Consonant Neutralization and Vowel Space Area in Tashlhiyt Berber’s Infant-Directed Speech

Abstract

Infant-Directed Speech (IDS) and Adult-Directed Speech (ADS) are two registers that can differ across multiple linguistic domains and social contexts. In languages where the phonemes’ acoustic clarity is modified, there is a typical assumption in the literature that these phonemes are either hyper- or hypo-articulated. These modifications have received considerable attention in language research, serving as the basis of proposals concerning the potential didactic and communicative (e.g., affective) functions of IDS. The current study adds to this literature by examining how vowels and consonants are modified in IDS in a previously unstudied African language with an unusual phoneme inventory: Tashlhiyt Berber. Seven caregivers were recorded interacting with: (a) infants between the ages of 0;6 and 1;5, and (b) an adult experimenter who is a native speaker of Tashlhiyt. In IDS, we found longer vowel duration for content words and exaggerated pitch levels across the board, successfully replicating cross-linguistic patterns. Counter to the hyper-articulation hypothesis, there was more vowel overlap in IDS, while no differences were observed between IDS and ADS for overall vowel space size. We additionally found that speakers tend to completely neutralize certain consonant contrasts in IDS, rendering them indistinguishable to the infant. Our results are most consistent with the proposal that caregivers use IDS to guide and maintain their infant’s attention and to convey positive affect.

Keywords

infant directed speech phonological development Tashlhiyt Berber hyperarticulation hypoarticulation

Introduction

One of the first tasks facing infant language learners is to identify a phonological system – the building blocks of language (Cutler, 2012). Human languages make use of the high flexibility of the vocal tract to make functional distinctions between sounds and assemble them to build larger, meaningful units of language (e.g., morphemes, words). Yet, how languages do this varies widely (Moran & McCloy, 2014). Polynesian languages, for instance, have very small phonological inventories (fewer than 20, Lynch, 1998). Other languages, such as Tashlhiyt Berber, which we study here, have an inventory of over 70 phonemes, most of which are consonants. Infants have no trouble negotiating the acquisition of typologically different phonological systems, but how they do so is still unclear. This is because our understanding of how and whether input is modified cross-culturally and in response to typologically different systems is very limited. In the current paper, we report on infant-directed speech (IDS) in Tashlhiyt Berber, investigating if and how caregivers modify their speech to infants.

Infant-Directed Speech

It is well established that speakers adapt to the needs of their interlocutor, including in speech directed to children. In many communities around the world, caregivers regularly use a special register when addressing children, which has been referred to in the literature as motherese (Newport & Gleitman, 1984), baby-talk (Ferguson, 1964), parentese (Ferjan Ramírez et al., 2024), IDS (Cox et al., 2023), and infant-directed language (Heymann et al., 2020). Many differences in multiple linguistic domains set IDS¹ apart from Adult-Directed Speech (ADS). Most obvious to naïve listeners, IDS is characterized by higher pitch (Cristia & Seidl, 2014), more repetitions (McRoberts et al., 2009), and shorter, less complex utterances (Genovese et al., 2019; Martin et al., 2016). More fine-grained differences include hyper-articulated (Lovcevic et al., 2025) and longer vowels (Marklund & Gustavsson, 2020), though these are not observed across the board (Cristia & Seidl, 2014).

IDS characteristics, and hyper-articulation in particular, are often argued to support language acquisition, where tokens in the input act as high-quality exemplars of the target segments and allow children to distinguish contrasts (e.g., corner vowels; Johnson et al., 1993). Additionally, syllable repetition and reduplication can scaffold word segmentation, one of the first challenges facing infants breaking into language (Ota & Skarabela, 2016, 2018; Stärk et al., 2022). It has also been shown that IDS carries more positive vocal affect linked to exaggerated prosody (Kalashnikova & Burnham, 2018; Kitamura & Lam, 2009) and accompanying facial gestures (Green et al., 2010). This may serve to capture and maintain the infant’s attention, with demonstrations that it successfully does so in newborns (Cooper & Aslin, 1990; Dunst et al., 2012). However, the role IDS plays in language acquisition continues to be debated, especially with regards to facilitating phonological development and specifically phonemic acquisition.

Although IDS exhibits some consistent crosslinguistic patterns, individual languages and cultures rarely show identical patterns of features (Kidd et al., 2025). There are likely many reasons underlying variability in IDS, but one possible source derives from typological variation. That is, differences in structural properties across languages (e.g., phoneme inventories) can lead to different modifications of the input and acquisition trajectories. Wang et al. (2016), for example, suggested that phonological structure might not only influence modification patterns in IDS, but also their extent. Pitch, for example, is the most salient IDS feature cross-linguistically, but is modified differently based on the language’s prosodic type (e.g., tonal, stress, or pitch-accented language). Thus, it is more exaggerated in stress-timed languages such as English than in tonal languages such as Mandarin, since allowing the caregiver to freely modify pitch in tonal languages risks impeding the transmission of tonal cues. Another common IDS feature is speech rate, which has also been shown to vary cross-linguistically. For example, Dutch-speaking but not Mandarin-speaking (Han et al., 2021) or Tamil-speaking mothers (Narayan & McDermott, 2016) slow down significantly when unfamiliar words are introduced. This could again be attributed to phonological characteristics (e.g., rhythmic class) or alternatively to cultural conventions (Han et al., 2021). Thus, it is important to consider language-specific phonological structure as well as culture-specific caregiving behaviors when studying the shape of IDS across different populations.

Another major way caregivers adapt their speech in IDS is by manipulating the degree of acoustic clarity. Evidence suggests that, when modified in IDS, phonemic categories can either be over- or under-specified (Lovcevic et al., 2025). This is observed in vowels, when they are pushed into more peripheral positions of the vowel space (Kuhl et al., 1997), in consonants where the durational differences in Voice Onset Time are exaggerated (Englund, 2005), and in tones (Xu Rattanasone et al., 2013). Such findings suggest that caregivers may attempt to maximize the salience of phonetic differences to facilitate phonemic perception (Kuhl et al., 2007), allowing infants to lay the foundations of their native language. This view is supported by findings that link hyper-articulation to later linguistic development (Kalashnikova & Burnham, 2018; Lovcevic et al., 2024), and by its manifestation when speakers address foreigners (Piazza et al., 2022), pets with some vocal ability (e.g., parrots, Xu et al., 2013), or under sub-optimal listening conditions (Garnier et al., 2018).

However, hyper-articulation in IDS has not been consistently observed cross-linguistically. Caregivers may not increase the salience of phonemes and might even hypo-articulate them instead. This has been observed for both salient (Norwegian /g/-/k/) and less salient (Nepali /ɡ/-/ɡʱ/-/k/-/kʱ/) consonant contrasts, as well as in larger (Cantonese) and smaller (Mandarin) vowel inventories (Benders et al., 2019; Rosslund et al., 2025; Tang et al., 2017; Xu Rattanasone et al., 2013). Much speculation has been put forward regarding the relationship between language-specific phonology and inventory size (Lovcevic et al., 2025), not only for IDS but also for other forms of hyper-speech (e.g., such as foreigner-directed speech, Piazza et al., 2022). Additionally, while cross-linguistic studies fail to reveal a consistent influence of phonological complexity on hyper-articulation’s emergence, it is possible that their mixed findings could be attributed to cultural or methodological variables (Wang et al., 2016). Indeed, any influence of language-type on IDS is filtered through a system of cultural practices related to child rearing. In sum, it seems that, unlike pitch, which is modified differently in IDS across languages based on structural features, the relationship between phonological structure and segment hyper/hypo-articulation patterns is unclear. Its variability across languages and cultures makes it important to study on a wider scale, if it is to be better understood (Lovcevic et al., 2025) and to overcome the challenges posed by the insufficient language coverage in the study of child language acquisition (Kidd & Garcia, 2022).

The Current Study

The goal of this paper is to contribute to the expansion and diversification of the range of languages covered in the IDS literature by describing the segmental characteristics of Tashlhiyt Berber’s IDS. Tashlhiyt is spoken in south-western Morocco. It is famous for its complex phonological system (Ridouane, 2014), comprising a large inventory of 71 consonants and a small vocalic system (three vowels: /a/, /i/, /u/) (Moran & McCloy, 2014). It also has a typologically rare syllable structure, which allows any segment to be in the syllable nucleus position. As a result, vowelless words (e.g., /tdːztstː/ “you compressed it”) comprise around 22% of all syntactic words in Tashlhiyt (Ridouane, 2008). Despite being well-described, Tashlhiyt remains understudied from a developmental perspective (but see Lahrouchi & Kern, 2018), and its complex phonological system represents an untouched corner of the linguistic design space in studies of acquisition (Passmore et al., 2025). Here, we ask whether the vowels and consonants of Tashlhiyt are modified by caregivers in IDS, and if so, how. That is, we ask (a) do caregivers hyper-articulate segments, and (b) do they modify vowels and consonants similarly? We pay particular attention to the role of phonological structure and whether it leads to different behavior by taking inventory size as a proxy of complexity. Our study represents an important addition to the literature: in research investigating hyper-articulation in IDS, Tashlhiyt is simultaneously the language with the largest consonant inventory (Table 1) and the smallest vowel inventory examined to date (Lovcevic et al., 2025).

Table 1.

Tashlhiyt’s Consonant Inventory Adapted From Ridouane (2014), Gray Areas Represent Neutralized Phonemes in a Maximally Reduced Inventory.^a.

Note. P = Plain; L = Labialized; E = Emphatic (Pharyngealized); IDS = Infant-Directed Speech.

Gray areas in the inventory represent the contrasts that may or may not be (subject to individual differences among caregivers) neutralized in IDS. Some caregivers, for example, may neutralize Emphatic dental consonants only, while others may neutralize all contrasts in gray.

There are at least two, mutually exclusive, possibilities in how Tashlhiyt-speaking caregivers could modify IDS in comparison to ADS. First, if phonological complexity drives the emergence of hyper-articulation, as discussed above, this should be (a) less pronounced in the comparatively smaller vocalic system but (b) more pronounced in the consonant system (to emphasize the differences between the less salient consonant contrasts). In contrast, if hyper-articulation is driven by convenience and is more likely to emerge when there is less risk of overlap between phonemic categories (Lovcevic et al., 2025), we expect the opposite pattern: consonant hypo-articulation and vowel hyper-articulation. Since there is no previous work on this topic in Tashlhiyt, we cannot make firm predictions and must be content with mapping out the hypothesis space in this manner.

We present several analyses bearing upon these predictions. For vowels, we asked whether vowels are acoustically more salient in IDS compared to ADS. We focused on a number of acoustic dimensions: pitch, duration, Vowel Space Area (VSA) size, vowel overlap, and individual formant characteristics. Observing a higher pitch in IDS would replicate cross-linguistic results and would be consistent with the suggestion that the register serves attention-getting and affective functions. With respect to vowel duration, we followed Bundgaard-Nielsen et al. (2023) in examining whether it is similarly modified in content words compared to function words, as well as in nouns compared to other word classes, controlling for word position. If longer vowel durations in IDS are observed, this would suggest that the register might serve a didactic function by enhancing vowel clarity, which could promote phonemic acquisition. Regarding VSA, a larger VSA and less vowel overall for IDS would indicate hyper-articulation, lending more support to the hyper-articulation hypothesis (Kuhl et al., 1997), while a smaller VSA and more vowel overlap would both suggest that vowels are less salient (e.g., hypo-articulated), going against the idea of facilitating perception and learning. Regarding formants, an overall increase in values would suggest that IDS is better understood as a means to convey vocal affect and/or imitation of infant vocalizations (Benders, 2013).

For consonants, we asked whether consonants are modified in IDS, similarly to vowels, in ways that could either increase or decrease their perceptual salience. We therefore compared the target phonemes to the actual phonemes produced by the caregivers and whether the phoneme’s surface realization matched its underlying form. We report that certain sets of contrasts in Tashlhiyt are systematically neutralized. That is, we find no evidence in favor of hyper- or hypo-articulation, but a completely different behavior that does not pattern with vowels in the language. Although “neutralization” is used in phonology to refer to the lack of phonetic contrast between phonemic categories in specific environments, we use this term to refer to neutralization regardless of context. To our knowledge, modifications such as the ones reported here have only been documented previously for IDS in Australian languages: Warlpiri (Laughren, 1984) and languages within the Arandic language family (Central Anmatyerr, Kaytetye, and Alywaar, Turpin et al., 2014).

Methods

Participants

Seven caregivers participated in this study (five females): two fathers, two mothers, two aunts, and one grandmother were recorded addressing three infants (one female) aged 6, 7, and 15 months. All caregivers lived in the same household as the infant, and none were first-time parents. No infants or caregivers had any reported health issues or language-related difficulties. All caregivers grew up as monolingual Tashlhiyt speakers. Three of them acquired Darija (Moroccan Arabic) after the age of 15 years, while others had no to very limited second language proficiency. Their education level ranged between illiterate (no formal schooling) and 6 years of schooling. The average household had six members, with at least two stay-at-home caregivers as well as one main breadwinner. The three participating infants all had at least one older sibling. All caregivers reported speaking only Tashlhiyt at home.

Materials

The caregivers were recorded using a head-worn cardioid microphone connected to a Dell XPS 15 laptop through a Focusrite Scarlett 2i2 audio interface. The recordings were saved as uncompressed WAV (i.e., Waveform Audio File Format) files at a sampling rate of 44.1 kHz. Videos depicting familiar objects and animals were used to elicit IDS and ADS, taken from open online sources.

Procedure

Each participant took part in two audio recording sessions in succession: one for IDS and another for ADS. Sessions took place in a quiet room in Agadir, Morocco. Infants took part in multiple sessions, but on different days to avoid fatigue and boredom, except for one infant who participated in two sessions with 30 min in between. Each session lasted 10 min, and recordings lasted, on average, 4 min per session. During the IDS sessions, infants sat on their caregiver’s lap facing a screen where a silent video was played. The video contained scenes of animals, babies, and cartoon animations. The caregivers were asked to draw the child’s attention to the video and describe what they saw to the child. Caregivers often engage in such “point and describe” activities in their daily life with their children, drawing their attention to objects or animals (e.g., “oh look, it’s a kitten! It’s playing with the ball”). During ADS sessions, infants were absent, and caregivers were asked to describe what they saw in the video to a Tashlhiyt-speaking adult male experimenter (the first author). The order of the sessions was counterbalanced. Four participants started with the IDS session, while the others started with the ADS.

Acoustic Analysis

Vowels

All recordings were annotated and analyzed in Praat (Boersma, 2001). We manually identified the onset and offset of target vowels (a/u/i) based on auditory information as well as visual inspection of the spectrogram. Vowels that were affected by noise (e.g., baby vocalizations), clipping issues, disfluencies, or were too short to establish a stable point for reliable formant estimation were not segmented. We also avoided segmenting vowels that were adjacent to labialized consonants, glides, and other vowels. These contexts can affect the F2 of the adjacent vowels, making it difficult to identify the vowel boundary or resulting in phonetically longer vowels, which might have skewed our durational measures. We coded each vowel for Word Position within the utterance (Initial, Medial, Final) and its word class (Content, Function). See Table B1 in Appendix B for the number of tokens for each condition.

A Praat script was used to automatically calculate the mean Pitch (f0), Duration, F1, F2 and F3 values from the middle 40% of each vowel to avoid any artifacts from the surrounding segments (e.g., nasals). Vowel durations were log-transformed for statistical analyses, while f0 values were converted to semitones, which better represent the logarithmic nature of the human perception of pitch. In Praat, we kept the default settings by setting formant ceilings at 5500 for Females and 5000 for Males. The formant values were normalized per speaker using Lobanov Transformation, which reduces the effects of talker-specific characteristics (e.g., vocal tract size) on the data. After normalization, we identified and excluded 38 vowel tokens as outliers because they fell outside the Inter-Quartile Range (IQR) by 1.5*IQR. Additionally, vowel tokens (N = 75) with a mean f0 higher than 350 Hz were excluded from the final dataset because they can potentially affect formant estimation accuracy in Praat. The final dataset included a total of 1051 vowel tokens.

Consonants

The recordings were transcribed for a second time in broad IPA (International Phonetic Alphabet) using Phon (Rose & MacWhinney, 2014), a software designed to analyze phonological data. Each target utterance was transcribed followed by its actual realization. We extracted all instances of the contrasts in gray in Table 1 (no tokens were excluded). A total of 1609 consonants were transcribed and analyzed. In this part of the study, we have based our results on transcriptions of two trained Tashlhiyt-speaking transcribers. Phon was then used to compute the extent of consonant neutralization using the Percentage of Consonants Correct score (PCC) analysis. The software returns the accuracy of each phoneme as well as how it was substituted. Agreement between the two transcribers was 100% for all contrasts except pharyngealized ones, which reached 79%.

Results

All statistical analyses were conducted in R (R Development Core Team, 2024) using the lme4 package (Bates et al., 2015). Emmeans (Lenth, 2024) was used for post-hoc tests. The ggplot2 (Wickham, 2011) and PhonR (Mccloy, 2012) packages were used to compute VSA size and visualize all plots. All code and data are available at https://osf.io/as98u/.

Vowels

In our first set of analyses, we focus on vowels, asking whether they are acoustically more salient in IDS compared to ADS. We describe several acoustic measures, including pitch, duration, VSA size, vowel overlap, and individual formant characteristics.

Pitch

We first asked whether Pitch was consistently different in IDS vowels compared to ADS. We fitted a linear mixed effects model with Pitch as the dependent variable (see Table 2 for model summary statistics), and with Register, Word Class, and Word Position included as fixed effects, Vowel identity and Speaker as random effects and with sliding-difference (repeated) contrast coding used for Word Position (Initial vowels were first compared to Medial vowels, then Medial vowels were compared to Final vowels). The only significant effect was Register (β = 7.72, SE = 0.57, t = 13.37, p < .001); pitch was consistently higher for IDS.

Table 2.

Summary of Statistical Models Estimating the Effect of Register, Word Class, and Vowel’s Word Position on Vowel Duration and Pitch.

Fixed effect	Duration model		Pitch model
Fixed effect	Beta^a	95% CI	Beta^a	95% CI
(Intercept)	4.43***	4.31, 4.55	16.7***	12.0, 21.5
Register	.02	−0.11, 0.16	7.72***	6.59, 8.85
Word Position
Utterance_Position.Initial.vs.Medial	.26*	0.04, 0.48	−1.07	−2.90, 0.76
Utterance_Position.Medial.vs.Final	.51***	0.31, 0.70	−.90	−2.56, 0.76
Word Class	.13***	0.06, 0.19	−.04	−0.57, 0.48
Register × Word Position
IDS× Utterance_Position.Initial.vs.Medial	.22	−0.21, 0.65	−2.73	−6.35, 0.89
IDS × Utterance_Position.Medial.vs.Final	.12	−0.27, 0.52	−1.75	−5.03, 1.53
Register × Word Class	.13*	0.01, 0.25	−0.83	−1.85, 0.18
AIC	1573		6511
BIC	1628		6567
R ² _conditional	.10		.72
R ² _marginal	.06		.23
ICC	.05		.63
RMSE	.46		3.86
RSE	.46		3.88

Note. CI = Confidence Interval; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; ICC = Intraclass Correlation Coefficient; RSE = Residual Standard Error; RMSE = Root-Mean-Square Error.

*p < .05. ***p < .001.

Duration

Here we compare vowel duration in IDS and ADS, and whether this is mediated by word class and word position (Table 3). We fit a mixed effects model with vowel duration (log-transformed) as the dependent variable, and with Register (sum-coded; ADS: −0.5, IDS: +0.5), Word Class (sum-coded; Content words: +0.5, Function words: −0.5), and Word Position (Initial/Medial/Final) included as fixed effects. Sliding-difference (repeated) contrast coding was used for Word Position (Initial compared to Medial and then Medial compared to Final). Vowel identity and Speaker were included as random effects. Table 2 presents the model summary results. There was no significant main effect of Register (β = .02, SE = 0.06, t = 0.33, p = .73), but there was a significant interaction between Register and Word Class (β = .13, SE = 0.06, t = 2.15, p = .03), suggesting that vowels in content words were longer in IDS than in function words. There was also a main effect of Word Position, indicating that word-final vowels were longer than word-medial vowels (β = .26, SE = 0.11, t = 2.33, p = .02) and that the latter were, in turn, longer than word-initial vowels (β = .51, SE = 0.10, t = 5.05, p < .001). However, there was no interaction with Register, suggesting that the final lengthening is not exaggerated in IDS compared to ADS.

Table 3.

Descriptive Statistics for Duration and Pitch Values (in ms and Hz) for Vowels in ADS and IDS Across Register, Word Class (Content and Function Words), and Word Position.

Register	Word class	Word position	Duration (SD)	Pitch (SD)
ADS	Content	Initial	74.78 (31.93)	181.21 (54.04)
IDS	Content	Initial	69.10 (20.56)	251.04 (64.20)
ADS	Function	Initial	71.44 (34.22)	169.98 (50.06)
IDS	Function	Initial	72.53 (19.72)	253.21 (68.50)
ADS	Content	Medial	87.79 (60.95)	181.07 (51.93)
IDS	Content	Medial	126.17 (120.93)	229.77 (70.94)
ADS	Function	Medial	83.74 (58.58)	182.01 (42.13)
IDS	Function	Medial	90.12 (88.93)	250.70 (65.63)
ADS	Content	Final	92.47 (44.59)	177.14 (49.39)
IDS	Content	Final	121.80 (101.52)	217.67 (66.50)
ADS	Function	Final	79.93 (33.56)	160.73 (53.36)
IDS	Function	Final	57.48 (13.48)	156.35 (60.47)

Note. ADS = Adult-Directed Speech; IDS = Infant-Directed Speech.

To explore these differences further, we investigated whether vowels in nouns were different in duration from those in other content words. Note that the latter category mostly consists of verbs because Tashlhiyt has very few adjectives. On a subset of data that only contained content words (see Table 4), we fitted a linear mixed effects model with Register (sum-coded; ADS: −0.5, IDS: +0.5), Content word class (sum-coded; Noun: +0.5 vs. Other content words: −0.5), and Word position (same coding structure as the previous model) as fixed effects, and with Speaker and Vowel Identity as random effects (see Table 5 for model summary statistics). A significant interaction between Register and Word Class (β = 0.27, SE = 0.07, t = 3.93, p < 0.001) showed that vowels in nouns in IDS were longer than those in Other Content words.

Table 4.

Descriptive Statistics for Duration (in ms) for Vowels in ADS and IDS Across Register, Word Class (Content Words Only), and Word Position.

Register	Word class	Word position	Duration (SD)
ADS	Noun	Initial	74.39 (15.71)
IDS	Noun	Initial	80.54 (26.42)
ADS	Other	Initial	74.91 (34.79)
IDS	Other	Initial	72.79 (23.40)
ADS	Noun	Medial	77.02 (39.36)
IDS	Noun	Medial	123.17 (111.89)
ADS	Other	Medial	99.56 (76.43)
IDS	Other	Medial	110.18 (113.67)
ADS	Noun	Final	113.23 (58.48)
IDS	Noun	Final	176.58 (119.50)
ADS	Other	Final	82.09 (32.04)
IDS	Other	Final	88.39 (65.59)

Note. ADS = Adult-Directed Speech; IDS = Infant-Directed Speech.

Table 5.

Summary of Statistical Models Estimating the Effect of Register, Word Class, and Vowel’s Word Position on Vowel Duration in Content Words.

Fixed effect	Beta^a	95% CI
(Intercept)	4.51***	4.38, 4.64
Register	.11	−0.05, 0.26
Word Position
Utterance_Position.Initial.vs.Medial	.16	−0.11, 0.44
Utterance_Position.Medial.vs.Final	.44***	0.21, 0.66
Word Class	.07	0.00, 0.14
Register × Word Position
IDS × Utterance_Position.Initial.vs.Medial	.18	−0.36, 0.72
IDS × Utterance_Position.Medial.vs.Final	.09	−0.35, 0.53
Register × Word Class	.27***	0.14, 0.41
AIC	1158
BIC	1210
R ² _conditional	.12
R ² _marginal	.07
ICC	.05
RMSE	.47
RSE	.48

***p < .001.

Vowel Space Area

Our third analysis evaluated the hyper-articulation hypothesis by examining whether Tashlhiyt-speaking caregivers expanded their VSA in IDS. A bigger VSA in IDS would indicate that caregivers hyper-articulated their vowels. The opposite pattern would instead suggest hypo-articulation. The descriptive statistics for mean formant values for each vowel across the two registers are shown in Table 6 (Raw) and Table 7 (Lobanov-normalized). We used the PhonR package (Mccloy, 2012) to compute and visualize the VSA (see Figure A1 in Appendix A for individual VSA plots). Figure 1 shows the acoustic vowel space triangle averaged across all caregivers for both registers (IDS vs. ADS). A 2-tailed paired-samples t-test with Register as the independent variable and VSA (Lobanov-normalized) as the dependent variable did not reveal any significant differences (t(6) = 0.29, p = 0.77, d = 0.12, 95% CI [−0.69, 0.92]) between the ADS (M = 1.57, SD = 0.23) and the IDS VSA (M = 1.52, SD = 0.34). These results do not support either of the previous predictions, revealing no evidence for either hyper-articulation or hypo-articulation of vowels in IDS. VSA expansion occurs when vowel formants are modified in ways that make the vowel triangle bigger (see Figure 1). However, these formants can be modified in other ways that do not make the triangle different in size but are still informative about IDS.

Table 6.

Descriptive Statistics of F1, F2, and F3 Values (in Hz) for All Vowels in ADS and IDS.

Register	Vowel	N	Mean F1 (SD)	Mean F2 (SD)	Mean F3 (SD)
ADS	/a/	338	650.83 (112.32)	1651.19 (215.55)	2866.06 (227.62)
IDS	/a/	199	631.89 (125.79)	1758.22 (221.71)	2898.01 (274.64)
ADS	/i/	235	448.21 (82.85)	2115.97 (208.74)	2906.32 (214.59)
IDS	/i/	117	435.91 (80.44)	2226.78 (245.34)	2982.71 (238.48)
ADS	/u/	85	502.51 (72.14)	1244.48 (217.96)	2700.13 (224.09)
IDS	/u/	77	496.40 (103.48)	1424.39 (303.87)	2822.10 (245.29)

Note. ADS = Adult-Directed Speech; IDS = Infant-Directed Speech.

Table 7.

Descriptive Statistics of F1, F2, and F3 Values (Lob-Normalized) for All Vowels by Register.

Register	Vowel	N	Mean F1 (SD)	Mean F2 (SD)	Mean F3 (SD)
ADS	/a/	338	0.65 (0.76)	−0.43 (0.53)	−0.11 (0.75)
IDS	/a/	199	0.73 (0.81)	−0.04 (0.61)	0.29 (0.95)
ADS	/i/	235	−0.91 (0.52)	0.90 (0.49)	0.09 (0.75)
IDS	/i/	117	−0.92 (0.61)	1.28 (0.59)	0.58 (0.81)
ADS	/u/	85	−0.44 (0.51)	−1.53 (0.62)	−0.69 (0.90)
IDS	/u/	77	−0.34 (0.56)	−0.92 (0.82)	0.13 (0.93)

Note. ADS = Adult-Directed Speech; IDS = Infant-Directed Speech.

Figure 1.

Vowel space area (Lobanov-normalized) for IDS (Dark Gray) and ADS (Light Gray).

Formant Analysis

Since we found no evidence for hyper- or hypo-articulation in the previous analysis, we asked whether vowel formants are modified in ways that indicate that caregivers are modifying affect instead of attempting to hyper-articulate vowels. To answer this question, we fitted three linear mixed models with, respectively, F1/F2/F3 (Lobanov-normalized) as the outcome variables (see Table 8 for model summaries). We included Register as a fixed effect (sum-coded; ADS: −0.5 vs. IDS: +0.5), while Speaker and Vowel Identity were added as random effects. There was a significant main effect of Register for the F2 (β = .44, SE = 0.03, t = 11.58, p < .001) and F3 models (β = .50, SE = 0.05, t = 9.46, p < .001), suggesting that, for all vowels, the second and third formant frequencies were higher across the board in IDS. For F1, however, the effect of Register was not significant (β = .06, SE = 0.04, t = 1.30, p = .19).

Table 8.

Summary of Statistical Models Estimating the Effect of Register on Vowel Formant Values.

Fixed effect	F1		F2		F3
Fixed effect	Beta^a	95% CI	Beta^a	95% CI	Beta^a	95% CI
(Intercept)	−.21	−2.22, 1.81	−0.12	−3.03, 2.79	.05	−0.73, 0.84
Register.IDS	.06	−0.03, 0.14	0.44***	0.36, 0.51	0.50***	0.40, 0.61
AIC	2191		1856		2604
BIC	2215		1881		2628
R ² _conditional	.59		.81		.18
R ² _marginal	.00		.02		.07
ICC	.59		.80		.12
RMSE	.67		.57		.82
RSE	.676		.574		.825

***p < .001.

Vowel Overlap

The next analysis tested whether vowel categories are more distinct in IDS by computing the degree of the overlap between any two given vowels. More overlap between vowels would be evidence for hypo-articulation, while less overlap indicates that the vowel contrasts are more salient in IDS. Both Pillai (computed through a MANOVA test) and Bhattacharyya scores in Figure 2 (Computed using the adehabitatHR package, Calenge, 2015) showed a greater mean overlap between the /u/-/i/ and /a-u/ contrasts in IDS than ADS (see Figure 3).

Figure 2.

Vowel category overlap scores for IDS and ADS: Higher Pillai scores indicate less overlap, while higher Bhattachariyya’s Affinity scores indicate more overlap.

Figure 3.

Ellipse plot visualizing mean vowel formant distributions for each of the three vowels in ADS (left plot) and IDS (right plot).

Consonant Contrasts

In our second set of analyses, we explore IDS modification in consonants. In particular, we report on a consonant neutralization behavior that characterizes IDS in Tashlhiyt, whereby caregivers seem to modify a set of phonemic contrasts (Figure 4) in IDS. These contrasts are phonemes that caregivers neutralize to varying degrees in the register. For example, one caregiver produced no /ʒ/ and /q/ tokens throughout the recordings, where they simply surfaced as /z/ and /k/, respectively. This behavior never occurred in the ADS recordings and is thus likely a property of Tashlhiyt IDS. When asked, all speakers indicated their awareness of this modification and said that they did it subconsciously, although it could be elicited as well.

We used Phon to calculate how often contrasts were neutralized. Since our transcriptions contained the target utterances as well as the actual utterances, we calculated how often a consonant’s surface realization matched its underlying representation (excluding mismatches resulting from other morpho-phonological processes). Our analyses in Phon returned an overall PCC of 65% (SD = 35%), which is the percentage of target phonemes that surfaced non-neutralized. Consonant contrasts varied in how often they were neutralized. As shown in Figure 4, average PCC scores for different contrasts ranged from 50% (SD = 31%) for the /r − l/ contrast to 87% for the /χ − ʜ/ contrast (SD = 17).

Figure 4.

Percentage of Consonant Correct for each consonant contrast (left panel) and speaker (right panel) in IDS. Higher PCC values indicate less neutralization occurrences.

Consonant contrast neutralization also varied by caregiver and ranged from 2% (SD = 2.5%) to 77% (SD = 22%), as can be seen in Figure 4. Similarly, speakers did not neutralize to the same extent across all contrasts. While two speakers completely neutralized /ʒ/ whenever it occurred in target words, another completely neutralized the /q/ phoneme. Additionally, pharyngealization (shown as “cˁ” in Figure 4), a secondary articulation that is distinctive in Tashlhiyt, was also subject to neutralization with an average PCC score of 52% (SD = 26%), with speakers ranging from 6% to 87% of pharyngealized phonemes. Regardless of the extent, all speakers engaged in some neutralization, which was never the case in ADS.

Discussion

In this paper, we asked whether segments – vowels and consonants – were realized differently in IDS compared to ADS in Tashlhiyt. With regards to vowels, we compared IDS and ADS vowels on measures of pitch, duration, VSA, vowel overlap, and individual vowel formants. Our results showed that vowels had consistently higher f0, making this the most salient marker of vowels in IDS in Tashlhiyt, in line with cross-linguistic findings (Cox et al., 2023). Additionally, vowel duration was longer in IDS, especially for content words and, within content words, especially for nouns. While VSA was constant across registers, Pillai and Bhattacharyya’s Affinity scores indicated that two of the three vowel Tashlhiyt contrasts have slightly more overlap in IDS. Our analysis of individual formant values also showed that F2 and F3 were consistently higher in IDS. With regards to consonants, we showed that caregivers neutralized certain consonant contrasts. These results indicate that caregivers adjust their speech when talking to infants, modifying vowels and consonants in different ways within the language.

Vowels

We found that VSA did not expand during IDS in Tashlhiyt, but that two out of three vowel categories showed more overlap. The result adds to the small number of studies showing that, in some languages, there is no clear-cut vowel hyper-articulation when caregivers engage with their infants (for English and Japanese, see Dodane & Al-Tamimi, 2007; and for Cantonese, see Xu Rattanasone et al., 2013). The lack of VSA expansion and increasing overlap does not support the hyper-articulation hypothesis, which argues that IDS is driven by the need to facilitate language development and that vowels should consequently be maximally distinct in the child’s input (Kuhl et al., 1997). Our VSA and vowel overlap results also suggest that hyper-articulation might not occur in systems with small vowel inventories, contrary to Lovcevic et al. (2025), who suggested that these languages are more likely to be hyper-articulated because they can afford to do so without risking too much overlap between categories. Instead, we found evidence for more overlap, even without VSA expansion, which is more consistent with the idea that acoustic salience in small vowel systems, and three-corner vowel systems specifically, does not necessarily increase in IDS. These results are consistent with vowel modifications in Warlpiri IDS, another language with a small vowel system (Bundgaard-Nielsen et al., 2023).

Cross-linguistically, vowels are often longer in IDS across all tokens. However, this might be simply because most studies only included content words in their data. We therefore compared content and function words. Indeed, vowels were longer in content words, which we followed up with a comparison of nouns to other content word classes. Interestingly, our results showed longer vowel durations in IDS for nouns than for other content words, which again replicates findings from Warlpiri (Bundgaard-Nielsen et al., 2023), a language with a similar vowel system. Cross-linguistically, noun production is associated with slower speech rates, which is argued to be due to their higher processing demands (Seifart et al., 2018). Warlpiri IDS also exhibits an expanded VSA for nouns, which we currently cannot directly examine for Tashlhiyt due to data limitations (e.g., not enough observations to calculate VSA for all conditions). Thus, considering that vowel duration and VSA expansion measures are highly correlated in IDS (Hartman et al., 2017; Shochi et al., 2009), it is very likely that vowels will also be hyper-articulated in Tashlhiyt IDS nouns, a result that would generally be consistent with a weaker version of the hyper-articulation hypothesis, where caregivers selectively adapt their speech to highlight the contrasts that children may struggle with instead of increasing acoustic salience across the board. Tashlhiyt and Warlpiri both have a compact vowel space with small inventories, which likely constrain, in comparable ways, the extent to which caregivers are able to modify speech when addressing children.

Our results also show an overall increase in pitch, F2 and F3 for all vowels in IDS. Increases in F2 and F3 are typically linked to a more advanced tongue position and more rounded lips, respectively. The literature suggests that these modifications result from speakers shortening their vocal tract, either in an attempt to mimic child speech or to seem non-threatening (Kalashnikova & Burnham, 2018). These modifications have also been argued to maintain infants’ attention (Segal & Newman, 2015; Trainor & Desjardins, 2002) and convey positive emotion (Benders, 2013; Trainor et al., 2000). Taken together, the VSA size, vowel overlap, pitch, duration, and vowel formant results are consistent with the view that, overall, the purpose of IDS is mainly to serve communicative (attentional and affective) functions.

Consonants

Our analysis of consonant modification in Tashlhiyt revealed an as-yet undocumented feature of IDS outside Australian languages (Laughren, 1984), in which some contrasts (e.g., /r/ and /l/) are completely neutralized to varying degrees by caregivers. For instance, when the underlying phoneme /r/ occurred in a word (e.g., /takurt/ “ball”), it was often realized as [l] (e.g., /takult/). This surprising behavior may result in the elimination of certain phonemes from the input, which in principle poses challenges for the acquisition of the full native inventory by the infant, at least through IDS. Note that there were considerable differences in consonant contrast neutralization behavior between contrasts and speakers (Figure 4). While contrasts such as /ʜ/ − /χ/ were only neutralized half of the time on average, others, such as /z/ − /ʒ/, were neutralized by some speakers all the time. In addition to the neutralization of these contrasts, secondary articulations such as pharyngealization were affected. Up to 92% of pharyngealized consonants were “de-pharyngealized” by some speakers. These results, similar to those of Benders et al. (2019) show that even when the phonological space is crowded (as is the case with the Tashlhiyt consonants), caregivers may avoid hyper-articulation.

This behavior in Tashlhiyt-speaking caregivers seems, at face value, to be incompatible with the idea of a listener-oriented didactic function of IDS. An alternative explanation could be that caregivers are imitating how infants (mis-)produce these sets of contrasts, a common form of caregiver-infant play. However, this behavior is present in IDS addressed very young infants, long before the infants’ first attempt at vocalizations. At the same time, neutralization appears to be a feature of IDS, since all the caregivers engaged in it to some degree. This may then reflect an implicit cultural understanding of the complexities of Tashlhiyt’s consonant inventory, with caregivers selectively neutralizing contrasts thought to be the most challenging sounds to their infant.

Similar consonant neutralization has been reported for IDS in some Australian languages. We again draw similarities with Warlpiri, which collapses all coronal Places of Articulation (PoAs) into a single category (Laughren, 1984). Warlpiri’s consonant inventory is not comparable in size to Tashlhiyt, but the coronal PoAs that are subject to neutralization can be perceptually challenging. It might then be the case that inventory size per se is not responsible for segmental modification in IDS, but rather the phonetic salience of specific contrasts. In addition to Nepali, which shows evidence for hypo-articulation, Warlpiri and Tashlhiyt IDS contain the most challenging consonants examined so far. All these contrasts, despite being less salient, were subject to hypo-articulation and even complete neutralization. And while we do not currently consider neutralization to be an extreme case of hypo-articulation (e.g., not a relative reduction of salience), but rather an example of a phonological substitution process (i.e., phonetic differences are made absent), acoustic data is needed to better understand how these phonemes surface in IDS. This is especially true for less salient contrasts, since they are difficult to judge impressionistically (e.g., agreement on pharyngealized consonants was 79%). Regardless of the nature of these modifications, the way these phonemes are realized in Tashlhiyt goes against a didactic view of IDS, which would only be supported by evidence for hyper-articulation, especially for challenging contrasts.

There were also differences in the amount of neutralization across individuals and phonemes. We do not know whether there are also such differences in Warlpiri’s IDS, since these were not reported, but we think it is likely that modification patterns, like neutralization, will also show variability between speakers and possibly phonemes in other languages beyond Tashlhiyt (Figure 4). This alters the input characteristics in different ways, including the frequency of phonemes (if at all) and even whole phoneme classes (e.g., rhotics). And while both languages similarly avoid rhotics, Tashlhiyt extends the neutralization behavior to PoAs other than coronals (e.g., uvulars). Despite these differences, the commonality is that caregivers from both language groups seem sensitive to the articulatory effort some phoneme contrasts place on infants and modify their productions by keeping the most peripheral of PoAs (Table 1). That is, modifications made in IDS may be influenced by both infants’ perceived biomechanical limitations and their perceptual ones. For Tashlhiyt, this is especially true with the neutralization of articulatorily complex pharyngealization and labialization contrasts (Buech et al., 2022), which, in addition to the consonant set in Figure 4, could (maximally) reduce the inventory size in Table 1 from 71 down to 32 phonemes when neutralized.

Prior findings link hyper-articulation to better language outcomes (Kalashnikova & Burnham, 2018; Lovcevic et al., 2024), while others do not find evidence for detrimental hypo-articulation effects (Rosslund et al., 2025). This poses the question of how neutralization might affect consonant acquisition in Tashlhiyt (and, indeed, Warlpiri). Undoubtedly, the behavior of Tashlhiyt caregivers may completely or partially neutralize certain consonants from the input. However, this does not mean Tashlhiyt-style IDS might not serve a facilitative role in learning. One possibility is that the temporary reduction in the number of consonant categories in the input allows children to break into the system more easily, and this early knowledge is used to bootstrap into the full adult system. The suggestion is that this pattern could be another example of how caregivers adapt to the developmental needs of their infants. In Cantonese, which has six tones, mothers show evidence for tone hyper-articulation before their infants reach 4 months of age, but stop doing so when they reach 11 months (Xu Rattanasone et al., 2013). Similarly, it could be that in Tashlhiyt, parents try to adapt to their infant’s perceived abilities by allowing them to acquire a certain number of segments before introducing them to the rest of the inventory.

Limitations

Our study has some limitations that may have affected participants’ behavior. Our experimenter was a male native speaker of Tashlhiyt and a community member. However, despite his status as an in-group member, in rural areas of Morocco, it is socially normative for males and females to exercise a degree of social distance, especially if they are not related. Thus, the experimenter’s gender might have inhibited IDS elicitation. This was clear in the audio recordings of one female caregiver, who was not related to the experimenter and who sporadically inhibited what they called “silly talk” throughout the recording (P002), unlike female caregivers who were close relatives of the experimenter (P008). The relatively formal nature of the recording context may have also influenced our results. Specifically, we anecdotally observed that caregivers engage in neutralization more during intimate moments with their infants, which further reinforces the possibility that it may have an affective function. However, the opportunity to engage in such intimate moments was limited in our study, where dyads were largely engaging in triadic communication via their focus on the video. Thus, our recording context may have inhibited neutralization due to the lack of face-to-face interaction between the caregiver and the infant and may have underestimated the extent of the neutralization phenomenon.

This leads to a suggestion for future research that is rarely incorporated into studies of IDS – the influence of non-verbal behavior. That is, if neutralizations are more frequent in face-to-face contexts that have distinct affective valency, they may also co-occur with non-verbal signals (e.g., haptic cues like touch and facial expressions like smiling) that cue children to language. This suggestion awaits further research. At the same time, our elicitation method was not without value: although it may have impacted the segmental characteristics and elicited IDS exhibiting less consonant neutralization than usual, we do not believe it impacted other features of IDS, such as prosody. This is because the point-and-describe interaction is a daily activity in the lives of Tashlhiyt-acquiring infants. Finally, it is worth noting that because we measured pitch on vowels, this meant that vowelless words, which Tashlhiyt allows, were automatically excluded. Therefore, the pitch results do not present the whole picture of how pitch is implemented in Tashlhiyt’s IDS using other pitch-bearing elements, as is the case with Tashlhiyt’s ADS (Grice et al., 2015).

Conclusion

In this paper, we have presented the first study of IDS in Tashlhiyt Berber, a language that is typologically interesting in the context of phonological acquisition due to the contrast in the size of its vowel (small) and consonant (very large) inventories. Thus, the language represents an important data point for the field, which has consistently studied languages from a narrow typological range (see Passmore et al., 2025). We found a number of notable results. Firstly, Tashlhiyt-speaking caregivers do not hyper-articulate vowel categories when addressing infants, despite (or perhaps, due to) the language having a compact vowel space. The size of vowel space was similar across IDS and ADS registers, suggesting that vowels are not particularly clearer in IDS. Rather, our results revealed evidence for more overlap in IDS, suggesting that the opposite pattern (hypo-articulation) may be true. Overall, the pattern of results suggested that, for vowels, modifications in IDS may be more geared to direct attention and convey positive emotion rather than to help children distinguish between different vowels. For consonants, our findings revealed a surprising pattern of modification where caregivers neutralized certain consonants from the input. These findings suggest that caregivers do not always hyper-articulate phonemic segments regardless of the inventory size and phonetic salience, and may choose to modify vowels and consonants differently within the same language. The lack of VSA expansion, the increasing vowel overlap, and in particular, the consonant neutralization provide strong evidence against the hyper-articulation hypothesis. However, possible facilitative strategies (e.g., the temporary reduction of the consonant inventory size) could still lend support for the view that IDS may have a didactic function even without hyper-articulation. Finally, the contrast neutralization behavior raises questions about the challenges infants face during consonant acquisition, as well as what motivates caregivers to follow such behavior.

Footnotes

Appendix A

Appendix B

Table B1.

Number of Tokens for Each Vowel, Register, and Word Class.

Vowel Identity	/a/		/i/		/u/		Total
Register	ADS (N = 33)	IDS (N = 19)	ADS (N = 23)	IDS (N = 11)	ADS (N = 85)	IDS (N = 77)	ADS (N = 658)	IDS (N = 393)
Word Class
Noun	89 (26.3%)	79 (39.7%)	44 (18.7%)	15 (12.8%)	51 (60.0%)	46 (59.7%)	184 (28.0%)	140 (35.6%)
Function	156 (46.2%)	60 (30.2%)	65 (27.7%)	33 (28.2%)	8 (9.4%)	5 (6.5%)	229 (34.8%)	98 (24.9%)
Other Content	93 (27.5%)	60 (30.2%)	126 (53.6%)	69 (59.0%)	26 (30.6%)	26 (33.8%)	245 (37.2%)	155 (39.4%)

Note. ADS = Adult-Directed Speech; IDS = Infant-Directed Speech.

Acknowledgements

We thank Esther Janse, Janine Berns, and Zara Harmon for their comments, and Salah Eddine Hafed for helping with IPA transcription. We would also like to thank the families who participated in this study.

ORCID iDs

Abdellah Elouatiq

Christina Bergmann

Evan Kidd

Caroline F. Rowland

Ethical Considerations

The Faculty of Social Sciences Ethics Review Committee at Radboud University approved our study (Approval Registration Number: ECSW-2022-052) on June 24th, 2022.

Consent to Participate

Respondents gave written and/or verbal (recorded) informed consent and signature before starting the recording session.

Author Contributions

Abdellah Elouatiq: Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Software; Visualization; Writing – original draft.

Christina Bergmann: Conceptualization; Methodology; Writing – review & editing.

Evan Kidd: Supervision; Validation; Writing – review & editing.

Caroline Rowland: Funding acquisition; Supervision; Validation; Writing – review & editing.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by the Max Planck Society.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

Data, materials, and analysis code are available at the Open Science Framework project’s page, at the permanent link: .

Notes

References

Bates

Mächler

Bolker

Walker

(2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1), 1–48.

Benders

(2013). Mommy is only happy! Dutch mothers’ realisation of speech sounds in infant-directed speech expresses emotion, not didactic intent. Infant Behavior and Development, 36(4), 847–862. https://doi.org/10.1016/j.infbeh.2013.09.001

Benders

Pokharel

Demuth

(2019). Hypo-articulation of the four-way voicing contrast in Nepali infant-directed speech. Language Learning and Development, 15(3), 232–254. https://doi.org/10.1080/15475441.2019.1577139

Boersma

(2001). Praat, a system for doing phonetics by computer. Glot International, 5(9/10), 341–345.

Buech

Ridouane

Hermes

(2022). Pharyngealization in Amazigh: Acoustic and articulatory marking over time. Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2022-September, Incheon, South Korea. pp.3448-3452. https://doi.org/10.21437/Interspeech.2022-10831

Bundgaard-Nielsen

R. L.

O’Shannessy

Wang

Nelson

Bartlett

Davis

(2023). Two-part vowel modifications in Child Directed Speech in Warlpiri may enhance child attention to speech and scaffold noun acquisition. Phonetica, 80(1–2), 1–42. https://doi.org/10.1515/phon-2022-0039

Calenge

(2015). Home range estimation in R: The adehabitatHR package. https://cran.r-project.org/web/packages/adehabitatHR/vignettes/adehabitatHR.pdf

Cooper

R. P.

Aslin

R. N.

(1990). Preference for infant-directed speech in the first month after birth. Child Development, 61(5), 1584–1595. https://doi.org/10.1111/j.1467-8624.1990.tb02885.x

Cox

Bergmann

Fowler

Keren-Portnoy

Roepstorff

Bryant

Fusaroli

(2023). A systematic review and Bayesian meta-analysis of the acoustic features of infant-directed speech. Nature Human Behaviour, 7(1), 114–133. https://doi.org/10.1038/s41562-022-01452-1

10.

Cristia

Seidl

(2014). The hyperarticulation hypothesis of infant-directed speech. Journal of Child Language, 41(4), 913–934. https://doi.org/10.1017/S0305000912000669

11.

Cutler

(2012). Native listening: Language experience and the recognition of spoken words. Mit Press.

12.

Dodane

Al-Tamimi

(2007). An acoustic comparison of vowel systems in adult-directed-speech and child-directed speech: Evidence from French, English & Japanese. In Proceedings of the 16th International Congress of Phonetics Sciences, (pp. 1573–1576).

13.

Dunst

C. J.

Gorman

Hamby

D. W.

(2012). Preference for infant-directed speech in preverbal young children. Center for Early Literacy Learning Reviews, 5(1), 1–13.

14.

Englund

(2005). Voice onset time in infant directed speech over the first six months. First Language, 25(2), 219–234.

15.

Ferguson

C. A.

(1964). Baby talk in six languages. American Anthropologist, 66(6), 103–114.

16.

Ferjan Ramírez

Weiss

Sheth

K. K.

Kuhl

P. K

. (2024). Parentese in infancy predicts 5-year language complexity and conversational turns. Journal of Child Language, 51(2), 359–384. https://doi.org/10.1017/S0305000923000077

17.

Garnier

Ménard

Alexandre

(2018). Hyper-articulation in Lombard speech: An active communicative strategy to enhance visible speech cues? The Journal of the Acoustical Society of America, 144(2), 1059. https://doi.org/10.1121/1.5051321

18.

Genovese

Spinelli

Romero Lauro

L. J.

Aureli

Castelletti

Fasolo

(2019). Infant-directed speech as a simplified but not simple register: A longitudinal study of lexical and syntactic features. Journal of Child Language, 47(1), 22–44. https://doi.org/10.1017/S0305000919000643

19.

Green

J. R.

Nip

I. S. B.

Wilson

E. M.

Mefferd

A. S.

Yunusova

(2010). Lip movement exaggerations during infant-directed speech. Journal of Speech, Language, and Hearing Research, 53(6), 1529–1542. https://doi.org/10.1044/1092-4388(2010/09-0005)

20.

Grice

Ridouane

Roettger

T. B.

(2015). Tonal association in Tashlhiyt Berber: Evidence from polar questions and contrastive statements. Phonology, 32(2), 241–266. https://doi.org/10.1017/S0952675715000147

21.

Han

de Jong

N. H.

Kager

(2021). Language specificity of infant-directed speech: Speaking rate and word position in word-learning contexts. Language Learning and Development, 17(3), 221–240. https://doi.org/10.1080/15475441.2020.1855182

22.

Hartman

K. M.

Ratner

N. B.

Newman

R. S.

(2017). Infant-directed speech (IDS) vowel clarity and child language outcomes. Journal of Child Language, 44(5), 1140–1162. https://doi.org/10.1017/S0305000916000520

23.

Heymann

Heflin

B. H.

Baralt

Bagner

D. M.

(2020). Infant-directed language following a brief behavioral parenting intervention: The importance of language quality. Infant Behavior and Development, 58, Article 101419. https://doi.org/10.1016/j.infbeh.2019.101419

24.

Johnson

Flemming

Wright

(1993). The hyperspace effect: Phonetic targets are hyperarticulated. Language, 69(3), 505–528. https://doi.org/10.2307/416697

25.

Kalashnikova

Burnham

(2018). Infant-directed speech from seven to nineteen months has similar acoustic properties but different functions. Journal of Child Language, 45(5), 1035–1053. https://doi.org/10.1017/S0305000917000629

26.

Kidd

Garcia

(2022). How diverse is child language acquisition research? First Language, 42(6), 703–735. https://doi.org/10.1177/01427237211066405

27.

Kidd

Hellwig

Garcia

Defina

Davidson

Allen

(2025). A comparative study of child-directed language across five cultures based on data from the Acquisition Sketch Project. Australian Journal of Linguistics, 45(3), 391–415 https://doi.org/10.1080/07268602.2025.2514825

28.

Kitamura

Lam

(2009). Age-specific preferences for infant-directed affective intent. Infancy, 14(1), 77–100. https://doi.org/10.1080/15250000802569777

29.

Kuhl

P. K.

Andruski

J. E.

Chistovich

I. A.

Chistovich

L. A.

Kozhevnikova

E. V.

Ryskina

V. L.

Stolyarova

E. I.

Sundberg

Lacerda

(1997). Cross-language analysis of phonetic units in language addressed to infants. Science, 277(5326), 684–686. https://doi.org/10.1126/science.277.5326.684

30.

Kuhl

P. K.

Conboy

B. T.

Coffey-Corina

Padden

Rivera-Gaxiola

Nelson

(2007). Phonetic learning as a pathway to language: New data and native language magnet theory expanded (NLM-e). Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1493), 979–1000. https://doi.org/10.1098/rstb.2007.2154

31.

Lahrouchi

Kern

(2018). From babbling to first words in Tashlhiyt language acquisition: Longitudinal two-case studies. Canadian Journal of Linguistics, 63(4), 493–526. https://doi.org/10.1017/cnj.2018.6

32.

Laughren

(1984). Warlpiri baby talk. Australian Journal of Linguistics, 4(1), 73–88. https://doi.org/10.1080/07268608408599321

33.

Lenth

R. V.

(2024). emmeans: Estimated marginal means, aka least-squares means (R Package Version 1.10.2.090002)

34.

Lovčević

Benders

Tsuji

Fusaroli

(2025). Acoustic exaggeration of vowels in infant-directed speech: A multimethod meta-analytic review. Psychological Bulletin, 151(6), 669–695. https://doi.org/10.1037/bul0000479

35.

Lovcevic

Burnham

Kalashnikova

(2024). Infants’ Lexical Processing: Independent Contributions of Attentional and Clarity Cues. Language Learning and Development, 20(1), 1–18. https://doi.org/10.1080/15475441.2022.2149402

36.

Lynch

(1998) Pacific Languages: An Introduction. Honolulu, HI: University of Hawai’i Press.

37.

Marklund

Gustavsson

(2020). The dynamics of vowel hypo- and hyperarticulation in swedish infant-directed speech to 12-month-olds. Frontiers in Communication, 5, 523768. https://doi.org/10.3389/fcomm.2020.523768

38.

Martin

Igarashi

Jincho

Mazuka

(2016). Utterances in infant-directed speech are shorter, not slower. Cognition, 156, 52–59. https://doi.org/10.1016/j.cognition.2016.07.015

39.

Mccloy

D. R.

(2012). Vowel normalization and plotting with the phonR package (Technical Reports of the UW Linguistic Phonetics Laboratory, 3).

40.

McRoberts

G. W.

McDonough

Lakusta

(2009). The role of verbal repetition in the development of infant speech preferences from 4 to 14 months of age. Infancy, 14(2), 162–194. https://doi.org/10.1080/15250000802707062

41.

Moran

McCloy

Wright

(Eds.). (2014). PHOIBLE Online. Max Planck Institute for Evolutionary Anthropology. http://phoible.org/

42.

Narayan

C. R.

McDermott

L. C.

(2016). Speech rate and pitch characteristics of infant-directed speech: Longitudinal and cross-linguistic observations. The Journal of the Acoustical Society of America, 139(3), 1272–1281. https://doi.org/10.1121/1.4944634

43.

Newport

E. L.

Gleitman

(1984). The current status of the motherese hypothesis. Journal of Child Language, 11(1), 43–79. https://doi.org/10.1017/S0305000900005584

44.

Ota

Skarabela

(2016). Reduplicated words are easier to learn. Language Learning and Development, 12(4), 380–397. https://doi.org/10.1080/15475441.2016.1165100

45.

Ota

Skarabela

(2018). Reduplication facilitates early word segmentation. Journal of Child Language, 45(1), 204–218. https://doi.org/10.1017/S0305000916000660

46.

Passmore

Hellwig

Garcia

Kidd

(2025). The scientific and cultural cost of convenience sampling in the face of rising language endangerment: Highlighting the role of language acquisition. Open Mind, 9, 501–514. https://doi.org/10.1162/opmi_a_00199

47.

Piazza

Martin

C. D.

Kalashnikova

(2022). The acoustic features and didactic function of foreigner-directed speech: A scoping review. Journal of Speech, Language, and Hearing Research, 65(8), 2896–2918. https://doi.org/10.1044/2022_JSLHR-21-00609

48.

R Development Core Team. (2024). R: A Language and Environment for Statistical Computing (Vol. 55).

49.

Ridouane

(2008). Syllables without vowels: Phonetic and phonological evidence from Tashlhiyt Berber. Phonology, 25(2), 321–359. https://doi.org/10.1017/S0952675708001498

50.

Ridouane

(2014). Tashlhiyt Berber. Journal of the International Phonetic Association, 44(2), 207–221. https://doi.org/10.1017/S0025100313000388

51.

Rose

MacWhinney

(2014). The Phonbank Project: Data and software-assisted methods for the study of phonology and phonological development. In Durand

Gut

Kristoffersen

(Eds.), The Oxford Handbook of Corpus Phonology (pp. 380–401). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199571932.013.023

52.

Rosslund

Varjola

Mayor

Kartushina

(2025). Longitudinal changes in consonant production in infant-directed speech and infants’ early speech production from 6 to 12 months. Infant Behavior and Development, 78, Article 102018. https://doi.org/https://doi.org/10.1016/j.infbeh.2024.102018

53.

Segal

Newman

R. S.

(2015). Infant preferences for structural and prosodic properties of infant-directed speech in the second year of life. Infancy, 20(3), 339–351. https://doi.org/10.1111/infa.12077

54.

Seifart

Strunk

Danielsen

Hartmann

Pakendorf

Wichmann

Witzlack-Makarevich

De Jong

N. H.

Bickel

(2018). Nouns slow down speech across structurally and culturally diverse languages. Proceedings of the National Academy of Sciences of the United States of America, 115(22), 5720–5725. https://doi.org/10.1073/pnas.1800708115

55.

Shochi

Sekiyama

Lees

Boyce

Göcke

Burnham

(2009). Auditory-visual infant directed speech in Japanese and English. Auditory-Visual Speech Processing 2009, AVSP 2009, University of East Anglia, Norwich, UK.

56.

Stärk

Kidd

Frost

R. L. A.

(2022). Word segmentation cues in German child-directed speech: A corpus analysis. Language and Speech, 65(1), 3–27. https://doi.org/10.1177/0023830920979016

57.

Tang

Xu Rattanasone

Yuen

Demuth

(2017). Phonetic enhancement of Mandarin vowels and tones: Infant-directed speech and Lombard speech. The Journal of the Acoustical Society of America, 142(2), 493. https://doi.org/10.1121/1.4995998

58.

Trainor

L. J.

Austin

C. M.

Desjardins

R. N.

(2000). Is infant-directed speech prosody a result of the vocal expression of emotion? Psychological Science, 11(3), 188–195. https://doi.org/10.1111/1467-9280.00240

59.

Trainor

L. J.

Desjardins

R. N.

(2002). Pitch characteristics of infant-directed speech affect infants’ ability to discriminate vowels. Psychonomic Bulletin and Review, 9(2), 335–340. https://doi.org/10.3758/BF03196290

60.

Turpin

Demuth

Campbell

A. N.

(2014). Phonological aspects of Arandic baby talk. In Pensalfini

Turpin

Guillemin

(Eds.), Language Description Informed by Theory (pp. 49–80). John Benjamins Publishing Company. https://doi.org/doi:10.1075/slcs.147.04tur

61.

Wang

Seidl

Cristia

(2016). Acoustic characteristics of infant-directed speech as a function of prosodic typology. In Heinz

Goedemans

van der Hulst

(Eds.), Dimensions of Phonological Stress (pp. 311–324). Cambridge University Press. https://doi.org/10.1017/9781316212745.012

62.

Wickham

(2011). Ggplot2. Wiley Interdisciplinary Reviews: Computational Statistics, 3(2), 180–185. https://doi.org/10.1002/wics.147

63.

Burnham

Kitamura

Vollmer-Conna

(2013). Vowel hyperarticulation in parrot-, dog- and infant- directed speech. Anthrozoos, 26(3), 373–380. https://doi.org/10.2752/175303713X13697429463592

64.

Xu Rattanasone

Burnham

Reilly

R. G

. (2013). Tone and vowel enhancement in Cantonese infant-directed speech at 3, 6, 9, and 12 months of age. Journal of Phonetics, 41(5), 332–343. https://doi.org/10.1016/j.wocn.2013.06.001