RE: [lojban] Re: cmegadri valfendi preti

["And Rosta" <a.rosta@lycos.co.uk>]

Lojbab:
> At 11:05 AM 12/7/02 +0000, And Rosta wrote:
> >As you may have noticed, English doesn't
> >have self-segmentation, yet you converse in it without problems with
> >determining word boundaries 
> 
> I'm a native speaker.  But people seem to have trouble learning 
> English.  And my limitations in learning Russian have been in part one of 
> picking words out.  Generally for me to understand Russian, I have to 
> identify one or two common words in the speech stream and resolve from 
> there.  If there are no familiar words in the speech stream, I cannot 
> resolve Russian into words.  Fluent Spanish I cannot resolve at all even if 
> I do occasionally recognize a word going by 

This is all true, but irrelevant, as far as I can see.

Me: "I think the rules could not be simplified to the point where they 
could be used in realtime, unless the entire morphology was discarded and 
redesigned from scratch."
You: "Since real human beings have spoken and understood Lojban is real 
time [...], this sounds like a post-Apollo claim that the moon is really a 
lump of green cheese 20 miles up."
Me: "As you may have noticed, English doesn't have self-segmentation, yet 
you converse in it without problems with determining word boundaries" 
You: It is hard for a learner to segment the speech stream into words.

Are you trying to argue that the people who have managed to understand
spoken Lojban have segmented the speech stream into words largely or
exclusively on the basis of the phonological segmentation rules? (The
alleged evidence being that learners would not otherwise be able to
segment.) That seems incredibly implausible to me.

> >As I said in messages of yesterday, I strongly suspect that with
> >spoken Lojban, as with natural languages, we rely mainly on pragmatics
> >rather than phonology to disambiguate word-boundaries. I acknowledge
> >that I have had almost nil experience of hearing spoken Lojban
> >while you have had a lot, so I may be wrong 
> 
> It should be easy to test.  Have the random sentence generator generate 
> something long and random but in theory grammatical.  Get someone to read 
> it fluently at varying speeds (a text-to-speech generator would be best to 
> prevent speaker cuing of word boundaries, and see if it can be resolved by 
> a listener 

I approve. And do the same for English, native speakers and foreign
learners, for comparison. Another good test would be to take a 
phonological string that makes good sense on an illegal segmentation,
and is grammatical nonsense on the legal segmentation; then see
which reading people tend to make. If anybody ever does the experiments 
for real, I'll be happy to put my money where my mouth is.

--And.