Message-Id: <9208041901.AA19115@relay1.UU.NET>
Date: Tue Aug  4 19:18:17 1992
Reply-To: cbmvax!uunet!math.ucla.edu!jimc
Sender: Lojban list <cbmvax!uunet!pucc.princeton.edu!LOJBAN>
From: cbmvax!uunet!math.ucla.edu!jimc
Subject:      Rafsi assignment comments
To: John Cowan <cowan@snark.thyrsus.com>
Status: RO

Thank you, Lojbab, for the massive effort of reviewing the rafsi
assignments.  The description was readable and comprehensible, and I hope
you get a good response from conlang about it.  The fact that we are
doing this kind of thing gives us credibility.

I am interested to see that John Cowan and Nick Nicholas are voting
in favor of most of the changes, even though there are some specific
items that they disagree with strongly.  It looks to me that they
value conservatism somewhat more than you did.  I look forward to
something quantitative showing what loss of coverage might occur if
their "no" votes prevailed.  Maybe the loss would be small.  I'm sure
we all agree that the "shifting sand dune" effect is something to
avoid.

Several people have mentioned that, although the amount of analysable
Lojban text is a *lot* larger than at any past time, it is still small
and idiosyncratic, e.g. crabs scraping their sides on the wet rock.
Thus one wonders if this corpus is appropriate as a basis for assigning
rafsi.  I would respond:

First, we have to assign the rafsi *now* so they can go in the
dictionary.  The new assignments seem substantially better than the old
ones, even if not optimal.

Second, my experience with lujvo is that they fall in two rough
categories. First are words or phrases which are specific to the topic
of the text, such as "flight path", "corner pocket", "fairy godmother",
"low crawlway" and so on.  In our present corpus of text the wide range
of possible such words will not be covered adequately.  Nick's comment
on a future avalanche of sexist terms using nakni or fetsi is well taken.
However, one can anticipate some broadly productive categories.  For
example, if taxfu-garment is interpreted as "x1 is a garment for body
part x2 of creature x3" then a whole constellation can be expected such
as "neck-garment" (tie), "knee-garment" (knee-pad), "finger-garment"
(those bandage jobbies that don't have an English name), etc. etc.
Even though the corpus may not include much discussion of garments,
it would be valuable to jack up the usage score of taxfu artificially
to anticipate such usage, and similarly where other clusters are
anticipated.

But there is also a wide range of modifications of gismu that are more
generic, such as transitive conversion, becoming, ceasing, and so on.
Just about any gismu whose x2 normally contains an abstraction is in
this category.  These constructions are very common in all kinds of
text, and my feeling is that regardless of the topic, the existing
corpus will exercise the important modifiers repeatedly.  It's this
kind of gismu that particularly needs good rafsi, and which will get
good rafsi in your reassignment.

The Lojban speakers also particularly need to be able to interpret
confidently and efficiently lujvo with modifying rafsi, even when
the modificand is a relatively rare or complicated concept.  But we
have been over this topic before, and it's not on the main line of
rafsi assignment.

I'm still snowed under language-wise, so I'm not able to give a point
by point analysis like Nick and John did.  But if you'll accept a
"warm and fuzzy" vote, mark me in the "yes" column with a bias toward
conservatism.

		-- jimc