Interpreting Language (with Logic)

Slides:



Advertisements
Similar presentations
Artificial Intelligence
Advertisements

Predicate Logic Colin Campbell. A Formal Language Predicate Logic provides a way to formalize natural language so that ambiguity is removed. Mathematical.
First Order Logic Resolution
CAS LX 502 Semantics 10b. Presuppositions, take
SEMANTICS.
L41 Lecture 2: Predicates and Quantifiers.. L42 Agenda Predicates and Quantifiers –Existential Quantifier  –Universal Quantifier 
F22H1 Logic and Proof Week 7 Clausal Form and Resolution.
CAS LX 502 8a. Formal semantics Truth and meaning The basis of formal semantics: knowing the meaning of a sentence is knowing under what conditions.
Albert Gatt LIN3021 Formal Semantics Lecture 5. In this lecture Modification: How adjectives modify nouns The problem of vagueness Different types of.
Knowledge Representation Methods
LTAG Semantics on the Derivation Tree Presented by Maria I. Tchalakova.
CAS LX 502 Semantics 1b. The Truth Ch. 1.
Logo Lesson 5 TBE Fall 2004 Farah Fisher. Prerequisites  Given a shape, use basic Logo commands and/or a procedure to draw the shape, with and.
Meaning and Language Part 1.
CAS LX 502 Semantics 2b. A formalism for meaning 2.5, 3.2, 3.6.
Survey of Mathematical Ideas Math 100 Chapter 3, Logic
Linguistic Theory Lecture 3 Movement. A brief history of movement Movements as ‘special rules’ proposed to capture facts that phrase structure rules cannot.
THE PARTS OF SYNTAX Don’t worry, it’s just a phrase ELL113 Week 4.
February 2009Introduction to Semantics1 Logic, Representation and Inference Introduction to Semantics What is semantics for? Role of FOL Montague Approach.
CAS LX 502 Semantics 3a. A formalism for meaning (cont ’ d) 3.2, 3.6.
1 LIN 1310B Introduction to Linguistics Prof: Nikolay Slavkov TA: Qinghua Tang CLASS 14, Feb 27, 2007.
Logical Equivalence & Predicate Logic
Syntax Lecture 8: Verb Types 1. Introduction We have seen: – The subject starts off close to the verb, but moves to specifier of IP – The verb starts.
CAS LX 502 8b. Formal semantics A fragment of English.
1 Chapter 7 Propositional and Predicate Logic. 2 Chapter 7 Contents (1) l What is Logic? l Logical Operators l Translating between English and Logic l.
Week 2 - Wednesday.  What did we talk about last time?  Arguments  Digital logic circuits  Predicate logic  Universal quantifier  Existential quantifier.
1 Knowledge Based Systems (CM0377) Lecture 4 (Last modified 5th February 2001)
Week 2 - Friday.  What did we talk about last time?  Predicate logic  Negation  Multiple quantifiers.
November 2003CSA4050: Semantics I1 CSA4050: Advanced Topics in NLP Semantics I What is semantics for? Role of FOL Montague Approach.
1 LIN 1310B Introduction to Linguistics Prof: Nikolay Slavkov TA: Qinghua Tang CLASS 13, Feb 16, 2007.
Chapter 1, Part II: Predicate Logic With Question/Answer Animations.
Slide 1 Propositional Definite Clause Logic: Syntax, Semantics and Bottom-up Proofs Jim Little UBC CS 322 – CSP October 20, 2014.
First Order Logic Lecture 2: Sep 9. This Lecture Last time we talked about propositional logic, a logic on simple statements. This time we will talk about.
October 2004CSA4050: Semantics III1 CSA4050: Advanced Topics in NLP Semantics III Quantified Sentences.
Artificial Intelligence: Natural Language
1 Predicate (Relational) Logic 1. Introduction The propositional logic is not powerful enough to express certain types of relationship between propositions.
Linguistic Essentials
Entity Theories of Meaning. Meaning Talk Theory should make sense of meaning talk Theory should make sense of meaning talk What sorts of things do we.
Albert Gatt LIN3021 Formal Semantics Lecture 4. In this lecture Compositionality in Natural Langauge revisited: The role of types The typed lambda calculus.
LECTURE 2: SEMANTICS IN LINGUISTICS
Rules, Movement, Ambiguity
Theory and Applications
1 LIN 1310B Introduction to Linguistics Prof: Nikolay Slavkov TA: Qinghua Tang CLASS 16, March 6, 2007.
Chapter 2 Logic 2.1 Statements 2.2 The Negation of a Statement 2.3 The Disjunction and Conjunction of Statements 2.4 The Implication 2.5 More on Implications.
CS 285- Discrete Mathematics Lecture 4. Section 1.3 Predicate logic Predicate logic is an extension of propositional logic that permits concisely reasoning.
SYNTAX.
◦ Process of describing the structure of phrases and sentences Chapter 8 - Phrases and sentences: grammar1.
Lecture 1 Ling 442.
1 LIN 1310B Introduction to Linguistics Prof: Nikolay Slavkov TA: Qinghua Tang CLASS 11, Feb 9, 2007.
CAS LX 502 9b. Formal semantics Pronouns and quantifiers.
Albert Gatt LIN3021 Formal Semantics Lecture 3. Aims This lecture is divided into two parts: 1. We make our first attempts at formalising the notion of.
1 Outline Quantifiers and predicates Translation of English sentences Predicate formulas with single variable Predicate formulas involving multiple variables.
NATURAL LANGUAGE PROCESSING
Meaning and Language Part 1. Plan We will talk about two different types of meaning, corresponding to two different types of objects: –Lexical Semantics:
Propositional Logic. Assignment Write any five rules each from two games which you like by using propositional logic notations.
Propositional and predicate logic
SYNTAX.
Logic II CSE 140 etc..
Syntax Lecture 9: Verb Types 1.
7/3/2018 EMR 17 Logical Reasoning Lecture 11.
Language, Logic, and Meaning
Chapter Eight Syntax.
Part I: Basics and Constituency
Chapter 3 Introduction to Logic 2012 Pearson Education, Inc.
CS201: Data Structures and Discrete Mathematics I
Chapter Eight Syntax.
Predicates and Quantifiers
CS201: Data Structures and Discrete Mathematics I
Semantics 2: Syntax-Semantics Interface
Starting out with formal logic
Presentation transcript:

Interpreting Language (with Logic) Semantics II Interpreting Language (with Logic)

Primary Objectives Continue our study of how meanings in natural language are Represented Constructed Two examples: More on adjectives, sets, etc. How quantification with every, some, etc. is represented

Background Recall that the guiding principle for studying objects larger than single terminals is that meanings are built out of their constituent parts: Principle of Compositionality: The meaning of a whole is a function of the meaning of the parts and of the way they are syntactically combined. (associated with Frege; cf. Partee reading) Note crucially that there are two components to this: What the parts mean How the parts are combined We will review these two components with reference to some of the adjective examples we studied last time

The Meaning of the Parts Recall that for some adjectives, we made use of the idea that the interpretation of ADJ N involved intersection: RULE (informal): When an adjective A modifies a noun N ([A N]), the interpretation of this object is the set defined by the intersection of A’s meaning with N’s meaning This rule accounts for the interpretation of e.g. red book, as the intersection of two sets.

Other types of adjectives We saw one type of adjective that is not (necessarily) intersective before: Larry is a skillful artist. Larry is a chess player Therefore: Larry is a skillful chess player. (Doesn’t work) So one thing that we have to know is what kind of adjective we are dealing with In addition, we’ll need to know what syntactic structure we have Remember, these are the two components of compositionality

Recall… From last time as well: Larry is a poisonous snake Larry is a chess player. Therefore: Larry is a poisonous chess player The phrase poisonous chess player is ambiguous…it can also mean that he’s not poisonous per se, but as a chess player, he is.

For example…(again) So, with poisonous chess player, it seems that some adjectives can be interpreted in either fashion. Recall also: Larry is a beautiful dancer. Meaning1: He dances beautifully Meaning2: He is beautiful, and he is a dancer (he might dance poorly) Question: Do these differences involve different structures, or just a lexically ambiguous set of adjectives?? I.e., different compositional analyses can be given, which differ in terms of how much they build into the lexical item, or into the structure

Still another type Consider a further type of adjective: John is a former chess player. Adjectives like former (including alleged, counterfeit, etc.) are: Not intersective: || former chess player|| is not the intersection of ||former|| and ||chess player|| Not like skillful type adjectives either: ||skillful chess player|| is a subset of ||chess player|| But ||former chess player|| is not a subset of ||chess player|| Aside: John is tall/skillful/*former All of these things are syntactically Adjectives; but how they combine to create larger meanings is determined in part by how they differ from one another How to represent such differences goes beyond what we’ll do; at this point, we will examine a second factor, syntactic structure

Structure One simple case illustrating structural differences involves adjectives combining with nouns either (1) in phrases, vs. (2) in compounds. Example: Phrase: black board. Meaning: here it is intersective (a thing that is both black and a board) Compound: bláckboard. Meaning: thing that we write on with chalk. Not intersective! A blackboard could be e.g. green. So: how things are put together is crucial. In a sense, this recapitulates what we saw in our study of word structure and syntax (remember ‘unlockable’ and ‘fix the car with a wrench’); the ambiguities correspond to different structures…

Other examples of structure Consider some further examples: John hammered the metal. John hammered the metal flat. In the second example, the adjective flat defines the state that the metal moves towards by being hammered. Now, how about: John hammered the metal naked. In the second example here, we understand the adjective as defining the state that John was in when he undertook the hammering of the metal However, the structural position of modifiers like the naked adjective is in principle compatible with both subject and object: John met Bill naked. (=John or Bill)

However…. When things like naked appear in the VP, they can be interpreted with either the subject or the object, if it makes sense Interestingly: Further examples show that the flat type adjectives and the naked type are in different syntactic positions: John hammered the metal flat naked. *John hammered the metal naked flat. The second example is deviant because it seems that the first of the two adjectives must go with the object; and in this case, that doesn’t make sense

Interim summary When it comes to building meanings, two primary factors must be taken into account: What individual elements (e.g. specific classes of adjectives in the examples above) mean What syntactic structures these elements appear in As we have been noting throughout, there are clear correlations between structure and meaning. What we have added in this discussion is the further idea that what the individual words mean can also have an effect on how meanings are derived.

Quantifiers: More Ambiguities Thus far we have seen different ways in which an ambiguity may arise: Structural ambiguity: Unlockable = [[un lock] able] or [un [lockable] Fix the car with a hammer = PP modifies VP, or PP modifies NP Lexical ambiguity (from homophony/polysemy) The pool made the party a lot better. = swimming pool = game of pool = entertaining rain puddle Another type of ambiguity is found in (certain) sentences with more than one quantifier (like every, etc.): Every student read some book.

The ambiguity Every student read some book. Reading1: Every student read some book or other (different books) Reading2: Every student read the same book Such ambiguities arise in other cases as well; consider: A student is certain to solve this problem. Reading1: Some student or other is going to solve this problem Reading 2: A particular student is going to solve this problem (e.g. Mary) (think about it…) These are often called scope ambiguities; see below In order to explain the nature of such ambiguities, we will look at some simple logic

Introducing logic We already saw some basic logic in the last slides In effect, we talked about the operation of intersection, which we associated with AND In order to make our logic powerful enough to talk about the ambiguity introduced above, we’ll have to add something to it

Interpreting Quantifiers Understanding the nature of the problem here requires some assumptions about quantifiers. In logical analysis, quantifiers are interpreted with respect to some domain; think of this as a world. We’ll introduce a restricted world below. Quantifiers don’t seem to refer to things in the way that things like cat do. Consider: No students went to the library. What would no students refer to?? In order to see how quantifiers are interpreted, it is useful to have a small domain (think of it like a model) to look at what our logical statements mean. I choose….

A restricted domain Let’s illustrate with respect to a simple domain how the quantifiers work. We have a domain (in this example, a set of characters from Sesame Street): Who is missing…?

Back to reality: some basic logic In our logic, we need names for individuals: ernie  Ernie bb  Big bird elmo  Elmo Etc… We also need predicates, which are sets of individuals: e.g., red, blue, googly-eyed; these apply to one argument (see below) These predicates represent sets, like in our adjective examples; in this world: ||blue|| = {grover, cookie monster…} ||googly-eyed} = {cookie monster} Etc. We can then write simple statements, and judge whether or not they are true with respect to our model

Example statements Some things that we could say (with truth value) Blue(cm) ‘cookie monster is blue’; true Red(bb) ‘big bird is red’; false And so on We can also have predicates with two places; e.g. Taller(x,y) for ‘x is taller than y’: Taller(bb,cm) true (big bird is taller than cookie monster) Within this system, we can also define “and”, “or”, “not”, “if…then”; e.g. blue(cm) AND red(elmo) How are we going to say things like ‘some things are red’, ‘no thing is an NBA player’, and so on? This is where we need a way of representing quantification

Two quantifiers Quantifiers come with variables, presented here as x, y, etc. Existential Quantification: This is written with a ‘backwards E’ It is read as ‘there exists an x such that…’ Example: x BLUE(x) This means ‘there exists an x such that x is blue’ In our model, this is true; we can find individuals in the denotation of BLUE The other quantifier we need is one that says ‘every…’

Universal Quantification Universal Quantification: Represents in logical the meaning of ‘every’ or ‘all’ This is written with an upside-down ‘A’ Example: (let the predicate Ses be ‘is a Sesame Street character’) x Ses(x) This is read as ‘for all x, x is a Sesame Street character’ This is true in our model, but not in other models, e.g. the real word. Meanings like ‘no’ involve the quantifiers above and negation

Now, Returning to two Quantifiers Remember that we launched into this investigation of logic in order to understand the two meanings of examples like Every student read some book. Simplifying: Every student read some book. To simplify, we’ll look at: Everyone saw someone Which has the same ambiguity In our logic, the two readings have unambiguous statements

Representing the readings Reading1: Everyone saw some person or other  x y (Saw(x,y)) Read as: ‘For all x, there exists some y such that x saw y’ Reading 2: Everyone saw the same person. y  x (Saw(x,y)) Read as: ‘There exists a y such that for all x, x saw y’ The question for research in natural language semantics is how a single sentence/structure like that of Everyone saw someone can have or correspond to these distinct logical representations That is, why is it that the single sentence, with someone as object, can correspond to a meaning in which the existential quantifier is outside of the universal?