Re: place structure of lujvo

[cbmvax!uunet!pucc.princeton.edu!jimc]

Godzilla strikes!  Run for the hills!  The dikyjvo are on the loose!

David Cortesi <infmx!godzilla!cortesi@UUNET.UU.NET> writes:
> ...
> Nick: The x1 of sudga becomes the x2 of sudri'a...[further explanation,
>         leading to]...sudga means is-dry;
>         sudri'a is a single word meaning to dry: x1 dries x2 of x3
> ...
> What I guess I don't understand is the range of possible differences.
> In particular, I am (like the 4-year-old) desperately looking for rules
> that I can apply.  When I meet a strange lujvo, like Nick's "sudri'a,"
> under what rules do I translate it?  If I cannot simply unpack it to
> make a tanru and translate that, what am I supposed to do?

Right on!  What are we supposed to do with a lujvo (or a tanru, for
that matter)?  In JCB's anciently made lujvo there were very definite
patterns such as the one Nick used in sudri'a.  They were never
officially written down and may have been unconscious.  In any case,
many years ago when rafsi were first invented by JCB I needed exactly
the rules that David Cortesi again asks for.  I wrote down what JCB was
doing, and those were the first dikyjvo rules.  They have gone through
many revisions and improvements since then, though there has been a
certain lack of acceptance by the powers that be.

Lojbab has asserted that sometime in the future he will issue
"guidelines" (not rules) for lujvo interpretation, but he thinks that
algorithmically analysing a lujvo to extract a unique meaning from it
is impossible in Lojban's present state, and unwise even if it were
possible.  He sees the following problems with dikyjvo:

1.  In a steady-state natural language the speaker community by
consensus of usage determines the meaning of each word.  The linguists
merely describe.  Introducing rigid rules (prescriptive linguistics)
goes against what are seen as the natural ungovernability of the
language process.

2.  The range of meanings to be covered by lujvo is broad.  The rules
select only a subset of possible meanings; for example, Dave's "dryly
caused" would be excluded in favor of "caused to be dry".  Restricting
ourselves to one meaning per word is too limiting, especially if the
allowed meaning is prespecified via an algorithm.

3.  There are several patterns -- three, in the most recent
formulation.  Lojban is not pre-designed to allow unambiguous
distinction between them.  (Whereas -gua!spi was designed from the
beginning with dikyjvo in mind.)  It would take coordinated language
features to resolve this ambiguity.

In addition, there are two related problems also mentioned by Dave.
First and simpler, there is a natural tendency to interpret a lujvo the
same as its underlying tanru.  Thus if rules are adopted to interpret
lujvo they would naturally be used equally on tanru.  Any tampering
with the florid metaphoricity of tanru is strongly resisted.  (Note
that it is possible to preserve florid metaphor despite rules, by using
a special cmavo to join truly metaphoric tanru terms.)

The second problem is the old nemesis of cleft places.  As an
illustration, take "x1 is a wannabe-soldier".  To make a lujvo for
this, let's analyse it into English phrases headed by gismu: "x1 wants
to be a soldier", or in Lojban, "x1 djica lodu'u sonci".  Now the hard
part: who is a soldier?  Lojbab says that you can pretty well figure it
out from context, which is true, but which sounds to me to be a lot
closer to English and other nasty natural languages than it is to a
"real" logical predicate language.

In my ideal language I would be able to say authoritatively, "x1 of the
containing bridi is the x1 of the abstraction (unless otherwise
specified with an explicit sumti), because that's how you always use
`want'".  Dave asked how the wheat got to be in x1 of "dry", and I'm
saying there should be an explicit rule.  Here's how it works.  (sudri'a
steps on too many other issues, so I'll stick with my own example.)

In wannabe-soldier, the tanru is sonci djica, the lujvo is soidji, and
in both forms you are supposed to interpret the phrase as "x1 djica
lodu'u x1 sonci".  The rule for interpretation comes as part of the
definition of djica -- for this and about 350 other words, there are
particular common patterns of usage where containing bridi arguments
get replicated into abstract sumti (and non-abstract ones too!) in ways
that vary with the individual word.  Actually there are only a few
patterns repeated over and over, but there are enough exceptions in the
place structures I have built that I found it useful to annotate the
words individually.

Lojbab resists such a plan because it designates specific argument
places for specific purposes, thus providing grounds for saying that
"cases" in the Latin sense are recognized in Lojban.  Lojbab has a
"case nihilism" policy (which I rather agree with) that the semantic
role of each place of each gismu is unique; the obvious parallels
within certain classes of places may be observed for learning aids, but
are not recognized formally within the language.  Thus there is no need
to specify as policy just what the semantic role categories (cases)
actually are.  While I understand Lojbab's concern and I agree with his
way of dealing with it for Lojban, I do not agree with carrying the
"head in sand" case policy so far as to sacrifice the very useful
possibilities of cleft places and dikyjvo.

                -- jimc