RESULTS AND DISCUSSION

Mean orientation times to the AP and PA lists were calculated for each infant. The data for the 10-month-olds (M _AP = 8·32 s, SD = 2·57 s; M _PA = 8·26 s, SD = 3·28 s), and for the 13-month-olds (M _AP = 6·38 s, SD = 2·61 s; M _PA = 9·69 s, SD = 2·25 s), are presented in Figure 1. A 2-way ANOVA with the between-subject factor of age (10 versus 13 months) and the within-subject factor of lexical structure (AP versus PA words) was conducted. The effect of lexical structure was significant (F(1,30) = 8·38, p = ·007, ηp ² = ·22), with infants having longer orientation times to PA than to AP lists. The effect of age was not significant (F(1,30) = 0·11, p = ·75). However, the interaction between age and lexical structure was significant (F(1,30) = 8·94, p = ·005, ηp ² = ·23), indicating that the effect of lexical structure changed with age. Planned comparisons showed that the lexical structure effect was not significant at 10 months (F(1,30) = 0·005, p = ·95), but was significant at 13 months (F(1,30) = 17·32, p < ·001, d = 1·36). A bias for PA stimuli was found in only 8 of the 16 10-month-olds (p = ·60, binomial test), but in 12 of the 16 13-month-olds (p = ·04, binomial test).

Fig. 1.

Mean orientation times (and standard error of the mean) to the AP and PA stimuli in Experiments 1 and 2, to the A-initial versus P-initial items in Experiment 3a, and to the P-final versus A-final items of Experiment 3b; ** p ≤ 0.01; *** p ≤ 0.001.

Experiment 1 first establishes the emergence of a perceptual posterior–anterior bias between the ages of 10 and 13 months. Importantly though, given that all adjacent regularities were fully controlled (that is, the Vow₁Cons and the ConsVow₂ were chosen so that there was no significant difference between the adjacent sequences of the PA and AP lists according to the Lexique 3 database), these results support the interpretation that, by 13 months, infants have learned these non-adjacent vocalic regularities present in the French lexicon. More specifically, they have learned the general predominance of non-adjacent sequences of PA vowels over AP vowels in French words.

Second, these results suggest the existence of a delay in the acquisition of non-adjacent vocalic regularities, compared to similar consonantal regularities which are acquired by 10 months of age (the LC bias; Gonzalez-Gomez & Nazzi, Reference Gonzalez-Gomez and Nazzi2012a). One possibility is that the acquisition of vocalic non-adjacent regularities is delayed in comparison to consonantal non-adjacent regularities. This is predicted by the third hypothesis outlined in the ‘Introduction’, according to which the LC and PA biases are the result of distributional computations at the lexical level, at which there is a processing advantage for consonants over vowels as stated by the CV hypothesis (Nespor et al., Reference Nespor, Pena and Mehler2003). However, an alternative interpretation for the present acquisition delay is that the stimuli in the present experiment were more complex than those on the LC bias used by Gonzalez-Gomez and Nazzi (Reference Gonzalez-Gomez and Nazzi2012a). Indeed, the present study contrasted five anterior vowels to five posterior vowels, while Gonzalez-Gomez and Nazzi only contrasted two labial consonants to two coronal consonants. To investigate this methodological possibility, a second experiment was conducted contrasting only two anterior vowels to two posterior vowels.

RESULTS AND DISCUSSION

Mean orientation times to the AP and PA lists were calculated for each infant. The data for the 10-month-olds (M _AP = 7·89 s, SD = 2·34 s; M _PA = 8·53 s, SD = 2·20 s), and for the 13-month-olds (M _AP = 6·39 s, SD = 2·16 s; M _PA = 9·11 s, SD = 2·00 s), are presented in Figure 1. As in Experiment 1, a 2-way ANOVA with the between-subject factor of age (10 versus 13 months) and the within-subject factor of lexical structure (AP versus PA words) was conducted. The effect of lexical structure was significant (F(1,30) = 13·39, p < ·001, ηp ² = ·31), infants having longer orientation times to PA than to AP lists. The effect of age was not significant (F(1,30) = 0·56, p = ·46). However, the interaction between age and lexical structure was significant (F(1,30) = 5·08, p = ·03, ηp ² = ·15), indicating that the effect of lexical structure changed with age. Planned comparisons showed that the lexical structure effect was not significant at 10 months (F(1,30) = 0·99, p = ·33), but was significant at 13 months (F(1,30) = 17·48, p < ·001, d = 1·31). A bias for PA stimuli was found in only 10 of the 16 10-month-olds (p = ·23, binomial test), but in 12 of the 16 13-month-olds (p = ·04, binomial test).

Experiment 2 replicates the results from Experiment 1, showing that the posterior–anterior bias emerges between the ages of 10 and 13 months, using the same number of contrasted elements as in Gonzalez-Gomez and Nazzi (Reference Gonzalez-Gomez and Nazzi2012a). Therefore, these results show that the delay found for vocalic acquisitions compared to consonantal ones did not result from differences in the number of vowels versus consonants used in the two studies. Importantly, these differences in acquisition timing appear to be related to the existence of functional differences between vowels and consonants, as stipulated by the CV hypothesis (Nespor et al., Reference Nespor, Pena and Mehler2003). Given the advantage for consonants, the comparison of Gonzalez-Gomez and Nazzi, (Reference Gonzalez-Gomez and Nazzi2012a) and the present study suggest that the LC and PA biases are learned at the lexical level.

In Experiments 1 and 2, because there is no frequency database for infant-directed speech, the stimuli were built controlling for adjacent regularities using the Lexique 3 database, with no full guarantee that the frequencies would be similar in both types of input. Moreover, we found a predominance of A-initial words over P-initial words, and a predominance of A-final words over P-final words (cf. Table 2), and at this point it cannot be excluded that infants would be influenced by these frequency/positional effects (e.g. prefer A-final words). In order to differentiate the relative contribution of the non-adjacent relationship between the two vowels and these potentially confounding adjacent and positional factors, two additional control sub-experiments were run at 13 months (following the same logic as Gonzalez-Gomez & Nazzi, Reference Gonzalez-Gomez and Nazzi2012a). For these control experiments, the stimuli of Experiment 1 were re-recorded, removing either the final vowel (leaving P-initial and A-initial Vow₁Cons items) or the initial vowel (leaving A-final and P-final ConsVow₂ items). This manipulation removed the non-adjacent regularity we were investigating, while adjacent regularities and positional properties between the two lists of stimuli remained identical to those in Experiment 1.

If 13-month-old infants were sensitive to non-adjacent vocalic regularities in Experiments 1 and 2, they should have no preference in either of these two additional conditions. By contrast, if they were reacting to differences in adjacent frequencies (present in the infants' input but not in the adult lexicon), then they should prefer P-initial over A-initial Vow₁Cons items in Experiment 3a, or A-final over P-final ConsVow₂ items in Experiment 3b, or both. Lastly, if infants were sensitive to the frequency/positional factors, they might prefer A-initial words in Experiment 3a (a frequency bias that would go against the observed preference for AP words), and A-final words in Experiment 3b (a frequency bias that would support a preference for PA words). Note that these frequency/positional predictions differ from the potential adjacent dependency predictions, and only one of them would support the observed PA bias at 13 months. Accordingly, two new groups of 13-month-olds were tested on either P-Cons versus A-Cons stimuli (Experiment 3a), or Cons-A versus Cons-P stimuli (Experiment 3b). Note that both P-Cons and Cons-A stimuli are in line with the posterior–anterior structure (which we will call PA-based structures), while both A-Cons and Cons-P stimuli are in line with the anterior–posterior structure (which we will call AP-based structures).

RESULTS AND DISCUSSION

Mean orientation times to the A-Cons and P-cons lists of Experiment 3a were calculated for each infant and are presented in Figure 1. Mean orientation times were 7·85 s (SD = 3·29 s) for the A-Cons list and 7·44 s (SD = 3·09) for the P-Cons list. Similarly, mean orientation times to the Cons-A and the Cons-P lists of Experiment 3b were calculated for each infant and are also presented in Figure 1. Mean orientation times were 7·98 s (SD = 2·34 s) for the Cons-A list and 8·17 s (SD = 3·38) for the Cons-P list.

A 2-way ANOVA with the between-subject factor of Experiment (3a versus 3b) and the within-subject factor of lexical structure (AP-based versus PA-based) was conducted. Both main effects were not significant (F(1,30) = 0·37, p = ·55, for lexical structure; F(1,30) = 0·20, p = ·66, for experiment). Additionally, the interaction between experiment and lexical structure was not significant (F(1,30) = 0·05, p = ·83). Planned comparisons showed that the lexical structure effect was not significant in either Experiment 3a (F(1,30) = 0·34, p = ·56; a preference for P-Cons items was found for only 8 of the 16 infants, p = ·60, binomial test) or Experiment 3b (F(1,30) = 0·08, p = ·78; a preference for Cons-A items was found for 10 of the 16 infants, p = ·23, binomial test).

The absence of preference in these control experiments establishes that the 13-month-olds in Experiments 1 and 2 were not responding to adjacent properties (between Vow₁Cons or ConsVow₂) of the stimuli, since these adjacent properties were also present in Experiments 3a and 3b. Furthermore, the lack of preference also rules out the possibility that the PA preference was due to a positional bias, such as an A-word-final bias. This suggests that the frequency controls we had made on the basis of the adult Lexique 3 database were appropriate for infant testing. More importantly, they establish that, in Experiments 1 and 2, infants were responding to non-adjacent properties, namely the predominance of PA words over AP words in the French lexicon.

GENERAL DISCUSSION

The goals of the present study were to investigate whether infants are sensitive to non-adjacent phonotactic regularities between vowels, and if so, whether these acquisitions are acquired at the same age as equivalent consonantal regularities. Accordingly, we investigated when French-learning infants develop a preference for Vow₁ConsVow₂ items with a posterior–anterior structure over items with an anterior–posterior structure, the posterior–anterior structures being comparatively more frequent in French than the anterior–posterior ones (cf. Table 1). The results of Experiment 1 showed that this bias emerges between 10 and 13 months of age, that is, three months later than what has been previously found for a consonant bias of similar magnitude (cf. Gonzalez-Gomez & Nazzi, Reference Gonzalez-Gomez and Nazzi2012a). These results were replicated in Experiment 2, using simpler stimuli (stimuli containing 4 instead of the 10 different vowels used in Experiment 1). Taken together, the results of Experiments 1 and 2 showed that, between 10 and 13 months of age, infants become sensitive to non-adjacent phonological regularities between vowels. This interpretation is reinforced by the way the stimuli were constructed, controlling for adjacent regularities using the adult database Lexique 3.

Nevertheless, as discussed earlier, two other effects could have affected the 13-month-olds' preferences. First, the adjacent frequency control could only be calculated on an adult database, since no suitable database is available for very young infants. If these adjacent frequencies were different for the infant input and were responsible for the observed posterior–anterior preference at 13 months, then presenting 13-month-olds with Vow₁Cons should result in a preference for posterior-initial items, and/or presenting them with ConsVow₂ should result in a preference for anterior-final items. Second, frequency/positional effects might also have contributed to the labial–coronal preference, at least when it comes to final anterior vowels, since anterior vowels are overall more frequent than posterior vowels in French, and this is true both in word-initial and word-final position. If 13-month-olds were reacting to these frequency/positional properties, then presenting them with Vow₁Cons should result in a preference for anterior-initial items (an effect that would go against a posterior–anterior preference), and presenting them with ConsVow₂ should result in a preference for anterior-final items (an effect that would support a posterior–anterior preference). These effects were evaluated in two additional control experiments presenting infants with either the first two phonemes (Vow₁Cons, Experiment 3a) or the last two phonemes (ConsVow₂, Experiment 3b) of the original stimuli used in Experiment 1.The results of Experiments 3a–b rule out these alternative interpretations, since no preferences were obtained at 13 months. These latter results further suggest that the adjacent regularities, controlled using the adult database Lexique 3, were appropriate for infant input. The present study thus provides new evidence showing that infants become sensitive to vocalic non-adjacent regularities between 10 and 13 months of age.

At this point, it is important to discuss the age difference in acquisition between the results of Van Kampen and colleagues (2008) and our results. While the results of Van Kampen et al., showed that Turkish-learning 6-month-olds are already sensitive to vowel harmony (a property present in their native language), it is not until seven months later (13 months) that we found evidence of infants' sensitivity to non-adjacent vocalic regularities. As discussed in the ‘Introduction’, several factors could explain this difference. First, crucial factors were not controlled in Van Kampen et al., such as the adjacent and positional frequencies of the stimuli used. Second, the prevalence of vocalic harmony in Turkish (90% of words according to Ketrez, Reference Ketrez2014) is far higher than the prevalence of the PA bias, which makes vowel harmony much more salient than the PA bias in French. Third, vocalic harmony in Turkish has a morphosyntactic component (Rose & Walker, Reference Rose, Walker, Goldsmith, Riggle and Yu2011), and might be learned at the morphosyntactic level rather than the lexical level, as the present PA bias. A related possibility is that vowel harmony is learned as an identity/repetition rule (applying at the level of a given feature), a type of rule that has been found to be learned in laboratory experiments by both 7-month-olds (Marcus et al., Reference Marcus, Vijayan, Rao and Vishton1999) and newborns (Gervain et al., Reference Gervain, Macagno, Cogoi, Pena and Mehler2008, 2012). Both possibilities would predict an advantage for vowels over consonants according to the CV hypothesis, which proposes that vowels play a more important role on extracting (morpho)syntactic-like generalizations (Nespor et al., Reference Nespor, Pena and Mehler2003), which might explain the time lag in the acquisition of both biases.

Given the pending questions regarding the interpretation of vocalic harmony acquisition (which could be further tested with a language in which vowel harmony has lower prevalence, such as Hungarian, which has 77% of harmonic words; Ketrez, Reference Ketrez2014), it is interesting that our results suggest the existence of a delay in the acquisition of vocalic non-adjacent phonological regularities compared to consonantal non-adjacent ones, which have been shown to be acquired between 7 and 10 months of age (cf. Gonzalez-Gomez & Nazzi Reference Gonzalez-Gomez and Nazzi2012a; Nazzi et al., Reference Nazzi, Bertoncini and Bijeljac-Babic2009a). It is important to emphasize that the prevalence of the vocalic (PA) and the consonantal (LC) biases in the French lexicon is similar (PA bias occurrence ratio = 1·79; LC bias occurrence ratio = 1·68). Thus, the asymmetry found in the acquisition of non-adjacent vocalic and consonantal regularities cannot be explained in terms of prevalence of occurrence. Moreover, as noted earlier, PA/AP structures are overall almost nine times more frequent than LC/CL structures, which should have favored, if anything, the acquisition of the PA bias.

The delay of acquisition of the vocalic PA dependency can be explained by the CV hypothesis (Nespor et al., Reference Nespor, Pena and Mehler2003), according to which vowels and consonants play different roles at different linguistic processing levels. More specifically, it has been proposed that vowels are given more weight in the encoding of information about prosodic/syntactic structure, while consonants are given more weight in the encoding of information about lexical identity. The CV hypothesis predicted two possible outcomes about infants' preference for PA sequences. The first possibility, if the PA and LC biases were learned as structural generalizations (i.e. as a rule about vowel or consonant identity and order), predicted an advantage for vocalic over consonantal sequences, thus earlier acquisition of the PA bias. The second possibility, if the biases were learned at the lexical level, predicted an advantage for consonantal acquisitions, hence later acquisition of the PA bias than the LC bias. Our results are in line with the second possibility, suggesting that these non-adjacent phonological regularities are lexically related acquisitions.

The possibility that these non-adjacent regularities would be learned at the lexical level raises the question of whether the acquisition of the PA and LC biases would have implications for lexical learning. Recent studies have found that these kinds of regularities help infants identify possible word-like units and learn new words. First, Gonzalez-Gomez and Nazzi (Reference Gonzalez-Gomez and Nazzi2013) conducted a word segmentation study exploring infants' ability to segment word forms having either a labial–coronal or a coronal–labial structure. Their results show that although infants were able to segment the words with an LC structure at 10 months, they were not able to segment the CL ones until 13 months. In a subsequent study, Gonzalez-Gomez, Poltrock, and Nazzi (Reference Gonzalez-Gomez, Poltrock and Nazzi2013) used animated cartoons in a word learning task to explore whether or not words having an LC structure are easier to learn than words having a CL structure. Their results show that 14-month-olds were able to learn only LC words, while 16-month-olds could learn both LC and CL words. These results are in line with evidence showing that infants use knowledge about native adjacent phonotactic regularities to learn the words of their native language. This has been shown in studies investigating word form segmentation (Mattys & Jusczyk, Reference Mattys and Jusczyk2001) and word learning (Graf Estes, Edwards & Saffran, Reference Graf Estes, Edwards and Saffran2011). All these studies show that words made up of frequent phonotactic sequences are segmented and learned more easily and/or at an earlier age than words made up of infrequent phonotactic sequences. Based on the findings of the present study, similar word segmentation and word learning advantages would be predicted around 13 months of age for words having a PA bias in French, which will have to be tested in the future. Such ‘advantages’ should also be language-specific, that is, based on the specific properties of the lexicon of a given language, as has been found using sequences of plosive consonants (labial /p/, /b/, and coronal /t/, /d/) by comparing French-learning infants (developing an LC bias) and Japanese-learning infants (developing a CL bias).

Importantly, the set of studies described above shows that infants' knowledge of their native language phonotactic patterns influences both their word segmentation and their word learning abilities, providing evidence that early speech perception and later lexical acquisition are closely related. However, the fact that by 9 months Turkish-learning infants use vowel harmony to segment words, and more particularly use non-harmonic sequences as a cue to find word boundaries (Van Kampen et al., Reference Van Kampen, Parmaksiz, van De Vijver, Höhle, Gavarró and Freitas2008), might suggest that whether these patterns are acquired at a lexical level or not may not be directly related to the fact that they would then be used at a lexical level. Further studies will be needed to explore this issue.

The difference in timing found between the acquisition of vocalic and consonantal non-adjacent regularities is in line with previous studies showing an advantage for consonants over vowels in lexically related tasks in both adults (Bonatti et al., Reference Bonatti, Pena, Nespor and Mehler2005; Cutler et al., Reference Cutler, Sebastián-Gallés, Soler-Vilageliu and van Ooijen2000; New et al., Reference New, Araújo and Nazzi2008) and toddlers/children (Havy & Nazzi, Reference Havy and Nazzi2009; Havy et al., Reference Havy, Bertoncini and Nazzi2011; Nazzi, Reference Nazzi2005; Nazzi, et al., Reference Nazzi, Floccia, Moquet and Butler2009b). However, this is the first time that this kind of asymmetry has been found at such a young age. Taken together, these results support the existence of an early privileged role of consonants in lexical processes in French, as proposed by Nespor et al. (Reference Nespor, Pena and Mehler2003). Future research will be needed to continue evaluating infants' acquisition of different non-adjacent consonantal and vocalic regularities in order to determine, on the one hand, the kinds of regularities that infants are sensitive to, and on the other hand, which regularities are treated as lexical acquisitions and which are treated as rule generalizations. At any rate, the present study establishes that infants learning French become sensitive to non-adjacent vocalic regularities in their native language between 10 and 13 months of age, hence later than their acquisition of equivalent consonantal non-adjacent regularities, as predicted by the CV hypothesis if these regularities are learned at the lexical level.