Suppose our initial hypothesis is that
"Kilroy" is a B1 innovation (this is the null hypothesis if there are no facts
forcing us to assume a pre-B1 age for "Kilroy"). If new evidence ("Kilroy was
here") falsifies this hypothesis, we abandon it and formulate a new hypothesis
that matches the facts more accurately. Hypotheses are not sacred (except to
people who feel about their pet theories the way religious fundamentalists feel
about their dogma). This is how science works. The strategy is to reduce
arbitrary speculativeness. You can't intuit the "real" state of affairs if no
convincing evidence for it is available -- you'd end up just fooling
yourself.
Parsimony is not the absolutely highest
priority either -- Occam's Razor is only a useful heuristic. A novel
proposal, even if slightly speculative, may lead to elegant
generalisations or to a more coherent reconstruction that acknowledges the
relevance of hitherto neglected facts. Here's a simple example. The
"Italo-Celto-Germanic" word for 'fish' has traditionally been reconstructed as
*pisk(o)- (Latin piscis, OIr íasc, Gothic fisks), but if we analyse it as
*pik^-sk(o)- with cluster simplification, an attractive hypothesis emerges:
*peik^- means 'paint, mark, decorate', and so the 'fish' word can be interpreted
as 'speckled, spotted' -- the original meaning being perhaps 'trout'. We violate
the principle of parsimony (*pik^-sko- is not the simplest analysis, though it
remains within the bounds of formal acceptability), but the payoff may
justify this offence: we find previously unrecognised cognates outside the
Western IE area that support the new analysis, like Slavic *pIstr-o~gU
'common trout' (with *pIstr- < *pisr- < *pik^-r- = *pIstrU 'piebald,
variegated'), and by finding a semantic derivation for *pisko- we arrive at a
more elegant and intellectually more satisfying etymology. Of course the next
step should be to show the new hypothesis to the public so that critics can
identify its weak points and possibly make us abandon it.
Piotr
----- Original Message -----
Sent: Friday, February 09, 2001 11:31 AM
Subject: [tied] Re: Occam's Razor
In this case, suppose we apply Occam in the version you
suggest. A "Kilroy" exists in language B1, but not in languages B2 and B3,
therefore (by Occam) not in their ancestor language *A. Suddenly we find a
"Kilroy was here" somewhere in the landscape.