Why learners privilege word-order over case-marking: A cross-linguistic meta-analysis, new data from Estonian, Finnish and Polish, and a discriminative learning model

Language acquisition is one of the crowning achievements of our species; though a longstanding and unresolved question is why many learners struggle with a particular core and fundamental sentence type. In English, a two-participant sentence like The dog chased the cat can mean only one thing. But in many languages worldwide, the meaning of the sentence can be flipped (e.g., to ‘the cat chased the dog’) even while the word-order stays the same, by the use of case-marked nouns (e.g., cat+NOMINATIVE, dog+ACCUSATIVE). First (Study 1), we confirm using meta-analysis that, across 21 languages, children show a significant disadvantage for non-canonical (“flipped”) versus canonical (“unflipped”) sentences (Cohen’s d=0.69, SE=0.09). Second (Study 2), we demonstrate with a new comprehension study of 3- to 6-year-old native learners of Estonian (N=87), Finnish (N=83) and Polish (N=82) that this effect is not due to (1) lack of knowledge of individual inflected words, (2) lack of linguistic context, (3) use of still pictures as opposed to animations, or (4) a focus on relatively young children. Third (Study 3), we show that this effect falls naturally out of a simple discriminative-learning model, in which the cue of case-marking struggles to compete with the more frequently available cue of word-order to predict which noun (e.g., dog, cat) is the AGENT (e.g., the chaser), and which the PATIENT (e.g., the one chased). Together, these findings demonstrate that languages which, on the surface, appear to be very different are learned in a similar way: by discriminating cues to meaning.