Re: [FACETIOUS] Please pass the text

Danny Yee (danny@STAFF.CS.SU.OZ.AU)
Wed, 21 Feb 1996 16:01:40 +1000

I wrote
> > I don't know about "early Computer Science", but modern linguistics,
> > using many ideas and techniques derived from computer science, has had
> > immense success in explaining features of natural language.

Thomas Rimkus replied:
> What successes and to what end, career support or real understanding?

Well, that all depends on how interesting you think natural language is.
(And I don't know enough about the sociology of academia to say much
about career support motivation.) You might like to check out Volume
1 of _An Invitation to Cognitive Science_ (which is on Language) --
there's what I consider some examples of good work on natural language
which uses a reasonable amount of formal theory. [ MIT Press 1995,
general editor Osherson, book review to appear sometime ]

> > "context-free grammar" and "stack" and mathematical definitions of
> > things like "information" *are* useful and *do* produce real results.

> *OK*, in formal contexts I suppose, but you left out mu-recursion and some
> other really neat buzzwords which I have not heard used in the same
> sentence in 30 years. Never late than better, I guess. (Oops, but you
> get the picture I suppose, which included a request for information about
> the age and maturity of your department).

My department? I work for an Anatomy department, so its age and
maturity is probably irrelevant to any discussion of modern linguistics.

> 30 years ago, natural
> languge processing and chess playing were in the same dilemna, too much
> stuff to deal with. Now, however, we have chess playing programs built
> on heuristics (Oh my god, another buzzword from the 60's) which can challenge
> the title while natural language languishes in the soup.

Well, natural language is harder than chess. That's one of the
big discoveries of the last 30 years -- it seems trivial now, but a
lot of important discoveries are like that. Personally I suspect that
full natural language comprehension requires strong AI and probably
*implies* the ability to learn chess.

> Could it be that
> formal languages and chess are relatively recent constructs which can be
> approached by formal systems while natural languages pass information by
> patterns which elude formal systems?

Hang on, there seems to be some confusion here. Formal languages *are*
formal systems! I'm not sure about the "recentness" being relevant,
either -- as I said above, I suspect that natural language is just
more complex. I can think of things much older than human natural
language (say the ontogeny of the nematode worm C. elegans) for which
we have fairly complete formal descriptions.

> > It seems to me that postmodernists reject all of this mostly because
> > they don't understand it and they hate the idea that anything they
> > don't understand could actually be so powerful.

> I do understand part of it (will the person who claims to understand it
> all, please stand up), dont hate it, do recognize its
> power in fueling the "machine", and dont think of myself as postmodern.

My comments weren't directed at you personally, though perhaps I
should have made that clear. I just associate postmodernism with
all that "everything is a text" business.

> Modern linguistics with all its "successes"
> has not moved very far. I stand to be corrected, but please give us an outline
> of the major successes in deeper understanding of natural languages which have
> been brought about by your beloved mechanistic approach.

If I get some time I'll summarise some of the examples from the book I
mentioned above. Methinks you are expecting things to happen too quickly,
though -- it took a long time between the beginnings of scientific
(mathematical/mechanistic) study of the movements of astronomical bodies
and an explanation of the perihelion of Mercury, for example. Computer
science and modern linguistics are so young as disciplines that expecting
them to be complete already is a bit unrealisitic.

> It appears that the
> underlying problem in natural languge processing is in pattern recognition,
> something in which parsing and logical analysis have had little success
> in dealing with.

Language processing is a vastly different kind of pattern
recognition to (say) visual recognition of objects. We now have a
good understanding of some of the differences between the two.

> The system to which you have given your loyalty cannot yet
> recognize women and children from soldiers at arms let alone the subtle
> patterns of natural language.

Que? I'm no reductionist, as most people on this list will realise,
though I am a materialist and an (instrumental) realist. My loyalty
is mostly to pragmatism -- if those pushing alternative bandwagons can
make progress, then that's great, and I'll start looking at what they
are doing.

> The problem, as I see it in modern science, is that the bandwidth is clogged
> with bureacracy and supporters of the standard path. If anyone dares to
> question the unspoken assumptions of the basis of some "scientific" area of
> study, the walls start to crawl with a cheerleading mentality.

Well, sometimes, yes. Though I think those sort of failing are far,
far worse in "non-scientific" areas of academia than in the scientific
ones. (At least judging by the situation here at Sydney Uni.)

> > This is the sort of vacuousness I hate about "postmodernism"

> "Hate", uh close-mindedness?!? Hatred, it seems to me, usually stems from
> our own insecurities. Maybe the vacuousness is inherent in the factionalized
> model of understanding you so adamantly embrace? There seeems to be a
> significant amount of pent-up seriousness underlying your facetiousness.

Sure! I wouldn't trouble anyone with anything *completely* facetious.

Humour isn't just for fun, it's a useful epistemological tool!

Danny Yee.