Print

Print


Catching up to do...

On Fri, 14 Jan 2011 08:57:33 -0600, Sai <[log in to unmask]> wrote:

>* direct expression of semantic relationships = harder to be Orwellian
>or obfuscatory, deprecation of paraphrasis (and attendant silly
>excesses of anti-'plagiarism'), clearer overall structure for
>comprehension

Welll... this one they all say.  I don't deny that it may well be possible
to obtain systemic improvements in clarity at the language level, say.  But
I think making it harder e.g. to be obfuscatory at the level of grammar is
essentially futile: if someone _wants_ to be obfuscatory, there will be a
way.  For the language to really prevent them having the resources for it,
I'd expect you'd need to weaken its expressive resources enough that it
probably wouldn't be a language anymore.  (Much like the systems Hofstadter
analogises to Goedel's theorem, or record-player-breaking records.)  

A lot of this rather just seems cultural to me (though that correlates).

>* (not in UNLWS, but in future computer-native ones) interactively
>'fractal' display of detail - so the zoomed out version displays
>overall story, zooming in gives you arbitrary level of component
>detail

I'd say that's basically orthogonal to nonlinearity.  In fact, you know,
it'd be cool if someone wrote one of those in a familiar natlang:  a
paragraph synopsis of a narrative, which, as you zoom in on various points,
starts squeezing new modifiers and clauses and sentences in at that point,
and so on recursively down, giving much the same effect.  

On Sat, 15 Jan 2011 06:37:23 -0600, Sai <[log in to unmask]> wrote:

>Would a nontemporal narrative be weird? Perhaps. But I still think
>it'd be a narrative.

I wonder whether Rosenberg's diagram poems
  http://www.well.com/user/jer/diags.html
are close enough to canonical exemplars of nontemporal narratives.  'Cause
they are plenty weird :-)

>>> What do you mean by 'orthography' here?
>>
>> The 'writing' done live is a kind of speech that unfolds in time.
>> Orthography translates speech into a permanent record that doesn't unfold in
>> time. Recording-and-playback of speech creates a permanent record, but one
>> that unfolds in time, without translation into a different mode.
>
>Ah. Recording-and-playback I think is the native form of a natively
>temporal conversational process. (As in, of ones that are--not as in a
>claim that all conversations are.)
>
>TTBOMK (Alex?) we have no plan for a flattening of that, though as I
>said, it would be possible to indicate this through e.g. the color
>channel, which we don't currently use. (We do use saturation, e.g. for
>cartouches.)

We don't currently use colour, and I thought we meant to avoid making
anything depend on it, for those times when you don't have an arsenal of
coloured pens.  (Therefore I see nothing wrong with using it to mark
speakers when it is available, and otherwise following the monochromatic mode.)

We do plan to have a 2-dimensional retranscription of conversations, though;
such a thing is implicit in the last entry currently in our table of glyphs,
the predicate glossed "converse" there.  

I was envisioning that one might reduce a conversation to a static form for
these purposes by carving it up into multiple cartouches, one for each
contribution of a speaker (to the extent that contributions are well
discretisable).  To indicate temporal sequence one might attach these to a
sparkline in order, if that's geometrically feasible; or else one might let
them overlap each other and use z-index, which ones are covering parts of
which others, to encode time.  

>>> We are completely discounting as desiderata any linearization, reified
>>> spoken mode, etc. This is meant to be, ahead of all other things, a
>>> writing system that takes maximal advantage of its medium.
>>
>> It does appear crucial to your scheme that it unfolds through time, so that
>> while the text itself is not linear, discourse still is. (That's an
>> observation, not a criticism, of course.)
>
>I don't think this is true, as I hinted above.
>
>Certain kinds of discourse, it's true, are temporal--ones that have
>temporally bound back-and-forth.
>
>But, just thinking offhand: suppose for instance conversational
>parties independently compose messages, which are then simultaneously
>merged. There is nothing intrinsically temporal to latch on to here,
>yet the result of such a process could be argued to be a conversation
>(inasmuch as what occurs at most conference panels is 'conversation'
>[ok so I'm a bit cynical]).

Well, there's at least a directed acyclic graph structure to this sort of
composition.  A given contribution can be sequenced after all the
contributions which its composer saw before it and is addressing in it.  (If
you want to insist on thinking of contributions as dynamic and possibly
responding to each other as they unfold, okay; you can still say take my
view on the level of edits or pushes or ...)  And a directed acyclic graph
is really just a possibly hairier version of And's multithreaded type.  

In the generic case, it seems to me that two turns in such a composition
won't have anything to merge unless they invoke the same
discourse-foregrounded entity or situation, in which case they most likely
have a common ancestor that introduced that entity or situation, and then
they can both be temporally latched onto it.  There are special cases, sure,
but this is enough to make me wonder whether such a conversation might be
relatively temporal in practice.  


>> I don't yet see how the grid-based system acts a constraint.
>
>* it wastes space by reserving it a priori for (ishly) semantic role slots
>* it constrains syntax by forcing directionality--as I think Sky
>discussed, syntax is going to be severely stressed as is just from
>problems like graph relaxation and crossing avoidance, so making that
>worse is bad
>* there are (in the naive case) only 8 (compass-oidal) syntactic roles
>something can fill; I imagine one might need far more

Different kind of grid, Sai.  A grid-based system only suffers from these
defects under the further constraint that two graphical elements can only
have syntactic relations if they're closer than some bounded distance. 
There's nothing local like that about Livagian's system.

If you want to compare it to our system, think not of a grid but of a
bipartite graph, whose two kinds of vertices are (1) our glyphs = And's
predicates, and (2) our overlaid binding points i.e. micropronouns = And's
argument values.  Put the glyphs along one axis of the grid, and the binding
points along the other.  Now remove all the lines; instead, whenever a glyph
used to be connected to a binding point, fill in the grid cell indexed by
them with the semantic role that the line used to carry.  That's the
Livagian system, as I understand it. 

I'd had this bipartite graph idea bouncing around in my head for awhile as a
good stripping down of much of language to its essentials, in fact, and had
it in the back of my mind as we broke the ground of UNLWS.  Livagian may
well have been my ultimate, mostly forgotten, inspiration.

>> UNLWS's sparklines and decorations are
>> optional extras, serving as abbreviations for what could otherwise in
>> principle be expressed by the basic glyph+binding point device.
>
>Sparklines could not easily be compressed into static glyphs. They're
>one of the stronger examples of what I mean by fungible symbols above;
>their form *is* data.

Easily, no, but in principle, yes.  It would be more than enough to say "at
time 0, the temperature was 68.3 degrees; at time 0.1, it was 68.2; at time
0.2, it was 68.0; ...".  It's primarily for Tuftean reasons that we don't do
it this way.  

>> A predicate pictogram must essentially consist of a set of binding points
>> such that the identities (i.e. semantic roles) of the binding points are
>> identifiably distinct from one another, and such that the binding points
>> form an identifiable whole that is distinct from the pictogram for other
>> predicates. -- Yes, this is lovely, reducing language to its purest form,
>> with the least possible distortion due to the constraints of the channel.

I'm delighted you think so.  

>I should also just stress, as I think this is likely to be another one
>of those things that's significant to our design and not easily clear
>given its universality in other systems, that NLWS symbols are *not
>pictograms*, in that they are not really "symbols" to begin with.
>
>That is, they are not arbitrary in form, nor are they iconic of what
>they represent IRL (rather, they are iconic of *semantic* properties);

Well, right now a lot of them _are_ iconic of their referent (in shape or
whatever), but it's intrinsic semantic properties we're aiming to represent,
ideally with some iconicity, yes.  

Though, when we have a large enough lexicon and this principle is properly
in play, we'll also have to consider whether brittleness under noise is a
concern.  Glyphs that both make sense in a given context probably share a
lot of inherent semantic properties, and thus look similar, and thus may be
confusible one for the other. 

>they are not self-contained (rather, they are fungible and permeable);

... again, as applications of such features arise.  We don't really have any
yet.  ... Or no, there was one old proposal, to do with locatives.  Whatever
came of it, Sai?


>On Fri, Jan 14, 2011 at 12:12 PM, Jim Henry <[log in to unmask]> wrote:
>> I don't see how a 2D writing system's root pictograms could really be
>> more advantageously 2D than a spoken language's strings of phonemes

Well, the greater iconicity of two dimensions for pictographs is something,
to whatever extent we mean to exploit it.  

>> In a 2D language's compound words, vertical vs.
>> horizontal links between elements might have significance,
>
>We decided against this because it entails a grid, 

No, it doesn't.  Only not having rotational symmetry.  But we did indeed
decide to have rotational symmetry.

> In a 2D language's compound words

It's also not clear whether we'll need to recognise the category "compound
word" at all.  Perhaps we can get away with on the one hand productive
syntax, and on the other hand non-pervasive lexical incorporation-type
behaviours of certain sets of words.  

But, yes, we're using primarily connections by lines so far, not mere
relative position without connection, so there's virgin territory here. 

Alex