Frames

Semantic Networks and Frames

Reference: Bratko ed. 4, section 15.6 / page 360-

Aim:

To describe semantic networks and frames. Semantic nets are a simple way of representing the relationships between entitities and concepts. Frames can do the things that semantics networks do, but take a more object-oriented type approach. Frames allow procedures called demons to be attached to their slots greatly increasing the power of this knowledge representation method.

Keywords: ako, default facet in a frame, demon, if_added demon, if_removed demon, if_replaced demon, if_needed demon, if_new demon, range demon, help demon, cache facet, multi-valued facet, facet, frame, generic frame, inheritance, instance frame, isa, procedural attachment, semantic network, slot

Plan:

commonsense reasoning using defaults and inheritance
semantic network examples
quantification and semantic networks
frames, slots, procedural attachment, demons
cylinder example
person, measure, ph_frame example

Logic and Rules

So far we have been dealing with rule-based systems. These have the following attributes:

Symbolic representation
Inference mechanism allows conclusions to be drawn from facts and rules.
Certainty factors allow rules to deal with uncertainty. [These will be covered later.]

Common Sense Reasoning

However, rules are cumbersome as a way of encoding relational knowledge and knowledge about objects.

People have vast background knowledge to cope with everyday situations.
We don't have to be told everything explicitly because we can call on the background knowledge.
We use `default' knowledge to handle situations where knowledge is incomplete.
This is called common sense reasoning.

Defaults and Inheritance

Defaults and inheritance are ways of achieving some commonsense:
Inheritance is a way of reasoning by default, that is, when information is missing, fall back to defaults.
Semantic networks represent inheritance.

Using Inheritance

To find the value of a property of e1, first look at e1.
If the property is not attached to that node, "climb" the isa link to the node's parent and search there.
- isa signifies set membership: ∈
- ako signifies the subset relation: ⊆
Repeat, using isa/ako links, until the property is found or the inheritance hierarchy is exhausted.
Sets of things in a semantic network are termed types.
Individual objects in a semantic network are termed instances.
Here is Prolog code for doing inheritance with semantic nets.

Examples of Semantic Networks

State: I own a tan leather chair.

Event: John gives the book to Mary.

Complex event:Bilbo finds the magic ring in Gollum's cave.

Frames

Frames implement semantic networks.
They add procedural attachment.
A frame has slots and slots have values.
A frame may be generic, i.e. it describes a class of objects.
A frame may be an instance, i.e. it describes a particular object.
Frames can inherit properties from generic frames.

Minsky

A major contributor to the invention of frames is Marvin Minsky, a founder of AI and of the MIT AI Lab. He is now is Toshiba Professor of Media Arts and Sciences and Professor of Electrical Engineering and Computer Science at MIT. He won the Turing Award in 1970.

Marvin Minsky

Demons

Demons are attached to slots to cause side effects when the slot is accessed.

if_added: demons are triggered when a new value is put into a slot.
if_removed: demons are triggered when a value is removed from a slot.
if_replaced: is triggered when a slot value is replaced.
if_needed: demons are triggered when there is no value present in an instance frame and a value must be computed from a generic frame.

More Demons

if_new: is triggered when a new frame is created.
range: is triggered when a new value is added. The value must satisfy the range
help: is triggered when the range demon is triggered and returns false.

Demon-related facets

cache: means that when a value is computed it is stored in the instance frame. condition.
multi_valued: means that the slot may contain more than one value.

A Simple Frame Example

Documentation for the frame language used in this example is available (but the system only works in iProlog: it's pretty unusual for a Prolog system to support frames directly).

cylinder ako object with
  height:
    range      number(new value) and new value > 0
    help       print("Height must be a positive number")
    if_needed  ask
    if_removed remove volume from this cylinder
    cache      yes;
  radius:
    range      number(new value) and new value > 0
    help       print("Radius must be a positive number")
    if_needed  ask
    if_removed remove cross_section from this cylinder
    cache      yes;
  cross_section:
    if_needed  pi * radius of this cylinder ^ 2
    if_removed remove volume from this cylinder
    cache      yes;
  volume:
    if_needed  cross_section of this cylinder *
                 height of this cylinder
    cache      yes!

A More Complicated Example

person ako object with
  name:
    range      string(new value)
    help       print("The name should be a string.")
    if_new     ask
    cache      yes;
  sex:
    range      new value in [male, female]
    help       print("Sex can only be male or female, not ", new value)
    if_needed  ask
    if_replaced print("Are you sure you want a sex change?")
    if_removed print("Are you sure you want the sex removed?")
    cache      yes;
  year_of_birth:
    range      year of current_date - 120 .. year of current_date
    help       print("Invalid year of birth.")
    if_needed  ask("Year of birth")
    cache      yes;
  age:
    cache      yes
    if_needed  year of current_date - year_of_birth of this person;
  parents:
    multivalued yes
    range      new value must_be_a person
    help       print("The value in a parents slot must be a person.");
  height:
    range      10..220
    help       if new value < 10 then
    		  print(new value, "cm is too short."),
    	       if new value > 220 then
    	 	  print(new value, "cm is too tall."),
    	       print("The height should be between 10 and 220cm.")
    if_needed  ask("What is the height of ", name of this person);
  weight:
    range      1..150
    if_needed  ask("What is the weight of ", name of this person)
    cache      yes
    if_added   if new value > 100 then
    		  print("Your ", this slot, " is too high!");
  occupation:
    range      string(new value)
    help       print("The occupation should be a string.")
    if_needed  ask("What is the occupation")
    cache      yes
    if_removed print("I used to be a ", old value, ".")!

measure ako object with
  current_value:
    range      allowable_low of this measure .. allowable_high of this measure
    help       print("The patient is dead!")
    if_new     ask(prompt of this measure)
    cache      yes
    if_added
    	if	new value < expected_low of this measure then
    		replace interpretation of this measure by low
    	else if new value > expected_high of this measure then
    		replace interpretation of this measure by high
    	else	replace interpretation of this measure by normal
    if_replaced replace last_value of this measure by old value!

Comment: The first reference to interpretation of this measure has the effect of dynamically adding a slot to the measure frame. Similarly, references to low , normal , and high dynamically add these as possible values.
end of comment

ph_frame ako measure with
    prompt:		default	"ph level";
    allowable_low:	default	6;
    allowable_high:	default	8;
    expected_high:	default	7.6;
    expected_low:	default	6.5!

'HCO3_frame' ako measure with
    prompt:		default	"HCO3 level";
    allowable_low:	default	6;
    allowable_high:	default	8;
    expected_high:	default	7.6;
    expected_low:	default	6.5!

paCO2_frame ako measure with
    prompt:		default	"paCO2 level";
    allowable_low:	default	6;
    allowable_high:	default	8;
    expected_high:	default	7.6;
    expected_low:	default	6.5!

patient ako person with
  ph:	
    cache		yes
    if_needed	make [ph_frame];
  'HCO3':	
    cache		yes
    if_needed	make ['HCO3_frame'];
  paCO2:
    cache		yes
    if_needed	make [paCO2_frame];
  diagnosis:
    multivalued	yes
    default	[];
  investigation:
    if_new	
    	if	interpretation of ph of this patient = low
    	then	add acidosis to diagnosis of this patient,

    	if	interpretation of ph of this patient = high
    	then	add alkalosis to diagnosis of this patient,

    	if	interpretation of paCO2 of this patient = low
    	then	add hypocarbic to diagnosis of this patient,

    	if	interpretation of paCO2 of this patient = high
    	then 	add hypercarbic to diagnosis of this patient,

    	if 	acidosis in diagnosis of this patient
    	and	interpretation of 'HCO3' of this patient = low
    	then	add primary_metabolic_acidosis
			to diagnosis of this patient!

current_date isa object with
    year:	2005;
    month:	4;
    day:	1!

Summary: Semantic Networks and Frames

Semantic networks start out by encoding relationships like isa and ako but can represent quite complicated information on this simple basis.
Frames organize knowledge around concepts considered to be of interest (like person and patient in the example code).
A frame can be a generic frame (or template) or an instance frame.
Frames also allow procedural attachment - that is, demons can be attached to slots so that the mere fact of creating a frame or accessing a slot can cause significant computation to be performed.

CRICOS Provider Code No. 00098G
Copyright (C) Bill Wilson, 2008, except where another source is acknowledged. Much of the material on this page is based on an earlier version by Claude Sammut. The image of Minsky came from his home page.