Do Infants Learn from Isolated Words? An Ecological Study

ABSTRACT

Researchers disagree as to the importance for infant language learning of isolated words, which occur relatively rarely in input speech. Brent and Siskind (2001) showed that the first words infants produce are words their mothers used most frequently in isolation. Here we investigate the long-term effects of presentation mode on recognition memory for word forms. In two experiments we assess whether 12-month-old infants remember novel words presented in the home, over a three-week period: (i) in isolation or (ii) sentence-finally. When tested with word lists infants recognize words that had been presented in isolation, but not those that had been presented sentence-finally. They fail to recognise the trained words when tested with a segmentation task, regardless of presentation mode during the training. Our results indicate that the relatively small proportion of words produced in isolation in the input likely play a disproportionate role in the early period of language learning.

Acknowledgments

We would like to thank Mariam Dar for running many of the experiments and the families for taking part in this study. We also thank Daniel Swingley for his thoughtful and supportive feedback on earlier versions of this paper.

Notes

¹ Unlike Brent and Siskind (2001), Swingley and Humphrey (2017) found a strong effect of overall frequency; they explained this discrepancy as compared with Brent and Siskind’s results as being due to their having used log-transformed rather than the raw frequencies that Brent and Siskind used. Furthermore, Brent and Siskind used two outcome measures, both CDI and the children’s actual productions, as transcribed by the experimenters. In contrast, Swingley and Humphrey based their outcome measures on CDI data only, despite the misgivings they express as to the potential unreliability of maternal reports.

² This segmentation test was presented as an additional training phase in the lab. The children showed recognition of the words by listening longer to the sentences containing the word they were trained on in the home than to sentences containing a new word. It therefore served as a segmentation or recognition test, in addition to being a further training opportunity.

³ In a pilot experiment 18 English-learning monolinguals were tested at 12 months (mean = 0;11.18, sd = 2.8 days; 9 females). Data from five additional infants were discarded due to crying or parental interference (n = 3) and experimental error (infants too old at time of testing, n = 2). The looking times to Isolated (Mean = 5421 ms, sd = 2743) vs. Sentential (Mean = 4705 ms, sd = 2984) or Unfamiliar (Mean = 4758 ms, sd = 2426) words were not significantly different (Repeated measures ANOVA, F = 1.12, df = 2, p = .34). Effect size was medium (Partial Eta Squared = .062). The three-way comparison made interpretation difficult. We therefore ran the experiments reported here as pairwise comparisons.

⁴ The age for testing was based on previous studies (e.g., Hallé & de Boysson-Bardies, 1994; Vihman et al., 2004, 2007). In these studies infants showed recognition of untrained words (made familiar through exposure to naturalistic input in the home) when tested in the lab on word lists at age 11 months but not earlier (at 10 months), nor later (at 12 months, according to Vihman et al., 2007; although 12-month olds did succeed at the task in Hallé & de Boysson-Bardies, 1994). Since in the present study the children had to be trained with new words, making the task more challenging, we set the age of testing at 12 months.

⁵ The last child to be tested was given 3 Isolated passages in a row by mistake and the mistake was discovered only after the study had ended. We conducted the analyses both with and without his data; the pattern of results remained the same.

⁶ We also find no evidence that number of readings had any effect on the ability to recognize and segment the words, which makes us more confident that the finding of a negative correlation in Experiment 1 must be spurious.

Additional information

Funding

We are grateful to the UK Economic and Social Research Council (grant reference ES/J004952/1) and the British Academy (grant reference SG132286) for funding.