[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Cowan's responses to the rafsi report (long)



The following constitutes my first response to lojbab's massive and
much-needed review of the rafsi system.  I will start by making various
points which have occurred to me as I read his 100K of information, and
then render my opinions on each of the proposed rafsi-chain changes.

I wish to make clear at the start that this is in no way a "minority
report".  I had absolutely nothing to do with the rafsi review, and
knew nothing of it -- save the bare fact that it was underway -- until
I saw lojbab's output.  I have spoken to him once by phone since then.

First, I simply do not have much faith in the statistics on which all
the proposals are based.  I do not believe that there is nearly enough
Lojban text, in quantity or in variety, to make sound statistical
judgments about "reduction efficiencies".  It is far too soon to do
that.  On the other hand, I strongly agree that we need to baseline the
rafsi system, and do it soon, since we will soon see book publication.

Originally, I had intended to oppose all the changes on the grounds that
what had already been placed in the public domain, to be used or misused
by people outside the control of la lojbangirz., was de facto baselined.
But on consideration, I decided to evaluate the changes as specified.
In particular, the de facto stability of the rafsi list despite the
addition of new gismu meant that the new gismu had no "fair chance" of
getting rafsi, and many of them had none.  Few, if any, of the new
gismu have been used in text, so probably the review did them no
direct good; however, some rafsi may have been (or may still be) freed
which could be assigned to them.

Casting around for a different basis for making case-by-case decisions, I
considered computing a benefit/cost ratio, figuring the cost of relearning
as 1 per reassigned rafsi.  I actually computed this for most of the
chains, but later decided it wasn't useful data.  As lojbab says, almost
nobody has learned the rafsi anyway.

Instead, I decided to look at "winner-loser pairs".  Not all chains have
these.  A "winner" is a gismu that formerly had no rafsi that would get
one as a result of lojbab's report.  A "loser", obviously, is a gismu
that had a rafsi but loses it (or them) in the new system.  I decided
to concentrate on the chains with winner-loser pairs (sometimes there
is a loser but no winner, or vice versa).  To my mind, a chain with no losers
is a reshuffle of existing stuff whose precise details nobody cares about,
and I support almost all such chains because they do no immediate harm
and nobody can predict their eventual outcome.

It seems to me that breadth of coverage is more important than depth.
No one can say how a loser gismu might come to be used in lujvo that
nobody has thought of.  (In particular, a usage score of zero means
simply that 1) the gismu refers to something not in the seventy-year-old
Eaton data, and 2) the gismu refers to something that I, Nick, lojbab,
and a handful of others haven't talked about in the language yet.)

Another point, procedurally perhaps the most important.  Hitherto, there
has been a policy to assign leftover rafsi to words that can take them
even if there's no (lojbab-type) evidence that those words need them.
I believe that it is time to change this policy.  New gismu are still
being created, albeit very slowly; with the rafsi list unbaselined, it
is no trouble to liberate a rafsi from some existing gismu "which has
never been used".  Once the rafsi list is baselined and published, however,
we no longer know what rafsi have or have not been used.  Deleting or
reassigning a rafsi becomes as impermissible as deleting or reassigning
gismu.

With "useless" rafsi unassigned, it becomes possible to create new gismu
in full confidence that a good rafsi will be available for that gismu;
if not for the top-scoring version spit out by the algorithm, for one of
the top few scorers.  (Many existing and baselined gismu were not the
top-scorers, but were selected either to avoid conflicts with other
"more important" gismu or to provide better rafsi.  As lojbab says, this
process was complete before baselining except for "mleca".)

In addition, the more rafsi there are lying about, the more likely that
a garbled lujvo will be changed into some completely different lujvo.
Keeping redundancy acceptably high -- and it is perilously low in the
realm of 2-term fully reduced (6-letter) lujvo -- demands caution in the
assignment of rafsi.  I do not yet have full information on which rafsi
I think should be stripped from existing gismu, but I have commented
below on some of what I think is appropriate.

This proposal on rafsi assignment represents a CHANGE IN POLICY for
la lojbangirz., and as such must be discussed publicly before baselining.
Please discuss.

On to lojbab's specific proposals.  The syntactic form "broda - brode"
means that in the specified chain, "broda" is the winner gismu and 
"brode" is the loser gismu.

1.  skicu - daski.  Vote PRO.  The obvious lujvo selsku (thing described)
and velsku (description) dominate any possible results from daski -- a
pocket is a pocket.

2 - 6.  Vote PRO.

7.  kalsa - karce.  Vote PRO.  The rafsi is a CVC in either case, and I
find it hard to believe in "car type of thingummy" as a reasonable
concept.

8.  dakfu - dakli.  Vote PRO.  "knife" and "sack" both seem rather useless
in lujvo to me.

9 - 12.  Vote PRO.

13.  cfipu - frinu.  Vote PRO with modifications.  I think that "frinu"
could keep "fin", leaving "finti" with "fi'i".  Supporting lujvo:
fraction-number, fraction-vote (proportional representation),
fraction-acre (a "section" or "hide").

14.  Vote PRO.

15.  Here the choice is between 15 and A; I think that
A is preferable because of the cinmo situation.  There may be lujvo with
citno in first place, but most of these are probably "parallel" lujvo
that could be easily reversed: bear-youngthing instead of young-bear.
Vote CON.

16 - 19.  Vote PRO.

20.  danti - darsi.  Undecided.

21.  The text refers to alternate A, but I assume  this is an error for B.
I think B is too radical, because it proposes "mau" for "cmalu", and the gap
between "cmalu" and "zmadu" is too great.  What about giving "mau" to "makcu"
(mature) and "ma'u" to "cmalu"?  If this were done, I would favor B more.
No vote yet.

22.  stici - snidu.  Vote PRO.  Nora's reasoning convinces me.

23.  snanu - navni.  Vote PRO.  I don't think much of navni in any event.

24 - 29.  Vote PRO.

30.  salpo - sralo.  Vote PRO, although I note that sralo was mentioned
as a "sacred" gismu in the text.

31.  bancu - baxso.  Vote PRO.  Cultural.

32 - 33.  Vote PRO.

34.  kamni loses, no winner.  This one is hyphenation driven.  Vote
PRO, reluctantly.

35 - 37.  Vote PRO.

38.  Vote PRO, except that I think "rog" should not
be assigned to "romge".

39.  Vote PRO, except that I think "goi" and "go'o"
should be saved for something better.

**** End of Section 1 ****

40.  pante - palta.  Vote PRO.  I think the statistics seriously
underrepresent pante, which is a new gismu.

41.  minra - jmifa.  Vote PRO.  I argue elsewhere that "minra" includes
English "echo".

42.  mapti - matra.  Undecided.

43.   Vote PRO.

44.  ranmi - pambe.  Vote PRO.

45 - 53.  Vote PRO.

54.  kandi - kadno.  Vote PRO.  Cultural.

55 - 62.  Vote PRO.

63.  stace - slanu.  Vote PRO.

64.  rismi - risna.  Vote PRO.  Rice is incredibly important worldwide,
and most "heart" metaphors are malglico anyway.

65.  Vote PRO, except that "so'u" for sorghum is not
justified.

66.  Vote PRO.

67.  sraku - rakso.  Vote PRO.  Cultural.

68 - 75.  Vote PRO.

76.  cutne - macnu.  Vote CON.  Despite the statistics, I think that
"chest" is generally not going to be that useful (most metaphors are
malglico) whereas "manual" gets manual-transmission, manual-operation,
manual-method, manual-labor.

77.  Vote PRO.

78. nimre loses, nothing wins.  "nimre" is a new gismu and probably
has not had a fair shot at usage:  yellow-citrus (lemon), green-citrus
(lime), dim-red-citrus (grapefruit), citrus-tree, citrus-fruit, citrus-
juice, etc.  Vote CON.  A possible reconstruction:  let nimre keep
mre, give mle to merli.

79 - 80.  Vote PRO.

81.  taske - staku.  Vote CON.  Supporting lujvo: ceramic-tile, ceramic-
block, ceramic-cup, ceramic-plate, hot-ceramic, ceramic-sculpture.

82.  sitna - srito.  Vote PRO.  Cultural.

83 - 84.  Vote PRO.

85.  zbani loses, no winners.  Vote PRO, very reluctantly.

86 - 94.  Vote PRO.

**** End of Section 2 ****

95.  dinko loses, no winners.  Vote CON.

96.  senci wins, no losers.  Vote PRO.

97.  Error:  is "mo'i" being given "mov" (which is free) or "mok"
(which is what the chain suggests)?  No vote.

98.  Vote CON, except that "jve" for "je" sounds good.  Otherwise,
unnecessary change.

99.  Vote CON.  Overkill.

100.  Vote CON.  I think decimal digit rafsi should be sacred, because they
appear in the draft lessons -- the only rafsi to be explicitly taught
there.

101 - 103.  Vote PRO.

104.  Vote CON.  See 100.

105.  Vote PRO, except that cokcu doesn't need cok.

106.  Vote PRO.

107.  "clo" is fine, "col" and "co'o" are overkill.  Mixed vote.

108 - 109.  Vote PRO.

110.  "jo'u" is fine.  "jov" is not free unless 98 passes, and I oppose
98.  "jmu" is overkill.

111 - 119.  Vote PRO.

120.  Vote CON.  Overkill.  Nobody will ever use anything but "sel".

121 - 122.  Vote PRO.

123.  Vote CON.  See 120.

124.  Vote CON.  See 120.

125.  Vote PRO.

126.  Vote CON.  See 100.

127.  Vote PRO.

128.  Vote CON.  See 120.

129.  Vote PRO.

130.  Vote CON.  See 100.

131 - 134.  Vote PRO.

135.  Vote PRO, except that "fod" and "foi" are overkill.

136.  Vote PRO, except that "foc" is overkill.

137.  Vote PRO, except that "vag" is overkill.

138.  Vote PRO, except that "diz" is overkill.

139.  Vote PRO.

140.  misro loses, nothing wins.  Vote CON. "misno" in first position
can probably be switched to last position easily.

141 - 146.  Vote PRO.

147.  Vote PRO, except that "sem" is overkill.

148.  Vote PRO.

149.  Vote PRO with modifications:  bolci gets blo only.

150 - 158.  Vote PRO.

158.  Vote PRO, except that "kef" is overkill.

159 - 161.  Vote CON.  Overkill.

A.  Vote PRO (see 15).

B.  Vote CON unless modified (see 19).

C.  Vote CON.

D.  Makes no sense as written, but I take the proposal to be moving
din to dinri.  If so, vote PRO.

**** End of Section 3 ****

Summary of other than PRO votes:  13, 15, 20, 21, 38 39, 42, 65, 76, 78,
81, 95, 97, 98, 99, 100, 104, 105, 107, 110, 120, 123, 124, 126, 128, 130,
135, 136, 137, 138, 140, 147, 149, 158, 159, 160, 161, B, C.  Errors in
this list are superseded by the above detailed listing.

-- 
John Cowan	cowan@snark.thyrsus.com		...!uunet!cbmvax!snark!cowan
			e'osai ko sarji la lojban.