Pierre Abbat:
> Another question: why is the cmegadri broken off from what precedes it,
> instead of just breaking between the cmene and the cmegadri and leaving
> the cmegadri to be found later? What about {MUstelaVIson}
> and {muSTElaVIson}? How
> should they be analyzed?
Because when you make that break you *must* flag the cmegadri part as a
cmegadri (otherwise a pause before the cmene would have been necessary),
whereas a further parsing may change what you thought was a cmegadri
to, say, a brivla ending... which would then invalidate your previously
parsed cmene! (I hope I am clear enough :-)
So: {MUstelaVIson} = {MUste la Vison} and {muSTElaVIson} is
rejected because a brivla cannot end with a stressed syllable.
Note that, while I consider that result correct, I find the error label
quite unlogical: a forward parsing would give an error after parsing
{muSTEla}, saying that a pause is needed before cmene {Vision},
which seems much more palatable: better error messages are another
advantage IMO of a change of the current backward algorithm
for a forward one.
-- Lionel