Poster #161 - Statistical Learning in Infancy - Transitional Probabilities vs. Repetitions

• In Event: 1-115 - Poster Session 03
  In Poster Session: PS 03 Section - Perceptual, Sensory, Motor

Integrative Statement

Humans receive a myriad of sensory inputs and a crucial task during infancy is to make sense of this input. One prominent learning mechanism available to infants is statistical learning, the extraction of associations among individual items in a sequence or array. For example, 8-month-olds can extract information about the sequential statistics of syllables in a continuous speech stream and use this information to segment the speech stream into its constituent units (Saffran et al., 1996). Moreover, infants as young as newborns can discriminate statistically structured from random visual sequences (Bulf et al., 2011; Kirkham et al., 2002). Performance in these studies was thought to stem from infants’ computation of transitional probabilities (TPs) between items in sequence (TPs are the conditional probabilities of the next item given the previous item). However, it is not clear from these studies whether infants younger than 8 months can learn TPs and use them to determine boundaries between units in an otherwise undifferentiated sequence. It is possible that unit frequency and unit repetitions in sequence can also provide information for unit boundaries. The goal of the present study was to examine TP- vs. repetition-based learning in young infants as they viewed sequences of audiovisual shapes.

We habituated 5-month-old infants to statistically predictable sequences of three two-item pairs of synchronous auditory and visual stimuli (6 unique items). Infants in the low repetition (Low Rep) condition (n = 28) viewed pairs defined by TPs between items (TP = 1.0). TPs between pairs were .50; repetitions were disallowed. For infants in the high repetition (High Rep) condition (n = 26), one pair was allowed to repeat up to three times. There were no pauses between items in sequences, and pair boundaries could only be determined by TPs or repetitions. After habituation infants viewed test sequences consisting of two segmented pairs of items. One pair was a unit from the habituation sequence (high TP or high repeating), and the other was composed of two items that were sometimes seen in sequence during habituation (low TP or infrequently appearing). We reasoned that if infants learned pairs from either TPs or repetitions, they would look longer at the novel test pair (low TP in the Low Rep condition; infrequent test pair in the High Rep condition).

A Condition x Test Display (familiar or novel) x Trial Pair mixed ANOVA on log-transformed looking times revealed a marginally significant Condition x Test Display interaction, F(1, 52) = 3.99, p = .051, and no other reliable effects. Infants in the High Rep condition looked longer at novel stimuli than infants in the Low Rep condition, t(52) = 1.95, p = 0.057, but looking to familiar stimuli did not differ between groups, t(52) = .70, ns (see Figure 1). These results suggest that young infants may use repetition as a way of segmenting streams of information more readily than TPs. Ultimately, this result clarifies the learning mechanisms available to young infants and it emphasizes the importance of repetition for future statistical learning research.

Authors

• Shannon Brady, University of California, Los Angeles (UCLA)
  Presenting Author

• Scott Johnson, University of California, Los Angeles (UCLA)
  Non-Presenting Author