Constituent length has been shown to influence syntactic decisions in a number of recent studies. For example, in a series of eye-tracking experiments using N1+de+N2+RC constructions, Pynte and Colonna (1998, 2000) found a high attachment preference for long RCs and a low attachment preference for short RCs. This result can be accounted for in terms of "silent prosody" (silent readers would be submitted to the prosody-motivated "same-size sister constituent constraint", see Janet Fodor, 1998).
Interestingly, a similar relation between length and attachment site was found in the new annotated corpus for French developed at Paris 7 University (1 Million words from the French newspaper le Monde, Abeillé & al, 2001). In this corpus, 52% RCs attach to N1 in contexts where another N is a structural candidate, and the proportion is reverse (43%)for short RCs (length less than 6 words). Although consistent with the notion of silent prosody (presumably prosodic constraints also apply to writing), this result suggests an alternative explanation, in line with experience-based models of sentence processing.
The relevance of both prosody-based and experience-based approaches was confirmed by a new corpus search carried out on a series of 932 "V+NP+PP" sequences. Whereas long PPs were found to be preferentially attached to VPs (as verb complements), "short" PPs were found to be preferentially attached to NPs (as noun complements). (Mean length = 6.31 words for verb complements vs. 5.09 words for noun complements). The possible influence of length constraints upon constituent order is currently under investigation. For example, a proportion of short verb-complement PPs can be expected to appear before the complement NP.
As far as comprehension is concerned, the possible interaction of constituent length and constituent order was examined in a questionnaire experiment conducted with a set of sentences similar to (1) and (2). The ambiguous PP "de la course" was preferentially interpreted as a verb complement in (1) (the verb "eliminer" is ditransitive) and as a noun complement in (2). Note that, in (2), interpreting the PP as a verb complement would result in a long+short sequence of complements. These preliminary results should be confirmed by an eye-tracking experiment currently in progress.
References
Abeillé, A., Clément, L., & Kinyon, A. (2001). Building a Treebank for French. In Building and using syntactically annotated corpora. Kluwer Academic Publishers.
Fodor, J.D. (1998). Learning to parse? Journal of Psycholinguistic Research, 27, 285-319.
Pynte, J., Colonna, S. (1998). French readers sometimes prefer to attach low. Paper presented at AMLaP, Freiburg, September, 24-26.
Pynte, J., & Colonna, S. (2000). Decoupling syntactic parsing from visual inspection: The case of relative clause attachment in French. In Kennedy, A., Radach, R., Heller, D., & Pynte, J. (Eds.). Reading as a perceptual process. pp. 529-547 Oxford: Elsevier