Multiple Multiple Spellout - meaghanfowlie.commeaghanfowlie.com/documents/fowlie_mms_handout.pdf · b. bayi the-nom yarangu man-gen dugumbiru woman-erg buran see-pres/pst wangal ...

Multiple Multiple Spellout∗

Meaghan Fowlie

Minimalism seminar, May 21 2014

1 Overview

Take-home: There are two separate spellout operations: linearise and transfer

• linearise assigns linear order to the terminal nodes of a constituent

– based on Fox and Pesetsky (2005)’s operation Linearise in combination withNunes and Uriagereka (2000)’s revised LCA

• transfer 1 Removes constituent from the derivation and sends it to PF and LF

– based on Nunes and Uriagereka (2000)’s Spellout

• 3 puzzles:

– CED (no movement out of non-objects) 6= PIC (only movement out of phaseedges)

– CED exceptions

– Free word order

∗This work is based on my undergraduate thesis written at McGill University in 2007. I would like tothank the many students and professors there who took time to advise me on this project, in particularJon Nissenbaum, Tobin Skinner, Raphael Mercado, and especially my undergraduate thesis supervisor LisadeMena Travis, who was also responsible for pointing out the possibilities this approach presents for freeword order. I extend many thanks as well to the Principles of Linearisation workshop at GLOW 31 and theUCLA syntax-semantics seminar for advice and questions and support, as well as to Ed Stabler for openingmy eyes to the importance of computational considerations.

1In the thesis and paper, this is called atomise.

1

2 3 Puzzles

2.1 CED and PIC Effects

Phase Impenetrability Condition (PIC) If H is the head of a phase, HP isnot accessible to operations outside HP; only H and its specifier are accessible(Chomsky, 2001).

Condition on Extraction Domains (CED) (Huang, 1982): Verbal comple-ments can be extracted from but adjuncts and specifiers cannot.

• Spellout solution to impenetrabiity: once something is spelled out, it is no longeraccessible to syntax.

• Usual PIC solution: phase head spells out its sister once phase is complete

• This approach to phases does not help with CED effects:

– Embedded CPs – classic phases – are often object CPs. The CED says objectsare easier, not harder, to escape from.

– CED means all extraction out of specs & adjuncts is impossible, not just non-edgeextraction.

vP

Phase?Subj

A B G

V’

V PhaseCP

escape hatchC TP

D E F

2.2 Exceptions to CED

Sometimes specifiers and adjuncts can be extracted from. For example, in some free wordorder languages like Dyirbal, noun markers, adjectives, and nouns can be separated.

Hindi: posessors extract from non-direct objects

(1) Kiskiiwhose

tumyou

soctethink

ho kithat

[SUB

[SUB

twh

twh

kitaab]book]

coriistolen

ho gayiiwas

‘Whose book do you think was stolen?’

2

OV languages should maybe show CED effects for D.O.s: Under Antisymmetry,direct objects in OV languages start as rightward complements of V, and then move to aposition to the left of V.

(2)DO

blahblah

VP

V tDO

All movement is necessarily to a specifier or adjunct position, so we expect direct objects inOV languages to show CED effects. They do not. Dutch:2:

(3) [Van[Of

wie]whom]

hebhave

jeyou

[DO

[DO

eena

fotopicture

t]t]

gezien?seen?

‘Who did you see a picture of?’

2.3 Free Word Order

Free Word Order: Optional and semantically vacuous word-order variation

Some “free word order” is clearly A-movement as it has binding effects (Saito, 1992).However, some free word order is very likely not movement at all, e.g.: Dyirbal (Dixon,1972).3

(4) a. bayithe-nom

wangalboomerang-nom

bangulthe-gen

yaranguman-gen

bulganubig-gen

banggunthe-erg

dugumbiruwoman-erg

buransee-pres/pst‘The woman saw the big man’s boomerang’

b. bayithe-nom

yaranguman-gen

dugumbiruwoman-erg

buransee-pres/pst

wangalboomerang-nom

banggunthe-erg

bangulthe-gen

bulganubig-gen

All word-orders are claimed to be grammatical.

A well-known linguist took exception to this, categorically denied that freedom ofword-order of this magnitude was possible in any language, and accused the writerof exaggerating. [(4-b)] was put to informants at the next opportunity, and theycastigated the writer for asking a trivial and unnecessary question – “you know that’salright!” (Dixon 1972 p. 107-8)

2Dutch data from Floris van Vugt p.c. 20103r = semi-retroflex continuant, d = lamino-palatal/alveolar. the in the glosses are noun-markers. Ab-

breviations are as follows: nom = nominative case gen = genitive case erg = ergative case pres/past =present or past tense

3

Optional movement?

• 8 words in the sentence, so 8! = 40 320 possible word-orders.

• → 40 320 optional movement combinations

• If Move is feature-driven, may need 40 320 separate numerations

• It’s semantically vacuous movement, but we expect same meaning ↔ same structure

• A better solution would derive all orders from one structure.

3 Background

3.1 Linear Correspondence Axiom (LCA)

Linear Correspondence Axiom (Kayne, 1994): For any pair of non-terminalnodes 〈X, Y 〉, if X asymmetrically c-commands Y then each terminal node dom-inated by X precedes each terminal node dominated by Y . Moreover, the set ofall such correspondences constitutes a total ordering on the terminal nodes.

Kayne assumes irreflexive dominance and that the terminal nodes (e.g. lexical items)project up to a syntactic head without branching. At least one of these two assumptions isnecessary to derive a total ordering on the terminal nodes.

3.2 Fox and Pesetsky

• Fox and Pesetsky (2003, 2005) propose that Spellout fixes the relative order of thelexical items in a Spelled-out domain.

• At the end of the construction of each Spellout Domain (SD) Di, a order is defined onthe elements of Di by some linearisation algorithm

• By order I mean an antisymmetric, transitive binary relation

• The final result must be a total order (essentially, defines a unique list of all theelements with no contradictions)

• Note this only makes sense if LIs are indexed, otherwise repetitions will look likecontradictions

• Note too that this doesn’t work with the copy theory of movement, as Edshowed us last class.

• After Linearise, constituents can still move out, but they cannot change their order.

4

This elegantly derives successive cyclicity effects such as Quantifier Movement in Scan-dinavian langauges and the Holmberg Generalisation, including aspects that are rarely ac-counted for. Fox and Pesetsky also derive other types of successive cyclicity. Consider, forexample, English object wh-movement. The object wh-word in its theta-position is after theverb and subject. For example:

(5) What did he read?

(6) [vP what he read what]]

whathe

read

(7) [CP what did [vP what he read what]]

what he read

did

Versus:

(8) [vP he read what]]

heread

what

(9) [CP what did [vP he read what]]

5

he

read

what

did

3.3 Nunes and Uriagereka

• assume bare phrase structure

• Remove dominance from LCA

Linear Correspondence Axiom (Nunes and Uriagereka 2000): A termi-nal node α precedes a terminal node β iff α asymmetrically c-commands β.

• No total ordering with:

– Complex specifier → Fix with spellout

– simple sisters → who knows?

VP

DP

the NP

man PP

in DP

the NP

yellow hat

V’

has DP

a monkey

Figure 1: Lexical Items of a complex specifier have no c-command relationship with LexicalItems of V’

Spellout (N&U):

6

1. Linearise with N&U LCA

2. Remove phase from derivation

3. Leave copy of root label

“Label”:

• Includes all info needed for syntax to manipulate phase as a unit

– So feature calculus must project unchecked features up structure, included in label

– {DP, wh}

{which, -N, DP, wh} {monkey, N}

• Acts as a bookmark for putting the tree back together again at LF, PF.

– Must be indexed because some labels-as-features could be identical

• Is now a terminal node, so it’s linearisable with its nieces

VP

〈The man in the yellow hat〉 V’

has DP

a monkey

• Specifiers and adjuncts must be spelled out before they are Merged, or elsethey can’t be linearied with nieces

• Spellout removes constituent from derivation, rendering it an island

• Derives subject, adjunct islands: CED effects

4 Proposal

• F&P: spelled-out domains accessible, just linearised

• N&U: spelled-out domains inaccessible

• But! They are looking at different domains

– F&P: spine

7

– N&U: satellites

Figure 2: Spine and Satellites

Solution: Both are correct but incomplete. Two separate operations apply: lineariseand transfer .

• transfer : Transfer everything to LF and PF. Leave the indexed label.

– Delayed by a phase4

• linearise : F&P’s Linearise, with ordering algorithm specified as N&U’sLCA

– occurs immediately

4.1 Interaction of linearise and transfer

• transfer satellites before Merging them so that linearise can apply

• linearise phases before transfer so that they have an order

• linearise without transfer : successive cyclicity etc. “Phase edge” effects areapproximate and epiphenomenal

• transfer without linearise : free word order4or maybe just until the full phrase of the phasehead is complete?

8

4.2 Derivational Workspaces:

• Build constituents in parallel in separate workspaces

• There’s a “main” workspace where the spine is being built

• When work in a workspace is complete, structure may be spelled out

• I think it comes down to: if you’re merging two complex structures, the one beingselected or the adjunct can be spelled out

4.3 Domains

This is given for the sake of concreteness, but the same idea of a split spelloutoperation could apply to any number of specific proposals about spellout domains

• Suppose the spine is naturally divided into domains:

– VP domain (where the arguments and manner adverbs go),

– TP domain (where case, agreement, tense etc and the sentential adverbs live),

– CP (left periphery, with Force and Focus and whatnot)

• Proposal: each domain is a phase.

• You know a phase is complete when a head belonging to a new domain is merged

4.4 Parameters

The basic model is this: each Spellout Domain D is specified for whether each operationoccurs, thus:

D[±L±T]

+L –L+A [+L+T] [–L+T]–A [+L–T] [–L–T]

Table 1: Four possibilities for each Spellout Domain

1. [+L+T]: Linear order is strict; constituent can move and still be linearised with newnieces. e.g. English Subjects

9

2. [+L–T]: Linear order is strict; constituent can’t move and still be lineasised with itsnew nieces.5 E.g. Scandinavian VP

3. [–L–T]: Linear order is not (yet) determined; can’t move and still be linearised withits new nieces. i.e. acts like non-phase

4. [–L+T]: No linear order determined; could move as a unit; order within constituent isfree. e.g. Dyirbal CPs

Languages are also parameterised for whether each spellout operation applies to a com-pleted workspace: W[±L±T]

These parameters combine to make Spellout Types. For example, supposing there arethree SDs A, B, C, we have:

W[±L±T] A[±L±T] B[±L±T ] C[±L±T]

4.4.1 The Procedure for Spellout

When a Spellout Domain D is complete:

1. Check for SDs embedded in D

2. For each embedded SD E, if E is [+T], transfer E

3. If D is [+L] linearise D

When you are finished with a workspace:

1. Check for SDs embedded in D

2. For each embedded SD E, if E is [+T], transfer E

3. If language is is W[+L], linearise D

4. If the language is W[+T], transfer D

When a sentence is finished:

1. If language is C[+L], linearise the finished sentence

2. Always transfer the finished sentence

5I just realised I had these descriptions wrong, so maybe this new one isn’t even true. Can v/VPs movearound in Scandinavian languages with Homlsber’s generalisation?

10

4.4.2 Free word order: Tagalog

(10) Nagbigaygave

[ng-libro[gen-book

sa-babaedat-woman

ang-lalaki]nom-man]

‘The man gave the woman a book’Nagbigay [ng-libro ang-lalaki sa-babae]Nagbigay [sa-babae ng-libro ang-lalaki]Nagbigay [sa-babae ang-lalaki ng-libro ]Nagbigay [ang-lalaki sa-babae ng-libro]Nagbigay [ang-lalaki ng-libro sa-babae]

(11) a. nggen

libro-ngbook-lk

malakibig

‘the big book’b. ng

genmalaki-ngbig-lk

librobook

(12) nakitasaw

sadat

babae-ngwoman-lk

masayabig

angnom

lalakiman

‘The man saw the happy woman’

4.5 An illustration

The man saw the happy woman/Nakita sa babae masaya ang lalaki

Step 1. Build the happy woman

5D

1the -N, D

4N

2happy A

3woman N

5D

1sa -N, D

4N

2masaya A

3babae N

Step 2. Merge of D signals end of N phase.

English: N[+L] Tagalog: N[–L]

2 → 3

Step 3. Meanwhile, build the man in a separate workspace

11

3aD,-nom

1athe -N,D,-nom

2aman N

3aD

1aang -N,D

2alalaki N

Step 4. Workspace finished with.

(a) Look inside for SDs.

English: N[–T] Tagalog: N[+T]

2a = lalaki

(b) English W[+L] Tagalog W[+L]

2 → 3

1a→ 2a 1a → 2a

(c) English W[+T] Tagalog W[+T]

1a = the 2a = lalaki

2a = man 1a = ang

3a = {1a,2a} 3a = {1a,2a}

Step 5. Merge verb

7-D,V

6saw -D,-D,V

5D

1the -N,D

4N

2happy A

3woman N

7-D,V

6nakita -D,-D,V-tns

5D

1sa -N,D

4N

2masaya A

3babae N

Step 6. Merge of V signals end of DP phase.

(a) Look inside DP for phases

English N[–T] N[+T]

1a = the 2a = lalaki

2a = man 1a = ang

3a = {1a,2a} 3a = {1a,2a}2 = masaya

3 = babae

4 = {2,3}

12

Remove 4 from tree

7-D,V

6nakita -D,-D,V-tns

5D

1ng -N,D

4N

(b) English: D[+L] Tagalog D[+L]

1→ 2 → 3 1→ 4

1a→ 2a 1a → 2a

Step 7. Merge subject (3a)

8V

3aD,-nom

7-D,V

6saw -D,-D,V

5D

1the -N,D

4N

2happy A

3woman N

8V

3aD

7-D,V

6nakita -D,-D,V-tns

5D

1sa -N,D

4N

Step 8. English: Merge T (9) Tagalog: Merge T, head-move V to T

10+nom,T

9-V,+nom,T

8V

3aD,-nom

7-D,V

6saw -D,-D,V

5D

1the -N,D

4N

2happy A

3woman N

10T

9nakita -V,T

8V

3aD

7-D,V

6nakita -D,-D,V-tns

5D

1sa -N,D

4N

Step 9. Merge of T indicates end of V phase.

Look inside VP for phases.

13

English D[+T] Tagalog D[+T]

1a = the 2a = lalaki

2a = man 1a = ang

3a = {1a,2a} 3a = {1a,2a}2 = happy 2 = masaya

3 = woman 3 = babae

4 = {2,3}1 = the 1 = sa

5 = {1,2,3} 5 = {1,4}Remove 5 from tree Remove 5 from tree

English V[+L] Tagalog V[–L]

3a→ 1 → 2 → 3 1→ 4

1a→ 2a 1a → 2a

Step 10. English: move subject up, merge C Tagalog: merge C, head-move T

13C

12-T,C

11T

3aD

10+nom,T

9-V,+nom,T

8V

3aD

7-D,V

6saw -D,-D,V

5D

13

12nakita C

10T

9nakita -V,T

8V

3aD

7-D,V

6nakita -D,-D,V-tns

5D

Step 11. Merge of C indicates end of TP phase.

(a) Look inside for phases.

English: V[–T] Tagalog: V[–T]

(b) English: T[+L] Tagalog: V[–L]

3a→ 1 → 2 → 3 1 → 4

↓9→ 6→ 5

1a→ 2a 1a → 2a

14

Step 12. Done sentence

(a) Look inside for phases

(b) English: T[–T] Tagalog: T[+T]

10 = {9,3a,6,5}Remove structure 10

13

12nakita C

10T

(c) English: C[+L] Tagalog: C[+L]

12→ 3a→ 2 → 3 1→ 4

↓ ↓9 → 6→ 5 1a → 2a

1a→ 2a 12 → 10

(d) English: C[+T] Tagalog: C[+T]

1a = the 2a = lalaki

2a = man 1a = ang

3a = {1a,2a} 3a = {1a,2a}1 = the 1 = sa

2 = happy 2 = masaya

3 = woman 3 = babae

4 = {2,3}10 = {9,3a,6,5}

5 = {1,2,3} 5 = {1,4}6 = saw 12 = nakita

13 = {12,9,3a,6,5} 13 = {12,9,3a,6,5}

Step 13. Make sequence: Substitute in sets for all nodes that stand for sets

{1a,2a} → 1→ 2 → 3 1 → {2,3}↓6→ {1,2,3} 1a → 2a

1a→ 2a 12 → {9,6,{1,{2,3}},{1a,2a}}

For each set, choose an order. If you choose an order that contradicts any other partof the order, the derivation will crash

15

1a→ 2a→ 6 → 1→ 3→ 2 12→ 9→ 6 → {1,{2,3}}→ {1a,2a}1 → {2,3} 1a → 2a

12→ 9 → 6 → 1 → 3→ 2→ 1a→ 2a

Sub in words:

the man saw the happy woman nakita ε ε sa babae masaya ang lalaki

5 Solutions to our Mysteries

5.1 CED vs PIC

• CED: no extraction from specifiers and adjuncts

• PIC: limited extraction from phases, including objects

• Problem: how can these both be effects of spellout?

• Solution:

– CED is an effect of transfer of workspaces (W[+T])

– PIC is an epiphenomenon of linearise : mostly it’s just the edge that can movebecause that’s most likely to preserve ordering

(13) a. [CPWhatj did [TP〈the monkey with pizza〉i[vP tj ti throw tj]PIC: linearise forces what to move through vP edge

b. *[CPWhatj did [TP〈the monkey with tj〉i[vP ti throw a slice]CED: subject is tranfered before it’s merged

5.2 CED violation:

• Puzzle: some languages allow extraction from subjects and adjuncts

• Solution: W[±T] is a parameter

5.3 OV Languages

• Puzzle: under antisymmetry, the Direct Object moves to become a satellite, so itshould be an island, but it isn’t.

• Solution: the direct object doesn’t start as a satellite, but rather as a complement, sothere is time for things to move out before it is tranfered

16

5.4 Free Word Order

Claim: If no linear order is determined on lexical items, order is free.

This approach derives multiple word orders from a single structure. LF gets the sameinformation no matter the order in which the Lexical Items (LIs) are pronounced. It is notnecessary to come up with multiple optional move operations to account for the optionalityof word order.

References

Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A Life inLanguage, pp. 1–52. Cambridge, MA: MIT Press.

Dixon, R. (1972). The Dyirbal Language of North Queensland. Cambridge: CambridgeUniversity Press.

Fox, D. and D. Pesetsky (2003, July). Cyclic linearization and the typology of movement.online.

Fox, D. and D. Pesetsky (2005). Cyclic linearization of syntactic structure. TheoreticalLinguistics 31, 1–45.

Huang, C.-T. J. (1982). Logical relations in Chinese and the theory of grammar. Ph. D.thesis, MIT.

Kayne, R. (1994). The Antisymmetry of Syntax, Volume 25 of Linguistic Inquiry Monographs.Cambridge, MA: MIT Press.

Nunes, J. and J. Uriagereka (2000, April). Cyclicity and extraction domains. Syntax 3 (1),20–43.

Saito, M. (1992). Long distance scrambling in Japanese. Journal of East Asian Linguistics 1,69–118.

17