[YG Conlang Archives] > [jboske group] > messages [Date Index] [Thread Index] >
On Mon, Dec 16, 2002 at 05:41:37PM -0000, And Rosta wrote:
> Jordan:
> > On Sun, Dec 15, 2002 at 09:27:39PM -0000, And Rosta wrote:
[...]
> > > My own intuitions do tell me that ka with multiple ce'u will be
> > > rare. But they tell me that coreferential sumti will be frequent
> >
> > I believe that there's enough text already to proove you wrong about
> > that, though
>
> I don't need to look at Lojban text. I know from English how common
> coreference is, and I know that even in my infrequent writing in
> Lojban issues of how best to handle coreference have loomed large.
We're not just talking about coreference. We're talking about
coreference within a ka abstraction at a subbridi level of more
than one. Otherwise there's no problem. Since that case is *so*
incredibly rare (I may even have not ever seen it), I don't think
there's an issue---especially since we can use subscripts or goi
ko'a if it ever happens.
> > > > Anyway; I don't find "goi ko'a" particularly difficult. I would
> > > > use ce'uxipa though
> > >
> > > Think about how your usage and others' handles coreference, and
> > > how much of it requires no glorking. Or, as an exercise, try
> > > writing in a way such that coreference is handled in a glorking-
> > > free way. I'd be interested to see the results and to see whether
> > > you are happy with the mechanisms Lojban provides
> >
> > I very rarely use glorking anaphora. I use goi ko'a, ri, and lerfu
> > sumti the majority of the time. I do use "ra", but not nearly as
> > frequently. I also use vo'a (but vo'[eoiu] not very often)
>
> I'm not sure whether ri involves glorking. What is the antecedent
> of ri in {le mamta be la djan ri}, and in {le mi mamta ri}? And is
In the first it is djan, in the second it is mamta, but in
"le la djan. mamta ri" it would be djan. See below for why.
> the antecedent of {le broda} = {ro da poi cmima le'i broda} {da}
> or is it {ro da poi cmima le'i broda}? I don't know. If those
I don't understand that question.
> questions have settled answers, then ri may be glork-free.
ri is intended to be glork free. The rule is that the referent of
the first "complete" sumti which you encounter when looking from
the ri leftwards, skipping all members of selma'o KOhA, is the
referent of the ri.
It probably is ambiguous about outter quantification, though. I
would just assume a rule of "it is the same as the last quantification
on the the thing refering to the same referent". So in "re le gerku
cu darxi ci ri ri", the third ri would be the same as "ci ri" (but
probably not neccesarily the same 3). This is a minor point though.
> As far as I know, lerfu sumti require glorking. Hang on while I
> check CLL.... Not clear from the book how recentness is ranked
> when sumti are within sumti, nor is it clear whether the antecedent
> is always the recentest sumti containing a name or description or
> whether it can be lerfu sumti ("le nanmu ... ny ... le ninmu ...
> ny"), nor is it clear whether the lerfu sumti repeats the
> quantification of the antecedent... and so forth.
Lerfu sumti are not supposed to require glorking. Xorxes has
indicated that he does not care to follow this rule, however.
The rule for lerfu is just like for ri except that it is based on
the first letter (you can use multi letters for tanru components
though) of the sumti and that it is overriden if someone explicitly
binds the lerfu using goi.
> Vo'V has one ambiguity that Nick intends the BF to resolve (sumti
> in root vs local bridi), and another ambiguity as to whether it
> repeats the quantifier of the antecedent, which ideally will also
> get settled. So ideally, the BF will make it glork-free.
vo'[aeiou] is glork-free, essentially. Knowledge that it is supposed
to be long distance is pretty widespread.
> > > It is not that "goi ko'a" is horrendously difficult, but are you
> > > happy to use it every time you introduce a new referent in a
> > > text or a paragraph or a sentence?
> >
> > Yes. It or lerfu sumti (which are also glork-free)
> >
> > If I am only refering to it once or twice I'll just use things like
> > "ri", "le go'i", etc though
>
> Does {le go'i} repeat the quantifier of the antecedent?
My guess is no, it starts it over with a "ro".
> Anyway, hopefully the BF will remove all this indeterminacy.
There's very little indeterminacy on these gadri. The quantifier
question is valid, though.
> One place where ko'a assignment is often required is in prenexes,
> which are currently underused (that is, we are not yet, I think,
> very careful about which bridi a sumti is quantified in).
I'm not sure what you mean by that.
[...]
> > > Whether it is nei or no'a that is subscripted, one will want a way
> > > to count inwards from the root bridi and outwards from the local
> > > bridi. Even then it won't help for antecedents embedded within
> > > sumti
> >
> > Outwards, yes (that's what no'a does). Inwards makes less sense,
> > and probably isn't worth it
>
> I remember now. I or someone else had proposed no'a xi ro, no'a
> xi da'a, etc. for inwards. The root bridi and local bridi being
> the two salient reference points.
>
> {no'a} is redundant, therefore.
Huh? How's that? I still don't see the use of inward counting
instead of outward. To me (and I *think* the book says this in 18)
no'a xi ro means the outermost bridi.
[...]
> > > You have some notion of an aesthetic for the grammar of Lojban that
> > > I am not aware of. I of course have a powerful aesthetic sensibility,
> > > but it evidently doesn't coincide with yours
> >
> > It's not aesthetic, it's about doing it right.
>
> What criteria do you use for rightness, if not aesthetic ones? You
> can't be thinking of functional criteria, because the 'hack' works.
Lots of hacks 'work'. In software development we still consider
them not to be 'right'. It's not as simple as aesthetics (which
implies to me that it's nonimportant); it impacts code maintainability
and such things.
> > If you weren't
> > working around not changing the grammar/preprocessor, you'd do those
> > things by having replacements at the preprocessor level before
> > things go into the grammar. That way if the grammar is doing
> > analysis of such things the code doesn't need to grok different
> > lexer tokens for KU and such things which are supposed to be purely
> > structural, and will instead just read a "goi ko'a" or whatever
>
> I see. Yes, in an ideal world there'd be no distinction between
> syntactic and logicosemantic structures, and the preprocessor
> would mediate between logicosemantic and phonological structures.
> But we're not an an ideal world, here in Lojbanistan.
Err, this is just a matter of what is being done at what layer of
the parser. It is easier/simpler/etc (esp. if you're doing some
real analysis of what ko'a happens to point to) to implement a kei'a
type thing as a preprocessor step on each line rather than as special
semantics for a particular kind of KU.
[...]
> > > > > Regarding the saving of a single syllable, there is a sense in
> > > > > which the saving of words is desirable, regardless of whether it
> > > > > saves syllables. There are two reasons for this. The first is that
> > > > > since syllable-count was ignored in the design, the best way to
> > > > > do justice to the design, when exploring it through usage, is to
> > > > > ignore syllable-count. The second reason is that one can measure
> > > > > length not only in syllables but also in words (which are the units
> > > > > of input that the parser operates on)
> > > >
> > > > I highly disagree. Unless there are stops between the words for
> > > > some reason it doesn't count. I see no benifit to being shorter
> > > > in terms of words (esp. since a lot of words can get compounded
> > > > together also ("goiko'a", "lenu"). And having a relatively hackish
> > > > shorting cmavo just to save *one* syllable is pretty laughable
> > >
> > > You disagree, but you haven't argued against my reasons, so I note
> > > that you disagree, but see no reason to change my mind
> >
> > You said:
> > - It is desirable to save words instead of (just) syllables, because
> > a) We can ignore syllable length ("like the original design[1]")
> > b) We can count length in words, instead of syllables
> > These are circular, so I don't consider them reasons...
>
> Where is the circularity? The reason for ignoring syllable length
> was given: honouring the design parameters. The reason for counting
> words was given: they're the units of post-phonological processing.
> From the perspective of the syntactic processor, utterance length
> is measured in words; so in that sense a dodecasyllabic lujvo is
> of equal complexity to a disyllabic gismu.
Ok you did say that (in parenthesis; it really didn't look like a
main claim). I don't think this really has anything to do with how
a parser (logically or actually) operates. If anything you could
argue that the size should be measured in kilobytes or something.
But seriously, this is a language for people, and to a human (and
even to a computer parser, technically), the smallest lexical token
consists of the sounds in a syllable, stress information, and whether
there's a pause after it.
> > [1] That the design ignored syllable length isn't even entirely
> > true btw; length in syllables wasn't a *primary* concern, but it
> > wasn't ignored---there's a reason why so many of the most common
> > cmavo are 1 syllable
>
> This is true. They were assigned based on the best guess, sans usage,
> of the optimal assignment. Usage that heeds that assignment will
> tend to entrench it. Usage that ignores it will test whether it
> was in fact optimal.
I've found that overall it's pretty good. The most common cmavo
(le, lo, la, loi, lei, mi, do, etc) are all single syllable.
--
Jordan DeLong - fracture@hidden.email
lu zo'o loi censa bakni cu terzba le zaltapla poi xagrai li'u
sei la mark. tuen. cusku
Attachment:
binqZDO_wFq9F.bin
Description: application/ygp-stripped