Representations for KBS: Semantic Networks 6.871 Lecture 7Outline • Quillian's foundations: associations • Implicit meanings for uniform links • Knowledge-related primitives [eg. CDs] • Concern for semantics of the language • Structured inheritance networks [eg. KL-ONE] • Where the field is today 6.871 – Lecture 7 2Preview • Semantic networks have evolved: – Shift in motivation from modeling cognitiveprocesses to addressing computational issues. – Shift in representation goals from "all human memory" to certain types of knowledge [eg.definitions vs. assertions, classes vs. instances] – Semantics of links have become less intuitive and more formally defined. – Shift in reasoning mechanisms suited to morecareful definitions of primitives. 6.871 – Lecture 7 3What is a Semantic Net? • What’s a net? • What a semantic net? • Where do the semantics come from? 6.871 – Lecture 7 4Questions for Semantic Nets • Regarding the original motivation – How should we view the world? – What are the recommended inferences? • Regarding the representation formalism: – (What) are the(re) primitives? • The primitives of a KR technology are those things “the interpreter is programmed in advance to understand”[Brachman] – What knowledge can we express? – What does a concept mean? • May be what the machine infers • May be a formal answer • Regarding the reasoning mechanism: – What are the easy/automatic inferences? – How efficient can we make these? 6.871 – Lecture 7 5Semantic Memory [Quillian, 1966] • Motivations – Understand the structure of human memory, and its use in language understanding – What sort of representational format can permit the “meanings” of words to be stored, so that humanlike use of these meanings is possible? • Psychological evidence that memory uses associative links in understanding words 6.871 – Lecture 7 6Semantic Memory [Quillian, 1966] • Motivations: – Claim that people use same memory structure for a variety of tasks • Wish to encode dictionary definition of words. • And then: • Comparing and contrasting meanings of two words • Generating quasi-English sentences to describe the comparison 6.871 – Lecture 7 7Semantic Memory Formalism • Plane: A network of nodes and links for representing the definition of a word "concept" • Nodes: – Type nodes: Direct representation of word [one per plane] – Token nodes: Denote a type node in some other plane • Link types – Type node A is a subclass of B – A,B, and C disjunctive [conjunctive] – A relates B and C – A is a token associated with type node A – A modifies B [an “escape hatch”] 6.871 – Lecture 7 8( ) j S Plant 1 PLANT Live Animal 10C1 or or and and and Leaf 12N1 People 12N1 Process In Industry or Get 3 From 3 FoodAir Earth or Plant or or Plant 2 Use For 5 IN 9 = B D Seed Earth Grow Plant 3 Put For 4 = B Plant 2 Plant 3 1. Living structure which is not an animal, frequently with leaves, getting its food from air, water, earth. 2. Apparatus used for any process in industry. 3. Put seed, plant, etc. in earth for growth. = A Structure With 3 17C1 = A = A Water Ob ect 12C1 = A Apparatus 13C1 Person T8C1 6.871 – Lecture 7 Figure by MIT OCW. 9Food FOOD Formor Meal = B Live = B Being 2 = B and or Drink Keep Grow = B Into 1. That which living being has to take in to keep it living and for growth. Things forming meals, especially other than drink. = A Thing Has-To To 7 = A Take 11 = A Other-Than 6.871 – Lecture 7 Figure by MIT OCW. 10Semantic Memory Formalism • Expressiveness: Any word with a dictionary definition • Meaning of a concept: two answers – dictionary definition in its plane. – "full concept": transitive closure of all links • Size ?? • Focus is on nodes: in use links are merely connections 6.871 – Lecture 7 11Semantic Memory Reasoning • Comparing meanings of two words: via “spreading activation” – Intersections in unguided breadth-first search • General purpose • Is this “closest path” the shared meaning? • Describing the comparison: – Trace the links leading to the intersections. 6.871 – Lecture 7 12Spreading Activation (A) (B) (C) P P P P P P P P P P P P P P P P P P P P P P P P ROBIN Robin is activated Robin primes its associates Robin Red Breast is ais a is a Blue Eggs Bird Canary Sings Fly Feathers Animal Skin Breathe ROBIN is ais a is a Blue Eggs Bird Canary Sings Fly Feathers Animal Skin Breathe ROBIN is ais a is a Blue Eggs Bird Canary Sings Fly Feathers Animal Skin Breathe Continued priming from Yellow Red Breast Yellow Red Breast Yellow 6.871 – Lecture 7 Figure by MIT OCW. 13Primitives? • What’s primitive in Quillian? • Why primitives? • Approaches to primitives: – Language independent: Conceptual dependencies – Language [English] dependent: OWL 6.871 – Lecture 7 14Conceptual Dependency • A strongly reductionist approach • Five primitive categories of knowledge – Actions [Eg. Propel, Ingest, Ptrans, Mtrans] – Tenses [Eg. Present, Fast, Future] – Objects [any noun] – Modifiers of actions: case frames [eg. object, subject, recipient] – Modifiers of objects • Combining primitives yields standard scenarios – building blocks world knowledge. 6.871 - Lecture 7 15Example CDs Basic dependency John PTRANS Combining 2 dependencies John PROPEL cart More Complex: John pI John INGEST doo o ice cream 6.871 - Lecture 7 spoon 16Conceptual Dependency • Motivation: Provide a canonical form for world knowledge expressible in any natural language. • Why a canonical form is valuable – Deciding whether two expressions have the same meaning. • If not, how close are they? – Understanding complex text [eg. stories] 6.871 – Lecture 7 17Conceptual Dependencies • Expressiveness: All world knowledge? • Not an intuitive means of communication, for us. 6.871 – Lecture 7 18Links: What Do They Mean? • IS-A – Clyde is-a elephant – Elephant is-a mammal • The World Wide (Non-Semantic) Web – What does a hyperlink mean? • What does that mean? • Eg: books on the web • Need to think about the semantics of the network notation, to minimize the “intuitive” meanings of links – Similarity to semantics in logic sense – Meaning arises from: • what the interpreter does (procedural semantics) • formal definitions 6.871 –