It has been found (Lovric et al., 2000) that the inclusion of 'od', a non-thematic preposition similar to English 'of', significantly lowers the preferred attachment of a relative clause (RC) in the ambiguous construction N1-(od)-N2[Gen]-RC in silent reading of Croatian. A significant effect of RC-length was also observed, with long RCs attaching higher than short RCs (both with and without 'od'). Lovric et al. noted that these effects cannot be explained by non-phonological accounts that have been proposed to date, since neither 'od' nor RC-length has any semantic, discourse or Gricean effects on the sentence.
The Implicit Prosody Hypothesis (Fodor, 1998; Quinn et al., 2000) suggests that attachment preferences in silent reading are sensitive to a language's default patterns of prosodic phrasing, projected onto sentences by the reader. As a way of establishing default prosody, the present study seeks regularities in the overt prosody of Croatian N1-(od)-N2-RC complexes. Their relation to the attachment data for silent reading is then assessed.
The materials were number-disambiguated sentences, forcing attachment high or low; the design factorially combined Preposition ('od' present vs. absent) and RC-Length (long vs. short), as in (1). Subjects read each sentence silently for meaning first, and then aloud for recording. In the acoustic analysis, length measurements were taken for N1 and N2 for all sentence versions.
The results showed lengthening of N1 when 'od' was present (113 msec on average, p<.01), suggesting that 'od' induces a prosodic break before it. There was also lengthening of N2 before long RCs compared with short RCs (110 msec on average, p<.01), suggesting a higher likelihood of a prosodic break before long RCs.
These results support a prosodic approach. Under the Implicit Prosody Hypothesis, prosodic breaks projected in silent reading encourage non-local syntactic attachment of the immediately following constituent, just as in overt prosody (see Maynell, 1999). The presence of 'od' affects the overt prosody of Croatian complex NPs in such a way as to separate N1 from N2, favoring a grouping of N2 with the RC; this would bias toward lower attachment, as was observed by Lovric et al.. As for the length-effect, for long RCs, unlike short RCs, a prosodic break separating N2 from the RC favors higher attachment, also as observed. The facts of overt prosodic phrasing, thus, explain the attachment findings in silent reading of Croatian.
References
Fodor, J. D. 1998. Learning to parse? Journal of Psycholinguistic Research, 27, 2, 285-319.
Lovric, N., Bradley, D. & Fodor, J. D. 2000. RC attachment in Croatian with and without preposition. Poster presented at AMLaP Conference, Leiden.
Maynell, L. A. 1999. Effect of pitch accent placement on resolving relative clause ambiguity in English. Poster presented at the 12th Annual CUNY Conference on Human Sentence Processing, New York.
Quinn, D., Abdelghany, H. & Fodor, J.D. 2000. More evidence of implicit prosody in silent reading: French, English and Arabic relative clauses. Poster presented at the 13th Annual CUNY Conference on Human Sentence Processing, San Diego.