An algebraic characterization of unary 2-way transducers Christian - - PowerPoint PPT Presentation

an algebraic characterization of unary 2 way transducers
SMART_READER_LITE
LIVE PREVIEW

An algebraic characterization of unary 2-way transducers Christian - - PowerPoint PPT Presentation

An algebraic characterization of unary 2-way transducers Christian Choffrut 1 and Bruno Guillon 1,2 1 LIAFA - Universit Paris-Diderot, Paris 7 2 Dipartimento di Informatica - Universit degli studi di Milano Septembre 17, 2014 ICTCS - Perugia


slide-1
SLIDE 1

An algebraic characterization of unary 2-way transducers

Christian Choffrut1 and Bruno Guillon1,2

1LIAFA - Université Paris-Diderot, Paris 7 2Dipartimento di Informatica - Università degli studi di Milano

Septembre 17, 2014 ICTCS - Perugia - 2014 work published in MFCS 2014

1 / 15

slide-2
SLIDE 2

2-way automaton over Σ

⊲ t h e i n p u t w o r d ⊳

Automaton

READ

← →

left endmarker right endmarker WRITE

→ A

  • Q, q-, F, δ
  • transition set: ⊂ Q × Σ⊲,⊳ × {−1, 0, 1} × Q

2 / 15

slide-3
SLIDE 3

2-way automaton over Σ

⊲ t h e i n p u t w o r d ⊳

Automaton

READ

← →

left endmarker right endmarker WRITE

→ A

  • Q, q-, F, δ
  • transition set: ⊂ Q × Σ⊲,⊳ × {−1, 0, 1} × Q

2 / 15

slide-4
SLIDE 4

2-way transducer over Σ, Γ

⊲ t h e i n p u t w o r d ⊳

Automaton

READ

← →

left endmarker right endmarker t h e

  • u

t p u t WRITE

→ (A, φ)

  • Q, q-, F, δ
  • production function: δ → Rat(Γ∗)

2 / 15

slide-5
SLIDE 5

A simple example: SQUARE = {(w, ww) | w ∈ Σ∗}

⊲ a b a c c a b ⊳

3 / 15

slide-6
SLIDE 6

A simple example: SQUARE = {(w, ww) | w ∈ Σ∗}

⊲ a b a c c a b ⊳ a b a c c a b ◮ copy the input word

3 / 15

slide-7
SLIDE 7

A simple example: SQUARE = {(w, ww) | w ∈ Σ∗}

⊲ a b a c c a b ⊳ a b a c c a b ◮ copy the input word ◮ rewind the input tape

3 / 15

slide-8
SLIDE 8

A simple example: SQUARE = {(w, ww) | w ∈ Σ∗}

⊲ a b a c c a b ⊳ a b a c c a b a b a c c a b ◮ copy the input word ◮ rewind the input tape ◮ append a copy of the input word

3 / 15

slide-9
SLIDE 9

A simple example: SQUARE = {(w, ww) | w ∈ Σ∗}

⊲ a b a c c a b ⊳ a b a c c a b a b a c c a b ◮ copy the input word ◮ rewind the input tape ◮ append a copy of the input word

3 / 15

slide-10
SLIDE 10

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳

4 / 15

slide-11
SLIDE 11

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a

copy the input word

4 / 15

slide-12
SLIDE 12

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a

copy the input word rewind the input tape

4 / 15

slide-13
SLIDE 13

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a a a a a

copy the input word rewind the input tape

4 / 15

slide-14
SLIDE 14

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a a a a a

copy the input word rewind the input tape

4 / 15

slide-15
SLIDE 15

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a a a a a a a a a

copy the input word rewind the input tape

4 / 15

slide-16
SLIDE 16

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a a a a a a a a a

copy the input word rewind the input tape accept and halt with nondeterminism

4 / 15

slide-17
SLIDE 17

Another example: UnaryMult =

  • (an, akn) | k, n ∈ N

a a a a ⊳ a a a a a a a a a a a a

copy the input word rewind the input tape accept and halt with nondeterminism

4 / 15

slide-18
SLIDE 18

Rational operations

◮ Union

R1 ∪ R2

◮ Componentwise concatenation

R1 · R2 = {(u1u2, v1v2) | (u1, v1) ∈ R1 and (u2, v2) ∈ R2}

◮ Kleene star

R∗ = {(u1u2 · · · uk, v1v2 · · · vk) | ∀i (ui, vi) ∈ R}

5 / 15

slide-19
SLIDE 19

Rational operations

◮ Union

R1 ∪ R2

◮ Componentwise concatenation

R1 · R2 = {(u1u2, v1v2) | (u1, v1) ∈ R1 and (u2, v2) ∈ R2}

◮ Kleene star

R∗ = {(u1u2 · · · uk, v1v2 · · · vk) | ∀i (ui, vi) ∈ R}

Definition (Rat(Σ∗ × Γ∗))

The class of rational relations is the smallest class:

◮ that contains finite relations ◮ and which is closed under rational operations

5 / 15

slide-20
SLIDE 20

Rational operations

◮ Union

R1 ∪ R2

◮ Componentwise concatenation

R1 · R2 = {(u1u2, v1v2) | (u1, v1) ∈ R1 and (u2, v2) ∈ R2}

◮ Kleene star

R∗ = {(u1u2 · · · uk, v1v2 · · · vk) | ∀i (ui, vi) ∈ R}

Definition (Rat(Σ∗ × Γ∗))

The class of rational relations is the smallest class:

◮ that contains finite relations ◮ and which is closed under rational operations

Theorem (Elgot, Mezei - 1965)

1-way transducers = = the class of rational relations.

5 / 15

slide-21
SLIDE 21

Hadamard operations

◮ H-product

R1 H R2 = {(u, v1v2) | (u, v1) ∈ R1 and (u, v2) ∈ R2}

6 / 15

slide-22
SLIDE 22

Hadamard operations

◮ H-product

R1 H R2 = {(u, v1v2) | (u, v1) ∈ R1 and (u, v2) ∈ R2}

Example: SQUARE = {(w, ww) | w ∈ Σ∗} = Identity H Identity

⊲ a b a c c a b ⊳ a b a c c a b a b a c c a b

◮ copy the input word ◮ rewind the input tape ◮ append a copy of the input word

6 / 15

slide-23
SLIDE 23

Hadamard operations

◮ H-product

R1 H R2 = {(u, v1v2) | (u, v1) ∈ R1 and (u, v2) ∈ R2}

◮ H-star

RH⋆ = {(u, v1v2 · · · vk) | ∀i (u, vi) ∈ R}

6 / 15

slide-24
SLIDE 24

Hadamard operations

◮ H-product

R1 H R2 = {(u, v1v2) | (u, v1) ∈ R1 and (u, v2) ∈ R2}

◮ H-star

RH⋆ = {(u, v1v2 · · · vk) | ∀i (u, vi) ∈ R}

Example: UnaryMult =

  • (an, akn) | k, n ∈ N
  • = Identity H⋆

⊲ a a a a ⊳ a a a a a a a a a a a a

copy the input word rewind the input tape accept and halt with nondeterminism

6 / 15

slide-25
SLIDE 25

H-Rat relations

Definition

A relation R is in H-Rat(Σ∗ × Γ∗) if R =

  • 0≤i≤n

Ai H BH⋆

i

where for each i, Ai and Bi are rational relations.

7 / 15

slide-26
SLIDE 26

Main result

When Σ = {a} and Γ = {a}:

Theorem (Elgot, Mezei - 1965)

1-way transducers = = the class of rational relations .

8 / 15

slide-27
SLIDE 27

Main result

When Σ = {a} and Γ = {a}:

Theorem (Elgot, Mezei - 1965)

1-way transducers = = the class of rational relations . T h i s t a l k 2

  • w

a y t r a n s d u c e r s H

  • R

a t r e l a t i

  • n

s

8 / 15

slide-28
SLIDE 28

Main result

When Σ = {a} and Γ = {a}:

Theorem (Elgot, Mezei - 1965)

1-way transducers = = the class of rational relations . T h i s t a l k 2

  • w

a y t r a n s d u c e r s H

  • R

a t r e l a t i

  • n

s

Proof

◮ ⊇: easy ◮ ⊆: difficult part

8 / 15

slide-29
SLIDE 29

Known results

◮ 2-way functional =

= MSO definable functions [Engelfriet, Hoogeboom - 2001]

9 / 15

slide-30
SLIDE 30

Known results

◮ 2-way functional =

= MSO definable functions [Engelfriet, Hoogeboom - 2001]

◮ 2-way general incomparable MSO definable relations

[Engelfriet, Hoogeboom - 2001]

9 / 15

slide-31
SLIDE 31

Known results

◮ 2-way functional =

= MSO definable functions [Engelfriet, Hoogeboom - 2001]

◮ 2-way general incomparable MSO definable relations

[Engelfriet, Hoogeboom - 2001]

◮ 1-way simulation of 2-way functional transducer:

decidable and constructible [Filiot et al. - 2013]

9 / 15

slide-32
SLIDE 32

Known results

◮ 2-way functional =

= MSO definable functions [Engelfriet, Hoogeboom - 2001]

◮ 2-way general incomparable MSO definable relations

[Engelfriet, Hoogeboom - 2001]

◮ 1-way simulation of 2-way functional transducer:

decidable and constructible [Filiot et al. - 2013] When Γ = {a}:

◮ 2-way unambiguous −

→ 1-way [Anselmo - 1990]

9 / 15

slide-33
SLIDE 33

Known results

◮ 2-way functional =

= MSO definable functions [Engelfriet, Hoogeboom - 2001]

◮ 2-way general incomparable MSO definable relations

[Engelfriet, Hoogeboom - 2001]

◮ 1-way simulation of 2-way functional transducer:

decidable and constructible [Filiot et al. - 2013] When Γ = {a}:

◮ 2-way unambiguous −

→ 1-way [Anselmo - 1990]

◮ 2-way unambiguous =

= 2-way deterministic [Carnino, Lombardy - 2014]

9 / 15

slide-34
SLIDE 34

From H-Rat to 2-way transducers (unary case)

Property

The family of relations accepted by 2-way transducers is closed under ∪ ,

H and H⋆ .

10 / 15

slide-35
SLIDE 35

From H-Rat to 2-way transducers (unary case)

Property

The family of relations accepted by 2-way transducers is closed under ∪ ,

H and H⋆ .

Proof.

◮ R1 ∪ R2:

◮ simulate T1 or T2 10 / 15

slide-36
SLIDE 36

From H-Rat to 2-way transducers (unary case)

Property

The family of relations accepted by 2-way transducers is closed under ∪ ,

H and H⋆ .

Proof.

◮ R1 ∪ R2:

◮ simulate T1 or T2

◮ R1 H R2:

◮ simulate T1 ◮ rewind the input tape ◮ simulate T2 10 / 15

slide-37
SLIDE 37

From H-Rat to 2-way transducers (unary case)

Property

The family of relations accepted by 2-way transducers is closed under ∪ ,

H and H⋆ .

Proof.

◮ R1 ∪ R2:

◮ simulate T1 or T2

◮ R1 H R2:

◮ simulate T1 ◮ rewind the input tape ◮ simulate T2

◮ RH⋆:

◮ repeat an arbitrary

number of times:

◮ simulate T ◮ rewind the input tape ◮ reach the right endmarker

and accept

10 / 15

slide-38
SLIDE 38

From H-Rat to 2-way transducers (unary case)

Property

The family of relations accepted by 2-way transducers is closed under ∪ ,

H and H⋆ .

Corollary

H-Rat ⊆ accepted by 2-way transducers

 

0≤i≤n

Ai H BH⋆

i

 

10 / 15

slide-39
SLIDE 39

From 2-way transducers to H-Rat (unary case)

A first ingredient, a preliminary result:

Lemma

With arbitrary Σ and Γ = {a}: H-Rat is closed under ∪ ,

H and H⋆ .

Proof.

Tedious formal computations. . .

11 / 15

slide-40
SLIDE 40

From 2-way transducers to H-Rat (unary case)

We fix a transducer T .

◮ Consider border to border run segments;

u

⊳ q1 q2 q3 q4

12 / 15

slide-41
SLIDE 41

From 2-way transducers to H-Rat (unary case)

We fix a transducer T .

◮ Consider border to border run segments;

u

⊳ q1 q2 q3 q4 R1 = {(u, v1)} R2 = {(u, v2)} R3 = {(u, v3)}

12 / 15

slide-42
SLIDE 42

From 2-way transducers to H-Rat (unary case)

We fix a transducer T .

◮ Consider border to border run segments; ◮ Compose border to border segments;

u

⊳ q1 q2 q3 q4 R1 = {(u, v1)} R2 = {(u, v2)} R3 = {(u, v3)}

12 / 15

slide-43
SLIDE 43

From 2-way transducers to H-Rat (unary case)

We fix a transducer T .

◮ Consider border to border run segments; ◮ Compose border to border segments;

u

⊳ q1 q2 q3 q4 R1 = {(u, v1)} R2 = {(u, v2)} R3 = {(u, v3)} R1 H R2 H R3 = {(u, v1v2v3)}

12 / 15

slide-44
SLIDE 44

From 2-way transducers to H-Rat (unary case)

We fix a transducer T .

◮ Consider border to border run segments; ◮ Compose border to border segments; ◮ Conclude using the closure properties of H-Rat.

u

⊳ q1 q2 q3 q4 R1 = {(u, v1)} R2 = {(u, v2)} R3 = {(u, v3)} R1 H R2 H R3 = {(u, v1v2v3)}

12 / 15

slide-45
SLIDE 45

From 2-way transducers to H-Rat (unary case)

q1 q2 ⊲

u

⊳ define a relation R bi , bj

13 / 15

slide-46
SLIDE 46

From 2-way transducers to H-Rat (unary case)

q1 q2 ⊲

u

⊳ define a relation R bi , bj

Q × {⊲, ⊳}

∈ ∈

13 / 15

slide-47
SLIDE 47

From 2-way transducers to H-Rat (unary case)

q1 q2 ⊲

u

⊳ define a relation R bi , bj

Q × {⊲, ⊳}

∈ ∈ R0,0 R0,1 · · · R0,2 R1,0 R1,1 · · · R1,2 · · · Ri,j · · · Rk,0 Rk,1 · · · Rk,k

                         

HIT =

2 |Q| 2 |Q|

13 / 15

slide-48
SLIDE 48

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ .

14 / 15

slide-49
SLIDE 49

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ).

14 / 15

slide-50
SLIDE 50

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ). ⊲ a a a a a a a a a a ⊳

q q

  • utput

14 / 15

slide-51
SLIDE 51

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ). ⊲ a a a a a a a a a a ⊳ q rational output

14 / 15

slide-52
SLIDE 52

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ). By closure property:

Corollary

Each entry of HIT H⋆ is in H-Rat.

14 / 15

slide-53
SLIDE 53

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ). By closure property:

Corollary

Each entry of HIT H⋆ is in H-Rat.

Remark

The relation accepted by T is a union of entries of HIT H⋆.

14 / 15

slide-54
SLIDE 54

From 2-way transducers to H-Rat (unary case)

Second ingredient: The behavior of T is given by the matrix HIT H⋆ . Third ingredient:

Lemma

Each entry Rb1,b2 of the matrix HIT is rational ( constructible ). By closure property:

Corollary

Each entry of HIT H⋆ is in H-Rat.

Remark

The relation accepted by T is a union of entries of HIT H⋆.

Corollary

accepted by 2-way transducers ⊆ H-Rat

14 / 15

slide-55
SLIDE 55

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations.

15 / 15

slide-56
SLIDE 56

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations. From our construction follows:

◮ 2-way transducers can be made sweeping .

15 / 15

slide-57
SLIDE 57

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations. From our construction follows:

◮ 2-way transducers can be made sweeping .

With only Γ = {a}:

◮ 2-way

    

deterministic unambiguous functional

    

accept rational relations.

15 / 15

slide-58
SLIDE 58

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations. From our construction follows:

◮ 2-way transducers can be made sweeping .

With only Γ = {a}:

◮ 2-way

    

deterministic unambiguous functional

    

accept rational relations.

◮ 2-way transducers are uniformizable by 1-way transducers.

15 / 15

slide-59
SLIDE 59

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations. From our construction follows:

◮ 2-way transducers can be made sweeping .

With only Γ = {a}:

◮ 2-way

    

deterministic unambiguous functional

    

accept rational relations.

◮ 2-way transducers are uniformizable by 1-way transducers.

Every thing is constructible .

15 / 15

slide-60
SLIDE 60

Conclusion

Theorem

When Γ = {a} and Σ = {a}: 2-way transducers accept exactly the H-Rat relations. From our construction follows:

◮ 2-way transducers can be made sweeping .

With only Γ = {a}:

◮ 2-way

    

deterministic unambiguous functional

    

accept rational relations.

◮ 2-way transducers are uniformizable by 1-way transducers.

Every thing is constructible . Thank you for your attention. 15 / 15