From Logic To MG

From Logic to Montague Grammar:
Some Formal and Conceptual Foundations of Semantic Theory
Course Handouts and Problem Sets
Seth Cable
University of Massachusetts Amherst
January 13, 2014
i
Forward
This document contains the class notes, handouts, and problem sets that were developed for the
Fall 2013 Proseminar in Semantics at the University of Massachusetts Amherst.
At the suggestion of course mentor Barbara Partee, as well as several students in the
class, I have collected these here and posted them publicly for anyone who may have an interest
in either Montague Grammar (MG) generally or the specific papers “Universal Grammar” (UG)
and “The Proper Treatment of Quantification in Ordinary English” (PTQ).
As to the content, structure, and goals of the course, I refer the reader to the course
syllabus that follows this forward. I will state briefly, however, that the primary original
contribution of these course notes to the already substantial didactic material on MG is its
presentation of the algebraic framework of UG, its discussion of the relationship between the UG
framework and the system presented in PTQ, and the introductory sections explaining and
motivating the development of model-theoretic semantics.
Given that these are course notes and handouts, much of the material that follows
borrows heavily from published texts and articles. A footnote at the beginning of each handout
lists the readings that those notes are based upon (and often borrow from). As a service to the
reader, however, I also list them comprehensively below:
Course Reading List
Unit 1: Formal Preliminaries

Heim & Kratzer. 1998. Semantics in Generative Grammar. Oxford: Dordrect.
Partee, Barbara, Alice ter Meulen, and Robert E. Wall. 1993. Mathematical Methods in
Linguistics. Dordrecht: Kluwer.
Stewart, Ian and David Orme Tall. 1977. The Foundations of Mathematics. Oxford: Oxford
University Press.
Unit 2: A Review of Propositional Logic and First Order Logic

Crossley, J.N., C.J. Ash, C.J. Brickhill, J.C. Stillwell, and N.H. Williams. 1972. What is
Mathematical Logic? New York: Dover.
Gamut, L.T.F. 1991. Logic, Language, and Meaning, Volume 1: Introduction to Logic.
Chicago: University of Chicago Press.
Unit 3: Algebras and Semantics

Dowty, David R., Robert E. Wall, and Stanley Peters (1981) Introduction to Montague
Semantics. Dordrecht: Kluwer.
Halvorsen, Per-Kristian and William Ladusaw. 1979. “Montague’s ‘Universal Grammar’: An
Introduction for the Linguist.” Linguistics and Philosophy 3: 185-223.
Montague, Richard. (1974) “Universal Grammar.” In Thomason, Richmond (ed) Formal
Philosophy: Selected Papers of Richard Montague. New Haven: Yale University Press.
ii
Unit 4: Montague’s Theory of Translation
Unit 5: An Algebraic Approach to Quantification and Lambda Abstraction

Unit 6: The Proper Treatment of Quantification in Ordinary English

Montague, Richard. (1974) “The Proper Treatment of Quantification in Ordinary English.” In
Thomason, Richmond (ed) Formal Philosophy: Selected Papers of Richard Montague.
New Haven: Yale University Press.
I wish to thank here the students and visitors who participated in the seminar and whose
comments and questions greatly improved the content of these notes: Elizabeth Bogal-Allbritten,
Ed Ferrier, Cameron Gibbs, Hannah Greene, Hsin-Lun Huang, Dennis Kavlakoglu, Stefan
Keine, Chisato Kitagawa, Jon Ander Mendia, Yangsook Park, Jeremy Pasquereau, Ethan Poole,
and Megan Somerday.
Special thanks are owed first and foremost to our course mentor, Barbara Partee. Besides
simply being an invaluable resource for us all, clarifying key points and drawing connections I
would never have dreamed of, she was a constant font of encouragement both for the students
and for myself. Without Barbara, this course never would have been possible.
Finally, if the reader notices any errors – either trivial or substantial – please do not
hesitate to contact me.
iii
Contents:
Given that the following material was originally separate documents, pagination begins afresh at
the beginning of each section. Consequently, specific materials cannot be located by page
number. However, I do provide here an overall outline of the material that follows.
0. Course Syllabus
1. Unit 1: Formal Preliminaries
1.1 Relations and Functions
1.2 Cardinalities, Infinities and Proof by Induction
1.3 Problem Set on the Formal Preliminaries
2. Unit 2: A Review of Propositional Logic and First Order Logic
2.1 Propositional Logic: Syntax and Natural Deduction
2.2 First Order (Predicate) Logic: Syntax and Natural Deduction
2.3 Problem Set on Syntax and Natural Deduction
2.4 Propositional Logic: Formal Semantics and Valuations
2.5 First Order Logic: Formal Semantics and Models
2.6 Problem Set on Formal Semantics (Valuations and Models)
2.7 Proving the Soundness and Completeness of Propositional Logic
2.8 Key Applications of Model Theoretic Semantics for First Order Logic
3. Unit 3: Algebras and Semantics
3.1 An Algebraic Perspective on Propositional Logic
3.2 Problem Set on Syntactic Operations, Semantic Operations, Homomorphisms
3.3 An Algebraic Perspective on the Syntax of FOL (Without Quantification)
3.4 Montague’s General Theory of Semantics
3.5 Problem Set on Languages and Interpretations
iv
4. Unit 4: Montague’s Theory of Translation
4.1 Laying the Groundwork
4.2 The Notion of a ‘Translation Base’
4.3 Translation Bases and Interpretations
4.4 Problem Set on Translation and Indirect Interpretation
5. Unit 5: An Algebraic Approach to Quantification and Lambda Abstraction
5.1 Preliminaries
5.2 Fregean Interpretations
5.3 Computing with ‘Logically Possible Partly-Fregean Interpretations’
5.4 Applications to the Analysis of English
5.5 Problem Set on the Analysis of Quantification
6. Unit 6: The Proper Treatment of Quantification in Ordinary English
6.1 First Steps Towards PTQ: A New Presentation of Our System for Quantifiers
6.2 The Fragment of English
6.3 Intensional Logic
6.4 The Translation System, Part 1
6.5 The Translation System, Part 2
6.6 Problem Set on the PTQ System
v
Seth Cable Proseminar on Semantic Theory
Fall 2013 Ling 720
From Logic to Montague Grammar:

Some Formal and Conceptual Foundations of Semantic Theory
Syllabus

Linguistics 720 Course Website:
Tuesday, Thursday 2:30 – 3:45 http://people.umass.edu/scable/LING720-FA13/
Room: Dickinson 110
Course Instructor: Seth Cable E-mail: [email protected]
Course Mentor: Barbara Partee E-mail: [email protected]

1. General Overview

In general, the semantics proseminar is intended to serve as a bridge between the introductory
graduate semantics courses (610, 620) and the more advanced semantics seminars. Typically, a
specific subject is covered in more depth than is normally done in the intro classes, but the
discussion is paced at a level appropriate for second- and third-year students.
The subject of this year’s proseminar is, loosely speaking, the formal semantic framework of
Montague Grammar. The overall goal of the course will be to provide students with enough
formal background to successfully navigate and critically evaluate the early literature of our
discipline. Consequently, this course will be closer in spirit to an introductory semantics course
than to a full-fledged seminar. It is my hope that – with input from the students and other
participants – this course may ultimately be developed into a regularly taught component our
graduate semantics curriculum.
1
Fall 2013 Ling 720
1.1 Goals of the Course
The primary goal of this course is to provide students who have taken 610 and 620 with (some
of) the background necessary to read papers written within the Montague Grammar tradition,
especially the classic papers in Montague (1974) and Partee (1976).
As the field of formal semantics has grown over the past forty years, it has become more tightly
integrated into the other subdisciplines of linguistics – especially syntax. A consequence of this
integration is that much semantic research nowadays assumes a GB-style syntactic architecture,
where LF tree-structures are separately generated by the syntax and then ‘input’ to a recursively
defined semantic interpretation operation. In addition, in much current work, the denotations
output by the interpretation operation are characterized purely via representations in a logical
metalanguage, one that is implicitly understood by readers.
Although there are many merits to these developments, a negative consequence is that one can be
expert in reading and evaluating current semantic research, while nevertheless finding great
difficulty in comprehending the earliest and most influential works of our discipline. In addition,
although the original ‘Montague Grammar’ framework is no longer widely used, much current
work is nevertheless still written in an explicitly ‘model-theoretic’ style akin to those early
works. Finally, the research done within frameworks that are ‘directly compositional’ is best
understood and contextualized as a development of the key ideas of Montague.
For these reasons, it is quite important for semantics students to have the ability to read and
comprehend the classic papers of Montague (1974) and Partee (1976). It is also extremely
important to be able to ‘translate’ proposals from one research tradition into another, to allow for
effective comparison and evaluation of different analyses. The primary goal of this seminar,
then, is to develop these skills, alongside a deeper understanding and appreciation of the seminal
works of Montague.
1.2 Structure of the Course
A fundamental concept in the work of Montague and others is that of ‘interpretation with respect
to a model’. While this concept is emphasized in some semantics textbooks (Chierchia &
McConnell-Ginet 2000), it is not emphasized in all introductory semantics curricula. In addition,
in many semantics textbooks, the concept is introduced somewhat by fiat, without much
motivating (or clarifying) context. For this reason, our course will begin by providing some
crucial historical and conceptual context for the tools of ‘model-theoretic semantics’.
In this first section, “Why Models?”, we begin by reviewing the syntax and proof system of two
fundamentally important logical languages: Propositional Logic (PL) and First Order Logic
(FOL). We will then see how several key questions about these systems motivate the
development of a mathematically precise characterization of what it means to be an
‘interpretation’ of these languages. We will then see how so-called ‘models’ can play this rule
for FOL. Finally, we will see how, once armed with the notion of a ‘model’, we can answer those
fundamentally important questions about FOL. This general plot structure is outlined below:
2
Fall 2013 Ling 720
Part 1: Why Models?
• Propositional Logic (PL): Syntax and Natural Deduction System

• First Order (Predicate) Logic (FOL): Syntax and Natural Deduction System
• Formal Semantics of PL: Valuations
• Proving Soundness and Completeness of PL Natural Deduction
• Formal Semantics of FOL: Models
• Proof Sketch of Soundness and Completeness of FOL Natural Deduction
• Some Other Neat Results of Model Theory
• A Model Theory for Natural Language? Benefits and Obstacles
Readings and Resources (Posted on Moodle)

Gamut, L.T.F. 1991. Logic, Language, and Meaning, Volume 1: Introduction to Logic.
Chicago: University of Chicago Press. (Chapters 1, 2, 3, 4)
Linguistics. Dordrecht: Kluwer. (Chapters 5, 6, 7)
Crossley, J.N., C.J. Ash, C.J. Brickhill, J.C. Stillwell, and N.H. Williams. 1972. What is
Mathematical Logic? New York: Dover. (Chapter 2)
As outlined above, once it’s clear what can be achieved with a model theoretic semantics for
FOL, the question will naturally arise of whether such a semantics could ever be given for a
natural language. We’ll examine the obvious benefits of developing such a semantic theory, as
well as some of the obvious obstacles that initially stood in its way.
Next, the class will veer sharply into the domain of abstract algebra. Many of the key ideas in
Montague’s work are based on the following core insight: interpreting a language with respect to
a model can be conceived of mathematically as a special kind of ‘mapping’ between algebras,
namely a ‘homomorphism’. Therefore, we will begin with an introduction to the concepts of an
‘algebra’ and a ‘homomorphism’. We’ll then see right away that a ‘valuation’ of PL is essentially
a homomorphism from a kind of ‘syntactic algebra’ (forming the sentences of PL) to a ‘semantic
algebra’ (consisting of operations over truth-values). We’ll then further develop this notion of
‘interpretation as homomorphism’ so that it can apply to FOL (without quantification).
Part 2: Algebras and Semantics
• Introduction to Algebras and Morphisms.

• PL as an Algebra, Valuations as Morphisms
• Algebraic Syntax/semantics of FOL (without quantification)

Linguistics. Dordrecht: Kluwer. (Chapter 9)
Semantics. Dordrecht: Kluwer (Chapter 8)
3
Fall 2013 Ling 720
At this point, we will have enough background to begin discussing Montague’s theory of
translation. Another key part of Montague’s (1974) framework is the notion that – under certain,
very special conditions – translation from one language into another can also be conceived of as
a homomorphism between (syntactic) algebras. We’ll see that this ‘homomorphic’ conception of
translation has a very crucial consequence: if a language L can be (homomorphically)
translated into another language L’, which has a defined (model-theoretic) semantics, then
you’ve also thereby provided L with a defined (model-theoretic) semantics. We will discuss the
central importance of this consequence for the general program of formal semantics.
Part 3: Algebras, Translations, and Indirect Interpretation
• Montague’s Theory of Indirect Interpretation

• Indirect Interpretation of English (without quantification)
• Algebraic Syntax and Semantics of FOL (with quantification)
• Indirect Interpretation of Quantification in English

Montague, Richard. (1974) “Univeral Grammar.” In Thomason, Richmond (ed) Formal
Once we’ve reached this point, we’ll actually have developed a significant purely ‘extensional’
fragment of English in Montague’s (1974) framework. However, many of the key advances in
Montague’s work stem from his use of a specially designed ‘Intensional Logic’ (IL). Thus, the
final stage of our gradual introduction to Montague’s system will be a study of his Intensional
Logic, as well as asome its basic applications to the analysis of English.
Part 4: Montague’s Intensional Logic (IL) and its Applications
• Key concepts behind the Intensional Logic (IL)

• Formal syntax and model-theoretic semantics of IL
• Algebraic characterization of IL’s model-theoretic semantics
• Indirect Interpretation of English via IL (Basic Examples)

Semantics. Dordrecht: Kluwer (Chapter 6, Chapter 7)
Gamut, L.T.F. 1991. Logic, Language, and Meaning, Volume 2: Intensional Logic and Logical
Grammar. Chicago: University of Chicago Press. (Chapter 5)
4
Fall 2013 Ling 720
Having come this far, it will now be possible for us to read and discuss much of Montague’s
seminal paper “Universal Grammar” (UG). Our guide to this work will be the excellent
overviews by Halvorsen & Ladusaw (1979) and Dowty et al. (1981). Having walked through the
most important parts of UG, we will also be well-equipped to read and discuss that most seminal
of Montague’s works, “The Proper Treatment of Quantification in Ordinary English” (PTQ). If
time permits, we will also read one or two other classic works in formal semantics, such as
Karttunen’s “Syntax and Semantics of Questions”.
Part 5: Classic Papers in Montague Grammar

• Montague’s “Universal Grammar”
• Montague’s “The Proper Treatment of Quantification in Ordinary English”
• (Perhaps Karttunen’s “Syntax and Semantics of Questions”)
• (Perhaps one more paper, from Partee (1976))

Montague, Richard. (1974) “The Proper Treatment of Quantification in Ordinary English” In
Thomason, Richmond (ed) Formal Philosophy: Selected Papers of Richard Montague.
New Haven: Yale University Press.
Gamut, L.T.F. 1991. Logic, Language, and Meaning, Volume 2: Intensional Logic and Logical
Grammar. Chicago: University of Chicago Press. (Chapter 6)
Partee, Barbara. 1975. “Montague Grammar and Transformational Grammar.” Linguistic Inquiry
6(2): 203-300
Partee, Barbara. 1976. Montague Grammar. New York: Academic Press
Some Additional Notes About Course Content and Difficulty
• This is not an introduction to semantics. I assume students have taken (the equivalent) of
610 and 620. Thus, many technical concepts from 610 and 620 will be assumed.
(e.g., extension, intension, possible world semantics, etc.)
• This is not an introduction to logic. As explained in the course announcement, I assume

that students have a basic background in logic, as would be obtained in a typical
undergraduate introduction to logic course.
(e.g. ‘translation’ into PL and FOL, truth-tables, natural deduction)
• Doing the reading is critical. I’m going to aim to move relatively quickly, particularly
through Part 1. This course is cumulative, and so it’s crucial that you keep up.
5
Fall 2013 Ling 720
General Piece of Advice: MEET WITH US!
At any point in the semester, please meet with Barbara and I regarding any issues at all,
particularly if you are having any kind of difficulties with the course. We are also very happy to
discuss anything at all, especially any interesting puzzles you happen to note along the way.
2. Course Requirements
As mentioned, this class will be closer in spirit to 620 than to a full-fledge seminar. Thus, the
course requirements will be similar to those of 620.
2.1 Problem Sets
Due to the technical nature of the course material, it is critical that students complete regular
weekly problems sets. These problem sets will be assigned on Thursday and due the following
Thursday. Students are permitted to collaborate on problem sets, as long as each student writes
up their work individually. The answers to the problem sets will not always be discussed in class.
In such cases, I will write up and distribute answer keys.
2.2 Final Presentation and Final Paper
Students will be required to complete a final project, which will be presented in two formats: an
in-class presentation at the end of the term, and a final paper.
The nature of this final project is rather open-ended, but any of the following types of
project would be acceptable:
• A critical discussion of some aspect of Montague's classic papers UG or PTQ which we

did not cover in class.
• A critical discussion of some paper in Partee (1976) (not covered in class)
• A critical discussion of some paper in Barker & Jacobson (2007) Direct Compositionality
• A Montague Grammar (MG) or Directly Compositional (DC) adaptation of an existing
semantic analysis
• An original analysis of some phenomenon, done within MG or DC
The in-class presentations will take place during a special meeting after the final class; ideally
during the week of December 9th – December 13th. The final paper will be due December 20th.
Students should decide upon a final project by November 14th (at the absolute latest).
6
Fall 2013 Ling 720
3. Various Dates of Interest
October 15th: No class (Monday schedule)
November 14th: Declare topic of final project
November 28th: No class (Thanksgiving break)
December 5th: Last day of class
December 9th-13th: Final Presentations
December 20th: Final Paper Due
December 23rd: Final Grades Due
7

Unit 1:
Formal Preliminaries
Fall 2013 Ling 720
Formal Preliminaries, Part 1: Relations and Functions 1
1. Basic Concepts of Set Theory
I assume that the reader is familiar with the following:

• Basic concept of a set
• Set abstraction notation, { x : x is a boy }
• The concepts of cardinality, subset, powerset, union, intersection, complementation
• Various key set theoretic identities (X ∪ (Y ∩ Z) = ((X ∪ Y) ∩ (X ∪ Z))
For a review of these key concepts, the reader is referred to Partee et al. (1993), Chapter 1
2. Relations and Functions
2.1 Ordered Tuples, Products, Projections
(1) Ordered Pair
The defining property of an ordered pair is that ‘order matters’. That is:
• If <x,y> = <y,x>, then x = y
• If x ≠ y, then <x,y> ≠ <y,x>
(2) Ordered n-Tuple
With the notion of an ordered pair, we can define the concept of an ordered triple:
<x, y, z> =def < <x,y> , z >
Clearly, this will generalize to the definition of an arbitrary n-tuple:

< x1 , … , xn > =def < <x1, … xn-1 > , xn >
(3) Cartesian Product
If we have two sets A, B, then A × B, the cartesian product of A and B is:
A × B =def { <x, y> : x ∈ A and y ∈ B }
Illustration: { a, b } × { c, d } = { <a,c>, <a,d>, <b,c>, <b,d> }

{ c, d, e } × { a, b } = { <c,a>, <c,b>, <d,a>, <d,b>, <e,a>, <e,b> }
Note: For any set A, A × ∅ = ∅
1
These notes are based upon material in the following required readings: Partee et al. (1993) Chapter 1, Chapter 2;
Heim & Kratzer (1998) pp. 29-32; Stewart & Tall (1977) pp. 105-106.
1
Fall 2013 Ling 720
(4) n-Ary Cartesian Product
Suppose that A1, …, An is a series of n sets:
A1 × … × An =def the set of all n-tuples <a1, …, an> such that a1 ∈ A1, … an ∈ An
=def { <a1, …, an> : ai ∈ Ai }
Illustration:
{ a, b } × { c, d } × {e , f } = {<ace>, <acf>, <ade>, <adf>, <bce>, <bcf> <bde>, <bdf>}
(5) Cartesian Power

• Let A be any set. A2 = A × A
• Let A be any set and n be any (natural) number. An = A × … × A (n times)
2.2 Relations, Domains, Ranges, Inverses
(6) (Binary) Relation

A set of ordered pairs is a (binary) relation.
Illustrations: { <x,y> : x is the mother of y }

{ <x,y> : x loves y }
{ <a,b>, <c,d>, <e,f>, … }
(7) Domain and Range

Let R be a relation.
a. The Domain of R: { x : <x,y> ∈ R }
b. The Range of R: { y : <x,y> ∈ R }
(8) Inverse of Relation

Let R be a relation. R-1 is the inverse of R, and is defined as follows:
R-1 =def { <y,x> : <x,y> ∈ R }
Illustration: Let R = { <a,b>, <c,d>, <e,f> } Note: (R-1)-1 = R

R-1 = { <b,a>, <d,c>, <f,e> }
(9) n-Ary Relation

An n-ary relation R is a set of n-tuples.
Illustrations: { <x,y,z> : x is between y and z } ternary relation

{ <x,y,z,s> : x bet y z dollars that s would lose } quaternary relation
{ <a,b,c,d,e>, <f,g,h,i,j> <k,l,m,n,o> } ‘5-ary’ relation
Note: Given the definition in (2), an n-ary relation R is equivalent to a binary relation
whose domain is a set of (n-1)-tuples, (i.e., an (n-1)-ary relation).
2
Fall 2013 Ling 720
2.3 Functions and Stuff
(10) Function
Let R ⊆ A × B. R is a function from A to B if the following conditions hold:

a. The domain of R is A
b. If <x,y> ∈ R and <x,z> ∈ R, then y = z.
(11) Key Notations Relating to Functions
a. f: A  B =def f is a function from A to B

b. f(x) =def the unique y s.t. <x,y> ∈ f
c. BA =def the set of all functions from A to B.
d. f: A  B  C =def f is a function from A to CB
(12) Surjection, Injection, Bijection
Let f: A  B.
a. f is a surjection (onto) =def the range of f = B

=def for every b ∈ B, there is an a ∈ A s.t. f(a) = b
b. f is an injection (one-to-one) =def if f(a) = f(a’), then a = a’

=def each a ∈ A is mapped to a different b ∈ B
c. f is a bijection =def f is a surjection (onto) and an injection (one-to-one)
Illustrations:
Surjection: a d (not an injection)

b
c e
Injection: a c (not a surjection)

d
b e
Bijection: a d
b e
c f
3
Fall 2013 Ling 720
(13) Inverses and Bijections
a. Since a function f is a relation, we can speak of its inverse f -1
b. The inverse of a function f isn’t necessarily a function (consider surjections)
c. If f: A  B is a bijection, then f -1 is a function. f -1 is also a bijection.
Proof:
(i) f -1 is a function.
Suppose it weren’t. Then there would be <x,y>, <x,z> ∈ f -1 where y ≠ z.
But this would entail f(z) = f(y) = x, and so f isn’t an injection, and so f
isn’t a bijection, contrary to assumption.
(ii) f -1 is an injection (one-to-one).

Suppose it weren’t. Then <x,y>, <z,y> ∈ f -1 where x ≠ z. But this would
entail that <y,x>, <y,z> ∈ f, where x ≠ z, and so f isn’t a function, contrary
to assumption.
(iii) f -1 is an surjection (onto).

Suppose it weren’t. Then there is an x ∈ A such that there is no y ∈ B
such that f -1(y) = x. But, then it follows that that there is an x ∈ A such
that there is no y ∈ B such that f(x) = y, and so the domain of f isn’t A,
contrary to assumption.
(14) n-Ary Function

• Note that the domain of a function can be a set of n-tuples. Such a function will be
dubbed an ‘n-ary function’.
f: (A1 × … × An)  B
Illustration: { < < x,y >, z > : z = x + y } binary function

{ < < x,y,z >, s > : s = x + y + z } ternary function
• Note that an n-ary function is a set of ordered pairs, the first member of which is an n-
tuple.
• Consequently, given the definition in (2), an n-ary function is equivalent to an

(n+1)-ary relation
Illustration:
{ < < x,y >, z > : z = x + y } = { <x,y,z> : z = x + y }
{ < < x,y,z >, s > : s = x + y + z } = { <x,y,z,s> : s = x + y + z }
Note: Sometimes the term ‘function’ is restricted to unary functions. Functions of arity greater
than 1 (i.e., binary, ternary, etc.) are sometimes dubbed ‘operations’.
4
Fall 2013 Ling 720
2.4 Function Composition
(15) (Basic) Function Composition
Let f: A  B and g: B  C be unary functions. The composition of f and g is
g°f = { <x,z> : for some y, <x,y> ∈ f and <y,z> ∈ g }
Illustration: From Partee et al. (1993):
Note: For all x, g°f(x) = g(f(x))
Note: Another way of writing ‘the composition of f and g’ is g<f>
Note: In the definitions above, the functions g and f are assumed to be unary…
We can generalize these definitions to functions of any arity…
5
Fall 2013 Ling 720
(16) (Generalized) Function Composition
Let g be an n-ary function, and let f1, …, fn be a series of n m-ary functions. The
composition of g and f1, …, fn is the m-ary function defined as follows:
g<f1, …, fn> =def the m-ary function such that for any m-ary sequence a1, … am
g<f1, …, fn>(<a1, … am>) =
g( f1(<a1, … am>), …, fn(<a1, … am>))
Illustration:
Let g = { <<x,y>,z> : z = x + y }, f = { <x,y> : y = x – 1 }, h = { <x,y> : y = x + 2 }
Then: g<f,h>(2) = g(f(2),h(2)) = g(1,4) = 5
g<f,h> = { <x,y> : y = (x-1) + (x+2) }
3. Characteristic Functions
(17) Characteristic Functions and Characteristic Sets
a. Characteristic Function fA of a Set A

For every set A ⊆ B, there is a unique function fA: B  {1, 0}, defined below:
For all b ∈ B, fA(b) = 1 iff b ∈ A
b. Characteristic Set Af of a Function f: B  {1,0}

For every function f: B  {1,0}, there is a unique set Af ⊆ B, defined below:
For all b ∈ B, b ∈ Af iff f(b) = 1
Note:
• Given this regular correspondence between sets and (characteristic) functions, we will
often shift freely between the two without comment.
• Note, however, that such equivocation is not entirely innocent: sets are distinct set-
theoretic objects from their characteristic functions.
{a,b,c} ≠ { <a,1>, <b,1>, <c,1>, <d,0>, <e,0>, <f,0> }
(18) Characteristic Functions of Relations

Recalling the definition in (9) of an n-ary relation, the definition in (17) entails that every
n-ary relation R ⊆ A1 × … × An has its own characteristic function:
For all <a1, … , an> ∈ A1 × … × An , fR(<a1, … , an>) = 1 iff <a1, … , an> ∈ R
Note: For this reason, we will also often shift freely between n-ary relations and n-ary
characteristic functions (even though they are, strictly speaking, different objects)
6
Fall 2013 Ling 720
(19) One Final Observation
• The characteristic function of an n-ary relation is an n-ary function.
• But, recall from (14), that every n-ary function is equivalent to an (n+1)-ary relation
• Thus, every n-ary relation corresponds to a unique (n+1)-ary relation
Illustration: { <a,b>, <bc>, <c,d> } ≈ { <a,b,1>, <b,c,1>, <c,d,1>, <b,a,0>,

<c,b,0>, <d,c,0>, … }
Note: Sometimes, Montague shifts freely between these two objects (so, watch out!)
4. Currying Functions 2
(20) Currying of Binary Functions
Let f be a binary function f: (A×B)  C. There’s a unique function curry(f): A  B  C

defined as follows:
For all <x,y> ∈ A × B, f(<x,y>) = c iff curry(f)(x)(y) = c
Illustration:
Suppose we have the following function f: { a, b } × { c, d }  {1, 0}
f: <a,c>  1 curry(f): a c1

<a,d>  0 d0
<b,c>  0
<b,d>  1 b c0
d1
Note: The lambda notation makes it quite easy to define curry(f):
curry(f) = [ λx : [ λy : f(<x,y>) ] ]
2
Given that the original discoverer of this technique was Moses Schönfinkel, some have proposed that the term
‘schönfinkeling’ be used instead of ‘currying’ (as Haskell Curry rediscovered the technique later). However, to my
knowledge, the term ‘schönfinkeling’ has not widely caught on in mathematics and computer science.
7
Fall 2013 Ling 720
(21) Currying n-Ary Functions
• Note that in the definition in (20), the set A could itself be a set of pairs.
o In such a case, f(<<x,y>,z>) = c iff curry(f)(<x,y>)(z) = c
• Thus, in such a case, curry(f) will itself be a binary function. Thus, we could easily
also speak of the function curry(curry(f)).
o In such a case f(<<x,y>,z>) = c iff curry(curry(f))(x)(y)(z) = c
o This reasoning clearly generalizes to the following:
Let f be an n-ary function f: A1 × … × An  C. There is a unique function CUR(f),

CUR(f): A1  …  An  C defined as follows:
For all <a1, … , an> ∈ A1 × … × An,

f(<a1, … , an>) = c iff CUR(f)(a1)…(an) = c
Note: The lambda notation again makes it quite easy to define CUR(f)
CUR(f) = [ λx1 : … [ λxn : f(<x1 , … xn >) ] … ]
(22) Key Consequence
Let R R ⊆ A1 × … × An be an n-ary relation. From (18) and (21), it follows that there is a
unique function CUR(fR) such that:
<a1, … , an> ∈ R iff CUR(fR)(a1)…(an) = 1
Note: Again, the lambda notation makes it quite easy to define CUR(fR)
CUR(fR) = [ λx1 : … [ λxn : fR(<x1 , … xn >) ] … ]
Note:
Putting all of this together, we will often shift freely between the following (distinct) set-
theoretic objects:
• Set of n-tuples (n-ary relations)
• Functions from n-tuples to {1,0} (characteristic functions of n-ary relations)
• Curried functions from n-tuples to {1,0}
• ABXC and (AC)B
8
Fall 2013 Ling 720
5. Indexing
(23) Indexed Family
• Throughout the notes above, I’ve made use of the following informal notation:
A = {a1, …, an}
‘A is a set consisting of n different elements ai , for all 0 < i ≤ n’
• Note that this informal notation implies the existence of a bijection f: {1, 2, ..n }  A
f(i) = ai
• Note that there’s no special reason why use have to use numbers as indices; it’s just
convenient. This sets up the following general definition.
Definition:
Let J and A be sets such that there is a bijection f: J  A. We can say that A is an
indexed family, and that J is the index set.
Notation:
Suppose that A is an indexed family, whose index set is J. We can represent A as follows:
(i) { a : there is a j ∈ J such that f -1(a) = j }
(ii) { aj }j ∈ J
(24) Indexing and Tuples
• Throughout the notes above, we’ve also used numerical indices to represent n-tuples:
< a1 , … , an >
• We can adapt our notation in (ii) above as means for compactly representing n-tuples.
Both of the following are equivalent to < a1 , … , an >:
(i) < ai >i ∈ {1, …, n }
(ii) < ai >i ≤ n
9
Fall 2013 Ling 720
(25) Union and Intersection of Indexed Sets
Let S be an indexed family of sets with index set A.
a. ∪S = { x : x ∈ Sa for some a ∈ A }
b. ∩S = { x : x ∈ Sa for all a ∈ A }
(26) Alternate Notations
a. If S is an indexed family of sets with index set A, then ‘∪S’ is sometimes written:
b. If S is an indexed family of sets with index set A, then ‘∩S’ is sometimes written:
c. If S is an indexed family of sets with index set {1, …, n }, then ‘∪S’ is sometimes
written:
c. If S is an indexed family of sets with index set {1, …, n }, then ‘∩S’ is sometimes
written:
10
Fall 2013 Ling 720
Formal Preliminaries, Part 2: Cardinalities, Infinities, and Proof by Induction 1
(1) Some Key Players
a. The Natural Numbers (ℕ)

• All the whole numbers greater than or equal to 0 , {0, 1, 2, … }
b. The Integers (ℤ)

• All the whole numbers (including those less than 0), { …, -2, -1, 0, 1, 2 … }
c. The Rational Numbers (ℚ)

• All the numbers that can be written as a fraction, { n/m : n, m ∈ ℤ & m ≠ 0 }
d. The Real Numbers (ℝ)

• All the numbers that can be represented by an infinite decimal expansion
• All the rational numbers and all the irrational numbers (e.g., π)
Note: ℕ ⊂ ℤ ⊂ ℚ ⊂ ℝ
1. Cardinalities and Infinities
(2) Cardinality (Informal) |A| = ‘the cardinality of A’

= ‘the number of elements in A’
(3) Cardinality, Injection, and Bijection
The following statements are intuitively true for finite sets. We’ll therefore assume they
are true for all sets (including infinite ones).
a. |A| ≤ |B| if and only if there is an injection f: A B
b. |A| = |B| if and only if there is an bijection f: A B
Note:
• Recall that if f is a bijection, then f -1 is a bijection too.
• Thus, if there is a bijection f: AB, then there is also a bijection f -1: B A
o Thus, by the definition in (3b), if |A| = |B|, then |B| = |A| (as desired)
o Also, as you can prove to yourself:
If |A| = |B| and |B| = |C|, then |A| = |C|
For any set A, |A| = |A|
1
These notes are based upon the following required readings: Partee et al. (1993), Chapter 4, pp. 192-198.
1
Fall 2013 Ling 720
(4) Key Consequence: Infinite Sets Can Have Same Cardinality as Proper Subsets
• Consider the following function: f(x) = 2x / f = { <x,y> : y = 2x }
• This function f is a bijection from ℕ to the set of even numbers!

o It’s an injection: every x is mapped to a different even number
o It’s a surjection: every even number is equal to 2x for some x ∈ ℕ
• Thus, even though { n : n ∈ ℕ and n is even } ⊂ ℕ,

|{ n : n ∈ ℕ and n is even }| = |ℕ|
• Intuitively, no finite set contains a proper subset of the same cardinality.

o Thus, we can take this interesting property of ℕ as characteristic of ‘infinities’
(5) Characterization of Non-Finite

A set S is infinite if and only if there is a proper subset S’ ⊂ S such that |S’| = |S|
(6) Transfinite Cardinals
• Although it seems sensible to speak of |ℕ|, there is clearly no finite cardinal number
n ∈ ℕ such that |ℕ| = n.
• It will be useful to introduce new, transfinite cardinal numbers to allow us give a

name to the cardinality of ℕ
• We introduce the special symbol ‘‫א‬0’ (aleph null) below to refer to this first
transfinite cardinal.
‫א‬0 = |ℕ|
(7) Countable and Countably Infinite
a. A set S is countably (denumerably) infinite iff |S| = ‫א‬0
b. A set S is countable iff S is finite or S is countably infinite.
(8) Demonstrating that an Infinite Set is Countable, Part 1
• To show that an infinite set S is countable, show that there is a bijection from S to ℕ
• After all, this would entail |S| = |ℕ| = ‫א‬0
2
Fall 2013 Ling 720
(9) The Natural Numbers Without Zero (ℕ – {0}) are Countable
Consider the following function: f(n) = n – 1
a. The function f is clearly an injection from ℕ–{0} to ℕ

(each number in ℕ–{0} is mapped to a different member of ℕ)
b. The function f is clearly a surjection from ℕ–{0} to ℕ

(every member of ℕ is equal to (n-1) for some element in ℕ–{0})
Thus, |ℕ – {0}| = |ℕ| = ‫א‬0
(10) The Integers ℤ are Countable
Consider the function f defined below:
f(x) = 0 if x = 0
2x-1 if x is positive
-2x if x is negative
Picture of f(x), from Partee et al. (1993):
a. f(x) is clearly a function from ℤ to ℕ

b. f(x) is an injection from ℤ to ℕ
• Each positive number is mapped to an odd number
• Each negative number is mapped to an even number (greater than 0)
• Only 0 is mapped to 0
c. f(x) is a surjection from ℤ to ℕ
• 0 is mapped to 0
• Every positive even number is equal to -2x for some negative integer
• Every positive odd number is equal to 2x-1 for some positive integer
Therefore, f(x) is a bijection, and so |ℤ| = |ℕ| = ‫א‬0

(Note, this is despite the fact that ℕ ⊂ ℤ)
3
Fall 2013 Ling 720
(11) Demonstrating that an Infinite Set is Countable, Part 2
Now that we know that ℤ and ℕ–{0} are countable, we can show that S is countable by
showing that there is a bijection from S to ℤ or from S to ℕ–{0}
• After all, this would entail |S| = |ℤ| = ‫א‬0 or |S| = |ℕ–{0}| = ‫א‬0
(12) The Positive Rationals are Countable
Usually, the following ‘intuitive’ (or ‘visual’) proof is used to show that there is a
bijection from { n : n ∈ ℚ and n > 0 } to ℕ–{0}
a. Step One:
We can arrange the set { n : n ∈ ℚ and n > 0 } into the following infinite table:
1/1 1/2 1/3 1/4 1/5 1/6 1/7 …
2/1 2/2 2/3 2/4 2/5 2/6 2/7 …
3/1 3/2 3/3 3/4 3/5 3/6 3/7 …
4/1 4/2 4/3 4/4 4/5 4/6 4/7 …

… … … … … … …
b. Step Two:
Some rationals appear more than once in this table (e.g., 1/1 = 2/2). We can fix
this by snaking around the grid (infinitely) in the way sketched below. Every time
we hit a number that we’ve already passed, we cross it out.
1/1 1/2 1/3 1/4 1/5 1/6 1/7 …
2/1 2/2 2/3 2/4 2/5 2/6 2/7 …
3/1 3/2 3/3 3/4 3/5 3/6 3/7 …
4/1 4/2 4/3 4/4 4/5 4/6 4/7 …
5/1 5/2 5/3 5/4 5/5 5/6 5/7 …

… … … … … … …
4
Fall 2013 Ling 720
c. Step Three:
We take the augmented grid resulting from Step 2 (where repeated rationals are
crossed out), and we snake through it again just as before, mapping the rationals
in the grid to ℕ–{0} in the following way:
(i) We map 1/1 to 1

(ii) We then proceed as follows:
 Suppose that at step m in the ‘snaking’ we’ve just mapped the rational
p/q to the natural number n.
 We next examine the rational r/s we come to at step (m+1).

• If r/s is not crossed off in the grid, we map it to (n+1).
• If r/s is crossed off, then we proceed to the next step in the
snaking…
1/1 1/2 1/3 1/4 1/5 1/6 1/7 …
2/1 2/2 2/3 2/4 2/5 2/6 2/7 …
3/1 3/2 3/3 3/4 3/5 3/6 3/7 …
4/1 4/2 4/3 4/4 4/5 4/6 4/7 …
5/1 5/2 5/3 5/4 5/5 5/6 5/7 …
6/1 6/2 6/3 6/4 6/5 6/6 6/7 …

… … … … … … …
f(1/1) = 1 f(2/3) = 7
f(1/2) = 2 f(3/2) = 8
f(2/1) = 3 f(4/1) = 9
f(3/1) = 4 f(5/1) = 10
f(1/3) = 5 f(1/5) = 11
f(1/4) = 6 f(1/6) = 12 …
d. Step Four:
The function f defined above is a bijection from { n : n ∈ ℚ and n > 0 } to
ℕ–{0}
(i) It is an injection:
No two rationals will end up mapped to the same number in ℕ–{0}
(ii) It is a surjection:
Since there are infinite number of positive rationals, every number
in ℕ–{0} will be equal to f(x) for some rational x.
5
Fall 2013 Ling 720
(13) The Rationals are Countable
• We can use the result in (12) to show that the entire set of rationals ℚ is countable.
• Consider the function h defined below (where f is the function in (12))
For all n ∈ ℚ, h(n) = 0 if n = 0

f(n) if n > 0
–f(–n) if n < 1
The function h above is a bijection from ℚ to ℤ
a. The function h is clearly an injection

o Because f is an injection to ℕ–{0} every positive rational will be mapped to a
different positive integer, and every negative rational will be mapped to a
different negative integer.
b. The function is clearly a surjection

o Because f is an surjection to ℕ–{0} every positive integer n will be equal to
h(x) (=f(x)) for some positive rational.
o Because f is an surjection to ℕ–{0} every negative integer n will be equal to
h(x) (= -f(-x)) for some negative rational).
Therefore, |ℚ| = |ℤ| = ‫א‬0
2. Uncountable (Non-denumerable) Sets
So far, we’ve seen that |ℕ| = |ℤ| = |ℚ| = ‫א‬0, even though ℕ ⊂ ℤ ⊂ ℚ
• This might lead one to wonder whether, in addition, |ℝ| = ‫א‬0
• In this section, we’ll see that this is not the case, |ℝ| ≠ ‫א‬0
o That is, there are some infinities that are uncountable (non-denumerable)
(14) Powersets and Cardinalities

For every set S, | S | < | ℘(S) |
• Suppose that | S | = | ℘(S) |. There must then be a bijection f: S  ℘(S)
• Now, for any element x ∈ S, it is clear that either x ∈ f(x) or x ∉ f(x)
• Thus, we can define the set R = { x : x ∉ f(x) }
• Now, R is a subset of S, and so there must be some y ∈ S such that f(y) = R
• Finally, it must be that either y ∈ f(y) or y ∉ f(y)
o Suppose y ∈ f(y). But then y ∉ R, and so y ∉ f(y). Contradiction.
o Suppose y ∉ f(y). But then y ∈ R, and so y ∈ f(y). Contradiction.
6
Fall 2013 Ling 720
(15) Key Consequence: | ℘(ℕ) | > | ℕ | = ‫א‬0

If a set S is such that | S | > ‫א‬0, then we say that S is uncountable (non-denumerable).
(16) Key Result: The Real Numbers Between 0 and 1 are Uncountable
a. Key Background Fact:

Every real number between 0 and 1 can be uniquely represented as a sequence
consisting of ‘0.’, followed by an infinitely long number of decimals:
0.13456789890989999….
0.57682838494827789…
o Thus, every real number between 0 and 1 uniquely corresponds to a sequence

of the form ‘0.a1a2a3a4a5a6 … ’, where each ai is a decimal numer.
b. The Proof;
o Suppose that |{ n : n ∈ ℝ and 0 < n < 1 }| = |ℕ|. Then there is a bijection f
from { n : n ∈ ℝ and 0 < n < 1 } to ℕ.
o Given this bijection f, it is possible to write an (infinitely long) list of all the
members of { n : n ∈ ℝ and 0 < n < 1 }. Given the key background fact in
(16a), this list will look as follows, where anm is the mth decimal in the nth real
number in the ordering:
1 0. a11 a12 a13 a14 a15 ...

2 0. a21 a22 a23 a24 a25…
3 0. a31 a32 a33 a34 a35…
4 0. a41 a42 a43 a44 a45…
…
o Now, we can use this list to define a real number r between 0 and 1 that is not
on this list:
• The integer component of r is 0
• The first decimal in r after 0 is different from a11
• The second decimal in r after 0 is different from a22
• The third decimal in r after 0 is different from a33
• (and so on…)
o The real number r is guaranteed not to appear anywhere on this list.

• After all, for any natural number n, r will differ from f(n) in the nth
decimal after 0.
o Therefore, this list doesn’t contain all the real numbers between 0 and 1.
Consequently, there is no bijection from { n : n ∈ ℝ and 0 < n < 1 } to ℕ.
Thus, |{ n : n ∈ ℝ and 0 < n < 1 }| ≠ |ℕ|. Thus, |ℝ| >|ℕ| = ‫א‬0
7
Fall 2013 Ling 720
(17) Additional Transfinite Cardinals
• For various reasons, it will be helpful to have a name for |℘(ℕ)|:
‘the cardinality of ℘(ℕ)’
• It is known that |ℝ| = |℘(ℕ)|
3. Proof by Mathematical Induction
(18) Key Axiom of Number Theory
Suppose that for some property P, we can show (i) and (ii):
(i) 0 has property P
(ii) For any n ∈ ℕ, if n has property P, then (n+1) has property P.
Then we can conclude that every n ∈ ℕ has property P
(19) Key Consequence of (18)
Suppose that for some property P, we can show (i) and (ii):
(i) 0 has property P
(ii) For any n ∈ ℕ, if every number m < n has property P, then n has P
Then we can conclude that every n ∈ ℕ has property P
(20) Some Terminology
a. An argument making use of the axiom in (18) is typically referred to as a proof by

(weak) induction
b. An argument making use of the consequence in (19) is typically referred to as

proof by strong induction.
c. In a proof by (weak/strong) induction,
(i) Proving that 0 has property P is called the ‘base step’ (‘base case’)
(ii) Proving either (18ii) or (19ii) is called the ‘induction step’.
Note: If the base case is some numeral n > 0, then a proof by induction demonstrates that P
holds for all m such that n ≤ m
8
Fall 2013 Ling 720
(21) Key Application

If S is a countable set – that is, if there is a bijection f: ℕ S – then we can use proofs by
induction to prove things about S!
(22) Illustration: Generalized Distributive Law
Claim:
For all n ∈ ℕ such that 2 ≤ n, A ∪ ( B1 ∩ … ∩ Bn ) = (A ∪ B1) ∩ … ∩ (A ∪ Bn)
Proof by Induction:
a. Base Step: n = 2
A ∪ ( B1 ∩ B2 ) = (A ∪ B1) ∩ (A ∪ B2)
This follows from the simple set-theoretic equivalences proven in Chapter 1 of

Partee et al. (1993).
b. Induction Step
Let n ∈ ℕ be such that: A ∪ ( B1 ∩ … ∩ Bn ) = (A ∪ B1) ∩ … ∩ (A ∪ Bn)
o By the associativity of intersection:

A ∪ ( B1 ∩ … ∩ Bn ∩ Bn+1 ) = A ∪ ( ( B1 ∩ … ∩ Bn ) ∩ Bn+1 )
o Next, by the base step in (22a):

A ∪ ( ( B1 ∩ … ∩ Bn ) ∩ Bn+1 ) = (A ∪ ( B1 ∩ … ∩ Bn )) ∩ (A ∪ Bn+1 )
o Next, by the induction assumption for n:

(A ∪ ( B1 ∩ … ∩ Bn )) ∩ (A ∪ Bn+1 ) =
((A ∪ B1) ∩ … ∩ (A ∪ Bn)) ∩ (A ∪ Bn+1 )
o Finally, by the associativity of intersection again:

((A ∪ B1) ∩ … ∩ (A ∪ Bn)) ∩ (A ∪ Bn+1 ) =
(A ∪ B1) ∩ … ∩ (A ∪ Bn) ∩ (A ∪ Bn+1 )
o Thus, A ∪ ( B1 ∩ … ∩ Bn ∩ Bn+1 ) =
(A ∪ B1) ∩ … ∩ (A ∪ Bn) ∩ (A ∪ Bn+1 )
Therefore, by (weak) induction, it follows that for all n ∈ ℕ such that 2 ≤ n:

A ∪ ( B1 ∩ … ∩ Bn ) = (A ∪ B1) ∩ … ∩ (A ∪ Bn)
9
Fall 2013 Ling 720
(23) Illustration of Strong Induction: Well Ordering Principle
Claim: If S ⊆ ℕ and S ≠ ∅, then there is an a ∈ S such that for all s ∈ S, a ≤ s.
Proof by Strong Induction:
• Suppose that there is an S ⊆ ℕ and S ≠ ∅. For a contradiction, suppose that there is

no a ∈ S such that for all s ∈ S, a ≤ s.
• By strong induction, we’ll show that for all n ∈ ℕ, n ∉ S, and so S = ∅, contrary to

assumption.
a. Base Step: n = 0
Clearly, 0 ∉ S. (After all, for all s ∈ S, 0 ≤ s)
b. Induction Step
Let n ∈ ℕ be such that for all m < n, m ∉ S. We will show that n ∉ S.
• Suppose that s ∈ S. Now, clearly (n-1) < s. (After all, if s ≤ (n-1), then s < n,
and so by the induction assumption s ∉ S, contrary to assumption.)
• Next, since (n-1) < s, it follows that n ≤ s. Since s was arbitrary, it follows that
for all s ∈ S, n ≤ s.
• Consequently, n ∉ S (After all, by assumption there is no a ∈ S such that for

all s ∈ S, a ≤ s.)
• Thus, by strong induction, for all n ∈ ℕ, n ∉ S, and so S = ∅, contrary to

assumption.
• Therefore, for any S ⊆ ℕ and S ≠ ∅, there is a ∈ S such that for all s ∈ S, a ≤ s.
10
Fall 2013 Ling 720
Problem Set on the ‘Formal Preliminaries’ 1
(1) Basic Comprehension Questions on Relations and Functions
a. Let A = {b,c} and B = {2,3}. State whether the following are true or false:
(i) (A×B) ∩ (B×A) = ∅

(ii) <c,c> ⊆ A2
(iii) {<b,3>,<2,a>} ⊆ (A×B) ∪ (B×A)
(iv) ∅ ⊆ (A×A)
b. Let A = {b,c} and B = {2,3}. Let R = {<b,b>, <b,2>, <c,2>, <c,3>}
(i) What is the range and domain of R?

(ii) What is R-1?
c. Let f: AB and g: BC both be bijections. Show that (g°f )-1 = (f -1)°(g -1)
d. Let R = { <x,y> : x, y ∈ { 1, 2 } and x < y }.
(i) Represent R as a set of pairs.

(ii) Represent the characteristic function of R as a set of pairs 2
(iii) Represent the characteristic function of R as a ternary relation
(iv) Represent the curried characteristic function of R as a matrix
Reminder: By ‘matrix’ I mean a diagram like the following:
a aa
bc
cb
b ab
bc
ca
c ac
bc
cc
1
Most of these exercises are taken from Partee et al. (1993), Chapters 2 and 4.
2
Assume that the domain of the characteristic function of R is {1,2}×{1,2}.
1
Fall 2013 Ling 720
(2) Proving that Sets are Countable
a. Show that the set of integer powers of ten {10, 100, 1000, 10000, … } is
countable.
b. Suppose that the following is true of English:
(i) There is a finite alphabet for writing sentences, consisting of 26 letters and
a space (forget punctuation marks for now)
(ii) Every sentence of English is a finite string in the alphabet in (i)
(iii) There is no upper bound on the length of sentences of English. That is, for
any sentence S of English, there is always a longer sentence S’.
Show that the set of English sentences is countably infinite.
Some Hints:
• To solve this, you simply have to show that there is a defined way of ordering
the sentences of English into an infinite ‘list’ (as we did for the rationals
greater than 0).
• One way of solving this makes use of ‘alphabetical order’.
• Another, more round-about way of solving this makes use of the following
key consequence of the ‘fundamental theorem of arithmetic’
Consequence of Fundamental Theorem of Arithmetic:

For every positive integer n > 1, there is exactly one way of representing n as
a product of powers of primes:
n = p1a1 × p2a2 × … × pkak where each pi is prime, each ai ∈ ℕ
(3) Proofs by Induction
Construct a proof by induction for the following general equivalence:
( X1 ∪ … ∪ Xn )´ = X1´ ∩ … ∩ Xn´
Hint:
You can appeal to the following general equivalences:
(i) X´´ = X
(ii) X1 ∩ … ∩ Xn = (X1 ∩ … ∩ X(n-1) ) ∩ Xn
(iii) X1 ∪ … ∪ Xn = (X1 ∪ … ∪ X(n-1) ) ∪ Xn
2

Unit 2:
Review of Propositional Logic and First Order Logic

Fall 2013 Ling 720
Propositional Logic: Syntax and Natural Deduction 1
The Plot That Will Unfold
• I want to provide some key historical and intellectual context to the ‘model theoretic’
approach to natural language semantics, context that is usually missing from introductory
semantics classes / texts.
• I want to begin by getting across the very motivation for having ‘models’ at all:
o Why did logicians develop this mathematical device?

o What is it good for? What did it originally do for us?
• Consequently, I’m going to start off by giving a purely syntactic introduction to the
systems of Propositional Logic and First Order (Predicate) Logic.
o We’ll then see that certain key questions about these systems will require us to
develop a mathematically precise characterization of what it is to be an
‘interpretation’ of these formal languages…
1. Some Historical Background
(1) Valid Argument

An argument is valid if whenever the premises are true, the conclusion must be true.
a. Valid Argument: All men are mortal. Socrates is a man.

Therefore, Socrates is mortal.
b. Invalid Argument: All cats are mortal. All dogs are mortal.
Therefore, all cats are dogs.
(2) Key Insight from Ancient Philosophers (Aristotle)

Many valid arguments seem to share the same general syntactic form.
a. Valid Arguments: All Ps are Q. S is a P. No Ps are Qs. All Ss are Ps.

Therefore, S is a Q. Therefore, No Ss are Qs.
(Barbara) (Celarent)
b. Aristotelian Logic (Syllogistic Logic):

First to characterize valid arguments purely in terms of their syntactic form.
o So successful that it was logic from 300BC to 1800s.
1
These notes are based upon material in the following required reading: Gamut (1991), Chapter 1, Chapter 2 pp. 28-
41, Chapter 4 pp. 128-141; Partee et al. (1993) Chapter 5, Chapter 6 pp. 97-98.
1
Fall 2013 Ling 720
(3) Major Weakness of Syllogistic Logic
• Only applies to ‘syllogisms’, a very restricted form of argument.
• Consequently, cannot capture valid arguments based on relations:
All horses are animals.
Therefore, every horse’s head is an animal’s head.
(4) Key Developments in the 19th Century
a. Early-to-Mid 1800’s
• Renewed activity in logic; first highly original work in centuries

(George Boole, Augustus DeMorgan)
• Renewed interest in whether mathematics derives from ‘pure logic’.
b. Frege’s Begriffsschrift (1879)
• Birth of modern formal logic.
• First appearance of quantified variables
• Essentially all of First Order Logic (with a little Second Order Logic, too)
• Highly idiosyncratic 2-dimensional notation2
• Reconstructs an unprecedented amount of mathematics and natural language

reasoning
o First real indication that math could just derive from pure logic
c. Further Developments After Begriffsshrift
• Other logicians develop and modify the key insights of Begriffsschrift
• Peano introduces many of our modern logical notations: ∀, ∃, ⊃
• Principia Mathematica (1910-1913; but begun much earlier)
o Could be called the ‘climax’ of the 19th century period
o Three-volume work reconstructing enormous amounts of mathematics

from a logical proof system close to what we have today.
2
For a nice overview of Frege’s funky logical notation: http://en.wikipedia.org/wiki/Begriffsschrift
2
Fall 2013 Ling 720
(5) The Core Idea Behind a Formal Logic
• A precisely defined formal notation for representing certain aspects (not all) of the
‘logical structure’ of an assertion.
• A set of syntactically defined rules for deriving formulas in the notation from other
formulas in the notation.
Illustration: Natural Deduction (Logic 101):
1. p  (~q ∨ r) Premise
2. q Premise
3. q  ~r Premise
4. ~r Modus Ponens 2,3
5. q & ~r Conjunction 2,4
6. ~~(q & ~r) Double Negation 5
7. ~(~q ∨ ~~r) DeMorgans 6
8. ~(~q ∨ r) Double Negation 7
9. ~p Contraposition 1,8
(6) An Immediate Problem for This Program (Foreshadowing)
• As with classical logic, the goal of the enterprise is to provide a purely syntactic
characterization of what it is to be a valid argument.
• But, the notion of ‘validity’ is a semantic one.

(if premises true, then conclusion must be true)
• So, how can you be sure that your syntactically defined system does what it’s
supposed to do?
o How do you show that it only derives valid inferences?
o How do you show that it derives all the valid inferences?
• To do this, you need a mathematically precise characterization of what it means

for a formula to be ‘true’ (relative to an ‘interpretation’).
o It wasn’t until about the 1930’s or so until this idea reached its modern form
(Löwenheim, Gödel, Tarski)
Let’s begin by putting ourselves in the position of a person in 1910 or so…

We’ll start off with a purely syntactic introduction to Propositional Logic
3
Fall 2013 Ling 720
2. A Review of Propositional Logic (PL): Syntax and Informal Semantics 3
The system of Propositional Logic (PL) is intended to capture the inferences that depend upon
the meaning of the so-called ‘sentential connectives’: and, or, if…then, and not
(7) The Vocabulary of Symbols
a. The Non-Logical Constants (a.k.a ‘The Logical Variables’)

An infinite set of proposition letters: {p, q, r, s, t, … p1, p2, p3, p4, …}
b. The Logical Constants:

(i) ~ Negation ‘It is not the case that…’
(ii) & Conjunction ‘and’
(iii) ∨ Disjunction ‘or’ (inclusive)
(iv)  (Material) Implication ‘if…then’
c. Syntactic Symbols: (,)
(8) The Definition of a ‘Well-Formed Formula’ (WFF) of PL

The set of ‘well-formed formulae’ of PL, WFF, is the smallest set such that:
a. If ϕ is a proposition letter, then ϕ ∈ WFF
b. If ϕ, ψ ∈ WFF, then
1. ~ϕ ∈ WFF
2. (ϕ & ψ) ∈ WFF
3. (ϕ ∨ ψ) ∈ WFF
4. (ϕ  ψ) ∈ WFF
(9) Using PL To Encode Sentences of English
• We can use the syntactic rules in (8) and the informal semantics in (7) to write PL
formulae that ‘encode’ certain statements of English:
a. Sentence: “Dave is tall, and if Dave isn’t tall, then Mary is dancing.”
Encoding: ( p & ( ~p  q ) )
b. Sentence: “If Bill or John is leaving, then Mary and Sue are not happy.”
Encoding: ( ( b ∨ j )  ( ~m & ~s ) )
• In setting up such encodings, it is critical to supply a ‘key’, indicating which English

assertions the propositional letters ‘stand for’:
c. Key: p: Dave is tall b: Bill is leaving m: Mary is happy

q: Mary is dancing j: John is leaving s: Sue is happy
3
My discussion here will assume prior familiarity with the overall system of Propositional Logic. Students are
referred to Partee et al. (1993), Chapter 6 for crucial background.
4
Fall 2013 Ling 720
(10) An Important Property of the PL Syntax
• In (8), we provide a recursive definition of the well-formed formulae of PL.
• This definition allows us to use mathematical induction to prove things about all
well-formed formulae of PL.
(11) Illustration: An Inductive Proof about PL Formulae
a. Claim: Every well-formed formulae of PL has an even number of parentheses.

(i.e., If ϕ ∈ WFF, then there are an even number of parentheses in ϕ)
b. Proof By Strong Induction

We’ll prove the claim by mathematical induction on the number of logical
constants in ϕ
(i) Base Step: 0

If ϕ contains no logical constants, then ϕ is a proposition letter, and so ϕ
contains 0 parentheses (and 0 is even).
(ii) Induction Step

Let n be such that for all m < n, if ϕ contains m logical constants, then ϕ
has an even number of parentheses.
• Now suppose that ϕ contains n logical constants. There are four

possible cases to consider:
1. ϕ is of the form ~ψ, where ψ contains (n-1) logical constants. By assumption,

then, ψ contains an even number of parentheses. Thus, so does ϕ.
2. ϕ is of the form (χ & ψ), where χ contains m < n logical constants, and ψ
contains j < n logical constants. By assumption, then, χ and ψ both contain an
even number of parentheses. Thus, so does ϕ.
3. ϕ is of the form (χ ∨ ψ), where χ contains m < n logical constants, and ψ contains
j < n logical constants. By argument parallel to (2), so does ϕ.
4. ϕ is of the form (χ  ψ), where χ contains m < n logical constants, and ψ

contains j < n logical constants. By argument parallel to (2), so does ϕ.
5
Fall 2013 Ling 720
(12) Proof by Induction on the Complexity of Formula
• The general structure of the proof in (11) is extremely common in (meta)logic. It’s
often called proof by induction on the complexity of formula.
• As the example in (11) makes clear, such ‘proofs by induction on complexity of

formula’ are really just a special case of ‘proof by (mathematical) induction’
Induction on the Complexity of Formula:

Suppose that you’ve shown both (i) and (ii) below. It follows that every formula ϕ in the
language has property P.
(i) Every primitive formula of the language has property P.
(ii) For every complex formula ϕ, if the immediate subformulae of ϕ have property P,
then so does ϕ
3. A Review of Propositional Logic (PL): Natural Deduction
(13) Major Goal of This Section

Let’s provide a purely syntactic characterization of ‘valid inference’ in the PL notation.
• This syntactic characterization will be embodied in a proof system (natural deduction)
• We’re going to lay out some rules – stated entirely in syntactic terms – for deriving
formulae in PL from other formulae.
o As we’ll see, these syntactic rules intuitively capture certain key aspects of the
everyday meaning of the English logical words ‘not’, ‘and’, ‘or’, ‘if..then’
(14) Definition of a Derivation

A derivation is a finite, numbered list of formulae, where each formula is accompanied
by a coded statement indicating how it entered the derivation (see (5))
Such ‘coded statements’ will be of the following two forms:
(i) ‘Assumption’ indicates that the formula is an assumption (premise)
(ii) ‘RULE n, …, m’ Where ‘RULE’ is the name of one of the derivation rules
below, and ‘n, …, m’ are the numbers of the formulae in
the derivation that are ‘input’ to the rule
6
Fall 2013 Ling 720
(15) Some Terminology and Notation
a. Assumptions of the Derivation:

The formulae in the derivation that are accompanied by ‘Assumption’
b. Conclusion of the Derivation: The final line in the derivation.
c. Key notation: { ϕ1 , ϕ2 , ϕ3 … } ⊢ ψ
‘There is a derivation where a subset of { ϕ1 , ϕ2 , ϕ3 … }
are the assumptions, and ψ is the conclusion’
Note: Given the way ‘⊢’ is defined in (15c), the set { ϕ1 , ϕ2 , ϕ3 … } could be infinite.
This will make certain proofs about the system easier later on…
Now let’s get to stating and illustrating some of the deduction rules!...
(16) The Rule of ‘&-Introduction’ (I&)

The following is an acceptable derivation:
1. …
… …
n1 ϕ
… …
n2 ψ
… …
m (ϕ & ψ) I& n1, n2
Illustration: { p, q, r } ⊢ ((p & q) & (p & r))
1. p Assumption
2. q Assumption
3. r Assumption
4. (p & q) I& 1,2
5. (p & r) I& 1,3
6. ((p & q) & (p & r)) I& 4,5
Intuitive Motivation:
If we can (in English) assert some sentence ‘S1’, and some sentence ‘S2’, then we can
also assert the sentence ‘S1 and S2’
Note:
From now on, I’ll leave ‘the following is an acceptable derivation’ implicit in the statement of
the deduction rules…
7
Fall 2013 Ling 720
(17) The Rule of ‘&-Elimination’ (E&)
1. … 1. …
… … … …
n (ϕ & ψ ) n (ϕ & ψ )
… … … …
m ϕ E& n m ψ E& n
If we can (in English) assert some sentence ‘S1 and S2’, then we can also assert ‘S1’.
If we can (in English) assert some sentence ‘S1 and S2’, then we can also assert ‘S2’.
Illustration: Proving the Associativity of ‘&’ (Part 1)

1. (p & (q & r)) Assumption
2. p E& 1
3. (q & r) E& 1
4. q E& 3
5. (p & q) I& 2,4
6. r E& 3
7. ((p & q) & r) I& 5,6
Note: By a similar derivation, we can show ((p & q) & r) ⊢ (p & (q & r))
(18) The Rule of ‘Repetition’

The following rule doesn’t add anything interesting, but it makes proofs easier.
1. …
… …
n. ϕ
… …
m. ϕ Repetition n
Illustration:
1. (p & (q & r)) Assumption
2. p E& 1
3. (q & r) E& 1
4. q E& 3
5. r E& 3
6. (p & q) I& 2,4
7. r Repetition 5
8. ((p & q) & r) I& 6,7
8
Fall 2013 Ling 720
(19) The Rule of ‘-Elimination’ (E)
The following derivation rule is often referred to as ‘modus ponens’
1. …
… …
n1 (ϕ  ψ)
… …
n2 ϕ
… …
m ψ E n1, n2
If we can (in English) assert the sentence ‘If S1 then S2’, and we can assert the sentence
‘S1’, then we can also assert ‘S2’
Illustration: { (p  (q  r)), p, q } ⊢ r
1. (p  (q  r)) Assumption
2. p Assumption
3. q Assumption
4. (q  r) E 1,2
5. q Repetition 3
6. r E  4,5
(20) The Rule of ‘-Introduction’ (I) and ‘Conditional Proof’
If whenever we assume ‘S1’ it follows that ‘S2’, then we can assert ‘if S1 then S2’.
The Rule:
The following is an acceptable derivation, as long as no line j > m makes reference to any
lines from n1 to n2.
1. …
… …
n1 ϕ Assumption
…
n2 ψ
m (ϕ  ψ) I
• In such a derivation, we say that the formulae occurring on lines n1 to n2 have been
dropped or withdrawn
• Note the key restriction that once a formula has been dropped, no subsequent lines
can make reference to it.
9
Fall 2013 Ling 720
Illustration: ( (p & q)  r ) ⊢ (p  (q  r))
1. ( (p & q)  r ) Assumption
2. p Assumption
3. q Assumption
4. (p & q) I& 2,3
5. r E 1, 4
6. (q  r) I
7. (p  (q  r)) I
(21) The Rule of ‘∨-Introduction’ (I∨)
1. … 1. …
… … … …
n. ϕ n. ψ
… … … …
m. (ϕ ∨ ψ) I∨ n m. (ϕ ∨ ψ) I∨ n
If we can (in English) assert a sentence ‘S1’, then for any sentence ‘S2’, we can assert ‘S1
or S2’ and ‘S2 or S1’ (inclusive ‘or’)
(22) The Rule of ‘∨-Elimination’ (E∨)
1. …
… …
n1 (ϕ ∨ ψ)
… …
n2 (ϕ  χ)
… …
n3 (ψ  χ)
… …
m χ E∨ n1, n2, n3
If we can (in English) assert ‘S1 or S2’, ‘if S1 then S3’ and also ‘if S2 then S3’, then we can
also assert ‘S3’
10
Fall 2013 Ling 720
(23) Illustration of the Rules for Disjunction: Associativity of ‘∨’ (Part 1)
(p ∨ (q ∨ r)) ⊢ ((p ∨ q) ∨ r)
1. (p ∨ (q ∨ r)) Assumption
2. p Assumption
3. (p ∨ q) I∨ 2
4. ((p ∨ q) ∨ r) I∨ 3
5. (p  ((p ∨ q) ∨ r)) I
6. (q ∨ r) Assumption
7. q Assumption
8. (p ∨ q) I∨ 7
9. ((p ∨ q) ∨ r) I∨ 8
10. (q  ((p ∨ q) ∨ r)) I
11. r Assumption
12. ((p ∨ q) ∨ r) I∨ 11
13. (r  ((p ∨ q) ∨ r)) I
14. ((p ∨ q) ∨ r) E∨ 6, 10, 13
15. ((q ∨ r)  ((p ∨ q) ∨ r)) I
16. ((p ∨ q) ∨ r) E∨ 1, 5, 15
(24) The Rule of ‘~ -Elimination’ (E~)

For various reasons, it will help to have a special symbol indicating that a contradiction
has been reached: ⊥ ‘falsum’
1. …
… …
n1 ϕ
… …
n2 ~ϕ
… …
m ⊥ E~ n1 n2
(25) The Rule of ‘~ -Introduction’ (I~)

With this special symbol, we can now state the following rule of I~
1. …
… …
n1 ϕ Assumption
…
n2 ⊥
m ~ϕ I~
If whenever we assume ‘S’ a contrdiction follows, then we can assert ‘not S’.
11
Fall 2013 Ling 720
(26) Illustration of the Rules for Negation: Double Negation (Part 1)
p ⊢ ~~p
1. p Assumption
2. ~p Assumption
3. ⊥ E~ 1,2
4. ~~p I~
(27) The Rule of ‘Ex Falso Sequitur Quodlibet’ (EFSQ)
Intuitive Motivation: Anything follows from a contradiction.
1. …
… …
n ⊥
m ϕ EFSQ
Illustration: { (p ∨ q) , ~p } ⊢ q
1. (p ∨ q) Assumption
2. ~p Assumption
3. p Assumption
4. ⊥ E~ 2,3
5. q EFSQ
6. (p  q) I
7. q Assumption
8. q Repetition 7
9. (q  q) I
10. q E∨ 1, 6, 9
(28) The Rule of ‘Double Negation’ (~~)
1. …
… …
n ~~ ϕ
… …
m ϕ ~~ n
Note:
Although our system without (28) can prove one half of the equivalence ‘ϕ iff ~~ϕ’ (26), we
need the rule in (28) for our system to derive the complete equivalence.
12
Fall 2013 Ling 720
4. The Power of Our Natural Deduction System for PL
Although we our system has just 11 (really, 10) relatively simple rules, it can capture a great
many intuitively valid inferences!
(29) Derivation of DeMorgans, Part 1 ~(p & q) ⊢ (~p ∨ ~q)
1. ~(p & q) Assumption

2. ~(~p ∨ ~q) Assumption
3. p Assumption
4. q Assumption
5. (p & q) I& 3,4
6. ⊥ E~ 1, 5
7. ~q I~
8. (~p ∨ ~q) I∨ 7
9. ⊥ E~ 2, 8
10 ~p I~
11. (~p ∨ ~q) I∨ 10
12. ⊥ E~ 2, 11
13. ~~(~p ∨ ~q) I~
14. (~p ∨ ~q) ~~ 13
(30) Derivation of DeMorgans, Part 2 (~p ∨ ~q) ⊢ ~(p & q)
1. (~p ∨ ~q) Assumption

2. ~p Assumption
3. (p & q) Assumption
4. p E& 3
5. ⊥ E~ 2, 4
6. ~(p & q) I~
7. (~p  ~(p & q)) I
8. ~q Assumption
9. (p&q) Assumption
10. q E& 9
11. ⊥ E~ 8,10
12. ~(p & q) I~
13. (~q  ~(p & q)) I
14. ~(p & q) E∨ 1, 7, 13
(31) The Big Question

Does our system offer a perfect syntactic characterization of ‘validity’ for PL?
• Does every derivation correspond to a valid inference?
• Does every valid inference in PL correspond to a derivation?
How would we even show this?...
13
Fall 2013 Ling 720
First Order (Predicate) Logic: Syntax and Natural Deduction 1
A Reminder of Our Plot
• I wish to provide some historical and intellectual context to the formal tools that logicians
developed to study the semantics of artificial languages.
• For this reason, I’m beginning with a purely syntactic presentation of two key logical
systems: Propositional Logic (PL) and First Order (Predicate) Logic (FOL).
• In our last notes, we covered PL. Now, we’ll get a (syntactic) introduction to FOL.
1. A Review of First Order (Predicate) Logic (FOL): Syntax and Informal Semantics 2
The system of First Order Logic (FOL) is intended to capture the inferences that depend upon:
(i) the meaning of the so-called ‘sentential connectives’: and, or, if…then, and not
(ii) the meaning of the quantifiers ‘every’ and ‘some’
(1) The Vocabulary of Symbols
a. The Logical Constants:

(i) ~ Negation ‘It is not the case that…’
(ii) & Conjunction ‘and’
(iii) ∨ Disjunction ‘or’ (inclusive)
(iv)  (Material) Implication ‘if…then’
(v) ∀ Universal Quantifier ‘for all…’
(vi) ∃ Existential Quantifier ‘there is an…’
b. Syntactic Symbols: (,)
a. The Non-Logical Constants (a.k.a ‘The Logical Variables’)
(i) An infinite set of predicate letters: {P, Q, R, B, … P1, P2, P3, P4, …}
• Each predicate letter has an associated ‘arity’ (unary, binary, etc.)
• Each predicate letter ‘stands for’ a property or a relation
(ii) An infinite set of individual constants: {a, b, c, …, a1, a2, a3 , … }

• The individual constants ‘stand for’ proper names (Bill, John, etc.)
(iii) An infinite set of variables: {x, y, z, …, x1, x2, x3, … }

1
These notes are based upon material in the following required reading: Gamut (1991), Chapter 3 pp. 65-83,
Chapter 4 pp. 128-148; Partee et al. (1993), Chapter 7 pp. 135-140.
2
My discussion here will assume prior familiarity with the overall system of First Order Logic. Students are
1
Fall 2013 Ling 720
(2) The Definition of a ‘Well-Formed Formula’ (WFF) in FOL
a. If ϕ is an n-ary predicate letter and each of α1, …, αn is either an individual

constant or a variable, then ϕα1…αn ∈ WFF
1. ~ϕ ∈ WFF
2. (ϕ & ψ) ∈ WFF
3. (ϕ ∨ ψ) ∈ WFF
4. (ϕ  ψ) ∈ WFF
c. If ϕ ∈ WFF and υ is a variable, then

1. ∀υϕ ∈ WFF
2. ∃υϕ ∈ WFF
Notes:
• The clause in (2a) creates the atomic formulae of FOL. The clause in (2c) creates the
universal formulae and existential formulae.
• The set WFF includes formulae with ‘free variables’ and ‘vacuous quantification’ (defined
properly later)
Hxb (free variables)
∃y∀xHxb (vacuous quantification)
∃yHxb (free variables and vacuous quantification)
(3) Using FOL To Encode Sentences of English

We can use the syntactic rules in (2) and the informal semantics in (1) to write FOL
formulae that ‘encode’ certain statements of English:
a. Sentence: ‘If Bill or John is leaving, then Mary and Sue aren’t happy.’
Encoding: ((Lb ∨ Lj)  (~Hm & ~Hs))
b. Sentence: ‘Every cat gave a book to Bill.’

Encoding: ∀x(Cx  ∃y(By & Gxyb))
In setting up such encodings, it is critical to supply a ‘key’, indicating what the predicate
letters and individual constants ‘stand for’:
Key Lx: x is leaving b: Bill

Hx: x is happy j: John
Cx: x is a cat m: Mary
Bx: x is a book s: Sue
Gxyz: x gave y to z
2
Fall 2013 Ling 720
An Important Note:
• In the key above, formulae of the form ‘Gxyz’ are interpreted so that x is the ‘subject’ of
gave, while ‘y’ is the direct object, and ‘z’ is the indirect object.
• Nothing forces this however. We could just have easily had the following in our key:
Gxyz: z gave y to x
• Such ‘right-to-left’ readings of atomic formulae will be useful to us later, when

we’re mechanically translating sentences of English into sentences of FOL…
(4) Key Definition: Scope

If ∀υψ is a subformula of ϕ, then ψ is the scope of (this occurrence of) ‘∀υ’ in ϕ
If ∃υψ is a subformula of ϕ, then ψ is the scope of (this occurrence of) ‘∃υ’ in ϕ
Illustration: In the formula ‘∃x(Qx & ∀y(Py  ∃zSxyz))’

(i) The scope of ∃x = (Qx & ∀y(Py  ∃zSxyz))
(ii) The scope of ∀y = (Py  ∃zSxyz)
(iii) The scope of ∃z = Sxyz
(5) Key Definition: Free and Bound Variables
a An occurrence of the variable υ in the formula ϕ is free in ϕ if (i) and (ii) hold:
(i) that occurrence of υ does not occur directly to the right of either ∃ or ∀
(ii) that occurrence of υ is not in the scope of any occurrence of ∃υ or ∀υ in ϕ
b. An occurrence of the variable υ is bound by ∀υ (∃υ) in ϕ if (i) and (ii) hold:

(i) ∀υψ (∃υψ) is a subformula of ϕ
(ii) That occurrence of υ is free in ψ
Illustration: In the formula ‘∀x(Px & ∃xBx)’
(i) The occurrence of ‘x’ in ‘Px’ is free in ‘(Px & ∃xBx)’

(ii) The occurrence of ‘x’ in ‘Bx’ is free in ‘Bx’
(iii) The occurrence of ‘x’ in ‘Bx’ is not free in ‘(Px & ∃xBx)’
(iv) The first occurrence of ‘x’ in ‘(Px & ∃xBx)’ is bound by ‘∀x’
(v) The occurrence of ‘x’ in ‘Bx’ is bound by ‘∃x’
(vi) The occurrence of ‘x’ in ‘Bx’ is not bound by ‘∀x’
(6) Key Definition: Sentence

ϕ is a sentence of FOL if (i) ϕ ∈ WFF, and (ii) there are no free variables in ϕ
3
Fall 2013 Ling 720
2. A Review of First Order Logic (FOL): Natural Deduction
(7) Major Goal of This Section

Let’s provide a purely syntactic characterization of ‘valid inference’ in the FOL notation.
• This syntactic characterization will be embodied in a proof system (natural deduction)
• We’re going to lay out some rules – stated entirely in syntactic terms – for deriving
formulae in FOL from other formulae.
o As we’ll see, these syntactic rules intuitively capture certain key aspects of the
everyday meaning of the English logical words ‘every’ and ‘some’
(8) Features Inherited from PL Natural Deduction

All the following will directly imported from our system of natural deduction for PL:
a. Definition of ‘derivation’
b. Turnstyle notation ‘⊢’
c. The rules I&, E&, Repetition, I, E, I∨, E∨, I~, E~, EFSQ, ~~
(9) Special Feature of FOL Natural Deduction

ϕ can be an assumption in an FOL derivation iff ϕ is a sentence.
Our natural deduction system for FOL adds four new rules.
(10) Special Notation for Statement of Deduction Rules for FOL
If ϕ ∈ WFF, α is an individual constant, and υ is a variable, then ‘[α/υ]ϕ’ is the formula

just like ϕ, except that every free instance of υ is replaced with an instance of α:
[b/x](Pcx & Dabx) = (Pcb & Dabb)
(11) The Rule of ‘∃-Introduction’ (I∃)
1. …
… …
n [α/υ]ϕ
… …
m ∃υϕ I∃ n
If we can (in English) assert for a particular thing α that ‘ϕ’ is true of α, then we can
assert that there is something that ‘ϕ’ is true of.
4
Fall 2013 Ling 720
(12) The Rule of ‘∀-Elimination’ (E∀)
1. …
… …
n ∀υϕ
… …
m [α/υ]ϕ E∀ n
Intuitive Motivation
If we can (in English) assert that ‘ϕ’ is true of everything, then for any particular thing α,
we can assert that ‘ϕ’ is true of α,
Illustration of I∃ and E∀: ∀xPx ⊢ ∃xPx

1. ∀xPx Assumption
2. Pb E∀ 1
3. ∃xPx I∃ 3
(13) The Rule of ‘∀-Introduction’ (I∀)
If we can show that ‘ϕ’ is true of an arbitrary entity α (‘arbitrary’ = we’ve not assumed
anything about α whatsoever), then we can assert that ‘ϕ’ is true of everything.
Key Definition:
If ‘[α/υ]ϕ’ appears in a derivation at line n, then α is arbitrary at line n if (i) and (ii) hold
(i) α does not appear in any (non-dropped) assumptions in the derivation
(ii) α does not appear in ϕ
The Rule:
The following is an acceptable derivation, as long as α is arbitrary at line n.
1. …
… …
n [α/υ]ϕ
… …
m ∀υϕ I∀ n
Illustration: ∀x∀yPxy ⊢ ∀xPxx
1. ∀x∀yPxy Assumption
2. ∀yPay E∀ 1
3. Paa E∀ 2
4. ∀xPxx I∀ 3
• Note: ‘Paa’ = ‘[a/x]Pxx’, and a is arbitrary (in ‘[a/x]Pxx’) in at line 3.
5
Fall 2013 Ling 720
(14) The Rule of ‘∃-Elimination’ (E∃)
If we can assert (i) that ‘ϕ’ is true of something, and (ii) that if ‘ϕ’ is true of an arbitrary
entity α, then ψ must be true (‘arbitrary’ = we’ve not assumed anything about α
whatsoever), then we can assert that ‘ψ’ is true.
Key Definition:
If ‘[α/υ]ϕ  ψ’ appears in a derivation at line n, then α is arbitrary at line n if (i)-(iii):
(i) α does not appear in any (non-dropped) assumptions in the derivation
(ii) α does not appear in ϕ
(iii) α does not appear ψ
The Rule:
The following is an acceptable derivation, as long as α is arbitrary at line n.
1. …
… …
n1 ∃υϕ
… …
n2 [α/υ]ϕ  ψ
… …
m ψ E∃ n1, n2
Illustration: ∃x∀yPxy ⊢ ∀y∃xPxy
1. ∃x∀yPxy Assumption
2. ∀yPby Assumption
3. Pba E∀ 2
4. ∃xPxa I∃ 3
5. ∀yPby  ∃xPxa I
6. ∃xPxa E∃ 1,5
7. ∀y∃xPxy I∀ 6
• Note:
‘∀yPby’ = ‘[b/x]∀yPxy’, and b is arbitrary (in ‘[b/x]∀yPxy’  ∃xPxa’) at line 5
• Note: ‘∃xPxa’ = ‘[a/y]∃xPxy’, and a is arbitrary (in ‘[a/y]∃xPxy’) at line 6.
Side Comment:
- The rules I∀ and E∃ add significantly to the complexity of our natural deduction system.
- However, they are crucial for system to capture all the valid inferences in FOL
- They also greatly complicate the proof that our natural deduction system for FOL is ‘complete’
6
Fall 2013 Ling 720
3. The Power of Our Natural Deduction System for FOL
Although we our system for FOL has just 15 (really, 14) relatively simple rules, it can capture a
great many intuitively valid inferences!
(15) Derivation of Quantifier Negation, Part 1 ~∃xPx ⊢∀x~Px
1. ~∃xPx Assumption
2. Pa Assumption
3. ∃xPx I∃ 2
4. ⊥ E~ 1, 3
5. ~Pa I~
6. ∀x~Px I∀ 6
• Note: ‘~Pa’ = ‘[a/x]~Px’ and a is arbitrary (in ‘[a/x]~Px’) at line 5
(16) Derivation of Quantifier Negation, Part 2 ∀x~Px ⊢ ~∃xPx
1. ∀x~Px Assumption
2. ∃xPx Assumption
3. Pa Assumption
4. ~Pa E∀ 1
5. ⊥ E~ 3,4
6. (Pa  ⊥) I
7. ⊥ E∃ 2, 6
8. ~∃xPx I~
• Note: ‘(Pa ⊥)’ = ‘[a/x](Px⊥)’ and a is arbitrary (in ‘[a/x](Px⊥)’ at line 6
(17) The Big Question

Does our system offer a perfect syntactic characterization of ‘validity’ for FOL?
• Does every derivation correspond to a valid inference?
• Does every valid inference in FOL correspond to a derivation?
How would we even show this?...
7
Fall 2013 Ling 720
Problem Set on Propositional Logic and First Order Logic:
Translation and Natural Deduction 1
(1) Translation from English to Propositional Logic
Please encode the following English statements as formulae in PL. Be sure to include a
‘key’ indicating what English statements each propositional letter ‘stands for’.
a. God willing, peace will come.

b. If it isn’t summer, then it is damp and cold, if it is evening or night.
c. If you do not help me if I need you, I will not help you if you need me.
d. John comes only if Peter does not come.
e. We are going, unless it is raining.
f. If Johnny is nice, then he will get a bicycle from Santa Clause, whether he wants
one or not.
(2) Derivations in Our Natural Deduction System for Propositional Logic
Please provide derivations in our natural deduction system for PL establishing each of the
following.
a. (p ∨ q) ⊢ ((p  q)  q)
b. (p & ~q) ⊢ ~(p  q)
c. (p  ~q) ⊢(q  ~p)
d. (p & (q ∨ r)) ⊢ ((p & q) ∨ (p & r))
e. ((p & q) ∨ (p & r)) ⊢ (p & (q ∨ r))
(3) Translations from English to First Order Logic
Please encode the following English statements as formulae in FOL. Be sure to include a
‘key’ indicating what the predicate letters and individual constants ‘stand for’.
a. Although John and Mary love each other deeply, they make each other unhappy.
b. It is not the case that all ambitious people are not honest.
c. Lyn got a present from John, but she didn’t get anything from Peter.
d. Nobody lives in Hadley who wasn’t born there.
e. People who live in Amherst or close buy own a car.
f. If somebody is noisy, then everybody is annoyed at him.
1
Most of these exercises are taken from Gamut (1991), Volume 1: Chapters 2, 3 and 4
1
Fall 2013 Ling 720
(4) Identifying Sentences of FOL
Please identify whether the following formulae of FOL are sentences or not.
a. (∃xPxa & Bx)

b. ∃x(∀yPxy  Bx)
c. ∀y~∃xPxy
d. ∃x(~Ba  (~∀y(~Pxy ∨ Qb)  Cy))
e. ∀x∀y((Pxy & By)  ∃wCxw)
f. (∀x∀yPyy  Bx)
g. ∀x(∀yPyy  Bx)
(5) Proof by Induction on the Complexity of Formula

Please show the following by constructing a proof on the complexity of formula:
Claim: No formula of FOL begins with a variable.
(6) Derivations in Our Natural Deduction System for First Order Logic
Please provide derivations in our natural deduction system for FOL establishing each of
the following.
a. ∀x(Px & Bx) ⊢ (∀xPx & ∀xBx)

b. ∀xPx & ∀xBx ⊢∀x(Px & Bx)
c. ∃x(Px & Bx) ⊢ (∃xPx & ∃xBx)
d. ∃x~Px ⊢~∀xPx
e. ~∀xPx ⊢∃x~Px
2
Fall 2013 Ling 720
Propositional Logic: Formal Semantics and Valuations 1
1. The Motivation for Developing a Formal Semantics for Propositional Logic
(1) A Central Goal of (Formal) Logic, Since Aristotle

To provide a purely syntactic characterization of valid argumentation.
(2) The Means of Achieving This in Formal Logic
a. A precisely defined formal notation for representing certain aspects of the ‘logical
structure’ of an assertion.
b. A set of syntactically defined rules for deriving formulas in the notation from
other formulas in the notation (e.g., our system of ‘Natural Deduction’).
(3) An Immediate Challenge for This Program
How do we know when we’ve succeeded? How do we know whether our formal system
indeed produces all and only the inferences that are valid (in the notation)?
a. Entailment:
A set of formulae S entails ψ, if ‘whenever’ every ϕ ∈ S is true, so is ψ
• Notation: S⊨ψ
• Note: If S entails ψ, then the inference of ψ from S is valid
b. Soundness:
We’ve established the soundness of our system, if we can show that every
derivation is / corresponds to a valid inference.
• Notation: If S ⊢ ψ, then S ⊨ ψ
c. Completeness:
We’ve established the completeness of our system, if we can show that every
valid inference (in the notation) corresponds to a derivation.
• Notation: If S ⊨ ψ, then S ⊢ ψ,
1
These notes are based upon material in the following required readings: Gamut (1991), Chapter 2 pp. 44-54; Partee
et al. (1993), Chapter 6 pp. 97-112.
1
Fall 2013 Ling 720
(4) The Main Obstacle to Proving Soundness and Completeness
The notion of ‘entailment’ in (3a) is currently too vague and informal for precise
(mathematical) argumentation.
• What, exactly, does it mean to say ‘whenever all the formulae in S are true’?
To have a precise proof of soundness/completeness, we need rigorous explication of

what it means for a formula of PL to be ‘true’.
(5) Towards a Formalization of ‘Truth’ for PL
Let us begin by considering our informal semantics for PL, and how it can be used to
determine a truth-value for a PL formula:
Example Sentence: (p & ~q)
Determining Its Truth (Informally):
a. Step One:
• A formula of PL isn’t true or false in and of itself, but only relative to an

(informal) interpretation (i.e., ‘key’).
• So the first thing an informal interpretation does is it maps the primitive

proposition letters to particular statements of English
p: ‘Seth lives in Northampton.’

q: ‘Rajesh lives in Northampton.’
b. Step Two:
Next, to determine whether the whole formula is true, we first consider the truth
of the English sentences that the proposition letters are mapped to:
‘Seth lives in Northampton’ is true

‘Rajesh lives in Northampton’ is true
c. Step Three:
Finally, once we have the truth-values of the English translations, the meanings of
the (English) logical connectives determine a truth-value for the whole formula.
~q = ‘It is not the case that Rajesh lives in Northampton’ = false

(p & ~q) = false
2
Fall 2013 Ling 720
It seems, then, that an informal interpretation of PL determines a truth-value for a PL formula by
• Mapping the primitive proposition letters to truth-values (indirectly, through English

translations)
• Then, for complex formulae, the main connective determines the truth-value of the whole
formula, based on the truth-values of the component formulae
Thus, it seems we can develop a more abstract, mathematically precise conception of ‘PL
interpretation’ in the following way...
2. A Review of Propositional Logic (PL): Formal Semantics and Valuations 2
(6) Definition of a ‘Valuation’ for Propositional Logic
A valuation V is a function from the well-formed formulae of PL to the set of truth-

values {1,0} (V: WFF  {0,1}) such that:
a. If ϕ is a proposition letter, then V(ϕ) ∈ {1,0} (redundant, but helpful)
1. V(~ϕ) = 1 iff V(ϕ) = 0
2. V((ϕ & ψ)) = 1 iff V(ϕ) = 1 and V(ψ) = 1
3. V((ϕ ∨ ψ)) = 1 iff V(ϕ) = 1 or V(ψ) = 1 (inclusive ‘or’)
4. V((ϕ  ψ)) =1 iff V(ϕ) = 0 or V(ψ) = 1 (inclusive ‘or’)
(c. V(⊥) = 0 (This is need for the proof of soundness))
(7) The Truth-Tables of the Logical Connectives
a. ϕ || ~ϕ b. ϕ | ψ || (ϕ & ψ) c. ϕ | ψ || (ϕ ∨ ψ)
1 0 1 1 1 1 1 1
0 1 1 0 0 1 0 1
0 1 0 0 1 1
d. ϕ | ψ || (ϕ  ψ) 0 0 0 0 0 0
1 1 1
1 0 0
0 1 1
0 0 1
2
My discussion here will assume prior familiarity with the formal semanitcs of Propositional Logic, particularly
truth-tables, validity, tautology, logical-equivalence, etc. Students are referred to Partee et al. (1993), Chapter 6 for
crucial background.
3
Fall 2013 Ling 720
(8) Calculating Truth-Tables for Complex Formulae
• In top row, list all the proposition letters in the formula (followed by double lines).
• Then, list all the sub-formulae, going from bottom-up (followed by single lines)
• Each row below corresponds to one of the possible valuations of the proposition
letters in the sentence (2n rows, where n = number of proposition letters)
Illustration: ((p & q)  ~r)
p| q | r || ~r | (p & q) | ((p & q)  ~r)

1 1 1 0 1 0
1 1 0 1 1 1
1 0 1 0 0 1
1 0 0 1 0 1
0 1 1 0 0 1
0 1 0 1 0 1
0 0 1 0 0 1
0 0 0 1 0 1
We can use this notion of a ‘valuation’ to provide a more precise characterization of our key
semantic concepts!
(9) Tautology
a. Informal Notion: S is a tautology if S is ‘necessarily true’
b. Formal Notion: ϕ is a tautology if for every valuation V, V(ϕ) = 1
Illustration: (p ∨ ~p) p || ~p | (p ∨ ~p)

1 0 1
0 1 1
(10) Contradiction
a. Informal Notion: S is a contradiction if S is ‘necessarily false’
b. Formal Notion: ϕ is a contradiction if for every valuation V, V(ϕ) = 0
Illustration: (p & ~p) p || ~p | (p & ~p)

1 0 0
0 1 0
(11) Contingent
a. Informal Notion: S is contingent if S is ‘possibly true and possibly false’
b. Formal Notion: ϕ is contingent if there is a valuation V such that V(ϕ) = 1,
and a valuation V’ such that V’(ϕ) = 0
4
Fall 2013 Ling 720
Key Consequences
• A formula ϕ of PL is a tautology iff ~ϕ is a contradiction.
• A formula ϕ of PL is contingent iff ~ϕ is contingent
(12) Logical Equivalence
a. Informal Notion: S and S’ are logically equivalent if S is true whenever S’ is.
b. Formal Notion:
ϕ and ψ are logically equivalent if for every valuation V, V(ϕ) = V(ψ)
Illustration: p, ~~p p || ~p | ~~p

1 0 1
0 1 0
Key Consequences
• Formulae ϕ and ψ of PL are logically equivalent iff ~ϕ and ~ψ are logically equivalent.
• If ϕ and ψ are both tautologies, then ϕ and ψ are logically equivalent.
• If ϕ and ψ are both contradictions, then ϕ and ψ are logically equivalent.
Given our definition of valuation in (6), our definition of ‘logical equivalence’ in (12b) captures
a wide variety of intuitive equivalences, many of which we also derived in our proof system!
(13) DeMorgans Laws: (p ∨ q) is Logically Equivalent to ~(~p & ~q)
p | q || (p ∨ q) | ~p | ~q | (~p & ~q) | ~(~p & ~q )

1 1 1 0 0 0 1
1 0 1 0 1 0 1
0 1 1 1 0 0 1
0 0 0 1 1 1 0
(14) Associativity of &: ((p ∨ q) ∨ r) is Logically Equivalent to (p ∨ (q ∨ r))
p | q | r || (p ∨ q) | ((p ∨ q) ∨ r ) | (q ∨ r) | ((p ∨ (q ∨ r ))
1 1 1 1 1 1 1
1 1 0 1 1 1 1
1 0 1 1 1 1 1
1 0 0 1 1 0 1
0 1 1 1 1 1 1
0 1 0 1 1 1 1
0 0 1 0 1 1 1
0 0 0 0 0 0 0
5
Fall 2013 Ling 720
(15) Key Result: Substitution of Logically Equivalent Formulae
Informal Statement:
If ϕ and ψ are logically equivalent, then replacing ϕ with ψ will have no effect on the
truth-value of a larger sentence.
Formal Statement:
Suppose that ϕ and ψ are logically equivalent, that ϕ is a subformula of χ, and that χ’ is
the formula just like χ, except that all instances of ϕ are replaced with ψ. It follows that χ
and χ’ are logically equivalent.
Illustration: (p  q) and (~~p  q): p | q || ( p  q ) | ~p | ~~p | ( ~~p  q)

1 1 1 0 1 1
1 0 0 0 1 0
0 1 1 1 0 1
0 0 1 1 0 1
(16) Key Result: Dropping Operators from Our System
• Informally speaking, what (15) shows is that if ϕ and ψ are logically equivalent, then
anything we can ‘express’ in PL with ϕ, we can also ‘express’ with ψ
• Now, recall the following key equivalences (proofs left to the student):
(i) (ϕ  ψ) is logically equivalent to (~ϕ ∨ ψ)
(ii) (ϕ ∨ ψ) is logically equivalent to ~(~ϕ & ~ψ)
(iii) (ϕ & ψ) is logically equivalent to ~(~ϕ ∨ ~ψ)
• It follows from (i) that anything we can ‘express’ in PL with ‘’, we can also express
with ‘~’ and ‘∨’
• It follows from (ii) that anything we can ‘express’ in PL with ‘∨’, we can also express
with ‘~’ and ‘&’.
• It follows from (iii) that anything we can ‘express’ in PL with ‘&’ we can also
express with ‘~’ and ‘&’
• Consequently, we can drop ‘’ from our PL system (leaving ‘~’, ‘&’ ‘∨’), and
still have an equivalently ‘expressive’ system
• Consequently, we can drop ‘∨’ from our system (leaving ‘&’ and ‘~’), and still
have an equivalently ‘expressive’ system
• Consequently, we can drop ‘&’ from our system (leaving ‘∨’ and ‘~’), and still
have an equivalently ‘expressive’ system
6
Fall 2013 Ling 720
(17) Key Result: Defining Operators in Terms of Other Operators
Given the result in (16), we could view the operators ‘’ and ‘∨’ (or ‘’ and ‘&’) not as
primitive operators, but as special abbreviations for more complex formulae
• (ϕ  ψ) is ‘shorthand’ for (~ϕ ∨ ψ)

• (ϕ ∨ ψ) is ‘shorthand’ for ~(~ϕ & ~ψ)
We’ll make use of this commonplace idea in (17) later on, as it will greatly simplify the
definitions of the logical languages we use for the semantic analysis of English…
Finally, the notion of ‘valuation’ provides us with a more rigorous definition of ‘entailment’…
(18) Preliminary Definition

If S is a set of formulae, then V is a valuation for S if for every ϕ ∈ S, V(ϕ) = 1
(19) Entailment
a. Informal Notation:
A set of formulae S entails ψ, if ‘whenever’ every ϕ ∈ S is true, so is ψ
b. Formal Notion:
A set of formulae S entails ψ, if every valuation V of S is such that V(ψ) = 1
(20) Key Consequence: Entailment and Tautology
Let S be a finite set of formulae { ϕ1, … , ϕn }. S entails ψ iff ((ϕ1 & … & ϕn)  ψ) is a
tautology.
Proof: Suppose that S entails ψ. Thus, for any valuation V, if V((ϕ1 & … & ϕn)) = 1 then
V is a valuation for S, and so V(ψ) = 1. Thus, for any valuation V, if V((ϕ1 & …
& ϕn)) = 1, then V(ψ) ≠ 0, and so V(((ϕ1 & … & ϕn)  ψ)) = 1.
Suppose that ((ϕ1 & … & ϕn)  ψ) is a tautology. Now suppose that V is a
valuation of S. It follows that V((ϕ1 & … & ϕn)) = 1. Moreover, since ((ϕ1 & …
& ϕn)  ψ) is a tautology, it follows that V(((ϕ1 & … & ϕn)  ψ)) = 1.
Consequently, it must be the case that V(ψ) ≠ 0, and so V(ψ) = 1.
(21) Major Consequence: Computability of Entailment for PL
One can ‘effectively’ compute whether a finite set of PL formulae S entail a formula ψ
• Just compute the truth-table for (S  ψ) and check whether it’s a tautology!
7
Fall 2013 Ling 720
Now that we have this notion of ‘valuation’, the key issues in (3) become much more tractable!
(22) Soundness of PL
• If S ⊢ ψ, then S ⊨ ψ
• If S ⊢ ψ, then if V is a valuation for S, V(ψ) = T
(23) Completeness of PL
• If S ⊨ ψ, then S ⊢ ψ
• If every valuation V of S is also a valuation of ψ, then S ⊢ ψ
In the next set of notes, we’ll see how (22) and (23) can be rigorously proved!
8
Fall 2013 Ling 720
First Order Logic: Formal Semantics and Models 1
(1) Goal for These Notes

Just as we did for PL, we need a mathematically rigorous explication of what it is for a
sentence of FOL to be ‘true’.
• This will give us an appropriately precise definition of ‘entailment’, which
will ultimately allow us to prove that FOL is sound and complete…
(2) Towards a Formalization of ‘Truth’ for FOL

Let’s begin by considering sentences of FOL without quantifiers, and how an informal
semantics (‘key’) determines a truth-value for them.
Example Sentence: (Fa & ~Pbc)
Determining Its Truth (Informally)
a. Step One: Consult the Key

A ‘key’ for an FOL encoding maps individual constants of FOL to proper names,
and predicate letters of FOL to properties/relations.
Fx: x is French a: Angelika Kratzer

Pxy: x is older than y b: Seth Cable c: Rajesh Bhatt
b. Step Two: Consult the Facts

Next, we consider whether the properties/relations denoted by the English
expressions actually hold of the entities named by the English names.
• Angelika Kratzer is not French.

• Seth Cable is not older than Rajesh Bhatt
This determines the truth-value of the atomic sentneces of FOL.
c. Step Three:
Finally, once we have the truth-values of the atomic sentences, the meanings of
the (English) logical connectives determine a truth-value for the whole formula.
~Pbc = ‘It is not the case that Seth is older than Rajesh’ = True
Fa = ‘Angelika Kratzer is French’ = False
(Fa & ~Pbc) = false
1
These notes are based upon material in the following required readings: Gamut (1991), Chapter 3 pp. 87-101;
Partee et al. (1993), Chapter 7 pp. 140-152, Chapter 13 pp. 321-331.
1
Fall 2013 Ling 720
It seems, then, that an informal interpretation of FOL determines a truth-value for an FOL
sentence (without quantification) by
• Mapping the individual constants to entities (indirectly, through English translations)
• Mapping the predicate letter to some set (property) or set of n-tuples (relation)
• For atomic sentences, the sentence is true if the individual(s) named by the constant are
‘in’ the set or relation denoted by the predicate letter.
• For complex sentences, the main connective determines the truth-value of the whole
formula, based on the truth-values of the component formulae
Thus, it seems we can develop a more abstract, mathematically precise conception of ‘FOL
interpretation’ in the following way...
2. A Review of First Order Logic (FOL): Formal Semantics and Models 2
(7) Definition of a ‘Model’ for First Order Logic
A model M is a pair <D, I> consisting of:
a. A non-empty set D, called the ‘domain of M’
b. A function I, whose domain is the individual constants and predicate letters, and
whose range satisfies the following conditions:
(i) If α is an individual constant, then I(α) ∈ D

(ii) If Φ is an n-ary predicate letter, then I(Φ) ⊆ Dn
Note:
• If Φ is a unary predicate letter, then I(Φ) ⊆ D1 = D (a subset of D)
• If Φ is a binary predicate letter, then I(Φ) ⊆ D2 = D×D (a set of pairs from D)
• If Φ is a ternary predicate letter, then I(Φ) ⊆ D3 = D×D×D (a set of triples from D), etc.
(8) Illustration:
The following is a model of FOL < { Angelika, Seth, Rajesh }, I >, where I consists of at
least the following mappings:
I(a) = Angelika I(b) = Seth I(c) = Rajesh
I(F) = ∅ I(P) = { <Angelika, Rajesh>, <Angelika, Seth>, <Rajesh, Seth>}
2
My discussion here will assume prior familiarity with the formal semantics of First Order Logic. Students are
2
Fall 2013 Ling 720
With this notion of ‘model’ in place, we can introduce the key formal semantic notion below:
(9) Valuation of FOL, Relative to a Model (To Be Revised)
Let M be a model <D, I>. Then the ‘valuation based on M’ (VM) is a function whose
domain is the set of FOL sentences (without quantifiers), whose range is {0,1}, and
which satisfies the conditions below:
(i) If ϕ = Φα1... αn, then VM(ϕ) = 1 iff <I(α1)... I(αn)> ∈ I(Φ)
(ii) If ϕ = ~ψ, then VM(ϕ) = 1 iff VM(ψ) = 0
(iii) If ϕ = (ψ & χ), then VM(ϕ) = 1 iff VM(ψ) = 1 and VM(χ) = 1
(iv) If ϕ = (ψ ∨ χ), then VM(ϕ) = 1 iff VM(ψ) = 1 or VM(χ) = 1
(v) If ϕ = (ψ  χ), then VM(ϕ) = 1 iff VM(ψ) = 0 or VM(χ) = 1
(10) Illustration
Let M be the model (partially) defined in (8). From (9), it follows that
• VM(Fa & ~Pbc) =1 iff (by 9iii)
• VM(Fa) = 1 and VM(~Pbc) = 1 iff (by 9ii)
• VM(Fa) = 1 and VM(Pbc) = 0 iff (by 9i)
• I(a) ∈ I(F) and <I(b), I(c) > ∉ I(P) iff (by definition of I in (8))
• Angelika ∈ ∅ and <Seth, Rajesh> ∉ { <Angelika, Rajesh>, <Angelika, Seth>,

<Rajesh, Seth>}
Thus, we can calculate that VM(Fa & ~Pbc) = 0
In this way, the definition of ‘model’ in (8) and ‘valuation’ in (9) mirrors the way our earlier
‘informal interpretation’ (key) maps the formula ‘(Fa & ~Pbc)’ to the truth-value false
PROBLEM: We forgot about quantifiers! How do we add them into the picture?
3
Fall 2013 Ling 720
(11) Key Problem: The Semantics of Formulae with Free Variables
Consider an FOL sentence like ‘∃xPx’.

• This is a complex formula made up of the quantifier ‘∃x’ and the atomic formula ‘Px’
• Moreover, we’re going to want this formula to end up being entailed by ‘Pa’.
• Thus, we’re going to want the truth-value of this formula to somehow be

determined by some kind of semantic value that the formula ‘Px’ has (and for
that value to ‘connected’ with the truth of Pa in some fashion…)
• BUT HOW IN GOD’S NAME DO WE DO THAT???
(12) A Naive Intuition, to Get Us Started
a. In English, “Something is red” is true iff there is a particular thing x such that we
could ‘point to it’ and truthfully say ‘THAT is red’.
b. In English, “Everything is red” is true iff for any thing in the world x, you could
‘point to it’, and truthfully say ‘THAT is red’.
(13) Building From That Naïve Intuition
a. ∃xϕ is true iff there is a thing α such that ϕ is true when x ‘picks out’ α
b. ∀xϕ is true iff for any thing α, ϕ is true when x ‘picks out’ α
But what do we mean by ‘when x picks out α’?!?

Well, this will be familiar to the semanticists…
(14) Variable Assignment

Let M be a model <D, I>. Then g is a variable assignment (based on M) if:
(i) g is a function whose domain is the set of all the variables in FOL,
(ii) and whose range is D
Note: g needn’t be an injection; it could map two different variables to the same α ∈ D!
(15) Notation
Let M be a model <D, I>, and let g be a variable assignment based on M, and let α ∈ D,
and let v be a variable of FOL.
g(v/α) is the variable assignment exactly like g except (at most) that it maps v to α
4
Fall 2013 Ling 720
Note: The following follow from the definition in (15)
• g(x/a)(y/b) is just like g except (at most) that it maps x to a and y to b.
• g(x/a)(x/b) = g(x/b)
(16) The Term ‘Term’

t is a ‘term (of FOL)’ iff t is either a variable or an individual constant
(17) Some More Notation

Let t be a term, M be a model <D, I> and g be a variable assignment based on M. Then
the interpretation of t relative to M and g [[ t ]]M,g is defined as follows:
a. If t is an individual constant, then [[ t ]]M,g = I(t)

b. If t is a variable, then [[ t ]]M,g = g(t)
With these notions in place, we can revise our notion of ‘valuation’ in (9) so that it now can map
every formula (without quantifiers) to truth values, even those with free variables!
(18) Valuation of FOL, Relative to a Model and a Variable Assignment (To Be Revised)
Let M be a model <D, I> and g be a variable assignment (based on M) Then the
‘valuation based on M and g’ (VM,g) is a function whose domain is the set of FOL
formulae (without quantifiers), whose range is {0,1}, and which satisfies the conditions
below:
(i) If ϕ = Φt1... tn, then VM,g(ϕ) = 1 iff < [[t1]]M,g , ..., [[tn]]M,g)> ∈ I(Φ)
(ii) If ϕ = ~ψ, then VM,g(ϕ) = 1 iff VM,g(ψ) = 0

(iii) If ϕ = (ψ & χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 1 and VM,g(χ) = 1
(iv) If ϕ = (ψ ∨ χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 1 or VM,g(χ) = 1
(v) If ϕ = (ψ  χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 0 or VM,g(χ) = 1
(19) Illustration
Let M be the model (partially) defined in (8) and let g be the following variable
assignment: {<x, Rajesh>, <y, Seth>, <z, Angelika>. From (18), it follows that
• VM,g(~Pxc) =1 iff (by 18ii)

• VM,g(Pxc) =0 iff (by 18i)
• < [[x]]M,g, [[c]]M,g > ∉ I(P) iff (by definition in (17))
• < g(x), I(c)> ∉ I(P) iff (by definition of g and I)
• < Rajesh, Rajesh> ∉ { <Angelika, Rajesh>, <Angelika, Seth>, <Rajesh, Seth>}
Thus, we can calculate that VM,g(~Pxc) = 1
5
Fall 2013 Ling 720
Our notion of ‘interpretation (valuation) with respect to a variable assignment’ explicates what
we mean by ‘when a variable picks out α’
• ‘Φ is true when x ‘picks out’ α’ ≈ VM,g(Φ) = 1 if g(x) = α
With this in mind we can extend our definition in (18) to include quantifiers!
(20) Extending (18) to Include Quantifiers
a. What We Ultimately Want: Two equations whose left hand side looks like:
(vi) If ϕ = ∃vψ, then VM,g(ϕ) = 1 iff …

(vii) If ϕ = ∀vψ, then VM,g(ϕ) = 1 iff …
b. Recall Our Naïve Intuition:
(i) ∃xψ is true iff there is a thing α such that ψ is true when x ‘picks out’ α
(ii) ∀xψ is true iff for any thing α, ψ is true when x ‘picks out’ α
c. Spelling Out the Naïve Intuition With Variable Assignments
(vi) If ϕ = ∃vψ, then VM,g(ϕ) = 1 iff there is an a ∈ D such that VM,g(v/a)(ψ) = 1
(vii) If ϕ = ∀vψ, then VM,g(ϕ) = 1 iff for every a ∈ D, VM,g(v/a)(ψ) = 1
 In (vi) and (vii), we interpret ψ relative to the variable assignment g(v/a);

thus, we interpret ψ with ‘v picking out a’
(21) Valuation of FOL, Relative to a Model and a Variable Assignment (Final Version)
Let M be a model <D,I> and g be a variable assignment (based on M). Then the
formulae, whose range is {0,1}, and which satisfies the conditions below:
(i) If ϕ = Φt1... tn, then VM,g(ϕ) = 1 iff < [[t1]]M,g , ..., [[tn]]M,g)> ∈ I(Φ)
(ii) If ϕ = ~ψ, then VM,g(ϕ) = 1 iff VM,g(ψ) = 0
(iii) If ϕ = (ψ & χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 1 and VM,g(χ) = 1
(iv) If ϕ = (ψ ∨ χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 1 or VM,g(χ) = 1
(v) If ϕ = (ψ  χ), then VM,g(ϕ) = 1 iff VM,g(ψ) = 0 or VM,g(χ) = 1
(vi) If ϕ = ∃vψ, then VM,g(ϕ) = 1 iff there is an a ∈ D such that VM,g(v/a)(ψ) = 1
(vii) If ϕ = ∀vψ, then VM,g(ϕ) = 1 iff for every a ∈ D, VM,g(v/a)(ψ) = 1
6
Fall 2013 Ling 720
(22) Illustration 3
Let M be a model <{Dave, Joe}, I>, where I(L) = {<Dave,Joe>, <Joe,Dave>}. Let g be a
variable assignment such that g(x) = Dave and g(y) = Joe.
a. Interpreting ‘∃yLxy’ with respect to M and g
1. VM,g (∃yLxy) = 1 iff (by (21vi))
2. There is an a ∈ D s.t. VM,g(y/a)(Lxy) = 1 iff (by (21i))
3. There is an a ∈ D s.t. < [[x]]M,g(y/a) , [[y]]M,g(y/a)> ∈ I(L) iff
4. There is an a ∈ D s.t. < g(y/a)(x) , g(y/a)(y)> ∈ I(L)

iff (by def. of g and I)
5. There is an a ∈ D s.t. < Dave , a > ∈ {<Dave,Joe>, <Joe,Dave>}
Thus, we can calculate that VM,g (∃yLxy) = 1
b. Interpreting ‘∀x∃yLxy’ with respect to M and g
1. VM,g (∀x ∃yLxy) = 1 iff (by (21vii))
2. For every a ∈ D, VM,g(x/a)(∃yLxy) = 1 iff (by (21vi))
3. For every a ∈ D, there is an a’ ∈ D s.t. VM,g(x/a)(y/a’)(Lxy) = 1

iff (by (21i))
4. For every a ∈ D, there is an a’ ∈ D s.t.

< [[x]]M,g(x/a)(y/a’) , [[y]]M,g(x/a)(y/a’)> ∈ I(L) iff
5. For every a ∈ D, there is an a’ ∈ D s.t.

< g(x/a)(y/a’)(x), g(x/a)(y/a’)(y) > ∈ I(L) iff (by def. of g and I)
6. For every a ∈ D ({Dave, Joe}), there is an a’ ∈ D ({Dave, Joe}) s.t.

< a, a’ > ∈ {<Dave,Joe>, <Joe,Dave>}
Thus, we can calculate that VM,g (∀x∃yLxy) = 1
• We’ve found a viable way of using models to assign truth-values to formulae of FOL
• So, we’ve also found a way of using models to assign truth-values to FOL sentences
o So, models seem like an excellent characterization of ‘interpretation’ for FOL
3
For another helpful illustration of the key definitions in (21), see Partee et al. (1993), Chapter 13 pp. 326-327.
7
Fall 2013 Ling 720
3. Some Important Consequences and Notations
(23) Key Fact: Sentences of FOL and Variable Assignments
Let M be any model and g, h be variable assignments based on M. If ϕ is a sentence of

FOL, then VM,g(ϕ) = VM,h(ϕ)
• That is, the truth-value of a sentence relative to a model and a variable assignment
doesn’t depend upon the variable assignment at all (since there are no free variables)
o Notice how in (22b), we don’t ever actually calculate the value of g for any
variable.
o Thus, even if we replaced g with some other variable assignment h in (22b),

we’d still get the same result!...
• Consequently, if a sentence ϕ is true with respect to a model M and a variable

assignment g, then it’s true with respect to M and any variable assignment h
So, even though we need to refer to a variable assignment to calculate the truth-value of a
sentence, the sentence’s truth-value doesn’t depend upon the assignment we pick…
For this reason, we can introduce the following terminology:
(24) Key Terminology: A Model of an FOL Sentence (Set of Sentences)
Let M be a model, ϕ be a sentence of FOL, and S be a set of FOL sentences.
a. ϕ is true in M iff for any variable assignment g, VM,g(ϕ) = 1
b. M is a model for ϕ iff ϕ is true in M
c. M is a model for S iff for all ϕ ∈ S, M is a model for ϕ
Also, if we want to – and some do – we can eliminate direct reference to VM,g in our semantics
for FOL, by extending the notation ‘[[ . ]]M,g’, defined in (17)
8
Fall 2013 Ling 720
(25) Interpretation With Respect to a Model
Let M be a model <D, I> and g be a variable assignment based on M. Then the
interpretation (a.k.a. denotation) of an ‘expression’ of FOL relative to M and g [[.]]M,g is
defined as follows:
a. If v is a variable, then [[ v ]]M,g = g(v)
b. If α is an individual constant, then [[ α ]]M,g = I(α)
c. If Φ is a predicate letter, then [[ Φ ]]M,g = I(Φ)
d. If ϕ = Φt1... tn, then [[ϕ]]M,g = 1 iff < [[t1]]M,g , ..., [[tn]]M,g)> ∈ [[ Φ ]]M,g
e. If ϕ = ~ψ, then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0
f. If ϕ = (ψ & χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 1 and [[χ]]M,g = 1
g. If ϕ = (ψ ∨ χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 1 or [[χ]]M,g = 1
h. If ϕ = (ψ  χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0 or [[χ]]M,g = 1
i. If ϕ = ∃vψ, then [[ϕ]]M,g = 1 iff there is an a ∈ D such that [[ψ]]M,g(v/a) = 1
j. If ϕ = ∀vψ, then [[ϕ]]M,g = 1 iff for all a ∈ D, [[ψ]]M,g(v/a) = 1
(26) Notation
Let ϕ be a sentence of FOL and M be a model. The interpretation (a.k.a. denotation) of ϕ
with respect to M, [[ϕ]]M, is [[ϕ]]M,g for an arbitrary variable assignment g.
Consequences:
• ϕ is true in M iff [[ϕ]]M = 1 iff M is a model for ϕ
M
• M is a model for S iff for all ϕ ∈ S, [[ϕ]] = 1
4. Using Models to Define Semantic Concepts for FOL
With our formal characterization of ‘truth’ for FOL, we can give precise definitions to the
concepts of ‘tautology’, ‘contradiction’, ‘logical equivalence’ and ‘entailment’….
These definitions are going to be restricted to sentences of FOL…
9
Fall 2013 Ling 720
(27) Tautology
Let ϕ be a sentence of FOL. ϕ is a tautology iff for every model M [[ϕ]]M = 1
Illustration: ∀x(Px ∨ ~Px)
Proof That It’s a Tautology

Let M be any model <D, I> and g be any variable assignment (based on M).
• Suppose that [[∀x(Px ∨ ~Px)]]M,g = 0

• Then it’s not the case that for all a ∈ D, [[Px ∨ ~Px]]M,g(x/a) = 1
• Then there is some a ∈ D such that [[Px ∨ ~Px]]M,g(x/a) = 0.
• Then there is some a ∈ D such that [[Px]]M,g(x/a) = 0 and [[~Px]]M,g(x/a) = 0
• Then there is some a ∈ D such that [[Px]]M,g(x/a) = 0 and [[Px]]M,g(x/a) = 1
• Then there is some a ∈ D such that g(x/a)(x) ∉ I(P) and g(x/a)(x) ∈ I(P)
• Then there is some a ∈ D such that a ∉ I(P) and a ∈ I(P) CONTRADICTION
Important Note:
It’s common for logicians to refer to FOL sentences true in every model as universally valid.
• The term ‘tautology’ is often restricted to FOL formulae that can be obtained by taking a
tautology of PL and replacing the propositional letters with FOL sentences.
(28) Contradiction
Let ϕ be a sentence of FOL. ϕ is a contradiction iff for every model M [[ϕ]]M = 0
Illustration: ∃x(Px & ~Px)
Proof That It’s a Contradiction

• Suppose that [[∃x(Px & ~Px)]]M,g = 1

• Then there is some a ∈ D such that [[Px & ~Px]]M,g(x/a) = 1.
• Then there is some a ∈ D such that [[Px]]M,g(x/a) = 1 and [[~Px]]M,g(x/a) = 1
• Then there is some a ∈ D such that [[Px]]M,g(x/a) = 1 and [[Px]]M,g(x/a) = 0
• Then there is some a ∈ D such that g(x/a)(x) ∈ I(P) and g(x/a)(x) ∉ I(P)
• Then there is some a ∈ D such that a ∈ I(P) and a ∉ I(P) CONTRADICTION
(29) Contingent
Let ϕ be a sentence of FOL. ϕ is contingent iff there are models M and M’ such that
[[ϕ]]M = 1 and [[ϕ]]M’ = 0.
Illustration: ∃x(Px & Qx) [proof trivial]
10
Fall 2013 Ling 720
(30) Consequences
• ϕ is a tautology (universally valid) iff ~ϕ is a contradiction
• ϕ is contingent iff ~ϕ is contingent
(31) Key Observation
• Note that in showing that a given sentence of FOL is a tautology or contradiction, we

need to use some (modest) mathematical ingenuity.
o Unlike PL, we don’t just ‘mechanically calculate’ it out (via truth-tables)
• This is no accident. A very deep and interesting theorem of computability theory

is that there can be no purely mechanical, algorithmic procedure for calculating
whether an FOL sentence is a tautology or not.
• Similarly, there is no purely mechanical procedure for calculating whether or not one
FOL sentence ‘entails’ another (defined below) or whether two sentences of FOL are
‘logically equivalent’ (defined below).
o Unlike PL, ‘you gotta use your noodle…’
(32) Logical Equivalence

Let ϕ, ψ be a sentences of FOL. ϕ and ψ are logically equivalent if for every model M
[[ϕ]]M = [[ψ]]M
Illustration: ∀x~Px and ~∃xPx
Proof of Logical Equivalence:

• [[∀x~Px]]M,g = 1 iff
• For all a ∈ D, [[~Px]]M,g(x/a) = 1 iff
• For all a ∈ D, [[Px]]M,g(x/a) = 0 iff
• For all a ∈ D, a ∉ I(P) iff
• It is not the case that there is an a ∈ D s.t. a ∈ I(P) iff
• It is not the case that there is an a ∈ D s.t. [[Px]]M,g(x/a) = 1 iff
• It is not the case that [[∃xPx]]M,g = 1 iff
• [[∃xPx]]M,g = 0 iff
• [[~∃xPx]]M,g = 1
(33) Consequences
• Any two tautologies (universal validities) are logically equivalent

• Any two contradictions are logically equivalent.
• If ϕ, ψ are logically equivalent, then so are ~ϕ, ~ψ
11
Fall 2013 Ling 720
(34) Consequence
The following two formulae are logically equivalent: ∃xPx and ~∀x~Px
• ∀x~Px and ~∃xPx are logically equivalent (proof in (32))

• ~∀x~Px and ~~∃xPx are logically equivalent (consequences in (33))
• ~∀x~Px and ∃xPx are logically equivalent (double negation)
(35) Key Result: Substitution of Logically Equivalent Formulae
Informal Statement:
If ϕ and ψ are logically equivalent, then replacing ϕ with ψ will have no effect on the
truth-value of a larger sentence.
Formal Statement:
Suppose that ϕ and ψ are logically equivalent, that ϕ is a subformula of χ, and that χ’ is
the formula just like χ, except that all instances of ϕ are replaced with ψ. It follows that χ
and χ’ are logically equivalent.
(36) Key Result: Dropping Operators (Quantifiers) from Our System
• Informally speaking, what (35) says is that if ϕ and ψ are logically equivalent, then
anything we can ‘express’ in FOL with ϕ, we can also ‘express’ with ψ
• Now, note the logical equivalence we just proved in (34):

‘∃xPx’ is logically equivalent to ~∀x~Px
• From (34), it follows that anything we can ‘express’ in FOL with ‘∃’ we can express
with ‘∀’ and ‘~’.
• Consequently, we could drop ‘∃’ from our FOL system, leaving just ‘∀’ and ‘~’,
and still have an equally ‘expressive’ system
• In addition, we could simply view ‘∃’ as a special abbreviation for more complex
formulae:
∃vψ is ‘shorthand’ for ~∀x~ψ
• (We’ll make use of these ideas later, where they will come in handy for the
definitions of logical languages we use for the semantic analysis of English…)
12
Fall 2013 Ling 720
(37) Entailment
Let ϕ, ψ be a sentences of FOL. ϕ entails ψ if every model M such that [[ϕ]]M = 1 is also
such that [[ψ]]M = 1
Illustration: (∀xPx ∨ ∀xQx) entails ∀x(Px ∨ Qx)
Proof of Entailment:
Let M be any model <D, I> such that [[(∀xPx ∨ ∀xQx)]]M = 1. Let g be any variable
assignment based on M.
• Thus, [[(∀xPx ∨ ∀xQx)]]M,g = 1

• Thus, either (i) [[∀xPx]]M,g = 1 or (ii) [[∀xQx)]]M,g = 1
• Suppose (i). Then for all a ∈ D, [[Px]]M,g(x/a) = 1
o Therefore, for all a ∈ D, either [[Px]]M,g(x/a) = 1 or [[Qx]]M,g(x/a) = 1
o Therefore, for all a ∈ D, either [[(Px ∨ Qx)]]M,g(x/a) = 1
o Therefore, [[∀x(Px ∨ Qx)]]M = 1
• Suppose (ii). Then for all a ∈ D, [[Qx]]M,g(x/a) = 1
o Therefore, for all a ∈ D, either [[Px]]M,g(x/a) = 1 or [[Qx]]M,g(x/a) = 1
o Therefore, for all a ∈ D, either [[(Px ∨ Qx)]]M,g(x/a) = 1
o Therefore, [[∀x(Px ∨ Qx)]]M = 1
• Thus, [[∀x(Px ∨ Qx)]]M,g = 1
Now that we have this notion of ‘interpretation with respect to a model’, it becomes possible to
prove rigorously that our proof system for FOL is sound and complete!
(38) Soundness of PL
• If S ⊢ ψ, then if M is a model for S, then [[ψ]]M = 1
(39) Completeness of PL
• If every model M for S is also a model for ψ, then S ⊢ ψ
13
Fall 2013 Ling 720
Problem Set on Propositional Logic and First Order Logic:
Formal Semantics 1
(1) Computing Entailment in Propositional Logic

Please use truth-tables to establish whether the following claims are accurate.
a. { (~p ∨ ~q) } ⊨ ~(p ∨ q)

b. { (p ∨ q), (~p  ~q) } ⊨ q
c. { ((p  q)  r) } ⊨ (p  (q  r))
(2) Computing Logical Equivalence in Propositional Logic

Please use truth-tables to show that the following pairs are logically equivalent.
a. (i) (ϕ & ψ) (ii) ~(~ϕ ∨ ~ψ)

b. (i) (ϕ  ψ) (ii) (~ψ  ~ϕ)
c. (i) (ϕ & (ψ ∨ χ)) (ii) ((ϕ & ψ) ∨ (ϕ & χ))
(3) Defining Operators in Propositional Logic

Please show how the operators ‘~’ and ‘∨’ can be defined using the ‘nor’-operator (↓)
defined below.
ϕ | ψ || (ϕ ↓ ψ)
1 1 0
1 0 0
0 1 0
0 0 1
(4) Computing Truth of Formulas Relative to a Model

Let M be the model < L, I>, where L is the set of English letters { a, b, c, d, e, f, g, h, I, j,
k, l, m, n, o, p, q, r, s, t, u, v, w, x, y, z } and I consists of at least the following mappings:
I(A) = { x : x ∈ L and x is a vowel }

I(R) = { <x,y> : x, y ∈ L and x precedes y in alphabetical order }
Please calculate the truth-values of the following sentences relative to M:
a. ∃x∃y∃z(Rxy & (Ay & (Rxz & ~Az)))

b. ∀x(Rxx  ~Ax)
c. ∀x(~Ax  ∃yRxy)
1
Most of these exercises are taken from Gamut (1991), Volume 1: Chapter 2, Chapter 3
1
Fall 2013 Ling 720
(5) Proving Logical Equivalence for First Order Logic Sentences
Please show that the following pairs of formulae are logically equivalent.
a. (i) ∀x∀yPxy (ii) ∀y∀xPxy

b. (i) ∃x∃yPxy (ii) ∃y∃xPxy
(6) Proving Entailment for First Order Logic Sentences

Please show that the following entailment relations hold.
a. ∀xPx ⊨ ∃xPx 2
b. ∃x(Px & Qx) ⊨ (∃xPx & ∃xQx)
c. ∃y∀xPxy ⊨ ∀x∃yPxy
2
Hint: Recall that the domain D of a model has to be a non-empty set.
2
Fall 2013 Ling 720
Proving the Soundness and Completeness of Propositional Logic:
Some Highlights 1
(1) A Summary of What We’ve Done So Far for PL
a. We’ve given a purely syntactic characterization of ‘valid inference’ in PL

S⊢ψ
b. We’ve given a formal semantics for PL notation, and used it to provide a (proper)
semantic definition of ‘valid inference’ in PL.
S⊨ψ
(2) The BIG Question: Do these two characterizations of validity coincide?

S⊢ψ iff(?) S⊨ψ
(3) The Theorems We Wish to Prove
Soundness of PL:
If ψ can be derived from S in our natural deduction system for PL, then S entails ψ
• If S ⊢ ψ, then if V is a valuation for S, V(ψ) = T
Completeness of PL:
If S entails ψ, then ψ can be derived from S in our natural deduction system.
• If every valuation V of S is also a valuation of ψ, then S ⊢ ψ
(4) Some History of the Proofs

• Soundness was basically proven rather early on (it’s easy, but tedious)
• Before there were proper proofs, people were largely convinced that PL and FOL
were ‘complete’
o After all, anything anyone ever wanted to prove could be proved!
• The first proper proof that FOL is complete was Gödel’s PhD thesis (1929)
o It’s crazy complicated, and nobody teaches it anymore
• In his PhD thesis, Henkin (1949) hit upon a much simpler, and just plain cooler proof
o This is the one everybody teaches to this day…
1
Crossley et al. (1972), Chapter 2; Partee et al. (1993) Chapter 8 pp. 225-227.
1
Fall 2013 Ling 720
For reasons of time, I won’t give the complete proof of either soundness or completeness:
• However, I will hit the ‘highlights’ of both…
• As we’ll see, the completeness proof is more involved, and more interesting!...
1. Proving the Soundness of Natural Deduction for Propositional Logic
(5) Theorem to Prove: Soundness If S ⊢ ψ, then S ⊨ ψ
(6) Key Observation

If S ⊢ ψ, then there is a finite subset S’ ⊆ S such that there is a derivation consisting of n
lines where each ϕ ∈ S’ appear as ‘Assumptions’ and where ψ appears on line n.
• Key Idea:
We can use (strong) induction to prove the following, which would prove (5):
(7) Restatement of Soundness (for Mathematical Induction)

For every natural number n > 0, if S ⊢ ψ with a proof consisting of n lines, then S ⊨ ψ
I won’t give the entire inductive proof of (7), but I’ll give you the main gist…
(8) Preliminary Observation
The following are, technically speaking, proofs in our natural deduction system.
a. 1. p Assumption
b. 1. p Assumption
2. (q & r) Assumption
3. (s  t) Assumption
• Proof (1a) consists of one single line. It terminates right after we add ‘p’ as an
assumption. Thus, {p} ⊢p (which intuitively should be the case)
• Proof (1b) has three lines. It terminates right after we add the third assumption
‘(st)’. Thus, {p, (q&r), (st)} ⊢(st) (which intuitively should be the case)
2
Fall 2013 Ling 720
(9) Proof of Soundness Theorem (7) By Strong Induction
a. Base Step: n=1

Suppose that S ⊢ ψ with a proof consisting of 1 line.
• Thus, for some finite subset S’ ⊆ S, S’ ⊢ ψ with a proof consisting of 1 line.
• Given our system, the proof in question must be a ‘degenerate’ case like (8a),
where ψ is an Assumption.
• Consequently, ψ ∈ S’, and so ψ ∈ S. Consequently, S ⊨ ψ
b. Induction Step:
Let n ∈ ℕ be such that for all m < n, if S ⊢ ψ with a proof consisting of m lines,
then S ⊨ ψ.
• We’ll now show that if S ⊢ ψ with a proof consisting of n lines, then S ⊨ ψ.
• We’ll show this by considering all the ways that a proof consisting of (n-1)
lines can be extended to a proof consisting of n lines.
• Given the structure of our system, there are 12 cases to consider:
1. Adding an assumption ψ
2. Deriving ψ by Repetition
3. Deriving ψ by I&
4. Deriving ψ by E&
5. Deriving ψ by I∨
6. Deriving ψ by E∨
7. Deriving ψ by E
8. Deriving ψ by I
9. Deriving ψ by E~
10. Deriving ψ by I~
11. Deriving ψ by ~~
12. Deriving ψ by EFSQ
For reasons of time, I won’t do all 12…

Just a few notable ones…
1. Adding ψ as an Assumption
Suppose that S ⊢ ψ with a proof consisting of n lines, where the final line has
‘Assumption’ as the justification. It follows that ψ ∈ S, and so S ⊨ ψ
3
Fall 2013 Ling 720
2. Deriving ψ by Repetition
‘Repetition’ as the justification.
• By definition of ‘Repetition’, S ⊢ ψ with a proof of length m < n.

• Therefore, by the Induction Assumption, S ⊨ ψ.
3. Deriving ψ by I&
‘I&’ as the justification.
• By definition of ‘I&’, ψ = (ϕ & χ), and S ⊢ ϕ with a proof of length m < n,

and S ⊢ χ with a proof of length m < n
• Therefore, by the induction assumption, S ⊨ ϕ and S ⊨ χ
• Therefore, S ⊨ (ϕ & χ) (= ψ)
4. Deriving ψ by E&
(Can be shown via an argument parallel to the one for I&)
5. Deriving ψ by I∨
(easily shown via an argument similar to those above)
6. Deriving ψ by E∨
(easily shown via an argument similar to those above)
7. Deriving ψ by E
‘E’ as the justification.
• By definition of ‘E’, it follows that S ⊢ (ϕ  ψ) with a proof of length m <

n, and S ⊢ ϕ with a proof of length m’ < n.
• Therefore, by the induction assumption, S ⊨ (ϕ  ψ) and S ⊨ ϕ.
• Therefore S ⊨ ψ.2
The other steps in the proof are basically parallel to this…
EXCEPT THAT: the steps for I and I~ rely upon a minor (trivial) lemma concerning
‘conditional proofs’…
2
After all, if there were a valuation V of S s.t. V(ψ) = 0, then since S ⊨ ϕ, this valuation would be s.t.V(ϕ  ψ) =
0, and so S wouldn’t entail (ϕ  ψ).
4
Fall 2013 Ling 720
(10) Important Note
As trivial as it is, this proof of the soundness of PL would not even get off the ground
without a clear, mathematically precise definition of what ‘entailment’ for PL is…
• And this requires a clear, mathematically precise definition of what an ‘interpretation’
of PL is…
2. Proving the Completeness of Natural Deduction for Propositional Logic
(11) Theorem to Prove: Completeness If S ⊨ ψ, then S ⊢ ψ.
The first crucial step to proving completeness is the ‘Key Lemma’ in (13).
• For reasons of time, I won’t review the demonstration here.
• Interested readers are referred to Gamut (1991), p. 150
(12) Key Preliminary Definition: Consistency

Let S be a set of formulae in PL. S is inconsistent if S ⊢⊥. S is consistent if S ⊬⊥
• Note that ‘(in)consistency’ here is a syntactic property.
(13) Key Lemma

Let S be a set of formulae in PL. S ∪ {ψ} is inconsistent iff S ⊢ ~ψ
• Note that (13) just states that S ∪ {ψ} ⊢⊥ iff S ⊢ ~ψ

• If you consider our rules of I~ and E~, you can see that (13) pretty trivially holds…
The second crucial step to proving completeness is seeing how the ‘Consistency Theorem’ in
(14) would entail Completeness in (11)
(14) The Consistency Theorem

If S is a consistent set of formulae in PL, then there is a valuation V of S.
(15) The Consistency Theorem Entails Completeness
• Suppose that S ⊨ ψ. It follows that S ∪ {~ψ} has no valuation.
• Therefore, by contraposition of (14), it follows that S ∪ {~ψ} is inconsistent.
• Therefore, by (13), it follows that S ⊢ ~~ψ, and so S ⊢ ψ.
5
Fall 2013 Ling 720
The third and most arduous step in the completeness is proof is proving (14).
And, the most arduous part of proving (14) is proving the lemma in (16)…
(16) Lindenbaum’s Lemma

Let S be a consistent set of formulae in PL. There is a consistent set S* such that S ⊆ S*
and S* has the following key ‘closure properties’.3
For any formulae ϕ and ψ of PL:

a. ϕ ∈ S* iff ~ϕ ∉ S*
b. (ϕ & ψ) ∈ S* iff ϕ ∈ S* and ψ ∈ S*
c. (ϕ ∨ ψ) ∈ S* iff ϕ ∈ S* or ψ ∈ S*
d. (ϕ  ψ) ∈ S* iff ϕ ∉ S* or ψ ∈ S*
Note:
For those who are interested, proving (16) isn’t intellectually all that difficult. It just takes time to
correctly lay out the procedure for constructing S* from S….
Now that we have this huge set S*, we’re home free!
(17) The Cool Central Insight of Henkin’s Proof

You can take a set S* with the properties in (16), and directly build a valuation for S*
from the formulas in S* itself!
(18) Model Existence Lemma

If a consistent set S* has the ‘closure properties’ in (16a-d), then S* has a valuation.
Namely, it has the valuation defined as follows:
a. The Valuation for S*:

Let the valuation V be such that for every proposition letter α of PL, V(α) = 1 iff
α ∈ S*
b. Claim:
The valuation V defined in (18a) is a valuation for S*.
The final step in the completeness theorem is proving the claim in (18).
• The proof will be by induction on the complexity of formulae…
Again, I won’t do the whole proof here, but I’ll review some key illustrative steps…
3
This set S* is commonly referred to as a ‘maximally consistent set’. Note, too, that since S* is consistent, ⊥ ∉ S*.
6
Fall 2013 Ling 720
(19) Proof of the Model Existence Lemma
Claim:
Let V be the valuation defined in (18a). For any formula ϕ of PL, ϕ ∈ S* iff V(ϕ) = 1.
Proof (by Induction on Complexity of Formulae):
a. Base Step: Proposition Letters

Suppose that ϕ is a proposition letter. Then, by the very definition in (18a), it
trivially follows that V(ϕ) = 1 iff ϕ ∈ S*
b. Induction Step:
Suppose that ϕ is a (complex) formula of PL, and that for any of its immediate
subformulae ψ ∈ S* iff V(ψ) = 1. We will now that that ϕ ∈ S* iff V(ϕ) = 1.
There are four cases to consider: ~, &, ∨, 
1. ϕ = ~ψ
 ~ψ ∈ S* iff (by closure property in (16a))
 ψ ∉ S* iff (by induction assumption)
 V(ψ) = 0 iff (by definition of a valuation)
 V(~ψ) = 1
2. ϕ = (ψ & χ)
 (ψ & χ) ∈ S* iff (by closure property in (16b))
 ψ ∈ S* and χ ∈ S* iff (by induction assumption)
 V(ψ) = 1 and V(χ) = 1 iff (by definition of a valuation)
 V(ψ & χ) = 1
3. ϕ = (ψ ∨ χ) Proof is parallel to those for 1. and 2.
4. ϕ = (ψ  χ) Proof is parallel to those for 1. and 2.
(20) Putting It All Together
a. Given the lemma in (16), we’ve shown that any consistent set S can be
‘expanded’ into a larger consistent set S* with the properties in (16a-d).
b. Given (18)-(19), we’ve shown that any such set S* with the properties in (16a-d)
has a valuation V.
c. Since V is a valuation for S*, and S ⊆ S*, it follows that V is a valuation for S.
d. Thus, any consistent set of formulae has a valuation V. QED (14).
7
Fall 2013 Ling 720
(21) Taking Stock of What We’ve Done
• In our last set of notes, we developed a mathematically rigorous characterization of

what it means for a formula of PL to be ‘true under an interpertation’ (valuation).
• We’ve just seen how these notions have allowed us to prove that our syntactic proof
system for PL is a perfect syntactic characterization of validity in PL
• For the first time in human history, we’ve shown that we can indeed give a
perfect, purely syntactic characterization of what it means for an inference to be
valid (in a specified language)
o That’s a huge achievement…

o And it’s even more an achievement when we do it for FOL…
8
Fall 2013 Ling 720
Key Applications of Model Theoretic Semantics for FOL:
Soundness, Completeness, Compactness, etc. 1
(1) A Summary of What We’ve Done So Far for PL
a. We’ve given a purely syntactic characterization of ‘valid inference’ in FOL

S⊢ψ
b. We’ve given a formal (model-theoretic) semantics for FOL notation, and used it
to provide a (proper) semantic definition of ‘valid inference’ in FOL.
S⊨ψ
In these notes, I’ll review some key applications of our model-theoretic semantics for FOL.
These results will provide further motivation for using ‘models’ as a mathematical
characterization of ‘interpretation’ for FOL…
1. Proving Soundness and Completeness for First Order Logic
(2) The Theorems We Wish to Prove
Soundness of FOL:
If ψ can be derived from S in our natural deduction system for FOL, then S entails ψ
• If S ⊢ ψ, then if M is a model for S, [[ψ]]M = T
Completeness of FOL:
If S entails ψ, then ψ can be derived from S in our natural deduction system.
• If every model M for S is also a model for ψ, then S ⊢ ψ
(3) On the Proof of Soundness for FOL

The proof of soundness for FOL is not importantly different from the proof of soundness
for PL.
• We simply extend the soundness proof of PL so that the induction step also
considers the cases where:
(i) ψ is derived by E∃ (ii) ψ is derived by I∃

(iii) ψ is derived by E∀ (iv) ψ is derived by I∀
1
Crossley et al. (1972), Chapter 2; Partee et al. (1993) Chapter 8 pp. 198-201.
1
Fall 2013 Ling 720
(4) Illustration of the Additional Steps for Proving Soundness of FOL
The following subcase can be added to the induction step in (9b) of the handout “Proving
the Soundness and Completeness of Propositional Logic”
13. Deriving ψ by I∃
‘I∃’ as the justification.
• By definition of ‘I∃’, ψ = ∃xϕ and S ⊢ [α/x]ϕ with a proof of length m < n.

• By the induction assumption, then S⊨[α/x]ϕ.
• Now, let M be any model <D,I> for S, and g be any variable assignment
based on M.
• It follows that M is a model for [α/x]ϕ.
• Therefore [[ [α/x]ϕ ]]M,g = 1
• Let I(α) = a. It follows that [[ ϕ ]]M,g(x/a) = 1.
• Therefore, there is an a ∈ D such that [[ ϕ ]]M,g(x/a) = 1
• Therefore, [[ ∃xϕ ]]M,g = 1.
• Since g was arbitrary, it follows that [[ ∃xϕ ]]M = 1, and so M is also a model
for ∃xϕ.
• Since M was arbitrary, it follow that any model of S is also a model of ∃xϕ,
and so S ⊨ ∃xϕ,
For reasons of time, I won’t walk through the other four additional cases…
Nevertheless, we can still see that this proof wouldn’t even get off the ground without our
formal, model-theoretic semantics for FOL…
(4) On the Proof of Completeness for FOL

The proof of completeness for FOL has the same general structure as the proof of
completeness for PL.
a. Key Lemma:
Let S be a set of sentences in FOL. S ∪ {ψ} is inconsistent iff S ⊢ ~ψ
b. Consistency Theorem:
If S is a consistent set of sentences in FOL, then there is a model M of S.
c. Consistency Theorem Entails Completeness:

• If S ⊨ ψ, then S ∪ {~ψ} has no model.
• Therefore, consistency theorem (4b) entails that S ∪ {~ψ} is inconsistent.
• Therefore, key lemma (4a) entails that S ⊢ ~~ψ, and so S ⊢ ψ
2
Fall 2013 Ling 720
Again, the most arduous step of the completeness proof is showing proving the Consistency
Theorem (4b)…
And the most difficult part of proving (4b) is proving the lemma in (5)…
Also, proving this lemma is much more involved for FOL than for PL.
(5) Lindenbaum’s Lemma

Let S be a consistent set of sentences in FOL. There is a consistent set S* such that S ⊆
S* and S* has the following key ‘closure properties’.2
For any formulae ϕ and ψ of PL:

a. ϕ ∈ S* iff ~ϕ ∉ S*
b. (ϕ & ψ) ∈ S* iff ϕ ∈ S* and ψ ∈ S*
c. (ϕ ∨ ψ) ∈ S* iff ϕ ∈ S* or ψ ∈ S*
d. (ϕ  ψ) ∈ S* iff ϕ ∉ S* or ψ ∈ S*
e. ∃xϕ ∈ S* iff there is an individual constant b such that [b/x]ϕ ∈ S*
f. ∀xϕ ∈ S* iff for every individual constant b, [b/x]ϕ ∈ S*
Again, now that we have this huge set S*, we’re home free!...
(6) The Cool Central Insight of Henkin’s Proof

You can take a set S* with the properties in (5), and directly build a model for S* from
the formulas in S* itself!
(7) The Model Existence Lemma

If a consistent set S* of FOL sentences has the closure properties in (5), then there is a
model M for S*.
a. The Model for S*

Let M be the model <D, I> where:
(i) D is the set of all the individual constants in the language.
(ii) The function I satisfies the conditions below.
1. If α is an individual constant, then I(α) = α
2. If Φ is a predicate letter of arity n, then I(Φ) is such that:

<α1, …, αn> ∈ I(Φ) iff Φα1 … αn ∈ S*
b. Key Claim: The model M defined in (7a) is a model for S*
2
Those who are intimately familiar with the completeness proof for FOL will know that I’m ‘fudging’ here on the
statement of Lindenbaum’s Lemma for FOL (since I’m not mentioning extending our FOL language by adding
infinitely many constants).
3
Fall 2013 Ling 720
To prove the key claim in (7), and thus the Model Existence Lemma, we again do (strong)
induction on the number of logical operators in an FOL sentence…
• The proof is basically the same as that for PL; the only new and interesting steps come
with the atomic formulae and the quantificational formulae
(8) Proof of the Model Existence Lemma
Claim:
Let M be the model defined in (7a). For any natural number n, if ϕ is a sentence of FOL
with n logical operators, then ϕ ∈ S* iff [[ϕ]]M = 1.
Proof (by Induction on Complexity of Formulae):
a. Base Step: Atomic Sentences (n = 0)

Suppose that ϕ is an atomic formula Φα1 … αn.
 Φα1 … αn ∈ S* iff (by condition 2 in definition of M)
 <α1, …, αn> ∈ I(Φ) iff (by condition 1 in definition of M)
 <I(α1), …, I(αn)> ∈ I(Φ) iff (by definition of a ‘model’)
 [[Φα1 … αn]]M = 1
b. Induction Step:
Let n be such that for all m < n, if ϕ is a sentence of FOL with m logical
operators, then ϕ ∈ S* iff [[ϕ]]M = 1. We will now show that if ϕ is a sentence of
FOL with n logical operators, then ϕ ∈ S* iff [[ϕ]]M = 1.
There are six cases to consider: ~, &, ∨, , ∃, ∀

1. ϕ = ~ψ Same as in completeness proof for PL
2. ϕ = (ψ & χ) Same as in completeness proof for PL
3. ϕ = (ψ ∨ χ) Same as in completeness proof for PL
4. ϕ = (ψ  χ) Same as in completeness proof for PL
5. ϕ = ∃xψ
 ∃xψ ∈ S* iff (by closure property (5e))
 There is an individual constant b such that [b/x]ψ ∈ S*

iff (by the induction assumption)
M
 [[ [b/x]ψ ]] = 1 iff
 [[ ψ ]]M,g(x/b) = 1, for an arbitrary variable assignment g iff
 [[∃xψ ]]M = 1
6. ϕ = ∀xψ Proof similar to the one in 5.
4
Fall 2013 Ling 720
(9) Putting it All Together
a. Given the lemma in (5), any consistent set S of FOL sentences can be ‘expanded’
into a larger consistent set S* with the properties in (5a-f).
b. Given (7)-(8), any such set S* with the properties in (5a-f) has a model M .
c. Since M is a model for S*, and S ⊆ S*, it follows that M is a model for S.
Thus, any consistent set of FOL sentences has a model M. QED (4b).
(10) Really Cool Thing to Notice
In the Henkin proof of FOL’s completeness, we construct a model structure directly from
the set of FOL sentences S*.
o The domain is the set of individual constants

o Each constant is interpreted as itself
o Each predicate letter Φ is interpreted as the relation holding of α1, …, αn iff the
sentence ‘Φα1 … αn’ is in S*
Thus, the very nature of models themselves – objects of the form <D, I> - factor into the
central core step of the proof.
o This could be viewed as giving ‘additional motivation’ for this particular

formalization of the notion of ‘FOL interpretation.’
(11) Taking Stock of What We’ve Done
• In our last set of notes, we developed a mathematically rigorous characterization of

what it means for a sentence of FOL to be ‘true under an interpretation’ (model).
• We’ve just seen how this notion has allowed us to prove that our syntactic proof
system for FOL is a perfect syntactic characterization of validity in FOL
• For the first time in human history, we’ve shown that we can indeed give a
perfect, purely syntactic characterization of what it means for an inference to be
valid (in a specified language)
5
Fall 2013 Ling 720
2. Some Other Important Results of our Model Theoretic Semantics
(12) Demonstrating Consistency

Suppose you want to know whether some set S of FOL formulae are consistent or not;
that is, you want to know whether S ⊢ ⊥ (inconsistent) or S ⊬ ⊥ (consistent).
a. The Challenge:
If all we have is the natural deduction system, if S ⊢ ⊥, then we can eventually
show that (we’ll have the proof). But if S ⊬ ⊥, there’s no way to conclusively
show this (in finite time) with just the natural deduction system.
b. The Solution:
• Given soundness, if S ⊢ ⊥, then S ⊨ ⊥, and so there is no model M for S.
• Thus, with our model theoretic semantics, we can show that S is
consistent (S ⊬ ⊥) by devising a model for S!
(13) Demonstrating Independence

Suppose you want to know whether some FOL sentence ϕ can be derived from S or not.
That is, you want to know whether S ⊢ ϕ or S ⊬ ϕ.
• In the latter case, we say that ϕ is ‘independent’ of S.
a. The Challenge:
Again, if all we have is the natural deduction system, there’s no way to
conclusively show (in finite time) that S ⊬ ϕ.
b. The Solution:
Given soundness again, if S ⊭ ϕ then S ⊬ ϕ. Therefore, if we can show that that
there is a model M of S such that [[ϕ]]M = 0, we’ve shown that S ⊬ ϕ.
(14) Historical Relevance of (12) and (13)
For centuries, mathematicians struggled to show whether the fifth axiom of Euclid’s
geometry was ‘independent’ of the other axioms (or not).
• People tried to derive contradictions from the negation of the fifth axiom, but couldn’t
succeed.
• Finally, however, mathematicians succeeded in showing that there were (informal)

models of geometric systems where (only) the fifth axiom is negated.
o Such (informal) models show that non-Euclidean geometries are consistent

(12) and that the fifth axiom is indeed independent (13).
6
Fall 2013 Ling 720
(14) Compactness Theorem
a. Claim:
A set of FOL sentences S has a model iff every finite subset of S has a model.
b. Proof:
(i) If S has a model, then of course every finite subset does (duh).
(ii) Suppose S doesn’t have a model.
• Thus, S⊨⊥ and so by completeness, S⊢⊥.
• Since derivations are finite, it follows that there is some finite subset
S’ ⊆ S such that S’⊢⊥.
• Thus, by soundness, S’⊨⊥, and so S’ doesn’t have a model.
• Thus, not every finite subset S’ ⊆ S has a model.
The theorem in (14) is a powerful tool in advanced meta-logic and model-theory, since it allows
for easier proofs that certain infinite sets of sentences are consistent…
(15) The Löwenheim-Skolem Theorems
a. The ‘Downward’ Löwenheim-Skolem Theorem

If there is a model M (= <D,I>) for S, then there is a model M’ (= <D’,I’>) for S
such that D’ is countable (finite or countably infinite).
b. The ‘Upward’ Löwenheim-Skolem Theorem

If there is a model M (= <D,I>) for S whose domain D is countably infinite, then
there is a model M’ (= <D’,I’>) for S such that D’ is uncountable.
The theorems in (15) entail that – although sentences of FOL can ‘say’ that there are infinitely
many things – they cannot say whether that infinity is countable or not…
• Some philosophers (Putnam) have also tried to connect (15) with philosophical problems
relating to the nature of reference and meaning…
(16) Lindstrom’s Theorem (Informally Put)

Any logic that satisfies conditions C (unnamed here), and satisfies compactness (14), and
also satisfies the Downard Löwenheim-Skolem Theorem just is FOL.
Lindstrom’s Theorem provides a powerful tool for showing that a given logical system – no
matter how superficially different from FOL it is – is ultimately just a notational variant of FOL.
7
Fall 2013 Ling 720
3. Models Beyond First Order Logic
As shown above, the introduction of ‘models’ as abstract characterizations of ‘interpretation’ for

FOL has been an extremely important and fertile development in logic and mathematics…
• Throughout the 50s and 60s, logicians developed model-theoretic semantics for logical
systems beyond FOL.
(17) Modal Logic

One of the first huge advances was Saul Kripke’s development of a model theoretic
semantics for modal logic, along with the first completeness proofs of such logics.
a. Syntax: PL/FOL + the unary sentence connectives ♢ ‘it is possible that’

and ☐ ‘it is necessary that’
♢(p & ☐q) ‘it’s possible that p and it’s necessary that q’.
b. Semantics: We add to the model structure a set of possible worlds

(and a relation over those worlds, a.k.a. ‘the modal base’)
(18) Temporal Logic
a. Syntax: PL/FOL + the unary sentence connectives P ‘it was the case that’
and F ‘it will be the case that’
F(p & Pq) ‘it will be the case that p and it was the case that q.
b. Semantics: We add to the model structure a set of times

(and a linear ordering over those times)
(19) Second Order Logic
a. Syntax: FOL + variables over n-ary relations (for every natural number n)
∀P (Pab  Pbc)
‘every relation that holds between a & b also holds between c & b’
b. Semantics: We extend variable assignments so that they can accommodate

these new variables.
When we combine these notations and model-theoretic structures, we seem to be approaching

the expressive capacity of a significant sub-part of human language!
‘It was once possible that Dave would have something in common with Mary’
P♢F∃P (Pd & Pm)
8
Fall 2013 Ling 720
(20) The Exciting Possibility (ca. 1950s and 1960s)

Could we develop a model theoretic semantics for a natural language (or at least a
significant part of one)?
(21) Why Would We Want a Model-Theoretic Semantics for English?
a. A Compositional Semantics for Natural Language

A model theoretic semantics is a compositional semantics.
• The definition of ‘interpretation w.r.t. a model M’ specifies how the semantic

value of a complex expression is computed in terms of (i) the semantic value
of its component expressions, and (ii) how the complex expression is
syntactically constructed.
• Why would we want a compositional semantics for a natural language? See

Linguistics 610….
b. A Theory of How Language Connects with ‘Reality’

Interpreting a language w.r.t. a model gives us a theory of how the language
‘hooks up’ with language-external reality.
• A model is a little mathematical representation of the universe (or

‘multiverse’, if we have possible worlds).
o The domain D are the entities that exist.

o The range of I gives us properties and relations between those entities
• The function I in the model maps linguistic expressions to ‘things’ in the

model of reality:
o Individual constants are mapped to entities
o Predicates are mapped to properties and relations holding of them
• Why would we want a semantic theory that maps language to language/mind-

external reality? (See Intro to Philosophy of Language)
c. Natural Languages are Not Fundamentally Different from Artificial Languages
• For most of the 20th century, it was commonly held that there is a fundamental
difference between natural languages and artificial languages (like FOL).
• This difference in type was held to entail that the mathematical tools for
analyzing artificial languages could not be applied to natural languages.
• But, if we could give a model theoretic semantics for NL, that would
weaken the idea that there is such a fundamental division (Montague)
9
Fall 2013 Ling 720
(22) Another, Partisan Comment
• When we semanticists lay out a model theoretic semantics for a natural language, we
thereby specify what (some of) the valid inferences in that language are.
o Under any model theoretic semantics, some sentences will end up being
predicted to be true whenever some other (set of) sentences are…
• Moreover, in as much as the theory is truly rigorous and formal, these predictions
can be calculated without any understanding of the (object) language itself.
Thus, we semanticists are the true inheritors of the original Artisotelian logical program:
providing a formal/predictive characterization of valid inference for (natural) language.
(23) Immediate Challenges For Developing a Model-Theoretic Semantics for English
Alas, in the world of the late 50s and early 60s, the following were some obvious
roadblocks to developing such a program for natural language semantics…
a. Semantic Ambiguity of (Seemingly) Syntactically Unambiguous Sentences
• Sentences like the following are semantically ambiguous, but they don’t seem
(at first glance) to be syntactically ambiguous.
(i) Some girl loves every man.

(ii) John wants to buy an ugly car.
• A model-theoretic semantics, though, will map every sentence structure to

exactly one interpretation in the model…
b. The Ungodly Complexity of Natural Language Syntax
• A model-theoretic semantics builds upon a recursive characterization of the

sentences of a language (e.g. FOL).
• Thus, to give a model-theoretic semantics for English, we need some kind of

recursive characterization of a ‘well-formed’ English sentence…
• But who in the world understands English grammar well enough to give a
recursive formal syntax for English?
10
Fall 2013 Ling 720
c. The Ungodly Complexity of the Syntax/Semantics Interface in NL
The main reason why people were so pessimistic about a formal semantics for
natural language is that there just seemed to be so many puzzles about the
syntax/semantics interface in natural language.
(i) John wants to find a unicorn and eat it.
• This sentence doesn’t commit us to there being a unicorn…

• So what the heck does ‘it’ refer to?
• Could ‘it’ function as a bound pronoun? How?
(ii) Every boy sings or dances.
• This sentence doesn’t mean the same thing as ‘every boy sings or
every boy dances’.
• Thus, this isn’t a case of two conjoined sentences with ellipsis.
• So, ‘or’ must join together the two VPs directly…
• BUT HOW IS THAT POSSIBLE, IF IT ALSO JOINS TOGETHER
TWO SENTENCES?
(iii) The temperature is 90 degrees and rising.
• The predicate “is 90 degrees” seems to want the extension of “the

temperature”. (~ the extension of “the temperature” at this time is 90).
• The predicate “is rising” seems to want the intension of “the

temperature”.
• But, in this sentence, the NP “the temperature” is argument to both

those predicate simultaneously… HOW?!
(iv) A man who is smoking walked in.
• The truth-conditions of this sentence seem to be:

∃x((Mx & Sx) & Wx)
‘there is an x s.t. x is a man and x is smoking and x walked in.’
• Clearly, ‘Mx’ is contributed by ‘man’ and ‘Wx’ by walked in…

o But how does ‘who is smoking’ contribute ‘(… & Sx)’?
o HOW DOES THAT WORK??
Of course, Montague (1974) showed how the issues in (23a-c) can be solved…
To see how he did this, though, we need to start with ALGEBRA…
11

Unit 3:
Algebras and Semantics

Fall 2013 Ling 720
An Algebraic Perspective on Propositional Logic:
The First Steps Towards Montague Grammar 1
• Many of the key ideas underlying Montague’s general semantic program stem from an
‘algebraic’ perspective on logic and its formal semantics.
• In these notes, we will begin developing such an algebraic perspective.
• However, it will require us to first make some superficial revisions to our syntax and
semantics for PL and FOL.
1. Prolegomena, Part 1: A New Presentation of PL
Recall our formal definition of a ‘WFF’ in Propositional Logic, below:

1. ~ϕ ∈ WFF
2. (ϕ & ψ) ∈ WFF
3. (ϕ ∨ ψ) ∈ WFF
4. (ϕ  ψ) ∈ WFF
(2) A More Precise Restatement of the Conditions in (1b)
a. If ϕ is a string in the vocabulary of PL which is in the set WFF, then the following
is also in WFF: The result of concatenating the symbol ‘~’ and the string ϕ
b. If ϕ and ψ are strings in the vocabulary of PL which are in the set WFF, then the
following is also in WFF: The result of concatenating ‘(’, ϕ, ‘&’, ψ and ‘)’.
c. If ϕ and ψ are strings in the vocabulary of PL which are in the set WFF, then the
following is also in WFF: The result of concatenating ‘(’, ϕ, ‘∨’, ψ and ‘)’.
d. If ϕ and ψ are strings in the vocabulary of PL which are in the set WFF, then the
following is also in WFF: The result of concatenating ‘(’, ϕ, ‘’, ψ and ‘)’.
1
These notes are based upon material in the following required readings: Partee et al. (1993) Chapter 9, Chapter 13
pp. 331-336. In addition, students are highly encouraged to begin reading Halvorsen & Ladusaw (1979) and Dowty
et al. (1981) Chapter 8, and to start gradually working through Montague’s original “Universal Grammar” paper.
1
Fall 2013 Ling 720
Just for fun – but with big implications for later – let’s introduce the ‘syntactic operations’
below:
(3) Syntactic Operations Over the Vocabulary of PL

The following operations (functions) take as input either strings of symbols over the
vocabulary of PL or pairs of such strings:
a. The Operation ‘Not’

Not(ϕ) = The result of concatenating the symbol ‘~’ and the string ϕ
b. The Operation ‘And’

And(ϕ,ψ) = The result of concatenating ‘(’, ϕ, ‘&’, ψ and ‘)’
c. The Operation ‘Or’

Or(ϕ,ψ) = The result of concatenating ‘(’, ϕ, ‘∨’, ψ and ‘)’
d. The Operation ‘If’

If(ϕ,ψ) = The result of concatenating ‘(’, ϕ, ‘’, ψ and ‘)’
With all this in place, we can offer the following equivalent restatement of the syntax of PL:

b. If ϕ ∈ WFF, then Not(ϕ) ∈ WFF
c. If ϕ, ψ ∈ WFF, then And(ϕ,ψ) ∈ WFF
d. If ϕ, ψ ∈ WFF, then Or(ϕ,ψ) ∈ WFF
e. If ϕ, ψ ∈ WFF, then If(ϕ,ψ) ∈ WFF
(5) Vocabulary
Let A be a set, and f be an n-ary function. We say that A is closed under f if
(i) Every <a1, …, an> ∈ An is in the domain of f
(ii) If <a1, …, an> ∈ An, then f(<a1, …, an>) ∈ A
(6) Observation
The set WFF defined in (4) is closed under the operations Not, And, Or, and If.
2
Fall 2013 Ling 720
(7) Abstracting Even Further
If we wanted to – and we will want to later on – we could schematically represent each of
the ‘syntactic rules’ in (4b-e) as the following tuples:
b. < Not, <WFF>, WFF>

‘The result of applying Not to a member of WFF is a WFF’
c. < And, <WFF, WFF>, WFF>

‘The result of applying And to a pair consisting of a WFF and a WFF is a WFF’
d. <Or, <WFF, WFF>, WFF>

‘The result of applying Or to a pair consisting of a WFF and a WFF is a WFF’
e. <If, <WFF, WFF>, WFF>

‘The result of applying If to a pair consisting of a WFF and a WFF is a WFF’
(8) Key Observation
• In this way of describing the syntax of PL, we importantly distinguish between:

(i) Syntactic Operations
(ii) Syntactic Rules
• Syntactic Operations (Neg, And, Or, If) freely apply to any strings of symbols, and
output strings that aren’t necessarily part of the language we want to define.
• Syntactic Rules (7b-e) make use of the syntactic operations to define categories of
strings in the vocabulary, i.e., syntactic categories (e.g. ‘WFF’)
This will be of fundamental importance to the Montague Grammar architecture…
Now, recall our formal definition of a ‘valuation’ for Propositional Logic, below:

A valuation V is a function from the well-formed formulae of PL to the set of truth-
values {1,0} (V: WFF  {0,1}) such that if ϕ, ψ ∈ WFF, then
a. V(~ϕ) = 1 iff V(ϕ) = 0

b. V((ϕ & ψ)) = 1 iff V(ϕ) = 1 and V(ψ) = 1
c. V((ϕ ∨ ψ)) = 1 iff V(ϕ) = 1 or V(ψ) = 1
d. V((ϕ  ψ)) =1 iff V(ϕ) = 0 or V(ψ) = 1
Again, just for fun – but with big implications for later – let’s introduce the following ‘semantic
operations’:
3
Fall 2013 Ling 720
(10) Semantic Operations Over {1,0}
The following operations (functions) take as input elements of either {1,0} or {1,0}2:
a. The Operation ‘Neg’

Neg = { <1,0>, <0,1>}
b. The Operation ‘Conj’

Conj = { <<1,1>,1>, <<1,0>,0>, <<0,1>,0>, <<0,0>,0> }
c. The Operation ‘Disj’

Disj = { <<1,1>,1>, <<1,0>,1>, <<0,1>,1>, <<0,0>,0> }
d. The Operation ‘Imp’

Imp = { <<1,1>,1>, <<1,0>,0>, <<0,1>,1>, <<0,0>,1> }
(11) Observation: {1,0} is closed under Neg, Conj, Disj, Imp
We can use the semantic operations in (10) and the syntactic operations in (3) to provide the
following re-statement of what a valuation is…
A function V: WFF  {0,1} is a valuation if it satisfies the conditions below:
a. If ϕ = Not(ψ), then V(ϕ) = Neg(V(ψ))
b. If ϕ = And(ψ,χ), then V(ϕ) = Conj(V(ψ),V(χ))
c. If ϕ = Or(ψ,χ), then V(ϕ) = Disj(V(ψ),V(χ))
d. If ϕ = If(ψ,χ), then V(ϕ) = Imp(V(ψ),V(χ))
Confirm For Yourself:

The new definition in (12) entails the key conditions of our earlier definition in (9):
a. V(~ϕ) = 1 iff V(ϕ) = 0

b. V((ϕ & ψ)) = 1 iff V(ϕ) = 1 and V(ψ) = 1
c. V((ϕ ∨ ψ)) = 1 iff V(ϕ) = 1 or V(ψ) = 1
d. V((ϕ  ψ)) =1 iff V(ϕ) = 0 or V(ψ) = 1
4
Fall 2013 Ling 720
2. Prolegomena, Part 2: A New Presentation of FOL
We’ll also slightly alter our presentation of FOL in a manner similar to what we just did for PL...
• Recall our earlier definition below:

The set of ‘well-formed formulae’ of FOL, WFF, is the smallest set such that:

constant or a variable, then ϕα1…αn ∈ WFF
1. ~ϕ ∈ WFF
2. (ϕ & ψ) ∈ WFF
3. (ϕ ∨ ψ) ∈ WFF
4. (ϕ  ψ) ∈ WFF
c. If ϕ ∈ WFF and υ is a variable, then

1. ∀υϕ ∈ WFF
2. ∃υϕ ∈ WFF
For our first minor change, we’ll slightly alter the syntax of the atomic formulae.
(14) Slightly Altered Syntax for Atomic Formulae

If ϕ is an n-ary predicate letter and each of α1, …, αn is either an individual constant or a
variable, then (…(ϕα1)…αn) ∈ WFF
Illustration: Earlier Syntax: Faxby

New Syntax: ((((Fa)x)b)y)
Now, let’s expand the syntactic operations in (3) by adding the following…
(15) Syntactic Operations Over the Vocabulary of FOL
a. The Operation ‘Concat’

Concat(ϕ,ψ) = The result of concatenating ‘(’, ϕ, ψ, and ‘)’
b. The Operation ‘All’

All(ϕ,ψ) = The result of concatenating ‘∀’, ϕ, and ψ
c. The Operation ‘Ext’

Ext(ϕ,ψ) = The result of concatenating ‘∃’, ϕ, and ψ
5
Fall 2013 Ling 720
With these syntactic operations, we can restate our definition of WFF in FOL as follows:

constant or a variable, then Concat(…(Concat(Concat(ϕ,α1), α2), …, αn) ∈ WFF
d. If ϕ, ψ ∈ WFF, then Or(ϕ,ψ) ∈ WFF
e. If ϕ, ψ ∈ WFF, then If(ϕ,ψ) ∈ WFF
f. If ϕ ∈ WFF and υ is a variable, then Ext(υ,ϕ) ∈ WFF
g. If ϕ ∈ WFF and υ is a variable, then All(υ,ϕ) ∈ WFF
We could also use our schematic notation in (7) to provide the following even more compact
presentation of the rules in (16b)-(16g).2
(17) Abstract Representation of Rules (16b)-(16g)
b. < Not, <WFF>, WFF>

c. < And, <WFF, WFF>, WFF>

d. <Or, <WFF, WFF>, WFF>

‘The result of applying Or to a pair consisting of a WFF and a WFF is a WFF’
e. <If, <WFF, WFF>, WFF>

‘The result of applying If to a pair consisting of a WFF and a WFF is a WFF’
f. <Ext, <VAR, WFF>, WFF>

‘The result of applying Ext to a pair consisting of a VAR and a WFF is a WFF’
g. <All, <VAR, WFF>, WFF>

‘The result of applying All to a pair consisting of a VAR and a WFF is a WFF’
2
Rule (16a) won’t be schematically representable in this way until we’ve made one last change, to come later.
6
Fall 2013 Ling 720
Given our new notation for atomic formulae and our semantic operations in (10), we’ll make a
slight change to our definitions of ‘model’ and ‘valuation w.r.t. a model and variable assignment’

(ii) If Φ is an n-ary predicate letter, then I(Φ) is the curried characteristic
function of an n-ary relation R ⊆ Dn
Note:
We are now interpreting predicate letters not as subsets of D or relations on D, but as the
curried characteristic functions of such sets and relations.
(19) Valuation of FOL, Relative to a Model and a Variable Assignment
Let M be a model <D,I> and g be a variable assignment (based on M). Then the
formulae, whose range is {0,1}, and which satisfies the conditions below:
(i) If ϕ = Concat(…(Concat(Φ,α1), …, αn), then

VM,g(ϕ) = I(Φ)([[α1]]M,g)…([[αn]]M,g) 3
(ii) If ϕ = Not(ψ), then VM,g(ϕ) = Neg(VM,g(ψ))
(iii) If ϕ = And(ψ,χ), then VM,g(ϕ) = Conj(VM,g(ψ), VM,g(χ))
(iv) If ϕ = Or(ψ,χ), then VM,g(ϕ) = Disj(VM,g(ψ), VM,g(χ))
(v) If ϕ = If(ψ,χ), then VM,g(ϕ) = Imp(VM,g(ψ), VM,g(χ))
(vi) If ϕ = Ext(υ,ϕ), then VM,g(ϕ) = 1 iff there is an a ∈ D such that VM,g(v/a)(ψ) = 1
(vii) If ϕ = All(υ,ϕ), then VM,g(ϕ) = 1 iff for every a ∈ D, VM,g(v/a)(ψ) = 1
3
Note that if I(Φ) is the curried characteristic function of R, then this is equivalent to saying < [[α1]]M,g , ...,
[[αn]]M,g)> ∈ R, as in our original definition of a valuation based on M and g.
7
Fall 2013 Ling 720
(20) Illustration of the New Definition
a. Illustration of the Definition of a Model:

The pair FOL < { Angelika, Seth, Rajesh }, I > is a model of FOL, where I
consists of at least the following mappings: 4
I(a) = Angelika I(b) = Seth I(c) = Rajesh
I(F) = { <Angelika, 0>, <Seth, 0>, <Rajesh,0> }
I(P) = Angelika  Angelika  0

Rajesh  1
Seth  1
Rajesh  Angelika  0
Rajesh  0
Seth  1
Seth  Angelika  0
Rajesh  0
Seth  0
b. Illustration of the Definition of a Valuation:

Let M be the model (partially) defined in (20a) and g be any variable assignment
based on M
• VM,g( ~((Pb)c) ) = (by definition of Not and Concat)
• VM,g( Not( Concat( Concat(P,b), c) ) ) = (by (19ii))
• Neg( VM,g(Concat( Concat(P,b), c) ) ) = (by (19i))
• Neg( I(P)([[b]]M,g)([[c]]M,g) ) = (by definition of [[.]]M,g)
• Neg( I(P)(I(b))(I(c)) ) = (by definition of I)
• Neg( I(P)(Seth)(Rajesh) ) = (by definition of I(P))
• Neg( 0 ) = (by definition of Neg)
• 1
4
Note that we are again basically interpreting ‘F’ as “is French” and ‘P’ as “is older than”.
8
Fall 2013 Ling 720
3. Algebras and Morphisms: The Key Concepts
Various advances in algebra in the 18th and 19th century lead mathematicians to develop a highly
general and abstract definition of what a system of ‘algebra’ is.
(21) Definition of an Algebra
An algebra is a tuple < A, f1, … fn > consisting of a set A together with one or more
operations (functions) f1, … , fn, where A is closed under each of f1, … , fn.
• Note: The operations f1, … , fn don’t have to be of the same arity, but they must be of
some finite arity.
(22) Illustrations of Algebras
a. < ℕ, +, × >
o The natural numbers are closed under addition and multiplication
Note: This is not an algebra: < ℕ, +, ×, – >, since ℕ isn’t closed under subtraction
b. < ℤ, +, ×, – >
o The integers are closed under addition, multiplication, and subtraction
Note: This is not an algebra: < ℤ, +, ×, –, ÷ >, since ℤ isn’t closed under division.
c. < ℚ, +, ×, –, ÷ >
o The rationals are closed under addition, multiplication, subtraction, division
d. < WFFPL , Not, And, Or, If >

o The WFF of PL are closed under Not, And, Or, and If (see (6))
e. < {1,0}, Neg, Conj, Disj, Imp >

o The set {1,0} is closed under Neg, Conj, Disj, Imp (see (11))
f. < {1,0}, Conj, Disj >
g. < {{a}, ∅}, ∩, ∪ >

o The set {{a}, ∅} is closed under intersection and union
{a} ∪ {a} = {a} {a} ∩ {a} = {a}
{a} ∪ ∅ = {a} {a} ∩ ∅ = ∅
∅ ∪ {a} = {a} ∅ ∩ {a} = ∅
∅∪ ∅= ∅ ∅∩ ∅= ∅
9
Fall 2013 Ling 720
The algebras < {1,0}, Conj, Disj > and < {{a}, ∅}, ∩, ∪ > are intuitively ‘similar’
< {1,0}, Conj, Disj > < {{a}, ∅}, ∩, ∪ >

Conj(1,1) = 1 ∩({a},{a}) = {a}
Conj(1,0) = 0 ∩({a}, ∅) = ∅
Conj(0,1) = 0 ∩(∅, {a}) = ∅
Conj(0,0) = 0 ∩(∅, ∅) = ∅
Disj(1,1) = 1 ∪ ({a},{a}) = {a}

Disj(1,0) = 1 ∪ ({a},∅) = {a}
Disj(0,1) = 1 ∪ (∅, {a}) = {a}
Disj(0,0) = 0 ∪ (∅, ∅) = ∅
This intuitive notion of ‘similarity’ is captured in the following notion of ‘isomorphism’.
(24) Isomorphic and Isomorphism

Let A <A, f1, …, fn> and B <B, g1, …, gn> be algebras. We say that A and B are
isomorphic if the following hold:
(i) There is a one-to-one correspondence between the operations f1, …, fn and

g1, …, gn
(ii) Each fi is of the same arity as gi
(iii) There is a bijection h: A B with the following key property:
For every operation fi in A, h( fi (a1, … am) ) = gi ( h(a1), … h(an) )
If A and B are isomorphic, then the bijection h: AB is an isomorphism.
Note: Informally, (iii) states that that:
If you apply bijection h to the result of fi applied to a1 through am, you get the same thing
in B as you’d get if
You applied the bijection h to a1 through am to get a sequence b1 through bm
You then applied the corresponding function gi in B to b1 through bm
Thus, if fi (a1, … am) = a, then gi ( h(a1), … h(an) ) = h(a)
o And so we see how the isomorphism h entails that the structure of the algebra
A is ‘mirrored’ in the algebra B
10
Fall 2013 Ling 720
(25) Illustration
The algebras < {1,0}, Conj, Disj > and < {{a}, ∅}, ∩, ∪ > are isomorphic.
Demontration: Assume the correspondence Conj ~ ∩ and Disj ~ ∪

Consider the bijection h = { < 1, {a} >, < 0, ∅ > }
Checking For Conj ~ ∩

h(Conj(1,1)) = h(1) = {a} = ∩({a},{a}) = ∩(h(1), h(1))
h(Conj(1,0)) = h(0) = ∅ = ∩({a}, ∅) = ∩(h(1), h(0))
h(Conj(0,1)) = h(0) = ∅ = ∩(∅, {a}) = ∩(h(0), h(1))
h(Conj(0,0)) = h(0) = ∅ = ∩(∅, ∅) = ∩(h(0), h(0))
Checking For Disj ~ ∪

h(Disj(1,1)) = h(1) = {a} = ∪ ({a},{a}) = ∪ (h(1), h(1))
h(Disj(1,0)) = h(1) = {a} = ∪ ({a}, ∅) = ∪ (h(1), h(0))
h(Disj(0,1)) = h(1) = {a} = ∪ (∅, {a}) = ∪ (h(0), h(1))
h(Disj(0,0)) = h(0) = ∅ = ∪ (∅, ∅) = ∪ (h(0), h(0))
Thus, the function h: A B is an isomorphism from A to B.
In addition to the notion of ‘isomorphism’ in (24), there’s also a weaker notion of ‘similarity’
between algebras, that of ‘homomorphism’.
(26) Homomorphic and Homomorphism

Let A <A, f1, …, fn> and B <B, g1, …, gn> be algebras. We say that A and B are
homomorphic if the following hold:
(i) There is a one-to-one correspondence between the operations f1, …, fn and

g1, …, gn
(ii) Each fi is of the same arity as gi
(iii) There is a function h: A B with the following key property:
For every operation fi in A, h( fi (a1, … am) ) = gi ( h(a1), … h(an) )
If A and B are homomorphic, then the bijection h: AB is an homomorphism.
Note:
A homomorphism differs from an isomorphism in that the former need not be an injection or a
surjection; it can map two different things in A to the same thing in B, or there could be things in
B that nothing in A gets mapped to…
11
Fall 2013 Ling 720
(27) Key Illustration
The algebras < WFFPL , Not, And, Or, If > and < {1,0}, Neg, Conj, Disj, Imp > are
homomorphic.
Demonstration:
Assume the correspondence Not ~ Neg, And ~ Conj, Or ~ Disj, and If ~ Imp.
Let V be any valuation, as defined in (12).
o Note, V is a function from WFFPL to {1,0}.

o Thus, we need only check the key correspondence property in (26iii)
Checking For Not ~ Neg: V(Not(ψ)) = Neg(V(ψ)) (by (12a))
Checking For And ~ Conj: V(And(ψ,χ)) = Conj(V(ψ),V(χ)) (by (12b))
Checking For Or ~ Disj: V(Or(ψ,χ)) = Disj(V(ψ),V(χ)) (by (12c))
Checking for If ~ Imp: V(If(ψ,χ)) = Imp(V(ψ),V(χ)) (by (12c))
4. Semantics and Homomorphism
(28) Key Consequence of (27)
A valuation V: WFFPL {1,0} is a homomorphism from < WFFPL , Not, And, Or, If >
(‘syntactic algebra) to < {1,0}, Neg, Conj, Disj, Imp > (‘semantic algebra’)
One of the crucial insights of Montague was to generalize from (28) in the following way…
(29) An ‘Interpretation’ of Propositional Logic (To Be Revised)
Let the language of PL be the algebra L = < WFFPL , Not, And, Or, If >. An interpretation
of PL is a structure < B, f1, f2, f3, f4, h> such that:
a. < B, f1, f2, f3, f4 > is an algebra
b. h is a homomorphism from L to < B, f1, f2, f3, f4 >
(30) Key Consequence:

If V is a valuation, then < {1,0}, Neg, Conj, Disj, Imp, V > is an interpretation of PL.
However, as shown below, the notion of ‘interpretation’ is more general than that of ‘valuation’
12
Fall 2013 Ling 720
(31) An Interpretation That is Not Based on Valuations
Let h: WFFPL  P({a,b,c}) be a function from WFFPL to the powerset of {a,b,c} with the
following key properties: 5
a. h(~ϕ) = h(Not(ϕ)) = h(ϕ)´
b. h((ϕ & ψ)) = h(And(ϕ, ψ)) = ∩( h(ϕ), h(ψ) )
c. h((ϕ ∨ ψ)) = h(Or(ϕ, ψ)) = ∪( h(ϕ), h(ψ) )
d. h((ϕ  ψ)) = h(If(ϕ, ψ)) = IMP( h(ϕ), h(ψ) )

• Where IMP(A,B) = A´ ∪ B
Illustration:
If h(p) = {b,c} and h(q) = {a,b}, then h((p&q)) = {b}, and h((p∨q)) = {a,b,c}
Key Consequence:
The structure < P({a,b,c}), ´, ∩, ∪, IMP, h > is an interpretation of PL.
• < P({a,b,c}), ´, ∩, ∪, IMP > is an algebra
• h is a homomorphism from L to < P({a,b,c}), ´, ∩, ∪, IMP >

o h is a function from WFFPL to P({a,b,c})
o The assumptions in (31a-d) show that it has the homomorphism property (26iii)
We could view an interpretation like < P({a,b,c}), ´, ∩, ∪, IMP, h > as being one in which the
WFFs of PL are mapped to the ‘possible worlds’ in {a,b,c} where they are true…
Finally, we can (try to) generalize the notion of ‘interpretation’ in (29) to any language…
(32) Generalizing ‘Interpretation’ for Any Language (To Be Revised)
Let L = < A, f1, …, fn > be a ‘syntactic algebra’ for a given language. An interpretation of
L is a structure < B, g1, …, gn, j> such that:
a. < B, g1, …, gn > is an algebra.

b. For any i, fi and gi have the same arity.
c. j is a function from the ‘lexical items’ (e.g. ‘non-logical constants’) in A to B.
If B = <B, g1, …, gn, j> is an interpretation of L, then the meaning assignment for L
determined by B is the unique homomorphism h from L to <B, g1, …, gn> such that j ⊆ h.
5
That such a function exists can be shown by giving a concrete example. Students familiar with modal
propositional logic will be familiar with many such concrete examples.
13
Fall 2013 Ling 720
(33) Semantics and Homomorphism: The Intuitive Relation
a. The Principle of Compositionality

The meaning of a complex expression is a ‘function of’ (i) the meanings of its
component expressions and (ii) their mode of syntactic composition.
b. Key Observation:
(33a) basically says that interpretation is homomorphism from syntax to meaning
• To get the meaning of a complex expression ϕ, you determine the syntactic

rule SynR that derives it from its component expressions ψ, χ
• Then, you compute the meanings [[ψ]], [[χ]]
• Finally, you input [[ψ]], [[χ]] into a semantic rule SemR that ‘corresponds’ to
the syntactic rule SynR
o Thus, [[ SynR(ψ,χ) ]] = SemR( [[ψ]], [[χ]] )
Syntax: Semantics
ϕ (= SynR(ψ, χ)) [[ϕ]] (= SemR([[ψ]], [[χ]])
ψ χ [[ψ]] [[χ]]
(34) Semantics and Homomorphism: Another Possible Connection

Suppose that interpretation is homomorphism from syntax to meaning.
• It follows that if a language has a semantics, there is a ‘similarity’ (homomorphism)

between its structure (syntactic algebra) and mind-external reality (semantic algebra)
a. Picture Theory of Meaning (Early Wittgenstein):

A meaningful expression ‘pictures’ a state of affairs / atomic fact.
(There is a structural similarity between the structure of the expression and
the ‘logical facts’ of the world.)
b. Correspondence Theory of Truth (Russell, et multia alia)

For a statement to be true, it must have a structural isomorphism with the state of
affairs that makes it true.
“ ‘A cat is on the mat’ is true iff there is in the world a cat and a mat, and
the cat is related to the mat in virtue of being on it. If any of the three
pieces is missing… then the statement is false.”
14
Fall 2013 Ling 720
5. Moving Beyond Propositional Logic: The Key Challenges
If the definition in (32) is truly viable, then all we need to do to provide a semantics for a
language L is the following:
• Characterize the language L as a ‘syntactic algebra’ < A, f1, …, fn >
• Find a structure < B, g1, …, gn, j> that satisfies the key properties of being an
‘interpretation’ of the language.
Thus, in order for the general semantic program embodied in (32) to be viable, we must be able
to characterize every (semantically interpreted) language as a syntactic algebra…
(35) Key Problem: First Order Logic!
• Given our presentation in (16), the key syntactic operations forming the WFFs of
FOL are: Concat, Not, And, Or, If, Ext, All.
• However, the structure < WFFFOL, Concat, Not, And, Or, If, Ext, All > is not an
algebra!
o WFFFOL is not closed under Concat, Ext, and All
• Concat( Pa , ((Rb)c) ) = (Pa((Rb)c)) ∉ WFFFOL
• Ext( Pa, ((Rb)c) ) = ∃Pa((Rb)c) ∉ WFFFOL
• All( Pa, ((Rb)c) ) = ∀Pa((Rb)c) ∉ WFFFOL
Main Issue:
In PL, applying any of the syntactic operations to a WFF creates a WFF. This just
isn’t the case for FOL
The Solution:
We’ll need to augment what we mean by an ‘algebraic characterization’ of a
language, and thus the definition in (32).
(36) Another Challenge Raised by FOL

Even once we resolve the issue in (35), to pursue the general program outlined in (32),
we’ll need to find some ‘semantic operations’ that ‘correspond’ to Ext and All
• Finding such operations is what logicians mean by ‘algebraization of FOL’.

• Doing this is far from trivial; Montague’s solution is not commonly used and difficult
15
Fall 2013 Ling 720
Problem Set on Syntactic Operations, Semantic Operations, and Homomorphisms
(1) Exercise on the Syntactic Operations for PL

Please show how the following formulae of PL can be constructed via applications of the
syntactic operations Not, And, Or, and If.
For example, if the formula were (p  ~(q & r)), the correct answer would be
If(p, Not(And(q, r))).
a. ~(p & (p  (q ∨ r)))
b. (((p ∨ r)  q) ∨ s)
c. (((p  q)  (~q  ~p)) & ((p  q)  (~p ∨ q)))
(2) Exercise on the Syntactic Operations for FOL

Please show how the following formulae of FOL can be constructed via applications of
the syntactic operations Concat, Not, And, Or, If, Ext, and All.
For example, if the formula were (∃x((Px)a) & (Rb)), the correct answer would be
And(Ext(x, Concat(Concat(P, x), a)), Concat(R, b)).
a. ∀x((Px)  ∃y((Lx)y))
b. ∃z(((Ra)z) ∨ ~((Ra)z))
c. ((Ab) ∨ ~∃x(Ax))
(3) Exercise on (Montagovian) Syntactic Rules

Let Merge be an operation that takes two strings ϕ and ψ and forms the string ‘ϕ ψ’.
Using this operation Merge, please show how each of the PS rules below can be
abstractly characterized as triple <Op, <Cat1, … Catn>, Cat>, where Op is an n-ary
syntactic operation, and each of Cati and Cat is some syntactic category label.
a. DP  D NP
b. PP  P NP
c. NP  A NP
d. S  NP VP
e. How could we capture the following PS rule using this ‘triplet’ notation? Please
do not in any way alter the notation:
VP  V (NP)
1
Fall 2013 Ling 720
(4) Our New Definition of a Model
Let M be a model <D,I> as defined in the handout “First Order Logic: Formal Semantics
and Models”, where D = { a, b, c }, and I consists of at least the mappings below:
I(P) = { x : x ∈ D and x is a vowel }
I(Q) = { <x,y> : x,y ∈ D and x precedes y in the alphabet }
a. Please convert M into a model as defined in (18) on the handout “An Algebraic
Perspective on Propositional Logic.” That is, state what D and I should be under
the new definition in (18).
b. Please use the new definition of ‘valuation’ in (19) on the handout “An Algebraic
Perspective on Propositional Logic” to show how the converted model M assigns
truth-values to the following formulae.
(i) ∃x(Px)
(ii) ∀x∃y((Qy)x)
(5) Exercise on ‘Interpretations’ for PL

Let <{0,1}, Neg, Conj, Disj, Imp, V > be an interpretation for PL, as defined in (29) of
“An Algebraic Perspective on Propositional Logic.” Moreover, assume that V is such that
V(p) = 1, V(q) = 0, and V(r) = 1. Please use the assumption that V is a homomorphism to
calculate truth-values of the following formulae.
For example, if the formula were (p  q), then the calculation would be:
(i) V(p  q) = (by definition of PL)
(ii) V(If(p,q)) = (by homomorphism property of V)
(iii) Imp(V(p),V(q)) = (by definition of V)
(iv) Imp( 1, 0 ) = (by definition of Imp)
(v) 0
a. ~(p & (p  (q ∨ r)))

b. (((p ∨ ~r)  q) ∨ p)
c. (((p  q)  (~q  ~p)) & ((p  q)  (~p ∨ q)))
(6) Exercise on Homomorphisms and Compositions

Let A < A, f1, … fn > and B < B, g1, … gn > and C < C, h1, … hn > all be algebras.
Assume that k: A B is a homomorphism from A to B, and that j: B  C is a
homomorphism from B to C. Please show that j°k is a homomorphism from A to C.
Note: To properly show this, you need to show the following:
(i) That j°k is a function from A to C [trivial]

(ii) That the operations f1, … fn can be put into a one-to-one correspondence with h1,
… hn, such that for all i ∈ ℕ, fi is the same arity as hi [trivial]
(iii) That for all i ∈ ℕ, j°k( fi(a1, …, am) ) = hi( j°k(a1), …, j°k(am) ) [important part]
2
Fall 2013 Ling 720
An Algebraic Perspective on the Syntax of First Order Logic (Without Quantification) 1
1. Statement of the Problem, Outline of the Solution to Come
(1) The Key Problem
• There is much to recommend an ‘algebraic’ treatment of interpretation, where it is

conceived of as a homomorphism from a ‘syntactic’ algebra to a ‘semantic’ one.
• Based on our algebraic semantics for PL, we were lead to the (preliminary)
generalization in (2) below.
• However, its not possible to directly extend this conception of interpretation to FOL:
o Unlike PL, the structure < WFFFOL , Concat, Not, And, Or, If, Ext, All > is not
an algebra, because WFFFOL isn’t closed under those operations…
(2) Algebraic Perspective on ‘Interpretation’ (To Be Revised)
Let L = < A, f1, …, fn > be a ‘syntactic algebra’ for a given language. An interpretation of
L is a structure < B, g1, …, gn, j> such that:
a. < B, g1, …, gn > is an algebra.

b. For any i, fi and gi have the same arity.
c. j is a function from the ‘non-logical constants’ in A to B.
If B = <B, g1, …, gn, j> is an interpretation of L, then the meaning assignment for L
determined by B is the unique homomorphism h from L to <B, g1, …, gn> such that j ⊆ h.
(3) General Outline of a Solution

We’re going to loosen what it means to be a ‘syntactic algebra’ for a given language.
• In a ‘syntactic algebra’ for L, < A, f1, …, fn >, the set A can be a strict superset of the
well-formed sentences of the language.
• For FOL, we’ll construct an algebra < A, Concat, Not, And, Or, If, Ext, All > such
that WFFFOL ⊂ A.
• We’ll then see how we can define/characterize WFFFOL in terms of this algebra (with
a few added formal ingredients)
• In the next notes, we’ll build an algebraic semantics for WFFFOL from these tools
1
These notes are based upon material in the following readings: Halvorsen & Ladusaw (1979), Dowty et al. (1981)
Chapter 8, and Thomason (1974) Chapter 7 (Montague’s “Universal Grammar”).
1
Fall 2013 Ling 720
(4) Zooming In: Obtaining WFFFOL From Our Syntactic Algebra
We’ll construct WFFFOL from < A, Concat, Not, And, Or, If, Ext, All > by using a
‘generate and filter’ approach not too unlike classic GB:
• In the algebra < A, Concat, Not, And, Or, If, Ext, All >, A will contain all the WFF’s
of FOL, but also a whole bunch of ‘syntactic garbage’ (e.g., (P & a), ~x, etc.)
• We’ll then show how syntactic rules function as a kind of filter over A (in the GB
sense), separating out the WFFs from the ‘syntactic garbage’.
• We’ll use these rules (‘syntactic filters’) to formally define that subset of A which is
the well-formed formulae of FOL.
• In the next notes, we’ll make a minimal change to (2) so that it will straightforwardly
extend to the complex system consisting of the algebra and these ‘filters’.
• Funky Property of Our Resulting Semantics:
o We’ll still keep the idea that an interpretation of FOL is a homomorphism

from < A, Concat, Not, And, Or, If, Ext, All > to a ‘semantic algebra’.
o Thus, our ‘algebraic semantics’ will interpret – not only the WFFs of
FOL – but also all the syntactic garbage (!!!)
As always, though, to make our lives easier, we’ll start out by considering only the set of
sentences of FOL that don’t have any quantifiers.
We’ll also assume a minimal set of sentence connectives (&, ~)…
(5) Fragment of FOL We Will Focus on For Now: ‘FOL-NoQ’

The set of well-formed formulae of FOL-NoQ, WFFFOL-NoQ, is the smallest set such that:
a. If ϕ is an n-ary predicate letter and each of α1, …, αn is an individual constant,

then Concat(…(Concat(Concat(ϕ,α1), α2), …, αn) ∈ WFFFOL-NoQ
Examples of WFF of FOL-NoQ: (((Qa)b) & ~(Rc))

~(~(Ab) & ~((Sc)d)) (≈ ((Ab) ∨ ((Sc)d)) )
~(((Tg)b) & ~(Lg)) (≈ (((Tg)b)  (Lg)) )
2
Fall 2013 Ling 720
2. An Algebraic Characterization of WFFFOL-NoQ
In this section, we’re going to show how the set WFFFOL-NoQ can be characterized in terms of a
syntactic algebra. There will be six major steps:
1. Building the Syntactic Algebra
2. Introducing the Syntactic Category Labels (i.e., the Types)
3. Using the Category Labels to Define the ‘Basic Expressions’ of FOL-NoQ (i.e.,
the non-logical constants)
4. Introducing the Syntactic Rules of FOL-NoQ
5. Using the Syntactic Rules and the Basic Expressions (non-logical constants) to
Obtain the Syntactic Categories of FOL-NoQ
6. Putting It All Together: Defining WFFFOL-NoQ in Terms of these Ingredients
Note:
As we’ll see, the resulting system generating WFFFOL-NoQ can be viewed as a kind of abstract
mathematical characterization of ‘what’s going on’ in recursive definitions like (5).
2.1 Building the Syntactic Algebra
(6) The Syntactic Algebra

Let < A, Concat, Not, And > be the algebra such that:
a. Concat, Not, And are the syntactic operations defined previously.
b. A is the smallest set such that

(i) If Φ is a predicate letter, then Φ ∈ A
(ii) If α is an individual constant, then α ∈ A
(iii) A is closed under Concat, Not, And
(7) Remarks
a. WFFFOL-NoQ ⊆ A
• Anything in WFFFOL-NoQ can be created by iterated application of Concat, Not,
And to the predicate letters and individual constants.
b. WFFFOL-NoQ ≠ A
• A includes such ‘syntactic garbage’ as (P & a), ~x, (ab).
c. Not every string in the vocabulary of FOL-NoQ is in A (e.g., P~, &(aB))
3
Fall 2013 Ling 720
(8) a. Question:
Since WFFFOL-NoQ ⊆ A, why not equate FOL-NoQ with the algebra <A, Concat,
Not, And>, like we did for PL?
b. Answer:
There’s not enough ‘information’ in <A, Concat, Not, And> alone to
construct/define the set WFFFOL-NoQ
• But, if we paired this algebra with some other formal devices that could be
used to construct/define WFFFOL-NoQ …
• …Then we could equate the language FOL-NoQ with that system!
This will probably become a bit clearer to you once you actually see how we do this…
2.2 Introducing the Syntactic Category Labels (The Types)
(9) Background: Syntactic Categories and Syntactic Category Labels
a. Syntactic Categories
We can view the ‘syntactic categories’ of a given language sets of strings in the
vocabulary of that language.
Illustration:
• We could say that the category ‘NP’ in English is the set of strings { dog,
student of history, ugly man with a telescope, big funny book about bees, … }
• We could say that the category ‘VP’ in English is the set of strings { runs, ate
a pickle, laughs in the face of danger, always sings horribly, …}
b. Syntactic Category Labels

If syntactic categories are sets of strings, we could view the ‘category labels’ as
being an index set, used to index the family of categories. (See Handout 1).
Illustration:
• Category Labels: {NP, N, VP, V, PP, P, DP, D, S, AP, A, …. }
• Syntactic Categories of English:
o CNP, CN, CVP, CN, CV, CPP, CP, CDP, CD, CS, CAP, CA, …
o { Cδ }δ ∈ {NP, N, VP, V, PP, P, DP, D, S, AP, A, …. }
4
Fall 2013 Ling 720
We’re now going to define a set of syntactic category labels and syntactic categories for FOL…
These will later be used to define a set of syntactic rules that can be used to ‘extract’
WFFFOL-NoQ from the set A….
The category labels will be very familiar to the semanticists…
(10) The Types
The set T of types is the smallest set such that:
a. e∈T
o This is the category label for expressions denoting ‘entities’; i.e., the terms
b. t∈T
o This is the category label for things denoting ‘truth-values’; i.e., the WFFs.
c. If σ ∈ T and τ ∈ T, then <σ, τ> ∈ T

o <σ, τ> will be the category label for things denoting functions from σ to τ
o Thus, <e,t> is the category label for functions from entities to truth-values
(i.e., the unary predicates)…
2.3 Defining the Basic Expressions of FOL-NoQ
(11) Basic Expressions of a Category

If δ is a category label, then Xδ are all the basic (primitive / lexical) expressions of
category δ.
Illustration for English:
• XN = The set of nouns in the English lexicon.

• XNP = { one }2
• XV = The set of verbs in the English lexicon.
• XVP = { so }3
• XP = { in, on, under, from, with … }
• XPP = { there, then, … }
• XD = {a, the, every, most, … }
• XDP = { he, she, it, they, … }
• XS = ∅
We can use these sets Xδ in conjunction with the syntactic rules (to be defined below) to
generate the full set of categories Cδ for the language.
2
This would be ‘anaphoric one’ in English, which some analyze as an NP pronouns (e.g. ‘a student of chemistry
from France, and one from Germany’).
3
This would be the VP-proform ‘so’, that appears in sentences like ‘Mary went to the park, and Frank did so too’.
5
Fall 2013 Ling 720
(12) The Basic Expressions of FOL-NoQ
a. Xe = The set of individual constants, CONS
b. X<e, …, t > = The set of n-ary predicate letters, where n = the number of times e
appears in the category label X<e, …, t >
Illustration: X<et> = The set of unary predicate letters.

X<e<et>> = The set of binary predicate letters.
X<e<e<et>>> = The set of ternary predicate letters, etc.
c. For all other types τ ∈ T, Xτ = ∅.
Note: This states that our ‘lexicon’ for FOL-NoQ contains (i) individual constants, and (ii) n-
ary predicate letters for every n ∈ ℕ, but nothing else. (i.e., the non-logical constants)
2.4 Introducing the Syntactic Rules of FOL-NoQ
Another important use of the syntactic category labels is in stating syntactic rules. Recall the
syntactic rules that we had constructed for PL and FOL in the last handout.
(13) Abstract Representation of Syntactic Rules for FOL (To Be Revised)
a. < Not, <WFF>, WFF>

b. < And, <WFF, WFF>, WFF>

Recall that the WFFs of FOL-NoQ are going to be of category t. We could then rewrite the key
rules in (13a,b) as follows:
(14) Syntactic Rules for FOL-NoQ (To Be Revised)
a. < Not, < t >, t>

‘The result of applying Not to a member of category t is also a member of
category t’.
b. < And, <t, t>, t>

‘The result of applying And to a pair consisting of a member of category t and a
member of category t is itself also a member of category t.
6
Fall 2013 Ling 720
Finally, in addition to the rules in (14a,b), let’s also add every rule of the general form in (15c),
where σ, τ ∈ T.
(15) Syntactic Rules for FOL-NoQ (Final Version)
a. < Not, < t >, t>
b. < And, <t, t>, t>
c. < Concat < <σ, τ>, σ >, τ >

‘The result of concatenating a member of category <σ, τ> with a member of
category σ will be a member of category τ.’
IMPORTANT NOTE:
(15c) is a kind of ‘meta-rule’, a short hand for representing an infinite set of rules. It
encompasses all the following (concrete) rules:
• < Concat < <e, t>, e >, t >

‘The result of applying ‘Concat’ to a member of category <e,t> and a member of
category e will be a member of category t.’
Illustration: If P is of category <et>, and a is of category e, then (Pa) is of category t
• < Concat < <e, <e, t>>, e >, <e,t> >

‘The result of applying ‘Concat’ to a member of category <e, <e,t>> and a member of
category e will be a member of category <e,t>.’
Illustration: If R is of category <e, <e, t>>, and a is of category e, then (Ra) is of

category <et>.
Consequently, if b is of category e, then ((Ra)b) is of category t
2.5 Using the Rules and Basic Expressions to Generate the Syntactic Categories
With our basic expressions in (12) and our syntactic rules in (15), we can define the syntactic
category Cδ for every category label δ.
(16) Definition of the Syntactic Categories Cδ

For any category label τ ∈ T, Cτ is the smallest set such that both the following hold:
a. Xτ ⊆ Cτ
b. If (i) < f , < σ1, …, σn>, τ > is a syntactic rule, and (ii) ϕ1, …, ϕn are such that
each ϕi ∈ Cσi, then f (ϕ1…ϕn) ∈ Cτ
7
Fall 2013 Ling 720
In plain English, the (full) category Cτ will contain all the basic (lexical) expressions of that
category, as well as all the complex expressions of that category that you can create from the
syntactic rules!
• Note, then, that the definition in (16) is simply a formal, mathematically precise
statement of how we’ve been informally reading rules like (15).
(17) Illustration of the Definitions in (16)
a. If P is a unary predicate letter, then P ∈ X<et> (12b), and so P ∈ C<et> (16a).
b. If Q is a binary predicate letter, then Q ∈ X<e,<e,t>>, and so Q ∈ C<e,<e,t>>
c. If a,b are individual constants, then a,b ∈ Xe (12a), and so a,b ∈ Ce (16a).
d. If P is a unary predicate letter, and a is an individual constant, then (Pa) ∈ Ct

 P ∈ C<et> (17a)
 a ∈ Ce (17c)
 < Concat < <e, t>, e >, t > is a syntactic rule (15c)
 Concat(P,a) ∈ Ct (16b)
 (Pa) ∈ Ct
e. If Q is a binary predicate letter, and a,b are individual constants, then ((Qa)b) ∈ Ct
 Q ∈ C<e,<e,t>> (17b)
 a,b ∈ Ce (17c)
 < Concat < <e, <e, t>>, e >, <e,t> > is a rule (15c)
 Concat(Q,a) ∈ C<et> (16b)
 (Qa) ∈ C<et>
 < Concat < <e, t>, e >, t > is a syntactic rule (15c)
 Concat((Qa),b) ∈ Ct (16b)
 ((Qa)b) ∈ Ct
f. If P is a unary predicate letter, and a is an individual constant, then ~(Pa) ∈ Ct

 (Pa) ∈ Ct (17d)
 < Not, < t >, t> is a syntactic rule (15a)
 Not((Pa)) ∈ Ct (16b)
 ~(Pa) ∈ Ct
g. If P is a unary predicate letter, Q is a binary predicate letter, and a,b are individual
constants, then (((Qa)b) & ~(Pa)) ∈ Ct
 ~(Pa) ∈ Ct (17f)
 ((Qa)b) ∈ Ct (17e)
 < And, <t, t>, t> is a syntactic rule (15b)
 And(((Qa)b), ~(Pa)) ∈ Ct (16b)
 (((Qa)b) & ~(Pa)) ∈ Ct
8
Fall 2013 Ling 720
(18) Introducing (Montagovian) Analysis Trees
o A handy way of representing calculations like those in (17a-g) is through the use of
trees of form in (18a).
o These trees represent how a particular syntactic rules works to create an expression of
a particular category.
a. General Form of An Analysis Tree

Suppose the rule < f , < σ1, …, σn>, τ > with (16) entails that f (ϕ1…ϕn) ∈ Cτ
This can be represented via a tree of the following form:
< f (ϕ1…ϕn), f > ‘Output’ of Rule, paired

with operation in Rule
ϕ1 … ϕn ‘Inputs’ to Rule
b. Illustrations
(i) < (Pa), Concat >
P a
(ii) < ((Qa)b), Concat >
< (Qa), Concat > b
Q a
(iii) < (((Qa)b) & ~(Pa)), And >
< ((Qa)b), Concat> < ~(Pa), Not >
< (Qa), Concat > b < (Pa), Concat >
Q a P a
Note: Although these are superficially similar to PS trees, they are importantly different, in that
the non-terminal nodes do not list the category of the resulting expression, but rather the
syntactic operation used to obtain it…
9
Fall 2013 Ling 720
2.6 Putting It All Together: Defining WFFFOL-NoQ in Terms of these Ingredients
(19) The Crucial Observation: Ct = WFFFOL-NoQ
That is, we can take our syntactic algebra < A, Concat, Not, And > and obtain that subset
of A equal to WFFFOL-NoQ by adding:
a. A set of category labels T (10)

b. A set of ‘basic expressions’ Xτ , for each τ ∈ T (12)
c. A set of syntactic rules (15)
Given the definition in (16) for creating Cτ from these ingredients (for every τ ∈ T), we
are able to obtain the set Ct = WFFFOL-NoQ
Thus, if we add the ingredients in (19a,b,c) to our syntactic algebra <A, Concat, Not, And>,
we will have ‘enough information’ to obtain the set WFFFOL-NoQ...
…Thus, given the comment in (8b), we could abstractly characterize FOL-NoQ as in (20)
(20) Algebraic Characterization of FOL-NoQ
The language FOL-NoQ is the structure < A, Concat, Not, And, Xτ, S, t >τ ∈ T , where:
a. < A, Concat, Not, And > is the algebra defined in (6)
b. < … Xτ … >τ ∈ T is shorthand for the infinite family of basic expressions Xτ

defined in (12).
c. S is the (infinite) set of syntactic rules defined in (15).
d. t is the category label τ such that Cτ is the set of WFFs of FOL-NoQ (19).
Again, we’ve seen how the elements of the structure < A, Concat, Not, And, Xτ, S, t >τ ∈ T can
be used to obtain WFFFOL-NoQ
o For this reason, we are now justified in equating FOL-NoQ with this structure.
o Note that < A, Concat, Not, And, Xτ, S, t >τ ∈ T is not actually itself an algebra. Rather, it
contains an algebra as one of its proper parts.
10
Fall 2013 Ling 720
3. Generalizing Our Treatment of FOL-NoQ to All Languages
In Montague Grammar, our method for defining FOL-NoQ by means of a ‘syntactic algebra’ is
generalized to all languages, in (approximately) the following way.
(21) General Definition of a Language (To be Revised)
A language L is a structure < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ such that:
a. < A , Fγ >γ ∈ Γ is an algebra.
o Note: the set Γ is an index set, used to index the (syntactic) operations in the
(syntactic) algebra < A , Fγ >γ ∈ Γ
b. A is the smallest set such that:

(i) For all δ ∈ Δ, Xδ ⊆ A.
(ii) A is closed under the operations Fγ for all γ ∈ Γ
o Note: The set Δ is the set of category labels, being used to index each of the
set of ‘basic categories’ (lexical items) Xδ
o Note: The conditions in (i) and (ii) are what give us that < A , Fγ >γ ∈ Γ is a
‘syntactic algebra’, where A contains all the well-formed structures of the
language L (as well as a ton of junk).
c. S is a set of sequences of the form < Fγ , < δ1, …, δn >, δ >, where γ ∈ Γ, Fγ is an
n-ary operation, and δ1, …, δn, δ ∈ Δ
o Note: This condition states that a syntactic rule consists of:
(i) some n-ary operation Fγ from the algebra < A , Fγ >γ ∈ Γ
(ii) a n-tuple of category labels from the set of labels Δ

(the categories of the expressions ‘input’ to Fγ)
(iii) some category label δ ∈ Δ

(the category of the resulting output)
d. δ0 ∈ Δ
o Note: δ0 will be the category label of the ‘sentences’ of L.
o Note: Given that a language L is often characterized as a ‘set of
sentences’, this element δ0 will allow us to define such a set L
given our structure above.
11
Fall 2013 Ling 720
Key Observation:
Our ‘algebraic characterization’ of FOL-NoQ in (20) is a specific instance of the general kind of
structure in (21).
With the general definition in (21) at our disposal, we can also offer the following general
definition of what it is to be a ‘syntactic category’ of the language L.
(22) General Definition of Syntactic Categories of a Language

Let L be a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ. Language L generates the family
CAT of syntactic categories iff:
a. CAT is a family of subsets of A, indexed by Δ
o Note: This means that the members of CAT are the sets Cδ , for all δ ∈ Δ
o Note: This all means that for all δ ∈ Δ, Cδ ⊆ A.
b. Xδ ⊆ Cδ , for all δ ∈ Δ
o Note: This means that for every category label δ ∈ Δ, the category Cδ
contains all the ‘basic expressions’ (lexical items) of that category.
c. If the sequence < Fγ , < δ1, …, δn >, δ > ∈ S, and ϕ1, …, ϕn are such that ϕi ∈ Cδi,
then Fγ(ϕ1…ϕn) ∈ Cδ
o Note: This is just a straightforward generalization of (16b)
o Note: This says that Cδ will contain – not just the basic expressions
lexical items Xδ - but also all the expressions of category δ you can
generate via the syntactic rules.
d. For each Cδ ∈ CAT, Cδ is the smallest set satisfying conditions (22a)-(22c).
(23) The ‘Meaningful Expressions’ of a Language L

Let L be a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ and let L generate the family CAT of
syntactic categories. The meaningful expressions of L (MEL) is the union of all the
categories in CAT ( i.e., ∪δ∈ΔCδ )
12
Fall 2013 Ling 720
4. Applying Our Definition of a ‘Language’ to (a Fragment of) English
If (21) truly is a viable definition of what a ‘language’ is, then it should be possible to represent a
natural language, like English, as such a system.
o In this section, we’ll show that this is possible for a subpart (fragment) of English,
roughly corresponding in its expressive power to FOL-NoQ
(24) Some Comments
a. In representing (a fragment of) English as a system like (21), we are basically

giving a theory of English syntax.
 I.e., the system responsible for constructing complex expressions of

English from the primitive lexical items.
b. One early and perennial criticism of Montague Grammar was its accompanying
theory of syntax, stated in (21).
c. Note, though, that the intellectual motivation behind Montague’s theory of syntax
was completely different from that of generative linguists.
 Montague wanted a theory broad enough to cover anything that we’d

reasonably call a ‘language’ (including artificial and non-human ones)
 Thus, it’s a feature – not a bug – that his system of syntax doesn’t
(necessarily) capture generalizations regarding the structure of human
languages.
(25) The Syntactic Category Labels

Let the set Δ consist of the following syntactic category labels:
a. NP (noun phrases; proper names for now)
b. IV (intransitive verbs and derived verb phrases)
c. TV (transitive verbs)
d. S (sentences)
(26) The Basic Expressions (The Lexicon)

For each of the category labels above, the basic expressions of that category are:
a. XNP = { Barack, Michelle, Mitt }
b. XIV = { smokes }
c. XTV = { loves }
d. XS = ∅
13
Fall 2013 Ling 720
(27) The Syntactic Operations
a. Merge:
Binary operation that maps two strings ‘x’, ‘y’ to the string ‘x y’
Illustration: Merge( loves, Mitt ) = loves Mitt
b. AndE
Binary operation that maps two strings ‘x’, ‘y’ to the string ‘x and y’
Illustration: AndE( Mitt, Barack ) = Mitt and Barack
c. NotE
Unary operation that maps a string ‘x’ to the string ‘it is not the case that x’
Illustration: NotE( Barack smokes ) = It is not the case that Barack smokes
(28) The Syntactic Rules

Let the set SE consist of the following tuples:
a. < Merge, < TV, NP >, IV >

‘The result of merging a TV with an NP is a IV’
b. < Merge, < NP, IV >, S >

‘The result of merging an NP with an IV is an S’
c. < AndE , <S , S> , S >

‘The result of applying AndE to an S and an S is an S’
d. < NotE, <S>, S >

‘The result of applying NotE to an S is an S’.
(29) The Definition of ‘Mini-English’
The language ‘Mini-English’ is the structure < A, Merge, AndE, NotE, Xδ , SE, S > δ ∈ Δ
such that:
a. A is the smallest set such that:

(ii) A is closed under the operations Merge, AndE and NotE
b. Merge, AndE, NotE, Xδ , SE, S, and Δ are all as defined in (25)-(29)
14
Fall 2013 Ling 720
(30) Some Remarks
a. The structure ‘Mini-English’ in (29) clearly satisfies our general definition of a

‘language’ in (21).
b. The set A contains infinitely many well-formed sentences of English (Mitt

smokes, Barack loves Michelle and Michelle smokes, etc.)
c. The set A also contains a whole bunch of ‘syntactic garbage’ (smokes loves, it is
not the case that Mitt, Michelle Barack, etc.)
d. We can use the rules in SE and the general definition in (22) to generate the
syntactic categories Cδ of English.
 As with FOL-NoQ, we can use ‘analysis trees’ as a handy way of
representing the calculations (31).
e. Note that given rule (28a) the string loves Michelle is the same category (IV) as
the intransitive verb smokes.
 This is a recurring idea in Montague’s papers and other early MG works
 That is, Montague and others don’t distinguish syntactically between
‘VPs’ and single intransitive verbs.
 I believe the main reason for this is that – from their perspective – this
is more elegant than having an extra rule converting all IVs to VPs…
(31) Some Sentences of Mini-English
a. Michelle loves Barack.
< Michelle loves Barack, Merge >
Michelle < loves Barack, Merge >
loves Barack
b. Barack smokes and it is not the case that Mitt smokes.
< Barack smokes and it is not the case that Mitt smokes, AndE >
< Barack smokes, Merge > < It is not the case that Mitt smokes, NotE >
Barack smokes < Mitt smokes, Merge >
Mitt smokes
15
Fall 2013 Ling 720
Montague’s General Theory of Semantics 1
1. Revising Our Algebraic Perspective on ‘Interpretation’
Last time, we motivated and illustrated the following general definition of what a ‘language’ is:
(1) The Montagovian Definition of ‘Language’ (To Be Slightly Revised)

(ii) A is closed under the operations Fγ for all γ ∈ Γ
b. δ0 ∈ Δ
Given this (re)definition of what a ‘language’ is, we must now slightly revise our earlier
definition of what an ‘interpretation’ of a language is.
(2) The Montagovian Definition of ‘Interpretation’ (To Be Slightly Revised)

Let L be a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ. An interpretation for L is a structure
< B, Gγ , f >γ ∈ Γ such that:
a. < B, Gγ >γ ∈ Γ is an algebra with the same number of operations as < A , Fγ >γ ∈ Γ
Note: This is basically already represented in the fact that that the operations Gγ
and Fγ are indexed by the same set Γ
b. Gγ is an operation of the same arity as Fγ , for all γ ∈ Γ
Note: As we’d expect, each Gγ is the ‘semantic operation’ corresponding to the

syntactic operation Fγ
c. f is a function from ∪δ∈ΔXδ into B.
Note: f is a function from the basic expressions (lexical times) to some meanings.
Thus, f represents the ‘lexical semantics’ of our language, and so is like the
function ‘I’ in a model <D, I>.
1
1
Fall 2013 Ling 720
(3) The Definition of ‘Meaning Assignment’

Let L be a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ. Let B = < B, Gγ , f >γ ∈ Γ be an
interpretation for L .
The meaning assignment for L determined by B is the unique homomorphism g from the
syntactic algebra < A , Fγ >γ ∈ Γ to the semantic algebra < B, Gγ >γ ∈ Γ such that f ⊆ g
(4) Some Remarks
a. The fact that an interpretation B does indeed determine such a (unique)

homomorphism g is not obvious, but follows from conditions (2a,b).
b. Under the definitions in (2) and (4), we still conceive of ‘meaning’ (i.e.,
semantics) as a homomorphism from a syntactic algebra to a semantic one.
c. Under these new definitions, however, a meaning assignment g will also map all
the ‘syntactic garbage’ in A to some kind of meanings in B.
• This consequence is definitely surprising. We’ll see in a bit, however, that

it’s basically ‘harmless’.
• Note, too, that it’s a necessary consequence of our general conception of

semantics as ‘homomorphism between algebras’.
o Since the set of all and only the well-formed expressions of a language
don’t generally form an algebra…
o In the general case, the ‘syntactic algebra’ for a language will contain
a bunch of ‘syntactic garbage’.
o Consequently, a semantics for the language will also end up

interpreting the ‘syntactic garbage’ in the algebra…
We’ll now illustrate the definitions in (2) and (3) by describing a general class of interpretations
for FOL-NoQ, as well as some specific instances of this class…
2
Fall 2013 Ling 720
2. Towards The Notion of a ‘Fregean Interpretation’
In his classic paper ‘Universal Grammar’, Montague introduces the term ‘Fregean Interpretation’
as a label for a specific kind of ‘interpretation’, as defined in (2).
• It’s a very complex definition, and we’ll begin to get a handle on it by introducing a core
sub-part of it for our language of FOL-NoQ
(5) The Language FOL-NoQ

The language FOL-NoQ is the structure < A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T
where:
a. < A, Fγ >γ ∈ {Concat, Not, And} is the algebra such that
(i) FConcat = Concat (from previous notes)

FNot = Not (from previous notes)
FAnd = And (from previous notes)
(ii) A is the smallest set such that:

1. It contains every individual constant and predicate letter.
2. It is closed under FConcat, FNot, FAnd
b. The basic categories Xτ are such that:
(i) Xe = The set of individual constants, CONS
(ii) X<e, …, t > = The set of n-ary predicate letters, where n = the number of
times e appears in the category label X<e, …, t >
(iii) For all other types τ ∈ T, Xτ = ∅.
c. The set S is the following (infinite) set of syntactic rules:

(i) < FNot , < t >, t>
(ii) < FAnd , <t, t>, t>
(iii) < FConcat , < <σ, τ>, σ >, τ >, for all σ, τ ∈ T
Illustrative Members of Ct :
(((Qa)b) & ~(Rc)) ~(~(Ab) & ~((Sc)d)) ~(((Tg)b) & ~(Lg))
We’ll begin by defining a family of sets that will serve as the semantic values in a ‘Fregean
interpretation’
3
Fall 2013 Ling 720
(6) The Set of ‘Possible Denotations’ (To be Revised)
Let T be the set of types and E be some non-empty set (of entities). The set Dτ, E of
denotations of type τ based on E is defined as follows:
(i) De,E = E
(ii) Dt,E = { 0, 1 }
(iii) If σ, τ ∈ T, then D<σ,τ>,E = the set of functions from Dσ ,E to Dτ,E
= Dτ,EDσ,E
Illustration: Let E = { Michelle, Barack, Mitt }.
De,E = {Michelle, Barack, Mitt}
Dt,E = { 0, 1 }
D<et>,E = All the functions from {Michelle, Barack, Mitt} to {0,1}

All the characteristic functions of members of P(E).
D<e,<et>>,E = All the functions from {Michelle, Barack, Mitt} to

some function from {Michelle, Barack, Mitt} to
{1, 0}
= All the characteristic functions of members of E × E
The family of sets defined in (6) will be a key ingredient in the definition of a ‘Fregean
Interpretation’ of FOL-NoQ…
The second key ingredient will be the semantic operations defined below…
(7) The Semantic Operations (To Be Revised)
a. The Operation ‘GNot’

GNot = { <1,0>, <0,1>}
b. The Operation ‘GAnd’

GAnd = { <<1,1>,1>, <<1,0>,0>, <<0,1>,0>, <<0,0>,0> }
c. The Operation ‘GConcat’ 2

If α ∈ D<σ, τ>, E and β ∈ Dσ, E, then GConcat(α,β) = α(β)
With these ingredients, we can now make a first attempt at defining what a ‘Fregean
Interpretation’ of FOL-NoQ is…
2
Note that GConcat is basically an operation of ‘function application’.
4
Fall 2013 Ling 720
(8) Fregean Interpretation of FOL-NoQ (To Be Substantially Revised)
Let E be a set (of entities). A Fregean Interpretation of FOL-NoQ based on E is an
interpretation of FOL-NoQ < B, Gγ , f >γ ∈ {Concat, Not, And} such that
(i) B = ∪τ∈T Dτ, E

Note: The set B of semantic values is the union of all the denotations of type τ
based on E.
(ii) The operations GConcat, GNot, GAnd are as defined in (7).
(9) An Immediate Problem for the Definition in (8)
• According to our definition in (2), in order for < B, Gγ , f >γ ∈ {Concat, Not, And} to be an
‘interpretation’, < B, Gγ >γ ∈ {Concat, Not, And} must be an algebra.
• However, if B = ∪τ∈T Dτ, E , then it isn’t closed under GConcat, GNot, GAnd
o GNot, GAnd are only defined for elements of B that are in Dt,E
o GConcat is only defined for pairs α, β such that α ∈ D<σ, τ>, E and β ∈ Dσ, E,
(10) A Related Problem
• If < B, Gγ , f >γ ∈ {Concat, Not, And} is to be an interpretation, then the meaning

assignment based on it must be homomorphism from A in (5) to B = ∪τ∈T Dτ, E
• But remember that A contains a whole bunch of ‘syntactic garbage’.
• Thus, there has to be ‘meanings’ in B that this syntactic garbage gets mapped to
• For example, since GNot, and FNot would ‘correspond’ in the homomorphism, we
need to have it that for an individual constant ‘a’ where f(a) = α ∈ De, E
g( ~a ) =
g(FNot(a)) =
GNot(g(a)) =
GNot(f(a)) =
GNot(α) = ??? = some kind of ‘semantic garbage’ corresponding

to the ‘syntactic garbage’ ~a
5
Fall 2013 Ling 720
(11) Incorporating ‘Semantic Garbage’ Into Our System
What is ‘semantic garbage’, however? And how do we incorporate it into our system?
• Montague is conspicuously silent on the matter in UG and PTQ. To my knowledge,

there’s no indication in his extant writings on how he thought the issue in (9)/(10)
could/should be concretely handled (probably it was too trivial to discuss…)
• Halvorsen & Ladusaw (1979) have some more focused discussion of the matter, but
they also leave the nature of the ‘semantic garbage’ entirely open (and cryptic).
• For better or worse, the following is my own proposal: we are going to slightly
redefine De so that it contains a special element ‘garbage’
o We are then going to redefine the operations Gγ in light of this…
(12) The Set of ‘Possible Denotations’ (To be Revised)

Let T be the set of types and E be some non-empty set (of entities). The set Dτ, E of
denotations of type τ based on E is defined as follows:
(i) De,E = E ∪ { garbage }

(ii) Dt,E = { 0, 1 }
= Dτ,EDσ,E
(13) The Semantic Operations
a. The Operation ‘GNot’
(i) If α ∉ Dt,E, then GNot(α) = garbage
(ii) If α ∈ Dt,E, then GNot(α) = 1, if α = 0

0, otherwise
b. The Operation ‘GAnd’
(i) If α ∉ Dt,E or β ∉ Dt,E, then GAnd(α,β) = garbage
(ii) If α, β ∈ Dt,E, then GAnd (α,β) = 1, if α = β = 1

0, otherwise
c. The Operation ‘GConcat’
(i) If α = garbage or β = garbage, then GConcat(α,β) = garbage
(ii) If not α ∈ D<σ, τ>, E and β ∈ Dσ, E, then GConcat(α,β) = garbage
(iii) Otherwise, if α ∈ D<σ, τ>, E and β ∈ Dσ, E, then GConcat(α,β) = α(β)
6
Fall 2013 Ling 720
(14) Illustration of These Definitions
Let f be the <et> function { <Barack, 0>, <Michelle, 1> }3
a. GNot(1) = 0 b. GNot(Barack) = garbage

c. GAnd(1,1) = 1 d. GAnd(1, f ) = garbage
e. GConcat(f, Barack) = f(Barack) = 0 f. GConcat(f, garbage) = garbage
g. GConcat(Michelle,Barack) = garbage h. GConcat(f, f ) = garbage
With the formalization of ‘semantic garbage’ in (12) and (13) at hand, we can now properly
define a ‘fregean interpretation’ for FOL-NoQ…
(15) Fregean Interpretation of FOL-NoQ (To Be Substantially Revised)
Let E be a non-empty set (of entities). A Fregean Interpretation of FOL-NoQ based on E

is an interpretation of FOL-NoQ < B, Gγ , f >γ ∈ {Concat, Not, And} such that
a. B = ∪τ∈T Dτ, E
Note: Again, the set B of semantic values is the union of all the denotations of
type τ based on E.
Note: This definition now makes use of the revised definition in (12), which adds
the special element garbage to De, E
b. The operations GConcat, GNot, GAnd are as defined in (13).
Note: Under the new definitions in (13), B is closed under GConcat, GNot, GAnd
c. The function f is such that

(i) For all τ ∈ T, if α ∈ Xτ then f(α) ∈ Dτ, E
(ii) For all α ∈ Xe , f(α) ≠ garbage
Note: The conditions in (15c) entail that:

• Every individual constant will be mapped to some entity in De
• No individual constant will be mapped to (interpreted as) garbage (in De)
• A n-ary predicate letter of type <e,…,t> will be mapped to a function in
D<e,…,t>,E
o Thus, an n-ary predicate letter is mapped to the curried characteristic
function of an n-ary relation! (Just as it should be).
3
Technically speaking f must also now be defined for garbage. We can simply assume that f(garbage) = 0 for all
<et> functions f. Similar such assumptions can be made for functions of the other types. From this point on, we’ll
simply ignore garbage in our specifications of the denotations of the basic lexical items.
7
Fall 2013 Ling 720
(16) A Concrete Instance of a Fregean Interpretation of FOL-NoQ
Let the set S = { Michelle, Barack, Mitt }. Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be the
Fregean interpretation based on S, such that f contains the following mappings:
a. f(a) = Michelle
b. f(b) = Barack
c. f(c) = Mitt
d. f(P) = h = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
e. f(Q) = j = Michelle  Michelle  1

Barack  1
Mitt  0
Barack  Michelle  1
Barack  1
Mitt  0
Mitt  Michelle  0
Barack  0
Mitt  1
Note: We’ve basically interpreted ‘P’ as the property ‘is president’, and we’ve
interpreted ‘Q’ as the relation ‘x is loved by y’.
We can now use the ‘meaning assignment g determined by B’ to interpret formulae in our FOL-
NoQ language!...
(17) Using the Fregean Interpretation in (16) to Interpret Sentences of FOL-NoQ

Let g be the meaning assignment for FOL-NoQ that is determined by B in (16).
(i) g( ~(Pa) ) = (by definition of FOL-NoQ)
(ii) g( FNot( FConcat(P,a)) ) = (by homomorphism property of g)

(iii) GNot( g(FConcat(P,a)) ) = (by homomorphism poperty of g)
(iv) GNot( GConcat( g(P), g(a) )) = (by definition of g) 4

(v) GNot( GConcat( f(P), f(a) )) = (by definition of f in (16))
(vi) GNot( GConcat( h, Michelle )) = (by definition of GConcat)

(vii) GNot( h(Michelle) ) = (by definition of h in (16d))
(viii) GNot( 0 ) = (by definition of GNot)
(ix) 1
4
Note that by the definition in (3), f ⊆ g, and so if α ∈ X then g(α) = f(α).
τ
8
Fall 2013 Ling 720
(18) Using the Fregean Interpretation in (16) to Interpret Sentences of FOL-NoQ
Let g be the meaning assignment for FOL-NoQ that is determined by B in (16).
(i) g( (((Qa)b) & (Pb))) = (by definition of FOL-NoQ)
(ii) g( FAnd( FConcat(FConcat(Q,a),b)), FConcat(P,b) ) ) = (by homomorphism property of g)
(iii) GAnd( g( FConcat( FConcat(Q,a),b)) ), g( FConcat(P,b) ) ) = (by homomorphism property of g)
(iv) GAnd( GConcat( g(FConcat(Q,a), g(b))) , GConcat( g(P),g(b) ) ) =

(by homomorphism property of g)
(v) GAnd(GConcat(GConcat(g(Q), g(a)), g(b)), GConcat( g(P),g(b) ) ) = (by definition of g)
(vi) GAnd( GConcat(GConcat( f(Q), f(a) ), f(b)), GConcat( f(P), f(b) ) ) = (by definition of f)
(vii) GAnd( GConcat(GConcat( j, Michelle ), Barack ), GConcat( h, Barack ) ) =

(by definition of GConcat)
(viii) GAnd( GConcat(GConcat( j, Michelle ), Barack ), h(Barack) ) = (by definition of h)
(ix) GAnd( GConcat(GConcat( j, Michelle ), Barack ), 1 ) = (by definition of GConcat)
(x) GAnd( GConcat( j(Michelle), Barack ), 1 ) = (by definition of j)
(xi) GAnd( GConcat( {<Michelle,1>,<Barack,1>,<Mitt,0>}, Barack ), 1 ) =

(by definition of GConcat)
(xii) GAnd( 1, 1 ) = 1 (by definition of GAnd)
(19) Using the Fregean Interpretation in (16) Interpret Syntactic Garbage
(i) g( (P~a) ) = (by definition of FOL-NoQ)
(ii) g( FConcat( P, FNot(a) ) ) = (by homomorphism property of g)
(iii) GConcat( g(P), GNot( g(a) ) ) = (by definition of g)
(iv) GConcat( f(P), GNot( f(a) ) ) = (by definition of f)
(v) GConcat(h , GNot(Michelle) ) = (by definition of GNot in (13a))
(vi) GConcat(h, garbage) = (by definition of GConcat in (13c))
(vii) garbage
9
Fall 2013 Ling 720
3. Fregean Interpretations and Models of FOL-NoQ
(20) Taking Stock of What We’ve Done So Far
• In the last set of notes, we developed the general definition of ‘language’ in (1),
according to which any language is built upon a ‘syntactic algebra’ < A , Fγ >γ ∈ Γ
• We developed a notion of ‘meaning’, according to which a semantics for a given

language is a homomorphism from the syntactic algebra < A , Fγ >γ ∈ Γ to a semantic
algebra < B , Gγ >γ ∈ Γ
• We developed a notion of ‘Fregean interpretation’, a special kind of interpretation for

our FOL-NoQ language.
• We saw how one specific such ‘Fregean interpretation’ can be used to map sentences
of FOL-NoQ to truth-values (just like a classical model!)
There’s an important relation between our concept of ‘Fregean Interpretation’ in (15) and our
concept of a ‘model’ for FOL, repeated below:

(ii) If Φ is an n-ary predicate letter, then I(Φ) is the curried characteristic
function of an n-ary relation R ⊆ Dn
(22) The Correspondence Between (Fregean) Interpretations and Models, Part 1
Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be a Fregean interpretation (for FOL-NoQ); let g be
meaning assignment determined by B. Let M = <D,I> be such that D = De,E and I = f.
a. Claim: <D,I> is a model (for FOL-NoQ)

• D = De,E is by definition a non-empty set.
• By definition in (15c), f is a function satisfying the condition in (21b).
b. Claim: For any sentence ϕ (of FOL-NoQ), g(ϕ) = [[ϕ]]M
(proof left as exercise for the student)
10
Fall 2013 Ling 720
(23) The Correspondence Between (Fregean) Interpretations and Models, Part 2
Let M = <D,I> be a model (for FOL-NoQ). Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be such
that B = ∪τ∈T Dτ, D and f = I.
a. Claim: <B, Gγ , f>γ ∈ {Concat, Not, And} is a Fregean interpretation (for FOL-NoQ)
• By definition, B = ∪τ∈T Dτ, E , for a (non-empty) set E = D
• By definition, GConcat, GNot, GAnd are as defined in (13).
• By definition (21b), function I satisfies the conditions in (15c); it is such that

(i) For all τ ∈ T, if α ∈ Xτ then f(α) ∈ Dτ, E
(ii) For all α ∈ Xe , f(α) ≠ garbage
b. Claim: Let g be the meaning assignment determined by B. For any sentence ϕ (of
FOL-NoQ), g(ϕ) = [[ϕ]]M
(proof left as exercise for the student)
(24) Some Remarks
a. Thus, for every Fregean interpretation B based on E, there is a corresponding

model M whose domain is E, and which makes exactly the same sentences true.
b. Similarly, for every model M = <D,I>, there is a Fregean interpretation B based

on D, and which makes exactly the same sentences true.
c. Given this relationship between models and Fregean interpretations, we can freely
shift between the two (cf. sets and their characteristic functions).
d. Similarly, we can view Fregean interpretations and models as being in essence

‘the same thing’ (even though they are different set-theoretic objects)
(cf. sets and their characteristic functions)
e. Thus, if we can provide a Fregean interpretation for a given language, we’ve

also thereby provided that language with a model-theoretic semantics.
f. Finally, given this correspondence between models and Fregean interpretations,

we can see that our general notion of ‘interpretation’ in (2) is truly a more general
concept of ‘interpretation’ than is found in our model-theoretic semantics of FOL.
11
Fall 2013 Ling 720
4. Extending The Framework to Natural Languages
(25) Burning Question:

Can we provide a (Fregean) interpretation to (a fragment of) a natural language, like
English?
A second fundamental contribution of Montague’s was the discovery that there are, in principle,
two different ways of providing an interpretation for a language.
(26) Method 1: Direct Interpretation of a (Natural) Language L
a. Step 1:
Provide an analysis of language L as a structure < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ
satisfying the core definition in (1).
b. Step 2: Specify a structure < B, Gγ , f >γ ∈ Γ satisfying the core definition in (2)
of an ‘interpretation’ for < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ
• This is the method pursued by Montague in his “English as a Formal Language”
(27) Method 2: Indirect Interpretation (To be Revised)
a. Step 1:
Provide an analysis of language L as a structure < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ
satisfying the core definition in (1).
b. Step 2:
Take an (artificial) language L’ = < A’ , F’γ’ , X’δ’ , S’, δ0’ >γ’ ∈ Γ’ , δ’ ∈ Δ’ such
that there is already a known interpretation B = < B, Gγ’ , f >γ’ ∈ Γ’ for L’.
c. Step 3: (To be Revised)

Give a homomorphism h from A (the structures of L) to A’ (the structures of L’)
o This homomorphism h is effectively a ‘translation’ from L into L’.
• This is the method pursued by Montague in his most seminal papers, UG and PTQ
(28) Key Skeptical Question

With this ‘indirect interpretation’, all we really do is supply a ‘translation’ h from one
language L into another L’. Have we really provided L with a semantics when we do this?
12
Fall 2013 Ling 720
(29) Answer to the Skeptical Question
• Recall that the composition of two homomorphisms is itself a homomorphism

(Assignment 4)
• Therefore, if B is the interpretation of L’, and g is the (homomorphic) meaning

assignment determined by B, and h is the (homomorphic) translation from L to L’,
then g°h is a homomorphism from A (the expressions of L) to B (the meanings
assigned to L’).
• Therefore, by providing an indirect interpretation for a language L, we’ve also

provided a (model-theoretic) semantics for L
Indirect Interpretation in a Picture (Oversimplified):
< A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ
h (the syntactic translation from L to L’)

g°h (the semantic
interpretation of L)
< A’ , F’γ’ , X’δ’ , S’, δ0’ >γ’ ∈ Γ’ , δ’ ∈ Δ’
g (the semantic interpretation of L’)
< B, Gγ’ , f >γ’ ∈ Γ’
In other words, with ‘indirect interpretation’, we get our semantics for L ‘indirectly’,
through the translation language L’. But we still do get a semantics for L.
This notion of ‘indirect interpretation’ will become clearer once we work through a concrete
example…
There are a few important properties of ‘indirect interpretation’ that we can go ahead
and observe, however….
(30) Indirect Interpretation is ‘Eliminable’
• Note that if we ‘indirectly interpreted’ language L via language L’, then we have also
thereby ‘directly interpreted’ L, via the composition of the two homomorphisms g°h
• Thus, indirect interpretation is never a necessary part of giving a semantics for a

language L; it’s ‘eliminable’:
o Once we’ve done ‘indirect interpretation’ we could recast everything we’ve
done as ‘direct interpretation’ via the composition g°h
13
Fall 2013 Ling 720
(31) Some Choice Quotes
• From Halvorsen & Ladusaw (1979), p. 210:

“An understanding of [the eliminability of indirect interpretation] is necessary to
understand the use of…logics of PTQ and most other analyses within Montague
grammar. As has been stated elsewhere, the [logical language] is an expository device
and is in no way a necessary part of an analysis of any language offered within this
theory. By using the easily interpreted [logical language] as a mediator, natural
languages can be analyzed syntactically and then provided with a translation…from
them to [the logical language] to induce their interpretation….This method of
analysis amounts to direct interpretation of natural language.” (emphasis mine)
• From Dowty et al. (1981), p. 263:

“Translating English into [a logical language] was therefore not essential to
interpreting the English phrases we generated; it was simply a convenient
intermediate step in assigning them meanings. This step could have been
eliminated had we chosen to describe the interpretation of English directly… This
point is important, because anyone who does not appreciate it may misunderstand the
role of [logical languages] in applications of Montague’s descriptive framework to
natural languages.” (emphasis mine)
The general point here is that, to give a semantics for (e.g.) English in Montague Grammar, you
don’t have to translate English into any logical language…
• It’s just that doing so can be a very handy, elegant means of specifying an
interpretation for the language.
Again, this will all become clearer when we’ve seen some concrete instances of it…
(32) Why Do Indirect Interpretation?
If the logical language is well-designed and familiar to readers, then it can provide a more
‘perspicuous’ representation (statement / name) of the meanings that we wish to be
assigned to the (natural) language expressions.
Illustration (From 610):

‘[ λxe : x smokes ]’ vs. ‘The function f from De to Dt such that for all x ∈ De,
f(x) = 1 iff x smokes.’

“[The purpose of indirect interpretation] was to have a convenient, compact notation
for giving a briefer statement of semantic rules than we were able to give in earlier
chapters of this book, where semantic rules were formulated rather long-windedly in
English… [The logical language] could provide us with names for meanings…”
14
Fall 2013 Ling 720
(33) Partisan Comment: Indirect Interpretation and Semantics in the 21st Century
• In a certain sense, much of the semantic literature nowadays is rife with ‘indirect
interpretation’, but without an explicit, model-theoretic semantics for the translation
language.
• When in semantics paper we write things like [[-er]] = [ λP<dt>: λQ<dt>: … ], we

assume that whatever appears to the right of ‘=’ is understood to our readers (and
ourselves).
o This isn’t always the case though… sometimes formulas and notations
creep in that seem to make sense, but turn out not to upon deeper probing…
o In the original Montague Grammar framework, there is much lower risk of

this happening, since one must also provide an explicit model-theoretic
semantics for their interpretation language.
So, this is the general gist of how ‘indirect interpretation’ in MG works…

Note, however, that the characterization in (27) has some gross oversimpliciations…
In the next set of notes, we’ll set the record straight by delving into Montague’s theory of
translation…
15
Fall 2013 Ling 720
Problem Set on Languages and Interpretations
(1) Basic Comprehension Questions on Our (Provisional) Definition of a ‘Language’

Let L be a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ as defined in (21) on the handout
“An Algebraic Perspective on the Syntax of First Order Logic”.
a. What are the syntactic rules of L?
b. What are the syntactic category labels of L?
c. What are the basic expressions (‘lexical items’) of L? (Please represent as an

indexed family of sets.)
d. Which elements in L form an algebra together?
f. What is the category label for the declarative sentences of L?
g. What are the syntactic operations of L? (Please represent as an indexed family of

sets).
h. What is the difference between A and the meaningful expressions of L? Can they
ever be the same?
i. Let CAT be the syntactic categories of L.

(i) Which element are the members of CAT subsets of?
(ii) Which element serves to index the members of CAT?
(iii) Please represent CAT as an indexed family of sets.
(iv) Which elements in L work together to generate CAT?
(2) Basic Exercise in Language Design

Let < A, Fγ , Xδ , SE, S >γ ∈ {Merge, And, Not} δ ∈ {NP, IV, TV, S} be the language ‘Mini-
English’, as defined in (29) on the handout “An Algebraic Perspective on the Syntax of
First Order Logic”.
a. Please alter this structure minimally, so that the category CS includes strings like
the following:
If Mitt smokes, then Barack smokes.
b. Please provide an analysis tree showing how the following string is derived.
If Barack loves Michelle, then it is not the case that Mitt smokes.
c. Are the following meaningful expressions of the language you defined? Why or
why not?
(i) Then Barack smokes.
(ii) If Barack loves Michelle.
(ii) And Barack smokes.
1
Fall 2013 Ling 720
(3) More Advanced Exercise on Language Design
Let First Order Logic (FOL) be the language defined in (16) on the handout “An
Algebraic Perspective on Propositional Logic.”
a. Represent FOL as a language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ , following the

definition in (21) on the handout “An Algebraic Perspective on the Syntax of First
Order Logic”. To do this, you will need to do the following:
(i) Identify a set of syntactic operations { Fγ }γ ∈ Γ that will generate the

WFFs of FOL (as defined in (16)).
(ii) Use these operations to define an algebra <A, Fγ>γ ∈ Γ such that the WFFs
of FOL are a subset of A.
(iii) Special Hint: Let the syntactic category labels Δ be T ∪ {var}, where T is
the set of types, and ‘var’ is the category label for variables.
(iv) Organize the basic expressions of FOL into sets { Xδ }δ ∈ Δ

Special Hint: Make sure that Xe contains both the individual constants
and the variables.
(v) Write out a set of syntactic rules S for FOL.

Special Hint: You might find it helpful to consult (17) on “An Algebraic
Perspective on Propositional Logic”
b. Given your representation of FOL as a language, please provide an analysis tree

showing how the following formula is generated by your system (where P is a
unary predicate letter and Q is a binary predicate letter):
∀x(Px & ~∃y((Qa)y))
c. Let R be a ternary predicate letter, a and c be individual constants, and x be a

variable.
(i) Please state whether the following are or are not meaningful expressions
of the language you defined.
1. ((Ra)x)
2. (((Ra)x)c)
3. ∃x((Ra)x)
(ii) If the string is a meaningful expression, please provide a calculation

showing what category it is a member of (follow the format in (17) on the
handout “An Algebraic Perspective on the Syntax of First Order Logic”.
(iii) If the string is not a meaningful expression, please provide a brief

explanation of why it isn’t.
2
Fall 2013 Ling 720
(4) Another Exercise on Language Design
Let Propositional Logic (PL) be the language defined in (4) on the handout “An
Algebraic Perspective on Propositional Logic.” Please represent PL as a language,
following the definition in (21) on the handout “An Algebraic Perspective on the Syntax
of First Order Logic”.
HINT: Adapt what you did in exercise (3).
(5) An Exercise on Models and (Montagovian) Interpretations

Let FOL-NoQ be the language < A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T defined in (5)
on the handout “Montague’s General Theory of Semantics” (MGTS).
a. Let E be a set of entities, and let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be a Fregean
interpretation for FOL-NoQ based on E (as defined in (15) on MGTS). Let M be
a model <E, I> (as defined in (21) on MGTS), where I = f.
Please show via induction on structural complexity that every ϕ ∈ Ct is such that
[[ϕ]]M = g(ϕ), where g is the meaning assignment based on B.
Some Hints:
1. First, show that if ϕ is an atomic formula of FOL-NoQ, then [[ϕ]]M = g(ϕ).
To show this, first use our definition of a model M to show:
[[ (…(Φα1)…αn) ]]M = g(Φ)(g(α1))…(g(αn))
Then, use the homomorphism property of g to show:
g(Φ)(g(α1))…(g(αn)) = g( (…(Φα1)…αn) )
2. Next, assume that ϕ is a conjunction (ψ & χ), and that ψ and χ are both such that [[ψ]]M
= g(ψ) and [[χ]]M = g(χ). Show that [[ϕ]]M = g(ϕ).
To show this, use the definition of a model M and the induction assumption to show:
[[(ψ & χ)]]M = GAnd( g(ψ), g(χ) )
Then use the homomorphism property of g to show:
GAnd( g(ψ), g(χ) ) = g( (ψ & χ) )
3. Next, assume that ϕ is a negation ~ψ, and that ψ is such that [[ψ]]M = g(ψ). Show that
[[ϕ]]M = g(ϕ).
To show this, follow the same general strategy laid out in 2.
b. Let M be a model <D, I>. Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be a Fregean
interpretation for FOL-NoQ based on D, where f = I.
Please show via induction on structural complexity that every ϕ ∈ Ct is such that
[[ϕ]]M = g(ϕ), where g is the meaning assignment based on B.
3

Unit 4:
Montague’s Theory of Translation

Fall 2013 Ling 720
Montague’s Theory of Translation: Laying the Groundwork 1
1. Some Context and Questions to Have in the Background
(1) Fundamental Fact: Translation ≠ Semantics
 Translating expressions from a language L into a language L’ doesn’t necessarily tell

you anything about the semantics of the expressions in language L.
o For example, simply knowing that (1a) can be translated as (1b) doesn’t tell
you anything about what (1a) means.
a. Tlingit Sentence: Ax éet yaan uwaháa

b. Haida Translation: Dii.uu q’wiidang gwaa.
 However, translating an expression from a language L into a language L’ whose

semantics are known can inform you of the semantics of the expressions in L.
o For example, given that you speak English, knowing that (1a) can be
translated as (1c) would inform you of the meaning of (1a).
c. English Translation: I’m hungry. (~ Hunger moves to me imperceptibly.)
• Of course, even such translations don’t necessarily mean one has a compositional
semantics for language L.
(2) Question (Not Necessarily Montague’s)

Given our background theory of language and meaning, under what conditions (if any)
can we guarantee that translating from one language L into another language L’ gives us a
compositional semantics for L.
(3) Some Historical Context: Semantics in Generative Grammar Before Montague

Prior to Montague (and for some time afterwards), generative linguists conceived of
natural language semantics as having the following goal:
• Develop a theory of the system that maps syntactic structures to some kind of
‘mentalese’ (conceptual structures) encoding the information in the sentence.
IIlustration: Jackendoff’s ‘Conceptual Semantics’
“A dog is a reptile”  [State Is-included in ( [Thing Type: Dog ] ) ( [Thing Type: Reptile ] ) ]
(Jackendoff 1983: 96)
1
1
Fall 2013 Ling 720
(4) Early and Perennial Criticism of Such Approaches
Given the fundamental fact in (1), simply translating a sentence of English to a sentence
of ‘mentalese’ isn’t (necessarily) providing a semantics for the English sentence.
• The problem of providing a semantics for English now becomes the problem of
providing a semantics for the ‘mentalese’ notation (one that had never been taken up)
(5) Question (Not Necessarily Montague’s)

Under what conditions can this problem in (4) be circumvented? Again, under what
conditions (if any) can we guarantee that translating from one language L into another
language (notation) L’ gives us a compositional semantics for L.
2. Key Ingredient: First Order Logic as a Family of Languages
Up to now, we’ve been using the term “First Order Logic” to refer to a single language…
• However, at this point, it will be important to view First Order Logic not as a single
language, but rather as a family of infinitely many different languages…
First Order Logic A: ∃x( (Px) & ((Qa)b) )

First Order Logic B: ∃x( (dog’ x) & ((loves’ bill’ ) mary’) )
First Order Logic C: ∃x( (  ) & ((  )  )
(6) First Order Language (Logic)
A first order language (first order logic) is a language whose vocabulary of symbols
satisfies the conditions in (6a) and whose WFFs satisfy the conditions in (6b).
a. The Vocabulary of a First Order Language (Logic):

(i) The Logical Constants: ~, &, ∀, (∨, , ∃ are ‘abbreviations’)
(ii) Syntactic Symbols: (,)
(iii) The Non-Logical Constants:
1. A countable set of predicate letters (with associated arities)
2. A countable set of individual constants
3. A countably infinite set of variables {x, y, z, …, x1, x2, x3, … }
b. The Well-Formed Formulas of a First Order Language (Logic):

(i) If ϕ is an n-ary predicate letter and each of α1, …, αn is either an
individual constant or a variable, then Concat(…(Concat(Concat(ϕ,α1),
α2), …, αn) ∈ WFF
(ii) If ϕ, ψ ∈ WFF, then ~ϕ ∈ WFF and (ϕ & ψ) ∈ WFF
(iii) If ϕ ∈ WFF and υ is a variable, then ∀υϕ ∈ WFF
2
Fall 2013 Ling 720
(7) Remarks
a. Note that a first order language only needs a countable set of predicate letters and
individual constants. Thus, in a first order language, those sets can be finite.
b. Note that we are requiring a first order language to use the vocabulary in (6ai, ii),
the (infinite) variables in (iii), and the exact syntax rules in (6b).
 Thus, a logic in ‘Polish notation’ wouldn’t be a FOL according to (6); nor

would one making use of the alternate symbols ‘¬’ and ‘∧’
 The reason for this is simply because we need to keep some things constant
between FOLs; it works for us right now to keep these constant.
c. The definition in (6) defines an infinite set of different languages/logics.
 We can think of the general term ‘First Order Logic’ as referring to this
infinite set of different languages.
(8) Illustration: The First Order Language ‘Politics’
The language ‘Politics’ is the first order language whose vocabulary is as in (8a) and
whose WFFs are defined in (8b).
a. The Vocabulary of ‘Politics’

(i) The Logical Constants: ~, &, ∀,
1. Predicate Letters:
 Unary Predicate Letters: { smokes’ }
 Binary Predicate Letters: { loves’ }
2. Individual Constants: { michelle’, barack’, mitt’ }
3. Variables: {x, y, z, …, x1, x2, x3, … }
b. The WFFs of ‘Politics’ (exactly as in (6b))
(9) Some Illustrative Formulae of Politics Notice how the vocabulary in our
logical language is boldfaced, and
a. ~(smokes’ barack’) followed by primes?...
b. ( (loves’ barack’) michelle’ ) Get used to that!...
c. ~( (smokes barack’) & (smokes’ mitt’) )
d. ∀x ~( (smokes’ x) & ~~((loves’ mitt’) x))
3
Fall 2013 Ling 720
3. Introducing the Central Characters
(10) A Roadmap of Where We’re Headed
We’re going to build up Montague’s theory of translation by showing how a fragment of

English can be ‘translated’ (formally) into a fragment of Politics, and then showing how
that (formal) translation also gives us a semantics for the fragment of English.
a. Step One: Build the relevant fragment of Politics
b. Step Two: Try to build the relevant fragment of English
c. Step Three: NOTICE A FUNDAMENTAL PROBLEM IN STEP TWO
d. Step Four: Fix that problem, leading to a further refinement of our definition
of what a language is…
As we’ve done before, we’re going to make our lives easier by putting aside quantification for
the moment…
(11) A Useful Fragment of Politics: Politics-NoQ

The language ‘Politics-NoQ’ is the language whose vocabulary is as in (11a) and whose
WFFs are defined in (11b).
a. The Vocabulary of ‘Politics-NoQ’

(i) The Logical Constants: ~, &
1. Predicate Letters:
 Unary Predicate Letters: { smokes’ }
 Binary Predicate Letters: { loves’ }
2. Individual Constants: { michelle’, barack’, mitt’ }
b. The WFFs of ‘Politics-NoQ’

(i) If ϕ is an n-ary predicate letter and each of α1, …, αn is an individual
constant, then Concat(…(Concat(Concat(ϕ,α1), α2), …, αn) ∈ WFF
(ii) If ϕ, ψ ∈ WFF, then ~ϕ ∈ WFF and (ϕ & ψ) ∈ WFF
(12) Some Illustrative Formulae of Politics-NoQ
a. ~(smokes’ barack’)
b. ( (loves’ barack’) michelle’ )
4
Fall 2013 Ling 720
(13) Remarks
 Politics-NoQ is not a ‘first order language’, as defined in (6).
 In terms of its structure, Politics-NoQ is quite similar to our language FOL-NoQ from
the last two handouts.
 Consequently, we can easily see how to characterize Politics-NoQ in terms of our
general (Montagovian) definition of a language.
(14) The Language Politics-NoQ (Montagovian Presentation)

The language Politics-NoQ is the structure < A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T
where:
a. < A, Fγ >γ ∈ {Concat, Not, And} is the algebra such that

(i) FConcat , FNot , FAnd are as defined previously.
(ii) A is the smallest set such that:
1. { smokes’, loves’, michelle’, barack’, mitt’ } ⊆ A
2. It is closed under FConcat, FNot, FAnd
b. The basic categories Xτ are such that:

(i) Xe = { michelle’, barack’, mitt’ }
(ii) X<e,t> = { smokes’ }
(iii) X<e,<e,t>> = { loves’ }
(iii) For all other types τ ∈ T, Xτ = ∅.
c. The set S is the following (infinite) set of syntactic rules:

(i) < FNot , < t >, t>
(ii) < FAnd , <t, t>, t>
(iii) < FConcat , < <σ, τ>, σ >, τ >, for all σ, τ ∈ T
(15) Some Illustrative Members of Category Ct of Politics-NoQ
a. ~(smokes’ barack’)
b. ( (loves’ barack’) michelle’ )
Given this structural similarity between Politics-NoQ and FOL-NoQ, it’s also rather easy to set
up a (Fregan) interpretation for Politics-NoQ!
5
Fall 2013 Ling 720
(16) A (Fregean) Interpretation of Politics-NoQ
Let the set S = { Michelle, Barack, Mitt }. Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be the
Fregean interpretation based on S, such that f consists of the following mappings:
a. f(michelle’) = Michelle
b. f(barack’) = Barack
c. f(mitt’) = Mitt
d. f(smokes’) = h = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
e. f(loves’)= j = Michelle  Michelle  1

Barack  1
Mitt  0
Barack  1
Mitt  0
Barack  0
Mitt  1
Note: We’ve basically interpreted ‘smokes’’ as the property ‘smokes’, and we’ve
interpreted ‘loves’’ as the (curried) relation ‘x is loved by y’.2
(17) Using the Fregean Interpretation in (16) to Interpret Sentences of Politics-NoQ
(i) g( ~(smokes’ barack’ ) ) = (by definition of Politics-NoQ)
(ii) g( FNot( FConcat(smokes’, barack’)) ) = (by homomorphism property of g)
(iii) GNot( g(FConcat(smokes’ , barack’)) ) = (by homomorphism property of g)
(iv) GNot( GConcat( g(smokes’), g(barack’) )) = (by definition of g)
(v) GNot( GConcat( f(smokes’), f(barack’) )) = (by definition of f in (16))
(vi) GNot( GConcat( h, Barack )) = (by definition of GConcat)
(vii) GNot( h(Barack) ) = (by definition of h in (16d))
(viii) GNot( 1 ) = (by definition of GNot)
(ix) 0
2
That is, in the notation of Heim & Kratzer (1998), we’re interpreting loves’ as [λy: [λx: x loves y]]. The
semanticists in the house can probably guess why we’re doing this ; )
6
Fall 2013 Ling 720
Finally, to get our third player on the field, let’s recall that fragment of English that we defined a
while back…
(18) The Definition of ‘Mini-English’

‘Mini-English’ is the structure < E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And}, δ ∈ Δ such that:
a. The Syntactic Categories: Δ = {NP, IV, TV, S}
b. The Syntactic Operations:

(i) KConcat = Merge (from previous notes)
(ii) KNot = NotE (from previous notes)
(iii) KAnd = AndE (from previous notes)
c. The Basic Expressions:

(i) XNP = { Barack, Michelle, Mitt }
(ii) XIV = { smokes }
(iii) XTV = { loves }
(iv) XS = ∅
c. The Syntactic Algebra:

E is the smallest set such that:
(i) For all δ ∈ Δ, Xδ ⊆ E.
(ii) E is closed under the operations KConcat, KNot and KAnd
d. The Syntactic Rules: The set SE consists of the following tuples:

(i) < KConcat, < TV, NP >, IV >
(ii) < KConcat, < NP, IV >, S >
(iii) < KAnd , <S , S> , S >
(iv) < KNot, <S>, S >
(19) Illustrative Sentence (Expression of Category CS) of Mini-English
It is not the case that Barack smokes and Mitt smokes.
< It is not the case that Barack smokes and Mitt smokes, KAnd >
< It is not the case that Barack smokes, KNot > < Mitt smokes, KConcat >
< Barack smokes, KConcat > Mitt smokes
Barack smokes
7
Fall 2013 Ling 720
Unfortunately, there’s a fundamental problem with the language as defined in (18). To see this,
recall our ultimate goal, informally sketched out below.
(20) Our Goal for a Theory of Translation

We want to develop a way of homomorphically mapping expressions of mini-English to
expressions of Politics-NoQ (so that we can ultimately get a semantics for English)
< E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And}, δ ∈ Δ
h (translation from mini-Eng to Pol-NoQ)

g°h (the semantic
interpretation of mini-Eng)
< A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T
g (semantic interpretation of Politics-NoQ)
<B, Gγ , f>γ ∈ {Concat, Not, And}
(21) Critical Problem

We currently conceive of the set E (expressions of mini-English) as consisting of strings
of English words.
 However, some such strings in our mini-English language can be created in more
than one way from the syntactic operations (and rules) of our language.
Example: < It is not the case that Barack smokes and Mitt smokes, KNot >
< Barack smokes and Mitt smokes, KAnd >
< Barack smokes, KConcat > < Mitt smokes, KConcat >
Barack smokes Mitt smokes
 Intuitively, we want these two different ways of constructing the mini-English

sentence to lead to two different Politics-NoQ translations:
( ~(smokes’ barack’) & (smokes’ mitt’) )
~( (smokes’ barack’) & (smokes’ mitt’) )
 But if the translation h is a mapping from strings of English to expressions of

Politics-NoQ, each such string will be mapped to only one translation!
8
Fall 2013 Ling 720
(22) Some More General Remarks
 Ultimately, we want there to be two different Politics-NoQ translations of sentence

(19) because we want this string to be paired with two different semantic values.
o There is a reading where (19) is true

≈ ~( (smokes’ barack’) & (smokes’ mitt’) )
o There is a reading where (19) is false

≈ ( ~(smokes’ barack’) & (smokes’ mitt’) )
 Also, we have the background belief that (19) has these two readings because of the
different ways that the sentence can be constructed in English
o (i.e., it’s not because of any ambiguity in what the words mean…)
 But, if we are semantically interpreting strings of English words – and ‘interpretation’

is conceived us as a homomorphism (function) from expressions to meanings – then
each string will be mapped to exactly one meaning.
 Thus, unlike with Politics-NoQ, it is not feasible to build a semantics that interprets
(directly or indirectly) English strings.
 In LING 610, this problem doesn’t even arise, because right from the start we’re
interpreting phrase structure trees
o After all, a given tree is only ever constructed in one way by Merge and
Move… hmm…
(23) Another Critical Problem

In the picture in (20), translation is a homomorphism from < E, Kγ >γ ∈ {Concat, Not, And} for
mini-English to the algebra < A , Fγ >γ ∈ {Concat, Not, And} for Politics-NoQ
 Under such a homomorphism, we’d naturally want KConcat and FConcat to correspond.
This will get the right interpretation for VPs (IVs) after all:
h( loves michelle ) = h ( KConcat ( loves, Michelle )) =

FConcat( h(loves), h(Michelle) = FConcat( loves’, michelle’ ) = (loves’ michelle’)
 However, this will get the wrong result for sentences! Sentences will end up mapped
to syntactic garbage in A.
h( Barack smokes ) = h ( KConcat ( Barack, smokes )) =

FConcat( h(Barack), h(smokes) = FConcat( barack’, smokes’ ) = (barack’ smokes’)
Syntactic Garbage!!!
9
Fall 2013 Ling 720
(24) More General Remarks
 Obviously, what we want is for h(Barack smokes) = (smokes’ barack’)
 But, there is no operation in the algebra for Politics-NoQ which will take as argument
the translation of Barack (barack’) and the translation of smokes (smokes’) and
return the formula (smokes’ barack’)
 So maybe our homomorphic translation function h shouldn’t actually be a

homomorphism to the algebra < A , Fγ >γ ∈ {Concat, Not, And}
o Maybe it should be a homomorphism to some other algebra that we can
construct from < A , Fγ >γ ∈ {Concat, Not, And}
In these notes, we’ll deal only with the problem in (21)-(22)…

In the next set of notes, we’ll tackle the problem in (23)-(24)…
4. Montague’s Notion of a ‘Disambiguated Language’
(25) What We Want
 We want it to be that the interpreted expressions of our language can only ever be
created from the syntactic operations (rules) in exactly one way.
 This way, we won’t ever have to worry about interpreting ‘syntactically ambiguous’
expressions (because they just won’t exist in our language).
 Preview of Where This is Going:

We’ll relate such ‘syntactically unambiguous’ expressions to sentence strings of
English via a special operation (akin to ‘linearization’ or ‘Spell Out’).
(26) Question
Below we have our earlier (Montagovian) definition of a language. What do we have to
add to this to ensure that no expressions are syntactically ambiguous?

(i) For all δ ∈ Δ, Xδ ⊆ A; (ii) A is closed under the operations Fγ for all γ ∈ Γ
b. δ0 ∈ Δ
10
Fall 2013 Ling 720
(27) Montague’s Answer
At a minimum, we need to ensure that:
a. Nothing in the basic expressions Xδ (lexical items) can also be constructed by the
syntactic operations.
That is: Xδ and the range of Fγ are disjoint for all δ ∈ Δ and γ ∈ Γ
b. No element of A will be the output of two different operations Fγ and Fγ’

c. No single operation Fγ will take two different inputs a, a’ ∈ A and give the same
output.
That is: For all sequences a1, …, an ∈ An and a’1, … a’m ∈ Am, if Fγ(a1, …, an) =
Fγ’(a’1, … a’m), then Fγ = Fγ’ and <a1, …, an> = <a’1, … a’m>.
Note that if the conditions in (27) hold, then every expression in A will either be (i) a basic
expression (lexical item), or (ii) constructible in exactly one way from the syntactic operations.
(28) Montagovian Definition of a ‘Diambiguated Language’

The following definition now replaces our earlier concept of a language, as well as its
concomitant definition.
A disambiguated language is a structure < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ such that:

(i) For all δ ∈ Δ, Xδ ⊆ A; (ii) A is closed under the operations Fγ for all γ ∈ Γ
c. Xδ and the range of Fγ are disjoint for all δ ∈ Δ and γ ∈ Γ
d. For all sequences a1, …, an ∈ An and a’1, … a’m ∈ Am, if Fγ (a1, …, an) =
Fγ ’(a’1, … a’m), then Fγ = Fγ ’ and a1, …, an = a’1, … a’m.
e. S is a set of sequences of the form < Fγ , < δ1, …, δn >, δ >, where γ ∈ Γ, Fγ is an
f. δ0 ∈ Δ
11
Fall 2013 Ling 720
(29) Remarks
a. Our language Politics-NoQ is such a disambiguaged language.
b. Our language ‘mini-English’ is not a disambiguated language.
c. Potential Problem:
If we assume that the expressions of mini-English (and English) are strings, then
we just aren’t going to be able to represent those systems as disambiguated
languages.
d. Solution:
Along with the concept of a ‘disambiguated language’ in (28), we need a more
general concept of a ‘language’.
(30) Montagovian Definition of a Language (Final Version)

A language is a pair < L, R >, where L = < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ is a
disambiguated language, and R is a binary relation whose domain is A.
 This relation R is an ‘ambiguating relation’.

o It maps expressions in the disambiguated language L to expressions not
necessarily from L.
 Importantly, R can be many-to-one (surjection), and so we can have more than one
expression from the disambiguated language being mapped to the same expression in
the range of R (hence, the term ‘ambiguating’)
(31) Illustration: Politics-NoQ-SansParens
a. Informal Definition
(i) Vocabulary: Same as Politics-NoQ
(ii) The WFFs of ‘Politics-NoQ-SansParens’
1. If ϕ is an n-ary predicate letter and each of α1, …, αn is either an
individual constant or a variable, then ϕα1…, αn ∈ WFF
2. If ϕ, ψ ∈ WFF, then ~ϕ ∈ WFF and ϕ & ψ ∈ WFF
Illustrative Formulae: ~ smokes’ barack’

loves’ barack’ michelle’
~ smokes barack’ & smokes’ mitt’
b. Formal Definition
The pair < < A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T, R >, where the structure
<A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T is Politics-NoQ, and R is the function
that takes any element of A and deletes every parenthesis.
R( ~( (smokes barack’) & (smokes’ mitt’) ) ) =

~ smokes barack’ & smokes’ mitt’
12
Fall 2013 Ling 720
(32) Remark
Every disambiguated language < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ can also be represented as
a language < < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, R>
 Simply let R be the identity function!!
(33) Some Concomitant Definitions

Let L be a language < < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, R>.
a. The proper expressions of L is the range of R.

b. The operation indices of L is Γ.
c. The category labels of L is Δ
d. The syntactic rules of L is S.
e. The basic expressions of L of category δ is { ϕ : ∃ψ ∈ Xδ such that ψRϕ }
f. The category δ of L is { ϕ : ∃ψ ∈ Cδ such that ψRϕ }, where Cδ is in the family

of categories generated by < A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ
g. The meaningful expressions of L is the union of all the categories δ of L.
h. The declarative sentences of L is the category δ0 of L
i. If ϕ is a meaningful expression of L, then ϕ is syntactically ambiguous if there are

distinct ψ, ψ’ ∈ ∪δ∈ΔCδ such that ψRϕ and ψ’Rϕ
j. The language L is syntactically ambiguous if there is a meaningful expression ϕ

of L which is syntactically ambiguous.
(34) New Goal

Given all that we’ve laid out, it seems that we now want to do the following:
a. Represent mini-English as a (syntactically ambiguous) language < L, R>, where

(i) L is some syntactically unambiguous language, and
(ii) R can ‘transform’ expressions of L into expressions of mini-English.
b. Translate mini-English into Politics-NoQ indirectly, via translation from L into

Politics-NoQ.
Mini-English ‘Disambiguated’ L Politics-NoQ

Ambiguating ‘R’ homomorphic translation
13
Fall 2013 Ling 720
5. Representing Mini-English Via a Disambiguated Language
(35) Key Question

Given our new goal in (34), what should the expressions of our ‘disambiguated mini-
English’ look like?
 Well, each complex expression must transparently reflect how it was constructed by
the syntactic operations…
 That is, for each complex expression, there should be exactly one analysis tree…
(36) Montague’s Core Insight
For a syntactically ambiguous natural language like English, we could assume that the
syntactically disambiguated expressions are the analysis trees themselves!!
 That is, (mini-)English is a pair < L, R>, where the expressions of L are analysis trees,
and the relation R simply maps an analysis tree to the string in its root node!
< Barack smokes , KConcat >

Barack smokes.
Barack smokes Ambiguating R
(37) Remarks
We’ll see in a moment how to actually implement the idea in (36). For the moment, let’s
notice the similarities and differences between this and an ‘LF’-based semantics.
a. Key Similarity:
Our semantics does not directly interpret surface strings of English. Rather, it
interprets abstract structures that represent how those strings can be derived.
b. Key Difference:
Unlike an ‘LF’-based semantics (like in 610), our system doesn’t first construct
the analysis tree (LF structure) for a whole sentence and then ‘input’ that into
semantic interpretation…
o That is, as will be clear in a few more classes, the syntax and semantics work
in tandem with one another…
 Every time a ‘move’ is made in the syntax to make a structure, a

corresponding ‘move’ is made in the semantics to determine a
meaning for that structure…
But how do we construct a ‘disambiguated language’ where the expressions are analysis trees?
14
Fall 2013 Ling 720
(38) Step One: The Category Labels
The syntactic categories of Disambiguated mini-English will be just the same as before:
Δ = {NP, IV, TV, S}
(39) The Basic Expressions

The basic expressions of Disambiguated mini-English will be ‘trivial trees’. The
following trees consisting of root-nodes without any daughters.
a. XNP = { < Barack, ∅>, < Michelle, ∅>, < Mitt, ∅> }
b. XIV = { < smokes, ∅> }
c. XTV = { < loves, ∅> }
d. XS = ∅

Our syntactic operations now take trees (including ‘trivial trees’) as input and output
other trees, as defined below.
 In the definitions below, α and β are trees whose root nodes are ordered pairs. In
addition α’ and β’ are the first members of the root nodes of α and β (respectively).
a. KConcat(α, β) = < α’ β’ , Concat >
α β
b. KNot(α) = < it is not the case that α’ , Not >
c. KAnd(α, β) = < α’ and β’ , And >
α β
Just for fun – since it will set us up for something important later, let’s also add the
following syntactic operation.
d. KIf(α, β) = < If α’ then β’ , If > Note:

The right-hand
α β member of a
node is now
(41) The Syntactic Algebra the index of
E is the smallest set such that: the operation,
a. For all δ ∈ Δ, Xδ ⊆ E. rather than the
b. E is closed under the operations KConcat, KNot, KAnd, and KIf operation
itself…
15
Fall 2013 Ling 720
We can retain much the same set of syntactic rules SE that we had before:
a. < KConcat, < TV, NP >, IV >

b. < KConcat, < NP, IV >, S >
c. < KAnd , <S , S> , S >
d. < KIf , <S , S> , S >
e. < KNot, <S>, S >
(43) The Definition of Our Language: ‘Disambiguated Mini-English’

The structure < E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And, If}, δ ∈ Δ where E, Kγ, Xδ , SE, and Δ
are as defined in (38)-(42).
Some Illustrative Members of Category CS
a. < It is not the case that Barack smokes and Mitt smokes, Not >
< Barack smokes and Mitt smokes, And >
< Barack smokes, Concat > < Mitt smokes, Concat >
< Barack , ∅ > < smokes , ∅ > < Mitt, ∅> < smokes, ∅ >
b. < It is not the case that Barack smokes and Mitt smokes, And >
< It is not the case that Barack smokes, Not > < Mitt smokes, Concat >
< Barack smokes, Concat > < Mitt, ∅> < smokes, ∅ >
< Barack , ∅ > < smokes , ∅ >
c. < Barack loves Michelle, Concat >
< Barack , ∅ > < loves Michelle, Concat >
< loves ∅ > < Michelle, ∅ >
(44) Remark Disambiguated Mini-English is indeed a disambiguated language.

 No syntactic operation will ever create a basic expression.
 Because of the way the trees are indexed, no two ops will ever have the same output
16
Fall 2013 Ling 720
Now, we can use Disambiguated Mini-English to characterize Mini-English as a language, in
the sense of (30)
(45) Montagovian Definition of Mini-English

Mini-English is the structure < < E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And, If}, δ ∈ Δ, R>, where
a. The structure < E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And, If}, δ ∈ Δ is Disambiguated
Mini-English, as defined in (43).
b. R is a function which takes as input a tree T in E, and returns as output the first
member of the root node of T.
< Barack loves Michelle, Concat >
R < Barack , ∅ > < loves Michelle, Concat > =
Barack loves Michelle
(46) Some Illustrative Members of the Category S for Mini-English

a. Barack smokes.
b. Barack loves Michelle.
c. It is not the case that Barack smokes and Mitt smokes.
(47) Remark Mini-English is a syntactically ambiguous language (33j)

 After all, let T be the tree in (43a), and T’ be the tree in (43b).
 R(43a) = R(43b) = It is not the case that Barack smokes and Mitt smokes.
What Coming Up Next:
 We now have the following two disambiguated languages:

o Politics-NoQ
o Disambiguated Mini-English
 We have an interpretation for Politics-NoQ
 We’re now going to try to find a way of homomorphically mapping expressions of

Disambiguated Mini-English to ones of Politics-NoQ
17
Fall 2013 Ling 720
Montague’s Theory of Translation: The Notion of a ‘Translation Base’ 1
1. Review of What We Have and Where We Want to Go
(1) Our Key Ingredients
a. Two Disambiguated Languages:

Both the languages below are such that every expression in the language is either
(i) a basic expression, or (ii) formed in exactly one way from the syntactic
operations, but (iii) not both.
(i) Politics-NoQ
The structure < A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T where the
algebra < A, Fγ >γ ∈ {Concat, Not, And}, and the sets Xτ and S are as before.
(ii) Disambiguated Mini-English (DME)

The structure The structure < E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And, If}, δ ∈ Δ,
where algebra <E, Kγ>γ ∈ {Concat, Not, And, If}, and the sets Xδ , SE as before.
b. An Interpretation for Politics-NoQ

Let the set S = { Michelle, Barack, Mitt }. Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be
the Fregean interpretation based on S, such that f is as defined before.
(2) What We Want to Make from Those Ingredients

We want to develop a way of homomorphically mapping expressions of DME to
< E, Kγ, Xδ , SE, S >γ ∈ {Concat, Not, And, If}, δ ∈ Δ
h (translation from DME to Pol-NoQ)

g°h (the semantic
interpretation of DME)
1
1
Fall 2013 Ling 720
Although we’ve come closer to our goal by restricting our attention to ‘disambiguated
languages’, there are still two key problems facing our project…
(3) Critical Problem 1:
• If translation h is to be a homomorphism from DME to a syntactic algebra for P-NoQ,

then there must be a syntactic operation OP in the latter that ‘corresponds’ to KConcat.
• Moreover, under this correspondence, it must be that:
h( KConcat.( Barack, smokes )) =

OP ( h(Barack), h(smokes)) =
OP ( barack’ , smokes’) = ( smokes’ barack’ )
h( KConcat.( loves, Barack )) =

OP ( h(loves), h(Barack)) =
OP ( loves’ , Barack’) = ( loves’ barack’ )
• But this seems inconsistent! How can OP(barack’ , smokes’) = (smokes’ barack’),
while OP(loves’, barack’) = (loves’ barack’)???
(4) Critical Problem 2 (New):
• Our syntactic algebra for DME contains the operation KIf.
• Again, if h is to be a homomorphism from DME to a syntactic algebra for P-NoQ,

there must be some syntactic operation in the latter that ‘corresponds’ to KIf
• But there isn’t any!
The Plan:
We’ll go halfway to fixing the problem in (3); at which point, the problems in (3) and (4) will
become the same. Then we’ll solve that more general problem by introducing a new, central idea
of Montague’s: the Translation Base.
2. Prolegomena: A Slight Change to Our Definition of Disambiguated Mini-English
We’re going to introduce a slight change to our definition of DME…

It will seem ad hoc for now, but we’ll see independent motivation later on (with quantification)…
(5) Step One: The Category Labels

The syntactic categories of Disambiguated Mini-English will be just the same as before:
Δ = {NP, IV, TV, S}
2
Fall 2013 Ling 720
(6) The Basic Expressions
The basic expressions of DME will be just the same as before, too.
a. XNP = { < Barack, ∅>, < Michelle, ∅>, < Mitt, ∅> }
b. XIV = { < smokes, ∅> }
c. XTV = { < loves, ∅> }
d. XS = ∅

Our set of syntactic operations for DME is going to be altered. We’re going to split
KConcat into two different operations: KMerge-S and KMerge-IV.
 In the definitions below, α and β are trees whose root nodes are ordered pairs. In
addition α’ and β’ are the first members of the root nodes of α and β (respectively).
a. KMerge-S(α, β) = < α’ β’ , Merge-S >
α β
b. KMerge-IV(α, β) = < α’ β’ , Merge-IV >
α β
c. KNot(α) = < it is not the case that α’ , Not >
d. KAnd(α, β) = < α’ and β’ , And >
α β
e. KIf(α, β) = < If α’ then β’ , If >
α β
Right now, the only difference between KMerge-S and KMerge-IV is the index on the root of the
output. Again, later on these operations will become more substantively different.
(8) The Syntactic Algebra

E is the smallest set such that:
a. For all δ ∈ Δ, Xδ ⊆ E.
b. E is closed under the operations KMerge-S, KMerge-IV, KNot, KAnd, and KIf
3
Fall 2013 Ling 720
With the changes to our syntactic operations in (7) come some concomitant changes to our
syntactic rules…

We can retain much the same set of syntactic rules SE that we had before:
a. < KMerge-IV, < TV, NP >, IV >

b. < KMerge-S, < NP, IV >, S >
c. < KAnd , <S , S> , S >
d. < KIf , <S , S> , S >
e. < KNot, <S>, S >
(10) New Definition for Disambiguated Mini-English

The structure < E, Kγ, Xδ , SE, S >γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ where E, Kγ, Xδ , SE,
and Δ are as defined in (38)-(42).
Illustrative Structure: < Barack loves Michelle, Merge-S >
< Barack , ∅ > < loves Michelle, Merge-IV >
(11) Half of Problem (3) Solved

Now we don’t need to find a single operation in Politics-NoQ that corresponds to KConcat
• In particular, we can now assume without problem that KMerge-IV corresponds to FConcat
h( KMerge-IV( loves, Michelle )) =
FConcat ( h(loves), h(Michelle)) =
FConcat ( loves’ , michelle’) = ( loves’ michelle’ )
(12) Remaining Problem
• Now, however, we have this additional operation KMerge-S. And, there doesn’t seem
to be an operation in Politics-NoQ that corresponds to it.
• And, we still need to find an operation in Politics-NoQ that corresponds to KIf !
(13) Key Idea Behind the Solution to (12)

Because of KIf and KMerge-S , there won’t be a homomorphism between the syntactic
algebras < E, Kγ>γ ∈ {Merge-S, Merge-IV, Not, And, If} and < A , Fγ>γ ∈ {Concat, Not, And}
• But maybe we can make a new algebra from < A , Fγ >γ ∈ {Concat, Not, And} that will
fit the bill!!!
4
Fall 2013 Ling 720
3. Polynomial Operations and Derived Syntactic Rules
(14) Road Map for This Section

In this section, we will see how to do the following:
a. From an algebra <A, Fγ>γ ∈ Γ, create new complex operations from the operations
{ Fγ }γ ∈ Γ.
b. From a (disambiguated) language <A, Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, create new

complex syntactic rules from the rules S.
This will provide us with the key tools for introducing Montague’s notion of a ‘translation base’.
Advisory:
In my own view, the technical concept of a ‘translation base’ is the least a priori intuitive
ingredient of Montague’s theory…
• Once you see what the thing does – and how it does what it does – it becomes easier
to get your mind around…
• But, it’s hard to build up piece-by-piece in a completely intuitive way…
• Thus, please have faith that we’re going somewhere interesting with all this…
3.1 Polynomial Operations Over an Algebra
In this section, we’ll cover the goal in (14a). The principle means of creating a complex
operation from simpler operations is function composition (Handout 1), repeated below.
(15) (Generalized) Function Composition

Let g be an n-ary function, and let f1, …, fn be a series of n m-ary functions. The
composition of g and f1, …, fn is the m-ary function defined as follows:
g<f1, …, fn> =def the m-ary function such that for any m-ary sequence a1, … am
g<f1, …, fn>(<a1, … am>) =
g( f1(<a1, … am>), …, fn(<a1, … am>))
Illustration:
Let g = { <<x,y>,z> : z = x + y }, f = { <x,y> : y = x – 1 }, h = { <x,y> : y = x + 2 }
Then: g<f,h>(2) = g(f(2),h(2)) = g(1,4) = 5
g<f,h> = { <x,y> : y = (x-1) + (x+2) }
5
Fall 2013 Ling 720
In addition to function composition, we’ll also make use of the special functions in (16) and (17).
(16) Identity Functions

The function Idn,m takes as argument an m-tuple α and returns the nth member of α
a. Id1,1(a) = a
Id1,1(b) = b
b. Id1,2(a,b) = a Id1,2(c,d) = c
Id2,2(a,b) = b Id2,2(c,d) = d
c. Id1,3(a,b,c) = a Id1,3(c,d,e) = c
Id2,3(a,b,c) = b Id2,3(c,d,e) = d
Id3,3(a,b,c) = c Id3,3(c,d,e) = e
(17) Constant Functions

The function Cα, m takes as argument an m-tuple β, and for any such m-tuple β, returns α
a. Ca,1(a) = a Cb,1(a) = b
Ca,1(b) = a Cb,1(b) = b
b. Ca,2(a,b) = a Cb,2(a,b) = b
Ca,2(c,d) = a Cb,2(c,d) = b
c. Ca,3(a,b,c) = a Cb,3(a,b,c) = b
Ca,3(c,d,e) = a Cb,3(c,d,e) = b
With these ingredients in place, we can introduce the key concept in (18).
(18) The Polynominal Operations Over an Algebra

Let A = <A, Fγ>γ ∈ Γ be an algebra. The class of polynomial operations over A is the
smallest class K such that the following all hold.
a. Fγ ∈ K for all γ ∈ Γ
Note: The polynomial operations over A include all the operations in A
b. Idn,m ∈ K, for all n, m ∈ ℕ

Note: Every possible identity function is also a polynomial operation over A.
c. Ca,m ∈ K, for all a ∈ A and m ∈ ℕ

Note: Every possible constant function (to A) is also a polynomial op. over A.
d. If G is an n-ary function in K, and F1, … Fn are n m-ary functions in K, then

G<F1, … Fn> ∈ K
Note: The polynomial operations over A are closed under function composition.
6
Fall 2013 Ling 720
(19) Remark
So, in other words, F is a polynomial operation over A (= <A, Fγ>γ ∈ Γ) if any of the
following hold:
a. F is one of the operations { Fγ }γ ∈ Γ.
b. F is an identity function.
c. F is a constant function (to A).
d. You can obtain F via iterated function composition from either (a), (b), or (c).
We’ll now illustrate the key concept in (18) by looking at some polynomial operations over the
syntactic algebra for Politics-NoQ, <A, Fγ>γ ∈ {Concat, Not, And}.
(20) C(smokes’ barack’), 1

This function takes any expression in A and returns the expression (smokes’ barack’)
C(smokes’ barack’), 1 ( (loves’ michelle’) ) = (smokes’ barack’)

C(smokes’ barack’), 1 ( mitt’ ) = (smokes’ barack’)
(21) Id2,3 This function takes any triple in A and returns the second member.
(22) Id1.2 This function takes any pair in A and returns the first member.
(23) Id2.2 This function takes any pair in A and returns the second member.
(24) FNot<FNot>
This function takes any expression in A and returns its double negation.
FNot<FNot>( (smokes’ barack’) ) =

FNot(FNot( (smokes’ barack’) ) ) =
~~(smokes’ barack’)
(25) FAnd< C(smokes’ barack’), 1, Id1,1 >

This function takes any expression α in A and returns the conjunction of (smokes’
barack’) with α.
FAnd< C(smokes’ barack’), 1, Id1,1 >( (smokes’ mitt’) ) =

FAnd( C(smokes’ barack’), 1((smokes’ mitt’)), Id1,1((smokes’ mitt’)) ) =
FAnd( (smokes’ barack’)), (smokes’ mitt’) ) =
( (smokes’ barack’) & (smokes’ mitt’) )
7
Fall 2013 Ling 720
(26) FNot<FAnd<Id1,2, FNot<Id2,2>>>
This function takes any pair of expressions α, β in A and returns ~(α & ~β).
FNot<FAnd<Id1,2, FNot<Id2,2>>>( (smokes’ barack’), (smokes’ mitt’) ) =
FNot( FAnd( Id1,2( (smokes’ barack’), (smokes’ mitt’) )

FNot( Id2,2( (smokes’ barack’), (smokes’ mitt’) ) ) … ) =
FNot( FAnd( (smokes’ barack’), FNot( (smokes’ mitt’) ) … ) =
FNot( FAnd( (smokes’ barack’), ~(smokes’ mitt’) ) … ) =
FNot( ( (smokes’ barack’) & ~(smokes’ mitt’) ) ) =
~( (smokes’ barack’) & ~(smokes’ mitt’) )

Recall that the formulae (ϕ  ψ) and ~(ϕ & ~ψ) are logically equivalent.
• Consequently, we are viewing (ϕ  ψ) as a special ‘abbreviation’ for ~(ϕ & ~ψ)
• Thus, the operation FNot<FAnd<Id1,2, FNot<Id2,2>>> would seem to be a good

‘translational correspondent’ of KIf in Disambiguated Mini-English
h( KIf (α, β) ) = FNot<FAnd<Id1,2, FNot<Id2,2>>>(h(α), h(β)) = ~(h(α) & ~h(β)) = (h(α)  h(β))
(28) FConcat<Id2,2, Id1,2>

This function takes any pair of expressions α, β in A and returns (β α)
FConcat<Id2,2, Id1,2>(barack’, smokes’) =
FConcat( Id2,2(barack’, smokes’) , Id1,2(barack’, smokes’) ) =
FConcat( smokes’ , barack’ ) =
( smokes’ barack’ )

It seems like FConcat<Id2,2, Id1,2> would be a good ‘translational correspondent’ of KMerge-S
in Disambiguated Mini-English.
h( KMerge-S (Barack, smokes)) = FConcat<Id2,2, Id1,2>(h(Barack), h(smokes)) =
FConcat<Id2,2, Id1,2>(barack’, smokes’) = (smokes’ barack’)
8
Fall 2013 Ling 720
(30) Summary Observation
• If we look to the polynomial operations over the syntactic algebra for Politics-NoQ,
<A, Fγ>γ ∈ {Concat, Not, And}, we will find syntactic operations over A that could viably
correspond to the syntactic operations KIf and KMerge-S over E
• How, though, does this help us in our quest for a homomorphism from E to A?...
(31) Polynomial Operations and Algebras
a. Key Fact:
Let A be an algebra <A, Fγ>γ ∈ Γ. If the set { Hγ’ }γ’ ∈ Γ’ consists of polynomial
operations over A, then A is closed under { Hγ’ }γ’ ∈ Γ’.
(proof left as an exercise to the student)
b. Key Consequence:
Let A be an algebra <A, Fγ>γ ∈ Γ, and let { Hγ’ }γ’ ∈ Γ’ consist of polynomial
operations over A. The structure <A, Hγ>γ’ ∈ Γ’, is an algebra.
Thus, given (31b), if <A, Fγ>γ ∈ {Concat, Not, And} is the syntactic algebra for Politics-NoQ, then the
following is also an algebra: <A, Hγ>γ ∈ {Merge-S, Merge-IV, Not, And, If}, where
• HNot and HAnd = FNot and FAnd, respectively
• HMerge-IV = FConcat
• HIf = FNot<FAnd<Id1,2, FNot<Id2,2>>>
• HMerge-S = FConcat<Id2,2, Id1,2>
And, it seems like it might be possible to have a homomorphism from the syntactic algebra
for DME <E, Kγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} to <A, Hγ >γ ∈ {Merge-S, Merge-IV, Not, And, If}
Before we can use all of this to lay out a theory of homomorphic translation between languages,
we also need to introduce a way of constructing complex syntactic rules (14b)…
9
Fall 2013 Ling 720
3.2 The Derived Syntactic Rules of a Language L
(32) Background Motivation
• Intuitively, a translation from one language L to another language L’ should always

map the well-formed expressions of L to well-formed expressions of L’
• As we’ll soon see, one way of ensuring this appeals to the notion of a ‘derived
syntactic rule’, defined in this section.
• For reasons that will also be clear shortly, this definition is going to closely
mirror the definition for the polynomial operations over an algebra
(33) Derived Syntactic Rules of a Language

Let L be a language << A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ , R>. The derived syntactic rules of
L is the smallest class K such that the following hold:
a. S⊆K
Note: Thus all the syntactic rules of L are also ‘derived syntactic rules’
b. For all n, m ∈ ℕsuch that n ≤ m, if <δ1, … , δn, … , δm> is an m-tuple of elements

from Δ, then < Idn,m , <δ1, … , δn, … , δm> , δn> ∈ K
Note:
This means that all the logically possible syntactic rules of the form below are
also ‘derived syntactic rules’:
< Id2,4 , < e, t, <e,<e,t>>, <e,t> >, t >

‘The result of applying Id2,4 to a quadruple consisting of an expression of type e,
one of type t, one type <e,<e,t>> and one of type <e,t> is an expression of type t.’
c. For all n ∈ ℕ, if a ∈ Cδ , and <δ1, …, δn> is an n-tuple of elements from Δ, then

the triple < Ca,n, <δ1, …, δn>, δ > ∈ K.
Note:
This means that all the logically possible syntactic rules of the form below are
also ‘derived syntactic rules’.
< C(smokes’ barack’), 2 , < <e,t>, <e,t> >, t >

‘The result of applying C(smokes’ barack’), 2 to a pair consisting of an expression of
type <e,t> and an expression of type <e,t> is an expression of type t.’
The fourth and final condition on the derived syntactic rules basically amounts to them being
closed under ‘composition’… it is rather complex to state formally, however…
10
Fall 2013 Ling 720
d. If <F , <δ1 , … δn>, δ> ∈ K, F is an n-ary operation, and each of G1, … Gn are an
m-ary operation such that <Gj , <δ’1 , … δ’m>, δj > ∈ K, then the following is also
a member of K:
<F<G1, … Gn>, <δ’1 , … δ’m>, δ>
Note: To get a sense of how this ‘composition’ operation on rules works, consider that (i) and
(ii) are rules in our language Politics-NoQ.
(i) < FNot , < t >, t >

(ii) < FConcat, < <e,t>, e >, t >
Thus, definition (33) would entail that the following is a ‘derived rule’ of Politics-NoQ.
(iii) < FNot<FConcat>, < <e,t>, e >, t >

‘The result of applying FNot<FConcat> to a pair consisting of an expression
of type <e,t> and an expression of type e is an expression of type t.’
Note, too, that this derived rule would intuitively ‘be true of’ for Politics-NoQ.
• This raises the following key generalization…
(34) Derived Syntactic Rules and the Syntactic Categories of a Language

Let L be a language <<A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, R>, and let <H, <δ1 , … δn>, δ> be
a derived syntactic rule of L.
a. Claim: If ϕ1, … ϕn are such that each ϕi ∈ Cδi , then H(ϕ1, … ϕn) ∈ Cδ
(That is, the derived syntactic rules will only ever generate ‘meaningful
expressions’ of a language.)
b. Proof: (left as an exercise to the student)
4. The Concept of a Translation Base
In the previous section, we developed the tools below. We also developed them so that they
mirror one another.
(35) Polynomial Operations (Over an Algebra)

A way of taking algebra <A, Fγ>γ ∈ Γ and creating new complex operations {Hγ’ }γ’ ∈ Γ’
from the operations { Fγ }γ ∈ Γ such that <A, Hγ’>γ’ ∈ Γ’ is also an algebra.
11
Fall 2013 Ling 720
(36) Derived Syntactic Rules
A way of taking a language <A, Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, and creating new complex
syntactic rules S’ from the polynomial operations over <A , Fγ>γ ∈ Γ.
We’re now going to use these tools to construct Montague’s general theory of translation…
To do this, let’s first consider some ideal properties of a translation function h from one language
<A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ to another language < A’ , F’γ’ , X’δ’ , S’, δ’0>γ’ ∈ Γ’ , δ’ ∈ Δ’
(37) Correspondence Between the Syntactic Categories

In Montague’s theory of translation, there must be a consistent mapping from the
syntactic categories of L to the syntactic categories of L’.
• That is, if δ ∈ Δ, then there must be a corresponding δ’ ∈ Δ’ such that if ϕ ∈ Cδ, then
h(ϕ) ∈ Cδ’
• For example, thinking of our languages DME and Politics-NoQ, such a mapping of
the categories would be as follows:
NP  e
TV  <e,<e,t>>
IV  <et>
St
• The reason why such a mapping is needed is ultimately tied to Montague’s (final)
definition of a ‘Fregean Interpretation’… Just go with it for now…
Consequence: A translation from L to L’ will need to specify a function g from Δ to Δ’
(38) Polynomial Operations

If we want the translation h: A  A’ from L to L’ to be a homomorphism, then we’re
going to need to find some operations F’ to correspond with the syntactic operations of L.
• We’ve already seen that in the general case the basic operations { F’γ’ }γ’ ∈ Γ’ of L’
are not going to be sufficient.
• We’ve also already seen that the polynomial operations over the syntactic algebra for
L’, < A’ , F’γ’ >γ’ ∈ Γ’ can supply us with such operations.
Consequence:
A translation from L to L’ must identify some polynomial operations { Hγ }γ ∈ Γ over the
syntactic algebra for L’ < A’ , F’γ’ >γ’ ∈ Γ’
12
Fall 2013 Ling 720
(39) Derived Syntactic Rules
A translation h from L to L’ should always map the ‘meaningful expressions’ of L to
meaningful expressions in L’.
• Now, recall that we’re going to want h to be a homomorphism, where every syntactic
operation Fγ in L corresponds with some polynomial operation Hγ over the syntactic
algebra of L’.
h( Fγ(α1, … αn) ) = Hγ( h(α1), … h(αn) )
• Consequently, we will want it to be that if Fγ(α1, … αn) is a meaningful expression of

L, then Hγ( h(α1), … h(αn) ) is a meaningful expression of L’ too.
Consequence:
Under a translation h from L to L’, if Fγ in L corresponds with Hγ (a polynomial operation
over L’), then if (a) is a syntactic rule of L, then (b) is a derived syntactic rule of L’
a. < Fγ , <δ1 , … δn>, δ >
b. < Hγ , <g(δ1), … g(δn)>, g(δ) >
• To see how the condition above works, recall the general result in (34): if the tuple
<Hγ , <g(δ1), … g(δn)>, g(δ) > is a derived syntactic rule, and ϕ1, … ϕn are such that
each ϕi ∈ Cg(δi) , then H(ϕ1, … ϕn) ∈ Cg(δ)
o Now, suppose that α is a meaningful expression of L, and Fγ(α1, … αn) = α,

where α1 ∈ δ1, …, αn ∈ δn
o Thus, the translation h(α) = h(Fγ(α1, … αn)) = Hγ(h(α1), … h(αn))
o Now, given our category correspondence (37), it follows that h(α1) ∈ Cg(δ1),
…, h(αn) ∈ Cg(δn)
o Therefore, from our general result in (34) – and the fact that (b) is a derived
rule of L’ – it follows that Hγ(h(α1), … h(αn)) ∈ Cg(δ)
o Thus, we have it that h(α) is also a meaningful expression of L’ !
With all of these ingredients on the table, we can now provide Montague’s general definition
of a ‘translation base’…
13
Fall 2013 Ling 720
(40) Translation Base from Language L to Language L’
Let L be a language <L, R> and L’ be a language <L’, R’>, where L is the disambiguated
language <A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ and L’ = <A’ , F’γ’ , X’δ’ , S’, δ’0>γ’ ∈ Γ’ , δ’ ∈ Δ’
A translation base from L to L’ is a structure <g, Hγ, j>γ ∈ Γ such that:
a. g is a function from Δ to Δ’ (37)
b. For all γ ∈ Γ, Hγ is a polynomial operation over the algebra <A’, F’γ’ >γ’ ∈ Γ’
sharing the same arity as Fγ (38)
c. If <Fγ , <δ1 , … δn>, δ > ∈ S, then the following is a derived syntactic rule for L’:
<Hγ , <g(δ1), … g(δn)>, g(δ) > (39)
d. j is a function whose domain is ∪δ ∈ Δ Xδ , and whenever ϕ ∈ Xδ, j(ϕ) ∈ C’g(δ).
Note: j is a function that maps the basic expressions of L of category δ to some

meaningful expressions of L’ of the corresponding category g(δ).
With the notion of a translation base, we can construct the following definition of a translation
function…
(41) Translation Function from Language L to Language L’

Let T be a translation base <g, Hγ, j>γ ∈ Γ from L to L’
The translation function determined by T is the unique homomorphism k from the

algebra <A, Fγ>γ ∈ Γ to the algebra <A’, Hγ>γ ∈ Γ such that j ⊆ k.
(42) Remarks
• That any translation base T determines such a homomorphism k is essentially

guaranteed by the conditions we placed on the polynomial operations { Hγ }γ ∈ Γ
• If k is a translation function from L to L’, then k is not necessarily a homomorphism

from the syntactic algebra of L to the syntactic algebra of L’
o Rather, it’s a homomorphism to an algebra we define on the basis of L’
o This allows translation to be a homomorphism even if two language algebras
are not themselves homomorphic!
14
Fall 2013 Ling 720
(43) Definition of Translation
Let T be a translation base from L to L’, and let k be the translation function determined
by T.
If α is an expression of L’ and β is an expression of L, then α is a translation of β if

there are α’ ∈ A’ and β’ ∈ A such that:
(i) α’R’α and β’Rβ
(ii) α’ is a meaningful expression of L’ and β’ is a meaningful expression of L
(iii) k(β’) = α’
5. Translating from Mini-English to Politics-NoQ
To round out these notes, we’ll use all these tools to spell out a translation base and
(homomorphic) translation function from Mini-English to Politics-NoQ
(44) Mini-English
Mini-English is the structure < < E, Kγ, Xδ , SE, S >γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ, R>,
where
a. The structure <E, Kγ, Xδ , SE, S>γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ is
Disambiguated Mini-English, as defined in (10).
b. R is a function which takes as input a tree T in E, and returns as output the first
member of the root node of T.
(45) Politics-NoQ
Politics-NoQ is the structure <<A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T , R’>, where:
a. The structure <A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T is the disambiguated

language Politics-NoQ, as defined previously.
b. R is the identity function.
We’ll now lay out each of the three main ingredients for a translation base from Mini-English to
Politics-NoQ…
15
Fall 2013 Ling 720
(46) Correspondence Between the Syntactic Categories
Given our discussion in (37), let us define the function g: Δ  T as follows:
g(NP) = e g(TV) = <e, <e,t>> g(IV) = <e,t> g(S) = t
(47) The Polynomial Operations

Given our discussion below (31), we will want to build our translation base from the
following polynomial operations over the algebra <A, Fγ >γ ∈ {Concat, Not, And}.
a. HNot and HAnd = FNot and FAnd, respectively
b. HMerge-IV = FConcat
c. HIf = FNot<FAnd<Id1,2, FNot<Id2,2>>>
d. HMerge-S = FConcat<Id2,2, Id1,2>
(48) Crucial Step We Will Usually Leave Implicit

If we build our translation base on the polynomial operations above, then condition
(39)/(40c) – combined with our category correspondence in (46) – requires the following:
a. < HNot , <t>, t > is a derived rule of Politics-NoQ
b. < HAnd , <t, t>, t > is a derived rule of Politics-NoQ
c. < HMerge-IV , < <e,<e,t>>, e >, <e,t> > is a derived rule of Politics-NoQ
d. < HIf , <t, t>, t> is a derived rule of Politics-NoQ
e. < HMerge-S , <e, <et>>, t> is a derived rule of Politics-NoQ
Showing that (48a-e) hold will be left as an exercise for the student. Note that (48a,b,c) are
trivial; only (48d,e) require some calculating out…
(49) The (Lexical Translation) Function j

Given the category correspondence in (46) – and what we obviously want to achieve – let
us define the function j as follows:
a. j(< Barack , ∅ >) = barack’

b. j(< Michelle , ∅ >) = michelle’
c. j(< Mitt , ∅ >) = mitt’
d. j(< smokes , ∅ >) = smokes’
e. j(< loves , ∅ >) = loves’
16
Fall 2013 Ling 720
(50) Putting It All Together: The Translation Base from Mini-English to Politics-NoQ
Let T be the structure < g , Hγ , j>γ ∈ {Merge-S, Merge-IV, Not, And, If}, where g, Hγ , and j are as
defined in (46)-(49). T is a translation base from Mini-English to Politics-NoQ.
We can now use the translation function k determined by T to homomorphically map expressions
of mini-English to expressions of Politics-NoQ.
(51) Translating from Mini-English to Politics-NoQ

Let k be the translation function determined by T, as defined in (50). Let T be the tree in
(10) above.
a. k(T) = (by definition of DME)
b. k( KMerge-S( < Barack , ∅ >,

KMerge-IV( < loves ∅ >, < Michelle, ∅ > ) ) ) =
(by homomorphism property of k)
c. HMerge-S( k(< Barack , ∅ >) ,

k( KMerge-IV( < loves ∅ >, < Michelle, ∅ > ) ) ) =
(by homomorphism property of k)
d. HMerge-S( k(< Barack , ∅ >) ,

HMerge-IV( k(< loves ∅ >), k(< Michelle, ∅ > ) ) ) =
(by definition of k and j)
e. HMerge-S( j(< Barack , ∅ >) ,

HMerge-IV( j(< loves ∅ >), j(< Michelle, ∅ > ) ) ) =
(by definition of j)
f. HMerge-S( barack’ , HMerge-IV( loves’, michelle’ ) ) =

(by definition of HMerge-IV)
g. HMerge-S( barack’ , (loves’ michelle’) ) =

(by definition of HMerge-S)
h. ((loves’ michelle’) barack’)
(52) Remark
Under the translation function k determined by T, and given the definition in (43), it
follows that there are two different formulae in Politics-NoQ that are translations of the
Mini-English sentence It is not the case that Barack smokes and Mitt smokes.
(exercise for the student!)
17
Fall 2013 Ling 720
(53) What We Wanted
We wanted to develop a way of homomorphically mapping expressions of DME to
<E, Kγ, Xδ , SE, S>γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ
k (translation from DME to Pol-NoQ)

g°k (the semantic
interpretation of DME)
(54) What We Have Now
< E, Kγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} < A , Fγ >γ ∈ {Concat, Not, And}
k g
<A, Hγ>γ ∈ {Merge-S, Merge-IV, Not, And, If} < B, Gγ >γ ∈ {Concat, Not, And}
o We have a way of homomorphically mapping expressions of Mini-English to

expressions of Politics-NoQ.
o However, our translation homomorphism doesn’t hold between the syntactic

algebra of Mini-English and the syntactic algebra of Politics-NoQ.
o Rather, it holds between the syntactic algebra of Mini-English and another
syntactic algebra that we construct on the basis of < A , Fγ>γ ∈ {Concat, Not, And}
o But, our interpretation homomorphism holds between the syntactic algebra of

Politics-NoQ and the interpretation <B, Gγ , f>γ ∈ {Concat, Not, And}
o Our interpretation function g is not a homomorphism from our derived
syntactic algebra <A, Hγ>γ ∈ {Merge-S, Merge-IV, Not, And, If}
o Therefore, the composition of k and g is not a homomorphism from our syntactic

algebra for Mini-English to the interpretation <B, Gγ , f>γ ∈ {Concat, Not, And}
o So how do we get what we want, a homomorphism from the syntactic
algebra of Mini-English to an interpretation structure?....
Tune in next time for Part 3 of ‘Montague’s Theory of Translation’…
18
Fall 2013 Ling 720
Montague’s Theory of Translation: Translation Bases and Indirect Interpretations 1
1. From Translation Functions to Interpretations
(1) What We Currently Have
a. Mini-English (and Disambiguated-Mini-English):

Mini-English is <<E, Kγ, Xδ , SE, S>γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ, R>
(i) < E, Kγ, Xδ , SE, S>γ ∈ {Merge-S, Merge-IV, Not, And, If}, δ ∈ Δ is Disambiguated
Mini-English
(ii) R maps trees in E to the first member of their root node.
b. Politics-NoQ
Politics-NoQ is the language << A , Fγ , Xτ , S , t >γ ∈ {Concat, Not, And}, τ ∈ T , Id>
(i) <A , Fγ , Xτ , S , t>γ ∈ {Concat, Not, And}, τ ∈ T is the disambiguated language
Politics-NoQ.
(ii) Id is the identity function.
c. Interpretation for Politics-NoQ

Let the set S = { Michelle, Barack, Mitt }. Let B = <B, Gγ , f>γ ∈ {Concat, Not, And} be
the Fregean interpretation based on S, such that f is as defined before.
d. Translation Base from Mini-English to Politics-NoQ

Let be T is the translation base < g , Hγ , j>γ ∈ {Merge-S, Merge-IV, Not, And, If}, where g,
Hγ , and j are as defined before.
Where k is the translation function determined by T and h is the meaning assignment

determined by B, we have the following homomorphisms.
k h
(2) Key Issue

• What we want is a homomorphism that maps E to B.
• But since the operation indices are different in <A, Hγ>γ ∈ {Merge-S, Merge-IV, Not, And, If}
and < A , Fγ >γ ∈ {Concat, Not, And} we don’t have such a homomorphism yet…
1
1
Fall 2013 Ling 720
(3) The Key Theorem Relating Translation to Interpretation

If <A, Fγ>γ ∈ Γ is an algebra, h is a homomorphism from <A, Fγ >γ ∈ Γ to some algebra
(<B, Gγ>γ ∈ Γ), and for each γ ∈ Π, Hγ is a polynomial operation over <A, Fγ>γ ∈ Γ ,
then there is exactly one algebra <B, Gγ >γ ∈ Π such that h is a homomorphism from
<A, Hγ >γ ∈ Π to <B, Gγ >γ ∈ Π (Montague 1974: 225)
(4) Restatement of the Key Theorem

Suppose the following conditions all hold:
a. There are two algebras A = <A, Fγ>γ ∈ Γ and B = <B, Gγ>γ ∈ Γ
b. There is a homomorphism h from A to B
c. There is a an algebra A’ = <A, Hγ>γ ∈ Π where { Hγ }γ ∈ Π are all polynomial
operations over <A, Fγ>γ ∈ Γ
There is therefore one (exactly one) algebra B’ = <B, Gγ>γ ∈ Π such that h is also a
homomorphism from A’ to B’.
(5) The Importance of the Key Theorem
• Conditions (4a-c) are exactly what we have in (1)!

a. <A, Fγ>γ ∈ {Concat, Not, And} and <B, Gγ>γ ∈ {Concat, Not, And} are algebras.
b. Meaning assignment h is a homomorphism from A <A, Fγ>γ ∈ {Concat, Not, And} to B
<B, Gγ>γ ∈ {Concat, Not, And}
c. <A, Hγ>γ ∈ {Merge-S, Merge-IV, Not, And, If} is an algebra where the operations are all
polynomial operations over <A, Fγ>γ ∈ {Concat, Not, And}
• Therefore, the key theorem in (3)/(4) guarantees us the following:

There is an algebra <B, Gγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} such that h is also a
homomorphism to it from <A, Hγ >γ ∈ {Merge-S, Merge-IV, Not, And, If}
k h
< B, Gγ >γ ∈ {Merge-S, Merge-IV, Not, And, If}
2
Fall 2013 Ling 720
(6) Remarks
• Note that in the algebra < B, Gγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} the set B is the set of
meanings in our interpretation of Politics-NoQ
• Although Montague doesn’t say it, the operations { Gγ }γ ∈ {Merge-S, Merge-IV, Not, And, If}
are all polynomial operations over <B, Gγ>γ ∈ {Concat, Not, And} whose definitions mirror
those of the operations in our translation base { Hγ }γ ∈ {Merge-S, Merge-IV, Not, And, If}
• As desired, hºk is a homomorphism from <E, Kγ>γ ∈ { Merge-S, Merge-IV, Not, And, If}
(Disambiguated-Mini-English) to < B, Gγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} (our derived
semantic algebra).
• Thus, thanks to our translation base T, we now have an interpretation for Mini-
English.
The Interpretation for Mini-English: < B, Gγ , hºj >γ ∈ {Merge-S, Merge-IV, Not, And, If}
(i) <B, Gγ >γ ∈ {Merge-S, Merge-IV, Not, And, If} is the derived algebra guaranteed by (3)/(4),
(ii) h is the meaning assignment determined by < B, Gγ >γ ∈ {Concat, Not, And}
(iii) j is the lexical translation function in our translation base T.
• Note that this structure will satisfy our general definition of an interpretation:
o Because hºk is a homomorphism, we know that for all γ ∈ {Merge-S,

Merge-IV, Not, And, If}, Kγ and Gγ are of the same arity.
o hºj is a function from the basic expressions of Mini-English into B.
Clearly, this result in (5)/(6) generalizes to all languages, allowing us to state the following
general theorem…
(7) General Theorem on Indirect Interpretation

Let L and L’ be languages such that there is an interpretation B for L’ and a translation
base T from L to L’. There is an interpretation B’ for L.
You can no doubt already see how (7) follows from what we’ve seen so far…
For those who are interested, we can give a more explicit proof of it as in (8).
3
Fall 2013 Ling 720
(8) Proof of the General Theorem on Indirect Interpretation
• Let h be the meaning assignment determined by B. Let k be the translation

function determined by T. Let j be the lexical translation function in T.
• By definition, h is a homomorphism from the syntactic algebra of L’ <A, Fγ>γ ∈ Γ

to the semantic algebra of B <B, Gγ>γ ∈ Γ
• By definition, k is a homomorphism from the syntactic algebra of L <E, Kγ>γ ∈ Π

to the algebra <A, Hγ>γ ∈ Π where {Hγ}γ ∈ Π are all polynomial operations over
<A, Fγ>γ ∈ Γ
• Therefore, by the theorem in (3)/(4), there is an algebra <B, Gγ>γ ∈ Π such that h
is also a homomorphism from <A, Hγ>γ ∈ Π to <B, Gγ>γ ∈ Π
• Consequently, hºk is a homomorphism from <E, Kγ>γ ∈ Π to <B, Gγ>γ ∈ Π
• Therefore, we know that for all γ ∈ Π, Kγ and Gγ have the same arity.
• Moreover, hºj is a function from the basic categories in L to B.
• Therefore, the structure <B, Gγ, hºj >γ ∈ Π is an interpretation for L.
(9) Remark
Furthermore, the meaning assignment determined by the interpretation <B, Gγ, hºj >γ ∈ Π
will be hºk (proof left as exercise for the student)
(10) The Big Upshot
a. Our Initial Question:

Given our background theory of language and meaning, under what conditions
can we guarantee that translating from one language L into another language L’
gives us a compositional semantics for L.
b. Answer:
If we can provide an interpretation for L’ and the translation from L to L’ satisfy
the conditions of a translation base, then we are guaranteed a compositional
semantics for L.
Thus, a new viable path to providing a semantics for a (natural) language is to provide a
translation base from that language to a logical language whose semantics is already defined.
4
Fall 2013 Ling 720
2. Illustration: Mini-English and Politics-NoQ
(11) First Observation

When we compose together k and h in (1), we get a mapping from sentences of
Disambiguated-Mini-English to meanings in B.
Illustration: Let T be the tree such that R(T) = Barack loves Michelle.
a. hºk(T) = (by definition of function composition)
b. h(k(T)) = (by definition of Mini-English)
c. h(k( KMerge-S( < Barack , ∅ >,

KMerge-IV( < loves ∅ >, < Michelle, ∅ > ) ) ) ) =
(by homomorphism prop.)
d. h(HMerge-S( k(< Barack , ∅ >) ,
HMerge-IV( k(< loves ∅ >), k(< Michelle, ∅ > ) ) ) ) =
(by definition of k)
e. h(HMerge-S( barack’, HMerge-IV( loves’, michelle’ ) ) ) = (by def. of HMerge-IV)
f. h(HMerge-S( barack’ , (loves’ michelle’) ) ) = (by definition of HMerge-S)
g. h(((loves’ michelle’) barack’)) = (by definition of Politics-NoQ)
h. h( FConcat( FConcat(loves’, michelle’), barack’)) = (by homomorphism prop.)
i. GConcat( GConcat(h(loves’), h(michelle’)), h(barack’))) = (by def. of h)
j. GConcat( GConcat( j , Michelle ), Barack )) = (by def. of GConcat)
k. GConcat( j(Michelle) , Barack )) = (by def. of GConcat)
l. j(Michelle)(Barack) = (by def. of j)
m. 1
(12) Second Observation

Observing the behavior of hºk over a range of examples, we can directly construct an
interpretation for Disambiguated-Mini-English, which will mirror the behavior of hºk
5
Fall 2013 Ling 720
(13) The Interpretation of Disambiguated Mini-English
Let < B, Gγ , l >γ ∈ {Merge-S, Merge-IV, Not, And, If} be the structure defined as follows:
a. The Definition of the Set B

The set B is the same set as the set B in B = <B, Gγ , f>γ ∈ {Concat, Not, And} , the
interpretation of Politics-NoQ we had defined previously.
• That is B = ∪τ∈T Dτ, {Michelle, Barack, Mitt}
b. The Definition of the Semantic Operations

The operations {Gγ}γ ∈ {Merge-S, Merge-IV, Not, And, If} are defined as follows:
(i) GNot = GNot (defined previously)
(ii) GAnd = GAnd (defined previously)
(iii) GIf = GNot<GAnd<Id1,2, GNot<Id2,2>>>
(iv) GMerge-IV = GConcat (defined previously)
(v) GMerge-S = GConcat<Id2,2, Id1,2>
Note:
• The definitions of the operations {Gγ}γ ∈ {Merge-S, Merge-IV, Not, And, If} mirror the
definitions of the operations {Hγ}γ ∈ {Merge-S, Merge-IV, Not, And, If} in our translation base.
• Moreover, {Gγ}γ ∈ {Merge-S, Merge-IV, Not, And, If} are all polynomial operations over the
semantic algebra <B, Gγ>γ ∈ {Concat, Not, And}
• Consequently, <B, Gγ>γ ∈ {Merge-S, Merge-IV, Not, And, If} is an algebra.
c. The Definition of the Lexical Interpretation Function

The lexical interpretation function l is defined as follows:
(i) l(< Barack , ∅ >) = Barack

(ii) l(< Michelle , ∅ >) = Michelle
(iii) l(< Mitt , ∅ >) = Mitt
(iv) l(< smokes , ∅ >) = the function h equal to f(smokes’)
(v) l(< loves , ∅ >) = the function j equal to f(loves’)
Note: Where g is the meaning assignment determined by B, l = gºj
(14) Remark < B, Gγ , l >γ ∈ {Merge-S, Merge-IV, Not, And, If} is an interpretation of Mini-English
6
Fall 2013 Ling 720
(15) Remark
The meaning assignment g determined by <B, Gγ, l>γ ∈ {Merge-S, Merge-IV, Not, And, If} is equal
to hºk
(16) Illustration
Let g be the meaning assignment determined by <B, Gγ l>γ ∈ {Merge-S, Merge-IV, Not, And, If}.
Let T be the tree such that R(T) = Barack loves Michelle.
a. g(T) = (by definition of Mini-English)
c. g( KMerge-S( < Barack , ∅ >,

KMerge-IV( < loves ∅ >, < Michelle, ∅ > ) ) ) =
(by homomorphism prop.)
d. GMerge-S( g(< Barack , ∅ >) ,
GMerge-IV( g(< loves ∅ >), g(< Michelle, ∅ > ) ) ) =
(by definition of g)
e. GMerge-S( Barack, GMerge-IV( j, Michelle ) ) = (by def. of GMerge-IV)
f. GMerge-S( Barack, j(Michelle) ) = (by def. of GMerge-S)
g. j(Michelle)(Barack) = (by def. of j)
m. 1
3. Indirect Interpretation: A Summary
(17) Direct Interpretation

Let L be a language. Direct interpretation of L is the specification of an interpretation B
of L.
(18) Indirect Interpreation

Let L be a language. Indirect interpretation of L is the specification of a language L’,
an interpretation B’ of L’ and a translation base T from L to L’.
(19) Indirect Interpretation Always Yields Direct Interpretation
• Suppose that we have indirectly interpreted the language L.

o That is, we have defined a language L’ an interpretation B’ of L’ and a
translation base T from L to L’
• Given the result in (7)/(8) it is thus trivial to construct an interpretation B of L.
7
Fall 2013 Ling 720
(20) Some Choice Quotes

“An understanding of [the eliminability of indirect interpretation] is necessary to
understand the use of…logics of PTQ and most other analyses within Montague
grammar. As has been stated elsewhere, the [logical language] is an expository device
and is in no way a necessary part of an analysis of any language offered within this
theory. By using the easily interpreted [logical language] as a mediator, natural
languages can be analyzed syntactically and then provided with a translation…from
them to [the logical language] to induce their interpretation….This method of
analysis amounts to direct interpretation of natural language.” (emphasis mine)

“Translating English into [a logical language] was therefore not essential to
interpreting the English phrases we generated; it was simply a convenient
intermediate step in assigning them meanings. This step could have been
eliminated had we chosen to describe the interpretation of English directly… This
point is important, because anyone who does not appreciate it may misunderstand the
role of [logical languages] in applications of Montague’s descriptive framework to
natural languages.” (emphasis mine)
(21) Why Do Indirect Interpretation?
It’s Just Sometimes Conceptually ‘Easier’

If the logical language is well-designed and familiar to readers, then it can provide a more
‘perspicuous’ representation (statement / name) of the meanings that we wish to be
assigned to the (natural) language expressions.
Illustration (From 610):

‘[ λxe : x smokes ]’ vs. ‘The function f from De to Dt such that for all x ∈ De,
f(x) = 1 iff x smokes.’

“Since the translation process is less involved than interpretation, presentation of
fragments of the languages becomes clearer.”

“[The purpose of indirect interpretation] was to have a convenient, compact notation
for giving a briefer statement of semantic rules than we were able to give in earlier
chapters of this book, where semantic rules were formulated rather long-windedly in
English… [The logical language] could provide us with names for meanings…”
8
Fall 2013 Ling 720
(21) Road Map of Where We’ve Been and Where We’re Going
• We’ve now covered the conceptual core of the Montague Grammar architecture, as
well as the key sections of Montague’s “Universal Grammar”.
o Sections of UG Covered: 1, 2, 3, 5
• What we haven’t covered from UG is:

o The general theory of Fregean interpretations
o Montague’s presentation of the syntax/semantics of his Intensional Logic
o Montague’s presentation of the syntax/translation of a fragment of English
• However, those sections of UG are largely superceded by Montague’s paper

“The Proper Treatment of Quantification in Ordinary English” (PTQ)
o If we had all the time in the world, we would cover both systems and compare
them (hint for final paper).
o Given our limited time, however, I’d like to now move us from UG to PTQ.
• What’s Next on the Agenda:

Extending this system to handle quantification in FOL and English.
• An algebraic treatment of FOL (with quantification)
• An translation base from a fragment of English to FOL.
• Following that, we’ll examine Montague’s Intensional Logic and its applications to
English (and various puzzles therein).
9
Fall 2013 Ling 720
Problem Set on Translation and Indirect Interpretation
(1) Polynomial Operations and Algebras

Let A be an algebra <A, Fγ>γ ∈ Γ , and let H be a member of the polynomial operations K
over A. Show that A is closed under H.
There are four steps to showing that A is closed under H, corresponding to the four
‘ways’ by which H could be a member of K.
a. Step One: Show that A is closed under Fγ for all γ ∈ Γ
b. Step Two:
Let Idn,m be any identity function (projection function). Show that A is closed
under Idn,m.
c. Step Three: Let a ∈ A. Show that A is closed under Ca,m.
d. Step Four:
Let G be an n-ary function that A is closed under. Let F1, … Fn be n m-ary
functions that A is closed under. Show that A is closed under G<F1, … Fn>.
(2) Derived Syntactic Rules and Meaningful Expressions

Let L be a language <<A , Fγ , Xδ , S, δ0>γ ∈ Γ , δ ∈ Δ, R>, and let <H, <δ1 , … δn>, δ> be
a derived syntactic rule of L. Show that if ϕ1, … ϕn are such that each ϕi ∈ Cδi, then
H(ϕ1, … ϕn) ∈ Cδ.
There are four steps to showing that H(ϕ1, … ϕn) ∈ Cδ, corresponding to the four ‘ways’
by which <H, <δ1 , … δn>, δ> could be a derived syntactic rule of L.
a. Step One:
Let <H, <δ1 , … δn>, δ> ∈ S. Show that if ϕ1, … ϕn are such that each ϕi ∈ Cδi,
then H(ϕ1, … ϕn) ∈ Cδ.
b. Step Two:
Let <H, <δ1 , … δn>, δ> be a rule of the form < Idn,m , <δ1, … , δn, … , δm> , δn>.
Show that if ϕ1, … ϕm are such that each ϕi ∈ Cδi, then Idn,m(ϕ1, … ϕm) ∈ Cδn .
c. Step Three:
Let <H, <δ1 , … δn>, δ> be of the form < Ca,n, <δ1, …, δn>, δ >, where a ∈ Cδ.
Show that if ϕ1, … ϕn are such that each ϕi ∈ Cδi, then Ca,n(ϕ1, … ϕn) ∈ Cδ.
Continued on next page…
1
Fall 2013 Ling 720
d. Step Four:
Let the rule <F , <δ1 , … δn>, δ> have the property that if ϕ1, … ϕn are such that
each ϕi ∈ Cδi, then F(ϕ1, … ϕn) ∈ Cδ. In addition, for each G1, … Gn, let the rule
<Gj , <δ’1 , … δ’m>, δj > have the property that if ϕ1, … ϕm are such that each ϕi ∈
Cδ’i, then Gj(ϕ1, … ϕm) ∈ Cδj.
Show that the rule <F<G1, … Gn>, <δ’1 , … δ’m>, δ> has the property that if ϕ1,
…, ϕm are such that each ϕi ∈ Cδ’i, then F<G1, … Gn>(ϕ1, … ϕm) ∈ Cδ.
(3) An Exercise in Indirect and Direct Interpretation of a Fragment of English
a. Minimally alter our language Mini-English so that its expressions now include
strings like Neither Mitt smokes nor Barack smokes.
Note:
Don’t worry if your system also produces such marginal strings as Neither it
is not the case that Mitt smokes nor Barack loves Michelle.
b. Take our translation base in (46)-(50) on the handout “The Notion of a

Translation Base”, and minimally alter it so that strings like Neither Mitt smokes
nor Barack smokes receive appropriate translations in Politics-NoQ.
Note:
Be sure to show that any new polynomial operations in your translation base
have the property in (40c) on handout “The Notion of a Translation Base.”
c. Please show how your new translation base, along with our interpretation for
Politics-NoQ, assigns a truth-value to the analysis tree for Neither Mitt smokes
nor Barack smokes.
d. Given your proposed translation base, construct a direct interpretation of

Disambiguated Mini-English, and show how it interprets Neither Mitt smokes nor
Michelle smokes.
2

Unit 5:
An Algebraic Approach to Quantification and
Lambda Abstraction

Fall 2013 Ling 720
An Algebraic Approach to Quantification and Lambda Abstraction: Preliminaries 1
(1) Our Analytic Toolbox (The Framework)
a. A Theory of Syntax
(i) ‘Disambiguated’ languages as quintuples <A, Fγ, Xδ, S, δ0>γ ∈ Γ , δ ∈ Δ
(ii) Languages as pairs < <A, Fγ, Xδ, S, δ0>γ ∈ Γ , δ ∈ Δ, R>
b. A Theory of Semantics
(i) Interpretations as structures <B, Gγ, f >γ ∈ Γ
(ii) Meaning assignments as homomorphisms from <A, Fγ>γ∈Γ to <B, Gγ>γ∈Γ
c. A Theory of Translations Inducing Interpretations

(i) Translation bases as structures < g , Hγ’ , j>γ’ ∈ Γ’
(ii) Translations as homomorphisms from <A’, Fγ,>γ’ ∈ Γ’ to <A, Hγ’>γ’ ∈ Γ’
(iii) Translation from L’ to L induces the interpretation <B, G’γ’, f’ >γ’ ∈ Γ’
We’re now going to apply these analytic tools (framework) to the phenomenon of quantification,
both in English and in logical languages…
(2) An Outline of the Plan
a. Step One: Introduce a logical language with quantification.
• Since we’re ultimately going to use this logical language to indirectly interpret
English quantificational NPs, this language will also include ‘λ’
b. Step Two: Introduce a model-theoretic semantics for the language
• This will give us a more comprehensible means for establishing certain key
semantic properties of the language.
• It will also provide the basis for Step Three
c. Step Three: Introduce a (Montagovian) interpretation structure for the language
• This is the intellectually hardest step.
d. Step Four: Introduce a translation base from English to the logical language.
• This takes work, but the key ideas will already be familiar to you.
1
These notes are based upon material in the following readings: Partee et al. (1993) Chapter 13, Dowty et al. (1981)
1
Fall 2013 Ling 720
1. The Logical Language TL: Non-Montagovian Presentation
I assume that everyone has a basic familiarity with lambda abstraction and its applications to
natural language semantics.
• For a review, the student is referred to Heim & Kratzer (1998: 34-40) and Partee et al.
(1993: 336-369).
(3) Typed Logic (TL)

A typed logic (language) is a language whose vocabulary of symbols satisfies the
conditions in (4), and whose syntax satisfies the conditions in (7).
(4) The Vocabulary of a TL
a. The Logical Constants: 2

(i) Sentence Connectives: ~, &, ∨, 
(ii) Quantifiers: ∀, ∃
(iii) Lambda Operator: λ
b. The Syntactic Symbols: (,)
c. The Non-Logical Constants:

(i) Constants:
For every type τ ∈ T, a countable set of constants of type τ, CONτ
(ii) Variables:
For every type τ ∈ T, a countably infinite set of variables of type τ:
VARτ = { vτ, n : n ∈ ℕ }
(5) Remarks
a. Note that in (4c), we are using the types to categorize expressions of our
language, just as in our Montagovian definitions of FOL-NoQ and Politics-NoQ.
b. Given (4ci), we will have individual constants (CONe) and for any n ∈ ℕ, a set of
n-ary predicate constants (CON<e,t>, CON<e,<e,t>>, CON<e<e<e,t>>>, …)
c. Given (4ci), we can also have constants of type t, <<e,t>,t>, <e,e>, etc.
o Given our intended applications, we won’t make use of such constants here.
d. Given (4cii), our variables now come with a subscript indicating their type.
• ve,3 v<e,t>,4 v<e,<e,t>>,2 v<<e,t>,t>,6 vt,190
2
We’ll now make use of a larger set of primitive logical operators. This will make the statement of our semantics
longer, but it will later simplify our analysis of English.
2
Fall 2013 Ling 720
(6) Meta-Language Abbreviations for Variables
Although our variables ‘officially’ all look like those in (5d), to save space we will make
use of the following meta-language abbreviations for variables of type e and <e,t>:
a. xn = ve,n b. Pn = v<e,t>,n
In our (relatively) informal syntaxes for PL, FOL, FOL-NoQ, Politics, and Politics-NoQ, our
syntax defined a set of ‘WFFs’…
• But, now that our logical language contains ‘λ’, it will be more efficient to arrange our
syntax so that it defines sets of ‘meaningful expressions’ of various types τ ∈ T (MEτ)
(7) The Syntax of a TL
a. If ϕ ∈ CONτ or ϕ ∈ VARτ, then ϕ ∈ MEτ
b. If ϕ ∈ ME<σ,τ> and ψ ∈ MEσ, then (ϕ ψ) ∈ MEτ
c. If ϕ, ψ ∈ MEt , then (i) ~ϕ ∈ MEt

(ii) (ϕ & ψ) ∈ MEt
(iii) (ϕ ∨ ψ) ∈ MEt
(iv) (ϕ  ψ) ∈ MEt
d. If v ∈ VARτ, and ϕ ∈ MEt , then (i) ∃vϕ ∈ MEt

(ii) ∀vϕ ∈ MEt
e. If v ∈ VARσ , and ϕ ∈ MEτ , then (λv ϕ) ∈ ME<σ ,τ >
(8) Illustration of a TL Language: Politics+λ
a. Vocabulary of Politics+λ
(i) Logical Constants: (as in (4a))
(ii) Syntactic Symbols: (, )
(iii) Non-Logical Constants
1. Variables (as in (4cii))
2. Constants: CONe = { mitt’, barack’, michelle’ }
CON<et> = { smokes’, man’, president’ }
CON<e<e,t>> = { loves’ }
For all other τ ∈ T, CONτ = ∅
b. Syntax of Politics+λ: (as in (7))
3
Fall 2013 Ling 720
(9) Some Illustrative Meaningful Expressions of Politics+λ
a. ~ ( ( loves’ mitt’ ) barack’ )
(i) loves’ ∈ CON<e<e,t>>, mitt’, barack’ ∈ CONe (8a)

(ii) loves’ ∈ ME<e<e,t>>, mitt’, barack’ ∈ MEe (7a)
(iii) ( loves’ mitt’ ) ∈ ME<e,t> (7b)
(iv) ( ( loves’ mitt’ ) barack’ ) ∈ MEt (7b)
(v) ~ ( ( loves’ mitt’ ) barack’ ) ∈ MEt (7c)
b. ∀x3 ( (smokes’ x3)  ~ ( ( loves’ mitt’ ) x3) )
(i) loves’ ∈ CON<e<e,t>>, smokes’ ∈ CON<e,t>, mitt’, barack’ ∈ CONe, x3 ∈ VARe (8a)
(ii) loves’ ∈ ME<e<e,t>>, smokes’ ∈ ME<e,t>, mitt’, barack’, x3 ∈ MEe (7a)
(iii) (loves’ mitt’) ∈ ME<e,t> (7b)
(iv) ((loves’ mitt’) x3) ∈ MEt (7b)
(v) ~((loves’ mitt’) x3) ∈ MEt (7c)
(vi) (smokes’ x3) ∈ MEt (7b)
(vii) ((smokes’ x3)  ~((loves’ mitt’) x3)) ∈ MEt (7c)
(viii) ∀x3 ( (smokes’ x3)  ~ ( ( loves’ mitt’ ) x3) ) ∈ MEt (7d)
c. ∃P4 (P4 michelle’)
(i) michelle’ ∈ CONe, P4 ∈ VAR<e,t> (8a)

(ii) michelle’ ∈ MEe, P4 ∈ ME<e,t> (7a)
(iii) (P4 michelle’) ∈ MEt (7b)
(iv) ∃P4 (P4 michelle’) ∈ MEt (7d)
d. ( (λx3 (man’ x3)) mitt’ )
(i) man’ ∈ CON<e,t>, mitt’ ∈ CONe, x3 ∈ VARe (8a)

(ii) man’ ∈ ME<e,t>, mitt’, x3 ∈ MEe (7a)
(iii) (man’ x3) ∈ MEt (7b)
(iv) (λx3 (man’ x3)) ∈ ME<e,t> (7e)
(v) ( (λx3 (man’ x3)) mitt’ ) ∈ MEt (7b)
e. (λP4 ∀x3 ( (man’ x3)  (P4 x3) ) )
(i) man’ ∈ CON<e,t>, P4 ∈ VAR<e,t>, x3 ∈ VARe (8a)

(ii) man’, P4 ∈ ME<e,t>, x3 ∈ MEe (7a)
(iii) (man’ x3), (P4 x3) ∈ MEt (7b)
(iv) ( (man’ x3)  (P4 x3) ) ∈ MEt (7c)
(v) ∀x3 ( (man’ x3)  (P4 x3) ) ∈ MEt (7d)
(vi) (λP4 ∀x3 ( (man’ x3)  (P4 x3) ) ) ∈ ME<<e,t>,t> (7e)
4
Fall 2013 Ling 720
2. The Semantics of a TL: Non-Montagovian Presentation
In this section, we will develop a model-theoretic semantics for TL languages.
• This model-theoretic semantics will provide the basis from which we will develop a
(Montagovian) interpretation for a TL language.
• Moreover, the model-theoretic semantics will allow for a more perspicuous

demonstration of a key validity in TL languages (‘lambda conversion’)
• Furthermore, seeing the relationship between this model-theoretic semantics and the later
interpretation will lay important groundwork for our presentation of PTQ
(10) The Denotations Based on a Set E 3

Let T be the set of types and E be some non-empty set (of entities). If τ ∈ T, then the set
Dτ, E of denotations of type τ based on E is defined as follows:
(i) De,E = E
(ii) Dt,E = { 0, 1 }
(11) Definition of a Model for a TL Language

A model M for a TL language L is a pair <E, I> consisting of: 4
a. A non-empty set E, called the ‘domain of M’
b. A function I, whose domain is equal to (i) and whose range satisfies the condition
in (ii).
(i) Domain of I: ∪τ ∈ T CONτ
(ii) Condition on Range of I: If α ∈ CONτ , then I(α) ∈ Dτ,E
(12) Remarks
If <E,I> is a model for a TL language, then:
a. If α is an individual constant (α ∈ CONe), then I(α) is a member of E (De,E)
b. If α is an n-ary predicate letter (α ∈ CON<e,…,t>), then I(α) is the curried

characteristic function of an n-ary relation in E (D<e,…,t>,E)
3
Note that since our model-theoretic semantics won’t ever interpret ‘syntactic garbage’, we needn’t add the special
element garbage to De,E.
4
To avoid confusion with the notation in (10), the domains of models will now generally be represented as ‘E’.
5
Fall 2013 Ling 720
(13) Illustration: A Model for Politics+λ
Let the model M be the pair <{Barack, Michelle, Mitt}, I>, where I consists of the
following mappings:
a. I(michelle’) = Michelle
b. I(barack’) = Barack
c. I(mitt’) = Mitt
d. I(smokes’) = h = { <Michelle, 0>, <Barack,1>, <Mitt,0> }

e. I(man’) = i = { <Michelle, 0>, <Barack,1>, <Mitt,1> }
f. I(president’) = k = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
e. f(loves’)= j = Michelle  Michelle  1

Barack  1
Mitt  0
Barack  1
Mitt  0
Barack  0
Mitt  1
Given that we now have variables for types other than e, we also need a concomitant change in
our definition of a variable assignment…
(14) Variable Assignment

Let M be a model <E, I> of a TL language. Then g is a variable assignment (based on
M) if its domain is equal to (i) and its range satisfies the property in (ii).
(i) Domain of g: ∪τ ∈ T VARτ
(iii) Condition on Range of g: If α ∈ VARτ , then g(α) ∈ Dτ,E
(15) Remarks
If M = <E,I> is a model for a TL language, and g is a variable assignment based on M
a. If α is an individual variable (α ∈ VARe), then g(α) is a member of E (De,E)
b. If α is an n-ary predicate letter variable (α ∈ VAR<e,…,t>), then g(α) is the curried

characteristic function of an n-ary relation in E (D<e,…,t>,E)
6
Fall 2013 Ling 720
With the definitions in (11) and (14), we can now define the notion of ‘interpretation with
respect to a model M and a variable assignment g, [[.]]M,g.’
(16) Interpretation With Respect to a Model and a Variable Assignment

Let M be a model <E, I> for a TL language L and g be a variable assignment based on
M. The interpretation (a.k.a. denotation) of a meaningful expression of L relative to M
and g [[.]]M,g is defined as follows:
a. If v ∈ ∪τ ∈ T VARτ , then [[ v ]]M,g = g(v)
b. If α ∈ ∪τ ∈ T CONτ , then [[ α ]]M,g = I(α)
c. If ϕ = (ψ χ), then [[ϕ ]]M,g = [[ψ]]M,g([[χ]]M,g)
d. If ϕ = ~ψ, then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0
e. If ϕ = (ψ & χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 1 and [[χ]]M,g = 1
f. If ϕ = (ψ ∨ χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 1 or [[χ]]M,g = 1
g. If ϕ = (ψ  χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0 or [[χ]]M,g = 1
h. If ϕ = ∃vψ and v ∈VARτ , then [[ϕ]]M,g = 1 iff

there is an a ∈ Dτ ,E such that [[ψ]]M,g(v/a) = 1
i. If ϕ = ∀vψ and v ∈VARτ , then [[ϕ]]M,g = 1 iff for all a ∈ Dτ ,E , [[ψ]]M,g(v/a) = 1
j. If ϕ = (λvψ), v ∈VARσ and ψ ∈ MEτ , then [[ϕ]]M,g =

The function p whose domain is Dσ ,E , whose range is Dτ ,E and for all a ∈ Dσ ,E ,
p(a) = [[ψ]]M,g(v/a)
(17) Remarks
a. Given the sorting of variables into types, along with the definition in (14), our
definitions in (16h,i) allow our language to quantify, not only over entities, but
also over objects of all other types.
b. Similarly, given the definition in (14), our definition in (16j) entails that the type
of the variable in a lambda expression will determine the domain of the function
denoted by the expression.
7
Fall 2013 Ling 720
I will now use the definitions in (16) to show how our model in (13) can be used to assign
semantic values to meaningful expressions of Politics+λ.
• For reasons of space, the calculations below are greatly appreviated…
• Students should work out for themselves the full calculations based on (16)…
(18) Interpreting Expressions of Politics+λ

Let M be the model defined in (13). Let g be some arbitrary variable assignment based
on M.
a. ∀x3 (man’ x3)
(i) [[∀x3 (man’ x3)]]M,g = 1 iff

(ii) For all a ∈ De,E , i(a) = 1 iff
(iii) For all a ∈ {Michelle, Barack, Mitt},
{ <Michelle, 0>, <Barack,1>, <Mitt,1> }(a) = 1
(iv) Thus, [[∀x3 (man’ x3)]]M,g = 0
b. ∃P4 (P4 michelle’)
(i) [[ ∃P4 (P4 michelle’) ]]M,g = 1 iff

(ii) There is an a ∈ D<et>,E such that a(Michelle) = 1
(iii) Thus, [[ ∃P4 (P4 michelle’) ]]M,g = 1
Note: Even though neither h, i, nor k map Michelle to 1, there are still other functions f ∈ D<et>,E
such that f(Michelle) = 1
c. (λx3 ((man’ x3) & (smokes’ x3)))
(i) [[(λx3 ((man’ x3) & (smokes’ x3)))]]M,g =
(ii) The function p with domain De,E , range Dt,E, and for all a ∈ De,E ,
p(a) = 1 iff i(a) = 1 and h(a) = 1 =
(iv) The function p with domain {Mitt, Barack, Michelle}, range {0,1}, and
for all a ∈ {Mitt, Barack, Michelle},
p(a) = 1 iff { <Michelle, 0>, <Barack,1>, <Mitt,1> }(a) = 1 and
{ <Michelle, 0>, <Barack,1>, <Mitt,0> }(a) = 1 =
(v) { <Michelle, 0>, <Barack,1>, <Mitt,0> } =
(vi) The characteristic function of the set of ‘men who smoke’
8
Fall 2013 Ling 720
d. (λP4 (P4 mitt’))
(i) [[(λP4 (P4 mitt’))]]M,g =
(ii) The function p with domain D<et>,E , range Dt,E and for all a ∈ D<et>,E ,
p(a) = a(Mitt) =
(iii) The function p with domain D<et>,E , range Dt,E and for all a ∈ D<et>,E ,
p(a) = 1 iff a(Mitt) = 1
(iv) The characteristic function of the set of <et>-functions f such that

f(Mitt) = 1
(v) The characteristic function of the set of ‘properties that Mitt has’.
Note:
Given the result in (18d), we also have it that:
• [([(λP4 (P4 mitt’)) smokes’)]]M,g = p([[smokes’)]]M,g) = p(h) = h(Mitt) = 0
M,g
• [([(λP4 (P4 mitt’)) man’)]] = p([[man’)]]M,g) = p(i) = i(Mitt) = 1
e. (λP4 ∀x3 ( (man’ x3)  (P4 x3) ) )
(i) [[(λP4 ∀x3 ( (man’ x3)  (P4 x3)))]]M,g =
(ii) The function p with domain D<et>,E, range Dt,E and for all a ∈ D<et>,E ,
p(a) = 1 iff for all a’ ∈ De,E, either i(a’) = 0 or a(a’) = 1 =
(iii) The function p with domain D<et>,E, range Dt,E and for all a ∈ D<et>,E ,
p(a) = 1 iff for all a’ ∈ De,E, if i(a’) = 1 then a(a’) = 1
(iv) The characteristic function of the set of <et>-functions f such that if x

is ‘a man’, then f(x) = 1
(v) The characteristic function of the set of ‘properties every man has’
Note:
Given the result in (18e), we also have it that:
• [[((λP4 ∀x3 ( (man’ x3)  (P4 x3))) smokes’)]]M,g = p([[smokes’)]]M,g) = p(h) = 0
• [[((λP4 ∀x3 ( (man’ x3)  (P4 x3))) man’)]]M,g = p([[man’)]]M,g) = p(i) = 1
9
Fall 2013 Ling 720
3. A Key Logical Equivalence
The model-theoretic semantics introduced above yields a key logical equivalence, one that will
be of much use to us shortly…
(19) Notational Preliminary

If ϕ, ψ are meaningful expressions of a TL language L, ϕ ∈ MEτ and v ∈ VARτ , then
‘[ϕ/v]ψ’ is the meaningful expression just like ψ except that every free instance of v is
replaced with ϕ.
[barack’/x3] ((man’ x3) & (smokes’ x3)) = ((man’ barack’) & (smokes’ barack’))
[smokes’/P4]∀x3((man’ x3)  (P4 x3)) = ∀x3((man’ x3)  (smokes x3))
(20) Key Validity: Lambda Conversion

Let (λvψ) and ϕ be meaningful expressions with no variables in common, and let ϕ ∈
MEτ and v ∈ VARτ. It follows that the meaningful expressions in (a) and (b) are logically
equivalent:
a. ((λvψ) ϕ) b. [ϕ/v]ψ
(21) Illustration of Lambda Conversion

Each of the following pairs of meaningful expressions are logically equivalent.
a. ( (λx3 ((man’ x3) & (smokes’ x3))) barack’ ) ⇔

((man’ barack’) & (smokes’ barack’))
b. ( (λP4 ∀x3 ( (man’ x3)  (P4 x3))) smokes’ ) ⇔

∀x3((man’ x3)  (smokes x3))
(22) Informal Proof of Lambda Conversion

Let M be a model <E, I> of a TL language and g be a variable assignment based on M.
• If (λvψ) is such that v ∈VARτ and ψ ∈ MEσ, then by (16j), [[(λvψ)]]M,g is the
function p with domain Dτ,E and range Dσ,E such that for all a in Dτ,E, p(a) =
[[ψ]]M,g(v/a)
• If ϕ ∈ MEτ, then by (16) it follows that there is an a ∈ Dτ,E such that [[ϕ]]M,g = a.
• By (16c), it follows that [[((λvψ) ϕ)]]M,g = [[(λvψ)]]M,g([[ϕ]]M,g) = p(a) = [[ψ]]M,g(v/a)
• Finally, it is intuitively the case that [[ψ]]M,g(v/a) = [[ [ϕ/v]ψ ]]M,g

Thus, [[ ((λvψ) ϕ) ]]M,g = [[ [ϕ/v]ψ ]]M,g
10
Fall 2013 Ling 720
(23) a. Question:
Why must we assume in (20) that (λvψ) and ϕ have no variables in common?
b. Answer:
If ϕ contains a free variable v’ which is bound in ψ, then it can sometimes happen
that v’ is free in ((λvψ) ϕ) but bound in [ϕ/v]ψ.
Illustration: ((λvt,3 ∃x2((Px2) & vt,3)) (Qx2)) is not logically equivalent to:
∃x2((Px2) & (Qx2))
(24) The Utility of Lambda Conversion
As the semanticists in the audience are no doubt aware, we will be making much use of
lambda conversion when it comes time to translate structures of English to structures of
Politics+λ.
• Our translation process will output rather complex formulae of Politics+λ.
• Lambda conversion will allow us to ‘convert’ those complex formulae into simpler,
logically equivalent formulae.
o These simpler formulae will more transparently represent the ‘meanings’ that
our translation ends up assigning to the expressions of English.
Rough Illustration:
In the rough illustration below, imagine that h is our translation function from English to
Politics+λ.
(i) h(every man smokes) =
(ii) HMerge-S( h(every man), h(smokes) ) =
(iii) HMerge-S( (λP4 ∀x3 ( (man’ x3)  (P4 x3))), smokes’ ) =
(iv) ( (λP4 ∀x3 ( (man’ x3)  (P4 x3))) smokes’ ) ⇔
(v) ∀x3((man’ x3)  (smokes x3))
11
Fall 2013 Ling 720
4. The Logical Language Politics+λ: Montagovian Presentation
In (8), we provide a relatively informal, non-Montagovian definition of the language Politics+λ.
• We will now represent Politics+λ as a ‘disambiguated language’, in the sense of

Montague (1974).
• In the next handout, we will develop a Montagovian interpretation for Politics+λ,

based upon this definition…
(25) The Vocabulary of Politics+λ


(i) Constants:
CONe = { mitt’, barack’, michelle’ }
(ii) Variables:
VARτ = { vτ, n : n ∈ ℕ }
(26) The Syntactic Operations of Politics+λ
a. FConcat(α, β) = (α β)
b. FNot(α) = ~α
c. FAnd(α, β) = (α & β)
d. FOr(α, β) = (α ∨ β)
e. FIf(α, β) = (α  β)
f. F∃ (α, β) = ∃α β
g. F∀ (α, β) = ∀α β
h. Fλ (α, β) = (λα β)
12
Fall 2013 Ling 720
(27) The Syntactic Algebra of Politics+λ
<A, Fγ >γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ} is the algebra such that:
(i) { Fγ }γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ} are as defined in (26)
(ii) A is the smallest set such that the following holds:
1. For all τ ∈ T, CONτ ⊆ A and VARτ ⊆ A
2. A is closed under { Fγ }γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ}
Note:
Thus, the set A contains all the constants and variables in (25), and is closed under the syntactic
operations in (26).
(28) The Syntactic Category Labels of Politics+λ

Let T be the set of types. Δ = T ∪ { <var,τ> : τ ∈ T }
Note:
Thus, the syntactic category labels include (i) all the types, and (ii) for every type τ ∈ T, the
‘variable type’ <var, τ>.
(29) The Basic Expressions of Politics+λ

For every type τ ∈ T:
a. X<var,τ> = VARτ
b. Xτ = CONτ ∪ X<var,τ>
Note: Thus, it follows that:

• X<var,e> = { ve, n : n ∈ ℕ } = { xn : n ∈ ℕ }
• Xe = { mitt’, barack’, michelle’ } ∪ { xn : n ∈ ℕ }
• X<var,<e,t>> = { v<e,t>, n : n ∈ ℕ } = { Pn : n ∈ ℕ }
• X<e,t> = { smokes’, man’, president’ } ∪ { Pn : n ∈ ℕ }
• X<var,<e,<e,t>>> = { v<e,<e,t>>, n : n ∈ ℕ }
• X<e,<e,t>> = { loves’ } ∪ { v<e,<e,t>>, n : n ∈ ℕ }
• For all other types τ ∈ T, X<var,τ> = VARτ = { vτ, n : n ∈ ℕ } = Xτ
13
Fall 2013 Ling 720
(30) The Syntactic Rules of Politics+λ
The (countably infinite) set S consists of (a) and (b) below:
a. The following triples: (i) <FNot , <t>, t>

(ii) <FAnd , <t,t>, t>
(iii) <FOr , <t,t>, t>
(vi) <FIf , <t,t>, t>
b. For every σ, τ ∈ T, a triple of the following form:

(i) < FConcat , < <σ, τ>, σ >, τ >
(ii) < F∃ , < <var,σ>, t >, t >
(iii) < F∀ , < <var,σ>, t >, t >
(iv) < Fλ , < <var,σ>, τ >, <σ, τ> >
Note: Rules (30bii), (30biii), and (30biv) can be informally read as the following, each of which
mirrors an informal syntactic rule in (7d,e).
• “Applying F∃ to a variable of category <var,σ> and an expression of category t forms an

expression of category t.”
• “Applying F∀ to a variable of category <var,σ> and an expression of category t forms an

expression of category t.”
• “Applying Fλ to a variable of category <var,σ> and an expression of category τ forms an

expression of category <σ, τ>.”
(31) The Disambiguated Language Politics+λ
Politics+λ is the structure <A, Fγ, Xδ , S, t>γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ}, δ ∈ Δ such that A,
{ Fγ }γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ}, { Xδ }δ ∈ Δ, S, Δ are all as defined in (25)-(30)
Note: Politics+λ is a disambiguated language.
We will now illustrate the definitions above by examining some meaningful expressions of (31),
with associated calculations.
14
Fall 2013 Ling 720
(32) Illustrative ME: ~ ( ( loves’ mitt’ ) x3 ) ∈ Ct
(i) loves’ ∈ X<e<e,t>>, mitt’, x3 ∈ Xe (by (29))

(ii) loves’ ∈ C<e<e,t>>, mitt’, x3 ∈ Ce (by def. of {Cδ}δ ∈ Δ)
(iii) < FConcat , < <e,<e,t>>, e >, <e,t> > ∈ S (by (30))
(iv) FConcat(loves’, mitt’) ∈ C<e,t> (by def. of {Cδ}δ ∈ Δ)
(v) (loves’ mitt’) ∈ C<e,t> (by def. of FConcat)
(vi) < FConcat , < <e,t>, e >, t > ∈ S (by (30))
(vii) FConcat((loves’ mitt’), x3) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(viii) ( ( loves’ mitt’ ) x3 ) ∈ Ct (by def. of FConcat)
(ix) <FNot , <t>, t> ∈ S (by (30))
(x) FNot((( loves’ mitt’ ) x3 )) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(xi) ~ ( ( loves’ mitt’ ) x3 ) ∈ Ct (by def. of FNot)
(33) Illustrative ME: ∀x3 ( (smokes’ x3)  ~ ( ( loves’ mitt’ ) x3) ) ∈ Ct
(i) smokes’ ∈ X<e,t>, x3 ∈ Xe , x3 ∈ X<var,e> (by (29))

(ii) smokes’ ∈ C<e,t>, x3 ∈ Ce, x3 ∈ C<var,e> (by def. of {Cδ}δ ∈ Δ)
(iii) < FConcat , < <e,t>, e >, t > ∈ S (by (30))
(iv) FConcat(smokes’, x3) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(v) (smokes’ x3) ∈ Ct (by def. of FConcat)
(vi) ~ ( ( loves’ mitt’ ) x3 ) ∈ Ct (by (32))
(vii) <FIf , <t,t>, t> ∈ S (by (30))
(viii) FIf((smokes’ x3), ~((loves’ mitt’) x3)) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(ix) ( (smokes’ x3)  ~ ( ( loves’ mitt’ ) x3) ) ∈ Ct (by def. of FIf)
(x) < F∀ , < <var,e>, t >, t > ∈ S (by (30))
(xi) F∀ (x3, ((smokes’ x3)  ~((loves’ mitt’) x3)) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(xii) ∀x3 ( (smokes’ x3)  ~ ( ( loves’ mitt’ ) x3) ) ∈ Ct (by def. of F∀)
(34) Illustrative ME: ∃P4 (P4 michelle’) ∈ Ct
(i) michelle’ ∈ Xe , P4 ∈ X<e,t>, P4 ∈ X<var,<e,t>> (by (29))

(ii) michelle’ ∈ Ce , P4 ∈ C<e,t>, P4 ∈ C<var,<e,t>> (by def. of {Cδ}δ ∈ Δ)
(iv) FConcat(P4, michelle’) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(v) (P4 michelle’) ∈ Ct (by def. of FConcat)
(vi) < F∃ , < <var,<e,t>>, t >, t > ∈ S (by (30))
(vii) F∃ (P4, (P4 michelle’)) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(viii) ∃P4 (P4 michelle’) ∈ Ct (by def. of F∃)
15
Fall 2013 Ling 720
(35) Illustrative ME: ((λx3 (man’ x3)) mitt’) ∈ Ct
(i) man’ ∈ X<e,t>, mitt’, x3 ∈ Xe , x3 ∈ X<var,e> (by (29))

(ii) man’ ∈ C<e,t>, mitt’, x3 ∈ Ce , x3 ∈ C<var,e> (by def. of {Cδ}δ ∈ Δ)
(iv) FConcat(man’, x3) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(v) (man’ x3) ∈ Ct (by def. of FConcat)
(vi) < Fλ , < <var,e>, t >, <e, t> > ∈ S (by (30))
(vii) Fλ(x3, (man’ x3)) ∈ C<e,t> (by def. of {Cδ}δ ∈ Δ)
(viii) (λx3 (man’ x3)) ∈ C<e,t> (by def. of Fλ)
(ix) FConcat((λx3 (man’ x3)), mitt’) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(x) ((λx3 (man’ x3)) mitt’) ∈ Ct (by def. of FConcat)
(36) Illustrative ME: (λP4 ∀x3 ((man’ x3)  (P4 x3))) ∈ C<<e,t>,t>
(i) man’ ∈ X<e,t>, x3 ∈ Xe , X<var,e> , P4 ∈ X<e,t>, X<var,<e,t>> (by (29))

(ii) man’ ∈ C<e,t>, x3 ∈ Ce , C<var,e> , P4 ∈ C<e,t>, C<var,<e,t>> (by def. of {Cδ}δ ∈ Δ)
(iv) FConcat(man’, x3) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(v) (man’ x3) ∈ Ct (by def. of FConcat)
(vi) FConcat(P4, x3) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(vii) (P4 x3) ∈ Ct (by def. of FConcat)
(viii) <FIf , <t,t>, t> ∈ S (by (30))
(viii) FIf((man’ x3), (P4 x3)) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(ix) ((man’ x3)  (P4 x3)) ∈ Ct (by def. of FIf)
(x) < F∀ , < <var,e>, t >, t > ∈ S (by (30))
(xi) F∀ (x3, ((man’ x3)  (P4 x3))) ∈ Ct (by def. of {Cδ}δ ∈ Δ)
(xii) ∀x3 ((man’ x3)  (P4 x3)) ∈ Ct (by def. of F∀)
(xiii) < Fλ , < <var,<e,t>>, t >, <<e,t>,t> > ∈ S (by (30))
(xiv) Fλ ( P4 , ∀x3 ((man’ x3)  (P4 x3))) ∈ C<<e,t>,t> (by def. of {Cδ}δ ∈ Δ)
(xv) (λP4 ∀x3 ((man’ x3)  (P4 x3))) ∈ C<<e,t>,t> (by def. of Fλ)
16
Fall 2013 Ling 720
An Algebraic Approach to Quantification and Lambda Abstraction:
Fregean Interpretations 1
(1) The Disambiguated Language Politics+λ

Politics+λ is the disambiguated language <A, Fγ, Xδ, S, t>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ}, δ∈Δ
such that
a. Non-Logical Vocabulary
(i) Constants:
CONe = { mitt’, barack’, michelle’ }
(ii) Variables:
VARτ = { vτ, n : n ∈ ℕ }
b. Syntactic Algebra:
<A, Fγ >γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ} is the algebra such that:
(i) { Fγ }γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ} are as defined on the previous handout
(ii) A is the smallest set such that the following holds:
1. For all τ ∈ T, CONτ ⊆ A and VARτ ⊆ A
2. A is closed under { Fγ }γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ}
c. Syntactic Categories: Δ = T ∪ { <var,τ> : τ ∈ T }
d. Basic Expressions: For every type τ ∈ T:

(i) X<var,τ> = VARτ
(ii) Xτ = CONτ ∪ X<var,τ>
e. Syntactic Rules:
S is the (countably) infinite set consisting of:
(i) <FNot , <t>, t> (ii) <FAnd , <t,t>, t>
(iii) <FOr , <t,t>, t> (vi) <FIf , <t,t>, t>
And, for every σ, τ ∈ T, a triple of the following form:
(i) < FConcat , < <σ, τ>, σ >, τ > (ii) < F∃ , < <var,σ>, t >, t >
(iii) < F∀ , < <var,σ>, t >, t > (iv) < Fλ , < <var,σ>, τ >, <σ, τ> >
1
These notes are based upon material in the following readings: Thomason (1974) Chapter 7 (Montague’s
“Universal Grammar”).
1
Fall 2013 Ling 720
(2) Our Present Goal: Define an interpretation for Politics+λ.
(3) The Crucial Challenge

To have an interpretation B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} for Politics+λ, we
need to state some semantic (algebraic) operations G∃ , G∀ , and Gλ , which will
‘correspond’ with F∃ , F∀ , and Fλ. What in the world could those be?
(4) The Plan
• As we did several weeks ago for FOL, we’re going to see our way to a solution by
reflecting informally on the meaning of deictic elements (akin to free pronouns).
• This will bring us to Montgue’s definition of a Fregean Interpretation
• We’ll use part of this definition to set up an interpretation structure for Politics+λ.
1. Some Reflections on the Nature of Meaning
(5) Illustrative Sentence with Free (Deictic) Pronoun: He smokes.
(6) a. Stupid Question

In a context where I’m pointing to Barack, does sentence (5) have a meaning?
b. Obvious Answer:
Sure; after all, in such a context, (5) seems to have the intension in (i) and the
extension in (ii).
(i) Intension of (5), Pointing to Barack: [λw : Barack smokes in w]
(ii) Extension of (5), Pointing to Barack: 1 (true)
(7) a. Difficult Question

Suppose I don’t specify a context for (5). Do we still say it has a meaning?
b. Negative Answer
Without specifying a context, (5) doesn’t have a defined intension or extension.
Thus, sentence (5) seems to be meaningless.
c. Positive Answer
To say that (5) has no meaning in such circumstances would fail to distinguish
between it and gibberish like “snoochie boochies”.
• Thus, its potential to have an intension/extension (relative to a context)
should be viewed as constitutive of its meaning.
2
Fall 2013 Ling 720
(8) A General Picture of Meaning, Fitting the Positive Answer

The ‘meaning’ – broadly construed – of an expression is a function from contexts to
intensions. (cf. Kaplan’s notion of “character” vs. “content”).
Meaning Intension (Sense) Extension (Reference)

Context World/time
of evaluation
• Some expressions have a meaning that maps every context to the same intension.
o We can say that such expressions have a meaning that is ‘not context-dependent’
Illustration: Barack smokes.

Meaning: Function that maps every context c to
[λw : Barack smokes in w]
• Some expressions have a meaning that maps different contexts to different intensions.
o We can say that such expressions have a ‘context-dependent’ meaning.
Illustration: I smoke.
[λw : the speaker of c smokes in w]
Illustration: He smokes.
[λw : the deictic focus of c smokes in w]
(9) Where We’re Going From This
• We’re going now to try to model meanings (in B) as functions from contexts (variable
assignments) to intensions (functions from possible worlds to extensions).
• We’ll begin by redefining our system of types and denotations, so that they now
include objects that can serve as ‘intensions’.
(10) The Types (New Definition)

a. e∈T (entities)
b. t∈T (truth-values)
c. If σ ∈ T and τ ∈ T, then <σ, τ> ∈ T (functions from σ to τ)
d. If σ ∈ T, then <s, σ> ∈ T (functions from world-times to σ)
3
Fall 2013 Ling 720
(11) The Denotations (New Definition)
Let T be the set of types, let E be some non-empty set (of entities), and let I be some
other non-empty set (of ‘indices’ / world-time pairs). The set Dτ, E, I of denotations of
type τ based on E and I is defined as follows:
a. De,E,I = E ∪ {garbage}
b. Dt,E,I = { 0, 1 }
c. If σ, τ ∈ T then D<σ,τ>,E,I = the set of functions from Dσ ,E,I to Dτ,E,I
d. If σ ∈ T then D<s,σ >,E,I = the set of functions from I to Dσ ,E,I
= (Dσ ,E,I)I
(12) Remark
The only real change to our earlier definition for the set of ‘denotations’ comes in (11d),
which yields the ‘intensional objects’, including the following:
a. D<s,e>,E,I = Functions from I (world-time pairs) to E (entities).

= Individual Concepts
b. D<s,t>,E,I = Functions from I (world-time pairs) to {0,1}

= Propositions
c. D<s,<e,t>>,E,I = Functions from I (world-time pairs) to D<e,t>,E,I

= Properties
(13) Some Considerations Towards ‘Meanings’
• Following (8) and (9), we want ‘meanings’ to be functions from contexts (variable
assignments) to intensions (functions from world-times to the denotations in (11)).
• Thus, if J is our set of contexts (variable assignments) and I is our set of world-time
pairs, then a meaning should be something from the following set:
( (∪τ ∈ T Dτ, E, I)I )J

Functions from J to functions from I to the set of all possible denotations
• But recall from our first week that – due to the nature of currying – we can easily shift
×
between the following two objects: (AB)C and AB C
• Thus, we can – and Montague does – state that meanings are elements from the
following set:
I×J
(∪τ ∈ T Dτ, E, I)
Functions from the pairs in (I×J) to the set of all possible denotations
4
Fall 2013 Ling 720
(14) The Set of Possible Meanings

Let T be the set of types, let E, I and J be non-empty sets (E = entities; I = indices/world-
times; J = contexts/variable-assignments). The set Mτ, E, I, J of meanings of type τ based on
I×J
E, I, and J is (Dτ, E, I)
I×J
Note: In UG, Montague sometimes shifts freely between (Dτ, E, I) and ( (Dτ, E, I)I )J
With this conception of meaning at hand, we can begin to define Montague’s notion of a
‘Fregean Interpretation’…
A key ingredient in that definition is the notion of a ‘type assignment’…..
(15) Type Assignment

Let T be the set of types in (10). Let L be a language <<A, Fγ, Xδ, S, δ0>γ ∈ Γ, δ ∈ Δ, R>.
A type assignment for L is a function σ: Δ  T such that σ(δ0) = t.
Note: Thus, a type assignment pairs each syntactic category label δ with exactly one type τ, and
it ensures that δ0 (the category label for ‘declarative sentences’) is paired with type t.
(16) The Core Idea Behind a Fregean Interpretation

Informally speaking, Montague’s Fregean interpretations are intended to be
interpretations B = <B, Gγ, f>γ ∈ Γ such that:
a. B is a set of ‘meanings’, as defined in (14); i.e., B ⊆ ∪τ ∈ T Mτ, E, I, J
b. Expressions of the same category are assigned meanings of the same type.
• If g is the meaning assignment determined by B, then for each syntactic
category Cδ, there is a single type τ such that if ϕ ∈ Cδ , then g(ϕ) ∈ Mτ, E, I, J
(17) Question: Why have this condition in (16b)?
• Right from the start, (16b) was criticized as too restrictive. For example, it entails that
“Barack” and “some man” must have the same type of denotation (and so the former
cannot directly refer to Barack).
• To my knowledge, the restriction in (16b) doesn’t ‘do’ anything.

o It isn’t necessary for the meaning assignment g to be a homomorphism…
o It’s just stipulated in UG and PTQ; there isn’t much independent motivation
(18) Possible Answer (?)

It makes the mapping between the syntax and semantics akin to that of a typed logic?
5
Fall 2013 Ling 720
(19) Official Formal Definition of a Fregean Interpretation
Let L be a language <<A, Fγ, Xδ, S, δ0>γ ∈ Γ, δ ∈ Δ, R>. A Fregean interpretation B of L
is an interpretation <B, Gγ, f>γ ∈ Γ of L such that for some type assignment σ and non-
empty sets E, I, J:
a. B ⊆ ∪τ ∈ T Mτ, E, I, J
Note: Thus, the semantic values in B will all be ‘meanings’ in the sense of (14).
Thus, the meaning assignment based on B maps expressions of L to
(i) functions from <context,world-time> pairs to ∪τ ∈ T Dτ, E, I
(ii) functions from contexts to
functions from world-time pairs to ∪τ ∈ T Dτ, E, I
• This clearly guarantees us property (16a).
b. For all δ ∈ Δ, if ϕ ∈ Xδ , then f(ϕ) ∈ Mσ(δ), E, I, J
Note:
That is, if category δ is paired with type τ by the type assignment σ, then the
lexical interpretation function f maps every basic expression of category δ to a
meaning of type τ.
• This, combined with the next condition, guarantees property (16b).
c. If < Fγ , <δ1, … δn>, δ> ∈ S, and b1, … bn are such that for all i, bi ∈ Mσ(δi), E, I, J,
then Gγ(b1, … bn) ∈ Mσ(δ), E, I, J.
Note:
That is, if categories δ1, … δn, δ are paired with types τ1, …, τn, τ by σ, then
If the result of applying syntactic operation Fγ to expressions of category
δ1, … δn is an expression of Cδ, then
The result of applying the corresponding semantic operation Gγ to
meanings of type τ1, …, τn is a meaning of type τ.
• This, combined with the preceding condition, guarantees (16b).
Note: We’ll also want in (19) to ensure that f and g do not end up mapping any meaningful
expressions of L to garbage, but we’ll leave the details of that implicit in what follows…
(20) Concomitant Definition

A Fregean interpretation for L connected with E, I, J, σ is an interpretation B such that
the conditions in (19a-c) hold.
6
Fall 2013 Ling 720
2. A ‘Partly-Fregean’ Interpretation for Politics+λ
Now, in Politics+λ there are no ‘intensional’ environments…

Thus, we won’t be making full use of the definition in (19) to interpret Politics+λ…
Let us, then, develop our own notion of a ‘Partly-Fregean’ interpretation, which will be quite
applicable to our language in (1)…
(21) The Key Ideas Behind a ‘Partly-Fregean’ Interpretation

Since we won’t be dealing with intensions for now, let us view meanings as functions
from contexts (variable assignments) to denotations.
Meaning Extension (Reference)

Context (Variable Assignment)
a. The Types (For Politics+λ):

(i) e∈T (entities)
(ii) t∈T (truth-values)
(iii) If σ ∈ T and τ ∈ T, then <σ, τ> ∈ T (functions from σ to τ)
b. The Possible Denotations (For Politics+λ):

Let T be the set of types and E be some non-empty set (of entities). The set Dτ, E
of denotations of type τ based on E is defined as follows:
(i) De,E = E ∪ {garbage}
(ii) Dt,E = { 0, 1 }
c. The Possible Meanings (For Politics+λ):

Let T be the set of types, and let E and J be non-empty sets (E = entities; J =
contexts/variable-assignments). The set Mτ, E, J of meanings of type τ based on E
J
and J is (Dτ, E)
(22) Partly-Fregean Interpretation

Let L be a language <<A, Fγ, Xδ, S, δ0>γ ∈ Γ, δ ∈ Δ, R>. A partly-Fregean interpretation B
of L is an interpretation <B, Gγ, f>γ ∈ Γ of L such that for some type assignment σ and
non-empty sets E, J:
a. B ⊆ ∪τ ∈ T Mτ, E, J
b. For all δ ∈ Δ, if ϕ ∈ Xδ , then f(ϕ) ∈ Mσ(δ), E, J
c. If < Fγ , <δ1, … δn>, δ> ∈ S, and b1, … bn are such that for all i, bi ∈ Mσ(δi), E, J,
then Gγ(b1, … bn) ∈ Mσ(δ), E, J.
7
Fall 2013 Ling 720
Note: Our definition of a ‘partly-Fregean’ interpretation simply takes Montague’s definition of

a Fregean interpretation (19) and removes the set I (indices / world-time pairs).
(23) Concomitant Definition

A partly-Fregean interpretation for L connected with E, J, σ is an interpretation B such
that the conditions in (22a-c) hold.
(24) Goal: Define a partly-Fregean interpretation for Politics+λ, where J is the set of possible
variable assignments.
(25) Immediate Issue:

• Up to now, variable assignments have been defined with respect to a given model.
• Thus, we need a new definition of ‘variable assignment’, one that makes no reference
to a model…
(26) New Definition of Variable Assignment (for Politics+λ)

Let E be a non-empty set (of entities). A variable assignment based on E is a function g
satisfying the conditions below:
(ii) Range of g: If α ∈ VARτ , then g(α) ∈ Dτ,E (and g(α) ≠ garbage)
Note: Variable assignment map variables directly to denotations, not to meanings.
(27) The Central, Scary Question that We’ve Been Putting Off Until Now
How do we build a semantics for Politics+λ, where the semantic values (‘meanings’) are
functions from variable assignments to denotations?
(28) Some Informal Considerations to Get Us Started
a. The meaning of a constant maps every variable assignment to the same denotation
• Just like the English name Barack, the individual constant barack’ should
map every variable assignment to the same individual, Barack
o After all, in our model-theoretic semantics, [[barack’]]M,g doesn’t
depend on g…
b. The meaning of a variable can map different variable assignments to different

denotations.
o After all, in our model-theoretic semantics, [[x3]]M,g does depend on g
8
Fall 2013 Ling 720
(29) Formalizing the Informal Considerations

a. If ϕ ∈ CONτ , then it should be that f(ϕ)(g) = f(ϕ)(g’), for all g, g’ ∈ J.
b. If ϕ ∈VARτ , then it should be that f(ϕ)(g) = g(ϕ) for all g ∈ J.
Note: If (29) holds, then

(i) The meaning of a constant maps every variable assignment g to the same denotation.
(ii) The meaning of a variable v maps a variable assignment to g(v).
The meaning of v can map different variable assignments to different denotations.
Let’s build on the ideas in (28)-(29), to define a special sub-class of partly-Fregean

interpretations…
2
(30) Logically Possible Partly-Fregean Interpretation of Politics+λ
A logically possible partly-Fregean interpretation of Politics+λ is a partly-Fregean

interpretation B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} for Politics+λ connected with E,
J, σ, such that:
a. J is the set of variable assignments based on E.
b. If ϕ ∈ CONτ , then f(ϕ)(g) = f(ϕ)(g’), for all g, g’ ∈ J.
c. If ϕ ∈VARτ , then f(ϕ)(g) = g(ϕ) for all g ∈ J.
d. …. To be filled in later …
Finally, to complete the definition in (30), we will specify what the semantic operations in
{Gγ}γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} must be…
This, of course, is the most challenging part…
• Rather than motivate these operations from the ‘ground up’, I’m simply going to
introduce them and show you how they work…
• Note that the set B = ∪ τ ∈ T Mτ , E, J is closed under each of the following

operations Gγ .
2
The term in (30) and its definition mimic Montague’s definition in UG of a ‘logically possible Fregean
interpretation’ of his Intensional Logic.
9
Fall 2013 Ling 720
(31) The Operation GConcat
a. If α ∈ M<σ,τ>, E, J and β ∈ Mσ, E, J, and β ≠ garbage, then GConcat(α,β) =

The function F-A such that if g ∈ J, then F-A(g) = α(g)(β(g))
b. Otherwise, GConcat(α,β) = garbage
(32) How Does This Work?
• Consider the following two sentences: (smokes’ barack’) (smokes’ x3)
• Since FConcat and GConcat should ‘correspond’, it follows for meaning assignment h:
h((smokes’ barack’)) = h((smokes’ x3)) =

h(FConcat(smokes’, barack’) = h(FConcat(smokes’, x3) =
GConcat(h(smokes’), h(barack’)) = GConcat(h(smokes’), h(x3)) =
GConcat(f(smokes’), f(barack’)) = GConcat(f(smokes’), f(x3)) =
• Now, given (22), in a partly-Fregean interpretation of Politics+λ:

f(smokes’) ∈ M<e,t>, E, J f(barack’), f(x3) ∈ Me,, E, J
• Thus, given (31a), it follows that:
GConcat(f(smokes’), f(barack’)) =
The function F-A such that if g ∈ J, then F-A(g) = f(smokes’)(g)(f(barack’)(g))
GConcat(f(smokes’), f(x3)) =
The function F-A’ such that if g ∈ J, then –FA’(g) = f(smokes’)(g)(f(x3)(g))
• Since smokes’ and barack’ are constants, and x3 is a variable, (30b,c) entails that:
o For any g, g’ ∈ J, f(smokes’)(g) = f(smokes’)(g’) = s ∈ D<e,t>, E
o For any g, g’ ∈ J, f(barack’)(g) = f(barack’)(g’) = b ∈ De, E
o For any g, ∈ J, f(x3)(g) = g(x3)
• Thus, it therefore follows that:
h((smokes’ barack’)) = GConcat(f(smokes’), f(barack’)) =

The function F-A such that if g ∈ J, then F-A(g) = s(b)
h((smokes’ x3)) = GConcat(f(smokes’), f(x3)) =

The function F-A’ such that if g ∈ J, then F-A’(g) = s(g(x3))
10
Fall 2013 Ling 720
(33) Remarks
• Thus, the ‘meaning’ of (smokes’ barack’) is a constant function from contexts

(variable assignments) to truth-values:
o It maps every variable assignment to the value that s yields for b.
• Thus, the ‘meaning’ of (smokes’ x3) is not a constant function from contexts (variable
assignments) to truth-values:
o It maps every variable assignment to the value that s yields for g(x3)
• Parallel with Demonstratives:

o The ‘meaning’ of Barack smokes maps every context to the same intension
o The ‘meaning’ of He smokes maps different contexts to different intensions
• Parallel with Models:

o No matter what variable assignment g is chosen [[(smokes’ barack’)]]M,g will
be the same value: I(smokes’)(I(barack’))
o The value of [[(smokes’ x3)]]M,g will equal I(smokes)(g(x3)), and so will vary
with the variable assignment g.
(34) The Operation GNot

a. If α ∈ Mt, E, J then GNot(α) =
The function Neg such that if g ∈ J, then Neg(g) = 1 iff α(g) = 0
b. Otherwise, GNeg(α) = garbage
(35) How Does This Work? (Part 1)
• Consider the following two sentences: ~(smokes’ barack’) ~(smokes’ x3)
• Since FNot and GNot should ‘correspond’, it follows for meaning assignment h:
h(~(smokes’ barack’)) = h(~(smokes’ x3)) =

h(FNot((smokes’ barack’)) = h(FNot((smokes’ x3)) =
GNot(h(smokes’ barack’)) = GConcat(h(smokes’ x3)) =
• Now, recall from (32) that h(smokes’ barack’), h((smokes’ x3)) ∈ Mt, E, J
Thus, it follows from (34) that:
GNot(h(smokes’ barack’)) =
The function Neg such that if g ∈ J, Neg(g) = 1 iff h(smokes’ barack’)(g) = 0
GConcat(h(smokes’ x3)) =
The function Neg’ such that if g ∈ J, Neg’(g) = 1 iff h(smokes’ x3)(g) = 0
11
Fall 2013 Ling 720
• Now, also recall from (32) that:

h((smokes’ barack’)) = The function F-A such that if g ∈ J, then F-A(g) = s(b)
h((smokes’ x3)) = The function F-A’ such that if g ∈ J, then F-A’(g) = s(g(x3))
• It thus follows that:
h(~(smokes’ barack’)) = GNot(h(smokes’ barack’)) =

The function Neg such that if g ∈ J, Neg(g) = 1 iff s(b) = 0
h(~(smokes’ x3)) = GConcat(h(smokes’ x3)) =

The function Neg’ such that if g ∈ J, Neg’(g) = 1 iff s(g(x3)) = 0
(37) Remarks
• Thus, the ‘meaning’ of ~(smokes’ barack’) is a constant function from contexts

o It maps every variable assignment to 1 iff s(b) = 0
• Thus, the ‘meaning’ of ~(smokes’ x3) is not a constant function from contexts
o It maps every variable assignment to 1 iff s(g(x3)) = 0
• Parallel with Demonstratives:

o The ‘meaning’ of Barack doesn’t smoke maps every context to the same
intension
o The ‘meaning’ of He doesn’t smoke maps different contexts to different
intensions

o No matter what variable assignment g is chosen [[~(smokes’ barack’)]]M,g
will be the same value: 1 iff I(smokes’)(I(barack’)) = 0
o The value of [[~(smokes’ x3)]]M,g will equal 1 iff I(smokes)(g(x3)) = 0, and so
will vary with the variable assignment g.
(38) The Operation GAnd
a. If α ∈ Mt, E, J and β ∈ Mt, E, J, then GAnd(α,β) =

The function Conj such that if g ∈ J, then Conj(g) = 1 iff α(g) = β(g) = 1
b. Otherwise, GAnd(α) = garbage
12
Fall 2013 Ling 720
• Consider the following two conjunctions:

((smokes’ barack’) & (smokes’ mitt’)) ((smokes’ barack’) & (smokes’ x3))
• Since FAnd and GAnd should ‘correspond’, it follows for meaning assignment h:
h(((smokes’ barack’) & (smokes’ mitt’))) =

GAnd( h((smokes’ barack’)), h((smokes’ mitt’))
h(((smokes’ barack’) & (smokes’ x3))) =

GAnd( h((smokes’ barack’)), h((smokes’ x3))
• Now, recall that h(smokes’ barack’), h(smokes’ mitt’), h((smokes’ x3)) ∈ Mt, E, J
Thus, it follows from (38) that:
GAnd( h((smokes’ barack’)), h((smokes’ mitt’)) =

The function Conj such that if g ∈ J, then Conj(g) = 1 iff
h((smokes’ barack’))(g) = h((smokes’ mitt’)(g) = 1
GAnd( h((smokes’ barack’)), h((smokes’ x3)) =

The function Conj such that if g ∈ J, then Conj(g) = 1 iff
h((smokes’ barack’))(g) = h((smokes’ x3)(g) = 1
• Next, recall that from (32) – and a parallel computation – that:

h((smokes’ barack’)) = The function F-A such that if g ∈ J, then F-A(g) = s(b)
h((smokes’ mitt’)) = The function F-A such that if g ∈ J, then F-A(g) = s(m)
h((smokes’ x3)) = The function F-A’ such that if g ∈ J, then F-A’(g) = s(g(x3))
• Consequently:
h(((smokes’ barack’) & (smokes’ mitt’))) =
The function Conj such that if g ∈ J, then Conj(g) = 1 iff s(b) = s(m) = 1
h(((smokes’ barack’) & (smokes’ x3))) =

The function Conj such that if g ∈ J, then Conj(g) = 1 iff s(b) = s(g(x3)) = 1
(40) Remarks
• Again, we have it that h(((smokes’ barack’) & (smokes’ mitt’))) is a constant

function from variable assignment to truth-values.
o Just as [[((smokes’ barack’) & (smokes’ mitt’))]]M,g doesn’t vary with g
• Again, we have it that h(((smokes’ barack’) & (smokes’ x3))) is not a constant
function from variable assignment to truth-values.
o Just as [[((smokes’ barack’) & (smokes’ x3))]]M,g does vary with g
13
Fall 2013 Ling 720
Given the discussion in (39)-(40), it can easily be seen that the following operations will
similarly serve as semantic correlates of FOr and FIf
(41) The Operation GOr

a. If α ∈ Mt, E, J and β ∈ Mt, E, J, then GOr(α,β) =
The function Disj such that if g ∈ J, then Disj(g) = 1 iff α(g) =1 or β(g) = 1
b. Otherwise, GOr(α) = garbage
(42) The Operation GIf

a. If α ∈ Mt, E, J and β ∈ Mt, E, J, then GIf(α,β) =
The function Cond such that if g ∈ J, then Cond(g) = 1 iff α(g) =0 or β(g) = 1
b. Otherwise, GIf(α) = garbage
Now, we finally come to the central problem of how to treat the variable-binding operators…
(43) The Operation G∃

a. If there is a type τ ∈ T and a variable v ∈ VARτ such that for all g ∈ J,
α(g) = g(v), and β ∈ Mt, E, J, then
G∃(α,β) = The function E such that if g ∈ J, E(g) = 1 iff
there is an x ∈ Dτ,E such that β(g(v/x)) = 1
b. Otherwise, G∃(α,β) = garbage
• First, note the requirement that there be a type τ ∈ T and a variable v ∈ VARτ such
that for all g ∈ J, α(g) = g(v).
• Given our conditions in (30b,c), this will only ever hold (in a logically possible
partly-Fregean interpretation) if α = f(v), i.e., the ‘meaning’ of the variable v.
• Thus, this condition ensures that if the first argument of G∃ is not a variable-meaning,
then the output will be garbage.
o Thus, our system will map syntactic garbage like ‘∃mitt’ (smokes’ mitt’) to
garbage
14
Fall 2013 Ling 720
• Now, consider the simple existential sentence: ∃x3 (smokes’ x3)
• Since F∃ and G∃ should ‘correspond’, it follows that for meaning assignment h:

h( ∃x3 (smokes’ x3) ) = h( F∃(x3 , (smokes’ x3)) = G∃( h(x3), h((smokes’ x3)) )
• Next, note that h((smokes’ x3)) ∈ Mt, E, J and for any g ∈ J, h(x3)(g) = f(x3)(g) = g(x3)
• Thus: G∃( h(x3), h((smokes’ x3)) ) =

The function E such that if g ∈ J, E(g) = 1 iff
there is an x ∈ De,E such that h((smokes’ x3))(g(x3/x)) = 1
• Now, recall that for any g ∈ J, h((smokes’ x3))(g) = s(g(x3))
• Thus: G∃( h(x3), h((smokes’ x3)) ) =

The function E such that if g ∈ J, E(g) = 1 iff
there is an x ∈ De,E such that s(g(x3/x)(x3)) = 1
• Finally, by definition of ‘g(x3/x)’, we have it that:

h( ∃x3 (smokes’ x3) ) = The function E such that if g ∈ J, E(g) = 1 iff
there is an x ∈ De,E such that s(x) = 1
(46) Remarks
• Thus, the meaning of ∃x3(smokes’ x3) is a constant function on variable assignments
o It maps any variable assignment g to 1 iff there is an x ∈ De,E such that the
‘denotation of smokes’ (s) maps x to 1

For any variable assignment g, [[∃x3(smokes’ x3)]]M,g = 1 iff there is an x ∈ De,E such
that I(smokes’)(x) = 1.
Given the discussion in (44)-(46), it can easily be seen that the following operations will
similarly serve as semantic correlates of F∀
(47) The Operation G∀

a. If there is a type τ ∈ T and a variable v ∈ VARτ such that for all g ∈ J,
α(g) = g(v), and β ∈ Mt, E, J, then
G∀(α,β) = The function A such that if g ∈ J, A(g) = 1 iff
for all x ∈ Dτ,E , β(g(v/x)) = 1
b. Otherwise, G∀(α,β) = garbage
15
Fall 2013 Ling 720
And, last but not least, how do we handle the semantics of lambda abstraction?...
(48) The Operation Gλ
a. If there is a type σ ∈ T and a variable v ∈ VARσ such that for all g ∈ J,

α(g) = g(v), and β ∈ Mτ, E, J, then
Gλ(α,β) = The function L such that if g ∈ J, L(g) =
The function p with domain Dσ,E such that for any x ∈ Dσ,E , p(x) = β(g(v/x))
b. Otherwise, Gλ(α,β) = garbage
• Again, the requirement that ‘for all g ∈ J, α(g) = g(v)’ ensures that if the first
argument α is not the meaning of a variable, then Gλ(α,β) = garbage
o Thus, ‘(λmitt’ (smokes’ mitt’))’ will end up being interpreted as garbage
• Now, consider the simple formula: (λx3 (smokes’ x3))
• Since Fλ and Gλ should ‘correspond’, it follows that for meaning assignment h:

h( (λx3 (smokes’ x3)) ) = h( Fλ(x3 , (smokes’ x3)) = Gλ( h(x3), h((smokes’ x3)) )
• Next, note that h((smokes’ x3)) ∈ Mt, E, J and for any g ∈ J, h(x3)(g) = f(x3)(g) = g(x3)
• Thus: Gλ( h(x3), h((smokes’ x3)) ) =

The function L such that if g ∈ J, L(g) =
The function p with domain De,E such that for any x ∈ De,E ,
p(x) = h((smokes’ x3))(g(x3/x))
• Now, recall that for any g ∈ J, h((smokes’ x3))(g) = s(g(x3))
• Thus: Gλ( h(x3), h((smokes’ x3)) ) =

The function L such that if g ∈ J, L(g) =
The function p with domain De,E such that for any x ∈ De,E ,
p(x) = s(g(x3/x)(x3))
• Finally, by definition of ‘g(x3/x)’, we have it that:

h( (λx3 (smokes’ x3)) ) =
The function L such that if g ∈ J, L(g) = the function p with domain De,E such that
for any x ∈ De,E , p(x) = s(x)
16
Fall 2013 Ling 720
(50) Remarks
• So, the meaning of (λx3(smokes’ x3)) is a constant function on variable assignments

o It maps any variable assignment g to the function p which takes an x ∈ De,E
and returns the value that the ‘denotation of smokes’ (s) maps x to.

For any variable assignment g, [[(λx3(smokes’ x3))]]M,g =
The function p whose domain is De,E and whose range is Dt,E, and for all x ∈ De,E
p(x) = I(smokes)(x).
(51) Summary
It seems, then, that the operations {Gγ}γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} will serve nicely as
‘semantic correlates’ of our syntactic operations {Fγ}γ∈{Concat, Not, And, Or, If, ∃, ∀, λ}
Consequently, let us wrap up our definition in (30) as follows…
(52) Logically Possible Partly-Fregean Interpretation of Politics+λ
A logically possible partly-Fregean interpretation of Politics+λ connected with E, J, σ is

a partly-Fregean interpretation B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} for Politics+λ
connected with E, J, σ, such that:
a. J is the set of variable assignments based on E.
b. If ϕ ∈ CONτ , then f(ϕ)(g) = f(ϕ)(g’), for all g, g’ ∈ J.
c. If ϕ ∈VARτ , then f(ϕ)(g) = g(ϕ) for all g ∈ J.
d. The operations {Gγ}γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} are as defined in (31)-(48)
17
Fall 2013 Ling 720
3. A Logically Possible Partly-Fregean Interpretation for Politics+λ
Let us now illustrate the key notion in (52) by concretely spelling out such an interpretation.
(53) A Logically Possible Partly-Fregean Interpretation for Politics+λ
Let E be the set {Barack, Michelle, Mitt}. Let J be the set of variable assignments based
on E. Let σ be the type assignment for Politics+λ such that for all τ ∈ T, σ(τ) =
σ(<var,τ>) = τ. Let B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} be the logically possible
partly-Fregean interpretation of Politics+λ connected with E, J, σ such that f consists of
the following mappings.
a. f(barack’) = The function b such that for all g ∈ J, b(g) = Barack.

b. f(michelle’) = The function m such that for all g ∈ J, m(g) = Michelle
c. f(mitt’) = The function mt such that for all g ∈ J, mt(g) = Mitt
d. f(smokes’) = The function s such that for all g ∈ J, s(g) =

o= { <Michelle,0>, <Barack,1>, <Mitt,0> }
e. f(man’) = The function mn such that for all g ∈ J, mn(g) =

i= { <Michelle,0>, <Barack,1>, <Mitt,1> }
f. f(president’) = The function pr such that for all g ∈ J, pr(g) =

k= { <Michelle,0>, <Barack,1>, <Mitt,0> }
g. f(loves’) = The function l such that for all all g ∈ J, l(g) =
j = Michelle  Michelle  1
Barack  1
Mitt  0
Barack  1
Mitt  0
Barack  0
Mitt  1
Finally, as we’ve done so many times before, we can use the meaning assignment h determined
by B to assign meanings to the expressions of Politics+λ
• To save space, I’ve not includes all the calculation steps below. Students are
encouraged to work out all the steps for themselves.
18
Fall 2013 Ling 720
In the calculations below, let h be the meaning assignment determined by the interpretation B
defined in (53).
(54) ∀x3 (man’ x3)
• h( ∀x3 (man’ x3) ) =
• The function A such that if g ∈ J, A(g) = 1 iff for all x ∈ De,E , i(x) = 1 =
• The function A such that if g ∈ J, A(g) = 0
Compare:
Relative to the model M defined in (13) on the previous handout, for any variable
assignment g, [[∀x3 (man’ x3)]]M,g = 0
(55) ∃P4 (P4 michelle’)
• h( ∃P4 (P4 michelle’) ) =
• The function E such that if g ∈ J, E(g) = 1 iff

there is an x ∈ D<e,t>,E such that x(Michelle) = 1 =
• The function E such that if g ∈ J, E(g) = 1
Compare:
assignment g, [[∃P4 (P4 michelle’)]]M,g = 1
(56) (λx3 ((man’ x3) & (smokes’ x3)))
• h( (λx3 ((man’ x3) & (smokes’ x3))) ) =
• The function L such that if g ∈ J, L(g) = the function p with domain De,E such that for
any x ∈ De,E , p(x) = 1 iff i(x) = 1 and o(x) = 1. =
• The function L such that if g ∈ J, L(g) = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
Compare:
assignment g, [[(λx3 ((man’ x3) & (smokes’ x3)))]]M,g =
{ <Michelle, 0>, <Barack,1>, <Mitt,0> }
19
Fall 2013 Ling 720
(57) (λP4 (P4 mitt’))
• h( (λP4 (P4 mitt’)) ) =
• The function L such that if g ∈ J, L(g) = the function p with domain D<e,t>,E such that
for any x ∈ D<e,t>,E , p(x) = x(Mitt)
Compare:
assignment g, [[(λP4 (P4 mitt’))]]M,g =
The function p with domain D<et>,E , range Dt,E and for all x ∈ D<et>,E , p(x) = x(Mitt)
(58) (λP4 ∀x3 ( (man’ x3)  (P4 x3) ) )
• h( (λP4 ∀x3 ((man’ x3)  (P4 x3))) ) =
for any x ∈ D<e,t>,E , p(x) = 1 iff for all y ∈ De,E , i(y) = 0 or x(y) = 1
for any x ∈ D<e,t>,E , p(x) = 1 iff for all y ∈ De,E , if i(y) = 1 then x(y) = 1
Compare:
assignment g, [[(λP4 ∀x3 ((man’ x3)  (P4 x3)))]]M,g =
The function p with domain D<et>,E, range Dt,E and for all x ∈ D<et>,E , p(x) = 1 iff
for all y ∈ De,E, if i(y) = 1 then x(y) = 1
(59) ( (λx3 ((man’ x3) & (smokes’ x3))) mitt’ )
• h( ((λx3 ((man’ x3) & (smokes’ x3))) mitt’) ) =
• The function F-A such that if g ∈ J, then F-A(g) =

h((λx3 ((man’ x3) & (smokes’ x3))))(g)( h(mitt’)(g) ) = (by (53), (56))
• The function F-A such that if g ∈ J, then F-A(g) = L(g)( mt(g) ) = (by (53), (56))
• The function F-A such that if g ∈ J, then F-A(g) =

{ <Michelle, 0>, <Barack,1>, <Mitt,0> }(Mitt) =
• The function F-A such that if g ∈ J, then F-A(g) = 0
Compare: Relative to the model M defined in (13) on the previous handout, for any
variable assignment g, [[( (λx3 ((man’ x3) & (smokes’ x3))) mitt’ )]]M,g = 0
20
Fall 2013 Ling 720
4. Relationship Between Models and Logically Possible Partly-Fregean Interpretation
We now have two ways of providing Politics+λ a formal semantics:

(i) The model-theoretic semantics from the handout ‘Preliminaries’
(ii) The logically possible partly-Fregean interpretation in (53)
(60) Fun Fact: There is (again) an important equivalence between these two systems!
(61) The Relationship Between Models and Interpretations (Part 1)

Let M be a model <E,I> for Politics+λ. Let J be the set of variable assignments based on
M. Let B = ∪τ ∈ T Mτ, E, J.
Finally, let f be the function whose domain is ∪δ∈Δ Xδ and for any ϕ ∈ ∪δ∈Δ Xδ,
f(ϕ) = the function m whose domain is J and for any g ∈ J, m(g) = [[ϕ]]M,g
a. Claim 1:
The structure B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} is a logically-possible
partly-Fregean interpretation of Politics+λ.
b. Claim 2:
Let h be the meaning assignment determined by B. Every meaningful expression
ϕ of Politics+λ is such that, for any variable assignment g ∈ J:
h(ϕ)(g) = [[ϕ]]M,g
(62) Remark
Thus, if we are given a model M for Politics+λ, we can construct a logically possible
partly-Fregean intepretation B that assigns equivalent values to the meaningful
expressions of Politics+λ.
(63) The Relationship Between Models and Interpretations (Part 1)

Let B = <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} be a logically possible partly-Fregean
interpetation of Politics+λ connected with E, J, σ. Let I be the function whose domain is
∪τ∈T CONτ and for any ϕ ∈ ∪τ∈T CONτ , I(ϕ) = f(ϕ)(g), for an arbitrary g.
a. Claim 1: The structure M = <E, I> is a model of Politics+λ.
b. Claim 2:
Let h be the meaning assignment determined by B. Every meaningful expression
ϕ of Politics+λ is such that, for any variable assignment g ∈ J:
h(ϕ)(g) = [[ϕ]]M,g
21
Fall 2013 Ling 720
(63) Remark
Thus, if we are given a logically possible partly-Fregean interpretation B of Politics+λ,
we can construct a model M for Politics+λ that assigns equivalent values to the
meaningful expressions of Politics+λ.
(64) Summary
• Given this relationship between models and logically possible partly-Fregean

interpretations of Politics+λ, we can freely shift between the two (cf. sets and their
characteristic functions).
• Similarly, we can view these two systems as being in essence ‘the same thing’ (even
though they are very different set theoretic objects).
• Thus, even when we’re presenting our semantics in model-theoretic terms, we

are also thereby providing a (Montagovian) interpretation for the language.
(65) The Importance of This Point
• As you can probably tell now, once we’ve got variable binding in the language, the
model-theoretic presentation of the semantics is much simpler and more
comprehensible than its presentation as a (Montagovian) ‘interpretation.’
• The difference in comprehensibility is even more profound for Montague’s

system of ‘Intensional Logic’, used in PTQ.
• For this reason, in PTQ (but not UG), Montague presents the semantics for IL in
strictly model-theoretic terms (no ‘interpretations’).
o I will be following suit, but do know that there is straight-forward way of

converting Montague’s model structures for IL into ‘interpretations’…
(just see UG, if you’re interested!...)
22
Fall 2013 Ling 720
Computing with ‘Logically Possible Partly-Fregean Interpretations’
Let h be the meaning assignment determined by the logically possible partly-Fregean

interpretation in (53) of the handout “Fregean Interpretations.”
(i) h((λP4 ∃x3 ((man’ x3) & (P4 x3)))) = (by definition of Politics+λ)
(ii) h( Fλ( P4 , F∃ ( x3 , FAnd( FConcat(man, x3), FConcat(P4, x3)…) = (by homomorph. of h)
(iii) Gλ( h(P4) , G∃ ( h(x3) , GAnd( GConcat( h(man), h(x3)), GConcat( h(P4), h(x3))…)
= (by def. of h)
(iv) Gλ( f(P4) , G∃ ( f(x3) , GAnd( GConcat( f(man), f(x3)), GConcat( f(P4), f(x3))…)
= (by def. of f)
(v) Gλ( f(P4) , G∃ ( f(x3) , GAnd( GConcat( mn , f(x3)), GConcat( f(P4), f(x3))…)
= (by def. of Gλ)
(vi) The function L such that if g ∈ J, L(g) =

The function p with domain D<e,t>,E such that for any x ∈ D<e,t>,E ,
p(x) = G∃ ( f(x3) , GAnd( GConcat( mn , f(x3)), GConcat( f(P4), f(x3))…)(g(P4/x))
= (by def. of G∃)
(vii) The function L such that if g ∈ J, L(g) =

p(x) = [The function E such that if g’ ∈ J, E(g’) = 1 iff there is a y ∈ De,E such that
GAnd( GConcat( mn , f(x3)), GConcat( f(P4), f(x3))…)(g’(x3/y)) = 1](g(P4/x))
= (by meta-logic)
(viii) The function L such that if g ∈ J, L(g) =

p(x) = 1 iff there is an y ∈ De,E such that
GAnd( GConcat( mn , f(x3)), GConcat( f(P4), f(x3))…)(g(P4/x)(x3/y)) = 1
= (by def. of GAnd)
(ix) The function L such that if g ∈ J, L(g) =

[The function Conj such that if g’ ∈ J, then Conj(g’) = 1 iff
GConcat( mn , f(x3))(g’) = 1 and GConcat( f(P4), f(x3))(g’) = 1](g(P4/x)(x3/y)) = 1
= (by meta-logic)
Fall 2013 Ling 720
(x) The function L such that if g ∈ J, L(g) =
GConcat( mn , f(x3)) )(g(P4/x)(x3/y)) = 1 and GConcat( f(P4), f(x3))(g(P4/x)(x3/y)) = 1
= (by def. of GConcat)
(xi) The function L such that if g ∈ J, L(g) =

mn((g(P4/x)(x3/y))(f(x3)(g(P4/x)(x3/y))) = 1 and
f(P4)((g(P4/x)(x3/y))(f(x3)(g(P4/x)(x3/y))) = 1 = (by (53))
(xii) The function L such that if g ∈ J, L(g) =

i(g(P4/x)(x3/y))(x3)) = 1 and g(P4/x)(x3/y))(P4)(g(P4/x)(x3/y))(x3)) = 1
= (by def. of ‘Logically Possible Partly-Fregean Interpretation)
(xiii) The function L such that if g ∈ J, L(g) =

p(x) = 1 iff there is an y ∈ De,E such that i(y) = 1 and x(y) = 1
Thus, h((λP4 ∃x3 ((man’ x3) & (P4 x3)))) is computed to be a constant function on variable
assignments.
• It maps every variable assignment to the function p, which is the characteristic
function of the set of <e,t> functions f which map some man to 1.
• Thus, it maps every variable assignment to the ‘set of properties that some man has’
Note the parallel to our model theoretic semantics from the handout “Preliminaries”:
Let M be the model defined in (13) of “Preliminaries”. Let g be any variable assignment based
on M.
[[(λP4 ∃x3 ((man’ x3) & (P4 x3)))]]M,g =
The function p with domain D<et>,E, range Dt,E and for all x ∈ D<et>,E ,
p(x) = 1 iff there is a y ∈ De,E such that i(y) = 1 and x(y) = 1
Fall 2013 Ling 720
An Algebraic Approach to Quantification and Lambda Abstraction:
Applications to the Analysis of English
(1) Ingredients on the Table

a. A logical language with both quantification and lambda abstraction (Politics+λ)
b. An interpretation <B, Gγ, f>γ∈{Concat, Not, And, Or, If, ∃, ∀, λ} for Politics+λ
(2) Goals for This Unit
a. Develop a fragment of English that contains quantificational NPs.
b. Develop a translation base from that fragment to Politics+λ
c. Examine the interpretation thereby induced for the fragment of English.
Note: The system developed here is a blend of the ones Montague develops in UG and PTQ.
To my knowledge, this system appears nowhere else in the literature on MG.
1. Towards a New Fragment of English: Quantificational NPs in Subject Position
In this section, we will work out at a relatively informal level how we should structure the
English fragment and its translation into Politics+λ
(3) Proximal Goal

Expand our fragment of English so it includes expressions like: some man, every man.
(4) a. Easy Question: What should the category of these new expressions be?
b. Easy Answer:
Well, their external syntactic behavior is just like the expressions Barack, Mitt,
Michelle. Thus, if the latter are NPs, then so should be some man and every man.
(5) First Steps Towards Implementation

We’re not ready to officially write out this new fragment of English, but it seems that it
should include the following pieces.
a. The New Category Labels:

(i) TV ‘transitive verb’
(ii) IV ‘intransitive verb’ (and ‘VPs’)
(iii) S ‘sentence’
(iv) T ‘term’ (formerly ‘NP’)
(v) CN ‘common noun’ (that is, the ‘Ns’)
1
Fall 2013 Ling 720
b. The Basic Expressions
(i) XTV = { < loves, ∅ > }
(ii) XIV = { < smokes, ∅ > }
(iii) XT = { < Barack, ∅ >, < Mitt, ∅ >, < Michelle, ∅ > }
(iv) XCN = { < man, ∅ >, < president, ∅ > }
(v) XS = ∅
c. The Syntactic Operations

In addition to the operations we introduced for DME, let’s add the operations
KSome and KEvery.
 In the definitions below, α and β are trees whose root nodes are ordered pairs.
In addition α’ and β’ are the first members of the root nodes of α and β.
(i) KSome(α) = < some α’ , Some >
(ii) KEvery(α) = < every α’ , Every >
d. The Syntactic Rules

Let’s add the following rules to the set of syntactic rules SE we had for DME.
(i) < KSome , < CN >, T >

‘The result of applying KSome to an expression of category CN is an
expression of category T.’
(ii) < KEvery , < CN >, T >

‘The result of applying KEvery to an expression of category CN is an
expression of category T.’
e. Illustration: The category CT includes the following trees:
< Barack, ∅ > < Mitt, ∅ > < Michelle, ∅ >
< every man, Every > < every president, Every >
< man, ∅ > < president, ∅ >
< some man, Some > < some president, Some >
< man, ∅ > < president, ∅ >
2
Fall 2013 Ling 720
(6) Handy Abbreviation

Since it’s clear from context here that the expressions of our fragment are trees, I’ll make
use of the convention below to save space:
expression = A tree whose root node is < expression , Index >
(7) Difficult Question:

What type of expressions in Politics+λ should our (quantificational) terms translate as?
(8) Towards the Answer (Linguistics 610)
• We don’t want the meaning of some man or every man end up being of type e
• Rather, we want it to be a function of type <<e,t>,t> (Generalized Quantifier)
hºk(some man) =
the function whose domain is D<et>,E, and for every f in D<et>,E maps f to 1 iff
there is an x in De,E such that x is a man and f(x) = 1
hºk(every man) =
the function whose domain is D<et>,E, and for every f in D<et>,E maps f to 1 iff
for all x in De,E , if x is a man then f(x) = 1
• Therefore, we’d like our translation function k to achieve the following mappings:
k(some man) = (λP0 ∃x0 ( (man’ x0) & (P0 x0) ) )
k(every man) = (λP0 ∀x0 ( (man’ x0)  (P0 x0) ) )
(9) A Challenge Appears!
• Thus, we want our translation function k to map some man and every man to
expressions in Politics+λ of type <<e,t>,t>
• But, recall that under our theory of translation, every expression of the same category
in English must be translated to an expression of the same type in Politics+λ
• Thus, the goal in (8) entails that every term of our language – even Barack,
Michelle, and Mitt – must be translated as <<et>,t> formulae.
3
Fall 2013 Ling 720
(10) The Solution to the Challenge (Linguistics 610)
The proper names Barack, Michelle, and Mitt will receive the translations below:
a. k(< Barack, ∅ >) = (λP0 (P0 barack’))

b. k(< Mitt, ∅ >) = (λP0 (P0 mitt’))
c. k(< Michelle, ∅ >) = (λP0 (P0 michelle’))
That is, the name (e.g.) Barack will end up translated as an expression denoting the
characteristic function of the set of <e,t> functions that map Barack to 1.
(11) Immediate Problem

In our translation base mapping Mini-English to Politics-NoQ, the syntactic operation
KMerge-S corresponds to the polynomial operation FConcat<Id2,2, Id1,1>.
• Given the translations in (8)-(10), if we maintain that correspondence, we’ll end up

mapping meaningful expressions of Mini-English to syntactic garbage in Politics+λ.
k( Barack smokes ) =
k( KMerge-S(< Barack, ∅ >, < smokes, ∅ >) ) =
FConcat<Id2,2, Id1,1>( k(< Barack, ∅ >), k(< smokes, ∅ >) ) =
FConcat<Id2,2, Id1,1>( (λP0 (P0 barack’)) , smokes’ ) =
( smokes’ (λP0 (P0 barack’)) ) = syntactic garbage!
(12) Potential Solution

For our new fragment of English, in the translation to Politics+λ, KMerge-S could now just
correspond to FConcat. That would yield the right results!
k( Barack smokes ) =
k( KMerge-S(< Barack, ∅ >, < smokes, ∅ >) ) =
FConcat( k(< Barack, ∅ >), k(< smokes, ∅ >) ) = (by (10))
FConcat( (λP0 (P0 barack’)) , smokes’ ) =
( (λP0 (P0 barack’)) smokes’ ) ⇔ (by lambda-conversion)
(smokes’ barack’)
4
Fall 2013 Ling 720
Our solution in (12) would also get the right results for sentences whose subjects are
quantificational terms!...
(13) Illustration: Every man smokes
k( every man smokes ) =
k( KMerge-S( every man , < smokes, ∅ >) ) =
FConcat( k(every man), k(< smokes, ∅ >) ) = (by (8))
FConcat( (λP0 ∀x0 ( (man’ x0)  (P0 x0) ) ) , smokes’ ) =
( (λP0 ∀x0 ( (man’ x0)  (P0 x0))) smokes’ ) ⇔ (by lambda-conversion)
∀x0 ((man’ x0)  (smokes’ x0))
(14) Illustration: Some man smokes
k( some man smokes ) =

k( KMerge-S( some man , < smokes, ∅ >) ) =
FConcat( k(some man), k(< smokes, ∅ >) ) = (by (8))
FConcat( (λP0 ∃x0 ((man’ x0) & (P0 x0))) , smokes’ ) =
( (λP0 ∃x0 ((man’ x0) & (P0 x0))) smokes’ ) ⇔ (by lambda-conversion)
∃x0 ((man’ x0) & (smokes’ x0))
(15) Summary
Well, it looks like we’re all done here! We just need to properly implement the ideas in
(5)-(12), and we can call it a day, put on our PJs, and get caught up on Breaking Bad!...
2. Towards a New Fragment of English: Quantificational NPs in Object Position
Oh wait… Crap! We forgot about the fact that quantificational terms can also be direct objects…
(16) New Goal

We would like for the sentences below to paired with the following translations:
a. Sentence: Michelle loves every president.

Translation: ∀x0 ((president’ x0)  ((loves’ x0) michelle’))
b. Sentence: Mitt loves some man.

Translation: ∃x0 ((man’ x0) & ((loves’ x0) mitt’))
5
Fall 2013 Ling 720
(17) The Obvious Problem
In our translation base mapping Mini-English to Politics-NoQ, the syntactic operation
KMerge-IV corresponds to the polynomial operation FConcat.
• Given the translations in (8)-(10), if we maintain that correspondence, we’ll end up

mapping meaningful expressions of Mini-English to syntactic garbage in Politics+λ.
o And, we definitely don’t get translations like (16a,b)
k( loves every president ) =
k( KMerge-IV( < loves, ∅ > , every president ) ) =
FConcat( k(< loves, ∅ >) , k(every president) ) = (by (8))
FConcat( loves’ , (λP0 ∀x0 ((president’ x0)  (P0 x0))) ) =
( loves’ (λP0 ∀x0 ((president’ x0)  (P0 x0))) ) = syntactic garbage!
(18) The More General Issue: Quantificational Terms in Object Position
• In the semantics for English that we get form our translation, we want terms like
every president to end up denoting functions of type <<e,t>,t> (GQs)
• But we also in our semantics want transitive verbs like smokes to end up denoting
functions of type <e,<e,t>> (curried characteristic functions of binary relations).
• But, how do we put together meanings of those two different types?
(19) The PTQ Solution, Rough Idea: Part 1
Our fragment of English will contain a syntactic operation that is like ‘QR-in-reverse’.
• This operation KQI (QI = ‘quantifying in’) will take (i) a quantificational term, and (ii)
a sentence containing a pronoun (cf. trace), and output a string where the pronoun (cf.
trace) is replaced with the quantification term.
Rough Idea Illustrated:
KQI( every president , Michelle loves him2 ) = Michelle loves every president.
6
Fall 2013 Ling 720
(20) Remark
Although the transformation of the strings is like ‘QR-in-reverse’, the actual analysis
trees output by KQI are geometrically similar to the PS-trees generated by QR.
< Michelle loves every president, QI >
< every president, Every > < Michelle loves him2 , Merge-S>
< president, ∅ > < Michelle, ∅ > < loves him2 , Merge-IV>
< loves, ∅ > < him2 , … >
(21) The PTQ Solution, Rough Idea: Part 2

The polynomial operation corresponding to KQI (HQI) will do all the following:
(i) Take the translation of the quantificational term

(ii) Take the translation of the sentence
(iii) Lambda-abstract over the variable in the translation of the sentence
(iv) Apply the translation of the quantificational term to the resulting λ-expression
k (KQI( every president , Michelle loves him2 ) ) =
HQI ( k(every president) , k(Michelle loves him2) ) =
The HQI ((λP0 ∀x0 ((president’ x0)  (P0 x0))) , ((loves’ x0) michelle’) ) =
Key
Step! ( (λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx0 ((loves’ x0) michelle’)) ) ⇔
(by alpha-conversion)
( (λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx1((loves’ x1) michelle’)) ) ⇔

(by lambda-conversion)
∀x0 ((president’ x0)  ((λx1((loves’ x1) michelle’)) x0))) ⇔

(by lambda-conversion)
∀x0 ((president’ x0)  ((loves’ x0) michelle’))
If we can flesh out the ideas in (19) and (21), we will have a system that properly translates /
interprets quantificational terms in object position.
7
Fall 2013 Ling 720
(22) Our To-Do List
a. Step One: Develop a theory of the syntax and semantics of pronouns in English
b. Step Two: Properly define the operation KQI
c. Step Three: Properly define the operation HQI
(23) Fun Fact!

Once we’ve set up everything correctly, our system will correctly predict quantifier-scope
ambiguities. That is, our system will predict that the English sentence in (a) can receive
either of the translations in (b).
a. Sentence of (Ambiguous) English: Some man loves every president.
b. Possible Translations into Politics+λ
(i) ∃x0 ((man’ x0) & ∀x1 ((president’ x1)  ((loves’ x1) x0)))
(ii) ∀x0 ((president’ x0)  ∃x1 ((man’ x1) & ((loves’ x0) x1)))
3. Developing The Analysis of English Quantification
3.1 The Syntax and Translation of English Pronouns
(24) A Radical, Crazy Idea Only Chomskians Could Come Up With

Perhaps pronouns are structurally complex, and consist of (i) a ‘pronominal core’
containing so-called ‘ϕ-features’ like Case, Gender, Number, and (ii) a pronominal index.
Hei = ProP
ϕP i
Nom, Masc, Sg
(25) Montagovian Precursor (in “Universal Grammar”)

Pronouns in English are actually syntactically derived from more primitive ‘indices’.
KHe(i) = Hei
(26) Adapting This Idea to Our System: The Syntax of Indices

We will assume that indices are trivial trees consisting of the following root-nodes:
IND = {<n,∅>:n∈ℕ }
8
Fall 2013 Ling 720
(27) The Translation of Pronouns into Politics+λ
The system we will go on to build will not translate the pronouns per se, but rather the
indices the pronouns are derived from.
a. Translation of Indices: For all n ∈ ℕ, k(< n , ∅ >) = xn ( = ve,n )
b. Polynomial Operation for KHe: HHe = Id1,1
c. Illustration: k( he 3 ) = k ( KHe(<3, ∅>) ) =

HHe( k(<3, ∅>) ) = (by (27a), (27b))
Id1,1(x3) = x3
(28) Partial Implementation of the Ideas in (25)-(27)
a. Expanding Our Set of Category Labels for English:

(i) TV ‘transitive verb’
(ii) IV ‘intransitive verb’ (and ‘VPs’)
(iii) S ‘sentence’
(iv) T ‘term’ (formerly ‘NP’)
(v) CN ‘common noun’ (that is, the ‘Ns’)
(vi) IN ‘indices’
(vii) PR ‘pronouns’
b. The Basic Expressions:

(i) XTV = { < loves, ∅ > }
(ii) XIV = { < smokes, ∅ > }
(iii) XT = { < Barack, ∅ >, < Mitt, ∅ >, < Michelle, ∅ > }
(iv) XCN = { < man, ∅ >, < president, ∅ > }
(v) XIN = {<n,∅ >:n∈ ℕ }
(vi) XPR = ∅
(vii) XS = ∅
c. The Syntactic Operations:

In addition to the operations we already have, let’s add the operations KHe and
KShe. We’ll also revise the operation KMerge-IV as in (iii) below:
 In the definitions below, α and β are trees whose root nodes are ordered pairs.
In addition α’ and β’ are the first members of the root nodes of α and β.
(i) KHe(α) = < he α’ , He >
(ii) KShe(α) = < she α’ , She >
9
Fall 2013 Ling 720
(iii) KMerge-IV(α, β) =
1. If β’ = he n, then < α’ him n , Merge-IV >
α β
2. If β’ = she n, then < α’ her n , Merge-IV >
α β
3. Otherwise < α’ β’ , Merge-IV >
α β
Note: To capture the effects of pronominal case, we now have independent, empirical
motivation for distinguishing the operations KMerge-IV and KMerge-S
d. The Syntactic Rules:

Let us remove from our fragment of English the rules <KMerge-IV, < TV, NP >, IV>
and < KMerge-S, < NP, IV >, S >. Let us add the following rules:
(i) < KHe , < IN >, PR >

‘Applying KHe to an expression of category IN (an index) yields an
expression of category PR (a pronoun).’
(ii) < KShe , < IN >, PR >

‘Applying KShe to an expression of category IN (an index) yields an
expression of category PR (a pronoun).’
(iii) < KMerge-IV, < TV, PR >, IV >

‘Applying KMerge-IV to an expression of category TV (transitive verb) and
one of category PR (pronoun) yields an expression of category IV.’
(iv) < KMerge-S, < PR, IV >, S >

‘Applying KMerge-S to an expression of category PR (pronoun) and one of
category IV yields an expression of category S.’
10
Fall 2013 Ling 720
(29) Illustration of the Resulting System
The system sketched in (28) will entail that CS contains trees such as:
a. < he 2 smokes, Merge-S >
< he 2, He > < smokes , ∅ >
<2,∅>
b. < she 5 loves him 7, Merge-S >
< she 5, She > < loves him 7 , Merge-IV >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
c. < he 38 loves her 2987, Merge-S >
< he 38, He > < loves her 2987 , Merge-IV >
< 38 , ∅ > < loves , ∅ > < she 2987, She >
< 2987 , ∅ >
Since we don’t actually pronounce indices in English, we will want our ‘ambiguating’ relation R
to delete them from the first member of the root-node of our trees.
(30) A Slight Change to Our ‘Ambiguating’ Relation R

R is a function which takes as input a tree T, and returns as output the first member of the
root node of T without any numerals.
< he 38 loves her 2987, Merge-S >
R < he 38, He > < loves her 2987 , Merge-IV > = he loves her
< 38 , ∅ > < loves , ∅ > < she 2987, She >
< 2987 , ∅ >
11
Fall 2013 Ling 720
(31) A Sketch of the Translation to Politics+λ
The correspondences in (31a), added with (27a), will yield the translations in (31b).
a. Correspondences Between Syntactic Operations and Polynomial Operations:

(i) KHe ~ HHe = Id1,1
(ii) KShe ~ HShe = Id1,1
(iii) KMerge-IV ~ HMerge-IV = FConcat
(iv) KMerge-S ~ HMerge-S = FConcat<Id2,2, Id1,1>
b. Illustrative Translations
(i) k ( he 2 smokes ) =
k ( KMerge-S ( KHe(<2, ∅>), < smokes , ∅ > ) ) =
HMerge-S ( HHe( k(<2, ∅> ), k(< smokes , ∅ >) ) =
HMerge-S ( HHe( x2 ), smokes’ ) =
FConcat<Id2,2, Id1,1>( Id1,1( x2 ), smokes’ ) =
FConcat<Id2,2, Id1,1>( x2 , smokes’ ) =
( smokes’ x2 )
(ii) k( she 5 loves him 7 ) =
k ( KMerge-S ( KShe(<5, ∅>),

KMerge-IV(< loves , ∅ >, KHe(<7, ∅>)))) =
HMerge-S ( HShe( k(<5, ∅>) ),

HMerge-IV( k(< loves , ∅ >) , HHe( k (<7, ∅>) ))) =
HMerge-S ( HShe( x5 ), HMerge-IV( loves’ , HHe( x7 ))) =
FConcat<Id2,2, Id1,1> ( Id1,1(x5) , FConcat( loves’ , Id1,1(x7))) =
FConcat<Id2,2, Id1,1> ( x5 , FConcat( loves’ , x7 ))) =
((loves’ x7) x5)
12
Fall 2013 Ling 720
3.2 The Syntactic Operation KQI (‘Quantifying In’)
(32) The Rough Idea Behind KQI

This will take (i) a quantificational term, and (ii) a sentence containing a pronoun, and
output a string where the pronoun is replaced with the quantification term.
KQI( every president , Michelle loves him2 ) = Michelle loves every president.
(33) Immediate Issue

To properly define such an operation, there must be some means of specifying which
pronoun in the sentential argument is to be replaced. Here, there are two possible paths:
a. First Implementation (“Universal Grammar”):

KQI is a ternary relation, whose first argument is the index of the pronoun to be
replaced in the sentential argument.
b. Second Implementation (“PTQ”):

Instead of a single operation KQI, we have an infinite number of operations KQI,n ,
where n is the index of the pronoun to be replaced in the sentential argument.
• In this unit, we’ll follow Montague in UG, and pursue the path in (33a).1
(36) Preliminary Notation: Pro = a meta-variable ranging over he, she, him, her
(37) The Operation KQI
Let α, β, γ be trees whose root nodes are ordered pairs. In addition, let α’ , β’ , γ’ be the
first members of the root nodes of α, β, γ (respectively).
KQI (α, β, γ) = < δ, QI >
β α γ
Where δ is the result of replacing in γ’ the left-most instance of Pro α’ with β’.
1
The statement here is somewhat misleading, since Montague didn’t introduce a ‘quantifying-in’ operation in UG.
However, the analysis Montague pursues for relative clauses in UG and PTQ faces the same issue in (33). The
general approach in (33a) was the one taken up by Montague in UG for the operation forming relative clauses. The
approach in (33b) was taken up in PTQ for both ‘quantifying-in’ and relativization.
13
Fall 2013 Ling 720
(38) Illustration of KQI
In the examples below, let T1 be the tree in (29a), T2 be the tree in (29b). Finally, let T3
be the tree: < every president, Every >
< president, ∅ >
a. KQI( <2, ∅> , T3 , T1 ) =
< every president smokes , QI >
< every president, Every > <2, ∅> < he 2 smokes, Merge-S >
< president, ∅ > < he 2, He > < smokes , ∅ >
<2,∅>
b. KQI( <5, ∅> , < Michelle, ∅> , T2 ) =
< Michelle loves him 7 , QI >
< Michelle, ∅> <5, ∅> < she 5 loves him 7, Merge-S >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
c. KQI( <7, ∅> , T3 , (38b) ) =
< Michelle loves every president, QI >
< every president, Every > <7, ∅> < Michelle loves him 7 , QI >
< president, ∅ > < Michelle, ∅> <5, ∅> < she 5 loves him 7, Merge-S >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
14
Fall 2013 Ling 720
(39) Remark
Notice how the analysis tree structures in (38) are geometrically similar to the LFs below,
which would be posited in a ‘Heim & Kratzer’-style system! (Hmmmm….)
a. S’’
DP S’
Every president 2 S
t2 VP
b. S’’’’
smokes
DP S’’’
Every president 7 S’’
DP S’
Michelle 5 S
t5 loves t7
(40) The Roots of Quantifier Scope Ambiguities
Our operation KQI will be able to output both the following analysis tree structures. Note
that the relation R in (30) would map both to the string some man loves every president.
a. < some man loves every president, QI >
< every president, Every > <7, ∅> < some man loves him 7 , QI >
< president, ∅ > < some man, Some> <5, ∅> < he 5 loves him 7, Merge-S >
< man, ∅ > < he 5, He > < loves him 7 , Merge-IV >
<5,∅> < loves , ∅ > < he 7, He >
b. < some man loves every president, QI > <7,∅>
< some man, Some > <5, ∅> < he 5 loves every president , QI >
< man, ∅ > < every president, Every > <7, ∅> < he 5 loves him 7, Merge-S >
< president, ∅ > < he 5, He > < loves him 7 , Merge-IV >
<5,∅> < loves , ∅ > < he 7, He >

<7,∅>
15
Fall 2013 Ling 720

In the system we will ultimately develop from this, the English string some man loves
every president is syntactically ambiguous.
• One analysis tree that R in (30) maps to this string is derived by QI-ing some man
before every president (40a).
• Another analysis tree that R in (30) maps to this string is derived by QI-ing every
president before some man (40b).
(42) Key Idea

We should design the polynomial operation HQI so that this syntactic ambiguity will lead
to a semantic ambiguity.
k ( (40a) ) = ∀x0 ((president’ x0)  ∃x1 ((man’ x1) & ((loves’ x0) x1)))
k ( (40b) ) = ∃x0 ((man’ x0) & ∀x1 ((president’ x1)  ((loves’ x1) x0)))
To close out this section, let us introduce the syntactic rule below.
(43) The Syntactic Rule Employing KQI
< KQI , < IN, T, S >, S >

‘Applying KQI to an expression of category IN (index), an expression of category T
(term), and an expression of category S (sentence), yields and expression of category S.’
Note: Given this rule, the following are members of CS: (38a), (38b), (38c), (40a), (40b)
3.3 The Polynomial Operation HQI , Corresponding to KQI
(44) What We HQI to Do For Us

In general, if HQI( xn , ϕ , ψ ) = ( ϕ (λxn ψ) ), that would work well for our translation.
k( every president smokes ) = (by (38a))
k ( KQI( <2, ∅> , every president , he 2 smokes ) ) = (by homomorphism property)
HQI( k(<2, ∅>) , k(every president) , k(he 2 smokes))) = (by (27a), (8), (31b))
HQI( x2 , (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (smokes’ x2)) = (by assumption)
Key
Part! ( (λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 (smokes’ x2)) ) ⇔ (by λ-conversion)
∀x0 ((president’ x0)  (smokes’ x0))
16
Fall 2013 Ling 720
(45) A Polynomial Operation that Would Do the Job
Let us define HQI as the following polynomial operation: FConcat< Id2,3 , Fλ< Id1,3, Id3,3 > >
Illustration:
FConcat<Id2,3, Fλ<Id1,3, Id3,3>>( x2 , (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (smokes’ x2)) =
FConcat ( Id2,3( x2 , (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (smokes’ x2)) ,
Fλ( Id1,3( x2 , (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (smokes’ x2))
Id3,3( x2 , (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (smokes’ x2)) ) ) =
FConcat ( (λP0 ∀x0 ((president’ x0)  (P0 x0))) , Fλ( x2 , (smokes’ x2) ) ) =
FConcat ( (λP0 ∀x0 ((president’ x0)  (P0 x0))) , (λx2 (smokes’ x2)) ) =
( (λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 (smokes’ x2)) )
We’re almost ready to set up our fragment of English and the translation base. The last
ingredient is to find polynomial operations corresponding to KSome and KEvery
(46) The Polynomial Operation HSome , Corresponding to KSome
a. What We Want HSome to Do For Us

In general, if HSome(ϕ) = (λP0 ∃x0 ((ϕ x0) & (P0 x0))), that would work well.
k( some man ) = k (KSome(<man, ∅>)) = HSome(k(<man, ∅>)) = HSome(man’) =

(λP0 ∃x0 ((man’ x0) & (P0 x0)))
b. A Polynomial Operation that Would Do the Job

Let us define HSome as the following polynomial operation:
Fλ< CP0,1 , F∃ < Cx0,1 , FAnd < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>
Illustration:
Fλ< CP0,1 , F∃ < Cx0,1 , FAnd < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>(man’) =
Fλ( CP0,1(man’) , F∃ ( Cx0,1(man’) ,
FAnd( FConcat ( Id1,1(man’), Cx0,1(man’) ) ,
FConcat ( CP0,1(man’), Cx0,1(man’) ) ) ) ) =
Fλ( P0 , F∃( x0 , FAnd( FConcat( man’, x0 ) , FConcat ( P0 , x0 ) ) ) ) =
Fλ( P0 , F∃( x0 , FAnd( (man’ x0) , (P0 x0) ) ) ) =
Fλ( P0 , F∃( x0 , ((man’ x0) & (P0 x0)) ) ) =
(λP0 ∃x0 ((man’ x0) & (P0 x0)))
17
Fall 2013 Ling 720
(47) The Polynomial Operation HEvery , Corresponding to KEvery
a. What We Want HEvery to Do For Us

In general, if HEvery(ϕ) = (λP0 ∀x0 ((ϕ x0)  (P0 x0))), that would work well.
k( every man ) = k (KEvery (<man, ∅>)) = HEvery(k(<man, ∅>)) = HEvery(man’) =

(λP0 ∀x0 ((man’ x0)  (P0 x0)))
b. A Polynomial Operation that Would Do the Job

Let us define HEvery as the following polynomial operation:
Fλ< CP0,1 , F∀ < Cx0,1 , FIf < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>
Illustration:
Fλ< CP0,1 , F∀ < Cx0,1 , FIf < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>(man’) =
Fλ( CP0,1(man’) , F∀ ( Cx0,1(man’) ,
FIf( FConcat ( Id1,1(man’), Cx0,1(man’) ) ,
FConcat ( CP0,1(man’), Cx0,1(man’) ) ) ) ) =
Fλ( P0 , F∀ ( x0 , FIf( FConcat( man’, x0 ) , FConcat ( P0 , x0 ) ) ) ) =
Fλ( P0 , F∀ ( x0 , FIf( (man’ x0) , (P0 x0) ) ) ) =
Fλ( P0 , F∀ ( x0 , ((man’ x0)  (P0 x0)) ) ) =
(λP0 ∀x0 ((man’ x0)  (P0 x0)))
We’re now ready to put all these ideas together into a semantic analysis of English
quantification!
4. The Fragment of English: ‘Mini-English+Qs’ (ME+Q)
(48) Step One: The Category Labels Δ = {TV, IV, S, T, CN, IN, PR}
(49) Step Two: The Basic Expressions
XTV = { < loves, ∅ > }

XIV = { < smokes, ∅ > }
XT = { < Barack, ∅ >, < Mitt, ∅ >, < Michelle, ∅ > }
XCN = { < man, ∅ >, < president, ∅ > }
XIN = {<n,∅>:n∈ℕ}
XPR = ∅
XS = ∅
18
Fall 2013 Ling 720
(50) Step Three: The Syntactic Operations
a. Operations Unchanged From Disambiguated Mini-English (DME)

KMerge-S = (as defined previously)
KNot = (as defined previously)
KAnd = (as defined previously)
KIf = (as defined previously)
b. New Syntactic Operations

KSome = (as defined in (5c))
KEvery = (as defined in (5c))
KHe = (as defined in (28c))
KShe = (as defined in (28c))
KMerge-IV = (as defined in (28c))
KQI = (as defined in (37))
KOr(α,β) = < α’ or β’, Or >

Another new one we’ll add,
α β just for fun!
(51) Step Four: The Syntactic Algebra

< E, Kγ>γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI} is the algebra such that:
a. { Kγ }γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI} are as defined above, and
b. E is the smallest set such that
(i) For all δ ∈ Δ, Xδ ⊆ E
(ii) E is closed under { Kγ }γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI}
(52) Step Five: The Syntactic Rules

The set SE consists of the following triples:
a. < KMerge-S, < PR, IV >, S >

b. < KMerge-IV, < TV, PR >, IV >
c. < KNot, < S >, S >
d. < KAnd, < S, S >, S >
e. < KIf, < S, S >, S >
f. < KOr, < S, S >, S >
g. < KSome , < CN >, T >
h. < KEvery , < CN >, T >
i. < KHe , < IN >, PR >
j. < KShe , < IN >, PR >
k. < KQI , < IN, T, S >, S >
Note: Contrary to the picture in (12)-(14), the rules above do not allow us to KMerge-S a term T
with an IV to form an S.
• We could easily add such a rule, but will refrain from doing so until later…
19
Fall 2013 Ling 720
(53) The Disambiguated Language: ‘Disambiguated Mini-English+Qs’ (DME+Q)
The structure < E, Kγ, Xδ , SE, S >γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI}, δ ∈ Δ
where E, Kγ, Xδ , SE, and Δ are as defined in (48)-(52).
(54) Some Illustrative Members of CS

Students are encouraged to work out for themselves how the rules in (52) entail that the
following are all members of CS in DME+Q.
a. < he 2 smokes, Merge-S >
< he 2, He > < smokes , ∅ >
<2,∅>
b. < she 5 loves him 7, Merge-S >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
c. < Michelle loves every president, QI >
< every president, Every > <7, ∅> < Michelle loves him 7 , QI >
< president, ∅ > < Michelle, ∅> <5, ∅> < she 5 loves him 7, Merge-S >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
d. < some man loves every president, QI >
< every president, Every > <7, ∅> < some man loves him 7 , QI >
< president, ∅ > < some man, Some> <5, ∅> < he 5 loves him 7, Merge-S >
< man, ∅ > < he 5, He > < loves him 7 , Merge-IV >
<5,∅> < loves , ∅ > < he 7, He >
<7,∅>
20
Fall 2013 Ling 720
(55) The Fragment of English: Mini-English+Qs (ME+Q)
Mini-English+Qs is the following language, where R is as defined in (30):
<< E, Kγ, Xδ , SE, S >γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI}, δ ∈ Δ, R>
(56) Some Illustrative Members of CS in ME+Q

a. he smokes
b. she loves him
c. Michelle loves every president
d. some man loves every president
Note: As was observed in (41), sentence (56d) of ME+Q is syntactically ambiguous.
5. The Translation Base from Mini-English+Qs to Politics+λ
(57) Step One: The Syntactic Category Mapping

Let the function g: Δ  T ∪ { <var,τ> : τ ∈ T } be defined as follows:
a. g(TV) = <e, <e,t>>

b. g(IV) = <e, t>
c. g(S) = t
d. g(T) = < <e,t>, t >
e. g(CN) = <e, t>
f. g(IN) = <var, e>
g. g(PR) = e
(58) Step Two: The Polynomial Operations

Let { Hγ }γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI} be the following polynomial
operations over the syntactic algebra <A, Fγ >γ ∈ {Concat, Not, And, Or, If, ∃, ∀, λ} for Politics+λ.
a. HMerge-S = FConcat<Id2,2, Id1,1>

b. HMerge-IV = FConcat
c. HNot = FNot
d. HAnd = FAnd
e. HOr = FOr
f. HIf = FIf
g. HSome =
Fλ< CP0,1 , F∃ < Cx0,1 , FAnd < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>
h. HEvery =
Fλ< CP0,1 , F∀ < Cx0,1 , FIf < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>
i. HHe = Id1,1
j. HShe = Id1,1
k. HQI = FConcat< Id2,3 , Fλ< Id1,3, Id3,3 > >
21
Fall 2013 Ling 720
(59) Step Three: Checking for Derived Syntactic Rules
In order to employ the polynomial operations above in our translation base, it must the
case that each of the following are derived syntactic rules of Politics+λ.
Students are encouraged to confirm for themselves whether they are.
a. < FConcat <Id2,2, Id1,1>, < e, <e,t> >, t > (~ Rule (52a))
b. < FConcat , < <e,<e,t>>, e >, <e,t> > (~ Rule (52b))
c. < FNot , < t >, t > (~ Rule (52c))
d. < FAnd, < t, t >, t > (~ Rule (52d))
e. < FIf, < t, t >, t > (~ Rule (52e))
f. < FOr, < t, t >, t > (~ Rule (52f))
g. < Fλ< CP0,1 , F∃ < Cx0,1 , FAnd < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>,
< <e,t> >, <<e,t>, t> > (~ Rule (52g))
h. < Fλ< CP0,1 , F∀ < Cx0,1 , FIf < FConcat< Id1,1, Cx0,1>, FConcat< CP0,1 , Cx0,1>>>>,
< <e,t> >, <<e,t>, t> > (~ Rule (52h))
i. < Id1,1 < <var,e> > e > (~ Rule (52i) and (52j)) 2
j. < FConcat< Id2,3 , Fλ< Id1,3, Id3,3>> , < <var,e>, <<e,t>, t>, t>, t > (~ Rule (52k))
(60) Step Four: The Lexical Translation Function

The function j has as its domain { Xδ }δ ∈ Δ , and consists of the following mappings:
a. j(< loves, ∅ >) = loves’

b. j(< smokes, ∅ >) = smokes’
c. j(< Barack, ∅ >) = (λP0 (P0 barack’))
d. j(< Mitt, ∅ >) = (λP0 (P0 mitt’))
e. j(< Michelle, ∅ >) = (λP0 (P0 michelle’))
f. j(< man, ∅ >) = man’
g. j(< president, ∅ >) = president’
h. For all n ∈ ℕ, j(< n, ∅ >) = xn ( = ve,n )
(61) Definition of the Translation Base from Mini-English+Qs to Politics+λ

Let T be the structure < g , Hγ , j>γ ∈ {Merge-S, Merge-IV, Not, And, If, Or, Some, Every, He, She, QI}, where
g, Hγ , and j are as defined in (57)-(60). T is a translation base from Mini-English+Qs to
Politics+λ.
We can now use the translation function k determined by T to homomorphically map expressions
of DME+Q (analysis trees) to expressions of Politics+λ.
2
There’s actually a problem here for our definition of ‘derived syntactic rule’. Although (59i) is true when
informally read (‘Applying Id1,1 to an expression of category <var,e> yields an expression of category e’), it is not
strictly speaking a ‘derived rule of Politics+λ’, since the label <var,e> ≠ e. I don’t myself know of any possible
solutions to this problem. (Montague himself doesn’t run into it in UG, for various independent reasons…)
22
Fall 2013 Ling 720
(62) A Critical Inference Rule When Translating: Alpha-Conversion

If variable v is bound in ϕ, and variable v’ appears nowhere in ϕ, then ϕ is logically
equivalent to [v/v’]ϕ
Illustration: (λP0 ∃x0 ((man’ x0) & (P0 x0))) ⇔ (λP1 ∃x1 ((man’ x1) & (P1 x1)))
(63) Illustration of the Translation Function k

Let T1 be the tree for some man loves every president in (40a) [every > some].
(i) k(T1) =
(ii) k( KQI(<7, ∅>, KEvery(<president, ∅>),

KQI(<5, ∅>, KSome(<man, ∅>),
KMerge-S( KHe(<5 , ∅>), KMerge-IV(<loves , ∅>, KHe(<7 , ∅>)))))) =
(iii) HQI( k(<7, ∅>), HEvery( k(<president, ∅>) ),

HQI( k (<5, ∅>), HSome( k(<man, ∅>) ),
HMerge-S( HHe( k(<5 , ∅>) ), HMerge-IV( k(<loves , ∅>), HHe( k(<7 , ∅>)))))) =
(iv) HQI( j(<7, ∅>), HEvery( j(<president, ∅>) ),

HQI( j(<5, ∅>), HSome( j(<man, ∅>) ),
HMerge-S( HHe( j(<5 , ∅>) ), HMerge-IV( j(<loves , ∅>), HHe( j(<7 , ∅>)))))) =
(v) HQI( x7 , HEvery(president’),

HQI( x5 , HSome(man’),
HMerge-S( HHe(x5), HMerge-IV( loves’ , HHe(x7))))) =
(vi) HQI( x7 , HEvery(president’),

HQI( x5 , HSome(man’), ((loves’ x7) x5) ) =
(vii) HQI( x7 , (λP0 ∀x0 ((president’ x0)  (P0 x0))),

HQI( x5 , (λP0 ∃x0 ((man’ x0) & (P0 x0))), ((loves’ x7) x5) ) =
(viii) HQI( x7 , (λP0 ∀x0 ((president’ x0)  (P0 x0))),

( (λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx5 ((loves’ x7) x5)) ) ) =
(ix) ((λP0 ∀x0 ((president’ x0)  (P0 x0)))

(λx7 ((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx5 ((loves’ x7) x5)))))
(64) Remark
The end product of our translation process is the complex lambda-expression in (63ix).
However, thanks to α-conversion and λ-conversion, we can ‘transform’ this into a
simpler, logically equivalent formula.
23
Fall 2013 Ling 720
Note:
Because of all the variables shared between the sub-formulae in (63ix), we cannot immediately
apply λ-conversion to ‘simplify’ the formula; instead, we must first apply α-conversion to
change the bound variables so that they aren’t shared between the lambda-expressions.
(64) Simplifying the Translation in (63)
(i) ((λP0 ∀x0 ((president’ x0)  (P0 x0)))

(λx7 ((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx5 ((loves’ x7) x5))))) ⇔ (α-conversion)
(ii) ((λP0 ∀x0 ((president’ x0)  (P0 x0)))

(λx7 ((λP1 ∃x1 ((man’ x1) & (P1 x1))) (λx5 ((loves’ x7) x5))))) ⇔ (λ-conversion)
(iii) ∀x0 ((president’ x0) 

((λx7 ((λP1 ∃x1 ((man’ x1) & (P1 x1))) (λx5 ((loves’ x7) x5)))) x0)) ⇔ (λ-conversion)
(iv) ∀x0 ((president’ x0) 

((λP1 ∃x1 ((man’ x1) & (P1 x1)))(λx5 ((loves’ x0) x5)))) ⇔ (λ-conversion)
(v) ∀x0 ((president’ x0) 

∃x1 ((man’ x1) & ((λx5 ((loves’ x0) x5)) x1))) ⇔ (λ-conversion)
(vi) ∀x0 ((president’ x0)  ∃x1 ((man’ x1) & ((loves’ x0) x1)))
(65) Remark
• Again, strictly speaking, the translation of tree (40a) is the complex lambda
expression in (63ix).
o The simpler formula in (64iv) is not the translation of (40a).
• (65vi) is however, a logically-equivalent formula, one that allows us to more easily
‘see’ what the semantic value induced by our translation for (40a) is…
(66) Illustration of the Translation Function k

Let T2 be the tree for some man loves every president in (40b) [some > every].
(i) k(T1) =
(ii) k( KQI(<5, ∅>, KSome(<man, ∅>),

KQI(<7, ∅>, KEvery(<president, ∅>),
KMerge-S( KHe(<5 , ∅>), KMerge-IV(<loves , ∅>, KHe(<7 , ∅>)))))) =
(iii) HQI( k(<5, ∅>), HSome( k(<man, ∅>) ),

HQI( k (<7, ∅>), HEvery( k(<president, ∅>) ),
HMerge-S( HHe( k(<5 , ∅>) ), HMerge-IV( k(<loves , ∅>), HHe( k(<7 , ∅>)))))) =
24
Fall 2013 Ling 720
(iv) HQI( j(<5, ∅>), HSome( j(<man, ∅>) ),
HQI( j (<7, ∅>), HEvery( j(<president, ∅>) ),
HMerge-S( HHe( j(<5 , ∅>) ), HMerge-IV( j(<loves , ∅>), HHe( j(<7 , ∅>)))))) =
(v) HQI( x5 , HSome(man’),

HQI( x7 , HEvery(president’),
HMerge-S( HHe(x5), HMerge-IV( loves’ , HHe(x7))))) =
(vi) HQI( x5 , HSome(man’),

HQI( x7 , HEvery(president’), ((loves’ x7) x5) ) =
(vii) HQI( x5 , (λP0 ∃x0 ((man’ x0) & (P0 x0))),

HQI( x7 , (λP0 ∀x0 ((president’ x0)  (P0 x0))), ((loves’ x7) x5) ) =
(viii) HQI( x5 , (λP0 ∃x0 ((man’ x0) & (P0 x0))),

((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx7 ((loves’ x7) x5)))) =
(ix) ((λP0 ∃x0 ((man’ x0) & (P0 x0)))

(λx5 ((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx7 ((loves’ x7) x5)))) =
Once again, we can apply α-conversion and λ-conversion to ‘transform’ the translation in
(ix) into a simpler, logically-equivalent formula
(x) ((λP0 ∃x0 ((man’ x0) & (P0 x0)))

(λx5 ((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx7 ((loves’ x7) x5)))) ⇔ (α-conversion)
(xi) ((λP0 ∃x0 ((man’ x0) & (P0 x0)))

(λx5 ((λP1 ∀x1 ((president’ x1)  (P1 x1))) (λx7 ((loves’ x7) x5)))) ⇔ (λ-conversion)
(xii) ∃x0 ((man’ x0) &

((λx5 ((λP1 ∀x1 ((president’ x1)  (P1 x1))) (λx7 ((loves’ x7) x5))) x0)) ⇔ (λ-conversion)
(xiii) ∃x0 ((man’ x0) &

((λP1 ∀x1 ((president’ x1)  (P1 x1))) (λx7 ((loves’ x7) x0))) ⇔ (λ-conversion)
(xiv) ∃x0 ((man’ x0) &

∀x1 ((president’ x1)  ((λx7 ((loves’ x7) x0)) x1))) ⇔ (λ-conversion)
(xv) ∃x0 ((man’ x0) & ∀x1 ((president’ x1)  ((loves’ x1) x0)))
Again, the formula in (63xiv) is not the translation of tree (40b); the translation is (63ix).
However (63xv) is a simpler, logically-equivalent expression…
Thus, it offers a more ‘transparent’ representation of the meaning assigned to (40b)
25
Fall 2013 Ling 720
6. Inducing an Interpretation for Mini-English+Q
If we compose together our translation function k and our meaning assignment h for Politics+λ,
we now (thanks to our general theory of translation), obtain a meaning assignment for DME+Q.
(67) Illustration of the Induced Interpretation of Mini-English+Q

Let T1 be the tree for some man loves every president in (40a) [every > some].
(i) h°k(T1) = (by definition of composition)
(ii) h(k(T1)) = (by (63))
(iii) h(((λP0 ∀x0 ((president’ x0)  (P0 x0)))

(λx7 ((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx5 ((loves’ x7) x5)))))) = (by (64))
(iv) h(∀x0 ((president’ x0)  ∃x1 ((man’ x1) & ((loves’ x0) x1))))
= (by definition of interpretation B in (53) of previous handout)
(v) The function A such that if g ∈ J, A(g) = 1 iff for all for all x ∈ De,E ,
if k(x) = 1 then there exists a y ∈ De,E such that i(y) = 1 and j(x)(y)
= (by definition of interpretation B in (53) of previous handout)
(vi) The function A such that if g ∈ J, A(g) = 1
Note:
In our induced interpretation of Mini-English, tree (40a) – the ‘inverse-scope reading of some
man loves every president’ – is interpreted as being true (relative to any variable assignment).
(68) Illustration of the Induced Interpretation of Mini-English+Q

Let T2 be the tree for some man loves every president in (40b) [some > every].
(i) h°k(T2) = (by definition of composition)
(ii) h(k(T2)) = (by (66))
(iii) h(((λP0 ∃x0 ((man’ x0) & (P0 x0)))

(λx5 ((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx7 ((loves’ x7) x5))))) = (by (66))
(iv) h(∃x0 ((man’ x0) & ∀x1 ((president’ x1)  ((loves’ x1) x0)))) = (by def. of B)
(v) The function E such that if g ∈ J, E(g) = 1 iff there is an x ∈ De,E such that
i(x) = 1 and for all y ∈ De,E, if k(y) = 1 then j(y)(x) = 1 = (by def. of B)
(vi) The function E such that if g ∈ J, A(g) = 1
26
Fall 2013 Ling 720
Note:
In our induced interpretation of Mini-English, tree (40b) – the ‘surface-scope reading of some
man loves every president’ – is interpreted as being true (relative to any variable assignment).
7. Summary
In these notes, we’ve accomplished the following:
• Developed a fragment of English (ME+Q) which contains quantificational terms

(quantificational NPs)
• Provided a translation base homomorphically mapping the expressions of DME+Q to

expressions of Politics+λ.
• Via our interpretation for Politics+λ defined in the last handout, obtained an ‘induced’
interpretation for ME+Q.
The resulting system has the following advantageous properties:
• Interprets sentences where quantificational terms are in direct object position
• Correctly / automatically predicts quantifier scope ambiguities in sentences containing

more than one quantifier.
• Correctly / automatically predicts quantificational binding of pronouns (HOMEWORK!)
What’s Next on the Agenda?
• Thus far, all our syntactic and semantic analyses have been within framework as
presented in Montague’s paper “Universal Grammar.”
• In his paper “PTQ”, however, Montague employs a relatively simplified presentation of

the system, one that allows for a much more transparent / readable treatment of opaque /
intensional contexts.
• Therefore, to build towards that, I will next ‘transform’ the system we developed in the
last few handouts into a system akin to that presented in PTQ.
o Again, the substance of the analysis will remain the same (as you’ll see)
o All that really differs is the notation / presentation employed…
27
Fall 2013 Ling 720
Problem Set on the Analysis of Quantification
The exercises below make reference to the handout “An Algebraic Approach to Quantification
and Lambda Abstraction: Applications to the Analysis of English.”
(1) Pronominal Binding in Our Fragment of English
a. Please show how our English fragment in (55) and our translation base in (61)
together predict that (i) receives a translation logically equivalent to (ii).
(i) Michelle loves some man and he smokes.
(ii) ∃x0 ((man’ x0) & (((loves’ x0) michelle’) & (smokes’ x0)))
b. Does the sentence below raise any problems for our system?
(i) Michelle loves every man and he smokes.
(2) Another Exercise in Indirect and Direct Interpretation
a. Minimally alter our English fragment in (55) so that its expressions now include
strings like no man smokes and Michelle loves no man.
b. Minimally alter our translation base in (61) so that strings like no man smokes
receive appropriate translations in Politics+λ.
Note: For this exercise, you don’t have to show that anything is a derived
syntactic rule.
c. Please show how the new translation base, along with the interpretation for
Politics+λ presented in (53) of the handout “Fregean Interpretations” assign a
meaning to the sentence no man smokes.
d. Given your proposed translation base, construct a direct interpretation of Mini-

English+Qs.
1
Fall 2013 Ling 720
(3) One Final Exercise on Direct Interpretation
a. Minimally alter our English fragment in (55) so that its expressions now include
strings like most man smokes and Michelle loves most man.
Note: For this exercise, please disregard the need for most in English to
appear with a plural NP.
b. Provide a direct interpretation of the resulting fragment.
HINT:
For this problem, you should aim to predict that most man smokes has a meaning that maps a
variable assignment g to 1 iff the set of men who smoke is larger than the set of men who don’t
smoke.
That is, it maps g to 1 iff: | {x : x in De, E & h(man)(g)(x) = 1 & h(smoke)(g)(x) = 1 }| >
| {x : x in De, E & h(man)(g)(x) = 1 & h(smoke)(g)(x) = 0 }|
Fun Fact:
• There is no formula of Politics+λ that will have the meaning outlined above.
• Consequently, we cannot use Politics+λ to provide an indirect interpretation of the

fragment in (3a).
• This highlights the way in which our ability to do indirect interpretation of natural
language is limited by the logical languages we have at our disposal, while there are no
such limits on our ability to do direct interpretation…
2

Unit 6:
The Proper Treatment of Quantification in Ordinary
English

Fall 2013 Ling 720
First Steps Towards PTQ: A New Presentation of Our System for Quantifiers
(1) From UG to PTQ
• By far and away, the most cited paper by Montague is “The Proper Treatment of
Quantification in Ordinary English” (PTQ).
• When compared to “Universal Grammar”, PTQ is a relatively informal presentation

of Montague’s key ideas.
o It was written as a paper for the 1970 Stanford Workshop on Grammar and
Semantics.
• Consequently, PTQ doesn’t hold to letter of the UG system and its notations, and
introduces certain simplifications…
o Many aspects of Montague’s analysis in PTQ can be easily rephrased in the
strictly algebraic UG framework
o Others, as we will see, cannot (without sacrificing some elegance)
In these notes, we will take the analysis of English quantification developed in the last few
handouts, and ‘convert’ it into a format closer to what is found in PTQ…
• Many of the changes are simply superficial ones of terminology and notation…
1. The Fragment ‘Mini-English+Q’: PTQ-Style Presentation
(2) Redefinition: Category

In PTQ, the indices that we’ve been referring to as ‘category labels’ (Δ in UG) are instead
dubbed simply ‘categories’. The set of ‘categories’ is represented as Cat (~Δ in UG)
o This more closely approximates the terminology in generative linguistics.
(3) The Categories of ME+Q: Cat = {TV, IV, S, T, CN, PR}
(4) New Notation: Basic Expressions

In PTQ, BA is the set of basic expressions of category A. (~Xδ in UG)
(5) The Basic Expressions of ME+Q

BTV = { loves }
BIV = { smokes }
BT = { Barack, Mitt, Michelle }
BCN = { man, president, woman }
BPR = { he n : n ∈ ℕ } ∪ { she n : n ∈ ℕ }
BS = ∅
1
Fall 2013 Ling 720
(6) Remarks
• Our English expressions are now boldfaced rather than italicized.
• Following PTQ (rather than UG), pronouns are now primitive expressions, rather than
ones syntactically derived from indices.
• To set up something interesting for later, we’ve also added woman to our set of CNs.
• Note that our English expressions are now strings again, rather than trees. As
we’ll see, this won’t pose any problems for the PTQ ‘version’ of MG….
(7) New Terminology: Phrases

In PTQ, the sets that we’ve been referring to as (e.g.) ‘the category δ’ (Cδ in UG) are
instead dubbed ‘the phrases of category δ’.
• PA = The phrases of category A

As in UG, the set ∪A∈Cat PA (~ ∪δ∈Δ Cδ in UG) will be defined via the syntactic rules.
• In PTQ, however, a ‘syntactic rule’ is a rather different object from that in UG.
• As shown below, in PTQ, the ‘rules’ incorporate both (i) the definition of the
syntactic operations in the language, and (ii) the information contained in a UG ‘rule’.
(9) The Rule S1
a. The Rule: BA ⊆ PA, for every category A.
b. Remarks:
• This is clearly not a ‘syntactic rule’, in the sense found in UG.
• What S1 does is move to the system of ‘syntactic rules’ a condition that was
(in UG) part of the general definition of the set of categories (∪δ∈Δ Cδ)
(10) The Rule S2
a. The Rule:
If ζ ∈ PCN, then F0(ζ), F2(ζ) ∈ PT, where F0(ζ) = every ζ and F2(ζ) = some ζ
b. Remarks:
• Again, this is not a ‘syntactic rule’, in the sense found in UG.
• What S2 does is combine together the information we had earlier factored out
into (i) the definitions of KEvery and KSome, and (ii) the rules <KEvery, <CN>, T>
and <KSome, <CN>, T>
2
Fall 2013 Ling 720
Note:
• Following Montague in PTQ (and UG), the indices on our rules will no longer be
evocative mnemonics. Rather, they will simply be numerals.
• I will also be numbering both rules and operations in a way that matches the
corresponding rules and operations in PTQ (thus, F0 for every and F2 for some).
(11) The Rule S4
a. The Rule: If α ∈ PPR and δ ∈ PIV, then F4(α,δ) ∈ PS, where F4(α,δ) = α δ
b. Remarks:
Again, as with S2 and the rules below, this ‘syntactic rule’ combines together the
information that in the UG system was factored out into:
• The definition of KMerge-S , and
• The rule < KMerge-S, < PR, IV >, S >
(12) The Rule S5
a. The Rule:
If δ ∈ PTV and β ∈ PPR, then F5(δ,β) ∈ PIV, where F5(δ,β) = δ him n if β has the
form he n , F5(δ,β) = δ her n if β has the form she n, and F5(δ,β) = δ β otherwise.
b. Remarks: Again, this rule combines together the following information:

• The definition of KMerge-IV
• The rule < KMerge-IV, < TV, PR >, IV >
(13) The Rule S11
a. The Rule:
If ϕ, ψ ∈ PS, then F8(ϕ, ψ), F9(ϕ, ψ) ∈ PS , where F8(ϕ,ψ) = ϕ and ψ, and F9(ϕ,ψ)
= ϕ or ψ.
b. Remarks:
• Again, we see how this one ‘syntactic rule’ contains both (i) the definition of
the syntactic operations F8 (~ KAnd) and F9 (~ KOr), and (ii) the UG-syntactic
rules references those operations.
• I will follow Montague in PTQ by only having these rules and operations; I
will henceforth drop the operations KIf, KNot, and the rules involving them.
• After all, they give us horrible analyses of conditionals and negation
in English ; )
3
Fall 2013 Ling 720
(14) The Rule S14
a. The Rule:
If α ∈ PT and ϕ ∈ PS, then F10,n(α,ϕ) ∈ PS , where F10,n(α,ϕ) comes from ϕ by
replacing the first occurrence of he n , him n , she n , or her n, by α.
b. Remarks:
• Again, this one ‘syntactic rule’ combines together both:
(i) The definition of the operations F10,n(α,ϕ) (~ KQI(<n,∅>,α,ϕ) )
(ii) The UG-style rule < F10,n , <T, S>, S>
• Note that this rule doesn’t appeal to a single syntactic operation, but a
whole infinite family of them, { F10,n : n ∈ ℕ }
With the rules above, we are able to offer the following definition of the set of phrases for ME+Q
(15) The Phrases of Mini-English+Q

{ PA }A∈Cat is the smallest family of sets indexed by Cat such that S1-S14 are true.
(16) The Meaningful Expressions of English

ϕ is a meaningful expression of ME+Q if there is an A ∈ Cat such that ϕ ∈ PA.
(17) Some Illustrative Members of PS
a. Every man smokes
(i) man ∈ BCN , smokes ∈ BIV , he 1 ∈ BPR (by (5))

(ii) man ∈ PCN , smokes ∈ PIV , he 1 ∈ PPR (by S1)
(iii) he 1 smokes ∈ PS (by S4)
(iv) every man ∈ PT (by S2)
(v) every man smokes ∈ PS (by S14)
Analysis Tree:
every man smokes , 10,1
every man , 0 he 1 smokes , 4
man he 1 smokes
4
Fall 2013 Ling 720
b. Michelle loves every president
(i) Michelle ∈ BT , president ∈ BCN , loves ∈ BTV , she 1, he 2 ∈ BPR (5)

(ii) Michelle ∈ PT , president ∈ PCN , loves ∈ PTV , she 1, he 2 ∈ PPR (S1)
(iii) loves him 2 ∈ PIV (S5)
(iv) she 1 loves him 2 ∈ PS (S4)
(v) Michelle loves him 2 ∈ PS (S14)
(vi) Michelle loves every president ∈ PS (S14)
Analysis Tree: Michelle loves every president, 10,2
every president, 0 Michelle loves him 2, 10,1
president Michelle she 1 loves him 2, 4
she 1 loves him 2, 5
loves he 2
c. Some man loves every president.
(i) man, president ∈ BCN , loves ∈ BTV , he 1, he 2 ∈ BPR (5)

(ii) man, president ∈ PCN , loves ∈ PTV , he 1, he 2 ∈ PPR (S1)
(iii) loves him 2 ∈ PIV (S5)
(iv) he 1 loves him 2 ∈ PS (S4)
(v) some man, every president ∈ PT (S2)
(vi) some man loves him 2 ∈ PS / he 1 loves every president ∈ PS (S14)
(vii) some man loves every president ∈ PS (S14)
Analysis Tree 1: some man loves every president, 10,2
every president, 0 some man loves him 2, 10,1
president some man, 2 he 1 loves him 2, 4
man he 1 loves him 2, 5
loves he 2
Analysis Tree 2: some man loves every president, 10,1
some man, 2 he 1 loves every president, 10,2
man every president, 0 he 1 loves him 2, 4
president he 1 loves him 2, 5
loves he 2
5
Fall 2013 Ling 720

As shown in (17c), there are meaningful expressions of ME+Q that are genuinely
syntactically ambiguous.
• ME+Q as defined above would not be a ‘disambiguated language’ in the sense of UG.
(19) Quote by Montague

“Thus our fragment admits genuinely (that is, semantically) ambiguous sentences. If it
were desired to construct a corresponding unambiguous language, it would be convenient
to take the analysis trees themselves as the expressions of that language; it would then be
obvious how to characterize…the structural operations of that language and the
correspondence relation between its expressions and those of ordinary English. For
present purposes, however, no such construction is necessary.” (Montague 1974; p. 23).
No such construction is necessary?...

As we will see shortly, translation in PTQ is simply a relation, and need not be a function
2. The Logical Language TL: PTQ-Style Presentation
(20) Logical Languages in PTQ: First Key Difference
• In PTQ, Montague does not represent the logical translation language as a

‘disambiguated language’ in the sense of UG.
o Rather, he gives a (relatively) simple, recursive definition of its syntax.
• In PTQ, Montague does not represent the semantics of the logical translation
language as a ‘(Fregean) interpretation’ in the sense of UG.
o Rather, he gives a (relatively) simple, model-theoretic semantics.
• However, given the equivalences between these systems for syntax and
semantics, it is possible (though laborious) to convert between them (see UG)
o Thus, there is no substantive difference in the theory of the syntax/semantics

of the logical translation language (the presentation in PTQ is just ‘simpler’)
(21) Logical Languages in PTQ: The Second Key Difference
• Up to now, we’ve been translating our natural language into a ‘tailor-made’ logical
language with a finite set of constants.
• Technically speaking, in PTQ (and UG), Montague doesn’t do that, but rather
lays out a logical language with an infinite set of constants
o That is, in PTQ and UG, man’ isn’t the translation of man, but rather a meta-
language abbreviation for “whatever constant c is the translation of man”
6
Fall 2013 Ling 720
In what follows, we define the syntax and semantics of a single language, dubbed ‘Typed Logic’.
(22) The Vocabulary of Typed Logic (TL)


(i) Constants:
For every type τ ∈ T, a countably infinite set of constants of type τ:
CONτ = { cτ, n : n ∈ ℕ }
(ii) Variables:
VARτ = { vτ, n : n ∈ ℕ }
(23) Meta-Language Abbreviations for Variables and Constants

Although our variables ‘officially’ all look like those in (22c), to save space we will make
use of the following meta-language abbreviations:
a. xn = ve,n b. Pn = v<e,t>,n
c. an = ce,n d. Qn = c<e,t>,n e. Rn = c<e, <e,t>>,n
(23) The Syntax of TL
a. If ϕ ∈ CONτ or ϕ ∈ VARτ, then ϕ ∈ MEτ
b. If ϕ ∈ ME<σ,τ> and ψ ∈ MEσ, then (ϕ ψ) ∈ MEτ
c. If ϕ, ψ ∈ MEt , then (i) ~ϕ ∈ MEt (ii) (ϕ & ψ) ∈ MEt

(iii) (ϕ ∨ ψ) ∈ MEt (iv) (ϕ  ψ) ∈ MEt
d. If v ∈ VARτ, and ϕ ∈ MEt , then (i) ∃vϕ ∈ MEt (ii) ∀vϕ ∈ MEt
e. If v ∈ VARσ, and ϕ ∈ MEτ , then (λv ϕ) ∈ ME<σ,τ>
7
Fall 2013 Ling 720
(24) Some Illustrative Meaningful Expressions of TL
a. ~ ((R3 a2) a1) [~ ((loves’ mitt’) barack’) in Politics+λ]
b. ∀x3 ( (Q2 x3)  ~ ((R3 a2) x3)) [∀x3((smokes’ x3)  ~ ((loves’ mitt’) x3))]
c. ∃P4 (P4 a3) [∃P4 (P4 michelle’) in in Politics+λ]
d. ( (λx3 (Q3 x3)) a2 ) [((λx3 (man’ x3)) mitt’) in Politics+λ]
e. (λP4 ∀x3 ( (Q3 x3)  (P4 x3) ) ) [(λP4 ∀x3 ((man’ x3)  (P4 x3))), Politics+λ]
Our semantics for TL is going to exactly follow our model-theoretic semantics for Politics+λ
(25) The Semantics of TL, Part 1: The Denotations Based on a Set E

Let T be the set of types and E be some non-empty set (of entities). If τ ∈ T, then the set
Dτ, E of denotations of type τ based on E is defined as follows:
(i) De,E = E
(ii) Dt,E = { 0, 1 }
(26) The Semantics of TL, Part 2: A Model for TL

A model M for TL is a pair <E, I> consisting of:
a. A non-empty set E, called the ‘domain of M’
b. A function I, whose domain is equal to (i) and whose range satisfies the condition
in (ii).
(i) Domain of I: ∪τ ∈ T CONτ
(ii) Condition on Range of I: If α ∈ CONτ , then I(α) ∈ Dτ,E
(27) The Semantics of TL, Part 3: Variable Assignments

Let M be a model <E, I> of TL. Then g is a variable assignment (based on M) if its
domain is equal to (i) and its range satisfies the property in (ii).
(iii) Condition on Range of g: If α ∈ VARτ , then g(α) ∈ Dτ,E
8
Fall 2013 Ling 720
(28) The Semantics of TL, Part 4: Interpretation w.r.t. Model and Variable Assignment
Let M be a model <E, I> for TL and g be a variable assignment based on M. The
interpretation (a.k.a. denotation) of a meaningful expression of TL relative to M and g
[[.]]M,g is defined as follows:
a. If v ∈ ∪τ ∈ T VARτ , then [[ v ]]M,g = g(v)
b. If α ∈ ∪τ ∈ T CONτ , then [[ α ]]M,g = I(α)
c. If ϕ = (ψ χ), then [[ϕ ]]M,g = [[ψ]]M,g([[χ]]M,g)
d. If ϕ = ~ψ, then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0
e. If ϕ = (ψ & χ), then [[ϕ]] = 1 iff [[ψ]]M,g = 1 and [[χ]]M,g = 1
M,g
f. If ϕ = (ψ ∨ χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 1 or [[χ]]M,g = 1

g. If ϕ = (ψ  χ), then [[ϕ]]M,g = 1 iff [[ψ]]M,g = 0 or [[χ]]M,g = 1
h. If ϕ = ∃vψ and v ∈VARτ , then [[ϕ]]M,g = 1 iff
there is an a ∈ Dτ ,E such that [[ψ]]M,g(v/a) = 1
i. If ϕ = ∀vψ and v ∈VARτ , then [[ϕ]]M,g = 1 iff for all a ∈ Dτ ,E , [[ψ]]M,g(v/a) = 1
j. If ϕ = (λvψ), v ∈VARσ and ψ ∈ MEτ , then [[ϕ]]M,g =
The function p whose domain is Dσ ,E , whose range is Dτ ,E and for all a ∈ Dσ ,E ,
p(a) = [[ψ]]M,g(v/a)
(29) Illustration of the Semantics: A Model for TL

Let the model M be the pair <{Barack, Michelle, Mitt}, I>, where I contains the
following mappings (amongst infinitely many others):
a. I(a1) = Barack
b. I(a2) = Mitt
c. I(a3) = Michelle
d. I(Q2) = h = { <Michelle, 0>, <Barack,1>, <Mitt,0> }

e. I(Q3) = i = { <Michelle, 0>, <Barack,1>, <Mitt,1> }
f. I(Q1) = k = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
e. f(R3) = j = Michelle  Michelle  1

Barack  1
Mitt  0
Barack  1
Mitt  0
Barack  0
Mitt  1
9
Fall 2013 Ling 720
(30) Illustration of the Semantics: Interpretation of Illustrative Expressions
Let M be the model defined in (29). Let g be some arbitrary variable assignment based
on M.
a. [[(λP4 (P4 a2))]]M,g =

The function p with domain D<et>,E , range Dt,E and for all a ∈ D<et>,E ,
p(a) = a(Mitt)
b. [[(λP4 ∀x3 ((Q3 x3)  (P4 x3)))]]M,g =

The function p with domain D<et>,E, range Dt,E and for all a ∈ D<et>,E ,
p(a) = 1 iff for all a’ ∈ De,E, either i(a’) = 0 or a(a’) = 1 =
The function p with domain D<et>,E, range Dt,E and for all a ∈ D<et>,E ,
p(a) = 1 iff for all a’ ∈ De,E, if i(a’) = 1 then a(a’) = 1 =
The characteristic function of the set of ‘properties every man has’
3. The Translation from ME+Q to TL: PTQ-Style Presentation
In PTQ, the system for translating from ME+Q to TL differs slightly from that in UG.
• As we will see, however, the differences are not that deep or fundamental…
(31) First Ingredient: Category-to-Type Mapping

Just like with the UG notion of a translation base, one of the key ingredients to the PTQ
translation system is a mapping from Cat to the set of types (in our logical language).
f(TV) = <e, <e,t>> f(IV) = <e,t>

f(S) = t f(T) = <<e,t>,t>
f(CN) = <e,t> f(PR) = e
Note: The mapping above is not at all the one that actually appears in PTQ.
Recall that we’re right now just ‘converting’ our system from the last handout
into the PTQ style.
(32) Second Ingredient: Lexical Translation Function

Again, just as in UG’s notion of a translation base, the second key ingredient to the PTQ
translation system is a function mapping the basic expressions of ME+Q to ones in TL.
Let g be a function whose domain is the set of basic expressions in (5) except for BT and
BPR, and for all A ∈ Cat, α ∈ BA, and α ∈ Domain(g), g(α) ∈ CONf(A)
o That is, g maps expressions of category A to constants of the corresponding type

o This restriction that g only maps to constants is unique to PTQ
10
Fall 2013 Ling 720
(33) Meta-Language Abbreviations
Recall that the constants of our language TL are all of the form cτ, n. In what follows,
we’ll make use of the following meta-language abbreviations.
a. loves’ = g(loves) [whatever <e,<e,t>> constant g maps loves to]

b. smokes’ = g(smokes) [whatever <e,t> constant g maps smokes to]
c. man’ = g(man) [whatever <e,t> constant g maps man to]
d. president’ = g(president) [whatever <e,t> constant g maps president to]
e. woman’ = g(woman) [whatever <e,t> constant g maps woman to]
(34) Third Ingredient: Translation Rules
In the PTQ-system, the work done in UG by the translation base and the definition of the
polynomial operations is instead done by a system of ‘translation rules’.
• These ‘translation rules’ stand in a one-to-one correspondence with the syntactic rules
in (9)-(14)
o Much as how in UG the polynomial operations are in correspondence with the
syntactic operations, and the derived syntactic rules with the syntactic rules.
• As we’ll see, these ‘translation rules’ do the work of both (i) defining the ‘polynomial
operation’ each syntactic operation corresponds to; (ii) putting the operations in
correspondence.
(35) The Rule T1
a. The Rule: Rule T1 consists of a collection of sub-rules.
(i) If α is in the domain of g, then α translates to g(α).
(ii) Barack translates to (λP4 (P4 barack’)), where barack’ ∈ CONe

Mitt translates to (λP4 (P4 mitt’)), where mitt’ ∈ CONe
Michelle translates to (λP4 (P4 michelle’)), where michelle’ ∈ CONe
(iii) he n and she n translate to: xn (= ve,n)
b. Remarks:
• This ‘translation rule’ incorporates the following information:
(i) The lexical translation function g is a subset of the full translation relation
(ii) Proper names translate as <<e,t>,t> formula; pronouns as type-e variables.
• In the UG system, the information in (ii)-(iii) is part of the lexical

translation function; in PTQ, though, that function can only map to
constants (and so we need to pack it in as a separate rule)
11
Fall 2013 Ling 720
(36) The Rule T2
a. The Rule: If ζ ∈ PCN and ζ translates to ζ’, then

(i) F0(ζ) translates to (λP0 ∀x0 ((ζ’ x0)  (P0 x0)))
(ii) F2(ζ) translates to (λP0 ∃x0 ((ζ’ x0) & (P0 x0))),
b. Remarks: As mentioned in (34), rule T2 does both the following:

(i) Defines the ‘polynomial operations’ corresponding to F0 (= HEvery) and F2
(= HSome)
(ii) Puts these polynomial operations ‘in correspondence’ with F0 and F2 in the
translation relation.
(37) The Rule T4
a. The Rule:
If δ ∈ PPR and β ∈ PIV, and δ, β translate to δ’, β’ respectively, then F4(δ,β)
translates to (β’ δ’)
b. Remarks: Again, as mentioned in (34), rule T4 does both the following:

(i) Defines the ‘polynomial operation’ corresponding to F4 (= HMerge-S)
(ii) Puts that polynomial operation in correspondence with F4 in the translation
relation.
(38) The Rule T5

If δ ∈ PTV and β ∈ PPR, and δ, β translate to δ’, β’ respectively, then F5(δ,β) translates to
(δ’ β’).
(39) The Rule T11

If ϕ,ψ ∈ PS, and ϕ, ψ translate to ϕ’, ψ’ respectively, then F8(ϕ,ψ) translates to (ϕ’ & ψ’)
and F9(ϕ,ψ) translates to (ϕ’ ∨ ψ’).
(40) The Rule T14
a. The Rule:
If α ∈ PT and ϕ ∈ PS, and α, ϕ translate to α’, ϕ’ respectively, then F10,n(α,ϕ)
translates to ( α’ (λxn ϕ’) )
b. Remarks:
• As in the corresponding syntactic rule S14, this translation rule covers not
simply one syntactic operation, but a whole infinite family of them.
• This translation rule implicitly does the work of defining an infinite family of
polynomial operations H10,n, each corresponding to F10,n
12
Fall 2013 Ling 720
(41) The Translation Relation
The translation relation translates to between expressions of ME+Q and those of TL is the
smallest binary relation satisfying T1-T14.
Given the correspondence between the syntactic rules S1-S14 and the translation rules T1-T14,
we can build up the translation for a sentence (rule-by-rule) as we construct it.
(42) Illustration of the Translation Rules, Part 1
(i) man, loves, president translate to man’, loves’, president’ respectively ((33), T1)
(ii) he 1 and he 2 translate to x1 and x2 respectively (T1)
(iii) F5(loves, he 2) translates to (loves’ x2) (T5)
(iv) loves him 2 translates to (loves’ x2) (def. of F5)
(v) F4(he 1, loves him 2) translates to ((loves’ x2) x1) (T4)
(vi) he 1 loves him 2 translates to ((loves’ x2) x1) (def. of F4)
(vii) F2(man) translates to (λP0 ∃x0 ((man’ x0) & (P0 x0))) (T2)
(viii) some man translates to (λP0 ∃x0 ((man’ x0) & (P0 x0))) (def. of F2)
(ix) F10,1(some man, he 1 loves him 2) translates to

((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx1 ((loves’ x2) x1))) (T14)
(x) some man loves him 2 translates to

((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx1 ((loves’ x2) x1))) (def. of F10,1)
(xi) F0(president) translates to (λP0 ∀x0 ((president’ x0)  (P0 x0))) (T2)
(xii) every president translates to (λP0 ∀x0 ((president’ x0)  (P0 x0))) (def. of F0)
(xiii) F10,2(every president, some man loves him 2) translates to

((λP0 ∀x0 ((president’ x0)  (P0 x0)))
(λx2 ((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx1 ((loves’ x2) x1))))) (T14)
(xiv) some man loves every president translates to

((λP0 ∀x0 ((president’ x0)  (P0 x0)))
(λx2 ((λP0 ∃x0 ((man’ x0) & (P0 x0))) (λx1 ((loves’ x2) x1))))) ⇔ (α- and λ-conv.)
(xv) ∀x0 ((president’ x0)  ∃x2 ((man’ x2) & ((loves’ x0) x2)))
13
Fall 2013 Ling 720
(43) Illustration of the Translation Rules, Part 2
Steps (i)-(vi) are exactly the same as those in (42):
(vii) F0(president) translates to (λP0 ∀x0 ((president’ x0)  (P0 x0))) (T2)
(viii) every president translates to (λP0 ∀x0 ((president’ x0)  (P0 x0))) (def. of F0)
(ix) F10,2(every president, he 1 loves him 2) translates to

((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 ((loves’ x2) x1))) (T14)
(x) he 1 loves every president translates to

((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 ((loves’ x2) x1))) (def. of F10,2)
(xi) F2(man) translates to (λP0 ∃x0 ((man’ x0) & (P0 x0))) (T2)
(xii) some man translates to (λP0 ∃x0 ((man’ x0) & (P0 x0))) (def. of F2)
(xiii) F10,1(some man, he 1 loves every president) translates to:

((λP0 ∃x0 ((man’ x0) & (P0 x0)))
(λx1 ((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 ((loves’ x2) x1))))) (T14)
(xiv) some man loves every president translates to:

((λP0 ∃x0 ((man’ x0) & (P0 x0)))
(λx1 ((λP0 ∀x0 ((president’ x0)  (P0 x0))) (λx2 ((loves’ x2) x1))))) ⇔ (α- and λ-conv.)
(xv) ∃x0 ((man’ x0) & ∀x2 ((president’ x2)  ((loves’ x2) x0)))

As shown by (42) and (43), the relation ‘translates to’ is not a function.
• When our syntactic derivation of some man loves every president follows the
procedure in Analysis Tree 1 in (17c), the translation is logically equivalent to:
∀x0 ((president’ x0)  ∃x2 ((man’ x2) & ((loves’ x0) x2)))
• When our syntactic derivation of some man loves every president follows the
procedure in Analysis Tree 2 in (17c), the translation is logically equivalent to:
∃x0 ((man’ x0) & ∀x2 ((president’ x2)  ((loves’ x2) x0)))
That is, syntactically ambiguous strings in ME+Q can be paired with more than one
translation, corresponding to the different ways the strings can be derived in the syntax.
• Of course, if the language were not syntactically ambiguous, then the translation
relation would be a function….
14
Fall 2013 Ling 720
4. Introducing ‘Meaning Postulates’
(45) A Potential Issue for Our System

Our semantics for TL set up in Section 2 was quite general and permissive.
Consequently, there are models for TL that would not be appropriate as (induced)
interpretations of ME+Q.
Illustration:
• It is perfectly consistent with the definitions in Section 2 for a model of TL to map
the constants man’ and woman’ to the same, or overlapping <et>-functions.
• However, it is (maybe) part of the grammar of English that these terms are antonyms.
(46) Relevant Quote from Montague

“The interpretations of intensional logic may, by way of the translation relation, be made
to play a second role as interpretations of English. Not all interpretations of intensional
logic, however, would be reasonable candidates for interpretations of English. In
particular, it would be reasonable in this context to restrict attention to those
interpretations of intensional logic in which the following formulas are true…”
(Montague 1974; p. 27)
(47) The Solution to Issue (45)

A ‘logically permissible model of TL’ is one in which formulae (47a,b) are true:
a. ∀x0 ((man’ x0)  ~(woman’ x0))
b. ∀x0 ((woman’ x0)  ~(man’ x0))
In our analysis of English, we only consider ‘logically permissible models of TL’.
(48) On ‘Meaning Postulates’
• The term ‘meaning postulate’ is often used (though not by Montague himself) to refer
to such conditions on models for a language.
• Though the example in (47) is dubious,1 it’s possible to imagine more plausible cases:
o ∀x0 ((bachelor’ x0)  (man’ x0))
o ∀x0 ((wine’ x0)  ∃x1 ((grape’ x1) & ((made-from’ x1) x0)))
• In general, such ‘meaning postulates’ allow us to encode facts about lexical semantics
into our overall analysis of the natural language.
o Whether this is a good theory of lexical semantics, though, is up for debate…
1
After all, one could argue that is simply a contingent biological fact that no men are women, rather than it being a
part of the lexical semantics of ‘man’ and ‘woman’.
15
Fall 2013 Ling 720
5. Direct and Indirect Interpretation in the PTQ-Style System
As should be obvious by now, given our model in (29), our translation system from Section 3
allows us to ‘indirectly interpret’ strings of ME+Q.
(49) The Interpretation of ME+Q Relative to a Model and Variable Assignment
Let ϕ be a meaningful expression of ME+Q, and let ϕ translate to ϕ’. Let M be a

(logically permissible) model for TL and g be a variable assignment based on M. We say
that [[ϕ’]]M, g is the interpretation of ϕ under translation ϕ’ relative to M and g.
Illustration:
• The interpretation of some man loves every president under translation (42) and
relative to the model in (29) is 1
o After all, Barack is the only president, and he loves himself.
• The interpretation of some man loves every president under translation (43) and
relative to the model in (29) is also 1
o Again, this is rendered true by the fact that Barack loves himself.
(50) The Equivalence of Indirect and Direct Interpretation in UG

Once a translation base T is is specified for L and L’, if L’ has an interpretation B’, it is
trivial (mechanical) to specify an interpretation B for L.
(51) a. Question: Is indirect interpretation similarly ‘eliminable’ in a PTQ-style system?
b. Answer:
Yes, but it is a little less trivial/mechanical to build the model M for natural
language L on the basis of model M’ for logical language L’ and the translation.
(52) A General Method for Recasting Indirect Interpretation as Direct Interpretation

For each ‘translation rule’ Tn , specify a new ‘interpretation rule’ In , where the output of
In is the model-theoretic interpretation of Tn
16
Fall 2013 Ling 720
(53) Direct Interpretation of ME+Q, Step 1: Defining the Models

A model M for ME+Q is a pair <E, I> consisting of (i) a non-empty set E, called the
‘domain of M’, and (ii) a function I, whose domain is (a) and whose range satisfies the
condition in (b).
(a) Domain of I: BTV ∪ BIV ∪ BCN
(b) Condition on Range of I:
1. If ϕ ∈ BTV, then I(ϕ) ∈ D<e,<e,t>>, E
2. If ϕ ∈ BIV, then I(ϕ) ∈ D<e,t>, E
3. If ϕ ∈ BCN, then I(ϕ) ∈ D<e,t>, E
(54) Illustration: A Model for ME+Q

Let the model M be the pair <{Barack, Michelle, Mitt}, I>, where I contains the
following mappings (amongst infinitely many others):
a. I(smokes) = h = { <Michelle, 0>, <Barack,1>, <Mitt,0> }

b. I(man) = i = { <Michelle, 0>, <Barack,1>, <Mitt,1> }
c. I(president) = k = { <Michelle, 0>, <Barack,1>, <Mitt,0> }
d. I(woman) = l = { <Michelle, 1>, <Barack,0>, <Mitt,0> }
e. f(loves) = j = Michelle  Michelle  1

Barack  1
Mitt  0
Barack  1
Mitt  0
Barack  0
Mitt  1
(55) Direct Interpretation of ME+Q, Step 2: Defining Variable Assignment

Let M be a model <E, I> for ME+Q. Then g is a variable assignment (based on M) if
its domain is equal to (i) and its range satisfies the property in (ii).
(i) Domain of g: ℕ
(iv) Condition on Range of g: For all n ∈ ℕ, g(n) ∈ E
We now must lay out a definition for ‘interpretation relative to a model and a variable
assignment’…
Notice how the definition below mirrors our translation rules T1-T14
17
Fall 2013 Ling 720
(56) Direct Interpretation of ME+Q, Step 3: Interpretation w.r.t. Model and Assignment
Let M be a model <E, I> for ME+Q and g be a variable assignment based on M. The
relation ‘X is an interpretation of ϕ relative to M and g’ [[.]]M,g is defined as follows:
a. Rule I1:
(i) If ϕ is in the domain of I, then [[ϕ]]M,g = I(ϕ)
(ii) [[Barack]]M,g =
the function p with domain D<e,t>,E such that for all f ∈ D<e,t>,E
p(f) = 1 iff f(β) = 1 (where β is some specified member of E)
[[Mitt]]M,g =
p(f) = 1 iff f(µ) = 1 (where µ is some specified member of E)
[[Michelle]]M,g =
p(f) = 1 iff f(ν) = 1 (where ν is some specified member of E)
(iii) [[he n]]M,g = [[she n]]M,g = g(n)
b. Rule I2:
(i) If ϕ ∈ PCN, then [[F0(ϕ)]]M,g =
p(f) = 1 iff for all x ∈ E, if [[ϕ]]M,g(x)= 1, then f(x) = 1
(ii) If ϕ ∈ PCN, then [[F2(ϕ)]]M,g =

p(f) = 1 iff there is an x ∈ E such that [[ϕ]]M,g(x)= 1 and f(x) = 1
c. Rule I4: If δ ∈ PPR and β ∈ PIV, then [[F4(δ, β)]]M,g = [[β]]M,g([[δ]]M,g)
d. Rule I5: If δ ∈ PTV and β ∈ PPR, then [[F5(δ, β)]]M,g = [[δ]]M,g([[β]]M,g)
e. Rule I11: If ϕ, ψ ∈ PS , then [[F8(ϕ,ψ)]]M,g = 1 iff [[ϕ]]M,g = [[ψ]]M,g = 1, and

[[F9(ϕ,ψ)]]M,g = 1 iff [[ϕ]]M,g = 1 or [[ψ]]M,g = 1
f. Rule I14:
If α ∈ PT and ϕ ∈ PS, then [[F10,n(α,ϕ)]]M,g =
[[α]]M,g(the function p with domain E such that for all x ∈ E, p(x) = [[ϕ]]M,g(n/x))
We can now use the definitions in (53)-(55) to directly interpret expressions of ME+Q!!
18
Fall 2013 Ling 720
(57) Illustration: Directly Interpreting a Sentence of ME+Q
Let M be the model defined in (54) and g be any variable assignment based on M.
(i) [[ every man smokes ]]M,g = (by definition of ME+Q)
(ii) [[ F10,1 ( F0(man), F4( he 1, smokes) ) ]]M,g = (by I14)
(iii) [[F0(man)]]M,g(the function p with domain E such that for all x ∈ E,

p(x) = [[F4( he 1, smokes)]]M,g(1/x)) = (by I4)
(iv) [[F0(man)]]M,g(the function p with domain E such that for all x ∈ E,

p(x) = [[smokes)]]M,g(n1x)([[he 1]]M,g(1/x))) = (by I1)
(v) [[F0(man)]]M,g(the function p with domain E such that for all x ∈ E,

p(x) = I(smokes)(g(1/x)(1)) = (by (54))
(vi) [[F0(man)]]M,g(the function p with domain E such that for all x ∈ E,

p(x) = h(x) = (by meta-logical reasoning)
(vii) [[F0(man)]]M,g(h) = (by I2)
(viii) (the function p with domain D<e,t>,E such that for all f ∈ D<e,t>,E
p(f) = 1 iff for all x ∈ E, if [[man]]M,g(x)= 1, then f(x) = 1)(h) = (by I1)
(ix) (the function p with domain D<e,t>,E such that for all f ∈ D<e,t>,E
p(f) = 1 iff for all x ∈ E, if i(x)= 1, then f(x) = 1)(h) = (by meta-logical reasoning)
(x) 1 iff for all x ∈ E, if i(x)= 1, then h(x) = 1 = (by (54))
(xi) 0
(58) Remark
• Note that for syntactically ambiguous strings, there can be two possible
interpretations w.r.t a model and variable assignment.
• Thus, as the language in (56) suggests, [[.]]M,g is not a function.
(59) What We’ve Done So Far

• Reviewed the UG architecture, and shown how it can provide a compositional
semantics for a fragment of English with quantificational ‘terms’ (NPs)
• Reviewed the PTQ architectures, and shown how it also can provide a compositional
semantics for a fragment of English with quantificational ‘terms’ (NPs)
• Examined how the systems in UG and PTQ relate to one another…
Now, let’s actually study Montague’s full semantic analysis of English in PTQ…
19
Fall 2013 Ling 720
The Proper Treatment of Quantification in Ordinary English, Part 1:
The Fragment of English
We will now explore the analysis of English that Montague puts forth in his seminal paper, PTQ.
As we’ve already seen, there are three principle parts to the analysis:
• The syntactic fragment of English (easy)
• The logical language, Intensional Logic (moderate)
• The translation from English to IL (difficult; after all, this is the actual semantic analysis)
(1) On UG vs. PTQ

“On their common domain of applicability, the three treatments [UG, PTQ, and English
as a Formal Language] essentially agree in the truth and entailment conditions imposed
on sentences… Nevertheless, the details of the present development possess certain
aesthetic merits, of coherence and conceptual simplicity, not to be found in the treatment
of English in [UG].”
1. The Syntactic Categories of English
(1) The Categories
• In PTQ, Montague employs a system of syntactic categories (i.e., category labels) for
English that are similar in structure to the types (i.e., a so-called categorical grammar)
• This allows for an elegant statement of the category-to-type correspondence between

English and IL.
Cat is the smallest set such that:

a. e, t ∈ Cat
b. If A, B ∈ Cat, then A/B, A//B ∈ Cat
(2) On Reading the Category Labels

From the point of view of our type notation, the category notation is ‘flipped backwards’
A/B and A//B = Expressions that when ‘combined’ with an expression of category
B yield an expression of category A.
Illustration:
t/e, t//e: Combines with an expression of category e to yield one of category t
(≈ <e,t>)
1
Fall 2013 Ling 720
(3) a. Question: What is the difference between A/B and A//B?
b. Answer: Nothing substantial.

As we’ll see in a second, it allows Montague to distinguish between CNs and IVs.
c. Quote:
“We shall regard the categories A/B and A//B as playing the same semantical but
different syntactical roles.”
(i.e., CN and IV have translations/meanings of the same type, but are different
syntactic categories in English.)
(4) Abbreviations for the Syntactic Categories
• Up until now, we’ve been treating the labels below as the actual category labels of
our language.
• In PTQ, though, Montague introduces them as meta-linguistic abbreviations for the

more complex category labels defined in (1).
a. Some New Definitions for Old Friends

(i) IV abbreviates t/e
(ii) T abbreviates t/IV ( = t/(t/e))
(iii) CN abbreviates t//e
(iv) TV abbreviates IV/T ( = (t/e)/(t/(t/e)) )
Note: TVs are expressions that combine with Terms (t/(t/e)) to yield IVs (t/e).
Thus, TVs will directly combine syntactically with quantificational terms…
b. Some New Friends
Category Informal Name Abbreviates Example

(i) IAV ‘IV-modifying adverb’ IV/IV slowly, allegedly
(ii) t/t ‘Sentence-modifying adverb’ necessarily
(iii) IAV/T ‘IAV-making preposition’ in, about
(iv) IV/t ‘Sentence-taking verb phrase’ believe that
(v) IV//IV ‘IV-taking verb phrase’ try to, wish to
Notes:
• Montague uses the term ‘verb phrase’ in (iv) and (v) a bit differently from syntacticians
• Just like TVs, prepositions in PTQ directly combine with quantificational terms.
• The only uses of the A/B and A//B distinction are:
o IV vs. CN (run vs. man)
o IAV vs. IV//IV (slowly vs. try to)
2
Fall 2013 Ling 720
(5) Remark
In PTQ, there are an infinite number of category labels for English, some of which don’t
seem to apply to any actual expression of English (e.g. e/t)
• Again, what this buys us is mainly just an elegant statement of the type-category
correspondence in the translation system…
2. The Basic Expressions of English
The following are the basic expressions of the English fragment, exactly as written out by
Montague in PTQ.
(6) Basic Expressions of the English Fragment
a. BIV = {run, walk, talk, rise, change}
b. BT = {John, Mary, Bill, ninety, he0, he1, he2, …}
c. BTV = {find, lose, eat, love, date, be, seek, conceive}
d. BIAV = {rapidly, slowly, voluntarily, allegedly}
e. BCN = {man, woman, park, fish, pen, unicorn, price, temperature}
f. Bt/t = {necessarily}
g. BIAV/T = {in, about}
h. BIV/t = {believe that, assert that}
i. BIV//IV = {try to, wish to}
j. BA = ∅ if A is any category other than those mentioned above. (In

particular, the sets Be of basic entity expressions and Bt of basic
declarative sentences are empty.)1
(7) Remark
In the PTQ system, pronouns are all terms. They are also all masculine. Finally, note that
the pronouns and indices are not syntactically separate expressions (unlike in UG).
1
Just to forestall any confusion, in the PTQ paper itself, Montague uses ‘Λ’ to denote the null set.
3
Fall 2013 Ling 720
(8) Key Observation, Previewing Some Fun to Come…
• Note that the following pairs of expressions are all members of the same category:
a. eat, seek [both TVs]

b. rapidly, allegedly [both IAVs]
c. in, about [both IAV/Ts]
• In each of these pairs, the second expression creates an ‘opaque’ environment,

whereas the fist one does not (i.e., the first creates a ‘transparent’ environment)
a. (i) John ate a unicorn. (entails ‘there is a unicorn’)

(ii) John seeks a unicorn. (doesn’t entail ‘there is a unicorn’)
b. (i) John rapidly danced. (entails ‘John danced’)

(ii) John allegedly danced. (doesn’t entail ‘John danced’)
c. (i) John talked in a house. (entails ‘there is a house’)

(ii) John talked about a unicorn. (doesn’t entail ‘there is a unicorn’)
• Given that seek, allegedly, and about create ‘opaque’ contexts, we’d ideally want
them to take intensions as arguments (LING 620).
o Seek takes as argument the intension of a unicorn
o Allegedly takes as argument the intension of danced.
o About takes as argument the intension of a unicorn.
• Recall, however, Montague’s requirement that expressions of the same category

map to translations/meanings of the same type.
• Consequently, since seek, allegedly, about have a meaning that takes intensions as
arguments, we’ll also need for eat, rapidly, and in to have such meanings…
3. The Syntactic Rules
Given the basic expressions in (6), the syntactic rules outlined in this section will simultaneously
define the set of meaningful expressions of English, ∪A∈Cat PA
(9) Rule S1 BA ⊆ PA, for every category A.
(10) Rule S2 (Forming Quantificational Terms)
If ζ ∈ PCN, then F0(ζ), F1(ζ), F2(ζ) ∈ PT,

where F0(ζ) = every ζ
F1(ζ) = the ζ
F2(ζ) is a ζ or an ζ according as the first word in ζ takes a or an
4
Fall 2013 Ling 720
(11) Remarks on Rule S2
a. Via the addition of F1, we are adding definite determiners to our fragment.
Montague will provide a ‘Russellian’ analysis of definite terms: 2
The man smokes is true iff ∃x0 ∀x1 (((man’ x1) ↔ x0 = x1) & (smokes’ x0))
There is exactly one man, and he smokes.
b. The definition of F2 in (10) appeals to a notion that Montague does not ever
explicitly define in the paper: whether a given word takes a or an
• In this case, though, we could easily revise (10) to explicitly contain the
generalization ‘an before a vowel’.
(12) Rule S3 (Relativization)
If ζ ∈ PCN and ϕ ∈ Pt , then F3,n(ζ,ϕ) ∈ PCN, where F3,n(ζ,ϕ) = ζ such that ϕ’, and ϕ’
comes from ϕ by replacing each occurrence of hen or himn by {he, she, it} or {him, her,
it}, respectively, according as the first BCN in ζ is of {masc., fem., neuter} gender.
Illustration:
(i) woman ∈ PCN, John loves him5 ∈ Pt (Rules)
(ii) F3,5 (woman , John loves him5) ∈ PCN (Rule S3)
(iii) woman such that John loves her ∈ PCN (def. of F3,5)
a. Rule S3 is the rule for forming relative clauses in PTQ. Note that it only forms
‘such that’ relatives; there are no mechanisms in PTQ for filler-gap dependencies.
b. Like F10,n from the last handout, F3,n is an infinite family of operations.
c. Also like F10,n , the translation operation corresponding to F3,n will lambda
abstract over the variable with index n.
Rough, Simplified Illustration:

F3,5 (woman , John loves him5) translates to: (λx5 ((woman’ x5) & ((loves x5) john’)))
woman such that John loves her translates to: (λx5 ((woman’ x5) & ((loves x5) john’)))
2
Note that, since Intensional Logic is being used as the translation language, it would not be possible in the PTQ
system to provide a Fregean/presuppositional analysis of definite terms.
5
Fall 2013 Ling 720
(14) Rule S4 (Subject-Predicate Rule)
If α ∈ Pt/IV and δ ∈ PIV, then F4(α,δ) ∈ Pt, where F4(α,δ) = α δ’ and δ’ is the result of
replacing the first verb (i.e., member of BIV, BTV, BIV/t, or BIV//IV) in δ by its third person
singular present.
a. In PTQ, the operation F4 doesn’t just combine a subject with a predicate, it is also
responsible for adding the tense and agreement morphology to the verb.
b. Due to its additional morpho-syntactic role, the definition of F4 appeals to two

important, fundamental concepts: (a) the notion of a ‘verb’, and (b) the notion of a
verb’s ‘third person singular present’.
c. The notion ‘verb’ is defined as part of the definition of F4. Notice, though, that
‘verb’ is not actually a syntactic category in the system.
• In fact, verb can’t be a syntactic category in MG, given the need for category-
to-type correspondence.
d. The notion ‘third person singular present’ is nowhere defined in the paper.
To properly implement this, though, we could imagine defining a function
‘3sgPRES’ which maps an English verb root to its 3rd singular present form:
3sgPRES(root) = does, if root = do

is, if root = be
has, if root = have
…
root+s otherwise
• With this function, we’d simly say that δ’ is obtained from δ by replacing the
first verb v in δ with 3sgPRES(v).
(15) Rule S5 (Direct Object Rule)
If δ ∈ PIV/T and β ∈ PT, then F5(δ,β) ∈ PIV, where F5(δ,β) = δ β if β does not have the
form hen, and F5(δ,hen) = δ himn
Remark:
Again, the definition of F5 captures the behavior of objective case on English pronouns.
6
Fall 2013 Ling 720
(16) Rule S6 (Prepositional Phrase Rule) If δ ∈ PIAV/T and β ∈ PT, then F5(δ,β) ∈ PIAV
Illustration: (i) in ∈ PIAV/T, he2 ∈ PT (Rule S1)

(ii) F5 (in, he2) ∈ PIAV (Rule S6)
(iii) in him2 ∈ PIAV (def. of F5)
Remark: S6 forms prepositional phrases in PTQ. Note its use of the operation F5
o This captures the presence of ACC on pronominal complements of Ps
o This also nicely illustrates the difference between syntactic
operations and syntactic rules in PTQ
(17) Rule S7 (Finite Complement Clause Rule)

If δ ∈ PIV/t and β ∈ Pt, then F6(δ,β) ∈ PIV, where F6(δ,β) = δ β
Illustration:
(i) believe that ∈ PIV/t , John runs ∈ Pt (Rule S1, S4)
(ii) F6(believe that, John runs) ∈ PIV (Rule S7)
(iii) believe that John runs ∈ PIV (def. of F6)
Question: Did Montague really need to introduce a new operation F6 here?

Couldn’t he have simply continued to use F5?
(After all, nothing in Pt will be of the form hen)
(18) Rule S8 (Infinitival Control Rule) If δ ∈ PIV//IV and β ∈ PIV, then F6(δ,β) ∈ PIV
Illustration:
(i) try to ∈ PIV//IV , find a unicorn ∈ PIV (Rule S1, S5)
(ii) F6(try to, find a unciorn) ∈ PIV (Rule S8)
(iii) try to find a unicorn ∈ PIV (def. of F6)
(19) Rule S9 (Sentential Adverbs Rule) If δ ∈ Pt/t and β ∈ Pt, then F6(δ,β) ∈ Pt
Illustration:
(i) necessarily ∈ Pt/t , John runs ∈ Pt (Rule S1, S4)
(ii) F6(necessarily, John runs) ∈ Pt (Rule S9)
(iii) necessarily John runs ∈ Pt (def. of F6)
(20) Rule S10 (Adverbs Rule)

If δ ∈ PIV/IV and β ∈ PIV, then F7(δ,β) ∈ PIV, where F7(δ,β) = β δ
Illustration:
(i) about a unicorn ∈ PIV/IV, talk ∈ PIV (Rule S1, S6)
(ii) F7(about a unicorn, talk) ∈ PIV (Rule S10)
(iii) talk about a unicorn ∈ PIV (def. of F7)
7
Fall 2013 Ling 720
(21) Rule S11 (Sentential Conjunction/Disjunction Rule)
If ϕ, ψ ∈ Pt, then F8(ϕ, ψ), F9(ϕ, ψ) ∈ Pt , where F8(ϕ,ψ)= ϕ and ψ and F9(ϕ,ψ)= ϕ or ψ.
(22) Rule S12 (IV Conjunction/Disjunction Rule)

If ϕ, ψ ∈ PIV, then F8(ϕ, ψ), F9(ϕ, ψ) ∈ PIV
Illustration:
(i) runs, loves Mary ∈ PIV (Rule S1, S5)
(ii) F8(runs, loves Mary), F9(runs, loves Mary) ∈ PIV (Rule S12)
(iii) runs and loves Mary, runs or loves Mary ∈ PIV (def. of F8 and F9)
(23) Remark
• Montague uses the same syntactic operations, F8 and F9, to do conjunction/disjuction

of sentences and conjunction/disjunction of IVs.
• As we’ll see later, this will cause some difficulty converting the PTQ system into the
algebraic UG format
o In the putative translation base, we’d need a single polynomial operation over IL
to correspond to F8 (F9)
o But a single such operation won’t give us the right translations for both sentential
conjunction (disjunction) and IV conjunction (disjunction)
(24) Rule S13 (Term Disjunction) If α, β ∈ PT , then F9(α, β) ∈ PT
Illustration:
(i) every man, John ∈ PT (Rule S1, S2)
(ii) F9(every man, John) ∈ PT (Rule S13)
(iii) every man or John ∈ PT (def. of F9)
(25) a. Question: Why didn’t Montague also include a rule for conjoining terms?
b. Educated Guess:
He didn’t want to have to deal with the collective reading of sentences like John
and Mary ate a unicorn.
And, at last we come to one of the centerpieces of the paper… the rules for ‘Quantifying In’…
8
Fall 2013 Ling 720
(26) Rule S14 (Quantifying In to Sentences)
If α ∈ PT and ϕ ∈ Pt , then F10,n(α,ϕ) ∈ Pt , where either:
(i) α does not have the form hek and F10,n(α,ϕ) comes from ϕ by replacing the first
occurrence of hen or himn by α, and all other occurrences of hen or himn by {he,
she, it} or {him, her, it} respectively, according as the gender of the first BCN or
BT in α is {masculine, feminine, neuter}, or
(ii) α = hek and F10,n(α,ϕ) comes from ϕ by replacing all occurrences of hen or himn
by hek or himk respectively.
(27) Remark 1
The definition of F10,n(α,ϕ) is disjunctive; its value depends upon whether the ‘term
argument’ α is a pronoun or not.
a. If α is not a pronoun, then we do the following:

(i) Replace the first instance of hen or himn with α, and
(ii) All subsequent instances of hen or himn with a pronoun agreeing in gender
with α
Illustration:
(i) a woman ∈ PT, he2 runs and John likes him2 ∈ Pt (Rules)
(ii) F10,2(a woman, he2 runs and John likes him2) ∈ Pt (Rule S14)
(iii) a woman runs and John likes her ∈ Pt (def. of F10,n)
b. If α is a pronoun, then we do the following:

Replace all instances of hen or himn with α or its accusative variant, respectively
Illustration:
(i) he3 ∈ PT, he2 runs and John likes him2 ∈ Pt (Rules)
(ii) F10,2(he3, he2 runs and John likes him2) ∈ Pt (Rule S14)
(iii) he3 runs and John likes him3 ∈ Pt (def. of F10,n)
(28) a. Question: Why did Montague use this disjunctive definition of F10,n?
After all, if we simply extended the first condition in (26ai) to pronouns, we’d end
up getting almost the same strings:
Illustration: (ii) F10,2(he3, he2 runs and John likes him2) ∈ Pt

(iii) he3 runs and John likes him ∈ Pt
b. Answer:
For whatever reason, condition (26ai) doesn’t copy the index onto the subsequent
pronouns. Consequently, if we quantified-in a woman on (28aiii), we wouldn’t
get gender agreement on the following pronouns.
9
Fall 2013 Ling 720
(29) Remark 2
The definition of F10,n(α,ϕ) again appeals to a notion that Montague doesn’t ever define
in the paper PTQ itself: the gender of a CN or T.
a. A Possible Implementation:
We could imagine defining a function GEN whose domain is BCN and BT and
whose range is {MASC, FEM, NEUT}.
GEN(John) = MASC
GEN(Mary) = FEM
GEN(ninety) = NEUT
…
GEN(man) = MASC
GEN(woman) = FEM
GEN(fish) = NEUT
…
We could then reformulate (26) so that it makes reference to this GEN function.
(30) Rule S15 (Quantifying In to NPs) If α ∈ PT and ϕ ∈ PCN , then F10,n(α,ϕ) ∈ PCN
Illustration:
(i) every man ∈ PT, woman such that he2 likes her ∈ PCN (Rules)
(ii) F10,2(every man, woman such that he2 likes her) ∈ PCN (Rule S15)
(iii) woman such that every man likes her ∈ PCN (def. of F10,n)
Remark: It still isn’t clear to me what use this rule is.

Neither Montague himself nor Dowty et al. (1981) discuss it in detail.
(31) Rule S16 (Quantifying In to VPs) If α ∈ PT and ϕ ∈ PIV , then F10,n(α,ϕ) ∈ PIV
Illustration:
(i) a unicorn ∈ PT, find him2 and eat him2 ∈ PIV (Rules)
(ii) F10,2(a unicorn, find him2 and eat him2) ∈ PIV (Rule S16)
(iii) find a unicorn and eat it ∈ PIV (def. of F10,n)
Remark: As we’ll see, this allows us to capture the opaque/bound reading of “John
wants to find a unicorn and eat it.”
Finally, to wrap things up, Montague introduces rules for adding negation and tense
morphology…
10
Fall 2013 Ling 720
(32) Rule S17 (Rules for Tense and Negation)
If α ∈ PT and δ ∈ PIV, then F11(α,δ), F12(α,δ), F13(α,δ), F14(α,δ), F15(α,δ) ∈ Pt, where:
(i) F11(α,δ) = α δ’ and δ’ is the result of replacing the first verb in δ by its negative
third person singular present.
Illustration: (i) John ∈ PT, run ∈ PIV (Rule S1)

(ii) F11(John, run) ∈ Pt (Rule S17)
(iii) John doesn’t run ∈ Pt (def. of F11)
(ii) F12(α,δ) = α δ’ and δ’ is the result of replacing the first verb in δ by its third
person singular future.

(iii) John will run ∈ Pt (def. of F12)
(iii) F13(α,δ) = α δ’ and δ’ is the result of replacing the first verb in δ by its negative
third person singular future.

(iii) John won’t run ∈ Pt ( def. of F13)
(iv) F14(α,δ) = α δ’ and δ’ is the result of replacing the first verb in δ by its third
person singular present perfect.

(iii) John has run ∈ Pt (def. of F14)
(v) F15(α,δ) = α δ’ and δ’ is the result of replacing the first verb in δ by its negative
third person singular present perfect.

(iii) John hasn’t run ∈ Pt (def. of F15)
11
Fall 2013 Ling 720
(33) Remarks
a. Again, Montague doesn’t ever properly define the terms negative…present,

future, negative…future, present perfect, or negative… present perfect.
b. But, again, we could imagine defining an array of morpho-syntactic functions

mapping every basic ‘verb’ to these values.
(i) Neg3sgPRES(root) = isn’t, if root = be

…
doesn’t root otherwise
(ii) 3sgFUT(root) = will root

(iii) Neg3sgFUT(root) = won’t root
(iv) 3sgPERF(root) = has root+ed 3
(v) NEG3sgPERF(root) = hasn’t root+ed
c. Montague doesn’t ‘decompose’ these morphological forms in the way we’ve

come to expect from transformational morpho-syntactic analyses of English.
o In PTQ, negative 3rd singular future and 3rd singular future are just two
different primitives; the ‘negative’ and ‘future’ isn’t separately factored out
d. Also, these features are introduced by the same rule that adds the subject with the
predicate.
o Thus, doesn’t run and won’t run aren’t themselves meaningful expressions.
o Thus, we can’t get conjunctions like John doesn’t run and won’t run.
(34) Snarky Side Remark

Despite how little Montague regards the syntactic research ‘emanating from MIT’, he
could have paid a bit closer attention to Syntactic Structures ; )
Those are all the rules! We can now use them to properly define the set of meaningful
expressions for the English fragment.
(35) The Phrases of the English Fragment

{ PA }A∈Cat is the smallest family of sets indexed by Cat such that S1-S17 are true.
(36) The Meaningful Expressions of the English Fragment

ϕ is a meaningful expression of the fragment if there is an A ∈ Cat such that ϕ ∈ PA.
3
Actually, what we really need here is a separate function PST-PARTICIPLE, mapping every verbal root to its past
participle.
12
Fall 2013 Ling 720
4. Some Illustrations of the Fragment
As the calculations below illustrate, the following are members of Pt
(37) every man loves a woman such that she loves him
(i) man, woman ∈ BCN, love ∈ BTV, he0, he1 ∈ BT (by (6))
(ii) man, woman ∈ PCN, love ∈ PTV, he0, he1 ∈ PT (by S1)
(iii) love him0 ∈ PIV (by S5)
(iv) he1 loves him0 ∈ Pt (by S4)
(v) woman such that she loves him0 ∈ PCN (by S3)
(vi) a woman such that she loves him0 ∈ PT (by S2)
(vii) love a woman such that she loves him0 ∈ PIV (by S5)
(viii) he0 loves a woman such that she loves him0 ∈ Pt (by S4)
(ix) every man ∈ PT (by S2)
(x) every man loves a woman such that she loves him ∈ PT (by S14)
Analysis Tree, Illustrating the Derivation
every man loves a woman such that she loves him, 10,0
every man, 0 he0 loves a woman such that she loves him0, 4
man he0 love a woman such that she loves him0, 5
love a woman such that she loves him0, 2
woman such that she loves him0, 3,1
woman he1 loves him0, 4
he1 love him0, 5
love he0
13
Fall 2013 Ling 720
(38) John seeks a unicorn.
a. Derivation / Analysis One

(i) unicorn ∈ BCN, seek ∈ BTV, John ∈ BT (by (6))
(ii) unicorn ∈ PCN, seek ∈ PTV, John ∈ PT (by S1)
(iii) a unicorn ∈ PT (by S2)
(iv) seek a unicorn ∈ PIV (by S5)
(v) John seeks a unicorn ∈ Pt (by S4)
John seeks a unicorn, 4
John seek a unicorn, 5
seek a unicorn, 2
unicorn
b. Derivation / Analysis Two

(i) unicorn ∈ BCN, seek ∈ BTV, John, he0 ∈ BT (by (6))
(ii) unicorn ∈ PCN, seek ∈ PTV, John, he0 ∈ PT (by S1)
(iii) a unicorn ∈ PT (by S2)
(iv) seek him0 ∈ PIV (by S5)
(v) John seeks him0 ∈ Pt (by S4)
(vi) John seeks a unicorn ∈ Pt (by S14)
John seeks a unicorn, 10,0
a unicorn, 2 John seeks him0, 4
unicorn John seek him0, 5
seek he0
(39) Remarks
o Again, we see that our English fragment admits of syntactically (and semantically)
ambiguous expressions.
o Again, this will present no problems for PTQ, where the translation relation needn’t
be a function.
o In addition, we’ll see that under the derivation in (38a), sentence (38) receives
the de dicto reading, while under the derivation in (38b), it receives de re reading
14
Fall 2013 Ling 720
(40) Some Optional Exercises for Students
Show how our English fragment predicts that each of the following are members of Pt
a. John doesn’t love the man such that he runs

(≈ John doesn’t love the man who runs.)
b. every man will talk about Mary
c. Bill has thought that a man loves him
d. every woman runs or loves Bill
15
Fall 2013 Ling 720
Intensional Logic
In these notes, we will explore the translation language employed in PTQ, Intensional Logic
(1) Why Intensional Logic?
• Our English fragment includes such words as seek, conceive, believe that, wish to,
allegedly, necessarily, and about
o Notably, these expressions must take the intensions of their complements as
argument (rather than the extensions of their complements)
• Consequently, we will want the induced semantics for the English fragment to end up
mapping some English expressions to their intensions.
• Thus, since we’re developing an indirect interpretation of the fragment, we’ll need a
logical language that can represent such intensions.
• In addition, this language will have the ability to represent the semantics of the modal
elements necessarily, present perfect, and future.
(2) Some Conceptual Background to Intensional Logic
• In LING 620, you learned that in a sentence like (2a) below, the extension of thinks
takes as argument the intension of Mary smokes.
a. John thinks [ that Mary smokes ]
• You may also have considered an analysis where the finite complentizer that has the
meaning in (2b), which entails that that Mary smokes will have the extension in (2c)
b. [[ that XP ]]w = [λw’ : [[XP]]w’ ]
c. (i) [[ that Mary smokes ]]w =

w’
[λw’ : [[Mary smokes]] ] =
[λw’ : Mary smokes in w’ ]
• Thus, the extension of that S is equal to the intension of S.
• It probably wasn’t noted at the time, but observe that that S in such a semantics also
itself has an intension:
d. [λw : [[ that Mary smokes ]]w ] = [λw : [λw’ : Mary smokes in w’]]
• Thus, the intension of that S is a constant function mapping every world to the
intension of S.
1
Fall 2013 Ling 720
(3) The Upshot of All This

The language IL will have an operator (^) that behaves just like that in our analysis above
1. The Syntax of Intensional Logic (IL)
To define the meaningful expressions of IL, we begin by defining its vocabulary. And, to define
its vocabulary, we must first define our system of (intensional) types.
(4) The System of (Intensional) Types The set Type is the smallest set such that:
a. e, t ∈ Type
b. If σ, τ ∈ Type, then <σ, τ> ∈ Type
c. If τ ∈ Type, then <s, τ> ∈ Type
(5) The Vocabulary of IL
a. The Logical Constants

(i) Sentence Connectives: ¬ Negation ‘it is not the case that’
∧ Conjunction ‘and’
∨ Disjunction ‘or’ (inclusive)
 Implication ‘if…then’
↔ Biconditional ‘if and only if’
(ii) Identity = Identity ‘equals’
(iii) Quantifiers V Existential ‘there is an…’

Λ Universal ‘for all…’
(iv) Lambda λ Lambda ‘the function that…’
(v) Modal Operators  Necessity ‘it is necessary that...’

The W Future ‘it will be the case that...’
important H Past ‘it has been the case that...’
new
additions! (vi) Intension/Extension Ops ^ Up ‘the intension of…’
∨
Down ‘the extension of…’
b. The Syntactic Symbols: (, ), [, ]
c. The Non-Logical Constants For every type τ ∈ Type

(i) An infinite set of constants of type τ, Conτ
(ii) A countably infinite set of variables of type τ, Varτ
Varτ = {vn, τ : n ∈ ℕ}
2
Fall 2013 Ling 720
(6) Remarks
• The variable vn, τ will sometimes be referred to as ‘the nth variable of type τ’
• For some reason, Montague doesn’t enumerate the constants (probably because he
doesn’t have to, unlike the variables).
• Montague also never specifies how the constants of IL should look.
o However, he uses j, m, b, n as meta-language labels for constants of type e
o Similarly, he uses, run’, man’, love’ as meta-language labels for constants of

predicative types
• We will follow suit, and temporarily assume the following:

o j, m, b, n ∈ Cone
o smoke’, run’, man’ ∈ Con<e,t>
o like’ ∈ Con<e,<e,t>>
o think’ ∈ Con<<s,t>,<e,t>>
(7) The Syntax of Intensional Logic
a. Every variable and constant of type τ is in MEτ

b. If ϕ ∈ MEτ and u is a variable of type σ, then λuϕ ∈ ME<σ,τ>
c. If ϕ ∈ ME<σ,τ> and ψ ∈ MEσ, then ϕ(ψ) ∈ MEτ
d. If ϕ,ψ ∈ MEτ , then ϕ = ψ ∈ MEt
e. If ϕ, ψ ∈ MEt and u is a variable, then (i) ¬ϕ ∈ MEt
(ii) [ϕ ∧ ψ] ∈ MEt
(iii) [ϕ ∨ ψ] ∈ MEt
(iv) [ϕ  ψ] ∈ MEt
(v) [ϕ ↔ ψ] ∈ MEt
(vi) Vuϕ ∈ MEt
(vii) Λuϕ ∈ MEt
(viii) ϕ ∈ MEt
(ix) Wϕ ∈ MEt
(x) Hϕ ∈ MEt
f. If ϕ ∈ MEτ, then [^ϕ] ∈ ME<s,τ>
∨
g. If ϕ ∈ ME<s,τ>, then [ ϕ] ∈ MEτ
(8) Meaningful Expression of IL ϕ is a meaningful expression of IL if ϕ ∈ ∪τ∈Type MEτ
3
Fall 2013 Ling 720
(9) Remarks
a. Note that in (7c) the notation for a predicate applied to its argument is ϕ(ψ), and
not (ϕ ψ), as in TL.
b. Note that ‘=’ can appear between any two expressions of the same type. Thus, all
the following are meaningful expressions:
(i) j=b
(ii) man’ = run’
(iii) man’(j) = run’(b)
Our semantics for ‘=’ will entail that man’(j) = run’(b) is logically equivalent to
[man’(j) ↔ run’(b)]
c. “The expression [^α] is regarded as denoting (or as having as its extension) the
intension of the expression α” (pp. 23-24)
o Thus, the operator ‘^’ is akin to English that in (2b)
∨
d. “The expression [ α] is meaningful only if α is an expression that denotes an
∨
intension or sense; in such a case [ α] denotes the corresponding extension.”
e. “In the presentation of actual expressions, of intensional logic, square brackets

will sometimes for perspicuity be omitted, and sometimes gratuitously inserted.”
(10) Some Illustrative Meaningful Expressions of IL
a. λv1,e [ smoke’(v1,e) ∧ run’(v1,e) ]

Informally: The characteristic function of the set of smokers that run.
b. λv1,<e,t> v1,<e,t>(j)
Informally: The characteristic function of the set of properties true of John.
(≈ (λP1 (P1 john’)) )
c. λv1,<e,t> Vv1,e [ man’(v1,e) ∧ v1,<e,t> (v1,e) ]

Informally: The characteristic function of the set of properties true of a man
(≈ (λP1 ∃x1 ((man’ x1) & (P1 x1))) )
d. think’([^like’(j)(m)])(b)
Informally: ‘Bill thinks that Mary likes John’
e. run’(j)
Informally: ‘It is necessary that John runs’
f. Λv1,e H smoke’(v1,e)
Informally: ‘For all x, it has been the case that x smokes.’
4
Fall 2013 Ling 720
2. The Semantics of Intensional Logic (IL)
In this section, we will develop a formal model-theoretic semantics for IL. As usual, we begin by
defining the notion ‘denotations of type τ’
• As the following definition suggests, our model structures will based on a set of
entities A, a set of worlds I, and a set of times J.
(11) The Denotations

Let A, I, and J be non-empty sets (A = set of entities; I = set of possible worlds; J = set of
moments of time). If τ ∈ Type, then the set Dτ, A, I, J of possible denotations of type τ
corresponding to A, I, J is defined as follows:
a. De,A,I J = A
b. Dt,A,I,J = { 0, 1 }
c. If σ, τ ∈ T then D<σ,τ>,A,I,J = the set of functions from Dσ ,A,I,J to Dτ,A,I,J
d. If σ ∈ T then D<s,σ>,A,I,J = the set of functions from I×J to Dσ ,A,I,J
×
= (Dσ ,A,I, J)I J
Notes:
• In PTQ, the set J is the set of times (moments of time); it’s not a set of
contexts/variable assignments (unlike in UG).
• In PTQ, a denotation of type <s,σ> is a function from world-time pairs to denotations
of type σ
In addition to the denotations in (11), our model theoretic semantics for IL will make reference
to a set of ‘senses’:
(12) The Senses

Let A, I, and J be non-empty sets (A = set of entities; I = set of possible worlds; J = set of
moments of time). If τ ∈ Type, then the set Sτ, A, I, J of possible senses of type τ
corresponding to A, I, J is equal to D<s, τ,>,A,I,J
With these ingredients, we can now define what a model of IL is…
(13) A Model of IL
An intensional model (or interpretation) of IL is a quintuple <A, I, J, ≤, F> such that:
a. A, I, J are non-empty sets

b. ≤ is a linear ordering of J
c. F is a function whose domain is the set of constants of IL
d. Whenever τ ∈ Type and α ∈ Conτ , F(α) ∈ Sτ, A, I, J
5
Fall 2013 Ling 720
(14) Remarks
Thus, an (intensional) model of IL is defined by: (i) a set of entities A, (ii) a set of worlds
I, (iii) a set of times J, (iv) a temporal ordering ≤, and (v) a ‘lexical’ interpretation
function F.
• Note that the function F maps the constants of IL to intensions, not denotations.
o Thus, type e constants are mapped to <s,e> functions (individual concepts)
o Type <et> constants are mapped to <s,<e,t>> functions (properties)
o Type <e,<e,t>> constants are mapped to <s,<e,<e,t>>> functions (‘relations in
intension’), etc….
• Models of IL are too complex to specify concretely. Nevertheless, to illustrate key

components of the definitions here, we can attempt the following partial description.
(15) Illustration of an Intensional Model (for IL)

Let M be an intensional model <A, I, J, ≤, F> such that:
a. A = {Barack, Michele}
b. I = { w1 , w2 }
c. J = { t1 , t2 , t3 }
d. ≤ = { <t1 , t2> , <t1 , t3> , <t2 , t3>, <t1 , t1> , <t2 , t2> , <t3 , t3> }
e. F comprises at least the following mappings:
F(b) = w1 , t1  Barack ; w2, t1  Barack

w1 , t2  Barack ; w2 , t2  Barack
w1 , t3  Barack ; w3 , t3  Barack
F(m) = w1 , t1  Michelle ; w2, t1  Michelle

w1 , t2  Michelle ; w2 , t2  Michelle
w1 , t3  Michelle ; w3 , t3  Michelle
F(smoke’) = w1 , t1  {<Barack,1>, <Michelle,1>}

w1 , t2  {<Barack,1>, <Michelle,0>}
F(think’) contains the following mapping:
w1 , t1  < < w1 , t1 >, 1 >

< < w1 , t2 >, 1 >
< < w1 , t3 >, 0 >  { <Barack,1>, <Michelle,1> }
< < w2 , t1 >, 0 >
< < w2 , t2 >, 0 >
< < w2 , t3 >, 0 >
6
Fall 2013 Ling 720
(16) Remarks Regarding the Model in (15)
• The type-e constant b (m) is interpreted as an <s,e> function (individual concept) that
maps every world-time pair to Barack (Michelle).
• The type-<e,t> constant smoke’ is interpreted as an <s,<e,t>> function (property).

Informally speaking, in this model:
o At world w1, Barack and Michelle both smoke at time t1, but then at time t2,
Michelle quits. Finally, at time t3, Barack also quits.
o At world w2, Barack is never a smoker, but Michelle is and never quits.
• The type-<<s,t>,<e,t>> constant think’ is interpreted as an <s,<<s,t>,<e,t>>>

function (the intension of thinks). Informally speaking, in this model:
o At world w1 and time t1, both Barack and Michelle think that Barack smokes.
Now that we have the definition of a model, the next step is to define the notion of a ‘variable
assignment’…
(17) Definition of an M -Assignment

Let M be an intensional model <A, I, J, ≤, F>. The function g is an M-assignment if:
a. It has as its domain the set of all variables of IL.
b. If u is a variable of type τ, then g(u) ∈ Dτ, A, I, J
Note: Variable assignments maps variables directly to denotations (extensions), not to

senses (intensions).
With these ingredients, we can now recursively define the central notion ‘extension with respect
to an intensional model, world, time, and variable assignment’
(18) Extension With Respect to Model, World, Time, and Variable Assignment
Let M be an intensional model <A, I, J, ≤, F> for IL, let i ∈ I and j ∈ J, and let g be an
M-assignment. If α is a meaningful expression of IL, then [[α]]M,i,j,g, the extension of α
with respect to M, i, j, and g, is defined as follows:
a. If α is a constant, then [[α]]M,i,j,g = F(α)(<i,j>)
Note: Thus, we obtain the extension of a constant α by applying the intension

F(α) to the world-time pair <i,j>.
b. If α is a variable, then [[α]]M,i,j,g = g(α)
7
Fall 2013 Ling 720
c. If α ∈ MEτ and u is a variable of type σ, then [[λuα]]M,i,j,g = the function h with

domain Dσ, A,I,J such that whenever x is in that domain, h(x) = [[α]]M,i,j,g(u/x)
d. If α ∈ ME<σ,τ> and β ∈ MEσ, then [[α(β)]]M,i,j,g = [[α]]M,i,j,g([[β]]M,i,j,g)
e. If ϕ,ψ ∈ MEτ , then [[ϕ = ψ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = [[ψ]]M,i,j,g
f. If ϕ,ψ ∈ MEt , then

(i) [[¬ϕ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = 0
(ii) [[ [ϕ ∧ ψ] ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = 1 and [[ψ]]M,i,j,g = 1
(iii) [[ [ϕ ∨ ψ] ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = 1 or [[ψ]]M,i,j,g = 1
(iv) [[ [ϕ  ψ] ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = 0 or [[ψ]]M,i,j,g = 1
(v) [[ [ϕ ↔ ψ] ]]M,i,j,g = 1 iff [[ϕ]]M,i,j,g = [[ψ]]M,i,j,g
Note: Thus, if ϕ,ψ ∈ MEt , then [[ [ϕ ↔ ψ] ]]M,i,j,g = 1 iff [[ϕ = ψ]]M,i,j,g = 1

Thus, if ϕ,ψ ∈ MEt , then [ϕ ↔ ψ] and ϕ = ψ are logically equivalent.
g. If ϕ ∈ MEt , and u is a variable of type τ, then

(i) [[ Vuϕ ]]M,i,j,g = 1 iff there is an x ∈ Dτ, A,I,J such that [[ϕ]]M,i,j,g(u/x) = 1
(ii) [[ Λuϕ ]]M,i,j,g = 1 iff for all x ∈ Dτ, A,I,J , [[ϕ]]M,i,j,g(u/x) = 1
h. If ϕ ∈ MEt , then
(i) [[ ϕ ]]M,i,j,g = 1 iff for all i’ ∈ I and j’ ∈ J, [[ ϕ ]]M, i’, j’, g = 1
Note: Thus, ‘ϕ’ is more aptly translated as ‘necessarily always ϕ’
(ii) [[ Wϕ ]]M,i,j,g = 1 iff for some j’ ∈ J such that j < j’, [[ ϕ ]]M, i, j’, g = 1
(iii) [[ Hϕ ]]M,i,j,g = 1 iff for some j’ ∈ J such that j’ < j, [[ ϕ ]]M, i, j’, g = 1
i. If α ∈ MEτ , then [[ [^α] ]]M,i,j,g = the function h with domain I×J such that if
<i’,j’> ∈ I×J, then h(<i’,j’>) = [[α]]M, i’,j’, g
Note: Thus, [[ [^α] ]]M,i,j,g is the function mapping a world-time pair to the
extension of α at that world-time. Thus, given (19) below, [[ [^α] ]]M,i,j,g is
the intension of α with respect to M and g, [[α]]M,g
∨
j. If α ∈ ME<s,τ>, then [[ [ α] ]]M,i,j,g = [[α]]M,i,j,g(<i,j>)
Note: Thus, if the extension of α at i and j is some intension, then the extension
∨
of [ α] at i and j is that intension applied to i and j.
8
Fall 2013 Ling 720
(19) Intension With Respect to a Model and Variable Assignment

Let M be an intensional model <A, I, J, ≤, F> for IL and let g be an M-assignment. If α
is a meaningful expression of IL, then [[α]]M,g, the intension of α with respect to M and
g, is the function h with domain I×J such that if <i’,j’> ∈ I×J, then h(<i’,j’>) = [[α]]M, i’,j’,g
(20) Truth With Respect to a Model, World, and Time

Let M be an intensional model <A, I, J, ≤, F> for IL and let i ∈ I and j ∈ J. If ϕ ∈ MEt ,
then ϕ is true with respect to M, i, and j iff for any M-assignment g, [[ϕ]]M,i,j,g = 1
Let us now illustrate these definitions by using them to interpret meaningful expressions of IL
(21) Illustrative Computations, Part 1

In the computations below, let M be an intensional model of the kind described in (15).
Let g be an arbitrary M-assignment.
a. smoke’(b)
(i) [[smoke’(b)]]M,w1,t1,g = (by (18d))

(ii) [[smoke’]]M,w1,t1,g([[b]]M,w1,t1,g) = (by (18a))
(iii) F(smoke’)(<w1,t1>)(F(b)(<w1,t1>)) = (by (15))
(iv) {<Barack,1>, <Michelle,1>}(Barack) =
(v) 1
b. [^smoke’(b)]
(i) [[ [^smoke’(b)] ]]M,w1,t1,g = (by (18i))
(ii) the function h with domain I×J such that

if <i’,j’> ∈ I×J, then h(<i’,j’>) = [[smoke’(b)]]M,i’,j’,g = (by (18a,d))
(iii) the function h with domain I×J such that

if <i’,j’> ∈ I×J, then h(<i’,j’>) = F(smoke’)(<i’,j’>)(F(b)(<i’,j’>))
= (by (15))
(iv) the function h with domain I×J such that
if <i’,j’> ∈ I×J, then h(<i’,j’>) = F(smoke’)(<i’,j’>)(Barack) = (by (15))
(v) <<w1 , t1>,1>, <<w1 , t2>,1>, <<w1 , t3>,0>,

<<w2 , t1>,0>, <<w2 , t2>,0>, <<w2 , t3>,0>
9
Fall 2013 Ling 720
(22) Illustrative Computations, Part 2
In the computations below, let M be an intensional model of the kind described in (15).
Let g be an arbitrary M-assignment.
a. b=b
(i) [[  b = b ]]M,w1,t1,g = 1 iff (by (18h))

M, i’, j’, g
(ii) for all i’ ∈ I and j’ ∈ J, [[ b = b ]] =1 iff (by (18e))
(iii) for all i’ ∈ I and j’ ∈ J, [[b]]M,i’,j’,g = [[b]]M,i’,j’,g iff (by (18a))
(iv) for all i’ ∈ I and j’ ∈ J, F(b)(<i’,j’>) = F(b)(<i’,j’>)
(v) Thus, [[  b = b ]]M,w1,t1,g = 1
b. H smoke’(b)
(i) [[H smoke’(b)]] M,w1,t2,g = 1 iff (by (18h))

(ii) for some j’ ∈ J such that j’ < t2 , [[smoke’(b)]]M, w1, j’, g = 1
iff (by (18a,d))
(iii) for some j’ ∈ J such that j’ < t2 , F(smoke’)(<w1, j’>)(F(b)(<w1, j’>)) = 1
iff (by (15))
(iv) for some j’ ∈ J such that j’ < t2 , F(smoke’)(<w1, j’>)(Barack) = 1
(v) Given that F(smoke’)(<w1, t1>)(Barack) = 1, [[H smoke’(b)]]M,w1,t2,g = 1
c. W smoke’(b)
(i) [[W smoke’(b)]] M,w1,t2,g = 1 iff (by (18h))

(ii) for some j’ ∈ J such that t2 < j’, [[smoke’(b)]]M, w1, j’, g = 1
iff (by (18a,d))
(iii) for some j’ ∈ J such that t2 < j’, F(smoke’)(<w1, j’>)(F(b)(<w1, j’>)) = 1
iff (by (15))
(iv) for some j’ ∈ J such that t2 < j’, F(smoke’)(<w1, j’>)(Barack) = 1
(v) Given that F(smoke’)(<w1, t3>)(Barack) = 0, [[W smoke’(b)]]M,w1,t2,g = 0
d. think’([^smoke’(b)])(m) (‘Michelle thinks that Barack smokes’)
(i) [[think’([^smoke’(b)])(m)]]M,w1,t1,g = (by (18d))

M,w1,t1,g M,w1,t1,g M,w1,t1,g
(ii) [[think’]] ([[^smoke’(b)]]] )([[m]] ) = (by (18a))
(iii) F(think’)(<w1,t1>)([[^smoke’(b)]]]M,w1,t1,g)(F(m))(<w1,t1>)) = (by (15))
(iv) F(think’)(<w1,t1>)([[^smoke’(b)]]]M,w1,t1,g)(Michelle) = (by (21b))
(v) F(think’)(<w1,t1>)
({<<w1 , t1>,1>, <<w1 , t2>,1>, <<w1 , t3>,0>,
<<w2 , t1>,0>, <<w2 , t2>,0>, <<w2 , t3>,0>})(Michelle) = (by (15))
(vi) 1
10
Fall 2013 Ling 720
(23) Optional Exercise for the Student

Compute the value of [[ λv1,<e,t> Vv1,e [smoke’(v1,e) ∧ v1,<e,t> (v1,e)] ]]M,w1,t1,g
Show that it is the characteristic function of <et>-functions that are true of some smoker.
3. Some Key Validities of Intensional Logic
In this section, we’ll make note of some key validities in IL, which we will use to ‘convert’ the
translations of English expressions to simpler, logically equivalent formulae.
(24) Alpha Conversion

If variable v is bound in ϕ, and variable v’ appears nowhere in ϕ, then ϕ is logically
equivalent to [v/v’]ϕ
Although alpha-conversion is just the same as before, lambda-conversion is now subject to an

important new restriction.
(25) Lambda Conversion

Let (λvψ) and ϕ be meaningful expressions with no variables in common, and let ϕ ∈
MEτ and v ∈ Varτ. If (a) or (b) hold, then the expressions in (c) are logically equivalent.
a. The variable v in ψ does not appear in the scope of ^, , W, F

b. The denotation of ϕ is the same for every world and time (in every model M)
(That is, for all M, i,i’ ∈ I and j,j’ ∈ J, [[ϕ]]M,i,j,g = [[ϕ]]M,i’,j’,g)
c. (i) [λv ψ](ϕ) (ii) [ϕ/v]ψ
(26) Why This New Restriction on Lambda Conversion?
• Consider, for example, the formulae [λv Wχ](ϕ) and [ϕ/v]Wχ
• Consider their extensions at a given world and time:

[[ [λv Wχ](ϕ) ]]M,i,j,g and [[ [ϕ/v]Wχ ]]M,i,j,g
• When we compute [[ [λv Wχ](ϕ) ]]M,i,j,g, we will compute [[ [λv Wχ] ]]M,i,j,g and
apply it to the value [[ϕ]]M,i,j,g
• However, when we compute [[ [ϕ/v]Wχ ]]M,i,j,g, we don’t compute [[ϕ]]M,i,j,g. Rather,

we will end up computing [[ϕ]]M,i,j’,g for some j’ > j.
• Therefore, unless ϕ has the same extension for every world and time, we can’t
guarantee that [[ [λv Wχ](ϕ) ]]M,i,j,g will be the same value as [[ [ϕ/v]Wχ ]]M,i,j,g
11
Fall 2013 Ling 720
(27) An Analogy to Maybe Make (25)-(26) More Intuitive
• In LING 620, we learned that (a) is ambiguous, and has both the readings in (b).
a. John thinks that the president smokes.
b. (i) De Dicto Reading:

In all of John’s ‘belief worlds’ w’, the president in w’ smokes in w’.
(ii) De Re Reading:
In all of John’s ‘believe worlds’ w’, the president in w0 smokes in w’.
• We also saw that we can get reading (i) from an LF where the phrase the president is
in the scope of ‘believes’, while we get (ii) from and LF where it’s moved above the
scope of believes.
c. (i) De Dicto LF: [ John [ believes [ the president smokes ] ] ]
(ii) De Re LF: [ the president [ 1 [ John [ believes [ t1 smokes ] … ]
• This is because in LF (i), the president ends up semantically evaluated with respect to
the belief worlds, while in LF (ii), it is evaluated with respect to the actual world.
• Consequently, we saw that for proper names, the two LFs in (c) end up mapped to the
same interpretation, because proper names have the same value in every world
d. Two LFs Mapped to the Same Truth-Conditions

(i) [ John [ believes [ Mary smokes ] ] ]
(ii) [ Mary [ 1 [ John [ believes [ t1 smokes ] … ]
• The same general issue here is also at play for IL in (25)-(26).
o The IL operators ^, , H, W shift the evaluation worlds and time.
o Thus, if ϕ is in the scope of ^, , H, W, we won’t interpret ϕ with respect to

the ‘actual’ world/time
o Consequently, if the variable v in ψ is in the scope of ^, , W, F, then in (i) ϕ

is interpreted with respect to the initial, ‘matrix’ evaluation world/time, while
in (ii), it is not.
(i) [λv ψ](ϕ)
(ii) [ϕ/v]ψ
o However, if ϕ has the same extension in every world and time, then this
difference won’t matter for the resulting interpretation, and (i) and (ii)
will be logically equivalent.
12
Fall 2013 Ling 720
(28) One Last Important Note
• Given the definition in (18b), it follows that if ϕ is a variable, then for all i, i’ ∈ I and
j, j’ ∈ J, [[ϕ]]M,i,j,g = [[ϕ]]M,i’,j’,g = g(ϕ)
• Therefore, if ϕ is a variable, then as long as [λv ψ](ϕ) and ϕ have no variables in

common, then the following will always be logically equivalent (even in IL):
(i) [λv ψ](ϕ) (ii) [ϕ/v]ψ
In addition to alpha-conversion and lambda-conversion, our translations in PTQ will also make
use of the key logical equivalence in (29)…
(29) Down-Up Cancellation (Dowty et al. 1981)

The following two expressions are logically equivalent:
a. α
∨
b. [ [^α]]
• That is, put informally, the extension of α at world i and time j will always be equal
to the intension of α applied to the world i and the time j.
(30) Optional Exercise for Students

Let M be an intensional model <A, I, J, ≤, F>, i ∈ I, j ∈ J, and g be an arbitrary M-
∨
assignment. Show that [[α]]M,i,j,g = [[ [ [^α]] ]]M,i,j,g
4. Some Useful Abbreviations in PTQ
In PTQ, Montague makes use of a large number of meta-linguistic abbreviatory conventions.
(31) Relational Notation: γ(α,β) abbreviates γ(α)(β)

∨
(32) Curly Bracket Notation 1: γ{α} abbreviates [ γ](α)
∨
(33) Curly Bracket Notation 2: γ{α,β} abbreviates [ γ](α,β)
(34) Key Illustration: Let P be a variable of type <s,<e,t>>
[λP P{b}](^smoke’) ⇔ (by λ-conversion)

[^smoke’]{b} = (by Curly Bracket Notation (CBN))
∨
[ [^smoke’](b) ⇔ (by Down-Up Cancellation (DUC))
smoke’(b)
13
Fall 2013 Ling 720
(34) The Curvy Hat
• Montague puts a curvy ‘hat’ over a variable to indicate lambda abstraction over that
variable.
• I will follow Dowty et al. 1981 in not making use of this notation
o After all, ‘λ’ is clear and easy enough…
(35) The Sharp Hat ûϕ abbreviates [^λuϕ]
• I will not make much use of this abbreviation.

• However, do familiarize yourself with it, as Montague uses it a lot.
14
Fall 2013 Ling 720
The Translation System, Part 1 1
In these notes, we will begin to explore Montague’s system for mapping expressions in our
fragment of English to formulae of Intensional Logic.
1. The Category-to-Type Mapping
The first key ingredient of a PTQ-style translation system (and a UG-style translation base) is a
function mapping the categories of the natural language to the categories of the logical language
(i.e.,, the types).
• It is at this first step that the semantic analysis developed in PTQ immediately gets rather
complicated.
• Understanding this mapping and why it is this way is key to understanding everything
that follows…
(1) The Core Issue: Opaque Environments in English
• Recall from the handout “The Fragment of English” that our fragment of English
contains the following basic expressions: seek, allegedly, about
• Recall also that these expressions create an ‘opaque’ (non-extensional) environment.

o “John seeks a unicorn” doesn’t entail “there is a unicorn.”
o “John talked about a unicorn” doesn’t entail “there is a unicorn.”
o “John allegedly danced” doesn’t entail “John danced.”
• Consequently, the meanings (extensions) of these lexical items ([[X]]) must combine
with the intension of their complements
o [[seek]] takes the intension of a unicorn as argument
o [[about]] takes the intension of a unicorn as argument
o [[allegedly]] takes the intension of dance as argument.
• Thus, [[seek]] is not of type <e,<e,t>>, and so in our indirect semantics, the
translation of seek cannot be of type <e,<e,t>>
o Similarly, the translation of about cannot be of type <e,<e,t>>
o Similarly, the translation of allegedly cannot be of type <<e,t>,<e,t>>
(2) Burning Question: What should the types of [[seek]], [[about]], [[allegedly]] be?
Before we develop an answer to this question in (2), let us notice two additional important issues
1
These notes are based upon material in Dowty et al. (1981) Chapter 7.
1
Fall 2013 Ling 720
(3) Issue 1: Higher, Intensional Types for Everything
• Since seek is a TV, and its translation is not of type <e,<e,t>>, then – due to the need
for category-to-type correspondence in the translation – no TV can have a
translation of type <e,<e,t>>
o Therefore, even the translations of find, eat, and love will be of the same
high, intensional type as that of seek
• Similarly, since allegedly is an IAV, it follows that no IAV (not even ‘rapidly’) can
have a translation of type <<e,t>,<e,t>>
o Therefore, the translation of rapidly must be of the same, high, intensional

type as allegedly
• Similarly, since about is an IAV/T (preposition), it follows that no IAV/T (not even
‘in’) can have a translation of type <e,<e,t>>
o Therefore, the translation of in must also be of the same, high, intnsional type
as about
(4) Immediate Problem: Failure to Predict Key Inferences
If the translations (and induced interpretations) of eat, rapidly, and in are of the same
high, intensional types as seek, allegedly, and about, our semantics will fail to predict
the validity of the following inferences:
a. “John ate a unicorn” does entail “There is a unicorn.”

b. “John talked in a house” does entail “There is a house.”
c. “John rapidly danced” does entail “John danced.”
(5) The Solution (Preview)
As we’ll see later, we can solve the problem in (4) by adding ‘meaning postulates’ for
eat, rapidly, in that guarantee the inferences in (4).
• Thus, by adding these meaning postulates, we guarantee that these lexical items
behave as if they received a purely extensional semantics (even though they don’t).
2
Fall 2013 Ling 720
(6) Issue 2: All ‘Function Application’ is Intensional Function Application
• Given (3), all TVs will have an (induced) meaning that takes the intension of a term
as its argument:
o Both seek and eat take the intension of a unicorn as argument.
• Given (3), all IAVs will have an (induced) meaning that takes the intension of an IV
as its argument.
o Both allegedly and rapidly take the intension of dance as argument.
• Given (3), all IAV/Ts will have an (induced) meaning that takes the intension of a
term as its argument.
o Both about and in take the intension of a unicorn as argument.
• Also, note that the (induced) meaning of IV/t verbs like believe that take as argument
the intension of their sentential complements.
• Similarly, the (induced) meaning of IV//IV verbs like try to take as argument the
intentsion of their IV complements
The General Pattern Emerging Here:

The translation (induced meaning) of a predicative expression in English always takes as
argument the intension of the translation (induced meaning) of its syntactic argument.
(7) Immediate Consequence: Higher Types for the Translations of Terms
• Recall that we view Ts (= t/IV) as being predicates of IVs
• Thus, the general pattern in (6) entails that, in sentence (7a) below, the translation of a
T like John should take as argument the intension of the translation of smokes
a. John smokes.
• Thus, the translation of John should be something like [λv0,<s,<e,t>> v0,<s,<e,t>>{j}]
Rough Illustration:
(i) The translation of John smokes = (by (6))
(ii) The translation of John taking as argument the intension of the translation of
smokes. = (by assumption, and notation)
(iii) [λv0,<s,<e,t>> v0,<s,<e,t>>{j}]([^smokes’]) ⇔ (by λ-conversion)
(iv) [^smokes’]{j} = (by Curly Bracket Notation)
∨
(v) [ [^smokes’]](j) ⇔ (by Down-Up Cancellation)
(vi) smokes’(j)
3
Fall 2013 Ling 720
All these observations taken together lead us (and Montague) to the following general view of
the category-to-type correspondence in PTQ…
(8) Montague’s Category-to-Type Correspondence in PTQ

The function f has Cat as its domain and is such that:
a. f(e) = e
b. f(t) = t
c. f(A/B) = f(A//B) = < < s, f(B) >, f(A) >
(9) Remarks
Thus, a predicative expression in English (i.e., one of category A/B or A//B) will always
get a translation that takes as argument an intensional expression (type <s,f(B)>)
• Thus, if ϕ in English combines syntactically with an expression of category B (i.e., ϕ

is of category A/B)…
o Then the translation of ϕ will combine syntactically with the intension of
the translation of an expression of category B (i.e., <s, f(B)>)
• Thus, we immediately capture the higher intensional types of the translations of IVs,
IAVs, IAV/Ts, IV/ts, and IV//IVs
(10) A Possible Snag

As beautiful as (8) is, it entails the following category-to-type correspondences.
a. f(CN) = f(t//e) = < <s,e> , t >

b. f(IV) = f(t/e) = < <s,e> , t >
c. f(T) = f(t/(t/e)) = < < s, < <s,e>, t > >, t >
• Thus, (8) predicts that the translation of walk will be a function of type <<s,e>,t>
o Thus, the translation (meaning) of walk takes individual concepts as argument
• Consequently, terms end up translated to expressions of extremely high type
(11) But, Is This Really a ‘Snag’
a. A ‘Pro’ of the Correspondences in (10):

By treating all IVs as (indirectly) denoting <<s,e>,t> functions, Montague can
analyze such puzzling sentences as “The temperature is 90 and rising.”
b. ‘Cons’ of the Correspondences in (10):

• As detailed by Dowty et al. (1981), there may be other, superior analyses of
“The temperature is 90 and rising.”
• The resulting translations for (relatively simple) expressions of English get
rather complicated.
4
Fall 2013 Ling 720
(12) A Simpler Category-to-Type Corespondence
As discovered by Bennett (1976) and detailed by Dowty et al. (1981), the following
category-to-type mapping necessitates very few changes to the overall PTQ system.
The function f has Cat as its domain and is such that:

a. f(e) = e
b. f(t) = t
c. f(CN) = f(IV) = <e,t>
d. For all other categories A/B, f(A/B) = f(A//B) = < <s, f(B)>, f(A) >
(13) Category-to-Type Correspondences, Under (12)
Category Corresponding Type Set-Theoretic Object

t (sentences) t truth-value
CN <e,t> function from entities to truth-values
IV <e,t> function from entities to truth-values
T (= t/IV) < <s,<e,t>>,t> function from properties to truth-values2
IV/t < <s,t>, <e,t> > function from propositions to <e,t>-functions
IV//IV < <s, <e,t>>, <e,t>> function from properties to <e,t>-functions
IAV (= IV/IV) < <s, <e,t>>, <e,t>> function from properties to <e,t>-functions
TV (= IV/T) < <s, < <s,<e,t>>,t>>, <e,t>> function from the intensions of GQs
to an <e,t>-function.
I will follow Bennett (1976) and Dowty et al. (1981) in adopting the category-to-type
correspondences in (12)-(13).
• Again, these will not necessitate any serious changes from what’s in the original article…
• The benefit is that the system ends up being relatively simpler…
2. The ‘Lexical’ Translation Function
The second main ingredient of a PTQ-style translation system (and the third main ingredient of a
UG-style translation base) is a function translating the basic expressions of the natural language.
• Again, in PTQ, the range of this function can only be constants of the logical language.
• Consequently, this function in PTQ will not have the following basic expressions in its
range (since we want them translated as either (i) variables, or (ii) complex expressions)
o BT The basic terms of English, {John, Mary, Bill, ninety, he0, he1, he2, …}
o necessarily
o be
2
Such functions, I will (somewhat misleadingly) refer to as ‘GQs’
5
Fall 2013 Ling 720
(14) The ‘Lexical’ Translation Function
Let g be a function such that:
a. The domain of g is the set of basic expressions of our fragment of English other
than be, necessarily, and the members of BT
b. Whenever A ∈ Cat, α ∈ BA, and α is in the domain of g, g(α) ∈ Conf(A)
(15) A Picture of the ‘Lexical’ Translation Function
a. The IVs: g(run), g(walk), g(talk), g(rise), g(change) ∈ Con<e,t>
b. The CNs: g(man), g(woman), g(park), g(fish), g(pen), g(unicorn),

g(price), g(temperature) ∈ Con<e,t>
c. The IV/t’s: g(believe that), g(assert that) ∈ Con<<s,t>,<e,t>>
d. The IV//IVs: g(try to), g(wish to) ∈ Con<<s, <e,t>>,<e,t>>
e. The IAVs: g(rapidly), g(slowly), g(voluntarily),

g(allegedly) ∈ Con<<s, <e,t>>,<e,t>>
f. The TVs: g(find), g(lose), g(eat), g(love), g(date),

g(seek), g(conceive) ∈ Con< <s, < <s,<e,t>>,t>>, <e,t>>
g. The IAV/Ts: g(in), g(about) ∈ Con< <s, < <s,<e,t>>,t>>, <<s,<e,t>>,<e,t>>>
(16) Some Useful Meta-Linguistic Abbreviations
• Recall that Montague never specifies what the constants of IL look like.
• Nevertheless, it will be very useful to have some elegant, compact way of referring to
the translations of various basic expressions of English.
• Montague therefore introduces the following convention for forming meta-linguistic

labels / abbreviations for various translations
a. The Convention: If α is in the domain of g, then α’ = g(α)
b. Illustration:
(i) run’ = g(run) [i.e., whatever Con<e,t> g maps run to]
(ii) man’ = g(man) [i.e., whatever Con<e,t> g maps man to]
(iii) find’ = g(find) [i.e., whatever Con< <s, < <s,<e,t>>,t>>, <e,t>> g maps find to]
(iv) believe that’ =
g(believe that) [i.e., whatever Con<<s,t>,<e,t>> g maps believe that to]
6
Fall 2013 Ling 720
(17) Some Further Useful Meta-Linguistic Abbreviations
a. In our translation rule for proper names, we will want to make reference to certain
specified members of Cone Montague introduces j, m, b, n as meta-linguistic
labels for such constants.
• Again, the letters ‘j’, ‘m’, ‘b’, and ‘n’ aren’t (necessarily) constants.
• Rather, j, m, b, n are simply labels we are using to refer to such constants.
b. In our translation rules, we will also want to use certain specific variables.
Therefore, to save space, Montague introduces the following abbreviations:
(i) x, y, xn abbreviates v1,e , v3,e v2n,e (entity variable)
(ii) p abbreviates v0,<s,t> (proposition variable)
(iii) P,Q abbreviate v0,<s,<e,t>> , v1,<s,<e,t>> (property variable)
(iv) P abbreviates v0,<s, <<s,<e,t>>,t>> (variable over intensions of GQs)
With all these ingredients in place, we are now ready to lay out the translation rules of PTQ!
3. Some Translation Rules of PTQ
We will begin with the first translation rule, which governs the basic expressions of the fragment
(18) Rule T1 (For Basic Expressions)
a. If α is in the domain of g, then α translates into g(α)
Illustration: run translates to run’ (= g(run))

man translates to man’ (= g(man))
find translates to find’ (= g(find))
believes that translates to believes that’ (= g(believes that))
∨ ∨
b. be translates into λP λx P { [^λy [ x = y]] }
Note: Don’t worry about this for now; we may discuss it later.
∨
c. necessarily translates into λp [  p ]
Note: Again, don’t worry about this one for now.
7
Fall 2013 Ling 720
d. (i) John translates into λP[P{j}]
(ii) Mary translates into λP[P{m}]
(iii) Bill translates into λP[P{b}]
(iv) ninety translates into λP[P{n}]
Note:
• In PTQ itself, the actual translation of (e.g.) John is λP[P{^j}]
• This is because in PTQ, Montague held IVs like walk to be of type <<s,e>,t>
• Since we’re assuming the simpler category-to-type mapping in (12), we can
use the simpler translations in (18d) above.
e. hen translates into λP[P{xn}]
Thus, translation rule T1 will cover the translations of all the basic expressions in our fragment.
• From this point on, our discussion of the translation rules will not follow the order of the
rules in PTQ.
• Rather, as with Dowty et al. (1981), we will discuss the rules in a more ‘pedagogically
oriented’ order.
• We’ll thus next turn directly for the translation rule handling subject-predicate structures.
(19) Rule T4 (Subject-Predicate Translation Rule)
If δ ∈ Pt/IV, β ∈ PIV, and δ, β translate into δ’, β’ respectively, then F4(δ, β) translates into
δ’([^β’])
Note: Just as promised in (7), the translation of a term δ will take as argument the
intension of the translation of an IV β.
(20) Illustration of Rule T4
a. Obtaining the Translation

(i) John translates into λP[P{j}], run translates into run’ (T1)
(ii) F4(John, run) translates into [λP[P{j}]](^run’) (T4)
(iii) John runs translates into [λP[P{j}]](^run’) (def. of F4)
b. Simplifying the Translation

(i) [λP[P{j}]](^run’) ⇔ (λ-Conv.)
(ii) [^run’]{j} ⇔ (CBN)
∨
(iii) [ [^run’]](j) ⇔ (DUC)
(iv) run’(j)
8
Fall 2013 Ling 720
(21) Remarks
• Thus, the translation of John runs is logically equivalent to run’(j)
• Thus, under our induced semantics for English, we predict that relative to a model
M, a world i, a time j, and a variable assignment g, John runs will be 1 iff
o The extension of run’ w.r.t. M, i, j, g maps the extension of j w.r.t. M, i, j, g

to 1
o (Informally speaking) ‘John is running’ in world i at time j (in model M)
• Thus, we find that our induced semantics correctly predicts the truth-conditions
of present tense sentences.
o Their truth at a particular world/time depends upon the extensions of the
predicates at that world/time
(22) An Important Addition: Meaning Postulates for Translations of Names
• Note that, in order to really get the semantics of John runs right, we’re going to want
the extension of the constant j to have the same extension in all possible worlds/times.
• That is, in the translations of John, Mary, Bill, and ninety in (18d), we’re going to
want j, m, b, n to behave as ‘rigid designators’.
• We can ensure this if we add the ‘meaning postulate’ below:
Meaning Postulate for Names

In a ‘logically possible’ interpretation for IL, the following formulae are true (at all
worlds and times):
Vx [x = α], where α is j, m, b, or n

Informally: ‘There is a (single) entity x s.t. in all possible worlds/times x = j’
‘There is a (single) entity x s.t. in all possible worlds/times x = m’
‘There is a (single) entity x s.t. in all possible worlds/times x = b’
‘There is a (single) entity x s.t. in all possible worlds/times x = n’
• Thus, the truth of the formula above (at worlds/times) ensures that the constant j
denotes the same entity in all possible worlds (and the same holds for m, b, n)
(23) Remark
Since we’ll only ever be considering ‘logically possible’ interpretations (models) for IL,
we can now always regard the constants j, m, b, n as having constant intensions in IL.
• Thus, we can now freely ‘λ-convert’ them into the scope of ^, , W, H
9
Fall 2013 Ling 720
Rule T4 handles the translation/interpretation of (positive) present-tense sentences;
Rule T17 handles the translation/interpretation of negative and past/future tense sentences…
(24) Rule T17 (Translation Rule for Negative, Perfect, Future Sentences)
If α ∈ PT, δ ∈ PIV, and α, δ translate into α’, δ’ respectively, then
(i) F11(α, δ) translates into ¬α’([^δ’])

(ii) F12(α, δ) translates into W α’([^δ’])
(iii) F13(α, δ) translates into ¬W α’([^δ’])
(iv) F14(α, δ) translates into H α’([^δ’])
(v) F14(α, δ) translates into ¬H α’([^δ’])
Illustrations:
a. (i) John translates into λP[P{j}], run translates into run’ (T1)
(ii) F11(John, run) translates into ¬[λP[P{j}]](^run’) (T17)
(iii) John doesn’t run translates into ¬[λP[P{j}]](^run’) (def. of F11)
(iv) ¬[λP[P{j}]](^run’) ⇔ ¬run’(j) (λ-Conv., CBN, DUC)
Note:
Thus, the translation of John doesn’t run is logically equivalent to ¬run’(j)
o Thus, informally speaking, John doesn’t run will be true at a world/time if
it’s false that John is running at that world/time
b. (i) John translates into λP[P{j}], run translates into run’ (T1)
(ii) F12(John, run) translates into W[λP[P{j}]](^run’) (T17)
(iii) John will run translates into W[λP[P{j}]](^run’) (def. of F12)
(iv) W[λP[P{j}]](^run’) ⇔ W run’(j) (λ-Conv., CBN, DUC)
Note:
Thus, the translation of John will run is logically equivalent to W run’(j)
o Thus, informally speaking, John will run is true at a world/time if John runs
at some future time at that world.
c. (i) John translates into λP[P{j}], run translates into run’ (T1)
(ii) F15(John, run) translates into ¬H[λP[P{j}]](^run’) (T17)
(iii) John hasn’t run translates into ¬H [λP[P{j}]](^run’) (def. of F15)
(iv) ¬H [λP[P{j}]](^run’) ⇔ ¬H run’(j) (λ-Conv., CBN, DUC)
Note:
Thus, the translation of John hasn’t run is logically equivalent to ¬H run’(j)
o Thus, informally speaking John hasn’t run is true at a world/time if there is
no previous time at that world where John runs.
10
Fall 2013 Ling 720
Thus far, we’ve only been illustrating the compositional rules with proper names…
Another easy set of rules to examine, though, are the translation rules for quantificational terms
(25) Rule T2 (Translation Rule for Quantificational Terms)
If ζ ∈ PCN and ζ translates into ζ’, then:

(i) F0(ζ) translates into λP Λx[ζ’(x)  P{x}]
(ii) F1(ζ) translates into λP Vy [Λx [ζ’(x) ↔ x = y] & P{y}]
(iii) F2(ζ) translates into λP Vx[ζ’(x) ∧ P{x}]
Illustrations:
a. (i) man translates into man’ (T1)

(ii) F0(man) translates into λP Λx[man’(x)  P{x}] (T2)
(iii) every man translates into λP Λx[man’(x)  P{x}] (def. of F0)
b. (i) man translates into man’ (T1)

(ii) F1(man) translates into λP Vy [Λx [man’(x) ↔ x = y] & P{y}] (T2)
(iii) the man translates into λP Vy [Λx [man’(x) ↔ x = y] & P{y}] (def. of F1)
c. (i) man translates into man’ (T1)

(ii) F2(man) translates into λP Vx[man’(x) ∧ P{x}] (T2)
(iii) a man translates into λP Vx[man’(x) ∧ P{x}] (def. of F2)
Note: As previewed a few classes ago, Montague adopts a ‘Russelian’ analysis of definite
descriptions like “the man”.
(26) Interactions Between Rules T2, T4, and T17

With rules T2, T4, and T17, we can now easily translate/interpret sentences where
quantificational terms occupy subject position.
a. (i) a man translates into λP Vx[man’(x) ∧ P{x}] (T2)

(ii) run translates into run’ (T1)
(iii) F4(a man, run) translates into [λP Vx[man’(x) ∧ P{x}]](^run’) (T4)
(iv) a man runs translates into [λP Vx[man’(x) ∧ P{x}]](^run’) (def. of F4)
(v) [λP Vx[man’(x) ∧ P{x}]](^run’) ⇔ (λ-Conv.)

(vi) Vx[man’(x) ∧ [^run’]{x}] ⇔ (CBN)
∨
(vii) Vx[man’(x) ∧ [ [^run’]](x)] ⇔ (DUC)
(viii) Vx[man’(x) ∧ run’(x)]
11
Fall 2013 Ling 720
b. (i) a man translates into λP Vx[man’(x) ∧ P{x}] (T2)
(ii) run translates into run’ (T1)
(iii) F15(a man, run) translates into ¬H[λP Vx[man’(x) ∧ P{x}]](^run’)
(T17)
(iv) a man hasn’t run translates into ¬H[λP Vx[man’(x) ∧ P{x}]](^run’)
(def. of F15)
(v) ¬H [λP Vx[man’(x) ∧ P{x}]](^run’) ⇔ (λ-Conv., CBN, DUC)

(vi) ¬H Vx[man’(x) ∧ run’(x)]
(27) Remarks
a. Thus, we correctly predict that a man runs is true at a world/time iff there is a
man x (at that world/time) who runs (at that world/time)
b. Thus, we also correctly predict that a man hasn’t run has a ‘reading’
(translation/interpretation) that is true at a world/time iff there is no earlier time at
that world where a man runs…
o Note that this amounts to the claim that there is a reading of a man hasn’t
run that is equivalent to no man has run
c. Of course, there is also a ‘wide-scope indefinite’ reading of a man hasn’t run,

where it is true if there is some (particular) man x such that x hasn’t run.
o Once we bring QI into the mix, we’ll see that our fragment predicts this
reading as well!
Finally, we’ll also examine the translation rules for conjunction and disjunction, since those are
also relatively easy.
(28) Rule T11 (Conjunction and Disjunction of Sentences)
If ϕ, ψ ∈ Pt and ϕ, ψ translates into ϕ’, ψ’ respectively, then

(i) F8(ϕ, ψ) translates into [ϕ’∧ ψ’]
(ii) F9(ϕ, ψ) translates into [ϕ’∨ ψ’]
Illustration: The student is asked to confirm that rule T11 entails the following:
a. The translation of John runs and Mary talks is logically equivalent to:
[run’(j) ∧ talk’(m)]
b. The translation of John runs or Mary talks is logically equivalent to:
[run’(j) ∨ talk’(m)]
12
Fall 2013 Ling 720
(29) Rule T12 (Conjunction and Disjunction of IVs)
If γ, δ ∈ PIV and γ, δ translate into γ’, δ’ respectively, then

(i) F8(γ, δ) translates into λx[ γ’(x) ∧ δ’(x) ]
(ii) F9(γ, δ) translates into λx[ γ’(x) ∨ δ’(x) ]
a. First Key Illustration
(i) run translates into run’, talk translates into talk’ (T1)
(ii) F8(run, talk) translates into λx[ run’(x) ∧ talk’(x) ] (T12)
(iii) run and talk translates into λx[ run’(x) ∧ talk’(x) ] (def. of F8)
(iv) John translates into λP[P{j}] (T1)
(v) F4(John, run and talk) translates into
[λP[P{j}]]([^λx[ run’(x) ∧ talk’(x)]]) (T4)
(vi) John runs and talk translates into
[λP[P{j}]]([^λx[ run’(x) ∧ talk’(x)]]) (def. of F4)3
(vii) [λP[P{j}]]([^λx[ run’(x) ∧ talk’(x)]]) ⇔ (λ-Conv.)

(viii) [^λx[ run’(x) ∧ talk’(x)]]{j} ⇔ (CBN)
∨
(ix) [ [^λx[ run’(x) ∧ talk’(x)]]](j) ⇔ (DUC)
(x) [λx[ run’(x) ∧ talk’(x)](j) ⇔ (λ-Conv.)
(xi) [ run’(j) ∧ talk’(j)]
Note: Thus, in our system, the translation of John runs and talk(s) is logically
equivalent to [run’(j) ∧ talk’(j)], which is also the translation of John
runs and John talks…
b. Second Key Illustration:
(i) every man translates into λP Λx[man’(x)  P{x}] (T1, T2)

(ii) run or talk translates into λx[ run’(x) ∨ talk’(x) ] (T1, T12)
(iii) every man runs or talk translates into
[λP Λx[man’(x)  P{x}]](^λx[ run’(x) ∨ talk’(x) ]) (T4)
(iv) [λP Λx[man’(x)  P{x}]](^λx[ run’(x) ∨ talk’(x) ]) ⇔ (λ-Conv., α-Conv.)

(v) Λx[man’(x)  [^λy[ run’(y) ∨ talk’(y)]]{x}] ⇔ (CBN, DUC)
(vi) Λx[man’(x)  [λy[ run’(y) ∨ talk’(y)]](x)] ⇔ (λ-Conv.)
(vii) Λx[man’(x)  [ run’(x) ∨ talk’(x)]]
‘For all x, if x is a man, then either x runs or x talks’
Note: In our system, the translation of every man runs or talk(s) is not logically
equivalent to the translation of every man runs or every man talks
3
Note that F4 is defined so that it only inflects the first verb in a conjunction of IVs. This is clearly a problematic
prediction of the PTQ system.
13
Fall 2013 Ling 720
(30) Rule T13 (Disjunction of Terms)
If α, β ∈ PT and α, β translate into α’, β’ respectively, then F9(α, β) translates into

λP [α’(P) ∨ β’(P)]
Illustration:
(i) John translates into λP[P{j}], Mary translates into λP[P{m}] (T1)
(ii) John or Mary translates into λP[ [λP[P{j}]](P) ∨ [λP[P{j}]](P)] (T13)
(iii) John or Mary runs translates into [λP[ [λP[P{j}]](P) ∨ [λP[P{j}]](P)]] (^run’)
(T4)
(iv) [λP[ [λP[P{j}]](P) ∨ [λP[P{j}]](P)]] (^run’) ⇔ (λ-Conv.)

(v) [ [λP[P{j}]](^run’) ∨ [λP[P{j}]](^run’)] ⇔ (λ-Conv.)
(vi) [ [^run’]{j} ∨ [^run’]{m} ] ⇔ (CBN)
∨ ∨
(vii) [ [ [^run’]](j) ∨ [ [^run’]](m) ] ⇔ (DUC)
(viii) [ run’(j) ∨ run’(m) ]
Note: Thus, our system predicts that the translation of John or Mary runs will be
logically equivalent to the translation of John runs or Mary runs.
Thus far, we’ve covered about as much of the translation system as we can without also
discussing the translations of ‘Quantifying In’ and ‘Direct Object’ structures…
• These will be the focus of the next handout…
• These are also, in a sense, the ‘analytic centerpiece’ of the PTQ system…
14
Fall 2013 Ling 720
The Translation System, Part 2 1
Thus far, our discussion of the translation system in PTQ has assiduously avoided transitive
verbs, and the whole issue of how the contrast between seek and eat is to be approached…
Before we can lay out Montague’s solution, however, we need to get two other items on the table.
1. Quantifying In and the De Re / De Dicto Ambiguity
Rule T7 below will allow us to translate/interpret sentences like John believes that Mary runs.
(1) Rule T7 (For Finite Complements of PA Verbs)
If δ ∈ PIV/t and β ∈ Pt , and δ, β translate into δ’, β’ respectively, then F6(δ, β) translates
into δ’(^β’)
(i) believe that ∈ PIV/t , Mary runs ∈ Pt (S1, S4)

(ii) believe that translates into believe that’ (T1)
(iii) Mary runs translates into [λP[P{m}]](^run’) (T1, T4)
(iv) F6(believe that, Mary runs) translates into
believe that’([^[λP[P{m}]](^run’)]) (T7)
(v) believe that Mary runs translates into
believe that’([^[λP[P{m}]](^run’)]) (def. of F6)
(vi) John ∈ PT, believe that Mary runs ∈ PIV (S1, S7)
(vii) F4(John, believe that Mary runs) translates into
[λP[P{j}]]([^believe that’([^[λP[P{m}]](^run’)])]) (T4)
(viii) John believes that Mary runs translates into
[λP[P{j}]]([^believe that’([^[λP[P{m}]](^run’)])]) (def. of F4)
(i) [λP[P{j}]]([^believe that’([^[λP[P{m}]](^run’)])]) ⇔ (α-conversion)

(ii) [λP[P{j}]]([^believe that’([^[λQ[Q{m}]](^run’)])]) ⇔ (λ-conversion)
(iii) [^believe that’([^[λQ[Q{m}]](^run’)])]{j} ⇔ (CBN, DUC)
(iv) believe that’([^[λQ[Q{m}]](^run’)])(j) ⇔ (λ-conversion, CBN, DUC)
(v) believe that’([^run’(m)])(j)
1
These notes are based upon material in Dowty et al. (1981) Chapter 7.
1
Fall 2013 Ling 720
A major component of the overall PTQ system is the translation rule for structures formed by
‘Quantifying In’ (Rules S14-S16). It’s none too different from the translation rule in our ‘toy’
PTQ system…
(3) Rule T14 (For Translating ‘Quantifying-In’ Structures)
If α ∈ PT, ϕ ∈ Pt , and α, ϕ translate into α’, ϕ’ respectively, then F10,n(α, ϕ) translates

into α’([^λxn ϕ’])
Note: Again, following our general pattern, in the translation of F10,n(α, ϕ), the
translation of α takes as argument the intension of [λxn ϕ’].
a. Obtaining the translation
(i) he3 translates into λP[P{x3}], run translates into run’ (T1)
(ii) he3 runs translates into λP[P{x3}](^run’) (T4, def. of F4)
(iii) a man translates into λP Vx[man’(x) ∧ P{x}] (T2, def. of F2)
(iv) F10,3(a man, he3 runs) translates into

λP Vx[man’(x) ∧ P{x}]([^λx3[λP[P{x3}](^run’)]]) (T14)
(i) λP Vx[man’(x) ∧ P{x}]([^λx3[λP[P{x3}](^run’)]]) ⇔ (α-conversion)
(ii) λP Vx[man’(x) ∧ P{x}]([^λx3[λQ[Q{x3}](^run’)]]) ⇔ (λ-conversion)
(iii) Vx[man’(x) ∧ [^λx3[λQ[Q{x3}](^run’)]]{x}] ⇔ (CBN, DUC)
(iv) Vx[man’(x) ∧ [λx3[λQ[Q{x3}](^run’)]](x)] ⇔ (λ-conversion)
(v) Vx[man’(x) ∧ [λQ[Q{x}](^run’)] ] ⇔ (λ-conversion)
(vi) Vx[man’(x) ∧ [^run’]{x}] ⇔ (CBN, DUC)
(vii) Vx[man’(x) ∧ run’(x)]
With Rules T7 and T14 on the table, we can now capture the well-known ‘de re / de dicto’
ambiguity in a sentence like John believes that a man runs.
2
Fall 2013 Ling 720
(4) The De Dicto Reading
a. a man runs translates into [λP Vx[man’(x) ∧ P{x}]](^run’) (T1, T2, T4)
b. believe that a man runs translates into

believe that’([^[λP Vx[man’(x) ∧ P{x}]](^run’)]) (T7)
c. John believes that a man runs translates into

[λP[P{j}]]([^believe that’([^[λP Vx[man’(x) ∧ P{x}]](^run’)])]) (T4)
d. [λP[P{j}]]([^believe that’([^[λP Vx[man’(x) ∧ P{x}]](^run’)])])

⇔ (α-conversion, λ-conversion, CBN, DUC)
e. believe that’([^[λQ Vx[man’(x) ∧ Q{x}]](^run’)])(j)

⇔ (λ-conversion, CBN, DUC)
f. believe that’([^Vx[man’(x) ∧ run’(x)]])(j)
Note:
Under this translation, John believes that a man runs ends up being true iff John stands
in the believe that’ relation to the proposition “there is a man who runs”.
• Under the auxiliary assumption that x stands in the believe that’ relation to p iff p is
true in all of x’s belief worlds, we see that this translation garners us the so-called de
dicto reading of the sentence:
o I.e., in all of John’s belief worlds, there is a man (in that world) who runs (in
that world).
(5) The De Re Reading, Part 1: Obtaining the Translation
a. he3 runs translates into λP[P{x3}](^run’) (T1, T4)
b. believe that he3 runs translates into

believe that’([^[λP[P{x3}](^run’)]]) (T7)
c. John believes that he3 runs translates into

[λP[P{j}]]([^believe that’([^[λP[P{x3}](^run’)]])]) (T1, T4)
d. F10,3(a man, John believes that he3 runs) translates into (T2, T14)
[λP Vx[man’(x) ∧ P{x}]]
([^λx3[λP[P{j}]]([^believe that’([^[λP[P{x3}](^run’)]])])])
e. John believes that a man runs translates into (def. of F10,3)

[λP Vx[man’(x) ∧ P{x}]]
([^λx3[λP[P{j}]]([^believe that’([^[λP[P{x3}](^run’)]])])])
3
Fall 2013 Ling 720
(6) The De Re Reading, Part 2: Simplifying the Translation
Note that the ‘simplification’ below proceeds in an ‘inside-out’ fashion:
a. [λP Vx[man’(x) ∧ P{x}]]

([^λx3[λP[P{j}]]([^believe that’([^[λP[P{x3}](^run’)]])])]) ⇔ (λ-conversion,
CBN, DUC)
b. [λP Vx[man’(x) ∧ P{x}]]

([^λx3[λP[P{j}]]([^believe that’([^run’(x3)])])]) ⇔ (λ-conversion,
CBN, DUC)
c. [λP Vx[man’(x) ∧ P{x}]]

([^λx3 believe that’([^run’(x3)])(j)]) ⇔ (λ-conversion)
d. Vx[man’(x) ∧ [^λx3 believe that’([^run’(x3)])(j)]{x}] ⇔ (CBN, DUC)
e. Vx[man’(x) ∧ [λx3 believe that’([^run’(x3)])(j)](x)] ⇔ (λ-conversion) 2
f. Vx[man’(x) ∧ believe that’([^run’(x)])(j)]
Note:
Under this translation, John believes that a man runs ends up being true iff there is a
particular man x such that John stands in the believe that’ relation to the proposition “x
runs.”
• Thus, assuming that y believe that’ p holds iff p is true in all of y’s belief worlds, this
translation garners us the so-called de re reading fot he sentence:
o It needn’t be the case that in all (or any) of John’s ‘belief worlds’ there is a
man running…
 All that is required is that in all John’s belief worlds, x runs, where x in
the actual (evaluation) world is some man.
(7) Remark
• Thus, the syntactic-semantic analysis in PTQ holds to the ‘scope theory’ of the De Re
/ De Dicto ambiguity…
• Thus, it is also subject to all the problems that we saw the scope theory is subject to in
LING 620…
2
Recall that we’re able to do λ-conversion into the scope of ‘^’ here because ‘x’ is a variable, and so will have the
same across in all possible worlds/times.
4
Fall 2013 Ling 720
2. The De Re / De Dicto Ambiguity with Seeks
Recall that, just like the sentence John believes that a man runs, the sentence John seeks a
unicorn is also syntactically ambiguous in our English fragment.
(8) The Syntactic Ambiguity of John seeks a unicorn
John seeks a unicorn, 4
John seek a unicorn, 5
seek a unicorn, 2
unicorn
John seeks a unicorn, 10,3
a unicorn, 2 John seeks him3, 4
unicorn John seek him3, 5
seek he3
In this section, we will see that this similar syntactic ambiguity also leads to a similar semantic
ambiguity:
• Under the derivation in (8a), John seeks a unicorn receives a de dicto reading (where
there need be no actual unicorns)
• Under the derivation in (8b), John seeks a unicorn receives a de re reading (where there
exists a particular, actual unicorn x such that John seeks x)
(9) Rule T5 (Translation Rule for TVs and Direct Objects)
If δ ∈ PTV and β ∈ PT , and δ, β translate into δ’, β’ respectively, then F5(δ, β) translates
into δ’(^β’)
Note: Again, following our general pattern, in the translation of F5(δ, β), the translation
of δ takes as argument the intension of the translation of β.
5
Fall 2013 Ling 720
(10) Illustration of Rule T5 (The De Dicto Reading of John seeks a unicorn)
(i) seek translates into seek’ (T1)
(ii) a unicorn translates into λP Vx[unicorn’(x) ∧ P{x}] (T2)
(iii) F5(seek, a unicorn) translates into

seek’([^λP Vx[unicorn’(x) ∧ P{x}]]) (T5)
(iv) John translates into [λP[P{j}]] (T1)
(v) John seeks a unicorn translates into

[λP[P{j}]](^seek’([^λP Vx[unicorn’(x) ∧ P{x}]])) (T4)
(i) [λP[P{j}]](^seek’([^λP Vx[unicorn’(x) ∧ P{x}]])) ⇔ (α-conversion)

(ii) [λQ[Q{j}]](^seek’([^λP Vx[unicorn’(x) ∧ P{x}]])) ⇔ (λ-conversion)
(iii) [^seek’([^λP Vx[unicorn’(x) ∧ P{x}]])]{j} ⇔ (CBN, DUC)
(iv) seek’([^λP Vx[unicorn’(x) ∧ P{x}]])(j)
(11) Remarks
a. Note that no further simplification can take place to the formula in (10b,iv).
• In our translation, seek’ is a constant of type <<s,<<s,<e,t>>,t>>, <e,t>>
• Thus, it takes as argument (i) and (ii) to yield an expression of type t
o (i) the intension of a generalized quantifier expression, and
o (ii) an entity
b. Thus, under the syntactic derivation in (8a), John seeks a unicorn receives a
translation which is true iff:
John stands in the seek’ relation to the intension of a unicorn
c. Finally, let us assume that x stands in the seek’ relation to P iff in all the world-
times <w’,t’> where x’s desires are met, P (w’,t’)(λy : x has y in w’ at t’)
• Thus, John stands in the seek’ relation to the intension of a unicorn iff in all
the world-times <w’,t’> where John’s desires are met:
There is an x such that x is a unicorn in w’ and t’ at John has x in w’ at t’
d. We see, then, that the translation we generate for parse (8a) amounts to the de
dicto reading of John seeks a unicorn.
6
Fall 2013 Ling 720
Now let’s see what happens when we derive the sentence along the lines in (8b):
(11) Obtaining the De Re Reading
(i) seek translates into seek’, he3 translates into λP[P{x3}] (T1)
(ii) F5(seek, he3) translates into seek’(^λP[P{x3}]) (T5)
(iii) seek him3 translates into seek’(^λP[P{x3}]) (def. of F5)
(iv) John translates into [λP[P{j}]] (T1)
(v) John seeks him3 translates into [λP[P{j}]](^seek’(^λP[P{x3}])) (T4)
(vi) a unicorn translates into λP Vx[unicorn’(x) ∧ P{x}] (T2)
(vii) John seeks a unicorn translates into

[λP Vx[unicorn’(x) ∧ P{x}]]
(^λx3[[λP[P{j}]](^seek’(^λP[P{x3}]))]) (T14, def. of F10,3)
(i) [λP Vx[unicorn’(x) ∧ P{x}]](^λx3[[λP[P{j}]](^seek’(^λP[P{x3}]))])

(ii) [λP Vx[unicorn’(x) ∧ P{x}]](^λx3 [seek’(^λP[P{x3}])(j)])
(iii) Vx[unicorn’(x) ∧ seek’(^λP[P{x}])(j)]
(12) Remarks
a. Thus, under the syntactic derivation in (8b), John seeks a unicorn receives a
translation which is true iff:
There is an x such that x is a unicorn, and John stands in the seek’ relation
to the GQ-intension ^λP[P{x}]
b. Again, let us assume that x stands in the seek’ relation to P iff in all the world-
times <w’,t’> where x’s desires are met, P (w’,t’)(λy : x has y in w’ at t’)
Thus, John stands in the seek’ relation to ^λP[P{x}] iff in all the world-
times <w’,t’> such that John’s desires are met, John has x.
c. We see, then, that the translation we generate for parse (8b) amounts to the de re
reading of John seeks a unicorn.
7
Fall 2013 Ling 720
3. Transitive Verbs that Don’t Seem to Create Opaque Contexts
(13) Interim Summary

Our translation system generates two different translations for John seeks a unicorn
a. Under one translation (reading), the sentence does not entail that any unicorns
actually exist (the de dicto reading)
b. Under another translation (reading), the sentence does entail that unicorns exist.
(the de re reading)
(14) An Immediate Problem: Parallel Ambiguity for John eats a unicorn.
• Note that in our English fragment, the sentence John eats a unicorn has the same
structural ambiguity as John seeks a unicorn.

John eats a unicorn, 4
John eat a unicorn, 5
eat a unicorn, 2
unicorn
John eats a unicorn, 10,3
a unicorn, 2 John eats him3, 4
unicorn John eat him3, 5
eat he3
• Furthermore, recall that – due to our category-to-type mapping – the English

transitive verb eat is also translated as a type <<s,<<s,<e,t>>,t>>, <e,t>> predicate.
• Consequently, our translation system predicts that John eats a unicorn will be
ambiguous in the same way as John seeks a unicorn.
o That is, John eats a unicorn will receive two translations, one
logically equivalent to (14c), and the other equivalent to (14d).
c. Translation One: eat’([^λP Vx[unicorn’(x) ∧ P{x}]])(j)
d. Translation Two: Vx[unicorn’(x) ∧ eat’(^λP[P{x}])(j)]
8
Fall 2013 Ling 720
(15) The Acute Empirical Problem: No Perception of Ambiguity
Contrary to the predictions above, English speakers don’t perceive an ambiguity in John
eats a unicorn, akin to that in John seeks a unicorn.
• More acutely, under the translation in (14c), John eats a unicorn can be true without
there being any actual unicorns (just like with (10b))
• However, English speakers universally agree that John eats a unicorn entails that
there does exist some unicorn (which John is eating).
• That is, of the two translations in (14c,d), only (14d) seems to align with the truth-
conditional judgments of English speakers.
(16) The Key Idea Towards a Solution

Suppose it were the case that the ‘lexical semantics’ of eat entailed that (16a) is true iff
(16b) is true.
a. eat’(P)(x)
∨
b. [ P]([^λx0 [eat’(^λP[P{x0}])(x)]])
As the computation below shows, it would predict that translation (14c) will be logically
equivalent to translation (14d).
c. (i) eat’([^λP Vx[unicorn’(x) ∧ P{x}]])(j) ⇔ (by assumption (16))
∨
(ii) [ [^λP Vx[unicorn’(x) ∧ P{x}]]]([^λx0 [eat’(^λP[P{x0}])(j)]])
⇔ (CBN, DUC)
(iii) [λP Vx[unicorn’(x) ∧ P{x}]]]([^λx0 [eat’(^λP[P{x0}])(j)]])

⇔ (λ-conversion)
(iv) Vx[unicorn’(x) ∧ [^λx0 [eat’(^λP[P{x0}])(j)]]{x}] ⇔ (CBN, DUC)
(v) Vx[unicorn’(x) ∧ [λx0 [eat’(^λP[P{x0}])(j)]](x)] ⇔ (λ-conversion)
(vi) Vx[unicorn’(x) ∧ eat’(^λP[P{x}])(j)]
If this were the case, then, we wouldn’t perceive an ambiguity in John eats a unicorn,
since the two translations (readings) would be truth-conditionally equivalent.
• Furthermore, both translations end up entailing that there are unicorns.
9
Fall 2013 Ling 720
In the PTQ system, we have a mechanism for encoding such aspects of ‘lexical semantics’ into
our analysis: the meaning postulates!
(17) The Solution: Meaning Postulate for ‘Non-Intensional’ Verbs

In a ‘logically possible’ interpretation for IL, the following formulae are true (at all
worlds and times): 3
∨
ΛxΛP  [ δ(P )(x) ↔ [ P]([^λx0 δ(^λP[P{x0}])(x)])
where δ translates any member of BTV other than seek or conceive
(18) Remark
If we restrict the interpretations of IL to only those that are ‘logically possible’, it will
follow (as shown in (16)) that translation (14c) is logically equivalent to (14d).
We will also predict the univocality of such sentences as:
a. John finds a unicorn. b. John loses a unicorn.

c. John loves a unicorn. d. John dates a unicorn.
(19) A General Summary of the Overall Story Regarding Transitive Verbs
a. We know that seek must receive a translation/meaning where it takes as argument

the intension of its complement.
b. Thus, the need for a category-to-type correspondence in the PTQ (and UG)
framework entails that all transitive verbs must take as argument the intension of
their complement.
c. Thus, even the translation/meaning of eat is a relation between an entity and the
intension of some GQ.
d. This, of course, raises the question of what the ‘lexical semantics’ of eat are.
When should we say that the eat relation holds between an entity and some GQ
intension?
∨
e. Well, suppose we say that eat holds between x and P exactly when [ P] holds of
the following crazy property: ([^λx0 [eat’(^λP[P{x0}])(x))
• That’s what the meaning postulate in (17) says…
f. As shown above, the resulting system correctly predicts that – while John seeks a
unicorn doesn’t entail the existence of unicorns – John eats a unicorn does.
3
Note that the actual formula used by Montague in PTQ is slightly different from the one in (17). However, as the
student can confirm, they amount to the same condition (see Dowty et al. (1981), pp. 219-227).
10
Fall 2013 Ling 720
(20) One Final Note
In PTQ, Montague extends the general strategy in (16)-(19) above to other categories
besides TV. In this way, the PTQ system is also able to capture the observed contrasts
between (i) rapidly vs. allegedly, and (ii) in vs. about.
11
Fall 2013 Ling 720
Final, Extended Problem Set on the PTQ System
1. Exercises Concerning the English Fragment
(1) Exercise on the Syntactic Rules of the Fragment

Please provide derivations showing that each of the following are meaningful expressions
of the English fragment in PTQ.
a. a woman hasn’t eaten the unicorn such that she has found it
b. John will talk about every woman
c. Bill doesn’t believe that every man or every woman dates him voluntarily
(2) Question Concerning Relative Clauses and Adverbs

An English sentence can contain an unbounded number of relative clauses and adverbs.
Does the English fragment in PTQ accurately capture this property of English? Why or
why not?
(3) Question Concerning Conjoined IVs

Does the English fragment in PTQ accurately predict that (3a) and (3b) are both
meaningful expressions of English, while (3c) is not? Why or why not?
a. John wishes to walk and talk
b. John walks and talks
c. * John wishes to walks and talk
2. Exercises Concerning Intensional Logic
In the exercises below, we will assume the following meta-language abbreviations and labels for
the variables and constants of IL.
Type Variables Constants

e x, y, z j, b, m
<s,e> r ----
<e,t> X run’, man’
<<s,e>,t> Q change’
<s, <e,t>> P ----
<e,e> ---- father-of’
<e,<e,t>> R love’
<s,t> p ----
<<s,t>,<e,t>> ---- believe’
1
Fall 2013 Ling 720
(4) Exercise on the Syntax of IL
For each of the formulae below, please state whether it is or is not a meaningful
expression of IL. If it is a meaningful expression, please state its type. If it is not a
meaningful expression, briefly explain why.
a. ^j f. λP P = ^run’
∨ ∨
b. change’(^j) g. [^ X]( ^j)
c. run’(^j) h. ^λx love’(j)(x)
d. λr change’(r) i. λp p
e. λP{j} j. λy father-of’(father-of’(y))
(5) Exercise on the Semantics of IL

Let M be any intensional model <A, I, J, ≤, F> for IL, let i ∈ I and j ∈ J, and let g be an
M-assignment. Prove each of the following: 1
a. [[ ¬W run’(j) ]]M,i,j,g = 1 iff

There is no j’ ∈ J such that j < j’ and F(run’)(<i,j’>)(F(j)(<i,j’>)) = 1
b. [[ [λP P{b}](^love’(m)) ]]M,i,j,g = 1 iff

F(love’)(<i,j>)(F(m)(<i,j>))(F(b)(<i,j>)) = 1
c. [[ Vx [man’(x) ∧ H run’(x)] ]]M,i,j,g = 1 iff

there is an x ∈ De, A,I,J such that F(man’)(<i,j>)(x) = 1 and
for some j’ ∈ J such that j’ < j, F(run’)(<i,j’>)(x) = 1
d. [[ believe’(^run’(j))(b) ]]M,i,j,g = 1 iff

F(believe’)(<i,j>)
(the function h with domain I×J such that if <i’,j’> ∈ I×J, then
h(<i’,j’>) = F(run’)(<i’,j’>)(F(j)(<i’,j’>)))
(F(b)(<i,j>)) = 1
3. Exercises Concerning the Translation System
(6) Exercise on the Analysis of Conjunction and Disjunction

Does the system in PTQ predict that (6a) and (6b) will ever receive logically equivalent
translat ions? Is this prediction accurate? [For the purposes of this exercise, please ignore
‘quantifying in’.]
a. a man or a woman walks and talks.

b. a man or a woman walks and a man or a woman talks
1
Note that for the purposes of this assignment, it is perfectly legitimate to show that a given formula of IL is
logically equivalent to another, syntactically simpler one, and then calculate the extension of the simpler formula.
2
Fall 2013 Ling 720
(7) Exercise on the Analysis of De Re / De Dicto Ambiguity
Show that the PTQ system predicts that (7a) should have a reading where it is true in
scenario (7b). Is this prediction accurate?
a. The Sentence: John believes that a man or a woman runs.
b. The Scenario:
John sincerely asserts “a man or the world’s best tennis player runs.” Little does
John know that the world’s best tennis player is actually a woman.
4. Thought Questions Concerning the MG Architecture
(8) Trees as Grammatical Objects

Please consider carefully the statement below. In what ways is it accurate, and in what
ways is it an (over)simplification? Please defend your answer.
“In Montague Grammar (MG), unlike in transformational frameworks like the

Minimalist Program, trees are never grammatical objects. That is, in MG, the
syntactic operations act on strings to create strings. Trees in MG are simply
diagrams that the analyst uses as an expository devise, to compactly represent the
derivations of the meaningful expressions, which are always strings.”
(9) Truth-Conditions and Model-Theoretic Semantics

In LING 610, we learned of the importance of truth-conditions to semantic theory, and
that semanticists seek to build systems that pair sentences of a natural language (e.g.)
English with their truth-conditions.
Do the algebraic and model-theoretic semantic systems we learned about in this
class – those in UG and PTQ – do this? That is, do they do they make predictions
regarding the truth-conditions of English sentences? Why or why not?

From Logic To MG

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

From Logic To MG

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

From Logic To MG

Uploaded by

Copyright:

Available Formats

From Logic to Montague Grammar:

Some Formal and Conceptual Foundations of Semantic Theory

Course Handouts and Problem Sets

January 13, 2014

Course Reading List

Unit 1: Formal Preliminaries

Unit 2: A Review of Propositional Logic and First Order Logic

Unit 3: Algebras and Semantics

Unit 5: An Algebraic Approach to Quantification and Lambda Abstraction

Unit 6: The Proper Treatment of Quantification in Ordinary English

1. Unit 1: Formal Preliminaries

1.1 Relations and Functions

1.2 Cardinalities, Infinities and Proof by Induction

1.3 Problem Set on the Formal Preliminaries

2. Unit 2: A Review of Propositional Logic and First Order Logic

2.1 Propositional Logic: Syntax and Natural Deduction

2.2 First Order (Predicate) Logic: Syntax and Natural Deduction

2.3 Problem Set on Syntax and Natural Deduction

2.4 Propositional Logic: Formal Semantics and Valuations

2.5 First Order Logic: Formal Semantics and Models

2.6 Problem Set on Formal Semantics (Valuations and Models)

2.7 Proving the Soundness and Completeness of Propositional Logic

3. Unit 3: Algebras and Semantics

3.1 An Algebraic Perspective on Propositional Logic

3.2 Problem Set on Syntactic Operations, Semantic Operations, Homomorphisms

3.3 An Algebraic Perspective on the Syntax of FOL (Without Quantification)

3.4 Montague’s General Theory of Semantics

3.5 Problem Set on Languages and Interpretations

4.1 Laying the Groundwork

4.2 The Notion of a ‘Translation Base’

4.3 Translation Bases and Interpretations

4.4 Problem Set on Translation and Indirect Interpretation

5. Unit 5: An Algebraic Approach to Quantification and Lambda Abstraction

5.2 Fregean Interpretations

5.3 Computing with ‘Logically Possible Partly-Fregean Interpretations’

5.4 Applications to the Analysis of English

5.5 Problem Set on the Analysis of Quantification

6. Unit 6: The Proper Treatment of Quantification in Ordinary English

6.2 The Fragment of English

6.3 Intensional Logic

6.4 The Translation System, Part 1

6.5 The Translation System, Part 2

6.6 Problem Set on the PTQ System

From Logic to Montague Grammar:

1.2 Structure of the Course

• Propositional Logic (PL): Syntax and Natural Deduction System

Readings and Resources (Posted on Moodle)

Part 2: Algebras and Semantics

• Introduction to Algebras and Morphisms.

Readings and Resources (Posted on Moodle)

Part 3: Algebras, Translations, and Indirect Interpretation

• Montague’s Theory of Indirect Interpretation

Readings and Resources (Posted on Moodle)

Part 4: Montague’s Intensional Logic (IL) and its Applications

• Key concepts behind the Intensional Logic (IL)

Readings and Resources (Posted on Moodle)

Part 5: Classic Papers in Montague Grammar

Readings and Resources (Posted on Moodle)