The computer tradition has always served as a great leveler between theoretical possibility and practical realization. One of the long-sought goals of that tradition, a world-championship level chess program, was gradually realized over the past few years. Can another long-sought goal, the computer capable of composing music at the level of a Bach or a Mozart, be far behind? Not according to the New York Times, whose recent cover story "Undiscovered Bach? No, a Computer Wrote It" (George Johnson, Science Times, November 11, 1997) coincided with the release of a new CD "Classical Music Composed by Computer" on Centaur Records. The composer is the computer program EMI (Experiments in Musical Intelligence), created by Professor David Cope of the Music Department of the University of California at Santa Cruz. Dr Cope originally designed the program to assist in the composition of his own works, but eventually, he became absorbed in defining the parameters of styles other than his own. EMI is not a repository for any particular repertoire, but rather, a program that takes, as its starting point, a single given work, and produces, as its endpoint, an imitation of the given work, in codifyably similar style. To date, the program has produced works ranging in style and medium from Palestrina thru Stravinsky, from solo instrument thru full orchestra. The Fifteen Two-Part Inventions were completed in 1996.

Just how well do they fulfill their intent with respect to Bach's Inventions? The answer depends upon how well one is able to gauge the stylistic "authenticity" of imitations in the first place. EMI's computer-generated imitations have been favorably received in performances at major universities and music festivals nationwide. In one instance, an EMI piece in the style of Bach was juxtaposed against another Bach-styled piece composed by a professional, as well as against an authentic piece by J. S. Bach himself. In an ensuing audience poll, EMI's piece was judged "authentic," while the professionally written piece was judged to have been written by the computer. It was not reported how Johann Sebastian's piece was judged, but clearly, vox populus is no gauge for stylistic authenticity. Let us consider, then, the role of the practical performer. In one instance, a period instrument specialist, speaking of a piano concerto created by EMI in imitation of Mozart, offered this appraisal:"It felt a little different than playing a normal Mozart work. But it was very much like a work of the same period. It was certainly in the ballpark." Somehow, the "ballpark," as in the music of a Fasch or a Dittersdorf, strikes one as a less worthy model than that of a Bach or a Mozart. It is the specific relationship to the major historical composers, rather than a generic relationship to lesser composers, that piques the interest in this project, and at any rate, the "feel" of a work is no substitute for a style-critical understanding.

Style criticism in the works of major historical composers is not an exact science, but it is a science nonetheless. It requires, first, a knowledge of the composers' complete known works in their authentic sources, and second, a strong theoretical apparatus. Style criticism is independent of the esthetics of performance, of the whims and vagaries of intervening executants, and of the consensual judgments of casual listeners. It is also the key determinant in the authentication of doubtful works, when source attributions remain unverifiable. In the European art community, major exhibitions of old master paintings and drawings are routinely inspected by committees of curators, academics, conservators, dealers, and expert collectors, that have sweeping powers to de-authenticate, reattribute, and even remove from exhibits, any works deemed to be of questionable status. These committees are a closed forum, but for that reason, can move quickly under pressure of the commercial nature of the exhibitions. The musicological community, by contrast, is a slow-moving, open forum, that speaks to a smaller, more disinterested audience. Since an exhaustive measure-by-measure analysis will neither be practicable within the present space, nor serviceable to the present readership without reference to the complete score (which remains unpublished), the following comments will deal with selected passages from selected pieces only.

Invention #1 in C major proceeds from phrase one to phrase two with a bass-formula that does not occur in Bach (c, g, b, c', where Bach would probably have written g, a, b, c'). There is no modulation to the dominant, although this might have taken place during the third phrase. The third phrase leads, instead, to a long digression in supertonic and subtonic regions, neither of which are key-relations that Bach would normally have emphasized. There are several harmonic non sequiturs, such as a progression from C minor, to D minor, to E minor within a three-measure span, and at the end, a confused succession of cadences, concluding on a six-four chord more akin to proto-classical style than to Bach. Invention #2 in C minor has, for a thematic model, Bach's Invention #6 in E major, BWV 777, but the initial chromatic motif is abandoned after the first phrase. The rising bass at the outset should probably have been interrupted at measure two, and the resulting tritone and diminished-seventh figures would have been more characteristic of the C minor tonality. In BWV 777, Bach's syncopation is relieved by a lively dactylic figure, but here, the writing carries the syncopation to excess, and is harmonically awkward. There is no modulation to the dominant (although a pair of accidentals at measure ten could have accomplished this), and no modulation whatsoever throughout the entire second half of the piece. Moreover, the final measures are all in so-called "natural" (Aeolian) minor, though Bach, whenever possible, uses raised sixths and sevenths in his minor mode. A sudden ascending melodic figure, five bars from the end, traverses nearly two-and-a-half-octaves within a two-measure span, and is, simply, poor writing. Invention #6 in E major has, for a thematic model, Bach's Invention #13 in A minor, BWV 784. There are several places in the bass where the accompanimental figures are unwieldy and uncharacteristic. Some of the sequences are not carried out as rigorously as one expects, and in one instance, Bach's permissible keyboard range with respect to Invention-writing is violated. There are several harmonic errors (doubled sevenths, missing root), and at least one far-fetched harmonic non sequitur. The development leads to a perfunctory transition that ought to have been extended further before returning to the opening theme. A V/I in G# minor, two measures from the end, is clearly out-of-place, and weakens the final cadence considerably. Invention #8 in F minor has, for a thematic model, Bach's Invention #9, BWV 780. The opening motif, the final cadence, and several other passages (about seven measures altogether) are essentially identical to Bach's. The opening harmonic progression, however (i, iv, VII, i, VI, VII), makes for an indecisive exposition, and sounds as if it were already the middle of the piece, rather than the beginning. Later, a sequence occurs, in which each two-measure group contains no fewer than four distinct figural ideas, whereas Bach's sequential patterns normally involve no more than two. The piece concludes with measures 1–7 in a literal da capo, but literal da capo is a luxury that Bach does not normally employ in short contrapuntal pieces. Invention #9 in G minor has a lively measure-long subject, but the ensuing countersubject is poor and arbitrary. In a two-part piece that moves mostly in eighths and sixteenths, it is best when one voice takes the eighths, and the other, the sixteenths. But where both voices have eighths, a certain stagnant quality is apt to result, and that is the case in at least two places here. An uncharacteristic bass passage occurs twice(including at the end): f, f#, d, g, where Bach would surely have written e, f#, d, g. Invention #11 in A minor begins with the unlikely bass-formula a, A, e, g#, where Bach would surely have written a, A, a, g#. At the same time, the opening melodic figuration might have been treated slightly differently had its idiomatically violinistic character been better taken into account. The initial statement of the theme might well have been conceived as a characteristic double-counterpoint combination, but this possibility was not realized. There are some untidy sequences, and several other unlikely figures in the bass. Toward the end, a tessitura problem arises, which Bach would surely have corrected, even at the expense of motivic consistency.

Again, these comments are offered only as representative examples of what a more exhaustive style-critical analysis might entail. The fact that these and similar examples fail under close examination does not, of course, suggest that either the theoretical or practical possibilities of the system are necessarily diminished. In most instances, the inconsistencies with the known work of J. S. Bach (or with compositional common-sense) are so transparent that they could easily have been observed and corrected tutorially. Had Dr. Cope employed EMI merely as a means of assisted composition, then these problems could have been worked out through the usual process of revision and refinement. But since he has pursued a different vision––following the discipline of totally computer-generated, rather than merely computer-assisted composition––then no such revisions are available to him, other than those that can be programmed into the system. From a programming point of view, one reasonably assumes that all such problems can eventually be worked out. But from a systems design point of view, if all such problems were eventually worked out, it is by no means clear that the system would then be capable of generating an appreciably different work than the model being furnished. In other words, if every conceivable stylistic parameter of a Bach composition were to have been programmed into the system, these very parameters, as a whole, might not then allow for any solution at all, other than the "solution" already found in the composition itself. Stated yet another way, Cope's Inventions, notwithstanding specific criticisms, do contain quite a bit of material that cannot, in the present writer's view, be distinguished from "authentic" Bach. If the ratio of objectionable passages to "authentic" passages were to be gradually reduced, one could eventually arrive at a perfect system. At the present time, however, no such system exists, and one can not say with certainty whether such a system would necessarily produce startling "new" Bach works at all, as opposed, merely, to obvious rehashings of the original models. Thus, it is important to Bach research that the experiment continue. There is, to be sure, an innate tendency among music lovers to treat such experimentation with skepticism. The New York Times, in the above-mentioned article, expressed the fear that "If a computer writes music, is music less than it appears?...If artificial composers can write 'the real thing,' then what has happened to creativity?" There is a history to that speculative viewpoint wherein Bach's music is held to embody a kind of divine presence, or other eternal reality, at a level unknowable, but higher than that of the music itself. This cherished belief is probably held by more music lovers throughout the world than most of us realize. On the other hand, many people feel that the time must eventually come when computers will generate a music that, to anyone's stylistic discernment, will be at the level of a Bach or a Mozart. It is a purely philosophical conundrum to try to imagine, for example, what Bach might have composed had he lived just a little longer, when, in reality, things did not unfold that way. But there is no doubt that Bach did write more music than has been transmitted to posterity, and computer-assisted processes could well be employed in attempting to logically recreate some of it. Skepticism and fear yield to fond hope, when one realizes that there is no real contest here at all; that the historic "chess match" is not between EMI and Bach, but rather, between EMI and a handful of exceptional musicians who might also be considered capable of recreating Bach's style, through composition, improvisation, and text-revision. At this point, we should be begging the question of just how close to perfect the EMI program presently rates with respect to recreating the style of Bach. The presence of a few trouble-spots, in and of itself, signifies nothing. The specific instances cited above, for example, constitute no more than about twenty percent of the work as a whole. If these were the only objectionable passages in the entire work, then one could credit EMI with an astounding eighty percent success rate in imitating Bach. But since a more thoroughgoing measure-by-measure analysis was omitted for lack of space, what, then, is the closer measure of EMI's actual success rate?

On the strength of the premise that four-measure phrases are, normally, the basic unit of speech in Bach's musical language––comparable, say, to discreet phrases (as opposed to single words) in spoken language––we can make an overall quantitative assessment as to how much of the work withstands close style-critical analysis by limiting positive assessments to passages of four measures and longer. Cope's Fifteen Two-Part Inventions comprise, altogether, 556 measures. A tabulation of the lengths of each of the individual pieces follows.

Invention #1, in its 50 measures, contains (among passages of at least four-measures length) seven measures which, in this writer's view, might pass for authentic. We might also include an additional four-measure passage, which, aside from its unlikely key-relationship to the rest of the piece (see commentary above), might also pass for authentic, if taken out-of-context. In Invention #2, it was not possible to discover any passages of least four-measures length that might pass for authentic. In Invention #3, at least sixteen measures were found passable; an additional six measures might, likewise, be found passable, allowing, however, for their poor connection with what precedes and follows. Inventions #4 and #5 showed nine measures each, with an additional four measures possible in Invention #4, although these measures emanate from an unlikely modulation to VI, where iv would have been more characteristic in Bach. In Invention #'s 6, 7, 9, and 10, no plausible four-measure passages were found. Invention #8 showed six good measures, several of them identical to their original model (see commentary above). Invention #11 showed fourteen good measures, plus another possible ten measures, in which the opportunity to employ the opening theme in characteristic double-counterpoint was missed (see commentary above).Invention #'s 12, 13, and 15 showed twenty-two, eleven, and nine good measures, respectively. Finally, Invention #14 showed four good measures, with an additional twenty-eight measures good in-and-of-themselves, though cast in a decidedly uncharacteristic subtonic key-relationship. In arriving at a quantitative assessment of EMI's present success rate, we therefore need to adopt a two-tiered view of what constitutes "good" writing, or writing that, to a high level of stylistic discernment, is indistinguishable from known, authentic Bach. On the one hand, we reckon as "good," any passages of four-measures or more that are irreproachable in terms of the kind of style-critical analysis articulated above. But on the other hand, we need to give consideration also to passages which, though faulty in the context of the piece as a whole, are otherwise "good" in-and-of-themselves. By this standard, the system may be judged surprisingly successful in imitating Bach, though final perfection, at present, remains more within the realm of theoretical possibility than current capability. We should, by all means, be ready and willing to harbor, someday, the hope of encountering "new" Bach works in this way, but not without careful empirical validation along style-critical lines.

[Home Page]
[Inquiries]