Chiastic Lambda-Calculi - Indiana University...
Transcript of Chiastic Lambda-Calculi - Indiana University...
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Chiastic Lambda-Calculi
wren gayle romano
Cognitive Science & Computational LinguisticsIndiana University, Bloomington
4 November 2014
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 0 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples and Motivation
Associative λ-calculi
Chiastic λ-calculi
〈〈λ→〉〉χL in action
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 1 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples and Motivation
Associative λ-calculi
Chiastic λ-calculi
〈〈λ→〉〉χL in action
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 1 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples
Scrambling in Japanese• Tarou
—-gaNom
honbook
-woAcc
yon-daread-Perf
• honbook
-woAcc
Tarou—
-gaNom
yon-daread-Perf
‘Taro read the book.’
Keyword arguments• yonda(wo=‘hon’, ga=‘Tarou’)• yonda(ga=‘Tarou’, wo=‘hon’)
Shorthands in category theory• FG where (FG )(X ) = F (GX)
• ηF where (ηF )(X ) = ηFX
• Fη where (Fη)(X ) = F (ηX
)
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 1 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
What do these have in common?
Juxtaposition is (essentially) associative
(f g) x ≈ f (g x)
Application is (essentially) commutative
f x y ≈ f y x
Our Goal convert those “≈” into “=”
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 2 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Scrambling in Japanese
Many languages have “free word order”
• Tarou—
-gaNom
honbook
-woAcc
yon-daread-Perf
• honbook
-woAcc
Tarou—
-gaNom
yon-daread-Perf
‘Taro read the book.’
• Both orders are normal and natural
• Both have the same propositional content
• Though, information structure may differ
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 3 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Arguments: Chomskian-style accounts
Tarou. . . . . .N
-ga. . . . . . . . .NPnom\N
JNPnom
hon. . . .N
-wo. . . . . . . .NPacc\N
JNPacc
yon-. . . . . . . . . . . . . . . .V \NPnom\NPacc
JV \NPnom
JV
-da. . . . .S\V
JS
hon. . . .N
-wo. . . . . . . .NPacc\N
JNPacc
Tarou. . . . . .N
-ga. . . . . . . . .NPnom\N
JNPnom
ITV /(V \NPnom)
yon-. . . . . . . . . . . . . . . .V \NPnom\NPacc
IBxV \NPacc
JV
-da. . . . .S\V
JS
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 4 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Adjuncts: Radical neo-Davidsonian accounts
Tarou. . . . . .N
-ga. . . . . . .S/S\N
JS/S
hon. . . .N
-wo. . . . . . .S/S\N
JS/S
yon-. . . .V
-da. . . . .S\V
JSI
SI
S
hon. . . .N
-wo. . . . . . .S/S\N
JS/S
Tarou. . . . . .N
-ga. . . . . . .S/S\N
JS/S
yon-. . . .V
-da. . . . .S\V
JSI
SI
S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 5 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Adjuncts: Radical neo-Davidsonian accounts
Tarou. . . . . .N
-ga. . . . . . .S/S\N
JS/S
hon. . . .N
-wo. . . . . . .S/S\N
JS/S
IBS/S
yon-. . . .V
-da. . . . .S\V
JSI
S
hon. . . .N
-wo. . . . . . .S/S\N
JS/S
Tarou. . . . . .N
-ga. . . . . . .S/S\N
JS/S
IBS/S
yon-. . . .V
-da. . . . .S\V
JSI
S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 6 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Arguments vs Adjuncts
Why prefer adjuncts?
• Avoids the need for T and Bx (they’re dangerous together)
• Syntax matches morphology/prosody
• Same parse tree for different word orders (commutativity)
• Online and partial parsing is easy (associativity)
Only moves the problem from syntax to semantics!• Also true of other CCG approaches to scrambling
Chiastic λ-calculi solve the problem(in the semantics)
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 7 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Arguments vs Adjuncts
Why prefer adjuncts?
• Avoids the need for T and Bx (they’re dangerous together)
• Syntax matches morphology/prosody
• Same parse tree for different word orders (commutativity)
• Online and partial parsing is easy (associativity)
Only moves the problem from syntax to semantics!• Also true of other CCG approaches to scrambling
Chiastic λ-calculi solve the problem(in the semantics)
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 7 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Arguments vs Adjuncts
Why prefer adjuncts?
• Avoids the need for T and Bx (they’re dangerous together)
• Syntax matches morphology/prosody
• Same parse tree for different word orders (commutativity)
• Online and partial parsing is easy (associativity)
Only moves the problem from syntax to semantics!• Also true of other CCG approaches to scrambling
Chiastic λ-calculi solve the problem(in the semantics)
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 7 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples and Motivation
Associative λ-calculi
Chiastic λ-calculi
〈〈λ→〉〉χL in action
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 8 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
What are functions?
Traditional λ-calculi intentionally confuse two ideas
Procedures operations mapping values to valuesData values representing procedures
Category theory keeps them distinct
Morphisms functions as proceduresExponentials functions as data
For associativity, we must keep them distinct too
(λx. e) Unbracketed abstractions are procedures〈〈λx. e〉〉 Bracketed abstractions are values
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 8 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Associative λ-calculi: 〈〈λ〉〉
Variables x , y, z ,. . .
Terms e, f, g ,. . . ::= x variables
| (λx. e) abstraction
| 〈〈e〉〉 bracketing
| e · f juxtaposition
Beta(λx. f ) · 〈〈e〉〉 {x 7→ e}f
Assoc(e · f ) · g ≡ e · (f · g)
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 9 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
What does juxtaposition mean?
Application (λx. e) · 〈〈f 〉〉
Composition (λx. e) · (λy. f )((λx. e) · (λy. f )
)· 〈〈g〉〉 ≡ (λx. e) ·
((λy. f ) · 〈〈g〉〉
)Tupling 〈〈f 〉〉 · 〈〈g〉〉
(λx.λy. e) ·(〈〈f 〉〉 · 〈〈g〉〉
)≡
((λx.λy. e) · 〈〈f 〉〉
)· 〈〈g〉〉
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 10 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
How powerful is it?
〈〈L 〉〉 is at least as powerful as L• Every L -term has an evaluation-equivalent 〈〈L 〉〉-term
JxK = x
J(λx. e)K = (λx. JeK)
...
Je · f K = JeK · 〈〈Jf K〉〉J(e)K = JeK
〈〈L 〉〉 can be more expressive than L• 〈〈λ→〉〉 has tuples, but they can’t be encoded in λ→
• Then again, almost everything stronger than λ→ has tuples
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 11 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples and Motivation
Associative λ-calculi
Chiastic λ-calculi
〈〈λ→〉〉χL in action
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 12 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Chiastic λ-calculi
The term level
Syntax — Two flavors of chiasmusEquivalenceReductionSanity check
The type level
SyntaxEquivalenceReductionSanity check — Well-formed types
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 12 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Formalizing restricted commutativity
Actually we don’t want full commutativity
• Tarou—
-gaNom
kurumacar
-gaNom
ar-uhave-Npst
‘Taro has a car.’
% kurumacar
-gaNom
Tarou—
-gaNom
ar-uhave-Npst
‘The car has a Taro.’
• Let a dimension denote a class of non-commuting elements
• Elements along different dimensions don’t interfere with one another
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 13 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Chiastic λ-calculi: 〈〈λ〉〉χVariables x , y, z ,. . .
Dimensions A, B, C ,. . .
Terms e, f, g ,. . . ::= x variables
| (λAx. e) abstraction
| 〈〈e〉〉A bracketing
| e · f juxtaposition
• Choose one
For this talk, we’ll only consider Chi L
A 6= BChi L
(λAx. λBy. e) ≡ (λBy. λAx. e)
A 6= BChi R
〈〈e〉〉A · 〈〈f 〉〉B ≡ 〈〈f 〉〉B · 〈〈e〉〉A
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 14 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Chiastic λ-calculi: 〈〈λ〉〉χVariables x , y, z ,. . .
Dimensions A, B, C ,. . .
Terms e, f, g ,. . . ::= x variables
| (λAx. e) abstraction
| 〈〈e〉〉A bracketing
| e · f juxtaposition
• For this talk, we’ll only consider Chi L
A 6= BChi L
(λAx. λBy. e) ≡ (λBy. λAx. e)
A 6= BChi R
〈〈e〉〉A · 〈〈f 〉〉B ≡ 〈〈f 〉〉B · 〈〈e〉〉A
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 14 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Chi L vs Chi R
Should we accept terms like this?
(λAx. λBy. e) · (λCz. 〈〈a〉〉A) · 〈〈b〉〉B · 〈〈c〉〉C
Should we accept sentences like this?• [sono
thathonbook
-woAcc
〈Hanako—
-ga〉Nom
Tarou—
-gaNom
kat-ta]buy-Perf
-toComp
omot-te iruthink.Prog
‘Hanako thinks that Taro bought that book.’
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 15 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Term equivalence for 〈〈λ〉〉χL
e ≡ f
A 6= B
(λAx. λB y. e) ≡ (λB y. λAx. e) e · (f · g) ≡ (e · f ) · g
e ≡ e ′
(λAx. e) ≡ (λAx. e′)
e ≡ e ′
〈〈e〉〉A ≡ 〈〈e ′〉〉Ae ≡ e ′ f ≡ f ′
e · f ≡ e ′ · f ′
e ≡ ef ≡ e
e ≡ f
e ≡ f f ≡ ge ≡ g
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 16 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Term reduction for 〈〈λ〉〉χL
e e ′
h ≡ (λAx. f )
h · 〈〈e〉〉A {x 7→ e}fe e ′
(λAx. e) (λAx. e′)
e e ′
〈〈e〉〉A 〈〈e ′〉〉Ae e ′
e · f e ′ · ff f ′
e · f e · f ′
e · f he · (f · g) h · g
f · g h
(e · f ) · g e · h
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 17 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our terms make sense?
Theorem Term reduction is weak Church–Rosser.
Proof There are no critical pairs.
Corollary Term reduction is Church–Rosser.
Proof Supposing we can prove strong normalization,then just use Newman’s lemma.
Conjecture Term reduction (for 〈〈λ→〉〉χL) is strongly normalizing.
Remark This is suspiciously difficult to prove.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 18 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
What are types?
The intrinsic view (a la Church)• Types are manifest in terms• Terms can have only one type• Ill-typed terms “don’t exist”
The extrinsic view (a la Curry)• Types characterize properties of terms• Terms could have multiple types• All terms exist, but we only care about the well-typed ones
Our view• Types give abstract interpretations of terms
Γ ` e B τ τ ∗ τ ′
∃e ′. e ∗ e ′ ∧ Γ ` e ′ B τ ′
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 19 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
What are types?
The intrinsic view (a la Church)• Types are manifest in terms• Terms can have only one type• Ill-typed terms “don’t exist”
The extrinsic view (a la Curry)• Types characterize properties of terms• Terms could have multiple types• All terms exist, but we only care about the well-typed ones
Our view• Types give abstract interpretations of terms
Γ ` e B τ τ ∗ τ ′
∃e ′. e ∗ e ′ ∧ Γ ` e ′ B τ ′
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 19 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Simply-typed left-chiastic λ-calculus: 〈〈λ→〉〉χL
Types σ, τ , υ,. . . ::= T primitive types
| σA→ τ arrow types
| 〈〈τ〉〉A bracketed types
| σ · τ juxtaposition
Γ ` e B τ
`ctx Γ Γ(x) ≡ τ
Γ ` x B τΓ, x :σ ` e B τ
Γ ` (λAx. e) B σA→ τ
Γ ` e B τΓ ` 〈〈e〉〉A B 〈〈τ〉〉A
Γ ` e B σ Γ ` f B τΓ ` e ·f B σ·τ
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 20 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Type equivalence for 〈〈λ→〉〉χL
τ ≡ σ
A 6= B
σA→ τ
B→ υ ≡ τB→ σ
A→ υ σ · (τ · υ) ≡ (σ · τ) · υ
σ ≡ σ′ τ ≡ τ ′
σA→ τ ≡ σ′
A→ τ ′
τ ≡ τ ′
〈〈τ〉〉A ≡ 〈〈τ ′〉〉Aσ ≡ σ′ τ ≡ τ ′
σ · τ ≡ σ′ · τ ′
τ ≡ τσ ≡ ττ ≡ σ
σ ≡ τ τ ≡ υσ ≡ υ
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 21 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Type reduction for 〈〈λ→〉〉χL
τ τ ′
ρ ≡ σA→ τ
ρ · 〈〈σ〉〉A τσ σ′
σA→ τ σ′
A→ τ
τ τ ′
σA→ τ σ
A→ τ ′
τ τ ′
〈〈τ〉〉A 〈〈τ ′〉〉Aσ σ′
σ · τ σ′ · ττ τ ′
σ · τ σ · τ ′
σ · τ ρ
σ · (τ · υ) ρ · υτ · υ ρ
(σ · τ) · υ τ · ρ
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 22 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
Theorem Type reduction for 〈〈λ→〉〉χL is strongly normalizing Proof
Theorem Type reduction for 〈〈λ→〉〉χL is Church–Rosser Proof
So we can define
Γ ` e B τ0 NF(τ0) ≡ τ `type τΓ ` e : τ
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 23 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
But, what does unresolved type juxtaposition mean?
Good 〈〈σ〉〉A · 〈〈τ〉〉BBad (σ
A→ 〈〈τ〉〉B) · 〈〈υ〉〉C where A 6= C
(σA→ τ) · 〈〈υ〉〉A where σ 6≡ υ
Ugly 〈〈σ〉〉A · (τB→ υ)
(ρA→ σ) · (τ B→ υ)
• If `type τ doesn’t accept ugly terms,then it doesn’t have the subterm property.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 24 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
But, what does unresolved type juxtaposition mean?
Good 〈〈σ〉〉A · 〈〈τ〉〉B
Bad (σA→ 〈〈τ〉〉B) · 〈〈υ〉〉C where A 6= C
(σA→ τ) · 〈〈υ〉〉A where σ 6≡ υ
Ugly 〈〈σ〉〉A · (τB→ υ)
(ρA→ σ) · (τ B→ υ)
• If `type τ doesn’t accept ugly terms,then it doesn’t have the subterm property.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 24 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
But, what does unresolved type juxtaposition mean?
Good 〈〈σ〉〉A · 〈〈τ〉〉BBad (σ
A→ 〈〈τ〉〉B) · 〈〈υ〉〉C where A 6= C
(σA→ τ) · 〈〈υ〉〉A where σ 6≡ υ
Ugly 〈〈σ〉〉A · (τB→ υ)
(ρA→ σ) · (τ B→ υ)
• If `type τ doesn’t accept ugly terms,then it doesn’t have the subterm property.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 24 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
But, what does unresolved type juxtaposition mean?
Good 〈〈σ〉〉A · 〈〈τ〉〉BBad (σ
A→ 〈〈τ〉〉B) · 〈〈υ〉〉C where A 6= C
(σA→ τ) · 〈〈υ〉〉A where σ 6≡ υ
Ugly 〈〈σ〉〉A · (τB→ υ)
(ρA→ σ) · (τ B→ υ)
• If `type τ doesn’t accept ugly terms,then it doesn’t have the subterm property.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 24 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Do our types make sense?
Every type has a normal form
But, what does unresolved type juxtaposition mean?
Good 〈〈σ〉〉A · 〈〈τ〉〉BBad (σ
A→ 〈〈τ〉〉B) · 〈〈υ〉〉C where A 6= C
(σA→ τ) · 〈〈υ〉〉A where σ 6≡ υ
Ugly 〈〈σ〉〉A · (τB→ υ)
(ρA→ σ) · (τ B→ υ)
• If `type τ doesn’t accept ugly terms,then it doesn’t have the subterm property.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 24 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Examples and Motivation
Associative λ-calculi
Chiastic λ-calculi
〈〈λ→〉〉χL in action
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 25 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Using 〈〈λ→〉〉χL to describe Japanese
Noun phrase scrambling
Tarou-ga hon-wo yondaHon-wo Tarou-ga yonda
Verbal morphology
“Paradoxical” behaviorResolving the paradox
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 25 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga
Tarou. . . . . . . . . . . . . . . .
〈〈Taro ′〉〉N :
N
-ga. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) :
S/S\NJ
(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) · 〈〈Taro ′〉〉N :
S/Sβ
(λS s. 〈〈s·〈〈Taro ′〉〉nom〉〉S) :
S/S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 26 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga
Tarou. . . . . . . . . . . . . . . .〈〈Taro ′〉〉N : N
-ga. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) : S/S\N
J(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) · 〈〈Taro ′〉〉N : S/S
β
(λS s. 〈〈s·〈〈Taro ′〉〉nom〉〉S) : S/S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 26 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga
Tarou. . . . . . . . . . . . . . . .〈〈Taro ′〉〉N
: N
-ga. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S)
: S/S\N
J(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) · 〈〈Taro ′〉〉N
: S/S
β
(λS s. 〈〈s·〈〈Taro ′〉〉nom〉〉S)
: S/S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 26 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga
Tarou. . . . . . . . . . . . . . . .〈〈Taro ′〉〉N
: N
-ga. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S)
: S/S\N
J(λN n. λS s. 〈〈s·〈〈n〉〉nom〉〉S) · 〈〈Taro ′〉〉N
: S/S
β
(λS s. 〈〈s·〈〈Taro ′〉〉nom〉〉S)
: S/S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 26 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of Tarou-ga hon-wo yonda
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈λnom n. n read ′ book ′〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λnom n. n read ′ book ′〉〉Sβ
〈〈(λnom n. n read ′ book ′) · 〈〈Taro′〉〉nom〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 27 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis of hon-wo Tarou-ga yonda
hon-wo
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S)
Tarou-ga
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S)
yonda
〈〈λacc a. λnom n. n read ′ a〉〉SI
(λS s. 〈〈s·〈〈Taro′〉〉nom〉〉S) · 〈〈λacc a. λnom n. n read ′ a〉〉Sβ
〈〈(λacc a. λnom n. n read ′ a) · 〈〈Taro′〉〉nom〉〉Sβ(χL)
〈〈λacc a. Taro′ read ′ a〉〉SI
(λS s. 〈〈s·〈〈book ′〉〉acc〉〉S) · 〈〈λacc a. Taro′ read ′ a〉〉Sβ
〈〈(λacc a. Taro′ read ′ a) · 〈〈book ′〉〉acc〉〉Sβ
〈〈Taro′ read ′ book ′〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 28 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
“Paradoxical” verbal morphology
Causative and passive verb forms• tabe-ru ‘to eat’• tabe-sase-ru ‘to cause to eat’• tabe-rare-ru ‘to be made to eat’
“Paradoxical” behavior of causative and passive
Morpho-phonologically behaves as a single wordSemantically behaves as if involving complementation• E.g., adverb scope ambiguity
But this “paradox” is due to traditional notions of constituency• Kubota 2008 vs GB, LFG, HPSG
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 29 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
General scheme for verbal morphology
Let the dimension E denote eventualities
Verbal roots have types of the general form
〈〈· · · → 〈〈τ〉〉E 〉〉V
Verbal inflections use “multicomposition”
(λVv. 〈〈(λEe. f ) · v〉〉S)
• v can have any arity
• The semantic content f , has access to the whole eventuality e• So if e is a compound eventuality, f can affect all or part of it
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 30 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Lexical entries for a few verbal inflections
Form = Semantics
Non-past -(r)u = λVv. 〈〈(λE e. e ∧Tense(e)=Npst) · v〉〉SPerfect -ta = λVv. 〈〈(λE e. e ∧Tense(e)=Perf) · v〉〉SCausative -(s)ase- = λVv. 〈〈λnom n. λdat d. (λE e. 〈〈e ∧Cause(e)=n〉〉E ) · v · 〈〈d〉〉nom〉〉VPassive -(r)are- = λVv. 〈〈λnom n. λdat d. (λE e. 〈〈e ∧ Exper(e)=n〉〉E ) · v · 〈〈d〉〉nom〉〉V
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 31 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis for yonda
yon-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
-da. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S
J(λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S) · 〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
β〈〈(λE e. e ∧Tense(e) =Perf) · (λacc a. λnom n. 〈〈n reads ′ a〉〉E )〉〉S
η
〈〈λacc a. λnom n. (λE e. e ∧Tense(e) =Perf) · 〈〈n reads ′ a〉〉E 〉〉Sβ
〈〈λacc a. λnom n. 〈〈n reads ′ a ∧ Tense(n reads ′ a) =Perf〉〉E 〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 32 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis for yonda
yon-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
-da. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S
J(λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S) · 〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
β〈〈(λE e. e ∧Tense(e) =Perf) · (λacc a. λnom n. 〈〈n reads ′ a〉〉E )〉〉S
η
〈〈λacc a. λnom n. (λE e. e ∧Tense(e) =Perf) · 〈〈n reads ′ a〉〉E 〉〉Sβ
〈〈λacc a. λnom n. 〈〈n reads ′ a ∧ Tense(n reads ′ a) =Perf〉〉E 〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 32 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis for yonda
yon-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
-da. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S
J(λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S) · 〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
β〈〈(λE e. e ∧Tense(e) =Perf) · (λacc a. λnom n. 〈〈n reads ′ a〉〉E )〉〉S
η
〈〈λacc a. λnom n. (λE e. e ∧Tense(e) =Perf) · 〈〈n reads ′ a〉〉E 〉〉Sβ
〈〈λacc a. λnom n. 〈〈n reads ′ a ∧ Tense(n reads ′ a) =Perf〉〉E 〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 32 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis for yonda
yon-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
-da. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S
J(λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S) · 〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
β〈〈(λE e. e ∧Tense(e) =Perf) · (λacc a. λnom n. 〈〈n reads ′ a〉〉E )〉〉S
η
〈〈λacc a. λnom n. (λE e. e ∧Tense(e) =Perf) · 〈〈n reads ′ a〉〉E 〉〉Sβ
〈〈λacc a. λnom n. 〈〈n reads ′ a ∧ Tense(n reads ′ a) =Perf〉〉E 〉〉S
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 32 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Semantic analysis for yonda
yon-. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
-da. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S
J(λV v. 〈〈(λE e. e ∧Tense(e) =Perf) · v〉〉S) · 〈〈λacc a. λnom n. 〈〈n reads ′ a〉〉E 〉〉V
β〈〈(λE e. e ∧Tense(e) =Perf) · (λacc a. λnom n. 〈〈n reads ′ a〉〉E )〉〉S
η
〈〈λacc a. λnom n. (λE e. e ∧Tense(e) =Perf) · 〈〈n reads ′ a〉〉E 〉〉Sβ
〈〈λacc a. λnom n. 〈〈n reads ′ a ∧ Tense(n reads ′ a) =Perf〉〉E 〉〉S
The η is a lie!
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 32 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
Conclusion
Associative λ-calculi• Justifies shorthands in category theory
Chiastic λ-calculi (namely 〈〈λ→〉〉χL)• Captures linguistic phenomena• Type reduction is CR and SN• Term reduction is WCR
Current work• Is term reduction SN?• Can we describe Γ ` e : τ more directly?• What about η?
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 33 / 34
Outline Examples and Motivation Associative λ-calculi Chiastic λ-calculi 〈〈λ→〉〉χL in action
∼fin.
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 34 / 34
Type reduction is SN Type reduction is CR
Type reduction is strongly normalizing
Type reduction is Church–Rosser
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 35 / 34
Type reduction is SN Type reduction is CR
Type reduction for 〈〈λ→〉〉χL is strongly normalizing
Definition The “length” of a type is the number of constructors
length(T ) = 1
length(σA→ τ) = 1 + length(σ) + length(τ)
length(〈〈τ〉〉A) = 1 + length(τ)
length(σ · τ) = 1 + length(σ) + length(τ)
Lemma Equivalent types have equal length.
Theorem Type reduction diminishes length; i.e.,∀τ, τ ′. τ τ ′ ⇒ length(τ) > length(τ ′)
Back
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 35 / 34
Type reduction is SN Type reduction is CR
Type reduction is strongly normalizing
Type reduction is Church–Rosser
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 36 / 34
Type reduction is SN Type reduction is CR
Type reduction for 〈〈λ→〉〉χL is Church–Rosser
Lemma Type reduction commutes with type equivalence; i.e.,τ
σ τ ′
σ′*
Theorem Type reduction is weak Church–Rosser.
Proof There are no critical pairs. Use the key lemma to resolvepotential conflicts between β and itself.
Corollary Type reduction is Church–Rosser
Proof By Newman’s Lemma.
Back
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 36 / 34
Type reduction is SN Type reduction is CR
Case 2
ρ · 〈〈σ〉〉A
τ ρ · 〈〈σ′〉〉A
ρ′ · 〈〈σ′〉〉A
β(e) (ρ · 〈〈−〉〉A)(s)
(− · 〈〈σ′〉〉A)(r)β(e′)
ρ ρ′
τA→ σ τ
A→ σ′
r
e
(τA→ −)(s)
e′
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 37 / 34
Type reduction is SN Type reduction is CR
Case 3a
ρ · 〈〈σ〉〉A
τ ρ′ · 〈〈σ〉〉A
ρ′ · 〈〈σ′〉〉A
β(e) (− · 〈〈σ〉〉A)(r)
(ρ′ · 〈〈−〉〉A)(s)β(e′)
ρ ρ′
τA→ σ τ
A→ σ′
r
e
(τA→ −)(s)
e′
wren gayle romano (Indiana University) Chiastic Lambda-Calculi 4 November 2014 38 / 34