1.1. Galician unstressed vowel system

Historically, the unstressed vowel system of Galician consists of units resulting from neutralization of the opposition between mid-low and mid-high vowels described by Veiga Arias (1965): “el gallego distingue entre /ę/ (abierta)-/ẹ/ (cerrada) y /ǫ/ (abierta)-/ọ/ (cerrada) solamente en las sílabas acentuadas, pero no en las demás, en las que consiguientemente hay una neutralización”(p. 397).[1] This neutralization had already taken place in Proto-Ibero-Romance, as Emiliano (2009) observes: “as sete vogais do latim vulgar tardio — [i], [e], [ɛ], [a], [ɔ], [o], e [u] — se reduziram no proto-ibero-romance em posição átona não final a cinco — [i], [e], [a], [o], e [u] — e em posição final a três –[e], [a], e [o]” (p. 260).[2]

This differs from the stressed system, where the opposition between /ɛ/ and /e/, and between /ɔ/ and /o/, remained in Vulgar Latin and continued into Galician (Lausberg, 1971, p. 253). The neutralization in the unstressed vowel system is confirmed by a comparison with the results in neighbouring languages where the development of unstressed vowels differs from that of the corresponding stressed vowels, e.g., Latin FĕRRUM produced Spanish hierro [ˈjero], with diphthongization of ĕ in stressed position, and Latin PŏRTA produced Spanish puerta [ˈpueɾta], but the derivative FĕRRāRĭUS became herrero [eˈreɾo], and PŏRTāRĭUS produced portero [poɾˈteɾo]. In both Spanish unstressed instances, we might have expected diphthongization of ĕ or ŏ, as occurred in stressed syllables, if their quality had been preserved in Vulgar Latin.

However, a more recent modification of the unstressed vowel system has been described. Santamarina (1974) was the first to suggest the existence of two different systems in present-day Galician: “uno, antiguo, en el que operaba la neutralización; otro, moderno, en el cual la neutralización ya no opera” (p. 12).[3] Other authors such as Porto Dapena (1977), Taboada (1979) and more recently Regueira (2008, 2009, 2010) support Santamarina’s view.

This change in the pretonic[4] system is also related to the inflection of verbs. In the “old” system referred to by Santamarina (1974), there is alternation in verbs with an open root vowel (e.g., eu voto [ˈew ˈbɔto̝], “I vote”) between the forms in the paradigm where the root vowel is stressed (in which case the mid-low quality of the vowel is maintained) and those where the root vowel is unstressed, in which raising of the root vowel occurs (e.g., nós votamos [ˈnɔs boˈtamos] “we vote”). In contrast to this traditional pattern, the new pattern described by Santamarina is characterized by the absence of these vowel alternations in many such verbs, so that the mid-low root vowel is kept throughout the paradigm here (e.g., eu voto [ˈew ˈbɔto̝] “I vote”, nós votamos [ˈnɔs bɔˈtamos] “we vote”), and verbs with a mid-high root vowel keep it throughout their paradigm, thereby giving rise to minimal pairs in pretonic position such as for example nós votamos [ˈnɔs bɔˈtamos] “we vote” versus nós botamos [ˈnɔs boˈtamos] “we throw”. This change in the system of unstressed vowels is also apparent in the way that new derived words are formed because “os derivados que se forman novos sobre palabras con vogal tónica media baixa manteñen a vogal na posición pretónica”[5] (Regueira, 2009, p. 152), as in rosa ~ roseira [ˈrɔsa̝ ~ rɔˈsejɾa̝], taberna ~ taberneiro [taˈβɛɾna̝ ~ taβɛɾˈneiɾo̝] or porta ~ porteiro [ˈpɔɾta̝ ~ pɔɾˈtejɾo̝] (Regueira, 2008).

Regueira (2008), from an Optimality Theory perspective, attributes this change to “un ascenso na xerarquía das constricións de Fidelidade (que obrigarían a que a raíz dos derivados fose ‘fiel’ á raíz da palabra primitiva) en detrimento das que prefiren vogais medias altas en sílabas átonas” (p. 278).[6] According to this idea, “os falantes máis conservadores antepoñen o proceso de elevación das vogais átonas á fidelidade á base léxica, mentres que nos falantes mais innovadores a fidelidade é prioritaria, e o proceso de elevación xa non actúa”[7] (Regueira, 2010, p. 21). However, this fails to account for the production of unstressed word-initial mid vowels as mid‑low, a recent innovation (generally occurring in high-register words), such as in omitir [ɔmiˈtiɾ] or emitir [ɛmiˈtiɾ], which contrasts with the realization of word-initial mid vowels in traditional words such as ovella [oˈβeʎa̝] (Regueira, 2010, p. 19).

Thirty years earlier, Porto Dapena (1977, p. 19) pointed out that the height of unstressed vowels depends crucially on the structure of the syllable they are part of. Porto Dapena (1977) distinguishes three different conditions affecting the realization of unstressed mid vowels in Galician:

1. En posición inicial absoluta y no trabada, tanto e como o tienden a hacerse abiertas […] 2) También se pronuncian abiertas en posición inicial absoluta, cuando van trabadas por una consonante que no sea n, m, s o x, y en interior de palabra […] 3) […] estas vocales se pronuncian cerradas en todos los demás casos […]. (pp. 20–21)[8]

Taboada (1979) adds that the quality of unstressed vowels is much more unclear and variable than that of stressed vowels, a phenomenon that he relates to a decrease in articulatory strength. The ideas expressed by Porto Dapena and Taboada are fundamental to our study, where we will look at the position of the syllable as a possible factor triggering the aforementioned change.

In the present study we will focus our attention on unstressed vowels in word-initial position in high-register words newly introduced in Galician, for which there seems to be no lexical conditioning, and which some scholars, notably Regueira (2009), treat as a special group in the change in the system of non-final unstressed vowels.

1.2. Proposal

The aim of this study is to take the ideas suggested by Taboada (1979) and Porto Dapena (1977) further by looking at the syllable as the principal factor involved in the new opposition between /e/ and /ɛ/ and between /o/ and /ɔ/ in word-initial position, from the point of view of Articulatory Phonology.

In Articulatory Phonology theory, articulatory gestures are the smallest units, the phonologicalatoms of speech. Gestures are defined as constriction units of action of the articulators, related to an ideal abstract target, but as mental units they are discrete and invariant (Hawkins, 1992, p. 13). Since gestures are naturally dynamical, context-dependent variation and coarticulation are inherent to them, so some parameters such as stiffness and duration of the articulatory gesture may be phonologically constrastive. As Browman and Goldstein pointed out (1993), “a time varying pattern of articulation motion (and its resulting acoustic consequences) is lawfully entailed by the dynamical systems themselves—they are self-implementing” (p. 54). This means that the duration of a gesture may be a distinctive feature, and also that surface alternation may be accounted for by the constriction degree and the dynamical consequences of interaction between the invariant gestural target and the variable duration of the gesture.

According to this theory, neutralizations are the consequence of the coordination of individual gestures in gesture constellations. For example, nasalization of a vowel that is followed by a nasal segment may occur because the timing of activation of the velum aperture gesture overlaps with the timing of the tongue body gesture, giving rise acoustically to a nasalized vowel. Similarly, it is not the quality of the vowel that changes but the timing of activation of the gesture for the vowel and the following gesture.

Our hypothesis is that the modification of the Galician unstressed vowel system in word-initial position is triggered by an increase of duration related to the articulatory gesture for the vowel, since greater duration allows for the existence of different ideal targets for mid-low and mid-high vowels. In turn, this increase in the duration of the gesture may be triggered by the particular syllable type, without onset nor coda and in word-initial position, described above for this phenomenon. As Goldstein, Chitoran and Selkirk (2007) say, syllable structure may lead to an increase of gesture duration which has acoustic consequences, since the articulator has more time to reach its ideal target.

Two experiments were performed to test our hypothesis. In the first experiment we tested whether there is a relationship between the duration of the gesture and the acoustic quality of mid vowels in Galician, by modifying, with Task Dynamics Application (TADA), the parameter of duration of gestures involved in the realization of the four Galician mid vowels: /ɛ/, /e/, /ɔ/ and /o/.

The second experiment is an acoustic one where we analyzed both mid-low and mid-high vowels in Galician in unstressed and stressed positions produced by native Galician speakers in order to compare the results with those obtained in the first experiment using articulatory synthesis.

2.1. The Task Dynamics Model

In order to test the theoretical assumptions of Articulatory Phonology, a computational system called Task Dynamics Application (TADA) was developed by Haskins Laboratories (2001–2006) which runs into MATLAB. TADA is based on the Task Dynamics Model explained by Hawkins (1992):

Task dynamics is a general model of skilled movement control that was developed originally to explain nonspeech tasks such as reading and standing upright, and has more recently been applied to speech. It is based on biological and physical principles of coordinated movement, but is couched in dynamical rather than anatomical or physiological terms. . . . Task dynamics describes movement in term of the tasks to be done, and the dynamics involved in doing them. (p. 9)

TADA incorporates three interrelated models. In the first, the system generates a coupling graph from an orthographic input introduced by the user. This graph specifies the gestures involved in the realization of the input and the synchronization links between them. The second model generates a gestural score, in which activation timings for each gesture are established. The third model recreates the movements in the vocal tract in the utterance, in a sagittal plane, and generates the acoustic output related to the current input.

Every gesture is specified by Constriction Degree, Constriction Location and Stiffness, in such a way that every modification in the parameters associated with the gesture will give rise to a different gestural score and, as a result, a different output, since any modification in the duration of a given gesture will result in a different coupling between that gesture and the next one.

2.2. Method for Galician mid vowels

In this first experiment we try to verify whether there is a relation between vowel height and duration of the gesture. For this purpose, we generated an input to TADA for each Galician mid vowel (/ɛ/, /e/, /ɔ/ and /o/). The words chosen as inputs to the model were emitir and omitir. These words were chosen because they exist in Galician and were also the tokens used in the acoustic experiment (see Section 3.). For each word we introduced two different inputs, one in which the constriction degree for the initial vowel had a mid-low target, and another in which it had a mid-high target. The resulting set of inputs were /emiˈtir/ and /ɛmiˈtir/ for the mid-front vowels, and /omiˈtir/ and /ɔmiˈtir/ for the mid-back vowels. From every input, TADA generates the gestural score, with the specific activation timing for every gesture and their dynamical parameters. Once those inputs were created, the durations of the tongue body gestures for /e/ and /ɛ/ and the durations of the tongue body and lip aperture gestures for /o/ and /ɔ/ were modified, from 10 to 45 ms in steps of 5 ms.

For every input entered, an acoustic output is produced from the gestural score; for this, TADA uses the synthesizer HLSync, as explained in the manual: “from the area functions, formant resonance frequencies and bandwidths are calculated, and these are used to generate an acoustic waveform using a fixed voiced source” (Nam & Goldstein, 2007, p. 4). Thus, a different acoustic output for every modification was generated by TADA, meaning that 8 different outputs were generated per vowel, that is, 32 outputs in total.

These acoustic signals (outputs) were analyzed with Praat (Version 5.3.51). Values of f₁, f₂ and duration for every vowel were taken into account. Both f₁ and f₂ were measured at the middle of the vowel, since this part is less influenced by transitions.

In order to test our hypothesis that greater duration of the gesture is related to a greater acoustical distinction of the height of mid vowels of Galician, two statistical analyses were carried out. One of these was a Regression Analysis for both f₁ and f₂ separately for each vowel. This analysis will reveal if there is a correlation between duration and formant frequencies. Since vowel quality in Galician is delimited by the distance between f₁ and f₂, a Regression Analysis for distance between these formant values for the pairs /e/–/ɛ/ and /o/–/ɔ/ was also performed. Distance was calculated using the Euclidean distance equation. Every Regression was analyzed using SPSS Statistics 20.

2.3. Results

The scatterplot in Figure 1 shows f₁ and f₂ values for every vowel generated with TADA. All vowels are more centralized at shorter durations. It is even possible to trace the trajectories for every vowel from the centre of the acoustical space (trapezoid) at shorter durations to the edges of the trapezoid at greater durations of the gesture. These centralized vowels at 15 ms may be due to the fact that the synthesizer model starts out from the neutral position of the tongue, that is, the position for schwa [ə].

Figure 1: Scatterplot of f1 and f2 for mid vowels of Galician within the different durations of the gesture tested. Circular dots represent /e/; squares represent /ɛ/; triangles represent /o/; and pentagons represent /ɔ/. The degree of transparency marks the duration of the gesture, with black representing greater durations of the vowel gesture and paler colours representing shorter durations.

For f₁ the scatterplot shows that the trajectories seem to delineate a movement from maximum aperture towards closure for every vowel as the duration of the gesture increases. There is more dispersion of f₁ for mid-high vowels than for mid-low vowels. Formant values of /e/_15ms, /e/_20ms and /e/_40ms interfere noticeably with those of /ɛ/.

2.4. Discussion

The results from the simulation study support the hypothesis that a greater duration of the gesture is related to a greater distinction between mid-low and mid-high vowels in both back and front pairs. However, the results do not agree completely with the hypothesis proposed.

We hypothesized that the new distinction between mid-low and mid-high vowels in word-initial position in Galician is due to the increase in the duration of the gesture in this position, making an opposition between mid-low and mid-high vowels possible, since greater gesture duration results in the articulator reaching its ideal target, reducing the amount of undershoot produced at shorter durations. We expected it to be the mid-low vowels that would change depending on the gesture duration, since at short gesture durations the articulator could not achieve its articulatory mid-low target, making any acoustic distinction between mid-low and mid-high vowels impossible. However, the results obtained in this simulation study indicate that only for the mid-high vowel /o/ is there a lowering related to the decrease in the duration of the gesture, whereas the results for /ɔ/ contradict our hypothesis, since there is no relation between the duration of the gesture and the formant values. This leads us to reject our expectation that it is the mid-low vowels, rather than the mid-high vowels, which changes depending on the duration of the gesture.

On the other hand, there is an unexpected relation between duration and centralization of the vowels, as shown by the results for the mid-high vowels. Although there is a significant relation between f₂ and the duration for the vowel /ɛ/, the different behaviour between this vowel and the mid-low back /ɔ/ leads us to consider the possibility that this significance may be due to the effect of both labialization and nasalization produced by contact with /m/.

Although there is ample evidence that the distance between /o/ and /ɔ/ increases as duration increases, this is due to the fact that only one of them changes its formant trajectory. From this study, we may assume that, at least for /o/, there is movement towards the raising, and for /e/ and /ɛ/ towards the extremes of the acoustical space.

In summary, we expected the mid-low vowels, and not the mid-high ones, that change depending on the duration of the gesture, but this was not supported by the results, since a relation between f₁ (taken as the acoustical cue for the vowel quality) and duration was only found for /o/.

3.1. Purpose

Since there are no acoustical descriptions of the Galician unstressed non-final vowel system, and given that the results from the first experiment were inconclusive, a second acoustic experiment was undertaken.

In this experiment, we looked at mid-low and mid-high vowels in Galician in order 1) to find out whether there are acoustic differences between unstressed and stressed mid vowels as Porto Dapena (1977) and Taboada (1979) claim; those studies were not conducted from an acoustical perspective and there is still no acoustic description of unstressed vowels in Galician, so this is a necessary preliminary step; 2) to compare the quality of mid-low and mid-high unstressed vowels to see whether they differ significantly; 3) to find out whether there is a relationship between the duration of the segment and formant values: if there proves to be, this will allow us to explain the lowering of initial unstressed vowels as a consequence of the increase in the duration of the gesture in this position, on account of which the articulators reach their target with an improvement in quality, notwithstanding the results obtained from the simulation study.

3.2. Methods

3.3. Results

The box plots in Figures 2 and 3 show the mean formant values for the four vowels studied in both Stressed and Unstressed syllables. Figure 2 shows that all the vowels studied have higher values of f₁ in Stressed than in Unstressed syllables. There also seems to be more variability in the quality of Unstressed mid-low vowels. Contrary to expectation, for the mid-high front vowel /e/ there appear to be no important differences between Stressed and Unstressed Position.

Figure 2: Box plots for f1 of the four vowels studied, in Unstressed and Stressed syllables.

Figure 3: Box plots for f2 of the four vowels studied, in Unstressed and Stressed syllables.

The second box plot (Figure 3) shows that mid-high vowels are more centralized in Unstressed than in Stressed syllable, since for front mid-high /e/ the mean values for f₂ are higher when Stressed than Unstressed, and for back mid-high /o/ the mean values for f₂ are lower in the Stressed than the Unstressed syllable. The reverse situation is found for mid-low vowels, which are more centralized in the Stressed syllable than in the Unstressed syllable, since for front mid-low /ɛ/ the mean values for f₂ are lowest in the Stressed syllable, and for back mid-low /ɔ/ the mean values for f₂ are higher in the Stressed syllable.

The third box plot (Figure 4) shows that there is ample variability in the Duration of vowels, especially for Unstressed mid-low vowels. However, this variability might be due to differences among speakers.

Figure 4: Box plot for the Duration of the four vowels studied, within the Stressed and Unstressed syllables.

3.3.1. Differences between Unstressed and Stressed vowels

The results obtained from the GLM Univariate Analysis for the variables f₁, f₂ and Duration and the three factors tested (Stress, Position and Sex) for each vowel studied (/e/, /ɛ/, /o/ and /ɔ/) are presented in Table 2.

Table 2: Results obtained from the GLM Univariate Analysis for the variables f₁, f₂ and Duration and the three factors tested (Stress, Position and Sex) for each vowel studied (/e/, /ɛ/, /o/ and /ɔ/). Factors with no significant effect are marked ns; otherwise, p values are indicated when significant.

Vowel	Variable	Stress		Position		Sex
		F	p	F	p	F	p
/e/	f1	34.303	.000	.837	ns	1.027	ns
	f2	12.149	.002	1.904	ns	49.413	.000
	Duration	24.781	.000	.004	ns	3.585	ns
/ɛ/	f1	8.411	.005	.363	ns	18.441	.000
	f2	4.333	.042	.275	ns	17.784	.000
	Duration	27.811	.000	1.082	ns	5.108	.028
/o/	f1	6.699	.014	.532	ns	.425	ns
	f2	4.237	.048	.840	ns	5.168	.030
	Duration	39.866	.000	.234	ns	.234	ns
/ɔ/	f1	7.804	.007	1.273	ns	29.845	.000
	f2	18.937	.000	21.783	.000	150.454	.000
	Duration	8.437	.006	.459	ns	.337	ns

For the vowel /e/, f₁ showed a significant effect of Stress (F = 34.303, p < .05). Post hoc analyses showed that the Stressed group has higher values of f₁ than the Unstressed group. This means that Unstressed /e/ is higher than Stressed /e/.

For f₂ there was a significant effect of Stress (F = 12.149, p < .05) and Sex (F = 49.413, p < .05). Post hoc comparisons showed that the Unstressed group had a significant lower f₂ than Stressed /e/. This means that Unstressed /e/ is more centralized than Stressed /e/.

Differences between Men and Women are related to the physical differences between male and female voices. For Duration there was a significant effect of Stress (F = 24.781, p < .05). Post hoc comparisons showed that the Stressed group had significantly greater durations than the Unstressed group.

For /ɛ/, f₁ showed a significant effect of Stress (F = 8.411, p < .05) and Sex (F = 18.441, p < .05). Post hoc analyses showed that the Unstressed group had a lower f₁ than the Stressed group, so Unstressed /ɛ/ is higher than Stressed /ɛ/. There was also a significant effect of the interaction of the two factors (F = 5.560, p < .05). As Figures 5 and 6 show, in the Women’s group there are no significant differences in height between Unstressed and Stressed /ɛ/, while in the Men’s group Unstressed /ɛ/ is higher than Stressed /ɛ/.

Figure 5: Interaction graph of Stress and Sex for f1 in /ɛ/.

Figure 6: Box plot of Stress and Sex for f1 in /ɛ/.

A possible explanation for this is that this difference between Men and Women is due to the coexistence of two models as described by Santamarina (1974), and that the men participating in the experiment have a more conservative unstressed vowel system as opposed to the women’s innovative system. However, there is insufficient evidence to back this hypothesis.

For f₂, there was a significant effect of Stress (F = 4.333, p < .05) and Sex (F = 17.784, p < .05). Post hoc analyses showed that Unstressed /ɛ/ had significant higher values of f₂ than Stressed /ɛ/, which means that, unexpectedly, Stressed is more centralized than Unstressed /ɛ/. On the other hand, the differences between Men and Women values for the f₂ are related to physical differences between female and male voices. For duration, there was a significant effect of Stress (F = 27.811, p < .05) and Sex (F = 5.108, p < .05). Post hoc comparisons showed that the Unstressed group had lower Duration than the Stressed one.

For the vowel /o/, the analysis showed a significant effect of Stress for f₁ (F = 6.699, p < .05), although there was no effect of any other factor or interaction. Post hoc comparisons showed that there was a significant difference between Stressed /o/, with higher values of f₁, and Unstressed vowels. This means that Stressed /o/ is lower than the Unstressed one. For f₂ there was a significant effect of Stress (F = 4.237, p < .05) and Sex (F = 5.168, p < .05), and the interaction between Stress and Sex (F = 4.307, p < .05) was also significant, meaning that there are significant differences in the second formant values between Men and Women that are related to physiological differences.

Figures 7 and 8 show that mean values of f₂ differ between Men and Women in Stressed syllable, since f₂ in the Men’s group is significant lower in Stressed than in Unstressed syllable; thus Men have a more centralized /o/ when it is Unstressed. For the Duration of /o/ there was a significant effect of Stress (F = 39.866, p < .05). Post hoc analyses showed that Stressed vowels had a significantly higher Duration than Unstressed vowels.

Figure 7: Interaction graph of Stress and Sex for f2 in /o/.

Figure 8: Box plot for f2 in /o/ for Stress and Sex variables.

For /ɔ/, f₁ shows a significant effect of both Stress (F = 7.804, p < .05) and Sex (F = 29.845, p < .05). Post hoc analyses showed that the Unstressed group had a significant higher f₁ than the Stressed group, which means that Unstressed is lower than Stressed /ɔ/.

For f₂, a significant effect of Stress (F = 18.937, p < .05), Position (F = 21.783, p < .05) and Sex (F = 150.454, p < .05) was found, and there was also a significant effect of the interaction between Stress, Position and Sex (F = 17.333, p < .05). Post hoc analyses showed that the group of Unstressed vowels in Boundary Position have significantly lower f₂ than the Stressed or Unstressed group in No Boundary Position. Figures 9 and 10 show f₂ differences in Unstressed and Stressed vowels between the Boundary and No Boundary Position for Men and Women. These results indicate that, while in the Men’s group there are lower values of f₂ in Unstressed /ɔ/, in the Women’s group there is a clear difference between the Boundary and No Boundary positions for both Unstressed and Stressed /ɔ/. Thus, unstressed /ɔ/ has a lower f₂ in Boundary position, meaning that the vowel is less centralized than in No Boundary position.

Figure 9: Interaction graphs showing the effect of the interaction between Sex (Men on the top graph, Women on the bottom one), Position and Stress on the f2 (on the y-axis) of /ɔ/.

Figure 10: Box plots for the f2 of /ɔ/ in Boundary and No Boundary Positions for Unstressed and Stressed syllables for Men (left column) and Women (right column).

For Duration, there was a significant effect of Stress (F = 8.437, p < .05). Post hoc analysis showed that the Stressed group had greater duration than the Unstressed one.

3.3.3. Relationship between vowel quality and Duration

In order to find out whether there is a relationship between formant values and the duration of the unstressed vowels, two Linear Regression analyses were performed for each vowel, one taking f₁ as the dependent variable and Duration as the independent variable, and the other taking f₂ as the dependent variable and Duration as the independent variable.

The results show that for the vowel /e/ there is no significant linear relation between f₁ and Duration (R² = .043, F = .631, p > .05) or between f₂ and Duration (R² = .175, F = 2.963, p > .05). Likewise, for /ɛ/ there is no significant linear relationship between f₁ and Duration (R² = .007, F = .211, p > .05), nor between f₂ and Duration (R² = .054, F = 1.700, p > .05). And also for /o/ there is also no significant linear relation between f₁ and Duration (R² = .005, F = .102, p > .05) or between f₂ and Duration (R² = .023, F = .521, p > .05).

However, for /ɔ/ the analysis showed a significant linear relation between f₁ and Duration (R² = .135, F = 4.683, p < .05), although there is no significant linear relation between f₂ and Duration (R² = .010, F = .229, p > .05). This means that the greater the duration of a segment, the higher the f₁, as is seen in the scatterplot in Figure 11.

Figure 11: Scatterplot for f1 and Duration (in ms) of /ɔ/.

3.4. Discussion

According to the results obtained from the first analysis (see section 3.3.1), we may conclude that there are significant acoustic differences, in terms of height and centralization, between Galician unstressed and stressed mid vowels, in agreement with Porto Dapena’s (1977) suggestion that unstressed vowels are more unclear and variable than stressed vowels. Broadly, vowels are longer in stressed syllables than in unstressed ones. The unstressed mid-high vowels /e/ and /o/ and mid-low /ɛ/ are higher and (at least for men) more centralized than their stressed counterparts. On the other hand, unstressed /ɔ/ is lower and less centralized than stressed /ɔ/, except in No Boundary Position in women.

The second analysis (see section 3.3.2.) showed significant differences in the height and centralization of the unstressed mid-front pair, /e/–/ɛ/, although there is no difference in duration. For the unstressed mid-back vowels, /o/–/ɔ/, there are also significant differences in height and centralization, plus greater duration for /ɔ/ than for /o/. Moreover, for unstressed /ɔ/ there is a significant effect of position on f₂ values, which means that this vowel is more centralized in medial position than in initial boundary position in women. On the other hand, the third analysis (see section 3.3.3) showed a significant relation between duration and height of the unstressed low mid-back vowel /ɔ/. Although the existence of a significant relation does not imply causality, this significance supports our hypothesis.

We can assume that most of the effects of the sex variable can be explained as physiological differences between male and female voices as Harrington (2010) states,

the formants of the female speaker are higher in frequency than those of the male, which is expected because female vocal tracts are on average shorter. But, as argued in Fant (1966), because the ratio of the mouth cavity to the pharyngeal cavity lengths is different in male and female speakers, the changes to the vowels due to gender are non-uniform: this means that the male–female differences are greater in some vowels than others. . . . Also the F2 differences are much more marked than those in F1 (p. 184).

However, it seems that for unstressed mid-low vowels the results point to the continued coexistence of the two models described by Santamarina (1974), one of which is more conservative, represented in this study by the men’s group, and the other more innovative, represented by the women’s group. Those results can be seen in the scatterplots in Figure 12.

Figure 12: Scatterplots of the mean formant values for the four vowels within two different kind of syllable. The circle represents the vowel /e/; the square represents /ɛ/; the triangle represents /o/; the pentagon represents /ɔ/. Grey markers represent stressed vowels, and black markers represent unstressed vowels.