quantitative methods in typology - uzh

Quantitative

Methods in Typology

Balthasar BickelUniversity of Leipzig

Rationale of this course

2

Anderson 2008: “The object of inquiry in linguistics is the human ability to acquire and use a natural language, and the goal of linguistic theory is an explicit characterization of that ability.”

20th century“Universalist Typology”

nature methodgoal formulating UGinterest in absolute implicational

universalspart of classical CogScicore question What’s a Possible/

Learnable Human Language?

A paradigm change

3

Three developments favor a move away from this

‣ Recognition of ‘universal areality’

‣ No sample can tell us what is (im)possible

‣ But samples allow probabilistic inference, using the same techniques as other disciplines (esp. bio-informatics)

‣ Recognition that answers to the “core question” may better come from empirical language acquisition research and comparative psychology than from linguistics per se

A paradigm change

4

20th century“Universalist Typology”

21st century“Distributional Typology”

nature method disciplinegoal formulating UG explaining distributionsinterest in absolute implicational

universalsdistributional probabilities

part of classical CogSci anthropology, including psychology

core question What’s a Possible/Learnable Human Language?

What’s Where Why?

What’s Where Why?

‣ what: what linguistic structures are there and how can we compare them?

‣ where: how are languages and their structures distributed in space and time, both locally and universally?

‣ why: what factors determine the distribution of languages and their structures in space and time?

5

What’s Where Why?

Any understanding of What’s Where Why involves quantitative methods:

‣what: measuring similarities

‣where: examining the distribution of similarities

‣why: developing statistical models of the distributions

6

Quantitative Methods in Typology I:

Comparing structures

Balthasar BickelUniversity of Leipzig

Comparing structures

8

•Bloomfield: “If you want to compare two languages, it helps to know one of them.”

•The Descriptive A Priori = analytical metalanguages or (when named and popular) ‘theoretical frameworks’

•Two necessary ingredients:

1.Absolute Universals = primitive concepts that are required for the analysis of every imagineable language (“Theoretical universal”)

2.Variables = sets (inventories, scales) of primitive concepts that are required for the analysis of at least one language

Other terms: tertium comparationis, comparative concepts, metrics, parameters, features

Bickel, to appear in Cambridge Encyclopedia of the Language Sciences, 2007

Too many options to take: how to justify metalanguages?

•One popular answer: claim the metalanguage to be psychologically and biologically real (Chomsky: “UG“, “S0”) (Descriptive A Priori = Ontological A Priori)

•Problems:

•if the metalanguage also includes variables (‘has vs. does not have clicks’), all values of the variable must be universally present in the brain, i.e. even in languages that don’t have evidence for them. How can such claims ever be tested? (Nichols 2008)

•the biological evidence is unclear (gap between genes, expression of genes, and behavior) (e.g. Müller 2009)

•language acquisition does not seem to rely on any of our linguistic metalanguages but on general learning mechanisms (Tomasello 2003)

9

A pragmatic approach to metalanguages

•The working linguist’s everyday answer: justify the metalanguage not by appeal to psychology or biology but only by standard principles of scientific inquiry:

•logical consistency•operationalizability (availability of proofs and

argumentation)•replicability•universal descriptive success: all structures can be

analyzed in a cross-linguistically consistent way, allowing comparison (compare “subject” and “proto-agent”)

•Compare this to the use of the metrical system as justified by its consistency, operationalizability, replicability, and descriptive and ‘typological’ success — and not by any innateness claim!

10

A pragmatic approach to metalanguages

•Conceived therefore as analytical instruments, the metalanguage

•makes no claim on genetic and/or cognitive reality (but may pick up signals that can be explained by cognitively and possibly even genetic patterns, cf. Hawkins 2004, Dediu & Ladd 207)

•makes no claim on the distribution of its denotata in the world (they could be all over, like “nasals”, or rare like “clicks”)

•therefore, it provides the right means to study actual distributions, i.e. what’s where why?

•So, let’s work out a good metalanguage!

11

A key problem for metalanguages:

•The structures we find often don’t quite fit with what we know: e.g. is X “really” a passive? “really” an affix? “really” subordinate?

•There are always many similarities but few identities!

12

‘many similarities, few identities’

•A case study (Bickel in press): “cosubordination”

13

! "#$%$ &'()*+,-$.,/01*$)2 3)4)*56 789$(:+ ;<=>2;?;@!";+

#"$%$%3A

&'("B+

)"CDE

*"F)(

+,-(8)4

")".'#8:G$(,B+HDII

)"CDE

%)%:G$($

)&HJ7H

/&0.&1.,.)1:,B+,H7KL

(,2(,$.,',)IM3N9$,LJ7;+LI

)3.-3.(F8#$A48.*,LI,B+

4%(%.-%.0,)F%1#9A0O,LI,;+

(3$%.5.")658,;4,PAHI'

QRG1%$ M .)1:$4 :G$($ ): :G$ #)1* (8)4 S8( 3)%8%8T+ F)( G$ F)#$ 48.*A M F%1#9$41* )*4 8U .$ .$*:AT

V ')0W) &'()*+,-$.,/01*$)2 3)4)*56 3)FL8*)%4 ;<<?2;?;@5,5,.-",#8++,"LKII

7"4358

84"&.".43'KC,B+,LI

03LK3

)%)%.9"FG1%4,'DH

4%&43+:1FX

:"430"/343.;"+G)(O$*,II

-&.".-"6415,B+,M-L

QIG$ .$*: ).)W 1*:8 :G$ #8++6 )*4 :G1+ 98W +G)(O$*$4 ) +:1FX )*4 405AT

!"#"$"%&"'

3)FL8*)%46 CA6 ;<<?A < '9"00"9 %: ="&7"A Y$(%1*2 380:8* 4$ /(0W:$(A

789$(:+6 ZA 7A6 ;<=[ <03$3A C8*48*2 E(88# \$%#A

‘many similarities, few identities’

•Universal definition: clause linkage with conjunct illocutionary scope (operator dependency) (e.g., Foley and Van Valin 1984)

14

! "#$%$ &'()*+,-$.,/01*$)2 3)4)*56 789$(:+ ;<=>2;?;@!";+

#"$%$%3A

&'("B+

)"CDE

*"F)(

+,-(8)4

")".'#8:G$(,B+HDII

)"CDE

%)%:G$($

)&HJ7H

/&0.&1.,.)1:,B+,H7KL

(,2(,$.,',)IM3N9$,LJ7;+LI

)3.-3.(F8#$A48.*,LI,B+

4%(%.-%.0,)F%1#9A0O,LI,;+

(3$%.5.")658,;4,PAHI'


V ')0W) &'()*+,-$.,/01*$)2 3)4)*56 3)FL8*)%4 ;<<?2;?;@5,5,.-",#8++,"LKII

7"4358

84"&.".43'KC,B+,LI

03LK3

)%)%.9"FG1%4,'DH

4%&43+:1FX

:"430"/343.;"+G)(O$*,II

-&.".-"6415,B+,M-L

QIG$ .$*: ).)W 1*:8 :G$ #8++6 )*4 :G1+ 98W +G)(O$*$4 ) +:1FX )*4 405AT

Y "#$%$ &'()*+,-$. /01*$)2 3)4)*5Z H)O0) -$. /01*$)Z 789$(:+ ;<==@<%O15

(&/"$3.-3.((0*A80:,LI,B+

1")"#)*

"'3BO

'(%.,'."G1:,BO,'AHI'

:%=[

QL14 :G$ O15 (0* 80: )*4 414 :G$ #$* X1%% 1:\T

] ')0W) &'()*+,-$. /01*$)2 3)4)*56 H)O0) -$. /01*$)Z 3)FL8*)%4 ;<<?2^^_@43;"&.:3.;"9($)X,H7 ,̀II

7"4358

84"&.">)"3=:G(8.,^aHDC"7A[

QL14 W80 9($)X 1: )*4 58 ).)W\T%9 QL14 W80 58 ).)W )b$( 9($)X1*5 1:\T &O($+0OO8+1*5 $1:G$( QW80 .$*: ).)WT 8( QW809(8X$ 1:T@

!"#"$"%&"'

3)FL8*)%46 CA6 ;<<?A ? '9"00"9 %: @"&7"A c$(%1*2 380:8* 4$ /(0W:$(A

789$(:+6 dA 7A6 ;<=e ?03$3A C8*48*2 E(88# f$%#A

789$(:+6 dA 7A6 ;<=g "#$%$ +.1:FG,($S$($*F$ )*4 :G$ :G$8(W 8S 5()##)(A A,)'&,/4,B C)D&,97 ;<6 hij _k

! "#$%$ &'()*+,-$.,/01*$)2 3)4)*56 789$(:+ ;<=>2;?;@!";+

#"$%$%3A

&'("B+

)"CDE

*"F)(

+,-(8)4

")".'#8:G$(,B+HDII

)"CDE

%)%:G$($

)&HJ7H

/&0.&1.,.)1:,B+,H7KL

(,2(,$.,',)IM3N9$,LJ7;+LI

)3.-3.(F8#$A48.*,LI,B+

4%(%.-%.0,)F%1#9A0O,LI,;+

(3$%.5.")658,;4,PAHI'


V ')0W) &'()*+,-$.,/01*$)2 3)4)*56 3)FL8*)%4 ;<<?2;?;@5,5,.-",#8++,"LKII

7"4358

84"&.".43'KC,B+,LI

03LK3

)%)%.9"FG1%4,'DH

4%&43+:1FX

:"430"/343.;"+G)(O$*,II

-&.".-"6415,B+,M-L

QIG$ .$*: ).)W 1*:8 :G$ #8++6 )*4 :G1+ 98W +G)(O$*$4 ) +:1FX )*4 405AT

Y "#$%$ &'()*+,-$. /01*$)2 3)4)*5Z H)O0) -$. /01*$)Z 789$(:+ ;<==@<%O15

(&/"$3.-3.((0*A80:,LI,B+

1")"#)*

"'3BO

'(%.,'."G1:,BO,'AHI'

:%=[

QL14 :G$ O15 (0* 80: )*4 414 :G$ #$* X1%% 1:\T

] ')0W) &'()*+,-$. /01*$)2 3)4)*56 H)O0) -$. /01*$)Z 3)FL8*)%4 ;<<?2^^_@43;"&.:3.;"9($)X,H7 ,̀II

7"4358

84"&.">)"3=:G(8.,^aHDC"7A[

QL14 W80 9($)X 1: )*4 58 ).)W\T%9 QL14 W80 58 ).)W )b$( 9($)X1*5 1:\T &O($+0OO8+1*5 $1:G$( QW80 .$*: ).)WT 8( QW809(8X$ 1:T@

!"#"$"%&"'

3)FL8*)%46 CA6 ;<<?A ? '9"00"9 %: @"&7"A c$(%1*2 380:8* 4$ /(0W:$(A

789$(:+6 dA 7A6 ;<=e ?03$3A C8*48*2 E(88# f$%#A

789$(:+6 dA 7A6 ;<=g "#$%$ +.1:FG,($S$($*F$ )*4 :G$ :G$8(W 8S 5()##)(A A,)'&,/4,B C)D&,97 ;<6 hij _k

Classical responses

•Basic principle: the universal definition follows from our metalanguage (the Descriptive A Priori), it cannot have exceptions.

•Therefore, what looks like an exception, isn’t! We can get rid of it, “explain it away”.

•In other words: Reduce the diversity before you study it!

•This is characteristic both of theory-of-grammar and typology approaches:

15

Classical theory-of-grammar approaches

•Search for a higher-ranking principle, e.g. a principle that explains why Tauya -pa allows disjunct scope although it “really” establishes cosubordination

—Problem: no principle known!

•Limit the scope of the definition, e.g. Tauya does not have cosubordination, but “sub-cosubordination”

—Problem: the structures are so similar to each other that one reading of cosubordination is the sole reading of “sub-cosubordination”.

16

Classical theory-of-grammar approaches (cont’d)

•Assume structural ambiguity: in Tauya, one reading ‘really’ reflects cosubordination while the other reflects something else (probably subordination, with disjunct scope) (Bickel 1998).

—Problem: No independent evidence for this

•Revise the definition, e.g. define cosubordination by constraint-free scope (Bickel 1991, Croft 2001)

—Problem: We can base the definition on any variable we want (e.g. finiteness, tense scope, extraction possibilities, assertedness, etc.) but we may always run into the same problems! “Methodological opportunism” (Croft 2001)

17

Classical typological approaches

•Typologize exemplars (“basic” cosubordination, like “basic word order“)

—Problem: which one to pick?

•Define “comparative concepts” that abstract away from language-particular details (Lazard 2006, Haspelmath 2007), perhaps via a “functional” definition (e.g. conjunct scope) —

—Problem: all language-specific analysis is informed by and benefits from cross-linguistically defined concepts.

‣All classical responses convey a sense of “After all, languages are not that different from each other”, and thereby prevent a deeper understanding of the human faculty for diversity (cf. Evans & Levinson 2009)

18

Bickel 2007, in press

An alternative: Multivariate Typology

•Diversity means that across languages, things are mostly similar and hardly ever identical.

•But similarity is nothing else but identity in some variables and difference in others.

‣For studying similarities, we need large systems of fine-grained variables that fully capture the range of known variation: Multivariate Typology.

‣And with this, we can describe and measure the variation, instead of reducing it — i.e. do what most other disciplines would do when confronted with variation.

19

An alternative: Multivariate Typology

•How many structures? — As many as are distinct in V1 ... Vk

•How many variables? — As many as are of interest to the research question, e.g. all variables needed to capture cross-linguistic differences in the syntax of clause linkage; or in morphological coding.

•Which variables? — Developed as needed for distinguishing structures during data collection (Autotypologizing Method: Bickel & Nichols 2002), or pre-determined by the research question.

•structural variables: morphosyntactic or semantic properties in which structures are alike or differ.

•denotation variables: denotations (stimuli, contexts, functions) in which structures are alike or differ

20

‘many similarities, few identities’: Multivariate typology

Example: variables of clause linkage

22

!"

# $%&'()$ *$+,-./* 0&/1$ 2$&,(0$ 3$'$4,*- ),-, 5,0 */- ,4,.',2'$ 5.-6 0(7&.$*- ,*,'80.09 # ,'0/ (0$

, 0.:1'.;$) 4$30./* /< -6$ ;*.-$*$00 4,3.,2'$9 =6$ </''/5.*+ )$;*.-./*0 /< '$4$'0 5$3$ (0$)> '.0-.*+

/*'8 '$4$'0 -6,- ,3$ ,&-(,''8 ,--$0-$) .* -6$ ),-,2,0$?

!""#$%&'() =6$ 0&/1$ /< .''/&(-./*,38 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

FGHCDB@E? $%-$*)0 -/ $.-6$3 -6$ :,.* /3 -6$ )$1$*)$*- &',(0$ 2(- *$4$3 -/ 2/-6

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

@ABHENJGBEOPNKK? .0 */- 3$+(',-$) 28 -6$ &',(0$ '.*Q,+$ -81$

*#$%&'() =6$ 0&/1$ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

+,-,.(-($$) =6$ )$1$*)$*- &',(0$ .0 6$,)$) 28 , 4$32 </3: -6,- .0

PGBGEK? ,- '$,0- ,0 :,*8 &,-$+/3.$0 :(0- 2$ :,3Q$) ,0 .* :,.* &',(0$0

BABPGBGEK? /*'8 <$5$3 &,-$+/3.$0 ,3$ ,''/5$)

JBR? $.-6$3 -6$ 0,:$ 3,*+$ /3 '$00 &,-$+/3.$0 &,* 2$ :,3Q$)

!""#/012) S,3Q.*+ /< .''/&(-./*,38 </3&$ /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- /*'8 .< .- :,-&6$0 -6$ :,3Q.*+ /* -6$ :,.* &',(0$

*#/012) S,3Q.*+ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- 0(2W$&- -/ &/*0-3,.*-0 2,0$) /* -6$ -$*0$ /3 0-,-(0 &6/.&$ .* -6$

:,.* &',(0$

34//(.14) =6$ 3,*+$ /< &,-$+/3.$0 -6,- &,* 2$ $%13$00$) /* '.*Q$) &',(0$0 .0

HRVVKENG@JI? :(0- :,-&6

JHRVVKENG@JI? &,* 2$ ).X$3$*-

PNKK? &,* 2$ ).X$3$*- ,*) &,* $4$* .*&'()$ $'$:$*-0 /< ).X$3$*- -81$ Y).X$3$*- 1,3-0 /<

01$$&6> &',(0$0 ,*) Z[0> $-&9\

56) ]($0-./* 5/3)0 ,*) &/*0-.-($*- </&(0 .*0.)$ )$1$*)$*- &',(0$0 ,3$

AT? ,''/5$)

MJBBKF? */- ,''/5$)

78.10%.,&-) ^%-3,&-./* /< $'$:$*-0 /< )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

+9:) _/&(0 :,3Q.*+ /* )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

;&$,.,&-) =6$ 1/0.-./* /< -6$ )$1$*)$*- &',(0$ 4.0O`O4.0 -6$ :,.* &',(0$ 5.-6 56.&6 .- $*-$30 ,

)$1$*)$*&8 3$',-./* .0

!"#$%"& #"'%$() * +,)" --. -//0


23

!"

# $%&'()$ *$+,-./* 0&/1$ 2$&,(0$ 3$'$4,*- ),-, 5,0 */- ,4,.',2'$ 5.-6 0(7&.$*- ,*,'80.09 # ,'0/ (0$

, 0.:1'.;$) 4$30./* /< -6$ ;*.-$*$00 4,3.,2'$9 =6$ </''/5.*+ )$;*.-./*0 /< '$4$'0 5$3$ (0$)> '.0-.*+

/*'8 '$4$'0 -6,- ,3$ ,&-(,''8 ,--$0-$) .* -6$ ),-,2,0$?

!""#$%&'() =6$ 0&/1$ /< .''/&(-./*,38 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

FGHCDB@E? $%-$*)0 -/ $.-6$3 -6$ :,.* /3 -6$ )$1$*)$*- &',(0$ 2(- *$4$3 -/ 2/-6

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$


*#$%&'() =6$ 0&/1$ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

+,-,.(-($$) =6$ )$1$*)$*- &',(0$ .0 6$,)$) 28 , 4$32 </3: -6,- .0

PGBGEK? ,- '$,0- ,0 :,*8 &,-$+/3.$0 :(0- 2$ :,3Q$) ,0 .* :,.* &',(0$0

BABPGBGEK? /*'8 <$5$3 &,-$+/3.$0 ,3$ ,''/5$)

JBR? $.-6$3 -6$ 0,:$ 3,*+$ /3 '$00 &,-$+/3.$0 &,* 2$ :,3Q$)

!""#/012) S,3Q.*+ /< .''/&(-./*,38 </3&$ /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- /*'8 .< .- :,-&6$0 -6$ :,3Q.*+ /* -6$ :,.* &',(0$

*#/012) S,3Q.*+ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- 0(2W$&- -/ &/*0-3,.*-0 2,0$) /* -6$ -$*0$ /3 0-,-(0 &6/.&$ .* -6$

:,.* &',(0$

34//(.14) =6$ 3,*+$ /< &,-$+/3.$0 -6,- &,* 2$ $%13$00$) /* '.*Q$) &',(0$0 .0



PNKK? &,* 2$ ).X$3$*- ,*) &,* $4$* .*&'()$ $'$:$*-0 /< ).X$3$*- -81$ Y).X$3$*- 1,3-0 /<

01$$&6> &',(0$0 ,*) Z[0> $-&9\

56) ]($0-./* 5/3)0 ,*) &/*0-.-($*- </&(0 .*0.)$ )$1$*)$*- &',(0$0 ,3$

AT? ,''/5$)

MJBBKF? */- ,''/5$)

78.10%.,&-) ^%-3,&-./* /< $'$:$*-0 /< )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

+9:) _/&(0 :,3Q.*+ /* )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

;&$,.,&-) =6$ 1/0.-./* /< -6$ )$1$*)$*- &',(0$ 4.0O`O4.0 -6$ :,.* &',(0$ 5.-6 56.&6 .- $*-$30 ,

)$1$*)$*&8 3$',-./* .0

!"#$%"& #"'%$() * +,)" --. -//0


24

!"

#""$ %&'( #)'*+, -(. /01+(*2 3*1+*+405/&'&6*+7 89&44 !:;<$

*= !"#"$"%$!&$'(9&>(5?,@AB8=C85CB@5!DE8

"$!"$)*+,-?,@5,((5!DEFG

#HIJKLMN$

HO, P( 9&>1+Q *+D .( ,(( 14RN

S= #"$.$/$0$(#'+,!,@5Q1T(5",%U)=C85CB@5!,E8FOV@

#"$!)$1$/-(*45%U)5!,5CBWX

#HIJKLMN$

H/1T( >( ,&>(4P1+Q *+D 4P(+ O .1YY (*4 14=N #2-/-Z H/1T( >( ,&>(4P1+Q 4& (*4[N$

)P( ,(+4(+9( 1+ #""*$ 1YY0,4'*4(, 4P( \*94 4P*4 >*1+ 9Y*0,( ]0(,41&+ >*'6(', P*T( 9&^0+94 ,9&_(Z

.P1Y( #""S$ ,P&., 4P*4 4P1, S*,19 ,4'0940'( D&(, +&4 S*+ 4P( &990''(+9( &\ *4 Y(*,4 1>_('*41T( >*'6(',

&+ 4P( D(_(+D(+4 9Y*0,(=

E +0>S(' &\ Y*+Q0*Q(, *YY&. 1YY&9041&+*'` >*'61+Q &+ D(_(+D(+4 9Y*0,(, &+Y` 1\ 14 >*49P(, 4P(

>*'61+Q 1+ 4P( >*1+ 9Y*0,(= O+ a(YP*'(Z \&' (b*>_Y(Z !259Y*0,(, #&\ 4P( 61+D 1YY0,4'*4(D S(\&'( S`

#?$ *+D #!;$$ 9*+ S( >*'6(D S` *+ 1>_('*41T(Z S04 &+Y` 1\ 4P( ,*>( >&&D 1, *Y,& >*'6(D &+ 4P(

>*1+ 9Y*0,(2

#"?$ a(YP*'(

3"1$"(*45OV@

!28BG

!4"5$"6Q&5OV@

#HIJKLMN$

HB*4 *+D Q&[N

O+ WP(9P(+Z 1YY&9041&+*'` >*'61+Q &+ 4P( D(_(+D(+4 9Y*0,( 1, _&,,1SY( 1+ ,&>( S04 +&4 &4P('

,4'0940'(,= %&' (b*>_Y(Z 4P( >*'61+Q 1, SY&96(D 1+ 9Y*0,(, >*'6(D S` 4P( ,(]0(+41*Y 9&+T('S 1+ $#"Z&' 4P( 9&+9(,,1T( 9&+T('S 1+ $#"$*4/41 4P*4 1, D('1T(D c&> 4P1,= O+ (*9P 9*,(Z 4P( 1+4(''&Q*41T( 9Y1419

#+22$ >0,4 &990' &+ 4P( >*1+ 9Y*0,( #"d*Z9$ *+D 9*++&4 S( *44*9P(D 4& 4P( D(_(+D(+4 9Y*0,( #"dSZD$2

#"d$ WP(9P(+ #/&&D "ee?Z V&Y&9P1(T* "ee<$

*= 7"822!"V=#f$=-gV

'9!"$#"+:",4&'(5CE)F88

;$"<$#"f5Q&5Wha

3:"$;$/:"$5+22=P&>(5f59&>(5i=@8)FG

#HIJKLMN$

HC1D V*Y16* 9&>( P&>(Z P*T1+Q Q&+( 4& 4P( ,4&'(RN #_'(,0__&,1+Q HP*T1+Q Q&+( 4& 4P(

,4&'(N$

>5 HC1D V*Y16* Q& 4& 4P( ,4&'( *+D 9&>( P&>(RN

?)' #>'@ HC1D V*Y16* Q& 4& 4P( ,4&'(Z P*T1+Q 9&>( P&>(RN

S= j7"822!"V=#f$=-gV

'9!"$#"+:",4&'(5CE)F88

;$"<$#"+22f5Q&5WhaFG

3:"$;$/:"$5P&>(5f59&>(5i=@8)

#HIJKLMN$

A#'/#B/B@ HC1D V*Y16* 9&>( P&>( P*T1+Q Q&+( 4& 4P( ,4&'(RN #_'(,0__&,1+Q HV*Y16*

9*>( P&>(N$

9= "41"8/=BA/

'>/<$#"$*4/41P145Wha5Wg-W

;$"<"$5+22f5Q&5i=@8)FG

C""5"k=#f$=-gV

'9!"$#"=,4&'(5CE)

#HIJKLMN$

HC1D k*'* Q& 4& 4P( ,4&'(Z (T(+ 4P&0QP `&0 P14 P('RN

#>' HC1D k*'* Q& 4& 4P( ,4&'(Z *+D 1, 14 (T(+ 4P( 9*,( 4P*4 `&0 P14 P('RN

D= j "41"8/=BA/

'>/<$#"$*4/41+22P145Wha5Wg-WFG

;$"<"$5"f5Q&5i=@8)

C""5"k=#f$=-gV

'9!"$#"=,4&'(5CE)

#HIJKLMN$

A#'/#B/B@ HC1D k*'* Q& 4& 4P( ,4&'(Z (T(+ 4P&0QP `&0 P14 P('RN

!"#$%"& #"'%$() * +,)" --. -//0

!"

#$%& '() #*+),-./01 2(3,0)4 5)3,),6(.27+78),9 ')3:), !;<=4"$!&

)> !"#$%?3@A*BC

&'$(')*'+*'&*[email protected]*.G?HI.!-JI

"'+"',[email protected]*.!-HKLM

#NOP.QRSTUV&

N*W0X 13YY @3D0 :0 ?7+8 ),Z 6W0, [ 13YY @7>V

\> ] !"#$%?3@A*BC

&'$(')*'+*'&*[email protected]*.G?HI.!-JI

"',-@7^/EF*_.!-HKLM

#NOP.QRSTUV&

./0,/&,&1 N*W0X 13YY @3D0 :0 ?7+8 ),Z -7 [ 10,6V` (-,-` N[ 10,6 \0a)(-0 6W0X 13YY @3D0:0 ?7+8>V

[, 7,0 6X?0 7b aY)(-0 Y3,8)@0` :)+80Z \X ), )@+00:0,6 ?)+)Z3@: 6W)6 )Y-7 -3@,)Y- -136aW.+0b0+0,a0#W0+0 3, b7+: 7b 6W0 6W3+Z ?0+-7,` Z3c0+0,6 -(\d0a6 :)+80+ '+*&` 60,-0 :)+83,@ 3- )YY710Z 3, 6W0Z0?0,Z0,6 aY)(-0 7,YX 3b 36 :)6aW0- 6W0 60,-0 :)+83,@ 7b 6W0 :)3, aY)(-0> *W3- 3- 6W0 a)-0 3, #$%)&`\(6 ,76 3, #$%\&> [, ),76W0+ 6X?0 7b aY)(-0 Y3,8)@0 3, '()` :)+80Z \X ), )@+00:0,6 ?)+)Z3@:136W7(6 -136aW.+0b0+0,a0 a7Z3,@ #W0+0 3, b7+: 7b 6W0 6W3+Z ?0+-7, ?Y(+)Y :)+80+ '$*&` ,7 -(aWa7,-6+)3,6 3- 3:?7-0Z> H0?0,Z0,6 aY)(-0- e00YX )YY71 60,-0 :)+83,@` +0@)+ZY0-- 7b :)3, aY)(-060,-0 aW73a0-4

#$<& '()!"#$%?3@A*BC

&'$('+*'$*'&*[email protected]*.G?IFf.!-JI

"',-@7^/EF*_.!-HKLM

#NghiTQV&

N*W0X 13YY @3D0 :0 ?7+8 ),Z -7 [ 10,6V` (-,-` N[ 10,6 \0a)(-0 6W0X 13YY @3D0 :0 ?7+8>V

F,a7,-6+)3,0Z 60,-0 :)+83,@ 3, Z0?0,Z0,6 aY)(-0- 3- 6+)Z3637,)YYX )--7a3)60Z 136W 036W0+ Na7.7+Z3,)637,V ),Z N-(\7+Z3,)637,V` 1W3Y0 6W0 ?+0-0,a0 7b a7,-6+)3,6- W)- -7:063:0- \00, -(@@0-60Z)- Z3)@,7-63a 7b NaW)3,3,@V ),Z Na7,D0+\V a7,-6+(a637,- #0>@> E7Y0X !;<j&> '710D0+` 3, ) ,(:\0+ 7bY),@()@0-` 3,aY(Z3,@ Y),@()@0- e7: 6W0 *+),-./01 2(3,0) b):3YX` -6+(a6(+0- 6+)Z3637,)YYX aY)--3.k0Z )- NaW)3,3,@V e00YX )YY71 60,-0 :)+83,@> J a)-0 3, ?73,6 3- 57+)b0` 1W0+0 6W0 Z3c0+0,6 -(\d0a6:)+80+ 3, 6W0 b7YY713,@ 0l):?Y0- 0-6)\Y3-W ?)-6 60,-0 +0b0+0,a0 3, 6W0 Z0?0,Z0,6 aY)(-0> *W0 :)3,aY)(-0 a), \0 3,Z0?0,Z0,6YX :)+80Z )- ?+0-0,6 7+ ?)-6 60,-04

#$;& 57+)b0 #*+),-./01 2(3,0)4 f3,),Z0+0),` C)?() /01 2(3,0)9 E)++ !;;;&

)> $"0',/%@3D0>!-.IKm>nKJM[I>!-HI

,2'(2*'2,-[CEo.@7>HFn>CnI>G-[/H.LFnnK/*>nKMKoJ/LK

#NQRSTUV&

N[ @)D0 36 ),Z W0 3- a(++0,6YX @73,@>V

\> $"0',/%@3D0>IKm>nKJM[I>!-HI

(')(2*-@7>HFn.E>CI*>G-[/H

#NQRSTUV&

N[ @)D0 36 ),Z W0 10,6 #617 7+ :7+0 Z)X- )@7&V

[, p):\(Y0` ) Y),@()@0 e7: 6W0 -):0 I3,7.*3\06), \+),aW )- f0YW)+0` 60,-0 :)+83,@ 3-)YY710Z 0laY(-3D0YX 7, aW)3,0Z` Z0?0,Z0,6 aY)(-0-> q)3, aY)(-0-` \X a7,6+)-6` a),,76 \0 :)+80Z b7+60,-0>r

r *W3- 3- Y380YX 67 +0-(Y6 e7: ) +0@(Y)+ ?+7a0-- 7b ,7:3,)Y3s3,@ :)3, aY)(-0-` b7YY713,@ ) @0,0+)Y 6+0,Z aW)+)a60+3-63a 7b

!"#$%"& #"'%$() * +,)" --. -//0


25

!"

# $%&'()$ *$+,-./* 0&/1$ 2$&,(0$ 3$'$4,*- ),-, 5,0 */- ,4,.',2'$ 5.-6 0(7&.$*- ,*,'80.09 # ,'0/ (0$

, 0.:1'.;$) 4$30./* /< -6$ ;*.-$*$00 4,3.,2'$9 =6$ </''/5.*+ )$;*.-./*0 /< '$4$'0 5$3$ (0$)> '.0-.*+

/*'8 '$4$'0 -6,- ,3$ ,&-(,''8 ,--$0-$) .* -6$ ),-,2,0$?

!""#$%&'() =6$ 0&/1$ /< .''/&(-./*,38 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

FGHCDB@E? $%-$*)0 -/ $.-6$3 -6$ :,.* /3 -6$ )$1$*)$*- &',(0$ 2(- *$4$3 -/ 2/-6

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$


*#$%&'() =6$ 0&/1$ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

+,-,.(-($$) =6$ )$1$*)$*- &',(0$ .0 6$,)$) 28 , 4$32 </3: -6,- .0

PGBGEK? ,- '$,0- ,0 :,*8 &,-$+/3.$0 :(0- 2$ :,3Q$) ,0 .* :,.* &',(0$0

BABPGBGEK? /*'8 <$5$3 &,-$+/3.$0 ,3$ ,''/5$)

JBR? $.-6$3 -6$ 0,:$ 3,*+$ /3 '$00 &,-$+/3.$0 &,* 2$ :,3Q$)

!""#/012) S,3Q.*+ /< .''/&(-./*,38 </3&$ /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- /*'8 .< .- :,-&6$0 -6$ :,3Q.*+ /* -6$ :,.* &',(0$

*#/012) S,3Q.*+ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- 0(2W$&- -/ &/*0-3,.*-0 2,0$) /* -6$ -$*0$ /3 0-,-(0 &6/.&$ .* -6$

:,.* &',(0$

34//(.14) =6$ 3,*+$ /< &,-$+/3.$0 -6,- &,* 2$ $%13$00$) /* '.*Q$) &',(0$0 .0



PNKK? &,* 2$ ).X$3$*- ,*) &,* $4$* .*&'()$ $'$:$*-0 /< ).X$3$*- -81$ Y).X$3$*- 1,3-0 /<

01$$&6> &',(0$0 ,*) Z[0> $-&9\

56) ]($0-./* 5/3)0 ,*) &/*0-.-($*- </&(0 .*0.)$ )$1$*)$*- &',(0$0 ,3$

AT? ,''/5$)

MJBBKF? */- ,''/5$)

78.10%.,&-) ^%-3,&-./* /< $'$:$*-0 /< )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

+9:) _/&(0 :,3Q.*+ /* )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

;&$,.,&-) =6$ 1/0.-./* /< -6$ )$1$*)$*- &',(0$ 4.0O`O4.0 -6$ :,.* &',(0$ 5.-6 56.&6 .- $*-$30 ,

)$1$*)$*&8 3$',-./* .0

!"#$%"& #"'%$() * +,)" --. -//0

!

! "#$%$ &'()$*+, -./01 -.//2

34 5 !"-,

#"6*$

$%&'()'(*+,(7$89:;9-,

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

)4 !"-,

#"6*$

$%&(+-("(7$89-,9B4C;B

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

DE %F+ +G$ 6*$ )<+ ,G$ @((A$? +G$ >((?4H

!"#"$"%&"'

IF@A$%1 I3%+G3,3*1 -..J I$%G3*$ ,<)(*?F83+F(8 38? +G$ +G$(*K (> +(7F@4 E8 L)$*+1 M3*$8 N4 &$?42

23%&+0/ +, 45"%/0 5+,6"-01 O= P QR ST*F@GU ";";4

IF@A$%1 I3%+G3,3* V W(X$8?*3 C4 WY?3Z31 O[[[4 " \$,G %((A 3+ X*3##3+F@3% *$%3+F(8, F8 E8?(9"*K384

7+,-%" --[1 =]= P =0J

^3@:(83%?1 _(*831 -..[4 8 -9"**"9 ': ;"%<"4 I$*%F8U ^(<+(8 ?$ `*<K+$*4

'()$*+,1 a(G8 '41 -./b 8*0504 _(8?(8U c*((# N$%#4

'()$*+,1 a(G8 '41 -./d "#$%$ ,eF+@G9*$>$*$8@$ 38? +G$ +G$(*K (> X*3##3*4 7+,-%+/3+4 =,>%+9< -.1]Q P fg

h38 h3%F81 '()$*+ :41 a*41 -..R B(e3*?, 3 ><8@+F(83%F,+ 3@@(<8+ (> ,(9@3%%$? $i+*3@+F(8 @(8,+*3F8+,4

E8 :$Z*F$8?+1 I$++K1 _(<F, `((,,$8,1 V a(G38 Z38 ?$* "<e$*3 &$?,42 ?'*@50A /39%43%90/B ":%,43+',"5+/3 @09/@043+C01 O. P f[4 I$*%F8U ^(<+(8 ?$ `*<K+$*4


26

!

! "#$%$ &'()$*+, -./01 -.//2

34 5 !"-,

#"6*$

$%&'()'(*+,(7$89:;9-,

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

)4 !"-,

#"6*$

$%&(+-("(7$89-,9B4C;B

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

DE %F+ +G$ 6*$ )<+ ,G$ @((A$? +G$ >((?4H

I J,38 &'$$,F8A -./0K L/=M 2

34 *%,',#38

+2"%?(N

3454*('5=,C4GF+9=,O4C;B

0,-BPC

%*('50+?F$9=,O4C;B

)+2'6Q

$%&'(

DRFS$8 +G3+ +G$ #38 GF+ +G$ ?(N1 ?F? F+ ?F$TH

)4 245(4.@(#$9;;

0,-BPC

20(,+7"8-,C9,+$74(S$*4;;

)%5#(8$U

.+-(4517<+97EVC

$%&'(

DE> U(< @(#$1 ,+$7 (S$* #$ 38? 7<+ U(<* #(8$U &F8 +G$ )3,A$+24H

@4 *%,',#38

0,-9BPC

3','%=,CP;;

*",N3*?$8

/'"%%38?,%F?$

+/('50+1N(4?(W89=,O4C;B

D", >(* +G$ #381 GF, N3*?$8 W$8+ ?(W8 F8 +G$ %38?,%F?$4H

!"#"$"%&"'

XF@A$%1 X3%+G3,3*1 -..Y X$%G3*$ ,<)(*?F83+F(8 38? +G$ +G$(*U (> +(7F@4 E8 Z)$*+1 [3*$8 \4 &$?42

:8%&+0/ +, ;<"%/0 <+,="-01 L= ] ^_ `a*F@GK ";";4

XF@A$%1 X3%+G3,3* b c(N$8?*3 C4 cd?3S31 Leee4 " f$,G %((A 3+ N*3##3+F@3% *$%3+F(8, F8 E8?(9"*U384

>+,-%" --e1 =g= ] =0Y

V3@:(83%?1 h(*831 -..e4 ? -5"**"5 '@ A"%2"4 X$*%F8K V(<+(8 ?$ R*<U+$*4

'$$,F8A1 R$* C41 -./i :85%;8%50/ ",& 8$0+5 @%,;8+',/ +, B/",4 "#,+$*?3#K X$j3#F8,4

'()$*+,1 k(G8 '41 -./i ?*0<04 h(8?(8K l*((# \$%#4

'()$*+,1 k(G8 '41 -./m "#$%$ ,WF+@G9*$>$*$8@$ 38? +G$ +G$(*U (> N*3##3*4 >+,-%+/8+; C,D%+52 -.1g^ ] no

p38 p3%F81 '()$*+ :41 k*41 -.._ B(W3*?, 3 ><8@+F(83%F,+ 3@@(<8+ (> ,(9@3%%$? $q+*3@+F(8 @(8,+*3F8+,4

E8 :$S*F$8?+1 X$++U1 h(<F, R((,,$8,1 b k(G38 S38 ?$* "<W$*3 &$?,42 E'*7<0F /85%;8%50/G "@%,;8+',"<+/8 705/70;8+H01 L. ] ne4 X$*%F8K V(<+(8 ?$ R*<U+$*4


27

!"

# $%&'()$ *$+,-./* 0&/1$ 2$&,(0$ 3$'$4,*- ),-, 5,0 */- ,4,.',2'$ 5.-6 0(7&.$*- ,*,'80.09 # ,'0/ (0$

, 0.:1'.;$) 4$30./* /< -6$ ;*.-$*$00 4,3.,2'$9 =6$ </''/5.*+ )$;*.-./*0 /< '$4$'0 5$3$ (0$)> '.0-.*+

/*'8 '$4$'0 -6,- ,3$ ,&-(,''8 ,--$0-$) .* -6$ ),-,2,0$?

!""#$%&'() =6$ 0&/1$ /< .''/&(-./*,38 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

FGHCDB@E? $%-$*)0 -/ $.-6$3 -6$ :,.* /3 -6$ )$1$*)$*- &',(0$ 2(- *$4$3 -/ 2/-6

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$


*#$%&'() =6$ 0&/1$ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

+,-,.(-($$) =6$ )$1$*)$*- &',(0$ .0 6$,)$) 28 , 4$32 </3: -6,- .0

PGBGEK? ,- '$,0- ,0 :,*8 &,-$+/3.$0 :(0- 2$ :,3Q$) ,0 .* :,.* &',(0$0

BABPGBGEK? /*'8 <$5$3 &,-$+/3.$0 ,3$ ,''/5$)

JBR? $.-6$3 -6$ 0,:$ 3,*+$ /3 '$00 &,-$+/3.$0 &,* 2$ :,3Q$)

!""#/012) S,3Q.*+ /< .''/&(-./*,38 </3&$ /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- /*'8 .< .- :,-&6$0 -6$ :,3Q.*+ /* -6$ :,.* &',(0$

*#/012) S,3Q.*+ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- 0(2W$&- -/ &/*0-3,.*-0 2,0$) /* -6$ -$*0$ /3 0-,-(0 &6/.&$ .* -6$

:,.* &',(0$

34//(.14) =6$ 3,*+$ /< &,-$+/3.$0 -6,- &,* 2$ $%13$00$) /* '.*Q$) &',(0$0 .0



PNKK? &,* 2$ ).X$3$*- ,*) &,* $4$* .*&'()$ $'$:$*-0 /< ).X$3$*- -81$ Y).X$3$*- 1,3-0 /<

01$$&6> &',(0$0 ,*) Z[0> $-&9\

56) ]($0-./* 5/3)0 ,*) &/*0-.-($*- </&(0 .*0.)$ )$1$*)$*- &',(0$0 ,3$

AT? ,''/5$)

MJBBKF? */- ,''/5$)

78.10%.,&-) ^%-3,&-./* /< $'$:$*-0 /< )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

+9:) _/&(0 :,3Q.*+ /* )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

;&$,.,&-) =6$ 1/0.-./* /< -6$ )$1$*)$*- &',(0$ 4.0O`O4.0 -6$ :,.* &',(0$ 5.-6 56.&6 .- $*-$30 ,

)$1$*)$*&8 3$',-./* .0

!"#$%"& #"'%$() * +,)" --. -//0

These are independent and additional variables for at least two reasons...


1.Constraints on WH may also hold in the absence of extraction (Foley & Van Valin 1984):

28

!"

#$ !"#$#%&'

%#&'()*

($#$"+,*)--

!)#.&/01'

*$)"2&'0,2301

+*,$-%+..4$045'6+,7$*-(

89, :4/ ;<1 %&'= <3, .&/01' 51'; 045' 3' ;<1 2&'0,2301$>

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

!"# $%&'( %) *'+(,-%.( -. /-.0+1 &/2'(+( 2.1 +3,)2&,-%. 4%((-5-/-,-+(

?' %&'P 2&'.E&.1, ,4%1 ;PF1, 4: 23'C10 #2&E,1, H&' ;<1 4##E//1'#1 4: IE1,;34' 54/0,$ ?; 3, ,4%16;3%1, #2&3%10 ;<&; ,E#< #4',;/&3';, #&' H1 01/3G10 Q4% .1'1/&2 1W;/&#;34' #4',;/&3';, @X4,, !NT"=S<4%,CP !N"+B 4/ Q4% 3':4/%&;34' ,;/E#;E/1 #4',;/&3';, @7421P Y Z&' Z&23' !NOA= Z&' Z&23'!NN[B= &'0 ;<&; 4' 13;<1/ &##4E';= ;<1 #4',;/&3';, 03\1/1';3&;1 H1;511' E'3G1/,&22P 01D'10 ;PF1, 4:#2&E,1 23'C&.1 @#44/03'&;34' G,$ ,EH4/03'&;34'B$ M451G1/= 3; ,11%, H1,; ;4 ;/1&; IE1,;34' &'0 :4#E,F4,,3H323;31, &, & G&/3&H21 ;<&; 3, 3'01F1'01'; 4: 1W;/&#;34' F4,,3H323;31,= &'0 '4; ;4 ;31 &'P 4: ;<1,1 ;44;<1/ G&/3&H21, 4: #2&E,1 23'C&.1 ,E#< &, 3224#E;34'&/P ,#4F1 H1<&G34/$ (<1 #/4,,623'.E3,;3# G&/3&;34'3, KE,; ;44 H3. :4/ &'P /10E#;34'3,; &FF/4&#<$ (<1/1 &/1 ;</11 F31#1, 4: 1G301'#1 :4/ C11F3'. ;<1G&/3&H21, 03,;3'#;$

73/,;= &, &/.E10 HP 7421P Y Z&' Z&23' @!NOAB &'0 Z&' Z&23' @!NN[B= & H&' 4' IE1,;34', 3'01F1'01'; #2&E,1 #&''4; H1 /10E#10 ;4 1W;/&#;34' #4',;/&3';, H1#&E,1 ,E#< & H&' %&P 4H;&3' 1G1'53;<4E; 1W;/&#;34'= 3$1$ 5<1/1 IE1,;34' 54/0, /1%&3' +# *+/"$ 7421P Y Z&' Z&23' @!NOAB &'0 Z&' Z&23'@!NN[B 01%4',;/&;1 ;<3, 53;< 0&;& Q4% ;<1 -34E&' 2&'.E&.1 ]&C<4;&= &'0 & ,3%32&/ ,3;E&;34' 4H;&3',3' (&EP&^

@+[B (&EP& @_&#`4'&20 !NNaB

&$ #%,#++,6bXc

(%5<4de)_f

/",),/%d+,*f.3G16+,6`-

0)",+1#%2d+,*f,116+Fg*9X9_b(X?S9]$h

@8ijklm>B

8n<4 030 <1 .3G1 3; ;4o &'0 ;<1P ,&5 <3%> @8n<4 030 <1 .3G1 3; ;4 5<1' ;<1P ,&5 <3%o>B

H$ p (%5<4de)_f

!%+<1/1

3$3%,),#)#+1-)#4%16+,69--g()*

4 @8qrsli>B

5#/%#6%67 8p3: 5<4 030 #4%1 <1/1t>

bW&%F21 @+[&B 3, H&,10 4' ;<1 ,&%1 8#<&3'3'.> #4',;/E#;34' ;<&; 5&, 322E,;/&;10 1&/231/ 3' @[B$ ?';<1,1 #4',;/E#;34',= +# *+/" IE1,;34' 54/0, &/1 F4,,3H21 3',301 ;<1 01F1'01'; #2&E,1 @<1&010 HP /")/%8<1 .&G1 3; ;4 ,4%14'1 &'0t> (<3, 3, 3' %3'3%&2 4FF4,3;34' ;4 8;4F3#> #2&E,1 #4',;/E#;34', 4: ;<1 C3'04H,1/G10 1&/231/ 3' @OB^ 3' ;<1,1 #4',;/E#;34',= ;<1 01F1'01'; #2&E,1 041, '4; &2245 IE1,;34' 54/0,$(<3, #4',;/&3'; 3, 3%F4,10 HP ;<1 ;PF1 4: #2&E,1 23'C&.1= &'0 041, '4; /1,E2; Q4% &'P 1W;/&#;34'#4',;/&3'; ,3'#1 IE1,;34' 54/0, &/1 '1G1/ 1W;/&#;10 3' ;<1 2&'.E&.1$

!"#$%"& #"'%$() * +,)" --. -//0


2.The domain relevant for extraction may differ from the domain relevant for WH

29

!

! "#$%$ &'()$*+, -./01 -.//2

34 5 !"-,

#"6*$

$%&'()'(*+,(7$89:;9-,

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

)4 !"-,

#"6*$

$%&(+-("(7$89-,9B4C;B

-.")<+

%-."=,

/".>((?

*",0(+("1@((A9-,9B4C;B

"#$

DE %F+ +G$ 6*$ )<+ ,G$ @((A$? +G$ >((?4H

I J,38 &'$$,F8A -./0K L/=M 2

34 *%,',#38

+2"%?(N

3454*('5=,C4GF+9=,O4C;B

0,-BPC

%*('50+?F$9=,O4C;B

)+2'6Q

$%&'(

DRFS$8 +G3+ +G$ #38 GF+ +G$ ?(N1 ?F? F+ ?F$TH

)4 245(4.@(#$9;;

0,-BPC

20(,+7"8-,C9,+$74(S$*4;;

)%5#(8$U

.+-(4517<+97EVC

$%&'(

DE> U(< @(#$1 ,+$7 (S$* #$ 38? 7<+ U(<* #(8$U &F8 +G$ )3,A$+24H

@4 *%,',#38

0,-9BPC

3','%=,CP;;

*",N3*?$8

/'"%%38?,%F?$

+/('50+1N(4?(W89=,O4C;B

D", >(* +G$ #381 GF, N3*?$8 W$8+ ?(W8 F8 +G$ %38?,%F?$4H

X YG$@G$8 &Z3AG9:3NG$,+38F38[ R((? L\\=1 V(%(@GF$S3 L\\/2

34 :";++<"V4&]24ZPV

$%WG3+

+0=(,")<U9Y^_

=>"G(<,$

#(0>"(5"6]9@(#$9 4̀C;B

D`G3+ ?F? V3%FA3 )<U 38? @3#$ G(#$TH

)4 ?""5"a4&]24ZPV

<'=$?*$,,4ZPV

0=(,")<U9Y^_

=>"G(#$

#(0>"(5"1]9@(#$9 4̀C;B

Da3*3 )(<NG+ 3 ?*$,, 38? @3#$ G(#$4H

@4 5 ?""5"@(/Aa4&]24&9b'R2

0=(,")<U9Y^_

=>"G(#$

#(0>"(=$"]9@(#$9Y^_

#(';%]9"Jc4CBYC

<'=$?*$,,

B,80,&0&C D+G$ ?*$,, +G3+ a3*3 )(<NG+ 38? @3#$ G(#$4H

_<+ +G$ ,3#$ @(8,+*<@+F(8, )%(@A $d+*3@+F(8 F8 +G$ >(*# (> *$%3+FS$ @%3<,$,K

e YG$@G$8 &V(%(@GF$S3 L\\/2

34 ?""5"a4&]24ZPV

<'=$?*$,,4ZPV

0=(,")<U9Y^_

=>"G(#$

#(0>"(5"1]9@(#$9 4̀C;B

()*'+

Da3*3 )(<NG+ 3 ?*$,, 38? @3#$ G(#$4H

)4 5 ?""5"@(/Aa4&]24&9b'R2

0=(,")<U9Y^_

=>"G(#$

#(0>"(=$"]9@(#$9Y^_

#(';%]9"Jc4CBYC

<'=$?*$,,

()*'+

B,80,&0&C D+G$ ?*$,, +G3+ a3*3 )(<NG+ 38? @3#$ G(#$4H

f _$%G3*$K g<$,+F(8 >(*#3+F(8


30

!"

# $%&'()$ *$+,-./* 0&/1$ 2$&,(0$ 3$'$4,*- ),-, 5,0 */- ,4,.',2'$ 5.-6 0(7&.$*- ,*,'80.09 # ,'0/ (0$

, 0.:1'.;$) 4$30./* /< -6$ ;*.-$*$00 4,3.,2'$9 =6$ </''/5.*+ )$;*.-./*0 /< '$4$'0 5$3$ (0$)> '.0-.*+

/*'8 '$4$'0 -6,- ,3$ ,&-(,''8 ,--$0-$) .* -6$ ),-,2,0$?

!""#$%&'() =6$ 0&/1$ /< .''/&(-./*,38 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

FGHCDB@E? $%-$*)0 -/ $.-6$3 -6$ :,.* /3 -6$ )$1$*)$*- &',(0$ 2(- *$4$3 -/ 2/-6

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$


*#$%&'() =6$ 0&/1$ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ :,.* &',(0$ .0

@ABCDB@E? $%-$*)0 -/ -6$ :,.* &',(0$ ,*) -6$ )$1$*)$*- &',(0$

IA@JI? .0 '.:.-$) -/ -6$ :,.* &',(0$

KLEKBHGMIK? $%-$*)0 -/ $.-6$3 -6$ :,.* &',(0$ ,'/*$ /3 -/ 2/-6 -6$ :,.* &',(0$ ,*) -6$

)$1$*)$*- &',(0$> 2(- *$4$3 -/ -6$ )$1$*)$*- &',(0$ ,'/*$

+,-,.(-($$) =6$ )$1$*)$*- &',(0$ .0 6$,)$) 28 , 4$32 </3: -6,- .0

PGBGEK? ,- '$,0- ,0 :,*8 &,-$+/3.$0 :(0- 2$ :,3Q$) ,0 .* :,.* &',(0$0

BABPGBGEK? /*'8 <$5$3 &,-$+/3.$0 ,3$ ,''/5$)

JBR? $.-6$3 -6$ 0,:$ 3,*+$ /3 '$00 &,-$+/3.$0 &,* 2$ :,3Q$)

!""#/012) S,3Q.*+ /< .''/&(-./*,38 </3&$ /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- /*'8 .< .- :,-&6$0 -6$ :,3Q.*+ /* -6$ :,.* &',(0$

*#/012) S,3Q.*+ /< -$*0$ /3 0-,-(0 /1$3,-/30 .* -6$ )$1$*)$*- &',(0$ .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

UJNVABG@? ,''/5$) 2(- 0(2W$&- -/ &/*0-3,.*-0 2,0$) /* -6$ -$*0$ /3 0-,-(0 &6/.&$ .* -6$

:,.* &',(0$

34//(.14) =6$ 3,*+$ /< &,-$+/3.$0 -6,- &,* 2$ $%13$00$) /* '.*Q$) &',(0$0 .0



PNKK? &,* 2$ ).X$3$*- ,*) &,* $4$* .*&'()$ $'$:$*-0 /< ).X$3$*- -81$ Y).X$3$*- 1,3-0 /<

01$$&6> &',(0$0 ,*) Z[0> $-&9\

56) ]($0-./* 5/3)0 ,*) &/*0-.-($*- </&(0 .*0.)$ )$1$*)$*- &',(0$0 ,3$

AT? ,''/5$)

MJBBKF? */- ,''/5$)

78.10%.,&-) ^%-3,&-./* /< $'$:$*-0 /< )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

+9:) _/&(0 :,3Q.*+ /* )$1$*)$*- &',(0$0 .0

AT? ,''/5$)

MJBBKF? */- ,''/5$)

;&$,.,&-) =6$ 1/0.-./* /< -6$ )$1$*)$*- &',(0$ 4.0O`O4.0 -6$ :,.* &',(0$ 5.-6 56.&6 .- $*-$30 ,

)$1$*)$*&8 3$',-./* .0

!"#$%"& #"'%$() * +,)" --. -//0

!

! "#$%#$& '()*+&) ,-.-%#+$/)0 12%23

)2 !"#$$%",245,

&'%"()"*+"67-*$89:;<==

,("-.()"/0*+"1>8?-8"@A'<B5"3

2+"#-C$

,(3+"(4"5>8%-C$8D2E=;

F,).+G) H$&7 7- 7#$ 67-*$ )&I 7#$& %)C$ J)%G #-C$2K

J2 !"#$$%",2'>3245,

.6H#)7

$32()"/0*+"1JLM8"@A'<B5"3

2+"#-L6$

,(3+"(4"7>8%-C$8D2E=;

FD#)7 I+I ,).+G) JLM )&I %)C$ #-C$NK

%2 !"#$$%",2'>3245,

.6H#)7

$32(2."*+"JLM8DOP4<B5"

2+"#-L6$

,(3+"(4"7>8%-C$8D2E=;

FD#)7 I+I ,).+G) JLM )&I %)C$ #-C$NK

!"#"$"%&"'

A+%G$.0 A).7#)6)*0 QRRS A$.#)*$ 6LJ-*I+&)7+-& )&I 7#$ 7#$-*M -T 7-1+%2 U& PJ$*70 V)*$& O2 '$I238&69$3: $) 2#"6:3 #$)%"-30 WX Y Z[ (\*+%#] :=:=2

A+%G$.0 A).7#)6)* ^ _-?$&I*) E2 _`I)/)0 Waaa2 : b$6# .--G )7 ?*)CC)7+%). *$.)7+-&6 +& U&I-8:*M)&2;$)-6" QQa0 XcX Y XdS

,)%9-&).I0 e-*&)0 QRRa2 < -4"=="4 >? @"6'"2 A$*.+&] ,-L7-& I$ f*LM7$*2

g$$6+&G0 f$* E20 QRhi 8&462&643: ")9 &.3$4 ?6)2&$>): $) A:")2 :C67$*I)C] A$j)C+&62

g-J$*760 >-#& g20 QRhi <=3#32 e-&I-&] "*--C O$.C2

g-J$*760 >-#& g20 QRhk :C$.$ 6H+7%#8*$T$*$&%$ )&I 7#$ 7#$-*M -T ?*)CC)*2 ;$)-6$:&$2 B)C6$4' QR0cZ Y lm

@)& @).+&0 g-J$*7 920 >*20 QRR[ ;-H)*I6 ) TL&%7+-&).+67 )%%-L&7 -T 6-8%)..$I $n7*)%7+-& %-&67*)+&762U& 9$/*+$&I70 A$77M0 e-L+6 f--66$&60 ^ >-#)& /)& I$* :LH$*) '$I623 D>=E#3F :&462&643:G "?6)2&$>)"#$:& E34:E32&$H30 WR Y la2 A$*.+&] ,-L7-& I$ f*LM7$*2


31

!

! "#$%#$& '()*+&) ,-.-%#+$/)0 12%23

)2 !"#$$%",245,

&'%"()"*+"67-*$89:;<==

,("-.()"/0*+"1>8?-8"@A'<B5"3

2+"#-C$

,(3+"(4"5>8%-C$8D2E=;

F,).+G) H$&7 7- 7#$ 67-*$ )&I 7#$& %)C$ J)%G #-C$2K

J2 !"#$$%",2'>3245,

.6H#)7

$32()"/0*+"1JLM8"@A'<B5"3

2+"#-L6$

,(3+"(4"7>8%-C$8D2E=;

FD#)7 I+I ,).+G) JLM )&I %)C$ #-C$NK

%2 !"#$$%",2'>3245,

.6H#)7

$32(2."*+"JLM8DOP4<B5"

2+"#-L6$

,(3+"(4"7>8%-C$8D2E=;

FD#)7 I+I ,).+G) JLM )&I %)C$ #-C$NK

!"#$%$"&' ;#$ 1-6+7+-& -Q 7#$ I$1$&I$&7 %.)L6$ /+68R8/+6 7#$ C)+& %.)L6$ H+7# H#+%# +7$&7$*6 ) I$1$&I$&%M *$.)7+-& +6STUVWXYZ[\8]^T_X +6 `a$I )&I +6 ).H)M6 )b$* 7#$ C)+& %.)L6$STUVWXYcV8]^T_X +6 `a$I )&I +6 ).H)M6 J$Q-*$ 7#$ C)+& %.)L6$SdVUTedV8^Wf^gV_\X %)& J$ J$Q-*$ -* )b$* 7#$ C)+& %.)L6$ JL7 CL67 J$ )Ih)%$&7 7-

+7SdVUTedV8cVd^\TZ_^dX %)& J$ J$Q-*$ -* )b$* 7#$ C)+& %.)L6$ )&I %)& J$ 6$1)*)7$I

i-C 7#$ C)+& %.)L6$ JM -7#$* I$1$&I$&7 %.)L6$6()*+,' ;#$ I$1$&I$&7 %.)L6$ )Ih-+&6

^W8@X 7- 7#$ 1*$I+%)7$ )&I %)& J$ %$&7$*8$CJ$II$I^W8=X 7- 7#$ %.)L6$ )&I %)&&-7 J$ %$&7$*8$CJ$II$IWV\^gjVWX 7- 7#$ %.)L6$ JL7 +6 6$1)*)7$I 6M&7)%7+%)..M )&I +&7-&)7+-&)..M

-+.+,+&/+#

A+%G$.0 A).7#)6)*0 kllm A$.#)*$ 6LJ-*I+&)7+-& )&I 7#$ 7#$-*M -Q 7-1+%2 n& PJ$*70 o)*$& O2 '$I238&69$3: $) 2#"6:3 #$)%"-30 pq r st (u*+%#X :=:=2

A+%G$.0 A).7#)6)* v w-?$&I*) E2 wxI)/)0 pyyy2 : i$6# .--G )7 ?*)CC)7+%). *$.)7+-&6 +& n&I-8:*M)&2;$)-6" kky0 qzq r q{m

,)%9-&).I0 |-*&)0 klly2 < -4"=="4 >? @"6'"2 A$*.+&X ,-L7-& I$ }*LM7$*2

~$$6+&G0 }$* E20 kl�Ä 8&462&643: ")9 &.3$4 ?6)2&$>): $) A:")2 :C67$*I)CX A$Å)C+&62

~-J$*760 >-#& ~20 kl�Ä <=3#32 |-&I-&X "*--C O$.C2

~-J$*760 >-#& ~20 kl�Ç :C$.$ 6H+7%#8*$Q$*$&%$ )&I 7#$ 7#$-*M -Q ?*)CC)*2 ;$)-6$:&$2 B)C6$4' kl0zs r ÉÑ

@)& @).+&0 ~-J$*7 920 >*20 kllt ;-H)*I6 ) QL&%7+-&).+67 )%%-L&7 -Q 6-8%)..$I $a7*)%7+-& %-&67*)+&762n& 9$/*+$&I70 A$77M0 |-L+6 }--66$&60 v >-#)& /)& I$* :LH$*) '$I623 D>=E#3F :&462&643:G "?6)2&$>)"#$:& E34:E32&$H30 pl r Éy2 A$*.+&X ,-L7-& I$ }*LM7$*2

!"

#$$% &'()'(* #+,,- !../%

01 !"#$$%"21342

&'%"()"*+"56,7(89:;<==

,("-.()">8?,8&@A

/.0,)"*+"B,,C1342<==

$01()"BDE8&@A

1+"',F(

,(0+"(2"3>8),F(8G1H=;

#IJKLMNO%

B1 !"#$$%"21342

&'%"()"*+"56,7(89:;<==

,("-.()">8?,8&@A

1+"',F(

,(0+"(2">8),F(8G1H=;

/.0,)"*+"B,,C1342<==

$01()"3BDE8&@A

#IJKLMNO%

)1 !"#$$%"21342

1+"',F(

,(0+"(2">8),F(8G1H=;

&'%"()"*+"56,7(89:;<==

,("-.()">8?,8&@A

/.0,)"*+"B,,C1342<==

$01()"3BDE8&@A

#IJKLMNO%

I20PQC0 R(*6 6, 6'( 56,7(S B,D?'6 0 B,,CS 0*- )0F( B0)C ',F(1O

;'( (T0FUP( Q* #$$)% 5',R5 6'06 0 ),*V(7B )P0D5( *((- *,6 B( 0-W0)(*6 6, 6'( F0Q* )P0D5( 6'06 Q6 7(X(75

6,Y '(7(S 6'( 5(ZD(*6Q0P ),*V(7B 5D[T ()" 5Q6D06(5 6'( (V(*6 ,X B,,C BDEQ*? Q*6, 0 -Q7()6 7(P06Q,*

6, 6'( (V(*6 (TU7(55(- Q* 6'( F0Q* )P0D5( #I),FQ*? ',F(OS '(7( Q*Q6Q0P%S *,6 Q* 6'( QFF(-Q06(PE

0-W0)(*6 )P0D5( #I?,Q*? 6, 6'( 56,7(OS '(7( 6'( 5(),*- )P0D5(%1 =D)' B('0VQ,7 Q5 670-Q6Q,*0PPE 60C(* 6,

B( 055,)Q06(- RQ6' I5DB,7-Q*06Q,*O 0*- )0* ,\(* B( ,B5(7V(- RQ6' R'06 670*5P06(5 ]*?PQ5' I0-V(7BQ0P

)P0D5(5O1

!"# $%&'( )* %++%,-.'/+

:-W,Q*(- )P0D5(5 07( *,6 (FB(--(- Q* 6'( 5(*5( ,X B(Q*? 07?DF(*65 ,X F0Q* )P0D5( U7(-Q)06(51 AD6

6'(E )0* 0660)' 6, V07Q,D5 P(V(P5 Q*5Q-( 0 F0Q* )P0D5(1 G'QP( F,7( ^*(8?70Q*(- -Q56Q*)6Q,* 07(

)(760Q*PE U,55QBP(S R'06 5((F5 6, B( 6'( F,56 QFU,760*6 -Q_(7(*)( Q5 B(6R((* ` )P0D5(5 0-W,Q*Q*?

6, 0 U7(-Q)06( ,7 V(7B #I0-8@O%S a )P0D5(5 0-W,Q*Q*? 6, (*6Q7( )P0D5(5 #I0-8=O%S ,7 bS )P0D5(5 6'06 07(

0-W,Q*(- 6, 5,F( 'Q?'(7 P(V(P 0*- 0UU(07 05 I-(60)'(-O c,F F0Q* )P0D5(5 Q* 6'( 5(*5( ,X de*Q? f

V0* -(7 :DR(70 #ghiiS ghh.%S de*Q? #ghh$% 0*- ,6'(75 R,7CQ*? ,* ]D7,U(0* )P0D5( PQ*C0?( U066(7*5

#0P5, )X1 6'( *,6Q,* ,X IP(\8-(60)'(-O 0*- I7Q?'68-(60)'(-O U,5Q6Q,* Q* j,P( 0*- j(X(7(*)( +70FF07k

@0* @0PQ* f l0H,PP0 ghhmS @0* @0PQ* !..$%1 : XD76'(7 U,55QBQPQ6E Q5 X,7 )P0D5(5 6, B( 0-W,Q*(- *,6

5, FD)' 6, 0 5(*6(*)( BD6 6, 0* D66(70*)(S 05 Q* 6'( )05( ,X 5U(()'80)6 F,-QnQ*? )P0D5(5 PQC( 42")%#'560"%$)-1

4X 0PP 6'(5( -Q56Q*)6Q,*5S 6'( ,*( 6'06 5((F5 6, B( ,X RQ-(5U7(0- 7(P(V0*)( Q5 6'06 B(6R((* 0-8@

0*- 0-8= ),*567D)6Q,*51 ;'( 670-Q6Q,*0P 6(7F I0-V(7BQ0P 5DB,7-Q*06Q,*O ),*o06(5 6'( 6R,S BD6 F0*E

P0*?D0?(5 F0C( 0 5'07U -Q56Q*)6Q,*5 #AQ)C(P ghhgS ghh/S ghhi%1 :-8@ )P0D5(5 U(7X,7F 6'( XD*)6Q,*

,X 0-V(7BQ0P F,-Q^(75k 0-8= )P0D5(5S BE ),*67056S U7,VQ-( ?(*(70P c0F(R,7C5 X,7 6'( F0Q* )P0D5(

#05 Q* IQX OS IR'(*O ,7 ?(*(70P I6,UQ)O )P0D5(5% ,7 5(ZD(*)(5 Q* 0* (V(*6 )'0Q*1 :-8@ )P0D5(5 6EUQ)0PPE

B('0V( PQC( ,7-Q*07E 0-V(7BQ0P ),*56Q6D(*65S ,\(* )05(8F07C(- PQC( 3H ),*56Q6D(*65 0*- (*6Q7(PE

670*5U07(*6 6, 6'( 055Q?*F(*6 ,X )05( 6, ,6'(7 5D)' ),*56Q6D(*651 ;'( )7Q6Q)0P 7(5DP6 ,X 0PP 6'Q5 Q5

6'06 0-8@ )P0D5(5 )0* B( )(*6(78(FB(--(-Y

#$p% A(P'07(

01 7.")%8&"91ql4&r

.$9('"%&("(#:%*&:q/5=8r56DFBP(8sHt@8H=;1=A>@8&42<t4&

%."2(03q/5=r?,8H=;

#IJuvO%

Iw( R(*6 6, 9'0*CD60 56DFBPQ*?1O

B1 6$&(1.$(#:),R8*58&42

"6(%."2(%0&3q/5=r),F(1,*16'(150F(1P(V(P8?,8;]2H

I='( Q5 U055Q*? BE RQ6' 6'( ),R51O

!"#$%"& #"'%$() * +,-. /0 1223


32

!"

#$%&'()*+,-./$0( 12 3456 786 12 79:7;2 7<5= >?5 @718 A97B25#$%&'()C&-./$0( 12 3456 786 12 79:7;2 D5EF=5 >?5 @718 A97B25#G&%$HG&-/'I/J&0,( A78 D5 D5EF=5 F= 7<5= >?5 @718 A97B25 DB> @B2> D5 76K7A58> >F 1>#G&%$HG&-C&G/,$*0/G( A78 D5 D5EF=5 F= 7<5= >?5 @718 A97B25 786 A78 D5 25L7=7>56 MF@ >?5

@718 A97B25 D; F>?5= 65L58658> A97B252!"#$%& N?5 65L58658> A97B25 76KF182

/'-O( >F >?5 L=561A7>5 786 A78 D5 A58>5=-5@D56656/'-P( >F >?5 A97B25 786 A788F> D5 A58>5=-5@D56656'&,/JQ&'( >F >?5 A97B25 DB> 12 25L7=7>56 2;8>7A>1A799; 786 18>F87>1F8799;

RF= AB==58> LB=LF252 S >=57> 799 >?525 T7=17D952 72 B82>=BA>B=56 @B9>18F@179 T7=17D952U N?12 127 21@L913A7>1F8 218A5 18 2F@5 A7252V >?5=5 @7; D5 78 B865=9;18W 2>=BA>B=5 X EF= 547@L95V Y9FA79Z786 Y54>5821D95Z 2AFL5 7=5 7=WB7D9; @F=5 21@197= >F 57A? F>?5= >?78 51>?5= 12 >F Y612KB8A>Z F= YAF8-KB8A>Z 2AFL5[ F= >?5 95T592 YD78856ZV Y?7=@F81AZV 786 YF\Z AFB96 D5 @F65956 18 >5=@2 FE 65W=552 FE=52>=1A>1T58522V 1U5U 72 7 =78\ T7=17D95U] S 957T5 54L9F=7>1F8 FE 2BA? LF221D191>152 >F EB>B=5 =5257=A?U

'() *%+,+,#-$. "/0 12"3$.

^ \5; _B52>1F8 12 >F :?7> 54>58> >?5=5 7=5 A=F22-918WB12>1A A9B2>5=2 F= YL=F>F>;L52Z FE A97B25 918\7W5L7>>5=82 X >?7> 12V >;L52 FE 2>=BA>B=52 >F :?1A? 978WB7W5-2L5A13A AF82>=BA>1F82 D57= 21@197=1>; >F7 _B78>137D95 65W=55U NF 386 FB>V :5 8556 @5>?F62 EF= @572B=18W 21@197=1>; D5>:558 1861T16B79AF82>=BA>1F82 786 EF= 7WW=5W7>18W >?5 21@197=1>152 FE 799 AF82>=BA>1F82 7> F8A5U

'()(4 5$,6+0.

^ 2>7867=6 @572B=5 FE 21@197=1>; B256 18 @78; 35962 12 >?5 18T5=25 =597>1T5 `7@@18W 612>78A5 a792F\8F:8 72 >?5 bF:5= AF5cA158>[ 255 d7BE@78 e fFB2255B: ghhij( >?5 612>78A5 F= 61221@197=1>;D5>:558 >:F AF82>=BA>1F82 12 5_B79 >F >?5 L=FLF=>1F8 FE 61k5=58> 95T592 18 799 8F8-5@L>; T7=17D952URF= 547@L95V 7AAF=618W >?5 L19F> 67>7D725 18 >?5 ^LL58614V >?5 61k5=58A5 D5>:558 >?5 ^@595 YDB>Z-AF82>=BA>1F8 786 YA?718Z-AF82>=BA>1F8 12 lm" n Uo!l 218A5 >?5=5 7=5 51W?> 8F85@L>; a8F8-p^j T7=1-7D952 786 3T5 FE >?5@ ?7T5 61k5=58> 95T592U N?5 612>78A5 D5>:558 >?5 ^@595 YDB>Z-AF82>=BA>1F8786 >?5 q59?7=5 Y786Z-AF82>=BA>1F8 12 2@7995=V T1rU sm" n UstlV 786 >?12 A7L>B=52 >?5 18>B1>1F8 >?7>>?525 AF82>=BA>1F82 7=5 =597>1T59; 21@197= >F 57A? F>?5=U PF@5 2>=BA>B=52 @7; 5T58 D5 1658>1A79U RF=547@L95V MF@ 799 :5 \8F:V u?78>;79 786 qB=v2?72\1 YA?718Z-AF82>=BA>1F82 ?7T5 1658>1A79 L=FL5=>15218 799 8F8-5@L>; T7=17D952[ >?12 =52B9>2 18 7 =597>1T5 `7@@18W 612>78A5 FE iU

P18A5 >?5 67>7D725 AF8>7182 oh AF82>=BA>1F82V AF@LB>18W `7@@18W 612>78A52 FE 799 L71=2 FE

AF82>=BA>1F82 =52B9>2 18 7 @7>=14 FE oh·(oh−g)! n!Vswo L71=2V 21@197= 18 \186 >F 612>78A5 A?7=>2 EFB86

18 =F76 @7L2U N?5 A?79958W5 12 8F: >F 386 L7>>5=82 18 >?525 !Vswo L71=2 D; 257=A?18W EF= A9B2>5=2

] N?78\2 >F x5k bFF6 786 OF9\5= b72> EF= =5@18618W @5 FE >?12U

!"#$%"& #"'%$() * +,-. /0 1223

!

! "#$%&'#

&( !"#$%&'#)(*+,-.

"()*+#%'*#*,-%*/012.03456$#2789:281;(1"<:2-,=

%"#.*/0*/01.>?281;

"#$

@A# B#C3 3? )%&CD43& 03456$EC>(F

6( &*1"-)/08,112G#0E'#

2-%3$##*/01.'E0#*1"<:.H;,8

!"#$%&'#)(*+,-.

%"#4*+&0*/01.>?2I81;

J( K!"#$%&'#)(*+,-.

&*1"-)/08,112G#0E'#

2-%3$##*/01.J?5#(4L*1"<:.H;,8

%"#4*+&0*/01.>?2I81;

@7M %# B&C30N %# BE$$ >? 3? )%&CD43&(F

!"#"$"%&"'

"EJD#$N "&$3%&0&'N OPPQ "#$%&'# 046?'GEC&3E?C &CG 3%# 3%#?'R ?M 3?LEJ( 7C S6#'3N T&'#C A( U#G(V

5'&6(/7 ($ 1,#&7/ ,($%#8/N W/ X YZ [\'EJ%] ^1^1(

"EJD#$N "&$3%&0&' _ ?̀>#CG'& 8( `aG&b&N Wccc( ^ d#0% $??D &3 >'&55&3EJ&$ '#$&3E?C0 EC 7CG?2^'R&C(

9($8&# OOcN /e/ X /fQ

g??GN <#hN WccQ -$&40# J?56ECEC> EC -%#J%#C( 5'&6(/7 ($ 9#$8&#8/ WfN OO/ X Ofc(

=&J)?C&$GN +?'C&N OPPc( : 8.#))#. -; <#&+#( "#'$EC] =?43?C G# g'4R3#'(

i##0ECDN g#' 8(N OPjk 5'.&1'&./7 #$6 '"/(. ;&$1'(-$7 ($ =7#$( ^503#'G&5] "#l&5EC0(

i?6#'30N <?%C i(N OPjk :)/,/( +?CG?C] -'??5 A#$5(

i?6#'30N <?%C i(N OPjm ^5#$# 0BE3J%2'#M#'#CJ# &CG 3%# 3%#?'R ?M >'&55&'( 9($8&(7'(1 >$?&(.+ OPNeY X no

:&C :&$ECN i?6#'3 )(N <'(N OPPZ ;?B&'G0 & M4CJ3E?C&$E03 &JJ?4C3 ?M 0?2J&$$#G #p3'&J3E?C J?C03'&EC30(

7C )#b'E#CG3N "#33RN +?4E0 g??00#C0N _ <?%&C b&C G#' ^4B#'& U#G0(V @-)2,/A 7'.&1'&./7B #;&$1'(-$#,(7' 2/.72/1'(C/N WP X nc( "#'$EC] =?43?C G# g'4R3#'(


•Further variables

•semantic relation between propositions

•syndesis and prosody

•cross-clausal coreference constraints (backward anaphora, switch-reference, logophorics etc)

33

Levinson et al. 2003, Cysouw 2007, Croft & Poole 2008, Bickel in press, etc.

But what about the good old typological generalizations?

•... can’t just get them as well, but better!

•For this we need...

A. Heuristics: data mining techniques, e.g.

•distance-based techniques for finding clusters of similar structures (split graphs, multidimensional scaling etc.) (‘prototypes’, bottom-up ‘canons’)

•entropy-based techniques for finding associations between variables (possibly weighted)

35

Cysouw 2003, Janssen et al. 2006, Bickel 2008, 2010

But what about the good old typological generalizations?

B. Test strategies: permutation-based statistics, e.g. of

•simple contingency tables

•generalized linear models

36

= next topic. For now, let’s explore some data mining techniques useful for comparing structures

R

Mining clause linkage data

•Are there any structures across languages that are more similar to each other than to others?

•A standard answer: compute the relative Hamming distance (proportion of different values in a set of variables)

37

Mining clause linkage data: measuring distances

38

Mining clause linkage data: measuring distances

39

Two issues

1.NAs reduce comparability (but perhaps we can extrapolate).

2.Variables with language-specific content (e.g. “Marker”, “Language”) are not interesting for this, they just increase the baseline distance

Mining clause linkage data: aggregating distances

3.

•How can we detect general patterns about what’s more similar to each other (e.g., having “low average distances”)?

•Aggregate and plot it!

•But there is a geometrical problem:

•DIST(A,B)=1, DIST(A,C)=1, DIST(B,C)=2/3

40

!"

#$ %&%'%()* )+,) ,-% .&#*%- )#/%)+%- ,.-#** ,&& 0,1-21*% 31*),(.%*456 7(% 2,8 #$ /#1(/ ,9#:) )+1* 1*98 0&#))1(/ ,&& 31*),(.%* ,) #(.% 1( , /-,0+4 ;#2%<%-= 31*),(.%* 1( , >" ·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

0.667

0.334

0.334

A

B

C

1

D?

C

B

!"# T1'0&% /-,0+

2

0.6670.084

0.25

0.167

0.084

0.167

A

C

D

B

!$# T0&1) /-,0+

%&'()* +, Y%0-%*%()1(/ ,&& 0,1-21*% 31*),(.%* 9%)2%%( $#:- 1)%'* 1( )2#A31'%(*1#(,& *0,.%4

B( ,&)%-(,)1<% *#&:)1#(= 2+1.+ P ,3#0) 1( )+% $#&&#21(/= .#(*1*)* 1( *1':&),(%#:*&8 0&#))1(/.#'0%)1(/ 31*),(.%* 1( 2+,) 1* U(#2( ,* , *0&1) /-,0+ HI,(3%&) Z Q-%** L""!J4 [+1* 1* *+#2( 1(C1/:-% L9= 2+%-% )+% &1(%* *+#2 )+% 0#**19&% 0,)+* #(% .,( ),U% 2+1&% -%*0%.)1(/ )+% )-:% 31*),(.%*\%,.+ )-:% 31*),(.% 9%)2%%( %&%'%()* 1* )+% *:' #$ )+% 1(31.,)%3 H-#:(3%3J &%(/)+* #( )+% *+#-)%*).#((%.)1(/ 0,)+4 B* , -%*:&)= )+% &%** 31*),(.%* H14%4 31**1'1&,-1)1%*J ,-% 1( .#(]1.)= )+% &%** )+%-%,-% *0&1)* ,(3 %&%'%()* .&:*)%- 1()# .&%,- /-#:0*4 P$ )+%-% 1* '#-% .#(]1.)= )+%-% ,-% '#-% *0&1)*=*:.+ ,* )+% 0,-,&&%&#/-,' 1( C1/:-% L9= ,(3 )+% %<13%(.% $#- .&:*)%-* 1* 2%,U%-\ 1( C1/:-% L9= $#-%?,'0&%= )+%-% 1* (# 0,-)1.:&,- .&:*)%- %?.%0) , )-%(3 )#2,-3* *%0,-,)1(/ I ,(3 Q ^#' ,&& %&*% H*1(.%DEFGHI=QJ 1* *',&&%- )+,( ,(8 #)+%- 31*),(.% 1( )+% %?,'0&%J4 T0&1) /-,0+* #$ )+1* U1(3 .,( 9%

56_?.%0) 2+%-% (#)%3= ,&& .#'0:),)1#(* +%-% ,(3 1( )+% $#&&#21(/ 2%-% 3#(% 1( HY Q%<%&#0'%() M#-% [%,' !WW@J= 21)+)+% ,331)1#(,& 0,.U,/%* HS,%.+&%- %) ,&4 !WW`J= HS%8%- %) ,&4 !WW",J= H;,:**%- Z T)-1''%-!WW"9J= ,(3 HS%8%- %) ,&4 !WW"9J4

!"#$%"& #"'%$() * +,)" --. -//0

·( − ) = ,


•now add D: dist(A,D)=1, dist(B,D)=1/3, dist(C,D)=1/2

41

0.667

0.334

0.334

A

B

C

1

D?

C

B

A B CBCD

1.0001.000 0.667 1.000 0.333 0.500


42

A B CBCD

1.0001.000 0.667 1.000 0.333 0.500

Hierarchical Clustering:stepwise grouping, minimizing distances between groups (e.g. via averages)

Multidimensional Scaling:find (algorithmically) lowest dimensionality with least distortion

A

B D

C

0.3

0.6

0.9

Clustering based on average dissimilarities

-0.2 0.0 0.2 0.4 0.6

-0.4

-0.2

0.0

0.2

A

B

C

D

Bandelt & Dress 1992, Bryant & Moulton 2004


•A more informative alternative: split graph methods, e.g. NeighborNet:

43

!"

#$ %&%'%()* )+,) ,-% .&#*%- )#/%)+%- ,.-#** ,&& 0,1-21*% 31*),(.%*456 7(% 2,8 #$ /#1(/ ,9#:) )+1* 1*98 0&#))1(/ ,&& 31*),(.%* ,) #(.% 1( , /-,0+4 ;#2%<%-= 31*),(.%* 1( , >" ·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

0.667

0.334

0.334

A

B

C

1

D?

C

B

!"# T1'0&% /-,0+

2

0.6670.084

0.25

0.167

0.084

0.167

A

C

D

B

!$# T0&1) /-,0+

%&'()* +, Y%0-%*%()1(/ ,&& 0,1-21*% 31*),(.%* 9%)2%%( $#:- 1)%'* 1( )2#A31'%(*1#(,& *0,.%4

B( ,&)%-(,)1<% *#&:)1#(= 2+1.+ P ,3#0) 1( )+% $#&&#21(/= .#(*1*)* 1( *1':&),(%#:*&8 0&#))1(/.#'0%)1(/ 31*),(.%* 1( 2+,) 1* U(#2( ,* , *0&1) /-,0+ HI,(3%&) Z Q-%** L""!J4 [+1* 1* *+#2( 1(C1/:-% L9= 2+%-% )+% &1(%* *+#2 )+% 0#**19&% 0,)+* #(% .,( ),U% 2+1&% -%*0%.)1(/ )+% )-:% 31*),(.%*\%,.+ )-:% 31*),(.% 9%)2%%( %&%'%()* 1* )+% *:' #$ )+% 1(31.,)%3 H-#:(3%3J &%(/)+* #( )+% *+#-)%*).#((%.)1(/ 0,)+4 B* , -%*:&)= )+% &%** 31*),(.%* H14%4 31**1'1&,-1)1%*J ,-% 1( .#(]1.)= )+% &%** )+%-%,-% *0&1)* ,(3 %&%'%()* .&:*)%- 1()# .&%,- /-#:0*4 P$ )+%-% 1* '#-% .#(]1.)= )+%-% ,-% '#-% *0&1)*=*:.+ ,* )+% 0,-,&&%&#/-,' 1( C1/:-% L9= ,(3 )+% %<13%(.% $#- .&:*)%-* 1* 2%,U%-\ 1( C1/:-% L9= $#-%?,'0&%= )+%-% 1* (# 0,-)1.:&,- .&:*)%- %?.%0) , )-%(3 )#2,-3* *%0,-,)1(/ I ,(3 Q ^#' ,&& %&*% H*1(.%DEFGHI=QJ 1* *',&&%- )+,( ,(8 #)+%- 31*),(.% 1( )+% %?,'0&%J4 T0&1) /-,0+* #$ )+1* U1(3 .,( 9%

56_?.%0) 2+%-% (#)%3= ,&& .#'0:),)1#(* +%-% ,(3 1( )+% $#&&#21(/ 2%-% 3#(% 1( HY Q%<%&#0'%() M#-% [%,' !WW@J= 21)+)+% ,331)1#(,& 0,.U,/%* HS,%.+&%- %) ,&4 !WW`J= HS%8%- %) ,&4 !WW",J= H;,:**%- Z T)-1''%-!WW"9J= ,(3 HS%8%- %) ,&4 !WW"9J4

!"#$%"& #"'%$() * +,)" --. -//0

0.667

0.334

0.334

A

B

C

1

D?

C

B

Amele .but

German.cause.detached

Tauya.topic

Kate.TOP

Belhare.chain

English.go.andUsan.causeAmele.purp

Tauya.inconseqKorafe.chain

Usan.topic

Russian.cvb.detached

English.cvb.detachedEnglish.cvb.adv

German.cvb.adV

German.purp

Belhare.com

Chechen.cvb.temp

Greek..Ancient. .part.coni.

Chechen.chain

Hua.a l ter . i ter

Godie.chain

Swahili .chainSwahil i .and.INF

Turkish.chainAmele.chain

Kate.serialKate.chain

Belhare.inconseq

Burushaski.chainUsan.chain.swr

Nepali .cvbBelhare.cvbBelhare.purp

Fore.chain

Hua.ds.chain

Lenakel.chain

English.fin.subYup.ik. .Central. .sub

Nepali . topic

Amele.cond

English.although

German.and

Belhare.andChechen.andAmele.or

Hua.inconseq

Hua.cond

Toura.chain

Hua.ss.chain

Newar. .Dolakha. .chain

Russian.and

German.cond.detached

Wambule.chain

Hua.topic

English.to.ing.detachedRussian.cvb.adV

Swahili . topicGerman.cvb

Turkish.serial

Chantyal.chain

Nepali .chain

Tauya.chain

Chechen.cvb

Belhare.sub

Godie.topic

German.sub

English.and

Swahil i .and

0.1

Neighbor-Net of clause linkage types (69 constructions, 24 languages, 11 variables)

Mining clause linkage data: type clusters (“prototypes”)

44

Mining clause linkage data: associations

•Many concepts of traditional metalanguages are absolute universals in disguise

•E.g., if S stands for ‘sentence (verb plus all arguments)’ and [S1 S2] represents subordination, then (as for example in RRG): [ [S1] S2]=ILL

therefore,

(1) ILL scope is not shared (disjunct)

(2) WH is banned

•(1) and (2) define the structure (“are diagnostic of it”), and

•the structure “predicts” (1) and (2), and (1) ↔ (2)

45


•Problem: Perhaps. Perhaps not!

•Unless the structure is “given” (descriptive A priori = ontological A priori!), we need empirical evidence.

•But rather than relying on a single datapoint, we want to estimate the degree to which the structure is supported from many datapoints.

•We know that single datapoints are prone to error!

•We will never know the truth, but we can try and separate signal from noise, statistically.

•Quantitative, rather than qualitative thinking.

46

•Entropy of X with estimated value probabilities px1...pxk:

•Joint entropy:

•Mutual information:

•Predictability of X given Y (‘Y → X’):

Hausser & Strimmer 2009, Margolin et al. 2006


47

( ) = log

= − log( )

( ) =−∑

= ( · ) · log( )

( ) = −∑

=

( ) · log( ( ))

=( − )

=−√ =

− µ̂√µ̂


•Estimate predictabilities between all pairs of variables

•Reduce triplets via weakest link deletion:

•Examine those with non-zero predictability:

48

!"

# $%&'()*+, -+, ./*0%1, 2'3 422/% 5*) 6%7%*18 9:;8)%5 !("|#)< ,2;=2/>+1*, %) 21?

@A""BCD E%*80*%1F EGH I 5*) !($|%)< ',& 8+7+91 J H I 7*% 2'; J H E< 7+K%*

!($|&) < !(%|&)? L2,, *8) J H I %*, I/)%32() &%/ M%N% J H E H ID 218+

8%)O%, 7*/ !($|&) = "F

I

E

J I

E

J

# E*1&',> -+, M+,)*,>%,O)23%1, O' 6%&%/ >%3',&%,%, 022/7%*8%, M+//%12)*+, ',&

I,21P8% &%/ 4%2/8+,Q$%8*&'%, @'AQE%*)/2> 6%&%/ R%11%D (),* = +),*−,̂),*√,̂),*

C O'/ E%8)*5Q

5',> &%/ S2/*2K1%,7%/)%< &*% &*% M+//%12)*+, -%/'/82;%,D O7%* E%*80*%1%F

-2.65

-2.00

0.00

2.00

3.80

Pearson

residuals:

p-value =1.0084e-08

blocked

flexible-relational

flexible-adjacent

pre-main

post-main

conjunct constraint-free disjunct

@2C T11+(')*+,88(+0'8 H U2)O8)%11',>

-2.03

0.00

2.00

Pearson

residuals:

p-value =0.0013239

post-main

ok

banned

pre-main flexible-adjacent flexible-relational

@KC U2)O8)%11',> HVWXY+('8 *5 ZU

IKK*1&',> AF $%8*&'%,2,21P8% @Y1[;%, 8*,& 0/+0+/)*+,21 O' &%, $%8*&'%,D +K%/921K E%Q

O'>81*,*%F 5%9/ 218 ',)%/ &%/ Z'119P0+)9%8% %/72/)%)D ',)%/921K &%/ E%O'>81*,*%F 7%,*>%/

218 %/72/)%)D U;2N*%/',>F *,&*-*&'%11% R%118*>,*\(2,O< 5*) - = ."]D ->1? =%P%/ %) 21?

A""B?C


49

π̂( | ) π̂( | )

↔→ →

→→↔ ↔

↔↔ ↔

↔↔ ↔

↔→↔ ↔→ →

↔ ↔↔

↔ ↔↔

↔ ↔→

But which values of these variables (features of these attributes) are associated?


•Pearson residuals:

•Expected rows: 50% each:

•Absolute residual (O-E)2 = (5-2.5)2 and (10-7.5)2

•Relative residual:

50

[,1] [,2] sum

[1,]

[2,]

sum

5 0 5

5 10 15

10 10 20

( ) = log

= − log( )

( ) =−∑

= ( · ) · log( )

= −∑

=

· log( )

=( − )

=−√ =

− µ̂√µ̂

[,1] [,2]

[1,]

[2,]

1.58 -1.58

-0.91 0.91[,1] [,2] sum

[1,]

[2,]

sum

2.5 2.5 5

7.5 7.5 15

10 10 20


•Association plots:

51

[,1] [,2]

[1,]

[2,]

1.58 -1.58

-0.91 0.91[2,]

[1,]

[,1] [,2]

Association plot, showing Pearson residuals


52

!"

Extraction

Illo

cu

tion

ary

fo

rce m

ark

ing

ok

harmonic

banned

blocked possible

!"# #$$%&'() → *+,('-,./0

Constituent question or focus

Illo

cu

tio

na

ry s

co

pe

local

extensibledisjunct

constraint-free

conjunct

banned ok

!$# #$$%1-/2* → 34

%&'()* +, 5*1.67'8 '0'891.1: ;/+ <*.=<,1 '(* 2(/2/(,./0'8 ,/ ,<* (*1.67'81 '06 ;/+ >.6,< ,/ ,<*07&;*( /? 6','2/.0,1@ ,<* 6.(*-,./0 /? ;/+*1 A/& ,<* ;'1*8.0* 1.=0'81 ,<*.( 1.=0 B('.1.0= ;/+*1?/( 2/1.,.C*@ ?'88.0= ;/+*1 ?/( 0*=',.C* (*1.67'81D@ 1<'6.0= .06.-',*1 ,<', (*1.67'81 '(* .06.C.67'8891.=0.E-'0, 706*( ' 2*(&7,',./0 ,*1,@ >.,< 8.=<, =(*9 ', ' FG '06 6'() =(*9 ', ' HG 1.=0.E-'0-* 8*C*8

B1** I*9*( *, '8J KLLMDJ N<* /(6*( /? 8*C*81 .0 ,<* 28/, .1 '(;.,('(9 B'82<';*,.-'8DJ

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

!"#$%"& #"'%$() * +,)" --. -//0

Examples

•Significantly more frequent than expected under H0:

No scope constraint, WH ok

53

!

!" !"#$ %&' (& )*)% )+,- .-,)/#0( #123 45-,6'55&6#0( ,#17,- !%&' *,01 )*)%3 &- !%&'.-&/, #138

9 :,;7)-, 4:#</,; =>>?8

)@ #$%"&'A?6BC(&DEBF

#(BGH

)*++%67&,6AIJKC

,,&(-&.$*++&$'&.%?6LD.'%D.-#0(@M&-DEBFDN6E

(/H

!"#$ 67, (& A17,-,C )0$ .'% %&' 67&,623!" !"#$ 67, .'% %&' 67&,6 *7,0 67, *,01 A17,-,C23 45-,6'55&6#0( ,#17,- !67,*,013 &- !67, .&'(7138

.@ 0%(+%";#(71,-

$'-'*7,-,

0'1+&$'&.%17-&*DEBFDN6L

#(BGH

230%(O)1<7,6

%4&+&*&.%/;#(71DIEBFD?6EDN6L

!P7,-, $#$ %&' 17-&* 17, ;#(71,- 6& 17)1 %&' 7)Q, 1& '6, O)1<7,623

!"#"$"%&"'

:#</,;R :);17)6)-R =>>S :,;7)-, 6'.&-$#0)1#&0 )0$ 17, 17,&-% &M 1&5#<@ T0 G.,-1R U)-,0 V@ 4,$@85+*6('2 (- 70%*2' 0(-#%.'R N? W XY Z[-#<7\ LBLB@

:#</,;R :);17)6)- ] &̂(,0$-) E@ ^_$)Q)R N```@ L a,67 ;&&/ )1 (-)OO)1#<); -,;)1#&06 #0 T0$&DL-%)0@8(-.*% ==`R ?b? W ?cS

K)<"&0);$R d&-0)R =>>`@ 9 ."%44%" !: ;%*<%@ :,-;#0\ K&'1&0 $, e-'%1,-@

f&.,-16R g&70 f@R =>hi 94'0'@ d&0$&0\ j-&&O V,;O@

f&.,-16R g&70 f@R =>hk LO,;, 6*#1<7D-,M,-,0<, )0$ 17, 17,&-% &M (-)OO)-@ 8(-.*(2+(7 =->*("< =>RbX W lm

!"

Extraction

Illo

cu

tio

na

ry f

orc

e m

ark

ing

ok

harmonic

banned

blocked possible

!"# #$$%&'() → *+,('-,./0


Illo

cutionary

scope

local

extensibledisjunct

constraint-free

conjunct

banned ok

!$# #$$%1-/2* → 34

%&'()* +, 5*1.67'8 '0'891.1: ;/+ <*.=<,1 '(* 2(/2/(,./0'8 ,/ ,<* (*1.67'81 '06 ;/+ >.6,< ,/ ,<*07&;*( /? 6','2/.0,1@ ,<* 6.(*-,./0 /? ;/+*1 A/& ,<* ;'1*8.0* 1.=0'81 ,<*.( 1.=0 B('.1.0= ;/+*1?/( 2/1.,.C*@ ?'88.0= ;/+*1 ?/( 0*=',.C* (*1.67'81D@ 1<'6.0= .06.-',*1 ,<', (*1.67'81 '(* .06.C.67'8891.=0.E-'0, 706*( ' 2*(&7,',./0 ,*1,@ >.,< 8.=<, =(*9 ', ' FG '06 6'() =(*9 ', ' HG 1.=0.E-'0-* 8*C*8

B1** I*9*( *, '8J KLLMDJ N<* /(6*( /? 8*C*81 .0 ,<* 28/, .1 '(;.,('(9 B'82<';*,.-'8DJ


!"#$%"& #"'%$() * +,)" --. -//0

Examples

•Significantly more frequent than expected under H0:

Local scope, no WH

54

!

!" !"#$ %&' (& )*)% )+,- .-,)/#0( #123 45-,6'55&6#0( ,#17,- !%&' *,01 )*)%3 &- !%&'.-&/, #138

9 :,;7)-, 4:#</,; =>>?8

)@ #$%"&'A?6BC(&DEBF

#(BGH

)*++%67&,6AIJKC

,,&(-&.$*++&$'&.%?6LD.'%D.-#0(@M&-DEBFDN6E

(/H

!"#$ 67, (& A17,-,C )0$ .'% %&' 67&,623!" !"#$ 67, .'% %&' 67&,6 *7,0 67, *,01 A17,-,C23 45-,6'55&6#0( ,#17,- !67,*,013 &- !67, .&'(7138

.@ 0%(+%";#(71,-

$'-'*7,-,

0'1+&$'&.%17-&*DEBFDN6L

#(BGH

230%(O)1<7,6

%4&+&*&.%/;#(71DIEBFD?6EDN6L

!P7,-, $#$ %&' 17-&* 17, ;#(71,- 6& 17)1 %&' 7)Q, 1& '6, O)1<7,623

R F)'%) 4K)<"&0);$ =>>S8

)@ -'-?5

4'(7,-,

4!4*-'&(&-%-(5"%6#1D?5DLBBTFJE

1!6'(&+(1);/DUJV

-'-5+*&'5-%'/?5T(#Q,DNTEJWLX@H

"#$%&

!F7,% 6)1 7,-, )0$Y.'1 $#$ %&' 1);/ 1& 17,O23!" !B#0<, 17,% 6)1 7,-,Z $#$ %&' 1);/ 1& 17,O23

.@ [ 7'*7&AIJKC

4'(7,-,

6!6'&%&-%-(5"%<&O,D?6DLBBTFJE

8 "#$%&

9-+'-:':; ![#M *7& $#$ <&O, 7,-,\3

!"#"$"%&"'

:#</,;Z :);17)6)-Z =>>] :,;7)-, 6'.&-$#0)1#&0 )0$ 17, 17,&-% &M 1&5#<@ ^0 G.,-1Z _)-,0 `@ 4,$@8<+*:('2 (- =0%*2' 0(-#%.'Z N? a bc de-#<7f LBLB@

:#</,;Z :);17)6)- g h&(,0$-) E@ hi$)Q)Z NSSS@ L j,67 ;&&/ )1 (-)OO)1#<); -,;)1#&06 #0 ^0$&DL-%)0@>(-.*% ==SZ ?k? a ?l]

K)<"&0);$Z W&-0)Z =>>S@ ? ."%44%" !6 @%*A%@ :,-;#0f K&'1&0 $, m-'%1,-@

X&.,-16Z n&70 X@Z =>op ?4'0'@ W&0$&0f U-&&O `,;O@

X&.,-16Z n&70 X@Z =>oq LO,;, 6*#1<7D-,M,-,0<, )0$ 17, 17,&-% &M (-)OO)-@ >(-.*(2+(= 9-B*("A =>Zkb a rs

!"

Extraction

Illo

cutio

na

ry fo

rce m

ark

ing

ok

harmonic

banned

blocked possible

!"# #$$%&'() → *+,('-,./0


Illo

cutionary

scope

local

extensibledisjunct

constraint-free

conjunct

banned ok

!$# #$$%1-/2* → 34

%&'()* +, 5*1.67'8 '0'891.1: ;/+ <*.=<,1 '(* 2(/2/(,./0'8 ,/ ,<* (*1.67'81 '06 ;/+ >.6,< ,/ ,<*07&;*( /? 6','2/.0,1@ ,<* 6.(*-,./0 /? ;/+*1 A/& ,<* ;'1*8.0* 1.=0'81 ,<*.( 1.=0 B('.1.0= ;/+*1?/( 2/1.,.C*@ ?'88.0= ;/+*1 ?/( 0*=',.C* (*1.67'81D@ 1<'6.0= .06.-',*1 ,<', (*1.67'81 '(* .06.C.67'8891.=0.E-'0, 706*( ' 2*(&7,',./0 ,*1,@ >.,< 8.=<, =(*9 ', ' FG '06 6'() =(*9 ', ' HG 1.=0.E-'0-* 8*C*8

B1** I*9*( *, '8J KLLMDJ N<* /(6*( /? 8*C*81 .0 ,<* 28/, .1 '(;.,('(9 B'82<';*,.-'8DJ


!"#$%"& #"'%$() * +,)" --. -//0

Examples

•A bit more frequent than expected under H0:

Disjunct scope, no WH

and many languages with WH extraction (English among them)

55

!

!" !"#$ %&' (& )*)% )+,- .-,)/#0( #123 45-,6'55&6#0( ,#17,- !%&' *,01 )*)%3 &- !%&'.-&/, #138

9 :,;7)-, 4:#</,; =>>?8

)@ #$%"&'A?6BC(&DEBF

#(BGH

)*++%67&,6AIJKC

,,&(-&.$*++&$'&.%?6LD.'%D.-#0(@M&-DEBFDN6E

(/H

!"#$ 67, (& A17,-,C )0$ .'% %&' 67&,623!" !"#$ 67, .'% %&' 67&,6 *7,0 67, *,01 A17,-,C23 45-,6'55&6#0( ,#17,- !67,*,013 &- !67, .&'(7138

.@ 0%(+%";#(71,-

$'-'*7,-,

0'1+&$'&.%17-&*DEBFDN6L

#(BGH

230%(O)1<7,6

%4&+&*&.%/;#(71DIEBFD?6EDN6L

!P7,-, $#$ %&' 17-&* 17, ;#(71,- 6& 17)1 %&' 7)Q, 1& '6, O)1<7,623

R F)'%) 4K)<"&0);$ =>>S8

)@ -'-?5

4'(7,-,

4!4*-'&(&-%-(5"%6#1D?5DLBBTFJE

1!6'(&+(1);/DUJV

-'-5+*&'5-%'/?5T(#Q,DNTEJWLX@H

"#$%&

!F7,% 6)1 7,-, )0$Y.'1 $#$ %&' 1);/ 1& 17,O23!" !B#0<, 17,% 6)1 7,-,Z $#$ %&' 1);/ 1& 17,O23

.@ [ 7'*7&AIJKC

4'(7,-,

6!6'&%&-%-(5"%<&O,D?6DLBBTFJE

8 "#$%&

9-+'-:':; ![#M *7& $#$ <&O, 7,-,\3

] W)/7&1) 4B#&')0Z )̂0 )̂;#0 =>>_8

)@ <=>#%$&(

#("GF

+?#*`I"GaYPb

@%A+?#%A?6c?6DC.#1,

$'/H

!P7)1 $#$ 17, $&( .#1,23 !"; !"#$ 17, $&( .#1, 6&O,17#0(23

.@ 7(B$?<%O)0

#("GF

+?#*,)1

@C+'`I"GaYPb

'B$C$DE*7#;,

+$%&7FB*?6EJBBD*#M,

#("GF

4-F*)1,-

(#FB(B*(,1@M&-

$'/H

!P7#;, 17, O)0 *)6 ,)1#0( 6&O,17#0(Z $#$ 7#6 *#M, (,1 *)1,- M&- 7#O23-!+; ![P7)1 $#$ 7#6 *#M, (,1 7#O *)1,-Z *7#;, 17, O)0 *)6 ,)1#0(23

!"#"$"%&"'

:#</,;Z :);17)6)-Z =>>d :,;7)-, 6'.&-$#0)1#&0 )0$ 17, 17,&-% &M 1&5#<@ `0 G.,-1Z e)-,0 b@ 4,$@8G+*:('2 (- H0%*2' 0(-#%.'Z N? f _g hi-#<7j LBLB@

:#</,;Z :);17)6)- k l&(,0$-) E@ lm$)Q)Z NSSS@ L n,67 ;&&/ )1 (-)OO)1#<); -,;)1#&06 #0 `0$&DL-%)0@I(-.*% ==SZ ?o? f ?pd

K)<"&0);$Z W&-0)Z =>>S@ J ."%44%" !6 K%*@%@ :,-;#0j K&'1&0 $, q-'%1,-@

X&.,-16Z r&70 X@Z =>st J4'0'@ W&0$&0j U-&&O b,;O@

X&.,-16Z r&70 X@Z =>su LO,;, 6*#1<7D-,M,-,0<, )0$ 17, 17,&-% &M (-)OO)-@ I(-.*(2+(H 9-L*("@ =>Zo_ f vw

!"

Extraction

Illo

cutionary

forc

e m

ark

ing

ok

harmonic

banned

blocked possible

!"# #$$%&'() → *+,('-,./0


Illo

cu

tio

na

ry s

co

pe

local

extensibledisjunct

constraint-free

conjunct

banned ok

!$# #$$%1-/2* → 34

%&'()* +, 5*1.67'8 '0'891.1: ;/+ <*.=<,1 '(* 2(/2/(,./0'8 ,/ ,<* (*1.67'81 '06 ;/+ >.6,< ,/ ,<*07&;*( /? 6','2/.0,1@ ,<* 6.(*-,./0 /? ;/+*1 A/& ,<* ;'1*8.0* 1.=0'81 ,<*.( 1.=0 B('.1.0= ;/+*1?/( 2/1.,.C*@ ?'88.0= ;/+*1 ?/( 0*=',.C* (*1.67'81D@ 1<'6.0= .06.-',*1 ,<', (*1.67'81 '(* .06.C.67'8891.=0.E-'0, 706*( ' 2*(&7,',./0 ,*1,@ >.,< 8.=<, =(*9 ', ' FG '06 6'() =(*9 ', ' HG 1.=0.E-'0-* 8*C*8

B1** I*9*( *, '8J KLLMDJ N<* /(6*( /? 8*C*81 .0 ,<* 28/, .1 '(;.,('(9 B'82<';*,.-'8DJ


!"#$%"& #"'%$() * +,)" --. -//0

Examples

•A bit less frequent than expected under H0:

Disjunct scope, WH ok

56

!

!" !"#$ %&' (& )*)% )+,- .-,)/#0( #123 45-,6'55&6#0( ,#17,- !%&' *,01 )*)%3 &- !%&'.-&/, #138

9 :,;7)-, 4:#</,; =>>?8

)@ #$%"&'A?6BC(&DEBF

#(BGH

)*++%67&,6AIJKC

,,&(-&.$*++&$'&.%?6LD.'%D.-#0(@M&-DEBFDN6E

(/H

!"#$ 67, (& A17,-,C )0$ .'% %&' 67&,623!" !"#$ 67, .'% %&' 67&,6 *7,0 67, *,01 A17,-,C23 45-,6'55&6#0( ,#17,- !67,*,013 &- !67, .&'(7138

.@ 0%(+%";#(71,-

$'-'*7,-,

0'1+&$'&.%17-&*DEBFDN6L

#(BGH

230%(O)1<7,6

%4&+&*&.%/;#(71DIEBFD?6EDN6L

!P7,-, $#$ %&' 17-&* 17, ;#(71,- 6& 17)1 %&' 7)Q, 1& '6, O)1<7,623

R F)'%) 4K)<"&0);$ =>>S8

)@ -'-?5

4'(7,-,

4!4*-'&(&-%-(5"%6#1D?5DLBBTFJE

1!6'(&+(1);/DUJV

-'-5+*&'5-%'/?5T(#Q,DNTEJWLX@H

"#$%&

!F7,% 6)1 7,-, )0$Y.'1 $#$ %&' 1);/ 1& 17,O23!" !B#0<, 17,% 6)1 7,-,Z $#$ %&' 1);/ 1& 17,O23

.@ [ 7'*7&AIJKC

4'(7,-,

6!6'&%&-%-(5"%<&O,D?6DLBBTFJE

8 "#$%&

9-+'-:':; ![#M *7& $#$ <&O, 7,-,\3

] W)/7&1) 4B#&')0Z )̂0 )̂;#0 =>>_8

)@ <=>#%$&(

#("GF

+?#*`I"GaYPb

@%A+?#%A?6c?6DC.#1,

$'/H

!P7)1 $#$ 17, $&( .#1,23 !"; !"#$ 17, $&( .#1, 6&O,17#0(23

.@ 7(B$?<%O)0

#("GF

+?#*,)1

@C+'`I"GaYPb

'B$C$DE*7#;,

+$%&7FB*?6EJBBD*#M,

#("GF

4-F*)1,-

(#FB(B*(,1@M&-

$'/H

!P7#;, 17, O)0 *)6 ,)1#0( 6&O,17#0(Z $#$ 7#6 *#M, (,1 *)1,- M&- 7#O23-!+; ![P7)1 $#$ 7#6 *#M, (,1 7#O *)1,-Z *7#;, 17, O)0 *)6 ,)1#0(23

d :,;7)-,

)@ -'&'"GKDWJU

@*,&%5-%%A?6BC6#1DB:e @̂EBFTFJE

4*-:$*1+&$'A?6BC<7)1DEBF

(/H

'()

!P7,0 7, *)6 7,-,Z $#$ 7, 6)% 6&O,17#0(23 4&- *)6 7, 6#;,0128!" !"#$ 7, 6)% 6&O,17#0( *7,0 7, *)6 7,-,23 4&- ;)1,- &0;%28G*+ -!+ !P)6 7, 7,-,Z )0$ $#$ 7, 6)% 6&O,17#0(23

.@ 2%&%*7&DA6CGXf

@%54&1$'#+&%&#5-%%<);;T?06LD<);;DB:e @̂EBFDNTFJE

,&#$%+:&%+&-(&.%#51$'/IGfD(&DEBFDIGfDNT`XX

'()

!g&' 7)$031 (&0, #M *7& 7)$ <);;,$ %&'23

!"

Extraction

Illo

cutionary

forc

e m

ark

ing

ok

harmonic

banned

blocked possible

!"# #$$%&'() → *+,('-,./0


Illo

cu

tio

na

ry s

co

pe

local

extensibledisjunct

constraint-free

conjunct

banned ok

!$# #$$%1-/2* → 34

%&'()* +, 5*1.67'8 '0'891.1: ;/+ <*.=<,1 '(* 2(/2/(,./0'8 ,/ ,<* (*1.67'81 '06 ;/+ >.6,< ,/ ,<*07&;*( /? 6','2/.0,1@ ,<* 6.(*-,./0 /? ;/+*1 A/& ,<* ;'1*8.0* 1.=0'81 ,<*.( 1.=0 B('.1.0= ;/+*1?/( 2/1.,.C*@ ?'88.0= ;/+*1 ?/( 0*=',.C* (*1.67'81D@ 1<'6.0= .06.-',*1 ,<', (*1.67'81 '(* .06.C.67'8891.=0.E-'0, 706*( ' 2*(&7,',./0 ,*1,@ >.,< 8.=<, =(*9 ', ' FG '06 6'() =(*9 ', ' HG 1.=0.E-'0-* 8*C*8

B1** I*9*( *, '8J KLLMDJ N<* /(6*( /? 8*C*81 .0 ,<* 28/, .1 '(;.,('(9 B'82<';*,.-'8DJ


!"#$%"& #"'%$() * +,)" --. -//0

Summary

•Recall that this was an exploratory study: large-scale multivariate typologies currently under development

- clause linkage

- grammatical relations

•Once the data exist and are mined, each associations needs theoretical interpretation and modeling so that it can be tested for universal validity

→ the topic of the next part

57

Compare to traditional approach

•Instead of empirical correlations, universal notions defining bundles of properties:

•cosubordination: {conjunct ILL scope, WH allowed}

•subordination: {disjunct or local ILL scope, WH banned}

•These are in fact absolute, exceptionless universals:

•conjunct ↔ WH ok

•disjunct/local ↔ WH banned

58


•Exceptions (as in Belhare) need to be “explained away”

•Or the absolute univeral needs to be limited in its predictive value, e.g.

•subordination = structure with disjunct or local ILL-scope.

•But then, no argumentation available of the kind:

“since X is subordinate, WH is not allowed” or “since X is subordinate, there is an intervening S node, and we cannot move WH”

59


•This is why in traditional approaches, metalanguages have a strong interest in maintaining absolute universals

•In traditional approaches, there is no way of understanding relations between structures or generalizations across structures without absolute universals

•Another example: branching direction (head parameter)

•In MULTIVARIATE TYPOLOGY, by contrast, all generalizations are probabilistic, and therefore have expections

•Instead of “if X bans disjunct scope, it must ban WH formation”, we get “... it is likely to ...”, i.e. statistical instead of absolute universals

60

Overall summary

•Absolute universals and variables are analytical tools, more or less successful in measuring variation and detecting distributional patterns, but with no claim on any deeper reality (i.e. Descriptive A priori ≠ Ontological A priori)

•Received absolute universals typically need decomposition into multiple variables allowing for much more variation

•Multivariate Typology allows analytical coverage of phenomena ‘in-between’ and captures similarities without forcing structures into pre-conceived universal slots

•Instead of asking “is X subordinate?”, we better ask questions like “does X enforce disjunct scope?”, “does X allow WH?” etc.

•Hundreds of variables, thousands of levels = the true diversity!

61

Overall summary

•These are the right questions for fieldwork.

•And, at the same time, these are the right questions for typology

•Answers lead to tables which we can mine for distributional patterns

•And so, the object of inquiry is these patterns, and the goal is to explain them.

(Contrast this with Anderson’s definition of the object and goal of linguistics: “The object of inquiry in linguistics is the human ability to acquire and use a natural language, and the goal of linguistic theory is an explicit characterization of that ability.”)

•Interestingly, more in line with other disciplines dealing with distributions (e.g. genetics, ecology, economics).

62

quantitative methods in typology - uzh

Documents