how good is the typological evidence for referential scales? · 2015. 7. 9. · some definitions,...

31
How good is the typological evidence for referential scales? Balthasar Bickel October 2007 Tuesday, October 23, 2007

Upload: others

Post on 20-Feb-2021

0 views

Category:

Documents


0 download

TRANSCRIPT

  • How good is the typological evidence

    for referential scales?

    Balthasar Bickel

    October 2007

    Tuesday, October 23, 2007

  • Some definitions, clarifications, and assumptions

    •A scale is an ordered factor in a predictive model. •In typology (like in corpus linguistics), many predictive

    models — and virtually all predictive models with scales — are statistical, not causal or deductive models.

    •Statistical models always need a theoretical interpretation that proposes a causal development of the distribution that is captured by the model.

    •Typologists’ favorite theoretical interpretation of scales is in terms of usage frequency that conventionalizes over time.

    2

    Tuesday, October 23, 2007

  • Some definitions, clarifications, and assumptions (cont’d)

    •A referential scale is a scale defined by referential categories, covering both ‘inherent’ referential categories like ‘animate’ and discourse-based referential categories like ‘speaker’ or ‘proximative’.

    •Referential categories are ultimately language-specific and can only be identified by language-specific criteria (cf. Silverstein’s term “inherent lexical content of noun phrases”.)

    3

    Tuesday, October 23, 2007

  • Some definitions, clarifications, and assumptions (cont’d)

    •Still, most linguists make the practical assumption that there is sufficient cross-linguistic overlap between referential categories so that we can legitimately speak of ‘the same scale in different languages’.

    1. largely unproblematic for, e.g., 2 > 3 (perhaps problematic for ‘CONJUNCT/DISJUNCT’ systems)

    2.more problematic for, e.g., proximative > obviative

    •Luckily, this is not much of a problem because I will only be concerned with scales of type (1), and no C/D-systems...

    •Specifically, I am concerned with the idea that the choice of grammatical relations is somehow affected by a universal referential scale of type (1). Call this is the RGR Idea.

    4

    Tuesday, October 23, 2007

  • Some versions of the RGR Idea and some their predictions

    1.Alignment (cf. Silverstein 1976, etc.)

    • For any given grammatical relation: odds(A=S) ~ rank(RH). 2.Marking (cf. Comrie 1981, etc.)

    • For A arguments: odds(zero) ~ rank(RH). • For P arguments: odds(zero) ~ - rank(RH).

    3.Direction (cf. Comrie 1981, DeLancey 1981, etc.)

    • For A arguments: odds(direct/antipassive) ~ rank(RH)• For P arguments: odds(passive/inverse) ~ rank(RH)

    4.Agreement triggers (cf. DeLancey 1981, etc.)

    • odds(agr) ~ rank(RH) (“hierarchical agreement”)

    5

    Tuesday, October 23, 2007

  • What can we test?

    •Good databases on alignment in case and agreement: AUTOTYP, WALS

    •Some preliminary surveys on agreement forms with passive/antipassive morphology.

    •However, RH coding limited to part of speech differences: alignment of case in pronouns; alignment of case in lexical nouns; alignment in verb agreement

    •See how far these data confirm the predictions•I will concentrate here on some aspects the Alignment and

    the Direction Hypotheses.

    6

    Tuesday, October 23, 2007

  • Alignment Hypothesis: case

    •Alignment hypothesis: odds(A=S case) ~ rank(RH)•Data: •remove zero-marked case, neutral alignment, and splits

    based on something other than RH (e.g. TAM splits)

    •merge WALS and AUTOTYP data (100% matching!)•Test the hypothesis, controlling for •genealogical factors: count datapoints only if they are not

    demonstrably related through inheritance (using Bickel’s 2007 g-sampling algorithm)

    •areal factors: test for the effect of macrocontinental impact because Bickel & Nichols 2007 have found that case alignment is affected by this

    7

    Tuesday, October 23, 2007

    http://www.uni-leipzig.de/~bickel/research/papers/gsampling.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/gsampling.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/gsampling.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/gsampling.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/case_geo_bickel_nichols2007.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/case_geo_bickel_nichols2007.pdf

  • Areal control: accusative vs. other non-neutral case alignment

    8

    • Africa and NG-Australia significantly different from the rest of the world, N = 190

    • Therefore, use a control factor specifying four macrocontinents

    Tuesday, October 23, 2007

  • Areal control: four macrocontinents

    9

    Tuesday, October 23, 2007

  • Alignment Hypothesis: case

    Two predictions of the Alignment Hypothesis

    1.If there is a split within a language, it follows the RH

    2.Across languages, odds(A=S) ~ rank(RH)

    10

    Tuesday, October 23, 2007

  • “If there is a case split within a language, it follows the RH.”

    11

    Africa Americas Eurasia NG-Australia

    predicted

    not predicted

    neutral

    0 2 2 80 0 0 0

    70 137 93 114

    Without genealogical control, N = 425:

    Tuesday, October 23, 2007

  • “If there is a case split within a language, it follows the RH.”

    12

    Africa Americas Eurasia NG-Australia

    predicted

    not predicted

    neutral

    0 2 2 20 0 0 0

    70 137 93 114

    After genealogical control (removing duplicate stock representatives of the same type), N = 419:

    • one could add languages like with splits between ‘no case’ and ‘acc’ (like English or its mirror-image, Iraqw)

    • one could add the Indo-Iranian counterexamples (‘erg’ pronouns, ‘acc’ nouns) discussed by Filimonova (2005)

    • but neither of these additions would change the figures to a degree that the statistical signal turns significant (after genealogical and areal control)!

    Tuesday, October 23, 2007

  • “Across languages, odds(A=S case) ~ rank(RH)”

    13

    Statistics (after removing Africa)

    Macrocontinent: χ2(2)=7.6, p=.023

    Part of Speech: χ2(1)=2.5, p=.114

    Interactions: n.s.

    Best-fitting model:

    Nagelkerke’s R2=.23

    Somer’s Dxy=.37

    N = 230 (genealogically

    controlled)

    macrocontinent = Africa

    noun

    acc

    oth

    er pronoun

    macrocontinent = Americas

    noun

    acc

    oth

    er

    pronoun

    macrocontinent = Eurasia

    noun

    acc

    oth

    er

    pronoun

    macrocontinent = NG-Australia

    noun

    acc

    oth

    er

    pronoun

    Tuesday, October 23, 2007

  • Summary on case alignment

    •No good evidence for RH effects if RH is assessed by the difference between pronouns and lexical nouns.

    •The distribution of case alignment might be better explained by diachrony:

    •source (e.g. ergatives from instrumentals are stuck with nouns: Garrett 1990)

    •inertia (e.g. ergative case might survive only in pronouns: Filimonova 2005)

    14

    Tuesday, October 23, 2007

  • Alignment Hypothesis: agreement

    •The Alignment-based Hypothesis also makes predictions for alignment in verb agreement

    •Silverstein’s example: Chinook:

    15

    A S PO id t(x!)-k- -tx!- ip l(x!-)-k- -lx!- 2s m- 2d mt-k- -mt- 2p m"-k -m"-k- es n- ~ ø- (>2) -n- ed nt-k- ~ q- (>2) -nt- ep n"-k- ~ q (>2) -n"- 3d "t-k- "t- -"- 3p t-k- tk- -t- 3coll #-k- -#- 3sf k- -a- ~ -ø- 3sm $- -i-

    Tuesday, October 23, 2007

  • Agreement alignment: Siewierska & Bakker’s database (N= 402)

    16

    Language Split, as coded in the database Stock LocationAinu 1s,2 accusative; 1p tripartite (isolate) E. Eurasia

    Chácobo 1s,2s,3s,1p tripartite; # 2p,3p accusative; Panoan S. America

    Comox 1,2 accusative; 3s tripartite Salish N. AmericaKamass 1,2 accusative; 3 tripartite Uralic Eurasia

    Paumarí 1,2,3p accusative; 3s,3d ergative Arauan S. America

    Seri 1s tripartite; 2s,p accusative; # 3 neutral (zero) (isolate) N. America

    Yimas 1,2 tripartite; 3 ergative Lower Sepik PNG? Maricopa 1,2 accusative; 3 neutral (zero) Yuman N. America? Maung 1,2 accusative; 3 ergative Iwaidjan Australia? Nez Perce 1,2 accusative; 3 ergative Plateau Penutian N. America? Tepehua 1,2 accusative; 3 neutral Totonac-Tepehuan C. America? Washo 1,2 accusative; 3 ergative (isolate) N. America? = alternative analyses available: Bickel 2007

    Tuesday, October 23, 2007

    http://www.uni-leipzig.de/~bickel/research/papers/rhgr_bb2007.pdfhttp://www.uni-leipzig.de/~bickel/research/papers/rhgr_bb2007.pdf

  • Counterexamples in Kiranti, e.g. Puma

    17

    Table 3. Split ergativity in Puma

    A S P

    1s -! (>3)

    -na (>2)

    -!a (NPST)

    -o! (PST)

    -ci~c"b 1da

    ni- -ci~c" (>2)

    1pa -m (>3)

    ni- -i~ni(n)~n" (>2) -i~ni(n)~n"

    2s t"-

    -na (1s>)

    2d t"- -ci~c"

    -na-ci (1s>)

    2p t"- -i~ni(n)~n"

    t"- -m (>3)

    ø- 3s

    p"- (>1)

    -u ~ -i

    3d p"- -ci~c"

    ni-p"- -ci~c" (>1)

    ni- -ci~c" (>2)

    m"-�

    p"- (>3s)

    ni-p"- (>1s)

    3p

    ni-p" -i~ni(n)~n" (>1ns)

    ni- -i~ni(n)~n" (>2)

    -ci

    a All first person nonsingular forms distinguish exclusive vs. inclusive forms, marked by –ka, but ommitted here since this does not affect alignment b For the conditions regulating allomorphies (marked here by a tilde), see Bickel et al. (2007a) Tuesday, October 23, 2007

  • Counterexamples in Kiranti

    18

    Branch With reverse ergativity split (at least in part)Without ergative alignment (i.e. with first person showing tripartite, accusative or neutral alignment)

    Eastern Athpare BelhareLimbu Mewahang Lohorung Chintang Yamphu

    Central Puma CamlingBantawaKulung

    Western Hayu WambuleKhaling ThulungBahing Jero

    Dumi

    Tuesday, October 23, 2007

  • Summary on agreement alignment

    •No good evidence for RH effects.•Whatever predicts the distribution of alignment in verb

    agreement, it is not the Referential Hierarchy; or if it does, the effects are so systematically blurred by competing factors that they do no show up as a statistical signal in a large set (N = 402) of the languages of the world.

    19

    Tuesday, October 23, 2007

  • Another hypothesis on alignment

    •Frequent observations (e.g. Croft 2003, Siewierska 2004, Haspelmath 2005):

    •verb agreement tends to arise through grammaticalization of pronouns;

    •pronouns are always higher on the RH than lexical nouns;•therefore, odds(A=S) ~

    (Alternatively: verb agreement tends to be favored by topical arguments; topical arguments rank higher)

    •Test this agains WALS and AUTOTYP data, controling for genealogy and geography (as before)

    •Merge data, favoring AUTOTYP in case of mismatches in the coding of agreement alignment (6%, N = 87, both directions)

    20

    Tuesday, October 23, 2007

  • “More A=S in agreement than in case”

    Again, two predictions:

    1.If there is a split within a language, agreement will have A=S and case will have another alignment.

    2.Across languages, odds(A=S) ~

    21

    Tuesday, October 23, 2007

  • If there is a split within a language, agreement favors A=S”

    22

    Africa Americas Eurasia NG-Australia

    predicted

    not predicted

    neutral

    0 7 4 150 4 0 2

    70 128 91 105

    Without genealogical control (N = 425):

    Tuesday, October 23, 2007

  • If there is a split within a language, agreement favors A=S”

    23

    Africa Americas Eurasia NG-Australia

    predicted

    not predicted

    neutral

    0 7 3 120 4 0 2

    70 128 91 105

    After genealogical control (removing duplicate stock representatives of the same type), N = 421:

    No evidence against independence of the odds ratio ‘predicted’ : ‘not predicted’ from Macrocontinents, Fisher Exact Test, p = .32

    Tuesday, October 23, 2007

  • ‘Across languages, A=S more like in agreement than in case’

    24

    Statistics (after removing Africa):

    Interaction: χ2(2)=11.1, p=.004

    Nagelkerke’s R2=.12

    Somer’s Dxy=.31

    N = 279 (genealogically

    controlled)

    Significant effects only in NG-

    Australia! (Fisher Exact, p < .001,

    N = 95)

    macrocontinent = Africa

    case

    acc

    oth

    er agreement

    macrocontinent = Americas

    case

    acc

    oth

    er

    agreement

    macrocontinent = Eurasia

    case

    acc

    oth

    er

    agreement

    macrocontinent = NG-Australia

    case

    acc

    oth

    er

    agreement

    Best-fitting model:Tuesday, October 23, 2007

  • Analysis of residuals

    25

    macrocontinent = Africa

    -0.825

    0.000

    0.592

    Pearsonresiduals:

    p-value =0.30022

    case

    acc

    oth

    er

    agreement

    macrocontinent = Americas

    -0.22

    0.00

    0.28

    Pearsonresiduals:

    p-value =0.67573

    case

    acc

    oth

    er

    agreement

    macrocontinent = Eurasia

    -0.436

    0.000

    0.401

    Pearsonresiduals:

    p-value =0.44082

    case

    acc

    oth

    er

    agreement

    macrocontinent = NG-Australia

    -2.40-2.00

    0.00

    2.00

    2.94

    Pearsonresiduals:

    p-value =1.7906e-06

    case

    acc

    oth

    er

    agreement

    Tuesday, October 23, 2007

  • Summary on alignment across construction

    •Clear statistical signal only in NG-Australia, but there it demonstrably independent of known genealogies

    •If there is also a trend outside NG-Australia, it must have been blurred by factors relating to the source of case markers, the development of paradigm structures etc.

    26

    Tuesday, October 23, 2007

  • Direction Hypothesis

    • For P arguments: odds(passive/inverse) ~ rank(RH)• Example: Picurís (Tanoan, N. America; Zaharlick 1982):

    27

    Himalayan Languages Symposium 2005, Bangkok, December 6 - 9

    Generics as First Person Undergoers and the Political History of the Southern Kirant

    Balthasar Bickel & Martin Gaenszle

    University of Leipzig www.uni-leipzig.de/~ff/cpdp

    1. Antipassives for first person object reference in Puma

    In Puma (Southern Kiranti) the same verb forms, marked by a prefix kha- and intransitive agreement inflection, are used for antipassives and for first person object (1P) agreement. This constellation is extremely unusual in the languages of the world: wherever diathesis has been reported to develop into 1P marking, it involves a passive, not an antipassive (e.g., DeLancey 1981, Aissen 1999, Bresnan et al. 2001, Zúñiga 2002, among others).

    a. ta-m!n-mia-"#n s$nene-pa. 1s-see-PASS-PST man-OBL

    The man saw me.’

    b. m!n-mia-"#n (s$nene-pa). see-PASS-PST man-OBL

    ‘S/he was seen (by the man).’

    c. ta-me-"#n. 1s-go-PST

    ‘I went.’

    (1) a. en-i. hear-3P[PST] (Active transitive inflection)

    ‘S/he heard him/her/it.’ (entails a specific undergoer referent)

    b. kha-en-a. ANTIP-hear-PST (Antipassive intransitive inflection)

    ‘S/he heard someone / people.’ or ‘S/he listened so as to find out whether or not there are people.’ (does not entail existence of a specific undergoer referent) OR: ‘S/he heard us (incl.).’

    (2) The Puma agreement paradigm: 1P and antipassive/intransitive forms (excerpt from the past affirmative paradigm of enma ‘to hear, listen’)

    1sP 1nsiP 1deP 1peP Antipassive/Intransitive 2sA t!eno" khat!ena khat!ena 2dA t!eno"c!" khat!enci khat!enci 2pA t!eno"n!" khat!ennin khat!ennin 3sA p!eno" khaena p!encika p!enninka khaena 3dA p!eno"c!" khap!enci nip!encika khap!enci 3pA nip!eno" kham!ena nip!enninka kham!ena

    Tuesday, October 23, 2007

  • Direction Hypothis: counter-example

    28

    a. ta-m!n-mia-"#n s$nene-pa. 1s-see-PASS-PST man-OBL

    The man saw me.’

    b. m!n-mia-"#n (s$nene-pa). see-PASS-PST man-OBL

    ‘S/he was seen (by the man).’

    c. ta-me-"#n. 1s-go-PST

    ‘I went.’

    (1) a. en-i. hear-3P[PST] (Active transitive inflection)

    ‘S/he heard him/her/it.’ (entails a specific undergoer referent)

    b. kha-en-a. ANTIP-hear-PST

    ‘S/he heard someone / people.’ OR: ‘S/he heard us (incl.).’ p!" akhani ohyatni kha-cop-a=ni. k!-nana-ci To"wama and.then hither thither ANTIP-look-PST-REP 3sPOSS-eS-ns T.

    Khiwama-ci-ya-tni kha-cop-a=ni, sa=cha metd!"ya"-ci=ni. Kh.-ns-ACROSS-ALL ANTIP-[3sS-]look-PST=REP anyone=ADD NEG.EXIST.PST-d-REP

    tonp!"=na khap-ma puss-i=ni. khap-a-"a khap-a-"a then.after=TOP weep-INF begin-3sP[PST]=REP [3sS-]weep-PST-IPFV [3sS-]weep-PST-IPFV

    kha#o=ni, k!-$a"ko"-di-tni kha-cop-a. !k-ta "aksi, when=REP 3sPOSS-pillow-UP-ALL ANTIP-[3sS-]look-PST one-CLF banana

    !k-ta berucha, !k-ta bechuk yu"a"a=ni (folk_tale.056ff.DR) one-CLF small.sickle one-CLF ginger PST.EXIST=REP

    ‘Then he looked (ANTIP) hither and thither (for them). He looked (ANTIP) in the direction of his sisters To!wama and Khiwama’s place, (but) there wasn"t any one of them! Then he started to cry. As he was crying and crying, he looked (ANTIP) up towards the pillow (having them in mind): there was one banana, one small sickle, and ginger (which the two had left for him).’

    a. (*thoro%-cha) kha-tat-o%. male-offspring ANTIP-bring-1sS.PST

    ‘I brought some people.’ (e.g. to help me work on this)

    b. thoro%-cha tat-o%. [zero-marked antipassive] male-offspring bring-1sS.PST

    ‘I brought some young man/men.’ (e.g. to help me work on this)!

    Bickel et al. 2007, in Journal of Himalayan Linguistics 7, 1 -19

    Himalayan Languages Symposium 2005, Bangkok, December 6 - 9

    Generics as First Person Undergoers and the Political History of the Southern Kirant

    Balthasar Bickel & Martin Gaenszle

    University of Leipzig www.uni-leipzig.de/~ff/cpdp

    1. Antipassives for first person object reference in Puma

    In Puma (Southern Kiranti) the same verb forms, marked by a prefix kha- and intransitive agreement inflection, are used for antipassives and for first person object (1P) agreement. This constellation is extremely unusual in the languages of the world: wherever diathesis has been reported to develop into 1P marking, it involves a passive, not an antipassive (e.g., DeLancey 1981, Aissen 1999, Bresnan et al. 2001, Zúñiga 2002, among others).

    (1) a. en-i. hear-3P[PST] (Active transitive inflection)

    ‘S/he heard him/her/it.’ (entails a specific undergoer referent)

    b. kha-en-a. ANTIP-hear-PST (Antipassive intransitive inflection)

    ‘S/he heard someone / people.’ or ‘S/he listened so as to find out whether or not there are people.’ (does not entail existence of a specific undergoer referent) OR: ‘S/he heard us (incl.).’

    (2) The Puma agreement paradigm: 1P and antipassive/intransitive forms (excerpt from the past affirmative paradigm of enma ‘to hear, listen’)

    1sP 1nsiP 1deP 1peP Antipassive/Intransitive 2sA t!eno" khat!ena khat!ena 2dA t!eno"c!" khat!enci khat!enci 2pA t!eno"n!" khat!ennin khat!ennin 3sA p!eno" khaena p!encika p!enninka khaena 3dA p!eno"c!" khap!enci nip!encika khap!enci 3pA nip!eno" kham!ena nip!enninka kham!ena

    2. The syntax and morphology of antipassives in Puma (Bickel et al. in prep.)

    Like in other Kiranti languages, a general properties of antipassives in Puma is that they have intransitive agreement morphology and intransitive case assignment (nominative on SA). But in Puma there are two kinds of antipassives: one marked by kha- for human undergoers, and an unmarked one for nonhuman undergoers.

    2.1 The unmarked antipassive: similar properties as in in other Kiranti languages (Angdembe 1998, Bickel 2004, 2006)

    • Objects are obligatory (incorporate-like) (3) a. !a redio en-!a. 1sNOM radio hear-1sS.NPST (Nonhuman antipassive intransitive inflection)

    ‘I do radio-hearing.’ (in general, does not entail the existence of a specific radio that the speaker has in mind)

    b. * !a en-!a.

    Tuesday, October 23, 2007

    http://www.linguistics.ucsb.edu/HimalayanLinguistics/http://www.linguistics.ucsb.edu/HimalayanLinguistics/

  • Direction Hypothesis: preliminary survey (N = 197)

    29

    Overlap w

    ith 1

    U a

    gre

    em

    ent m

    orp

    holo

    gy

    Antipassive Passive

    none

    som

    e

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    Tuesday, October 23, 2007

  • Summary on direction

    •No apparent universal trend favoring passives vs. antipassives for P arguments depending on the RH.

    •Case study on Puma suggests that antipassives for 1P in Puma reflects a impersonalization strategy, ultimately based on face saving strategies borrowed from Maithili (Bickel & Gaenszle 2007)

    •Further work on the Direction Hypothesis might want to test for competing effects of politeness strategies, suggesting a multifactorial model.

    30

    Tuesday, October 23, 2007

    http://www.uni-leipzig.de/~bickel/research/presentations/Kiranti1U_ALT2007.pdfhttp://www.uni-leipzig.de/~bickel/research/presentations/Kiranti1U_ALT2007.pdfhttp://www.uni-leipzig.de/~bickel/research/presentations/Kiranti1U_ALT2007.pdfhttp://www.uni-leipzig.de/~bickel/research/presentations/Kiranti1U_ALT2007.pdf

  • Conclusions

    •Given the prominence it received, the hypothesized universal effects of the referential hierarchy have surprisingly little statistical support.

    •What I did not and could not test is •RH effects beyond the pronoun vs noun opposition•the effects on Agreement triggers (hierarchical agreement)

    31

    Tuesday, October 23, 2007