1.1. The phonetics of French R

In most varieties of convergent French, /ʀ/ is articulated as a fricative or approximant, having a uvulo-velar or velar place of articulation, and may be voiced or voiceless, as in (1) (Tranel, Reference Tranel1987, forthcoming; Fougeron and Smith, Reference Fougeron and Smith1993; Walker, Reference Walker2001; Russell Webb, Reference Russell Webb2002, Reference Russell Webb2004a, Reference Russell Webb2004a; Rose Reference Rose2003).Footnote ² In (1a) and throughout this article, the diacritics [_˔] and [_˕] indicate a relatively narrower and wider aperture at the point of oral constriction, corresponding to fricative and approximant articulations, respectively. In (1b), [˯] and [˳] indicate voiced and voiceless surface forms, respectively.

1. (1) Surface forms of /ʀ/: Convergent French

   1. a. aperture

      

   2. b. voicing

      

Most speakers produce surface forms of /ʀ/ as fricative [] in word initial onsets and as approximant [] in codas, as in (1a). In intervocalic environments a good deal more variability is observed. In a phonetic study involving both spontaneous and careful speech tokens, Russell Webb (Reference Russell Webb2002) notes that speakers produce more approximant like segments in these contexts, an observation also echoed in Tranel (Reference Tranel1987, forthcoming) and O'Shaughnessy (1982), among others. At the same time, it is noted that fricative [] in intervocalic positions is far from extraordinary, especially in emphatic or careful speech styles.Footnote ³ /ʀ/ aperture also varies in hyper- or hypoarticulate speech styles, e.g. final /ʀ/ may be fricated for emphasis, whereas initial /ʀ/ may be lenited due to enchainment, i.e. the sequential resyllabification of speech. As noted in (1b), /ʀ/ is also subject to patterns of passive voicing (Tranel, Reference Tranel1987: 21; Walker, Reference Walker2001: 137–139; Russell Webb, Reference Russell Webb2004b).

Regional forms of French testify to additional surface forms and distributions. For many Brussels French (BF) speakers, for instance, aperture variation is largely absent (Grootaerts, Reference Grootaerts1953; Baetens Beardsmore, Reference Baetens Beardsmore1971: 84–85, 437–439); /ʀ/ almost always surfaces as a fricative, regardless of position or context, as in (2). Trilled [R] is also noted, although with far less regularity.

1. (2) Brussels French

   

Tranel (PC) observes that trilled [R] is marginal in modern French (see also Delattre, Reference Delattre and Delattre1966: 206), although it does occur in a number of highly marked speech styles, e.g. singing. Other examples of dialectal particularity are noted in Southwestern France, where /ʀ/ is often realised as a voiceless fricative (Tranel, Reference Tranel1987: 143); in Montréal, where /ʀ/ was until recently produced as [R] or [r], although fricatives are now most common (Tousignant, Sankoff and Santerre, Reference Tousignant, Sankoff, Santerre, Fasold and Schiffrin1989; cf. Sankoff and Blondeau, Reference Sankoff and Blondeau2007); and in Francophone Africa, where /ʀ/ is most often produced as an apical vibrant [ɾ], although velar articulations are not unknown (Dumont, Reference Dumont and Valdman1979; Duponchel, Reference Duponchel and Valdman1979; Manessy, Reference Manessy1984; Renaud, Reference Renaud and Valdman1979; see also Walter, Reference Walter1977).

Coda /ʀ/ vocalisation is noted in a handful of dialects, e.g. Quebecois (QF) and St. Thomas French (STF). In QF, /ʀ/ can surface as a homorganic vowel in syllabic codas.Footnote ⁴ In STF (3b), /ʀ/ vocalisation only occurs in post-vocalic clusters (i.e. complex codas); in simple codas, /ʀ/ is elided.

1. (3) /ʀ/ vocalisation

   1. a. Quebecois French (Côté, Reference Côté2004: 169-170)

      

   2. b. St. Thomas French (Highfield, Reference Highfield1979: 45)Footnote ⁵

      

Similar to STF data in (3b), coda /ʀ/ vocalization or erasure is common in regional French, as well as in non-standard registers and fast speech. This phenomenon is especially frequent when /ʀ/ is found in final, post-obstruent /ʀ/ position, e.g. cadre ‘frame’ [kad].

1.2. R distribution

A crucial issue with regard to the phonological status of /ʀ/ is the participation of this segment in phonetic and phonological classes, i.e. groups of sounds which behave in similar manners or take part in common processes. Like voiced fricatives, but distinct from /l/ and glides, /ʀ/ triggers phonetic vowel lengthening, as in (4).

1. (4) Vowel lengthening (Colantoni and Steele, Reference Colantoni and Steele2005; Laeufer, Reference Laeufer1992; Walker, Reference Walker2001: 42-44)

   

The question of whether vowel lengthening is best attributed to phonetic implementation or the phonological grammar is not taken up here (see e.g. Lindblom, Reference Lindblom2006 for discussion of the phonetics-phonology divide). Rather, what can be noted here is that /ʀ/ is associable with voiced fricatives and not /l/ and glides.

Despite the above, /ʀ/ distribution is most closely aligned with that of /l/, as exemplified in (5). Both segments frequently occur as the rightmost constituent in complex onsets; both also precede obstruents in complex codas or occur in final, post-obstruent codas.Footnote ⁶

1. (5) Liquid Distribution

   1. a. complex onset

      

   2. b. coda

      

   3. c. final post-obstruent

      

It should be noted that in complex codas such as (5b), /ʀ/ and /l/ also pattern with fricative /s/, e.g. vaste ‘vast’ [vast].Footnote ⁷

Other continuants surface as the rightmost element of onsets following schwa elision, e.g. s'maine (semaine ‘week’), d'sous (dessous ‘under’). Likewise, /ʀ/ may appear as the leftmost element in a surface onset, e.g. r'venir (revenir ‘return’). Onset tokens such as these do not obviate the regularities noted in (5), however, as similar processes result in clusters of any consonant types, including plosives (e.g. d'venir/devenir ‘become’) and nasals (e.g. n'veu/neveu ‘nephew’, v'nu/venu ‘come-PART’). With the exception of certain loans and scholarly words (e.g. pseudonyme, tsar), which may be analysed as affricates, fricatives are only the rightmost member of surface branching onsets resulting from schwa elision.

Similarities seen in (5) notwithstanding, /ʀ/ and /l/ distributions are not identical. /ʀ/ occurs in complex onsets following any plosive, regardless of the latter's place specification, while /l/ does not occur following coronal plosives. Conversely, /l/ may follow coronal fricative /s/ in a complex onset (e.g. slave, [slav] ‘Slavic’). /ʀ/ only appears in these positions due to schwa elision (e.g. s'ra/sera ‘be-3Sing’), as well as in a handful of loanwords (e.g. Srebrenica). While many /sl/ tokens are indeed loanwords, these are arguably more frequent and better integrated in the contemporary lexicon than are /sʀ/ tokens. Tellingly, data from Russell Webb (Reference Russell Webb2002) indicates that approximant [] is heavily favoured in both isolated tokens and sequential speech when such onsets arise from schwa elision.

Additional phonological data regarding possible statuses of /ʀ/ are observed in several regional varieties. Côté (Reference Côté1997, Reference Côté2000, Reference Côté2004) presents expanded analyses of QF final cluster simplification, as in (6), which is blocked by /ʀ/.

1. (6) QF Cluster Simplification (Côté, Reference Côté2004: 154-155, 168-169)

   

In coda clusters the final consonant is not elided when preceded by /ʀ/; according to her analysis, /ʀ/ patterns with /j/ and is distinct vis-à-vis /l/, nasals and fricatives.Footnote ⁸

/ʀ/ is an oddity in the phonological inventory of French, highlighting a central question of the present work. This is the only segment whose surface forms alternate along a continuum from fricative to approximant, also inheriting voice quality from the phonotactic environment. /ʀ/ is historically one of the few segments which have undergone a radical shift in place of articulation (i.e. from coronal to dorsal) and which is phonetically and phonologically variable in regional forms and registers. This is not to say that other French consonants are invariable. Liquids and nasals are also subject to voice assimilation, for example, but do not show fortis-lenis variability like /ʀ/. By the same token, intervocalic voiceless obstruents are often partially voiced (e.g. été [et ˯ e] ‘summer’) and final voiced obstruents may be partially devoiced (e.g. cède [sɛd ˳] ‘cede-3Sing’) (Russell Webb, Reference Russell Webb2002, Reference Russell Webb2004a). These data are, however, phonologically distinct from the above, as obstruents are contrastive for voicing (viz. aider [ede] ‘aid-INF’; cette [sɛt] ‘demonstrative-FEM’), whereas liquid voicing and /ʀ/ aperture are never contrastive.

Systemic associations become more complicated when similarities between /ʀ/ and glides are taken into account. Approximant [] and glides share similar phonetic profiles (vowel like formant structure with the highest resonance peaks below 3000 Hz), but, unlike both stable and derived glides, /ʀ/ is never a constituent of the syllable nucleus. And whereas glide formation involves changes to an underlying vowel, /ʀ/ lenition results in the partial vocalisation of a consonant (and complete vocalisation in certain regional forms and registers). It is also worth noting that [j] (the only glide licensed in codas) blocks vowel lengthening, whereas the opposite obtains in the case of coda [], as in (4).Footnote ⁹

1.3. Accounting for R: featural theory

It emerges from the above discussion that /ʀ/ is somewhat of a chameleon. Phonetically, it resembles either an obstruent or a sonorant, depending on its position and (for some speakers at least) speech style, while phonologically it participates in natural classes with fricatives and liquids, as well as with glides in some regional forms and registers. At the same time, /ʀ/ is unique vis-à-vis both obstruents and all other sonorants, which are resistant to aperture variation. The variability characterised by French /ʀ/ is hardly unique and, perhaps for that reason, interest in rhotics is recurrent within the phonological community. This section provides a brief overview of selected works which account for behaviour of /ʀ/ and the cross-linguistic grouping of rhotics in different languages. Those which do not address representational and/or featural issues (e.g. Lindau, Reference Lindau and Fromkin1985; Widdison, Reference Widdison1997) are not reviewed, although their consultation may be useful to readers unfamiliar with questions of rhotic classhood.

Most phonological accounts of French /ʀ/ surface patterns are founded on abstract representations constructed from a relatively small set of distinctive features, e.g. Chomsky and Halle (Reference Chomsky and Halle1968; see also Jakobson, Reference Jakobson1939; Jakobson, Fant and Halle, Reference Jakobson, Fant and Halle1952). A survey of phonological literature from the past three decades reveals that the putative structure assigned to /ʀ/ is far from uniform. Three examples, among the dozens of possible accounts, are provided in (7).

1. (7) Featural Specifications of /ʀ/

   

Each underlying representation calls upon different units in its account of /ʀ/. Bibeau (Reference Bibeau1975) groups /ʀ/ with non obstruents by virtue of the feature and value [+vocalic]. On the other hand, Dell (Reference Dell1980) employs more traditional featural notation ([+sonorant]) to facilitate this classhood, whereas Brousseau and Nikiema (Reference Brousseau and Nikiema2001) provide that /ʀ/ is [+syllabic], alluding to autosegmental issues. Little discussion of /ʀ/ surface variability is undertaken by these authors, despite acknowledgement of the intermediate nature of /ʀ/. The exception to this is voice assimilation, attributed to sonorant underspecification for laryngeal features by Dell (Reference Dell1980: 57–58).

A handful of recent works address the phonological status of /ʀ/ with more limited focus. Côté is primarily interested in cluster simplification in QF, ignoring both distribution and representation (Reference Côté2004: 159). Her classification of /ʀ/ as a glide considers only final and pre-final positions, rather than onsets (in which she acknowledges /ʀ/ to pattern with fricatives). Colantoni and Steele (Reference Colantoni and Steele2005) also question the grouping of /ʀ/ and /l/ among liquids. Given historical asymmetries between the two sounds, as well as frequent vowel intrusion in QF complex onsets, they conclude that /ʀ/ is a fricative (2005). Russell Webb presents a distinct approach to French /ʀ/ representation, taking into account voice (Reference Russell Webb2004b) and aperture (Reference Russell Webb2004a), as well as the systemic relations between phonemes and their input (2002). He assumes that underling representations may be minimal, but proposes no formal structure to this end.

Looking beyond the question of French /ʀ/, a review of works focusing on the cross linguistic status of rhotics underscores the difficulties associated with the phonological status and content of these segments. Walsh Dickey (Reference Walsh Dickey1997) argues for an abstract class [rhotic] as a structural byproduct. According to her analysis, formalised along the lines of Clements’ (Reference Clements1985) feature geometric model, different rhotics share a secondary laminal node under the coronal sub-node. The relatively rich featural content of rhotics explains their cross-linguistic patterning, most notably resistance to palatalisation. In her analysis, dorsal rhotics are specified as having both coronal and dorsal place specification, as well as pharyngeal specification, as in Figure 1. Under such an interpretation, all rhotics contain a secondary laminal node and are structurally both coronal and dorsal.

Figure 1. Dorsal Rhotic Featural Representation (Walsh Dickey, Reference Walsh Dickey1997: 138, her 3.18, modified).

A distinct approach is taken by Wiese (Reference Wiese and Hall2001), who accounts for cross-linguistic patterning as the result of rhotics’ unique position on a sonority scale, derived from Clements (Reference Clements, Kingston and Beckman1990), as in (8).

1. (8) Rhotic Sonority (Wiese, Reference Wiese and Hall2001: 355, emphasis of the original)

2. obstruent < nasal < lateral < /ʀ/ < glide < vowel

Following his formalisation, rhotic patterning is attributable to this segment's situation intermediate to /l/ and glides. As a logical extension, distinct sonorities might be combined or interleaved if a given rhotic patterns with another segment (e.g. /l/) in a given language. It is, however, unclear what precise representational structure should be viewed as the foundation of rhotic sonority.

Van Oostendorp (Reference Oostendorp, van de Velde and van Hout2001) presents the case for the Dutch rhotic as a placeless, empty consonant in Brabant and Limburg dialects, also reviewing varieties of English. He argues that rhotics do not occupy a fixed point on the sonority scale, but are chameleonic with regard to consonantality, i.e. may or may not be a consonant and thus may pattern with sonorants. Variability is attributed to the relative ranking of the markedness constraint Final-C (‘a phrase (word) should end in a consonant’), and proposes that the Dutch rhotic is ‘particularly sensitive to this because /r/ is not predetermined for place’ (2001: 116, 114). In essence, van Oostendorp posits that rhotics lack featural content and that their surface forms are determined by the grammar. While he alludes to representational emptiness or minimalism in other languages, he makes no formal proposal to this end. A similar proposal is made by Rose (Reference Rose2003) concerning QF /ʀ/ based on child acquisition data, although it is unclear as to whether the placelessness of a child's representation changes over time, eventually including specification for dorsality or the like. It should be noted that, upon grammatical maturation, the vast majority of French speakers do not show place variation similar to that seen among Dutch speakers.

2.1. Theoretical Background

The traditional and nearly unanimous assumption (see e.g. Clements and Hume, Reference Clements, Hume and Goldsmith1995: 245) that features are universal has come under increasing scrutiny in recent years. Boersma, for example, acknowledges the quasi-universal nature of articulatory and perceptual space, but denies the universality of featural values. Accordingly, the phonological building blocks of a language are defined by use and segmental representations include only relevant values (Reference Boersma1998: 172). A similar approach is taken by Mielke, who argues against the primacy of phonetic universals in the establishment of features, positing the latter as ‘abstract categories based on generalisations that emerge from phonological patterns’ (Reference Mielke2004: 7). According to EFT, features are born of generalisations made during acquisition, based on phonological evidence, and are only indirectly grounded in diachronic change or phonetic form (2004: 118–121, 126–127). The apparent universality of certain featural distinctions such as place (e.g. [labial]) or manner (e.g. [continuant]) is motivated by segmental participation in phonologically active classes, and putative cross linguistic universals are considered epiphenomenal, rather than innate (2004: 141–142).

Focusing on the question of rhotic cross linguistic patterning, Russell Webb (Reference Russell Webb2002) begins with the presupposition that features are not universal, based in large part on Boersma (Reference Boersma1998). Similar to EFT, he proposes that features derive from learned patterns of similarity and difference within a system, essentially from the holistic productive and receptive input made available to language users. It is hypothesised that a given segment may be defined according to intrinsic features (e.g. backness or dorsality), as well as the absence of features (e.g. the lack of high frequency noise), both of which provide for systemic contrast. Following the stated denial of featural universality, the rhotic of one language cannot be directly defined in reference to another or its form motivated by cross linguistic patterns. Cross linguistic comparison may be made between inventories and phonological dynamics (i.e. systems of similarity and distinction), rather than unique segments, but this only indirectly establishes rhotic classhood (2002: 163–175).

The implications of abandoning innate features are far reaching, especially in cases such as the one at hand. Rather than depend upon a fixed set of universal phonological units, requiring that a segment be fit to the set, it is possible to construct a phonological representation dependent on surface data, essentially deriving the feature set from surface forms and their observable phonetic and phonological characteristics. Unlike approaches founded on innate features and values, where function is induced from form, an emergent approach provides that form is deduced from function (function being allegorically understood as linguistic performance). From this perspective, it is perfectly feasible that one speaker's /ʀ/ is distinct from that of another, even if both are speakers of a common language. Similarly, the features and values assigned to the rhotic of one language or regional variety of a language need not be assumed of others. In the following section, it is argued that the chameleonic nature of /ʀ/ in convergent French leads to the establishment of a minimally specified abstract representation, i.e. one which contains little featural information. It is further proposed that different surface forms and distributions seen in regional forms of French may lead to the emergence of variant representations in some cases.

Before examining emergent features and their application to French, brief mention should be made of the repercussions of this approach for contemporary phonological analysis. The presumption that features are not innate poses a distinct problem for OT, especially given its principle of Richness of the Base (ROTB; Prince and Smolensky, Reference Prince and Smolensky1993: 207). This provides that no constraint holds at the level of input and implies that all grammars should contend with all possible input. At the heart of this matter are representational underspecification and the status of redundant phonological features in the input. Itô, Mester and Padgett (Reference Itô, Mester and Padgett1995) demonstrate that OT's non-derivational framework denies any serial ordering or minimalism, providing that underspecification is an output – rather than an input – property. This approach to indeterminate representations is distinct from that taken by Kirchner (Reference Kirchner1998) and Russell Webb (Reference Russell Webb2004a), who argue that input may be assumed to lack structural information as a means of avoiding circuitousness (see also Boersma Reference Boersma1998). This implies a weak understanding of ROTB. At a lexical level, input is unconstrained, but productive input, such as that under discussion here, is finite. The present approach assumes full specification at a lexical evaluative level, which is filtered by declarative constraints (see Myers, Reference Myers1999; Gess, Reference Gess and Holt2003, Reference Gess2004). While full treatment of this question far surpasses the scope of the present work, it should be noted that emergent features are compatible with OT and the proposal made here does not, a priori, invalidate ROTB, although this is necessarily considered a theorem, rather than a principle of linguistic production (see e.g. Hale & Reiss, Reference Hale and Reiss1998; Reiss, Reference Reiss2000; see also McMahon, Reference McMahon2000, Reference McMahon2007).

2.2. Emergent R

In order to contextualise discussion of /ʀ/ and related segments, the consonantal phoneme inventory of French is provided in Table 1, noting place and manner characteristics. A first step in accounting for /ʀ/ particularity is the establishment of features and featural values promoting its distinction and similarity vis-à-vis other segments. Given the phonological and phonetic evidence confronting language users and focusing for the moment on convergent French, a series of unary place and binary manner and major class features is employed. Foremost among these is [vowel], accounting for segments which may feature in syllabic nuclei (note that glides are specified as [+vowel] and [+sonorant]).

Table 1. French Consonant Inventory: adapted from Walker (Reference Walker2001: ix)

Place (e.g. [coronal]) and most manner features (e.g. [nasal]) are straightforwardly posited from the assumptions outlined above, i.e. that language users posit features and the values assigned to them based on their acquisition and use of the language. For example, a user is attentive to the contrast between poux [pu] ‘lice’, tout [tu] ‘all’ and cou [ku] ‘neck,’ leading to distinct place values. Likewise, the feature [vowel], is assigned to segments which appear in syllabic nuclei (i.e. vowels and semi-vowels) and which thus constitute the necessary element of a rhyme. Voicing is assumed to be specified only for those segments that are contrastive in this regard, i.e. fricatives and plosives. This is presented as [±voice], although emergent values might also be subsumed under privative headings such as [voice] and [spreadglottis] (see e.g. Jessen and Ringen, Reference Jessen and Ringen2002).

The manner feature [sonorant] is posited based on phonotactic evidence available to users, notably the ability of some segments to take part in complex onsets and codas. From this it is assumed that a language user distinguishes between nasals and liquids on the one hand and fricatives and plosives on the other. The former are frequently in complex syllabic structures and are relatively unrestricted in this regard; by contrast, the latter are far more restricted in their distribution (the question of /s/ not withstanding). This represents a departure from traditional definitions of sonorance, which derive from notions of spontaneous voicing (e.g. Chomsky and Halle, Reference Chomsky and Halle1968: 302) or aperture (e.g. Clements, Reference Clements, Kingston and Beckman1990). The major manner feature [obstruent] is associated with the class of segments whose surface forms require constriction in the oral cavity. Complementary to the definition of [sonorant], [obstruent] is further assumed to emerge from the relative restriction placed on a sound's distribution and potential to constitute complex syllabic structures. Finally, [continuant] is assumed to emerge from the distributional, articulatory and perceptual properties of sounds that do not require an oral cavity constriction. This is similar to Chomsky and Halle, who ascribe [continuant] to segments having an unimpeded oral airstream, providing uninterrupted source noise (1968: 317–318), although considerations outlined below add precision as to the nature of constriction in question.

A preliminary featural specification of French continuants is given in Table 2. In this matrix, values are assigned following Radical Underspecification Theory (Archangeli, Reference Archangeli1988; Stemberger, Reference Stemberger1991; Cohn, Reference Cohn1995), according to which an underlying representation contains only those specifications promoting distinction, with surface values being filled in by the grammar. As noted by question marks in Table 2, the status of /ʀ/ and /l/ with regard to [continuant] requires additional discussion. In the case of convergent French /ʀ/, language users are confronted with a sound which can never be in a syllable nucleus, i.e. a sound which does not contain the value [+vowel], but which differs from all other non-vowel segments as it does not involve specific constriction or closure target and is contextually variable for both manner and voice quality. Importantly, only /ʀ/ implies surface forms whose aperture or major manner characteristics track closely to context. Distinction of /ʀ/ from fricatives and /l/ derives from the latter sounds’ relatively stable articulatory targets and contextual aperture invariability.

Table 2. French Consonant Inventory: features and values

The status of /ʀ/ and, especially, /l/ with regard to [continuant] has long been a subject of debate in the phonological community (see e.g. van de Weijer, Reference Weijer1995) and serves as an important illustration of EFT (Mielke, Reference Mielke2004: 241–246, 267–270). In the present work, [continuant] is posited based on vowel lengthening and cluster restrictions, indirectly reflecting phonetic considerations. In the first instance, as in (4a), it is noted that vowels are lengthened before voiced fricatives and /ʀ/, but not before voiceless fricatives and /l/. Assuming that differential patterning involving fricatives is due to anticipatory laryngeal constriction impeding vowel lengthening, the crucial distinction lies between /ʀ/ and /l/. /ʀ/ surface forms do not result in complete contact between the articulator and articulatory target, allowing a central airstream channel, similar to that of vowels. In contrast, surface forms of /l/ (as well as /j/) involve mid-sagittal closure and a lateral air channel, more closely resembling that of plosives (Hirose, Reference Hirose, Hardcastle and Laver1997; Stevens, Reference Stevens1998: 543–555). Additional phonological evidence for /l/ specification as [−continuant] is gleaned from its distribution within onset and coda clusters, as in (5). /ʀ/ is more widely licensed in triple branching onset clusters as compared to /l/. Confronted with phonological input similar to this, it is plausible that learners posit /ʀ/ and fricatives as [+continuant] and /l/ and [−continuant]. These considerations lead to the representational values given in Table 3, where /l/ is specified as [−continuant, +sonorant] and /ʀ/ as [+continuant, +sonorant].

Table 3. /ʀ/ and /l/ specification

A visual model accounting for phonological classes is provided in Figure 2. Using only the values [+sonorant], [+continuant], [+obstruent] and [+vowel], it is possible to account for the phonetic and phonological patterns discussed in §1. Consistent with Table 3, a liquid is any non-nasal, non-vowel segment containing the value [+sonorant], whereas a fricative is any non-sonorant specified as [+obstruent, +continuant]. /ʀ/ patterns with fricatives, as both are specified as [+continuant], while the conjunction of /ʀ/, /l/ and glides (exemplified by /j/) is attributed to the value [+sonorant]. /ʀ/ may also be represented in a feature geometric model, following Clements (Reference Clements1985), as in Figure 3. It is worth noting that minimalism in this instance should not be conflated with emptiness, akin to van Oostendorp (Reference Oostendorp, van de Velde and van Hout2001) or Rose (Reference Rose2003), as /ʀ/ is specified for a major root, as well as place and manner features. Furthermore, /ʀ/ does not demonstrate the phonological behavior characteristic of so called empty consonants, e.g. epenthetic status or major place variation, as in Dutch, where rhotics can be coronal or dorsal.

Figure 2. Phonological classes.

Figure 3. Feature geometric representation of /ʀ/.

3.1. Accounting for Surface Forms

Constraint interaction in French highlights two important points concerning the formalisation of linguistic competence in a production grammar. Firstly, the representational minimalism of /ʀ/ renders it particularly sensitive to effort reduction and positional augmentation. Secondly, other segments, whose representations are relatively richer, are unaffected by these factors. Discussion and tableaux presented in this section concern the relative ranking of productive markedness and faithfulness, highlighting the effect of these on /ʀ/ inputs and their lack of effect on other segments, notably fricatives, /l/ and glides.

The relative ranking of markedness constraints is noted in Tables 4a through d (here, voiceless /ʀ/ is represented by [χ] in the output). In Table 4a, only [gy] satisfies -Effort(glot); approximant and vocalised candidates are eliminated by Onset(obs). The converse obtains in Table 4b, where the optimal output contains a voiceless fricative, this a consequence of the relative ranking of -Effort(glot). Interaction in Table 4c establishes the crucial ordering of −Effort(V_V), −Effort(glot) » Onset(obs), predicting the attested output [aab] for input /aʀab/ and blocking onset fortition in intervocalic environments (e.g. *[aab]). Here, the grammar predicts an outcome mirroring data from Russell Webb (Reference Russell Webb2002) and Tranel (Reference Tranel1987). In syllabic codas, as in Table 4d, the grammar predicts that the minimally specified /ʀ/ will be interpolated as an approximant [].

Table 4a. /ʀ/ in complex onset following voiced obstruent, grue ‘crane’

Table 4b. /ʀ/ in complex onset following voiceless obstruent, cru ‘raw’

Table 4c. Intervocalic /ʀ/, arabe ‘arab’

Table 4d. Word-final /ʀ/, perte ‘loss’

Table 5 presents an alternative possibility involving the ranking of −Effort(V_V) vis-à-vis Onset(obs). This grammar predicts equally optimal outputs, [aab] and [aab]. The question is purposefully left unanswered as to whether these constraints should be coranked or variably ranked, following proposals concerning output variation, such as Auger (Reference Auger2001) and the possibility of floating constraints or Espinosa (Reference Espinosa2004), who follows integrates work by Boersma (Reference Boersma1997), Hayes (Reference Hayes, Dekkers, van der Leeuw and van de Weijer2000) and Boersma & Hayes (Reference Boersma and Hayes2001) by situating competing constraints within a constriction band (see also Anttila, Reference Anttila, Hinskens, van Hout and Wetzels1997, Reference Anttila, Chambers, Trudgill and Schilling-Estes2002). These sources consider style shifting in registers or regional forms of a heteronymous language, while the present work endeavours to articulate a framework within which either constriction bands or floating constraints can be accommodated.

Table 5. Intervocalic /ʀ/, arabe ‘arab’: Co-optimality

Because the /ʀ/ input representations contain no information relative to manner of articulation, Pres has little effect on the selection of productive output structural descriptions. This constraint proves crucial for prediction of attested outputs involving /l/, /j/ and fricative inputs, however, as shown in Tables 6a through e. Following Tables 2 and 3, fricatives, glides and /l/ have relatively richer phonological content. Highly ranked Pres is crucial to output selection in these instances, as this constraint militates against the non-correspondence of input features to output form. In Tables 6a and b, Pres violations eliminate candidates in which the feature [obstruent] is unfaithful to input values. Pres also rules out possible changes in voicing and militates against infidelity to input featural information, e.g. candidates having a fricated semi-vowel, as in Table 6c, or those containing a lenited fricative, as in Tables 6d and e.

Table 6a. Word-initial /l/, blanche ‘white-FEM’

Table 6b. Word-final /l/, calte ‘flee-3S’

Table 6c. /j/, paye ‘pay-3S’

Table 6d. /s/, peste ‘plague’

Table 6e. /s/, plaça ‘place-3S/PAST’

3.2. Regional Variation

This section discusses the compatibility of the proposed grammatical model with data from regional forms of French. While it is far beyond the scope of the present article to address all regional variation, it is noted that the approach to both emergent representation and grammatical formalism is amenable to different input and divergent constraint rankings.

One possible solution to the question raised by alternate surface forms is predicated on the assumption that the phonological /ʀ/ representations may be more richly specified in some instances. The lack of aperture allophony in BF, for example, might be accounted for by /ʀ/ being specified as [+obstruent], as would obtain were learners to include /ʀ/ in the category of sounds having specific articulatory targets involving oral cavity stricture. This follows from data in (2) and the discussion of the feature [obstruent] in §2. Assuming for the purpose of discussion that the underlying representation of BF /ʀ/ contains the values [dorsal], [+continuant] and [+obstruent], no changes to the grammar articulated in Tables 4 or 6 need be made; this input would not be insensitive to phonetically-based constraints such as -Effort(V_V), as these are dominated by Pres. Questions of register shifting notwithstanding, this approach requires that BF speakers always produce /ʀ/ as a fricative and that positional or stylistic variation is absent, a possibility supported by the cited sources as well as anecdotal evidence. Assumption of more richly specified representations does not account for the surface forms of /ʀ/ in QF or STF, however, indicating that the difference between these and other varieties French may be due to grammatical distinction (i.e. a constraint ranking that is different from that given in Tables 4 through 6).

In her analysis of QF cluster simplification, Côté (Reference Côté2004) refers to the relative perceptual salience of consonants in the output. She notes that consonants adjacent to a vowel or to a consonant which is vowel like are maintained. This is explicated in reference to the acoustic transition from one segment to another in output, which either provides cues to consonant identity (e.g. from vowel or approximant to stop) or does not inhibit the transmission of segment internal cues (e.g. from vowel or approximant to fricative). She also shows that elision of plosives is more frequent and is more likely in the case of input sequences having the same place of articulation (2004: 178–180). This is analysed in an OT grammar comprising constraints targeting the optimal position of consonants (2004: 185–190).

While the approach to features and featural specification in Table 3 and Figure 2 is distinct from that included in Côté (Reference Côté2004), emergent, minimal specification of the input is not incompatible with her analysis (theoretical and formal differences regarding features notwithstanding). Following the proposal, post-vocalic /ʀ/ - obstruent clusters do not violate markedness constraints promoting output well-formedness as a measure of perceptual salience. /ʀ/ minimal specification presents the added advantage of capturing both the vowel like qualities of /ʀ/ (i.e., /ʀ/ is not an obstruent, but is a continuant), which are directly called upon in her perceptual account of cluster simplification (Côté, Reference Côté2004: 176). Minimal representation also facilitates the distinction of /ʀ/ vis-à-vis /l/, accounting for the differential treatment of clusters containing /ʀ/ and /l/.

A final issue that can be addressed is the vocalization of /ʀ/ in syllabic codas in QF and STF. Côté (Reference Côté2004) analyses QF vocalisation as the effect of markedness constraints promoting featural similarity among adjacent segments, input-output correspondence targeting featural specification (e.g. stop vs. non-stop) and position (notably vowel adjacency), and output-input uniformity, militating against multiple output-input correspondents. Data such as (3a) are predicted in her analysis by the ranking of Max-C (‘input consonants should have an output correspondent’) above Uniformity-V (‘No vowel in the output corresponds to itself and another segment in the input’) and C↔V (‘a consonant is adjacent to a vowel’), the latter of which are co-ranked (2004: 187–190). While the featural terminology and formalisation are different from those advanced above, a minimally specified /ʀ/ representation is compatible with Côté’s analysis. The distinction between QF and STF vis-à-vis French is captured in the relative ranking of either Pres (as in Table 4a) or Uniformity.