The Acoustic Properties of Vowels in Foreigner-Directed Speech: Insights from Speech Directed at Foreign Domestic Helpers ^†

Abstract

This study examines the acoustic properties of vowels in foreigner-directed speech (FDS) in interactions between female Omani-Arabic-speaking employers and their foreign domestic helpers (FDHs). Particularly, it investigates whether Arabic corner vowels /i:/, /a:/, and /u:/ undergo acoustic adaptations in FDS. The study also explores the influence of foreign interlocutors’ psycholinguistic characteristics, such as degree of foreign accent, religion, and length of residence (LoR), on the extent of these adaptations. Data were collected from 22 Omani-Arabic-speaking women interacting with their 22 FDHs and with a native speaker (NS) confederate using a spot-the-difference task. Acoustic measures including vowel space area, formant frequency measures (F1 and F2), fundamental frequency (f0), intensity, and duration were compared across speech directed at FDHs and the NS. The results revealed that FDS exhibited greater vowel space expansion, higher F1, and increased pitch (f0) and intensity compared to speech directed at the NS confederate. However, FDS did not significantly affect F2 values. Unexpectedly, vowel duration in FDS was shorter than in speech directed at the NS. Furthermore, the psycholinguistic factors of foreign interlocutors had no significant effect on vowel space expansion in FDS. These findings provide evidence that FDS is characterized by heightened prosodic and acoustic features, potentially contributing to clearer speech. Additionally, the study highlights that NSs employ FDS when interacting with foreigners perceived to have a foreign accent.

1. Introduction

Native speakers (NSs) frequently adjust their pronunciation when interacting with non-native speakers (NNSs). In particular, foreigner-directed speech (FDS) often involves modifications to prosodic and segmental aspects of speech compared to native-directed speech (NDS). Understanding these adjustments is crucial, as they are believed to enhance the comprehensibility and intelligibility1 of the target language. However, the segmental and prosodic characteristics of FDS remain understudied.

FDS is often compared to infant-directed speech (IDS) because foreigners, like infants, have limited linguistic competence in the target language and may benefit from modified input (Bradlow & Bent, 2002; Uther et al., 2007). Similarly to IDS, FDS is believed to serve a linguistic-didactic function (Uther et al., 2007; Scarborough et al., 2007) and an attentional role through pitch and pitch contours (Rothermich et al., 2019). Piazza et al. (2023) also provided evidence of FDS facilitating both word recognition and word learning.

FDS is generally characterized by clearly articulated segments (Hatch et al., 1978), a slower speech rate (Hatch et al., 1978; Biersack et al., 2005), segment lengthening (Biersack et al., 2005), and vowel space expansion (Uther et al., 2007; Scarborough et al., 2007; Hazan et al., 2015) compared to NDS. However, findings on the acoustic-phonetic and prosodic features of FDS have been inconsistent, primarily due to methodological differences (see Piazza et al., 2022, for a review). Some studies examined natural speech produced by NSs interacting with a foreigner confederate (Uther et al., 2007; Knoll & Costall, 2015; Kangatharan, 2015; Rodriguez-Cuadrado et al., 2018), while others analyzed elicited speech from speakers imagining themselves addressing a foreigner (Knoll & Scharrer, 2007; Scarborough et al., 2007). Several studies have specifically investigated the acoustic properties of vowels in FDS to determine whether it exhibits vowel hyperarticulation compared to NDS (e.g., Uther et al., 2007; Knoll & Scharrer, 2007; Scarborough et al., 2007).

One theoretical model that may explain variation in the quantity and quality of phonetic adjustments in FDS is Communication Accommodation Theory (CAT). A core concept of this theory is that individuals adjust their speech using communicative strategies such as convergence and divergence to reflect their attitudes toward their interlocutors and social groups (Giles & Ogay, 2007). According to CAT, interaction is goal-driven, with speakers aiming to achieve objectives such as effective communication, enhanced comprehensibility, social approval, or maintaining social distinctiveness (Coupland & Giles, 1988; Giles & Gasiorek, 2013). Several studies suggest that FDS reflects these goals, particularly the NS’s desire for mutual intelligibility and effective communication (e.g., Tarone, 1980; Freed, 1981).

Another key aspect of CAT concerns how speakers perceive their interlocutors, which may be based on observed behavior or stereotypes (Coupland & Giles, 1988). Zuengler (1991) argues that NSs’ perceptions of foreigners are shaped by two main factors: ethnicity and linguistic competence. Ethnic and cultural differences may lead NSs to either maintain distinctiveness by avoiding FDS or reduce social distance by adopting it. However, foreigners may not always be associated with a specific ethnic group but rather viewed collectively as “foreign” (Harder, 1980; Janicki, 1986). Beebe and Giles (1984) suggest that NSs may perceive NNSs as having lower social status due to their foreignness.

The interactional goals of NSs can also shift based on the perceived linguistic competence of the NNSs. Studies indicate that speakers are more likely to use FDS with NNSs who have limited communicative abilities than with those who are more proficient (Scarcella & Higa, 1982; Long, 1983; Ellis, 1985). Although the current study does not specifically examine the role of ethnicity or social status in FDS adjustments, CAT provides a useful framework for interpreting modifications influenced by the perceived foreign accent of the interlocutors. The central objective of this study is to determine whether speech directed at FDHs exhibits phonetic adjustments aimed at enhancing clarity due to their perceived foreign accent.

Another useful model for explaining phonetic modifications in FDS is the hyper- and hypo-articulation (H&H) theory, which posits that speakers adjust their linguistic output based on their communicative goals and the listener’s needs. Lindblom (1990) argues that speakers continuously assess their listeners’ level of understanding and modify their speech accordingly, adjusting phonemes, words, and syllables as needed. This adaptation occurs along a continuum, with highly articulated forms (hyper-speech) at one end and more reduced, less effortful articulations (hypo-speech) at the other (Lindblom, 1990).

Hyper-speech is characterized by increased duration and amplitude of articulatory gestures, with minimal overlap between them, whereas hypo-speech involves greater temporal overlap and more co-articulation (Tabain, 2003; Lindblom, 1990). In hyper-mode, speech segments are more distinct and reach their target form, while hypo-mode results in more reduction. Based on this theory, NSs interacting with foreigners likely assess their interlocutors’ linguistic needs and adjust their speech along the hyper/hypo continuum to enhance comprehension.

Guided by these theoretical insights, this study examines the acoustic properties of vowels in FDS within an understudied context: interactions between female employers and their foreign domestic helpers (FDHs) in a near-naturalistic setting. This context is unique as it reflects real-life interactions between NSs and foreigners who communicate daily. The study investigates whether FDS in this setting exhibits vowel modifications or acoustic enhancements compared to NDS.

2. The Acoustic Properties of Vowels in Foreigner-Directed Speech

Among the pioneering studies on FDS using natural interactions is that of Uther et al. (2007), who investigated whether vowel hyperarticulation (i.e., vowel space expansion) occurs in FDS compared to IDS and adult-directed speech (ADS). They recorded speech from ten Southern British-English mothers interacting with their infants (aged 4 months to 1 year), a British adult confederate, and two foreign-accented Chinese adult confederates. The results revealed significant vowel hyperarticulation in both IDS and FDS relative to ADS, with no significant difference between the vowel spaces of FDS and IDS. The authors concluded that vowel hyperarticulation, and its role in facilitating language learning in IDS, is generalizable to FDS. They also found that vowels in IDS were significantly longer in duration than those in ADS, while vowel duration in FDS was not significantly different from ADS. Additionally, FDS exhibited lower pitch (f0) and less positive affect than IDS, but FDS had equal pitch and significantly lower positive affect than ADS.

In contrast, Knoll and Costall (2015) found no evidence of vowel hyperarticulation in FDS, despite using natural interactions to elicit data. The authors examined vowel space expansion, mean f0, mean duration, f0 range, and intensity in FDS compared to ADS and hearing-impaired directed speech (HIDS). Ten Scottish females were recorded interacting with an adult Scottish female, one of three foreign-accented interlocutors with 2.5 years of residence in Scotland, and a hearing-impaired female using the DiapexUK task. They found that vowel space was significantly expanded in HIDS compared to ADS but not in FDS. There was no significant difference between the three speech registers in mean intensity, mean duration, or mean f0 range. However, FDS and HIDS had a significantly higher mean f0 than ADS.

Other studies examined FDS using hypothetical listeners, i.e., speech addressed to imaginary partners (e.g., Knoll & Scharrer, 2007; Scarborough et al., 2007; Knoll et al., 2009; Knoll & Costall, 2015). For example, Knoll and Scharrer (2007) replicated Uther et al.’s (2007) study but used imaginary interlocutors instead of real ones and a laboratory setting instead of the home environment. Ten British female students with a Southern English accent were asked to imagine talking to an infant, a British adult, and a foreigner in different sessions. Unlike Uther et al.’s study, the results indicated a significantly narrower vowel space in IDS relative to ADS or FDS. FDS and ADS were not significantly different in terms of vowel space size. This study also did not find any significant difference between the three registers regarding vowel duration. Vowel pitch was significantly higher in IDS compared to ADS, but IDS and FDS were not statistically different in terms of pitch, while the difference between FDS and ADS approached significance (p = 0.054). Similar to Uther et al.’s findings, IDS had a significantly higher positive vocal affect compared to the adult conditions, while FDS had lower positive affect compared to ADS, though this difference was not significant. Based on these results, Knoll and Scharrer (2007) concluded that their findings generally did not approximate those of Uther et al. (2007), except for vocal positive affect. They proposed that this discrepancy was due to the absence of “two-way dynamic feedback between speaker and listener”, which is fundamental for generating appropriate speech adjustments.

Additionally, some studies compared simulated vs. natural FDS (e.g., Scarborough et al., 2007; Knoll & Costall, 2015). Scarborough et al. (2007) compared speech directed at real interlocutors with that directed at imaginary ones. They found that speech directed at imaginary interlocutors had a significantly more expanded vowel space and longer vowels compared to speech directed at real foreign interlocutors. Their results also showed that speech directed at real interlocutors had significantly longer vowels than in ADS, and the speech rate was significantly slower. The authors concluded that FDS is comparable to speech directed at hearing-impaired listeners or speech produced in noisy settings.

Previous studies on FDS, which focused on speech addressed to foreign interlocutors (real or imaginary) with perceived foreign accents, have often overlooked crucial factors such as the interlocutors’ degree of foreign accent, L2 proficiency, or length of residence. Few studies have considered the impact of these characteristics on the extent of acoustic adjustments in FDS. One study by Kangatharan (2015) examined how interlocutors’ appearance and accent influence vowel hyperarticulation in FDS. Fifty-two white British adults participated in a spot-the-difference task with interlocutors from four groups: (1) native-looking and native-sounding (NLNS), (2) native-looking and foreign-sounding (NLFS), (3) foreign-looking and native-sounding (FLNS), and (4) foreign-looking and foreign-sounding (FLFS). She analyzed adaptations in vowel space, mean f0, intensity, vowel duration, and word duration in FDS. The results showed that vowel hyperarticulation was present in the NLFS and FLFS conditions, where the interlocutor had a foreign accent. Speech directed at foreign-accented interlocutors exhibited longer vowels and words compared to ADS. However, the interlocutor’s appearance did not affect vowel space expansion. In contrast, mean intensity was significantly higher in speech directed at foreign-looking interlocutors, regardless of accent, while f0 was consistent across all conditions. Another study by Kangatharan et al. (2022) examined the impact of foreign listeners’ characteristics on the comprehension of hyperarticulated speech, considering factors such as L2 learning onset (early vs. late), perceived accent, and appearance (foreign vs. native). Although both studies explored the influence of foreign interlocutors’ characteristics on FDS or comprehension of hyperarticulated speech, further research is needed to clarify the role of these factors in eliciting FDS.

Inconsistent results from previous studies regarding the expansion of vowel space area, vowel duration, and f0, as well as the scarcity of studies using natural interactions, highlight the need for further investigation of FDS. Therefore, the current study aims to further explore features examined in FDS, including vowel space area, vowel length, intensity, and f0, through a naturalistic interaction while considering the role of the foreign interlocutors’ characteristics and degree of foreign accent.

3. The Present Study

The current study seeks to investigate FDS in an understudied setting and a language other than English, which has dominated most FDS studies. Systematic analyses of FDS in languages other than English remain scarce, with one notable exception being Al-Kendi and Khattab (2019)’s research on Arabic, which the present study builds upon. Although previous research has examined specific phonetic aspects of native/non-native interactions in other languages, such as word reduction in Spanish (Rodriguez-Cuadrado et al., 2018), articulation rate variation in English and French (Kühnert & Antolík, 2017), and prosodic accommodation in French (Smith, 2007), these studies do not provide a comprehensive analysis of FDS vowels. This study contributes to filling this gap by exploring the acoustic-phonetic and prosodic properties of Arabic vowels in speech directed at live-in FDHs in Oman. In most Gulf countries in the Middle East, FDHs are employed on two- or three-year contracts, which they can extend up to more than five years. Some FDHs arrive in the Arab world without any prior knowledge of Arabic and start learning the language through naturalistic interactions with members of their host family. Others arrive with basic knowledge acquired through exposure to Islamic rituals in their home countries. This study examines whether the speech of female native Omani speakers, in interactions with their FDHs, exhibits any of the acoustic-phonetic adjustments reported in previous FDS research. Unlike prior studies, the current study also investigates the role of the foreign interlocutors’ characteristics on the extent of these acoustic-phonetic adjustments. The main aims of the study are (1) to investigate the acoustic-phonetic and prosodic properties of vowels in FDS, (2) to examine whether hyperarticulation occurs in FDS, and (3) to explore whether the acoustic-phonetic and prosodic adjustments in FDS correlate with the psycholinguistic characteristics of the foreign interlocutors. Three key characteristics were identified for analysis: degree of foreign accent, religion, and length of residence (LoR). To assess the degree of foreign accent, foreign accent ratings were collected from ten native Arabic listeners, who rated a selection of Arabic words produced by FDHs and two NSs (see Al-Kendi & Khattab, 2021, for more details). The foreign accent ratings indicated a significant difference between the ratings assigned to FDHs and NSs (β = 5.31, SE = 0.22, p < 0.01, see

Table 1. Descriptive statistics of foreign accent ratings (adopted from Al-Kendi & Khattab, 2021).

Group	Mean	Median	SD	Variance
FDH	3.64	3	2.62	6.86
NS	8.61	9	1.22	1.49

for descriptive statistics). FDHs were perceived to have a strong foreign accent compared to NSs (median = 3 for FDHs vs. median = 9 for NSs). While this finding is notable, variation was also observed in the ratings for FDHs compared to those for NSs (SD = 2.62 vs. SD = 1.22, respectively). Consequently, the foreign accent rating scores for FDHs were used as a covariate in the analyses to examine whether adjustments in FDS were influenced by the degree of the FDHs’ foreign accent (see Figure 1).

4. Materials and Methods

In what follows, I present the task that was used to elicit data from native Omani speakers while interacting with their FDHs.

4.1. Participants

Twenty-two female native Omani-Arabic speakers and twenty-two female FDHs participated in this study. Background information was collected through questionnaires. The Omani speakers, who were the employers and hosts of the FDHs, were from the city of Nizwa in A’Dakhilia Governorate. Their ages ranged from 19 to 50 years (mean = 34 years), and all were college graduates, except for one who was still in college. The FDHs came from various countries, including Tanzania (5), Indonesia (2), Sri Lanka (4), Bangladesh (2), the Philippines (5), Nigeria (1), India (1), Ethiopia (1), and Uganda (1), with corresponding L1 backgrounds such as Tagalog, Bengali, Swahili, Telugu, Yoruba, Sinhala, and Oromo. Their LoRs in the Arabic-speaking world ranged from 0.7 to 21 years (mean = 6 years). Most FDHs migrated as adults (mean age at arrival = 27 years), with some having prior exposure to Arabic through Islamic rituals (e.g., Qur’an recitation, prayers). Sixteen FDHs identified as Muslims, while six identified as non-Muslim. For this study, FDHs were classified by religion as either Muslim or non-Muslim. All participants reported being primarily addressed in Arabic by the families they worked for. To compare FDS to NDS, one native Arabic speaker (the author) participated to elicit NDS. The author, from the same city as the NS participants, speaks the same dialect. All participants reported no hearing or articulation issues.

It is important to note that, while the Omani NS participants may be more familiar with their FDHs than with the NS confederate, the current study did not examine the effect of familiarity on phonetic adjustments, as this was not the primary focus. To investigate the effect of familiarity, a separate group of participants would need to be recruited, such as adults who are familiar with the Omani NS participants (e.g., a sister or mother).

4.2. Method

A “spot the difference” task was used to elicit FDS and NDS, with resources similar to those used in the DiapixUK task (Baker & Hazan, 2011). This task was well suited for the current study as it allowed for the examination of specific acoustic features by manipulating picture differences based on keywords containing the target segments. By specifying speech targets, the task minimizes potential acoustic variability that may arise from differences in phonetic context across speech styles (Ferguson & Kewley-Port, 2002). Additionally, the task can be adjusted to encourage one participant to speak more than the other by instructing them to “take the lead” in the interaction (Baker & Hazan, 2011, p. 763).

4.3. Stimuli

For the original study, three lists of keywords containing the target segments were created. The first two lists included segments not relevant to this study, so we will not elaborate on them here. The third list contained nine content words featuring the three Arabic corner vowels /a:/, /i:/, and /u:/ (see

Table 2. The keywords containing the three corner vowels used in the diapex task to elicit FDS and NDS.

Vowel	Keyword
/i:/	/fi:l/ ‘elephant’2	/ti:n/ ‘fig’	/ħali:b/ ‘milk’
/a:/	/ba:b/ ‘door’	/ta:g/ ‘crown’	/kta:b/ ‘book’
/u:/	/fu:l/ ‘chickpeas’	/tu:t/ ‘berries’	/χaru:f/ ‘sheep’

). To ensure the FDHs were able to engage with the task, the selected keywords represented household objects.

4.4. Task Design and Procedures

The pictures for the task were designed to depict household scenes that would be easy for the NSs to describe and familiar to the FDHs. This was crucial for fostering a relaxed, natural interaction. Additionally, the task was structured to align with the criteria used in designing the DiapixUK task, considering factors like style, complexity, and the number and type of differences (Baker & Hazan, 2011).

The task involved six picture pairs, each depicting one of three different scenes: a kitchen, a living room, and a bedroom. Each scene included two distinct picture pairs—one designed to elicit FDS and the other to elicit NDS—to prevent repetition. The scenes featured objects representing specific keywords (Table 1), along with some distractors.

Each picture pair differed in terms of color, quantity, or the presence/absence of an object (e.g., a blue door in Picture X vs. a white door in Picture Y, or three books in X vs. two in Y). When the difference was based on the presence or absence of an object, the NS always had the version where the object was present, while the FDH had the version where it was absent—especially if the object represented a keyword. This design ensured that the NS produced all vowels of interest in the study. Each picture pair contained twelve differences, serving as the study’s stimuli (see Appendix A).

Each scene’s two picture pairs contained the same keywords from Table 2 (represented by objects in the pictures) but differed in the types of differences (e.g., number, size, color), the arrangement of objects, and the nature of the distractors. This design ensured the elicitation of comparable speech samples across interlocutor groups while maintaining the task’s complexity.

Speech recordings took place in a quiet room within the household of the female NS employers. The FDH and NSs were seated opposite each other and given three pictures depicting three different scenes, arranged in the same order. They were instructed to discuss the differences between their pictures without looking at each other’s images. The NS led the discussion by asking the FDH questions and describing the scenes. The NS then completed the same task with the author using the other set of picture pairs. To control for potential condition order effects that might have influenced the NS’s interaction, half of the NSs completed the task with their FDHs first, followed by the NS confederate, while the other half followed the reverse order. Interactions were recorded using a Roland Edirol R-09HR digital recorder with a sampling rate of 44,100 Hz in WAV-16bit format. The NSs wore a Sennheiser radio microphone to ensure high-quality audio capture.

4.5. Acoustic and Statistical Analysis

The elicited data were manually transcribed and labeled by the author using the speech analysis software Praat (Boersma & Weenink, 2009). All instances of the target vowels that were free of background noise, tongue stuttering, or speaker overlap were included in the analysis. The onset and offset of the vowels /a:, i:, and u:/ were determined based on the last periodic pulse in the waveform and the vertical consistency of the first three formants within the 0–5500 Hz range of the spectrogram (Al-Tamimi, 2017). When vowels were followed by a sonorant (e.g., in /fi:l/ ‘elephant’, /ti:n/ ‘fig’), vowel boundaries were identified based on intensity drop and visual assessment.

A Praat script was used to automate acoustic measurements, including F1, F2, f0, duration, and intensity, for all target vowels. All acoustic measurements were taken at the vowel midpoint, except for duration, as this point is least influenced by neighboring segments. Formant calculation was configured using a 25 ms Kaiser2 window with a 5 ms formant time step. The default Burg algorithm estimated a maximum of five formants, with a 5500 Hz upper frequency limit for female speakers. To ensure accuracy, all automatically extracted formant values were manually verified and corrected as needed. After extraction, formant frequencies were converted to the psychoacoustic Bark scale using the formula Z_n = {26.81/(1 + 1960/f_n0} − 0.53, where Z_n is the Z value of a formant n and f_n is a formant’s frequency in Hz (Traunmüller, 1990). Sound duration was measured in millisecond and intensity in decibel.

Statistical analyses were conducted in R using various packages (R Core Team, 2012). The PhonR package (McCloy, 2016) was used to plot the first and second formant frequencies (F1, F2) of the Arabic vowels /i:, a:, and u:/ in both FDS and NDS. The vowel space area was calculated using the convex hull area.

Two different analyses were conducted to examine vowel space. First, the vowel space area was calculated separately for each participant. Statistical differences between the vowel space areas of FDS and NDS were then analyzed using a linear mixed-effects model (LMEM), implemented via the lmerTest package (Kuznetsova et al., 2017). In this model, vowel space was the dependent variable, speech type (FDS vs. NDS) was the independent variable, and speaker was included as a random intercept.

The second analysis focused on overall vowel space area at the group level rather than individual vowel space. The vowel space area for each group was first calculated, and the difference between FDS and NDS was then assessed using a t-test, which compared the mean vowel space areas of the two groups.

To examine changes in F1 and F2 across speech registers, separate LMEMs were applied to each vowel. F1 was analyzed for all three vowels, as it primarily reflects vowel height, a crucial feature for distinguishing vowel qualities in this study. F2 was analyzed for the front vowel /i:/ and the back vowel /u:/, following prior research (Ferguson & Kewley-Port, 2002), since F2 is closely linked to vowel frontness and backness—key distinctions for these vowels in this context. The vowel /a:/ was excluded from the F2 analysis due to its relatively low sensitivity to front–back contrasts in Arabic, making F2 less informative for this vowel.

An additional LMEM was used to assess the effect of speech register on intensity, duration, and f0 for all three vowels. In this model, speech register was treated as a fixed effect, while speaker and target word were included as random intercepts. To evaluate the influence of FDH characteristics (degree of foreign accent, LoR, and religion) on the extent of acoustic changes, LMEMs were applied across all acoustic metrics. LoR and degree of foreign accent were treated as continuous variables, while religion was categorized as Muslim vs. non-Muslim. These factors were included as fixed effects, with speaker as a random intercept.

To determine whether changes in vowel space area, F1, F2, duration, intensity, and f0 were influenced by phrase boundary position, each vowel was assigned a binary value (0 or 1) based on whether it appeared at a phrase boundary, following Miyazawa et al. (2017). Two linear mixed-effects models were used to assess the effect size: a full model including speech register and phrase boundary as fixed effects and a reduced model excluding speech register. These models were estimated separately for the F1 and F2 values of each vowel, as well as for duration, intensity, and f0 across all three vowels (16 models in total). The same random effect structure as in previous models was applied. A likelihood ratio test (Barr et al., 2013) was used to compare the full and reduced models and determine the statistical significance of speech register effects.

5. Results

5.1. Vowel Space

The analysis of individually calculated vowel space areas for each participant revealed a significant difference between FDS and NDS (p < 0.01,

Table 3. LMEM results for the effect of speech register on vowel space area.

Predictor	Estimate	Std. Error	df	t-Value	Pr (>|t|)
Intercept	12.93	0.56	21.84	22.93
Speech: FDS	2.67	0.12	650.36	20.9	<2 × 10−16 ***

The three stars ‘***’ indicate that a p-value is less than 0.001.

). The vowel space area in FDS was larger by 2.67 Barks (±0.56 SE) compared to NDS, indicating clear evidence of vowel space hyperarticulation in FDS.

The results also showed that the vowel space in FDS did not vary significantly based on the FDH’s characteristics (

Table 4. LMEM results for the effect of foreign interlocutors’ characteristics in the adjustments of vowel space area in FDS.

Predictor	Estimate	Std. Error	df	t-Value	Pr (>|t|)
Intercept	14.38	2.15	21.0006	6.66
Accent rating	0.2	0.59	21	0.34	0.73
LoR	−0.14	0.14	21.0031	−1.005	0.32
Religion: non-Muslim	1.23	2.06	21.0053	0.59	0.55

). This suggests that NSs did not adjust their vowel space in response to the foreign interlocutor’s degree of foreign accent, LoR, or religion.

Based on the second method of analysis, a t-test revealed that the vowel space area in FDS was significantly larger than in NDS (p = 0.01, Figure 2). On average, the vowel space area in FDS was 0.5 Barks larger than in NDS. Visual inspection of Figure 2 shows that the vowel space area for FDS is larger, primarily due to an observable increase in the F1 values, with little to no change in the F2 values. This observation will be further supported statistically in the following section, where the overall F1 and F2 values of vowels in FDS and NDS are compared.

5.2. Formant Frequencies (F1 and F2)

The LMEM analyses showed that the mean F1 values for the vowels /i:/, /u:/, and /a:/ were significantly higher in FDS than in NDS (p < 0.01), as illustrated in

Table 5. LMEM results for the effect of speech register on F1 and F2 values of FDS and NDS vowels.

Metric	Vowel	Predictor	Estimate	Std. Error	df	t Value	Pr (>|t|)
F1	/i:/	NDS (Baseline)	4.43	0.07	26.53	58.17
		FDS	0.11	0.04	221.52	2.605	0.009 **
	/u:/	NDS	4.62	0.12	3.1003	38.38
		FDS	0.15	0.04	188.63	3.25	0.001 **
	/a:/	NDS	6.59	0.09	24.35	69.99
		FDS	0.205	0.04	213.02	4.58	<0.0001 ***
F2	/i:/	NDS	15.02	0.09	11.33	154.71
		FDS	−0.08	0.04	218.76	−1.77	0.07
	/u:/	NDS	7.81	0.19	4.58	40.46
		FDS	0.09	0.09	194.26	0.99	0.32

Two stars ‘**’ indicate that a p-value is less than 0.01; Three stars ‘***’ indicate that a p-value is less than 0.001.

and Figure 3. Since F1 is inversely related to tongue height, these results suggest that all three vowels were produced with greater tongue lowering in FDS compared to NDS. In contrast, the LMEM analyses revealed no significant effect of speech register on the F2 values of the high front vowel /i:/ and the high back vowel /u:/ (p > 0.05), as shown in Table 5 and Figure 4. The F2 value for /i:/ in FDS was slightly lower than in NDS by 0.08 Barks (±0.04 SE), suggesting a minor degree of tongue retraction in FDS, though the difference was negligible. Similarly, /u:/ in FDS appeared to be produced with a slightly advanced tongue position, as its F2 value was 0.09 Barks (±0.09 SE) higher than in NDS. However, this difference was also not statistically significant.

To rule out the effect of phrase boundary, the likelihood ratio tests showed that speech register contributed significantly to the full models, beyond the influence of phrase boundary (p < 0.01,

Table 6. Results of likelihood ratio tests for the effect of phrase boundary on F1 values.

Vowel	Model	Df	AIC	BIC	LogLike	Deviance	Chisq	Df	Pr (>Chisq)
/i:/	Reduced Model	5	201.72	219.14	−95.85	191.72
	Full Model	6	197.06	217.97	−92.53	185.06	6.65	1	0.001 **
/u:/	Reduced Model	5	180.48	197.1	−85.24	170.48
	Full Model	6	171.16	191.1	−79.57	159.16	11.32	1	<0.0001 ***
/a:/	Reduced Model	14	222.32	239.60	−106.162	212.32
	Full Model	15	203.74	6829.5	−95.872	191.74	20.58	1	<0.0001 ***

Two stars ‘**’ indicate that a p-value is less than 0.01; Three stars ‘***’ indicate that a p-value is less than 0.001.

). This suggests that the higher F1 values for the three vowels in FDS were a result of NSs hyperarticulating their speech based on the recipient’s status as a foreigner or an NS, rather than being merely a by-product of other factors.

Analyses of the effect of the foreign interlocutors’ characteristics on the mean F1 and F2 values of the vowels revealed no significant influence of the degree of foreign accent, LoR, or religion (

Table 7. LMEM results for the effect of foreign interlocutors’ characteristics on vowel F1 and F2 adjustments in FDS.

Metric	Vowel	Predictor	Estimate	Std. Error	df	t-Value	Pr (>|t|)
F1	/i:/	Intercept	4.69	0.23	21.19	19.64
		Accent rating	−0.05	0.06	21.18	−0.81	0.42
		LoR	0.02	0.01	22.31	1.27	0.21
		Religion: non-Muslim	−0.31	0.23	23.99	−1.35	0.18
	/u:/	Intercept	4.92	0.19	25.04	25.7
		Accent rating	−0.03	0.05	23.78	−0.72	0.47
		LoR	0.02	0.01	23.35	1.92	0.06
		Religion: non-Muslim	−0.19	0.17	21.61	−1.11	0.27
	/a:/	Intercept	6.83	0.25	20.92	26.73
		Accent rating	0.01	0.07	20.82	0.15	0.87
		LoR	−0.003	0.01	21.42	−0.2	0.83
		Religion: non-Muslim	−0.05	0.24	22.14	−0.22	0.82
F2	/i:/	Intercept	15.09	0.19	20.82	76.48
		Accent rating	0.03	0.05	20.9	0.68	0.49
		LoR	0.008	0.01	22.51	0.64	0.52
		Religion: non-Muslim	−0.06	0.19	25.15	−0.33	0.74
	/u:/	Intercept	7.98	0.25	28.36	31.59
		Accent rating	−0.01	0.06	23.1	−0.19	0.84
		LoR	0.01	0.01	24.68	1.18	0.24
		Religion: non-Muslim	0.04	0.22	21.18	0.18	0.85

). This indicates that the NSs did not alter the acoustic properties of their vowels based on the foreign interlocutor’s degree of foreign accent, LoR, or religion.

5.3. Prosodic and Temporal Features (f0, Intensity, and Duration)

The LMEM analyses showed that the mean f0 and intensity of the vowels in FDS were significantly higher than in NDS, as illustrated in

Table 8. LMEM results for the effect of speech register on vowel f0, intensity, and duration.

Metric	Predictor	Estimate	Std. Error	df	t-Value	Pr (>|t|)
f0	NDS (Baseline)	221.72	4.34	32.53	51.06
	FDS	16.42	2.22	657.42	7.37	5.05 × 10−13 ***
Intensity	NDS (Baseline)	62.65	0.85	27.39	73.54
	FDS	1.66	0.28	653.27	5.73	1.46 × 10−8 ***
Duration	NDS (Baseline)	145.57	5.63	19.45	25.83
	FDS	−9.41	2.82	659.089	−3.33	0.0009 ***

Three stars ‘***’ indicate that a p-value is less than 0.001.

and Figure 5 and Figure 6. However, unexpectedly, the mean duration of the FDS vowels was significantly shorter than that of NDS (Table 8 and Figure 7).

To rule out any influence of phrase boundary, the likelihood ratio tests indicated that the observed differences in f0, intensity, and duration between FDS and NDS were primarily due to the interlocutor being addressed by the NS, rather than being a mere result of prosodic boundary, as shown in

Table 9. Results of likelihood tests for the effect of phrase boundary on adjustments in vowel f0, intensity, and duration.

Metric	Model	Df	AIC	BIC	LogLike	Deviance	Chisq	Df	Pr (>Chisq)
f0	Reduced Model	5	6566.6	6589.3	−3278.3	6556.6
	Full Model	6	6516.2	6543.3	−3252.1	6504.2	52.443	1	4.43 × 10−13 ***
Intensity	Reduced Model	5	3778	3800.6	−1884	3768
	Full Model	6	3746.5	3773.6	−1867.2	3734.5	33.55	1	6.929 × 10−9 ***
Duration	Reduced Model	5	6812	6834.6	−3401	6802
	Full Model	6	6800.6	6827.8	−3394.3	6788.6	13.33	1	0.0002 ***

Three stars ‘***’ indicate that a p-value is less than 0.001.

.

6. Discussion and Conclusions

The current study examined the acoustic properties of vowels in FDS within a naturalistic interaction setting between female employers and their FDHs. The results of the study will be discussed in the following sections.

6.1. Vowel Space Area and Formant Frequencies

The difference in vowel space area between FDS and NDS in the current study was significant, with FDS exhibiting a larger vowel space. Expanding vowel space area has often been associated with enhanced intelligibility, as seen in clear speech (e.g., Ferguson & Kewley-Port, 2002), and plays a didactic role in both IDS (e.g., Kuhl et al., 1997; Trainor & Desjardins, 2002; Werker et al., 2007) and FDS (e.g., Uther et al., 2012; Uther et al., 2007). In clear speech, speakers typically enhance the distinction between vowel contrasts to produce clearer articulations, especially in the presence of a communicative barrier (e.g., noise) or when speaking to hearing-impaired listeners. In IDS, mothers reduce the challenges of language learning for infants or foreign interlocutors (Kuhl et al., 1997; Uther et al., 2007). In FDS, it has been assumed that hyperarticulation benefits listeners (e.g., Rothermich et al., 2019; Bobb et al., 2019; Uther et al., 2007). Piazza et al. (2023) provided evidence that acoustically modified FDS aids listeners in recognizing novel words and eventually producing intelligible L2 sounds. Additionally, Kangatharan et al. (2022) assessed whether hyperarticulated speech improves comprehension for both L1 and L2 listeners in noisy and quiet conditions. Their findings showed that all foreign listener groups rated hyperarticulated speech as more comprehensible than non-hyperarticulated speech in both conditions.

Additional analyses of the collective mean F1 and F2 values for all speakers in the two groups (FDS and NDS) revealed a larger vowel space in FDS. This expansion was driven by an increase in F1 values across all examined vowels, while the mean F2 values remained statistically similar between the two groups. Consistent with the LMEM results based on individual vowel spaces, FDS speakers exhibited higher mean F1 values than NDS speakers, suggesting that these vowels were likely produced with a more open jaw and greater vocal effort in FDS. This finding aligns with previous research indicating that increased F1 values reflect heightened jaw opening and vocal effort (Liénard & Di Benedetto, 1999). These results underscore the importance of carefully selecting analysis methods when calculating vowel space area. Notably, while the mean F2 values were similar across speech registers, different calculation methods yielded varying results. When vowel space area was determined based on individual vowel spaces, FDS had a significantly larger mean vowel space area than NDS. However, when using the collective mean F1 and F2 values for all individuals in each group, the expansion in FDS vowel space was primarily attributed to changes in F1, not F2. Future research could further explore these differences.

6.2. Temporal Properties

Temporal adjustments (i.e., duration of segments) in FDS are typically understood in relation to speech rate. A slow speech rate has been described as a robust characteristic of IDS, clear speech, and FDS (Hatch, 1983; Munro & Derwing, 1999; Biersack et al., 2005; Kangatharan, 2015). Wesche (1994) suggests that clear articulations and the production of vowels in their full form contribute to a slower speech rate. Furthermore, the H&H theory (Lindblom, 1990) posits that careful enunciations of speech result in the adoption of a hyper-mode, in which vowels and consonants are expected to reach their target form. In studies of IDS and clear speech, longer vowels and consonants are typically associated with slower speech delivery. However, in the current study, vowel duration in FDS was found to be significantly shorter than in NDS, contrary to the predictions of the H&H theory and the findings in IDS and clear speech.

The findings on vowel duration in studies of FDS have been inconsistent. Some studies have reported no significant difference in vowel length between FDS and NDS (e.g., Knoll et al., 2009; Knoll & Scharrer, 2007; Uther et al., 2007; Biersack et al., 2005), while others have reported significantly longer vowels in FDS (e.g., Scarborough et al., 2007; Kangatharan, 2015). The absence of a difference in vowel duration between FDS and NDS in previous studies is unexpected, particularly if we assume that FDS serves a similar function to IDS or clear speech, in which speech delivery is slower, and segments are typically longer.

The results of the current study are even more surprising, as they show significant vowel shortening or reduction in FDS. This finding could be interpreted in light of the NS-FDH context. It is possible that the NSs in this study did not feel the need to slow down their speech for their FDHs, as they were likely more familiar with them compared to the NS confederate, who was a stranger to most of the female employers. However, this interpretation remains speculative and would require further investigation. Future research could examine whether the reduction in vowel duration is due to familiarity with the interlocutor or other factors, such as the foreignness of the interlocutor (see Piazza et al., 2022, for a review).

It should also be noted that, although segment duration is a key factor in measuring speech rate, it is not the only factor. Other candidates include the length and number of words or syllables per second, the length of the utterance, and the length and frequency of pauses (Bradlow et al., 2003; Narayanan & Wang, 2005). Therefore, the significant reduction in vowel duration in FDS does not necessarily imply a lack of evidence for a slower speech rate. Since these other factors were not examined in the present study, it remains unclear whether FDS is slower than or comparable to NDS in terms of overall speech rate. This is an area that warrants further exploration in future research.

6.3. Prosodic Properties

The current study found that vowels in FDS exhibited higher intensity compared to NDS. Intensity is known to correlate with loudness (Fernald & Mazzie, 1991), and an increase in respiratory effort can also contribute to heightened sound intensity (Lehiste, 1976). This suggests that vowels in FDS were produced with more powerful articulation. Additionally, the study observed that vowels in FDS had increased f0 relative to NDS. A rise in f0 results from an increase in vocal fold tension (Monsen et al., 1978). Combined with other acoustic cues like intensity and duration, f0 adaptations in English help emphasize stressed syllables or words in a sentence (Lehiste, 1976). Specifically, f0 increases during the production of stressed syllables due to higher subglottal air pressure and vocal fold tension (Monsen et al., 1978).

Thus, one interpretation for the heightened intensity and f0 in FDS vowels could be an increase in vocal effort. This is further supported by the observed increase in F1 values for FDS vowels. In a study examining the effect of vocal effort on vowel spectral properties, Liénard and Di Benedetto (1999) found that both F1 and f0 increase when speakers exert more vocal effort. Similarly, Ferguson and Kewley-Port (2002) demonstrated that the rise in F1 in clearly produced vowels was a consequence of increased vocal effort. They also observed an increase in the intensity of clear speech during recording and digitization, attributing it to increased vocal effort. Picheny et al. (1986) further showed that clear speech is typically 5–8 dB more intense than conversational speech. Taken together, these findings suggest that the increased F1, f0, and intensity in FDS vowels in the present study likely reflect an attempt by NSs to speak clearly or with extra vocal effort.

Prosodic modifications in speech directed at infants have been well established in many studies (e.g., Fernald & Kuhl, 1987; Cooper & Aslin, 1990; Broesch & Bryant, 2015; Quigley et al., 2019). IDS is characterized by f0, expanded pitch contours, higher amplitude, and slower tempo compared to ADS (e.g., Suttora et al., 2017; Adriaans & Swingley, 2017). These heightened prosodic aspects of IDS are believed to be a major reason for infants’ preference for IDS (Fernald & Simon, 1984; Fernald & Kuhl, 1987; Cooper et al., 1997). Prosodic modifications in IDS are thought to serve two primary functions: convey affect (Fernald & Kuhl, 1987; Knoll & Scharrer, 2007) and aid language learning (Adriaans & Swingley, 2017; Han et al., 2022; Shi et al., 2023). Second language acquisition researchers have confirmed that preverbal infants tend to identify words and syllables within a sentence that are highlighted by prosodic cues, including pitch and stress (Gleitman & Wanner, 1982; Butler et al., 2014). This ability helps children recognize boundaries between syllables and words, as languages like English do not have a fixed set of acoustic markers for utterance boundaries (Albin & Echols, 1996). Thus, exaggerated prosodic cues in IDS can be beneficial for children in identifying these boundaries.

While FDS cannot be directly compared to IDS in terms of emotional content, the prosodic exaggerations in FDS are best interpreted as serving a didactic role. The target words used to elicit FDS in this study were likely candidates for sentential stress (focus), as they represented the objects in the pictures used in the eliciting task. Therefore, it is likely that NSs heightened the prosodic cues of focus words when interacting with their FDHs to draw attention to them (Albin & Echols, 1996).

6.4. The Role of Foreigners’ Psycholinguistic Factors

One of the aims of this study was to assess whether the psycholinguistic factors of foreigners influence the degree of acoustic adjustments in FDS. Regarding the impact of the foreigners’ degree of foreign accent, the findings showed that the FDHs’ foreign accent ratings did not affect the extent of acoustic adjustments in FDS vowels. Despite individual variation in foreign accent ratings, adjustments in the vowels of FDS did not depend on the degree of foreign accent perceived in the FDHs. One interpretation of this finding is that FDHs were consistently rated as having a strong foreign accent, regardless of their LoR in the Arab World (Al-Kendi & Khattab, 2021). Thus, it is possible that NSs employed hyperarticulated speech with all FDHs, based on their perceived foreign accent. Another explanation could be that the ratings of FDHs’ accents did not align with the female employers’ personal perceptions, which may have been influenced by their familiarity with the FDHs. This limitation presents an interesting area for future research, focusing on how employers perceive the accents of FDHs they interact with regularly.

LoR was assumed to correlate with FDHs’ exposure to Arabic, with longer stays expected to lead to greater language experience. However, the results showed that LoR did not significantly impact the level of hyperarticulation in FDS. This suggests that LoR might not be a reliable measure of language experience, as FDHs vary widely in the quantity and quality of linguistic input they receive. Additionally, the frequency and nature of interactions with family members could significantly differ across FDHs, further complicating the relationship between LoR and language proficiency. Previous studies, such as Moyer (2004)’s work, have suggested that factors such as language use, motivation, frequency of contact with NSs, and identity within the L2 community should also be considered when assessing language experience. Thus, future studies could explore these variables in more detail.

Muslim FDHs were expected to trigger less hyperarticulation than non-Muslim FDHs due to their background in Arabic. However, religion did not significantly influence the extent of hyperarticulation in this study. Overall, psycholinguistic factors such as religion, foreign accent, and LoR did not appear to play a role in the amount of hyperarticulation observed in FDS in this study.

In conclusion, this study provides evidence that speech directed at FDHs exhibits vowel space hyperarticulation similar to other speech registers, such as FDS, IDS, and clear speech. However, the hyperarticulation in this study was primarily driven by an increase in F1 values, reflecting increased vocal effort. Prosodic adaptations, such as higher f0 and intensity, further support the interpretation that speech directed at FDHs involves heightened vocal effort. These adaptations, along with the larger vowel space area, suggest that NSs in this interactive setting were speaking louder, with more vocal effort, and a more open jaw. Such strategies may be used to emphasize key linguistic elements and facilitate language learning in this context. While psycholinguistic factors like religion, foreign accent, and LoR did not influence the degree of hyperarticulation, the findings suggest that the perception of FDHs as foreigners with foreign accents may trigger such adaptations in FDS.