Lexical Acquisition Through Category Matching: 12-Month-Old Infants Associate Words to Visual Categories

Abstract

Although it is widely recognized that human infants build a sizeable conceptual repertoire before mastering language, it remains a matter of debate whether and to what extent early conceptual and category knowledge contributes to language development. We addressed this question by investigating whether 12-month-olds used preverbal categories to discover the meanings of new words. We showed that one group of infants (n = 18) readily extended novel labels to previously unseen exemplars of preverbal visual categories after only a single labeling episode, but two other groups struggled to do so when taught labels for unfamiliar categories (those who had been previously exposed, n = 18, or not exposed, n = 18, to category tokens). These results suggest that infants expect labels to denote categories of objects and are equipped with learning mechanisms responsible for matching prelinguistic knowledge structures with linguistic inputs. This ability is consistent with the idea that our conceptual machinery provides building blocks for vocabulary and language acquisition.

Keywords

word learning word extension categorization cognitive development infancy open data open materials

Language not only pervades human communication and social interactions but also provides us with powerful cognitive tools. In particular, language enables the development of new systems of knowledge (Spelke, 2003) and new means of transmission (Gelman & Roberts, 2017). Important concepts are lexicalized (e.g., “coffee,” “computer,” “DNA”), and labels are used in generic statements to convey semantic knowledge that extends beyond the here and now (e.g., “Coffee keeps people awake”). The development of these devices of cultural transmission starts in the form of word learning in early infancy. Although infants witness only particular things being named, their mental lexicon is not a catalogue of sounds paired with those items. Rather, it contains labels that represent categories of objects, such as body parts, food items, or artifacts (Bergelson & Swingley, 2012; Parise & Csibra, 2012). To date, however, it remains unknown how word meanings that enable the storage and generalization of cultural knowledge develop. Two chief possibilities have been suggested.

Some researchers have argued that the discovery of word meanings is initially guided by attentional biases (Smith, Jones, & Landau, 1996) that make children attend to salient perceptual features and interpret words as names for objects sharing these features (Smith, 2003). For example, toddlers tend to generalize new words to objects of the same shape (a phenomenon known as shape bias; Colunga & Smith, 2005). However, experience with language is needed for this strategy to emerge (Smith, Jones, Landau, Gershkoff-Stowe, & Samuelson, 2002; Smith & Samuelson, 2006), and children below 2 years of age struggle to use it (Son, Smith, & Goldstone, 2008). This observation is in stark contrast to the evidence that preverbal infants successfully generalize familiar words (Bergelson & Swingley, 2012; Parise & Csibra, 2012), suggesting that before shape bias develops, other cognitive mechanisms must support the acquisition of word meanings.

On the other hand, it has been proposed that language builds on preexisting conceptual knowledge (Clark, 2017; Mandler, 2004). Indeed, it has been widely documented that infants have access to some innate concepts (e.g., object, agent; Carey, 2009) and readily learn a variety of novel categories, both with help from communication (Ferguson & Waxman, 2016; Ferry, Hespos, & Waxman, 2013) and, more importantly, without it (Mandler, 2004; Mareschal & Quinn, 2001). Preverbal categories could become the meanings of new words, thanks to a specialized word-learning mechanism referred to as category matching (Macnamara, 1982; Nelson, 1973). According to this account, infants seek to map new words onto categorical representations already available in their conceptual repertoire. Once the new word becomes linked to a specific category, word generalization beyond the labeling context follows automatically, as all objects identified as members of this category fall under the label describing it. Such a mechanism would ensure fast word learning from minimal input. Despite the consensus that language must, to some degree, be built on our prelinguistic representations (Clark, 2004), to date there is no direct evidence that category knowledge can be directly used as a source of word meanings in infancy.

Here, we asked whether human infants possess special cognitive mechanisms, such as category matching, that ensure efficient word learning and open the way for the acquisition of knowledge conveyed through language. More specifically, across three experiments, we investigated whether preverbal categories can be accessed by infants and interpreted as word meanings when only one of the category tokens is named. First, we tested whether 12-month-olds use preverbal category knowledge to identify the meanings of new nouns and extend them to previously unseen category exemplars (Experiment 1). Then we explored whether preverbal categories were necessary and sufficient to explain the observed word-extension effects (Experiments 2 and 3).

Experiment 1

Twelve-month-olds viewed objects from two new categories, either in a category-training procedure or in a manner not conducive to category learning. Then they saw a single exemplar of each category being labeled. We reasoned that when given the opportunity to learn categories before labeling, infants would engage in word learning through category matching and extend the new labels to previously unseen exemplars of these categories.

Method

Participants

In total, 36 healthy, full-term 12-month-olds participated in the experiment. Eighteen infants were assigned to the blocked-categories group (10 females; average age = 12.11 months, range = 11.64–12.76 months) and 18 to the interleaved-categories group (8 females; average age = 12.06 months, range = 11.73–12.65 months). An additional 7 infants were tested but excluded from the final sample because they either did not provide enough data (n = 4; see Data Analysis) or failed to complete the experiment (n = 3). Families were recruited on a voluntary basis through advertisement in the local press. All infants were raised in monolingual English-speaking families, and all caregivers provided written informed consent. The target sample size of 36 was determined a priori on the basis of previous studies that have investigated word learning and word extension in infancy using eye tracking (e.g., Yin & Csibra, 2015). We acknowledge, however, that more optimal approaches for determining sample sizes, such as taking into account publication biases, have recently been suggested (e.g., Anderson, Kelley, & Maxwell, 2017).

Apparatus

Infants’ binocular gaze data were acquired using a Tobii TX 300 eye tracker (Tobii Technology AB, Danderyd, Sweden) with a 23-in. integrated monitor (sampling rate: 120 Hz, resolution: 1,920 × 1,080 pixels). The sound was delivered via external stereo speakers. The experiment was administered using custom-built MATLAB scripts (The MathWorks, Natick, MA) with the Psychophysics Toolbox for stimuli presentation (Brainard, 1997) and the Tobii Pro Analytics software development kit for data collection (https://www.tobiipro.com/product-listing/tobii-pro-sdk/).

Stimuli

Pictorial stimuli

We selected two types of objects that were unfamiliar to the infants: coffee makers and staplers (Fig. 1). Prior to the experimental session, the caregivers confirmed that the infants had not been previously exposed to these objects. Fourteen color photographs were selected for each type of object. The depicted items differed in color, orientation, and shape but had a similar surface area. All stimuli were presented on a white background about 15 cm by 15 cm in size that subtended approximately 14° of visual angle. For each object category, a subset of 13 images was randomly selected (different for every participant): 8 images for familiarization, 1 for word learning, and 4 for the word-extension test.

Fig. 1.

Schematic visualization of the experimental design in Experiment 1. The task had three phases: familiarization, word training, and word-extension test. Only the familiarization phase differed between the conditions with respect to whether the tokens of the two target categories were blocked into separate familiarization streams (the blocked-categories group) or not (the interleaved-categories group). Speech stimuli were played only during the word-training phase and the word-extension test (see descriptions at the bottom of each phase). The depicted object images represent a sample stimuli sequence corresponding to one experimental session. Each category token was presented only once.

Speech stimuli

The speech stimuli were two phonetically distinct pseudowords that conformed to English phonotactics: rif and toma. Because infants process labels that are presented in sentence frames more efficiently (Fennell & Waxman, 2010), the words were embedded in carrier phrases (see Design and Procedure). All speech stimuli were recorded by a female native speaker of British English, using infant-directed speech.

Design and procedure

The experiment consisted of three phases administered without pause—familiarization, word training, and word-extension test—for a total length of about 4 min. Each infant was randomly assigned to one of the two groups, which differed only with respect to the type of familiarization (i.e., blocked or interleaved categories). Infants were tested individually in a soundproof, dimly lit room. They sat in a car seat approximately 60 cm away from the monitor. To obtain a reliable eye-tracking signal, we used a five-point calibration sequence prior to the recording, which was repeated until all points were successfully calibrated.

Familiarization

Each infant was exposed to a set of objects consisting of 16 visual items, with 8 tokens per category. Images were introduced by sliding them into a preselected area (1,580 × 620 pixels) in the geometrical center of the display. Each image appeared for 1.5 s, zooming in and out and then sliding out of the display.

In both groups, infants received two familiarization streams containing 8 items each. In the blocked-categories group, one familiarization stream consisted of 8 items from category A and the other of 8 items from category B (Fig. 1). The order of item appearance and the order of category presentation (A vs. B) were randomized. In the interleaved-categories group, both familiarization streams included four items from one category interleaved with four items from the other category. The items were randomized with the constraint that no more than two tokens of the same category could be presented in succession.

Younger and Fearing (1999) demonstrated that, when presented with items from two distinct real-world categories (i.e., cat and horse) interleaved within a single familiarization stream, infants had difficulty forming two differentiated categories. In contrast, infants excel at forming categories when familiarized with items of a single real-world category (i.e., cat and horse; see Eimas & Quinn, 1994).

Moreover, grouping or highlighting various exemplars of the same category is a strategy employed by teachers when they intend to demonstrate category-specific, as opposed to exemplar-specific, properties (Shafto, Goodman, & Griffiths, 2014). Therefore, presenting category A and category B tokens in two separate familiarization streams should lead to formation of two distinct categories, whereas presenting tokens from category A interleaved with tokens from category B should lead to weaker categorization performance or none at all (see also Experiment 3).

Word training

Following familiarization, infants were introduced to two novel words, each paired with a previously unseen exemplar of one of the target categories. Each object was labeled six times: “This is a [name]! [Name]! Do you see the [name]? Look at the [name]! Find the [name]! [Name]!” The pictures were kept still during the delivery of the auditory stimuli but moved to the four corners and the center of the screen (randomized) between the labeling phrases to maximize the infants’ interest in the procedure. The label–category pairings and the order of presentation of a particular label–category pair during the training were counterbalanced across infants.

Word-extension test

The test phase comprised 4 trials, or 2 per word. All test trials had the same structure: After fixating on a centrally displayed attention getter, infants were exposed to two novel images (one from each category), different at each of the four test trials. The images were displayed side by side for 10 s. Only one label was provided during each test trial. The label was uttered three times—first embedded in a carrier phrase and then twice as a single word, with short periods of silence between presentations (“Look at the [name]! [Name]! [Name]!”).

Images were presented in silence for 2 s before the onset of speech so that we could measure the baseline preference for the two objects. The three tokens of the target label were delivered after 2.8 s, 5 s, and 7.5 s, respectively. The side on which the images appeared and the order of word and image presentation were counterbalanced across participants. There were two possible word orders: ABAB and BABA.

A short attention getter was presented in the center of the display before every familiarization stream, every word-training trial, and every word-extension test. The attention getter was gaze-controlled by the infants; that is, its duration was not predefined but depended on the infants’ looking behavior. The attention getter was displayed on the screen until infants fixated on it continuously for 500 ms.

Data analysis

Infants had to fulfill two criteria to be included in the final sample: minimum 50% attendance to the screen during each of the three experimental phases and contribution of a minimum of one valid test trial for each word (i.e., a trial with more than 50% attendance to the test stimuli computed across both pre- and postnaming). Infants in both groups spent a comparable amount of time attending to the stimuli (blocked-categories group: M = 77% of total session time, SD = 11%; interleaved-categories group: M = 78%, SD = 9%), t(34) = 0.09, p = .930, and provided a comparable number of valid test trials (blocked-categories group: M = 3.61 trials, SD = 0.70, range = 2–4; interleaved-categories group: M = 3.78 trials, SD = 0.55, range = 2–4), t(34) = 0.80, p = .430.

To analyze infants’ looking behavior during test trials, we defined two regions in the test display, one corresponding to the target and one to the distractor object. Following the terminology used in the eye-tracking literature, we refer to these regions as areas of interest: the target area of interest and the distractor area of interest, respectively. We investigated two indices of word comprehension: the proportion of target looking and the longest-look difference score. The proportion of target looking was computed by dividing the amount of time spent in the target area of interest by the total amount of time spent in the target and distractor areas of interest. Using proportional rather than absolute looking times ensured that the effects were not driven by infants who display overall longer looking times. The longest look was defined as the longest time spent within a single visit to an area of interest (a visit corresponds to the sum of fixations within the area of interest not separated by a fixation outside of the area of interest). The longest-look difference score was calculated by subtracting the longest look to the target from the longest look to the distractor and then dividing this difference by the sum of longest looks to both the target and distractor. The difference scores ranged from −1 to 1, with positive values indicating that infants directed their longest look to the target.

To assess the impact of the speech stimuli delivered during the test on the infants’ looking behavior, we divided test trials into prenaming and postnaming segments, and both word-comprehension measures were derived separately for each segment (for the time course of infants’ target fixation, see Fig. S1 in the Supplemental Material available online). The postnaming segment started 367 ms after the onset of the first token of the target word (Swingley, Pinto, & Fernald, 1999) and lasted until the end of the trial (i.e., the prenaming segment lasted 3.167 s, whereas the postnaming segment lasted 6.833 s). An increase from pre- to postnaming in looking toward the target, as indexed by the proportion of target looking or longest-look difference score, was taken as evidence for word extension.

Results

To assess infants’ word-extension performance, we entered the mean proportions of target looking time during the pre- and postnaming segments of the test into a two-way mixed-model analysis of variance (ANOVA) with group (blocked-categories familiarization vs. interleaved-categories familiarization) as a between-subjects factor and segment (prenaming vs. postnaming) as a within-subjects factor (Fig. 2). This analysis yielded a significant main effect of segment, F(1, 34) = 4.29, p = .046, η_p² = .11, and an interaction between group and segment, F(1, 34) = 12.39, p = .001, η_p² = .27. In line with our predictions, follow-up paired-samples t tests revealed that this interaction was due to increased target looking from the pre- to the postnaming segment in the blocked-categories group, t(17) = 3.69, p = .002, d = 0.87, 95% confidence interval (CI) = [.04, .14] (prenaming: M = .51, SD = .08; postnaming: M = .60, SD = .12), with 15 of 18 infants showing this effect, but not in the control group, t(17) = 1.11, p = .283, d = 0.26, 95% CI = [−.02, .07] (prenaming: M = .52, SD = .17; postnaming: M = .49, SD = .16). Moreover, following naming, only infants who received category training looked at the named object significantly more than expected by chance (.50), t(17) = 3.58, p = .002, d = 0.84, 95% CI = [.54, .66]; interleaved-categories group: t(17) = 0.23, p = .82, d = 0.05, 95% CI = [.41, .57]. Prior to naming, neither group displayed a preference for either of the test stimuli—blocked-categories group: t(17) = 0.60, p = .558, d = 0.14, 95% CI = [.47, .55], interleaved-categories group: t(17) = 0.41, p = .690, d = 0.09, 95% CI = [.43, .60]. Note that for two-sample tests, we report 95% CIs for the difference in the mean of the dependent variable and for one-sample, tests we report 95% CIs for the mean.

Fig. 2.

Results of Experiment 1: (a) mean proportion of target looking time and (b) mean longest-look difference score during test in the prenaming and postnaming segments, separately for the blocked- and interleaved-categories groups. Diamonds indicate means, and black horizontal lines indicate medians. The bottom and the top of the boxes represent the first and the third quartiles. Whiskers extend from the middle quartiles to the smallest and largest values within 1.5 times the interquartile range. Black points connected across boxes represent the data sets of individual participants. Dashed lines represent chance performance. Significant effects are indicated with asterisks (p < .01). Positive values in (b) indicate that the longest looks were directed at the target; negative values indicate that the longest looks were directed at the distractor.

A two-way mixed-model ANOVA on the average longest-look difference score yielded a significant interaction between group and segment, F(1, 34) = 15.37, p < .001, η_p² = .31. Infants’ looks toward the target increased in duration after labeling in the blocked-categories group, t(17) = 4.06, p = .001, d = 0.96, 95% CI = [.13, .42] (prenaming: M = −.01, SD = .20; postnaming: M = 0.26, SD = 0.26), with 16 of 18 infants displaying this pattern of results, but not in the interleaved-categories group, t(17) = 1.49, p = .154, d = 0.35, 95% CI = [−.04, .25] (prenaming: M = .06, SD = .35; postnaming: M = −.04, SD = .37). Before labeling, the difference scores were at chance in both groups—blocked-categories group: t(17) = 0.29, p = .776, d = 0.07, 95% CI = [−.11, .09]; interleaved-categories group: t(17) = 0.76, p = .458, d = 0.18, 95% CI = [−.11, .24]. After labeling, the longest-look difference scores increased above chance in the blocked-categories group, t(17) = 4.32, p < .001, d = 1.02, 95% CI = [.13, .39], but not in the interleaved-categories group, t(17) = 0.46, p = .65, d = 0.11, 95% CI = [−.22, .14].

Discussion

Our results suggest that preverbal categories are used by infants as meanings of newly encountered labels and enable them to determine the extensions of these labels after only a single naming episode. However, two alternative explanations of our findings should be considered. First, it remains possible, although unlikely (Son et al., 2008), that infants could generalize new words without prior category knowledge. We addressed this possibility in Experiment 2 by testing whether removing the category training would affect infants’ word-extension performance. Second, the chance performance in the interleaved-category group might have been due to learning a single superordinate-like category that included both coffee makers and staplers. Children are reluctant to attach multiple names to a single object (a phenomenon known as mutual exclusivity; Markman & Wachtel, 1988), so they might have disregarded the naming events because objects from the single superordinate category were given two different names. We investigated this issue in Experiment 3 by testing whether the interleaved-category presentation of two kinds of objects leads to formation of a single category.

Experiment 2

To establish whether word generalization depends on category knowledge, we modified the task used in Experiment 1 by removing the familiarization phase. Furthermore, we administered an additional word-recognition test that followed the word-extension test. By using the same objects that were presented during the labeling events, this test investigated whether it is specifically the ability to generalize words, and not the ability to map words onto the specific objects, that is impaired without category knowledge.

Method

Participants

Eighteen healthy full-term 12-month-olds (6 females; average age = 12.08 months, range = 11.5–12.9 months) participated in the experiment. Thirteen other infants were excluded from the analysis because they did not provide enough data (n = 10, following the same exclusion criteria as in Experiment 1), because the experiment was interrupted (n = 1), because of a technical failure (n = 1), or because they were born preterm (n = 1). The procedure for recruiting families and obtaining their consent to participate was the same as that of Experiment 1, and sample size was selected to match Experiment 1 as well.

Apparatus and stimuli

The experimental setup and stimuli were identical to those used in Experiment 1. For each object category (coffee makers and staplers), a subset of five pictures was randomly selected (i.e., different for each participant) that included one picture for the word-learning phase and the word-recognition test and four for the word-extension test. The order of image presentation was randomly determined for each participant.

Design and procedure

The design and procedure were the same as in Experiment 1, with two exceptions. First, the word-training phase was not preceded by any familiarization. Instead, to ensure that the infants attended to the screen before the word training started, we presented them with a 5-s video clip depicting flower buds opening to the sound of soft music. Second, we administered two types of word-learning tests of four trials each to independently assess two dimensions of infants’ word-learning performance: word extension and word recognition. The word-extension test was identical to the one in Experiment 1 and consisted of two measurement periods: prenaming and postnaming (see Experiment 1).

The word-recognition test had the same structure as the word-extension test, except that instead of using novel category exemplars, we used the same tokens as in the word-training phase. The order of presentation of the two tests was fixed to enable direct comparisons with Experiment 1: that is, the word-extension test was always administered immediately after the word training and before the word-recognition test. Word-training and test trials were presented after a short gaze-controlled attention getter (minimum duration = 500 ms).

Data analysis

The participant-inclusion criteria and data-analysis plan were identical to those used in Experiment 1. We conducted two separate analyses for each type of test. Word-extension analysis consisted of 18 data sets, while word-recognition analysis consisted of 15 data sets, as 3 participants did not contribute at least two valid word-recognition trials. On average, the included data sets contained 3.78 valid word-extension test trials (SD = 0.43, range = 3–4) and 3.33 valid word-recognition test trials (SD = 0.82, range = 2–4).

Results

Word extension

Unlike in Experiment 1, infants did not increase their looking to the named category exemplars from pre- to postnaming, as revealed by a paired-samples t test comparing the proportions of target looking between pre- and postnaming, t(17) = 0.36, p = .723, d = 0.08, 95% CI = [−.06, .08] (prenaming: M = .49, SD = .08; postnaming: M = .50, SD = .10; see Fig. 3). Their looking level remained at chance throughout the test trial—prenaming: t(17) = 0.62, p = .545, d = 0.14, 95% CI = [.45, .53]; postnaming: t(17) = 0.02, p = .98, d < 0.01, 95% CI = [.45, .55]. The duration of the infants’ looks directed to the target and distractor objects was not affected by labeling, t(17) = 0.86, p = .402, d = 0.20, 95% CI = [−.07, .16] (prenaming: M = −.05, SD = .13; postnaming: M = −.01, SD = .21), with difference scores remaining at the chance level during both prenaming, t(17) = 1.73, p = .101, d = 0.41, 95% CI = [−.12, .01], and postnaming, t(17) = 0.14, p = .888, d = 0.03, 95% CI = [−.11, .10]. Thus, in this experiment, infants did not generalize novel labels. This result is in line with the literature documenting difficulties in word extension without relevant category knowledge in older children (Son et al., 2008).

Fig. 3.

Results of Experiment 2: (a) mean proportion of target looking time and (b) mean longest-look difference score during test in the prenaming and postnaming segments, separately for each test type (word-extension test, word-recognition test). Diamonds indicate means, and black horizontal lines indicate medians. The bottom and the top of the boxes represent the first and the third quartiles. Whiskers extend from the middle quartiles to the smallest and largest values within 1.5 times the interquartile range. Black points connected across boxes represent the data sets of individual participants. Dashed lines represent chance performance. Significant effects are indicated with symbols (^†p ≤ .10, *p < .05, **p < .01). Positive values in (b) indicate that the longest looks were directed at the target; negative values indicate that the longest looks were directed at the distractor.

Word-extension results were compared across experiments using mixed-model ANOVAs with group (blocked-categories familiarization in Experiment 1 vs. interleaved-categories familiarization in Experiment 1 and vs. no familiarization in Experiment 2) as a between-subjects factor and segment (prenaming vs. postnaming) as a within-subjects factor. The analysis on the proportions of target looking yielded a significant interaction between group and segment, F(2, 51) = 4.51, p = .016, η_p² = .15. This interaction occurred because speech affected infants’ looking patterns differently across groups. That is, before naming, the duration of infants’ target looking was comparable across groups, F(2, 51) = 0.30, p = .741, η_p² = .01, while after naming it varied across groups, F(2, 51) = 4.19, p = .021, η_p² = .14). Post hoc comparisons indicated that, after naming, blocked-categories familiarization led to longer target looking than interleaved-categories familiarization, p < .036 (Bonferroni corrected), and marginally longer than no familiarization, p < .061 (Bonferroni corrected). The analysis of longest-look differences scores yielded a similar pattern of results: the interaction between group and segment, F(2, 51) = 8.96, p < .001, η_p² = .26, was due to differences in how naming affected the duration of longest looks across groups—prenaming: F(2, 51) = 1.04, p = .362, η_p² = .04; postnaming: F(2, 51) = 5.97, p = .005, η_p² = .19. Blocked-categories familiarization resulted in longer looks to the target than in each of the control groups (vs. interleaved-categories familiarization, p = .003; vs. no familiarization, p < .007; both values Bonferroni corrected). These results confirm that there is a difference in how infants with different levels of category knowledge extended novel words: Namely, infants who received category training (i.e., blocked-categories familiarization, Experiment 1) performed better than infants who did not receive such training (i.e., interleaved-categories familiarization, Experiment 1, and no familiarization, Experiment 2).

Word recognition

Infants displayed a tendency to increase their target looking between pre- and postnaming, t(14) = 1.75, p = .101, d = 0.45, 95% CI = [−.02, .16] (prenaming: M = .49, SD = .12; postnaming: M = .56, SD = .09). Eleven out of 15 infants showed this effect. In addition, the proportion of target looking was significantly above chance during the postnaming period, t(14) = 2.62, p = .020, d = 0.68, 95% CI = [.51, .61], but not during the prenaming period, t(14) = 0.38, p = .710, d = 0.10, 95% CI = [.42, .56]. The duration of infants’ longest looks toward the target object increased significantly from pre- to postnaming, t(14) = 2.27, p = .039, d = 0.59, 95% CI = [.01, .35] (prenaming: M = −.02, SD = .26; postnaming: M = .16, SD = .21), with 12 of 15 participants displaying this effect. The difference scores were significantly above chance during the postnaming, t(14) = 3.06, p = .008, d = 0.79, 95% CI = [.05, .28], but not during the prenaming, t(14) = 0.26, p = .800, d = 0.07, 95% CI = [−.16, .13].

Discussion

Without nonverbal category knowledge, infants failed to extend novel labels to further exemplars of the labeled category, while showing a tendency to identify by name the specific objects that were used during labeling. Although we cannot conservatively conclude that infants mapped the trained words onto the particular objects that were labeled during the training, as the increase in the proportion of target looking after labeling failed to reach significance, the longest-look results suggest that such an association has been formed. This discrepancy between the two measures can be explained by the fact that the longest-look measure is not affected by the decrease of attention over time, as is the case for total duration of looking (Schafer & Plunkett, 1998). Note also that the word-recognition test was always administered following the word-extension test, meaning that infants had to (a) maintain the newly formed label–object mappings over the short delay required to complete the first test phase and (b) handle possible interference with processing it.

To sum up, word extension does not occur spontaneously following exposure to labeling and suggests that word extension is independent from word mapping. In particular, witnessing an unfamiliar object being labeled did not provide infants with sufficient information to reliably and rapidly extend the label to other objects.

Experiment 3

As we did not directly assess category learning in Experiment 1, it remains possible that infants formed the new categories only during the labeling, prompted by the presence of labels. Furthermore, infants in the interleaved-categories group, who saw staplers and coffee makers mixed in a single familiarization stream, might have formed one extensive category containing both kinds of objects. These possibilities were tested in Experiment 3.

Method

Participants

Thirty-six 12-month-old infants were included in the final sample. Infants were randomly assigned to two experimental groups, 18 to the blocked-familiarization group (10 females; average age = 12.11 months, range = 11.57–12.49 months) and 18 to the interleaved-familiarization group (8 females; average age = 12.15 months, range = 11.62–12.51 months). An additional 9 infants were excluded from the analysis because they did not complete the calibration routine (n = 1), failed to provide enough data (n = 6), or failed to complete the experiment (n = 2). Sample size was selected to match those of Experiments 1 and 2. Families were recruited and gave consent to participate as in Experiments 1 and 2.

Apparatus

Experiment 3 was conducted in a different laboratory from Experiments 1 and 2. The experimental setup was the same as for Experiments 1 and 2, except that infants’ gaze data were acquired using a Tobii Pro X2-60 eye tracker (sampling rate: 60 Hz) and the visual stimuli were displayed on an external 23-in. monitor (resolution: 1,920 × 1,080 pixels).

Stimuli

In addition to the two types of objects used in the previous experiments (familiarized categories: staplers and coffee makers), we selected one more type of object that was unfamiliar to the infants (novel category: garlic press), as confirmed through parental reports. For the familiarized categories, we used the same photographs as in Experiment 1, out of which a subset of 10 was randomly selected for each participant: eight images for familiarization and two images for the categorization test. We selected four additional photographs depicting exemplars of the novel category from which a subset of two images was used during the test (randomly determined). The additional pictures varied in color, orientation, and shape but were matched in size and surface area to the pictures used in Experiment 1.

Design and procedure

The task included only two phases—familiarization and the categorization test—which were administered without pause. The familiarization phase was identical to that of Experiment 1, with half of the infants tested on blocked-categories familiarization and the other half on interleaved-categories familiarization. During the categorization test phase, infants were shown two pairs of test pictures, displayed side by side in silence: a previously unseen token from the familiarized category (stapler or coffee maker, in a randomized order) and a token of a novel category (garlic press). Each test pair was displayed for 6 s and then repeated with the objects’ locations (left vs. right) swapped, in accordance with the designs used in the infant-categorization literature (e.g., Plunkett, Hu, & Cohen, 2008). Each familiarization stream and each test trial were preceded by a short gaze-controlled attention getter (minimum duration = 500 ms). The procedure was the same as for Experiment 1.

Data analysis

The inclusion criteria were the same as in Experiment 1. Note, however, that the categorization test trials were shorter than the word-extension word trials: 50% of on-screen time corresponded to 3 s. Infants had to provide at least two valid test trials (one trial per familiarization category) to be included in the final sample. Infants in both groups attended equally to the display (blocked-categories group: M = 83% of total session time, SD = 8%; interleaved-categories group: M = 82% of total session time, SD = 12%) t(34) = 0.02, p = .984, and provided a comparable number of valid test trials (blocked-categories group: M = 3.89, SD = 0.32, range = 3–4; interleaved-categories group: M = 3.94, SD = 0.24, range = 3–4), t(34) = 0.59, p = .559.

Following the methodology established in the field of infant categorization, we assessed category formation by measuring infants’ preference for the novel category tokens operationalized as the proportion of time spent looking at the novel category relative to the total looking time. Novelty preference is considered to be an index of category formation (e.g., Ferry et al., 2013; Plunkett et al., 2008; Younger & Fearing, 1999). This reflects the fact that infants discriminate between the test images and perceive one of them as familiar and the other as novel (on the basis of their freshly formed category knowledge); they orient preferentially to the novel stimulus. To determine whether infants reliably displayed a looking preference for the novel category, we divided the amount of time they spent in the novel category’s area of interest by the total amount of time spent in the novel and familiarized categories’ areas of interest (refer to the Supplemental Material for an additional analysis of the longest looks).

Results

Infants in the blocked-categories group looked longer at the novel category tokens (M = .58, SD = .16) than did infants in the interleaved-categories group (M = .48, SD = .06), t(23.25) = 2.34, p = .028, d = 0.78, 95% CI = [.01, .18], according to a Welch’s t test. Only in the blocked-categories group did infants look toward the novel object more than expected by chance, t(17) = 2.14, p = .047, d = 0.50, 95% CI = [.51, .66]; interleaved-categories group: t(17) = 0.96, p = .351, d = 0.23, 95% CI = [.45, .52] (see Fig. 4).

Fig. 4.

Results of Experiment 3: average proportion of looking toward novel category tokens during test, separately for the blocked and interleaved categories. Diamonds indicate means, and black horizontal lines indicate medians. The bottom and the top of the boxes represent the first and the third quartiles. Whiskers extend from the middle quartiles to the smallest and largest values within 1.5 times the interquartile range. The dashed line represents chance performance. The asterisk indicates a significant effect (p < .05).

Discussion

First, category formation occurred in the absence of language in the blocked-categories group under conditions identical to those in Experiment 1, indicating that in Experiment 1, categorical representations were available to infants before labeling. Second, there was no evidence for category formation in the interleaved-categories group. It is therefore unlikely that infants’ failure to generalize new words in Experiment 1 could be explained by the formation of a single superordinate category spanning both target categories (staplers and coffee makers), which interfered with the word-mapping processes.

General Discussion

The current results provide the first experimental evidence that preverbal category knowledge bootstraps the discovery of word meanings. We found that 12-month-olds rapidly grasped the meanings of two novel labels, each introduced with an object representing a different basic-level category. Infants’ understanding of new words was revealed through their word-extension performance: They succeeded at extending the new labels to previously unseen tokens of the referent categories. Importantly, infants could do so only when given the chance to learn the two categories before the new labels were introduced (as in the blocked-categories familiarization phase in Experiment 1). Infants who did not have relevant category knowledge before labeling, either because they were exposed to the exemplars of labeled categories in a manner not conducive to category formation (e.g., the interleaved-categories familiarization phase in Experiment 1) or because they were presented with the word training without any prior exposure to the category (Experiment 2), failed to extend novel words (see also Son et al., 2008).

Our results suggest that specialized cognitive mechanisms allow preverbal infants to rapidly acquire basic linguistic devices of cultural transmission. In particular, infants rely on category matching to identify the meanings of new labels. We demonstrated that even newly formed visual categories can be linked to novel words, as their meanings can guide the infants’ word extensions. Our findings by no means exclude the possibility that, later in development, attentional biases developing in children because of their experience with language (Gershkoff-Stowe & Smith, 2004) play a role in word learning (Smith et al., 2002; Son et al., 2008), but our results do highlight the fact that knowledge of preverbal categories contributes to word-generalization processes before that happens. Note that in the absence of relevant category knowledge, infants successfully associate new labels with particular objects (Pruden, Hirsh-Pasek, Golinkoff, & Hennon, 2006; Woodward, Markman, & Fitzsimmons, 1994), as Experiment 2 suggests. Therefore, nonverbal category knowledge is not a prerequisite for associating labels and the specific objects used during labeling, but enables infants to go beyond these associations.

How is category matching achieved? One possibility is that infants expect objects introduced through communication to represent categories. According to Csibra and Shamsudheen (2015), the labeled object is treated as a symbol standing for its category and the new label is directly attached to a category-level representation. In this scenario, category knowledge is used during labeling. Alternatively, the category knowledge might be used only at the stage of word extension. Initially, infants associate the label with a particular object, but on hearing the new label at test without its original referent, they retrieve the category to which this object belongs and extend the label along the boundary of this category. Further research is needed to establish which of these two strategies is at play.

We propose that the category-matching mechanism is available for word learning in the real world but remain aware that the categories used by infants in our study were taught in a lab in the somewhat artificial manner of assembling category tokens. However, category tokens also occur together outside labs (e.g., crossing a London street crowded with cars, strolling in a park full of dogs and trees, or just walking through a supermarket filled with products presented in an orderly fashion). This is sometimes due to the structuring of our environment for learning purposes. For example, caregivers who are asked to teach the properties of a category automatically select numerous category exemplars rather than a single one to highlight the category-diagnostic properties (Rhodes, Gelman, & Brickman, 2010). Infants, when allowed to freely manipulate objects, tend to explore exemplars of only one category in a sequence rather than alternating between exemplars of different categories (i.e., sequential touching; Rakison & Butterworth, 1998), thus selectively sampling the information in their environment. Even if this particular category-learning strategy were never available in real life, our finding still stands, as category-matching mechanisms should operate regardless of the way in which categories are learned (Waxman & Booth, 2003; Yin & Csibra, 2015).

In the past, emphasis has been placed on demonstrating that labeling guides category learning (Ferguson & Waxman, 2017). Our study provides the first evidence that visual categories themselves are directly exploited during word learning as sources of word meanings. One important implication of this observation is that individual differences in category learning may be a source of individual differences in the rate of language acquisition. For example, poor category learning in certain developmental disorders, such as autism (e.g., Davis & Plaisted-Grant, 2015), may explain the slower rate of vocabulary growth (Hudry et al., 2014).

To conclude, the current results not only constitute the earliest evidence for the generalization of newly learned words but also indicate the cognitive mechanisms underlying this ability. Infants use nonverbal category knowledge as a source of word meanings, and in its earliest stages, the cognitive machinery responsible for category formation is a key component of language development. Therefore, nonverbal conceptual resources provide a stepping stone to language and culture transmission.

Supplemental Material

PomiechowskaOpenPracticesDisclosure – Supplemental material for Lexical Acquisition Through Category Matching: 12-Month-Old Infants Associate Words to Visual Categories

Supplemental material, PomiechowskaOpenPracticesDisclosure for Lexical Acquisition Through Category Matching: 12-Month-Old Infants Associate Words to Visual Categories by Barbara Pomiechowska and Teodora Gliga in Psychological Science

Supplemental Material

PomiechowskaSupplementalMaterial – Supplemental material for Lexical Acquisition Through Category Matching: 12-Month-Old Infants Associate Words to Visual Categories

Supplemental material, PomiechowskaSupplementalMaterial for Lexical Acquisition Through Category Matching: 12-Month-Old Infants Associate Words to Visual Categories by Barbara Pomiechowska and Teodora Gliga in Psychological Science

Footnotes

Acknowledgements

We thank Katarina Begus, Gergely Csibra, Johannes Mahr, Eugenio Parise, and Denis Tatone for comments on the manuscript; Sinead Rocha for recording the speech stimuli; Eszter Körtveylesi, Krisztina Andrasi, Maria Toth, and Iti Arora for research assistance; and Chloe Taylor for proofreading the manuscript.

Action Editor

Rebecca Treiman served as action editor for this article.

Author Contributions

B. Pomiechowska and T. Gliga conceived and designed the study, collected and analyzed the data, and wrote the manuscript.

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 the European Commission Marie Curie Initial Training Networks (Grant 264301), the UK Medical Research Council (G0701484), and a European Research Council Starting Grant (284236).

Supplemental Material

Additional supporting information can be found at

Open Practices

All data and materials have been made publicly available via the Open Science Framework and can be accessed at https://osf.io/9cndv and https://osf.io/auxs4/, respectively. The design and analysis plans for the experiments were not preregistered. The complete Open Practices Disclosure for this article can be found at http://journals.sagepub.com/doi/suppl/10.1177/0956797618817506. This article has received badges for Open Data and Open Materials. More information about the Open Practices badges can be found at .

References

Anderson

S. F.

Kelley

Maxwell

S. E.

(2017). Sample-size planning for more accurate statistical power: A method adjusting sample effect sizes for publication bias and uncertainty. Psychological Science, 28, 1547–1562.

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.

Brainard

D. H.

(1997). The Psychophysics Toolbox. Spatial Vision, 10, 433–436.

Carey

(2009). The origin of concepts. New York, NY: Oxford University Press.

Clark

E. V.

(2004). How language acquisition builds on cognitive development. Trends in Cognitive Sciences, 8, 472–478.

Clark

E. V.

(2017). Semantic categories in acquisition. In Cohen

Lefebvre

(Eds.), Handbook of categorization in cognitive science (pp. 397–421). Amsterdam, The Netherlands: Elsevier.

Colunga

Smith

L. B.

(2005). From the lexicon to expectations about kinds: A role for associative learning. Psychological Review, 112, 347–382.

Csibra

Shamsudheen

(2015). Nonverbal generics: Human infants interpret objects as symbols of object kinds. Annual Review of Psychology, 66, 689–710.

Davis

Plaisted-Grant

(2015). Low endogenous neural noise in autism. Autism, 19, 351–362.

10.

Eimas

P. D.

Quinn

P. C.

(1994). Studies on the formation of perceptually based basic-level categories in young infants. Child Development, 65, 903–917.

11.

Fennell

C. T.

Waxman

S. R.

(2010). What paradox? Referential cues allow for infant use of phonetic detail in word learning. Child Development, 81, 1376–1383.

12.

Ferguson

Waxman

(2017). Linking language and categorization in infancy. Journal of Child Language, 44, 527–552.

13.

Ferguson

Waxman

S. R.

(2016). What the [beep]? Six-month-olds link novel communicative signals to meaning. Cognition, 146, 185–189.

14.

Ferry

A. L.

Hespos

S. J.

Waxman

S. R.

(2013). Nonhuman primate vocalizations support categorization in very young human infants. Proceedings of the National Academy of Sciences, USA, 110, 15231–15235.

15.

Gelman

S. A.

Roberts

S. O.

(2017). How language shapes the cultural inheritance of categories. Proceedings of the National Academy of Sciences, USA, 114, 7900–7907. doi:10.1073/pnas.1621073114

16.

Gershkoff-Stowe

Smith

L. B.

(2004). Shape and the first hundred nouns. Child Development, 75, 1098–1114.

17.

Hudry

Chandler

Bedford

Pasco

Gliga

Elsabbagh

. . . Charman

(2014). Early language profiles in infants at high-risk for autism spectrum disorders. Journal of Autism and Developmental Disorders, 44, 154–167.

18.

Macnamara

(1982). Names for things: A study of human learning. Cambridge, MA: MIT Press.

19.

Mandler

J. M.

(2004). Thought before language. Trends in Cognitive Sciences, 8, 508–513.

20.

Mareschal

Quinn

P. C.

(2001). Categorization in infancy. Trends in Cognitive Sciences, 5, 443–450.

21.

Markman

E. M.

Wachtel

G. F.

(1988). Children’s use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology, 20, 121–157.

22.

Nelson

(1973). Some evidence for the cognitive primacy of categorization and its functional basis. Merrill-Palmer Quarterly of Behavior and Development, 19, 21–39.

23.

Parise

Csibra

(2012). Electrophysiological evidence for the understanding of maternal speech by 9-month-old infants. Psychological Science, 23, 728–733.

24.

Plunkett

J. F.

Cohen

L. B.

(2008). Labels can override perceptual categories in early infancy. Cognition, 106, 665–681.

25.

Pruden

S. M.

Hirsh-Pasek

Golinkoff

R. M.

Hennon

E. A.

(2006). The birth of words: Ten-month-olds learn words through perceptual salience. Child Development, 77, 266–280.

26.

Rakison

D. H.

Butterworth

G. E.

(1998). Infants’ use of object parts in early categorization. Developmental Psychology, 34, 49–62.

27.

Rhodes

Gelman

S. A.

Brickman

(2010). Children’s attention to sample composition in learning, teaching and discovery. Developmental Science, 13, 421–429.

28.

Schafer

Plunkett

(1998). Rapid word learning by fifteen-month-olds under tightly controlled conditions. Child Development, 69, 309–320.

29.

Shafto

Goodman

N. D.

Griffiths

T. L.

(2014). A rational account of pedagogical reasoning: Teaching by, and learning from, examples. Cognitive Psychology, 71, 55–89.

30.

Smith

L. B.

(2003). Learning to recognize objects. Psychological Science, 14, 244–250.

31.

Smith

L. B.

Jones

S. S.

Landau

(1996). Naming in young children: A dumb attentional mechanism? Cognition, 60, 143–171.

32.

Smith

L. B.

Jones

S. S.

Landau

Gershkoff-Stowe

Samuelson

(2002). Object name learning provides on-the-job training for attention. Psychological Science, 13, 13–19.

33.

Smith

L. B.

Samuelson

(2006). An attentional learning account of the shape bias: Reply to Cimpian and Markman (2005) and Booth, Waxman, and Huang (2005). Developmental Psychology, 42, 1339–1343.

34.

Son

J. Y.

Smith

L. B.

Goldstone

R. L.

(2008). Simplicity and generalization: Short-cutting abstraction in children’s object categorizations. Cognition, 108, 626–638.

35.

Spelke

E. S.

(2003). What makes us smart? Core knowledge and natural language. In Gentner

Goldin-Meadow

(Eds.), Language in mind: Advances in the study of language and thought (pp. 277–311). Cambridge, MA: MIT Press.

36.

Swingley

Pinto

J. P.

Fernald

(1999). Continuous processing in word recognition at 24 months. Cognition, 71, 73–108.

37.

Waxman

Booth

(2003). The origins and evolution of links between word learning and conceptual organization: New evidence from 11-month-olds. Developmental Science, 6, 128–135.

38.

Woodward

A. L.

Markman

E. M.

Fitzsimmons

C. M.

(1994). Rapid word learning in 13- and 18-month-olds. Developmental Psychology, 30, 553–566.

39.

Yin

Csibra

(2015). Concept-based word learning in human infants. Psychological Science, 26, 1316–1324.

40.

Younger

B. A.

Fearing

D. D.

(1999). Parsing items into separate categories: Developmental change in infant categorization. Child Development, 70, 291–303.

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

0.23 MB

0.40 MB