Definitions In The Knowledge Representation Philosophy Essay
We seem that last few year notation of a prototype used in research of knowledge representation field. Prototype also allows specifying default properties for instance of description, prototype represent also allow overriding or cancelling of properties that apply in some case and don't apply in other cases. There is one belief that makes representation exceptions is easy, but it's make imposable one crucial type of representation which has composite description and its meaning dependent on function of that structure and parts. In this article we try to expiring this and other ramifications of the emphasis on default properties and typical object.
Knowledge representation community spent so many behind dissecting clyde the elephant and his friend but this elephant jokes is last laugh because clyde revenge comes in front of knowledge representation community in from of elephant jokes that is something like
Q: what's the big and gray, has a trunk, lives in the tree
A: An Elephant--- "I lied about the tree"
In above question and answer we laying about properties where elephant live, this is basic feature for other many knowledge representation schemes, mostly semantic net and frame systems comes with inheritance of properties, in this paper we are trying to explain how we can cancelling uniformly applicable properties which is inherited. According to this scheme, it limited us to interpreted node/frame and slot strictly default condition, to exclude definitional condition or contingent universal. To express simple composite description like "Elephant whose colour is gray," "polygon with four sides" using frame, frame requires some definitional capability.
This says that artificially intelligent system gets far with default handle and exception handle, but our aim is try to achieve default oriented frame notations to complete job of knowledge representation that is for frame to represent arbitrary concept of interrelation, this paper say that some of the consequence of succumbing to this temptation
Let us assume that we are talking about basic representation based on notation which has frame and slot, frame means conceptual unit, and slot means role description or relationship among conceptual units, in semantic net conceptual unit are called nodes, but in other system conceptual are called concept,frame,prototype, or units. In this paper we used frame to representative of conceptual unit. All the cases IS-A is principal connective and IS-A make one frame subcategaratization of another frame in short it IS-A make inheritance of properties, such type of cases sub-frame inherit all the properties of parent frame for that we take simple example
Consider the simple frame, "ELEPHANT," which we might portray graphically and lexically as in Figure 1.The notation used in the figure is intended to be intuitive, and the details aren't important, the KR literature tell us that elephant's properties and that something like this elephant are mammals and they are gray colour they have cylinders shape trunk, they have four legs. To representation in knowledge base we used IS-A relation, to represent a frame using ellipse and slot represent using small square, hear isa relation connection between Elephant frame and mammal frame with a slot on elephant or four of them it expressing that elephant has four lags, a slot trunk expressing it has cylinder shape trunk or trunk shape is cylinder, a colour slot say that elephant colour is gray. We might reading this scenario in informal way like this "an ELEPHANT is a MAMMAL that has four legs and a trunk that is a cylinder, and whose colour is gray," or, alternatively, " Every ELEPHANT is a MAMMAL that has four legs and a trunk that is a cylinder, and whose colour is gray."
In above example we show that using properties inheritance we can represent about elephant all properties and we also know that elephant's basic type in animal is mammal so elephant has also some similar properties as other mammal animal. Now we have to represent that some "Clyde" is elephant for that instead of doing same procedure, we can do one this take on frame of Clyde and inherit with elephant frame and make elephant has parent frame for clyde. So Clyde get all the properties his parent frame (Elephant).
For the above scenario is get all properties of parent frame so all instance of that frame get default properties, now we have describe that some elephant colour is not gray due some reason it become yellow for that we have to override the colour properties of default elephant parent frame. Representation schemes that suffer from definitional deficiency tend to try to hide the internal structure of descriptions within node/frame names.
4.0 Literature review
During the last few decades, research in knowledge representation and research in computational linguistics have been getting closer to each other, but in two different ways. On the one hand, the frame-based and object-oriented knowledge representation languages used in AI have widened their grasp on linguistic knowledge: not only domain knowledge has been 'framed', but also syntax, morphology, phonology, and the lexicon (Daelemans, 1987, 1988; De Smedt, 1984, 1990). On the other hand, dedicated linguistic formalisms have been enriched by ideas coming from established work in knowledge representation. The incorporation of inheritance in unification-based formalisms (e.g. Shieber, 1986) is an example of such enrichment .
After this successfully publics this paper some many suggestion and research was done some them are given hear with year and researcher name also given and also given his valuable research in this field. Name of those researchers are Brachman,1983, Brachman, Fikes, & Levesque, 1983; see also Brachman & Schmolze, 1985 for hints on how to integrate definitional representations with typicality statements). In addition, there have recently been many other interesting and relevant developments in representation. For example, with respect to the careful understanding of defaults and the like within semantic net and frame systems, I would direct the interested reader to at least Etherington and Reiter (1983) and Touretzky (1984), and most of the papers in the special issue of Artificial Intelligence on Non-Monotonic Logic (Bobrow, 1980). Elsewhere, Cherniak (1984) has made an interesting empirical argument for integrating both prototypical and deductive reasoning. With respect to the question of just what a "prototype" is, see at least Winograd (1978), Lehnert and Wilks (1979), and Bobrow and Winograd (1979). And for some related criticisms of representation systems and wishful thinking in AI, don't miss McDermott (1981). Finally, in retrospect, I think there are two principal reasons for the kind of confusion addressed here-one that so often seems to accompany knowledge representation work. First, the different sources from which the field has sprung are generally incompatible (as detailed somewhat in Brachman, 1979; see also Winograd, 1978); those using representations as psychological models have very different concerns and criteria for adequacy from those who want special-purpose formal logics with interesting computational properties. Second, there seem to be rampant level confusions throughout the history of knowledge representation, in particular, between what Newell (1981) calls the knowledge level and the symbol level. For example, when discussing what a formal representation system is capable of representing, the notion of inheritance mechanisms really should not come up; this is usually an implementation, or symbol level concern (although it could be a concern of yet a third kind, about readability or maintainability of networks). Conversely, when discussing inheritance mechanisms, LLactive slot-values," and the like, what it is that is being represented is irrelevant. (This accounts for the confusion surrounding Fahlman, Touretzky, and van Roggen (1981), wherein the authors of an elaborate inheritance mechanism went searching for what it meant, and couldn't find a consistent, intuitive interpretation for their mechanism.) The distinctions raised and discussed in this little paper are principally addressed at the knowledge level. For more on this distinction and its impact on representation, please consult Brachman and Levesque (1984), Hayes (1979)) Levesque (1984b), and Levesque (1985).
5.0 How topic related with KRR
We observed that information about a given object we might care about could be scattered amongst any number of seemingly unrelated sentences. With production system rules and the procedures in procedural systems, we have the corresponding problem: knowledge about a given object or type of object could be scattered around the knowledge base. A more representational motivated approach would be to group facts or rules in terms of the kinds of they pertain to. Indeed it is very natural to think of knowledge itself not as a mere collection of sentences, but rather as structured and organized in terms of what the knowledge is the objects of knowledge. Hear we used the term for the data structure used to represent these situations. While the original intended application of frames as a knowledge representation was for recognition, the idea of grouping related procedures in this way for reasoning has much wider applicability.
Reasoning within a frame system usually starts with the system's "recognizing" an object as an instance of a generic frame, basic reasoning in the frame system done in three step
In a frame system knowledge representation declare using object, there by instantiating some generic frame
Any slot filers can be explicitly inherited by new frame and instance inherited
For each slot with filers, if need procedure can be inhered run possibly causing a new slot filed or declare instance for new frame
The inheritance of properties value is done at the time individual frame was created, the local reasoning involving a single frame, when constructing a frame knowledge base, one would also think about the global structure of the KB and how computation should produce the desired overall reasoning. For any reasoning generic frame (In example mammal frame) must be create to solving problem of KB reasoning and representation
6. Work Summery
6.1 Property Inheritance and Necessary Conditions
As we discuss above in this representation system property inheritance is the principal interference. If Clyde is asserted to be an elephant by some kind of an ISA connection from CLYDE to ELEPHANT, then CLYDE would act as if it were a copy of the entire structure at ELEPHANT. In this case we conclude that clyde is mammal andhe has a trunk and it colour is gray. These procedures add property to clyde in virtual of his being an elephant. We corresponds like slot inherent link isa to the if-than statement, if we put under parent frame then it got all properties for any instances thus it implied as universal quantification . Other kind of interfrance can be drawn using network representation, its clear that mostly people think about property inheritance first, when they think of semantic net, they called inheritance network
Necessary condition for property inheritance: one we know that clyde is an elephant then what properties of elephant frame clyde frame can take by inheritance note that typically we do not use elephant properties to whether clyde is elephant but once we connect clyde with elephant using IS-A link than clye become elephant. The main use of frame representation for necessary condition is support with proof in various places in the literature, charniak proposes notation calculation technique using combination frames and predicates in this technique he translate frame into typed universal statement hear using charniak technique we translate clyde frame into universal statement 
Hear principal use of frame would be to determine clyde's properties once (clyde know that he is elephant), similar vein Hayes also interpret slot of frame explicitly in right hand side of condition, according to Hayes view frame represent implicitly universally quantified this concept we understand using one example C- represent conceptual frame, with relationship "R1.......R " this represent as follow
5.2 Is Nothing Sacred?
Clyde and his friend will tell that all instance of frame did not require feature specified in a frame like ELEPHANT from above example. For example among a all elephant some of elephant suffering from bad case of jaundiced, and its colour become yellow, to handle this type of exception frame notation provide overriding or cancelling of properties that would effect to particular instance or frame which inherited. To understand this concept we take one example clyde suffering with jaundiced so it's colour is changed to yellow even though generally elephant colour is gray so as par inheritance rule clye colour is gray but actually clyde colour is yellow, to handle this exception hear we override(Cancelling) clyde colour property. In Fig-2 the crossed line indicates an explicit connection between the new slot and its overridden parent-the so-called "cancel link" provided by some systems
must be inherited from conceptual units, which we call as typical for example clyde colour is yellow Hear we override normal colour of elephant in favor of clyde's special case it means intuitive of ELEPHANT that "every elephant is gray" is not right in some case. Researcher Hayes' and Fikes and Kehler's interpretation are true in general case as par this interpretation elephant has such properties rather than overridden properties it's call default. Fahlman one of the researcher of AI. He given a to node which has default properties as a TYPICAL NODE for example in above example elephant frame (node) which has default properties of elephant it's called TYPCAL-ELEPHANT. This system is not bad, this type of frame is used to represent stereotypical situation, and any notation that allows us represent the exceptions using overriding properties that properties which is inherited with TYPCAL ELEPHANT properties. We have to careful about this thing in representation
The implied meaning of typically, hear is strictly follow the default properties in the absent of any extra information given to the frame "typically" is more closely synonymous with "usually." In the manner used in frame notations, however typical properties are erase for each and every single case! Because in each single case they as par user required they change the property of particular instance of frame which is overridden like clyde and yellow elephant, we do not treat the "usually" issue further, but merely note that it is yet another type of adverb of quantification that we might need to express.
5.3 It Ain't Necessarily So
We used default interpretation of description is the slot notation for the properties cannot be used without alternation to make an unequivocal universal statement, since we hedge on the "every." The minute I assume from the ELEPHANT frame that all elephants are gray, when we declare this interpretation immediately we find a exception in below that frame, and we are trying to make universal quantifier without exception, since there are many domain and situations for this frame hold instances that satisfied without exception, for example, as I look out of my office window at the parking lot, I can easily see that all of the vehicles parked here are cars we can represent this fact using single statement, in fact it might be necessary to represent that how many car in parking lot since I might not have individual representative for individual instance in question, I don't have idea about how may car in the lot it in between fifty to hundreds so I need to find some another way to find about this universal quantifier, we might use one solution for this problem is something like we used default frame notation like above and for every set instance of frame continuously examine instance and check whether instance's description represent universal true or not, for that we create dummy instance for the vehicle I don't know much about individuals but this technique would be useful only in knowledge base where only information was excepted to complete. One of the pro-net researcher suggest a way to explicitly represent exceptionalness properties that is as follows "some fact can be declared sacred (uncancellable) and therefore true of all elephants without exception", if the notation allows cancellation but when we put some important or noting fact as uncancellable as we cannot get universal truths. One is try to go in wrong direction hear rise further difference between contingent (Logic) universal truths (like in our example what's in parking lot) and necessary once (truths of arithmetic) I would hardly lift an eybrow if someone told me that there was now a big rig outside of my office. On the other hand, I would think that something was fundamentally wrong if someone tried to convince me that two and two made five. While this is also an important difference and does lead to on import of cancellation making (e.g., contrast here happen to be no Exceptions" with "it is inconceivable that there could be any exceptions") There is a philosophical gold mine (or endless maze, depending on your perspective) of subtleties here, in which the interested reader can find some wonderful epistemological and metaphysical nuggets.
5.4 Kinds of cancellation
Before going to deep in this topic, let us revised "cancel-link", using cancelling on slot we have to answer this question "What is begin cancelled, the value(like cylindrical trunk) or the attribute(trunk) " example clyde has all the properties of the typical elephant, except that his trunk is hexagonal. Here we're cancelling just the value on the trunk slot. But we might also want to say that Clyde has no trunk, in which case, we want to cancel the attribute. And in which case we have to cancel value of attribute, there may be two possibly for clyde trunk one is clyde lost his trunk and another is clyde trunk lost all shape. But trunk is a physical object so it can't lose his properties of shape or mass, he may able to lose physical parts so the slot cancelling may have to distinguish between the value's being uncancellable and attribute being uncancellable. In the presence of explicitly cancel link may add something for example clyde elephant having no trunk it lose in war, says this is different than "A bird having no trunk" so cancelled attribute show history or basic applicability of property. There are feature possibility comes, One such is illustrated by the proposed representation in Figure-3. Cancellation means not to delete property or attribute permanently instead of this only delete the link between the frame and properties for some time, if we need this properties in feature we using this by giving link with frame to properties via slot.
6.5 Sufficiency and Compositionality: The Bad News Bared
From our reading of frames, and it realize that ISA links we cannot represent in simple continues global that they looks like, one would defendant that even stronger statements like honest-to-goodness definitions are totally out of the question in standard frame systems. In the suspect ion first we get its money but in the second one is dead wrong. For that we investigate further.
For our investigation imagine to define the two possible similar frames, which are Quadrilateral and Elephant. For Quadrilateral its means to it is polygon and it continent four sides. In our approach elephant are mammal and they have four legs. It means not like every four legs mammal are elephant. That is, regardless of whether or not the complex property, "four-legged mammal," is necessary, it not to means it is an elephant. In fact, it is strongly considered that no combination of properties is fulfill the it means to make an elephant in other words, "natural kind" construction cannot be fixed. In the opposite side, there is nothing any story of quadrilateral than four-sidedness on the top of the polygonicity.
Now this dissimilarity may not appear any effects, in fact the differences comes here may seem like logical nit-picking. We would not directly an argument which can be borrowed from language. Those language not continent lexical items of natural language with complete definition of the arguments. We can argue like these, which AI system in that majority of terms of interest to AI only as "elephant", for which critical definition is impossible. In many case we argue like these just add insult to injury. In the above figure we can defined that if any properties are canceled then we cannot represents the necessary condition.
So that, we have to represent that condition from AI languages with strongly nonmathematical cases. Also we had good reasons: Why we worry about definition if the only quadrilaterals we defined. From that we consider that: Once we have the concept of the elephant from that we can construct many numbers of composite concepts. Each of those concepts is relevant to the ELEPHANT which is surely definitional. For example, we consider the elephant with three legs- call our frames for it "ELEPHANT-WITH-THREE-LEGS" is the simple merging of the two attributes, each one is necessary and its pair is sufficient. As we know that it is impossible to elephant with three legs if it is then it is not an elephant. Also it cannot possible with both attributes that it was an elephant and also have exactly three legs. It goes to fail to fall under ELEPHANT-WITH -THREE-LEGS. So it is the irrefutable like ELEPHANT-WITH-THREE-LEGS will be useful. We can make any attributive changes in any frames, which reflect the results of analytical related to frames.
In fact frames like ELEPHANT-WITH -THREE-LEGS which should be a quite correspondent to the QUADRILATERALS. In the furthered document it nicely happened to declare the term which atomically describes the composite description and also its parent happens to be clean, with mathematical concepts. No one will deny that the natural kind parent of ELEPHANT-WITH-THREE-LEGS is significantly different than the non-natural kind of parent (POLYGON) of QUADRILATERAL. But we cannot deny that an elephant with three legs is an elephant. ELEPHANT-WITH-THREE-LEGS and its compositional religious order are more likes QUADRILATERAL than they are like ELEPHANT.
From that we can infer that knowledge representation system to be able to handle any reasonable range of descriptions, also the simplest composites creation from natural kind-like concepts which some type of definitional structuring capability is necessary. We can represent any knowledge inference mechanism worth its salt must be able to inferred -without fail-that a three-legged elephant has exactly three legs. The inner body structure of the nonatomic concepts like ELEPHANT-WITH-THREE-LEGS must be transparent to the system's inference mechanism, or it can't tell if the network developer is lying about the frame's apparent composition. Form the above it's not mean that any lexical items need have definition, only that which should not be to cast any question on the three legs of elephant it being an elephant or vice versa. In many cases many of the frames we needed to reason with consanguine to noun phrases in natural language. It is also silly to assume that we will struck just with the similar of the simple nouns.
6.6 Oblivious to the obvious
It is filling not well to writing about an elephant with three legs. Most frames and semantic net representation are force fully to miss such immediate inferences. The reason behind that the way of the construction. As mention that any property meaning is almost taken as it expense of any serious definition mechanism. The slots which can be overridden can also be used to represent only that properties which typically follow from being an instance of the frames. So that we cannot use the properties in bi-directions. In all the cases we can use the attributes which can follows some property it should be defined after that frames should be used to taken. And also we cannot differentiate to draw the inference-which we cannot confidently conclude that Clyde is a mammal even if we have stated that he is an elephant, because all its property should me cancellable, and we all show that ISA link represent that all elephant are a mammal.
As the result of that we say that system cannot use the structure of the frames to describe the whether the frames are more general or important to the other one. It means that in typical frames system we cannot say about that these frames is more important than the other one. Assume we wants create the frame corresponding to the composite concepts of the polygon with the four equal-length sides. Whatever we infer that or we assume system can blindly trust that. All we can do attach the RHOMBUS frame to the POLYGON frame and also mention that each four in equal size of length, here we can describe those property because those are incidental property of class of objects. It makes worse means when any part from my rhombus system we remove but system infer that it's also there but reality are totally different like concept of three legs elephant.
From above we can say that all of this is primate description. Each frames look like as the complex, but when the system goes to unluckily failure to express real universals and sufficient conditions. In such case the inadequate user always explain the system immediately superdescription of a newly supplied description, even that composite system also same thing should be done. From frames these are-they really can't hold much of anything. In many scenario frames are represented which not to be structured objects at all; their wholes seem to be less than the sums of their parts.
6.8 Hyphens to the Rescue?
Representation scheme suffer from definitional defect for solution of this problem try to hide internal structure of description of node or frame name, for example polygon with four side is compositionally structure description this we would used for concept of quadrilateral, but in actually it just primitive as ELEPHANT. We are assigning any meaning to frame by assign set of atomic symbol, in a system without true compositional structuring; there is no notion of representation by structured correspondence.
We cannot lie about the property which is already known by the system likes rhombus is a quadrilateral. Every composite structure has its own place. Based on that internal structure every frames has own place in the network. Also each frames places are to be fixed in network with respect to other predefined frames. Example we use rhombus then system automatically understand as four side shape but condition is that we are not allow to edit its internal structure. Main burden of the user to classification and also where actually it's placed, it's also helps to defined the knowledge learning and its maintenance. As any point of representation our description have being natural composition meaning but we don't know the how complex it is which we make, unless the system can differentiate between definition and descripti
6.9 Truth or consequence
Node in the semantic net and frame in frame system if both are not able to represent universal often we call as "prototype" because this somehow belong to typical object(TYPICAL-ELEPHANT) Unfortunately, this whole area is little bit confectioning, there two researcher suggest a of these semantic creatures that provide more information and analysis of a part fill with information. We here conclude our little diatribe against prototypes-at-the expense- of-definitions with a few of the (perhaps) less serious consequences.
6.10 When is the typical mammal is not a TYPICAL-MAMMAL
If we classified Natural animal than it look like:
INDIAN-ELEPHANT ISA ELEPHANT ISA MAMMAL
This kind of hierarchy is common in the literature, and looks very clearly reasonable. But not taken in false sense of security; if all of our descriptions arc of typical objects, this one should really read,
TYPICAL-INDIAN-ELEPHANT ISA TYPICAL-ELEPHANT ISA TPICAL-MAMMAL
Now we are facing some problem about relationship:
The typical elephant [some mythical abstract individual] is a typical mammal.
The elephant is the typical mammal. [In fact, the horse or the cow is probably the typical mammal,]
A typical elephant [some real individual] is a typical mammal. Or even, The typical elephant is a mammal. Or An elephant is a mammal.
The odd thing is that none of the prototype based formalism unable to express unless otherwise specified properties of mammals, hear little bit confusion is which ISA relation is first INDIAN-ELEPHANT and ELEPHANT or ELEPHANT and MAMMAL, we thinking as inter-node relation as a ISA relation it specific: It doesn't say whether elephants are mammals or not,; it just says that elephants have mammal-properties unless told otherwise, typical elephant me elephant with all default properties hear we point out fundamental problem of typical node in proto-network: using default node-type, there cannot be independent concept of elephant that is without default properties so we cannot difference between normal sence of elephant and typical elephant. In general it is p robably a good idea to keep "typical" out of the names of our nodes.
6.11 Cancel and Re-cancel.
Let's return to "cancel" links for a moment before wrapping up. Since we can override an attribute-say, the colour of an elephant-why not turn around and override it again? Figure 4 shows a situation that notations that allow cancelling do not rule out. And why not continue to do so, alternating between GRAY and YELLOW to our heart's content? There's nothing to stop this kind of behaviour, since the "implicit universals" aren't really universal!
In this paper author try to represent statement in the form of frame and slot in frame system and using node in semantic net system. Also author trying to handle exception in the frame system and try to represent universal truth for system.
Q: What's big and gray, has a trunk, and lives in the trees?
A: A giraffe-I lied about the colour, the trunk, and the trees.
If we represent this statement in frame system but user can also instances giraffe to elephant properties so its meaning is wrong but syntactically it's true. Author focus on overriding of properties or cancelling of properties, when override properties or cancel the properties it might be given also wrong meaning of sentences. In frame system any one can override the properties of generic frame; it might actually its not right properties for that frame (like elephant and its override properties colour and its value yellow) for this problem we can restrict uncancellable on some property of default frame. We also override or cancel the attribute of frame like elephant trunk attribute cancel by giraffe neck but both properties same, hear we used cancel link and again we cancel link with other link, There are thus three simple ideas to take away from this paper: Frames are typically not very frame-like, definitions are more important than you might think, and cancellation is worse than it looks. If we succeeded here, anyone can give own definition to system by cancellation, so before taking represent think twice.
If you are the original writer of this essay and no longer wish to have the essay published on the UK Essays website then please click on the link below to request removal: