Commentary on James R.
Hurford
Abstract: 50 words
Main Text: 1095 words
References: 322 words
Total text: 1095 words
+49-211-81-11473
werning@phil-fak.uni-duesseldorf.de
http://www.phil-fak.uni-duesseldorf.de/thphil/werning
Experimental data suggest that the division
between the visual ventral and dorsal pathways may indeed indicate that static
and dynamical information is processed separately. The ventral pathway
primarily generates representations of objects, while the dorsal pathway
produces representations of events. The semantic object/event distinction
relates to the morpho-syntactic noun/verb distinction.
By
presuming that expressions of language exhibit certain logical forms and by
mapping these forms onto regions of cortex, Hurford tries to reduce conceptual
structures to neural events. Although I agree to this two-step method, I
recently presented a methodically similar approach (Werning, 2001, 2003a) that
arrives at different conclusions with regard to the cortical realization of
object and property concepts: (i) A property concept
is identified with the column(s) of neurons that code for the property in
question. The application to the ventral/dorsal division leads to the
conjecture that property concepts are located both in the ventral and the
dorsal stream. However, the ventral stream predominantly hosts representations
for properties of objects expressed by objectual predicates (e.g., by
adjectives like red, vertical or, in more complex cases, by nouns like square,
circle), while columns in the dorsal stream predominantly represent
properties of events, e.g., the direction or speed of motion. These are
expressed by eventual predicates (e.g., by adverbs like downward, quickly
or, if complex, by verbs like to fall, to move). (ii)
Individual (object and event) concepts are identified with synchronous
oscillations. (iii) One or more property concepts F1, ..., Fn
are predicated of an individual concept x, i.e., F1(x) & ...
& Fn(x), just in case neurons of the columns that code for the
properties represented by F1, ... , Fn fire in synchrony,
i.e., with the same oscillation function. On the basis of these assumptions it
was possible to show how cortical structure semantically realizes a predicate
language (Werning, 2003b).
To
highlight the differences to Hurford’s proposal, let
me first turn to what I call his object-by-position hypothesis (OP): An
object concept amounts to the representation of the object's position in space.
OP expresses a fairly rich notion of what object representations are because
falling under an object concept would be equivalent to having all the
positional properties the object concept represents. In other words, being
object x requires the possession of certain essential
properties, if only positional ones. My view, in contrast, is less demanding
and follows Quine’s (1961) dictum: “To be [an individual] is to be the value of
a [bound] variable.” Accordingly, an individual concept would be the neural
event that binds together neural representations of properties. Following
Singer and Grey’s (1995) theory of binding by synchrony, individual concepts
should be identified with synchronous oscillations.
Hurford
conjoins OP with what I call his position-through-dorsal
hypothesis (PD): The dorsal stream (plus superior collicus
and pulvinar) is predominantly engaged in the
processing of positional information; and particularly not in the processing of
property information. A rival view holds that the dorsal stream primarily
processes motional information (Wurtz & Kandel, 2000). According to Merigan
and Maunsell (1993), the dorsal stream (involving,
inter alia, the thick stripes, the middle temporal
(MT), medial superior temporal, lateral parietal and ventral parietal areas and
area 7a) receives input predominantly from the magno
pathway via lateral geniculate nucleus. Recall that magno cells have excellent dynamic (temporal) resolution,
while the parvo cells contributing mainly to the
ventral stream have much better static (spatial) resolution. Furthermore, MT of
the dorsal stream seems to be paradigmatically involved in motion processing.
The dorsal pathway can thus be regarded as carrying mainly dynamic, i.e.,
eventual information (prototypically motion), while the ventral stream seems to
be preoccupied with static, i.e., objectual information (prototypically color
and form). It thus remains an empirically still unsettled issue whether the
ventral/dorsal division corresponds to a distinction between property (“what”)
information and positional (“where”) information or to one between objectual and
eventual information. Wurtz and Kandel
(2000) review a large amount of data from lesions in humans and monkey that
support the second option.
Even
if one accepts OP and PD, it would be rash to conclude that object concepts are
delivered exclusively by the dorsal stream. For, if property concepts are
processed by the ventral stream, what then is the mechanism of predication,
i.e., the mechanism of binding an object concept to a property concept? Hurford
gives no answer. Theorists who identify individual (object and event) concepts
with oscillation functions, in contrast, have shown in detail how the neurons
of one column can be modeled as oscillators so that the Gestalt principles are
honored, according to which neighboring elements with similar properties are
likely to belong to one and the same individual (Schillen
& König, 1994, Maye,
2002, Werning, 2003b). According to this view an individual concept is
generated within hyper-columns by synchronizing and de-synchronizing
connections. Since columns serve as property concepts, there is no anatomical
separation between the processing of property and individual concepts.
Summing
up the critical arguments, one may contrast Hurford’s
view with an alternative hypothesis: Property concepts and individual concepts
alike are processed in the ventral and dorsal stream. However, the dorsal
stream is predominantly occupied with the representation of events, which are
dynamic in nature. It hosts concepts of eventual properties and generates
individual event concepts. The ventral stream, on the other hand, tends to
produce representations of objects. It hosts objectual property concepts, which
are static in nature, and generates individual object concepts. This
ontological division in objects and events reflects a structure that is well
known from the logical analysis of language and thought (Varzi & Pianesi, 2000). A sentence of the form “A red circle is
slumping” has to be analyzed by quantification over an object (the red circle)
and an event (the slumping): ($x$e)(RED(x)
& CIRCLE(x) & SLUMPING(e) & AGENT_OF(x,e).
The mental representation expressed by the sentence, hence, consists of two
objectual property concepts (RED and CIRCLE), one individual object concept
(x), one eventual property concept (SLUMPING) and one individual event concepts
(e). According to the alternative hypothesis, the neural realization of RED and
CIRCLE should be columns of neurons in the ventral stream (e.g. V4). There
should be an oscillation among them that corresponds to the object concept x.
Furthermore, the property concept SLUMPING is expected to be realized by
columns of neurons in the dorsal stream (e.g. MT) and those neurons are
predicted to oscillate synchronously in a way described by the oscillation
function e.
The
alternative theory would, moreover, allow us to aim at a neurobiologically
founded explanation of the origin of the morpho-syntactic noun/verb dichotomy.
Although its universality has bee disputed, there seems to be rich evidence
that it holds (Mithun, 2000, Croft, 2000). Nouns and
their modifiers, adjectives, prototypically denote objects and their
properties, while verbs and their modifiers, adverbs, prototypically refer to
events and their properties. Since the alternative hypothesis suggests that the
semantic object/event distinction correlates with the ventral/dorsal division,
one might conjecture that this division, at least in evolutionary terms, is the
origin of the noun/verb distinction.
Croft, W. (2000). Parts of speech as language universals and as language-particular categories. In P. M. Vogel & B. Comrie (Eds.), Approaches to the typology of word classes (p. 65-102). Berlin: de Gruyter.
Maye, A. (2002). Neuronale Synchronität, zeitliche Bindung und Wahrnehmung. Ph.D. thesis, TU Berlin, Berlin.
Merigan, W. H., & Maunsell, J. H. (1993). How parallel are the primate visual pathways? Annual Review of
Neuroscience, 16, 369-402.
Mithun, M. (2000). Noun and verb in Iroquoian
languages: Multicategorisation from multiple
criteria. In P. M. Vogel & B. Comrie (Eds.),
Approaches to the typology of word classes (p. 397-420).
Pianesi, F., & Varzi,
A. C. (2000).
Events and event talk. In J. Higginbotham, F. Pianesi,
& A. C. Varzi (Eds.), Speaking of events (p.
3-47).
Quine,
W. V. (1961). On what there is. In From a logical point of
view (2nd ed., p. 1-19).
Schillen, T. B., &
König, P. (1994). Binding by temporal structure in multiple feature domains of an
oscillatory neuronal network. Biological Cybernetics, 70, 397-405.
Singer, W., & Gray, C. M. (1995). Visual feature
integration and the temporal correlation hypothesis. Annual
Review of Neuroscience, 18, 555-86.
Werning, M. (2001). How to solve the
problem of compositionality by oscillatory networks. In
J. D.Moore & K. Stenning
(Eds.), Proceedings of the twenty-third annual conference of the cognitive
science society (p. 1094-1099).
Werning,
M. (2003a). Synchrony and composition: Toward a cognitive architecture between
classicism and connectionism. In B. Loewe,
Werning,
M. (2003b). The temporal dimension of thought: Cortical foundations of
predicative representation. Synthese. (in press)
Wurtz, R. H., & Kandel, E. R. (2000).
Perception of Motion, Depth, and Form. In E. R.
Kandel, J. H. Schwartz, & T. M. Jessell (Eds.), Principles of neural science (p.
548-571).