On the Prosody/Lexicon Interface in Learning Word Order
Recent literature in developmental psycholinguistics has emphasized the role of prosodic bootstrapping in discovering syntactic constituents and word order regularities in early grammar.1 One of the puzzles this theoretical paradigm leaves open is the “perception (or comprehension) prior to production” maxim.2 In many cases a puzzling discrepancy between the child’s perceptive capacity and speech production is observed which goes beyond the well-known phenomenon of “truncated grammatical morphemes”. It is, for instance, by no means clear why children are capable of establishing trochees as a prosodic unit for the purpose of speech segmentation at the age of 7 months, but fail to establish the trochaic template until late in the second year at the level of production.3 The present paper addresses the problem of “perception prior to production”. It focuses on the relationship between access to prosodic information in the input for the purpose of discovering the rule of object placement in early German and the application of the same rule in speech production. A careful, computer-aided evaluation of the data from both normally developing and language impaired children reveals a surprising mismatching between the syntactic and the prosodic data. While all children target-consistently apply the OV rule of object placement at the syntactic level from early on, they all fail to apply the prosodic rule of relative prominence within the same phrases in speech production. <br /> Assuming the Rhythmic Activation Principle of Nespor et al. (1996) and Guasti et al. (this volume), we will argue that this data reflect a genuine discrepancy between perception and production. We will claim that this discrepancy is not simply amenable to a performance-motivated lag, but rather follows in a predictable way from learnability constraints such as the “Avoid Irreversible Wrong Decisions” maxim. <br /> The paper is organized as follows. Section 1 introduces the learning algorithm, focusing on the prosody/syntax mapping account of Nespor et al. (1996) and Guasti et al. (this volume). Section 2 is an analysis of the earliest object verb constructions in German. The data base consists of 7 corpora (3 normally developing children and 4 language impaired children). Section 3 proposes an account of the discrepancy between syntax and prosody in speech production in terms of learnability constraints.