To resolve ambiguities, there is some kind of system state table, or bag of objects. These relate alternative interpretations of natural language statements or factual statements about a model of reality. A specialized natural language parser such as a link grammar parser may supply facts derived from different interpretations of a sentence. need to explore the wrinkles of context-sensitivity and go over problem spots.
Each word can only be of a fixed number of possible types of syntactical elements, eg: "the"=definite article, "trace"=n. or v.
Yet, it is connected with relates to participates with other words in the sentence; alternative views connect different objects.
Nouns and noun phrases become objects and attributes. Verbs and verb phrases become messages defining and manipulating objects and attributes.
Concepts of Agent and Patient case-Roles figure strongly in the resolution heuristics.
Fuzzy associations would select instances or classes of objects:
. Take any ball and describe it to me. > Object/Class "Ball" has no attributes defined. . All balls are round, three-dimensional and have a size. > ok. . Most balls are solid, coloured, and have a smooth texture. > ok. . Some balls are attached to things. > ok. . Some balls roll, some bounce, some float or fly through the air. > ok.
These definitions are fuzzy because they are incomplete, ambiguous, or assign a partial set of attributes to some members of the class, or attributes whose properties are partially defined, or hidden.
A large library of object classes and generative grammar rules would be required to implement a system that could have even crude object manipulation capability.
. Solid objects have a mass. . Coloured objects can be seen.
Object-manipulation as an activity would need to anchor on some aspect of the external universe, so that the system could "discover" things about objects. To be completely self-contained is limited and boring.
. Take any ball. > ok. . Throw it to Clara.
Contextual state would enable "it" to refer to the selected ball. There are many special objects.. "me", "you", "it". The system could have to ask the operator questions to help resolve ambiguities.
. Bring the ball to me. > which ball? . Any one. > ok. * Fraser brings you a ball. . /me takes the ball. * Shields takes the ball.
A heap of contexts must be managed for a full conversation. Scoping rules exist in prose (sentence, paragraph, parallel structure, etc.) to help choose among them in order to disambiguate but free speech is fast and loose.
. give it to me. > you already have the ball. . what did she tell you? > Clara said nothing.
What set of rules is necessary and sufficient to define physical objects? non-physical?
Fractal structure could give an object infinite complexity. Want the system not to try to give a complete description of an instance of an object.. and therefore to be self-limiting.
System must keep references to the origin of the knowledge..
> Warm fuzzy's are cuddly. . Who told you that? > Paul did.
Is it sentence-argument or sentence-punishment? Each has its own set of links.. its own mechanisms, structures, usages, as if each is really a separate and distinct word.
Is "execution" fatal or destinic? Ask the executor.
To say "this is nonsense", or "this makes no sense", says that the thread of meaning is too thin to see.. therefore cannot be traced.. but does not say there was no thread there in the first place, whereas an absurdity speaks of things that clash with world-view.
Religion is, to you perhaps, nonsense, but to me, sublime.
So different structures can exist with the same name. Which says name-collisions are invited rather than shunned. In fact in poetic usage, such things are welcomed openly. "it" could refer to any structure. "she" could refer to any female structure. How to choose which structure? Syntax gives references, but need these references to stay alive from one sentence to the next.