Is KO OK? Segmentation of Consonant- and Vowel-Initial Words in German-Learning Infants

Abstract

Infants develop the ability to segment words from continuous speech from the age of 6 months onwards. However, in French- and English-learning infants, this ability initially appears to be restricted to consonant-initial words. Vowel-initial word segmentation only emerges at around 13.5 months of age, or at 11 months when additional segmentation cues are provided. Previous research has attributed this asymmetry to phonological and acoustic differences between word-initial consonants and vowels. The present study examined whether distributional properties such as the likelihood of encountering vowel-initial words also contribute to this developmental pattern. For this, 114 infants learning German, a language with a higher proportion of vowel-initial words than English and French, were tested. Using the head-turn preference procedure, 11-month-olds (Experiment 1) and 8-month-olds (Experiment 2) were familiarized with isolated words beginning with either consonants or vowels. Subsequently, their preference for listening to text passages containing these words was assessed. While the 8-month-olds showed no evidence of segmentation for either word type, the 11-month-olds successfully segmented both consonant- and vowel-initial words, showing no advantage for either of the word types. The results support an input-based account: Experience with a language richer in vowel-initial words may facilitate earlier segmentation of vowel-initial forms. Overall, the findings highlight the influence of language-specific distributional properties on early segmentation biases and underscore the need for research with infants learning structurally diverse languages.

1 Introduction

One of the key milestones in infants’ first year of life is learning to segment words from continuous speech. It is assumed that the ability to determine where words begin and end in the speech stream is crucial for lexical acquisition (see Jusczyk, 1997). However, this task is non-trivial, as the speech signal lacks clear word boundary markers such as pauses. Evidence suggests that by the age of 7.5 months, infants rely on multiple cues available in their ambient language to detect word boundaries in fluent speech. For example, they show the ability to identify words embedded in spoken text passages if these words occur next to high-frequency words from around the age of 6 months (e.g., mommy: Bortfeld et al., 2005; determiners: Shi et al., 2006). In addition to such lexical cues contained in their language input, infants are sensitive to language-specific acoustic/phonological segmentation cues provided by prosodic patterns (e.g., stress: Echols et al., 1997; Jusczyk et al., 1999b; prosodic boundaries: Christophe et al., 2003), allophonic variation (e.g., aspiration: Jusczyk, Hohne, & Bauman, 1999a), and phonotactic regularities (e.g., phoneme co-occurrence probability: Mattys & Jusczyk, 2001b; vowel harmony: Omane et al., 2025). Most of these studies have focused on infants’ segmentation of consonant-initial (henceforth, C-initial) words, based on findings that vowel-initial (henceforth, V-initial) words are more difficult to segment early on (e.g., Mattys & Jusczyk, 2001a). Despite abundant evidence on infants’ segmentation abilities, relatively little is known about the mechanisms underlying this ability, and how properties of the input language influence infants’ integration of these cues over the course of development. The present study addresses this by examining the assumed bias to segment C- but not V-initial words in 8- and 11-month-old infants learning German, a language differing from previously studied ones in its distribution of C- vs. V-initial words.

Regarding the development of infants’ ability to segment words from speech, young infants’ initial segmentation routines reflect a strong reliance on cues favoring the segmentation of words composed of sound patterns that are predominant in the input language, and only later, they are able to segment words that deviate from the default pattern. A prominent example of this is Jusczyk et al.’s (1999b) seminal study that found that English-learning 7.5-month-olds successfully segmented trochaic (strong-weak) bisyllabic words such as hamlet but failed to segment iambic (weak-strong) words. In fact, when familiarized with iambs such as guitar, they mis-parsed them (e.g., “guitar is” as “taris”). Only 3 months later—at the age of 10.5 months—did the English-learning infants succeed in segmenting iambic words. These results indicate a robust bias toward segmenting trochaic words, which is a favorable strategy, as English has relatively fewer iambic than trochaic words. However, these results also indicate that infants gradually adjust their reliance on segmentation cues to match the properties of their language input. In the present study, we investigate factors that contribute to infants’ adjustment of language-inherent “default” segmentation cues, focusing on the contrast between C- and V-initial words.

Previous research suggests that infants’ ability to segment words from speech is, at first, restricted to C-initial words. For example, when familiarized with the words ice and ash, English-learning infants do not show any signs of recognizing these monosyllabic words when subsequently hearing them embedded in text passages (e.g., in dice or cold ice) before the age of 16.5 months (Mattys & Jusczyk, 2001a), while C-initial words are already segmented at the age of 8 months (e.g., Jusczyk & Aslin, 1995). Follow-up research revealed that contextual factors, that is, the presence of additional segmentation cues, can facilitate the segmentation of V-initial words, leading infants to spot them at an earlier age. Nazzi and colleagues (Nazzi et al., 2005) reported that English-learning 13.5-month-olds segmented bisyllabic V-initial words if these followed the trochaic rather than the iambic stress pattern. Moreover, Seidl and Johnson (2008) showed that English-learning 11-month-olds successfully segmented V-initial words if they occurred in phrase-initial or phrase-final positions, presumably because in these positions, acoustic cues to the vowels are stronger than in the phrase-medial position. Furthermore, Kim and Sundara (2015) found that English-learning 11- but not 8-month-olds successfully segmented V-initial words in the phrase-medial position in the context of a lexical cue, namely if they occurred after the determiner the /ði/ (Kim & Sundara, 2015). In the same study, Kim and Sundara also probed whether the presence of an allophonic glottal stop preceding the word-initial vowel, which in English often naturally occurs in such contexts, would enable 11-months-olds to segment V-initial words, but they found no evidence for (or against) this. These results suggest that compared to the segmentation of C-initial words, the ability to segment V-initial words is developed later and depends on the infants’ ability to flexibly integrate multiple speech segmentation cues.

Another factor that relates to infants’ ability to segment V-initial words is their native language and its phonological properties. French-learning infants only segment V-initial words at the age of 24 months (Babineau & Shi, 2014, 2016). At 20 months, they still mis-segment V-initial target words as C-initial. This indicates that French-learning infants maintain a strong bias favoring a C-initial segmentation for longer than English-learning infants, and it is likely that this relates to the phenomenon of liaison: In French, word-final silent consonants are pronounced when followed by a V-initial word (e.g., les = [le], amis = [ami], but les‿amis [lezami]). This process can obscure word boundaries, making the segmentation of V-initial words more difficult for French-learning infants. Babineau and Shi (2014) propose that 20-month-old French-learning infants misinterpret the coda in liaison as the onset of the succeeding word, and thereby fail to segment V-initial words. To shed further light on the question of how input language characteristics impact the ability to segment C- and V-initial words, the present study will extend this research to infants learning German.

Three main explanations have been proposed to account for the delay in segmenting V- compared to C-initial words: (a) phonological, (b) acoustic, and (c) input-based accounts. First, the phonological account is based on a proposed universal bias against onsetless syllables (i.e., V-initial) and a preference for syllables to start with consonants in phonological grammar (e.g., Jakobson, 1962, p. 527; Kager, 1999, p. 92). These claims are based on typological evidence that most languages favor words starting with consonants (e.g., Blevins, 1995). Babineau and Shi (2014, 2016) interpret their findings as supported by this account.

Second, the acoustic account suggests that the delay in segmenting V-initial words arises because vowel onsets are acoustically less salient and lack clear, abrupt temporal boundaries characteristic of consonant onsets, making them more difficult to perceive and recognize (Mattys & Jusczyk, 2001a; Seidl & Johnson, 2008). V-initial words also tend to undergo resyllabification with a preceding C-final word (Nazzi et al., 2005; Seidl & Johnson, 2008), and these initial vowels are more affected by coarticulation from neighboring sounds than initial consonants (Seidl & Johnson, 2008). Seidl and Johnson (2008), who directly tested the acoustic account, interpret infants’ earlier ability to segment V-initial words in acoustically salient positions as supporting evidence for this explanation. Both Mattys and Jusczyk (2001a) and Nazzi et al. (2005) interpret their findings as consistent with this account, while noting that further research is needed: they predict that under the acoustic account, the segmentation of C-initial words beginning with continuant consonants (i.e., fricatives, nasals, liquids, glides) should be as difficult as segmenting V-initial words because similar to V-initial words, continuant-initial words lack clear acoustic discontinuities that would facilitate segmentation, unlike plosive-initial words. However, Kim and Sundara (2015) found no evidence that infants’ segmentation performance improves when word-initial vowels were preceded by a glottal stop, an acoustic marker that disrupts vowel continuity.

Finally, the input-based account claims that it is more difficult to segment V- compared to C-initial words because the former are relatively rarer in infant- and adult-directed speech (ADS; Mattys & Jusczyk, 2001a mention that only 19%–20% of all English words begin with vowels, referring to Swingley, 1999, and unpublished data by Brent & Siskind). As infants are sensitive to the positional and sequential phonotactic probabilities of their ambient language (e.g., Jusczyk et al., 1994; Piot et al., 2024), their minimal exposure to words starting with vowels might result in delayed segmentation (Mattys & Jusczyk, 2001a). Again, Mattys and Jusczyk (2001a) and Nazzi et al. (2005) indicate that their findings are in line with this account.

Taken together, studies that have investigated infants’ ability to segment V-initial words consistently showed that this ability emerges later than the segmentation of C-initial words. In principle, all existing findings are compatible with a phonological account that assumes a universal bias favoring C-initial segmentation. To date, only Seidl and Johnson’s (2008) study provides evidence specifically supporting the acoustic account. In principle, however, all existing findings are also all consistent with an input-based account given that both English and French contain a higher proportion of C- than V-initial words. Hence, more research is needed to further disentangle whether phonological, acoustic, or input-based accounts better explain infants’ segmentation of C- and V-initial words.

To shed further light on the mechanisms underlying infants’ initial bias toward C-initial segmentation and their delayed development of V-initial segmentation, the present study addressed two gaps: We examined the segmentation of V-initial words in a language with a higher proportion of V-initial words than in those previously studied and compared infants’ segmentation of words starting with continuant consonants to their segmentation of V-initial words. To this end, we tested infants acquiring German, which—like English—contains more C- than V-initial words, but crucially a higher proportion of V-initial words than English.

1.1 The Current Study

1.1.1 Research Questions, Hypotheses, and Predictions

In this study, we asked whether the bias toward C-initial segmentation would extend to 11-month-old infants acquiring a language with a higher proportion of V-initial words than English. We hypothesized that under the phonological and the acoustic account, 11-month-old infants should perform better at segmenting C- than V-initial words. In contrast, under the input-based account, their segmentation performance should reflect the relative distribution of C- vs. V-initial words in the input language; accordingly, infants learning a language such as German with a comparatively higher proportion of V-initial words than English or French might be able to segment V-initial words earlier than English-learning infants, for example, by 11 months of age.

The above-stated hypotheses reveal that we cannot distinguish between the phonological and the acoustic account when using plosive-initial words, as these are favored by both accounts. Therefore, we asked a second question of whether the bias toward C-initial segmentation would extend to words beginning with continuant consonants. We hypothesized that under the phonological account, 11-month-old infants should generally find it easier to segment C-initial words than V-initial words. Under the acoustic account, however, if 11-month-old infants have difficulties with segmenting V-initial words, they should face similar difficulties segmenting C-initial words beginning with continuant consonants.

To summarize, if German-learning infants show better performance when segmenting C- rather than V-initial words, both posed questions would be answered with yes, but they would mainly support the phonological account. Whereas if infants do not show the bias for a C-initial segmentation, both answers will be no, but they would mainly support the acoustic and the input-based account.

To approach these hypotheses, we used the head-turn preference procedure (Kemler Nelson et al., 1995) and measured infants’ listening preferences as indexed by their looking times (LTs) for text passages that either contained or did not contain previously familiarized target words (V- or C-initial). In Experiment 1, 11-month-olds were tested. In Experiment 2, 8-month-olds were added to explore developmental change. For experiments using the head-turn preference procedure, it is difficult to predict whether infants will show a familiarity (longer LTs to text passages with familiarized words) or a novelty preference (longer LTs to text passages without familiarized words). Therefore, in line with standards in the field, we consider LTs revealing a significant (α < .05) discrepancy between passages containing versus not containing familiarized target words to indicate successful segmentation.

1.1.2 Experimental Approach: Segmentation of Nonce Text Passages

Before turning to the experiments, it will be necessary for the reader to understand that we took a slightly different approach than prior studies when constructing the speech stimulus materials: We used text passages made up entirely of nonce words. This was done to prevent infants from relying on potential segmentation cues unrelated to our target manipulation, namely cues from glottal stops and from high-frequency function words, both of which typically precede V-initial nouns in German.

Phonetically, pre-vocalic glottal stops function as consonantal onset, which causes a problem when aiming to test for a C-initial segmentation bias. In German, syllable-initial vowels are often preceded by a glottal stop (Kohler Klaus, 1994; Krech, 1968). Zygis and Pompino-Marschall (2010) report the highest rate of V-initial glottal stops (in 75%) if word-initial syllables are stressed (as intended in our target words), and if speech rate is slow (as typical for speech stimuli for infant experiments). In adult processing, German listeners assign glottal stops perceptual weight comparable to that of phonemic consonants such as /h/ (Mitterer & Reinisch, 2015). Moreover, adults have been shown to use glottal stops as cues to word boundaries (e.g., Altenberg, 2005). Therefore, to obtain a cleaner test of a potential C-initial segmentation bias, we sought to prevent infants from relying on a consonantal glottal cue for segmenting V-initial words.

This choice may seem overly cautious, as currently there is no direct evidence that infants are sensitive to glottal stop cues. While English-learning infants are sensitive to some allophonic boundary cues by 10.5 months of age (Jusczyk et al., 1999a), they do not appear to benefit from glottal cues for segmenting V-initial words at 11 months (Kim & Sundara, 2015), and it is unclear whether English-learning infants’ success in segmenting V-initial words at later ages (Mattys & Jusczyk, 2001a; Nazzi et al., 2005) was driven by the presence of glottal stops. However, we reasoned that possibly, even though German-learning infants should also hear (word-initial and word-internal) vowels that are not preceded by this phone, they might be more sensitive to glottal stops as boundary cues than English-learning infants in light of the higher consistency of glottal stop realization in their ambient language.

The other cue we wanted to prevent infants from using when segmenting V-initial words was preceding high-frequency function words. Previous studies have shown that infants can use high-frequency determiners to segment adjacent nouns by 8 months of age (Shi et al., 2006; Shi & Lepage, 2008). Moreover, as discussed above, high-frequency function words facilitate infants’ segmentation of subsequent V-initial words (Kim & Sundara, 2015).

We solved both issues by using text passages composed entirely of nonce words. We familiarized infants with isolated nonce words (V-initial: otel, uger; C-initial: notel, luger, counterbalanced across infants), and subsequently tested their recognition of these when embedded in nonce text passages. In the passages, target words were always preceded by nonce particles, pronounced with an ambisyllabic intervocalic consonant and without any glottal stop (panotel, tiluger). This allowed us to use identical passages for both the V- and C-initial test conditions while eliminating potential confounds from function words and glottal stops. The nonce particle-target combinations were embedded in German-like nonce text passages that preserved German phonotactics, prosody, and some derivational morphology, yielding stimuli that were lexically controlled yet ecologically valid.

We are not aware of any prior infant speech segmentation studies that used nonce text passages. However, segmentation research commonly employs even more reduced stimuli to isolate specific cues while controlling for confounds. For example, artificial language learning paradigms (following pioneering work by Saffran et al., 1996) expose participants to continuous syllable streams designed to isolate particular segmentation cues. Importantly, such studies still yield cross-linguistic differences, with segmentation performance reflecting language-specific cue use (e.g., Vroomen et al., 1998), suggesting that highly controlled stimuli can provide insights into natural language processing and language development.

Our use of more naturalistic nonce passages rather than fully artificial languages reflects a compromise between experimental control and ecological validity. In our experience, naturalistic stimuli are typically easier to use with infants, and prior work shows that infants remain sensitive to native segmentation cues even when listening to unfamiliar languages (Houston et al., 2000). The nonce passages therefore allowed us to control potential confounds while maintaining speech that is phonetically, phonotactically, and prosodically natural and engaging for infants.

Next, before turning to the experiments probing our hypotheses, we present lexical statistics comparing the likelihood of words to start and end with consonants or vowels in German and English to get a better impression of how input characteristics might shape infants’ biases toward a preference for C-initial segmentation routines in these languages.

1.1.3 Lexical Statistics: Consonant- and Vowel-Initial Words in German and English

For ADS, we used the lemma databases from the WebCelex lexical database of English and German (Baayen et al., 1995, retrieved from https://webcelex.ivdnt.org/) to calculate the number of C- and V-initial lemmas in the two languages (using the PhonCV column). For English, we counted the “S” labels (indicating syllabic consonants) in the PhonCV column as “C” (consonant).

For child-directed speech (CDS), we used Stärk et al.’s (2022) phonetic annotation of 20 German datasets in the CHILDES database (MacWhinney, 2000). This database includes seven and a half hours of input data from 19 adults to 10 children aged 0;6 to 1;8. In addition, we used the English data analyzed in Saksida et al.’s (2017) study, based on Swingley’s (2005) phonetic annotation of the Korman (1984) corpus. This corpus includes input data to six children aged 0;1 to 0;4.

Frequency counts (see Table 1) show that both German and English have more C- than V-initial words. However, two-sample z-tests conducted in R (R Core Team, 2022) showed that German has a significantly higher proportion of V-initial words than English (ADS: z = 951.03, 95% CI [0.07, 1.00], p < .01; CDS: z = 6.91, 95% CI [0.01, 1.00], p < .01). For the word-final position, there are also more consonant-final than vowel-final words in both languages. However, in contrast to the initial position, German has a significantly lower proportion of V-final words than English in ADS (z = 567.18, 95% CI [-1.00, -0.05], p < .01), but there was no significant difference in CDS (z = 1.41, 95% CI [-1.00, 0.04], p = .88).

Table 1.

Type Frequency (Raw Counts) of German and English Lemmas in the CELEX Database (ADS: Adult-Directed Speech), and German (Stärk et al., 2022) and English (Swingley, 2005) Lexemes in Child-Directed Speech (CDS) Starting and Ending With Consonants and Vowels and Their Relative Likelihood of Occurrence (Expressed in Percentage, in Brackets).

	ADS		CDS
	German	English	German	English
Total number of types	51,728	52,447	1,660	1,748
C-initial	39,062 (75.5%)	43,658 (83.2%)	1,422 (85.7%)	1,550 (88.7%)
V-initial	12,666 (24.5%)	8,789 (16.8%)	238 (14.3%)	198 (11.3%)
C-final	44,225 (85.5%)	41,911 (79.9%)	1,310 (78.9%)	1,408 (80.5%)
V-final	7,503 (14.5%)	10,536 (20.1%)	350 (21.1%)	340 (19.5%)

These lexical statistics suggest cross-linguistic differences in infants’ input languages. Specifically, these results indicate that in terms of the input-based account, V-initial segmentation might be expected more in German than in English given the higher proportion of V-initial words in German than in English. Moreover, compared to English, vowels in German occur less often in the final position, which might even lead German-learning infants to favor a V-initial segmentation at CV transitions in speech. To shed more light on these issues, we conducted the following experiment.

2 Experiment 1

2.1 Methods

2.1.1 Participants

A total of 56 eleven-month-old infants (mean age = 11.03, range = [10;15–11;26], 28 females, 28 males) from monolingual German-speaking families were included in the analyses. Three additional infants were tested, but their data were not included in the final sample because of fussiness/sibling interference. All infants were in good health, had been born full-term (36–41 weeks of gestation), and had no known hearing or vision impairments. They were all raised in a monolingual German-speaking environment. Approximately half of the infants were familiarized in the C-initial condition (N = 27; 15 females, 12 males), whereas the other half was familiarized in the V-initial condition (N = 29; 13 females, 16 males). Families were contacted via the Potsdam BabyLAB’s participant database. They received a small gift (i.e., a colorful diploma with the picture of their infant) and a small monetary compensation for their participation.

2.1.2 Materials

Auditory Stimuli. We created isolated word lists for two different familiarization conditions, namely V- and C-initial conditions and a set of text passages for a test phase that contained these words. All stimuli (isolated words and text passages) were German-sounding. They were made up of nonce words to reduce the impact of word familiarity. All stimuli were composed of phonemes that belonged to the German inventory and adhered to the phonotactics and prosody of German. They were all spoken by the same female native speaker of German.

Nonce Target Words. Four bisyllabic nonce target words (henceforth: target words) that infants were familiarized with and had to be segmented from spoken text passages were created: C-initial notel /noːtl̩/ and luger /luːgɐ/; V-initial otel /oːtl̩/ and uger /uːgɐ/. The first syllable was a stressed open syllable with a long back vowel (/uː/, /oː/). In the C-initial condition, sonorant alveolar consonants (/n/, /l/) were selected as word onsets, as these phonemes (other than obstruents, due to the final-obstruent devoicing rule in German) are both licit word-beginnings and word-endings in German, hence allowing for them to be segmented as the initial segment of the target word but also as the final segment of a preceding word. The second syllable was unstressed and contained two phonemes: an initial plosive (/t/, /ɡ/), followed by a syllabic consonant (/l̩/) or a reduced vowel (/ɐ/). This stress pattern was selected, as it is the predominant one in German, which German-learning infants prefer (e.g., Höhle et al., 2009) and can successfully segment from continuous speech by the age of 9 months (Bartels et al., 2009; Hamburger et al., 2025; Vanoncini et al., 2022 the segmentation of unfamiliar monosyllabic words has not yet been demonstrated though they segment high-frequency monosyllabic determiners; Höhle & Weissenborn, 2003). For the recordings, the speaker produced the target words following a small carrier phrase and a long pause (e.g., “Schau!___Otel!” [“Look!___Otel”]; “Wirklich?___Otel?” [“Really?___Otel?”]; “Schon wieder.___Otel.” [“Again.___Otel.”]) to yield tokens with variable prosody (i.e., tokens with falling, rising, or level intonation). The long pauses were included to prevent that target word onsets would be affected by co-articulatory cues from the preceding carrier phrase.

Familiarization trials were constructed on the basis of these tokens: four lists of 16 tokens each were created for notel, luger, otel, and uger. The order of the tokens was arranged in a way that the different intonation patterns alternated, making the auditory stimulus more engaging for the infant. Within a trial, the onset-to-onset interval between tokens was 1.4 s, resulting in a total duration of 22.4 s (token duration varied between 714 ms and 979 ms; inter-stimulus intervals between 421 ms and 702 ms).

Nonce Particles. To create a context that would be ambiguous about whether the target word started with a consonant or a vowel in the continuous speech stream, we embedded it in a nonce carrier so that every time the target word occurred in the text passage, it would form part of a trisyllabic sequence: tiluger or panotel. The carriers, composed of an unstressed syllable preceding the target words, mimicked the function of a particle (for a similar approach, see Omane et al., 2025; van Kampen et al., 2008) and could be either interpreted as pa or pan, and as ti or til, depending on the preferred segmentation (C-initial segmentation: e.g., pa_notel or pan_notel; V-initial segmentation: e.g., pan_otel). Both nonce particles began with a voiceless plosive (C1 = /p/, /t/), as these are phonemes that infants easily perceive. This was followed by a long vowel (V1 = /iː/, /aː/), as a coda is possible but not required after these vowels, which contributes to the possibility that the following intervocalic consonant could be segmented as the coda of the nonce particle, or as the onset of the target word.

Nonce Text Passages. Two German-sounding nonce text passages that were entirely composed of nonce words were created (see Appendix Table A1). The passages consisted of six declarative nonce sentences each. Following Mattys and Jusczyk (2001a), the target words were embedded in various sentence positions (2x sentence initial, 2x final, 2x medial). Each of them contained one of the familiarized targets. To generate the panotel and tiluger nonce text passages, we first wrote two short text passages in real German, each consisting of six simple main clauses, with one specific bisyllabic real word and determiner (i.e., die Sonne, die Mate) occurring in positions where our nonce particle + target word sequence was supposed to be placed. The average number of words per sentence was 5.8 for the panotel text and 6.2 for the tiluger text. Next, these two texts were transformed by constructing nonce words that closely resembled the German ones in structure and sound so that nonce texts matching the original in syllable, word, and sentence length were rendered. Similar as in Lewis Carroll’s Jabberwocky, we maintained some derivational and inflectional morphemes (e.g., -t marking third-person singular on verbs; -er marking masculine, nominative case, singular on adjectives) so that the nonce texts would sound more natural and German-like, providing the impression of hearing fluctuations of adjectives, nouns, and verbs (English-learning infants are sensitive to inflectional morphemes in speech; Kim & Sundara, 2021). However, no real function words were included, as infants can use them as a cue for speech segmentation (Kim & Sundara, 2015; Shi et al., 2006). Hence, the position before the target word that would naturally be filled by a determiner (der, die, das) was filled with the nonce particle described earlier.

When recording the nonce text passages, the speaker used the real German sentences as a model: She always first read a real sentence, and immediately afterwards, a nonce sentence. This made it easier to produce the nonce sentences with natural prosody regarding features such as tempo, rhythm, pitch contours, and pauses. Six versions of both the panotel and tiluger text passages were created as test stimuli by concatenating the recorded sentences. They all had the same total duration (23.96 s) but varied in the order of the sentences such that during the test phase, each sentence appeared once in every possible position within the passage. To control for the total duration, we varied the length of the silent pauses between the sentences between 505 ms and 808 ms.

2.1.3 Procedure

Each child was assigned to either the C-initial or the V-initial experimental condition. We used a head-turn preference paradigm, with a familiarization phase followed by a test phase. During testing, caregivers were seated in a sound-attenuated testing booth with their infants on their laps, facing forward. Loudspeakers were mounted on the walls of the two side panels at about the level of the infants’ heads. There were three lights mounted on the walls: a small green light directly in front of the infants, and two small red lights on either side of the infants, close to the two speakers. A video camera was also connected from below the central light (i.e., in front of the infants) to a monitor in an adjacent control room where the experimenter was located.

The experiment took place as follows. The caregiver was instructed not to interact with the infant during testing and was then seated in the testing booth with the infant on the lap. During the experiment, the caregiver wore headphones playing experimental stimuli overlaid over music to mask the test stimuli. The experimenter, who was outside the testing booth in a control room, recorded the infant’s looking behavior via button presses. Each trial began by drawing the infant’s attention to the center by flashing the central light. Once achieved, the central light was turned off, and one of the two side lights started flashing. Once the infant turned and looked at it, the stimulus was played, along with the continuous flashing of the side light. The trial ended when the entire trial had been played or when the infant turned away for more than 2 s. Infants’ attention to the stimuli was measured based on their summed LT toward the target side light on a given trial (excluding the times when the infant was not looking at the light during the trials).

The experimental procedure is illustrated in Figure 1. Each experimental session began with a familiarization phase in which infants heard different tokens of one of the targets (i.e., luger/uger/notel/otel) until they accumulated 30 s of LT. Note that after 30 s of looking, the familiarization phase did not end automatically. Rather, the implemented procedure required the experimenter to pay attention to the LT counter and terminate the trial manually upon reaching the required time. Since the primary focus of the experimenter was the coding of the looking behavior of the child, the trial termination could take longer, sometimes until the current sound file was fully played (~47 s in total, which happened in four infants). In the familiarization phase, we counterbalanced across participants whether they were presented with a V- (otel, uger) or a C-initial target word (notel, luger). The test phase began immediately after the familiarization criterion was attained. It consisted of 12 test trials, in each of which the six-sentence passages were presented: six trials containing the familiar word, and the other six trials containing the unfamiliar word (thus, each infant could hear panotel and tiluger up to 36 times per test session). Condition and target word during familiarization were counterbalanced across participants; thus, whether the passage containing panotel or tiluger was novel for the infant during the test phase was also counterbalanced. Both the sequence of sentences within passages and the presentation of the panotel and tiluger texts were randomized, with the constraint that no passage was played more than twice in a row on the same side of the head-turn experiment setup. To be included in the analyses, infants had to complete at least two familiar test trials and two unfamiliar test trials.

Figure 1.

Graphic illustration of the experimental procedure.

2.1.4 Data Pre-Processing and Analysis

Before running the analysis, we excluded all test trials during which the infant’s LT ended before the onset of the target word within a passage (N = 57 test trials out of 663, 8.6%). All analyses were conducted in R-studio (R core team, 2022, version 4.2.2). We used linear mixed-effect models using the function lmer of the R package lme4 (Bates et al., 2015), the package lmerTest (Kuznetsova et al., 2017) to obtain p-values, and the package emmeans (Lenth & Piaskowski, 2026) for post hoc tests. We fitted a linear mixed-effect model, with infants’ log-transformed LTs (in seconds) as the dependent variable. As fixed factors, we included trial type (Familiar vs. Unfamiliar) and trial number (1–12, 1 being the baseline) as within-participant factors, and condition (C-initial vs. V-initial) as a between-participant factor. All binary factors were sum-contrasted for the model (i.e., effect coding: 1 vs. -1). Individual participant intercepts and by-participant random slopes for trial number were included in the random effects structure. By-participant random slopes for trial type were not added as the model issued singularity warnings when doing so. The full equation was as follows:

$$\mathrm{Log}\left(\mathrm{LT}\right)~\mathrm{trial}\mathrm{type}\times \mathrm{condition}+\mathrm{trial}\mathrm{type}\times \mathrm{trial}\mathrm{number}+\left(1+\mathrm{trial}\mathrm{number}|\mathrm{participant}\right).$$

2.2 Results

An illustration of the raw data can be viewed in Figure 2. Results of the statistical model are presented in Table 2. There was a main effect of trial type (β = 0.118, SE = 0.043, p = .006), with longer LTs for familiar trials (M = 7.928s, 95% CI [7.323s, 8.534s]) than for unfamiliar trials (M = 7.263s, 95% CI [6.648s, 7.877s], see Figure 2). There was also a main effect of trial number (β = -0.061, SE = 0.008, p < .001), indicating that participants’ LTs decreased over the duration of the experiment. No further significant main effects or interactions were found.

Figure 2.

Infants’ looking times in Experiment 1 and 2.

Table 2.

Results of the Linear Mixed Effects Model.

	log(LT)
Predictors	Estimates	CI	p
(Intercept)	2.10	1.99 to 2.22	<.001
trial type	0.12	0.03 to 0.20	.006
condition	0.07	−0.02 to 0.17	.132
trial number	−0.06	−0.08 to -0.05	<.001
trial type × condition	−0.02	−0.06 to 0.03	.482
trial type × trial number	−0.01	−0.02 to 0.00	.090
Random effects
σ2	0.33
τ00 id	0.09
τ11 id.trialnumber	0.00
ρ01 id	−0.16
ICC	0.26
N id	56
Observations	606
Marginal R2 / Conditional R2	0.111 / 0.340

Values highlighted in bold indicate statistically significant effects.

2.3 Discussion

In Experiment 1, we observed two significant effects: a main effect of trial number and, critically, a main effect of trial type. The effect of trial type indicates that 11-month-old German-learning infants successfully segmented target words from continuous speech irrespective of their initial segment (C- or V-initial). Their LTs were longer when they listened to passages that contained familiar words as compared to unfamiliar words, which suggests they had a familiarity preference, as observed in most infant speech segmentation experiments using similar methods (Bergmann & Cristia, 2016). The effect of trial number reflects the expected decline in infants’ attention over the course of the experimental session. This effect did not interact with trial type and therefore does not alter the interpretation of the segmentation findings.

The results suggest that structural differences between German and previously tested languages (English and French) affect infants’ readiness to segment V-initial words. Unlike the infants tested here, English-learning infants at the same age required additional cues to word boundaries to succeed in segmenting V-initial words in previous studies. Thus, with respect to our first research question of whether the bias toward C-initial segmentation extends to infants acquiring a language with a higher proportion of V-initial words, the answer appears to be no. This finding supports the input-based account, which predicts that infants exposed to a language richer in V-initial words, such as German, can segment such words earlier than English-learning infants.

Our second question of whether a potential C-initial segmentation bias would extend to continuant consonants cannot be evaluated on the basis of the present findings. Under the phonological account, 11-month-olds were expected to show a bias toward C-initial segmentation, whereas under the acoustic account, difficulties segmenting both continuant C- and V-initial words were expected due to the absence of clear temporal boundary cues. In the present experiment, however, infants segmented both word types, and there is no evidence that would suggest that one of the conditions was more difficult for them. Hence, the present finding does not provide any evidence for (or against) the phonological or the acoustic account.

This result does not rule out the possibility that a C-initial segmentation bias is present earlier in German-learning infants’ development. Under a phonological account, infants might initially rely on a general preference for C-initial segmentation but adjust this strategy as they accumulate language-specific distributional knowledge. Possibly, German-learning infants no longer rely on this bias by 11 months because it weakens more rapidly in German than in languages with fewer V-initial words, such as English or French.

This possibility raises the question of whether a C-initial segmentation bias is observable at an earlier age in German-learning infants, and whether this bias would extend to continuant consonants. Prior studies have shown that German-learning infants segment plosive-initial words from text passages as of the ages of 8 (Altvater-Mackensen & Mani, 2013; Hamburger et al., 2025) and 9 months (Bartels et al., 2009; Zahner et al., 2016; in other studies, segmentation effects only appeared when variables accounting for individual variability were added as, e.g., in Vanoncini et al., 2022). So far, only Zahner et al. (2016) included test words starting with continuant consonants (gune, sino, none, rade). In Experiment 2, we therefore tested 8-month-old infants to explore whether younger German-learning infants would show a bias toward continuant C-initial over V-initial segmentation when tested with the same task and stimuli

3 Experiment 2

3.1 Methods

3.1.1 Participants

A total of 58 eight-month-old infants (mean age = 7.94, range = [7;15–8;14], 30 females, 28 males) from monolingual German-speaking families were included in the analyses. Six additional infants were tested, but their data were not included in the final sample because of fussiness/sibling interference (N = 2) or technical/experimenter error (N = 4). As in Experiment 1, all infants were monolingually raised, in good health, had been born full-term, and had no known hearing or vision impairments. Approximately half of the infants were familiarized in the C-initial condition (N = 26; 13 females, 13 males), whereas the other half was familiarized in the V-initial condition (N = 32; 17 females, 15 males). Recruitment and compensation procedures were the same as in Experiment 1.

3.1.2 Materials and Procedure

Identical to Experiment 1.

3.1.3 Data Pre-Processing and Analysis

Before running the analysis, we excluded all test trials during which the infant’s LT ended before the onset of the target word within a passage (N = 33 test trials out of 689, 4.8%). For the analysis, we conducted two different linear mixed-effects models. The first model included data of the 8-month-olds group and had the same structure as in Experiment 1, with the notable exception that by-participant random slopes for trial number were not included in the model because the model failed to converge when doing so. The second model included data from both the 8-month-olds and 11-month-olds groups and explored whether the two age groups statistically differed from each other. This model included the between-participants factor age (11 months vs. 8 months, sum-contrasted), in interaction with trial type and condition, as well as trial number in interaction with trial type. The full equation was:

$$\mathrm{Log}\left(\mathrm{LT}\right)~\mathrm{age}\times \mathrm{trial}\mathrm{type}\times \mathrm{condition}+\mathrm{trial}\mathrm{number}\times \mathrm{trial}\mathrm{type}+\left(1+\mathrm{trial}\mathrm{number}|\mathrm{participant}\right).$$

3.2 Results

The results of the statistical model of the 8-month-olds are presented in Table 3. There was a main effect of condition (β = 0.101, SE = 0.040, p = .014), with overall longer LTs during test trials for participants who were familiarized with the C-initial condition (M = 9.152s, 95% CI [8.508s, 9.795s]) compared to participants who were familiarized with the V-initial condition (M = 7.711s, 95% CI [7.186s, 8.236s], see Figure 3). There was also a main effect of trial number (β = -0.064, SE = 0.006, p < .001), indicating that participants’ LTs decreased over the duration of the experiment. However, no significant main effect of trial type and no significant interactions were found.

Table 3.

Results of the Linear Mixed Effects Model, Including Data From the 8-Month-Olds Group.

	log(LT)
Predictors	Estimates	CI	p
(Intercept)	2.27	2.13 to 2.37	<.001
trial type	−0.04	−0.12 to 0.03	.264
condition	0.10	0.02 to 0.18	.014
trial number	−0.06	−0.08 to -0.05	<.001
trial type × condition	−0.01	−0.05 to 0.03	.604
trial type × trial number	0.01	−0.00 to 0.02	.222
Random effects
σ2	0.30
τ00 id	0.06
ICC	0.17
N id	58
Observations	656
Marginal R2 / Conditional R2	0.139 / 0.290

Values highlighted in bold indicate statistically significant effects.

Figure 3.

LTs (Means and CIs) by 8-month-old infants during the test phase, broken down by familiarization condition. Dots indicate individual infants.

Next, we explored how the 8-month-old infants tested in Experiment 2 fared in comparison to the 11-month-olds tested in Experiment 1, pooling their data. The results of the statistical model are presented in the Appendix (Table A2). There was a main effect of condition (β = 0.087, SE = 0.031, p = .006), with overall longer LTs during test trials for participants who were familiarized with the C-initial condition (M = 8.733s, 95% CI [8.265s, 9.201s]) compared to those for participants who were familiarized with the V-initial condition (M = 7.347s, 95% CI [6.973s, 7.721s]) across trial type and age. There was also a main effect of age (β = -0.74, SE = 0.031, p = .020), with overall longer LTs in the 8-month-olds group (M = 8.368s, 95% CI [7.955s, 8.780s]) than in the 11-month-olds group (M = 7.601s, 95% CI [7.170s, 8.032s]). There was a significant main effect of trial number (β = -0.063, SE = 0.005, p < .001), indicating that participants’ LTs decreased over the duration of the experiment. No further significant main effects or interactions were found. However, the interaction between trial type and age approached significance (β = -0.030, SE = 0.016, p = .054).

3.3 Discussion

In Experiment 2, we observed a main effect of trial number and a main effect of condition, but no main effect of trial type, and no interactions. In light of the lack of a trial-type effect, the present findings do not provide any evidence that 8-month-old German-learning infants can segment words from continuous speech, regardless of whether the words are C- or V-initial. This result is in contrast with results from prior studies that documented infants’ segmentation of C-initial words at similar or younger age (e.g., German: Altvater-Mackensen & Mani,2013; English: Juszcyk et al., 1999; Jusczyk & Aslin, 1995; Spanish and Catalan: Bosch et al., 2013). We will further discuss this discrepancy between the present and prior findings in the General Discussion. The effect of trial number reflects the expected decline in infants’ attention over the course of the experimental session. This effect did not interact with trial type and therefore does not alter the interpretation of the segmentation findings. The effect of condition, although not central to our predictions, might relate to 8-month-olds’ differential processing of C- versus V-initial words. We therefore return to this effect in the General Discussion, where we will also discuss the results of the exploratory analysis.

4 General Discussion

The goal of the present study was to examine the early emergence of infants’ bias toward segmenting C- over V-initial words and to test the extent to which this bias is shaped by phonological principles, acoustics, or by distributional properties of the input language. In German, V-initial words are less common than C-initial words; however, German-learning infants experience a significantly higher proportion of V-initial words in their input language than infants learning English or French. Across two experiments, we asked whether German-learning infants, who would have similar developmental delays in segmenting V- as compared to C-initial words as reported in prior studies, would have the same difficulties with words beginning with continuant consonants. Our results were that 11-month-old infants were able to segment familiarized words from text passages regardless of the status of the word-initial phoneme (as indicated by longer LTs to the familiar vs. the unfamiliar word). However, we found no evidence for (or against) a successful segmentation of neither C- nor V-initial words in 8-month-olds. Instead, we found a main effect of condition at 8 months, which indicated that infants preferred listening to text passages (indicated by longer LTs) when they had previously been familiarized with C- rather than V-initial words. Below, we will discuss potential explanations relating to the phonological, acoustic, and input-based accounts for why German-learning infants at 11 months succeeded in segmenting both continuant- and V-initial words, while at 8 months, we found no evidence for (or against) their success at segmenting either. We also discuss why 8-month-old infants showed longer listening behavior when listening to text passages after having been exposed to C- rather than V-initial words. Afterwards, we will discuss how methodological factors may have impacted the results. Finally, before presenting our conclusion, we will discuss some limitations of our study.

4.1 Phonological, Acoustic, or Input-Based Accounts?

The present results of German-learning infants segmenting both V- and continuant C-initial words at 11 months are in contrast with results previously reported for English- and French-learning infants, who only develop the ability to segment V-initial words at a later age. Interestingly, the results match those of findings of early German-learning children’s productions, as these do not seem to show a bias toward producing C-initial words either (Grijzenhout & Joppen-Hellwig, 2006). Which of the three accounts—phonological, acoustic, or input-based—best explains these results?

The findings challenge phonological accounts that predict a general advantage for C-initial segmentation. Phonological accounts (supported by Babineau & Shi, 2014, 2016) assumes a universal bias against onsetless syllables (e.g., Jakobson, 1962). However, this bias may be weaker or more language-specific than previously assumed given that many languages—such as several Australian languages—frequently begin words with vowels (Breen & Pensalfini, 1999), which again may have evolved due to acoustic factors that have shaped these patterns (e.g., Blevins, 2001; Tabain et al., 2004). At the very least, we were not able to measure effects of this universal bias in the present study because neither the 11- nor the 8-month-olds reacted differently to C- as compared to V-initial words.

The findings also challenge acoustic accounts (supported by Seidl & Johnson, 2008) that attribute infants’ delay in segmenting V-initial words to the weaker salience of vowel than consonant onsets. On the one hand, the absence of differential performance by German-learning infants for V-initial versus continuant C-initial words does not support an acoustic account positing a general C-initial segmentation bias that extends to continuant onsets. On the other hand, it may seem plausible that neither age group in our study showed the asymmetry between C- and V-initial words previously reported for English-learning infants because both types of onsets have weak acoustic salience in comparable ways, as predicted by some acoustic accounts (Nazzi et al., 2005). However, if reduced acoustic salience of continuant consonants underlies the present findings, its impact may be relatively modest, considering that by 9 months, German-learning infants have been shown to segment words beginning with continuants (Zahner et al., 2016; although, in that study, segmentation was additionally cued by pitch accents). In any case, if acoustic factors alone were responsible for delayed segmentation of V-initial words, it remains unclear why German- but not English-learning infants succeed segmenting words with acoustically less-salient onsets at 11 months.

Thus, the findings most strongly support an input-based account, suggesting that by 11 months, the processing procedure of assuming word-initial boundaries at consonants is language-specific and tailored to the distributional characteristics of the input language. Lexical statistics of both ADS and CDS (see Table 1) showed that while German has more C- than V-initial words, the proportion of V-initial words is higher in German than in English. Moreover, while German has more C- than V-final words, the proportion of V-final words is lower in German than in English (at least in ADS). Together, in line with the input-based account, the cross-linguistic differences in infants’ input languages make it more likely for German infants to show an ability for V-initial segmentation.

Note that the discrepancies between German and English were significant in both ADS and CDS. Numerically, however, the differences were larger in ADS, according to which German contains approximately 8% more V-initial and 6% fewer V-final words than English, whereas in CDS, the corresponding differences amount to only 3% more V-initial and 2% fewer V-final words. We consider the estimates derived from the CELEX database to be more reliable than those based on the CDS corpora, as the latter were much smaller than the former and may therefore not be fully representative of the linguistic input available to infants. In addition, it remains unclear to what extent infants develop their speech-processing routines on the basis of CDS as opposed to overheard ADS. While research has shown that properties of infant-directed speech, specifically the clarity of mothers’ phoneme articulation, are associated with infants’ phoneme-discrimination ability (Cristia, 2011; Liu et al., 2003), the relationship of properties of their ADS and their infants’ phoneme perception has as yet not been investigated. Taken together, these considerations suggest that large lexical databases such as CELEX may provide a more stable estimate of the distributional properties of the input relevant for early speech segmentation.

Based on the result that German-learning infants did not show any indication of favoring a C-initial segmentation routine, we wondered whether the input-based account would make any further predictions consistent with our results if more fine-grained input characteristics, namely the positional probabilities of the specific phonemes in word-initial versus word-final position, were considered. Table A3 in the Appendix displays lexical statistics for German words beginning and ending with the phonemes that occurred at the critical boundary in our stimuli for both ADS and CDS. These values indicate that, based on word-initial positional frequencies, a C-initial segmentation of panotel as pa_notel and tiluger as ti_luger should have been favored because /n/ and /l/ occur more frequently as word onsets than /o/ and /u/. However, based on word-final positional frequencies, a V-initial segmentation (pan_otel, til_uger) should have been favored, as /n/ and /l/ are also more common word-finally than /a/ and /i/. Thus, although German shows a slight overall skew toward C-initial words, there is no clear tendency for either a C- or V-initial segmentation bias on the basis of detailed phonotactic probabilities. Experimental evidence suggests that adult listeners can use both abstract and detailed phonotactic cues for speech segmentation (Boll-Avetisyan & Kager, 2016). Hence, it is possible that infants integrate cues from broad distinctions between consonants, as well as from positional probabilities of specific phonemes during speech segmentation. Altogether, these results most strongly support input-based accounts.

One unpredicted finding, however, may be taken as some support for a phonological or acoustic account of C-initial words being favored: Eight-month-olds familiarized with C-initial isolated words showed longer LTs when listening to the text passages than infants familiarized with V-initial words, an effect that also emerged when data from both age groups were pooled. Since the effect occurred while infants listened to the text passages that were identical across conditions, it seems, at a first glance, that explanations related to stimulus properties or information density are unlikely. Considering that condition was a between-participants factor, and the effect was unpredicted, the finding might be spurious. However, one speculative interpretation is that C-initial words are easier for infants to encode and maintain in short-term memory, reducing cognitive load, which would then allow them to sustain attention for longer during the test phase. Thus, although this attentional difference did not translate into differences in segmentation performance, it may reflect subtle processing advantages for C-initial over V-initial words.

4.2 Methodological Considerations

In designing the stimuli, we intentionally constructed text passages composed entirely of nonce words. This ensured that embedded target words were preceded neither by glottal stops, which could provide allophonic consonantal cues to word onsets, nor by function words that might facilitate segmentation. Although these choices allowed us to test our hypotheses more directly, they may also have increased the difficulty of the experimental task. In follow, we consider how these methodological choices might have influenced the observed results.

In Experiment 1, 11-month-old German-learning infants were able to recognize both familiarized C- and V-initial words in the passages, which suggests the general suitability of this method. However, the lack of evidence for or against 8-month-olds’ segmentation of either C- or V-initial words was unexpected given that prior studies using similar familiarization–test designs have reported successful segmentation of C-initial words at comparable ages in German (Altvater-Mackensen & Mani, 2013; Hamburger et al., 2025; Zahner et al., 2016, the latter including continuant-initial words) and even younger ages in English (e.g., Jusczyk & Aslin, 1995; Jusczyk et al., 1999b), Spanish, and Catalan (Bosch et al., 2013). It is possible that infants in these earlier studies benefited from real-word passages containing high-frequency function words. Because infants use such lexical cues for segmentation early in development (Shi et al., 2006), we deliberately eliminated these cues by using nonce passages. However, this may have made the task more demanding. This limits the direct comparability with previous findings, but it also highlights the need for more research using nonce materials to determine at what age infants rely on acoustic or phonological cues in the absence of lexical information.

Another factor that may have increased the difficulty of the task is that target words were not preceded by a glottal stop but rather by an ambisyllabic consonant. Prevocalic glottalization is more systematic in German than in English and may therefore play a larger role in German-learning infants’ segmentation in natural input. The lack of evidence for English-learning infants benefiting from glottal cues (Kim & Sundara, 2015; Mattys & Jusczyk, 2001a) might not necessarily transfer to German-learning infants. Future studies interested in language-specific speech segmentation cues may therefore want to investigate whether German-learning infants use the glottal stop as a cue for segmenting V-initial words.

Finally, there was one inconsistency between the familiarization and test stimuli. During familiarization, the V-initial targets were preceded by a glottal stop (/ʔotel/, /ʔuger/), whereas at test, these words followed an ambisyllabic consonant. From a developmental perspective, it is unclear whether infants encoded the familiarization items as V-initial: If infants perceived the glottal stop (which has, to our knowledge, not yet been tested), it is not obvious why they would treat it as an allophonic rather than a phonemic consonant, particularly given evidence that adult listeners of German process glottal stops in a similar way as the phoneme /h/ (Mitterer & Reinisch, 2015). One possible interpretation would therefore be that infants simply recognized the familiar V-initial strings when they occurred after the ambisyllabic consonant (e.g., /panotel/, /tiluger/), rather than segmenting them from the speech stream. However, extensive evidence from prior segmentation studies found no evidence that infants merely rely on surface pattern recognition in such tasks. For example, English-learning infants did not segment tar from guitar is (Jusczyk et al., 1999b), and they did not detect dice or ice in cleared ice (Mattys & Jusczyk, 2001a). Similarly, Akan- and Turkish-learning infants do not recognize similar-sounding patterns unless they are supported by phonotactic segmentation cues (Omane et al., 2025; van Kampen et al., 2008), and German-learning infants failed to recognize such patterns unless they are marked by pitch accent (Zahner et al., 2016). In light of this evidence, we consider it unlikely that the present results reflect simple pattern recognition and instead assume that infants successfully segmented both C- and V-initial words at 11 months.

4.3 Limitations

While the presented experiments allowed us to address our primary research questions, several additional factors could not be examined within the scope of this study and should be explored in future work. First, we were not able to measure effects of a universal C-initial segmentation bias because neither the 11- nor the 8-month-olds reacted differently to C- as compared to V-initial words. It is, however, possible that phonological factors (generally favoring C-initial segmentation) and input-based factors (weakening the C-initial bias in German) interact in the development of infants’ segmentation routines. It is therefore possible that effects of a C-initial segmentation bias would be measurable in German-learning infants at the interim ages of 9 and 10 months.

Second, and relatedly, it is also possible that acoustic and input-based factors interact. Thus, while it may speak against acoustic accounts that infants in our study succeeded in segmenting words beginning with acoustically weak vowels at a younger age than in English, this may again be because infants receive more evidence for words to begin with these phonemes in German than in English. This raises the question of whether 8-month-olds would have succeeded when stop consonants had been used as initial consonants, following the acoustic account that these provide stronger cues than continuants and vowels. Future work should therefore test whether German-learning 8-month-olds would identify nonce words such as kotel in pakotel, and puger in tipuger.

Last, our familiarization–test structure employed isolated words for familiarization and continuous speech for testing. While this approach had the advantage that we could use the same stimuli for a test phase in both the C- and V-initial conditions, it may have led infants to approach the task using their fresh memory of the familiarized words for lexical matching of the familiarized words with what they perceived in the text passages. A reversed order of familiarizing infants with text passages and testing them on isolated words with both C- and V-initial targets in addition to unfamiliarized novel words in a within-participant design could provide a more direct test of infants’ segmentation routines and clarify whether they show an inherent bias toward C-initial segmentation when no lexical scaffolding is available.

5 Conclusions

In sum, the results show that infants acquiring German, a language with a higher proportion of V-initial words than English, can segment V-initial words at an earlier age than previously attested. Together, the results provide new insight into how language-specific input characteristics shape early segmentation processes during the first year of life. The findings highlight the importance of research on infants learning languages with differing structural properties to better understand how acoustic, phonological, and distributional cues interact in shaping infants’ language development.