A Comparative Study of Armenian Vowels Production and Perception

A Comparative Study of Armenian Vowels Production and Perception

Tabita Toparlak

December 3, 2019

Contents

1 Introduction … 1
2 Background … 2
3 Measuring the Vocalic Space of Armenian Vowels … 3
3.1 Discoveries … 3
3.2 Material and Methodology … 3
3.3 Results … 4
3.4 Discussion … 7
3.5 Interim summary … 9
4 Autoditory Space of Armenian Speakers … 9
4.1 Introduction … 9
4.2 Material and Methodology … 10
4.2.1 Speakers … 10
4.2.2 Material … 10
4.3 Results … 12
4.4 Conclusion … 14

1 Introduction

This study measures the vowel space differences between two standard Armenian dialects. Modern Armenian has two standardized dialects: Standard Eastern Armenian (henceforth EArm) which is mostly spoken in Armenia, Iran, and Russia, and Standard Western Armenian (henceforth WArm) which is mostly in spoken in the Diaspora: France, the United States, Turkey, Syria and Lebanon. Each dialect is spoken in a wide range of geographic locations with many subdialects. For our

study, we focus on EArm as spoken in Yerevan, and on WArm as spoken in Beirut. Even though the two standard varieties are mutually intelligible, it is a known fact that they show differences in every linguistic domain, be it lexical, morphological, or syntactic. This dialectal difference is claimed to go back to $10^{\text {th }}$ century (Donabédian and Feydit, 2000):
“It is not until the tenth century that the first evidence of a language closer to the vernacular appeared. Middle Armenian began to be used in administrative, legal and practical texts. It demonstrates the first signs of a separation in two dialectal families: the Armenian of Cilicia marks the appearance of characteristic grammatical forms of Western Armenian. From the $12^{\text {th }}$ to the $19^{\text {th }}$ century, next to literary or religious texts in Classical Armenian, we gradually see books, especially related to practical life, but also popular literature, written in a language often unstable, but relatively close to the spoken language and which clearly foreshadows what will be modern Armenian. In fact, until the genocide of 1915, Armenian is spoken on a vast territory corresponding to historical Armenia, and includes a large number of dialects. As spoken languages with little access to writing, these dialects gave birth to two literary languages which, in the second mid-nineteenth century, had supplanted the classical literary language, not without having given rise to hard debates among the different intellectual currents”.

In this study, we examine the vowel spaces of each dialect and measure their first and second formants. We apply Lobanov’s z-score normalization period which is regarded as one of the most reliable normalization methods by the researchers in this domain (Adank, 2003; van der Harst, 2012).

Both dialects have the same underlying vowel inventory: /a,o,e,o, $\mathrm{i}, \mathrm{u} /$. To our knolwedge, there is some phonetic work on the vowel space of EArm (Xačatryan, 1988), but none on WArm. We fill this gap. We provide acoustic measurements for both dialects. We find formant differences based on dialect and gender. We also find that the schwa has variable pronucation.

2 Background

In sociophonetics, several studies have been conducted to examine the vowel spaces across the varieties of a language. These studies have been done for Dutch (Adank, 2003; van der Harst, 2012), Saterland Frisian and Low-High German (Schoormann et al., 2017), and Greek (Schoormann et al., 2017). Besides dialect-based variation, there is also work on gender differences in the production and perception of vowels. These differences are generally based on F0, articulatory variation, and F1/ F2 differences (Kachel et al., 2017). Linguistic variation in Armenian, however, has received limited attention in the literature. Adjarian (1909, 1952), as the first scholar work on this topic, examined several aspects of linguistic variation in Armenian. He provides a linguistic description for each documented variety. He describes lexical, morphological, syntactic, and also phonetic particularities for each dialect. But despite the importance of his work, Adjarian’s works date back to at least a century in the early 1900’s.

Most modern work on Armenian is by Bert Vaux (1998) for phonology and Amalya Xačatryan (1988) for phonetics. Vaux has worked extensively on Armenian, especially dialectology and phonology (Vaux, 2007). Vaux (1998) gives detailed description of the syllable structure and phonological inventory of Armenian from diverse dialects. In phonetics, Xačatryan (1988) describes the phonetic features such as the vowel space and gives average formant values of F1/ F2 for each vowel for EArm. Her vowel space is shown in Figure 1. There are some studies of Armenian stress and intonation (Toχmayjan, 1971; Athanasopoulou et al., 2017).

Figure 1: Vowel space distribution of Xačatryan (1988).

3 Measuring the Vocalic Space of Armenian Vowels

3.1 Discoveries

Because there is little to no work on Armenian vowel spaces, we did not have any hypotheses. Instead, this research is descrptive and exploratory. We collecting and analyzing acoustic data on the vowel space of Western and Eastern Armenian. We had the following findings.

1. Discoveries on the Armenian vowel space:

(a) The vowel [a] is lower in EArm than in WArm for both male and female speakers.
(b) The female speakers have a wider vowel space for [a] than the male speakers for both varieties.
© The schwa [ə] is more back in WArm than in EArm; the schwa of EArm is more central.
(d) Male speakers tend to produce front vowels [i,e] as more front than female speakers.
(e) The vowel space of the schwa is closer to [e] for EArm speakers and to [u] for WArm speakers.
(f) /e/ for EArm is an open-mid vowel than WArm /e/.
(g) Judging from the overlap on the F1/ F2 chart, phonologically, the schwa is treated as an epenthetic vowel, whereas, in WArm, the schwa is treated as a full vowel with clear boundaries.

3.2 Material and Methodology

For our data, we recorded 12 people in total. For each variety, 6 people ( 3 males and 3 females) were recorded. When choosing participants, the most important criterion was the participant’s proficiency in Armenian. The study was held in France. We chose participants who were the least affected from the dominant language. We also recorded a speaker from Turkey in order to see if there was a significant effect of Turkish central vowels of Turkish in her Armenian speech. The speakers’ age ranged between 21-40. Most of the recordings were done with a microphone and a portable sound card (Edirol) borrowed from Phonetics and Phonology Laboratory of Institut de Linguistique et Phonétiques Générales et Appliquées (ILPGA), Paris. The recordings were made in Praat (Boersma and Weenink, 2001) with a 44.100 Hz sampling rate. Recordings were done in a silent room in LPP. For the corpus, we created 32 utterances for each dialect. We slightly adapted each utterance according to dialect because the dialects had minor differences in syntax. We also controlled the positions of the vowels in the syllable and utterance. The sentences are grouped according to the target vowel. In sum, we had six set of utterances. The following example set demonstrates three sentences from our stimuli. The vowels in red were the target vowels. Speakers did not know which vowel was the target vowel.
2. Stimuli Set for [a]: This Utterance is identical for both dialects

The stimuli illustrated above, is the set for the vowel /a/. The speakers were recorded each set respectively. The recording had two parts: in part one, the speakers were asked to utter each sentence twice. In part two, the speaker were asked to utter only the words that contain the target vowel. Each word is uttered three times. Only the target vowels have been measured. The target vowels were in one of three positions: word-initial, word-medial and word-final. We asked our speakers to utter each sentence twice. For the labeling of the vowel, the first and the last cycle of the speech signal of the vowel was decided as the vowel onset and offset.

At the end of each sentence, the speakers were asked to utter each target word three times.
The target vowels were calculated automatically with Praat scripts. All the data was transformed to Excel files. The files were uploaded to visiblevowels.org in order to generate visualizations of the data (Heeringa, 2018).

Due to erroneous annotation, we had to exclude 293 tokens from 6093 tokens in total. For the most reliable results, we used Lobanov’s z-score normalization throughout the evaluation process of the data. Lobanov’s normalization method is a formant intrinsic and vowel extrinsic method. Its formula is the following:

$$D i=\frac{F i-\mu i}{\sigma i}$$

In the precedent equation, the mean formant frequency $\mu_{i}$ is calculated using all the vowels of a speaker for a formant $i$, while $\sigma_{i}$ refers to the standard deviation (Esfandiaria and Alinezhadb, 2014). Among various normalization methods, Lobanov’s method was the most appropriate for our purposes. In Lobanov’s method, it is assumed that the required information is distributed across more than one vowel category for each speaker (Flynn, 2011). Hence, the speaker variation is kept but minimised in an objective scale. The z-scores shows the distance of the standard deviation from the mean of distribution. This enables us to do a coherent comparison of the data.

3.3 Results

The following charts show the distribution of the vowels based on dialect and gender. The general vowel space distribution for the two dialects is shown in Figures 2a and 2b. The charts show all 2500-3000 vowel tokens spoken by the 6 speakers of each dialects. For each dialect,the data was normalized with Lobanov’s method. I show the average F1 and F2 found for each vowel. A small number of tones $(77,16)$ were removed because they were unexpectedly pronounced as $[\mathrm{y}]$.

Figure 2: General vowel space distribution for each dialect by pooling all participants

(a) Vowel space distribution for WArm $\mathrm{N}^{\mathrm{a}}$ of tokens: 2983, 77 tokens of $[\mathrm{y}]$ removed .

Eastern Armenian General

(b) Vowel space distribution for EArm $\mathrm{N}^{\mathrm{a}}$ of tokens: 2610 .

For each dialect and for each vowel, we calculated the average, mean, and standard deviations of F0, F1, F2, and F3. Tables (1 and (2) show the statistical data for WArm and EArm respectively.

Table 1: Average and standard deviation for F0 and F1-3 for Western Armenian vowels.

Table 2: Average and standard deviation for F0 and F1-3 for Eastern Armenian vowels.

The vowel spaces differed slightly by gender. For each dialect, 3 participants were female while 3 were male. Figures 3a and 3b show the vowel space for the female and male speakers of Western Armenian. There were 1668 tokens of female WArm speech, and 1315 tokens of male WArm speech. A small number of vowel tokens $(47,30)$ were removed because were unexpectedly $[\mathrm{y}]$.

Figure 3: General vowel space distribution for WArm based on gender

Similarly for Eastern Armenian, the vowel space for male and female speakers are given in Figures 4a and 4b. There 1299 tokens of female EArm speech, and 1311 tokens of male EArm speech. A small number of vowel tokens $(47,30)$ were removed because were unexpectedly $[y]$.

Figure 4: General vowel space distribution for EArm based on gender

Pooling together both dialects, male and female speakers had similar vowel spaces. Figures 5a and 5b show the vowel space for females and males. There 2967 tokens of female speech, and 2626 tokens of male speech. A small number of vowel tokens $(56,37)$ were removed because were unexpectedly $[y]$.

Figure 5: General vowel space distribution based on gender for the two dialects combined

(a) Female speakers of WArm and EArm N*of tokens: 2967
56 tokens of $[\mathrm{y}]$ removed

(b) Male speakers of WArm and EArm N*of tokens: 2626
37 tokens of $[\mathrm{y}]$ removed

3.4 Discussion

The acoustic measurements displayed the following findings.
3. Discoveries on the Armenian vowel space: (repeated)
(a) The vowel [a] is lower in EArm than in WArm for both male and female speakers.
(b) The female speakers have a wider vowel space for [a] than the male speakers for both varieties.
© The place of articulation for the schwa [ə] is more back in WArm than in EArm; the schwa of EArm is more central.
(d) Male speakers tend to produce front vowels [i,e] as more front than female speakers.
(e) The vowel space of the schwa is closer to [e] for EArm speakers and to [u] for WArm speakers.
(f) /e/ for EArm is an open-mid vowel than WArm /e/.
(g) Judging from the overlap on the F1/ F2 chart, phonologically, the schwa is treated as an epenthetic vowel, whereas, in WArm, the schwa is treated as a full vowel with clear boundaries.

We show the vowel charts for both varieties based on average F1-F2. ${ }^{1}$

^[1]

Figure 6: General vowel space distribution charts of the two dialects

(a) Western Armenian vowel chart.

(b) Eastern Armenian vowel chart.

The first aim of this study was to describe the Armenian vowels of both varieties on F1/ F2 scale. As we have summarized in section 3.1, the vowel [a] is lower in EArm than in WArm, for both male and female speakers (for EArm male and female speakers the mean values of F1 is 566 and 675 Hz , and 572 and 727 Hz for WArm speakers respectively). The vowel [a] is classified as low back vowel in Eastern Armenian (Xačatryan, 1988). Indeed, our data suggest the same features for [a] in EArm. Though the average formant frequencies indicate that the mean F1 and F2 values are slightly different from what we found, we have remarked that the jaw is closer and that the tongue has been moved more forward for the production of the vowel [a] (2a and 2b). Xačatryan speculated that some speakers have dialectal traces in speech; they produce [a] as a mid-vowel (Xačatryan, 1988). She also underlines the fact that, this speculation is unproven. Nevertheless, the analysis of the male and female speakers of EArm suggest that, even though the vowel distribution is remarkably distinct, there do exist some tokens of [a] and [o] whose distribution intersect with each other. This is a likely effect of coarticulation than any other dialectal differences.

The production of [a] varies by gender. The female speakers for both varieties tend to have a wider space for [a] than the male speakers. Male tokens indicate an average of $566,1450 \mathrm{~Hz}$ for F1,F2 in EArm and 564, 1321 Hz in WArm. On the other hand, female tokens indicate an average of $680,1413 \mathrm{~Hz}$ for F1, F2 in EArm and 696, 1509 Hz in WArm. Taking into account the effect of the vocal folds for the first and second formant frequencies, and judging from figures 5 a and 5 b , we can also conclude that the vowel [a] is produced more front in female speakers.

Xačatryan classifies [e] as a front mid-vowel for EArm. In her data, the average F1 and F2 were receptively 305 Hz and 1600 Hz . What is different in Xačatryan’s data than in ours is that the central vowel [a] has a greater value of F2 (1734 Hz) than [e] (1600 Hz) (Xačatryan, 1988). These values place the schwa as more front, and thus the vowel [e] becomes “more central”. In our data, there is variation in the vowel space based on dialect and gender. Some tokens intersect or even overlap with each other, but it is clear that the vowel space of the vowel [e] is always more front than the schwa [a]. Figures 2 a and 2 b suggest that the [e] is closed in WArm speakers more than in EArm speakers. The tongue is also back In both dialects, the range of [e] for male speakers is denser than for female speakers. The [e] of female speakers from both dialects can be characterized in the mid-vowel range, whereas the frequency range of [e] for male speakers of both varieties is closer to [a]. The [e] can thus be classified as a more central in males than that of the females’ (figures 5 a and 5 b ).

Because of coarticulation and schwa epenthesis we had to dismiss 211 tokens of schwa. The vowel space of the schwa remains difficult to demarcate. However, the range of the schwa draws a quite clear boundary for the schwa. When we look at the general distributions of the vowels, we can conclude from Figures 2a that the schwa is more prone to be a high-back vowel [ $\mathrm{m}]$ than a central [a] in WArm. On the other hand, the schwa [a] of EArm speakers shows a canonical frequency space of [a] in the F1/F2 dimensions (Figure and 2b). In EArm, the schwa [a] has a more front production in female than in male speakers. But on the other hand in WArm, [a] of the female speakers is more back and the F2 range is larger than in male speakers. In other words, In WArm, the [a] of male speakers is more centralized than the [a] of the females. The female speakers of WArm have a [a] range which is closer to [o], whereas the male speakers have a schwa range which is closer to [e]. For the WArm speakers, we can assume that

the vowel space of the schwa is more central. The female speakers of WArm, like the male speakers of EArm, have a schwa range that is closer to [e].

The vowel [i] is a front high vowel. Though, its vowel space is located practically the same place in the F1/ F2 axis, the subtle difference between the figures is sufficient to have an idea about the variation of articulation. The frequency space for the vowel [i] for WArm speakers is between 2000-3000 Hz, whereas this range is reduced to 2000-2500 Hz for EArm speakers. This indicates that the WArm speakers have a more front [i] than EArm speakers. We can also conclude that the vowel space of the WArm speakers is wider than for the EArm speakers. The EArm female speakers have an F2 value around 2500 Hz , while this average goes up to 3000 Hz for the females of WArm. For the male speakers of EArm, on the other hand, the F2 value goes up to 2500 Hz . This means that the jaw is closer to produce the vowel [i].

According to Xac̆atryan, the vowels [u] and [o] are similar in terms of articulation. She compares the width of the mouth for [o] with [a]; she finds that the width of the mouth for [o] is smaller than for [a]. Taking into account these two explanations for [o], we may suggest that the /o/ is a close-mid vowel in Armenian vocalic system. In fact, it is not completely implausible to analyze back mid-vowel /o/ as the open-mid vowel [3]. It is a known fact that the diphthong in /aw/ in $5^{\text {th }}$ century had become to be noted as $<0>$ in the beginning of the $13^{\text {th }}$ century (Macak, 2017). Until the expansion of this use, manuscript writers would use the Greek letter omega $\omega$ to point out two different pronunciations of /aw/ ([av] and [i] $)^{2}$. The words such as awur (day), hawr (father acc.), awgostos (august). By the time this use became more common, the additional omega were removed, and the letter $o$ was borrowed from Greek into the Armenian alphabet system. Putting aside the historical background of the difference between close-mid and open-mid back vowels, today, more such difference is highly improbable, specifically in our data. The only trace of this vowel shift is that word-initial /o/ is pronounced as [vo], while the contracted form of ancient /aw/ diphthong is realized as [o] word-initially. Based on our data, the most significant articulation difference between [o] and [u] is that the vowel [o] has an F1 around only 100 Hz higher than the former. Otherwise the tongue is in practically the same position for the two vowels. The vowel space of [o] is closer.

3.5 Interim summary

In this fieldwork, we tried to draw a general sketch of the vowel spaces of two main varieties of Armenian. We proved that the place of articulation of the schwa in Armenian never precedes [e]. The schwa of the WArm speakers occupies a space that is “closer” and more “near back” than in EArm. In general, female speakers tend to have a wider vowel space for the vowel [a] than the male speakers. The general F0 values of EArm speakers is lower than the WArm speakers. We believe that this study should also be conducted for the non-standard varieties of Armenian.

The next section describes a perception study of Armenian vowels. The results are descriptive of Armenian vowels, proposing general vowel spaces on F1/ F2 scale for both varieties. The schwa is not treated identical and EArm vowels tend to be darker than WArm vowels. Furthermore, the vowel spaces on production basis should also be supported by perception studies.

4 Autoditory Space of Armenian Speakers

4.1 Introduction

The previous study was a production study where we measured the vowels produced by Armenian speakers. This study provided a baseline of average formant values for Armenian vowels. We followed up a perception study using synthesized speech. The goal of the perception study is to examine of how different formant values affected the perception of Armenian vowels.

^[1]

We will first introduce the speakers, the material we used, the test parameters and finally, we will discuss the results.

4.2 Material and Methodology

4.2.1 Speakers

The most reliable results for the perception study would have been collected if we used the same participants from the production study. Unfortunately, the majority of the participants from the production study had left Paris at the time. We had to use new participants for the perception study. Despite the change in participants, the results are consistent and they are also consistent with the results from the production study.

6 speakers in total (4 WArm, 2 EArm) participated in this study. The age range of speakers is between 20-34. And just as for the production study, we consulted speakers from Beirut and Yerevan. Half of the participants (2 WArm, 1 EArm) had also participated in the production study. The WArm speakers lived in France and knew French, while the EArm speakers either did not live in France or did not know French.

4.2.2 Material

We used the JmH software developed by René Carré, CNRS. The test is constructed in the following way. A set of synthesized vowels are introduced to the participant. After the presentation of each stimulus on a computer screen, the participant must click on a box where a vowel appears which they consider that the stimulus is a realization. In case they do not identify any vowel, they point to an empty box. The duration of the vowel can be set by the experimenter (in this case 300 ms .). The formant space varies as follows: the first formant varies between 250 - 750 Hz with a pitch of 100 Hz ; the increment of the second formant is 200 Hz in a range between 650 and 2350 Hz ; for the anterior and posterior vowels, a rounded and non-rounded series are represented in the synthetic stimuli by varying the third formant. The vowel space is trapezoidal, which means that the values that correspond to the lower corners of the square have been discarded.

The entire test is presented 5 times with breaks between each test if necessary. With the exception of the first, all the results are elaborated. The first is not taken into account lest it be biased by familiarization with the trial. Therefore, from these 4 tests, we sorted the correct answers, i.e. the answers that were clicked 3 times correctly by the speaker. The responses are formatted in the visiblevowels.org format for visualization (Heeringa, 2018).

Figure 7: Screenshot of the perception test software.

Armenian does not have any contrast between long and short vowels or between half-open and half-closed vowels as in French. On the basis of production data we expect that the speakers of WArm will locate /a/ at a point where F1 is around $500-600 \mathrm{~Hz}$ because our first results suggest the /a/ more open for the speakers of EArm. We assume the absence of [y] and [oe] in the EArm speakers, because these vowels are still not part of their vowel system ${ }^{3}$. On the other hand, we assume that [y] will be present in WArm speakers as they are exposed to Turkish or French at an early age. Finally, we assume that the schwa, i.e. the central vowel, will appear close to the mid-back vowel.
${ }^{3}$ In contrast to the speakers of WArm, /iw/ is pronounced as [ju] by the speakers of EArm and [oe] only appears in the borrowings, for example, [boerek] (paste) and [boen] (butter)

4.3 Results

Figure 8: Perception Results for WArm

(a) Auditory space of a female speaker of WArm.

(b) Auditory space of a male speaker of WArm.

Figure 9: Perception Results for WArm

(a) Auditory space of a female speaker of WArm.

(b) Auditory space of a female speaker of WArm.

Figure 10: Perception Results for EArm

(a) Auditory space of a female speaker of EArm.

(b) Auditory space of a female speaker of EArm.

The results show the following:
For the speakers of WArm, the auditory space of [i] is 250 Hz for the F1 and 2000-2500 Hz for the F2. As for the space of [y] which in general is introduced into the system via the neighboring languages, it is characterized by an F2 between 1500 and 1900 Hz . The auditory space of the cardinal vowels is coherent among the speakers. This is the case for [u], as suggested by the tables, the distribution is between 250 Hz for F1 and 900 Hz for F2. There is one point (F1 $350 \mathrm{~Hz}-\mathrm{F} 2900 \mathrm{~Hz}$ ) that is labeled as [u] by WArm speakers. This area is usually labeled as [o] by speakers (including EArm women). However, as our first corpus suggests, [n] and [o] may overlap in this area.

Since there is no difference between open-mid and close-mid vowels, the judgment of the speakers for [o] varies from 350 to 600 Hz for the F1, and around 1100 Hz for the F2. For open-mid / close-mid vowels [e, c]. The distribution ranges from 300 to 700 Hz for F1 and from 1500 to 2000 Hz for F2.
[a]: As already mentioned, the frequencies of the first two formants for [a] are 650 and 1100 Hz , according to Xačatryan’s data. Our results suggest that this vowel presents differences in terms of articulation between the two standard varieties (in WArm the mandible is higher and the tongue is slightly ahead than in EArm). On the other hand, regarding the values of the frequencies, the speakers chose the farthest stimuli to label as [a] ${ }^{4}$. Figure 8a shows the frequencies from 750 Hz to 850 Hz as [a]. These frequencies are perceived as [a] by others, however, the rest of WArm speakers have larger auditory space for [a] from 650 Hz to 750 Hz for F1 and 850 Hz to 1050 Hz for the F2.
[ə]; Finally, the suggested auditory space for [oe] and/or [ə] corresponds to the central part of the distribution in the tables. This can be explained in several ways. In the first place, the target group of this test corresponds to the speakers of French, thus, the monosyllabic words were derived from French. The word that corresponds to [oe] is <leu>. For the central vowel, the software proposes <the> (marked as [ə] on the tables) that we took as “the vowel of the Armenian” whose characteristics are roughly similar to that of Turkish [u] (Özsoy, 2004, 36). As a result, speakers were asked to click on «the» when they heard something similar to the Armenian schwa. The main index for the detection of this vowel is the fact that the first formant of sound refers to a non-rounded vowel. The labeling of the third speaker 9a proposes the ideal location of the Armenian schwa. We had the results corresponding to this same zone, but they were detected only two times maximum, thus not enough to appear on the tables. This procedure also explains the gap in the auditory space of the second speaker of WArm (which is supposed to be filled by [oe] or the schwa like the others).

^[1]

[i,u]: The speakers of EArm have a smaller space for the vowel [i]. Only the stimuli with the highest F2 are correctly identified by EArm speakers. The perception of [u] is also consistent, with the exception of a second speaker who labeled only the $250 \mathrm{~Hz}-850 \mathrm{~Hz}$ point (F1/F2) as [u], while the first has a larger space for that.
[e]: We see two different labels for some identical coordinates in two speakers. Since Armenian does not distinguish between close-mid and open-mid vowels, we can conclude that the frequencies of 350 Hz to 1950 Hz of F1 and F2 are perceived as $[\mathrm{e}]$.
[o]: The first speaker has the following distribution for [o]: from 350 Hz to 650 Hz for F1, and from 850 Hz to 1050 Hz for F2. ${ }^{5}$.
[y]: As for the vowel [y], the two speakers show the erroneous identification results. The first speaker labeled the frequencies of 250 Hz and 850 Hz and 250 Hz and 1250 Hz as [y]. In contrast, the second speaker labeled the entire high distribution as $[\mathrm{y}]^{6}$. We hypothesized that EArm speakers are not able to categorize the vowel [y] as do WArm speakers. The results suggest either an exaggerated use or the complete abandonment of this use. The exaggerated use must be triggered by the will of the speaker to categorize all the sounds he hears in the inventory. Thus, among the group of WArm we see the identical distribution labeled by question marks because these sounds seemed to him impossible to identify. As a result, the speaker preferred to click on <ly> instead of <?> To categorize sounds that are not known in her phonological inventory. In a way, she selected a label with a set of sounds that are not usual to her ears.

We also see that they labeled the points that belong to [y], in principle with [oe]. This fact can be explained either by the spelling or the labeling of an unknown sound by a label that belong to unknown sounds of another language. As for the central vowel, in principle, the middle of the F1/F2 range is reserved for the schwa, and contrary to the WArm results, the lower zone of the table whose frequencies correspond to +700 Hz for F1 and 1500 - 2000 Hz for the F2, is categorized as central. The choice of labels for the central vowel is arbitrary (between [oe] and [a] of course) for the reasons we have already presented. For the vowel [a], the first speaker has marked a /a/ clear for the stimulus [o] (the box <la>), this result can be considered in agreement with the choice of the second speaker whose the perception of [a] is scattered more around the same area.

4.4 Conclusion

We have attempted to sketch the perception and production spaces of Armenian speakers. By this study, we see that the vowels whose frequencies are on the points furthest away from each other are the best categorized, the categorization of [u] and [o] on some measurement points as an exception. Duration may play a larger role in categorizing round and unrounded vowels. It is likely that the isolated listening of the vowels made the process of categorization more complicated. Exposure to French spelling was a variable that had a fairly significant impact on the results. As for the comparison of the two results, it is quite visible that the vowels in the tables of the production section are very concentrated towards the center, and the values of the vowels are much closer to each other.

The question of the auditory space of the central vowel of Armenian is still not completely resolved. Based on our corpus of production and perception, we will not be wrong if we propose that it is further back and higher, because the majority of erroneous entries are due to the central vowel.

References

Adank, P. (2003). Vowel Normalization: A Perceptual-acoustic Study of Dutch Vowels. UB Nijmegen [Host].

^[1]

Adjarian, H. (1909). Classification des dialects arméniens. Bibliothèque de l’Ecole des hautes études: Sciences historiques et philologiques. H. Champion.

Adjarian, H. (1952). Knnutjun Vani Barbari. Erevan : Erevani Hamalsarani Hratarakčowt’yown.
Athanasopoulou, A., I. Vogel, and H. Dolatian (2017). Acoustic properties of canonical and non-canonical stress in french, turkish, armenian and brazilian portuguese. In INTERSPEECH, pp. 1398-1402.

Boersma, P. and D. Weenink (2001). Praat, a system for doing phonetics by computer. Glot International 5(9/10), $341-345$.

Donabédian, A. and F. Feydit (2000). La langue: origine et étapes du développement de la langue (grabar, arménien moyen, arménien moderne). op. cit, 47-66.

Esfandiaria, N. and B. Alinezhadb (2014). Evaluating normalization procedures on reducing the effect of gender in persian vowel space. International Journal of Sciences: Basic and Applied Research (IJSBAR) 13(2), 303-316.

Flynn, N. (2011). Comparing vowel formant normalisation procedures. York papers in linguistics 2(11), 1-28.
Heeringa, W. . V. d. V. (2018). Visible vowels: a tool for the visualization of vowel variation. Proceedings CLARIN Annual Conference 2018, 8 - 10 October.

Johnson, K., E. Flemming, and R. Wright (1993). The hyperspace effect: Phonetic targets are hyperarticulated. Language, 505-528.

Kachel, S., A. P. Simpson, and M. C. Steffens (2017). Acoustic correlates of sexual orientation and gender-role self-concept in women’s speech. The Journal of the Acoustical Society of America 141(6), 4793-4809.

Macak, M. J. (2017). The phonology of Classical Armenian. In J. S. Klein, B. D. Joseph, and M. Fritz (Eds.), Handbook of Comparative and Historical Indo-European Linguistics, pp. 1037-1079. Walter de Gruyter.

Özsoy, A. S. (2004). Türkçe’nin yapısı, Volume 1. Boğaziçi Üniversitesi Yayınevi.
Schoormann, H. E., W. J. Heeringa, and J. Peters (2017). A cross-dialectal acoustic study of saterland frisian vowels. The Journal of the Acoustical Society of America 141(4), 2893-2908.

Toz̆maz̧jan, R. M. (1971). zamanagagits hajereni barajin J̧eftə. ləraper hasarakakan gitutšjunneri 2, 60-64.
van der Harst, S. (2012). The Vowel Space Paradox: A Sociophonetic Study on Dutch. LOT. Unknown Publisher.
Vaux, B. (1998). The Phonology of Armenian. Phonology of the world’s languages. Clarendon Press.
Vaux, B. (2007). Homshetsma: The language of the armenians of hamshen. In The Hemshin, pp. 311-332. Routledge.
Xačatryan, A. (1988). zamanakakits hajezeni hntfunabanutjun. Erevan : Haykakan Gitowtyownneri Akademia.