Knowledge Representation in 1960’s Networks and Meaning

1. Lectures on FramesArtificial Intelligence (CS 364)Khurshid Ahmad,Professor of Artificial Intelligence

2. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • The problem with the nets had been the interpretation associated with the nodes which in turn relates to the two problems of 'logical' and 'heuristic' adequacy. There are five major area of concerns here: • • First, what does or should the node represent: a class of objects or does the node represent an instance of an object?   • • Second, it is not clear whether the nodes represent the canonical instance of a concept or does the node represent the set of all instances of the object. • • Third, the semantics of a link that define new objects and a link that relate existing objects, particularly those dealing with 'intrinsic' characteristics of a given object. • Fourth, how does one deal with the problems of comparison between objects (or classes of objects) through their attributes: essentially the problem of comparing object instances: • Fifth, what mechanisms there are to handle 'quantification' in semantic network formalisms

3. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING The above five problem lead to the conclusion that the semantic representation is beset by the twin problems of logical and heuristic inadequacy: •Logical inadequacy: A semantic network is representationally inappropriate because the semantic nets could not make many of the distinctions, even pretty simple logical systems can make: between a specific instance of an object, a class of objects, all objects, no object, some objects, etc. • Heuristic inadequacy: Semantic networks do not contain the knowledge which helps in searching a given network

4. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames The schema for the psychologist Otto Selz is a network of concepts for 'guiding the thinking process', and for the experimental psychologist Fredrick Bartlett it was an active organisation of past experiences and reactions used in thinking and in perception. In its later rendering this notion of schema was taken over by AI researchers during the 1950's and 1960's as a basic building block for organising, storing and retrieving knowledge. There are two dominant and interlinked themes to be found in the knowledge representation literature of that time.

5. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema and Frames • The term frames appears to have at least five senses in The Oxford Dictionary of Computing: • The total amount of information presented in a display at any one time. • …………………………………………….. • A frame is a list of named SLOTS. Each slot can hold a fact, a POINTER to a slot in another frame, a RULE for deriving the value of the slot, or a PROCEDURE for calculating the value. • Frames can be used to represent the knowledge about a particular object or event

6. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames Consider the following objects and events: 1. Bill is a cat; 2. Opus is a penguin 3. The year 2000 flood in Chichester 4. Posh and Beck’s wedding.

7. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames Living objects 1 & 2 can be described as follows: member subset Opus Penguin Birds subset Animal subset Bill member Cat subset Mammals

8. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames Events 3 & 4 can be described as follows: a kind of a kind of Chichester Flood Flood Disaster a kind of Event a kind of Posh and Beck’s Wedding a kind of a kind of Celebration Wedding

9. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames Penguin A semantic network member likes Opus Bill A frame network Penguin member likes Opus Bill

10. Frame Name Value #1 Slot # 1 Value #2 Slot # 2 Value #3 Slot # 3 Value #4 Slot # 4 KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • A frame is a knowledge representation formalism based on the idea of a frame of reference. A frame carries with it a set of slots that can represent objects that are normally associated with a subject of the frame.

11. Animals subset 2 legs flight Yes lays eggs KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • A frame is a knowledge representation formalism based on the idea of a frame of reference. A frame carries with it a set of slots that can represent objects that are normally associated with a subject of the frame. Yes

12. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Note that there are two types of frame: • Class Frame • Individual or Instance Frame • They identical in construction, but a class frame describes entire classes like Progress Report or wedding, whilst individual frames describe individual objects or events like Progress Report No.6 or Sharon Bloggs wedding with John Smith

13. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames: Representing Instances Birds Penguin subset subset 2 Legs flight No flight Yes Animal Opus subset member vitality Yes likes Bill flight No

14. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Schema and Frames: Handling Exceptions Mammals Cats subset subset 4 Legs Climbs trees Yes Feeds young Yes Bats Animal subset subset vitality Yes Legs 2 flight No flight Yes

15. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema: The origins of Frames • Frames were originally proposed by Marvin Minksy (http://www.media.mit.edu/people/minsky/) in its current computational manifestation comprising slots, values and fillers. The origins can be traced back in Britain to the late Sir Fredrick Bartlett FRS, a Cambridge psychologist. • Bartlett talked about schemata (singular: schema) comprising structured groups of concepts which constitute the generic knowledge about events, scenarios, actions, and objects that has been acquired from past experience. • SAITO, A. (Ed.) (2000) Bartlett, Culture and Cognition. Cambridge: Cambridge University Press.

16. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema: The origins of Frames • Bartlett argued that when someone attempts to understand and remember a story he or she, perhaps, reconstructs the story to fit in with expectations based on their prior knowledge and past experience: • When you enter a room you expect to see some furniture – household or office furniture for instance – but seldom you expect to hear oceans or lions roaring in the room • When you go to a shop you expect goods/services on sale and somebody (or some machine) to transact business.

17. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema: The origins of Frames • Schemata can be used to recognise that the capital letter ‘A’ has two vertical lines, joined together at an angle, and a horizontal line, roughly half the length of the vertical lines, connecting the two. The schema for the letter ‘A’ helps us to recognise the following different forms • A A AAAAAA A • We have a similar schema for the small ‘a’: • aaaaaaaa a

18. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema: The origins of Frames • Schemata can be used to recall theories of physics (apples falling to the ground), facts of chemistry (Periodic table arrangement of elements), events in history (World War II), happenings in society and so on. • A schema can be linked together onto other schemata:

19. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Schema: The origins of Frames • Schemata can be organised hierarchically: High level schema subsuming the more specific lower level schemata

20. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Frames • A frame is a knowledge representation formalism based on the idea of a frame of reference. A frame carries with it a set of slots that can represent objects that are normally associated with a subject of the frame. • The slots can then point to other slots or frames. That gives frame systems the ability to carry out inheritance and simple kinds of data manipulation. The definition of an object is much more generic in the context of frames than, say, that of nodes in a semantic network: "an object is an instance of a concept if it resembles the characteristic prototypes of c more closely than the prototypes of concepts other than c" (Sowa, 1985:17).

21. Frame Name Value #1 Slot # 1 Value #2 Slot # 2 Value #3 Slot # 3 Value #4 Slot # 4 KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • The use of procedures in a frame system, perhaps, distinguishes this knowledge representation formalism from semantic networks most. Procedure # 1 Procedure # 2 Procedure # 3

22. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Frames • The use of procedures -also called demons in the literature - helps in the incorporation of substantial amounts of procedural knowledge into a particular frame-oriented knowledge base. • These procedures perform a variety of computational tasks, like requesting and reading in data from outside the frame system, over-writing data within the system (or over-riding, for example, default values or read-in values), or outputting messages, that is data, from the system to the outside world. The read-in and write-out procedures maintain constraints on the knowledge represented in a frame system.

23. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Frames The use of the term demon in frame based systems is used to relate and to distinguish frames from semantic networks. Demons are called demons 'because they lurk about doing nothing unless they see the request, read, or write operations they were designed to look for. In contrast to ordinary demons these [.....] demons are entirely friendly.'!(Winston[i] 1992:197). When the demons are inactive then a frame system is very much like a semantic network, it is the invocation of the demons that distinguishes the frames from the networks. [i]Winston, Patrick H. (1992) Artificial Intelligence (Third Edition). Reading (MASS.): Addison-Wesley Pub. Co.

24. Primitive Action System • Frames can be used to represent actions and their consequences. • Using two types of Frame: • Action Frame • State Change Frame • And a list of 14 (+1!) primitives:

25. Primitives

26. Action Frame Primitive Agent Object Destination Result Move_Object Chris £50 Note State-Change Frame Object Destination Pocket Chris’s_Mood Happy Example Consider the sentence: “Pocketing the fifty pounds made Chris happy.”

27. Action Frame Primitive Agent Result Speak HOD Action Frame Primitive Agent Object Destination Move_Concept Chris AI Students One action causing another “The Head of Department asked Chris to deliver the AI lectures.”

28. Subactions Action Frame Primitive Agent Object Destination Subaction Subaction Subaction Action Frame Primitive Object Destination Move_Object Move_Body_Part Chris Fingers £50 note Open Pocket Action Frame Primitive Object Move_Body_Part Hand Action Frame Primitive Object Destination Move_Body_Part Fingers Closed

29. ‘Do’ • Sometimes actions are not described explicitly: “Chris upset John” Action Frame Primitive Agent Result Do State-Change Frame Primitive Destination Chris John’s Mood Unhappy We need to use ‘Do’!

30. Summary of Primitive Action System • Action frames contain a primitive slot which must be filled by one of the 15 primitives. • State-change frames contain an object slot which is filled with an application specific value. • An action frame can be connected to one or more other action frames via a ‘subaction’ slot. • Action frames and state change frames may be connected via a ‘result’ slot • Other slots and slot values are application specific

31. Prototypical Situations • Many situations are prototypical: • Weddings • Birthday parties • Diner parties • Visiting restaurant / doctor / shop /cinema • Lectures • Disasters

32. Prototypical Events • Prototypical situations share a common set of attributes: • A Wedding has: • Time • Place • Date • Bride • Groom • . . . . . . . . • . . . . . . . .

33. Prototypical Events • However some of these attributes are shared with other prototypical situations. A Birthday party has: • Time • Place • Date • Host • . . . . .

34. Prototypical Events • We can use the powerful inheritance capabilities of the frames representation to help represent these situations:

35. Celebration AKO Host Guests Wedding AKO Bride Groom Birthday Party AKO Celebrant Age Event AKO Time Date Place

36. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Frames • Frames, cannot be used efficiently to organise 'a whole computation' for storing the essentials of a conventional knowledge base. For instance, rules and procedures for selecting appropriate rules for solving a given problem have to be stored in formalism better suited to the so-called rule-based programming, like production rules. And, frames have to be interfaced with programs that decide when the goal of a query to a knowledge base has been satisfied. • Frames, therefore, have been used in conjunction with other, less well-grounded, representation formalisms, like production systems, when used to build to pre-operational or operational expert systems.

37. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Frames and CENTAUR • CENTAUR, an expert system for diagnosing pulmonary infection diseases uses a 'hybrid' system, knowledge of the domain is contained in frames. • The frames used in CENTAUR not only store the knowledge of the objects of the domain, like pulmonary disease taxonomy and a causal chain for these diseases, the frame slots were also used to 'embed' rules. • The production rules were regarded as 'values' for a slot and retrieved by the appropriate invocation of the demons (Aikins[i] 1983a). • CENTAUR uses frame-like structures and production rules. Frame-like structures were used to represent prototypes, components and facts. The production rules are embedded in the prototypes. The prototypes contain domain specific information and domain independent information. The latter contains control information related to the execution of CENTAUR.

38. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING • Prototypical Frames • In a psychological experiment subjects were asked to classify 3 objects of different sizes (small or large), shapes (square or circle) and colours (red or green). Furthermore, the subjects were also given information about where to place the objects in relation to a line drawn on a surface. The subjects were asked to use two prototypes for their classification: Prototype A is red, square in shape and is small; Prototype B is green, circular in shape and is large: • Object 1 is green, circular and small and has to be placed on the left; • Object 2 is also green and circular, but it is large and has to be put on the right; • Object 3 is red, square shaped and is large has to be placed on the right. • Show how the objects can be placed in a frame taxonomy and how the objects inherit properties from their prototypical parents:

39. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Frames Frames were conceived to represent prototypical objects or stereotyped situations. The incorporation of frames-related ideas in a number of knowledge representation languages, in natural language processing systems and in theories of semantic and lexical organisation, and their incorporation in a number of knowledge engineering shells, pay tribute to the robustness and the simplicity of the original idea. (Lehrer and Kittay 1992)[ii] •[ii]Lehrer, Adrienne and Kittay, Eva Feder. (Editors). (1992). Frames, Fields and Contrasts - New Essays in Semantic and Lexical Organisation. Hillsdale (New Jersey, USA): Lawrence Erlbaum Associates.

40. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Frames The development of frames started the move away from a totally declarative approaches of how to represent the knowledge of the world or a domain without recourse to procedures that operate on descriptions used in a formalism. Knowledge in a frame system is represented through taxonomies, complete with inheritance and default values that typify taxonomies, and through the invocation of a set of procedures that operate on the descriptions in the taxonomy.

41. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Trees  The simplest hierarchy is a rooted tree. It is an acyclic graph with further constraints that untangle the hierarchy: There is a single general type at the top, and every other type has exactly one immediate supertype

42. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies A hierarchy of types and subtypes of "concepts" is a standard feature of most knowledge representation theories and their implementations, particularly the so-called systems of Semantic Networks.

43. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING

44. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Genetic Hierarchies: Relating couples by marriage;parents to children; and siblings to one another.

45. Polygon Quadrilateral Trapezium Parallelogram Rectangle Square KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Classification Hierarchies:

46. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Disease Hierarchies: Lung Heart Angina Bronchitis TB Angina

47. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Acyclic Graphs Every partial ordering can be drawn as an acyclic graph, a graph with no cycles. Although an acyclic graph has no cycles, it may have branches that separate and come back together again, permitting some nodes to have more than one parent. Such graph are sometimes called tangled hierarchies.

48. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies A Mathematical Description of Hierarchies The term 'hierarchy' usually means "partial ordering", where some types are more general than others. The ordering is only partial because many types are not comparable

49. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Acyclic Graphs & Tangled Hierarchies

50. KNOWLEDGE REPRESENTATION: 1960’S NETWORKS & MEANING Type Hierarchies Lattices Some hierarchies are ordered in a way that there is a single generic type at the top of the hierarchy, and every other type has exactly one parent y, which is the minimal supertype of x. Contrariwise, there may be "hierarchies: (or more accurately "hetrarchies") where they may (not) be a single generic type, and some (or every) other types may (not) have (more than) one parent y, which is not the minimal supertype of x. Lattice: Unlike trees, lattices may have nodes with multiple parents. But they impose other constraints: every pair of types x and y must have a unique minimal common supertype x >> y and a unique maximal common subtype y<<x. These constraints cause a lattice to look like a tree from both ends. At the top it has a maximal node T that is a supertype of all others, and at the bottom, it has a minimal node ^ that is a subtype of all others. It is also possible to have infinite lattices with no top or bottom.