In D. Recasens, F. Sanchez Miret & K. Wireback (eds.) (2010, to appear), Experimental phonetics and sound change. München, Lincom Europa.

RHOTIC RETROFLEXION IN ROMANCE. ACOUSTIC DATA FOR AN ARTICULATION-DRIVEN SOUND CHANGE. Chiara Celata, Scuola Normale Superiore, Pisa,

Abstract The retroflexion of /t(˘)r/ clusters in the Sicilian dialect is examined from an acoustic and historical point of view. The topic of consonant retroflexion has been widely investigated by Romance dialectologists and historical linguists, as well as by experimental phoneticians, but the convergence of the two sub-disciplines has only been episodic so far. This research aims at filling this gap by exploiting the resources of experimental phonetics for the purpose of diachronic reconstruction. Most studies on consonant retroflexion in the world’s languages have the explicit purpose of drawing a link between the phonological status of this consonant class and the phonetic invariance that should unambiguously define it. In these approaches, however, the large amount of variation that actually shapes the reality of phonetic events is reduced to bare labels such as ‘apicality’ or ‘falling F3’, that are fairly useless for diachronic purposes. On the other hand, it has been widely demonstrated that the only way to convincingly describe retroflex articulation and capture the multiplicity of phenomena hidden behind this label is to posit a continuum of places of articulation (Ladefoged & Bhaskararao 1983, and subsequent works). Following this latter approach, this research is not aimed at providing a phonetic and phonological description of a sound class, but rather at analyzing processes of sound change that affect retroflex and other close places of articulation.

1. Introduction This paper investigates the process of consonant retroflexion in rhotic context in some Italo-Romance dialects. Acoustic data on the Sicilian cluster /t(˘)r/ are provided which may shed light on the sound change mechanisms giving rise to the retroflex outcome in several areas of the (Italo)-Romance domain.

1

The topic of consonant retroflexion has been widely investigated by Romance dialectologists and historical linguists as well as by experimental phoneticians, but the convergence of the two sub-disciplines (i.e., historical dialectology and experimental phonetics) has been episodic so far. This research aims to fill this gap by exploiting the resources of experimental phonetics for the purpose of diachronic reconstruction. Empirical evidence is adduced for a crucial and much disputed, though overlooked, class of sounds using spontaneous speech produced by Sicilian speakers. In particular, acoustic data for the etymological cluster /t(˘)r/, mostly realized as [ˇ(˘)ß], are presented in comparison with those for /p(˘)r/ and /t(˘)s t(˘)S/. The experimental findings are discussed with reference to the acoustic and articulatory features of retroflex consonants for several non-Romance languages documented in the literature. A diachronic perspective is adopted with respect to the phonetic triggers of retroflexion in a rhotic context within the Romance domain. In contrast to some recent perceptually oriented accounts, articulatory reduction will emerge as the fundamental conditioning factor. One of the most remarkable achievements of modern instrumental phonetics is the observation of the great amount of variation characterizing linguistic production. Frequent sources of variability are found not only among different speakers of the same language, but also in the speech of a single speaker. Based on this simple but essential phenomenon, the Ohalian theory of language change (since Ohala 1981) predicts that a careful investigation of the acoustic, articulatory and aerodynamic characteristics of speech sounds should raise fundamental questions about the nature of sound change. In other words, sound change is rooted in sound variation, insofar as the latter is embodied in sociolinguistic and dialectal variation, morphophonemic alternations and typological preferences. On different grounds, most studies on consonant retroflexion in the world’s languages have the explicit goal of identifying a link between the phonological status of this consonant ‘class’ and its invariant phonetic properties. According to this approach, however, the large amount of the variation which shapes the reality of phonetic events is reduced to bare labels such as ‘apicality’ or ‘falling F3’, which are fairly useless for diachronic purposes. Moreover, it has been widely demonstrated that, especially from an articulatory point of view, the only way to convincingly describe retroflex articulations while capturing the multiplicity of phenomena hidden behind this label is to posit a continuum of places of articulation (Ladefoged & Bhaskararao 1983, Ladefoged & Maddieson 1996: 2130; see below, §.3).

2

Following this latter approach, this research focuses on the fine-grained acoustic description of the retroflex realization of the etymological cluster /tr/ as produced by several Sicilian speakers. The main goal, though, is not a phonetic and phonological description of retroflexes as a sound class, but rather an analysis of the sound change processes that affect retroflex and other close places of articulation.1

2. Retroflex consonants in Romance Retroflex articulation is generally said to be highly marked in terms of production mechanisms and distributional properties across languages. This is one reason why consonant retroflexion is usually analyzed at the typological level and studied for the purpose of taxonomic representation. The groundbreaking study by D.N.S. Bhat, “Retroflexion: an areal feature”, published within the Working Papers on Language Universals series in 1973, was the first and most important in a fruitful tradition of studies that were all more or less influenced by the classical typological framework developed in the course of the 1960s and 1970s. In this work, more than 150 languages are mentioned which have different retroflex segments in their sound inventory. These languages are said to belong to four major geographic areas (the Indian peninsula, the Australian area, the Pacific coast of America and Central Africa), as well as to three minor groups (some languages in Southern Africa, some Northern Germanic languages such as Swedish, Norwegian and English, and some Caucasian languages). This areal typological perspective was particularly successful and directly or indirectly adopted by all scholars working on consonant retroflexion. For example, the most detailed and exhaustive analysis of the phonetic and phonological characteristics of retroflexes in the world’s languages recently carried out by Hamann has been explicitly inspired by Bhat’s taxonomy (Hamann 2003: 2). Although the author states that her study is based on a detailed scrutiny of grammars from linguistic families exhibiting consonant retroflexion processes, no reference to Romance languages is ever made in spite of the fact that Romance dialectologists and philologists have been well aware of the existence of some retroflex pronunciations in Romance – at least since Schneegans (1888) or Merlo (1925) just to mention the most renowned scholars of the past centuries. Romance 1

This work is part of my PhD Dissertation (Celata 2006, available online at http://alphalinguistica.sns.it /tesi/celata/tesi_Celata.htm). The project is an investigation of several retroflex outcomes in different Italo-Romance languages. See also Celata (2002-2003) for related issues on Corsican liquids in consonantal clusters. The retroflexion of geminate laterals in Sicilian is analyzed in Celata (forthcoming). I would like to thank P.M. Bertinetto, M. Loporcaro, M. Żygis and D. Recasens for their comments and support.

3

retroflexion has been dealt with recently in essays going beyond the limits of local dialectology such as Dalbera-Stefanaggi (1991) and Maiden & Parry (1997). These circumstances, though surprising, are quite understandable if one considers that an interest in careful phonetic investigation, as well as in the phonological status, of the Italo-Romance retroflex sounds has belatedly arisen – even among scholars especially concerned with Romance languages in general and with Southern Italian varieties in particular. On the other hand, these are prolific research areas for historical linguists, philologists, phoneticians and dialectologists (see Celata 2006: 42-44 for details). Retroflex consonants are observed in a relatively wide area of the Romance domain. As far as Italo-Romance is concerned, they occur in many Southern dialects spoken in Calabria, Puglia, Abruzzo and Campania, as well as in Sicilian, Sardinian, Corsican and some varieties of Northern Tuscan. Retroflex realizations are also found in Western Asturian and were probably present in Old Gascon. Voiced retroflex stops and affricates developed mostly from the geminate lateral in word medial position (e.g., Ragusa Sicilian [ka»Va͢o] standing for It. cavallo ‘horse’) and sporadically at word boundaries (e.g., Minucciano Tuscan [kweÍ »omo] for It. quell’uomo ‘that man’; [vak a »Íet˘o] for It. vado a letto ‘I go to bed’, where the geminate lateral is triggered by Raddoppiamento Fonosintattico). The output of this process can be a stop [͢] or an affricate [͢Ω]. Gemination is generally preserved, but in some dialects a degemination process has occurred. There are also varieties in which the proto-Romance cluster /lj/, after having developed into a palatal lateral ([¥˘]), underwent a retroflexion process (e.g., Southern Corsican [»a͢a] for It. aglio ‘garlic’). In some Sardinian, Sicilian and Calabrian dialects, the retroflex stop may also extend to etymological /d˘/ (e.g., Sardinian [»tu˜Íu] for It. tondo ‘round’), though this process is said to be lexically determined. Aside from /l˘/, clusters composed of an alveodental stop followed by a rhotic, i.e., /t(˘)r/ but also /str/ and /ntr/, is the other major target of retroflexion. This target output is normally a retroflex voiceless affricate that can be simple or geminate, e.g., Sicilian [»ˇßEni] for It. treni ‘trains’ vs. [aˇ˘ßo»vaRe] for attrovare, It. trovare ‘to find’. 2 As far as the Sicilian dialect is concerned, /l˘/ retroflexion is attested all over Sicily, with the exception of some villages in the East (e.g., Bronte, Francavilla) and some Gallo-Italian colonies (such as Randazzo and Novara). Other GalloItalian colonies, on the contrary, show an hyper-extension of [͢] even to contexts 2

For a general survey of the developments of /l˘/ and /tr/ in Italo-Romance, see Rohlfs (1966: 257259, 263-264, 325-333, 368-369, 379-380). For more specific bibliographical references, see Celata (2006: 25-35).

4

not undergoing retroflexion in the other areas of the island (e.g., [͢agrima] for lacrima ‘tear’; Rohlfs 1966: 330). In some areas, a degemination process has lead to the phonetic realization [Í]. Retroflexion of /tr/ clusters is even more widespread across the island of Sicily; for example, it is documented even in Randazzo and Novara (Millardet 1933: 720). Also in Sicilian, as well as in Cosentian Calabrese and possibly other dialects, intervocalic or word-initial /r˘/ is said to be realized as a fricative postalveolar (Ruffino 1991). Retroflex pronunciation is a typical phonological feature of the local Sicilian dialect, but it is also found in the regional variety of Italian spoken in the island (Ruffino 2001). Sicilian speakers are aware of this phonetic peculiarity which is perceived as a marker of ethno-linguistic identity, and are able to avoid it when speaking regional Italian at least among the younger generations and educated people (Clemente et al. 2003: 307-314). Though broadly substantiated, retroflex outcomes are homogeneous across the island both regarding context of application and phonetic output. Variation may obviously affect both fine phonetic detail and socio-linguistic distribution (see for example Tropea, 1963, for a detailed analysis of the phonetic fluctuation of retroflexion among different social classes and different towns of Western Messinese). Since the beginning of the twentieth century many Romance philologists have attributed the retroflex consonants to a substratum origin (Guarnerio, 1902, Merlo 1925, Bottiglioni 1927, Millardet 1933, Schmeck 1952, Menéndez Pidal 1954, Rohlfs 1955). At the end of the 1960s, the substratum hypothesis was still most accredited (Blaylock 1968). Recently, however, three scholars have shown independently from one another that retroflexion in Romance must be conceived of as a modern development taking place later than the eighteenth century. These studies are based on philological evidence and deal with Northern Tuscany dialects (Savoia 1980), Sicilian (Caracausi 1986) and Sardinian (Contini 1987).

3. Retroflexion in rhotic context. Acoustic and articulatory features of rhotics and retroflexes The class of rhotic consonants subsumes a large variety of possible realizations differing greatly from one another at the articulatory, acoustic and aerodynamic levels. This group of sounds cannot be unambiguously defined along the manner or place of articulation dimensions. The great deal of variation characterizing rhotic consonants across different languages (as well as across speakers and prosodic conditions) is so striking that it has led some scholars to the conviction that rhotics tend to behave in similar ways phonologically, but do not form a 5

sound class on phonetic grounds (see, for example, Ladefoged & Maddieson 1996: 215-217, and several papers in Van De Velde & Van Hout 2001). Variation among rhotics is large even within one and the same linguistic community, as demonstrated by numerous studies on American English /®/ (see Labov et al. 2006 and Hashi et al. 2003). Rhotics also display a considerable number of contextdependent allophones stemming mostly from stylistic and/or phonotactic variation. In spite of this great amount of articulatory variability, only one acoustic correlate has been clearly identified as a unifying property of the class of rhotics, i.e., the lowering of the third formant (F3) in the transition from and to the adjacent sound. F3 lowering is mainly related to tongue dorsum retraction being triggered not only from place of articulation backing (as in the case of uvular or velar sounds) but also from apicality. Apicality and tongue dorsum retraction are articulatory features that define both rhotics (with the only exception of uvulars) and retroflexes. In order for the apex to rise and vibrate, the tongue body and most especially the tongue dorsum has to be retracted and raised against the pharynx. Similarly to rhotics, retroflex consonants are also a highly variable sound class. Although the IPA conventional classification revised to 2005 still considers ‘retroflex’ as one of the possible places of articulation occurring between the postalveolar and the palatal places, many scholars have suggested that the term in question corresponds rather to a complex gesture or, more generally, to an articulatory shape (see Ohala 1983 and Dixit 1990). Evidence for articulatory complexity in retroflex consonants is provided by the following comparison between retroflex and palatoalveolar non-retroflex sibilants taken from Ladefoged (1975): “Retroflex sounds are made by curling the tip of the tongue up and back so that the underside touches or approaches the back part of the alveolar ridge. […] Because the undersurface of the tip of the tongue is touching the back of the alveolar ridge, the blade (the upper surface of the tip) of the tongue is usually a considerable distance from the roof of the mouth. As a result the tongue is somewhat hollowed. […] The palato-alveolar sounds [š, ž] differ from retroflex sounds in the part of the tongue involved. In palato-alveolar sounds the upper surface of the tip of the tongue is near the roof of the mouth. In addition, the front of the tongue is slightly domed, as opposed to being hollowed. In both [ß] and [š] the maximum constriction of the vocal tract occurs near the back of the alveolar ridge. But these two sounds have different places of articulation, because the term specifying the place of articulation designates both what part of the roof of the mouth is involved 6

and what part of the tongue is involved. In retroflex sounds the underside of the tip of the tongue forms the articulation, but in palato-alveolar sounds the articulation is made by the upper surface of the tip of the tongue” (Ladefoged 1975: 145-147). The level of articulatory variability documented for retroflex consonants across the world’s languages is very high, particularly with respect to the typical curling back of the tongue tip against the roof of the mouth, but also to other articulatory parameters. For example, based on a comparative study of Hindi and Tamil retroflex stops, Ladefoged & Maddieson (1996: 25-27) established a distinction between two major subclasses of retroflex consonants, i.e., apical and sub-apical. As far as the constriction location is concerned, retroflex sounds can be alveolar, postalveolar, prepalatal and palatal across languages and speakers. Tongue tip configuration and place of articulation are often correlated; thus, Ladefoged & Bhaskararao (1983) show that Telugu retroflexes are subapical and prepalatal, while Hindi retroflexes are apical and postalveolar. Moreover, the authors suggest that these two articulations are specific examples of what they call a continuum of possible retroflex realizations. The notion of an articulatory continuum is particularly suitable for retroflex sounds given the multiplicity of factors (language, speaker, style, speech rate, vowel context, etc.) along which variation can apply. In a recent survey by Hamann (2003), retroflex consonants are said to be characterized by four principal articulatory properties: apicality, posteriority, sublingual cavity and retraction. This characterization is intended to admit feature optionality since it provides for the possibility of including elements that may or may not fulfil the above mentioned articulatory criteria. On the other hand, even though none of these four articulatory features is specific to retroflex sounds, their combination represents the set of those fundamental properties that define most realizations within the class of retroflexes. In particular, apicality (which includes subapicality as well) refers to the tongue tip involvement in linguopalatal contact; retroflexes are never laminal sounds. Posteriority refers to the place of articulation which, even if variable, is always comprised between the postalveolar and the palatal region. The sublingual cavity is in principle inherent to an articulation that combines apicality and posteriority. Finally, retraction refers to the typical pharyngealization or velarization effect resulting from tongue dorsum displacement, which, in its turn, arises from a requirement for retroflexes to be apical and involving a posterior place of articulation. The acoustic properties of this articulatory configuration have been carefully investigated for many languages of the world. There is general consensus about 7

the fact that the most relevant acoustic feature of retroflexion is the lowering of the third formant in VC and CV vowel transitions (e.g., see Stevens & Blumstein 1975, Dart & Nihalani 1999, Hamann 2003). F3 lowering has been attributed to posteriority and, to a minor extent, to velarization and rounding. Many scholars have found that even F4 is characteristically lower in the production of a retroflex consonant (e.g., Fant 1974, Stevens & Blumstein 1975, Spajić et al. 1996). Other studies have highlighted that articulatory retraction affects the F2 trajectory, which tends to raise toward F3 values (Ohala & Ohala 2001), while others have found that F2 does not provide any information about retroflexion compared to other coronal articulations possibly because it is strongly influenced by the quality of the surrounding vowel (see Krull et al. 1995, Żygis 2005). Various scholars have emphasized that, from a perceptual point of view, the acoustic cues of retroflexion are asymmetrically distributed. In particular, VC transitions are said to be more informative than CV transitions; consequently, the phonological contrast between retroflex and non-retroflex coronals is most likely to be preserved in intervocalic position and neutralized word-initially and postconsonantally (Steriade 1995, Blevins 2004: 119-120). The argument on the asymmetric distribution of retroflexion cues finds support in studies such as those by Dave (1977) on Gujarati and Anderson (1997) on Arrernte. According to other studies, however, the hypothesis of interest cannot predict entirely the speakers’ perceptual behaviour since the perceptual advantage of the VC transitions appears to operate in specific vocalic contexts only (Ohala & Ohala 2001, Hamann 2003b) and evidence for the asymmetric role of perceptual cues can be found for dental consonants as well (Hamann 2003b).

4. An acoustic investigation of Sicilian /tr/ 4.1 Methodology The present investigation is based on acoustic recordings of dialectal speech by five Sicilian speakers from the provinces of Trapani (located in the northwest area of the island), Ragusa (in the south) and Enna (in the centre). Since the retroflex pronunciations of both /tr/ and /l˘/ are known to be evenly distributed across the island (see above), there were no particular expectations based on the different geographical origin of the speakers. On the other hand, since sociolinguistic factors such as age and sociocultural background are said to exert a possible influence on the speakers’ control over the retroflex pronunciation in Sicily, we tried to control for these factors by recruiting 20 to 30 year old undergraduate students of Pisa University. In order to elicit natural dialectal speech, pairs of 8

subjects coming from the same dialectal area whenever possible were asked to talk to each other about topics suggested by the experimenter. Sporadic shifts to the local variety of Italian (“italiano regionale”) were observed. These shifts, however, did not have any effect on the elicitation of retroflex productions since retroflex consonant realizations occur in the dialectal and regional Italian speech of Sicilian speakers provided that they do not try to avoid them purposely. Data were recorded in the anechoic chamber of the Scuola Normale Superiore of Pisa. The speech corpus included 650 relevant lexical forms, containing either [ˇ(˘)ß], or [t(˘)s], [t(˘)S], [tr] and [pr]. The focus of the analysis was on the voiceless retroflex affricate [ˇ(˘)ß]. Acoustic analysis was performed on the affricate frication phase whose quality and tonality is said to contain much information about the presence or absence of the retroflex character of the consonant. The formant values of the ingoing and outgoing vowel transitions were also analyzed in order to determine the degree of velarization of retroflex consonants. The four parameters under investigation are therefore: i. The duration of the affricate, and of its closure and frication phases. ii. The intensity level of the frication phase. iii. The spectral characteristics of the frication phase through analysis of COG and spectral peak frequencies. iv. The F3 and F4 from and to preceding and following vowels, respectively. The durational values were normalized with respect to the duration of a long sequence composed of the five syllables preceding and following the target consonant. The same parameters were also used by Żygis (2005) in her analysis of Polish and Czech alveolar, palato-alveolar, retroflex post-alveolar, alveolopalatal and palatal affricates. The software Praat 5.0.23 and SPSS for Windows 11.5 were used for acoustic and statistical analysis respectively. The analysis of the relevant acoustic parameters was carried out using the same procedure as Żygis (2005) in order to facilitate a comparison between our results and hers. 4.2. Results 4.2.1. General remarks

9

As other authors have already pointed out, the retroflex realization corresponding to the etymological cluster /t(˘)r/ is in most cases a voiceless affricate, which will be represented as [ˇ(˘)ß] in accordance with suggestions made, among others, by Loporcaro (2000) and Sorianello & Mancuso (1998) about the transcription of the retroflex output from several Southern peninsular dialects. The waveform and spectrographic displays in Figure 1 show a representative example of this complex articulation composed of a closing phase followed by a frication phase. [Insert Fig_1 here] Fig. 1 Waveform and spectrogram of [»paˇße] < /patre/ for St. Italian padre, ‘father’, as produced by a male speaker from Trapani.

The percentage of retroflex affricate realizations of the etymological cluster /t(˘)r/ by our speakers has been assessed vis-à-vis that of the bi-segmental realization [t(˘)r]. Overall, [ˇ(˘)ß] represents about 69,2% of the total available /t(˘)r/ contexts, while the option [t(˘)r] occurs in the remaining 30,8%. Retroflex pronunciation is therefore the major variant, though speakers tend to avoid it in a remarkable number of cases, at least in the speech elicited for analysis in the present study which cannot be characterized as entirely natural. As shown in Table 1, large differences in occurrence of the retroflex realization are found across subjects which is consistent with the idiosyncratic and variable character of the /tr/ retroflexion rule in Sicilian.

Subject

Gender

% [ˇ(˘)ß]

% [t(˘)r]

S1

male

100

0

S2

female

33,4

66, 6

S3

female

85,2

14,8

S4

male

13,1

86,9

S5

female

90,2

9,8

Table 1. Percentages of occurrence of the retroflex realization for the cluster /tr/ as a function of speaker

10

4.2.2 Duration Before conducting an analysis of retroflex productions, the duration of the realization [tr(:)] was compared to that of the non-homorganic cluster /pr/. As far as cluster duration is concerned, the geminate sequence [t˘r] turns out to be significantly longer than the geminate cluster [p˘r] (F (1, 30) = 2.795, p < .05), while their singleton counterparts display an almost equivalent duration. It is noteworthy, however, that closure duration shows the opposite pattern with respect to total cluster duration: as for [p˘r], the labial stop takes up a significantly greater temporal slot than [t˘] in the [t˘r] cluster (F (1,30) = 3.234, p < .05), thus meaning that the rhotic tends to occupy a comparatively larger portion of the overall consonant sequence in homorganic vs. heterorganic segmental combinations. This conclusion is supported by duration data for singletons mirroring those for their geminates correlates: [t] in [tr] sequence is significantly shorter than [p] in [pr] (F (1, 118) = 5.067, p < .05). Therefore, we can provisionally conclude that the relatively longer duration of the cluster [t(˘)r] is not due to intrinsic durational differences between the articulatory gestures for the alveolar stop compared to the labial, but rather to the fact that the rhotic portion of the cluster is relatively longer in the alveolar cluster than in the labial cluster (especially in the case of the geminate clusters). This finding will be related to relevant acoustic properties later on. The absolute, non-normalized durational values for the clusters of interest are shown in Table 2. The rhotic/stop ratio confirms that, in comparison to the non-alveolar cluster, the alveolar cluster is made up of a stop followed by a noticeably long rhotic.

Cluster

Stop duration

Rhotic duration

Total cluster duration

Rhotic/stop ratio

[p˘r]

109

33

142

0.303

[t˘r]

90

70

160

0.778

[pr]

59

44

103

0.746

[tr]

50

52

102

1.06

Table 2. Duration values of the clusters /t(˘)r/ and /p(˘)r/ and of their components (in ms).

11

A comparison between the two allophones of the cluster /tr/ shows that the retroflex allophone [ˇ(:)ß] is significantly shorter than the non-retroflex allophone [t(:)r], both for the geminate clusters (F (1,38) = 4.190, p < .05) and the singleton ones (F (1, 164) = 7.890, p < .000). A similar trend is at work for the closure period alone at least in the case of the geminate clusters, even though this duration difference does not reach statistical significance. As for the singleton clusters, on the other hand, the alveolar and retroflex stops are approximately equally long. In summary, some shortening appears to occur as we proceed from the bi-segmental cluster to the retroflex affricate. The retroflex affricate [t(˘)ß] tends to be shorter than the dentoalveolar affricate [t(˘)s] and the palatoalveolar affricate [t(˘)S]. This difference is statistically significant when the consonants under comparison are geminate (F (2,139) = 8.123, p < .001) and approaches significance in the case of singletons. The dentoalveolar [t(˘)s] is the longest of all affricates, while the palatoalveolar [t(˘)S] shows an intermediate duration between [ˇ(:)ß] and [t(:)s]. Moreover, the relatively short duration of the retroflex affricate appears to be entirely due to the frication phase which is shorter than that of other affricates, and not to the closure phase which tends to be longer. A comparison among affricate closures achieves significance for the singleton consonants only (F (2, 168) = 14.765, p < .001); as far as the frication phase is concerned, both singleton and geminate retroflex affricates are significantly shorter than the dentoalveolar and the alveolopalatal ones (F (2, 278) = 13.713, p <.001 for the geminate; F (2, 186) = p < .001 for the singleton). As shown in Table 3, the frication/closure ratio is much lower for the retroflex than for the other affricates. We may therefore provisionally conclude that, when compared to the other affricates under analysis, the retroflex affricate is characterized by a comparatively shorter frication period.

Affricate

Stop duration

Frication duration

Total duration

Frication/stop ratio

[ˇ˘ß]

78

59

137

0.76

[t˘S]

68

83

151

1.22

[t˘s]

73

96

169

0.76

[ˇß]

51

37

88

0.72

[tS]

27

65

92

2.41

[ts]

40

65

105

1.62

Table 3.

12

Duration values for the clusters /ˇ(˘)ß/, /t(˘)S/ and /t(˘)s/ and their components (in ms).

Similar results were also reported by Żygis (2005) for Polish, i.e., a shorter frication phase for the retroflex affricate than for the dentoalveolar, alveolopalatal and palatalized affricates. The short acoustic duration of the obstruent release can be attributed to the specific production mechanisms for retroflex postalveolar affricates vis-à-vis those for non retroflex affricates, and in particular to the fact that tongue tip movements generally involve shorter temporal intervals than tongue blade movements. The Sicilian and Corsican voiced retroflex stop (usually transcribed as [Í(˘)]) appears to be also characterized by a salient release that tends to be comparatively longer than that of the dentoalveolar voiced stop [d˘], while being shorter than that of the affricates [d˘Z] or [d˘z] (Celata 2006 and in press). In many instances of [Í(˘)], the release was clearly fricative-like with a frequency noise about 1600 Hz for male speakers and 2000 Hz for female speakers. These values indicate a rather retracted point of articulation, probably postalveolar. Moreover, such a long and noisy release suggests that voiced retroflex consonants are apical articulations, i.e., apicoalveolar or apico-postalveolar.

4.3. Intensity of the frication phase The average intensity of the frication phase has been calculated from the end of the burst until the first vowel pitch pulse. Data were split by the quality of the following vowel. Frication intensity relative to the adjacent vowel intensity has been shown to be relevant for place identification for fricatives: in addition to the primary frication spectral peak, a secondary frication-by-vowel intensity peak is indeed relevant for fricative perception (and production) (see for example Hedrick & Ohde 1993). Data were also split by subjects in order to account for the small number of male and female speakers subject to analysis in the study. Table 4 displays the cross-speaker average intensity values of all consonants as a function of following low, mid front, high front, mid back and high back vowels. Some vocalic contexts were not included in the statistical analysis because they were not consistently represented in the corpus. Statistical tests were run on geminate affricates followed by /a/, /i/ and /u/ and on singleton affricates followed by /a/ and /i/.

13

Geminate

Singleton

V2

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

/a/

62 (10)

58 (11)

55 (9)

57 (9)

49 (6)

42 (5)

/e/, /E/

68 (2)

58 (12)

51 (4)

53 (6)

56 (11)

48 (8)

/i/

67 (6)

60 (11)

51 (9)

53 (12)

56 (11)

55 (11)

/o/, /ç/

56 (9)

56 (10)

53 (10)

54 (5)

54 (4)

52 (2)

/u/

70 (6)

51 (7)

42 (3)

54 (7)

48 (3)

50 (8)

Table 4. Frication intensity values for singleton and geminate affricates split by V2 quality (in dB). Standard deviations are given in brackets.

Overall, the retroflex affricate tends to exhibit significantly higher intensity values than the dentoalveolar and the alveolopalatal affricate. This is especially evident for the geminate cognates before /a, i, u/ (F (2, 74) = 3.820, p < .05; F (2, 76) = 5.025, p < .05; F (2, 21) = 4.084, p =.06), but also for the singleton consonants before /a/ (F (2, 74) = 5.997, p < .01). Table 5 displays the cross-speaker intensity averages split by speaker. A quite coherent picture emerges. For one subject (S4), the geminate retroflex affricate exhibits a significantly higher frication intensity than both the alveolopalatal and the dentoalveolar (F (2, 40) = 8.124, p < .05; post-hoc Tukey HSD test reveals that all comparisons are significant at the p <.05 level). For three subjects (S1, S2, S3), the intensity values for the retroflex and alveolopalatal consonants are significantly higher than for the dentoalveolar one (for all three subjects, the retroflex vs. alveolopalatal comparison is significant according to the post-hoc Tukey HSD test at the p < .05 level of significance). Finally for one subject (S5), the intensity level for the retroflex affricate is non-significantly lower than that for the alveolopalatal and non-significantly higher than that for the dentoalveolar. As for the singleton consonants, no consistent or significant difference was found to hold for any affricate pair.

14

Geminates

Singleton

Subjects

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

S1

68 (2)

67 (5)

64 (3)

68 (3)

67 (2)

69 (1)

S2

48 (2)

45 (3)

42 (2)

41 (4)

44 (2)

40 (5)

S3

48 (3)

48 (3)

44 (3)

49 (3)

48 (2)

45 (1)

S4

74 (2)

70 (4)

62 (4)

72 (1)

72 (3)

65 (1)

S5

52 (2)

55 (3)

50 (4)

55 (3)

54 (4)

53 (2)

Table 5. Frication intensity values for singleton and geminate affricates split by speaker (in dB). Standard deviations are given in brackets.

These data appear to be consistent with those presented in Table 4: the frication phase for the retroflex affricate is in most cases significantly more intense than that for the dentoalveolar affricate and may also be more intense than the frication phase for the alveolopalatal affricate. This pattern is mostly evident in the case of geminates. For Polish, frication intensity does not contribute to differentiate retroflex [ˇß] from dentoalveolar [ts], palatal [t˛], and palatalized [tSj] (Żygis 2005). It should be recalled however that Polish affricates are singleton consonants, their durations being about 110 ms for the retroflex, 130 ms for the dentoalveolar and 125 for the palatal. It is then evident that the duration of Polish affricates is in some respects intermediate between that of the Sicilian singleton and geminate affricates (see Table 3 above for the average duration values for the latter). The short duration of the Polish affricates might explain the absence of significant intensity differences among them.

4.4. Spectral mean of the frication phase The centre of gravity (COG) of the fricative period of an affricate has been shown to correlate with the size of the front cavity: the smaller the front cavity, the higher the COG value (see Gordon et al. 2002, Kim 2001). Retraction in place of articulation is expected to lead to an increase in the size of the front cavity and therefore to a decrease in COG. In the acoustic analysis of Polish and Czech

15

affricates conducted by Żygis (2005), COG variations were expected to correlate with place variations. Another relevant articulatory feature of consonant retroflexion, i.e., the presence of a sublingual cavity (see section 3), could in principle be at the origin of an increase in front cavity size. As previously mentioned, sublingual cavity size and constriction retraction often correlate with each other in retroflex consonants. Nevertheless, at the articulatory level, these two parameters need to be conceived of as distinct and independent from each other such that the presence of the former does not imply ipso facto the presence of the latter. Acoustic cues that are potentially opaque with respect to the distinction of interest require further disambiguation. For the purposes of the present study, COG values were calculated a 26 ms Hanning FFT window centered around the point of maximum intensity within the frication phase. Table 6 displays cross-speaker COG values for each consonant split by V2. It may be seen that COG values for the retroflex affricate are always lower than those for the dentoalveolar affricate, but undistinguishable from those for the alveolopalatal affricate thus suggesting that retroflex and alveolopalatal affricates are closely articulated.

Geminate

Singleton

V2

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

/a/

3908 (130)

4399 (490)

6435 (965)

3956 (438)

4377 (289)

6448 (374)

/e/, /E/

3854 (101)

4586 (287)

5718 (320)

4279 (539)

4258 (385)

5989 (811)

/i/

3663 (102)

4217 (616)

6471 (904)

4261 (432)

4224 (457)

6137 (627)

/o/, /ç/

4356 (552)

4159 (605)

5977 (766)

4009 (454)

4189 (478)

6201 (701)

/u/

4395 (211)

4614 (439)

5799 (524)

4070 (647)

4591 (399)

6199 (519)

Table 6. COG values for singleton and geminate affricates split by V2 quality (in Hz). Standard deviations are given in brackets.

16

Due to quantitative limitations and in parallel to the intensity analysis, COG statistical calculations were run only on the data for geminates in the context of /a/, /i/ and /u/ and for singletons in the context of /a/ and /i/. ANOVAs yielded a highly significant effect for all five contexts mentioned, i.e., F (2, 74) = 21.740, p < .001 (geminates, V2 = /a/), F (2, 76) = 3.532, p < .001 (geminates, V2 = /i/), F (2, 21) = 8.541, p <.05 (geminates, V2 = /u/), (2, 76) = 44.232, p < .001 (singletons, V2 = /a/) and F (2, 40) = 30.409, p < .001 (singletons, V2 = /i/). According to Sidak post-hoc tests, COG values for the retroflex and the alveolopalatal affricates do not differ significantly from each other with the only exception of singletons before /a/ (p <. 05), while being significantly lower than those for the dentoalveolar affricate at the p <.001 level. Statistical results for the individual speakers are in line with those across speakers (see Table 7). One-way ANOVAs reveal a main effect of consonant for all speakers for both singletons and geminates (p < .001). Post-hoc Sidak tests yielded a significant difference for [ˇ(˘)ß] vs [t(˘)s] and [t(˘)S] vs [t(˘)s] in all instances (p < .001), and a non-significant difference between [ˇ(˘)ß] and [t(˘)S] except for singleton consonants in the case of one subject (S5, p < .01).

Geminate

Singleton

Subjects

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

S1

4013 (422)

3496 (442)

5805 (560)

3758 (366)

3576 (132)

5964 (720)

S2

5023 (456)

4679 (379)

6237 (659)

3910 (461)

4486 (321)

6219 (461)

S3

4808 (375)

4863 (190)

6799 (535)

4431 (347)

4551 (449)

6384 (247)

S4

4567 (356)

4404 (410)

7581 (635)

4061 (502)

4319 (220)

7409 (664)

S5

4091 (281)

4263 (202)

5678 (392)

3796 (385)

4311 (115)

5679 (267)

Table 7. COG values for singleton and geminate affricates split by speaker (in Hz). Standard deviations are given in brackets.

Furthermore, it must be acknowledged that the COG values of the retroflex consonant in Sicilian are always much higher than the F4 values measured at the following vowel steady state period. Consider the following examples: when preceded by [ˇ˘ß], F4 is 3295 Hz for /a/, 3559 Hz for /i/ and 3289 Hz for /u/; after 17

[ˇß], on the other hand, F4 amounts to 3831 Hz (/a/), 3667 Hz (/i/) and 3675 (/u/). A comparison between these frequency values with those for COG in Table 6 reveals that the spectral mean of the affricate fricative portion is consistently higher than F4 of the following vowel. Sicilian affricates differ from Polish affricates analyzed in Żygis (2005) in this respect since, according to the latter study, COG values for [ˇß] (and for [t˛] and [tSj]) approximate those of either F3 or F4 of the following vowel while those for [ts] were well above F5 of the following vowel. Either way, the most important difference between Sicilian and Polish retroflexes lies in the fact that, while in Sicilian COG for [ˇ(˘)ß] lies close to that for [t(˘)S] and is significantly lower than that for [t(˘)s], in Polish, the spectral mean for [ˇß] is lower than that for [ts], [t˛] and [tSj] and even lower than that for Czech [tS] (Żygis 2005). The Polish data could look rather surprising inasmuch as, as argued by Żygis, COG values are assumed to correlate directly with place of articulation and, therefore, should reflect the existence of a more posterior place of articulation for the retroflex than for the palatal or palatalized postalveolar affricates. The argument seems difficult to maintain in such terms, however. Instead, as pointed out at the beginning of this section, one should assume that, an increase in front cavity size for a retroflex articulation can be attributed to a typical sublingual cavity independently of lingual constriction fronting degree. The low COG value for Polish [ˇß] might henceforth be related to the fact that, besides being articulated at the postalveolar region, the affricate of interest is likely to be featured by the presence of a sublingual cavity. This would result in an increase in the size of the oral cavity and would cause lower COG values than even those for the palatal and palatalized cognates. To summarize, as the spectral mean of the frication phase suggests, Polish and Sicilian retroflex affricates are likely to differ from one another regarding at least two articulatory features. Sicilian [ˇ(˘)ß] is indistinguishable from the alveolopalatal affricate at the spectral level because it is not produced with either a more posterior point of articulation nor a sublingual cavity (recall, however, the effect of apicality on frication duration and intensity discussed in sections 4.2 and 4.3). By contrast, Polish retroflexes are likely to be featured by either a more posterior constriction location, the presence of a sublingual cavity or both, the last option being the most probable.

18

4.5. Spectral peaks of the burst (P1) and the frication period (P2) The highest spectral peak at the burst whenever present (P1) and at the point of maximum intensity of the frication phase (P2) were derived by means of an LPC peak-picking algorithm with a 50 Hz pre-emphasis, a 25 ms analysis window and a 16 prediction order. In parallel to the centre of gravity computation procedure, the frequency of the highest spectral peak should be inversely related to the size of the oral cavity during fricative or affricate articulation (Kim 2001, Żygis 2005). Since spectral peaks need to be evaluated in relation to the formant frequencies of the following vowel, it was essential here more than in the previous analyses to split the data by vowel quality. For this reason, data for the different vocalic contexts were evaluated separately even for each subject. However, this procedure curtailed the numerical consistency of many data subsets thus disallowing any data statistical scrutiny. When data for all subjects are considered together, a rather coherent picture emerges especially for geminate consonants (see Table 8). At P1, the geminate retroflex affricate shows a significantly lower spectral peak frequency than the alveolopalatal and the dentoalveolar, at least in the context of /a/ (F (2, 42) = 19.067, p < .001) and /u/ (F (2, 12) = 6.755, p < .05). As for V2 = /i/, the corresponding comparison turned out to be non significant. As for the singleton consonants, frequency values for the retroflex do not differ from those for the alveolopalatal in the context of /i/, and are roughly equivalent to those for the alveolopalatal and the dentoalveolar in the context of /a/. No consistent consonant-dependent trend occurs in the /e/ and /o/ contexts. At P2, on the other hand, the scenario for geminates is most coherent in that the retroflex shows again the lowest frequency value of all affricates before /a/ and /i/ (F (2, 74) = 25.373, p < .001; F (2, 76) = 26.561, p < .001) and to some extent before /e/, but not before /u/. As for the singletons, the lowest frequency value corresponds to the retroflex affricate only when a central vowel follows (F (2, 70) = 30.281, p < .001).

19

Burst (P1) Geminate

Frication period (P2) Singleton

Geminate

Singleton

V2

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

[ˇ˘ß]

[t˘S]

[t˘s]

[ˇß]

[tS]

[ts]

/a/

3324 (168)

3768 (427)

5256 (669)

3445 (604)

3520 (687)

3481 (765)

3429 (91)

4154 (558)

6715 (606)

3737 (475)

4107 (818)

6910 (1022)

/e/, /E/

3643 (322)

3723 (170)

4666 (645)

3643 (321)

3223 (413)

3775 (541)

3439 (229)

4211 (645)

5661 (397)

4107 (55)

4016 (852)

5353 (654)

/i/

3266 (324)

3639 (318)

4588 (843)

3503 (525)

3487 (288)

4947 (290)

3413 (377)

3672 (399)

6948 (1578)

3992 (884)

3941 (718)

5819 (787)

/o/, /ç/

2472 (441)

3660 (125)

4730 (985)

2362 (359)

2654 (675)

2990 (387)

4669 (1645)

3813 (1804)

5874 (1886)

5161 (1014)

4762 (787)

5898 (998)

/u/

2511 (189)

2804 (228)

3620 (487)

2093 (159)

2012 (365)

3654 (441)

4412 (876)

5938 (1060)

4673 (712)

4146 (691)

3549 (661)

6014 (503)

Table 8. Spectral P1 and P2 frequency values for singleton and geminate affricates split by V2 (in Hz). Standard deviations are given in brackets.

Splitting the data by subjects reveals that not all tendencies illustrated thus far hold for all individual speakers. In particular, lowest P1 and P2 values are detectable for S1, S3 and S5 before /a/, for S1 before /i/ and /u/, for S4 before /u/ (geminate) and before /i, u/ (singleton), and for S5 before /u/ (geminate). In summary, data for the burst and frication period spectral peak confirm that constriction location is more posterior for the Sicilian retroflex than for the dentoalveolar and, to a more limited extent, the alveolopalatal as well. A similar pattern was found to occur for the COG data. Results were less clear-cut and more subject to speaker-dependent variability for the former measure than for the latter. Comparable results were obtained by Żygis (2005) inasmuch as the burst and frication spectral peaks did not show significant differences between the postalveolar affricates of Czech ([tS]) and Polish ([ˇß]). In sum, spectral peak analysis data provides support for a large degree of variability in front cavity size in the case of both Polish and Sicilian affricate realizations.

4.6. Vowel formant transitions F3 and F4 of the preceding and following vowels were semi-automatically extracted from LPC spectra using a peak-picking algorithm set over the same 20

default values mentioned above for burst and frication peak extraction. The speech acoustic material was previously downsampled to 11 KHz for female speakers and to 10 KHz for male speakers. Formant values were obtained manually at four temporal points: (i) V1 steady-state; (ii) V1 offset; (iii) V2 onset; (iv) V2 steady-state. In retroflex articulations, F3 and F4 lowering is generally said to be related to place of articulation and tongue dorsum retraction; moreover, retroflex sounds are expected to show an asymmetric distribution of the formant cues with the VC transitions being more prominent than the CV transitions (see section 3; Stevens & Blumstein 1975, Spajić et al. 1996). In order to determine consonant-dependent differences in the frequency extent of the formant transition, F3 and F4 values were measured at the steady-state of V1 and V2 and compared statistically by means of paired-samples t-tests with the corresponding formant frequency values at VC transition offset and CV transition onset, respectively. This statistical procedure proved fruitful for formant transition analyses of Sicilian and Corsican [Í(˘)] (see Celata, 2006). With respect to more traditional analysis procedures, where formant values at the VC transition offset in sequences with a retroflex consonant were directly compared with those in sequences with a non-retroflex coronal consonant, this method has several advantages. First, it avoids relying on the F3 and F4 absolute values, which are known to show a great deal of asystematic and unpredictable variation, even when the vocalic environment is kept constant. Second, it actually focuses on the dynamic aspects of the vowel transitions since each VC or CV formant endpoint value is directly compared to the corresponding vowel steadystate value. Finally, matched-samples statistics reduces cross-subject variability. There was consequently no need for splitting the data by subject /gender; only variation associated with vowel context was taken into account. Results for the VC and CV transitions are shown in Figures 2(a, b, c) and 3(a, b, c). As in the previous analyses, data were preliminarily split by consonant length. Since, however, no relevant differences were found between geminate and singleton consonants, the average values across singleton and geminate retroflex, alveolopalatal and dentoalveolar affricates were plotted.

[Insert Fig_2a, Fig_2b and Fig_2c here]

21

Fig. 2 (a, b, c) F3 and F4 values at the V1 steady-state period and at the VC transitions offset plotted for [ˇ(˘)ß], [t(˘)S] and [t(˘)s] as a function of vowel context.

[Insert Fig_3a, Fig_3b and Fig_3c here] Fig. 3 (a, b, c) F3 and F4 values at the CV transitions onset and at V2 steady-state period plotted for [ˇ(˘)ß], [t(˘)S] and [t(˘)s] as a function of vowel context.

The VC transitions endpoints for the retroflex affricate (see Fig. 2a) are significantly falling for F3 in the context of /o/, /ç/ (df = 19; t = 4.317, p < .05) and /u/ (df = 12; t = 3.122, p < .05), and for F4 only in the context of /o/ or /ç/ (df = 17; t = 2.976, p < .05). No effects were found in the other vowel contexts, with the exception of the /u/ where F4 shows unexpectedly a significantly rising effect from V1 steady-state to C onset (df = 12; t = -7.056, p < .01). Overall, this pattern differs from a trend for the dentoalveolar affricate not to exhibit falling transitions in any vowel environment (see Figure 2c). However, a significant F4 falling transition is detectable (see Figure 2b) for the alveolopalatal affricate in VC sequences with /u/ (df = 13; t = 4.977, p < .05), thus suggesting that high formants for posterior consonants may lower sporadically. A similar pattern emerges in the case of CV formant transitions. For the retroflex affricate, a significant rising effect from consonant offset to the steady-state period of the following vowel is detectable for F4 in the /o/, /ç/ context (df = 26; t = -3.111, p < .01), for F3 in the /u/ context (df = 10; t = -3.770, p <. 05), and to some extent for F4 in the /a/ and /u/ contexts (df = 36, t = -1.987, p = .078, and df = 10, t = -2.312, p = .069, respectively). This rising effect also occurs in the case of the alveolopalatal affricate in the /u/ context, which was statistically significant for F3 (df = 14; t = -2.738, p <.05) and marginally significant for F4 (df = 12; t = 1.656, p = .071). As for the dentoalveolar affricate, the formant transitions were almost flat in all vowel environments. The spectral data presented so far allow concluding that the place of articulation of Sicilian [ˇ(˘)ß] is close to that of [t(˘)S] and significantly different from that of

22

[t(˘)s]. Moreover, the retroflex affricate is especially retracted when preceded and/or followed by back vowels. This finding can easily be attributed to vowelconsonant coarticulation since both back vowels and retroflex consonants are typically produced with a tongue dorsum retraction which causes a mutual enhancement of their articulatory characteristics. Preference for retroflex consonants to occur in back vowel contexts has been observed in several other languages (see Celata 2006: 23 for a review). While the picture emerging from these findings is too blurred to allow drawing generalizations about the supposed asymmetry in the distribution of the formant cues, it seems implausible that a noticeable difference between the VC and CV transition could occur. No F3 falling effects were found for Polish retroflex affricates according to Żygis (2005) (F4 was not included in the analysis). This result may be ascribed either to the non-retracted quality of the Polish retroflex consonant or to the fact that /a/ was the only contextual vowel included in the experimental design. By contrast, the voiced retroflex stop [͢] of Sicilian and Corsican dialects analysed in Celata (2006) showed more systematic contextually-determined articulatory retraction cues. Falling and rising F3 and F4 effects were consistently found for both VC and CV transitions when the contextual vowel was /u/; moreover, similar formant movements happened to take place in a relevant number of cases in the context of /i/ and /e/ in the case of the Sicilian data and in the context of /a/ in the case of the Corsican data. Finally, the vowel formant trajectories turned out to interact with prosodic factors such as stress, i.e., they were more consistently falling and rising when the target vowel was stressed than when it was unstressed. The latter factor was also investigated for Sicilian [ˇ(˘)ß] in the present paper. Considering only those words with stressed V1 for VC transition analysis or those with stressed V2 for CV transitions analysis reduces significantly the dimension of the corpus, thus disallowing statistical testing. However, a comparison between the plots in Figure 4(a, b) with those in Figures 2(a) and 3(a) reveals that lexical stress does not seem to shape the results as far as [ˇ(˘)ß] is concerned. This is particularly evident for the CV transitions (see Figure 4b), where a trend towards a falling transition is found for F3 in the context of /u/ and for F4 in the context of /o/ and, less so, in the context of /a/.

[Insert Fig_4a and Fig_4b here] Fig. 4(a, b)

23

F3 and F4 values for [ˇ(˘)ß] at the VC transitions and V1 steady-state (a), and at the CV transitions and V2 steady-state (b). Data correspond to the stressed vowel condition.

4.2.7 Summary The acoustic data reported in this study has shed light on the following characteristics of the Sicilian voiceless retroflex affricate [ˇ(˘)ß]: (i) It is shorter than etymological [t(˘)r], and also shorter than the other affricates. (ii) Its frication period is shorter but more intense than the frication period of the other affricates. (iii) According the spectral data, it exhibits essentially the same place of articulation as the other postalveolar consonant [t(˘)S]. (iv) In the light of the data for the formant transitions, retraction in place of articulation is enhanced by back vowels. All these findings indirectly confirm the general opinion that articulatory differences among affricates are better specified at the frication period than at the closure period or at the vowel transition (see Recasens & Espinosa 2007, Żygis & Padgett 2007). Among the four parameters proposed by Hamann (2003: 19) for characterizing the articulation of retroflex consonants (i.e., apicality, posteriority, sublingual cavity and retraction; see section 3), apicality and posteriority appear to be relevant for Sicilian [ˇ(˘)ß] since the consonant has been found to be palatoalveolar and retraction is also present but with specific contextual limitations.

5. The origin of [ˇ(˘)ß] derived from [t(˘)r] A previous attempt to explain phonetically the process by which /t(˘)r/ may become a retroflex affricate was made by Sorianello & Mancuso (1998) in a paper where the voiceless retroflex consonant of the Calabrian variety spoken in Cosenza was submitted to spectrographic analysis. A highly similar proposal can

24

be found in Hamann (2003: 83-89) on the retroflexion of both /tr/ and /rt/ in the world’s languages. Sorianello & Mancuso (1998) hypothesized that the cluster was first affected by a process of rhotic retroflexion ([r] > [}]), which was assumed to be very likely to occur due to the strong acoustic similarity between rhotics and retroflexes. The second stage of the sound change was supposed to involve an anticipatory assimilation by which [t}] turned into [ˇ}]. Finally, a rhotic assibilation process occurred ([}] > [ß]) which, according to the authors, could be related to a general tendency for retroflex rhotics to exhibit a sibilant pronunciation in several Southern dialects of Italy. This trend is in fact well documented in the Cosentinian dialect where the geminate /r˘/ is transcribed as [ߢ] by Sorianello & Mancuso (1998: 154), as well as in Sicilian dialects (Ruffino 1991). The entire process may be summarized as in (1): (1)

/tr/

> /t}/

> /ˇ}/

> /ˇß/

(or /ˇ/)

The hypothesis by Sorianello & Mancuso (1998) (see also Hamann 2003: 83-89) is based on three fundamental assumptions, which are reviewed below. (i) Retroflexion is basically achieved through a process of regressive assimilation by which the postalveolar point of articulation of the rhotic would spread to the preceding alveolar stop. In the light of this possibility, this hypothesis will be referred to as the “assimilation hypothesis”. We will show in the following paragraphs that a phonetic mechanism differing from assimilation stricto sensu is more likely to account for /t(˘)r/ retroflexion. (ii) The rhotic turns out to be the first segment of the consonant cluster to develop a retroflexion feature as revealed by rhotic and retroflex consonants involving a falling F3 vowel transition. It is then possible to claim that the assimilation hypothesis has an acoustic-perceptual motivation. As a matter of fact, Hamann (2003) assumes the proposal originally put forward by Bhat (1973: 43) that, even if non-retroflex, an alveolar trill or a flap/tap can induce retroflexion in an adjacent consonant. In a following passage, the author states that the perceptually-based assimilation hypothesis may be a valid explanation for the retroflexion of rhotic-plosive clusters, and that «processes with a reversed order of rhotic and retroflex [i.e., /Cr/ as opposed to /rC/ clusters] can be treated identically» (Hamann 2003: 174; parentheses are mine). As shown in the course of this study, this statement must be revised. In the diachronic analysis of rhotic retroflexion, several aspects cannot be ignored such as that there are phonotactically different rhotic contexts (preceding /r/, following /r/), as well as 25

different rhotic places and manners of articulation. The two aspects are related to each other since variation in rhotic articulation is often a consequence of a change in the phonotactic structure of the word (just take, as a familiar example the tendency for rhotics produced as trills intervocalically to reduce to flaps or taps in postconsonantal position). (iii) The intermediate retroflex sequence [ˇ}] raises further questions. First of all, among the large number of retroflex variants used for the cluster /tr/ in the southern dialects of Italy (see section 2), we are not informed about the existence of any [ˇ}] pronunciation which also questions the actual existence of this phonetic realization in the past as a preliminary stage to other phonetic changes. Second, the tendency for rhotics to exhibit a sibilant character is documented in postvocalically Calabrian and Sicilian, and, as pointed out above, in particular in the case of the geminate trill. An extension to the postconsonantal position, as the assimilation hypothesis requires, is not obvious at all. In the study of sound change, any dissociation between the acoustic and perceptual facts and the articulatory events tends to reduce the complexity of the diachronic changes to a mere juxtaposition of two or more synchronic stages. As far as /tr/ retroflexion in the Romance dialects is concerned, an articulatory explanation has the advantage of taking into account an aspect which has been neglected in perception-based accounts, that is, rhotics undergo a process of articulatory reduction in specific contexts. Articulatory reduction is rooted in a universal trend by which rhotics tend to be strongly coarticulated with the adjacent segments and, in some cases, even to lose their segmental status. In particular, any reconstruction based on the mere acoustic similarity between rhotics and retroflexes does not handle the Romance data where three contextual conditions appear to be involved in the rhotic-induced retroflexion process: (1) cluster tautosyllabicity (/rt/ does not undergo retroflexion); (2) cluster homorganicity (retroflexion does not occur in the sequences /kr/ and /pr/); (3) rhotics occupy a comparatively large slot within the cluster as shown by the contrast between /tr/ and /pr/ referred to in section 4.2. A reliable explanation should then take into account these contextual restrictions applying to the retroflexion process in Sicilian and other Romance domains. In the present proposal, the acoustic similarity between rhotics and retroflexes is still a relevant factor. However, the focus shifts from changes in rhotic place of articulation to changes in rhotic manner of articulation. In addition, the notion of assimilation as a fundamental mechanism for the development of the retroflex affricate is rejected here.

26

As already mentioned (see above, section 3), apical trills are realized through a complex production mechanism requiring synchronization of different articulatory and aerodynamic parameters (Recasens & Pallarès 1999, Solé 2002). Articulatory complexity explains several aspects typically associated with rhotic production, such as that rhotics are acquired later than other consonants and are often produced inaccurately, show a high degree of coarticulatory resistance in VCV sequences, exhibit much context-dependent articulatory and aerodynamic variation, and reduce frequently to non-vibrating allophones in diachronic and synchronic alternations. Articulatory reduction is known to be sensitive to syllabic position: consonants are reduced in syllable final position more than in initial position and in tautosyllabic sequences more than in heterosyllabic ones (Straka 1964, Ohala & Kawasaki 1984, Fougeron 1999; and see Recasens 2004 for rhotic reduction in Catalan). As for rhotics, trills undergo different degrees of articulatory reduction depending on syllable position. As familiar examples take the alternation between [rosa] and [tRes] (*[tres]) in Spanish (Bakovic 1994), the difference between [mçrso] and [tREno] in Standard Italian (Farnetani & Kori 1986), and the change /r/ > [}], [R] in syllable-final and word-final position in Brazilian Guaranì (Dietrich 2002). According to the articulation-based explanation of /tr/ retroflexion, the crucial point is the progressive articulatory reduction undergone by the apical trill in the tautosyllabic and homorganic cluster /tr/ more than in any other contextual condition. The temporal compression effect by which the rhotic ends up occupying a relatively long portion of the cluster (see section 4.2) turns out therefore to be articulatorily grounded. More importantly, reduction may yield a shift in both place and manner of articulation. As for manner, this implies a reduction of the number of contacts. An apical trill is therefore expected to reduce firstly to an alveolar or postalveolar flap or tap, and secondly to an alveolar or postalveolar approximant. This development is summarized in (2). Given that rhotics are strongly characterized by variation in place of articulation (Ladefoged & Maddieson 1996: 215-217) and in view of the similarity between rhotics and retroflexes, [R] and a [}] are likely to alternate. As articulatory reduction becomes the rule, an approximant can arise thus causing [®] and [”] to alternate in free variation. (2)

[r]

>

[R] ~ [}] >

[®] ~ [”]

At this stage, the rhotic tends to lose its consonantal status and to reduce to a voiced appendix of the preceding stop. In other words, as shown in (3) below, an

27

affrication process applies which is consistent with diachronic developments widely attested in Romance phonology. (3) [tr]

>

[tR] ~ [t}]

>

[t®] ~ [t”]

> [ˇß]

The affrication process in question parallels, for example, the process [pla.tĕ.a] Latin platěa > [pla.tja] > [plat.tja] > [pjat˘sa] Italian piazza ‘square’. Retroflexion thus stems from affrication, starting out in a cluster and giving rise to a monosegmental affricate through resyllabification. The postalveolar place of articulation emerging from /tr/ affrication can easily be explained through articulatory blending. Both realizations [t®] and [t”] can cause the alveodental stop [t] to achieve a posterior place of articulation resulting from an increase in coarticulation between the two consonants of the cluster. In the course of the affrication process, the tongue body reaches a retracted configuration, allowing the apex to rise against the palate; this definitely leads to an apical pronunciation. Moreover, articulatory blending easily accounts for retroflexion not applying to non-homorganic /r/-clusters such as /pr/ and /kr/. In sum, the explanatory advantages of the articulatory proposal with respect to the previous acoustic-perceptual hypotheses are as follows. First, the phonetic mechanism by which retroflexion arises is the affrication of a stop-approximant sequence, which is highly unmarked and applies very frequently synchronically and diachronically in Romance. Second, affrication affects specifically strongly reduced rhotic realizations yielding an approximant postconsonantally in a tautosyllabic cluster. Third, coarticulation is particularly strong in the context of retroflexion since the two consonants are homorganic in this case. Fourth, the rhotic does not change into a separate fricative ([}] > [ß]), but is involved in the development of a sibilant affricate as a result of articulatory blending with the preceding stop.

6. Conclusions The retroflexion of the cluster /t(˘)r/ in the Sicilian dialect has been examined from an acoustic and historical point of view. The voiceless retroflex affricate [ˇ(˘)ß] in the speech of some Sicilian subjects appears to be well differentiated acoustically from other affricates showing an anterior ([t(˘)s]) or a posterior place of articulation ([t(˘)S]), especially regarding duration and apicality. However, it turns out to be articulated at a place that is no significantly more posterior than that for [t(˘)S] and exhibits velarization in specific contexts only. Many of these articulatory characteristics, as well as the contextual and phonosyntactic 28

restrictions involved in retroflexion, call for an articulatory account of the affricate retroflexion process through rhotic reduction and articulatory blending.

References

Anderson, V. 1997. “The perception of coronals in Western Arrernte”. Proceedings of Eurospeech, Rhodes, Greece, 22-25 September 1997. Vol. 1: 389392. Bakovic, E. 1994. “Strong onsets and Spanish fortition”. MIT Working Papers in Linguistics 23: 21-39. Bhat, D.N.S. 1973. “Retroflexion: An areal feature”. Working Papers on Language Universals 13: 27-67. Blaylock, C. 1968. “Latin L-, -LL-, in the Hispanic dialects: Retroflexion and lenition”. Romance Philology 21: 392-409. Blevins, J. 2004. Evolutionary Phonology. The Emergence of Sound Patterns. Cambridge: Cambridge University Press. Bottiglioni, G. (1927), “La penetrazione toscana e le regioni di Pomonte nei parlari di Corsica (II parte)”. L’Italia Dialettale 3: 1-69. Caracausi, G. 1983. Arabismi medievali di Sicilia. Palermo: Centro Studi Filologici e Linguistici Siciliani. Celata, C. 2002-2003. “Polimorfismo nei dialetti della Corsica. Gli esiti di lj nel dialetto del Fiumorbu”. L’Italia dialettale 63-64: 51-82. Celata, C. 2006. Analisi del processo di retroflessione consonantica in area romanza, con dati sperimentali dal còrso e dal siciliano. PhD unpublished dissertation, Pisa, Scuola Normale Superiore. Celata, C. (forthcoming). “Retroflexion of geminate laterals in Sicilian and Corsican”. Clemente, G., S. Serio & O. Manfrellotti. 2003. “Alcuni dati sul consonantismo palermitano tra dialetto e varietà regionale”. In G. Marotta & N. Nocchi (eds.), La Coarticolazione. Atti delle XIII Giornate del GFS . Pisa, ETS: 59-77. Contini, M. 1987. Etude de géographie phonétique et de phonétique instrumental du sarde. Alessandria: Edizioni dell’Orso. 29

Dalbera-Stefanaggi, M.J. 1991. Unité et diversité des parlers corses. Alessandria: Edizioni dell’Orso. Dart, S.N. & P. Nihalani. 1999. “The articulation of Malayalam coronal stops and nasals”. Journal of the International Phonetic Association 29: 129-142. Dave, R. 1977. “Retroflex and dental consonants in Gujarati: a palatographic and acoustic study”. Annual Report of the Institute of Phonetics, University of Copenhagen 1: 27-156. Dietrich, W. 2002. “A situação linguística na "zona guaranítica" brasileira (1)”. Revista Portuguesa de Filología 24. Dixit, R.P. 1990. “Linguotectal contact patterns in the dental and retroflex stops of Hindi”. Journal of phonetics 18: 189-201. Fant, G. 1974. Speech Sounds and Features. Cambridge: MIT Press. Farnetani, E. & S. Kori. 1986. “Effects of syllable and word structure on segmental durations in spoken Italian”. Speech communication 5: 17-34. Fougeron, C. 1999. “Prosodically conditioned articulatory variation: A review”. UCLA Working Papers in Phonetics 97: 1-73. Gordon M. & P. Barthmaier & K. Sands. 2002. “A cross-linguistic acoustic study of voiceless fricatives”. Journal of the International Phonetic Association 32: 141-171. Guarnerio, P.E. 1902. “Il sardo e il còrso in una nuova classificazione delle lingue romanze”. Archivio Glottologico 16: 491-516. Hamann, S. 2003. The phonetics and phonology of retroflexes. Utrecht: LOT. Hamann, S. 2003b. “Norwegian retroflexion – licensing by cue or prosody?” In A. Dahl, K. Bentzen & P. Svenonius (eds.) Special Issue of the Proceedings of the 19th Scandinavian Conference of Linguistics. Nordlyd 31,1: 63-77. Hedrick, M.S. & R.N. Ohde. 1993. “Effect of relative amplitude on perception of place of articulation”. Journal of the Acoustical Society of America 94: 20052026. Kim H. 2001. “The place of articulation of the Korean plain affricates in intervocalic position: an articulatory and acoustic study”. Journal of the International Phonetic Association 31: 227-257. Krull D., B. Lindblom, B.-E. Shia & D. Fruchter. 1995. “Cross-linguistic aspects of coarticulation: An acoustic and electropalatographic study of dental and

30

retroflex consonants”. In Proceedings of the XIII International Congress of Phonetic Sciences, Stockholm. Vol. IV: 436-439. Labov W., S. Ash & C. Boberg. 2006. Atlas of North American English: Phonology and Phonetics. Berlin: Mouton/de Gruyter. Ladefoged, P. 1975. A course in phonetic. New York: Harcourt Brace Jovanovich. Ladefoged, P. & P. Baskararao. 1983. “Non-quantal aspects of consonant production: A study of retroflex consonants”. Journal of Phonetics 12: 291-302. Ladefoged, P. & I. Maddieson. 1996. The sounds of the world’s languages. Oxford: Blackwell. Loporcaro, M. 2001. “Le consonanti retroflesse nei dialetti italiani meridionali: articolazione e trascrizione”. Bollettino del Centro di Studi Filologici e Linguistici Siciliani 19: 207-233. Maiden M. & M. Parry, eds.1997. The Dialects of Italy. London: Routledge. Menéndez Pidal, R. 1954. “A propósito de ll y l latinas. Colonización suditálica en España”. Boletín de la Real Academia Española 34: 165-216. Merlo, C. 1925. “Concordanze corse – italiane centromeridionali”. L’Italia Dialettale 1: 238-251. Millardet, G. 1933. “Sur un ancien substrat commun à la Sicile, la Corse et la Sardaigne”. Révue de Linguistique Romane 9 : 346-369. Ohala, J.J. 1981. “The listener as a source of sound change”. In C.S. Masek, R.A. Hendrik & M.F. Miller (eds.), Papers from the Parasession on Language and Behaviour. Chicago: Chicago Linguistic Society: 178-203. Ohala, J.J. 1983. Aspects of Hindi phonology. Delhi: Motilal Banarsidad. Ohala, J.J. & H. Kawasaki. 1984. “Prosodic phonology and phonetics”. Phonology Yearbook 1: 113-128. Ohala, M. & J.J. Ohala. 2001. “Acoustic VC transitions correlate with degree of perceptual confusion of place contrast in Indi”. In N. Grø & J. Rischel (eds.) To honour Eli Fischer-Jorgensen. Copenhagen: Reitzel: 265-284. Recasens, D. 2004. “The effect of syllable position in consonant reduction. Acoustic and articulatory evidence from heterosyllabic consonant clusters”. Journal of Phonetics 32: 435-453. Recasens, D. & A. Espinosa. 2007. “An experimental study of affricate consonants in Majorcan and Valencian Catalan”. Journal of the International Phonetic Association 37: 143-172. 31

Recasens, D. & M.D. Pallarès. 1999. “A study of /tap/ and /trill/ in the light of DAC coarticulation model”. Journal of Phonetics 27: 143-170. Rohlfs, G. 1955. “Vorrömische Lautsubstrate auf der Pyrenäenhalbinsel?”. Zeitschrift für Romanische Philologie 71: 408-413. Rohlfs, G. 1966 [=1949]. Grammatica storica dell’italiano e dei suoi dialetti.Fonetica. Torino: Einaudi. Ruffino, G. 1991. Dialetto e dialetti di Sicilia. Palermo: CUSL. Savoia, L.M. 1980. “Fonologia delle varietà apuane e garfagnine: Consonantismo”. Studi Urbinati di Storia, Filosofia e Letteratura. Supplemento Linguistico 2: 233-293. Schmeck, H. 1952. “Probleme der korsischen Konsonantismus. Phonologische Darstellung”. Zeitschrift für Romanische Philologie 68: 49-72. Schneegans, H. 1888. Laute und Lautentwickelung des sicilianischen Dialectes nebst 1 Mundarten Karte und aus dem Volksmunde gesammelten Sprachproben. Strassburg: K. J. Trübner. Solé, M.J. 2002. “Aerodynamic factors characteristics of trills and phonological patterning”. Journal of Phonetics 30: 655-688. Sorianello, P. & A. Mancuso. 1998. “Le consonanti retroflesse nel cosentino: analisi preliminare”. In Atti delle VIII Giornate di Studio del GFS, Pisa 18-19 dicembre 1997: 142-154. Spajić, S., P. Ladefogd & P. Bhaskararao. 1996. “The trills of Toda”. Journal of the International Phonetic Association 26: 1-21. Steriade, D. 2001. “Directional Asymmetries in Place Assimilation: A Perceptual Account”. In E. Hume & K. Johnson eds., Phonology: The Role of Speech Perception in Phonology. San Dieg: Academic Press: 219-250. Stevens, K. & S. Blumstein. 1975. “Quantal Aspects of Consonant Production and Perception: a Study of Retroflex Stop Consonants”. Journal of Phonetics 3: 215233. Straka, G. 1964. “L’évolution phonétique du latin au français sous l’effect de l’énergie et de la faiblesse articulatoires”. Travaux de Linguistique et de Littérature, Université de Strasbourg 2 : 17-98. Tropea, G. 1963. “Pronunzia maschile e pronunzia femminile in alcune parlate del messinese occidentale”. L’Italia Dialettale 25: 1-29.

32

Van de Velde, H. & R. Van Hout eds. 2001. “R-atics”. Sociolinguistic, phonetic and phonological characteristics of /r/. Bruxelles: Institut des Langues Vivantes et de Phonétique. Zhou, X., C.Y. Espy-Wilson, M. Tiede & S. Boyce. 2007. “An articulatory and acoustic study of “retroflex” and “bunched” American English rhotic sound based on MRI”. In Proceedings of Interspeech 2007, Antwerp, Belgium: 54-57. Żygis, M. 2005. “(Non)retroflexivity of Slavic affricates”. ZAS Papers in Linguistics 42: 69-115. Żygis, M. & J. Padgett. 2008. “A perceptual study of Polish fricatives, and its relation to historical sound change”. http://people.ucsc.edu/~padgett/papers.html

[Fig.1]

33

Retroflex affricate 4200

Frequency (Hz)

4000 3800 3600 3400 3200 F3_V1 3000 F3_VC 2800 F4_V1

2600 2400

F4_VC

a

e

i

o

u

V1

[Fig.2a]

Alveolopalatal affricate 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_V1 3000 F3_VC 2800 F4_V1

2600 2400

F4_VC

a

i

o

u

V1

[Fig. 2b]

34

Dentoalveolar affricate 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_V1 3000 F3_VC 2800 F4_V1

2600 2400

F4_VC

a

e

i

o

u

V1

[Fig. 2c]

Retroflex affricate 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_CV 3000 F3_V2 2800 F4_CV

2600 2400

F4_V2

a

e

i

o

u

V2

[Fig. 3a]

35

Alveolopalatal affricate 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_CV 3000 F3_V2 2800 F4_CV

2600 2400

F4_V2

a

e

i

o

u

V2

[Fig. 3b]

Dentoalveolar affricate 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_CV 3000 F3_V2 2800 F4_CV

2600 2400

F4_V2

a

e

i

o

u

V2

[Fig. 3c]

36

VC transitions (Retroflex affricate) 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_V1 3000 F3_VC 2800 F4_V1

2600 2400

F4_VC

a

e V1 (stressed)

[Fig.4a]

CV transitions (Retroflex affricate) 4200

3800 3600 3400

(Hz)

Frequency (Hz)

4000

3200

F3_CV 3000 F3_V2 2800 F4_CV

2600 2400

F4_V2

a

i

o

u

V2 (stressed)

[Fig.4b]

37