🚀 go-pugleaf

In article <37530426.0@news.kabsi.at>, Stefan M�hlbacher <e&s@kabsi.at> wrote:
>
>Patrick Juola schrieb in Nachricht <7iu7i3$pl0$1@quine.mathcs.duq.edu>...
>>In article <01bea8ef$3ca8fb30$54c011ac@ws6147>,
>>Stefan MUEHLBACHER <SMuehlba@frequentis.com> wrote:
>>>> So, how can you implement this maxim with computer programs?  i.e. How
>>>> are  you  going  to  represent  these  concepts in  a  computer  in  a
>>>> relatively neutral manner?  Should the computer represent "I'm hungry"
>>>> as "I AM hungry" or "I HAVE hunger", or neither (e.g. "I --> <state of
>>>> being hungry>")?
>>>
>>>of For me it is clear that the neutral manner to represent this sentence
>>>consists
>>>of two global symbols (it does not matter if those symbols consist of
>>>letters or
>>>are just pictograms).
>>>
>>>1.) State of being hungry: So we need a symbol for a relation that can be
>>>used together with one operand: Lets call it SYMBOL_FOR_BEING_HUNGRY
>>>
>>>2.) The other symbol is needed for the person that is in that state:
>>>Lets call it SYMBOL_FOR_PERSON.
>>>
>>>And the relation may be represented like:
>>>SYMBOL_FOR_BEING_HUNGRY (SYMBOL_FOR_PERSON).
>>>
>>>The two sentences "I am hungry" or "I have hunger" are just two ways to
>>>translate the same relation into english.
>>
>>Doesn't quite work, I'm afraid.
>>
>>Presumably you'll postulate a similar SYMBOL_FOR_BEING_TIRED?
>>
>
>You are right, I would postulate this.
>
>>Unfortunately, although the sentence "I am tired and hungry" works
>>well in English with a single verb and compound adjective, there is
>>no (structural) equivalent in French (due to the different verbs.
>>
>>As was pointed out a little bit further downthread, you're using a
>>fairly simple first order predicate logic here(*).  Let me switch into
>>Prolog notation for brevity....
>>
>>The problem is that we need some way to get from
>>
>>is_tired(person).
>> and
>>is_hungry(person).
>> to
>>
>>is_tired_and_hungry(person).
>>
>
>There is a base principle for the abstract representation of knowledge with
>global symbols:
>In this model there are two distinct domains:
>- One is the knowledge representation domain (this is not the communication
>form, it is just used to store knowledge).
>- One is a natural language (similar) domain where syntax will be regarded
>(used to tell knowledge to a human).
>
>In the knowledge representation domain it is normally not a good idea to use
>many predicates in a single expression: So in this domain it would not
>represent
>the knowledge as 'is_tired_and_hungry(person)'.
>It is better to represent the knowledge just as two independent pieces:
>is_tired(person), is_hungry(person).
>In this domain there is no need to merge the two pieces!

The problem is that the division you are proposing is itself a knowledge
domain in need of representation -- as is the knowledge that it is
or is not needed to merge the two pieces.

For example, "I am tired and hungry" implies both that "I am tired"
and "I am hungry."  

*BUT* "A zebra is black and white" does not imply the two individual
statements "A zebra is black" and "A zebra is white." -- most native
speakers would agree with the first, but disagree with the 2nd and 3rd.

So somehow you need to represent what is different about the first
properties (tired, hungry) and the second (black, white).

As I pointed out, on that path lies madness.

>You are right, rules like that are language dependent.
>
>>So the next step would be to break the problem down further; to noted
>>that hunger and tiredness are themselves both atomic properties, to
>>note that
>>
>>has_property(person,tired).
>>has_property(person,hungry).
>>
>>And that these can be conflated to a property list
>>
>>has_property(person,[tired,hungry]).
>>
>>This can capture both the English and French syntax but at the
>>cost of another level of abstraction in the representation.  These
>>representational levels unfortunately add up; unless you're a great
>>deal cleverer than I, you'll probably find yourself afloat in a
>>sea of abstract properties trying to rationalize to yourself
>>what property lamps have that no other objects in the world (including
>>flashlights and campfires) have, because it's necessary to find such
>>a distinguishing property in order to capture the syntax of
>>Chinese classifiers.
>
>
>
>This is a bit like to say: It is true that it is true that it is true that
>paris is the capital of france.

>I see no reason to need that in the knowledge representation domain.

You will when you look far enough.....  That's *THE* fundamental problem
with this approach to KR/MT.

	-kitten

>>
>>(*) Let me also inquire why you choose to make property symbols
>>into functions and object symbols into arguments.  Why
>>
>>is_tired(person).
>>
>>and not
>>
>>person(is_tired).
>>
>
>The reason is that there could be additional parameters like:
>
>is_tired(person,today)
>
>When I write: persion(is_tired,today) then predicates will be mixed up with
>operands.

I still don't understand.  Predicates and operands are syntactic properties
of the representational system you choose -- person(is_tired,today) is
a two-place predictate (person), with two operands.

You could just as easily have 

today(person, is_tired).

My question is why you chose to make verbs (and hence copula/adjective
pairs) into predicates *INSTEAD* of operands.

	-kitten