March 1, 2009 Dr. Muhammed Al-Mulhem 1 ICS 482 Natural Language Processing Semantics (Chapter 17) Muhammed Al-Mulhem March 1, 2009

2Dr. Muhammed Al-Mulhem Meaning So far, we have focused on the structure of language – not on the So far, we have focused on the structure of language – not on the meaning. Words have different meaning, depending on the context in which they are used. Words have different meaning, depending on the context in which they are used. We are going to cover : We are going to cover : What is the meaning of a word. What is the meaning of a word. How can we represent the meaning. How can we represent the meaning. What formalisms can be used. What formalisms can be used.

March 1, Dr. Muhammed Al-Mulhem Meaning Representations Were going to take the same basic approach to meaning that we took to syntax and morphology Were going to take the same basic approach to meaning that we took to syntax and morphology Were going to create representations of linguistic inputs that capture the meanings of those inputs Were going to create representations of linguistic inputs that capture the meanings of those inputs But unlike parse trees and the like these representations arent primarily descriptions of the structure of the inputs… But unlike parse trees and the like these representations arent primarily descriptions of the structure of the inputs… In most cases, theyre simultaneously descriptions of the meanings of utterances and of some potential state of affairs in some world In most cases, theyre simultaneously descriptions of the meanings of utterances and of some potential state of affairs in some world

March 1, Dr. Muhammed Al-Mulhem Meaning Representations Meaning representations are representations of linguistic inputs that capture the meanings of those inputs Meaning representations are representations of linguistic inputs that capture the meanings of those inputs They permit or facilitate semantic processing: They permit or facilitate semantic processing: Permit us to reason about their truth (relationship to some world) Permit us to reason about their truth (relationship to some world) Permit us to answer questions based on their content Permit us to answer questions based on their content Permit us to perform inference (answer questions and determine the truth of things we dont actually know) Permit us to perform inference (answer questions and determine the truth of things we dont actually know)

March 1, Dr. Muhammed Al-Mulhem Common Meaning Representations Sample meaning representation for the sentence: I have a car First Order Predicate Calculus (FOPC): x,yHaving(x) Haver(Speaker,x) HadThing(y,x) Car(y) Semantic Net: Having HaverHad-Thing SpeakerCar

March 1, Dr. Muhammed Al-Mulhem Common Meaning Representations Conceptual Dependency Diagram: Conceptual Dependency Diagram:Car Poss-By Poss-BySpeaker Frame-based Representations Frame-based Representations Having Haver: Speaker HadThing: Car

March 1, Dr. Muhammed Al-Mulhem Correspondence Between Representations They all share a common foundation: They all share a common foundation: meaning representation consists of structures composed of sets of symbols. meaning representation consists of structures composed of sets of symbols. Symbols include Objects Objects properties of objects properties of objects relations among objects relations among objects

March 1, Dr. Muhammed Al-Mulhem Two Distinct Perspectives These four representations can be viewed from two distinct perspectives: These four representations can be viewed from two distinct perspectives: All represent the meaning of a particular linguistic input All represent the meaning of a particular linguistic input I have a car I have a car All represent the state of affair in some world All represent the state of affair in some world

March 1, Dr. Muhammed Al-Mulhem Requirements of meaning representations Requirements that a meaning representation must fulfill: Verifiability Ambiguity Canonical Form Inference Expressiveness

March 1, Dr. Muhammed Al-Mulhem Verifiability The systems ability to compare the representation of the meaning of a sentence with the representation in a knowledge base. The systems ability to compare the representation of the meaning of a sentence with the representation in a knowledge base. Example Example Does Herfi serve vegetarian food? Serves (Herfi, vegetarian food) Match this representation with the knowledge base: Match this representation with the knowledge base: If found return true If found return true Otherwise return false. Otherwise return false.

March 1, Dr. Muhammed Al-Mulhem Ambiguity The system should allow us to represent meanings unambiguously. The system should allow us to represent meanings unambiguously.

March 1, Dr. Muhammed Al-Mulhem Canonical Form Distinct inputs could have the same meaning Distinct inputs could have the same meaning Does Herfi serve vegetarian dishes? Does Herfi serve vegetarian dishes? Do they have vegetarian food at Herfi? Do they have vegetarian food at Herfi? Are vegetarian dishes served at Herfi? Are vegetarian dishes served at Herfi? Does Herfi serve vegetarian fare? Does Herfi serve vegetarian fare?

March 1, Dr. Muhammed Al-Mulhem Canonical Form Solution: Solution: Inputs that mean the same thing should have the same meaning representation Inputs that mean the same thing should have the same meaning representation Vegetarian dishes, vegetarian food, vegetarian fare. Vegetarian dishes, vegetarian food, vegetarian fare. Have, serve Have, serve Relations among objects to be identical Relations among objects to be identical syntactic role analysis (e.g., subjects and objects) syntactic role analysis (e.g., subjects and objects) Herfi serves vegetarian dishes Herfi serves vegetarian dishes Vegetarian dishes are served by Herfi Vegetarian dishes are served by Herfi

March 1, Dr. Muhammed Al-Mulhem Inference Consider a more complex request Can vegetarians eat at Herfi? It would be a mistake to invoke the canonical form to force the system to assign the same representation to this request as those of: Does Herfi serve vegetarian food? Both results in the same answer, not because they mean the same thing, but because there is a common sense between what vegetarian eats and what vegetarian restaurants serve.

March 1, Dr. Muhammed Al-Mulhem Inference Inference: systems ability to draw valid conclusions based on the meaning representation of inputs and its store of background knowledge. Inference: systems ability to draw valid conclusions based on the meaning representation of inputs and its store of background knowledge. The system must draw conclusions about the truth of propositions that are not explicitly represented in the knowledge base, but that are logically derivable from the propositions that are present. The system must draw conclusions about the truth of propositions that are not explicitly represented in the knowledge base, but that are logically derivable from the propositions that are present.

March 1, Dr. Muhammed Al-Mulhem Variables for inference Id like to find a restaurant where I can get vegetarian food First observation: First observation: The request does not make reference to any particular restaurant The request does not make reference to any particular restaurant Use of variables since we do not know the name of restaurant Use of variables since we do not know the name of restaurant A representation can be: A representation can be: Serves(X, vegetarianFood) Serves(X, vegetarianFood)

March 1, Dr. Muhammed Al-Mulhem Expressiveness Must accommodate wide variety of meanings Must accommodate wide variety of meanings First Order Predicate Calculus (FOPC) is expressive enough to handle many of the NLP needs. First Order Predicate Calculus (FOPC) is expressive enough to handle many of the NLP needs.