Prosody and Function Words Cue the Acquisition of Word Meanings in 18-Month-Old Infants

Abstract

Language acquisition presents a formidable task for infants, for whom word learning is a crucial yet challenging step. Syntax (the rules for combining words into sentences) has been robustly shown to be a cue to word meaning. But how can infants access syntactic information when they are still acquiring the meanings of words? We investigated the contribution of two cues that may help infants break into the syntax and give a boost to their lexical acquisition: phrasal prosody (speech melody) and function words, both of which are accessible early in life and correlate with syntactic structure in the world’s languages. We show that 18-month-old infants use prosody and function words to recover sentences’ syntactic structure, which in turn constrains the possible meanings of novel words: Participants (N = 48 in each of two experiments) interpreted a novel word as referring to either an object or an action, given its position within the prosodic-syntactic structure of sentences.

Keywords

phrasal prosody function words language acquisition syntactic acquisition lexical development open data open materials preregistered

Humans acquiring language face the challenging task of learning the meanings of words: They must map the sounds of each word to a possible meaning. Given that in fluent speech, words are not separated from one another by clear acoustic markers (such as a silent pause), and given that for each spoken sentence, the world offers a wide array of possible referential intentions, how do babies manage to achieve this sound-to-meaning mapping? A central problem in language acquisition is to determine what sources of information infants can exploit to go from sound to meaning.

Syntactic structure, which governs the organization of words into sentences, has been proposed to be a universal and reliable source of information that children may exploit to discover the meaning of words (e.g., Gleitman, 1990). For instance, by 2 years of age, children infer that a novel word, such as dax, refers to an action when it occupies a verb position in a sentence, as in He is daxing that, or to an object when it occupies a noun position, as in This is a dax (e.g., Bernal, Lidz, Millotte, & Christophe, 2007; Waxman, Lidz, Braun, & Lavin, 2009). These findings demonstrate that the syntactic structure in which the words occur is an important source of information for children: They exploit the syntactic environment of a word to determine its syntactic category (e.g., a noun or a verb) and use the syntactic category to restrict the kind of meaning the novel word can have (e.g., verbs refer to actions or events).

This ability to exploit and learn from syntactic structures so early, although impressive, seems rather counterintuitive. Given that syntactic structure defines the relationships among words in a sentence and allows listeners to compute the meaning of a sentence from the meaning of the individual words that compose it, one would expect that infants would first need to learn the words and their meanings to then be able to learn how to organize words into sentences. We are thus faced with a chicken-and-egg problem: Children seem to need words to learn syntax and to need syntax to learn words. How can infants avoid this circularity? Here, we experimentally tested whether 18-month-olds can compute the syntactic structure of a sentence by relying on phrasal prosody and function words, two sources of information that are available early during language acquisition and convey information about syntactic structure.

Phrasal prosody is the rhythm and melody of speech: When we speak, words are not pronounced one after the other in a monotone way; rather, they are grouped together into intonational units (i.e., prosodic phrases). For example, the sentence “The little cat is running fast” tends to be spoken in two prosodic units: “the little cat” and “is running fast.” In all of the world’s languages, the boundaries between prosodic units always coincide with syntactic boundaries¹ (e.g., Shattuck-Hufnagel & Turk, 1996). In our example, this boundary appears between the noun phrase “the little cat” and the verb phrase “is running fast.” Infants are sensitive to phrasal prosody from birth (e.g., Mehler et al., 1988), and they perceive prosodic cues marking the boundaries between groups of words a few months afterward (Männel & Friederici, 2009; Shukla, White, & Aslin, 2011; Soderstrom, Seidl, Kemler Nelson, & Jusczyk, 2003). Thus, if they can pay attention to salient prosodically conditioned acoustic information (e.g., phrase-final lengthening, pauses, pitch contour discontinuity) correlating with syntactic constituent boundaries, they might be able to not only identify potential subdivisions in fluent speech but also to infer the location of some syntactic boundaries.

Function words and morphemes are elements that serve certain grammatical functions (e.g., articles, auxiliaries, pronouns). They are acquired within the first year of life because they are highly frequent (much more than content words: nouns, verbs, adverbs) and possess perceptual and distributional characteristics that distinguish them from content words (e.g., Gervain, Nespor, Mazuka, Horie, & Mehler, 2008; Shi, Morgan, & Allopenna, 1998; Shi, Werker, & Morgan, 1999). Because functional elements tend to consistently co-occur with content words from specific word classes (e.g., determiners such as the or a typically co-occur with nouns, whereas pronouns such as she and they tend to co-occur with verbs), infants could use statistical or distributional information in their input to learn about function words and to identify which words or sets of words co-occur with words from specific categories (e.g., Mintz, 2003). Validating this hypothesis, previous studies have shown that infants between 12 and 24 months old can use function words to categorize content words (e.g., “the blick” vs. “I blick”; Cauvet et al., 2014; He & Lidz, 2017; Shi & Melançon, 2010). However, in real life, not all content words are preceded by function words (e.g., in “The baby flies,” “flies” can be either a noun or a verb). In such cases, infants would need to integrate additional information into their distributional analysis and take into account syntactic constituents to constrain their parsing (i.e., “fly” is a noun in “The baby flies” but a verb in “The baby flies his kite”). Thus, function words/morphemes and phrasal prosody, together, may allow young infants to build at least a rudimentary representation of the syntactic structure of sentences (Christophe, Dautriche, de Carvalho, & Brusini, 2016; Christophe, Millotte, Bernal, & Lidz, 2008). Supporting this hypothesis, computational work demonstrates that models relying on phrasal prosody, function words, and a minimal semantic knowledge successfully predict the syntactic category of prosodic-syntactic units and unknown words (Christodoulopoulos, Roth, & Fisher, 2016; Gutman, Dautriche, Crabbé, & Christophe, 2015).

The situation is thus as follows: Phrasal prosody and function words are jointly predictive of syntactic structure in natural languages, and young infants are sensitive to each of these sources of information. What has never been investigated is whether infants can jointly use phrasal prosody and function words to access the syntactic structure of sentences and constrain their acquisition of word meanings. This ability would be crucial for language acquisition because it would allow infants to break free of the chicken-and-egg problem because phrasal prosody and function words are acquired well before infants know many words.

In the current study, we experimentally tested this hypothesis, investigating whether 18-month-old French-learning infants are able to exploit function words (Experiment 1) and phrasal prosody together with function words (Experiment 2) to constrain the acquisition of nouns and verbs.

Experiment 1: Function Words Constrain the Acquisition of Word Meanings

This experiment tested whether 18-month-olds are able to infer that a novel word such as bamoule refers to an object when they listen to sentences such as “It is a bamoule” and refers to an action when they listen to sentences such as “She is bamouling.”

Method

The entire method, data-analysis plan, and criteria for exclusion of participants for Experiment 1 were preregistered on the Open Science Framework before the experiment was conducted (osf.io/9n4u3). Materials and data can be found at osf.io/u2xct.

Participants

Forty-eight French-learning 18-month-olds participated in the study (24 in each experimental group; 27 girls; age: M = 18.1 months, SD = 0.2, range = 17.7–18.6). An additional 32 infants came to the lab but were not included in the final sample for one of the following reasons: not being able to finish the experiment because of fussiness (n = 13), failing to meet the preset habituation criterion (n = 6), parental interference (n = 3), a technical problem (n = 1), or crying during the experiment (n = 9). Parents provided informed consent. This research was approved by the local ethics committee.

Parents were asked to complete a French adaptation of the long version of the MacArthur-Bates Communicative Development Inventory (Kern & Gayraud, 2010) within 1 week of participation, as a measure of infants’ receptive and productive vocabularies. Scores were obtained for 34 of 48 infants. Reported comprehension vocabulary ranged from 10 to 502 words (M = 184, SD = 98.6), and reported production ranged from 4 to 128 words (M = 24, SD = 29.1). No correlation was found between infants’ vocabulary size and performance in this experiment (see “Supplementary Material” on the Open Science Framework).

At the beginning of this project, we had planned (and preregistered our intention) to test 16 participants in the experimental group (32 total) on the basis of the number of participants tested in previous studies using the same design (e.g., He & Lidz, 2017; Werker, Cohen, Lloyd, Casasola, & Stager, 1998). However, recent studies investigating the effects of sample size and statistical power in infant looking-time research (e.g., Oakes, 2017) have suggested that infant studies with sample sizes smaller than 24 per cell can be underpowered and result in false-positive and false-negative results. We thus ran a power analysis based on the effect size observed by He and Lidz (2017; Cohen’s d = 0.802315), which indicated that 25.38 participants would be needed in each condition to reach a significance level of less than .05 and a power of 80%, with a two-tailed alternative. Because counterbalancing required a number of participants divisible by 8, we tested a final sample of 24 infants in each condition (the closest number to 25.38).

Design

A habituation-switch paradigm (e.g., He & Lidz, 2017; Werker et al., 1998; cf. Fig. 1) was used to habituate infants to two video stimuli showing a penguin doing two different actions (one-participant agentive actions, e.g., spinning, cartwheeling), one in each video. During the presentation of one of the videos (e.g., a penguin spinning), infants heard sentences using a novel word as a noun, for example, “Regarde! C’est une bamoule!” (“Look! It’s a bamoule!”), in which bamoule is naming an object, the penguin), and during the presentation of the other video (e.g., a penguin cartwheeling), they heard sentences presenting another novel word as a verb, for example, “Regarde! Elle doripe!” (“Look! She is doriping!”), in which doripe is naming an action, cartwheeling). Note that the syntactic category of the novel words and the associations with the videos were counterbalanced across participants. Thus, half of the participants had bamoule as a noun and doripe as a verb, and half had the reverse. Half had spinning as the verb meaning, and half had cartwheeling as the verb meaning. This habituation phase gave the infants the opportunity to guess a possible meaning for each of the two novel words: If they exploited the linguistic context provided by the function words as adults would, then they would infer that the novel word employed as a noun (e.g., “C’est une bamoule”) referred to the penguin (the only object present in the video) and that the novel word employed as a verb (e.g., “Elle doripe”) referred to the action that the penguin was doing (e.g., spinning or cartwheeling, counterbalanced across participants). When infants reached a predefined habituation criterion (three consecutive trials for which the average looking time was less than 65% of the average looking time for the most-attended three consecutive trials), the habituation phase stopped and the test phase started immediately.

Fig. 1.

Experimental design: habituation-switch paradigm used in Experiments 1 and 2. All the infants were habituated to the same two video stimuli showing a penguin doing two different actions (e.g., spinning, cartwheeling—one in each video), while they listened to noun sentences, in which a novel word was used as a noun, and to verb sentences, in which another novel word was used as a verb. Then, as soon as infants reached a predefined habituation criterion, a test phase began with a switch between the sentences and the videos, in which half of the children saw a noun switch and half saw a verb switch. Given that the noun referred to an object (the penguin) whereas the verb referred to an action, if children correctly used function words (Experiment 1) and phrasal prosody (Experiment 2) to infer the meanings of the novel noun and the novel verb during the habituation phase, at test, they should have been more surprised (and looked more toward the videos) in the verb-switch condition (because the action changed, which is problematic for the verb interpretation) than in the noun-switch condition (in which they could still see a penguin in the video, and the fact that the action had changed was irrelevant). Note that the syntactic category of the novel words and the associations with the videos were counterbalanced across participants. Thus, half of the children tested had bamoule as a noun and doripe as a verb, and half had the reverse. Half had spinning as the verb meaning, and half had cartwheeling as the verb meaning.

At test, all infants were presented with two trials in which the audio tracks of the videos were switched (as illustrated in Fig. 1): Half of the infants heard a noun switch (e.g., they heard noun sentences with bamoule while watching the penguin cartwheeling) and half heard a verb switch (e.g., they heard sentences with the verb doripe while watching the penguin spinning). Given that the noun sentences refer to an object (i.e., a penguin, present in both videos) and the verb sentences refer to an action (i.e., either spinning or cartwheeling), if children correctly used the linguistic context to infer the meanings of the novel noun and the novel verb, they should be more surprised (look more toward the videos) in the verb-switch condition than in the noun-switch condition. Indeed, changing the action in the verb-switch condition violates the inference constructed about the verb meaning: Cartwheeling and spinning are different actions. However, the noun-switch condition did not violate the inference constructed about the noun meaning because infants could still see a penguin in the video (and the fact that the action had changed was irrelevant).

In other words, if children correctly exploited the linguistic context of the novel words to infer their meaning, we expected an asymmetrical pattern of results: more dishabituation in the verb-switch condition than in the noun-switch condition. Such an asymmetry between conditions, if we observed it, could only be due to the asymmetry in the linguistic contexts in which the novel words were presented (verb context vs. noun context). Indeed, if infants attempted to associate the novel words with some aspect of the video without taking into account the linguistic context, then the results should come out symmetrically (either dishabituation in both conditions or no dishabituation in both conditions).

Material

Two novel words in French (bamoule, doripe) were used as target words. For each novel word, two sentences were created: one sentence using it as a noun, for example, “Oh regarde! C’est une bamoule! Tu la vois la bamoule?” (“Oh look! It’s a bamoule! Do you see the bamoule?”) and another one using it as a verb, for example, “Oh regarde! Elle bamoule! Tu la vois qui bamoule?” (“Oh look! She is bamouling! Do you see her bamouling?”). Each sentence was repeated 12 times to create the audio tracks of the videos, resulting in a 37-s passage for each target word in each condition; each repetition was introduced by an audio prompt (e.g., “Oh,” “Wow,” “Hey”). All the passages had exactly the same audio prompts. The assignment of target words to syntactic categories was counterbalanced across participants, such that half had the target word bamoule as a noun and doripe as a verb, and half had the reverse. All the stimuli were recorded by a female native speaker of French (the last author), who recorded the sentences in a child-directed register.

The audio tracks of these passages were paired with two video stimuli showing a penguin doing two different actions (e.g., spinning, cartwheeling), one in each video. These videos had exactly the same duration as the audio tracks (37 s). Additionally, a silent video of a butterfly perched on a leaf was used to recapture infants’ attention when they looked away for more than 2 s.

Apparatus and procedure

Infants were tested individually in a sound-attenuated double-walled booth; each infant sat on a parent’s lap, facing a 27-in. TV screen positioned 70 cm away. A camera positioned on the top of the TV screen was connected with an LCD monitor placed outside the booth, where the experimenter sat. The experimenter observed the infant’s eye fixations to the screen and coded the infant’s looking behavior on-line by pressing a button on a keyboard when the infant was looking at the screen and releasing it as soon as the infant looked elsewhere. Parents wore headphones and listened to masking music during the entire experiment. The presentation of the stimuli and the on-line coding were controlled and recorded through the Habit program (Version 1.0; Cohen, Atkinson, & Chaput, 2004).

The experiment was composed of two phases: habituation and test. The experimenter was blind to the type of trial (habituation vs. test). The procedure started by displaying the attention getter on the screen (i.e., the silent video of a butterfly perched on a leaf). Once the child looked toward the screen, the experimenter initiated the first trial. If the toddler reoriented toward the screen within 2 s, the trial continued to play, but the time spent looking away was subtracted from the looking time. Each trial lasted until the child looked away for more than 2 s or until the maximum length of the trial was reached (i.e., 37 s).

During the habituation phase, the videos were presented repeatedly one after the other, for as much time as the child wanted to look at the TV screen, for a minimum of 4 trials and a maximum of 12 trials, depending on how fast the child reached the predefined habituation criterion. This criterion was reached when an infant’s average looking time during any block of 3 consecutive trials dropped to less than 65% of the average looking time for the most-attended block (i.e., the 3-trial block that had the longest total looking time). Habituation trials were presented in random blocks of two to avoid the same action–sentence pair occurring more than twice in a row and to ensure that the number of trials for each action–sentence pair was as balanced as possible, independently of the duration of the habituation phase. When infants reached the habituation criterion, the habituation phase stopped and the test phase started immediately.

At test, all infants were presented with a fixed number of two trials in which the audio sentences of the videos played during the habituation phase were switched. In other words, participants assigned to the noun-switch condition saw two trials in which the noun sentence was presented together with the video previously associated with the verb sentence, whereas participants assigned to the verb-switch condition saw two trials in which the verb sentence was presented together with the video previously associated with the noun sentence. Half of the participants were assigned to the noun-switch condition and half to the verb-switch condition.

Data processing and analyses

Data analyses and graphics were performed with R software (Version 3.2.2; R Core Team, 2015). We used the average looking time of the last two trials of the habituation phase and of the two test trials, and we compared the increase in looking time from habituation to test in the two experimental conditions (noun switch vs. verb switch). If infants were able to exploit function words to access the syntactic structure of the sentences in which the novel words occurred and if they could use this information to infer the syntactic category of the novel words and constrain their possible meaning, we expected a greater increase in looking time from habituation to test in the verb-switch condition than in the noun-switch condition. To test this, we performed an analysis of variance (ANOVA) with log-transformed mean looking times as the dependent measure, participants as the random factor, condition (noun switch vs. verb switch) as a between-participants factor, and phase (habituation vs. test) as a within-participants factor. The expected effect should have appeared as a significant interaction between condition and phase. Note that looking times were log-transformed before the ANOVA was run because the data did not follow a normal distribution, which is a necessary condition to conduct an ANOVA.

Results

The results of Experiment 1 are shown in Figure 2. Infants’ looking times increased more between habituation and test in the verb-switch condition than in the noun-switch condition: An ANOVA on log-transformed mean looking times revealed a significant interaction between condition and phase, F(1, 46) = 5.65, p = .022, d = 0.665, which confirms that, relative to the habituation phase, at test, children looked more toward the videos in the verb-switch condition than in the noun-switch condition. This is consistent with the interpretation that during the habituation phase, infants inferred that the novel verb referred to the action (e.g., cartwheeling), whereas the novel noun referred to the object (the penguin). Thus, at test, infants in the verb-switch condition were surprised when watching the penguin performing the other action (e.g., spinning) while listening to sentences containing the verb that they had associated with another action (e.g., cartwheeling) during the habituation phase. For instance, because the penguin was not cartwheeling but spinning, there was a discrepancy between the original meaning that they had inferred for this word and the current situation. Note that this increase in looking time can be explained either by the fact that infants thought the target word was not used correctly at test (i.e., the speaker was making a mistake and using the word incorrectly) or by the fact that infants realized that they had to broaden the meaning they had initially inferred for this word (e.g., instead of meaning cartwheeling specifically, doripe might refer to a broader class of movements, perhaps involving rotation, which is present in both actions). In contrast, infants tested in the noun-switch condition, who listened to noun sentences with bamoule while watching the penguin cartwheeling, did not show surprise during the test: Indeed, the meaning that they had inferred during habituation (that bamoule means penguin) was perfectly consistent with the test video they were watching because there was still a penguin on the screen.

Fig. 2.

Results of Experiment 1. The graph on the left shows mean looking time toward the videos during the last two trials of the habituation phase and during the two trials of the test phase, separately for children in the noun-switch condition (n = 24) and the verb-switch condition (n = 24). Error bars represent standard errors of the mean difference (test – habituation). The box-and-whiskers plots on the right show the increase in log-transformed mean looking times from habituation to test in each group. Each red dot represents the average for one participant in each group. Within each plot, the white dashed line represents the mean of the distribution, and the solid black line in the middle represents the median. The bottom of each box represents the 25th percentile (or first quartile), and the top represents the 75th percentile (or third quartile); the bottom whisker extends to the smallest value (unless it is an outlier), and the top whisker extends to the largest value (unless it is an outlier).

Consistent with the predictions from distributional-learning theories mentioned in the introduction, these results suggest that infants exploited the information carried by function words to constrain their interpretation of novel content-word meanings. Given that both groups were exposed to exactly the same videos and sentences during habituation, the only way to explain the asymmetry observed at test is that infants paid attention to the syntactic context instantiated by function words to correctly assign a syntactic category to the novel words and constrain their meanings. When we switched the audio tracks of the videos, this caused a violation of the inference that infants constructed about the verb meaning (i.e., cartwheeling and spinning are different actions) but not the inference about the noun meaning (i.e., it is still the same penguin in both videos).

Discussion

Experiment 1 shows that 18-month-olds use function words to compute the syntactic category of novel words and to constrain their probable meanings. However, because not all content words are directly preceded by function words, this strategy may not always be sufficient. Because the prosodic structure of an utterance correlates with its syntactic structure, listeners could exploit prosodic boundaries together with function words to constrain syntactic analysis (Christophe et al., 2016; Christophe et al., 2008; Morgan & Demuth, 1996), a hypothesis tested in Experiment 2.

Experiment 2: Phrasal Prosody and Function Words Constrain the Acquisition of Word Meanings

This experiment investigated whether infants take into account the position of a word within the prosodic structure of a sentence when computing its syntactic category (noun vs. verb).

Method

The entire method, data-analysis plan, and criteria for exclusion of participants for Experiment 2 were preregistered on the Open Science Framework before the experiment was conducted (osf.io/9n4u3). Materials and data can be found at osf.io/u2xct.

Participants

Forty-eight French-learning 18-month-olds participated in the study (24 in each experimental group; 24 girls; age: M = 18.2 months, SD = 0.2, range = 17.8–18.8). An additional 23 infants came to the lab but were not included in the sample for one of the following reasons: not being able to finish the experiment because of fussiness (n = 11), failing to meet the preset habituation criterion (n = 4), parental interference (n = 1), a technical problem (n = 2), or crying during the experiment (n = 5). Parents provided informed consent. This research was approved by the local ethics committee.

Parents were asked to complete a French adaptation of the long version of the MacArthur-Bates Communicative Development Inventory (Kern & Gayraud, 2010) within 1 week of participation as a measure of each infant’s receptive and productive vocabularies. Scores were obtained for 35 of 48 infants. Reported comprehension vocabulary ranged from 13 to 684 words (M = 213, SD = 125.3), and reported production ranged from 1 to 225 words (M = 33, SD = 43.3). No correlation was found between infants’ vocabulary size and performance.

Design

We used exactly the same paradigm as in Experiment 1 but investigated whether, in addition to function words, infants were able to rely on the relationship between the prosodic and syntactic structure of sentences to guide their syntactic interpretation and constrain the acquisition of word meanings. Instead of hearing a function word (e.g., article or pronoun) immediately preceding the to-be-learned word that would unambiguously cue its syntactic category, children had to take into account the prosodic structure in which a novel word appeared to compute its syntactic category. For instance, the novel word bamoule was presented as a noun in the sentence “[Regarde la petite bamoule]!” (“[Look at the little bamoule]!”), and the novel word doripe was presented as a verb in the sentence “[Regarde]! [La petite] [doripe]!” (“[Look]! [The little one] [is doriping]!”; brackets indicate prosodic phrase boundaries). As an illustration, note that we can find similar examples in English, with sentences such as “[Do you see the baby flies?],” in which “flies” is a noun, naming the insect, versus “[Do you see?] [The baby] [flies]!,” in which “flies” is a verb naming the action that the baby is doing.

Because both sentences in this experiment were composed of the same words and functional elements in the same order (regarde-la-petite-bamoule/doripe), an analysis in terms of which words precede bamoule or doripe is not sufficient to determine its syntactic category (because they are the same in both conditions); rather, the syntactic difference between these two sentences is reflected in their different prosodic structures. When doripe is a verb, there is a prosodic boundary preceding it (i.e., the boundary between the noun phrase and the verb phrase), and when bamoule is a noun, it is embedded in a single prosodic unit together with the other words of the sentence, corresponding to the verb look and the following noun phrase. If infants are able to use the information provided by the prosodic structure of a sentence to access its syntactic structure as adults and preschoolers do (de Carvalho, Dautriche, & Christophe, 2016; de Carvalho, Lidz, Tieu, Bleam, & Christophe, 2016; Millotte, Wales, & Christophe, 2007; Snedeker & Yuan, 2008) and if they can use this information to constrain the meaning of novel words, during the habituation phase, they should infer that the novel word used as a noun refers to the penguin (the only object present in the video) and that the novel word used as a verb refers to the action that the penguin is doing (e.g., spinning or cartwheeling, counterbalanced across participants). Thus, as in Experiment 1, during the test phase, we expected infants to look more toward the video (being more surprised) in the verb-switch condition than in the noun-switch condition.

Material

The same two novel words in French (bamoule, doripe) were used as target words to create minimal pairs of sentences that differed only in their prosodic structures. Thus, for each novel word, two sentences were created: one presenting the target word in a noun position within the prosodic-syntactic structure—“[Regarde la petite bamoule]! [Tu vois la petite bamoule]?” (“[Look at the little bamoule]! [Do you see the little bamoule]?”) and another one presenting the novel word in a verb position—“[Regarde]! [la petite] [bamoule]! [Tu vois]? [La petite] [bamoule]!” (“[Look]! [The little one] [is bamouling]! [Do you see]? [The little one] [is bamouling]!”). An example of each kind of sentence is depicted in Figure 3.

Fig. 3.

An example of the minimal pair of sentences created for Experiment 2. The x-axis represents the time course for each sentence. For each sentence type, the figure shows the waveform and the pitch contour (blue curve). Both sentences were composed of exactly the same words (la-petite-bamoule/doripe) but differed only in their prosodic structure, which reflected their different syntactic structures. In noun sentences, all the critical words were grouped together into a single prosodic unit: “[Regarde la petite bamoule]!” “([Look at the little bamoule]!”) In verb sentences, these words were spread into three different prosodic units: “[Regarde]! [La petite] [bamoule]!” (“[Look]! [The little one] [is bamouling]!”) The syntactic category of the novel words and the associations with the videos were counterbalanced across participants. Thus, half of the participants heard bamoule as a noun and doripe as a verb, and half heard the reverse. Half saw spinning as the verb meaning, and half saw cartwheeling as the verb meaning.

Sentences uttered in the verb condition had a phonological phrase boundary before the target word (i.e., corresponding to the boundary between the noun and the verb phrase). In contrast, in sentences uttered in the noun condition, all the words were grouped together into a single prosodic unit; these prosodic structures are consistent with theoretical descriptions of the relationship between prosodic and syntactic boundaries (e.g., Nespor & Vogel, 1986).

Each sentence was repeated 12 times to create the audio tracks, resulting in a 50-s passage for each target word in each condition; each repetition was introduced by an audio prompt (e.g., “Oh,” “Wow,” “Hey”). As in Experiment 1, the assignment of a syntactic category to the two novel words was counterbalanced across participants. All the stimuli were recorded by the same speaker as in Experiment 1, in child-directed register. The audio tracks were paired with the same video stimuli as those used in Experiment 1 (but with a duration of 50 s). The same silent video of a butterfly perched on a leaf was used as an attention getter.

Acoustic analyses

To assess prosodic differences between the two conditions, we conducted acoustic measurements (duration and pitch) on each of the eight exemplars of the test sentences (bamoule as a noun and bamoule as a verb, doripe as a noun and doripe as a verb, with the novel word repeated twice in each case; see Fig. 3). As expected from the literature (e.g., Jun & Fougeron, 2002), the analysis of duration revealed a significant preboundary lengthening: The rhyme of the word preceding the target word (e.g., “ite” from “petite”) in the verb condition (in which it was placed just before the prosodic phrase boundary) was lengthened by 211% compared with this same segment in the noun condition (in which it was placed in the middle of a prosodic unit; 381 vs. 130 ms; see Table 1). A silent pause of 62 ms preceding the target word (i.e., between petite and bamoule) was observed in the verb condition, whereas there was no pause between these words in the noun condition.

Table 1.

Mean Duration and Mean Pitch Change for the Stimuli in Experiment 2

Dependent variable	Noun sentence(e.g., [la petite bamoule])	Verb sentence(e.g., [la petite] [bamoule])	Comparison (two-tailed)
Duration (in milliseconds)
Rhyme of word preceding target (e.g., “ite” from petite)	130 (2.3)	381 (33.9)	t(3) = −7.35, p = .005
Pause before target (e.g., between petite and bamoule)	0 (0)	62 (10.9)	t(3) = −5.74, p = .010
Onset of target word (e.g., “b” from bamoule)	97 (10.8)	127 (5.9)	t(3) = −1.79, p = .171
Rhyme of target word (e.g., “oule” from bamoule)	533 (35.8)	443 (8.6)	t(3) = 2.71, p = .073
Pitch change (in Hertz, from the beginning to the end of the target words)
Word preceding target (e.g., last pitch value at the last vowel from petite minus first pitch value from the first vowel of petite)	−29.2 (3.1)	150 (27.7)	t(3) = −6.39, p = .008
Target word (e.g., last pitch value of “u” from bamoule minus first pitch value of “a” from bamoule)	66.7 (42.2)	25 (12.7)	t(3) = 0.77, p = .497

Note: Values in parentheses are standard errors for the segments around the prosodic boundaries.

The analysis of pitch contours in both sentence conditions revealed a significant difference between conditions (see Table 1), consistent with the literature describing French as having a tendency for a rising pitch contour toward the end of prosodic units (e.g., Welby, 2006). Thus, the word preceding the target word (i.e., petite) exhibited a greater rising pitch pattern in the verb prosody condition (+150 Hz; because of its position at the end of a prosodic unit) than in the noun prosody condition (−29 Hz; when it was placed in the middle of a prosodic unit). Given that in both conditions, the target word was placed at the end of a prosodic unit, no particular hypothesis was made regarding their differences in pitch or duration at the final position. The target word in the noun prosody condition (e.g., bamoule) seemed to exhibit a greater rising pitch pattern in the noun-prosody condition (+66 Hz) than in the verb-prosody condition (+25 Hz), but this difference was not significant.

Note that in previous studies (e.g., Fisher & Tokura, 1996; Soderstrom, Blossom, Foygel, & Morgan, 2008), acoustic analysis of mothers’ spontaneous speech addressed to American English-learning infants showed that they naturally produce the kind of acoustic cues associated with prosodic boundaries (both utterance- and internal phrase-level boundaries), such as the one we exploited here, including reliable prosodic cues to grammatical units such as boundaries between subject and verb phrases. In French, a study with adults found that they naturally produce the acoustic cues associated with the prosodic boundaries between noun and verb phrases that we exploited in the current experiment (Millotte et al., 2007). Taken together, these studies suggest that the relationship between prosodic and syntactic boundaries in our experiment may be present in children’s everyday spoken input and could be learned through exposure to language.

Apparatus and procedure

The apparatus and procedure were similar to those in Experiment 1. The only difference concerns the sentences uttered during the presentation of the videos. During the presentation of one of the videos (e.g., the penguin spinning), infants listened to sentences presenting a novel word in a noun position within the prosodic-syntactic structure; during the presentation of the other video (e.g., the penguin cartwheeling), they listened to sentences presenting the novel word in a verb position within the prosodic-syntactic structure. As in Experiment 1, in each trial, infants had the opportunity to listen to a maximum of 12 repetitions of the test sentences (for a total duration of 50 s because adding the word petite made the sentences longer). Half of the children were tested in the noun-switch condition, and the other half were tested in the verb-switch condition. Data processing and analyses were conducted the same way as in Experiment 1.

Results

The results of Experiment 2 are shown in Figure 4. Infants’ looking time increased more between habituation and test in the verb-switch condition than in the noun-switch condition: An ANOVA on log-transformed mean looking times revealed a significant interaction between condition and phase, F(1, 46) = 5.09, p = .029, d = 0.632, showing that infants looked longer (were more surprised) when tested in the verb-switch condition than in the noun-switch condition. This behavior, as in Experiment 1, suggests that the switch of the actions led to a violation of the inference constructed about the verb meaning but not about the noun meaning; consequently, infants were more surprised when listening to verb sentences than to noun sentences during the test phase.

Fig. 4.

Results of Experiment 2. The graph on the left shows mean looking time toward the videos during the last two trials of the habituation phase and during the two trials of the test phase, separately for children in the noun-switch condition (n = 24) and the verb-switch condition (n = 24). Error bars represent standard errors of the mean difference (test – habituation). The box-and-whiskers plots on the right show the increase in log-transformed mean looking times from habituation to test in each group. Each red dot represents 1 participant. Within each plot, the white dashed line represents the mean of the distribution, and the solid black line in the middle represents the median. The bottom of each box represents the 25th percentile (or first quartile), and the top represents the 75th percentile (or third quartile); the bottom whisker extends to the smallest value (unless it is an outlier), and the top whisker extends to the largest value (unless it is an outlier). Black dots below and above the whiskers indicate outliers.

Comparing the results of Experiments 1 and 2, one may note a lesser increase in looking times between habituation and test in Experiment 2 relative to Experiment 1, which surfaces in Experiment 2 as a small decrease between habituation and test in the noun-switch condition and a smaller increase than in Experiment 1 between habituation and test in the verb-switch condition. This might be because experimental trials were longer in Experiment 2 than in Experiment 1 (50 s vs. 37 s), which means that toddlers could reach the habituation criterion with longer average looking times (Experiment 2: 14.68 s, SD = 6.6; Experiment 1: 12.94 s, SD = 4.86), which would leave them more room to go down further at test. This remains speculative, as we do not know whether such a difference would be replicable (and had not stated we would make such a comparison in our preregistration). Note, however, that this does not impact our main result because the expected effect was a significant interaction between condition and phase, revealing a greater increase in looking time from habituation to test in the verb-switch condition compared with the noun-switch condition.

The present results thus show that at 18 months, infants are already able to use phrasal prosody as a cue to interpret a novel word as either a noun or a verb depending on its position within the prosodic structure of sentences. When listening to minimal pairs of sentences such as “Regarde la petite bamoule,” which can be produced either as a single prosodic unit, as in “[Regarde la petite bamoule]!” in which bamoule is used as a noun, or as three prosodic units, as in “[Regarde]! [La petite] [bamoule]!” in which bamoule is used as a verb, 18-month-olds correctly interpreted the target word as either a noun or a verb, depending on its position within the prosodic structure of the sentence, and they used this information to constrain the meaning of this novel word.

It is important to note that the noun and verb sentences had exactly the same words; thus, a distributional analysis tracking which functional element was preceding or following the target word would not have been sufficient to constrain infants’ interpretation in this experiment. This does not suggest, however, that infants did not exploit function words together with prosodic information to constrain their interpretations. Rather, we believe that prosody and function words were jointly exploited in this experiment because whereas prosodic boundaries allowed infants to group words into syntactic constituents and informed them about the location of syntactic boundaries, the prosodic boundaries, per se, did not directly provide the syntactic labels of constituents (e.g., noun phrase, verb phrase). To interpret the novel words as nouns or as verbs, infants had to use the information carried by the function words inside the prosodic units to determine the syntactic nature of these constituents delimited by prosody. For instance, when participants heard a sentence such as “[Regarde]! [La petite] [bamoule]!” the prosodic boundary before the target word bamoule signaled the presence of a syntactic constituent boundary. Given that the first prosodic unit (“la petite”) started with an article (“la”), this unit could be identified as a noun phrase: “la” (determiner) “petite” (noun). Having identified the first unit as a full noun phrase, infants might have expected it to be followed by a verb phrase, which allowed them to interpret bamoule as a verb. In the noun condition, in contrast, infants interpreted the novel word as a noun because all three words (“la” + “petite” + “bamoule”) appeared together in a single prosodic unit with the well-known verb regarde (“Regarde la petite bamoule!”).

Importantly, these results show that even in a situation in which the information provided by function words alone was not sufficient to compute the syntactic category of the novel words, infants were able to exploit prosodic information to recover the syntactic structure of sentences and, in combination with the other words, infer the syntactic category of novel words and therefore constrain their meanings.

Discussion

These results show that 18-month-olds are able to use phrasal prosody, together with function words, to recover the syntactic structure of sentences and to interpret a novel word as either a noun (referring to an object) or a verb (referring to an action), depending on its position within the prosodic-syntactic structure.

General Discussion

Across two experiments, we demonstrated that 18-month-olds, who are still in the process of learning the syntax and building up the lexicon of their language, can rely on function words and phrasal prosody to access the syntactic structure of sentences and guide their discovery of novel word meanings. In Experiment 1, French infants exploited the functional elements in a sentence to assign a syntactic category to a novel word and to constrain its meaning. In Experiment 2, in which the information provided by function words alone was not sufficient to compute the syntactic category of the novel words, infants simultaneously exploited prosodic information together with function words to recover the syntactic structure of sentences. In our study, infants used prosodically conditioned acoustic information to parse spoken sentences into groups of words and identify possible syntactic constituents; they exploited the function words inside these prosodic-syntactic constituents to determine the syntactic nature of these constituents, which in turn allowed them to infer the syntactic category of novel words and therefore constrain their meanings (associating nouns with objects and verbs with actions).

Infants’ ability to use function words and phrasal prosody to identify noun and verb contexts is truly impressive and raises the question of how infants may have learned which contexts go with nouns and which go with verbs. Some studies have suggested that infants could rely on a handful of known words to act as a seed for learning noun and verb categories (Brusini, Amsili, Chemla, & Christophe, 2011; Christophe et al., 2016; Gutman et al., 2015; see also Yarowsky, 1995). This hypothesis relies on two established premises: First, infants know the meaning of a few highly frequent content words (e.g., Bergelson & Swingley, 2012); second, they group concepts into semantic categories (e.g., objects, actions, agents; Carey, 2009). If, in addition, infants expect words from similar conceptual categories to occur in similar syntactic environments (Gleitman, 1990; Pinker, 1984), they could exploit the syntactic contexts in which the few words they already know appear, to infer information about other unknown words that appear in the same contexts. For instance, having observed that the words bottle and teddy bear label an object kind and often occur after the, as in “the bottle” and “the teddy bear,” infants could infer that other words occurring in that syntactic environment will share conceptual properties as well (e.g., blick in “the blick” might also refer to an object kind). Taken together, phrasal prosody, function words, and a distributional analysis of the contexts in which known words appear may be extremely useful to access the syntactic category of unknown words and allow infants to bootstrap their acquisition of word meanings.

This powerful mechanism might provide infants with a tool to construct a first-pass syntactic structure of spoken sentences, during the first steps of language acquisition. In this sense, our answer to the chicken-and-egg problem of learning word meanings through syntax and learning syntax through word meanings is that by providing information about the syntactic structure of sentences, phrasal prosody and function words can work as anchors to help infants access syntactic information. Crucially, because infants are sensitive to phrasal prosody and function words during the first year of life, the ability to jointly exploit phrasal prosody and function words to access syntactic structure may be in place even before infants know many words.

Our results were obtained with French, but we expect that phrasal prosody and function words should support early access to syntax in many different languages. Although prosodic information and functional elements can surface differently across languages, this information is present in all of the world’s languages (Dryer, 1992; Shattuck-Hufnagel & Turk, 1996). Overall, we suggest that phrasal prosody and function words may well represent a universal and extremely useful tool for infants to access syntactic information through a surface analysis of the speech stream and to bootstrap their way toward successful language acquisition.

Supplemental Material

deCarvalhoOpenPracticesDisclosure – Supplemental material for Prosody and Function Words Cue the Acquisition of Word Meanings in 18-Month-Old Infants

Supplemental material, deCarvalhoOpenPracticesDisclosure for Prosody and Function Words Cue the Acquisition of Word Meanings in 18-Month-Old Infants by Alex de Carvalho, Angela Xiaoxue He, Jeffrey Lidz and Anne Christophe in Psychological Science

Footnotes

Acknowledgements

We thank Alejandrina Cristia, Sid Kouider, and Alexander Martin for their thoughtful comments on previous versions of the manuscript.

Action Editor

Rebecca Treiman served as action editor for this article.

Author Contributions

All the authors designed the study. A. de Carvalho performed the research, analyzed the data, and wrote the manuscript. A. Christophe, J. Lidz, and A. X. He provided critical revisions. All the authors approved the final manuscript for submission.

Declaration of Conflicting Interests

The author(s) declared that there were no conflicts of interest with respect to the authorship or the publication of this article.

Funding

This research was supported by a PhD fellowship from the École Normale Supérieure and by a postdoctoral fellowship from the Fyssen Foundation to A. de Carvalho. It was also supported by grants from the Région IIe-de-France, Fondation de France, LabEx IEC (ANR-10-LABX-0087), Initiatives for Excellence PSL (ANR-10-IDEX-0001-02), and the ANR “Apprentissages” (ANR-13-APPR-0012).

Open Practices

All materials, collected data, and data analyses have been made publicly available via the Open Science Framework and can be accessed at osf.io/u2xct. The design and analysis plans were preregistered and can be accessed at osf.io/9n4u3. The complete Open Practices Disclosure for this article can be found at http://journals.sagepub.com/doi/suppl/10.1177/0956797618814131. This article has received the badges for Open Data, Open Materials, and Preregistration. More information about the Open Practices badges can be found at .

Notes

References

Bergelson

Swingley

(2012). At 6–9 months, human infants know the meanings of many common nouns. Proceedings of the National Academy of Sciences, USA, 109, 3253–3258. doi:10.1073/pnas.1113380109

Bernal

Lidz

Millotte

Christophe

(2007). Syntax constrains the acquisition of verb meaning. Language Learning and Development, 3, 325–341. doi:10.1080/15475440701542609

Brusini

Amsili

Chemla

Christophe

(2011). Learning to categorize nouns and verbs on the basis of a few known examples: A computational model relying on 2-word contexts. Paper presented at the biennial meeting of the Society for Research in Child Development, Montreal, Québec, Canada.

Carey

(2009). The origin of concepts. Oxford, England: Oxford University Press.

Cauvet

Limissuri

Millotte

Skoruppa

Cabrol

Christophe

(2014). Function words constrain on-line recognition of verbs and nouns in French 18-month-olds. Language Learning and Development, 10, 1–18. doi:10.1080/15475441.2012.757970

Christodoulopoulos

Roth

Fisher

(2016). An incremental model of syntactic bootstrapping. In Proceedings of the 7th Workshop on Cognitive Aspects of Computational Language Learning (pp. 38–43). Stroudsburg, PA: Association for Computational Linguistics.

Christophe

Dautriche

de Carvalho

Brusini

(2016). Bootstrapping the syntactic bootstrapper. In Scott

Waughtal

(Eds.), Proceedings of the 40th annual Boston University Conference on Language Development (pp. 75–88). Somerville, MA: Cascadilla Press.

Christophe

Millotte

Bernal

Lidz

(2008). Bootstrapping lexical and syntactic acquisition. Language and Speech, 51, 61–75. doi:10.1177/00238309080510010501

Cohen

Atkinson

Chaput

(2004). Habit X: A New Program for Obtaining and Organizing Data in Infant Perception and Cognition Studies (Version 1.0) [Computer software]. Austin: University of Texas.

10.

de Carvalho

Dautriche

Christophe

(2016). Preschoolers use phrasal prosody online to constrain syntactic analysis. Developmental Science, 19, 235–250. doi:10.1111/desc.12300

11.

de Carvalho

Lidz

Tieu

Bleam

Christophe

(2016). English-speaking preschoolers can use phrasal prosody for syntactic parsing. The Journal of the Acoustical Society of America, 139, EL216-EL222. doi:10.1121/1.4954385

12.

Dryer

(1992). The Greenbergian word order correlations. Language, 68, 81–138.

13.

Fisher

Tokura

(1996). Prosody in speech to infants: Direct and indirect acoustic cues to syntactic structure. In Morgan

J. L.

Demuth

(Eds.), Signal to syntax: Bootstrapping from speech to grammar in early acquisition (pp. 343–363). Hillsdale, NJ: Erlbaum.

14.

Gervain

Nespor

Mazuka

Horie

Mehler

(2008). Bootstrapping word order in prelexical infants: A Japanese–Italian cross-linguistic study. Cognitive Psychology, 57, 56–74. doi:10.1016/j.cogpsych.2007.12.001

15.

Gleitman

(1990). The structural sources of verb meanings. Language Acquisition, 1, 3–55.

16.

Gutman

Dautriche

Crabbé

Christophe

(2015). Bootstrapping the syntactic bootstrapper: Probabilistic labeling of prosodic phrases. Language Acquisition, 22, 285–309. doi:10.1080/10489223.2014.971956

17.

A. X.

Lidz

(2017). Verb learning in 14- and 18-month-old English-learning infants. Language Learning and Development, 13, 335–356. doi:10.1080/15475441.2017.1285238

18.

Jun

Fougeron

(2002). Realizations of accentual phrase in French intonation. Probus, 14, 147–172. doi:10.1515/prbs.2002.002

19.

Kern

Gayraud

(2010). Inventaire Français du Développement Communicatif (IFDC). [French Communicative Development Inventory]. Grenoble, France: La Cigale.

20.

Männel

Friederici

A. D.

(2009). Pauses and intonational phrasing: ERP studies in 5-month-old German infants and adults. Journal of Cognitive Neuroscience, 21, 1988–2006. doi:10.1162/jocn.2009.21221

21.

Mehler

Jusczyk

Lambertz

Halsted

Bertoncini

Amiel-Tison

(1988). A precursor of language acquisition in young infants. Cognition, 29, 143–178. doi:10.1016/0010-0277(88)90035-2

22.

Millotte

Wales

Christophe

(2007). Phrasal prosody disambiguates syntax. Language and Cognitive Processes, 22, 898–909. doi:10.1080/01690960701205286

23.

Mintz

T. H.

(2003). Frequent frames as a cue for grammatical categories in child directed speech. Cognition, 90, 91–117. doi:10.1016/S0010-0277(03)00140-9

24.

Morgan

Demuth

(1996). Signal to syntax: Bootstrapping from speech to grammar in early acquisition. Mahwah, NJ: Erlbaum.

25.

Nespor

Vogel

(1986). Prosodic phonology. Dordrecht, The Netherlands: Foris.

26.

Oakes

L. M.

(2017). Sample size, statistical power, and false conclusions in infant looking-time research. Infancy, 22, 436–469. doi:10.1111/infa.12186

27.

Pinker

(1984). Language learnability and language development. Cambridge, MA: Harvard University Press.

28.

R Core Team. (2015). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.

29.

Shattuck-Hufnagel

Turk

A. E.

(1996). A prosody tutorial for investigators of auditory sentence processing. Journal of Psycholinguistic Research, 25, 193–247. doi:10.1007/BF01708572

30.

Shi

Melançon

(2010). Syntactic categorization in French-learning infants. Infancy, 15, 517–533. doi:10.1111/j.1532-7078.2009.00022.x

31.

Shi

Morgan

J. L.

Allopenna

P. D.

(1998). Phonological and acoustic bases for earliest grammatical category assignment: A cross-linguistic perspective. Journal of Child Language, 25, 169–201. doi:10.1017/S0305000997003395

32.

Shi

Werker

J. F.

Morgan

J. L.

(1999). Newborn infants’ sensitivity to perceptual cues to lexical and grammatical words. Cognition, 72(2), B11–B21. doi:10.1016/S0010-0277(99)00047-5

33.

Shukla

White

K. S.

Aslin

R. N.

(2011). Prosody guides the rapid mapping of auditory word forms onto visual objects in 6-mo-old infants. Proceedings of the National Academy of Sciences, USA, 108, 6038–6043. doi:10.1073/pnas.1017617108

34.

Snedeker

Yuan

(2008). Effects of prosodic and lexical constraints on parsing in young children (and adults). Journal of Memory and Language, 58, 574–608. doi:10.1016/j.jml.2007.08.001

35.

Soderstrom

Blossom

Foygel

Morgan

(2008). Acoustical cues and grammatical units in speech to two preverbal infants. Journal of Child Language, 35, 869–902. doi:10.1017/S0305000908008763

36.

Soderstrom

Seidl

Kemler Nelson

D. G.

Jusczyk

P. W.

(2003). The prosodic bootstrapping of phrases: Evidence from prelinguistic infants. Journal of Memory and Language, 49, 249–267. doi:10.1016/S0749-596X(03)00024-X

37.

Waxman

S. R.

Lidz

J. L.

Braun

I. E.

Lavin

(2009). Twenty four-month-old infants’ interpretations of novel verbs and nouns in dynamic scenes. Cognitive Psychology, 59, 67–95. doi:10.1016/j.cogpsych.2009.02.001

38.

Welby

(2006). French intonational structure: Evidence from tonal alignment. Journal of Phonetics, 34, 343–371. doi:10.1016/j.wocn.2005.09.001

39.

Werker

J. F.

Cohen

L. B.

Lloyd

V. L.

Casasola

Stager

C. L.

(1998). Acquisition of word–object associations by 14-month-old infants. Developmental Psychology, 34, 1289–1309. doi:10.1037/0012-1649.34.6.1289

40.

Yarowsky

(1995). Unsupervised word sense disambiguation rivaling supervised methods. In Proceedings of the 33rd annual meeting of the Association for Computational Ling-uistics (pp. 189–196). Morristown, NJ: Association for Com-putational Linguistics. doi:10.3115/981658.981684

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.20 MB