Rare co-occurrences but frequent visual objects and object names: the at-home data for object name learning

• In Event: Infants’ learning about object properties and categories in diverse environments

Abstract

Most accounts of infants’ early object name learning focus on moments of co-occurrence between the name and its referent as the key contributor to learning. However, studies have suggested that co-occurrence alone may not be sufficient (Yu & Smith, 2008; Pereira, Smith, & Yu, 2014). The benefit of object or word familiarity on forming word-object mappings (Graf Estes, Evans, Alibali, & Saffran, 2007; Kucker & Samuelson, 2012; Mayor & Plunkett, 2010; Swingley, 2007) suggests that the role of co-occurrences may be mediated by object and word frequency. As such, the present study examines co-occurrences in infants’ egocentric views and the distribution of objects and their names individually in these views.
Videos containing mealtime events from 7 to 11-month-old infants (n = 14) were selected for analysis from a larger corpus of head-camera videos collected in children’s homes with no experimenter present. The sample contained 16.04 hours of footage comprising 344 mealtime events. Naïve adults labeled the five most obvious objects in still images from the recordings. Trained coders transcribed all speech in the target infant’s immediate vicinity, and naming moments were extracted from the transcripts. Concrete nouns on the MacArthur-Bates Communicative Development Inventory were included in the analysis if they occurred as an object and a name in the dataset.
The frequency distributions of visual objects and their names were both skewed with some very frequent categories and many more that are rare; however, the distribution of objects was more skewed than that of names (D = .4, p < .01). Indeed, only 9.49 hours of video contained any speech. The nature of visual objects was fundamentally different from that of object names, and the objects that were most frequently in view were not necessarily those with the most frequent object names. These measures were significantly but weakly correlated (rs(88) = .31, p < .01). When only the most frequent categories were considered, three frequency groups emerged: categories which are frequently seen but rarely named, frequently named but rarely seen, and both frequently seen and named (see Figure 1).
Co-occurrence was calculated liberally as the number of naming instances in a mealtime event in which both the object and its name ever occurred. For these frequent categories, co-occurrences were rare in the video (M = 0.71 instances/hour) (see Figure 2a). When only hours containing speech were considered, they were somewhat less rare (M = 1.19 instances/hour). On average, the categories co-occurred in 1.2% of the 344 mealtime events (Mdn = 0.9%) (see Figure 2b). For mealtimes in which the name occurred, however, the average proportion of mealtimes in which the object was also there was 64%. Thus, the base rate of co-occurrences was very low, but when an object name was said, it was likely that the object was present.
The rarity of co-occurrences of early-learned object-name pairs suggests that infants may be learning from minimal co-occurrence data as long as half of the pair (the object or its name) is highly frequent.

Authors

• Elizabeth Murphy Clerkin, Indiana University - Bloomington
  Presenting Author

• Linda B. Smith, Indiana University - Bloomington
  Non-Presenting Author