Cognitive bias overrides syntactic bootstrapping in novel verb learning

Acquiring novel verb meanings is challenging because events in the world are inherently multi-interpretable (Quine, 1960). While the acquisition of motion verbs has been well studied (e.g. Allen et al., 2007), only few studies investigated how lexicalization biases are acquired when children learn the meaning of change-of-state verbs (Bunger et al., 2016; Papafragou et al., 2002). Focusing on novel change-of-state verbs, our study investigates the role of syntactic bootstrapping (Gleitman, 1990; Naigles, 1996) in the development of lexicalization biases in learners of Dutch and English, both claimed to be manner languages (Talmy, 1985).Dutch preschoolers (N=39, MAge=3;11) and older English children (N=51, MAge=7;08) participated in a forced-choice task using an adjusted version of Wagner’s (2010) method. Dutch adults (N=8) served as controls. During the training phase participants watched animations of an agent performing an action on an object causing a change-of-state in the object, while they heard a novel verb presented in either an intransitive or transitive sentence frame. At test participants saw two animations - one showing a similar manner, the other a similar result - and had to choose which one depicted the novel verb (Figure 1). If sentence frame plays a role in determining manner/result readings, it was expected that the intransitive condition would trigger mostly manner readings and the transitive condition result readings (Wagner, 2010).A two-way repeated-measures ANOVA showed that Dutch adults (F(1,10)=5.07, p<.05) and English children (F(1,49)=34.8, p<.0001) were strongly result biased. None of the participants were manner biased. A logistic mixed-model revealed no effect of sentence frame: the result bias occurred equally often in both conditions (Figure 2). Furthermore, there was a strong positive relationship between age and the degree to which participants were result biased (b=.03, SE=.006, p<.0001). A distribution analysis shows that this effect was not due to a lack of task understanding in the younger children. Moreover, a stimuli validation study showed that the animations span the entire range with respect to the bias they induce.We discuss three main findings. 1) The result bias in our participants is surprising given that both languages are claimed to be manner languages. 2) The lack of effect of sentence frame is not in line with Wagner (2010). We conclude, contra Wagner, that syntactic bootstrapping does not affect this bias, and argue that this was not to be expected given that there are no strong links between transitivity and resultativity: both transitive and intransitive sentences can refer to resultative and non-resultative events (Borer, 2004). 3) Seeing the effect of age we tentatively suggest that this result bias emerges over the course of development. This will need to be confirmed by testing a wider range of ages in each language.To conclude, we propose that a result bias in verb learning arises because learners are intrinsically goal-directed (Carpenter et al., 2005, Gergely et al., 2002), which leads them to represent multi-interpretable scenes in terms of event outcomes, not manners of actions.