From: Iain Strachan <igd.strachan@gmail.com>

Date: Mon Feb 12 2007 - 18:49:01 EST

Date: Mon Feb 12 2007 - 18:49:01 EST

On 2/12/07, Freeman, Louise Margaret <lfreeman@mbc.edu> wrote:

*>
*

<my original snipped ... see earlier in thread>

*> I don't know much more than basic statistic for psychology majors, so I
*

*> can't evaluate the example you give, but I can understand simplifying a
*

*> mathamatical model to make calculations workable. I think the key to your
*

*> example is that the model works in practice. Having to assume that the
*

*> marine reptiles you are studying lived, breathed then went extinct
*

*> 64,990,000 years before you believe the planet they lived on was created
*

*> does not strike me as a practical, real-world solution.
*

*>
*

Louise , I agree with you that this assumption looks pretty silly,

and I am not trying to defend YECism, but it's a question only of

degree, not of concept.

And though as I say it looks silly, let me elaborate a little on what

the assumption behind the Hidden Markov Model means in practice,

because that looks pretty silly as well. Imagine that you are trying

to get a machine to recognise the spoken word "cat". The time when

you say the phoneme /a/ is one of the so-called "states" of the model.

The mathematics of the Hidden Markov Model assumes a distribution of

the length of time the /a/ sounds for in the word is exponentially

distributed. What does this mean in practice? It means that if you

took lots of examples of people pronouncing the word "cat" and

measured the length of time it took to say /a/ that the distribution

would be such that it turned out to be more likely that it took 10

milliseconds than 100 milliseconds, and more likely still that it took

1 millisecond than 10 millisecond, and more likely that it took 1

microsecond than 1 millisecond. In other words, the shorter the

duration the more likely it is. That clearly is a very silly

assumption indeed - no-one pronounces the word "cat" where the "a"

lasts only 1/100th of a second, and yet we build a model based on the

idea that it's more likely to last 1/100th of a second than a 1/10th

of a second.

The parameters of the Hidden Markov Model are chosen to give the most

probable set of values, given these assumptions, and the most probable

set are computed by a strict mathematical algorithm that can be

rigorously proven to maximise the probability, given this set of

assumptions, but we know perfectly well that the assumptions are plain

wrong.

Yes, it works in practice - who knows, maybe a YEC scientist will come

out with some predictions from their equally silly looking model that

also work in practice. I doubt it, but it can't be discounted.

To summarise my point, however - we often do research based on

assumptions that we know are wrong, so one can't really criticise Ross

for that.

Iain

To unsubscribe, send a message to majordomo@calvin.edu with

"unsubscribe asa" (no quotes) as the body of the message.

Received on Mon Feb 12 18:49:46 2007

*
This archive was generated by hypermail 2.1.8
: Mon Feb 12 2007 - 18:49:46 EST
*