Uninversion error in English-speaking children’s wh-questions: Blame it on the bigrams?

The aim of the present study was to investigate whether and how English-speaking children’s uninversion errors with wh-questions (e.g., *Who he can draw; c.f., Who can he draw?) are influenced by the surface frequency of individual bigrams and trigrams in the input, as predicted by input-based approaches. Production methods were used to elicit nonsubject wh-questions from 67 children aged 3;1 to 4;8 (M=4;0, SD=4 months). No support was found for the preregistered prediction that children will produce more uninversion errors when those errors incorporate – in the Bigram 3 position – high-frequency bigrams from uninverted structures (e.g., *Who he can draw?) than lower-frequency bigrams from uninverted structures (e.g., *Who he can name?). Importantly, when testing this prediction, all other bigrams and unigrams (i.e., single words) are either identical (e.g., Who+he, he+can, he, can) or closely matched for frequency (e.g., draw and name [as verbs] are of approximately equal corpus frequency). However, a non-preregistered exploratory analysis found a facilitatory effect on correct-question production of the frequency of the second and third bigrams from inverted structures (e.g., can he…he draw), even after controlling for unigram frequency. This analysis also found that rates of uninversion error (e.g., *Who he can draw?) were higher when the first uninverted bigram (e.g., Who he…) is of higher frequency in the input. We conclude that while input-based accounts are correct to highlight the importance of n-gram input frequencies on rates of correct production versus uninversion error, it is unclear on current evidence which n-grams are driving errors and why. In particular, the special emphasis placed by some such accounts on n-grams at the left-edge of the utterance (e.g, Who can…) may be unwarranted.