Acquisition of Desires before Beliefs: A Computional Investigation

Proceedings of the Seventeenth Conference on Computational Natural Language Learning, pages 231–240, Sofia, Bulgaria, August 8-9 2013. c⃝2013 Association for Computational Linguistics Acquisition of Desires before Beliefs: A Computational Investigation Libby Barak, Afsaneh Fazly, and Suzanne Stevenson Department of Computer Science University of Toronto Toronto, Canada {libbyb,afsaneh,suzanne}@cs.toronto.edu Abstract The acquisition of Belief verbs lags be- hind the acquisition of Desire verbs in children. Some psycholinguistic theo- ries attribute this lag to conceptual differ- ences between the two classes, while oth- ers suggest that syntactic differences are responsible. Through computational ex- periments, we show that a probabilistic verb learning model exhibits the pattern of acquisition, even though there is no dif- ference in the model in the difficulty of the semantic or syntactic properties of Be- lief vs. Desire verbs. Our results point to the distributional properties of various verb classes as a potentially important, and heretofore unexplored, factor in the ob- served developmental lag of Belief verbs. 1 Introduction Psycholinguistic studies have shown great inter- est in the learning of Mental State Verbs (MSVs), such as think and want, given the various cogni- tive and linguistic challenges in their acquisition. MSVs refer to an entity’s inner states, such as thoughts and wishes, which the language learner must be able to perceive and conceptualize appro- priately. Moreover, such verbs often appear in a Sentential Complement (SC) construction, which is complex for children because of the embedded clause. Despite some shared properties, MSVs are a heterogeneous group, with different types of verbs exhibiting different developmental patterns. Specifically, a wealth of research shows that chil- dren produce Desire verbs, such as want and wish, earlier than Belief verbs, such as think and know (Shatz et al., 1983; Bartsch and Wellman, 1995; Asplin, 2002; Perner et al., 2003; de Vil- liers, 2005; Papafragou et al., 2007; Pascual et al., 2008). Some explanations for this pattern posit that differences in the syntactic usages of Desire and Belief verbs underlie the observed develop- mental lag of the latter (de Villiers, 2005; Pas- cual et al., 2008). In particular, Desire verbs oc- cur mostly with an infinitival SC (as in I want (her) to leave), while Belief verbs occur mostly with a finite SC (a full tensed embedded clause, as in I think (that) she left). Notably, infiniti- vals appear earlier than finite SCs in the speech of young children (Bloom et al., 1984, 1989). Others suggest that Desire verbs are conceptu- ally simpler (Bartsch and Wellman, 1995) or prag- matically/communicatively more salient (Perner, 1988; Fodor, 1992; Perner et al., 2003). Propo- nents of the conceptual and pragmatic accounts ar- gue that syntax alone cannot explain the delay in the acquisition of Belief verbs, because children use finite SCs with verbs of Communication (e.g., say) and Perception (e.g., see) long before they use them with Belief verbs (Bartsch and Wellman, 1995). We use a computational model of verb argu- ment structure acquisition to shed light on the fac- tors that might be responsible for the developmen- tal …

Content truncated to 500 words