Phi-features and the Modular Architecture of - UMR 7023 - CNRS
Phi-features and the Modular Architecture of - UMR 7023 - CNRS
Phi-features and the Modular Architecture of - UMR 7023 - CNRS
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
12<br />
1.2 <strong>Modular</strong> architectures<br />
In syntax <strong>and</strong> its adjacent modules <strong>of</strong> realization <strong>and</strong> interpretation, different<br />
mechanisms see <strong>and</strong> operate on phi-<strong>features</strong>. <strong>Phi</strong>-<strong>features</strong> st<strong>and</strong> out an alphabet<br />
shared across <strong>the</strong> modules. Through <strong>the</strong>m, <strong>the</strong> distinctive character <strong>of</strong> each module<br />
may be discerned in <strong>the</strong> character <strong>of</strong> its phi-feature phenomena, <strong>and</strong> <strong>the</strong> interpretability<br />
<strong>of</strong> <strong>the</strong> output <strong>of</strong> one module to ano<strong>the</strong>r investigated.<br />
The modularity hypo<strong>the</strong>sis in (5) is a proposal about <strong>the</strong> architecture <strong>of</strong> human<br />
cognition:<br />
(5) <strong>Modular</strong>ity hypo<strong>the</strong>sis: cognition is organized into modules that are characterized<br />
by <strong>the</strong> following properties, <strong>the</strong>ir modular signature:<br />
a. partly different information types (domain specificity)<br />
(e.g. syntactic vs. phonological <strong>features</strong>)<br />
b. partly different mechanisms/principles<br />
(e.g. unbounded hierarchical vs. linearly adjacent dependencies)<br />
c. partial encapsulation from (inaccessibility to) o<strong>the</strong>r modules<br />
(e.g. phonology does not look into syntax <strong>and</strong> change an intermediate<br />
computation or an intermediate representation)<br />
The hypo<strong>the</strong>sis admits a range <strong>of</strong> executions that differ in <strong>the</strong> type <strong>and</strong> degree<br />
<strong>of</strong> uniqueness <strong>and</strong> autonomy <strong>of</strong> <strong>the</strong> modules. It characterizes <strong>the</strong> way cognition<br />
functions, without commitments to <strong>the</strong> neurological localisability or innate character<br />
<strong>of</strong> modules. It is applicable to a variety <strong>of</strong> cognitive domains, without asserting<br />
<strong>the</strong> nonexistence <strong>of</strong> nonmodular systems. The formulation in (5) is suggested by<br />
much recent inquiry, notably Chomsky (1980), Marr (1983), Pinker (1994),<br />
Jackend<strong>of</strong>f (2002), Col<strong>the</strong>art (1999), Sperber (2002), Barrett (2005), Barrett <strong>and</strong><br />
Kurzban (2006), <strong>and</strong> <strong>the</strong> articles in Carru<strong>the</strong>rs, Laurence <strong>and</strong> Stitch (2005, 2007).<br />
These build on but also differentiate <strong>the</strong>mselves from Fodor's (1983) pioneering<br />
proposal where modules are fur<strong>the</strong>rmore innately specified, fast-<strong>and</strong>-automatic,<br />
neurologically localisable input transducers. A brief look at modularity in ano<strong>the</strong>r<br />
cognitive domain provides a perspective for its manifestations in language.<br />
Low-level vision has been a paradigm case <strong>of</strong> modularity. Pylyshyn (1999)<br />
presents an overview highlighting its high degree <strong>of</strong> uniqueness <strong>and</strong> autonomy<br />
from <strong>the</strong> "central" cognitive system <strong>of</strong> goals, beliefs, <strong>and</strong> utilities (cf. also<br />
Raftopoulos 2001). These characteristics <strong>of</strong> vision may be illustrated through visual<br />
illusions such as those in Figure 1.1. In <strong>the</strong> Kanisza rectangle, interrupted contours<br />
are automatically completed to build <strong>the</strong> percepts <strong>of</strong> a foregrounded rectangle<br />
<strong>and</strong> four occluded circles, although <strong>the</strong> image can reflect a scene without<br />
corresponding objects. In <strong>the</strong> Penrose triangle, <strong>the</strong> spatial relationships between<br />
any two bars are locally computed from <strong>the</strong>ir joints, <strong>and</strong> <strong>the</strong>n maintained when<br />
<strong>the</strong>y are all assembled into a single physically impossible object. In <strong>the</strong> Müller-