Cope's EMI recombined a composer's own signatures

The system

Experiments in Musical Intelligence, usually abbreviated EMI, is composer David Cope's software for generating new music in the style of an existing composer's catalogue. Cope's own account of the project says he 'began Experiments in Musical Intelligence in 1981 as the result of a composer's block,' and that the method he eventually settled on is documented in detail in his book Experiments in Musical Intelligence, published by A-R Editions in 1996, the record this entry is about rather than any single concert or the later Emily Howell persona.

What the documents establish

Cope describes two failed approaches before arriving at the book's method. First he hand-coded part-writing rules, which produced music he calls 'vanilla,' 'lifeless,' and 'without much musical energy.' He then rebuilt the program to work from a database of existing scores, based on what he calls 'recombinancy,' defined in his own words as 'a method for producing new music by recombining extant music into new logical successions.' He states the process explicitly as three steps: 'deconstruction (analyze and separate into parts), signatures (commonality, retain that which signifies style), and compatibility (recombinancy, recombine into new works).' Cope's own bibliography page confirms the 1996 A-R Editions publication and lists an earlier peer-reviewed paper, 'Computer Modeling of Musical Intelligence in Experiments in Musical Intelligence,' in Computer Music Journal, documenting the same method before the book's full treatment.

Craft and rights

Recombinancy, as Cope defines it, depends on analysing a specific composer's existing corpus and identifying the 'signatures' that make a passage recognisably theirs, then recombining those extracted fragments into new sequences. That is structurally the same core operation, analyse a style, extract identifying patterns, recombine them, that sits underneath much later corpus-trained generation, though Cope built and ran it by hand on scores he selected himself rather than on data licensed or scraped at scale. The rights question this raises is not new: whose 'signature' is being extracted, and does stylistic recombination of a long-dead composer's public-domain scores raise the same consent question as recombination of a living artist's catalogue.

Outcomes and open questions

Cope's own account is explicit that this page describes the method, not a verdict on whether any resulting piece truly 'sounds like' the source composer; he writes that reactions to EMI's output have ranged from delight to anger, and predicts audiences will judge future algorithmic composition differently than they judged his.

• Does a description of a style-imitation system explain how it extracts a 'signature,' or only assert that the output sounds authentic?
• Does the corpus a recombination system draws from belong to a living artist, a public-domain historical composer, or something in between?
• What separates a hand-run recombination process from a system trained automatically on the same kind of material at much larger scale?

Cope's own three-step account, deconstruction, signature extraction, recombination, remains the clearest plain-language description available of what his 1996 book actually documents, independent of any later debate about Emily Howell or any single performance.