[lojban] Re: morphology paper announced

["Jorge Llambías" <lojban-out@lojban.org>]

On 2/22/07, Cyril Slobin <slobin@ice.ru> wrote:
> I think the idea behind is that recognition algorithm must be as easy as
> possible and must not require a complex analysis.

That would be wonderful. Unfortunately, the tosmabru and slinku'i tests
make this almost impossible. You can't recognize tosmabru and slinku'i
failure without some relatively complex analysis.

> You argue about {iglu}
> because you SEE that {glu} is not a brivla.

If you think of {.} as an ordinary consonant, {.iglu} is just like a gismu.

> Nobody mistakes one-syllable
> piece of text for a word. What about {adjgadja}? Do you still SEE that
> {djgadja} is not a brivla?

Yes, immediately, because {djg} for me is not a valid initial cluster.

But let's consider {.asprapra} instead. This is harder to figure out, because
{spr} is a valid initial, but then {sprapra} fails slinku'i.

If you think of {.} as an ordinary consonant, {.asprapra} is just like a
lujvo {pasprapra}, and {.adjgadja} is just like a lujvo {padjgadja}.

Detecting that the {.a} in {.asprapra} doesn't break off is no harder than
detecting that the {pa} in {pasprapra} doesn't break off.

> You suggestion is to cut off an initial cmavo
> and then recursively apply word recognition algorithm to the rest.

That's unavoidable. How else do you detect tosmabru failure?


>  On the other hand, my approach is
> to apply a short sequence of simple patterns (specified in brkwords).

I haven't studied brkwords very carefully, but as I pointed out, brkwords
says the string {.iglu} contains a brivla, doesn't it?

mu'o mi'e xorxes


To unsubscribe from this list, send mail to lojban-list-request@lojban.org
with the subject unsubscribe, or go to http://www.lojban.org/lsg2/, or if
you're really stuck, send mail to secretary@lojban.org for help.