Fig. 2

Experiment 1 results: the morpheme, digit, and class error rate in each condition. The best performance was in condition B. In this condition, each stimulus included 2 grammatical segments with 3–4 words in each; this seems to provide the optimal balance between the number of chunks and the chunk size. The performance was poorer in conditions C and D, which seem to support too little chunking as the stimuli have short grammatical segments; and in condition A, which seems to encourage too much chunking as the stimulus has a single long grammatical segment. This clear syntactic chunking effect shows that the participants represented the number’s syntactic structure. In the statistics table, Δ, χ², and p refer to the effect size and significance of the Condition factor in the linear mixed model described in the “Statistical analysis” section