Print

Print


On Thu, Jan 20, 2011 at 2:37 AM, Peter Bleackley
<[log in to unmask]> wrote:
> staving Logan Kearsley:
>>>
>>> As a retired computer programmer with an interest in artificial
>>> intelligence, chat-bots, and machine translation, I can use a narrow,
>>> and purely utilitarian definition of "better" by scoring proposed
>>> changes on whether or not, and to what extent those changes would make
>>> the task of computerized parsing and comprehension easier. Difficult
>>> sentences like "Time flies like an arrow, but fruit flies like a
>>> banana." must be made unambiguous and easy to parse.
>>
>> My own experience in this realm indicates that "easy for a computer to
>> parse" and "easy for a human to parse" are only loosely correlated.
>> Care must be taken to not make it too difficult for humans to use
>> while still improving things for the computer (although, reading
>> further, it seems like you're aware of that).
>>
>
> This reminds me of Jeffrey Henning's Fith, an exolang based on FORTH. It's
> an exolang whose grammar is based on a LIFO stack, and is impossible for
> humans to parse in real time, since a LIFO stack simply isn't a natural way
> for the human brain to work. LIFO stacks are pretty easy for computers to
> parse, however.
>
> I do wonder it it would be any easier for alien kangaroos to mentally parse
> a LIFO stack than it would be for humans. It doesn't seem very organic.

I was inspired by that a long time ago. The basic idea can be fixed-up
for easier human comprehension (even maintaining compatibility with
the original rules, up to a certain extent) by tricks like limiting
the stack depth or what classes of phrases are allowed to go beyond
certain depths. Basing a new language on that idea gives you an RPN
grammar, which can be arranged nicely for humans, but can also result
in nested structures that get very confusing. It helps a lot if you
sacrifice a little bit of word order for a little bit of explicit case
marking, so you can move around some of those nested trees.

My overall conclusions are that computers get by better with a smaller
set of rules, but humans have a balance point somewhere where fewer
rules result in at least as much difficulty as more.

-l.