LingLunch : Ewan Dunbar

Jeudi 19 Janvier 2017, 12:00 to 13:00
Organisation: 
Pascal Amsili (LLF)
Lieu: 

ODG – Salle du conseil (533)

Ewan Dunbar (LSCP)
Geometric tendencies in the typology of sound inventories

This talk investigates the sound inventories of the world's languages, looking at typological tendencies in what we call "geometric" properties: the abstract "shape" of the inventory as defined by the relatedness of the sounds they contain. A classic example of a geometric property of sound inventories is Economy (Martinet 1939, Clements 2003). Economy measures how "efficiently" an inventory uses the phonetic sub-space it sits in, so that a vowel inventory that makes contrasts in height ([i] versus [e]), rounding ([i] versus [y]), and backness ([y] versus [u]) is counted as having greater Economy the more sounds it has that can be contrasted only along these three dimensions (such as [Þ], which is like [e] except that it is rounded, and [o], which is like [Þ] except that it is a back vowel); it will be counted as having less Economy if it has sounds that contrast on other dimensions, like nasality (such as [ẜ], which is like [e] except that it is nasalized). It has been demonstrated previously that Economy is statistically significantly higher in human sound inventories than would be expected by chance if sounds combined totally independently into inventories (Mackie and Mielke 2011; see also Clements 2003, 2009). We replicate this result using a slightly more robust methodology and demonstrate that two geometric measures that have not been previously remarked upon are also at above-chance levels in sound inventories: Local Symmetry and Global Symmetry. Local Symmetry measures the number of "oppositions": pairs of sounds that contrast minimally, such as [t] and [d], which contrast minimally for voicing. The number of such oppositions, as measured independently of Economy, is greater than would be predicted by chance. Global Symmetry measures how well "balanced" an inventory. For example, if a vowel inventory has five front vowels and only one back 
vowel, or ten oral vowels and two nasal vowels, it scores low in Global Symmetry. Global Symmetry, as measured independently of both Economy and Local Symmetry, is also higher in natural inventories than would be predicted by chance. I discuss the methodology used to study these typological properties, which need to rely on some theory of how to define the "relatedness" of sounds and the "space" in which they lie (in this case, a particular set of binary phonological features), but which makes an attempt to "average over" as many theoretical assumptions as 
possible. I suggest some possible origins for these tendencies and ways of investigating them experimentally and using statistical modelling.