(as in the HSOM^31 ) and time (as proposed in Ref. 102), followed by a later integration step in
order to simulate the development of expectations. The network modelling shall allow fur-
ther investigation of how pitch and time are combined—a question that arises for learning,
perception, and memorization (cf. Ref. 103 for visual stimuli). In addition, networks
integrating both pitch and time create the possibility to study neuropsychological cases. The
goal would be to simulate neuropsychological cases with amelodia without arythmia (and
vice versa) by creating artificial lesions to the networks of normal perception (see Ref. 6 for
applications in other domains).
Even if the HSOM model is conceived on an abstract level and not in relation to brain
structures or neural circuitry, it can be a source of inspiration to generate hypotheses about
the neural circuitry underlying music perception. For example, the self-organizing algo-
rithm conforms to principles of cortical information processing, such as spatial ordering in
sensory cortex or tonotopic organization in auditory cortex. The outcome of the simula-
tions, together with aspects of cortical organization, leads to the question whether there
exist higher order maps such as a tonotopic organization of key centres.
In the domain of music perception, simulations are restricted to learning and perception
in combination with behavioural data. Future developments of networks, however, should
also have the goal of suggesting simulations that make the bridge to actual neural structures
and circuitry in the brain. In other domains, connectionist models in computational neuro-
sciences are created on the bases of existing knowledge about cerebral structures and neural
circuitry.104,105In visual perception, Otto et al.^105 investigated invariant object recognition
(same object despite changes in size or location). The creation of the network was based on
a set of neurophysiological constraints known about the ventro-lateral and dorsal-parietal
pathways associated with the processing of object identification (‘what’) and spatial localiza-
tion (‘where’). Based on the simulations’ outcome, the authors derived new predictions about
interactions between the neural pathways and motivated neurophysiological, biological and
behavioural experiments. Recently, the hypothesis of two types of pathways (‘what’, ‘where’)
has been proposed for auditory perception.106,107Once more details are known about their
functioning, simulations following the line proposed by Otto et al.^105 will enhance the under-
standing of their interaction and stimulate further explorations.
References
1.Christiansen, M. H.and N. Chater(2001) Connectionist Psycholinguistics. Westoirt, CT: Ablex.
2.McClelland, J. L.and D. E. Rumelhart(1981) An interactive activation model of context effects in
letter perception: part 1. An account of basic findings.Psychol. Rev.86, 287–330.
3.Rumelhart, D. E.and J. L. McClelland(1982) An interactive activation model of context effects in
letter perception. part 2.Psychol. Rev.89, 60–94.
4.McClelland, J. L.and J. L. Elman(1986) The TRACE model of speech perception.Cognit. Psychol.
18, 1–86.
5.Seidenberg,M.S.and J. L. McClelland(1989) A distributed, developmental model of word recog-
nition and naming.Psychol. Rev.96, 523–68.
118