Musical Jabberwocky ?

Imagine a game in which you are asked to listen to a few musical selections, some of which are human compositions and some of which were created by a computer algorithm. Your job is to discriminate between the two, in what amounts to a musical Turing test. This is precisely the way in which David Cope ends the first chapter of his book Virtual Music: Computer Synthesis of Musical Style, inviting the reader to follow along with the anonymous scores while listening to the accompanying CD. In the interest of fairness, both the human and computer pieces are ‘performed’somewhat mechanically by another computer program, in order to deprive both of the meaningful nuances that human musicians would otherwise provide [1]. The exercise is thought-provoking: at least on first hearing, all of the works are passable as human and the categorization is not at all trivial. After twenty years of work on his program Experiments in Musical Intelligence (EMI), David Cope has created a means of creating compositions that come close to passing a musical Turing test, and in the process has raised interesting questions about how music is perceived and whence it derives meaning. The book presents an overview of the program and samples of its compositions, along with commentaries from an eclectic group of musicologists and cognitive scientists: Douglas Hofstadter, Eleanor Selfridge-Field, Bernard Greenberg, Steve Larson, Jonathan Berger, and Daniel Dennett. EMI (or ‘Emmy’, its nickname in the book) composes by taking as input a database of compositions by a particular composer, such as Bach or Chopin, and giving as output a new composition in the style of the input composer. The resulting works are compelling because Emmy abstracts syntactic regularities from the input corpus at a very local level, imitating the texture and voice-leading habits of the to-be-emulated composer, and, at increasingly global levels, creating harmonic patterns of tension-resolution that are consistent with the originals. Because Emmy abstracts, with a sufficiently large musical database the results are not plagiarized, but rather seem to capture the essence of the composer’s style, and in some cases could pass for previously undiscovered Inventions or Mazurkas. One of the ways in which Emmy is successful at mimicking global musical structure is through ‘SPEAC’analysis, which segments a piece according to the functions that its phrases and sections play in the global syntax: statement, preparation, extension, antecedent or consequent. The idea stems from the work of music theorist Heinrich Schenker, who argued that the surface features of music are elaborations of a more basic underlying structure. That Emmy’s compositions depend on such an analysis for their success is consistent with theories such as that of Lerdahl and Jackendoff, who argued that such abstractions are fundamental to unconscious musical parsing [2]. Conversely, some of the other features that the program finds and reproduces might come to inform theories of music perception. Among these are what Cope calls ‘signatures’and ‘earmarks’. Signatures are short melodic fragments that are characteristic of a composer or a period, such as the pattern of a rising scale followed by a downward leap found ubiquitously in the cadences of Mozart piano sonatas. Although the same signatures occur throughout a piece, earmarks occur less frequently in predictable places to highlight important transitions in the global structure of the work. Because Emmy forces a reflection on how music is parsed and structured in the mind of the listener, it could be of more consequence to our understanding of music perception than it is to musical creativity and composition. The commentaries in Virtual Music raise many interesting issues, but a common theme is the nature of musical meaning. In particular, Leonard Meyer’s [3] distinction between absolute and referential meaning comes to mind, or in the terms of Juslin and Sloboda, intrinsic and extrinsic meaning [4]. Referential or extrinsic meaning derives from associations with extramusical concepts. Such associations are sometimes the will of the composer, as in Berlioz’s ‘March to the Scaffold’ in Symphonie Fantastique, and in other cases form as the result of individual or cultural history, as when a rather banal English drinking tune gained new emotional significance for Americans as the setting of their national anthem. Given the nature of Emmy, its compositions do not have these extrinsic references, a flaw that Bernard Greenberg points out in his discussion of its religiously uninspired ‘Bach’compositions. Although the mechanisms of such extrinsic meaning might be more accessible to conscious introspection, a greater and more interesting source of meaning is that intrinsic to the structure of the music itself. Appreciation of music’s structure does not require the explicit knowledge of the musically trained; all members of Western culture implicitly come to understand the syntactic regularities of its music [5], and are capable of being moved by them. In language, it has been argued that form (syntax) and meaning (semantics) are at least partly dissociable, allowing for sentences like those of Lewis Carroll’s ‘Jabberwocky’, in which correct syntax is preserved but semantics is considerably reduced [6]: ’Twas brillig, and the slithy toves Did gyre and gimble in the wabe: All mimsy were the borogoves, And the mome raths outgrabe. Could there be such a thing as a musical Jabberwocky, with syntax but no semantics? On one hand, the answer would seem to be no; given that a significant portion of musical meaning is intrinsic, the relationship between its syntax and semantics may be too tight. If one correctly grasps and imitates great compositions at all of the many hierarchical levels of their musical syntax, intrinsic sources of musical semantics will come along for the ride. On the other hand, all music is in some sense Jabberwocky music, in that we perceive a not-quitetangible meaning lying underneath the surface elements, by virtue of the patterns