Normally the syntax of English is very insistent that the direct object precede any time adverbials, as seen in the
contrast (20a, b). But if the object is very long and the time adverbial short, the reverse order (20d) is far more
acceptable. The reason is that this permits the prosody to satisfy conditions (18b, c) much better: the IntPs are much
closer in length, with the longer at the end. So evidently the needs of prosody are forcing a non-optimal syntactic
structure.^57 In short, the interaction betweenintonational phrasing and syntax resembles the interactionbetween stress
and syllabification seen in the previous section.
To su mup thissection: therelationof syntax tophonology, especiallytointonation, is qualitativelyjust liketherelation
amongthetiers ofphonology. Syntaxand phonologyareindependentcombinatorialsystems, builtfromdistinctsetsof
primitiveelements combined by distinct sets of formation rules. Neither can be reduced to or derived from the other.
Certain aspects of each component are related to certain aspects of the other through interface constraints. For
instance, small-scale syntactic units such as Nouns and Verbs correspond in linear order to morphophonological
Words and Clitics, and larger-scale syntactic phrases constrain Intonational Phrases. But the actual segmental content
of phonological words and their syllabification and stress play no role in constraining syntax, and most details of
syntactic embedding are invisible to phonology.^58 This sort of relation has in fact been strongly favored among
phonologists for some years (Zwicky and Pullum 1983); Inkelas and Zee 1990).
If we take this conclusion seriously, we ought to go on to the further conclusion that syntactic trees do not contain
phonologicalinformation. As far as syntax is concerned,star, galaxy, nebula, andcometare identical:they are justsingular
count nouns. It is this conclusion that justifies the for mof syntactic trees in Fig. 1.1—trees like (21a) rather than the
traditional (21b).
THE PARALLEL ARCHITECTURE 121
(^57) The traditional approach to this situation is to say that the direct object in (20d) is generated in normal direct object position, like (20c), but it is moved to the end by a
derivational rule called Heavy NP Shift. I a maware of no account in the literature that does it exactly the way proposed here, na mely through an override of syntactic
ordering constraintsin theinterests of prosody. For a discussionof“heaviness”constraints on syntacticorder, see Arnoldet al. (2000).Hawkins (1994) proposes an account
oftheseheaviness constraints interms ofparsing:basically, bothspeechproductionand speechperceptionfavorholding off longconstituentsas long as possible,preferring
to integrate shorter constituentsfirst. I don't think his accountis necessarily incompatible with the one offered here: it is important that the reordering involved,even if for
processing purposes, be stated in terms of intonational constraints. However, a closer comparison of the two views is warranted.
(^58) The problems here are the occasional cases where syntactic category has a bearing on word stress. A well-known example is Englishpermit (verb) versuspérmit (noun).