Listening Through Voices: Infant Statistical Word Segmentation Across Multiple Speakers

Graf Estes, K. & Lew-Williams, C. (2015). Listening Through Voices: Infant Statistical Word Segmentation Across Multiple Speakers. Developmental Psychology, 51(11), 1517-1528.

Statistical language learning refers to learning aspects of language based on the patterns of the language. For example, sounds within a word are more likely to occur together than are sounds that cross word boundaries. These patterns might help infants learn to distinguish words from word boundaries. In this study, Graf Estes and Lew-Williams focused on the statistical learning in infants.

In a real environment, infants are exposed to multiple voices, each with a unique speaking style and rate. Graf Estes and Lew-Williams used multiple voices in a monotone speech stream to mimic this natural environment. Infants from this study listened to a language produced by eight different voices that changed frequently for six minutes. Results suggested that infants were able to learn the artificial language when tested with a common voice or a novel voice, also suggesting that infants can generalize representations.

In the second series of experiments, infants listened to two different voices. The researchers suggested that the use of two dominant voices may better mimic the infant’s environment (e.g. two parents). Infants, however, failed to display signs of language acquisition. Graf Estes and Lew-Williams suggested two possible explanations: Infants were able to learn both the words and part-words, thus showing no discrimination during test phase, or the used of two voices resulted in a focus on discerning the voices, thus inhibiting language acquisition.

Although the mixed results require further investigation, the findings highlight the efficiency of learning in the context of variability (see also, Plante et al. 2014).

Blogger: Hosung (Joel) Kang is a neuroscience student completing his undergraduate thesis project in the Language and Working Memory Lab.

Constraints on Implicit Learning of Grammatical Form-Meaning Connections

Leung, J. H., & Williams, J. N. (2012). Constraints on Implicit Learning of Grammatical Form‐Meaning Connections. Language Learning, 62(2), 634-662.

Implicit learning is learning that takes place without intention or conscious awareness. Humans are able to extract and learn from patterns in the environment, without any realization of this learning.  In this study, Leung and Williams focused on the implicit learning of grammatical form-meaning connections – an area of implicit learning research where there is still much to investigate. 

A form-meaning connection takes place when the assignment of a meaning to an unfamiliar word is made.  In the case of this study, Leung and Williams used an artificial language with four determiner-like words.  These determiner words were used in front of a noun to encode whether the noun was near or far, or animate or inanimate.  Participants in this study were told all four of the novel determiners along with the near or far rule, however, they were not told about the animate or inanimate rule. 

The researchers conducted two experiments in which participants were shown two side-by-side pictures on a computer screen.  Participants were asked to click the correct image after hearing the corresponding phrase.  Each phrase involved one of the four determiner words, followed by the noun of one of the picture on the screen.  After training, participants completed a test phase in which participants who had learned the implicit rule for animate or inanimate markers would be at some advantage in terms of their reaction time. Results provided implicit learning of the animate/inanimate rule in this experiment. In the second experiment, involving implicit learning of relative size, no learning was observed. Leung and Williams posited that this might have been because some meanings are more susceptible to implicit learning than others, based on the characteristics of the language being learned.

Although the results provide some evidence of implicit learning of form-meaning connections, it is clear that this method of learning is slower than explicitly teaching the rule.    

Blogger: Alisha Johnson; Alisha is an undergraduate these student in the Language and Working Memory Lab