Ileana Grama

Posted on

Ileana Grama

Utrecht University


Learning and Generalizing Non-Adjacent Dependencies in 18-month-olds: A Mechanism for Language Acquisition?

Artificial Grammar Learning (AGL) experiments use miniature, artificially created ‘pseudo-languages’ to investigate infant and adult abilities to pick up patterns and regularities from the input (in an often unconscious manner). Successful learning of a miniature language in a laboratory setting is construed of as ‘proof of concept’ that these learning abilities might be involved in the process of early first language acquisition.

In this study I focus on a specific learning mechanism (the ability to pick up non-adjacent dependencies between a and b in an aXb configuration) and explore the claim that this ability could subserve language acquisition. AGL experiments have shown that both adults and infants can learn these a_b dependencies from aXb strings (Gomez, 2002; Gomez & Maye, 2005; Kerkhoff et al., 2015), but many have used aXb stimuli where the target elements a/b were very prosodically salient. In natural languages dependencies like The princess is always kissing frogs, on the other hand, dependencies are often instantiated between prosodically ‘reduced’ functional morphemes, and spanning more prosodically prominent lexical words/stems. Furthermore, in order to show mature mastery of dependencies like is_ing in language acquisition, children must learn to generalize this dependency to novel contexts: The general is harshly berating his troops.

In two separate experiments, using the Headturn Preference Procedure, I tested 18-month-old’s ability to (i) generalize the dependencies in aXb strings to novel aYb contexts, and (ii) to learn dependencies in aXb strings where a/b are prosodically ‘reduced’ (like functional morphemes), and the X element is prosodically salient (like lexical elements, see Grama, Kerkhoff & Wijnen 2016 for findings with adult participants). The procedure was similar to Gomez & Maye, 2005 and Kerkhoff et al., 2015.

There was no evidence that infants could generalize the dependencies to novel contexts, despite the extensive evidence that at 18 months, and in experimental setups very similar to the current one, learners can detect these dependencies in familiar contexts. This is consistent with findings like Soderstrom et al. (2007) should that, around this age, dependencies in natural languages are more easily tracked in familiar rather than novel contexts.

The results of the second experiment suggested that infants could track dependencies between prosodically ‘reduced’ elements, but that their phonological representations of these elements decayed quickly. Out of the 8 test trials, infants used the first test trial to refamiliarize themselves with these a_b dependencies. Across the subsequent trials 2-8, infants showed a significant preference for trials consistent with the dependencies from the first test trial. This is in line with Gómez, Bootzin & Nadel (2006), who argue that infants only retain an abstract representation of a_b dependencies from familiarization, and subsequently use the first trial to refamiliarize themselves with the item-specific ai_bi pairings. Furthermore, girls show a more marked learning effect than boys in this sample, although the effect of gender did not reach significance.

The results are discussed from the perspective of the role of non-adjacent dependency-learning (NADL) in language acquisition, differences between adult and infant learners, the role of abstract and item-specific information in NADL, as well as the advantages and methodological limitations of using the Headturn Preference Procedure.

Return to the list of speakers