Fragile Long-Term Retention Following Statistical Learning in Infancy

• In Event: 1-153 - Retention of statistical learning

Integrative Statement

Saffran, Aslin, and Newport (1996) demonstrated that infants are able to segment fluent speech into words using transitional probabilities, or the statistical probability that one syllable will follow another. Since then, researchers have extended this line of research to segmentation of complex and natural languages. However, few have investigated long-term retention of words acquired through segmentation via transitional probabilities. To examine this issue, we introduced delays between learning and test. In both Experiments 1 and 2, 6.5-month-old infants listened to an artificial language containing four disyllabic words for approximately 7 minutes. The language contained no pauses, intonation, or pitch changes in the stream. Therefore, the only cue to the location of the word boundaries was the transitional probabilities between syllables. In Experiment 1, half (n=16) of the infants listened to the language then took their typical morning nap. These infants were tested 15 minutes after waking up naturally. Infants in the wakefulness group (n=16) were yoked controls who listened to the language then stayed awake for an equal amount of time as one of the napping infants slept, plus 15 minutes. The average delay between learning and test in both conditions was 80 minutes. We tested all infants using the Head Turn Preference Procedure and measured difference in looking time between words (e.g. dapu, dobi) and part-words (e.g. diti, bugo) from the language. The part-words consisted of two syllables that spanned a word boundary, so while the syllables occurred in sequence throughout the language, they did not make up a legal word. Infants who stayed awake after exposure showed no difference in looking time to words (MBlock1=21.39s; MBlock2= 9.23s) vs. part-words (MBlock1=21.6s; MBlock2= 9.85s) in either block of testing. Infants who slept after learning showed a fragile memory for the words from the language, and we observed a significant two-way interaction indicating that these infants looked longer to the part-words (MBlock1=27.9s) vs. the words (MBlock1=19.32s) but only in the first block of testing (F(1,15)=7.44, p=0.016). This may indicate the fragility of infants’ memory at this age. We also observed a correlation between frontal slow-wave activity and retention in Block 1 (r=0.507, p<0.01) suggesting a potential role of sleep in memory retention early in life. In Experiment 2, we considered the notion that testing infants 80 minutes after exposure interfered with consolidation of the original memory, and we questioned whether overnight sleep stabilizes infants’ memory for the segmented words. Therefore, all infants napped (n=15) in the 4-hour window after exposure to the artificial language and were tested using the Head Turn Preference 24 hours later. The familiarization and test stimuli in this Experiment were identical to that in Experiment 1. Overall, infants did not distinguish words (M=10.13s) from part-words (M=12.54s, t=-0.97, p=0.35), suggesting they did not retain the words over the 24-hour period after learning. Moreover, overnight sleep does not seem to offer additional benefit to long-term memory retention in comparison to a single nap.

Authors

• Lucia Sweeney, University of Arizona
  Presenting Author

• Katharine Simon, The University of Arizona
  Non-Presenting Author

• Rebecca Gomez, The University of Arizona
  Non-Presenting Author