6.4: Musical Romanticism and Connectionism

The eighteenth-century Industrial Revolution produced profound changes in the nature of European life, transferring power and wealth from the nobility to the commercial class (Plantinga, 1984). Tremendous discontentment with the existing social order, culminating in the French revolution, had a profound influence on political, intellectual, and artistic pursuits. It led to a movement called Romanticism (Claudon, 1980), which roughly spanned the period from the years leading up to the 1789 French revolution through to the end of the nineteenth century.

A precise definition of Romanticism is impossible, for it developed at different times in different countries, and in different arts—first poetry, then painting, and finally music (Einstein, 1947). Romanticism was a reaction against the reason and rationality that characterized the Enlightenment period that preceded it. Romanticism emphasized the individual, the irrational, and the imaginative. Arguably music provided Romanticism’s greatest expression (Einstein, 1947; Plantinga, 1984), because music expressed mystical and imaginative ideas that could not be captured by language.

It is impossible to provide a clear characterization of Romantic music (Einstein, 1947; Longyear, 1988; Plantinga, 1984; Whittall, 1987). “We seek in vain an unequivocal idea of the nature of ‘musical Romanticism’” (Einstein, 1947, p. 4). However, there is general agreement that Romantic music exhibits,

a preference for the original rather than the normative, a pursuit of unique effects and extremes of expressiveness, the mobilization to that end of an enriched harmonic vocabulary, striking new figurations, textures, and tone colors. (Plantinga, 1984, p. 21)

The list of composers who were musical Romanticism’s greatest practitioners begins with Beethoven, and includes Schubert, Mendelssohn, Schumann, Chopin, Berlioz, Liszt, Wagner, and Brahms.

Romantic music can be used to further develop the analogy between classical music and cognitive science. In particular, there are several parallels that exist between musical Romanticism and connectionist cognitive science. The most general similarity between the two is that both are reactions against the Cartesian view of the mind that dominated the Enlightenment.

Romantic composers wished to replace the calculated, rational form of music such as Bach’s contrapuntal fugues (Gaines, 2005; Hofstadter, 1979) with a music that expressed intensity of feeling, which communicated the sublime. “It was a retrogression to the primitive relationship that man had had to music—to the mysterious, the exciting, the magical” (Einstein, 1947, p. 8). As a result, musical Romanticism championed purely instrumental music; music that was not paired with words. The instrumental music of the Romantics “became the choicest means of saying what could not be said, of expressing something deeper than the word had been able to express” (p. 32). In a famous 1813 passage, music critic E. T. A. Hoffman proclaimed instrumental music to be “the most romantic of all the arts—one might almost say, the only genuinely romantic one—for its sole subject is the infinite” (Strunk, 1950, p. 775).

Connectionist cognitive science too is a reaction against the rationalism and logicism of Cartesian philosophy. And one form of this reaction parallels Romantic music’s move away from the word: many connectionists interpreted the ability of networks to accomplish classical tasks as evidence that cognitive science need not appeal to explicit rules or symbols (Bechtel & Abrahamsen, 1991; Horgan & Tienson, 1996; Ramsey, Stich, & Rumelhart, 1991; Rumelhart & McClelland, 1986a).

A second aspect of musical Romanticism’s reaction against reason was its emphasis on the imaginary and the sublime. In general, the Romantic arts provided escape by longingly looking back at “unspoiled,” preindustrial existences and by using settings that were wild and fanciful. Nature was a common inspiration. The untamed mountains and chasms of the Alps stood in opposition to the Enlightenment’s view that the world was ordered and structured.

For example, in the novel Frankenstein (Shelley, 1985), after the death of Justine, Victor Frankenstein seeks solace in a mountain journey. The beauty of a valley through which he traveled “was augmented and rendered sublime by the mighty Alps, whose white and shining pyramids and domes towered above all, as belonging to another earth, the habitations of another race of beings” (p. 97). To be sublime was to reflect a greatness that could not be completely understood. “The immense mountains and precipices that overhung me on every side—the sound of the river raging among the rocks, and the dashing of the waterfalls around, spoke of a power mighty as Omnipotence” (p. 97).

Sublime Nature appeared frequently in musical Romanticism. Longyear’s (1988, p. 12) examples include “the forest paintings in Weber’s Der Freischütz or Wagner’s; the landscapes and seascapes of Mendelssohn and Gade; the Alpine pictures in Schumann’s or Tchaikovsky’s Manfred” to name but a few.

Musical Romanticism also took great pains to convey the imaginary or the indescribable (Whittall, 1987). In some striking instances, Romantic composers followed the advice in John Keats’ 1819 Ode on a Grecian Urn, “Heard melodies are sweet, but those unheard / Are sweeter.” Consider Schumann’s piano work Humoreske (Rosen, 1995). It uses three staves: one for the right hand, one for the left, and a third—containing the melody!—which is not to be played at all. Though inaudible, the melody “is embodied in the upper and lower parts as a kind of after resonance—out of phase, delicate, and shadowy” (p. 8). The effects of the melody emerge from playing the other parts.

In certain respects, connectionist cognitive science is sympathetic to musical Romanticism’s emphasis on nature, the sublime, and the imaginary. Cartesian philosophy, and the classical cognitive science that was later inspired by it, view the mind as disembodied, being separate from the natural world. In seeking theories that are biologically plausible and neuronally inspired (McClelland & Rumelhart, 1986; Rumelhart & McClelland, 1986c), connectionists took a small step towards embodiment. Whereas Descartes completely separated the mind from the world, connectionists assume that brains cause minds (Searle, 1984).

Furthermore, connectionists recognize that the mental properties caused by brains may be very difficult to articulate using a rigid set of rules and symbols. One reason that artificial neural networks are used to study music is because they may capture regularities that cannot be rationally expressed (Bharucha, 1999; Rowe, 2001; Todd & Loy, 1991). These regularities emerge from the nonlinear interactions amongst network components (Dawson, 2004; Hillis, 1988). And the difficulty in explaining such interactions suggests that networks are sublime. Artificial neural networks seem to provide “the possibility of constructing intelligence without first understanding it” (Hillis, 1988, p. 176).

Musical Romanticism also celebrated something of a scale less grand than sublime Nature: the individual. Romantic composers broke away from the established system of musical patronage. They began to write music for its own (or for the composer's own) sake, instead of being written for commission (Einstein, 1947). Beethoven’s piano sonatas were so brilliant and difficult that they were often beyond the capabilities of amateur performers who had mastered Haydn and Mozart. His symphonies were intended to speak “to a humanity that the creative artist had raised to his own level” (p. 38). The subjectivity and individualism of musical Romanticism is one reason that there is no typical symphony, art-song, piano piece or composer from this era (Longyear, 1988).

Individualism was also reflected in the popularity of musical virtuosos, for whom the Romantic period was a golden age (Claudon, 1980). These included the violinists Paganini and Baillot, and the pianists Liszt, Chopin, and Schumann. They were famous not only for their musical prowess, but also for a commercialization of their character that exploited Romanticist ideals (Plantinga, 1984). Paganini and Liszt were “transformed by the Romantic imagination into a particular sort of hero: mysterious, sickly, and bearing the faint marks of dark associations with another world” (Plantinga, 1984, p. 185).

Individualism is also a fundamental characteristic of connectionism. It is not a characteristic of connectionist researchers themselves (but see below), but is instead a characteristic of the networks that they describe. When connectionist simulations are reported, the results are almost invariably provided for individual networks. This was demonstrated in Chapter 4; the interpretations of internal structure presented there are always of individual networks. This is because there are many sources of variation between networks as a result of the manner in which they are randomly initialized (Dawson, 2005). Thus it is unlikely that one network will be identical to another, even though both have learned the same task. Rather than exploring “typical” network properties, it is more expedient to investigate the interesting characteristics that can be found in one of the networks that were successfully trained.

There are famous individual networks that are analogous to musical virtuosos. These include the Jets-Sharks network used to illustrate the interactive activation with competition (IAC) architecture (McClelland & Rumelhart, 1988); a multilayered network that converted English verbs from present to past tense (Pinker & Prince, 1988; Rumelhart & McClelland, 1986a); and the NETTALK system that learned to read aloud (Sejnowski & Rosenberg, 1988).

Individualism revealed itself in another way in musical Romanticism. When Romantic composers wrote music for its own sake, they assumed that its audience would be found later (Einstein, 1947). Unfortunately, “few artists gained recognition without long, difficult struggles” (Riedel, 1969, p. 6). The isolation of the composer from the audience was an example of another Romantic invention: the composer was the misunderstood genius who idealistically pursued art for art’s sake. “The Romantic musician . . . was proud of his isolation. In earlier centuries the idea of misunderstood genius was not only unknown; it was inconceivable” (Einstein, 1947, p. 16).

The isolated genius is a recurring character in modern histories of connectionism, one of which is presented as a fairy tale (Papert, 1988), providing an interesting illustration of the link between Romanticism and connectionism. According to the prevailing view of connectionist history (Anderson & Rosenfeld, 1998; Hecht-Nielsen, 1987; Medler, 1998; Olazaran, 1996), the isolation of the neural net researcher began with a crusade by Minsky and Papert, prior to the publication of Perceptrons (Minsky & Papert, 1969), against research funding for perceptron-like systems.

Minsky and Papert’s campaign achieved its purpose. The common wisdom that neural networks were a research dead-end became firmly established. Artificial intelligence researchers got all of the neural network research money and more. The world had been reordered. And neurocomputing had to go underground. (Hecht-Nielsen, 1987, p. 17)

Going underground, at least in North America, meant connectionist research was conducted sparingly by a handful of researchers, disguised by labels such as “adaptive pattern recognition” and “biological modelling” during the “quiet years” from 1967 until 1982 (Hecht-Nielsen, 1987). A handful of neural network researchers “struggled through the entire span of quiet years in obscurity.” While it did not completely disappear, “neural-net activity decreased significantly and was displaced to areas outside AI (it was considered ‘deviant’ within AI)” (Olazaran, 1996, p. 642). Like the Romantic composers they resemble, these isolated connectionist researchers conducted science for science’s sake, with little funding, waiting for an audience to catch up—which occurred with the 1980s rise of New Connectionism.

Even though Romanticism can be thought of as a musical revolution, it did not abandon the old forms completely. Instead, Romanticist composers adapted them, and explored them, for their own purposes. For example, consider the history of the symphony. In the early seventeenth century, the symphony was merely a short overture played before the raising of the curtains at an opera (Lee, 1916). Later, the more interesting of these compositions came to be performed to their own audiences outside the theatre. The modern symphony, which typically consists of four movements (each with an expected form and tempo), begins to be seen in the eighteenth-century compositions of Carl Philip Emmanuel Bach. Experiments with this structure were conducted in the later eighteenth century by Haydn and Mozart. When Beethoven wrote his symphonies in the early nineteenth century, the modern symphonic form was established—and likely perfected. “No less a person than Richard Wagner affirmed that the right of composing symphonies was abolished by Beethoven’s Ninth’” (p. 172).

Beethoven is often taken to be the first Romantic composer because he also proved that the symphony had enormous expressive power. The Romantic composers who followed in his footsteps did not introduce dramatic changes in musical form; rather they explored variations within this form in attempts to heighten its emotional expressiveness. “Strictly speaking, no doubt, musical Romanticism is more style than language” (Whittall, 1987, p. 17). Romantic composers developed radically new approaches to instrumentation, producing new tone colours (Ratner, 1992). The amount of sound was manipulated as an expressive tool; Romanticists increased “the compass, dynamic range, and timbral intensity of virtually all instruments” (Ratner, 1992, p. 9). New harmonic progressions were invented. But all of these expressive innovations involved relaxing, rather than replacing, classical conventions. “There can be little doubt that ‘romantic’ musical styles emanate from and comingle with ‘classic’ ones. There is no isolable time and place where one leaves off and the other begins” (Plantinga, 1984, p. 22).

Connectionist cognitive science has been portrayed as a revolution (Hanson & Olson, 1991) and as a paradigm shift (Schneider, 1987). However, it is important to remember that it, like musical Romanticism, also shares many of the characteristics of the classical school that it reacted against.

For instance, connectionists don’t abandon the notion of information processing; they argue that the brain is just a different kind of information processor than is a digital computer (Churchland, Koch, & Sejnowski, 1990). Connectionists don’t discard the need for representations; they instead offer different kinds, such as distributed representations (Hinton, McClelland, & Rumelhart, 1986). Connectionists don’t dispose of symbolic accounts; they propose that they are approximations to subsymbolic regularities (Smolensky, 1988).

Furthermore, it was argued earlier in this book that connectionist cognitive science cannot be distinguished from classical cognitive science on many other dimensions, including the adoption of functionalism (Douglas & Martin, 1991) and the classical sandwich (Calvo & Gomila, 2008; Clark, 1997). When these two approaches are compared in the context of the multiple levels of investigation discussed in Chapter 2, there are many similarities between them:

Indeed, the fact that the two can be compared in this way at all indicates a commitment to a common paradigm—an endorsement of the foundational assumption of cognitive science: cognition is information processing. (Dawson, 1998, p. 298)

Copland (1952, pp. 69–70) argued that the drama of European music was defined by two polar forces: “the pull of tradition as against the attraction of innovation.” These competing forces certainly contributed to the contradictory variety found in musical Romanticism (Einstein, 1947); perhaps they too have shaped modern connectionist cognitive science. This issue can be explored by considering connectionist approaches to musical cognition and comparing them to the classical research on musical cognition that was described earlier in the current chapter.