Sentence first, arguments after: Mechanisms of morphosyntax acquisition

Abstract

Natural languages contain complex grammatical patterns. For example, in German, finite verbs occur second in main clauses while non-finite verbs occur last, as in 'dein Bruder möchte in den Zoo gehen' (“Your brother wants to go to the zoo”). Children easily acquire this type of morphosyntactic contingency (Poeppel & Wexler, 1993; Deprez & Pierce, 1994). There is extensive debate in the literature over the nature of children’s linguistic representations, but there are considerably fewer mechanistic ideas about how knowledge is actually acquired. Regarding German, one approach might be to learn the position of prosodically prominent open-class words (“verbs go 2nd or last”) and then fill in the morphological details. Alternatively, one could work in the opposite direction, learning the position of closed-class morphemes (“-te goes 2nd and -en goes last”) and fitting open-class items into the resulting structure. This second approach is counter-intuitive, but I will argue that it is the one learners take.Previous research suggests that learners focus distributional analysis on closed-class items because of their distinctive perceptual properties (Braine, 1963; Morgan, Meier, & Newport, 1987; Shi, Werker & Morgan, 1999; Valian & Coulson, 1988). The Anchoring Hypothesis (Valian & Coulson, 1988) posits that, because these items tend to occur at grammatically important points in the sentence (e.g., phrase edges), focusing on them helps learners acquire grammatical structure. Here I ask how learners use closed-class items to acquire complex morphosyntactic patterns such as the verb form/position contingency in German. Experiments 1-4 refute concerns that morphosyntactic contingencies like those in German are too complex to learn distributionally. Experiments 5-8 explore the mechanisms underlying learning, showing that adults and children analyze closed-class items as predictive of the presence and position of open-class items, but not the reverse. In these experiments, subtle mathematical distinctions in learners’ input had significant effects on learning, illuminating the biased computations underlying anchored distributional analysis. Taken together, results suggest that learners organize knowledge of language patterns relative to a small set of closed-class items—just as patterns are represented in modern syntactic theory (Rizzi & Cinque, 2016).