The volume consists largely of papers based on contributions to the workshop 'r-atics: Sociolinguistic, phonetic and phonological characteristics of /r/, held at the University of Nijmegen from 25 to 27 May 2000; only the paper by Sankoff, Blondeau & Charity was originally presented elsewhere. As the editors explain in the introductory chapter ‘Patterns of /r/ variation’, the coin 'r-atics refers both to the ‘chameleonic nature of the rhotics’ and to ‘the fact that a lot of erroneous views on the characteristics and distribution of different variants of /r/ are spread among both lay-men and linguists’ (p. 1). The major aim of this volume is to provide a better insight into the patterns of variation and change of /r/ in the Western European languages from a variety of perspectives. Moreover, the editors express the hope that ‘this volume will be the starting point of an international and multidisciplinary network of linguists studying the characteristics of the rhotics in the languages of the world’ (p. 2).

Verstraeten & Van de Velde’s contribution is one of the first results of a large-scale study of the sociolinguistic, phonological and phonetic characteristics of /r/ in Dutch, currently conducted at the Universities of Brussels, Nijmegen and Utrecht. In order to establish the ‘standard’ realisations of /r/ across the Dutch language area, the authors investigate recordings of nonsense words read by language teachers. In comparison with previous work on /r/ in Dutch (e.g. Vieregge & Broeders 1993, van Reenen 1994, Van de Velde & van Hout 1999), the resulting picture is a rather static one, with relatively low intra- and interspeaker variability and virtually no effect of gender, age or phonological context on the realisation of /r/. Moreover, the findings seem to confirm ‘old’ standards of pronunciation, with innovative realisations reported in the literature (see references above, as well as Collins & Mees 1996 and Plug 2002), such as alveolar and retroflex approximants, conspicuous in their low frequencies. This raises the issue of how to define the standard variety of a language in linguistic research (see e.g. Honey 1997). Unfortunately the authors do not address this issue. If Standard Dutch is defined as any variety of Dutch without marked regional characteristics, Verstraeten & Van de Velde’s study may not have been the best way of finding out what /r/ in Standard Dutch sounds like. Still, their contribution is a promising point of departure for more extensive investigation.

Among the ‘phonetic’ contributions I will discuss those by Demolin, Denton, and Docherty & Foulkes, since these present results of acoustic and instrumental investigations.

Demolin presents some preliminary results of a study on /r/ in Belgian French. His contribution is short, but contains some interesting observations. In particular, Demolin’s finding that one of the variants of /r/ is a creaky quality of the ‘preceding’ vowel – a variant that is not mentioned in any of the cross-linguistic studies of /r/ mentioned above – should interest anyone working on the interface between phonetics and phonology.

Denton demonstrates how detailed phonetic investigation can benefit historical phonology. By combining evidence of sound changes triggered by /r/ from historical studies and insights into coarticulatory effects of /r/ from recent phonetic research, Denton presents a plausible case for refining current views of the realisation of /r/ in early West Germanic. 

Docherty & Foulkes’s contribution accompanies the article by the same authors discussed above; here the authors describe the phonetic properties of anterior approximant variants of /r/ in British English, and argue that the rise of labiodental realisations of /r/ reflects a change in progress whereby the labial component of the articulation of /r/ is retained at the cost of the lingual component. The combination of variationist and detailed phonetic investigation must be applauded, since much sociolinguistic work on /r/ depends on the auditory classification of segmental variants alone. It is well known that variants of /r/, especially when ‘vocalised’ or ‘elided’, can be extremely difficult to classify reliably. Moreover, in various languages of Western Europe, /r/ has strong effects on vowel quality, which raises the question of what counts as a realisation of /r/ and what does not. For example, in Dutch, vowels preceding /r/ often end in a mid-centralising off-glide, even when /r/ is ‘deleted’ (Collins & Mees 1996). It is therefore surprising that studies differentiating between ‘schwa-realisation’ and ‘non-realisation’ of /r/ (e.g. Vieregge & Broeders 1993, Van de Velde & van Hout 1999, Verstraeten & Van de Velde this volume) pay little or no attention to the phonetic definition of the categories set up (cf. van den Heuvel & Cucchiarini this volume, p. 194). In these cases, a closer consideration of the acoustic and articulatory aspects of the variation may provide the analyst with more robust classification criteria, and will certainly make the resulting analysis more transparent to the reader.

Among the ‘phonological’ contributions I am including here those by Wiese, van Oostendorp, and van den Heuvel & Cucchiarini, since these are concerned with the phonological representation of /r/ and phonological processes /r/ can undergo. I will also consider the contributions by Goeman & Van de Velde and Torp, since the findings of these articles raise some interesting phonological questions.

Wiese addresses the well-known problem that ‘Phonologically, rhotics tend to behave in similar ways and participate as a class in phonological rules’ while ‘Phonetically, the rhotics form a heterogeneous group’ (Lindau 1985:158). He argues that the considerable variability in the realisation of /r/, exemplified here by data from German, can be accounted for if it is assumed that the class of rhotics is a phonotactic, rather than a segmental class. To be more precise, he claims that /r/ is a point on the sonority scale. As Wiese points out in a footnote, this entails that the sonority scale cannot be defined in featural terms, since the hypothesis is based entirely on the impossibility of coming up with a universal feature representation of /r/. Unfortunately Wiese does not go into the questions of what such an ‘abstract’ sonority scale should look like, and how phonological representations of individual rhotics should be established under his proposal.

Van Oostendorp shows that in Tilburg and Maasbracht Dutch, /r/ sometimes patterns with the sonorant consonants and sometimes with the obstruents. He takes this as evidence that /r/ has a relatively ‘empty’ phonological representation and that a constraint operates according to which phonological constituents should end in a consonant. The second argument, in favour of a FinalC constraint, rests on the assumption that sonorants are not consonants, and it is therefore somewhat odd that van Oostendorp calls them ‘sonorant consonants’ (p. 117) and at one point considers nasals as a subclass of obstruents (p. 117, ex. 8). In addition, van Oostendorp’s structural explanation of the patterns – /r/ behaves as an obstruent when word-final and as a sonorant elsewhere – surely requires a consideration of all words ending in /r/. However, the data from Tilburg Dutch compare stem-internal and stem-final /r/; both are word-internal.

With regard to the first argument, van Oostendorp merely speculates on how the ‘placelessness’ of /r/ causes its complex behaviour, and fails to note that if, as is widely accepted, the primary function of phonological features is to specify natural classes, membership of several classes should be accounted for in terms of the presence of several features, rather than the absence of one. Moreover, if the default place is ‘velar’, how is the correct phonetic interpretation of /r/ achieved? Unfortunately both van Oostendorp and Wiese fail to spell out what they consider to be the function of phonological representations, and, crucially, how much phonetic information they should contain.

Similar questions regarding the status of phonological representations and processes are raised by van den Heuvel & Cucchiarini’s contribution. The authors present results of their study of the apparent non-realisation of postvocalic /r/ in Dutch and its implications for automatic speech recognition. While in earlier publications (Cucchiarini & van den Heuvel 1995, 1998) the authors have shown that ‘/r/-deletion’ is more likely to be applied following schwa than it is to occur following a full vowel, here they set out to determine whether its application is constrained by the phonological length of the preceding vowel and its degree of stress. The authors present judgements of the application of /r/-deletion in a corpus of spontaneous Dutch by an automatic speech recogniser and by human transcribers which do not suggest any effects of vowel length or stress on /r/-deletion.

Van den Heuvel & Cucchiarini’s research (see also Kessens, Wester & Strik 1999) is very interesting in that it demonstrates how descriptive generalisations can be made over large amounts of unscripted speech material. Unfortunately, however, the authors suggest by their use of the term ‘/r/-deletion’ and of slant brackets for transcriptions that these generalisations amount to a phonological analysis, although the fact that no segmental realisation of /r/ can be observed does not entail the deletion of the phoneme /r/ in the phonological computation (Plug 2002). Interestingly, the authors note that the human transcribers tend to perceive fewer occurrences of /r/-deletion than the automatic speech recogniser, and suggest that this may be related to the fact that for humans, but not for the machine, ‘the way one sound is perceived very much depends on the identity of the adjacent sounds and the transitions between the sounds’ (p. 195). This in itself indicates that despite the authors’ implication that /r/-deletion in Dutch is ‘reality’, further research is necessary in order to accommodate van den Heuvel & Cucchiarini’s findings in an explicit model of Dutch sound structure.

Goeman & Van de Velde test Walsh Dickey’s (1997) hypothesis that a uvular trill and a uvular fricative or approximant cannot co-occur in the sound system of a language; according to Walsh Dickey, these are phonetically too similar to maintain a phonological contrast. Although in the Dutch language area, a uvular trill is a possible realisation of /r/ and a uvular fricative is a possible realisation of /G/, Goeman & Van de Velde find that in their database, varieties of Dutch in which relatively ‘back’ realisations of /r/ are common, realisations of /G/ tend to be relatively ‘front’, and vice versa. It must be noted that ‘back’ realisations of /r/ appear to be on the increase (e.g. van Reenen 1994), especially in The Netherlands. Therefore, it would be interesting for the authors to repeat their study in a number of years’ time, in order to gain direct insight into the extent to which a threatening merger prohibits sound change. It would also be useful to consider the number of minimal pairs such a merger would affect; if this is relatively small, the pressure to prohibit a merger may not be as strong as Walsh Dickey and others have suggested. 

Torp provides an overview of the realisation of /r/ in the Scandinavian language area, and makes the intriguing observation that retroflex consonants, many of which originate historically as /r/-obstruent sequences, and dorsal realisations of /r/ appear to be mutually exclusive; in Torp’s words, ‘the presence of retroflexes in a dialect represents some kind of “linguistic vaccine” against the intrusion of dorsal /r/’ (p. 81). Co-occurrence constraints such as those discussed by Torp and by Goeman & Van de Velde raise important theoretical questions: how should such dependencies be represented in the phonology (cf. Carter forthcoming), in a model of speech perception (cf. Hawkins & Nguyen forthcoming), or at the level of phonetic implementation (cf. Stevens 1998)?

To sum up, many of the contributions to this volume contain interesting observations and raise interesting questions. Although I have been critical of some of the contributions and was somewhat disappointed not to find an articulatory study of /r/ in the volume, given the success of the gestural model associated with Articulatory Phonology in accounting for various sound patterns associated with the rhotics (Browman & Goldstein 1990, McMahon, Foulkes & Tollfree 1994, Gick 1999, Gick & Wilson 2001, Plug 2002, Carter forthcoming), I believe the volume is a useful ‘starting point of an international and multidisciplinary network of linguists studying the characteristics of the rhotics in the languages of the world’ (p. 2). As suggested by some of my criticisms, I applaud the multidisciplinary approach taken by the organisers of the workshop, and hope that this will further encourage ‘phonologists’ and ‘sociolinguists’ to pay attention to the phonetic details of the patterns they refer to, and ‘phoneticians’ to address the theoretical implications of their findings.

As an overview of current work on /r/ and a starting point for further research, this volume is well worth owning. The quality of proof-reading of the volume is not of a professional standard (e.g. Torp’s phonetic symbols appear somewhat different from those in other papers; Sankoff, Blondeau & Charity symbolise an arrow by the sequence ‘->’ in their title, while van Oostendorp uses a much nicer arrow symbol, e.g. on p. 120; the format of in-text references and placement of acknowledgements are not consistent across papers; some contributions contain noticeably more errors, omissions and inconsistencies than others, but a flawless presentation would arguably be too much to expect from a volume that is available without charge from the first editor (e-mail hsl-shl@hum.leidenuniv.nl).

Leendert Plug, Department of Language and Linguistic Science, University of York