two types of japanese scrambling in combinatory categorial grammar

87
JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion Two Types of Japanese Scrambling in Combinatory Categorial Grammar Daisuke Bekki Ochanomizu University / CREST, Japan Science and Technology Agency / National Institute of Advanced Industrial Science and Technology / National Institute of Informatics Empirical Advances in Categorial Grammar (CG2015) ESSLLI2015, Barcelona, August 11th (Tue), 2015. http://www.slideshare.net/kaleidotheater/cg2015slide 1 / 51

Upload: daisuke-bekki

Post on 18-Aug-2015

212 views

Category:

Science


1 download

TRANSCRIPT

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Two Types of Japanese Scrambling inCombinatory Categorial Grammar

Daisuke Bekki

Ochanomizu University / CREST, Japan Science and Technology Agency /National Institute of Advanced Industrial Science and Technology / National

Institute of Informatics

Empirical Advances in Categorial Grammar (CG2015)ESSLLI2015, Barcelona,

August 11th (Tue), 2015.

http://www.slideshare.net/kaleidotheater/cg2015slide

1 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What this talk is about

I Target phenomena: scrambling in Japanese

I Syntactic framework: combinatory categorial grammar (CCG:Steedman (1996), Bekki (2010))

1. A full-fledged categorial analysis of Japanese scrambling (cf.Deep and surface scrambling (Ueyama, 1998, 2003))

2. Advantage of categorial grammar (in general) over minimalism

3. Methodological difference between CCG and type-logicalgrammars

2 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Combinatory categorial grammar (CCG) (1/2)

Definition (Function application rules)

X/Y : f Y : aX : fa

>Y : a X\Y : f

X : fa<

Definition (Function composition rules)

X/Y : f Y/Z : gX/Z : λx.f(gx)

>BY \Z : g X\Y : fX\Z : λx.f(gx)

<B

Definition (Coordination rule)

S : f1 · · · CONJ : ◦ S : fm

S : f1 ◦ · · · ◦ fm〈Φ〉

3 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Combinatory categorial grammar (CCG) (2/2)

Definition (Functional crossed composition rule)

X/Y : f Y \Z : gX\Z : λx.f(gx)

>B×Y/Z : g X\Y : fX/Z : λx.f(gx)

<B×

Definition (Functional crossed substitution rules)

(X/Y )\Z : f Y \Z : gX\Z : λx.fx(gx)

>S×Y/Z : g (X\Y )/Z : f

X/Z : λx.fx(gx)<S×

Definition (Type raising rules)

X : aT /(T \X) : λf.fa

>TX : a

T \(T /X) : λf.fa<T

4 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Two types of scramblingconstructions in Japanese

5 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

First glance: Is word order in Japanese free?

Word-orders of ditransitive sentences: all 3!=6 patterns areallowed in Japanese:

(1) a. NPNOM NPDAT NPACC DTV

b. NPNOM NPACC NPDAT DTV

c. NPDAT NPNOM NPACC DTV

d. NPDAT NPACC NPNOM DTV

e. NPACC NPNOM NPDAT DTV

f. NPACC NPDAT NPNOM DTV

(when neither quantification nor binding is involved).

6 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Is Japanese ’non-configurational’?

Configurationality parameters: Hale (1980)

Japanese is classified as a nonconfigurational language, with theother free word order language like Warlpiri.

Under this view, the derivation of (1) has the flat structure asshown in (2).

(2) S → NP∗ V

7 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Japanese as a configurational language

Weak Crossover (in canonical word-order)

(3) English (Postal (1971))

a. ok Every boy loves his father.

b. * His father loves every boy.

=⇒ Subject-object assymmetry

(4) Japanese in canonical word-order (Hoji (1985))

a. ok QPNOM [...it...]DAT|ACC V

b. * [...it...]NOM QPDAT|ACC V

=⇒ Subject-object assymmetry (*flat structures)

Then, how the scrambling constructions are derived?

8 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Scrambling as movement

Harada (1977) pointed out that the scrambling in Japanese obeysthe subjacency, and thus is an instance of syntactic movement.

(5) Long-distance scrambling and complex-NP constraint

a. NPDAT|ACC NPNOM [NPNOM tx V]-that V

b. * NPDAT|ACC NPNOM [NPNOM tx V]-rel-NDAT|ACC V

1. JP scrambling is unbounded (thus Hale (1980)undergenerates) → Puzzle 0

2. JP scrambling subjects to some form of subjacency (thoughthe status of (5b) is controversial).

9 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Puzzle 1: Reconstruction effects

Reconstruction effects in long-distance scrambling:

(6) a. [... it ...]DAT|ACC NPNOM [QPNOM tx V]-that V

b. * QPDAT|ACC [...it...]NOM [NPNOM tx V]-that V

In long-distance scrambling, a dislocated NP is obligatorilyinterpreted in the trace position (Saito (1992)).

Long-distance scrambling is generated by a movement in PF?(Ueyama (1998, 2003))

10 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Puzzle 2: Absence of WCO

Weak Crossover (in canonical word-order): repeated

(7) English (Postal (1971))

a. ok Every boy loves his father.

b. * His father loves every boy.

=⇒ Subject-object assymmetry

(8) Japanese in canonical word-order (Hoji (1985))

a. ok QPNOM [...it...]DAT|ACC V

b. * [...it...]NOM QPDAT|ACC V

=⇒ Subject-object assymmetry (*flat structures)

11 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Puzzle 2: Absence of WCO

Weak Crossover (in scrambled word-order)

(9) Japanese in scrambled word-order (Hoji (1985), Ueyama(1998))

a. ok QPDAT|ACC [...it...]NOM tx V

b. ok [... it ...]DAT|ACC QPNOM tx V

=⇒ Subject-object assymmetry dissapears!=⇒ (9b) shows a reconstruction effect, but how (9a) is licenced?

12 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the puzzles

I JP scrambling seems to be an unboundedsyntactic movement.I Why reconstruction effect?

I Why ok?: QPDAT|ACC [...it...]NOM tx V

13 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the puzzles

I JP scrambling seems to be an unboundedsyntactic movement.I Why reconstruction effect?

I Why ok?: QPDAT|ACC [...it...]NOM tx V

13 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the puzzles

I JP scrambling seems to be an unboundedsyntactic movement.I Why reconstruction effect?

I Why ok?: QPDAT|ACC [...it...]NOM tx V

13 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Non-uniform analysis (Saito (1992), Ueyama (1998, 2003))

There are two types of scrambling constructions in Japanese:

Surface scrambling: Generated by the unbounded scramblingoperation (e.g. PF movement), in which a dislocatedNP is interpreted at its trace position, thusreconstruction effect obtains.

Deep scrambling: Generated by the clause-bounded scramblingoperation (e.g. operator movement), in which adislocated NP is interpreted at its landing site.

14 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Solution to Puzzle 1 and 2 in minimalism

(10) Japanese in canonical word-order (Hoji (1985))

a. ok QPNOM [...it...]DAT|ACC V

b. * [...it...]NOM QPDAT|ACC V (WCO)

(11) Japanese in scrambled word-order (Hoji (1985), Ueyama(1998))

a. ok QPDAT|ACC [...it...]NOM tx V

(Deep scrambling)

b. ok [... it ...]DAT|ACC QPNOM tx V

(Surface scrambling)

c. ok NPDAT|ACC NPNOM tx V

(Ambiguous between Surface and Deep) 15 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Puzzle 3: ‘Non-standard’-constituent coordination andmixed scrambling

Kang (1987) observed that the coordination between‘non-standard’-constituents of canonical and scrambled orders isavailable in Korean (which is also the case in Japanese).

(12) [NPACC NPNOM] and [NPNOM NPACC] V

Surface scrambling (ok):

(13) [ [... it ...]DAT|ACC QPNOM t ]x and [NPNOM NPACC] V

Deep scrambling (ok):

(14) [ QPDAT|ACC [...it...]NOM t ]x and [NPNOM NPACC] V

How can they be generated?16 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Puzzle 4: One deep scrambling per clause?

Ueyama (1998) observed that only one deep scrambling is allowedper clause.

(15) a. * QPDAT [...it...]-rel-NACC-Q [...it...]NOM tx t V

b. * QPACC [...it...]-rel-NDAT-Q [...it...]NOM tx t V

(i.e. In “NPDAT NPACC NPNOM V” order, either NPDAT or NPACC

is dislocated by surface scrambling. But why?)

17 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Not a matter of complexity

(16) a. * QPDAT [...it...]-rel-NACC-Q [...it...]NOM tx t V

b. * QPACC [...it...]-rel-NDAT-Q [...it...]NOM tx t V

c. ok QPNOM

[...it...]-rel-NDAT

-Q [...it...]ACC

tx t V

18 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of Japanese scrambling

I Ueyama (1998, 2003)’s non-uniform analysis(deep and surface scrambling) succeeds in givinga solution to Puzzle 1 and 2, but not to Puzzle3 and 4.I Puzzle 3 is fatal for minimalism.

I Two problems awaiting solution:I “Translation” of Ueyama (1998)’s analysis to categorial

grammar is not obvious: PF scrambling?I How can Puzzle 3 and 4 be solved?

19 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of Japanese scrambling

I Ueyama (1998, 2003)’s non-uniform analysis(deep and surface scrambling) succeeds in givinga solution to Puzzle 1 and 2, but not to Puzzle3 and 4.I Puzzle 3 is fatal for minimalism.

I Two problems awaiting solution:I “Translation” of Ueyama (1998)’s analysis to categorial

grammar is not obvious: PF scrambling?I How can Puzzle 3 and 4 be solved?

19 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of Japanese scrambling

I Ueyama (1998, 2003)’s non-uniform analysis(deep and surface scrambling) succeeds in givinga solution to Puzzle 1 and 2, but not to Puzzle3 and 4.I Puzzle 3 is fatal for minimalism.

I Two problems awaiting solution:I “Translation” of Ueyama (1998)’s analysis to categorial

grammar is not obvious: PF scrambling?I How can Puzzle 3 and 4 be solved?

19 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of Japanese scrambling

I Ueyama (1998, 2003)’s non-uniform analysis(deep and surface scrambling) succeeds in givinga solution to Puzzle 1 and 2, but not to Puzzle3 and 4.I Puzzle 3 is fatal for minimalism.

I Two problems awaiting solution:I “Translation” of Ueyama (1998)’s analysis to categorial

grammar is not obvious: PF scrambling?I How can Puzzle 3 and 4 be solved?

19 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of Japanese scrambling

I Ueyama (1998, 2003)’s non-uniform analysis(deep and surface scrambling) succeeds in givinga solution to Puzzle 1 and 2, but not to Puzzle3 and 4.I Puzzle 3 is fatal for minimalism.

I Two problems awaiting solution:I “Translation” of Ueyama (1998)’s analysis to categorial

grammar is not obvious: PF scrambling?I How can Puzzle 3 and 4 be solved?

19 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface scrambling in CCG

20 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface Scrambling by CB

We claim that there exists a pair of (language-specific)combinatory rules <CB and >CB, which we call permutedfunctional composition rules:

Definition (Permuted functional composition rules)

Y/Z : g X/Y : fX/Z : λx.f(g(x))

>CBX\Y : f Y \Z : gX\Z : λx.f(g(x))

<CB

We assume that each language optionally employs one or both ofthe CB rules, with Japanese (and possibly other languages withscrambling constructions such as Korean and Turkish) employingonly >CB as a rule for deriving surface scrambling.

21 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface Scrambling by CB

Definition (The reduction rule for the combinator C)

Cfxy = fyx

I ∴ CBfgx = Bgfx = f(g(x))I The CB rules are variations of the functional composition rule

B with its arguments scrambled by the C combinator: the>CB (and <CB) rule swaps the applicable order of twoadjacent functions g and f so that the function g on the left(right) applies first to the rightward (leftward) argument, towhich the function f on the right (left) applies latter.

22 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface Scrambling by CB

Definition (The reduction rule for the combinator C)

Cfxy = fyx

I ∴ CBfgx = Bgfx = f(g(x))I The CB rules are variations of the functional composition rule

B with its arguments scrambled by the C combinator: the>CB (and <CB) rule swaps the applicable order of twoadjacent functions g and f so that the function g on the left(right) applies first to the rightward (leftward) argument, towhich the function f on the right (left) applies latter.

22 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface Scrambling by CB

Definition (The reduction rule for the combinator C)

Cfxy = fyx

I ∴ CBfgx = Bgfx = f(g(x))I The CB rules are variations of the functional composition rule

B with its arguments scrambled by the C combinator: the>CB (and <CB) rule swaps the applicable order of twoadjacent functions g and f so that the function g on the left(right) applies first to the rightward (leftward) argument, towhich the function f on the right (left) applies latter.

22 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P1) Surface scrambling by CB and reconstruction effect(without movement)

Linguistically, the left function corresponds a dislocated constituentthat is yet interpreted in its original position, thus yielding areconstruction effect.

NPACC

dislocated NP

T /(T \NPo): λp.λ~x.Q2(y)(py~x)

NPNOM

T / (T \NPga)

: λp.λ~x.Q1(x)(px~x)T /(T \NPga\NPo)

: λp.λ~x.Q1(x)(Q2(y)(pyx~x))

>CBV

S\NPga\NPo

: λy.λx.V(x, y)S

: Q1(x)(Q2(y)(V(x, y)))

>

23 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P0) Long-distance scrambling by CB

... iti ...DAT

dislocated NP

T /(T \NPni)

NPNOM

T /(T \NPga)[QPi

NOM

T /(T \NPga)

T /(T \NPga\NPga)>B

T /(T \NPga\NPga\NPni)>CB

V1]S\NPga\NPni

-thatSto\S

Sto\NPga\NPni

<B V2

S\NPga\Sto

S\NPga\NPga\NPni

<B

S

24 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P0) Long-distance scrambling by CB

V1]S\NPga\NPni

-thatSto\S

Sto\NPga\NPni

<B V2

S\NPga\Sto

S\NPga\NPga\NPni

<B

25 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P0) Long-distance scrambling by CB

... iti ...DAT

dislocated NP

T /(T \NPni)

NPNOM

T /(T \NPga)[QPi

NOM

T /(T \NPga)

T /(T \NPga\NPga)>B

T /(T \NPga\NPga\NPni)>CB

26 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P0) Long-distance scrambling by CB

... iti ...DAT

dislocated NP

T /(T \NPni)

NPNOM

T /(T \NPga)[QPi

NOM

T /(T \NPga)

T /(T \NPga\NPga)>B

T /(T \NPga\NPga\NPni)>CB

V1]S\NPga\NPni

-thatSto\S

Sto\NPga\NPni

<B V2

S\NPga\Sto

S\NPga\NPga\NPni

<B

S

27 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P3) ’Non-standard’ constituent and mixed coordination byCB

NPACC

dislocated NP

T /(T \NPo)NPNOM

T /( T \NPga )

T /(T \NPga\NPo)>CB

NPNOM

T /(T \NPga)NPACC

T /(T \NPo)

T /(T \NPga\NPo)>B

T /(T \NPga\NPo)〈Φ〉 V

S\NPga\NPo

S>

28 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling in CCG

29 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling by B×

I Functional crossed composition rules >B× and <B× arenon-harmonic rules often used for extraction fromnon-right-peripheral cases in English.

I The >B× rule is also suggested as a source of (deep)scrambling construction in Steedman (2000).

Definition (Functional crossed composition rules)

X/Y : f Y \Z : gX\Z : λx.f(g(x))

>B×Y/Z : g X\Y : fX/Z : λx.f(g(x))

<B×

30 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling by B×

I Functional crossed composition rules >B× and <B× arenon-harmonic rules often used for extraction fromnon-right-peripheral cases in English.

I The >B× rule is also suggested as a source of (deep)scrambling construction in Steedman (2000).

Definition (Functional crossed composition rules)

X/Y : f Y \Z : gX\Z : λx.f(g(x))

>B×Y/Z : g X\Y : fX/Z : λx.f(g(x))

<B×

30 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling by B×

I Functional crossed composition rules >B× and <B× arenon-harmonic rules often used for extraction fromnon-right-peripheral cases in English.

I The >B× rule is also suggested as a source of (deep)scrambling construction in Steedman (2000).

Definition (Functional crossed composition rules)

X/Y : f Y \Z : gX\Z : λx.f(g(x))

>B×Y/Z : g X\Y : fX/Z : λx.f(g(x))

<B×

30 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling by B×

NPACC

dislocated NP

T /(T \NPo): λp.λ~x.Q2(y)(py~x)

NPNOM

T / (T \NPga)

: λp.λ~x.Q1(x)(px~x)

V

S\NPga \NPo

: λy.λx.V(x, y)Sterm\NPo

: λy.Q1(x)(V(x, y))

>B×

S: Q2(y)(Q1(x)(V(x, y)))

>

31 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Problem of B× as a souce of deep scrambling

I Coordinations of the form OS and OS V are not generated.

NPACC

dislocated NP

T /(T \NPo)NPNOM

T /(T \NPga)

T /(T \NPo\NPga)>B

NPNOM

T /(T \NPga)NPACC

T /(T \NPo)

T /(T \NPga\NPo)>B

32 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P2) Deep scrambling by Bu× rules

I Unary functional crossed composition rules >Bu× and <Bu

×are variations of Geach rule corresponding to B×.

I (Potentially problematic to parsing)

Definition (Unary functional crossed composition rules)

X/Y : fX\Z/(Y \Z) : λg.λx.f(g(x))

>Bu×

X\Y : fX/Z\(Y/Z) : λg.λx.f(g(x))

<Bu×

cf. Functional crossed composition rules

X/Y : f Y \Z : gX\Z : λx.f(g(x))

>B×Y/Z : g X\Y : fX/Z : λx.f(g(x))

<B×

33 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P2) Deep scrambling by Bu× rules

NPACC

dislocated NP

T /(T \NPo): λp.λ~x.Q2(y)(py~x)

NPNOM

T /(T \NPga)

: λp.λ~x.Q1(x)(px~x)

T \T ′/( T \NPga \T ′): λg.λy.λ~x.Q1(x)(gyx~x)

>Bu× V

Sterm\NPga\NPo

: λy.λx.V(x, y)Sterm\NPo

: λy.Q1(x)(V(x, y))

>

S: Q2(y)(Q1(x)(V(x, y)))

>

34 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P3) ’Non-standard’ constituent and mixed coordination byBu×

NPACC

dislocated NP

T /(T \NPo)

NPNOM

T /(T \NPga)

T \T ′/( T \NPga \T ′)>Bu

×

T /(T \NPga\NPo)>B

NPNOM

T /(T \NPga)NPACC

T /(T \NPo)

T /(T \NPga\NPo)>B

T /(T \NPga\NPo)〈Φ〉 V

Sterm\NPga\NPo

S>

35 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

(P4) Why only one deep scrambling per clause?

(16) a. * QPDAT|ACC [...it...]-rel-NACC|DAT-Q [...it...]NOM tx t V

NPNOM

T /(T \NPga)

: λp.λ~x.Q1(x)(px~x)

T \T ′/( T \NPga \T ′): λg.λy.λ~x.Q1(x)(gyx~x)

>Bu× V

Sterm\NPga\NPni\NPo

: λz.λy.λx.V(x, y, z)∗

The >Bu2× rule (or even the generalized >Bun

× rule) that licences(16) is also definable (but we claim that it is not employed inJapanese).

36 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

37 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Not completely satisfied

I Problem 1: Only scrambling constructions use CB and Bu×

I Problem 2: The Bu× rules are not safe in parsing.

I Problem 3: Why is Japanese scrambling optional? Whatsurface scrambling constructions are for?

38 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Not completely satisfied

I Problem 1: Only scrambling constructions use CB and Bu×

I Problem 2: The Bu× rules are not safe in parsing.

I Problem 3: Why is Japanese scrambling optional? Whatsurface scrambling constructions are for?

38 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Not completely satisfied

I Problem 1: Only scrambling constructions use CB and Bu×

I Problem 2: The Bu× rules are not safe in parsing.

I Problem 3: Why is Japanese scrambling optional? Whatsurface scrambling constructions are for?

38 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Not completely satisfied

I Problem 1: Only scrambling constructions use CB and Bu×

I Problem 2: The Bu× rules are not safe in parsing.

I Problem 3: Why is Japanese scrambling optional? Whatsurface scrambling constructions are for?

38 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Derivations in Hybrid TLCG

39 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling in Hybrid TLCG

Hybrid TLCG: Kubota (2015), Kubota and Levine (2015)

Definition (Deep and surface scrambling rules in Hybrid TLCG)

f ;S|(S|NP);Q

λσ.f(id) • σ(ε);S|(S|NP);Q

ds

f ;S|(S|NP);Q

λσ.f(id) • σ(id)(ε);S|(S|(S|(S|NP)));λp.p(Q)

ss

40 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Deep scrambling in Hybrid TLCG

λσ.σ(NPACC);S|(S|NPo);Q

λσ.NPACC • σ(ε);S|(S|NPo);Q

ds

NPNOM;NPga ;t

ϕ;NPo ;y

1V ;S\NPga\NPo ;v

ϕ • V ;S\NPga ;vy

\E

NPNOM • ϕ • V ;S;vyt

\E

λϕ.NPNOM • ϕ • V ;S|NPga ;λy.vyt

|I ,1

NPACC • NPNOM • ε • V ;S;Q(λy.vyt)

|E

41 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Multiple deep scrambling in a clause!

λσ.σ(NPDAT);S|(S|NPni);P

λσ.NPDAT • σ(ε);S|(S|NPni);P

ds

λσ.σ(NPACC);S|(S|NPga);Q

λσ.NPACC • σ(ε);S|(S|NPo);Q

ds

NPNOM;NPga ;t

ψ;NPni ;z

2

ϕ;NPo ;y

1V ;S\NPga\NPni\NPo ;v

ϕ • V ;S\NPga\NPni ;vy

\E

ψ • ϕ • V ;S\NPga ;vyz

\E

NPNOM • ψ • ϕ • V ;S;vyzt

\E

λϕ.NPNOM • ψ • ϕ • V ;S|NPga ;λy.vyzt

|I ,1

NPACC • NPNOM • ψ • V ;S;Q(λy.vyzt)

|E

λψ.NPACC • NPNOM • ψ • V ;S|NPo ;λz.Q(λy.vyzt)

|I

NPDAT • NPACC • NPNOM • V ;S;P(λz.Q(λy.vyzt))

|E

42 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Surface scrambling in Hybrid TLCG

λσ.σ(NPACC);S|(S|NPo);Q2

λσ.NPACC • σ(id)(ε);S|(S|(S|(S|NPo)));λp.p(Q2)

ss

λσ.σ(QPNOM);S|(S|NPga);Q1

χ;S|(S|NPo);f

3

ψ;NPga ;x

2

ϕ;NPo ;y

1V ;S\NPga\NPo ;v

ϕ • V ;S\NPga ;vy

\E

ψ • ϕ • V ;S;vyx

\E

λϕ.ϕ • ϕ • V ;S|NPo ;λy.vyx

|I ,1

χ(λϕ.ψ • ϕ • V );S;f(λy.vyx)

|E

λψ.χ(λϕ.ψ • ϕ • V );S|NPga ;λx.f(λy.vyx)

|I ,2

χ(λϕ.QPNOM • ϕ • V );S;Q1(λx.(f(λy.vyx)))

|E

λχ.χ(λϕ.QPNOM • ϕ • V );S|(S|(S|NPo));λf.Q1(λx.(f(λy.vyx)))

|I ,3

NPACC • NPNOM • V ;S;Q1(λx.Q2(λy.vyx))

|E

43 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Discussion

44 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I In CCG, the rule Bu× and the rule Bu2

× can be distinguished,which may not be the case in TLGs.

I Some quantifiers allow inverse scope reading, while others arenot (at least in Japanese).

45 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I In CCG, the rule Bu× and the rule Bu2

× can be distinguished,which may not be the case in TLGs.

I Some quantifiers allow inverse scope reading, while others arenot (at least in Japanese).

45 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I In CCG, the rule Bu× and the rule Bu2

× can be distinguished,which may not be the case in TLGs.

I Some quantifiers allow inverse scope reading, while others arenot (at least in Japanese).

45 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I Extractions in Lambek/ACG/Hybrid TLCG is the/\-introduction rules.

I CCG is a substructural/subdirectional combinatory logic,where the (weak version of) /\-introduction rules hold as apartial deduction theorem: Buszkowski (2010), Ozaki andBekki (2012)

I The fact that the CCG axioms can be derived as theorems inTCG DOES NOT mean that TCG is simpler than CCG: (aweak-version of) /\-introduction rules can be proven as atheorem in combinatory logic.

46 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I Extractions in Lambek/ACG/Hybrid TLCG is the/\-introduction rules.

I CCG is a substructural/subdirectional combinatory logic,where the (weak version of) /\-introduction rules hold as apartial deduction theorem: Buszkowski (2010), Ozaki andBekki (2012)

I The fact that the CCG axioms can be derived as theorems inTCG DOES NOT mean that TCG is simpler than CCG: (aweak-version of) /\-introduction rules can be proven as atheorem in combinatory logic.

46 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I Extractions in Lambek/ACG/Hybrid TLCG is the/\-introduction rules.

I CCG is a substructural/subdirectional combinatory logic,where the (weak version of) /\-introduction rules hold as apartial deduction theorem: Buszkowski (2010), Ozaki andBekki (2012)

I The fact that the CCG axioms can be derived as theorems inTCG DOES NOT mean that TCG is simpler than CCG: (aweak-version of) /\-introduction rules can be proven as atheorem in combinatory logic.

46 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

CCG (family) vs. TCG (family)

I Extractions in Lambek/ACG/Hybrid TLCG is the/\-introduction rules.

I CCG is a substructural/subdirectional combinatory logic,where the (weak version of) /\-introduction rules hold as apartial deduction theorem: Buszkowski (2010), Ozaki andBekki (2012)

I The fact that the CCG axioms can be derived as theorems inTCG DOES NOT mean that TCG is simpler than CCG: (aweak-version of) /\-introduction rules can be proven as atheorem in combinatory logic.

46 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

What JP scrambling means for Categorial Grammars?

I Categorial grammars have succeeded in explaining both:I Unbounded A’-movements by (/I) or (>B) rules (in other

words, by the deduction theorem)I Clause-bounded A-movements, such as in control structures.

I The status of JP scrambling has been an important topic ingenerative grammar: can it be reduced to either A’-movementor A-movement, or is it another type of movement?

I General question: can some version of CG analyse JPscrambling? How are the surface and deep scramblinggenerated therein?

47 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Summary of the CCG analysis: repeated

I CB rules for surface scrambling :

P0 Long-distance scramblingP1 Reconstruction effectP3 ’Non-standard’-constituent and mixed

coordination

I Bu× rules for deep scrambling :

P2 Absence of WCOP3 ’Non-standard’-constituent and mixed

coordinationP4 The ’one-per-clause’ constraint

48 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Reference I

Bekki, D. (2010) Nihongo-Bunpoo-no Keisiki-Riron -Katuyootaikei, Toogohantyuu, Imigoosei - (trans. ’FormalJapanese Grammar: the conjugation system, categorial syntax,and compositional semantics’). Tokyo, Kuroshio Publisher.

Buszkowski, W. (2010) “Categorial Grammars and SubstructuralLogics”.

Hale, K. (1980) “Remarks on Japanese phrase structure:comments on the papers on Japanese syntax”, MIT WorkingPapers in Linguistics 2, pp.185–203.

Harada, S.-I. (1977) “Nihongo-ni ‘Henkei’-wa Hituyoo-da(‘Transformation’ is Necessary for Japanese)”, Gengo 6(11-12).

49 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Reference II

Hoji, H. (1985) “Logical Form Constraints and ConfigurationalStructures in Japanese”, Doctoral dissertation, University ofWashington.

Kang, B.-m. (1987) “Functional Inheritance, Anaphora, andSemantic Interpretation in a Generalized Categorial Grammar”,Ph.d. thesis, Brown University.

Ozaki, H. and D. Bekki. (2012) “Extractability as DeductionTheorem in Subdirectional Combinatory Logic”, In theProceedings of Logical Aspect of Computational Linguistics(LACL2012). Nantes, France.

Postal, P. (1971) Cross-over Phenomena. New York, Holt,Reinhart and Winston.

Saito, M. (1992) “Long Distance Scrambling in Japanese”, Journalof East Asian Linguistics 1(1), pp.69–118.

50 / 51

JP Scrambling Surface scrambling Deep scrambling Hybrid TLCG Discussion

Reference III

Steedman, M. J. (1996) Surface Structure and Interpretation.Cambridge, The MIT Press.

Steedman, M. J. (2000) The Syntactic Process (Language,Speech, and Communication). Cambridge, The MIT Press.

Ueyama, A. (1998) “Two Types of Dependency”, Doctoraldissertation, University of Southern California. distributed byGSIL publications.

Ueyama, A. (2003) “Two Types of Scrambling Constructions inJapanese”, In: A. Barss (ed.): Anaphora: A Reference Guide.Cambridge, Blackwell.

51 / 51