Handle: Engineering Artificial Musical Creativity at the “Trickery” Level - Computational Creativity Research: Towards Creative Machines - page 297

14.5.3.3 Music Generation
Once the story has been formally represented, we can use a theorem prover to deter-
minewhat emotions or beliefs are present at any given timepoint by simply requesting
a series of proofs. One challenge in this method is that if we ask the theorem prover
to prove that an emotion is present at a given timepoint when it is not, the theorem
prover may run forever. However, since our stories are relatively simple, a proof
is usually found quickly if one exists we can confidently declare that an emotion
is absent if it is not found after a reasonable time period, so we still obtain good
results. As we scale up to more complex stories, this technique may not be valid
unless we are willing to wait a long time for every proof, but for now it achieves
our desired goal. Once the emotions, beliefs, or other factors important to the music
are determined for every timepoint, we query a knowledge-base using these descrip-
tive factors to determine what song is most appropriate at that time. Currently, our
knowledge-base is in accord with Motion Picture Moods for Pianists and Organists ,
a topic of songs for silent films like one mentioned previously [ 23 ]. For the fireworks
scene, Handle knows that the Triangle will perceive fireworks going off and can
therefore use the formulae presented above to infer that the Triangle will believe
there is a celebration and that the Triangle will be happy. Handle then queries the
knowledge-base for music that best matches the “happy” and “celebration” descrip-
tors and returns Giacomo Meyerbeer's Coronation March , the first line of which can
be seen in Fig. 14.14 .
As we move toward modern-day film music, the interaction between the story
and the music becomes much more complex, and therefore our method for gener-
ating music will have to become correspondingly intelligent. For example, while
most silent music is parallel in nature—meaning that it directly correlates with what
is happening in the story—it is just as common for modern-day film music to be
counterpunctual—meaning that it contradicts what is happening in the story, often
to emphasize some specific aspect. To cope with such techniques, we need to expand
our representation of the story to include not only what is happening in the story,
but what the story is trying to convey. We could then reason about what music best
matches the purpose of the story rather then just its contents. Our hope is that a
“reversal” of the story-generation approach in [ 7 ] will serve as a basis for this deeper
form of story understanding.
Fig. 14.14 Music selected by Handle for the fireworks scene