Laryngeal Features of Armenian Dialects

Laryngeal features of Armenian dialects

Giancarlo Schirru University of Cassino This paper concerns the laryngeal features of the Armenian stop system in a comparative perspective. First we will summarize a well-known problem in Armenian historical phonology, concerning the status of the series traditionally called "voiced. " Then we will present the acoustic data from field research, carried on in the Republic of Armenia, on different Armenian dialects, and we will discuss a possible phonological analysis, in both synchronic and diachronic terms. In the end we will analyze the position of the Armenian language in an Indo-European perspective, with special regard to Indo-Aryan reflexes of Indo-European stops.

The aim of this study is to motivate a line of research in the area of Indo-European consonantism, based on the hypothesis that the Indo-European series of stops traditionally called "voiced aspirated" could be better analyzed if specified by the [+slack vocal folds] or [-stiff vocal folds] features. For this purpose, we will make use of phonetics observations on living Indo-European varieties as arguments for their phonological analysis, and we will treat phonology of modern languages (the only ones investigable by "experimental phonology") as an object of comparison within the perspective of Indo-European phonology, together with arguments coming from historical phonology of singular historical traditions, comparative phonology and internal reconstruction.

Phonology of Armenian stops

Modern Standard Eastern Armenian, based on the Erevan variety, has a system of 15 stops; it can be classified in 3 series and 5 classes (see Table 1);1 1 See Ladefoged & Maddieson 1996: 66-68;Vaux 1998: 12-13 and bibliography quoted therein; Hacopian 2003. in Armenian writing, each stop in this system is expressed by a different graphical sign:

voiceless aspirated plain voiceless voiced bilabial /p / <փ> / p / < պ> /b/ <բ> alveolar /t / <թ> / t / < տ> /d/ <դ> alveolar affricate /ts / <ց> / t s / < ծ> /dz/ <ձ> postalveolar affricate /tʃ / <չ> / t ʃ / < ճ> /d / <ջ> velar /k / <ք> / k / < կ> / / <գ> Table 1 Stop system of Modern Eastern Armenian

The 5 classes refer to the place of articulation and to the presence of affrication. The 3 series are defined in terms of laryngeal features: in many descriptions, they are named "voiceless aspirated, " "plain voiceless" and "voiced. "

In the traditional reconstruction of historical phonology,2 Old Armenian has the same stop system as Modern Standard Eastern Armenian. The origin of the 3 series is linked with the Proto-Indo-European system through a chain shift, which may be summarized as follows: On the basis of these facts, Bolognesi argues that the input to (3) must be different from the output to (4):5 and concludes, following Holger Pedersen (1906: 336-41), Antoine Meillet (1922: 12-13;see Bolognesi 1987: 128-129), Émile Benveniste (1959), andHans Vogt (1938: 326-27;, that the Old Armenian so-called "voiced" series of stops, must instead be considered as voiced aspirated; therefore, in this reconstruction, the development (1.c) should be rewritten as (5): 5 PIE *dh > Arm. [d ] 3 "Après nasale, les sourdes arméniennes p, t, c, č, k subsistent de date ancienne, mais, de bonne heure, tendent à devenir sonores dans certains parlers" (Meillet 1936: 29). 4 Bolognesi 1960: 19-21; see also Godel 1975: 9-10 and n. 7 p. 9;Belardi 2006: 205-16. 5 An anonymous reviewer suggests another perspective to solve the paradox: namely a development *rt > *rth (aspiration of voiceless stops, like in other positions) > *rdh (voicing) > *rd. We can observe that such reconstruction has the disadvantage of postulate a voicing in *rth (where *th is in Armenian, at least in modern varieties as we will see, a true aspirate stop, and not a fricative like in Germanic) but not in rt, although such process is more predictable in a consonant with a shorter VOT than in a consonant with a longer one. Our reconstruction is consistent with a splitting of the PIE voiceless series in different Armenian reflexes: aspiration in strong position (fortition), voicing after sonorants (lenition), and further lenition until the status of approximants after vowels.

The same reflex -a voiced aspirated -should be reconstructed after sonorants: for ex. in O.Arm. որբ orb 'orphan' < PIE *orb -o-(cfr. Lat. orbus, Gr. ὀρφανός); O.Arm. բարձր barjr 'high' < PIE *b r ĝ u-(cfr. Skr. br h-ánt-; O.H.G. berg etc.). So, the process described in (3) would be better expressed as (6), thus solving the paradox:

The presence of voiced aspirated stops was already pointed out in modern Armenian dialects by a tradition of studies that began with Eduard Sievers (1901: § § 436, 442) and continued on to Hratchia Adjarian (1909) and Sidney Allen (1951). Articles by Benveniste (1959) and Vogt (1959) gave rise to broad debate on the nature of these sounds, which were considered either primary or secondary in the historical development.6 Nor is there agreement in the literature on the phonetic and phonological analysis of the Eastern Armenian "voiced" series for both the standard and the local dialects: thus, we have calls for new experimental research (e.g. in Lamberterie 1984: 227;Pisowicz 1997: 215;1998: 45).

The field research

Here we present data from field research carried out in the Republic of Armenia in May 2006: speech signals were collected by means of a list of 34 minimal pairs, read by a total of 27 speakers.7 We will consider just 15 of them here -native speakers of three different regional dialects: the variety of Erevan, the country's capital, which belongs to the Ararat dialect and is the basis for the modern standard of Eastern Armenian; the variety of Gyumri (the former Leninakan, and earlier Alexandropol), the country's 6 See references in Pisowicz 1976: 20-27;Kachaturian 1983;Vaux 1998: 211-41;Bolognesi 1998: 162-69. A new argument, supporting a reconstruction of the Proto-Armenian third series of stops as breathy, is offered by Garrett 1998. 7 The data were recorded with a Marantz PMD 670 digital recorder connected to a head-mounted, Shure unidirectional microphone. Acoustic analysis was made using Praat 5.1.03 software (www.praat.org). The analysis discussed below considers only a sub-set of the minimal pairs originally present in the list read by the speakers: 15 minimal pairs for the VOT values ( § 2.1) and 6 minimal pairs for the spectral tilt parameters ( § 2.1, 2.2.).

second largest city and main centre of the Širak region, in the Republic's northwestern corner, where a dialect usually classified as part of the Western Armenian group is spoken; and a group of speakers studied in Aštarak, a town situated around 30 kilometres northwest of Erevan, with a dialect of the Ararat group, where the so-called "voiced aspirated stops" were already documented.8 Investigations were made on site ( fig. 1).

Figure 1

Republic of Armenia mostly based on experimental observations illustrated in Allen (1951) and in Hacopian (2003).9 Our investigation is intended as a field research, and not a laboratory research: thus it can give just a first approximation to the linguistic reality, that could be better illustrated by a larger experimental observation: data are collected by mean of a technique currently used in similar researches.10

Erevan

For the Erevan dialect, we considered 5 speakers: speaker 1 (F, 34 years); speaker 2 (F, 53 years); speaker 3 (F, 25 years); speaker 4 (M, 72 years); and speaker 5 (M, 56 years).

Voice Onset Time (VOT) is the main acoustic correlate distinguishing the voiceless aspirated series from the plain voiceless one (see Kachaturian 1983: 60;Hacopian 2003). A longer VOT is attested in all the aspirates contrasting with a plain voiceless: this fact is constant in the 9 minimal pairs attesting to this opposition listed in (7a). Voiced stops have a voice bar (the so-called voice-lead) with a negative VOT, although this is not everpresent and, in some cases, is very weak: 6 minimal pairs (7b) display such opposition;11 the data are summarized in Table 2: 7 a պայտ payt /pajt/ 'horseshoe, cleat' ~ փայտ ayt /p ajt/ 'wood' տանկ tank /tank/ 'tank (armored vehicle)' ~ թանկ ṭank /t ank/ 'expensive' տարմ tarm /taɾm/ 'flock (of birds)' ~ թարմ ṭarm /t aɾm/ 'fresh, new' տող toł /to / 'line' ~ թող ṭoł /t o / 'let, allowed' 9 In Allen (1951) are analysed data recorded in London from one speaker of New Julfa; Hacopian (2003) is a laboratory research on the VOT parameter, with data offered by three native speakers of Teheran Eastern Armenian living in Los Angeles. 10 See, for example, Maddison & Ladefoged (1985); Cao & Maddison (1992). 11 Since the six minimal pairs listed in (7b) were also used for the spectral structure analysis of the first vowel pulses as described below, they were chosen among the 3 classes of stops without affrication phase, in which it is easier to identify the consonantal release; moreover, they all contain postconsonantal vowel /a/, in order to control for the influence of (differences in) vowel formant frequencies on the spectral tilt.

տուրք towrḳ /tuɾk / 'duty, tax' ~ թուրք ṭowrḳ /t uɾk / 'Turk' տուփ tow /tup / 'small box, carton' ~ թուփ ṭow /t up / 'bush, shrub, vein' ծեղ ceł /tse / 'stem, stalk' ~ ցեղ c eł /ts e / 'tribe, race' ճաղ čał /tʃa / 'knitting needle, rod, bar' ~ չաղ č ał /tʃ a / 'fat, stout, fertile'

voiceless aspirated plain voiceless voiced VOT (msec) 104.4 26 -115.9 Table 2 Average VOT values in Erevan (5 speakers, 15 minimal pairs)

Since the opposition between voiceless aspirated and plain voiceless stops is always implemented by bigger values of VOT, this parameter will not be discussed further, and the situation of the Erevan dialect may be taken as representative of the picture for the other varieties as well. But VOT is not the only acoustic correlate distinguishing the voiced series from the plain voiceless one. There is also a different quality of voice in the beginning of the post-consonantal vowel. If we take the first 50 msec. of the vowel signal following the consonantal release, and we compare this acoustic signal to the portion extracted after the voiceless stop, we find a smoother increase of the intensity of the vowel (already observed in Kachaturian 1983: 58), a lower pitch12 and a slacker voice (as opposed to a stiffer voice after voiceless stops).

This last property, the difference of voice quality, can be analysed in quantitative terms: one of the phonetic differences between slack and stiff voice is the spectral tilt, i.e. "the degree to which intensity drops off as frequency increases" (Gordon & Ladefoged 2001: 397). This property, already 12 We do not make further observations on the pitch, since the technique of recordings collection is not able to give valid data in this topic. Maddison (1992).

As a result, we may take the spectral tilt as one of the main acoustic correlates distinguishing between the plain voiceless and the voiced stops: the two series are thus characterized as having a stiff vs. slack voice quality in the first vocal pulses after the release of the consonant. Since many scholars have already described Armenian plain voiceless stops as "ejectives" (see, e.g., Allen 1951: 188), it may be stressed that in our corpus there are no more than 4 instances (in the set of 225 tokens of plain voiceless stops considered here) of ejective consonants, similar to ejectives in Georgian, and having the acoustic correlate of a double release (the glottal and the oral one; see Ladefoged & Maddieson 1996: 78-81). These may be considered as a minority type of phonetic realization of this series, perhaps influenced by contact with Kartvelian languages, an element quite strong in the history of the Standard Eastern Armenian variety which was originally greatly influenced by the dialect of the Tbilisi Armenian community.17 Therefore, we may draw a clear distinction between glottalized consonants, like ejectives, and consonants characterized by an increased stiffness (or tenseness) of vocal folds.

Aštarak and Gyumri

The distinction between aspirated voiceless and plain voiceless stops in the Aštarak and Gyumri dialects does not differ from that illustrated for the Erevan dialect in the previous section; no further acoustic accounts of it need be provided. Here we present the data for the distinction between the plain voiceless and voiced series.

In Aštarak, we recorded data from 5 native speakers: speaker 6 (M, 16 years), speaker 7 (M, 16 years), speaker 8 (M, 41 years), speaker 9 (M, 30 17 For this explanation for the presence of ejectives in some Armenian varieties, cf. Pisowicz 1989: 218-19;1997: 217;Rasmussen 1989: 160;Lamberterie 1984: 229-30;1994: 154;Belardi 2006: 214-16. The development of an ejective series of stops in contact with Caucasian languages is clearly attested in Ossetic (Abaev, Belardi & Minissi 1965: 62;Rasmussen 1989: 160;Thordarson 1989: 462); on the role played by the Tbilisi Armenian dialect in the making of Standard Eastern Armenian, see Nichanian 1989: 283-330. For a different interpretation of Armeninan ejective stops in the framework of the glottalic theory, see Kortlandt 1978: 110-11, and references cited therein; Gamkrelidze & Ivanov 1995: 36. years), and speaker 10 (M, 31 years). The realization of voiced stops is through two different allophones. A first phonotype does not display any voice bar during the occlusion of the consonant: it is, strictly speaking, voiceless in initial position. A different phonotype may occasionally be found, characterized by the voice bar and a clear aspiration phase across the consonant release: stops of this group can probably be connected to the "voiced aspirated" already heard by E. Sievers and A. Pisowicz. The main acoustic cue distinguishing the voiceless aspirate series and the voiced one is in the spectral tilt of the following vowel (50 msec after the release of the consonant): all the parameters are significant, apart from the difference between A3 and H1, which in one speaker (speaker 9) presents inverted values in comparison to the others.18 Data are summarized in Table 4, collected from the performance of the 6 minimal pairs already listed in (7b) This observation is consistent with the traditional classification of Gyumri dialect (and the rest of Širak dialects) in the western group, on the basis of morphological features (see, e.g. Adjarian 1909: 45-46). In this variety, the first step in the western consonantal shift -a strong devoicing of the voiced series -is attested. As a result, in both the Aštarak and Gyumri varieties, the distinction between voiceless and voiced series is mostly a matter of a difference between stiff vs. slack voice in the first vocalic portion immediately following the consonant.

Phonological analysis

In order to provide a phonological analysis of the phonetic observations illustrated above, we may distinguish between two different dimensions: each may be considered as an equipollent opposition. The first dimension regards the difference between aspirated, plain and ejective (glottalized) consonants, and may be treated through the [spread glottis] and [constricted glottis] features in the following way (cf. Halle & Stevens 1971;Stevens 1977;Kenstowicz 1994: 38- This dimension seems to be responsible for the opposition between the plain voiceless series and the voiceless aspirated series of the Armenian varieties, which, as we illustrated in § 2.1, is phonetically implemented by VOT. Therefore we may identify the following specifications for Eastern Armenian: This phonological analysis can be proposed not only for the modern varieties, but for Old Armenian as well (cf. Garrett 1991: 797-98;: the [spread glottis] feature can be used for marking the opposition between aspirated and plain stops, and either the [stiff vocal folds] feature or the [slack vocal folds] feature can be seen as responsible for the opposition between plain voiceless and the traditionally called "voiced" stops. Therefore, the three series of stops can be specified, for the oldest literary language, in two alternative sets: The difference between (13a) and (13b) could be viewed as merely nominalistic; but the two solutions differ in one point: they offer an opposite solution on which series (between plain voiceless and "voiced") must be considered as the marked one (and thus specified by the sign +), and which is the neutral one. Therefore, the final decision on which of the two representations is preferable can be taken after a more general consideration of the status of plain voiceless and "voiced" stops in Armenian phonology. The analysis in (13) has two advantages. First of all, it allows a straightforward phonological and phonetic derivation of all the three dialects analyzed, which represent as many historical evolutions of Old Armenian: and one of them, the Gyumri dialect, belongs to the Western Armenian varieties. Therefore the schemas in (13) can represent the original stage from which all the modern dialects may be derived.

As a second goal, the analysis in (13) is able to solve the Bolognesi paradox illustrated in § 1.

Armenian is the only Indo-European group, beside Indo-Aryan, in which the reflexes of the traditionally called "voiced aspirates" of PIE share three crucial properties: they are clearly distinct from the reflexes of PIE voiced stops; they are not voiceless but can be associated with some kind of laryngeal activity; in both groups, the third PIE series can be associated to "slack" or "murmured" voice quality.20

First of all, the voiced aspirated stops are actually murmured (breathy) in many modern Indo-Aryan languages, like Hindi, Marathi and Bengali; they are realized with a breathy voice vibration throughout the closure of the consonant, although a clear aspiration phase is observable across the consonantal release.21

In Gujarati, vowels adjacent to "voiced aspirated" are murmured and have the following general properties: they have a sharper spectral tilt than "clear" vowels; when they follow the voiced aspirated consonant, they have a small tone dip in the beginning and an increase in pitch throughout the development of the vowel; they sometimes have a longer duration than "clear" vowels, as is usual for non-modal vowels contrasting with modal ones.22

Tonal properties associated with originally "voiced aspirated" consonants can be better observed in many Indo-Aryan varieties, where such a series of stops can merge with other series (either the voiceless or the voiced one), while its properties were transferred to adjacent vowels in terms of tonal features: in the most widespread pattern, the vowels following the consonant develop a rising tone, and the vowels preceding it develop a falling tone. In such a process the depressive consonants (i.e. causing a fall in pitch of the adjacent vowel portions) are not simply voiced, as is observable in many tonal languages, but associated with breathy or 20 For a phonological comparison of Armenian and Indo-Aryan reflexes of PIE third series, see, among others, Benveniste 1959;Garrett 1991: 797-98; Hombert, Ohala & Ewan 1979: 47-48;Yip 2002: 36-38), at least at some stage of previous historical derivation. This evolution is attested, in the north-western area, in Panjabi (included the northern side of the so-called Lahndi varieties), where Indo-Aryan "voiced aspirated" series is merged with voiceless, and in many Dardic languages (with different patterns); the same process can be found in some varieties of Bengali for the eastern area, with merging of "voiced aspirated" series and voiced one.23 All these facts allow a reconstruction of a Proto-Indo-Aryan consonant system with a contrast between a series of voiced and a series of murmured stops, apart from the relevance of the [spread glottis] feature at this stage of derivation and for this opposition.24 In such a case, the basis of comparison for a series of voiced aspirated consonants in Proto-Indo-European becomes weak; instead the concordance of Indo-Aryan and Armenian offers an argument for reconstructing the opposition between the traditional PIE "voiced" and "voiced aspirated" series in terms of vocal tenseness, and specifies it, in phonological terms, either with the [stiff vocal folds] feature, or with the [slack vocal folds] feature.

It must be observed that such a solution differs from the glottalic theory in a crucial point: for the glottalic theory, the series of the traditional "voiced" is marked in terms of glottis adduction, and not vocal tenseness.25 Therefore, it is consistent with the use of the [constricted glottis] feature.

23 For a description of the phenomenon in an Indo-Aryan context , Elizarenkova 1990: 151-54;Masica 1991: 102; for Panjabi and so-called Lahndi (on this denomination see Rossi 1974), see Bahl 1957;Bahl 1969: 160-61;Wells & Roach 1980;Shackle 1980;Shackle 1994 Pal 1965. 24 For a more extensive analysis of Proto-Indo-Aryan phonology, we should consider the restructuring of the system in four different series of stops, with a series of voiceless aspirates progressively increasing. This is the stage of derivation already attested by Sanskrit, where voiced and voiceless aspirates seem to form, at least in some processes, a natural class; for a debate on this topic, cf. Joseph & Janda 1988;Janda & Joseph 2002, and literature cited therein. 25 Gamkrelidze & Ivanov 1973;Hopper 1973;Haudricourt 1975;Kortlandt 1978;Gamkrelidze & Ivanov 1995: 5-70. For further Armenian counter-evidence to the glottalic theory, see Pisowicz 1989: 217-20;Lamberterie 1994: 151-54;Olsen 1999: XL.

The proposal developed here is similar to the observations by James Clackson (2007: 48), who already recalled the continuum of voicing types discussed by Peter Ladefoged and Ian Maddieson (1996: 49): 14 breathy slack modal stiff creaky However, two difficulties can be found in the full applicability of this model to Proto-Indo-European: here vocal tenseness and glottis constriction are plotted in the same dimension; so creaky voice (and any kind of glottalization) is considered an increase of stiff voice. But creaky and glottalized voice is usually associated with a fall in pitch, while stiff (or tense) voice causes a rise in pitch, as we can see in Armenian and Indo-Aryan.26 It seems preferable to distinguish between two different dimensions: the first one concerns glottis spreading or constriction, and is responsible for aspiration; the second is connected with vocal tenseness, and is reflected in the difference between slack and stiff voice.

Conclusions

Comparison between Armenian and Indo-Aryan phonology leads to the following hypothesis: the Proto-Indo-European consonants traditionally called "voiced aspirates" may be characterized in terms of vocal tenseness; they may be considered as specified, within the consonant system, with the [+slack vocal folds] or [-stiff vocal folds] features. This hypothesis also has the advantage of better explaining the evolution of PIE "voiced aspirates" as voiceless in many Indo-European groups where the opposition between PIE "voiced" and "voiced aspirated" series was not dephonologized: not only in Modern Western Armenian (characterized by such evolution of Old Armenian "voiced stops") and in many modern Indo-Aryan languages (like Panjabi), but also in Ancient Greek and, at least in many contexts, in Latin and Italic.

Such arguments can motivate further research to investigate the entire phonological system more extensively; to decide which specification -26 For a discussion on this topic, see Henderson 1977, Gordon & Ladefoged 2001 or [-stiff v.f.] -can be more adequate; and to find other evidence within Indo-European comparative phonology.

Table 3

Table 4