XGitUrl: http://lambda.jimpryor.net/git/gitweb.cgi?p=lambda.git;a=blobdiff_plain;f=week1.mdwn;h=c68da8a0e636b1ceba1cdfbf955905168ee076dc;hp=be1b714dd9c3b6e5801d786f827b7e89c3042d19;hb=87483b8ff52adf85fd8d80060427f9e67f698b8a;hpb=545b8774052183ad91afd0b7dc741162af7ac405
diff git a/week1.mdwn b/week1.mdwn
index be1b714d..c68da8a0 100644
 a/week1.mdwn
+++ b/week1.mdwn
@@ 30,297 +30,317 @@ From philosophy
* issues in that same neighborhood will help us better understand proposals like Kit Fine's that semantics is essentially coordinated, and that `R a a` and `R a b` can differ in interpretation even when `a` and `b` don't
+Declarative/functional vs Imperatival/dynamic models of computation
+===================================================================
+Many of you, like us, will have grown up thinking the paradigm of computation is a sequence of changes. Let go of that. It will take some care to separate the operative notion of "sequencing" here from other notions close to it, but once that's done, you'll see that languages that have no significant notions of sequencing or changes are Turing complete: they can perform any computation we know how to describe. In itself, that only puts them on equal footing with more mainstream, imperatival programming languages like C and Java and Python, which are also Turing complete. But further, the languages we want you to become familiar with can reasonably be understood to be more fundamental. They embody the elemental building blocks that computer scientists use when reasoning about and designing other languages.
+Jim offered the metaphor: think of imperatival languages, which include "mutation" and "sideeffects" (we'll flesh out these keywords as we proceeed), as the pate of computation. We want to teach you about the meat and potatoes, where as it turns out there is no sequencing and no changes. There's just the evaluation or simplification of complex expressions.
1. Declarative vs imperatival models of computation.
2. Variety of ways in which "order can matter."
3. Variety of meanings for "dynamic."
4. Schoenfinkel, Curry, Church: a brief history
5. Functions as "firstclass values"
6. "Curried" functions
+Now, when you ask the Scheme interpreter to simplify an expression for you, that's a kind of dynamic interaction between you and the interpreter. You may wonder then why these languages should not also be understood imperatively. The difference is that in a purely declarative or functional language, there are no dynamic effects in the language itself. It's just a static semantic fact about the language that one expression reduces to another. You may have verified that fact through your dynamic interactions with the Scheme interpreter, but that's different from saying that there are dynamic effects in the language itself.
1. Beta reduction
1. Encoding pairs (and triples and ...)
1. Encoding booleans
+What the latter would amount to will become clearer as we build our way up to languages which are genuinely imperatival or dynamic.
+
+Many of the slogans and keywords we'll encounter in discussions of these issues call for careful interpretation. They mean various different things.
+
+For example, you'll encounter the claim that declarative languages are distinguished by their **referential transparency.** What's meant by this is not always exactly the same, and as a cluster, it's related to but not the same as this means for philosophers and linguists.
+
+The notion of **function** that we'll be working with will be one that, by default, sometimes counts as nonidentical functions that map all their inputs to the very same outputs. For example, two functions from jumbled decks of cards to sorted decks of cards may use different algorithms and hence be different functions.
+
+It's possible to enhance the lambda calculus so that functions do get identified when they map all the same inputs to the same outputs. This is called making the calculus **extensional**. Church called languages which didn't do this "intensional." If you try to understand this in terms of functions from worlds to extensions (an idea also associated with Church), you will hurt yourself. So too if you try to understand it in terms of mental stereotypes, another notion sometimes designated by "intension."
+
+It's often said that dynamic systems are distinguished because they are the ones in which **order matters**. However, there are many ways in which order can matter. If we have a trivalent boolean system, for exampleeasily had in a purely functional calculuswe might choose to give a truthtable like this for "and":
+
+ true and true = true
+ true and * = *
+ true and false = false
+ * and true = *
+ * and * = *
+ * and false = *
+ false and true = false
+ false and * = false
+ false and false = false
+
+And then we'd notice that `* and false` has a different intepretation than `false and *`. (The same phenomenon is already present with the mateial conditional in bivalent logics; but seeing that a nonsymmetric semantics for `and` is available even for functional languages is instructive.)
+
+Another way in which order can matter that's present even in functional languages is that the interpretation of some complex expressions can depend on the order in which subexpressions are evaluated. Evaluated in one order, the computations might never terminate (and so semantically we interpret them as having "the bottom value"we'll discuss this). Evaluated in another order, they might have a perfectly mundane value. Here's an example, though we'll reserve discussion of it until later:
+
+ (\x. y) ((\x. x x) (\x. x x))
+
+Again, these facts are all part of the metatheory of purely functional languages. But *there is* a different sense of "order matters" such that it's only in imperatival languages that order so matters.
+
+ x := 2
+ x := x + 1
+ x == 3
+
+Here the comparison in the last line will evaluate to true.
+
+ x := x + 1
+ x := 2
+ x == 3
+Here the comparison in the last line will evaluate to false.
+One of our goals for this course is to get you to understand *what is* that new
+sense such that only so matters in imperatival languages.
+Finally, you'll see the term **dynamic** used in a variety of ways in the literature for this course:
+* dynamic versus static typing
+* dynamic versus lexical scoping
 Order matters
+* dynamic versus static control operators
Declarative versus imperative:
+* finally, we're used ourselves to talking about dynamic versus static semantics
In a pure declarative language, the order in which expressions are
evaluated (reduced, simplified) does not affect the outcome.
+For the most part, these uses are only loosely connected to each other. We'll tend to use "imperatival" to describe the kinds of semantic properties made available in dynamic semantics, languages which have robust notions of sequencing changes, and so on.
(3 + 4) * (5 + 11) = 7 * (5 + 11) = 7 * 16 = 112
(3 + 4) * (5 + 11) = (3 + 4) * 16 = 7 * 16 = 112
+Map
+===
In an imperative language, order makes a difference.
x := 2
x := x + 1
x == 3
[true]
+Rosetta Stone
+=============
x := x + 1
x := 2
x == 3
[false]
Declaratives: assertions of statements.
No matter what order you assert true facts, they remain true:
The value is the product of x and y.
x is the sum of 3 and 4.
y is the sum of 5 and 11.
The value is 112.
+Basics of Lambda Calculus
+=========================
Imperatives: performative utterances expressing a deontic or bouletic
modality ("Be polite", "shut the door")
Resourcesensitive, order sensitive:
+The lambda calculus we'll be focusing on for the first part of the course has no types. (Some prefer to say it instead has a single typebut if you say that, you have to say that functions from this type to this type also belong to this type. Which is weird.)
Make x == 2.
Add one to x.
See if x == 3.
+Here is its syntax:

+
+Variables: x
, y
, z
...
+
Untype (monotyped) lambda calculus
+Each variable is an expression. For any expressions M and N and variable a, the following are also expressions:
Syntax:
+
+Abstract: (λa M)
+
Variables: x, x', x'', x''', ...
(Cheat: x, y, z, x1, x2, ...)
+We'll tend to write (λa M)
as just `(\a M)`, so we don't have to write out the markup code for the λ
. You can yourself write (λa M)
or `(\a M)` or `(lambda a M)`.
Each variable is a term.
For all terms M and N and variable a, the following are also terms:
+
+Application: (M N)
+
(M N) The application of M to N
(\a M) The abstraction of a over M
+Some authors reserve the term "term" for just variables and abstracts. We won't participate in that convention; we'll probably just say "term" and "expression" indiscriminately for expressions of any of these three forms.
Examples of terms:
+Examples of expressions:
x
(y x)
(x x)
(\x y)
(\x x)
(\x (\y x))
(x (\x x))
((\x (x x))(\x (x x)))
+ x
+ (y x)
+ (x x)
+ (\x y)
+ (\x x)
+ (\x (\y x))
+ (x (\x x))
+ ((\x (x x)) (\x (x x)))
Reduction/conversion/equality:
+The lambda calculus has an associated proof theory. For now, we can regard the proof theory as having just one rule, called the rule of **betareduction** or "betacontraction". Suppose you have some expression of the form:
Lambda terms express recipes for combining terms into new terms.
The key operation in the lambda calculus is betaconversion.
+ ((\a M) N)
((\a M) N) ~~>_beta M{a := N}
+that is, an application of an abstract to some other expression. This compound form is called a **redex**, meaning it's a "betareducible expression." `(\a M)` is called the **head** of the redex; `N` is called the **argument**, and `M` is called the **body**.
The term on the left of the arrow is an application whose first
element is a lambda abstraction. (Such an application is called a
"redex".) The beta reduction rule says that a redex is
betaequivalent to a term that is constructed by replacing every
(free) occurrence of a in M by a copy of N. For example,
+The rule of betareduction permits a transition from that expression to the following:
((\x x) z) ~~>_beta z
((\x (x x)) z) ~~>_beta (z z)
((\x x) (\y y)) ~~>_beta (\y y)
+ M [a:=N]
+
+What this means is just `M`, with any *free occurrences* inside `M` of the variable `a` replaced with the term `N`.
+
+What is a free occurrence?
+
+> An occurrence of a variable `a` is **bound** in T if T has the form `(\a N)`.
+
+> If T has the form `(M N)`, any occurrences of `a` that are bound in `M` are also bound in T, and so too any occurrences of `a` that are bound in `N`.
+
+> An occurrence of a variable is **free** if it's not bound.
Beta reduction is only allowed to replace *free* occurrences of a variable.
An occurrence of a variable a is BOUND in T if T has the form (\a N).
If T has the form (M N), and the occurrence of a is in M, then a is
bound in T just in case a is bound in M; if the occurrence of a is in
N, than a is bound in T just in case a is bound in N. An occurrence
of a variable a is FREE in a term T iff it is not bound in T.
For instance:
T = (x (\x (\y (x (y z)))))
The first occurrence of x in T is free. The second occurrence of x
immediately follows a lambda, and is bound. The third occurrence of x
occurs within a form that begins with "\x", so it is bound as well.
Both occurrences of y are bound, and the only occurrence of z is free.
+> T is defined to be `(x (\x (\y (x (y z)))))`
Lambda terms represent functions.
All (recursively computable) functions can be represented by lambda
terms (the untyped lambda calculus is Turning complete).
For some lambda terms, it is easy to see what function they represent:
+The first occurrence of `x` in T is free. The `\x` we won't regard as being an occurrence of `x`. The next occurrence of `x` occurs within a form that begins with `\x`, so it is bound as well. The occurrence of `y` is bound; and the occurrence of `z` is free.
(\x x) the identity function: given any argument M, this function
simply returns M: ((\x x) M) ~~>_beta M.
+Here's an example of betareduction:
(\x (x x)) duplicates its argument:
((\x (x x)) M) ~~> (M M)
+ ((\x (y x)) z)
(\x (\y x)) throws away its second argument:
(((\x (\y x)) M) N) ~~> M
+betareduces to:
and so on.
+ (y z)
+
+We'll write that like this:
+
+ ((\x (y x)) z) ~~> (y z)
+
+Different authors use different notations. Some authors use the term "contraction" for a single reduction step, and reserve the term "reduction" for the reflexive transitive closure of that, that is, for zero or more reduction steps. Informally, it seems easiest to us to say "reduction" for one or more reduction steps. So when we write:
+
+ M ~~> N
+
+We'll mean that you can get from M to N by one or more reduction steps. Hankin uses the symbol →
for onestep contraction, and the symbol ↠
for zeroormore step reduction. Hindley and Seldin use ⊳_{1}
and ⊳
.
+
+When M and N are such that there's some P that M reduces to by zero or more steps, and that N also reduces to by zero or more steps, then we say that M and N are **betaconvertible**. We'll write that like this:
+
+ M <~~> N
+
+This is what plays the role of equality in the lambda calculus. Hankin uses the symbol `=` for this. So too do Hindley and Seldin. Personally, I keep confusing that with the relation to be described next, so let's use this notation instead. Note that `M <~~> N` doesn't mean that each of `M` and `N` are reducible to each other; that only holds when `M` and `N` are the same expression. (Or, with our convention of only saying "reducible" for one or more reduction steps, it never holds.)
+
+In the metatheory, it's also sometimes useful to talk about formulas that are syntactically equivalent *before any reductions take place*. Hankin uses the symbol ≡
for this. So too do Hindley and Seldin. We'll use that too, and will avoid using `=` when discussing metatheory for the lambda calculus. Instead we'll use `<~~>` as we said above. When we want to introduce a stipulative definition, we'll write it out longhand, as in:
+
+> T is defined to be `(M N)`.
It is easy to see that distinct lambda terms can represent the same
function. For instance, (\x x) and (\y y) both express the same
function, namely, the identity function.
+We'll regard the following two expressions:

Dot notation: dot means "put a left paren here, and put the right
+ (\x (x y))
+
+ (\z (z y))
+
+as syntactically equivalent, since they only involve a typographic change of a bound variable. Read Hankin section 2.3 for discussion of different attitudes one can take about this.
+
+Note that neither of those expressions are identical to:
+
+ (\x (x w))
+
+because here it's a free variable that's been changed. Nor are they identical to:
+
+ (\y (y y))
+
+because here the second occurrence of `y` is no longer free.
+
+There is plenty of discussion of this, and the fine points of how substitution works, in Hankin and in various of the tutorials we've linked to about the lambda calculus. We expect you have a good intuitive understanding of what to do already, though, even if you're not able to articulate it rigorously.
+
+
+Shorthand
+
+
+The grammar we gave for the lambda calculus leads to some verbosity. There are several informal conventions in widespread use, which enable the language to be written more compactly. (If you like, you could instead articulate a formal grammar which incorporates these additional conventions. Instead of showing it to you, we'll leave it as an exercise for those so inclined.)
+
+
+**Dot notation** Dot means "put a left paren here, and put the right
paren as far the right as possible without creating unbalanced
parentheses". So (\x(\y(xy))) = \x\y.xy, and \x\y.(z y) x =
(\x(\y((z y) z))), but (\x\y.(z y)) x = ((\x(\y(z y))) x).
+parentheses". So:
+
+ (\x (\y (x y)))
+
+can be abbreviated as:
+
+ (\x (\y. x y))
+
+and:
+
+ (\x (\y. (z y) z))
+
+would abbreviate:
+
+ (\x (\y ((z y) z)))
+
+This on the other hand:

+ (\x (\y. z y) z)
Church figured out how to encode integers and arithmetic operations
using lambda terms. Here are the basics:
+would abbreviate:
0 = \f\x.fx
1 = \f\x.f(fx)
2 = \f\x.f(f(fx))
3 = \f\x.f(f(f(fx)))
...
+ (\x (\y (z y)) z)
Adding two integers involves applying a special function + such that
(+ 1) 2 = 3. Here is a term that works for +:
+**Parentheses** Outermost parentheses around applications can be dropped. Moreover, applications will associate to the left, so `M N P` will be understood as `((M N) P)`. Finally, you can drop parentheses around abstracts, but not when they're part of an application. So you can abbreviate:
+ = \m\n\f\x.m(f((n f) x))
+ (\x. x y)
So (+ 0) 0 =
(((\m\n\f\x.m(f((n f) x))) ;+
 \f\x.fx) ;0
 \f\x.fx) ;0
+as:
~~>_beta targeting m for beta conversion
+ \x. x y
((\n\f\x.[\f\x.fx](f((n f) x)))
 \f\x.fx)
+but you should include the parentheses in:
\f\x.[\f\x.fx](f(([\f\x.fx] f) x))
+ (\x. x y) z
\f\x.[\f\x.fx](f(fx))
+and:
\f\x.\x.[f(fx)]x
+ z (\x. x y)
\f\x.f(fx)
+**Merging lambdas** An expression of the form `(\x (\y M))`, or equivalently, `(\x. \y. M)`, can be abbreviated as:
+ (\x y. M)
+Similarly, `(\x (\y (\z M)))` can be abbreviated as:

+ (\x y z. M)
A concrete example: "damn" side effects
1. Sentences have truth conditions.
2. If "John read the book" is true, then
 John read something,
 Someone read the book,
 John did something to the book,
 etc.
3. If "John read the damn book",
 all the same entailments follow.
 To a first approximation, "damn" does not affect atissue truth
 conditions.
4. "Damn" does contribute information about the attitude of the speaker
 towards some aspect of the situation described by the sentence.
+Lambda terms represent functions
+
+All (recursively computable) functions can be represented by lambda
+terms (the untyped lambda calculus is Turing complete). For some lambda terms, it is easy to see what function they represent:
+
+(\x x) represents the identity function: given any argument M, this function
+simply returns M: ((\x x) M) ~~> M.
+
+(\x (x x)) duplicates its argument:
+((\x (x x)) M) ~~> (M M)
+
+(\x (\y x)) throws away its second argument:
+(((\x (\y x)) M) N) ~~> M
+
+and so on.
+
+It is easy to see that distinct lambda expressions can represent the same
+function, considered as a mapping from input to outputs. Obviously:
+ (\x x)

Old notes, no longer operative:
+and:
1. Theoretical computer science is beautiful.
+ (\z z)
 Google search for "anagram": Did you mean "nag a ram"?
 Google search for "recursion": Did you mean "recursion"?
+both represent the same function, the identity function. However, we said above that we would be regarding these expressions as synactically equivalent, so they aren't yet really examples of *distinct* lambda expressions representing a single function. However, all three of these are distinct lambda expressions:
 Y = \f.(\x.f (x x)) (\x.f (x x))
+ (\y x. y x) (\z z)
+ (\x. (\z z) x)
1. Understanding the meaning(use) of programming languages
 helps understanding the meaning(use) of natural langauges
+ (\z z)
 1. Richard Montague. 1970. Universal Grammar. _Theoria_ 34:37598.
 "There is in my opinion no important theoretical difference
 between natural languages and the artificial languages of
 logicians; indeed, I consider it possible to comprehend the
 syntax and semantics of both kinds of languages within a
 single natural and mathematically precise theory."
+yet when applied to any argument M, all of these will always return M. So they have the same extension. It's also true, though you may not yet be in a position to see, that no other argument can differentiate between them when they're supplied as an argument to it. However, these expressions are all syntactically distinct.
 2. Similarities:
+The first two expressions are *convertible*: in particular the first reduces to the second. So they can be regarded as prooftheoretically equivalent even though they're not syntactically identical. However, the proof theory we've given so far doesn't permit you to reduce the second expression to the third. So these lambda expressions are nonequivalent.
 Function/argument structure:
 f(x)
 kill(it)
 pronominal binding:
 x := x + 1
 John is his own worst enemy
 Quantification:
 foreach x in [1..10] print x
 Print every number from 1 to 10
+There's an extension of the prooftheory we've presented so far which does permit this further move. And in that extended proof theory, all computable functions with the same extension do turn out to be equivalent (convertible). However, at that point, we still won't be working with the traditional mathematical notion of a function as a set of ordered pairs. One reason is that the latter but not the former permits uncomputable functions. A second reason is that the latter but not the former prohibits functions from applying to themselves. We discussed this some at the end of seminar (and further discussion is best pursued in person).
 3. Possible differences:
 Parentheses:
 3 * (4 + 7)
 ?It was four plus seven that John computed 3 multiplied by
 (compare: John computed 3 multiplied by four plus seven)
 Ambiguity:
 3 * 4 + 7
 Time flies like and arrow, fruit flies like a banana.
 Vagueness:
 3 * 4
 A cloud near the mountain
 Unbounded numbers of distinct pronouns:
 f(x1) + f(x2) + f(x3) + ...
 He saw her put it in ...
 [In ASL, dividing up the signing space...]


2. Standard methods in linguistics are limited.
 1. Firstorder predicate calculus
 Invented for reasoning about mathematics (Frege's quantification)
 Alethic, order insensitive: phi & psi == psi & phi
 But: John left and Mary left too /= Mary left too and John left
+Booleans and pairs
+==================
 2. Simplytyped lambda calculus
+Our definition of these is reviewed in [[Assignment1]].
 Can't express the Y combinator
3. Meaning is computation.
 1. Semantics is programming
 2. Good programming is good semantics
 1. Example
+1. Declarative vs imperatival models of computation.
+2. Variety of ways in which "order can matter."
+3. Variety of meanings for "dynamic."
+4. Schoenfinkel, Curry, Church: a brief history
+5. Functions as "firstclass values"
+6. "Curried" functions
+
+1. Beta reduction
+1. Encoding pairs (and triples and ...)
+1. Encoding booleans
 1. Programming technique
 Exceptions
 throw (raise)
 catch (handle)
 2. Application to linguistics
 presupposition
 expressives
 Develop application:
 fn application
 divide by zero
 test and repair
 raise and handle
 fn application
 presupposition failure
 build into meaning of innocent predicates?
 expressives
 throw
 handle
 resume computation
