Message-Id: <9207140418.AA10830@relay1.UU.NET>
Date:         Mon, 13 Jul 1992 16:31:04 EDT
Reply-To: John Cowan <cbmvax!snark.thyrsus.com!cowan>
Sender: Lojban list <cbmvax!uunet!CUVMB.BITNET!pucc.Princeton.EDU!LOJBAN>
From: John Cowan <cbmvax!snark.thyrsus.com!cowan>
Subject:      Cowan on relative clauses
To: John Cowan <cowan@SNARK.THYRSUS.COM>,
        Eric Raymond <eric@SNARK.THYRSUS.COM>,
        Eric Tiedemann <est@SNARK.THYRSUS.COM>
Status: RO

After having read Colin's thoughtful paper on relative clauses, I started
out by believing that he was right.  Then I thought that his objections were
correct but that his solutions were wrong, based on a few errors.  Finally, I
have concluded that he is wrong almost altogether.

First, I need to correct a minor error in section 1.  "lo sipna noi melbi"
asserts that only those sleepers who are actually referred to are beautiful,
not that all sleepers are beautiful.   As Colin correctly notes in Section 3,
"ci lo sipna noi melbi" is meant to be understood as "[three out of all
sleepers], who incidentally are beautiful".  But "lo sipna" means "su'o lo
sipna", and therefore must be interpreted the same way.

I believe that Colin's main error lies in ignoring the uses of relative
clauses with non-description sumti.  If anything, the use of relative clauses
with da-series variables is even more important.  Colin's proposal to separate
incidental and restrictive clauses, and to place the latter within the scope
of "le...ku", does nothing for "da poi" constructions.  I also believe that
the notion of "restrictive relative clause" is far more semantically deep than
can be reasonably addressed by mere syntactic manipulations, requiring its
own semantic processing module.

First, it seems clear (and Colin implicitly recognizes) that all talk of
relative clauses and phrases can be reduced to "noi" and "poi" only.  The
alternatives are "voi" clauses (which Colin ignores) and relative phrases
with "ne", "pe", "ne" + BAI, "pe" + BAI, "po", "po'e", "no'u", and "po'u".
All of these may be reduced as follows:

	voi -> poi mi skicu fo da poi
	ne -> noi srana
	pe -> poi srana
	ne + BAI -> poi BRIVLA [where BRIVLA is the source of BAI]
	pe + BAI -> noi BRIVLA [ditto]
	po -> poi stici
	po'e -> poi ponse [with additional connotation of inalienability]
	no'u -> noi du
	po'u -> poi du

These transformations are not necessarily claimed to be exact or to work in
all cases, but they indicate the basic mechanism involved.

Now part of the work which eliminated the ziheks as logical connectives,
retaining only "zi'e" as a purely grammatical particle for conjoining
relative clauses, was the showing that the other connectives really connect
the embedded sentences.  So the former sentence

	le nanmu poi xekri zi'a poi blabi
	the men which-are black or which-are white

was equivalent to

	le nanmu poi ga xekri gi blabi
	the men which-are either black or white

(Forethought connection must be used because afterthought connection between
sentences involves ".i" and cannot be used in an embedded sentence.)  This
mechanism only breaks down when the joined clauses are mixed restrictive and
incidental, and in all cases the connective is then "zi'e".  Therefore, we
scrapped all the other ziheks.  "zi'e" connection between clauses of the
same type was retained for convenience and backward compatibility, but is
understood to indicate sentence connection of the embedded sentences.
(This also answers the question of whether multiple restrictive clauses are
commutative and associative, raised in Section 2: they are, because logical
conjunction is commutative and associative.)

Therefore, a semantic processor would take all the relative clauses as a unit
(precisely as the current syntax reflects) and sort them out, logically
conjoining all the restrictive ones into a single "poi" and all the incidental
ones into a single "noi".  Now the question is, how are these compound sentences
to be used to understand the sumti to which they attach?

Here is where reasoning from "da poi ..." comes into play.  Restrictive clauses
have a deep effect on "da"; they do not simply say that in addition to fitting
into its existing bridi "da" must also fit into another bridi; instead, the
>meaning< of "da" is changed from "some object" to "some object chosen from
the universe specified by the {poi}".  This is shown by the fact that
"da" thereafter has a meaning incorporating the restriction: it is not
local to the current sumti, but is pervasive until another "da poi" appears.

By similar reasoning, "lo mi ci sipna",
which means "lo ci sipna [ku] pe mi" exactly, and is roughly equivalent to
"lo ci sipna [ku] poi [ke'a] srana mi", asserts that the number of sleepers
is three within the domain "things associated with me", as opposed to "lo
ci sipna" by itself, which claims that there are three sleepers within the
general (unrestricted) domain.  (In either case, the quantification claim
is incidental.)

Once this domain restriction has been done, the meaning of the sumti can be
evaluated.  At this time, the incidental clause can be understood as applying
to the sumti in its entirety, and making a subordinate bridi (possibly compound)
which is incidentally asserted.  Note that this analysis implies that "ke'a"
means different things within restrictive and incidental clauses: in a
restrictive clause, it refers to the meaning the sumti would have if no
restriction were in effect; in an incidental clause, it refers to the sumti
as-is with any restriction in effect.  Therefore,

	ro da poi mlatu cu mabru
	all things which-are cats are-mammals

has an utterly different meaning from

	ro da noi mlatu cu mabru
	all things (which incidentally are cats) are mammals

which says that everything is a mammal, and what's more, everything is a
cat, too.

I suspect, however, that the current attachment point of "relative-clauses"
is too far down in the sumti hierarchy: the fact that it appears twice is
ipso facto suspicious.  I will make an attempt to do the necessary YACCing
to determine if the connection point can be moved up closer to, or within,
"sumti-3<93>".  External quantifiers should be processed either before
or on the same level as relative clauses.

As others have said, "lo mi ci broda" can't be changed to "lo ci mi broda"
because the latter means "the broda associated with the three of us [ci mi]".
However, the "mi" gets pulled out at an early stage and transformed into a
"[zi'e] pe mi" within the relative-clauses block, so I do not think that
its exact position matters.

I agree that "<quantifier> <quantifier> selbri [ku]" is peculiar, and I do not
really know what it means.  I would not mind seeing it go.  As far as I know,
the only usage has been for "Three out of five dentists" = "ciboi mu denmikce",
which is really a kind of fraction, since it does not assert that there are
five dentists!  Probably "picifi'umu loi" would do better:  three-fifths of
the mass of dentists.

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