Syntactic bootstrapping attitude verbs despite impoverished morphosyntax

ABSTRACT

Attitude verbs like think and want describe mental states (belief and desire) that lack reliable physical correlates that could help children learn their meanings. Nevertheless, children succeed in doing so. For this reason, attitude verbs have been a parade case for syntactic bootstrapping. We assess a recent syntactic bootstrapping hypothesis, in which children assign belief semantics to verbs whose complement clauses morphosyntactically resemble the declarative main clauses of their language, while assigning desire semantics to verbs whose complement clauses do not. This hypothesis, building on the cross-linguistic generalization that belief complements have the morphosyntactic hallmarks of declarative main clauses, has been elaborated for languages with relatively rich morphosyntax. This article looks at Mandarin Chinese, whose null arguments and impoverished morphology mean that the differences necessary for syntactic bootstrapping might be much harder to detect. Our corpus analysis, however, shows that Mandarin belief complements have the profile of declarative main clauses, while desire complements do not. We also show that a computational implementation of this hypothesis can learn the right semantic contrasts between Mandarin and English belief and desire verbs, using morphosyntactic features in child-ambient speech. These results provide novel cross-linguistic support for this syntactic bootstrapping hypothesis.

Acknowledgement

Special thanks to Yuhan Zhang for research assistance. For advice, assistance, and feedback, we are also grateful to Rachel Dudley, Philip Resnik, the University of Maryland’s Project on Children’s Language Learning, and the audiences at MACSIM 6, BUCLD 42 and Selection Fest, where earlier versions of this work were presented. This material is based upon work supported by the National Science Foundation under Grant No. 1449815.

Declaration of interest statement

The authors have no conflicts of interest to report.

Notes

1 Abbreviations used in glosses: IND: indicative; PART: particle; PROG: progressive aspect; SBJ: subjunctive; 3s: third person singular pronoun.

2 As Hacquard & Lidz note, beyond providing a principled link between subcategorization properties and semantics, this pragmatically rooted syntactic bootstrapping hypothesis can help make sense of a set of experimental results in which young children incorrectly treat think as being incompatible with false beliefs (for discussion of English and German, see Perner et al. 2003; but see Lewis, Hacquard & Lidz 2012, 2017 for results showing adult-like behavior in English-learning children with think). For illustration, consider the following scenario: Children are told that Dora is playing, but Boots mistakenly believes that she is asleep. In this context, young children usually (incorrectly) judge a statement like “Boots thinks that Dora is asleep” to be false. This fact is often linked to the development of a theory of mind in young children or children’s (in)ability to encode false beliefs (e.g., de Villiers 2005; de Villiers & de Villiers 2000). However, Hacquard & Lidz argue that this un-adult-like behavior arises because children take the statement to be an indirect assertion that Dora is asleep, which is inconsistent with the fact that Dora is actually playing.

3 The pronoun ni ‘you’ in (10b) is ambiguous between a subject and a vocative reading. It can be disambiguated by adding a proper name as a vocative: e.g., Zhangsan, ni chi shuiguo! ‘You eat fruit, Zhangsan!’

4 For the corpus analysis, the following items and their negated forms were defined as modal auxiliaries and adverbs (i) and aspect markers (ii).

(i)   a.   Epistemics: keneng “might”, yiding “must”

   b.   Roots: (bi)xu, dei, yiding “must”; ken “be willing to”; ke(yi) “can, be allowed to”; hui, neng(gou) “able to”;   (ying)gai, ying “should”

   c.   Future: hui, jiang

   d.   Others: shi (focus)

(ii)  a.   Experiential: -guo

   b.   Progressive: (zheng)zai

   c.  Negated perfective: mei(you)

5 We also tested an alternative but mathematically related version of this penalty on the English data set. For each verb, we first calculate two Kullback-Leibler divergence measures, D(Belief_v||Not-Desire_v) and D(Desire_v||Not-Belief_v). The penalty is calculated as the mean of the two measures. For space reasons, we do not report results obtained with this alternative penalty, except to note that they are very similar to the English results reported below in Section 5.4.