2.1 A class of rhotics?

Hyper-variation in R spans virtually every aspect of its articulation (Lindau 1985; Ladefoged & Maddieson 1996). In terms of manner of articulation, all non-stops are possible: from trills and taps/flaps to fricatives, approximants, and even vowels. In terms of place, R can be either coronal, articulated by the front part of the tongue against the dental/alveolar or post-alveolar region, or dorsal, articulated by the tongue dorsum against the velum or uvula. And while most R-sounds are voiced, voiceless allophones are commonplace, and voiceless phonemes are not unknown either. Such variation is present across languages as well as across dialects of the same language. Also within a single variety, it is usual to find free and phonologically-conditioned allophony for the usual single R-phoneme (Wiese 2011), or less usually two or more contrasting R phonemes (Maddieson 1984).

A key question is, then, do these phonetically diverse R-sounds form a class? From the above, it is neither manner nor place of articulation that unifies them (Ladefoged & Maddieson 1996: 215). Nor is there a single acoustic property, either. Lindau (1985), for instance, rejects a lowered third formant as a common denominator. So, with no phonetic similarity among them, rhotics have a merely conventional class membership based on orthography and/or diachrony.

However, despite their large phonetic diversity, rhotics exhibit unmistakable phonological homogeneity, often behaving as a class in terms of distribution and phonological activity; see Wiese (2001; 2011). Phonotactically rhotics appear in positions reserved for sonorants, even when the actual realization is non-sonorant (Chabot 2019: 8). For instance, they tend to have syllabic variants, and to be closer to the syllable nucleus when part of complex onsets or codas (Proctor 2009) – two properties they share with laterals and nasals. They also tend to merge with contiguous vowels, and to alternate with other rhotics (Ladefoged & Maddieson 1996). On that account, the term “rhotic” is warranted.

But even if rhotics seem to work as a class, a feature like [+rhotic] (Hall 1997) appears circular and ad hoc, as it merely offers a label with no phonetic meaning; see Wiese (2001) and Sebregts (2014). Chabot (2019: 5) argues, rightfully, that there is nothing exceptional about the behavior of R that would justify the introduction of a new phonological feature. Rhotics, he continues, “are sonorant phonemes like any other”, and they are only special because of their phonetic heterogeneity. After all, if there is no exclusive set of rhotic sounds (Scobbie 2006), then why should there be a feature, or set of features, associated exclusively with rhotics? This appears like a representational conundrum, at least for theories that require universal, phonetically-grounded features; but less so for theories that assume phonologically abstract, non-universal features (see Chabot 2019), such as those used in this study.

2.2 Minimalist feature geometry

To offer a representational analysis of Arabic R, I will adopt a fairly standard version of the Parallel Structures Model (PSM; Morén 2003; 2006; 2007; inter alia). The PSM is a modular, minimalist, and contrast-based model of feature geometry in which consonants and vowels exhibit parallel structures and identical features for place, manner, and laryngeal articulations. It incorporates insights from various other proposals, particularly Unified Place Theory (Clements 1991; Clements & Hume 1995), Dependency Phonology (Anderson & Ewen 1987), and Element Theory (Harris & Lindsey 1995).

Features in the PSM are abstract entities in that they are not phonetically determined, and they are specified in the geometry only if there is positive phonological evidence for their existence (cf. Anderson 1981; Clements 2001). In other words, only distinctive features are ever used in the grammar, whereas redundant features that make no difference in phonological patternings are eliminated. It follows that we should not expect identity across languages, or varieties of the same language, with respect to the composition of particular segments, even though we do find overwhelming similarity due to the universal phonetic properties of speech sounds. By parallelism in features we mean that they refer to broadly defined phonetic properties that may describe consonants and vowels alike. All features are supposed to be learned, rather than hard wired, à la Emergent Feature Theory (Mielke 2008). They are treated as cognitive categories, “discovered” on the basis of actual linguistic input.

How a particular terminal feature is interpreted in the PSM depends on its relationship to a superordinate node in the representational hierarchy. As illustrated in Figure 1, each place, manner, or laryngeal feature can be associated with a C-class or a V-class node, and the V-node is dependent on the C-node. The motivation behind this architecture is to create a unified machinery that captures consonant-vowel interactions and acoustic/articulatory parallelisms in natural language. To account for their asymmetries, consonants are allowed to have both C- and V-terminal features, while vowels can only have the latter. Another architectural mechanism of the model is building complex structures from simpler ones which, together with the dependency principle, ensures an effective degree of representational economy.

Figure 1

PSM basic geometry.

The three basic node types in the PSM deserve some elaboration here. Under the Place tier, we will use the articulator-based place features [labial], [coronal], and [dorsal] under the C-place node and its daughter V-node (cf. Clements 1991). Simple consonants have one place feature; a plain /r/, for example, could be specified for C-place [coronal]. Complex consonants have multiple features on the same place node (Morén 2003: 233); so a more complex type of /r/ could have C-place [coronal] and C-place [dorsal]. Finally, consonants with secondary articulation have features on both C-place and V-place nodes; hence a pharyngealized /rˤ/ could be specified for V-place [dorsal] in addition to C-place [coronal].

The structure of the Manner tier parallels that of the Place tier: a V-manner node is nested in a C-manner node, and both nodes make use of the loosely defined features [open] and [closed]. This arrangement neatly reflects the relative sonority of segments (Morén 2003: 222–223); so for instance a sonorant R could be represented by a combination of C-manner [open] and V-manner [closed], whereas a fricative R would have only C-manner [open]. Finally, the Laryngeal tier has the same architecture. For present purposes, however, C-laryngeal [lax] will be sufficient to distinguish voiced from voiceless obstruents (see Morén 2003: 230).

When we discuss segmental representations in the PSM, it is crucial to separate underlying from surface levels. Our concern is contrastive phonemes, not predictable allophones and phonetically-driven changes, which can be supplied by rule. The feature composition of these phonemes is based on their phonological distribution and activity, as directly observable in the surface data. As a consequence, the PSM forces us to look carefully at a wide range of phonetic details and (morpho)-phonological alternations in individual languages and to take into account the fine interactions between different processes. And this is what I will attempt to do here.

3.1 Phonetic variation

In most Arabic dialects, the rhotic is realized as a voiced dental/alveolar tap or trill (Younes 1994; Watson 2002). The singleton R is often a tap [ɾ], a single short apical closure, whereas the geminated R is a trill [r]. According to Ladefoged & Maddieson (1996: 219), this is generally the case for languages that regularly distinguish between singleton and geminated consonants, such as Italian and most varieties of Arabic. Arabic trills are found where the templatic pattern dictates a geminate in morphologically and semantically predictable classes. And phonotactically, they arise only where geminates are licensed: in intervocalic and final post-vocalic positions. Taps have no such restrictions. And it is reasonable, therefore, to treat the trills as geminated taps instead of independent phonemes, in parallel with other doubled consonants.¹ For simplicity, I will use [r] for the single tap and [rr] for the geminate.

In addition to the alveolar tap/trill realizations, we find R realized with no contact between the articulators. One type is a posterior fricative [ʁ] or [ɣ] – with close approximation between the back of the tongue and velar/uvular region – which is characteristic of specific speech communities (see Section 7). Another less common type is a frictionless continuant (Shaheen 1979) – either an alveolar approximant [ɹ] or a postalveolar retroflex [ɹ] – which is characteristic of certain sociolects, or of individual speakers where it is often viewed as a deviation from the norm (Younes 1994). Further, any of the rhotic variants can be devoiced, particularly in pre-pausal position, for example in [mitr̥] ‘meter’. And all the above variants exist as singleton and geminate.²

A major issue related to Arabic R is that of emphasis (also known as pharyngealization). As a matter of fact, each of the variants above can, in turn, occur either as pharyngealized or non-pharyngealized (Shaheen 1979: 145–146), thereby doubling the possible range of variants. For present purposes, however, it is sufficient to consider the voiced alveolar tap/trill realizations. Emphatic consonants are typically coronals characterized by a secondary articulation involving the retraction of the tongue dorsum toward the pharyngeal wall, and they are known to extend their pharyngealization to neighboring vowels and consonants within a certain domain; see Kaye & Rosenhouse (1997) and references therein. Arabic R is typically allied with the emphatic consonants (O’Leary 1923), but it stands out in two respects: (1) its influence on neighboring segments seems to be more restricted, and (2) it is known to alternate with a non-emphatic R cognate. This subtle behavior has compounded difficulties for researchers, so much so that Ghazeli (1977: 151) remarks: “In almost every investigation of pharyngealized consonants in Arabic there is a final section on “unresolved problems”, most of which are about the irregular behavior of [r]”.

The differentiation between emphatic (mufaxxama) and non-emphatic (muraqqaqa) R has been documented by Classical Arab grammarians such as Sibawayh, al-Zamaḫšari, and Ibn Yaʿīsh (Cohen 1969; Al-Nassir 1993). As it was never considered a phonemic distinction, they set to carefully determine the predictable conditions under which each variant appears. These conditions persist, more or less, in most eastern Arabic dialects today (Cantineau 1960: 50), and can be summarized as follows. The pharyngealized [rˤ] is found contiguous with back vowels [ɑ/ɑː, oː, u/uː] and in proximity to emphatic obstruents /tˤ, dˤ, sˤ, zˤ, ðˤ/ or uvulars /χ, ʁ, q/. The plain [r] is found adjacent to front vowels [a/aː, eː, i/iː] if no other emphatic is present. In most western dialects, on the other hand, the opposition takes on a distinctive value so clear that we can speak of two, plain versus emphatic, R phonemes. The essence of these generalizations will be elaborated and discussed in Sections 4–7 below.

3.2 A phonological micro-typology of Arabic R

Various Arabic consonants, with recognizably diverse reflexes, have been studied from a typological perspective; for example, the interdentals /θ, ð/, uvular /q/, velar /k/, affricate /ʤ/, but never R (see Holes 1995; Versteegh 1997; Watson 2011). The lack of dialect maps or isoglosses that demarcate variants of Arabic R does not reflect a denial of variation, but rather the fact that R’s behavior displays seemingly random variation and rarely conforms to clear-cut divisions. In my attempt to set up the R-typology, therefore, I had to closely examine this behavior in a wide range of Arabic dialects: six based on first-hand research and no less than twenty based on secondary sources. My goal was to answer three sets of questions:

1. How many R sounds exist in a given variety? Which patterns exhibit contrast and which exhibit structurally conditioned allophony?

2. What is the underlying form for each phoneme? And what evidence exists for assigning it a specific representation in the PSM?

3. What are the patterning similarities and differences across Arabic dialects? Do the differences give rise to concrete isoglosses?

Last question first: Yes, they do. Figure 2 shows four major dialect types based on the quality and number of R phonemes: the split-R in purple (Type I), the emphatic-R in green (Type II), the plain-R in yellow (Type III), and the scattered uvular-R dialects in orange (Type IV). The answer to questions (i) and (ii) will be different among these groups. Types II and III, for instance, have one R phoneme each; but the former is underlyingly emphatic while the latter is plain and dorsal. Types I and IV, on the other hand, each have two contrasting R phonemes: plain-emphatic and plain-uvular respectively. The representations of those R phonemes are clearly non-identical; and as stated in Section 2, they are entirely determined by the distribution and phonological activity of R on a dialect-by-dialect basis. We therefore typologize the rhotics phonologically, not phonetically. I argue that adopting this approach allows us to draw sharp distinctions between dialect types as well as between representational primitives, without which it would be impossible to construct the typology.

Figure 2

Distribution of R types in spoken Arabic dialects.

But before embarking on the details, a few disclaimers are in order. First, the map offers a simplified representation of the groups, which are not meant as highly precise or exhaustive isoglosses. Secondly, disregarding any phonetic and sociolinguistic variation, it is possible, and likely, that there is also some phonological variation within these large dialect groups. Finally, all generalizations in this article build on the six dialects studied at first hand. The placement of other dialects in any group is done to the best of my knowledge, on the basis of data and descriptions in secondary literature. Now we move on to examine the groups.

4.1 The distributional facts of R

The literature shows substantial agreement on a phonemic distinction between a plain /r/ and an emphatic /rˤ/ in Egyptian Cairene (Schulte 1985; Watson 2002; Woidich 2006; Youssef 2014) and Moroccan Arabic (Singer 1980; Heath 1987; 1997; Caubet 1993; Aguadé 2003; Gouma 2013). Despite some partial complementary distribution between plain and emphatic R (see below), the R type is not always predictable from the phonological environment, and two separate phonemes are postulated in order to handle the existing patterns. The most compelling evidence is the minimal pairs contrasting /r/ and /rˤ/, exemplified in (1a–b) for ECA and MCA, where the R’s show up in different positions in the word and the syllable.

In view of the extension of emphasis over whole words (indicated by dots underneath symbols), one wonders why these are considered contrasts between /r/-/rˤ/ segments, rather than between prosodic units – both valid ways of analysis. Proponents of the latter approach (e.g. Lehn 1963; Broselow 1976; Tsereteli 1982) recognize no underlying emphatic segments in Arabic, and treat emphasis as a floating prosodic feature that is supplied redundantly in the consonantal and vocalic systems. It has been observed, however, that emphasis (and any plain-emphatic contrasts) is found only in words containing one of the pharyngealized coronal consonants /tˤ, dˤ, sˤ, zˤ, rˤ/ or the marginal “secondary emphatics”. Younes (1982: 40) argues that this must be treated as accidental in the suprasegmental account because the actual phonemic content of the word should be irrelevant. More problematically, this approach grants the prosodic structure access to featural information that is assumed to be lower in the hierarchy, i.e. skipping over the segmental level of representation. For these reasons, a segmental approach where the pharyngealized coronals spread their emphatic feature is more widely accepted (see Youssef 2013: 94–96). Section 4.2 looks more into this process.

All other Type-I dialects are claimed to have two separate /r/-/rˤ/ phonemes on the basis of minimal pair evidence, e.g. [ʒra] ‘he ran’– [ʒ̣rˤạ] ‘it happened’ in Tunisian (Cohen 1970), [naːr] ‘it was lit’– [ṇạːrˤ] ‘fire’ in Libyan (Pereira 2009), [rafraf] ‘he quivered’– [rˤạf̣rˤạf̣] ‘he floated’ in Algerian (Georgin 1980), [sədraːja] ‘jujube’– [ṣə̣ḍrˤạːj̣ạ] ‘tree’ in Mauritanian Ḥassānīya (Cohen 1963), [ʤabar] ‘a broken bone’– [ʤ̣ạḅạrˤ] ‘he forced’ in Sudanese (Dickins 2007), [ḳạrˤrˤạ] ‘he dragged’– [karra] ‘he tore’ in Nigerian (Owens 1993), and [ḳạrˤạ] ‘he has seen’– [kara] ‘he rented’ in Mardin (Jastrow 2006).

It is worth mentioning in this connection that MCA has a more established /r/-/rˤ/ opposition than any other dialect in this group. This is due to paradigm regularization, where we find that a radical /rˤ/ or /r/ of a certain root persists in all words of that root: across inflections, ablauts, and play speech transformations (Heath 1987: 298). The paradigms in (2a), for example, have a consistent emphatic [rˤ] throughout, whereas those in (2b) have a plain [r] throughout, regardless of the nature of the neighboring vowels.³ Heath (1997) compares verbal active participle forms like [ʃ̣rˤə̣ḅ] – [ʃ̣ạrˤə̣ḅ] in (2a) with their corresponding forms in Saharan or Tunisian Arabic which alternate between [rˤ] and [r], and maintains that the leveling has gone farther in Morocco than elsewhere in northwest African Arabic. Gouma (2013: 90) concludes that the /r/-/rˤ/ distinction in MCA is a property of the root, just like the uncontroversial plain-emphatic pairs /t/-/tˤ/, /d/-/dˤ/, /s/-/sˤ/, and /z/-/zˤ/.

Some distributional facts further corroborate the status of /rˤ/ as a phonemically distinct sound in ECA and MCA. While a rhotic in the neighborhood of an obstruent emphatic /tˤ, dˤ, sˤ, zˤ/ must be [rˤ], never [r], irrespective of adjacent vowels (Shaheen 1979; Heath 1987), [rˤ] is retained in numerous words that do not contain those consonants. And while in MCA a uvular or pharyngeal consonant /q, x, ɣ, ʕ, ħ/ in the same stem tends to favor [rˤ], there are cases of plain [r] as well (Heath 2002: 152). Most importantly, the occurrences of [rˤ] in this group are not restricted to the environment of a low vowel /a/: [rˤ] also appears in the onset or coda of a syllable containing non-low vowels without the presence of other emphatics. The data in (3a–b) give instances of [rˤ] adjacent to [ụ, ụː, ọː], in ECA and MCA respectively.⁴

The point here is that if [rˤ] and [r] are found in minimal pairs and in parallel phonological environments beyond an accompanying obstruent emphatic or low vowel, then the two sounds exist in parallel distribution. Cohen (1963) makes much the same arguments for Mauritanian Ḥassānīya. In the Tunis or Sahel dialects of Tunisia, Ghazeli (1977: 161) acknowledges the occurrence of [rˤ] in context-free environments (i.e. independently of the presence of a back segment) or even next to a front vowel in the pair [ḅrˤẹːṭ] ‘I recovered’– [breːt] ‘I sharpened’.⁵

While the prior facts suggest that emphatic /rˤ/ in this dialect group emulates the obstruent emphatics, there remains one distributional fact that is unique to /rˤ/. In ECA, [rˤ] does not occur with tautosyllabic front vowels [i/iː, eː], in which case it loses the emphatic feature and becomes a plain [r] (Broselow 1976; Watson 2002). A similar, albeit less robust, generalization can be made about MCA, whereby a tautosyllabic [i/iː] prevents the realization of emphatic [rˤ] (Heath 1987; Gouma 2013). This distribution, known as /rˤ/ de-emphasis, leads to root-internal allomorphy between emphatic and plain R, if there is no other obstruent emphatic in the word. ECA abounds with alternations like those in (4a), whereas MCA has kept only a few in plural ablauts, some of which are given in (4b). Recall that MCA has leveled out most [rˤ]-[r] alternations within paradigms in favor of one or the other; therefore, some derivatives of a /rˤ/-type stem may have [rˤ] contiguous to a front vowel in the same syllable (i.e. without de-emphasis), as shown in (2a) above. Heath (1997) actually suggests that an originally allophonic distinction in MCA – with [r] next to [i] and [rˤ] next to back vowels – has become phonemic via neutralizations in the conditioning vocalic environments, resulting in the present /r/-/rˤ/ split.

Other Type-I dialects show a parallel correlation between a plain [r] and front vowels, leading to [rˤ]-[r] alternations within word families, e.g. [rˤạː] ‘he saw’– [riːt] ‘I saw’ in Tunisian (Ghazeli 1977), [ḥrˤạḅ] ‘he fled’– [haːrib] ‘fleeing’ in Mauritanian Ḥassānīya (Cohen 1963), and [ʤ̣ạrˤrˤạːħ̣] ‘surgeon’– [ʤaːriħ] ‘wound’ in Sudanese (Dickins 2007). This de-emphasis pattern has led some researchers (e.g. Ghazeli 1977) to reject the two-phoneme analysis of R advanced here, and led others (e.g. Heath 1987) to declare a conflict between the phonemic analysis and an allophonic one in which [rˤ] and [r] are conditioned realizations of a single phoneme. The two analyses are in fact compatible if we accept de-emphasis as a case of contextual neutralization (Kenstowicz & Kisseberth 1979: 211), where the underlying /r/-/rˤ/ contrast is merged in front vowel contexts but is maintained in all other contexts. We simply set up an archiphoneme /r/ in just those contexts where the phonological opposition is suspended.

The last pattern comes from loanword phonology, and is exclusive to MCA. In French loanwords borrowed into MCA, the French rhotic /ʁ/ is systematically interpreted as an alveolar tap /r, rˤ/ (5a), despite the fact that MCA has a phonetically equivalent velar-uvular fricative phoneme /ʁ/ (Paradis & LaCharité 2001; Lahrouchi 2018). Meanwhile, when Arabic words containing /ʁ/ are borrowed into French, they do not appear with the French rhotic /ʁ/, but more willingly with a velar stop /ɡ/, as exemplified in (5b).

Both studies concur that the choice is based on the rhotic’s phonological identity – in this case distinctive features and phonotactic role – rather than its surface phonetic form. More specifically, MCA speakers identify the French rhotic as a sonorant, which for example occupies the nucleus-adjacent position in onset and coda clusters, a description that maps into their native /r, rˤ/, but not their /ʁ/ phoneme. The Arabic /ʁ/, on the other hand, is maintained as an obstruent with a similar place of articulation, /ɡ/, when borrowed into French. The idea is that one and the same phonetic form, say [ʁ], may sometimes be classified as a rhotic (as in French) and sometimes as a fricative (as in MCA): a fact that is hardly puzzling if rhotics are defined in terms of their phonological behavior alone (Wiese 2011; Chabot 2019). This will be the assumption of the featural analysis below.

The distributional facts in this section have confirmed that the presence or absence of emphasis is contrastive for R in Type-I dialects. Minimal pairs, paradigm uniformity, and variety of contexts provided the clues. Moreover, we have seen that /r/ and /rˤ/ can lead lives as autonomous phonemes even though they do not contrast in all vocalic environments. In the next section, we review some phonological processes in which /r/-/rˤ/ participate actively, as a way to understanding their featural representations.

4.2 Phonological processes involving R

In Type-I dialects, not only does /rˤ/ bear an emphatic feature contrastively, but it also extends this feature over adjacent strings of segments in a process known as emphasis spread (ES for short). ES can be defined as a bidirectional, long-range pharyngealization effect. Target vowels often become retracted or more centralized than their non-emphatic counterparts, and are marked acoustically by F₂ lowering (Card 1983; Shahin 2002), while target consonants can be verified by the formant onsets and/or offsets of their neighboring vowels (El-Dalee 1984; Norlin 1987). As a widespread phenomenon across Arabic dialects, ES displays considerable variation with regard to its domain of application, its directionality, as well as the identity of the contrastive emphatic triggers and, if any, opaque segments.

In ECA /rˤ/ has frequently been designated a “secondary emphatic” (e.g. in Schulte 1985; Watson 2002), citing its unique de-emphaticizing behavior and a limited spreading effect. However, Youssef (2014) has shown that /rˤ/ has the same ES domain as /tˤ, dˤ, sˤ, zˤ/, and must therefore be classified as a “true” emphatic. Examples (6a–b) illustrate that /rˤ/ triggers rightward and leftward ES within the prosodic word (including affixes), and that non-tautosyllabic /i, iː/ do not block ES from /rˤ/, as in [rˤạḥịːḅạ] and [ħ̣ịrˤạːṣạ].

In MCA, there is consensus that /rˤ/ belongs with the true (obstruent) emphatics in that it has the same long-range influence they have. The ES domain here is the uninflected word; spreading continues further to prefixes, but not to inflectional suffixes unless they are vowel-initial and surface as tautosyllabic with a stem-final [rˤ] (Benhallam 1980; Heath 1987), as shown in (7). Segments like the front vowel /i/ and the postalveolar consonants /ʃ, ʒ, j/ do not systematically block ES, although they are reported to have limited blocking effects in the left-to-right direction (Heath 1987: 324). This, together with the noted discrepancy between prefix and suffix targets, suggests that ES in MCA is more regressive than it is progressive.

Emphasis spread from /rˤ/ obtains in all dialects of this group; however, some variation in its scope has been noted. For example, Owens (1993) does not differentiate between /rˤ/ and /sˤ, dˤ/ in Nigerian Arabic, and states that ES from those segments extends throughout the word domain, e.g. in [f̣ạrˤẉạ] ‘animal skin’. Abumdas (1985), on the other hand, states that /rˤ/’s influence in Libyan is “almost as strong as” the obstruent emphatics, while Ghazeli (1977) proclaims that /rˤ/ in Tunisian has only a limited, local spreading effect.

Aside from these differences, what does ES tell us about the feature makeup of /rˤ/? Emphatic consonants are distinguished from their plain counterparts by an additional non-primary back articulation (Davis 1995: 472). And since consonants with secondary articulations in the PSM have both C-place and V-place terminal features (Morén 2003: 199), we can posit V-place [dorsal] to indicate pharyngealization in the class of emphatics /tˤ, dˤ, sˤ, zˤ, rˤ/.⁶ The advantage of a V-place feature is that it can spread to consonant and vowel targets of ES, whereas a C-place feature can only spread to consonants (see Section 2.2). Moreover, the V-place [dorsal] on /rˤ/ will be delinked next to a front vowel and no ES takes place. This last ban can be formulated in autosegmental theory as an adjacency restriction on certain feature combinations, e.g. V-place [dorsal] and V-place [coronal], within the syllable.

Another phonological process involving R is the assimilation of the definite article /l/, attested in nearly all Arabic dialects. The definite /l/ assimilates totally to a following word-initial coronal (alveolar/palato-alveolar) – /t, tˤ, d, dˤ, s, sˤ, z, zˤ, ʃ, ʒ, l, n, R/ – and the outcome is a geminate copy of the trigger. Thus in both ECA and MCA (8a–b), plain /r/ and emphatic /rˤ/ triggers produce [rr] and [rˤrˤ] respectively (preceded by epenthesis in pausal position). This is in contrast to words with initial non-coronals which preserve [l] in their definite form, as in [ʔil-bard] ‘the cold’ and [ʔil-ɣasiːl] ‘the laundry’ (ECA). The assimilation is obligatory; and although heavily morphologized, it is synchronically active, as evident from assimilations to initial R’s in recent loanwords (see Watson 2002: 219).

The inference here is that /r, rˤ/, along with the other triggers of /l/-assimilation, are coronal. Since the triggers form a mixture of plain and emphatic consonants, this must be their primary place of articulation. The appropriate feature for this natural class is, therefore, C-place [coronal]. Watson (2002: 220–221) suggests that the main motivation behind this process is an OCP violation on the coronal tier, rejecting representations in which /l/ and the trigger are each linked to its own copy of the feature. The resulting structure with merged C-place [coronal] prompts other features to spread to the /l/, creating a fake geminate.

The third R-related process is coronal sonorant assimilation, by which /n/ and /l/ undergo total regressive assimilation to a following /r, rˤ/ across words and morpheme boundaries, again producing a geminate copy of the trigger. The nasal /n/ may also assimilate to /l/. Assimilations of this type prevail in both ECA and MCA (9) except in very careful speech (Harrell 1957; Heath 1987; Youssef 2013). The underlying forms for phrase-initial words or morphemes in (9) are the same as the surface forms in isolation, i.e. [miːn] ‘who’, [waːkil] ‘eating’, etc. Note that the converse process, assimilation of /r, rˤ/ to /n, l/, is not attested.

These assimilations are common across Arabic dialects. They are described both briefly in grammars and extensively in some studies of segmental assimilation, e.g. Abumdas (1985) and Elramli (2012) on Libyan Arabic. The process verifies that trigger /r, rˤ/ and target /l, n/ are all coronal sonorant segments. (Another indication of the sonorancy of /r, rˤ/ in MCA is the loanword pattern in Section 4.1). Let us assign a combination of C-manner [open] and V-manner [closed] for the natural class of sonorants in the PSM (see Morén 2006: 1210). These features correspond to liquids and nasals being continuants (open) and vowel-like (sonorous). Thus, coronal sonorant assimilation takes place when the trigger and target merge their C-place [coronal] and their manner features to avoid multiple OCP violations.

To summarize, Type-I dialects established a two-way contrast for their R’s: a plain /r/ vs. an emphatic /rˤ/. Despite an apparent hazy region, the two phonemes emerge as independent in minimal pairs and in multiple phonetic environments. Their phonological activity offers support for the feature representations in Figure 3. In terms of place of articulation, /r, rˤ/ have “primary” C-place [coronal] while /rˤ/ has an extra “secondary” V-place [dorsal] to mark emphasis. In terms of manner of articulation, they are both sonorants, with C-manner [open] and V-manner [closed] features. Voicing is redundant and therefore excluded. The following section presents a group of dialects with only a /rˤ/ phoneme.

Figure 3

Representation of /r/ and /rˤ/ in Type-I dialects.

5.1 The distributional facts of R

There are no true minimal pairs contrasting emphatic and plain R in RPA (Younes 1982: 55; Herzallah 1990: 107). All three subdialects consider /rˤ/ to be an underlying emphatic on a par with /sˤ, ðˤ, tˤ/, but with no non-emphatic counterpart. Younes (1982; 1994) provides explicit distributional evidence, both phonological and morphological, of the R phoneme behaving as a “primary” emphatic. The first fact is that /rˤ/ generally co-occurs with a backed [ạ], rather than a front [a], variant of the low vowel /a/. A comparison of the pairs in (10) reveals that /rˤ/ patterns with emphatic /sˤ, tˤ/ but not with the plain consonants in the right-hand column. Remember that these emphatic consonants also occur adjacent to other vowels, as in Type I, so it is unacceptable to assume that they are underlyingly plain, with emphatic surface forms in [ạ] contexts.

The second fact is that /rˤ/ behaves like the “primary” emphatics and the velar obstruents /k, x, ɣ/ in a morphological rule of ablaut. Only when a verb has one of these as the second or third root consonant does the imperfect (roughly the present tense) surface with [u]; otherwise the vowel is [i]. Consider the perfect-imperfect alternations in (11a) vs. (11b).

The final evidence concerns the distribution of the feminine suffix allomorphs. As in many Levantine Arabic dialects, this suffix turns up as [a/ạ] if preceded by an emphatic, velar, pharyngeal, or laryngeal consonant (12a), or as [i] (i.e. with imāla) if preceded by a front consonant (12b). Again, /rˤ/ behaves like the former group in selecting the /a/-type allomorph.

Central to Type-II dialects is the full predictability of the R type from the phonological context. If we can predict all instances of R in which emphasis does not appear, and if the emphatic [rˤ] is the elsewhere variant, then it is reasonable to argue for a single emphatic /rˤ/ phoneme that is de-emphaticized to [r] in those specific contexts. This is the line maintained by Younes, Herzallah, and Shahin for their respective subdialects. All three agree that a non-emphatic [r] is a de-emphaticized /rˤ/ because it is predictable in precisely three well-defined phonological environments.

The first is before a velar /k, x, ɣ/ in the same consonantal root, where we only find the plain [r] variant. Besides /rˤ/, the emphatic fricatives /sˤ, ðˤ/ are also de-emphaticized in this context. Boldface consonants in (13a) are arguably underlyingly emphatic, which means that emphatic [rˤ, sˤ, ðˤ] do not occur to the left of the velars, but occur freely to their right (13b); hence the distribution is “directionally conditioned” (Herzallah 1990: 122–123).

Another context that conditions a plain [r] in RPA is the strict adjacency to one of the non-emphatic coronals /θ, t, d, s, z, n, ʃ, ʧ, ʤ, j/ in the same non-affixed stem (14a). (This has also been reported by Bani-Yasin & Owens (1987) in a Northern Jordanian Arabic dialect).⁷ Unlike the previous context, however, /rˤ/ is not de-emphaticized if it is separated from the coronal trigger by an intervening vowel. Thanks to this, we get [r]~[rˤ] alternations between nominals and their morphologically related verb forms, as in (14b). Similar forms in Type-I dialects, such as ECA, display paradigm uniformity with a plain [r] throughout.

According to Herzallah (1990), the consonants that never cause /rˤ/ de-emphasis are the labials, pharyngeals, and laryngeals; that is, peripheral consonants whose articulation does not involve the tongue blade nor dorsum. The dorsal and coronal contexts above are not absolute either: if the triggers do not fulfil the adjacency and directionality requirements, they still co-occur with [rˤ]. Together these facts imply that the emphatic [rˤ] is the default allophone, a conclusion reinforced by the following, last environment that conditions [r].

In RPA, only a non-emphatic [r] appears “in the neighborhood of a non-inflectional, non-epenthetic, non-low front vowel” (Younes 1994: 220); that is to say, /rˤ/ is de-emphaticized in words containing a stem-internal /i, iː, eː/. R and the vowel may be immediately adjacent in either direction (15a), or they may be separated from each other by no more than one consonant (15b). The observation is that tautosyllabicity is not required here, in contrast to Type-I dialects.

Adjacent epenthetic [i] and inflectional (suffixal or prefixal) [i/iː], on the other hand, have no effect on /rˤ/. The examples in (16a–b) show an emphatic [rˤ] in these two respective contexts. Inflectional morpheme boundaries in (b) are shown with a dash.

An often-discussed detail in this frame is the discrepancy between two homophonous derivational and inflectional /i/ suffixes: the former gives rise to [r], while the latter retains the preceding [rˤ]. As (17) illustrates, this has produced apparent contrasts between words ending in the adjectival (relational) suffix –i and the identical first singular possessive suffix –i ‘my’.⁸ Notice that some of these pairs may or may not be semantically related, as the glosses in (17b) reveal, in which case they are derived from different roots.

This discrepancy can be easily understood in a rule ordering account. Drawing on the work of Younes (1982) and Watson (2002), I assume that /rˤ/ de-emphasis takes place prior to inflectional affixation and [i]-epenthesis; and once de-emphasis fails to materialize at this stage, emphatic [rˤ] survives on the surface. Consider, for example, the apparent minimal pair [ʃaʤari] ‘arboreal’– [ʃ̣ạʤ̣ạrˤ-i] ‘my trees’, which can be disambiguated via the derivation in Table 1. Rule ordering here is used for illustrative purposes, rather than for any theoretical preference.

As a restatement of the derivation in Table 1, Younes (1993) proposes that morpheme boundaries block de-emphasis, but inflectional affixes alone create no such boundaries. Younes thus treats the derivational suffix as part of the stem. A competing view, advanced by Herzallah (1990), is to propose two underlying /i/ phonemes: one specified (for roots and derivational suffixes) and one underspecified (for epenthesis and inflectional suffixes). The former de-emphaticizes /rˤ/ whereas the latter does not. These two proposals have their merits and drawbacks, but assessing them is outside the scope of the article.

The most noteworthy reference to R in other Type-II dialects is made in Syrian Arabic. Cowell (1964) states that in Damascus and some parts of Greater Syria there is no contrast between plain [r] and emphatic [rˤ] in the same contexts, such that words like ‘flowing’ and ‘my neighbor’ are pronounced identically as [ʒaːri]; compare with (17b). He adds that “in other regions the r/rˤ distinction – though not obliterated – is often subject to local and individual variations to such an extent that its importance is very slight” (Cowell 1964: 8). In Lebanese, Obegi (1971: 17–18) lists a single R phoneme with plain and emphatic allophones, and refers to “the severe distributional restrictions” on emphatic [rˤ]. Conversely, Obrecht (1968) and Al Mashaqba (2015) propose two phonemes, /r/ and /rˤ/, in Lebanese and Southern Bedouin Jordanian respectively, based on pairs like [ʒaːri] – [ʒ̣ạːrˤ-i], but we have already seen that these are not genuine minimal pairs.

So far, we have shown that R acts like the “primary” emphatics in three distributional patterns, and that it surfaces as non-emphatic in exactly three contexts. From these findings, we conclude that there is only one R phoneme in Type-II dialects, which is fundamentally emphatic, ergo /rˤ/, and which has a phonologically predictable plain allophone [r]. Emphatic [rˤ] as an allophone of /r/ is not predictable in the same way, and we infer that the inventory lacks a non-emphatic /r/ phoneme. Let us now examine evidence from phonological activity that supports these conclusions.

5.2 Phonological processes involving R

In most Levantine Arabic dialects, bidirectional emphasis spread is triggered by /rˤ/, among other emphatics (Grotzfeld 1980: 180). We have established in Section 4.2 that ES targets consonants and vowels within a certain domain, and is characterized articulatorily by pharyngealization and acoustically by F₂ lowering. Younes (1993) and Davis (1995) discussed a peculiar directionality effect in RPA, where rightward ES is more restricted than leftward ES. Right-to-left spreading has no blockers, and the minimum domain is the uninflected word (18a). Left-to-right spreading, on the other hand, continues up to the first low vowel, including any intervening consonants (18b), and is blocked by an immediately following /ʃ, j, w/ or a non-low vowel (18c).⁹ While the domain applies to all emphatics, note that some of the opaque contexts, namely /ʃ, j/ and the front vowels /i, iː, eː/, already cause /rˤ/ to lose its emphatic character (see Section 5.1).

Furthermore, ES from /rˤ/ in both directions is more limited than ES from the other emphatics. More specifically, morpheme boundaries block ES from a /rˤ/ source (to suffixes and prefixes), but not from /sˤ, ðˤ, tˤ/, unless /rˤ/ is contiguous to a low vowel or /h/ (Younes 1993: 127). By comparing the effect of the triggers in (19a) versus (19b), one can see that the uninflected word (minimum) domain mentioned above applies to /rˤ/.

Card (1983) examined emphasis in the Palestinian Arabic of Jerusalem, and argued that the R is not emphatic based on three facts: (1) the F₂-lowering influence of R on other segments is weaker than that of the so-called true emphatics; (2) the more restricted domain of ES from R just named; and (3) the distributional restrictions of R next to front vowels (Section 5.1). Without entirely ruling out the phonetic distinction in (1), Younes (1993) asserts that R must be treated phonologically as an emphatic in Palestinian even if it does not measure up to the more established emphatic phonemes. Other Type-II dialects where /rˤ/ is reported to cause bidirectional ES include Syrian (Cowell 1964), Southern Bedouin Jordanian (Al Mashaqba 2015), and Northern Rural Jordanian (Bani-Yasin & Owens 1987); with examples like [ḅạrˤrˤạ] ‘outside’, [f̣ạrˤẉạ] ‘hide’ (noun), and [x̣ạṇʤ̣ạrˤ] ‘dagger’.

Moving on to the featural composition of emphatic /rˤ/ in RPA, I will largely reconstruct Herzallah’s (1990) analysis in the PSM. As a member of the natural class of ES triggers, /rˤ/ will be assigned a “secondary” V-place [dorsal] feature. As a blocker to ES from /rˤ/, the /w/ is assigned V-place [labial], which is incompatible with a tier-adjacent V-place [dorsal]. Non-emphatic coronal consonants and front vowels are also opaque in that they cause an adjacent /rˤ/ to de-emphaticize. These can be split into a V-place [coronal] class containing /ʃ, ʧ, ʤ, j/ as well as /i, iː, eː/, and a C-place [coronal] class containing /θ, t, d, s, z, n/. Again, we can assume an incompatibility condition between [coronal] and /rˤ/’s V-place [dorsal]. Velar /k, x, ɣ/, too, are opaque in that they de-emphaticize a preceding /rˤ, sˤ, ðˤ/. If the former are assigned C-place [dorsal] and the latter V-place [dorsal], then de-emphasis can be explained as an OCP [dorsal] effect across nodes that will delink the emphasis feature.

Beyond emphasis spread, the /rˤ/ phoneme triggers complete assimilation of the definite article /l/, a process known to be initiated by coronals. Similar to (8) in Type-I dialects, both emphatic [rˤ] and non-emphatic [r] realizations are active in the process, suggesting that the rhotic belongs to the natural class of coronals, and hence is specified for C-place [coronal]. Herzallah (1990: 115–116) takes the fact that de-emphaticized [r] still triggers /l/-assimilation – just as any other coronal – as an indication that /rˤ/ does not lose its [coronal] feature upon de-emphasis, or else it would not trigger this phrase-level process.

Finally, /rˤ/ in RPA partakes in coronal sonorant assimilation, by which /l/ fully assimilates to /rˤ/ and /n/ to /rˤ, l/; see (9). The process is also reported in Syrian (Cowell 1964), Urban Jordanian (Zuraiq & Zhang 2006), Northern Rural Jordanian (Bani-Yasin & Owens 1987), and Southern Rural Jordanian (Al Huneety 2015), among other dialects in this group. A related observation by McCarthy (1986), cited in Herzallah (1990), is that the consonants /n, l, rˤ/ do not co-occur in the same root. What this all entails is that /n, l, rˤ/ form a class of coronal sonorants, and are consequently specified for C-manner [open] and V-manner [closed]. The existence of a V-manner feature is, then, what differentiates /rˤ/ from the emphatic obstruents, and is responsible for its discrete performance.

In summary, it has been shown that [r] and [rˤ] are realizations of one phoneme in this group, and that this phoneme is underlyingly emphatic, despite two types of patterning. On the one hand, /rˤ/ patterns with other emphatic coronals in its distribution and in inducing leftward and rightward ES. On the other, it partially differs from those emphatics due to its vulnerability to be de-emphaticized and to its more limited ES domain. I proposed that /rˤ/ has the representation in Figure 4, with C-place [coronal] and V-place [dorsal] plus C-manner [open] and V-manner [closed]. This is the same as emphatic /rˤ/ for Type-I dialects (Figure 3). When de-emphaticized under various conditions, the V-place node is delinked. After the emphatic-R dialects, we examine dialects with a single, plain /r/ phoneme.

Figure 4

Representation of /rˤ/ in Type-II dialects.

6.1 The distributional facts of R

The literature on MBA lists one plain /r/ phoneme (see e.g. Erwin 1963; Blanc 1964; Altoma 1969); and no minimal pairs that suggest a phonemic split of R are ever given. There are nonetheless two R allophones in complementary distribution: one emphatic and one plain (Shamdin Agha 1969: 54). According to Rahim (1980: 245–246), the non-emphatic [r] is found before a high front vowel [i/iː] or glide [j] in the same syllable (20).

Otherwise, the R phoneme is realized as an emphatic or pharyngealized [rˤ]. The emphatic gesture is most audible in the vicinity of obstruent emphatics /sˤ, ðˤ, tˤ/ (21a) and low back vowels [a/aː] (21b), but also occurs next to other vowels (21c). Notice that in (21b) only the adjoining low vowel is backed; in (21a) more vowels and consonants are backed due to the influence of ES from /sˤ, ðˤ, tˤ/; while in (21c) no other segment is audibly backed.

Based on these facts alone, it appears that R is an emphatic phoneme that undergoes de-emphasis to [r] in limited environments – seeing that [rˤ] is the elsewhere allophone. Youssef (2015), however, describes a distributional pattern in MBA that invalidates the underlying emphatic nature of R. In this pattern, [i] and [u] exist in complementary distribution in specific morphological classes, e.g. in epenthetic vowels of derived nouns and in initial stem vowels of perfect/imperfect Measure-I verbs, among others. Although /i/ and /u/ are otherwise contrastive, the appearance of [u] in those classes is determined by two neighboring labial and dorsal consonants, and R is a participant. Let us take the synchronic distribution of [i]~[u] in epenthetic vowels as an example.

Like many Arabic dialects, MBA has a tendency to break up a final consonant cluster with an epenthetic vowel (EV). As a result, derived nouns of the morphological pattern CaCC – wherein the stem vowel is always /a/ – will surface as CaCvC via epenthesis, and the EV may be [i] or [u] depending on the consonantal environment.¹⁰ If the cluster consists of a labial /b, p, f, m/ plus a velar /k, ɡ, x, ɣ/, uvular /q/, emphatic /tˤ, sˤ, ðˤ, lˤ/, or an R (collectively, back consonants) in either order, then the EV is always [u] (Blanc 1964: 55–56), as exemplified in (22a). The same is observed when the cluster consists of a labial and some other consonant provided the initial consonant is an emphatic (22b). Apart from these [u]-domains, the EV is [i]. And as a rule, [i] surfaces if only one of the cluster consonants (C₂/C₃) is a labial or a back consonant, or if the cluster consists of two back consonants (22c).

The distribution shows that the EV splitting final clusters of derived nouns is predictable in MBA, and that epenthetic [u] can be construed as a contextual variant of /i/ in those cases (described as labialization). Here, some comments are in order concerning the behavior of R. First of all, R functions as a trigger of labialization in (22a), in [ʔamurˤ] and [ħafurˤ], where [u] must be flanked between a labial and a back, emphatic or non-emphatic, consonant. R is not, however, found in (22b), where [u] must be adjacent to the labial but not to the back (emphatic) trigger, which can extend its backness from a distance via ES. Indeed, what we see in (22c) is that an initial R and a C₂/C₃ labial do not cause labialization, i.e. they produce epenthetic [i], as in [rˤạmiz] and [rˤạʔif]. This suggests the characterization of R as underlyingly non-emphatic in MBA, or else it would perform like the emphatics in (22b). We return to this issue in Section 6.2, where emphasis spread provides more definitive clues. Now we turn briefly to the treatment of R in other Type-III dialects.

According to Watson (2002: 16), R cannot be described as pharyngealized in several varieties of Arabic. Yemeni (Sanʿāni) /r/, for instance, has the allophone [rˤ] only next to, or no more than one vowel from an emphatic obstruent, as in [ṃạħ̣dˤạrˤ] ‘official report’ and [ɡ̣ụrˤtˤ] ‘earring’; elsewhere it is realized as a plain [r], even in words where other Peninsular dialects have an emphatic realization, such as [raːs] ‘head’, [ħarr] ‘hot’, and [saːra] ‘Sarah’ (Qafisheh 1990: 173). Another dialect where the distribution of emphatic [rˤ] is severely restricted is Bedouin Hijazi (Al-Mozainy 1981), in which /r/ appears as [rˤ] only when the adjacent vowel is [u/uː], and plain [r] when it is [i/iː] or [a/aː]. Lastly, in their investigation of a wide range of Peninsular dialects, Johnstone (1967), Prochazka (1988), and Holes (1990) list only non-emphatic /r/ as a distinctive phoneme.¹¹

Peripheral Arabic dialects that have lost the emphatic consonants obviously belong to the plain-R type. Consider the forms in (23a), where emphatic /sˤ, tˤ, dˤ, rˤ/ in ECA correspond to their non-emphatic cognates /s, t, d, r/ in Maltese. What is interesting in many such dialects is that the historical emphatics have left traces in vocalism. The distribution of the diachronic low vowels /a, aː/ in Maltese is a case in point (Cowan 1966; Mifsud 2008). Low vowels are retained when contiguous to pharyngeal consonants [ħ, ʕ] or an earlier emphatic, or else realized with raising as [e] or [ie] (23b). Cowan (1966) maintains that an originally allophonic variation of vowels based on emphatic-plain contexts was followed by phonemicization of these variants when the emphatics were no longer present as a conditioning factor. Or, put differently, the emphasis component has remained in the vowels after disappearing from the consonantal system; see also Walter (2002).

Overall, then, this section has given distributional evidence of a single R phoneme in Type-III dialects, mainly by showing that its [r] and [rˤ] realizations are phonologically predictable. The labialization pattern has additionally shown that R behaves more like velar-uvular than emphatic triggers. The next section explores a different type of evidence.

6.2 Phonological processes involving R

The triggers of emphasis spread in MBA are limited to /sˤ, ðˤ, tˤ, lˤ/, in which case we observe a bidirectional long-distance pharyngealization effect within the uninflected word domain (24a); see Youssef (2013: 126–139) for details. Where none of these emphatics exists, three other consonants, /r, q, ɣ/, seem to have a local backing influence on subsequent low vowels (24b); see also (21b) for more examples on R.

Central to the current discussion is why these segments cause backing, and whether this effect is ES or something else. On the one hand, Al-Ani (1970: 32–35) has pointed out an F₂ lowering effect of /q, ɣ/ on following [a/aː] and [i/iː] vowels, which does not extend farther. Influence on preceding vowels is insignificant. What we also notice is that these consonants have some posterior (or dorsal) articulation: the stop /q/ is uvular, while the fricative /ɣ/ is realized as uvular [ʁ] in most contexts (Rahim 1980: 242–243; Giannini & Pettorino 1982: 24). We infer that the backing is no more than low-level coarticulation, not ES, which is a long-distance effect by definition. Now the same may be said about R in MBA. The rhotic is described as a pharyngealized tap/trill – unless adjacent to [i/iː] – that results in lowering F₂ of a following [a/aː] (Al-Ani 1970: 33). Again, this impact is rarely attained in distant vowels, which points toward phonetic conditioning, rather than phonological ES. The conclusion is that /r, q, ɣ/ are not emphatics, even though they have a dorsal articulation.

Three important facts about R have so far been presented. First, there is one R phoneme; given that the distinction between [r] and [rˤ] is allophonic. Second, this phoneme is a non-emphatic /r/; it does not behave like the emphatic triggers of labialization and long-range ES. And third, it is dorsal; it does pattern (phonologically) with the velar-uvular /q, k, ɡ, x, ɣ/ in triggering labialization and (phonetically) with /q, ɣ/ in having a coarticulatory backing effect. Having established this, R must be specified for a “primary” C-place [dorsal] feature together with the velar-uvular consonants, but lacks V-place [dorsal] which is reserved for emphatics. So as long as a Type-III dialect exhibits the labialization pattern, its plain /r/ phoneme will be featurally different from the /r/’s in Type I (above) and Type IV (discussed below).

The specification of a C-place [dorsal] feature on the phonetically alveolar rhotic calls to mind the case of velar laterals in the Papuan languages Yagaria and Kanite, discussed in Blevins (1994). Blevins argues for a coronal (featural) component in these articulatorily dorsal laterals, citing only phonological evidence – from alternations, distribution, and loanword phonology. What this suggests is that the loose relationship between phonetics and phonology holds not only for rhotics, but potentially for the larger class of liquids.¹² That said, the /r/ in Type-III dialects is specified for other straightforward place and manner features, phonological support for which is by now familiar.

First, just as in the previous dialect types, /r/ is among the coronal consonants that trigger assimilation of the definite article /l/ in MBA, as in /l-rizma/ → [r-rizma] ‘the parcel’ (Erwin 1963); see also (8). The process is attested in all dialects of this group, counting peripheral ones like Maltese (Borg 1997). And it follows that /r/ is a member of the natural class of C-place [coronal] segments.

The next process is coronal sonorant assimilation, by which /n, l, lˤ/ regressively assimilate to /r/ across morpheme and word boundaries in MBA, and also in Southern Najdi (Al-Qahtani 2004) and Sanʿāni Arabic (Watson 2002), among others. See similar examples in (9) above. Akin to this process are the alternations between /r/ and /n, l/ in a number of Type-III dialects. In Najdi Arabic (Al-Qahtani 2004), for example, some loanwords exhibit metathesis between /r/ and /n, l/ (25a). And in Maltese, Borg (1997: 256–257) notes a “widespread synchronic trend […] towards reshuffling of liquids” in words of Italian origin, such that /l/ arises in syllable-initial and /r/ in syllable-final position, with some free variation at syllable codas (25b). On the cross-linguistic tendency of liquids to undergo metathesis, see Ultan (1978).

Together the last two facts point to a natural class of coronal sonorants comprising /r, n, l, lˤ/. This means that /r/ has additional C-manner [open] and V-manner [closed] to account for its sonorancy. See Youssef (2013: 77–78) for a detailed autosegmental analysis.

To sum up, there is one non-emphatic /r/ phoneme in this group, with two allophones. This phoneme patterns with both coronals and non-emphatic dorsals; hence it is characterized by a double place of articulation, C-place [coronal] and C-place [dorsal], but no secondary articulation. In addition, it patterns with the coronal sonorants; and so it also gets C-manner [open] and V-manner [closed] features. Compare the representation of Type-III /r/ in Figure 5 with that of Type-I /r/ in Figure 3, where there is only [coronal] under the C-place node.

Figure 5

Representation of /r/ in Type-III dialects.

7.1 The distributional facts of R

Both MLA and CBA belong to the Tigris branch of qeltu-dialects. Besides the characteristic “perseveration” of /q/ (where we find /ɡ/ in the gilit-dialects), uvular R is one of the most notable features of this branch which, according to Jastrow (1978), serves as a distinguishing criterion to separate it from the other two, Euphrates and Anatolian, qeltu-dialects. There is little doubt that uvular R originates from Old Arabic alveolar /r/, possibly first as a substrate influence from Aramaic. Jastrow (2007) states that it dates back to the medieval Abbasid period; ever since it has become a frequent sound as it also fell into harmony with inherited /ʁ/. The merged phoneme has been described as a voiced post-velar fricative (Watson 2002: 16); and although I prefer to use the uvular [ʁ] symbol, a velar [ɣ] is an equally plausible choice. (According to Paradis & LaCharité 2001: 278, uvular and velar fricatives do not contrast in any Arabic dialect).

A uvular [ʁ] occurs in the majority of MLA and CBA words with etymological R, though not necessarily in identical words in the two dialects. Example set (26) illustrates that these are well established, everyday words. But note that, contrary to Jastrow (1979), lexical forms with original geminate (trilled) [rr] retain the gemination in [ʁʁ], e.g. in [bạʁʁạ] ‘outside’ and [ħạʁʁ] ‘heat’.

On the other hand, several studies have remarked a tendency to re-establish alveolar [r] in the two dialects (e.g. Blanc 1964; Jastrow 1979; Abu-Haidar 1991). This is especially true for loanwords from foreign languages like Turkish, Persian, and English (27a), and from Literary Arabic or one of the Arabic dialects (27b). The Arabic words with [r] incorporate religious vocabulary, proper names, or terms for abstract ideas, although some are not even manifestly loanwords, such as the verbs in (27c). The disposition toward [r] in loanwords was also observed in Jewish Baghdadi (Mansour 1957), and the dialects of Tikrit (Johnstone 1975) and Djidjelli (city in northeastern Algeria) (Marçais 1956).

At the same time, there are a few minimal pairs contrasting /r/ with /ʁ/, both inherited and new (28a–b), which are sufficient to establish them as separate phonemes in Type-IV dialects; and “once a phoneme, always a phoneme”.¹⁴ Mansour (1957), by the same reasoning, advocates the two-phoneme hypothesis in Jewish Baghdadi. He lists more contrasts than those found in MLA and CBA (28c), and suggests that the forms with /r/ are recent Literary Arabic loans whereas those with /ʁ/ have already existed with a different, more basic meaning in the dialect. However, he carefully delineates this as a by-product of borrowing rather than a semantic restriction, since there is no independent support for the latter; see Note 14.

From the Maghrebi subgroup, the Arabic spoken in Djidjelli was reported to have a weakened version of R, which is non-trilled and articulated by the back of the tongue against the uvula, ergo [ʁ̝] (Marçais 1956: 17). This [ʁ̝] sound is described as phonetically similar but not identical to the uvular fricative [ʁ], i.e. with less friction noise and a further back point of articulation. And it is asserted that speakers do not confuse them. Hence in this subclass we have a /r/ phoneme that can potentially be realized as [ʁ̝], and a separate /ʁ/ phoneme.

One last relevant pattern is vowel raising (imāla) of the feminine suffix -a (cf. (12) above). The suffix surfaces as [i] after a final “front” consonant (29a) and as [a/ạ] after an emphatic or back consonant, [ʁ] and [r] included (29b). There are two exceptions to imāla, known in both MLA (Jastrow 1979: 40) and CBA (Abu-Haidar 1991: 30). First, a raised [i] appears after a [ʁ] reflex of old r where the preceding stressed vowel is front (29c). Also, no imāla is attested in loanwords ending in /a/ regardless of which consonant comes before, e.g. in [sˤọːḍạ] ‘soda’ and [doːndirma] ‘ice cream’. The implications of imāla will be dealt with in Section 7.2.

In this section we have examined distributional evidence, including from minimal pairs, for two R phonemes in these dialects: /ʁ/ which is a reflex of old r merged with inherited /ʁ/, and /r/ found principally in loans. While /r/ is exclusively rhotic, /ʁ/ cannot be claimed so, given that it stands equally for historical r and ʁ. So do we expect, along with various researchers, that the diachronic change will not lead to changing the phonological behavior of the rhotic? To answer this, we need to examine phonological processes that show whether /ʁ/ behaves as a coronal sonorant or as a dorsal fricative. This is the purpose of the next section.

7.2 Phonological processes involving R

Blanc (1964: 21) points out that the diachronic shift from [r] to [ʁ] has resulted in the latter sound appearing in “positions strange to it in Old Arabic […] notably in contact with /q, x/”. These abnormal contacts have led to further phonetic change in this group, for instance in the form of total assimilation. As noted by Abu-Haidar (1991: 10), expected sequences like /qʁ/ and /xʁ/, where /ʁ/ is a reflex of etymological r, will surface as [qq] and [xx] respectively (30). And although this is a clear pattern in the two dialects under scrutiny, Blanc (1964: 22) gives some unassimilated forms in CBA, such as [Ɂaqʁab] ‘nearer’ or the optional [ʕaqʁabi] ~ [ʕaqqabi] ‘scorpion’, adding that assimilated forms are generally more common in Jewish Baghdadi. In Djidjelli Arabic, [ʁ̝] and [ʁ] are “difficult to differentiate when they appear adjacent to each other” (Marçais 1956: 17), which can also be interpreted as assimilation.

In some contexts, again following Blanc, there is a tendency toward vocalization, whereby a [w]-like articulation of /ʁ/ is identified. For instance, the [ʁ] in [quʁuːn] ‘horns’, [fuʁħaːn] ‘happy’, and [Ɂakbaʁ kaððaːb] ‘biggest liar’ is perceived with little or no audible velar constriction due to the influence of the adjacent /uː/, /ħ/, and /k/ sounds. There are occasions on which full vocalization takes place, too. As demonstrated in (31a), a long back rounded vowel [oː] or [uː] appears in MLA instead of /u/ followed by a uvular /ʁ/, but only if the preceding consonant is a back /q, x, ʁ/ (Jastrow 1979; Tawfiq 2010). These changes seem to be connected to speech rate, and so in careful speech no assimilation occurs, hence the alternate forms in (31a). An outdated feature of MLA is the vocalization to [oː] in words that have the digit 4 built in (e.g. 14, 40, 400), where the modern pronunciations have re-established the [ar] sequence, as indicated in (31b). (Blanc remarked back in 1964 that one of his informants considers the [oː] forms old fashioned, which was echoed by my informants).

Marçais (1956) mentions a comparable tendency in Djidjelli Arabic, whereby a final uvular R is not pronounced, and the preceding consonant or vowel seems to be prolonged with an [a]-like element, e.g. [biː^a] ‘a well’, [tuː^a] ‘bull’, [sˤạḅ^a] ‘patience’.

In contrast with assimilation, there is also dissimilation: a tendency to realize the old r as [r], rather than [ʁ], in close proximity to an inherited /ʁ/ and a back or pharyngealized vowel (32a). But where the two root consonants are separated by a front vowel or palatal /j/, the co-occurrence restriction does not apply, and so two instances of [ʁ] occur (32b). This curious arrangement is found in MLA (Tawfiq 2010) and CBA (Abu-Haidar 1991), as well as in Jewish Baghdadi Arabic (Blanc 1964).

Assimilation, vocalization, and dissimilation can thus be viewed as ways to resolve the newly created, unusual contacts between the uvular R and the back consonants /q, x, ʁ/. Assuming these are active phonological processes, they are apparently motivated by an OCP violation, and we can propose that /q, x, ʁ/ are all specified for C-place [dorsal]. In the case of total assimilation, the OCP violation on this tier will lead to merger of other consonantal features. In vocalization, the [dorsal] feature on /ʁ/ will merge with the vocalic [dorsal] on an adjacent /u/. And in dissimilation, where the only trigger is the inherited fricative /ʁ/, we need to propose two additional features that exclude other dorsal triggers. These are C-manner [open], designating friction, and C-laryngeal [lax], designating voice (see Morén 2003: 222–233). The feature [lax], in the sense of “lenis”, is distinctive as it differentiates /ʁ/ from /x/.

To work out the feature composition of /r/, let us examine assimilation of the definite /l/. It comes as no surprise that triggers are the coronal consonants (see (8) above); hence /r/ is included but not /ʁ/ (irrespective of its origin), as exemplified in (33). My data exhibit no discrepancy between the two dialects in this respect: in accordance with Jastrow (1979) for MLA, but in conflict with Abu-Haidar (1991) who states that the triggers in CBA are not limited to coronals, and so include both /r/ and /ʁ/. It is possible that this is a fast-speech-only phenomenon which I was not able to elicit in my pilot study, so I will not rule it out.

What /l/-assimilation tells us about /ʁ/ is that it is a non-coronal. The /r/ phoneme is one of the coronal triggers and so must be specified for C-place [coronal]. But should it be specified for [dorsal] too? For one thing, the imāla data in (29) suggest that /ʁ/ and /r/ behave like emphatic or back consonants. This matches the behavior of /ʁ/ in other patterns, and thus suits its characterization as C-place [dorsal]. /r/ does not display similar signs of dorsality. It does, however, cause backing of a following low vowel along with the non-emphatic consonants /ʁ, x, q, ɡ, h, ħ, ʕ/, as many of the transcriptions indicate. Because the effect is apparently gradient and very local, in contrast with the categorical long-range emphasis spread from /sˤ, ðˤ, tˤ/, I argue that backing from /r/ and those consonants is phonetic coarticulation rather than ES (see Section 6 for a full discussion on MBA /r/). The same goes for imāla. And on that account /r/ is a non-emphatic segment, and has no [dorsal] feature.

Lastly, we should substantiate sonorancy. For /ʁ/ there is none. /r/, however, triggers coronal sonorant assimilation at normal speech rate – see (9) – although it was limited to /l/ targets in my data, as in /ji-ʃtiɣal rassaːm/ → [ji-ʃtiɣar rạssaːm] ‘he works as a painter’. Further, Jastrow (1979: 41) refers to an interesting pattern in the dialect of Baḥzāni (a town near Mosul) by which nominal forms of the C₁VC₂C₃ shape break up the final coda cluster only if C₃ is a sonorant /m, n, r, l/. I will take these as indications that /r/ is generally sonorant in this group, and assign it the usual C-manner [open] and V-manner [closed] features.

In short, based on the evidence presented here, there is a coronal sonorant /r/ and a dorsal fricative /ʁ/ in Type-IV dialects, and they are distinct phonemes. It is both the rarity of languages with more than one R-phoneme and those with uvular R (Maddieson 1984) that make the dialects in this group particularly remarkable. The suggested representation of a non-emphatic /r/ in Figure 6 (a) is by now standard: C-place [coronal], C-manner [open], and V-manner [closed]. The uvular /ʁ/, on the other hand, was assigned C-place [dorsal], C-manner [open], and C-laryngeal [lax] (b). This distinction is in line with Maddieson’s (1984: 88) generalization that a language with two R phonemes is “unlikely to restrict their contrast to place of articulation”.

Figure 6

Representation of /ʁ/ and /r/ in Type-IV dialects.

Turning back to the question at the end of Section 7.1: Can we contend that the uvular R remained a rhotic in this group? Our phonological evidence implies otherwise. The new /ʁ/ sound has wholeheartedly melted with the fricative /ʁ/ phoneme, and no longer behaves as a sonorant. Wiese (2001), Scobbie (2006), and Chabot (2019) may be correct about the diachronic stability of R across many languages, but what we saw here poses a challenge to their hypothesis.