taken to be correlated with reduced phonological complexity, though the pre-
cise mechanism by which this transpires is not specified. We are inclined to
the opinion that data on the sizes of speaker populations are far too compro-
mised by the distortions due to decimation of indigenous populations through
disease, displacement, assimilation, and other tribulations following European,
Chinese, and other colonial expansions for any reliable “natural” signal of a
link between phoneme inventory size and population size to be recoverable.
The most severely affected peoples include those of the Americas, Australia,
Taiwan, and Siberia and the Russian Far East (cf. Butlin 1983, Stannard 1992,
Shepherd 1993), all areas quite distant from Africa. In view of this scepti-
cism, we have not conducted any reanalysis related to the population figures
that Atkinson used. Several authors (e.g., Haudricourt 1961, Trudgill 2004,
McWhorter 2008) have suggested that population size in itself does not predict
linguistic complexity, but rather the nature of the broad social setting of lan-
guage use may have varied impacts. In fact, small populations may favor the
preservation of complex and idiosyncratic linguistic features.
Figure 1. Sum of normalized consonant, basic vowel inventory, and tone values against
best-origin distance; linear and cubic spline fits
Figure 2. Sum of normalized consonant, total vowel inventory, tone, and syllable com-
plexity values against best-origin distance; linear and cubic spline fits
ity. In South America, the Caribbean coast and the Amazon basin generally
have low phonological complexity but along the Andean spine and the Pa-
cific coast the phonological complexity is higher. In the Southern cone (Chile
and Argentina) where there are very few indigenous languages phonological
complexity is high. If these local variations are overlooked to concentrate on
a broad overall pattern, a general decline in phonological complexity corre-
lated with distance from an assumed point of entry in the Bering Strait would
be found, given that eastern locations are further from the Bering Strait than
locations at the same latitude in the west, and Atkinson (2011: 348) indeed
specifically notes that “distance from the Bering Strait is inversely correlated
with phoneme inventory size within the Americas after controlling for popula-
tion size (rdistance = −0.173, P = 0.043)”.
An alternative model might focus on noting that the eastern side of the Amer-
icas is less complex than the west; where the land-mass is narrow in Meso-
America the east-west division is neutralized. This might be readily congru-
ent with scenarios for the peopling of the Americas that posit multiple early
routes of entry (for discussion see Dillehay 2009), with therefore likely dif-
ferent linguistic characteristics. As a simple-minded exploratory examination
of this idea, we analyzed the two phonological indices that show a systematic
relationship with Atkinson’s best-fit distance in the Americas, the size of the
consonant inventory and the syllable index (the vowel inventory measures ac-
tually show an increasing trend with distance from the Bering Strait, tone
shows no correlation) using the latitude and longitude coordinates of the lan-
guages. The sum of the standardized values of these variables correlates pos-
itively (R2 = .190, p < .0001, n = 116) with higher latitude, which is a good
approximation to strictly north-south distance. Since the western coast of the
Americas follows a broadly north-west/south-east slant, we used a slanted line
fitted approximately along the coast to “correct” the longitude values for each
language based on its latitude, so that the adjusted values are roughly propor-
tionate to distance from the west coast in degrees. We then used a linear approx-
imation to distance per degree of longitude at a given latitude (95 % accurate)
to estimate east-west distance. There is a highly significant correlation between
these calculated distances and the phonological index (R2 = .209, p < .0001)
reflecting lower complexity at more easterly locations. In a stepwise multiple
regression with distance from the Bering Strait, latitude, and “corrected” lon-
gitudinal distances as predictors of the summed consonant/syllable index, the
approximated east-west distance is the first to enter, a significant improvement
to the model is made by adding latitude (cumulative R2 = .336, p < .00001),
whereas distance from the Bering Strait makes no contribution (p = .5239).
There are hints here that the single cline found by Atkinson might better be
“parsed” into two separate ones.
It is obvious that there are other factors underlying the pattern of distribution
of phonological complexity beyond a straightforward reduction in complexity
with distance from Africa. It is even possible that the “African origin” effect is a
proxy for some combination of these other factors. One possibility is related to
the acoustic adaptation hypothesis (Morton 1975, Wiley & Richards 1978, Pe-
ters & Peters 2010) which suggests that animal vocal communication systems
(avian and mammalian) are adapted to the climatic/ecological environment in
which they operate, optimizing for the transmission characteristics of the envi-
ronment. Several studies have argued for a similar effect in human languages
(e.g., Munroe et al. 2009). Temperate environments with open vegetation facil-
itate transmission of higher frequency signals more than warmer more densely
Figure 3. Sum of normalized consonant inventory and syllable complexity values against
distance from temperate zone (centered at ±45° decimal latitude); linear fit
References
Atkinson, Quentin D. 2011. Phonemic diversity supports a serial founder effect model of language
expansion from Africa. Science 332. 346–349.
Blench, Roger. 2006. Archaeology, language and the African past. Lanham, MD: AltaMira.
Butlin, Noel. 1983. Our original aggression: Aboriginal populations of southeastern Australia
1788–1850. Sydney: Allen & Unwin.
Dillehay, Tom D. 2009. Probing deeper into first American studies. PNAS 106(4). 971–978.
Dryer, Matthew S. & Martin Haspelmath (eds.). 2011. The world atlas of language structures
online. München: Max Planck Digital Library. http://wals.info
Dryer, Matthew S. 2011. Order of subject, object and verb. In Dryer & Haspelmath (eds.) 2011,
Chapter 81. http://wals.info/chapter/81 (accessed 4 June 2011)
Haudricourt, André-Georges. 1961. Richesse en phonèmes et richesse en locuteurs. L’Homme 1.
5–10.
Hay, Jennifer & Laurie Bauer. 2007. Phoneme inventory size and population size. Language 83.
388–400.
Maddieson, Ian. 1984. Patterns of sounds. Cambridge: Cambridge University Press.
McWhorter, John. 2008. Why does a language undress? Strange cases in Indonesia. In Matti Mi-
estamo, Kaius Sinnemäki & Fred Karlsson (eds.), Language complexity: Typology, contact,
change, 167–190. Amsterdam: Benjamins.
Morton, Eugene S. 1975. Ecological sources of selection on avian sounds. American Naturalist
109(965). 17–34.
Mous, Maarten. 2003. Loss of linguistic diversity in Africa. In Mark Janse & Sijmen Tol (eds.),
Language death and language maintenance: Theoretical, practical and descriptive ap-
proaches, 157–170. Amsterdam: Benjamins.
Munroe Robert L., John G. Fought & Ronald K. S. Macaulay. 2009. Warm climates and sonority
classes: Not simply more vowels and fewer consonants. Cross-Cultural Research 43. 123–
133.
Nettle, Daniel. 1999. Linguistic diversity. Oxford: Oxford University Press.
Peters, Gustav & Marcell K. Peters. 2010. Long-distance call evolution in the Felidae: Effects
of body weight, habitat, and phylogeny. Biological Journal of the Linnean Society 101(2)
487–500.
Ramachandran, Sohini, Omkar Deshpande, Charles C. Roseman, Noah A. Rosenberg, Marcus
W. Feldman & L. Luca Cavalli-Sforza. 2005. Support from the relationship of genetic and
geographic distance in human populations for a serial founder effect originating in Africa.
PNAS 102(44). 15942–15947.
Sands, Bonny, 2009. Africa’s linguistic diversity. Language and Linguistics Compass 3(2). 559–
580. http://onlinelibrary.wiley.com/doi/10.1111/j.1749-818X.2008.00124.x/full
Shepherd, John R. 1993. Statecraft and political economy on the Taiwan frontier, 1600–1800.
Stanford, CA: Stanford University Press.
Shosted, Ryan. 2006. Correlating complexity: A typological approach. Linguistic Typology 10.
1–40.
Stannard, David. 1992. American Holocaust: Columbus and the conquest of the New World. New
York: Oxford University Press.
Trudgill, Peter. 2004. Linguistic and social typology: The Austronesian migrations and phoneme
inventories. Linguistic Typology 8. 305–320.
Wiley, R. Haven & Douglas G. Richards. 1978. Physical constraints on acoustic communication in
the atmosphere: Implications for the evolution of animal vocalizations. Behavioral Ecology
and Sociobiology 3. 69–94.