Here’s a quick summary of an article about phrasal ordering preferences by H. Yamashita and F. Chang, published 2001 in Cognition. The title of the paper is “Long before short” preference in the production of a head-final language. Enjoy.
Summary: The authors conduct a set of production experiments confirming results from corpus studies which show a preference for long-before-short phrasal orderings in Japanese (Hawkins 1994). This is the opposite of the preference in English heavy NP shift constructions for short-before-long orders (Arnold et al 2000), posing a potential problem for the accessibility-based accounts of phrasal ordering which have been used to explain the findings in English.
Experiment 1: Subjects were shown three boxes of text containing a subject NP, object NP, and a transitive verb in different corners of the screen and were asked to create a sentence using those constituents. The weight of the NPs was manipulated by adding a modifier to the subject NP, the object NP, or to neither. The dependent variable was the percentage of trials in each condition in which subjects produced the shifted DO-Sub-V order. Both orders (i.e. canonical sub-NP obj-NP V and shifted obj-NP sub-NP V) were possible in all trials. While the canonical order was always preferred to the shifted order, object-fronting was more likely when objects were long than when short.
Experiment 2: Same basic procedure as in Exp. 1, but here subjects were shown a direct object-NP, indirect object-NP, and ditransitive V. Weight of the NPs was manipulated by adding a relative clause to the DO-NP, the IO-NP, or neither. In each of these trials, four orderings are possible:
1. Canonical: Sub-NP + IO-NP + DO-NP + Verb
2. IO-shift: IO-NP + Sub-NP + DO-NP + Verb
3. DO-shift: DO-NP + Sub-NP + IO-NP + Verb
4. Sentence-internal DO-shift: Sub-NP + DO-NP + IO-NP + Verb
Both IO-shift and DO-shift were more common when the IO-NPs and DO-NPs are long, respectively. Sentence-internal DO-shift was also more common when DO-NPs were long than when DO- and IO-NPs were short. Finally, when DO-NPs were long, sentence-internal shifting was more common than sentence-initial shifting.
Discussion: The long-before-short preference in Japanese is not consistent with accessibility-based accounts of phrasal ordering, which predict that shorter phrases—having fewer words to retrieve and therefore fewer orderings to select from—are more accessible to speakers, and therefore are likely to be produced earlier in an incremental production system.
To account for the findings in Japanese, the authors appeal to the theory of grammatical encoding (Bock and Levelt 1994; Garrett 1980), and propose that the contradictory patterns in English and Japanese are a function of different levels of the production system, or different levels of constraints within the production system. Specifically, they propose that short-before-long orderings in English are a result of form-based decisions, whereas the long-before-short preference in Japanese is a result of decisions which are more conceptually driven. Long phrases are semantically richer than short phrases, making them more salient and therefore more conceptually accessible than short phrases. On the other hand, long phrases are more costly for retrieval, so short phrases are likely to be available before long ones. Phrase ordering decisions then are subject to two possibly competitive sets of constraints, which may be weighed relative to each other in different ways, depending on the language or grammatical construction under consideration. There is evidence that form-based considerations are more influential post-verbally (Stallings et al. 1998), which could tip the scales in favor of conceptually driven as opposed to form-based decisions in verb-final languages such as Japanese.