[lojban] modularity & grammar (was: RE: Re: nai in UI (was: BPFK phpbb)

["And Rosta" <a.rosta@lycos.co.uk>]

jimc:
> On Thu, 1 May 2003, And Rosta wrote:
> > But I opine that a syntax tree is valid only if it is the syntax tree of
> > some sentence, and since a sentence is a pairing of a sound and a meaning,
> > a syntax tree is valid only if it yields a meaning 
> 
> Well, that's where we differ, and I think the difference is important 
> Take for example the distinction between:
> 
>     54 68 65 72 65 20 69 73 20 61 20 63 72 6f 63 6f
>     64 69 6c 65 20 69 6e 20 74 68 65 20 62 61 74 68
>     74 75 62
> 
> and "There is a crocodile in the bathtub".  There is a whole lot of
> processing between the first one and the second.  To demolish a straw man,
> the representation in octets is certainly valid (as a representation of a
> byte string in hex).  You're not going to deny that, are you, despite the
> fact that there's a subtle flaw that invalidates the utterance at the final
> stage of processing 

I don't understand what this example is getting at, or why we differ.
 
> A major lesson of 20th century system design is, you get a lot more value
> for your effort if you make things modular, with well-defined interfaces
> that are not penetrated.  

Sure. In the case of language, phonology seems to belong to a separate
module from syntax, for example. The well-formedness conditions that
apply to phonology and those that apply to syntax seem mutually 
independent. I'm not attacking the notion of modularity; I'm attacking
the notion of a syntax module that doesn't produce meaningful output.

Here's how I think the modules should work:

         phonetic interpretation interface
       ______|____________________________________________
       | phonology module [complex well-formedness rules] |
       |     |                                            |
grammar|     |simple correspondence rules                        |
       |     |                                            |
       | syntax module [complex well-formedness rules]    |
       ------|---------------------------------------------
         ideational interpretation interface

> In our context, lexing and parsing should be independent, 

Desirable but not crucial -- but that's a different discussion topic.

> and both should be isolated from semantic analysis.  

But the well-formedness conditions that operate on syntactic structures
produce syntactic structures that express well-formed formulas. That is
the rationale for having syntactic structures in the first place. Although
some constraints on syntactic structures don't have a semantic rationale,
they don't seem to be functionally modular.

> <phma>
> made a related point in his reply.  In natlangs, meaning "informs" grammar,
> but Chomsky's lesson is that natlangs have a grammar that can be expressed
> with minimal reference to meaning.  

That is the view Chomsky took in 1957, but since the early 1960s he has
being saying pretty much what I say above. Essentially this was because
in 1957 he hadn't applied his ideas to actual language, but by the
early 60s that was being done, and it became apparent that most elements
of syntactic structure (which turns out to be richer than the mere
strings of Chomsky 1957) have ramifications for semantic interpretation.

> Not to say that every natlang conforms
> perfectly to this ideal, but even English comes pretty close.  The Lojban
> grammar was created to refer to meaning only at the level of the syntactic
> category of the various words, and I think that design principle should
> remain unchanged 

To me it seems like a gang of computer programmers were let loose on
designing a language. But think back to the very beginnings of Loglan
when briefly it literally was (so I am told) speakable predicate logic.
At that stage it was perfectly transparent that the output of the
syntax was a well-formed logical formula. When it turned out that that
syntax needed to be made more user-friendly, what should have been
done was to preserve the essential conception of the design, but to
make much more complex the mapping between strings of phonological
words and the logical formulas that they ultimately express. Instead,
the mapping to logical formulas was turned into something vague and
a new 'pseudosyntax' was developed.
 
> > I contend (a) that if we are dealing with language then we have to
> > engage with this notion of meaningfulness, since language is intrinsically
> > meaningful -- if you don't have meaning then you don't have language --
> > and (b) that a logical language like Lojban ought to spend a lot of
> > energy on firming up the definition of meaningfulness (e.g. a well-formed
> > logical formula, augmented by extralogical stuff like attitudinality
> > etc.) 
> 
> Oh, yes, certainly, I agree with you 100%, and I think that there hasn't
> been enough attention paid to the last (and hardest) step, semantic
> analysis.  But General February advises, don't split your force, or
> conversely, each module should deal with only the task assigned to it, and
> not fritter away its unity and effectiveness dealing with the tasks of
> other modules.  Specifically, the semantic analyser depends on the parser
> realizing the agreed-on grammar, and that's all the parser should do.  The
> parser should not be called on to recognize a misplaced kau; that
> positively ought to be left to a later stage, semantic analysis 

At least we're not talking at cross purposes. My difficulty with your
view is that if you ignore the contribution that kau makes to a logical
formula, then there are no nonarbitrary criteria for whether it is or
isn't misplaced. You can fabricate some criteria, which is what Lojban
has done, but they are arbitrary and redundant. To use your terminology
above, we need the semantic analyser but the parser is redundant --
everything it does is actually part of the job of the semantic analyser
(or whatever you want to call the module that maps from phonological
strings to logical formulas).

--And.