Developmental Aspects of Greek Vowel Reduction in Different Prosodic Positions

Abstract

This study investigates the development of Greek vowel reduction across different prosodic positions (stressed, pre-stressed, post-stressed), examining normative data from 72 participants aged 3 years to adulthood and balanced for gender. Participants performed a delayed repetition task, producing real trisyllabic words with the vowels [i, ε, ɐ, o, u] examined in the second syllable. Measurements included relative vowel duration, normalized acoustic vowel space areas, and Euclidean distances of vowels from the centroid of the acoustic space. Our findings show that changes in speech motor control, system stiffness, and stress marking with age, along with children’s prosody sensitivity, contributed to several developmental milestones: the completion of the developmental trajectory of relative vowel duration and temporal vowel reduction at early adolescence; the attainment of adult-like spatial vowel characteristics and their reduction at preschool age; and the early acquisition of the prosodic strength of the stress conditions, leading to vowel reduction from the stressed to pre-stressed to post-stressed conditions. The correlation strength between temporal and spatial vowel reduction across ages revealed age-related differences in spatiotemporal speech organization, with significant gender-related differences observed only in vowel space areas, where females exhibited larger areas possibly related to sociophonetic factors. Intrinsic vowel duration appeared from age 3.

1. Introduction

Vowels carry key linguistic information, related to quality, stress, rhythm, and intonation (Nespor et al. 2003), and assume a pivotal role in facilitating effective communication. Consequently, an examination of their characteristics, alongside phenomena like vowel reduction, is important both for theory and applications in clinical practice and technology.

Vowel reduction is a phenomenon that can take on phonological or phonetic dimensions. Phonological vowel reduction is categorical, resulting in the substitution of a vowel in weak syllables. As proposed by Miller (1972), this process may entail the complete neutralization of the reduced vowel (e.g., change to [ə]) or the retention of certain phonetic features, such as rounding (e.g., in Eastern Ojibwa) or palatality (e.g., in certain English dialects). In contrast, phonetic vowel reduction is gradable and results in shifts of vowels within the acoustic vowel space, without altering the phonological quality of the vowels. Barnes (2006) emphasizes that a key difference between phonological and phonetic vowel reduction is that phonological vowel reduction involves changes in vowel quality that directly affect vowel duration, whereas phonetic vowel reduction involves changes in vowel duration that, in turn, influence the spectral characteristics of the vowels.

1.1. Phonetic Vowel Reduction

Phonetic vowel reduction, which is the focus of this study, is characterized by a general displacement of reduced vowels within the acoustic space. High vowels are particularly susceptible due to their inherently short duration and are also prone to whispering or even deletion in certain languages like Adrean Spanish (Delforge 2008). Different degrees of phonetic vowel reduction can be explained in relation to variation in the coordination of articulatory gestures within the framework of Articulatory Phonology (Browman and Goldstein 1992). Specifically, any deviation from the ideal acoustic target, ranging from slight alteration to complete deletion, stems from changes in gesture magnitude and/or an increase in inter-gestural overlap (Browman and Goldstein 1992).

While various factors may influence articulatory coordination (Kuehn and Moll 1976; Nord 1986; Flege 1988), Moon and Lindblom (1994), drawing from Lindblom’s (1963) vowel reduction model, propose that vowel undershoot arises from three main factors: 1. reduced vowel duration, 2. large spectral distance between the target vowel and neighboring speech sounds, and/or 3. slow rate of formant changes relative to articulatory movement speed. Therefore, articulators fail to reach the ideal target, leading to vowel reduction 1. when vowel duration is reduced, e.g., due to faster speech rate or other prosodic factors; 2. when there is a considerable difference between the vowel’s F2 value and the locus of the adjacent consonants, e.g., between [i] and [p]; and 3. when a speaker’s articulatory movements are slow during transitions between sounds due to their speaking style. In earlier studies, it was posited that reduced vowels tend to occupy central positions in the acoustic vowel space (Joos 1948; Tiffany 1959), but research on Swedish, English, and French suggests that they assimilate to neighboring speech sounds in terms of their spectral characteristics (Lindblom 1963; Fourakis 1991; Moon and Lindblom 1994; Gendrot and Adda-Decker 2005). Thus, in contexts without schwa-like elements, reduced vowels move away from the acoustic center.

Research across various languages has indicated a direct link between vowel reduction and several duration-related linguistic factors. These factors include stress (e.g., in Turkish: Sabev and Payne 2019), tempo (e.g., in Swedish: Lindblom 1963; in English: Fourakis 1991; in Brazilian Portuguese: Oh 2019), speaking style (e.g., in English: Chen 1980; Moon and Lindblom 1994; in Dutch: Koopmans-van Beinum 1980; in Spanish: Harmegnies and Poch-Olivé 1992), lexical length (e.g., in English: Moon and Lindblom 1994), and prosodic position (e.g., in Brazilian Portuguese: Oh 2019). However, these factors may not uniformly impact vowel reduction. For instance, Fourakis et al. (1999) showed that differences in stress conditions cause more pronounced changes in the acoustic vowel space than in tempo. It should also be noted that while stress- and tempo-related vowel reduction can shrink the acoustic vowel space, it does not necessarily make all vowels appear more centralized (Fourakis 1991). As aforementioned, the consonantal context plays a crucial role in how vowels shift within the acoustic vowel space.

However, according to the listener-oriented H&H Theory (Lindblom 1990), a key precondition for vowel reduction to occur is the presence of conducive conditions for effective communication. Essentially, the speaker adjusts their production, either under-articulating or over-articulating, in an effort to strike a balance between conserving energy and ensuring that the listener comprehends the message. Factors that may influence the speaker’s behavior, based on the perceived difficulty of message processing from the listener’s standpoint, encompass, among others, lexical frequency in conjunction with lexical neighborhood density (Lindblom 1990; Chopper and Turnbull 2018), word category (Meunier and Espesser 2011), and new/given information (Kul 2010). For instance, since listeners find it easier to process frequent and low-neighborhood-density words, as well as grammatical words and elements conveying already given information, speakers might opt for under-articulation in such cases.

1.2. Phonetic Vowel Reduction in Greek

Greek is a language with a simple vowel system, consisting of 5 vowel qualities, [i, ε, ɐ, ο, u] (Arvaniti 1999). Early research on the spectral characteristics of the Greek vowels produced by adults showed no evidence of phonological vowel reduction in Standard Modern Greek (Dauer 1980a; Nicolaidis 1990). However, it was observed that high vowels [i, u] could undergo devoicing or even complete deletion, depending on factors such as consonantal environment, prosodic position, tempo, morpheme status, lexical frequency, and specific intonation patterns, as shown by Dauer (1980b).

Further studies on adult speech (Fourakis et al. 1999; Nicolaidis 2003; Baltazani 2007; Lengeris 2012) have shed light on the nature of phonetic vowel reduction across all Greek vowel qualities. When vowel duration is shortened, an upward shift of the vowel space and compression along the front–back axis in the acoustic space have been reported (Fourakis et al. 1999). Specifically, such reduction is evident in fast compared to slow speech (Fourakis et al. 1999), in the non-focused compared to the focused condition (Fourakis et al. 1999), in spontaneous compared to read speech (Nicolaidis 2003; Lengeris 2012), in the post-stressed compared to the pre-stressed position (Baltazani 2007), in longer words compared to shorter words (Baltazani 2007), and in the unstressed compared to the stressed condition (Fourakis et al. 1999; Nicolaidis 2003; Lengeris 2012). Stress-induced vowel reduction, in particular, has been a focal point of research, resulting in its examination under various conditions: 1. Fourakis et al. (1999) used a reading task with controlled words in a carrier sentence, observing a 30% reduction in the acoustic vowel space from the stressed to unstressed conditions at a slow tempo, and a 23% reduction at a fast tempo; 2. Nicolaidis (2003) investigated this phenomenon in monologs, confirming a compression of the acoustic vowel space when duration decreased due to various factors, including stress; and 3. Lengeris (2012) examined vowel reduction in dialogs and extended text reading, and reported that the acoustic vowel space was compressed in the unstressed condition compared to the stressed one in both conditions, with the stress-induced degree of vowel reduction being only slightly more pronounced in read speech than in spontaneous speech.

1.3. Vowel Production and Vowel Reduction in Children

Children pass through various developmental stages before mastering adult speech patterns. With reference to vowels, the phonological acquisition of English non-rhotic vowels is typically completed at 3 or 4 years of age (Kent 1992; Stoel-Gammon and Pollock 2008), with an error rate in their production not exceeding 4% (Pollock and Berni 2003). Similar observations have been made for Greek, which has a simpler vowel system. Papadopoulou (2000), as cited by Mennen and Okalidou (2007), found that by age 4, 90% of Greek monolinguals have mastered Greek vowel monophthongs, while Babatsouli (2019) documented the accurate production of these vowels by ages 2;06–2;07. Additionally, Babatsouli (2020, 2021) noted an early acquisition of vowel hiatus in Greek, with its emergence by age 2;08 in disyllabic words and by age 2;10 in multisyllabic words. Such findings suggest that regardless of the complexity of the vowel system, phonological vowel acquisition is typically completed by a similar age across languages. This early vowel acquisition seems to be attributed to 1. the nature of vowels, as they are speech elements that carry prominence and are easily perceptible (Lee et al. 2010), and 2. the continuous improvement in tongue control (Kent 1992).

However, the gradual maturation of the speech motor control system until pre-adolescence/adolescence (Gerosa et al. 2006; Lee et al. 1999) and the ongoing development of the vocal tract into adulthood (Boë et al. 2007; Story et al. 2018) require continuous compensation in speech production, with phonetic vowel development ultimately reaching completion around adolescence. For example, research on English-speaking children demonstrates that due to the increasingly improved coordination of articulatory gestures over time, there is a decrease in vowel duration and temporal variability by age 12 (Gerosa et al. 2006; Lee et al. 1999), and a decrease in formant frequency variability by age 14 (Lee et al. 1999). Less intra-vowel stability and greater inter-vowel overlap in the acoustic vowel space of children compared to adults (Yang and Fox 2013) are attributed to children’s less precise jaw, lip, and tongue coordination. Interestingly, the magnitude of variability is not only influenced by speech motor control skills but also by anatomy, with increased variability being observed during periods of extensive anatomical changes (von Hofsten 1989; Smith et al. 1996; Vorperian 2000; Vorperian et al. 2005), revealing children’s efforts to adapt their articulatory targets. These anatomical changes, such as the increase in vocal fold length and width or in the length and volume of the vocal tract, also impact the acoustic outcomes, leading to a decrease in fundamental frequency and formant frequencies with age, as shown in studies on English-, Cantonese-, and Greek-speaking children (Hillenbrand et al. 1995; Huber et al. 1999; Lee et al. 1999; Lee 2016; Kelmali 2020; Kent and Rountrey 2020). Focusing on formant frequencies, Sfakianaki (2002) and Kelmali (2020), who examined Greek-speaking children up to age 10 and adults of both sexes using reading and repetition tasks, found that the higher formant frequencies in children of both sexes also result in their acoustic vowel space being positioned further to the left and lower compared to adults’.

It is also noteworthy that as individuals age, particularly during adolescence, sex-related differences become more pronounced (Fitch and Giedd 1999; Boë et al. 2007). Μales experience greater growth than females accompanied by a sharper drop in fundamental frequency and formant frequencies, resulting in sex-related differences in these acoustic characteristics from the age of 12 onward (Huber et al. 1999; Lee et al. 1999). However, some studies report significant differences in formant frequencies between male and female children before the age of 12 (Busby and Plant 1995; Lee 2016). Such evidence together with findings showing that adult females sometimes exhibited longer vowel duration (Hillenbrand et al. 1995; Ericsdotter and Ericsson 2001) and larger vowel space areas than adult males (Sfakianaki 2002; Vorperian and Kent 2007; Kelmali 2020) has been attributed to sociophonetic reasons, which is particularly a tendency of female speakers to produce clearer speech (Simpson 2009).

Despite the protracted phonetic vowel development and the observation that vowel undershoot is a late-acquired phenomenon, demanding extensive practice and refined motor abilities (Kapatsinski et al. 2020), vowel undershoot has not been extensively studied in children. Previous studies have predominantly focused on analyzing selected acoustic properties of vowels, such as duration, intensity, and fundamental frequency, to understand how children mark stress, without examining their formant frequencies (Allen and Hawkins 1980; Pollock et al. 1993; Kehoe et al. 1995; Schwartz et al. 1995). The effect of stress on spectral vowel characteristics was investigated in three German-speaking children aged 2–6 years old, but no common pattern of reduction from the stressed to the unstressed condition was identified among them (Schneider and Μöbius 2006). Stress-induced vowel reduction was also studied in 18 Greek-speaking children aged 3–7 years old and adults, and the findings showed that 1. there was centralization of unstressed vowels in the acoustic vowel space across all age groups examined; 2. there was significantly less stress-induced temporal vowel reduction in the youngest children compared to adults; 3. there was a similar degree of stress-induced spatial vowel reduction in children of all ages compared to adults; and 4. there was a weaker correlation between temporal and spatial vowel reduction in three-year-olds compared to the older age groups, due to the significantly less stress-induced temporal vowel reduction in the youngest children mentioned above (Christodoulidou et al. 2023).

The scarcity of studies on the developmental course of vowel reduction necessitates further research across different age groups including older children for two main reasons: 1. the developmental trajectory of duration, a main determinant of vowel undershoot, is not completed before adolescence (Lee et al. 1999), and 2. age-related differences in stress marking as manifested in durational differences are observable up to middle childhood. Specifically, although some studies on English-speaking children suggest that the stress contrast appears to be weaker in children only up to the age of 2–4 years (Allen and Hawkins 1980; Pollock et al. 1993; Schwartz et al. 1995), Ballard et al. (2012) have provided evidence of weak stress contrast and the preservation of long duration in the unstressed condition for iambic words even in their eldest child group, i.e., seven-year-olds.

To this end, the objectives of the present study were the following:

• To investigate the development of vowel production and vowel reduction across a wide range of ages (from 3 years old to adults) of both genders under various stress conditions (stressed, pre-stressed, and post-stressed vowels) due to their differing prosodic strengths;
• To examine the relationship between temporal and spatial stress-induced vowel reduction across ages.

In accordance with the previous literature, an effect of stress on the duration and spectral characteristics of vowels across ages is predicted. This is because children are highly sensitive to stress patterns even from infancy (Jusczyk et al. 1993, 1999), leading us to assume that children in the studied age range will have already acquired that, in Greek, the unstressed conditions are prosodically weaker compared to the stressed one (Fourakis et al. 1999; Nicolaidis 2003; Lengeris 2012) and that the post-stressed condition is prosodically weaker than the pre-stressed one (Baltazani 2007). Nevertheless, we anticipate age-related differences in the degree of stress-induced vowel reduction in the measurements of this study (relative vowel duration, normalized acoustic vowel space areas, and Euclidean distances of vowels from the centroid of the normalized acoustic vowel space). In particular, given that the developmental trajectory of absolute vowel duration is not completed before the age of 12 (Lee et al. 1999), it is likely that this will influence the developmental trajectory of relative vowel duration. As a result, we predict age-related differences in both duration and the degree of stress-induced temporal vowel reduction up to the age of 11. Moreover, even though we do not expect strong age-related variations in spectral vowel characteristics due to the use of normalized measurements, we anticipate that the age-related differences in the degree of temporal vowel reduction will also lead to age-related differences in the degree of spatial vowel reduction up to the age of 11 years, as phonetic vowel reduction involves changes in spectral vowel characteristics associated with changes in duration (Lindblom 1963; Moon and Lindblom 1994; Barnes 2006). Additionally, given the observed age-related differences in the spatiotemporal organization of speech (Christodoulidou et al. 2023), we hypothesize that young children will exhibit a weaker correlation between temporal and spatial vowel reduction compared to older children and adults. Finally, due to the use of normalized measurements, we do not anticipate strong gender-related differences, although differences in the positioning of vowels may be found. Pettinato et al. (2016), for example, indicated that females have more peripheral vowels in the normalized acoustic vowel space compared to males.

Overall, the ultimate goal of this study was to establish the first norms for vowel reduction across various ages, enhancing our understanding of the developmental path of vowel production, and contributing to theory and applied research and practice in clinical intervention and speech technology applications. In these domains, research on vowel reduction is important as it addresses the challenges encountered by 1. specific clinical populations, such as children with childhood apraxia of speech (Tubi et al. 2024), who confront difficulties in the reduction of unstressed vowels, and 2. speech synthesizers, which tend to over-articulate (Nord 1986).

2. Materials and Methods

2.1. Participants

In this cross-sectional study, we recorded a total of 72 participants. Greek-speaking children were mainly recruited from schools in the Prefecture of Thessaloniki. They were distributed in nine age groups with an equal number of participants in each, i.e., 8 individuals, 4 males, and 4 females in each age group. There were three- (3;01–3;11), five- (5;02–5;11), seven- (7;00–7;11), nine- (9;00–9;11), eleven- (11;00–11;11), thirteen- (13;00–13;07), fifteen- (15;00–15;11), and seventeen-year-olds (17;00–17;03), with adults (20–26 years old) serving as the control group. Demographic questionnaires, completed by parents, or the adult participants themselves, confirmed that all participants had been lifelong residents of the Prefecture of Thessaloniki, had Standard Modern Greek as their only mother tongue, and had no history of hearing or speech/articulation problems. Hearing screening tests were conducted using the MADSEN AccuScreen (TEOAE) by Otometrics on 70 participants (two participants were not available for the hearing screening test), and the Phonetic and Phonological Development Test (Panhellenic Association of Logopedists 1995) was administered to all participants up to the age of 7 years given that Greek phonological development stages typically conclude by the age of 6 years (Panhellenic Association of Logopedists 1995). For this test, children engaged in a picture-naming task, producing words with various phonological structures, speech sounds, and consonant clusters, to assess whether their performance matched the expected stage of phonetic and phonological development for their age.

2.2. Speech Material

The speech material consisted of 15 trisyllabic words of the form CV.CV.CV, grouped into 5 triplets. Within each triplet, comprising 3 phonetically similar words, one of the 5 Greek vowel qualities [i, ε, ɐ, o, u] was studied in the second syllable of each word. The phonetic context adjacent to the target vowel remained constant across all words in each triplet, and stress was placed on the first, second, or third syllable, leading to the examination of the target vowels in post-stressed, stressed, and pre-stressed positions (see

Table 1. Triplets of trisyllabic target words with the Greek vowels examined in pre-stressed, post-stressed, and stressed positions. Target vowels are underlined.

Vowel	Pre-Stressed	Post-Stressed	Stressed
[i]	[kodiˈnɐ] ‘nearby’	[ˈkodinɐ] ‘I got shorter’	[koˈdino] ‘to shorten’
[ε]	[polɛˈmo] ‘I battle’	[ˈpolɛmo] ‘war’	[poˈlɛmɐ] ‘(you) battle’
[ɐ]	[ɣɐlɐˈtɐ] ‘milkman’	[ˈɣɐlɐtɐ] ‘cartons of milk’	[sɐˈlɐtɐ] ‘salad’
[o]	[bɐloˈnɐ] ‘balloon seller’	[ˈbɐlonɐ] ‘I mended’	[bɐˈlono] ‘I mend’
[u]	[mɐɣuˈlɐ] ‘with huge cheeks’	[ˈmɐɣulɐ] ‘cheeks’	[mɐˈɣulɐ] ‘huge cheek’

). Each word was repeated 5 times, resulting in the analysis of a total of 5400 vowels (72 participants × 3 stress conditions × 5 vowel qualities × 5 repetitions).

Based on the phonological development stages reported by the Panhellenic Association of Logopedists (1995), the speech sounds and the phonotactic structure of the target words were suitable for all age groups under investigation. Due to constraints imposed by the aforementioned criteria for the design of the speech material, the lexical frequency of the target words varied. Further investigation of the target CV frequency in HelexKids (Terzopoulos et al. 2024), a corpus comprising 1.3 million words sourced from Greek primary education textbooks across all grades, revealed high frequency for all target syllables, i.e., at least 4630 instances for each syllable (see

Table A1. Number of observations in HelexKids for the lexical syllable CV containing the studied vowel of every test word.

Test Words	HelexKids
	CV Containing the Examined Vowel
	Unstressed	Stressed	Total
[ˈɣɐlɐtɐ]	21,615	13,883	35,498
[sɐˈlɐtɐ]	21,615	13,883	35,498
[ɣɐlɐˈtɐ]	21,615	13,883	35,498
[ˈpolɛmo]	16,506	11,613	28,119
[poˈlɛmɐ]	16,506	11,613	28,119
[polɛˈmo]	16,506	11,613	28,119
[ˈkodinɐ]	6752	3280	10,032
[koˈdino]	6752	3280	10,032
[kodiˈnɐ]	6752	3280	10,032
[ˈbɐlonɐ]	20,217	8853	29,070
[bɐˈlono]	20,217	8853	29,070
[bɐloˈnɐ]	20,217	8853	29,070
[ˈmɐɣulɐ]	2353	2277	4630
[mɐˈɣulɐ]	2353	2277	4630
[mɐɣuˈlɐ]	2353	2277	4630
MINIMUM	2353	2277	4630
MAXIMUM	21,615	13,883	35,498

in the Appendix A).

2.3. Recordings

All participants took part in a delayed repetition task. To enhance motivation, especially among younger age groups, the task was presented as a game in PowerPoint on a personal computer. During this game, participants acted as travelers who had to travel to eight islands recently discovered by a pilot. At the start of the journey, the pilot explained the game and informed participants that she would produce some words in sentences, similar to the response she typically gives when asked if she uses any particular words on these islands. The target words in the carrier phrase Λέω το ____ παντού [ˈlεo to ____ pɐˈdu] ‘I say the (word) ____ everywhere’ were played out to participants in random order. Participants listened to the pre-recorded audio prompt with the target word in the carrier phrase and saw a picture of the target word on the computer screen. Literate participants additionally saw the sentence in orthographic form for convenience, but they were not asked to read the item. The pre-recorded pilot’s question Τι τους απαντώ; [ˈti tus ɐpɐˈdo] ‘What do I answer to them?’ followed, after which participants produced the experimental sentence with the target word at a comfortable speech rate and intensity level. Upon producing a series of words within the carrier phrase, participants completed their journey on each island and were then able to proceed to the next one.

After a familiarization phase with the speech material and the task, recordings were conducted in small quiet rooms either at schools or in participants’ homes. For the recordings (sampling rate: 44,100 Hz), a Marantz PMD661 MKII recorder and an AKG C1000 S microphone were used, placed approximately 15 cm from the participants’ mouths.

The experiment adhered to the principles of the Declaration of Helsinki and received approval from the Research Ethics Committee of the Aristotle University of Thessaloniki (Protocol Code: 296445/2021; Date of Approval: 8 December 2021), and the Directorate of Primary Education of Western Thessaloniki (Protocol Code: 2125; Date of Approval: 14 February 2022) for conducting the research in school premises. Furthermore, informed consent forms were signed by either the participants themselves or their guardians (in the case of minors).

2.4. Measurements

2.4.1. Acoustic Analysis

The acoustic analysis was performed using Praat (Boersma and Weenink 2023). Word and vowel segmentation were carried out manually. Vowel segmentation was based on typical criteria, i.e., the onset and offset of vowel formant structure, with segmentation lines inserted at the nearest zero crossing (Barbier et al. 2015; Abakarova et al. 2018; Noiray et al. 2018). Target words were segmented from the end of the formant structure of [to] in the carrier phrase to the end of the formant structure of the final vowel of the test word (see Figure 1). Based on these boundaries, we extracted the absolute vowel and word duration to calculate normalized/relative vowel duration using the following formula: (absolute vowel duration/absolute word duration) × 100.

In addition, the formant frequencies F₁ and F₂ were extracted from the temporal midpoint of each target vowel. These formant frequencies were then normalized using the normLobanov function in R (R Core Team 2022), and the code employed for this process can be found in the Supplementary Material. The Lobanov normalization method was used, because, according to Adank et al. (2004), it is the most effective method for controlling for anatomical differences and preserving phonetic variation. Subsequently, the normalized formant frequencies were rescaled to Hertz-like values using the formulas proposed by Thomas and Kendall (2024):

F′₁ = 250 + 500 (F^N₁ − F^N_1MIN)/(F^N_1MAX − F^N_1MIN)

(1)

F′₂ = 850 + 1400 (F^N₂ − F^N_2MIN)/(F^N_2MAX − F^N_2MIN),

(2)

where F′ denotes the rescaled normalized formant frequencies, and F^N represents the normalized formant frequencies of each vowel observation. These rescaled normalized formant frequencies resemble the non-normalized formant frequencies of adult males reported by Nicolaidis (2003).

The extraction of rescaled normalized F′₁ and F′₂ values was carried out to calculate normalized vowel space areas, following the method outlined by Fourakis et al. (1999). This involved partitioning the acoustic vowel space into three triangles and determining the total area by aggregating the areas of these individual triangles. Through this approach, a total of 1080 normalized vowel space areas were computed and subjected to analysis (72 participants × 3 stress conditions × 5 repetitions).

Since areas can only measure the size of the acoustic vowel space without providing information about vowel centralization, we also computed the Euclidean distance of each vowel token from the centroid of the acoustic space using the rescaled normalized F′₁ and F′₂ values. The centroid of the acoustic space was determined by calculating the mean values of F′₁ and F′₂ for all vowel tokens, which were consistent for all speakers ($\overline{{F^{\prime }}_1}$ = 432.3; $\overline{{F^{\prime }}_2}$ = 1483.7) because F′₁ and F′₂ were Lobanov-normalized and subsequently rescaled. The Euclidean distances were derived using the formula

$$ed=\sqrt{{({F^{\prime }}_1-\overline{{F^{\prime }}_1})}^2+{({F^{\prime }}_2-\overline{{F^{\prime }}_2})}^2},$$

(3)

where $\overline{{F^{\prime }}_1}$ and $\overline{{F^{\prime }}_2}$ represent the mean values of F′₁ and F′₂ for all vowel tokens of each speaker separately. Larger Euclidean distances indicated that the vowel token was farther away from the centroid of the acoustic space.

As mentioned above, normalized values were chosen for all measurements (duration, vowel space areas, and Euclidean distances) to examine vowel reduction. They were deemed the most appropriate for the following reasons:

• By controlling for differences in anatomy and speech rate in the normalized values, we can directly infer in which stress condition children face the greatest difficulties in reduction;
• Without controlling for differences in anatomy and speech rate that normalized values allow, the correlation between temporal and spatial vowel reduction is problematic because each speaker has a different speech rate and different sizes in the acoustic space, making it impossible to see in the overall data of each age group how a change in duration due to the stress conditions results in a change in spectral characteristics;
• Examining only the normalized values for the phenomenon of vowel reduction is sufficient because, as shown in Christodoulidou et al. (2023), the degree of vowel reduction between the stress conditions and the observed age-related differences in this degree do not depart extensively from non-normalized measurements.

2.4.2. Statistical Analysis

The data underwent statistical analysis in R (R Core Team 2022), with the code provided in the Supplementary Material. We employed linear mixed-effects models (nlme::lme function), incorporating age, gender, stress, and vowel (not for normalized vowel space areas) as fixed-effects factors, with speaker as a random-effects factor and repetitions as cases. Word was not included as a random-effects factor because it did not enhance the model for duration, caused singularity issues in the model for Euclidean distances, and was not applicable to the model for vowel space areas since calculating an area involved combining data from 5 Greek vowels and thus data from 5 different words. Among the fixed-effects factors, age and gender were between-subjects factors, while stress and vowel were within-subjects factors. In these models, some of the fixed-effects factors exhibited heterogeneity, which was accounted for using the varIdent function. To refine the models, a full-factorial analysis was conducted, finally retaining the statistically significant variables with an effect size (η_p²) of 0.01 or greater, as determined using the effectsize::eta_squared function (η_p² values of 0.01 indicate a small effect, 0.06 a medium effect, and 0.14 a large effect). Further analysis was carried out using Tukey’s post hoc tests through the emmeans function.

To explore the relationship between the dependent variables under investigation and age, regression analyses were employed using the lm function. In these analyses, age was included as a numerical variable, with values assigned as follows: 1 for three-year-olds, 2 for five-year-olds, and so forth, up to 9 for adults.

Additionally, the degree of vowel reduction between the stress conditions—defined as the decrease in the values of the dependent variables from one stress condition to another (stressed vs. pre-stressed, stressed vs. post-stressed, and pre-stressed vs. post-stressed)—was calculated as percentages. For normalized vowel duration, this percentage change for each pair of stress conditions was computed per speaker and vowel (8 speakers per age group × 5 vowels = 40 observations per age group), and for normalized acoustic vowel areas, it was calculated per speaker and repetition (8 speakers per age group × 5 repetitions = 40 observations per age group). This dataset size enabled comparisons to detect significant age-related differences between adults and each of the other age groups using Wilcoxon tests with the wilcox.test function (small effect: r < 0.3; moderate effect: 0.3 ≤ r ≤ 0.5; strong effect: r > 0.5).

Finally, the relationship between relative vowel duration and normalized vowel space areas and the relationship between relative vowel duration and Euclidean distances were examined using Pearson’s correlations with the cor.test function (8 participants per age group × 3 stress conditions × 5 repetitions = 120 observations per age group). The correlation coefficient (r) was interpreted as follows: r < 0.2 as very low correlation; 0.2 ≤ r < 0.4 as low correlation; 0.4 ≤ r < 0.6 as moderate correlation; 0.6 ≤ r < 0.8 as strong correlation; and r ≥ 0.8 as very strong correlation.

3. Results

3.1. Normalized/Relative Vowel Duration

The regression analysis showed a significant decrease in relative vowel duration as age increased (p = 0.0001, R²_adj = 0.8801) (see Figure 2). The main effect of age was significant in the linear mixed-effects model analysis (F (8, 63) = 4.714, p = 0.0001, η_p² = 0.37), with Tukey’s post hoc tests showing that adults exhibited significantly shorter relative vowel duration than three-, five-, and nine-year-olds.

The main effect of gender was not statistically significant, but the main effect of stress was (F (2, 5306) = 5835.93, p < 0.0001, η_p² = 0.69), as relative vowel duration decreased across the stress conditions in the following order: stressed (20) > pre-stressed (14) > post-stressed (12.1). Figure 3 illustrates the effect of stress across all age groups (Age × Stress interaction: F (16, 5306) = 23.434, p < 0.0001, η_p² = 0.07), revealing that the degree of temporal reduction between the stress conditions was not consistent. The degree of vowel reduction from the stressed to unstressed conditions increased until the age of 11 years, dropped sharply at the age of 13 years, and remained relatively similar at the older ages, while the degree of vowel reduction between the unstressed conditions showed continuous fluctuations with an increase in age. Wilcoxon tests indicated the following: 1. The degree of vowel reduction between the stressed and pre-stressed conditions (18.4–39.2%) was significantly different in adults (27.7%) compared to three- (18.4%, W = 440, p = 0.00054, r = 0.387), nine- (33.3%, W = 1007, p = 0.047, r = 0.223), and eleven-year-olds (39.2%, W = 1222, p < 0.0001, r = 0.454). 2. The degree of vowel reduction between the stressed and post-stressed conditions (26.9%–48.2%) significantly differed in adults (39.8%) compared to three- (26.9%, W = 333, p < 0.0001, r = 0.502) and eleven-year-olds (48.2%, W = 1179, p = 0.00027, r = 0.408). 3. No significant age-related differences in the degree of vowel reduction between the unstressed conditions (10.4–18.5%) were noted. The age-related differences in the degree of vowel reduction between the stressed and unstressed conditions stemmed from the gradual decrease in relative vowel duration in the unstressed conditions and the gradual increase in the stressed condition until 11 years. Tukey’s post hoc tests revealed that in the unstressed conditions (pre-stressed and post-stressed vowels), three-year-olds had significantly longer relative vowel duration than adults, while in the stressed condition, seven-, nine-, and eleven-year-olds notably prolonged vowel duration, resulting in a significantly longer relative vowel duration than adults.

Finally, the main effect of vowel reached statistical significance (F (4, 5306) = 232.838, p < 0.0001, η_p² = 0.15). Specifically, Tukey’s post hoc tests demonstrated that relative vowel duration decreased in the order [ɐ] (16.9) > [o] (15.8) > [ε] (15.4) > [i] (14.5) = [u] (14.3), where only the difference between [i] and [u] was not statistically significant. No interaction with stress or age group was statistically significant.

3.2. Normalized Vowel Space Areas

A significant increase in normalized vowel space areas with age was shown by the regression analysis (p = 0.0047, R²_adj = 0.6619) (see Figure 4). The main effect of age was statistically significant in the linear mixed-effects model analysis (F (8, 54) = 4.411, p = 0.0004, η_p² = 0.4), with Tukey’s post hoc tests revealing that three-year-olds exhibited significantly smaller areas than adults. Moreover, the main effect of gender reached statistical significance (F (1, 54) = 4.936, p = 0.0305, η_p² = 0.08), with females (74,779) exhibiting significantly larger normalized vowel space areas than males (71,469). The statistically significant Gender × Stress interaction (F (2, 972 = 6.497, p = 0.0016, η_p² = 0.01) showed that these gender-related differences were significant only in the stressed condition.

Additionally, the main effect of stress was statistically significant (F (2, 972) = 1502.453, p < 0.0001, η_p² = 0.76). Normalized vowel space areas decreased in the order stressed (112,412) > pre-stressed (61,624) > post-stressed condition (45,336), with vowels in the unstressed conditions falling within the normalized vowel space of the stressed condition. The pattern of vowels being compressed along both the high–low and front–back axes in the unstressed conditions compared to the stressed one was also observed across both genders (see Figure 5) and all age groups. The statistically significant Age × Stress interaction (F (16, 972) = 2.215, p = 0.0039, η_p² = 0.04) indicated that normalized vowel space areas decreased in the order stressed > pre-stressed > post-stressed condition at every age except for three- and five-year-olds, where the difference between the unstressed conditions did not reach statistical significance (see Figure 6). The degree of vowel reduction between the stress conditions varied with age (stressed vs. pre-stressed conditions: 29.9–48.5%; stressed vs. post-stressed conditions: 50–61.8%; pre-stressed vs. post-stressed conditions: 13.7–27.4%) (see Figure 6), but Wilcoxon tests did not reveal significant differences in vowel reduction between children of any age group and adults due to the high within-subject variability in reduction percentages. Despite the lack of statistically significant age-related differences in the degree of vowel reduction, changes in normalized vowel space areas were observed with age across the stress conditions. It is interesting to note that in both the stressed and post-stressed conditions, a gradual increase in normalized vowel space areas was observed up to the age of 11 years, with slight fluctuations at older ages. However, significant age-related differences were only present in the stressed condition, with three-year-olds exhibiting significantly smaller normalized vowel space areas than adults.

The significant Age × Gender × Stress interaction (F (16, 972) = 1.911, p = 0.0164, η_p² = 0.03) finally showed that in most age and gender groups, the pattern stressed > pre-stressed = post-stressed condition was observed, even though in each age and gender group, the pre-stressed condition tended to have larger areas than the post-stressed one. The reduction in statistically significant differences between the unstressed conditions at this point of the analysis seems to be attributed to the high variability observed in normalized vowel space areas.

3.3. Euclidean Distances

The results showed that the main effect of vowel was statistically significant (F (4, 5157) = 6178.43, p < 0.0001, η_p² = 0.83); Tukey’s post hoc tests showed that Euclidean distances from the centroid of the acoustic vowel space decreased in the order [i] (432.5) > [u] (389.5) > [o] (166.4) > [ε] (142.9) > [ɐ] (125.6), where all differences among vowels were statistically significant. However, different interactions (Age × Vowel: F (32, 5157) = 5.58, p < 0.0001, η_p² = 0.03; Age × Gender × Vowel: F (45, 5157) = 2.54, p < 0.0001, η_p² = 0.02) indicated various patterns in Euclidean distances among vowels across different ages and genders. The only consistent observation across age and gender groups was that the high vowels [i, u] had significantly larger Euclidean distances compared to the other vowel qualities, [o, ε, ɐ] (see Figure 7 (bottom) for the mean Euclidean distances across different age groups).

Additionally, the main effect of stress was statistically significant (F (2, 5157) = 1500.59, p < 0.0001, η_p² = 0.37), with Tukey’s post hoc tests showing significantly larger Euclidean distances for vowels in the stressed condition (304.5) compared to the pre-stressed (239.6) and the post-stressed condition (209.9). In addition, vowels in the pre-stressed condition showed significantly larger distances compared to vowels in the post-stressed condition. The statistically significant Stress × Vowel interaction (F (8, 5157) = 36.13, p < 0.0001, η_p² = 0.05) confirmed this pattern for [i, ε, u]; for [o, ɐ], there were no significant differences between the two unstressed conditions (i.e., stressed > pre-stressed = post-stressed). Tukey’s post hoc tests for the statistically significant Age × Stress × Vowel interaction (F (80, 5157) = 1.64, p = 0.0003, η_p² = 0.02) revealed the tendency of [i, u] to have significantly larger Euclidean distances from the non-high vowels [o, ε, ɐ] in each stress condition, significantly smaller Euclidean distances for unstressed than stressed vowels, and in cases larger Euclidean distances for pre-stressed than post-stressed vowels across all ages. Figure 8 and Figure 9 illustrate the F′₁ × F′₂ acoustic vowel spaces and the mean Euclidean distances of vowels by stress condition, as well as by stress condition and age group, respectively.

Finally, the main effects of age and gender did not reach significance, but as shown above, their interactions with other variables demonstrated distinct patterns concerning stress conditions and vowels across different age groups and genders.

3.4. Relationship between Temporal and Spatial Vowel Reduction

To explore the relationship between temporal and spatial vowel reduction, this subsection presents Pearson’s correlations 1. between relative vowel duration and normalized vowel space areas and 2. between relative vowel duration and Euclidean distances from the centroid of the acoustic space, calculated using rescaled Lobanov-normalized F′₁ and F′₂ values. Normalized values (for duration, vowel space areas, and Euclidean distances) were used to control for variations in speech rate and anatomical differences.

3.4.1. Correlations between Relative Vowel Duration and Normalized Vowel Space Areas

Figure 10 shows strong to very strong Pearson’s correlations between relative vowel duration and normalized vowel space areas across ages. In particular, strong correlations were evident in the younger age groups, namely in three- and five-year-olds (r = 0.62 and 0.64, respectively), followed by adults (r = 0.71), while very strong correlations were found in the remaining age groups (r = 0.8–0.87). In spite of the differences in the correlation coefficient across ages, a consistent pattern emerged across all age groups: For both relative vowel duration and normalized vowel space areas, the stressed condition consistently exhibited the highest values, followed by the pre-stressed condition, and then the post-stressed one.

3.4.2. Correlations between Relative Vowel Duration and Euclidean Distances

The Pearson’s correlation results between relative vowel duration and Euclidean distances were similar to those of the correlations between relative vowel duration and normalized vowel space areas. As shown in Figure 11, strong correlations were evident for three- and five-year-olds (r = 0.57 and 0.62, respectively), followed by adults (r = 0.72), and then very strong correlations were found for the other child groups (r = 0.81–0.85). Moreover, across all age groups, higher values were observed for both relative vowel duration and Euclidean distances in the stressed condition, followed by the pre-stressed and then the post-stressed conditions.

4. Discussion

The present study examined temporal and spatial reduction in Greek across a wide age range (3 years to adulthood) and for both genders. We analyzed vowels under three stress conditions, i.e., stressed, pre-stressed, and post-stressed, embedded in trisyllabic words of the form CV.CV.CV. To investigate vowel reduction minimizing the potential effects of individual differences in speech rate and anatomy, we examined vowel duration, acoustic vowel space areas, and Euclidean distances using normalized measurements.

4.1. Vowel Development

The findings highlight significant differences between children and adults in terms of both temporal and spatial vowel characteristics. These differences are discussed in detail below, providing a deeper understanding of how vowel production varies across age groups.

Regarding relative vowel duration, the results show a general decrease with age, both in the overall dataset (see Figure 2) and in the unstressed conditions (see Figure 3), suggesting advancements in the neuromuscular control of speech articulators. This maturation of the speech motor control system is further supported by other acoustic studies that revealed a decrease in segmental duration and its variability (Smith et al. 1996; Lee et al. 1999; Gerosa et al. 2006), along with an increase in articulation rate with age (Mahr et al. 2021; Kallay et al. 2022). Articulatory research provides more direct evidence of this maturation by showing increased velocity and reduced token-to-token variability in the movement of speech articulators with age (Green et al. 2002; Walsh and Smith 2002; Cheng et al. 2007; Green and Nip 2010; Zharkova 2017). Studies on anticipatory coarticulation further highlight progress in the coordination of articulatory movements as children grow older (Zharkova et al. 2008; Zharkova 2017, 2018), reflecting not only advancements in speech motor control but also improvements in cognitive skills such as organization and speech planning (Barbier et al. 2015). Despite this ongoing improvement with age, the stressed condition exhibited a gradual increase in relative vowel duration up to age 11. This finding suggests children’s growing awareness of the role of duration in marking stress in Greek (Botinis 1989; Arvaniti 2000), leading to an enhancement of the stress contrast. Due to this contrasting developmental pattern in relative vowel duration between the stressed and unstressed conditions, only three-, five-, and nine-year-olds differed significantly from adults in the pooled data. Nevertheless, significant differences between children and adults persisted up to age 11 in the stressed condition, suggesting that the developmental trajectory for relative vowel duration is generally completed at age 13, which is consistent with the findings for absolute vowel duration by English-speaking children (Lee et al. 1999). Finally, it is worth noting that in spite of age variations in relative vowel duration, a common pattern ([ɐ] > [i, u]) also emerged across ages, indicating that even younger children have acquired the intrinsic duration of vowels, which is consistent with Lee et al. (1999), Buder and Stoel-Gammon (2002), and Christodoulidou et al. (2023). This early acquisition may be attributed to the fact that intrinsic vowel duration is related to physiological factors: The longer duration observed in the low vowel [ɐ] results from the extensive jaw-lowering movement required for its articulation (Lehiste 1970).

Concerning spatial vowel characteristics, our results show an increase in the acoustic vowel space across the dataset (see Figure 4) and specifically in the stressed condition (see Figure 6), revealing the production of increasingly more peripheral vowels with age. These findings contrast with those of Vorperian and Kent (2007), who reported a decrease in the non-normalized acoustic vowel space with age. This discrepancy can be attributed to our use of normalized data, which minimized variation due to anatomical differences. The increase in the normalized acoustic vowel space with age in our study likely reflects the ongoing improvement in children’s speech motor control skills, as discussed above, with the increasingly more peripheral vowel positions revealing the gradual decrease in the stiffness of the motor system with age. In this developmental trajectory, Tukey post hoc tests indicated significant differences between three-year-olds and adults in normalized vowel space areas, both across the dataset and in the stressed condition, while Euclidean distances did not show any significant age-related differences, likely because they assessed vowel centralization by vowel quality individually, which may have obscured the broader trend. In the unstressed conditions, neither measurement revealed significant age-related differences, probably because the absence of stress is linked to increased system stiffness (Cooke 1980), which is a typical characteristic of less mature speech motor control (Redford 2019) and therefore does not pose notable difficulties for children. Consequently, both measures indicate that the developmental trajectory for normalized spatial vowel characteristics is completed very early, by around age 5. These results align with those of Ostry et al. (1984), who found that stiffness regulation development is completed at ages 5 or 6, and suggest that the developmental trajectory for normalized spatial vowel characteristics is completed earlier than for relative vowel duration, suggesting developmental differences in timing control versus the attainment of vowel targets in the F′₁ × F′₂ acoustic space.

4.2. Development of Vowel Reduction

Children of all ages in this study differentiated the stressed and unstressed conditions and, in line with Dauer’s (1980b), Nicolaidis’s (1990), and Baltazani’s (2007) findings for adults, they also differentiated the pre-stressed and post-stressed conditions. Relative vowel duration, normalized vowel space areas, and Euclidean distances decreased from the stressed to pre-stressed to post-stressed conditions, showing that the prosodic strength of the stress conditions—where the stressed condition is prosodically the strongest and the post-stressed condition is the weakest (Baltazani 2007)—is acquired very early. This early acquisition of stress-related prosodic strength aligns with the sensitivity to stress found in infants (Jusczyk et al. 1993, 1999), which can be explained by the exposure to prosody, including stress patterns, that children experience even while in the womb (DeCasper and Spence 1986). The difference between the stressed and unstressed conditions was consistently significant across all ages, but the difference from the pre-stressed to post-stressed conditions was not significant in normalized vowel space areas for all age groups (i.e., three- and five-year-olds), and was often not significant in Euclidean distances. However, given the similar degree of spatial vowel reduction between the age groups of three- and five-year-olds (13.9–24.3%) and the older ones (13.7–27.4%), the lack of significant differences between these stress conditions in vowel space areas is due to the increased variability in vowel space areas among younger children (see Figure 6). For Euclidean distances, it can be due to the analysis being performed based on both age and vowel quality, rather than age alone, which can result in a weakening of the observed differences.

The degree of vowel reduction between the stress conditions was studied separately for each age group, focusing on both temporal and spatial vowel characteristics. Similar to Christodoulidou et al. (2023), who examined vowel reduction in Greek-speaking children up to age 7 years, significant age-related differences were found in stress-induced temporal vowel reduction, but not in spatial vowel reduction. It should be noted that 3-year-olds did show considerable differences in spatial vowel reduction (stressed vs. pre-stressed: 29.9%; stressed vs. post-stressed: 50%) compared to adults (stressed vs. pre-stressed: 43.2%; stressed vs. post-stressed: 61%); however, no significant differences in vowel reduction between children of any age group and adults were found due to the high within-subject variability in reduction percentages. This variability stems from the high variability in vowel space areas across all the stress conditions, which is likely a product of the coarticulation of vowels with the neighboring lingual consonant(s). In terms of temporal vowel reduction, as described in Section 4.1, children progressively reduced relative vowel duration in the unstressed conditions and increased it in the stressed condition up to age 11, leading to significant age-related differences in vowel reduction being observed only when comparing the stressed to unstressed conditions (no age-related differences in vowel reduction were observed between the unstressed conditions). In particular, there was a continuous increase in the degree of vowel reduction between the stressed and unstressed conditions up to age 11 (stressed vs. pre-stressed: 18.4–39.2%; stressed vs. post-stressed: 26.9–48.2%), which revealed a trend toward less vowel reduction in younger ages and overmarking of the stressed/unstressed contrast in pre-adolescence. By age 13, children achieved adult-like relative vowel duration across all the stress conditions, completing the developmental trajectory of stress-induced temporal vowel reduction (stressed vs. pre-stressed: 29.4% for 13-year-olds and 27.7% for adults; stressed vs. post-stressed: 38.3% for 13-year-olds and 39.8% for adults). Further research on vowel reduction under different conditions, such as fast tempo or spontaneous speech—which have shown different characteristics from careful speech (Fourakis et al. 1999; Nicolaidis 2003; Lengeris 2012) and further challenge children’s neuromuscular skills—is needed to provide a more comprehensive examination of the developmental trajectory of vowel reduction.

Moreover, in this study, strong to very strong positive correlations between relative vowel reduction and normalized acoustic vowel space areas were observed across all ages. While weaker correlations were evident in three- and five-year-olds, stronger correlations prevailed in other age groups compared to adults, suggesting, similar to Christodoulidou et al. (2023), variations in the spatiotemporal organization of speech across ages. In spite of these age-related differences in the organization of speech, Lindblom’s (1963) and Moon and Lindblom’s (1994) models regarding the pivotal role of duration in vowel undershoot found empirical support even beyond adult populations: A decrease in duration from the stressed to pre-stressed to post-stressed conditions resulted in a decrease in acoustic vowel space areas in these conditions across all ages. Further correlations between relative vowel duration and Euclidean distances of vowels from the centroid of the acoustic vowel space for all ages revealed that there was centralization of vowels when duration decreased. In all age groups, vowels centralized from the stressed to unstressed conditions, and in some cases, from the pre-stressed to post-stressed conditions, resulting in an overall decrease in Euclidean distances in the following sequence: stressed > pre-stressed > post-stressed conditions. Specifically, in line with Fourakis et al. (1999), Nicolaidis (2003), Lengeris (2012), and Christodoulidou et al. (2023), an upward shift in the acoustic vowel space, particularly for non-high vowels, and a compression of the acoustic space along the F′₂ axis were observed in the unstressed conditions compared to the stressed one (see Figure 8 and Figure 9). According to Dauer (1980b), the upward shift of the acoustic vowel space in the unstressed conditions is likely due to physiological factors: The high tongue position associated with the neighboring lingual consonants, along with the reduced vowel duration in the unstressed conditions, contributes to a decreased extent of jaw drop for the vowel. The compression of the acoustic space along the F′₂ axis can also be related to reduced duration influencing the degree of tongue displacement toward more peripheral target-like positions along the horizontal axis. It can also be related to increased gestural overlap between consecutive segments in the unstressed condition in line with Moon and Lindblom’s (1994) model, given that the aforementioned studies, including the present one, examined consonantal environments that do not inhibit centralization. For example, in this study, the presence of at least one alveolar consonant—with an F₂ locus around 1700–1800 Hz (Ladefoged and Johnson 2010)—in the immediate environment of all examined vowels, combined with a bilabial consonant—characterized by a low F₂ locus (Ladefoged and Johnson 2010)—in the case of [ε] may drive all unstressed vowels, depending on their vowel quality and corresponding formant values, toward the center of the acoustic vowel space. Even though further research is needed to pinpoint the cause of F₂ axis compression, it is important to note that these findings contradict Fourakis’ (1991) claim, since they show that stress can affect not only vowel space size but also the centralization of vowels.

4.3. Gender-Related Differences

In this research, gender-related differences emerged in normalized acoustic vowel space areas only, with females exhibiting larger areas than males, particularly in the stressed condition where vowels occupy peripheral positions. Typically, these differences in normalized vowel space areas are attributed to sociophonetic factors, with females employing more extensive articulatory movements to enhance speech clarity (Pettinato et al. 2016). Nonetheless, anatomical considerations may provide an alternative explanation. Males may need to execute more extensive articulatory movements than females to achieve the desired acoustic targets due to their larger vocal tracts (Boë et al. 2007; Vorperian and Kent 2007). This requirement might pose challenges, despite males’ faster articulatory movements compared to females’ (Kuehn and Moll 1976).

5. Conclusions

This study explores the developmental trajectory of vowel production, focusing on stress-induced vowel reduction in Greek, and provides insights into potential factors contributing to age-related differences in the production of vowels. It shows that the continuous changes in speech motor control, system stiffness, and stress marking management with age lead to the acquisition of vowel production and reduction by early adolescence, with adult-like spatial vowel characteristics emerging earlier than temporal ones. Additionally, it reports significant correlations between temporal and spatial vowel reduction, in line with the undershoot model (Lindblom 1963; Moon and Lindblom 1994). The findings show that a decrease in vowel duration in the unstressed conditions results in a decrease in the vowel space area and a displacement in a higher position. The shrinkage of the vowel space is exhibited as increased vowel centralization (but not schwalization) from the stressed to pre-stressed to post-stressed conditions across ages.

Despite the relatively small number of participants per age group, this study shows consistent developmental patterns in vowel production and stress-induced vowel reduction. Moreover, its findings are particularly important as it is the first time that stress-induced temporal and spatial vowel reduction was examined from early childhood to adulthood in Greek, providing norms crucial for both clinical intervention and speech technology. Future research, including additional acoustic parameters (F0, intensity, examination of variability, etc.) across different languages, will offer further insights into the mechanisms underlying the developmental path of vowel production and vowel reduction, elucidating ‘universal’ vs. language-specific patterns.