Implicit Prosody in Silent Reading: Relative Clause Attachment in Croatian

When a relative clause (RC) follows two nouns (N1, N2) in a complex noun phrase such as that contained in the example English sentence below, the preferred interpretation has been found to differ across languages. (1) Someone shot the servant [N1] of the actress [N2] who was on the balcony [RC]. In some languages (e.g., English), readers preferentially interpret the RC as attaching to (i.e., modifying) N2. In other languages (e.g., Spanish), there is a preference for attachment to N1. This cross-linguistic variation is the only known counterevidence to the claim that the human sentence processing routines are universal, and could therefore be innate. Five major explanations have been proposed in the literature to account for cross-linguistic differences in RC-attachment: the Construal/Gricean account, Attachment-Binding Dualism, the Predicate Proximity/Recency model, the Tuning Hypothesis, and the Implicit Prosody Hypothesis (IPH). This thesis examines RC-attachment preferences in Croatian. The data reported address four of the above proposals: they argue against the first three and support the fifth, the IPH. (The data do not bear on the Tuning Hypothesis, which holds that there is no principled parsing explanation for the cross-linguistic facts.) The IPH claims that a default prosodic pattern, which may differ across languages, is mentally computed during silent reading, influencing syntactic attachment decisions. Croatian has a distinctive pattern of prosodic phrasing in that in sentences comparable to (1), it favors a prosodic break before a long RC as compared with a short RC, and a prosodic break before the preposition in the complex NP construction as compared with the non-prepositional variant of the same construction. (Both variants of the construction, prepositional and non-prepositional, are acceptable in Croatian.) The experimental findings reported here document these prosodic characteristics, and show that they are reflected in syntactic attachment preferences in silent reading. This evidence is important because if the IPH is correct, then cross-language RC-attachment differences are attributable to independently demonstrable differences in the prosodic principles in the grammar. That is, they fall into line with other observed parsing preferences in that they do not reflect non-universal processing routines