[YG Conlang Archives] > [engelang group] > messages [Date Index] [Thread Index] >
Mike:
> --- In engelang@y..., "And Rosta" <a-rosta@a...> wrote:
> > Mike:
> >
> > I will have to give this more thought -- to think about how my
> > own goals compare to the ones you list. But in the meantime
> > I will add one less personal one, parsability. That is,
> > whether an explicit parsing algorithm can be stated, and
> > if so, how simple is it.
>
> I did not name parsability as a goal, but if I had, I would
> have included it under syntactic non-ambiguity. I also forgot
> to mention morphological self-segregation. This too falls under
> the non-ambiguity category. At least, this is how I would have
> seen things before you posted this message.
It's true that nonambiguity takes in various factors, such as
whether word-internal morphological ambiguity is allowed,
whether lexical self-segmentation is present, whether
homonymy is allowed, and so on. But, as I think you implicitly
acknowledge, parsability is distinct from nonambiguity. -- Well,
we could say that an efficient parser requires nonambiguity,
but one can have nonambiguity without an efficient parsing
algorithm.
> > A goal of Livagian is that the output of the parser is an
> > explicit logically unambiguous representation. Furthermore,
> > the parser cannot look ahead beyond the current word, and
> > nor can it backtrack and alter structure it has already
> > built. (I am slightly simplifying the lookahead contraint,
> > but not misrepresenting the essential principle.) These
> > principles usefully constrain the range of possible syntactic
> > structures, and they also constrain the incidence of
> > phonologically null words. Further, they act as the guarantor
> > of nonambiguity and also of simplicity.
>
> You raise a few issues, some of which I believe I am clear on,
> and others of which I am afraid I am not. What I am clear on
> is that languages can indeed be designed with differing levels
> of complexity with regards to parsibility. The necessity to
> look ahead, the necessity to backtrack, and the overall number
> of parsing rules can certainly vary wildly, with strict head-
> initial grammars generally yielding the simplest parsing algorithms.
Indeed so.
> > A related issue is whether parsing the syntax gives you the
> > logical structure. For example, although Lojban claims to
> > have an explicit and unambiguous grammar, this claim is
> > tantamount to being bogus, or at least fatuous, since all
> > the grammar does is generate the set of well-formed word
> > strings; the structure imposed by the grammar on the word
> > strings does not give you the logical structure of the
> > sentence. The rules for getting from the lexicosyntactic
> > structure to the logical structure have not been created
> > yet. (I exaggerate slightly: the grammar does yield predicate-
> > argument structure, but not much beyond that.)
>
> It is here that I am a little cloudy. Can you elaborate a bit
> more on the distinction between the logical structure of the
> language and the lexicosyntactic structure as you see it?
> Since you mentioned Lojban, would it be possible for you to
> give an illustration in that language, assuming time permits?
Well, for example "ro nanmu cu prami lo ninmu" ("every man
loves a woman") parses with a structure reflecting the
predicate-argument structure, "[ro nanmu] [cu prami] [lo
ninmu]" (I am simplifying here, but not gratuitously
distorting the facts), according to the official grammar.
So nothing in the grammatical representation reflects the
quantifier scope, which in this instance is deemed to follow
from linear order.
But I would have thought that my overall point was fairly
clear, for we have a clear idea of what a logical structure
looks like: it encodes predicate-argument structure,
binding of variables, and scope relations among quantifiers
and connectives.
> I am studying Lojban off and on, and I can't say that I like
> all its features, but the one thing I was willing to concede
> to it was that it was unambiguous (other than a few minor quirks,
> such as the fact that some of the attitudinals are in effect
> mood markers). It would be quite interesting to discover if
> this were not the case.
As I said, this alleged nonambiguity is of rather a fatuous
sort. The grammar assigns to every licit word string exactly
one structure. But it doesn't say what that structure means
(apart from predicate-argument structure). By fiat, it is
held that every sentence has exactly one meaning, but as
I said, the mappings from sentences to meanings have not
been created yet, and it has certainly not been proved that
this is unambiguous.
The whole point of a grammar is to map sentence-sounds to
sentence-meanings. Any so-called grammar that doesn't do that
is not a grammar, and its claims to be unambiguous are
hollow.
It is not asking too much of an engelang that it have a
real grammar, for, thanks to logicians, we already know and
understand the structure of sentence-meanings. So the
goal of the loglang inventor is to take logical structures
as the starting point and work out the most efficient way
of encoding them so as to satisfy the other design goals.
--And.