Re: [lojban] Re: valfendi algorithm

[Pierre Abbat <phma@webjockey.net>]

On Thursday 23 January 2003 15:13, Robin Lee Powell wrote:
> Hoo-boy.
>
> What, exactly, do the proofs demonstrate?

Hopefully, that all well-formed Lojban words can be lexed correctly by the 
algorithm, regardless of what Lojban words precede and follow them, as long 
as the syllables are stressed correctly and pauses are inserted where 
required.

> I hereby offer to attempt to debug any reasonably rigorous proof you
> come up with.

Lemma: All brivla contain a consonant followed two letters later, ignoring
apostrophes, by a vowel. If a brivla contains 'y', it contains such a
consonant after the 'y'.

Proof: If a brivla contains no 'y', it contains two adjacent consonants
in the first five letters, ignoring apostrophes. Find the first vowel
after this consonant cluster. Two letters before it is a consonant.
If a brivla contains 'y', there is at least one rafsi after 'y'. Consider
the last rafsi. Either it is a CVV or CCV rafsi of a gismu, in which case
the first and last letters of the rafsi are the consonant and the vowel
two letters later, or it is the final long rafsi of a gismu or fu'ivla,
in which case, being identical to the selrafsi, it has such a consonant
by the first part.

Theorem: If two lerpoi R and S which both lack 'y' are such that for all i
R[i:i+1] is a valid initial consonant pair, valid consonant pair, valid lujvo
diphthong, fa'u valid fu'ivla diphthong iff S[i:i+1], and R[i] is a vowel,
consonant, fa'u y'ybu iff S[i] is, then R is a valid brivla iff S is,
regardless of whether for some i R[i] is 'n', 'r', or 'l' and S[i] isn't.

The proof of this was discussed on the list, and I think it's right, but 
haven't written it down as a proof yet.

Also some theorems about rafsi fu'ivla and slinku'i. I found a bug in the 
previously posted algorithm: in step 3.E.VI I have to check whether there is 
a consonant cluster in the first five letters (otherwise {lebicyCTIcpi} gets 
broken as {lebi cyCTIcpi}) and whether the part beginning with the first 
consonant cluster is monosyllabic, as well as whether it is a slinku'i or 
begins with a non-initial consonant cluster. There were also some bugs 
causing the program to crash or hang.

Please send me any test data you can think of. This should be Lojban or 
not-quite-Lojban text with words run together and stress indicated, such as

rotu'urselDE'iCUrikteROpu

which produces the output

-ro (tu'u,rse,LDE'i) -CU (ri,kte,RO,pu)

(I'm going to leave off syllabification of output or make it an option, but 
brivlavau must be syllabified to lex them.) Note that the pause after {CU} 
was left out, but it still was lexed. Lines beginning with a number sign are 
comments. You may put commas in weird places to see what happens, or even 
feed it garbage to try to crash it. I'll put the program on the Web soon.

I'd also like any experimental gismu you know of, with rafsi if they have 
them, and rafsi fu'ivla (the program will, when building the cdb, ignore 
fu'ivla that don't have rafsi). I don't need this now; it's for a part of the 
program I haven't started writing yet, which will take a lujvo (including one 
made with fu'ivla rafsi, if given a command-line option to do so) and output 
its tanru.

phma