Artificial grammar learning in Williams syndrome and in typical development: The role of rules, familiarity, and prosodic cues.

Stojanovik, V., Zimmerer, V., Setter, J., Hudson, K., Poyraz-Bilgin, I., & Saddy, D. (2017). Artificial grammar learning in Williams syndrome and in typical development: The role of rules, familiarity, and prosodic cues. Applied Psycholinguistics, 1-27.

I heard about Williams syndrome (WS) for the first time when I was a linguistics student in Düsseldorf. The genetic disorder was interesting for one perceived dissociation: People with WS, so the view, had typical language capacities, but impairments of general cognitive abilities, demonstrating that the language system was independent, modular in the sense of (early) Chomsky and Jerry Fodor. Today we know this to be wrong. The language in WS may appear typical at a quick glance, but there are production as well as comprehension deficits, especially at a grammatical level. This study, led by Vesna Stojanovik at the University of Reading, aimed to understand what underlies the language profile in WS.

The study used an artificial grammar learning paradigm. Because of my PhD project I am quite familiar with these experiments. In artificial grammar learning, the researcher designs a language, exposes participants to it and then tests whether they successfully distinguish grammatical and ungrammatical "sentences". It is an attempt to study language acquisition and processing under very controlled conditions. Most of these experiments are very short (less than an hour), so their insights are mostly limited to a first contact with the artificial language. However, because researchers can manipulate so many factors, the paradigm can reveal language processing biases in different populations.

Here's what I think is the most important finding in this study: In typically developing children, we saw a processing bias towards familiarity in the younger subgroup (mean age 5 years). Children judged artificial sentences (for example, "rana subi pafil") as correct when they had heard them, or very similar sentences, before. In the older subgroup (mean age almost 9 years) there was more awareness of grammatical rules. Children judged sentences as correct when they agreed with the rules of the language, regardless of whether the exact sentence, or a similar sentence, was presented before. My interpretation of this result falls in line with what people like Elena Lieven or Michael Tomasello claim, namely that children start with very conservative, item-based grammatical knowledge, and only through maturation and exposure become more aware of the rules that allow language to be so creative. According to our findings, children aged nine are so accustomed to this abstract linguistic knowledge that they approach a novel language with such abstraction in mind. In contrast, in our WS sample, which on average was about 9.5 years, we only found evidence of familiarity-based learning. The "switch to rules" had not taken place.

Findings like these may explain why language in WS is similar, but different. People with WS may be good at learning sentence chunks and exemplars, but have difficulties when sentence structures require more abstract knowledge. Complex constructions, which other children learn using their "knack" for abstraction, are approximated with less sufficient cognitive tools.

I contributed to this paper by helping with data analysis and putting results in the context of the paradigm. This was the first time I got quite intimate with data I had not collected, from an experiment I had not designed. It's tricky. But I was fascinated with the methods, as Vesna and her colleagues came up with a very child-friendly artificial grammar learning experiment which used wizards casting spells as a metaphor.

The study has some issues with power. I am fairly confident about results from the typically developing group, but WS is a rare condition, and we wish we'd had bigger numbers here.