Armenian (Yerevan Eastern Armenian and Beirut Western Armenian)

Armenian (or , /hɑjeˈɾen/, ISO 639-1 hy) comprises an independent branch of the Indo-European language family.1 Its earliest attested ancestor is Classical Armenian in the fifth century CE (see Godel 1975; Thomson 1989; DeLisi 2015; Macak 2016). Modern Armenian is classified into two dialect families: Eastern Armenian (ISO 639-3 hye) and Western Armenian (ISO 639-3 hyw). Eastern Armenian is spoken in modern-day Armenia, and large speaker communities also exist in Georgia, Russia and Iran (shown in Figure 1). Western Armenian was historically spoken in the Ottoman Empire, but now includes varieties spoken throughout the Armenian diaspora in the Middle East, Europe, and the Americas (Donabédian 2018). There are substantial Western Armenian speaker communities in Turkey (Istanbul), Lebanon (Beirut), Syria (Aleppo, Damascus), California (Fresno, Los Angeles County), France (Marseilles), Australia (Sydney) and Argentina (Buenos Aires). There are also recent diaspora communities of Eastern Armenian speakers in California (Karapetian 2014), as well as communities of Western Armenian speakers in Armenia who escaped the Armenian genocide during World War I, who repatriated after World War II, or who fled the ongoing Syrian civil war. UNESCO lists Western Armenian as an endangered language in Turkey, and there are significant language promotion efforts in many diaspora communities that are intended to combat declining use by speaker generations born in the Americas and Europe (Al-Bataineh 2015; Chahinian & Bakalian 2016).


Introduction
Armenian (հայերէն or հայերեն, /hAje»Ren/, ISO 639-1 hy) comprises an independent branch of the Indo-European language family.1 Its earliest attested ancestor is Classical Armenian in the fifth century CE (see Godel 1975;Thomson 1989;DeLisi 2015;Macak 2016).Modern Armenian is classified into two dialect families: Eastern Armenian (ISO 639-3 hye) and Western Armenian (ISO 639-3 hyw).Eastern Armenian is spoken in modernday Armenia, and large speaker communities also exist in Georgia, Russia and Iran (shown in Figure 1).Western Armenian was historically spoken in the Ottoman Empire, but now includes varieties spoken throughout the Armenian diaspora in the Middle East, Europe, and the Americas (Donabédian 2018).There are substantial Western Armenian speaker communities in Turkey (Istanbul), Lebanon (Beirut), Syria (Aleppo, Damascus), California (Fresno, Los Angeles County), France (Marseilles), Australia (Sydney) and Argentina (Buenos Aires).There are also recent diaspora communities of Eastern Armenian speakers in California (Karapetian 2014), as well as communities of Western Armenian speakers in Armenia who escaped the Armenian genocide during World War I, who repatriated after World War II, or who fled the ongoing Syrian civil war.UNESCO lists Western Armenian as an endangered language in Turkey, and there are significant language promotion efforts in many diaspora communities that are intended to combat declining use by speaker generations born in the Americas and Europe (Al-Bataineh 2015; Chahinian & Bakalian 2016).
Each of the two dialect families has dozens of documented varieties (Adjarian 1909;Vaux 1998:ch1.1;Sayeed & Vaux 2017). 2,3Instrumental phonetic research on the Armenian languages began with Adjarian 1899, which first proposed the concept of voice onset time to differentiate plosive voicing contrasts prior to its independent development in Lisker & Abramson 1964(Khachatryan & Airapetyan 1987;Braun 2013).Xačatryan 1988 provides a detailed phonetic description of Eastern Armenian including acoustic measurements, palatography, tracings of mid-sagittal X-ray images of the vocal tract, and discussion of perceptual experiments.Khachatryan & Ayrapetyan 1971 report instrumental articulatory and acoustic descriptions of Eastern Armenian consonants.Both works focus on a high-register literary variety spoken by students and broadcast announcers.Fairbanks 1948, Johnson 1954, and Dum-Tragut 2009 are general linguistic grammars of standardized Eastern and Western dialects with substantial phonetic material, and Allen 1950 is a phonetic description of an Eastern Armenian speaker who had grown up in Iran.
This illustration describes and compares the phonetics of two Armenian varieties: the Western variety spoken in Beirut, and the Eastern variety spoken in modern Yerevan and surrounding regions.The Beirut Western Armenian recordings were made by Hossep Dolatian (HD), a twenty-eight-year-old male speaker who was born in Beirut, Lebanon.He grew up in an Armenian-speaking community in Beirut and moved to the USA at the age of twentyone.In addition to Western Armenian, he is fluent in English and has advanced proficiency in Arabic.His recordings were made with a stand-mounted MXL 770 microphone in a quiet environment.The Yerevan Eastern Armenian recordings were made by Susanna Khechoyan (SK), a fifty-six-year-old female speaker who was born in Artsvashen.She lived in Yerevan, Armenia from the age of six up to thirty-eight, and has lived in Los Angeles since then.She grew up speaking Eastern Armenian and continues to use it daily, and also speaks English fluently and has advanced proficiency in Russian.She taught Armenian as an elementary school teacher in Armenia for twelve years.Her recordings were made with a head-mounted Shure SM10A microphone in a sound isolation booth.
One coauthor, Tabita Toparlak (TT), grew up speaking Western Armenian and Turkish in Istanbul and also speaks French and English.Another coauthor, Peter Guekguezian (PG), identifies as a heritage speaker of Western Armenian with intermediate proficiency and grew up in the western USA speaking English as a first language.When we had metalinguistic questions about pronunciation, we also discussed them with several other Eastern and Western Armenian speakers that we knew in the USA, Canada, Istanbul, Yerevan and Beirut.The speakers who contributed recordings, judgments and commentary to this illustration are Armenian language instructors, linguistics researchers and educated professionals in other fields.As such, the Armenian varieties that we describe are high-register varieties.In particular, HD describes the variety in his contributed recordings as a standardized variety of Western Armenian as it is spoken in Beirut,4 though his illustrated plosive pronunciations differ from more Arabic-dominant Western Armenian speakers still living in Beirut (discussed in Beirut Armenian plosives; see also Godson 2004;Kelly & Keshishian 2021;Tahtadjian 2021).The consonant inventory of the two Armenian varieties is given in the 'Consonants' table above (see also Xačatryan 1988: 85;Vaux 1998: 16;Dum-Tragut, 2009: 13).Consonants in parentheses are found only in Yerevan Armenian.Yerevan Armenian and most Eastern varieties have a three-way laryngeal contrast for plosives and affricates (voiced, voiceless unaspirated and voiceless aspirated), while Beirut Armenian and most Western varieties make only a two-way distinction for these sounds (Vaux 1998: 7ff).Yerevan Armenian also has a phonemic trill, which is merged to the tap in Beirut Armenian. 5he Yerevan voiceless unaspirated plosives correspond with Beirut voiced ones, and the Yerevan voiced plosives correspond with Beirut aspirated ones (Baronian 2017).These correspondences are shown below for labial plosives in word-initial position.The orthography is given with the Soviet-era reformed system that is currently used in Armenia.In cases where this differs from the Classical orthography used by the Western Armenian diaspora, the Classical orthography is listed second.Like the plosives, the Yerevan voiceless unaspirated affricates correspond with Beirut voiced ones, but Yerevan voiceless aspirated and voiced affricates are both voiceless unaspirated in HD's Beirut Armenian variety.These correspondences are shown below for dental affricates in word-initial position.Plosives and affricates in recent loanwords do not respect these correspondences.In recent loanwords, these sounds are typically borrowed using the same pronunciation (voicing or aspiration) in both dialects, and are spelled differently in each dialect.For example, the borrowed city name /bej»RutH/ 'Beirut' has a word-initial voiced plosive in both Yerevan and Beirut Armenian, but is spelled Բեյրութ in the Eastern orthography and Պէյրութ in the Western orthography.The accompanying recording of the Yerevan form was contributed by Vahagn Petrosyan (VP), a thirty-five-year-old male speaker who has lived in Yerevan since birth.

Affricate correspondences in Yerevan and Beirut Armenian
The consonant inventory is illustrated in the table below.While almost all of the words in the table exist in both dialects, the dialectal forms that do not illustrate the target sound in each row are omitted.

Yerevan Armenian plosives
Yerevan Armenian has three labial, three dental, and three velar plosives that contrast in both onset and coda position: With respect to their distribution, final /b/ is very rare and attested mostly in loanwords, such as /»StAb/ 'headquarters' borrowed from Russian xtab (originally German Stab). 6The Armenian orthography does not reliably index final voicing due to diachronic change and perhaps cross-dialectal borrowings. 7For example, the form ձագ 'young of an animal' (in table 6 See also Syllable structure on the pronunciation of sibilant-initial consonant clusters in Yerevan Eastern compared to Beirut Western Armenian.Prothetic schwa is variable in Yerevan Armenian and may be uncommon especially in Russian loanwords such as շտաբ 'headquarters'.This word is not used in Beirut Armenian but HD suggests that it might be pronounced /´S»tHApH/ with an invariable initial schwa. 7The Armenian alphabet was first developed around the fifth century CE.Although it has been modified since then, the spelling of Armenian words does not perfectly correspond with modern pronunciation, particularly with regard to obstruent voicing and aspiration (Johnson 1954: 5;Vaux 1998: 17-18, 237;Dum-Tragut 2009: 24-27).Affricate correspondences in Yerevan and Beirut Armenian) is written with final <գ>.This letter typically corresponds to /g/ in Yerevan Armenian and /kH/ in Beirut Armenian, but this word is pronounced /»dÉ zAkH/ in Yerevan and /» tÉ sAkH/ in Beirut with the same final voiceless aspirated plosive in both varieties.Some Yerevan Armenian speakers report a (lamino-)alveolar pronunciation of the dental /d, t, tH/ plosives, such as for the accompanying recording of /tH´»tHu/ թթու 'sour' contributed by speaker VP.The velar /k/ plosive can be placed farther back toward [k= , q].This is illustrated in the accompanying recording of /»ko"/ կող 'rib' pronounced [k= o"], which was contributed by a twenty-year-old female speaker of Eastern Armenian who grew up in Yerevan.This pronunciation is optional, and the X-ray tracings in Xačatryan 1988 show a clearly velar [k] between two /A/ vowels.
Figure 2 shows spectrograms and waveforms of SK's recordings of the labial plosives in word-initial (upper row) and word-final (lower row) positions.Figure 3 shows the mean and standard deviation of closure duration, closure voicing duration, and aspiration duration for all nine plosives in each position.These measurements are from eight speakers who grew up in Yerevan, each reading aloud 155 unique words in a carrier phrase with nuclear accent (collected in Seyfarth & Garellek 2018).
The voiced series /b, d, g/ typically has breathy voicing during the closure and could be closely transcribed as [b-, d-, g_ ] (Macak 2017; Seyfarth & Garellek 2018).Seyfarth & Garellek (2018) report that 15% of word-initial /b, d, g/ tokens lack voicing during the closure; the accompanying recording of /gA»Ri/ գարի 'barley' contributed by SK is an example of such a token.In syllable onset position, the breathy voice quality can be measured in the initial portion of the following vowel as a louder first harmonic (2-3 dB) and lower cepstral peak prominence (1 dB) relative to the voiceless unaspirated plosives (Schirru 2012;Seyfarth & Garellek 2018).These measures index a more open glottis (Klatt & Klatt 1990; However, Khachaturian (1983Khachaturian ( , 1992) ) indicates that speakers of some other regional Eastern Armenian varieties produce more extended and audible aspiration after voiced plosive onsets than in the Yerevan variety.In syllable coda position, voiced plosives more often have short post-aspiration (< 50 ms in citation forms) in addition to reliable closure voicing and a longer nucleus vowel (Adjarian 1899;Lisker & Abramson 1964;Xačatryan 1988;Hacopian 2003;Seyfarth & Garellek 2018).
The other voiceless series /pH, tH, kH/ is defined by post-aspiration in both syllable onset and coda positions, along with a correspondingly higher f0 during the initial portion of a following vowel in onset position (Allen 1950;Lisker & Abramson 1964;Khachatryan & Airapetyan 1987;Xačatryan 1988;Hacopian 2003;Schirru 2012;Toparlak 2017;Seyfarth & Garellek 2018).Yerevan Armenian is thus typologically rare in that it maintains a threeway voicing and aspiration contrast in both initial and final position using virtually the same acoustic cues in both positions (Hacopian 2003).

Beirut Armenian plosives
Western Armenian is described as having a two-way contrast between voiceless aspirated and voiced plosives (Fairbanks 1948;Vaux 1998;Baronian 2017).
ˈɡɑɾ կար 'seam' ˈkʰɑɾ քար 'rock' Among Western Armenian speakers in Anglophone Canada and the USA, the voiceless plosives are aspirated, and the other plosive series varies phonetically between voiceless unaspirated and voiced (Kelly & Keshishian 2021;Tahtadjian & Kochetev 2021;Tahtadjian 2021; also compare to Balabanian 2020 with Armenian-English-French speakers in Quebec).This pattern is similar to North American English, though closure voicing may be somewhat more common in North American Western Armenian than English (compare to e.g., Lisker & Abramson 1964: 395;Schertz 2013: 254;Davidson 2016).
Among speakers of Western Armenian in Lebanon, the plosive voicing contrast is between voiceless unaspirated and voiced plosives (Kelly & Keshishian 2019;Tahtadjian 2021).This pattern corresponds to what has been described for Lebanese Arabic stops (Bellem 2014;Al-Tamimi & Khattab 2018).As Lebanese Arabic was the most dominant non-Armenian language for the speakers in Kelly & Keshishian 2019 and Tahtadjian 2021, it is possible that the phonetic contrast used by these speakers is influenced by their multilingualism.
HD grew up in Beirut, Lebanon but has lived in the USA for the past eight years at the time of recording.His Armenian plosives currently follow the Western Armenian pattern reported for speakers in Anglophone Canada and the USA: one series is voiceless aspirated in most environments while the other can be either voiced or plain voiceless.Figure 4 shows the mean and standard deviation of closure duration, closure voicing duration, and aspiration duration for all six plosives in prevocalic word-initial (upper row) and postvocalic wordfinal (lower row) position.In this intervocalic context, HD consistently produces the voiced series with closure voicing, but in other environments he often produces these plosives as voiceless unaspirated, especially in utterance-initial position.For example, the initial sound of the accompanying recording of /bARdkH/ [pA® 8 3 tkH] պարտք 'debt' is voiceless (see also Figure 9 in Syllable structure).
In HD's Beirut Armenian variety, the voiceless plosives are not aspirated adjacent to voiceless sibilants (see Fairbanks 1948: 4-5), with word-medial examples shown below: Word-final voiceless plosives are also unaspirated after voiceless sibilants, though utterance-final plosives are typically still aspirated.This is illustrated in utterance-final Aspiration is variable in sequences of two plosives, where the first plosive is often unaspirated even when it has a definite release, such as in /jeRekHSApH»tHi/ [jeRekSAp»tHi] երեքշաբթի 'Tuesday' in the table above (see Fairbanks 1948: 4-5).However, when a sequence of two voiceless plosives has the same place of articulation, it is consistently pronounced as a long aspirated plosive and not rearticulated.This is illustrated in the form /pHAtHtH´»vAdÉ z/ [pHAtH˘´»vAdÉ z] փաթթուած 'wrapped' as compared with the borrowing /pHAtHA»tHes/ [pHAtHA»tHes] փաթաթէս 'potato'.
With respect to place of articulation, we perceived the Western Armenian speakers in Lebanon recorded by Kelly & Keshishian (2021) as using a dental articulation for the coronal plosives.Impressionistically, some of the speakers in the USA pronounced dental plosives while others pronounced alveolar ones.
Figure 5 illustrates the closure, sibilant frication and aspiration intervals of the three dental affricates in a carrier phrase for Yerevan Armenian: 'your (way of) saying "tree"' 'your (way of) saying "pain"' 'your (way of) saying "young of an animal"' Although all three affricates show some carry-over voicing from the preceding vowel, it is more robust for the voiced affricate in the right panel of Figure 5.The closure is longer in the unaspirated affricate than in the aspirated one (Khachatryan & Ayrapetyan 1971;Xačatryan 1988).For the aspirated affricate (center panel), the aspiration interval begins at the 0.2-second mark.There is not always such a clear boundary between the sibilant frication and the aspiration in the aspirated affricates.However, the interval of combined frication and aspiration in the aspirated affricates is reliably longer than the frication in the unaspirated affricates (Khachatryan & Ayrapetyan 1971;see Xačatryan 1988: 139 for perceptual evidence).Khachatryan & Airapetyan (1987) report substantially more overlap in the distributions of voicing lag time for the voiceless aspirated and unaspirated affricates than we observed, but they also show that the voiceless aspirated and unaspirated affricates can nevertheless be distinguished by a linear combination of voicing lag time and the intensity of noise above 2 kHz after the closure.
Beirut Armenian has a two-way voicing contrast for both affricate places.The broader Western affricate contrast is commonly reported as voiceless aspirated and voiced, matching the Western plosive contrast (Vaux 1998, in contrast with Fairbanks 1948).Because HD's voiceless affricate pronunciations are unaspirated (shown in Figure 6), which is different than his plosive pronunciations, we categorize the voiceless affricates in his Western variety as /tÉ s, tÉ S/. Kelly & Keshishian (2019) also found that the voiceless affricate series is unaspirated for Lebanese Western speakers more broadly.
The dental series /dÉ z, tÉ s, tÉ sH/ (/dÉ z, tÉ s/ for Beirut Armenian) may be lamino-alveolar apico-dental, with the tongue tip incidentally contacting the upper teeth.Based on palatography and X-ray images, Xačatryan (1988: 170) and Khachatryan & Ayrapetyan (1971: 296) describe these affricates as apical and post-dental, having gingival contact which differs from the greater dental contact in the plosives.Other descriptions have labeled these affricates as alveolar (Fairbanks 1948;Dum-Tragut 2009) and dental (Allen 1950;Johnson 1954).Our speakers express disagreement and uncertainty as to whether the tongue usually makes contact with the back of the upper teeth.The X-ray tracings in Xačatryan 1988 do show dental contact during the affricate closure (Figures 29-30).In recordings of connected speech, we perceive these sounds as having a dental closure after vowels when the frication release is removed from the audio.
The sibilants /s, z/ are described as 'dental' (Allen 1950;Khachatryan & Ayrapetyan 1971), 'post-dental' (Johnson 1954: 26;Xačatryan 1988: 85, 110), or 'alveolar' (Fairbanks 1948;Dum-Tragut 2009).SK, TT, and a Yerevan speaker report that the tongue tip is held behind the upper teeth, while HD and three Yerevan speakers report that the tongue tip touches the back of the lower teeth during these sounds.The sagittal X-ray tracings for /s, z/ in Xačatryan 1988 show a lamino-alveolar place of articulation with an inconsistent tongue tip position (Figures 23-24, as described on p. 178).
Figure 7 shows the log power spectral density for the voiceless fricatives and affricates for SK and HD.For /s/ and /tÉ s/, SK has a high spectral peak at roughly 9 kHz (as in Khachatryan & Ayrapetyan 1971: 302;Xačatryan 1988: 178) which may index an apico-dental articulation (cf.lower peak for alveolar /s/ in Jongman, Wayland, & Wong 2000;Maniwa, Jongman, & Wade 2009).HD has a lower spectral peak for /s/ around 7 kHz, consistent with a laminoalveolar but not apico-dental articulation.For /S/ and /tÉ S/, both speakers have a relatively low peak around 3 kHz (as in Khachatryan & Ayrapetyan 1971: 304), and SK has a second mode at 9 kHz.Each panel shows a long-term average spectrum which is calculated over the center 30 milliseconds of sibilant energy (for fricatives), or over the 30 milliseconds immediately following the closure release (for affricates), omitting the release transient if one is present.Measurements are taken from the illustrative recordings accompanying the consonant table above.
Voiceless labiodental /f/ (grapheme ֆ) is rare and occurs primarily in borrowed words, though many borrowings date to the twelfth century or earlier.Dum-Tragut (2009: 20) transcribes the voiced labiodental /v/ as the approximant [V] when it occurs in word-initial position before /o/, as well as in medial and final position after /e/.Our speakers have the voiced fricative [v] in all positions.Xačatryan (1988: 176) measures a wider labiodental constriction for /AvA/ compared to /AfA/ from X-ray tracings (ibid., Figures 21-22).This may be a categorical manner difference ([f] vs. [V]), or the wider /v/ constriction may simply facilitate voicing, which is unnecessary for voiceless /f/.
The examples listed below illustrate the two sounds adjacent to the near-back vowel /A/ and the front vowel /e/ in syllable onsets and codas: The voiced back continuant varies between the fricative ["] and the approximant ["4 ].The preceding Beirut Armenian recording of /"AzA»Ros/ Ղազարոս 'Lazarus (name)', as well as the accompanying recording of Yerevan Armenian /s´»"el/ սղել 'to increase a price' contain approximant realizations.

Obstruent cluster devoicing
All obstruents participate in a regressive devoicing pattern when a cluster of two obstruents is formed by compounding, suffixation, or other morphological alternations.In both dialects, the first obstruent is devoiced if the second obstruent is voiceless (see Xačatryan 1988: 104;Vaux 1998: 17-18).For example, the suffix /-tÉ sHi/ (Yerevan), /-tÉ si/ (Beirut) -ցի is used to form demonyms (rows 1-2 of the table below).When this voiceless-initial suffix follows a word such as /pHA»Riz/ Փարիզ 'Paris' that has a final voiced obstruent, the stem-final obstruent is voiceless (rows 3-4).The Yerevan Armenian recordings in this subsection were contributed by VP.
'America' 'American' 'Paris' In addition to this regressive devoicing pattern, Beirut Armenian also has a progressive devoicing pattern: if the first obstruent is voiceless, then the second obstruent is devoiced.For example, the Beirut Armenian suffix /-god/ -կոտ is used to form adjectives (following a stem-final sonorant in row 1, and a voiced obstruent in rows 2-3 of the table below).When it combines with a word that has a final voiceless obstruent, such as /»vAX/ վախ 'fear', the suffix-initial obstruent is voiceless (rows 4-5).

Beirut
In Yerevan Armenian, we observed optional progressive devoicing of /v/ following a voiceless obstruent.For example, when the diminutive suffix /-ik/ -իկ combines with a word ending in /u/, the final syllable is pronounced /-vik/ (rows 1-2 of the table below; see Syllable structure on vowel hiatus).However, the syllable onset varies between [v] and [f] when it follows a voiceless obstruent in the stem word (rows 3-4 below).

Yerevan
VP indicates that both the voiced [v] and devoiced [f] pronunciations are natural in թթվիկ , as well as in the /v/-initial passive suffix when it follows a voiceless obstruent (not illustrated here).However, other than obstruent sequences involving /v/, the morphophonological context where progressive devoicing could potentially occur is very rare in Yerevan Armenian, and it is thus not clear whether progressive devoicing is generally optional, obligatory, or idiosyncratic in this dialect.
The rhotics /r, R/ are contrastive in Yerevan Armenian, with minimal pairs such as the following: The tap is often spirantized in syllable codas (Fairbanks 1948: 8;Allen 1950: 195;Xačatryan 1988: 186;Toparlak 2017: Appendix 2).We observed this especially in citation forms and in emphatic speech. 8For example, SK's recordings contrast /»seR/ and /»ser/ primarily via louder frication noise for the tap /R/ with a low spectral peak around 4 kHz (perhaps non-sibilant postalveolar fricative [® 8 3 ])9 .However, word-final /R/ is often unambiguously tapped or approximated in connected speech, such as in the accompanying recordings of the North Wind and Sun by HD and SK (see Transcription of recorded passage).Tap and approximant variants are both common in intervocalic position.The recordings of /kHA»Ri/ քարի 'rock (DAT.GEN)' and /XA»RisX/ խարիսխ 'anchor' contributed by SK illustrate approximant variants of /R/.
The recording of /»ser/ is not trilled, but has a falling third formant that is excited by aspiration without voicing, and only weak or absent oral frication noise.Most of the coda trills in the accompanying recordings contributed by SK are voiceless and/or spirantized.For example, the recording of /» tÉ sAr/ ծառ 'tree' in the section Vowels (below) includes a voiceless trill, while the recording of the same word in the Consonants table is more similar to the variant in /»ser/.A voiced trill coda is illustrated in the recording of /»kor/ կոռ 'coerced labor (archaic)', from the speaker who contributed /»ko"/ կող 'rib' in the subsection Yerevan Armenian plosives.
Although the two rhotics are contrastive in Yerevan Armenian, the tap is sometimes trilled before coronals (Xačatryan 1988: 108;Vaux 1998: 19; also noted by speaker VP), and wordinitial trills are often reduced to the tap. 10n Western Armenian, both sounds are merged to the tap /R/ in all environments (Vaux 1998), though they are prescriptively taught as contrastive in Canadian Armenian language schools (Talia Tahtadjian, p.c.) and some Western Armenian dictionaries (Sak'apetoyean 2011; also in Fairbanks 1948).For the Western Armenian community in Canada, Tahtadjian (2020) reports that the Western Armenian tap and trill are acoustically distinguishable: the Western trill sometimes has multiple articulator contacts (occurring in about 30% of onset trills produced by older speakers, and <15% of other trills), and the trill is about 2 ms longer in onset position and 4 ms longer in coda position.These differences are probably too small or too variable to reliably index a category difference.In the accompanying recording of Beirut Armenian [rAzmA»gAn] ռազմական 'military' contributed by HD, the onset consonant has multiple contacts, but HD indicates that the tap and trill were perceptually indistinguishable for his peer group in a Lebanese Armenian school.
In both dialects, the two rhotics are rare word-initially.While there are some native rhoticinitial words such as Yerevan /rAmkA»kAn/ ռամկական 'democratic', most such words are names like Yerevan /rAfA»jel/ Ռաֆայել 'Raphael (name)' and borrowings such as Yerevan /rA»bi/ ռաբբի 'rabbi'.For these words, the Beirut forms use a tap instead of the trill: /RAmgA»gAn/, /RAfA»jel/ (Ռաֆայէլ ), /RA»pHi/.Both rhotics are more frequent medially and finally, and occur in Yerevan minimal pairs such as the above.

Word-initial vowels
Toparlak ( 2019) measured vowels that were elicited from six speakers each of Yerevan Armenian and Beirut Armenian, including three female and three male speakers per variety.All speakers were aged 21-40 and living in Paris, and Armenian was their dominant language.The speakers each read aloud thirty-two dialect-appropriate sentences containing target words with initial, medial, and final vowels, and then repeated the target words in isolation.The sentences and isolated words were repeated multiple times by each speaker for a total of approximately 5,800 measured vowel tokens.Figure 8 shows a graphical summary of formant measurements for the six Armenian vowels, with values given in Table 1 (female speakers) and Table 2 (male speakers).The mid-front vowel /e/ varies acoustically between /e/ and /E/ and could be narrowly transcribed as [e4 ].As an example, the recording of /A"ju»se/ աղյուսե 'made of brick' contributed by SK has a final vowel closer to open-mid [E] while the corresponding vowel in the recording contributed by HD is closer to [e].On the other hand, the vowel in the recording of /»jet, »jed/ ետ 'backward' in Consonants contributed by HD is lower than in the corresponding recording by SK.Godson (2004) shows that this vowel is lower for Western Armenian speakers in southern California with increasing English dominance, closer to the California English open-mid front vowel.
The mid-back vowel is categorized as /ç/ by some grammars (e.g., Dum-Tragut 2009: 13), but it has a low F1 distribution that partially overlaps with the /u/ category (Toparlak 2019), which makes the broad symbol /o/ more appropriate.
Although all six vowels occur in all three environments, word-initial mid vowels are uncommon (see below on Schwa).Word-initial /e/ occurs in forms of the copula է as found throughout the Transcription of recorded passage as well in as a few other native words such as /»eS/ էշ 'donkey', but is mostly used in loanwords.In word-final position, most roots with final /o/ are loanwords such as Yerevan /k ( H ) i»lo/ կիլո or Beirut /kHi»lo/ քիլո, քիլոյ 'kilogram'.However, the native Armenian hypocoristic suffix /-o/ is used widely, as in /»mARo/ or /mA»Ro/ Մարո , Մարօ from /mAR»jAm/ Մարիամ 'Mariam (name)'.Schwa Khachaturian (1985) and Dum-Tragut (2009) treat the Armenian schwa as a phoneme, but Vaux (1998) and Allen (1950) analyze the schwa as purely epenthetic, and Hovhannisyan (2014: 89) argues that it is excrescent in some environments such as before final rhotics.Acoustically, schwa is mid-central for Yerevan Armenian, but might be transcribed as [´3 ] for Beirut Armenian (cf.lower F1 in Figure 2 and Tables 1-2; Toparlak 2019; see also Gordon et al. 2012 andSeyfarth &Garellek 2018).The two different acoustic distributions suggest that it has an acoustic or articulatory target in at least one of the dialects, in at least some environments.
In monosyllabic non-onomatopoeic free-standing words, schwas are unattested except for the name of the alphabet letter /»´tH/ ը.Schwas are also used in initialisms, in which consonant letters are pronounced as the corresponding consonant sound plus schwa, as in /i.i»h´/ ԻԻՀ 'I.I.H., the Islamic Republic of Iran'.Polysyllabic words generally have at least one other non-schwa vowel with word-level stress (see Word-level prosody below).A few polysyllabic words have only schwa vowels, but many of these are onomatopoeic or nativized loanwords, such as /f´s»t´X/ ֆստըխ 'pistachio' (cf.Turkish /fµstµk/ fıstık).
Schwa often appears in morphophonological alternations involving stress changes or consonant clusters (Vaux 1998(Vaux , 2003)).For example, Yerevan /tÉ s´»rel/, Beirut /dÉ z´»Rel/ ծռել 'to lean' is a verb form of the adjective /» tÉ sur/, /»dÉ zuR/ ծուռ 'crooked' in which the stressed high vowel alternates with an unstressed schwa.The first-and second-person possessor suffixes /-s/, /-t/ (Yerevan), /-tH/ (Beirut) appear directly after vowel-final stems, but are preceded by a schwa when they are used with consonant-final stems.The presence of schwa in these alternations is not predictable from the phonology alone (Dolatian 2021): high vowels do not generally alternate with schwa when they are unstressed (see the Vowels table for examples), and /-s/ can occur in other word-final coda clusters such as in /» tÉ sAXs/, /»dÉ zAXs/ ծախս 'cost' without schwa (see Syllable structure).Moreover, some free-standing forms like Yerevan /v´»kA/ վկա and Beirut /v´»gA/ վկայ 'witness' have no related forms without schwa (Vaux 1998), and our speakers indicate that it is impossible or unnatural to elide the schwa ( * vkA, * vgA) in isolation or in citation form.

Syllable structure
Armenian generally allows up to CjVCCkH syllables, with all elements optional except for V.
There are some exceptional complex onsets other than /Cj/ which are never pronounced with an intervening schwa, primarily in non-nativized loanwords ( Ġaragyowlyan 1974: 65).

Eastern Western
When vowel hiatus would occur between a stem-final vowel and a suffix-initial vowel, various repair strategies are attested (Vaux 1998: 27ff;Dolatian 2020: 35ff).Vowel repair strategies include [j]-insertion between vowels and /u/ devocalizing to [v].Different words and different vowel sequences show different types of vowel hiatus repairs.In some cases, multiple vowel hiatus repair strategies are possible for the same word.For example, the instrumental case form of the vowel-final stem /le»zu/ լեզու 'tongue' with the vowel-initial suffix /-ov/ -ով is variably ).The form with [j]-insertion is more typical in Western Armenian and preferred by HD, while the form with /u/ changing to [v] is more typical in Eastern Armenian and preferred by SK.
Words with initial sibilant-stop clusters are variably pronounced with an initial schwa, such as in [(´)stA»nAl] ստանալ 'receive'.In citation form, word-initial sibilant-stop clusters are obligatorily pronounced with an initial schwa in Western Armenian.Such schwas used to be obligatory in earlier stages of Eastern Armenian too ( Ġaragyowlyan 1974: 139ff), but the absence of a schwa in initial sibilant-stop clusters has now become more common, arguably due to contact with Russian (Dum-Tragut 2009: 31ff; see also Avetisyan 2011: 14).For both dialects, the presence of the schwa in word-initial sibilant-stop clusters varies in connected speech.For example, Ġaragyowlyan (1974: 143ff) reports that the presence of Eastern wordinitial schwa before sibilant-stop clusters depends on the preceding word's coda.A wordinitial invariant schwa is found before other consonant clusters in a handful of words, such as in Yerevan [´N»keR] and Beirut [´N»geR] ընկեր 'friend'.
A variety of coda clusters are found in word-medial and word-final positions.Almost all two-consonant coda clusters with falling sonority are attested without schwa epenthesis.However, nasal + obstruent clusters are homorganic, and coda /l/ generally does not occur in native coda clusters except after /j/.A few exceptional two-consonant coda clusters with rising or level sonority are attested, with examples below: The accompanying recording of Yerevan /pAte»RAzm/ is pronounced [pAte»RAsm8 ], and SK sometimes deletes the final /m/ in this word.The accompanying recording of Beirut /me»dAkHs/ is pronounced [me»dAks] with an unaspirated [k] due to the following sibilant (see Beirut Armenian plosives) though still without schwa.The orthography indicates other coda clusters with rising or level sonority, but many of these are pronounced with an intervening schwa (Vaux 1998: 26-27).Stem-final /-kH/ is an exception to these generalizations about coda clusters: it can occur after any singleton consonant or two-consonant cluster without any degree of schwa epenthesis, regardless of sonority, in both dialects (see Vaux 1998: 83; Dolatian 2021 for phonological analyses).Figure 9 shows spectrograms of /pARtkH, bARdkH/ [pA® 8 3 tkH, pA® 8 3 tkH] պարտք 'debt' from SK and HD, which each show three final consonant articulations -a spirantized /R/, a /t, d/ constriction and release, and a final aspirated /kH/ release -with no acoustic evidence for schwa. 11

Word-level prosody
Words in citation form are typically stressed on the final syllable, unless the final syllable nucleus is schwa.If the final syllable nucleus is schwa, then stress is on the last non-schwa vowel, which is typically in the penultimate syllable.The table below illustrates these stress generalizations for the word /tÉ SA»kAt, dÉ ZA»gAd/ ճակատ 'forehead' and several of its derived forms.In the forms with a full vowel in the final syllable, stress is word-final, no matter whether the word ends in an open syllable (row 2) or a closed syllable (rows 1, 3 and 5), or whether the word is unsuffixed (row 1), suffixed (rows 2-3) or a compound (row 5).The form /tÉ SA»kAt´t, dÉ ZA»gAd´tH/ ճակատդ (row 4) has a final schwa and penultimate stress.'is a forehead' ճակատ է ճակատ For words with only schwa nuclei, our speakers produce final stress: /f´s»t´x/ ֆստըխ 'pistachio'.Ač . aṙyan (1971: 194) reports that these words have final stress, while Vaux (1998: 133) reports initial stress.This disagreement may reflect inter-speaker variation or differences in (loan)word origin.There are very few words with only schwa nuclei in Armenian and many of these words are onomatopoeic, which tend to be exceptional cross-linguistically.
Some notable classes of exceptions to the general stress patterns include ordinal numbers, vocatives, and hypocoristics, as well as some common adverbs and a few other idiosyncratic words, suffixes, and clitics, though exceptional non-final stress can be variable (Vaux 1998:ch4;Dum-Tragut 2009).For example, the Yerevan Armenian clitic /el/ էլ 'also' can carry stress, while the Beirut cognate /Al/ ալ cannot, as in Yerevan /tÉ SAkAt »el, tÉ SA»kAt el/ էլ contributed by VP versus Beirut /dÉ ZA»gAd Al/ ճակատ ալ 'also a forehead'.
Exceptional stress also occurs in some morphological alternations, which is also dialectspecific.For example, in subjunctive past imperfective forms in HD's Beirut Western Armenian, stress exceptionally occurs on a theme vowel which precedes the tense-agreement suffix, regardless of where the theme vowel is in the word.This means that the Beirut Armenian subjunctive second-person singular form has stress on the penultimate syllable -which contains the theme vowel -even though the final syllable has a peripheral vowel that would normally be stressed (cf.rows 2-4 in the table below).In Yerevan Eastern Armenian, the final syllable (the tense-agreement suffix) receives expected stress in all inflected forms.

Yerevan
Beirut Additionally, in negative past perfective verb forms, the finite verb negation prefix չ is /tÉ S-, tÉ S´-/ in Western Armenian, and /tÉ SH-, tÉ SH´-/ in Eastern Armenian.The prefix has exceptional initial stress in Western Armenian, but is unstressed as expected in Eastern Armenian.
The recorded sentences that accompany the subsection Acoustics of stress and focus illustrate the contrast of Yerevan Eastern /tÉ SH(´)-kAr»tʰɑtsʰ ͡ / and Beirut Western /»t ͡ ʃə-gARtHAtÉ É s/ չկարդաց 'to read (3SG PST PRF NEG)'.The difference in stress correlates with the variable deletion of initial schwa in the Yerevan Eastern forms (final stress) but not the Beirut Western forms (initial stress).

Sentence-level intonation and focus
In Eastern Armenian, broad focus (neutral context) sentences can have either SVO or SOV word order (Samvelian, Faghiri & Khurshudyan 2023).In broad-focus SVO sentences in Eastern Armenian, each prenuclear prosodic constituent (i.e., S and V) has a rising -L H-contour aligned with its right edge.Each prenuclear constituent has a successively lower pitch, i.e., there is downstep or declination between successive H tone targets (Haghverdi 2016;Skopeteas 2021).Skopeteas (2021) also finds sentence-final lowering in Eastern Armenian, which he analyzes as an L-% boundary tone at the right edge of the intonational phrase.In Western Armenian, broad focus sentences have an unmarked SOV order, and Toparlak (2019) reports the same intonational patterns as Skopeteas (2021). 13Western Armenian listeners perceive nuclear accent on the pre-verbal element -usually the object -and the verb undergoes post-focal compression or deaccenting (Toparlak 2019;Toparlak & Dolatian 2022;Dolatian 2022).Skopeteas (2021) gives experimental evidence that narrow or contrastive focus is marked by a falling contour H * +L on the stressed syllable of the focused word. 14The H * toneindicated by maximal f0 -is aligned with the left edge of the stressed syllable, followed by the L tone, indicated by a steep fall starting either immediately after the H * or at the right edge of the prosodic word.For questions, Skopeteas (2021) finds a sharp rise L+H * on the stressed syllable of either the focused element in narrow focus questions or the verb in broad focus questions.After this rise, he finds post-focal deaccenting.Toparlak & Dolatian (2022) find similar results for unmarked, SOV sentences with narrow object focus in Western Armenian: a pitch rise on the focused word, followed by post-focal deaccenting.
Simple yes-no questions with the unmarked SOV order have a final pitch rise on the verb (H%) in both dialects, with no change in word order.Wh-questions have a pitch rise on the wh-word followed by post-focal deaccenting.In Eastern Armenian, wh-questions have a sentence-final fall (L%) while in Western Armenian these questions have a pitch rise (H%) instead (Fairbanks 1948: 29;Johnson 1954Johnson : 15, Ġowkasyan 1990;;Dum-Tragut 2009: 54ff;Toparlak & Dolatian 2022).

Acoustics of stress and focus
Previous research on Eastern Armenian reports the acoustic correlates of stress as duration, intensity, and/or f0 including a final pitch rise (T'oxmaxyan 1983: 119;Xačatryan 1988: 76-79), but most of this earlier work did not clearly differentiate word-level stress and intonation.Skopeteas (2021) reports that a word-final pitch rise is a cue to the right edge of prenuclear constituents in Yerevan Armenian (see also Sentence-level intonation and focus and Haghverdi 2016).Both final and non-final stressed syllables can host a nuclear pitch accent that is aligned earlier within the syllable than this right edge tone.
For Western Armenian, Athanasopoulou et al. (2017) report that word-level stress is cued only by mean f0 and f0 excursion.They find that word-level stress has similar acoustic cues regardless of whether stress is final or non-final (due to a final schwa nucleus).Using the same data, Vogel & Athanasopoulou (2018) also report that stress does not involve significant enhancement in duration.To measure a word-level stress and focus paradigm, we elicited four target word forms in the following two carrier contexts, adapted from Athanasopoulou et al. (2017)  The four target word forms are listed below: As described in Word-level prosody, stress is final in these forms except /tÉ SA»kAt´/, /dÉ ZA»gAd´/ ճակատը, which has a final schwa nucleus.The first and second row contrast final and non-final stress.The third row has a trisyllabic form with final stress, for comparison   with the non-final stress in the second row.The fourth row contrasts unstressed word-medial /A/ with the stressed word-medial syllable in the second row.Vowel measurements for these elicitations are shown in Figures 10-11.HD (Beirut Western Armenian) produced stressed /A/ vowels with longer duration, higher F1 and lower F2 (i.e., more peripheral).In the focus condition, stressed /A/ also had a higher f0 and a wider range.SK (Yerevan Eastern Armenian) produced stressed /A/ vowels with a similar effect on F2, but duration and f0 were affected only in the focus condition (see Skopeteas 2021).This suggests that duration may be an acoustic correlate of stress for HD's Beirut Armenian, while vowel space expansion may be a stress correlate for both dialects.Both speakers showed postfocal compression or deaccenting.In two of four tokens, word-final schwa had an f0 rise even though it followed the stressed syllable (Figure 13).As Figures 10-11 and 12-15 illustrate, the f0 of stressed syllables is higher under the focused condition than either unstressed or stressed syllables in the non-focused condition, except in SK's recording of bisyllabic /tÉ SA»kAt/.In our elicitations, all vowels were also longer in words under focus.

Figure 1 (
Figure 1 (Colour online) Map of the distribution of Armenian in the Southern Caucasus (CC-BY-SA 4.0 figure created by Wikimedia Commons user GalaxMaps, retrieved June 20, 2021).

Figure 2
Figure 2 Yerevan Armenian labial plosives in words produced in isolation.Upper row shows voiced, voiceless unaspirated, and voiceless aspirated plosives in word-initial position; lower row shows the same plosives in word-final position.All spectrograms are calculated with a 5-millisecond Gaussian window and a 2-millisecond window advance.

Figure 3
Figure 3 Mean (filled bars) and standard deviation (whiskers) for closure duration, closure voicing duration, and aspiration duration for the nine Yerevan Armenian plosives in word-initial and word-final position, based on measurements from eight speakers in Seyfarth & Garellek 2018.Each bar includes between 32-176 tokens.

Figure 4
Figure 4 Mean (filled bars) and standard deviation (whiskers) for closure duration, closure voicing duration, and aspiration duration for the six Beirut Armenian plosives in word-initial and word-final position.Measurements are from recordings of HD reading aloud ten unique words containing each plosive twice in one or two carrier phrases, collected and annotated using the procedure in Seyfarth & Garellek 2018.Each bar includes twenty tokens.

Figure 5
Figure 5 Yerevan Armenian dental affricates in a carrier phrase.Spectrograms and waveforms show a 300-millisecond excerpt beginning with a preceding /o/ vowel, then the target dental affricate indicated above the panel, and then the following /A/ vowel.

Figure 6
Figure 6 Beirut Armenian affricates in words produced in isolation.

Figure 8
Figure 8 Average first and second formant frequencies for the six vowels in Yerevan and Beirut Armenian, based on measurements from three female and three male speakers per language variety in Toparlak 2019.Ellipses cover the central 50% of the observations for a vowel type.

Figure 10
Figure 10 Measurements for /A/ vowels for SK (Yerevan Armenian).Connected points show the actual token measurements and large open circles show group means for each measure.Gray circles (/A/ vowels without primary stress) include six tokens per group, and black circles (/A/ vowels with primary stress) include three tokens per group.

Figure 11
Figure 11 Measurements for /A/ vowels for HD (Beirut Armenian).Connected points show the actual token measurements and large open circles show group means for each measure.Gray circles (/A/ vowels without primary stress) include six tokens per group, and black circles (/A/ vowels with primary stress) include three tokens per group.

Table 1
Formant frequency mean and standard deviation for the six vowels in Yerevan and Beirut Armenian, based on measurements from Toparlak 2019 with three female speakers per language variety.Mean values are rounded to the nearest 10 Hz.

Table 2
Formant frequency mean and standard deviation for the six vowels in Yerevan and Beirut Armenian, based on measurements from Toparlak 2019 with three male speakers per language variety.Mean values are rounded to the nearest 10 Hz.
Clitics are unstressed, as illustrated in the table below (contributed by VP), as well as in the Yerevan form /vi» tÉ Sum ejin/ in the first phrase of the North Wind and Sun (Transcription of recorded passage), which contains a bisyllabic clitic.