Print

Print


On Sat, Sep 24, 2011 at 2:16 PM, Logan Kearsley <[log in to unmask]> wrote:
> On Sat, Sep 24, 2011 at 11:52 AM, Gary Shannon <[log in to unmask]> wrote:
>> On Sat, Sep 24, 2011 at 10:23 AM, Mechthild Czapp <[log in to unmask]> wrote:
---
>>
>> Altering the prescribed argument order might make deeper nesting
>> easier or more difficult to parse. That would have to be tested. E.G.:
>>
>> John book cummings poetry OF ABOUT Mary GIVE
>> John Mary poetry cummings OF book ABOUT GIVE
>>
>> John mother OF Mary SEE theater AT PAST
>> Theater Mary John mother OF SEE PAST AT
>
> Indeed; there we see an example of easily-lost-track-of-list problem.
>
>> Perhaps there is a way to ease understanding with some "optimum"
>> canonical argument order for each function word.
>
> That would be neat, but I am sceptical. Recognizing that the plural of
> anecdote is not data, my own experience in trying to comprehend
> languages with extremely simple grammars indicates that the best order
> to maintain interspersion of function words and arguments and to keep
> predicates close to their arguments changes depending on the
> sub-structure of the arguments. Adding case marking, or some other
> means of identifying argument roles that allows us to free up the word
> order, does a lot to address the problem by allowing one to pick the
> optimal order for every individual instance.
>

Given ease of parsing by the human brain as the criteria, is there an
optimum mix of prefix, infix, and postfix structure, and is there an
optimum argument order.

For example, is the infix "'S" in "John 'S mother" easier, harder, or
of the same difficulty as the infix "of" in "mother of John". Clearly
both appear in many natlangs, so probably not. But using postfix (RPN)
notation when the function word is an adjective, although popular
among natlangs (e.g. "casa blanca") seems less optimal since there is
nothing to easily delimit a list of concatenated (or nested)
adjectives. In a prefix structure like BIG( FAT( JUICY( steak))), even
without the parens or commas ("big fat juicy steak') the list clearly
comes to an end the moment some noun is mentioned. (Aside from
adjectival nouns like "auto" and "parts" in "The rich famous auto
parts dealer" But even there, the final noun of the collection clearly
delimits the NP.)

Suppose instead the model is more like a template with some number of
slots that can be filled with the appropriate phrases:

"the happy_____"  (the happy clown)
"_____ went-to _____" (He went to the store.)
"____ prefer _____ to _____" (I prefer tacos to burritos.)
"____ give_____ to _____" (John gives the book to Mary.)
"_____ is_____as____ as _____" (This book is nearly as old as that one.)

Then we can dispense with the whole notion of prefix vs infix vs
postfix and just deal with a catalog of templates.

Parsing is then simply a matter of looking up in our mental data base
a template that matches the sentence in question, and pattern matching
is something at which the human brain has been shown to excel.

--gary