Rhotic relationships: diachrony vs. synchrony in representing /r/

Abstract

The phonetic diversity of r-sounds, or rhotics, both cross-linguistically and within many languages, e.g. Spanish (Lipski 1994), German (Ulbrich 1972), and Swedish (Muminovic and Engstrand 2002), presents a challenge for phonologists. If there are no phonetic features that are shared by all rhotics (Lindau 1985; Ladefoged and Maddieson 1996:244), it is hard to assign them a shared featural representation that accounts for their similar patterning across and within languages, and does justice to their variable nature. Phonologists have attempted to characterise rhotics in highly abstract ways, as featurally (almost) empty (e.g. Harris 1994; Giegerich 1999; Wiese 2001), or by assigning them an abstract feature [rhotic] (Hall 1997) or a shared structure (Walsh Dickey 1997). Such top-down approaches, however, are inevitably ad- hoc, largely circular, and sometimes not even descriptively adequate (Gąsiorowski 2006). Lindau’s (1985) solution is to characterise rhotic unity in terms of “family resemblance” (see also Magnuson 2007 for an expanded version). That is, while it is not true that all r-sounds share some phonetic property, it is true that every r-sound shares something with at least one other: rhotics form a network of speech sounds with overlapping features. However, while less arbitrary than the top-down approaches, it is equally unrestrictive: many other speech sounds share properties with some rhotics, but do not function as rhotics themselves. Without a theory of which sounds can be in the rhotic set, “family resemblance” fares no better than calling rhotics “empty” or “[+rhotic]”. Ladefoged and Maddieson in fact suggest giving up the search for unity among rhotics altogether, as it “seems to rest mostly on the historical connections between these subgroups and the choice of the letter r to represent them all” (1996:245), a solution dubbed “shockingly informal” by Gąsiorowski (2006). I propose a formalisation of Ladefoged and Maddieson’s suggestion, by characterising rhotics not in terms of family resemblances, but family relationships. This is more restrictive, in that relationships between r-sounds are not established on the basis of phonetic similarity, but a diachronic link between two variants. Such links can be established by examining languages with extensive r-variation and inspecting very closely the phonetic detail of r- sounds, their linguistic distribution over contexts, and any sociolinguistic variation. The primary data in this talk come from a large-scale corpus of urban-accented Standard Dutch containing over 20,000 r-tokens from ~400 speakers from six Dutch and four Flemish cities. The realisation of /r/ in Dutch is notoriously variable across speakers and linguistic contexts, and in line with recent studies (Smakman 2006) the corpus distinguishes between some twenty r-variants. Age-related and geographical patterns in the data enable establishing the origin of specific variants in others. Specifically, this origin often lies in what happens to particular r- sounds in casual speech processes. For example, in casual speech, an alveolar trill may fail to be realised as such, given that it requires very precise articulatory settings in combination with narrow aerodynamic margins (Solé 2002). The result can be an alveolar approximant or fricative, which may subsequently come to function as an r-variant itself. This approach anchors the unity among r-variants within a theory of sound change, but locates it outside the phonology. I suggest that this is where it should be, and that top-down featural approaches to r are best replaced by a bottom-up one. Instead of requiring, a priori, that r-sounds form a category /r/ within a language, such a category is formed by speakers on the basis of multi-level evidence, including phonological alternations and sociolinguistic variation. For example, when speakers encounter two apparently phonetically unrelated phones, such as [ʁ] and [ɻ], in an allophonic relationship, as is the case for many Dutch speakers in the corpus, they may set up representations at multiple levels in a hierarchical system (as per Ladd 2006), reflecting both their phonetic differences and their phonological identity. In other words, while the phonetic unity between such different r-sounds is one that can be traced diachronically (as a family relationship), their phonological unity (in representation) need not refer to this relationship, but is forged synchronically by speakers.