A Changing Role for Transitional Probabilities in Word Learning Across Development

• In Event: How do the statistics of the environment shape the infant mind across domains?

Abstract

Speech unfolds continuously, making word segmentation and language learning a challenge. Infants readily track transitional probabilities (TPs), which tend to dip at word boundaries, providing them with information about which syllable sequences form a word, and which span word boundaries (Saffran et al., 1996). TPs also play an important, facilitative role in learning word meanings, at least in 17-month-olds, such that high-TP (HTP) Italian words are better learned as labels than low-TP (LTP) words (Hay et al., 2011). This effect appears to reflect an advantage for mapping HTP words relative to other sequences, rather than resistance to mapping LTP words, because 17-month-olds also fail to learn unfamiliar Italian words as labels in the same task. However, remarkably little is known about the role of TPs in word learning as infants get older and develop stronger mapping skills. In fact, by adulthood, the HTP-advantage disappears (Mirman et al., 2008). Here, we tested whether infants’ HTP advantage diminishes as lexical development progresses, and infants become better at forming word-referent mappings.

We familiarized two groups of English-learning 24-month-olds (N=64) to naturally spoken Italian sentences containing equally frequent HTP and LTP words. One group (Intact-TPs condition) was tested on how well they mapped the HTP and LTP words from their familiarization (i.e., casa, bici, fuga, and melo). Infants in a second group (Violated-TPs condition) were familiarized to the identical Italian sentences and tested on how well they mapped sequences that involved violations of the HTP and LTP words. For example, if HTP words were CAsa and BIci, violated-HTP words were CAci and BIsa (TP=0). Comparing performance on HTP and LTP words allows us to test whether 24-month-olds show an HTP advantage for familiar sequences, like younger infants. Comparing performance on violated-HTP and violated-LTP words to their HTP and LTP counterparts allows us to test whether infants at this age learn HTP and/or LTP sequences better than unfamiliar ones.

We found that infants learned both HTP and LTP words (Figure 1), and that they did not demonstrate an HTP advantage at the group level, t(31)=1.29, p=.208, d=.23. Interestingly, there were individual differences in the degree to which HTP words were learned better than LTP words, such that 24-month-olds infants with smaller vocabularies, like younger infants, were more likely to show an HTP-advantage (Figure 2). We also found that violating the statistics of the LTP words had little impact on performance (Figure 1). This suggests that infants do not necessarily learn familiar sequences better than unfamiliar ones. However, violating the statistics of HTP labels significantly reduced learning (HTP vs violated-HTP, t(62)=2.96, p=.004, d=.74; see Figure 2).

In sum, 24-month-olds can learn both HTP and LTP words, but only infants with smaller native-language vocabularies show an HTP-advantage. HTP words are nonetheless represented differently than LTP words, as infants only resist mapping sequences that violate HTPs. These results suggest that sensitivity to TPs plays an important but changing role across development, initially facilitating mapping HTP words, and eventually helping infants avoid mapping violations of good word forms.

Authors

• Jill Lany
  Presenting Author

• Ferhat Karaman

• Jessica Sari Fleming Hay, University of Tennessee - Knoxville