Mississippihas a regular alternating stress pattern. Consequently, the principles matching the grid to the syllabic tier
prefer thefirst syllableto be heavy and the second light. And so the syllabification adjusts to meet this preference: the
firstssound isfelttobelongtotheendofthefirstsyllable(and possiblyalsotothebeginning ofthesecond),givingthe
first syllable a coda and therefore making it heavy. By contrast, the secondssound does not belong to the end of the
second syllable, but only to the beginning of the third. This enables the second syllableto be light, comportingwith its
light stress. This situation is interesting to compare withMissouri, whosefirst syllable is unstressed. Here thezsound
(spelledss) is felt to belong to the beginning of the second syllable, so as to keep thefirst syllable light.
Thus thecombinationof syllabic/segmentaland metrical structures results in a complex interaction, of whichwe have
seen here only the tiniest fragment. The principles of stress across the languages of the world have been studied in
intense detail and consist in large part of elaborations of principles of the sort presented here: languages vary, for
instance, in what grids are allowable and preferable (e.g. are there ternary as well as binary grids?), in exactly what
makes a syllable count as heavy, in how stress clashes are mitigated, and in what special provisions can be made at the
beginnings and ends of words (Kager 1995); Halle and Idsardi 1995).^51
The overallarchitecture of this fragment of the grammar can be schematically diagrammed as Fig. 5.2. Let us next see
how this fragmentfits into the larger system. The metrical grid has a number of interfaces with other linguistic and
non-linguistic subsystems. For a simple case, the main stress in a word or phrase determines the point in time when
language will be coordinated with a hand gesture; syllabic content plays no role. Thus beat gestures with the hand can
be performed at the positions in sentence (9) marked in line(a), but notat the positions in line(b).The beats neednot
be perceived visually—they can be performed by pounding on the table (auditory) or poking the hearer (haptic).
114 ARCHITECTURAL FOUNDATIONS
(^51) Giventheessentiallyfinitescope of stress systems, and theclarity withwhichtheyare understood crosslinguistically, stress wouldseem to be a prime candidate for attempts
at computer modeling of language learning—far better than the syntactic and semantic phenomena usually attempted. Stress is of course a lot less sexy, and requires of
modelers an engagement withlinguisticmaterials in exoticlanguages. But I thinktheattemptwould be wellworth theeffort, in terms of theinsightitwouldyield about the
possibilitiesand/or limitationsofvariousmodelsof learning. Inparticular, thesorts ofinteractionsobservedin theexamplesabove,whichare typical,are reminiscentofthe
“Harmony Theory”of Smolensky (1999).