Re: Rafsi assignments (Pt 1)

[cbmvax!uunet!bradford.ac.uk!C.J.Fine]

Generally, I echo Cowan's concerns: I don't think we have meaningful
statistics yet, and so whatever we do is tinkering. At the same time, I
recognise that we want to have the best baseline we can make, so it is
worth considering individual cases. My salute to you doi bab for doing the
work.

My general response is to be conservative. Where there is a good argument for
change, we should consider it (and the statistics, flawed as they are, can
be one input to the consideration), but should err in the direction of not
changing.

I also concur that we should not give extra rafsi to words that don't need
them just because they are hanging around. Keep them back for future use, and
incidentally increase redundancy for the time being.

There is another different measure of rafsi assignments that I have been
thinking of: niceness. This is a purely subjective measure (though I guess
that others will concur at least to some degree) whereby I prefer some
letter-combinations from the gismu to others.
eg. I like rafsi that preserve three consecutive letters from the gismu, and
those that retain the initial letter; I dislike the swapped forms.

On this basis I have given arbitrary scores to patterns of rafsi as follows:

gismu	4 pts	3 pts	2 pts	1 pt

steci	ste	tec	sec	te'i
		se'i
satre	sat	tre	sar	sta
	sa'e
(I have not distinguished tei and te'i, because tei is a nicer word but for
me less reminiscent of the gismu).

I have marked the cycles according to their net change in this score.
This is obviously a much less significant measure than others, even if it should
appear that other people agree with my scoring; but some of the changes appeal
to me quite strongly on this basis.  (Note that, since 'no rafsi' scores 0,
the introduction of a new rafsi into a cycle can cancel out several reduced
nicenesses. Also, I have ignored cmavo rafsi except for CV`V getting themselves
as rafsi, which I count at  4 pts.)

Again, I am not suggesting using this as a measure for evaluating rafsi change,
just as an additional minor criterion to consider in cases of doubt.



So, to detail (format: change no; my score; my score ignoring klugi'u; comment

1 +1	YES - skicu deserves it more than daski
2 +4	YES - bajra deserves baj, and bal is nicer for balji
3 -2 +2	YES - give snada a rafsi and don't worry about either klugi'u
4 +2	SHRUG
5 +1	SHRUG
6 -3	SHRUG	Not sure cmila needs a rafsi
7 0	NO	What about kackla? It may be that kalsa does need the rafsi -
		I don't recall it. Does kancu?
8 0	YES
9 +6	YES	This is a tasty set of changes (though pity about pensi)
10 0	SHRUG	Does dacru really need a rafsi??
11 -2 	SHRUG
12 -4	MILD YES	untasty, but probably makes sense
13 +1	SHRUG	frinu doesn't need 2, but does cfipu need one?
14 0	MILD YES
15 -4	SHRUG	tid and cki are probably good, but is it worth all the hassle?
		I don';t find the alternative any better.
16 0	YES	Give an extra word a rafsi
17 2	YES	ditto
18 -1 SHRUG
19 +2	YES	This would be SHRUG, but the nicer CVC's swing it.
20 +2	SHRUG

More comments to follow next week.