other interfaces we have seen, reading implements a partial homology between its input and output structures.^115
Sign language presents a more complex case. It evidently replaces one of the phonological tiers—the segmental
structure—with a new one linked to visual input and motor output; its distinctive features are things like hand shape
and hand movement. But the rest of phonological organization—syllabic, prosodic, and morphophonological
structure—is preserved. Again, it appears that the replacement parts in the system are still highly domain-specific and
encapsulated.
Thus we have at least four distinct interfaces between aspects of the visual syste mand aspects of phonology. Each is
exquisitely specialized in the mapping it performs; each is bi-domain-specific in its own particular way. They cannot
together be subsumed under a“general vision–phonology”processor.
What begins to bother me here is the proliferation of special-purpose interface processors. The situationis better than
an indiscriminate general-purpose processor, because each processor is so limited in what it can do. On the other
226 ARCHITECTURAL FOUNDATIONS
(^115) SupposeFodor were rightthat the language module's input is transducers at the ear and its output is syntacticstructure. Since phonological structure is a levelintermediate
betweenthesetwo, itisbyhypothesis informationallyencapsulated.Thatmeansthatthevisualsystemshouldbeunabletoaffectphonologicalstructure—as itclearlydoesin
reading. In otherwords, if we take Fodor's claim of informational encapsulationliterally, there should be no such thing as reading.Fodor might reply (and has done so) that
we really don't know the exact extentof the language module, and that a full characterizationwould include this second class of possible inputs. Still, such an answer has to
finesse a serious proble mof how orthographic infor mation is segregated and processed differently fro mthe rest of visual input in thevisualmodule.Garfield (1987), a
volume devoted to discussion of Fodorian modularity, contains a number of papers that mention reading, but curiously, none observes this challenge to Fodor's position.
For instance, Clifton and Ferreira (1987) speak of a“lexical-processing module (which processes both visual and auditory information)”(279), which sneaks in reading
without further notice.Yetthevisualinformationinvolved inreadingis notraw visual input:identifyinglettersand their linearorder wouldappear torequirea levelofvisual
processing at which size and shape constancy have been established, and at which shapes can be abstracted away fro mthe peculiarities of typeface. Moreover, the linear
order ofwrittenwords can't be detected by a purely“lexical”module.Hencemuch of the“vision module”must be invokedin reading—oneinput module coupled intothe
middle of another.Moreover,Clifton and Ferreira (1987), Carroll and Slowiaczek (1987), and Frazier (1987) all discuss the effects of parsing difficultyon eye movementsin
reading, without noticing that such effects constitutea strong violation of the tenets of Fodorian modularity. Within Fodor's conception,how can the inner workingsof the
domain-specific and informationallyencapsulatedlanguage module interact with theinnerworkingsof thedomain-specific and informationallyencapsulated vision module,
suchthat,say,“theeye-movementcontrolsysteminterruptstheword-recognitionprocessorand switchesintoreanalysismode,under thecontrolofthelanguage processor”
(Carroll and Slowiaczek 1987: 2.34)? Unlike the other two papers, Carroll and Slowiaczek at least recognize vision and language as“two systems that are biologically and
functionally distinct”(2.35); but they overlook the severe consequences for Fodorian modularity.