--- In
cybalist@yahoogroups.com, "mkelkar2003" <swatimkelkar@...>
wrote:
> http://www.hjholm.de/SLR-update%20web.htm
>
> What would the Indo Hittite tree look like using SLR (Separation
Level
> Recovery method)?
> "Holm, Hans J.: The Indo-Hittite Hypothesis Falsified by the
> Separation-Level
> Method. In preparation."
I can think of two flaws that the web page shows no sign of having
been corrected.
Firstly, how do you know whether a feature (e.g. a root) is
retained? You can't do it on the basis of just two languages, you
have to bring others into consideration. I followed his suggestion
for modelling the method and generalised it as follows:
Consider 3 languages, A, B and C, with independent retention
probabilities pA, pB, and pC. A feature counts as retained by
language A if both language A and one or both of B and C retain it,
and similarly for B. For pC = 1.0 you get the simple case that he
invites us to simulate.
I then considered 3 languages with pA = pB = 0.6 and pC = 0.8, and
did a thousand replications starting with 1000 features. (Private
note: Program slr.f) The number of features shared by the ancestor
of A and B was then calculated as 846.7 (s.d. 19.3), and from another
thousand replications the number of features shared by A and C was
772.4 (s.d. 14.8). From this one confidently concludes that
langauges A and C are more closely related than A and B - until you
also discover that B and C are more closely related than A and B!
Secondly, the method assumes that retentions are uncorrelated, in
particular that knowing that language A has retained a feature tells
you nothing about the relative probability of language B retaining
the feature. This hypothesis is untrue. Consider Swadesh word-
lists: the meanings 'five' and 'you' generally keep their primary
word far better than the meaning 'road'. This will make pairs of
languages that have simply accelerated their rate of change seem more
closely related!
> "SLR could not achieve the correct subgrouping of the Mixe-Zoque
> language group used as a test case."
In other words, the method has already been discredited!
Richard.