Seeking Allies - Natural and Artificial Reasoning - page 177

in which we posited a mechanism of inferential semantics. The mechanism was
proposed on the observation that communication of music was rich with metaphor as a
descriptive aid, metaphors that can be drawn from more explicit and positive domains
such as war or nature. Meaning was thus inferred from relationships evoked by these
metaphors that can then be applied to describing unobserved and less concrete ideas
such as music perception. The problem with metaphors is that they are a culturally
based and a dynamic trope (a figure of speech that changes in context or time). It
is because of this that language falls firmly into the area of irrational sets with the
consequent difficulties. A technical solution invokes the process of tracking meaning
through a belief system (Chap. 12, Addis et al. 2004 ; Billinge and Addis 2004 ;
Stepney et al. 2004 ).
I described in Chap. 12 how an inferential semantics that uses metaphor might
work. Such a process required two distinct mechanisms. The first was a belief system
originally created to show how scientists decide what experiments to perform or
with whom to communicate in order to find out which of several possible hypotheses
about the world is workable. Workable here means making the world more predictable
(Chaps. 5 and 6). The second mechanism was the internal modelling of other people's
beliefs derived from conversation (Chap. 12). This latter process was bypassed in the
original belief system by assuming that the model would be the same as the actual
perceptions (Chap. 7). These actual perceptions were made accessible within the
computer model by allowing the computer actors (agents) to have partial information
of another's perceptions. How this might be accomplished with people was not
considered until the second mechanism of modelling people's beliefs was designed.
These processes were emulated as computer programs to show how they might work
in practice. For such complex processes involving many different scenarios, it was
only through running such computer models that these processes could be tested
for coherence.
Constructing a computer program as though it were a theory is not enough. The-
ories also have to be tested against the world, and the effectiveness of a theory can
only be assessed in its ability to make successful predictions. Any theory that makes
the world predictable is useful, and a better theory will improve on this by making
more secure or detailed predictions. A theory can also provide a framework in which
to design experiments and recognise significant features. Even a completely inade-
quate theory can play this latter role, and without designed experiments and puzzling
observations, a new theory cannot be coherently created. To this end we produced a
computer model of music communication not to say 'This is how it is with people'
but to say 'This makes the world a less surprising place' (Peirce, in Weiner 1966 ).

14.2 The Experiment

The experiment is described in detail elsewhere (Billinge and Addis 2003 , 2004 ).
A summary of the process is that four people (A1-A4) were asked to listen to four
pieces of classical (but little known) music. Each piece of music lasted about 10-15