Learning From Positive Evidence: The Acquisition Of Verb Argument Structure

This dissertation investigates how children acquire verb argument structure using positive evidence from the linguistic input. I discuss three case studies: the acquisition of raising and control verbs, the acquisition of causatives, and the acquisition of passives. When discussing these case studies, I address previous theories of verb argument structure learning that attempt to account for verb argument structure learning through indirect negative evidence (e.g., Pinker 1989), and I show that these approaches are inadequate in accounting for the developmental trajectory of the learner. Instead, I defend two learning models throughout the dissertation: the Sufficiency Principle (Yang 2016) and the Active Mapping Model. The Active Mapping Model of Language Acquisition illustrates how the learner uses conceptual and structural information in their language to form verb classes. In this model, there is no innate mapping between the conceptual and structural cues. Chapter 2 discusses the acquisition of raising and control verbs. Given three kinds of verbs, verbs that are purely raising, purely control, or verbs that can be both, the literature has foregrounded a learnability problem (Becker 2006). If some verbs can be both raising and control, then what prevents the learner from assuming that all verbs can take either structure? This learnability problem, known as the problem of overgeneralization (Baker 1979), then puts forth the question of how the learner retreats from this hypothesis to arrive at the adult grammar that also has pure raising and control verbs. Using the Sufficiency Principle, I show that the problem of overgeneralization does not arise, as the number of verbs that can be both raising and control do not meet the threshold of generalization determined by the Sufficiency Principle. I also argue against theories that propose indirect negative evidence to retreat from overgeneralization, and instead, I argue that raising and control verbs are learned from positive evidence in the input, which in this case is in the form of non-referential subjects. In Chapter 3, I discuss the acquisition of the causative alternation rule, which is a true case of overgeneralization, as seen by the errors made by children in their production data. In the acquisition of causatives, I demonstrate that the overgeneralization errors are predicted under the Active Mapping Model where the learner categorizes verbs into classes based on conceptual and structural cues. Given the learner\u27s vocabulary size and verb classes, the causative alternation rule is found to be productive when the input the learner receives is examined. Thus, under this learning model, the child errors are predicted. Moreover, using the Sufficiency Principle to determine the threshold of generalization, I show that the learner retreats from overgeneralization when their vocabulary size increases, as the rule is then no longer productive when the input is examined. In this chapter, I also test the Sufficiency Principle experimentally, and find support for it in the results obtained from child participants. In Chapter 4, I use the Sufficiency Principle and the Active Mapping Model to examine the developmental trajectory of children\u27s acquisition of passives. This chapter answers two questions: whether the passive construction is productive for the learner early on, and whether the asymmetry in the acquisition of actional and non-actional passives (e.g., Pinker et al. 1987) can be accounted for under the models of language learning assumed in this dissertation. Using the child production data and the input data, I show that the passive construction is productive in the input, and productive for the English-learning child. By examining Adam\u27s vocabulary and verb classes, the passive rule is found to be productive for Adam given the number of passives present in the input. Moreover, under the Active Mapping Model, I show that the asymmetry in the acquisition of passives is predicted. This dissertation investigates whether the learner\u27s developmental trajectory can be accounted for solely from the input data. In each of the cases discussed in the chapters of the dissertation, I demonstrate the role of the input in language acquisition. Furthermore, using the Active Mapping Model, I show that the innate linking between the verbs semantics and the syntactic structure is not only unnecessary, but also does not account for the full range of facts found in the child data