Morphosemantics, constructions, algorithmic typology and parallel

62
1 Morphosemantics, constructions, algorithmic typology and parallel texts Bernhard Wälchli [email protected] Stockholm University Bern (CH), September 20, 2012 Berner Zirkel für Sprachwissenschaft Universität Bern, Institut für Sprachwissenschaft

Transcript of Morphosemantics, constructions, algorithmic typology and parallel

Page 1: Morphosemantics, constructions, algorithmic typology and parallel

1

Morphosemantics,constructions,algorithmictypologyandparalleltextsBernhardWä[email protected](CH),September20,2012BernerZirkelfürSprachwissenschaftUniversitätBern,InstitutfürSprachwissenschaft

Page 2: Morphosemantics, constructions, algorithmic typology and parallel

2

AbstractUnlikemorphology(theinternalformalstructureofwords)andsemantics(thestudyofthemeaningofwordsandsentences),morphosemanticsisconcernedwiththelinkbetweenmarkerandmeaning.Traditionalapproachestomorphosemanticssuchassemioticsandconstructiongrammararguethattherelationshipbetweenimageacoustiqueandconceptissymbolic.Thisworkswellifthelinksareknown(inthe“proficiencymode”).InthistalkIarguethatthereisastatisticalalternativewhichisparticularlyusefulifthelinksarenotknown(inthe“discoverymode”).Meaningsandmarkersformcollocationsintextswhichcanbemeasuredbymeansofcollocationmeasures.However,thereisaconsiderablenon‐isomorphismbetweenmarkerandmeaning.Asiswellknownamarkercanhavemanydifferentmeanings(polysemy).Somewhatlesswellknownisthatameaningisoftenexpressedbymanydifferentmarkers,bothparadigmaticallyandsyntagmatically(polymorphy). Tomakemeaningsandmarkerscommensurable,theymustbeconvertedintounitsofthesamekind.Thissamekindisthesetofcontextsinatextorcorpuswhereamarkerormeaningoccurs.Ifthedistributionofameaninginacorpusisknown,itscorrespondingmarkercomplexcanbedeterminedwhichconsistsofaparadigmaticallyandsyntagmaticallyorderedsetofsimplemarkers.Themarkersconsideredherearesurfacemarkersoftwotypes:wordformsandmorphs

Page 3: Morphosemantics, constructions, algorithmic typology and parallel

3

(continuouscharacterstringswithinwordforms).Moreabstractmarkertypessuchaslexemes,grammaticalcategoriesandwordclassesmightoftenbebettermarkersthansurfacemarkers,buttheyarenotavailableinthediscoverymode. Markercomplexesareasimpleconstructiontype.Aproceduralapproachtoconstructiongrammarisadoptedwheremarkercomplexesareviewedasanintermediatestageinaprocessingchainofincreasinglymorecomplexconstructiontypesfromsimplemarkersviamarkercomplexestosyntacticconstructions.Markercomplexeshavetheadvantagethattheycanbeextractedautomaticallyfrommassivelyparalleltexts,i.e.translationsofthesametextintomanylanguages,suchastheNewTestamentusedhere.Inparalleltextsthesamemeanings(withcertainrestrictions)areexpressedacrossdifferentlanguages.Thismeansthatafunctionaldomaincanbedefinedasasetofcontextswhereacertainmeaningoccurs. Thesameprocedureisappliedtocross‐linguisticallysimilarmaterialandtheprocedureappliedtocross‐linguisticdataisfullyexplicitandthereforereplicable.Itcanbeimplementedinacomputerprogramandrunwithouttheinterventionofatypologist(algorithmictypology).Theunderlyingideaisthattheprocedureofextractionisinvariant(proceduraluniversal)whereastheextractedstructurescanbehighlyvariabledependingonthetextsandlanguagestowhichtheyareapplied.

Page 4: Morphosemantics, constructions, algorithmic typology and parallel

4

Thetalkconsiderstowhatextentsurfacemarkersaresufficientasinputfortheidentificationofconstructionsinarangeofgrammaticalandlexicaldomainsinaworld‐wideconveniencesampleofsomewhatmorethan50languages.Oneofthedomainsconsideredinmoredetailiscomparisonofinequality.Comparisonofinequalityisexpressedinmostlanguagesofthesamplebyanatleastbipartitemarkercomplexconsistingofthepartsstandardmarker(‘than’)andpredicateintensifier(‘more’,‘‐er’).Itwillbearguedherethatbothofthemareintrinsicpartsofthecomparativeconstruction.ThesefindingsarenotfullyinaccordancewithLeonStassen’stypologyofcomparison–aclassicalstudyinfunctionaldomaintypology–whichisbasedexclusivelyontheencodingofthestandardNP.Otherdomainsconsideredinthetalkincludenegation,‘want’,future,andpredicativepossession.

Page 5: Morphosemantics, constructions, algorithmic typology and parallel

5

TraditionalapproachestomorphosemanticsDeSaussure:Semiotics Croft:RadicalConstructionGrammar

Morphology theinternalformalstructureofwordsSemantics thestudyofthemeaningofwordsandsentencesMorphosemantics thelinkbetweenmarkerandmeaning

CONSTRUCTION

Imageacoustique

syntacticpropertiesmorphologicalpropertiesphonologicalproperties

semanticpropertiespragmaticproperties

discourse‐functionalproperties

Concept FORMsymboliccorrespondence(link)

(CONVENTIONAL)MEANING

Page 6: Morphosemantics, constructions, algorithmic typology and parallel

6

Analternativetosymboliclinks:collocations“[a]collocationisanexpressionconsistingoftwoormorewordsthatcorrespondtosomeconventionalwayofsayingthings”(Manning&Schütze1999:151)

strongtea powerfuldrugMeaningandmarkersaredifferentkindsofthings.However,intextstheybothmanifestthemselvesasdistributions.Distributionisthemediuminwhichmeaningcanbeturnedintomarkerandviceversa.MeaningsandmarkerscollocateCollocationmeasures,e.g.,t‐score

na

n

ny

nx

na

BAprobn

BprobAprobBAprobT

1),(1)()(),(

Page 7: Morphosemantics, constructions, algorithmic typology and parallel

7

Naturalandnon‐naturalmeaning(Grice1957,written1948)Naturalmeaning Non‐naturalmeaning

“Thosespotsmeanmeasles” “Thelightedsignontheroofmeansthat

thecabisfree”Canberestatedas“Thefactthathehasthosespotsmeansthathehasmeasles”xmeansthatpentailsp

Iscompatiblewith“Butitisn’tinfactfree–thedriverhasmadeamistake”.xdoesnotmeanthatpentailsp

Collocationsarenaturalmeaning

Page 8: Morphosemantics, constructions, algorithmic typology and parallel

8

ConstructionsinConstructionGrammarGoldberg(2006:5)“[A]LLLEVELSOFGRAMMATICALANALYSISINVOLVECONSTRUCTIONS:LEARNEDPAIRINGSOFFORMWITHSEMANTICORDISCOURSEFUNCTION,includingmorphemesorwords,idioms,partlylexicallyfilledandfullygeneralphrasalpatterns”(emphasisremoved,BW)morpheme: e.g.pre‐,ingword e.g.avocado,anaconda,andidiom(partlyfilled) e.g.jog<someone’s>memory,send<someone>tothecleanersDitransitive SubjVObj1Obj2(e.g.hegaveherafishtaco,hebakedheramuffin)Passive SubjauxVPpp(PPby)(e.g.thearmadillowashitbyacar)

Adynamicapproachtoconstructions Marker Markercomplex SyntacticconstructionProcessingchainofincreasinglymorecomplexconstructions

Page 9: Morphosemantics, constructions, algorithmic typology and parallel

9

Anonomasiologicalapproach Meaning FormOnomasiological given wantedSemasiological wanted given

Intypology:functionaldomainAccordingtoMiestamo(2005:293)afunctionaldomainis“anydomainofrelated(semanticorpragmatic)functionsthat(oneormore)language(s)encodewiththeformalmeanstheypossess”

Non‐isomorphismofmarkersandmeanings:PolysemyandPolymorphy ameaningisoftenexpressedbymanydifferentmarkers

Spanishquererhasmorethanonemeaning‘want’,‘love/desire’

Negation(informalsemanticssimply)inFrenchhasmorethanonemarker:ne,pas,point,non,rien,sansetc.Syntagmaticandparadigmaticpolymorphy

Page 10: Morphosemantics, constructions, algorithmic typology and parallel

10

Markercomplex:paradigmaticallyandsyntagmaticallyorderedsetofsimplemarkers

Basicconventions

readbottom‐upleft‐to‐right

Slots:columns,lefttorightAmplitude:verticalextensionDedication:horizontalextensionWordform:greenMorph:yellow,#iswordboundary

Wordorder:doesnotfigureSyntax:doesnotfigureLexemes,grams:donotfigure0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - French

ne

n'

nonsans

pas

point

rienpersonne

niaucun

plus

mais

Page 11: Morphosemantics, constructions, algorithmic typology and parallel

11

French: [ne|n'|non|sans]1[pas|point|rien|personne|ni|aucun|plus]2[mais]3Alemannic: [nit|kei|nigs|nemads|nimi|keini]1[sundern]240005017 [ne]1croyez[pas]2quejesoisvenupourabolirlaloioulesprophètesjesuisvenu[non]1

pourabolir[mais]3pouraccomplir

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - French

ne

n'

nonsans

pas

point

rienpersonne

niaucun

plus

mais

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - Alemannic

nit

keinigs

nemadsnimikeini

sundern

Page 12: Morphosemantics, constructions, algorithmic typology and parallel

12

English: [not|no|nothing|lest|neither|cannot|none|except|never]1[but]2English2: [not|no|never|nothing|lest|unless|neither]1[but]2[do|did|does]340005017 think[not]1thaticame...icame[not]1todestroy[but]2tofulfil40005017 [do]3[not]1thinkthatihave...ihave[not]1cometodestroythem[but]2tofulfillthem

Page 13: Morphosemantics, constructions, algorithmic typology and parallel

13

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - Wolof

ul

du

uñu#

umabañ

w aaye

kenn

dara

a

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - Moore

ye

sãlaa

ka

da

kõnra

baaned

Wolof: [>ul<|du|>uñu#<|>uma<|bañ]1[waaye|kenn]2[dara|a]3Moore: [ye|sã|laa]1[ka|da|kõn|ra|zɩ]2[baa]3[ned]440005017 b[ul]1eendefeneñëw...ñëw[uma]1ngirdindileen[waaye]2ngirñuamciman40005017 [da]2tags‐y...noy[ye]1mam[ka]2wannansãam‐b[ye]1layaatɩpids‐ba

Page 14: Morphosemantics, constructions, algorithmic typology and parallel

14

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - Mari

огыл

ок

у кеида

огытылом

огыдаогытынжеит

омылогынаотогыдалогешогыналынышт

но

а

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Negation - Ewe

o

oa

a�ke

meny e

#ny em

wom

boŋ

ga

ke

#m

Mari: [огыл|ок|уке|ида|огытыл|ом|огыда|огыт|ынже|ит|омыл|огына|от|огыдал|огеш|огынал |ынышт]1[но|а]2Ewe: [o|oa]1[aɖeke|menye|>#nyem<|>wom<]2[boŋ|>ga<]3[ke]4[>#m<]5

Page 15: Morphosemantics, constructions, algorithmic typology and parallel

15

AlgorithmictypologyandproceduraluniversalsThesameprocedureisappliedtocross‐linguisticallysimilarmaterialandtheprocedureappliedtocross‐linguisticdataisfullyexplicitandthereforereplicable.Itcanbeimplementedinacomputerprogramandrunwithouttheinterventionofatypologist(algorithmictypology).Theunderlyingideaisthattheprocedureofextractionisinvariant(proceduraluniversal)whereastheextractedstructurescanbehighlyvariabledependingonthetextsandlanguagestowhichtheyareapplied.

Page 16: Morphosemantics, constructions, algorithmic typology and parallel

16

ComparisonofInequalityStassen(1985)ComparisonandUniversalGrammarTypologybasedonStandardofComparisonLocative ‘Elephantbigat/onhorse’ Separative ‘Fromhorseelephantbig’ Allative ‘Bigelephanttohorse’Particle ‘Elephantbigthanhorse’Exceed ‘Elephantbigexceedshorse/exceedshorseinsizeConjoinedA ‘Elephantbig,horsesmall’ B ‘Elephantbig,horsenotbig’Functionaldomain.Stassen,definedintensionally:Aconstructionhavingthesemanticfunctionofassigningagradedpositiononapredicativescaletotwoobjects,standardandcompareeareNPsHere,definedextensionally:Forconvenience,anyversecontainingEnglishthan

Page 17: Morphosemantics, constructions, algorithmic typology and parallel

17

Whataboutthepredicateintensifier“more”,“‐er”?“amajorityofthelanguagesdonotusesuchanovertmarking”(Stassen1985:27)“...Ihavenotsucceededinfindinganexplanatoryprincipleonthebasisofwhichthepresenceorabsenceofthismarkingcanbepredicted.Hence,Iwillassumethatthephenomenonofcomparative‐markingisirrelevanttoourtypologyofcomparativeconstructions,andthatitmustbeexplainedintermsof(asyetunknown)regularitieswhichareindependentofthosethatdeterminethechoiceofaparticulartypeofcomparativeconstruction.Therefore,Iwillnotindicatesystematicallywhetherornotagivenlanguagerequiresmorphologicalmarkingofthecomparativepredicate”(Stassen1985:28).‐>ThedatacollectioninStassen’stypologyisexplanation‐driven.Inthepresentapproach,datacollectionisindependent

Page 18: Morphosemantics, constructions, algorithmic typology and parallel

18

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Spanish

más

may or

mejor

que

es

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Portuguese

domais

maior

melhor

que

é

Portuguese: [do]1[mais|maior|melhor]2[que]3[é]4Spanish: [más|mayor|mejor]1[que]2[es]341001007 vemaquele[que]3[é]4[mais]2poderoso[do]1[que]3eudequemnãosoudigno41001007 vienetrasmíel[que]2[es]3[más]1poderoso[que]2yoalcualnosoydignode

Page 19: Morphosemantics, constructions, algorithmic typology and parallel

19

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Norwegian

enn

mere

større

ere#

bedre

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Lithuanian

negu

esn

daugiau

geriau

labiau

lengv iau

Norwegian: [enn]1[mere|større|>ere#<|bedre]2Lithuanian: [negu|už]1[>esn<|daugiau|geriau|labiau|lengviau]241001007 ermigkommerdensomersterk[ere]2[enn]1jeghanhvisskoremjegikk41001007 skelbėpomanęsateinagaling[esn]2is[už]1maneašnevertasnusilenkęs

Page 20: Morphosemantics, constructions, algorithmic typology and parallel

20

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Tsimane

muju'cha'

mọjo'chas

mọjo'cha'

v a'cac

uy a'y a'

ọy a'y a'

dadam'

jam

Tsimane: [muju'cha'|mọjo'chas|mọjo'cha'|va'cac]1[uya'ya'|ọya'ya']2[dadam']3[jam]4Yanesha: [ello|tama]1[atarr]2

Page 21: Morphosemantics, constructions, algorithmic typology and parallel

21

Nointensifierintheconjoinedtype?No!Rathernostandardmarkerandmorethanoneintensifierinstead.Yanesha’(Matthew3:11)Ña‐pa' nent [atarr]2 ahuamencat‐esha' na‐ña‐pa' ama [tama]1 ahuamencat‐eyay‐no.he‐TOP REL much strong‐PROP I‐SEQ‐TOP,not that.much mighty‐NEG.SUFF‐MIDD‘that{comethafterme}ismightierthanI’

atarrP1;amatamaP2‘muchP1,notthat.muchP2’atarrP1;ellometan(err)anP2‘muchP1,more/again/separatedsurpass(again)P2’Tsimane(Matthew3:11)Mu’ qui ra' atsij [muju'cha']1 fer bu'yi‐ty,he/thatso.that FUT come more strong be.in.a.position‐MASC [jam]4 jeñej yụ, [uya'ya']2 yụ... not like I less I

TsimanehasthreeintensifierslotsP1[dadam']3[muju'cha']1...[ọya'ya']2P2‘P1bettermore...lessP2’42012007[dadam']3mu'[muju'cha']1arajjinacmi'in[jam]4jenejjaijtyi'in[ọya'ya']2ma'jotacsi

Page 22: Morphosemantics, constructions, algorithmic typology and parallel

22

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Maori

atu

ake

erangi

nuike

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Tagalog

kaylaloŋ

higit

pa

dakila

Maori: [atu|ake|erangi]1[nui]2[ke]3 atu‘away’,ake‘up’,nui‘big’,/kee/‘different’Tagalog: [kay]1[laloŋ|higit]2[pa|dakila]3

41001007 ...iahautetahihekaharawa[ake]1iahauekoreahauetauki41001007 ...sumusunodsahulihankoaŋ[laloŋ]2makapaŋyarihan[kay]1saakinhindiako41012033 ...sakaniyaŋsariliay[higit]2[pa]3[kay]1salahatnaŋmaŋahandog...

Page 23: Morphosemantics, constructions, algorithmic typology and parallel

23

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Ewe

wu

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Haitian

pasepi

pito

plis

41001007mounk'apvinaprem'langen[plis]3pouvwaanpil[pase]1m'mwenpabon40010031noupabezwenpèmenmnouvo[pi]2[plis]3[pase]1anpiltizwazo40011009twimwenmenmmwendinouli[pi]2[plis]3[pase]1yonpwofèt41009043l'jete[pito]2ouantrenanlaviaakyonsèlmen[pase]1pououreteaktoudemen

Page 24: Morphosemantics, constructions, algorithmic typology and parallel

24

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Wolof

gën

ëpp

sut

moo a

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - Erzya

седе

де

40003011 kiyñëwsamagannaaw[moo]2ma[ëpp]1kàttan41010025 giléemjaarcibën‐bënupusa[moo]2[gën]1[a]3yombboroomalalduggci40003011 мельгансыцясьмон[де]2нь[седе]1пеквиев

Page 25: Morphosemantics, constructions, algorithmic typology and parallel

25

“Doublemarking”incomparisonisdescribedinAnsaldo(1999)asonetypeofcomparisonthatneedstobeaccountedforbyparticularexplanations.However,“doublemarking”isclearlytheruleratherthantheexception.Inavastmajorityofthelanguagessurveyedtherearebothstandardmarkersandintensifiersextractedincomparison unlessthereisastandardmarkerextendedintotheintensifyingdomainor

unlesstherearetwodifferenttypesofintensifiersintheconjoinedtype

Inaclearmajorityoflanguages,comparisonissyntagmaticallypolymorphous

Page 26: Morphosemantics, constructions, algorithmic typology and parallel

26

Borrowingoffunctionwordsandpolymorphy

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Comparison - QuechuaCajamarca

mas

maski

manda

0 1 2 3 40.

00.

20.

40.

60.

81.

0

Comparison - CakchiquelCentral

más quechuvech

rukij

40003011 illapapirurinnamshamuqnoqa[manda]2suq[mas]1pudirniyuq41010025 xa[más]1laek...camello...junbak[que]2[chuvech]3junbeyonnoc

“Doppelsetzung”(Stolz&Stolz1996,Wiemer&Wälchli2012)

Page 27: Morphosemantics, constructions, algorithmic typology and parallel

27

Nextdomain:WANT

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Want - Somali

#doonay

doonijeclaan

inuu

inaan

inaad

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Want - Czech

chtel

#nech

chteli

chcete

chcichce

chcešchceš-lichtíti

chtejechtejí

iti#

Page 28: Morphosemantics, constructions, algorithmic typology and parallel

28

Somemarkercomplexesfor‘Want’:Czech: [chtěl|>#nech<|chtěli|chcete|(5moreforms)]1[>iti#<]2Zulu: [>thand<]1[>#uku]2Wolof: [bëgg|>#bëgg<]1[a]2Greek(Modern): [να|θέλω]1[θέλει|θέλεις|θέλω|θέλετε|(6moreforms)]2Somali: [>#doonay<|dooni|jeclaan]1[inuu|inaan|inaad]2Saramaccan: [kë]1GreenlandicWest: [>uma<]1Maltese: [>rid<|ried|riedx]1Haspelmath’s(2005)typologyof‘want’(simplified)andmarkercomplexes Complementsubject

implicitComplementsubjectexpressedovertly

Desiderativeaffix

Markercomplexwithoutsubordinator

Saramaccan Maltese WestGreenlandic

Markercomplexwithsubordinator

Czech,Zulu,Wolof ModernGreek,Somali

Page 29: Morphosemantics, constructions, algorithmic typology and parallel

29

Khanina(2008,2010)vs.GoddardandWierzbicka(2010). Khaninaclaimsthat‘want’isnotuniversalinthesenseof“beingtreatedonlyasparticulartypeofamoregeneralsituation”(2008:845).

Inhervarietysample,shefindsthat95of136desideratives(hercovertermforall‘want’expressions)“areregularlyusedtoexpressothersituationsthanpure‘want’”(2008:847).

AccordingtoherthisisachallengeforNaturalSemanticMetalanguagewhere‘want’isconsideredtobeasemanticprime,“i.e.anindivisibleunitofmeaningwithalexicalexponentinalllanguages”(GoodardandWierzbicka2010:108).

ForNaturalSemanticMetalanguageitisimportanttodistinguishbetweenpolysemyandsemanticgenerality.Forinstance,Spanish,quererhastwo(ormore)meanings‘want’and‘love/like’ratherthanonegeneralmeaning‘want/love/like’.Khanina,however,arguesthatmultiplemeaningsof‘want’expressionsarebestanalyzedasmacrofunctionbydefault.

Khaninaquestionstheuniversalityof‘want’.AccordingtoGoodardandWierzbicka(2010)thisisdueonlytoanunderestimationofpolysemy.

Page 30: Morphosemantics, constructions, algorithmic typology and parallel

30

Parallelstothepresentapproach

GoddardandWierzbicka(2010:114):“asemanticallyprimitivemeaningwillalwaysbeexpoundedbymeansofspecificallylexicalmaterial,bya‘segmentalsign’,andnot(forexample)byreduplication,orablaut,orsolelythroughagrammaticalconstruction.”

Khanina’sapproachissimilartothepresentoneinthatsheexplicitlychoosesdesiderativesinEuropeanlanguages(thepracticalmeta‐languagesofmostdescriptivegrammars)asherpointofdeparture.ShealsospeaksoftranslationalequivalentsofStandardAverageEuropean.IfIherechooseClassicalGreekethelo‘want’todefinethedomain,thebasicrationaleisverysimilar.

NaturalSemanticMetalanguageproceedstoalargeextentonomasiologicallyasfarassemanticprimesareconcerned.Forallconceptswhichareclaimednottobesemanticprimes,however,NaturalSemanticMetalanguagerathertakesasemasiologicalstance,butthisdoesnotneedtoconcernusheresince‘want’isclaimedtobeasemanticprime.

Page 31: Morphosemantics, constructions, algorithmic typology and parallel

31

Discoverymodevs.proficiencymodeIntheapproachtakenhere,weoperateinthediscoverymode.Thismeansthatwecannotmakeanydistinctionbetweenpolysemyandmacrofunctionsincethereisnoestablishedmarker‐meaningrelationship.Ifwewanttofindouthowameaningisexpressedcross‐linguistically,whatisgivenisonemeaningandallpotentialmarkercandidates.Themarker‐meaningrelationshipcannotbegiven,otherwisewewouldnotfindout.Ifthemarker‐meaningrelationshipisgiven,wealreadyknowwhatthemeaningofaformis.However,Idonotassumethatthereshouldbeanisomorphismbetweenmarkerandmeaning,aslongasthereisacollocationofmeaningandmarker,arelationshipcanbeestablishedirrespectiveofwhetherthereispolysemyinanarrowsenseormacrofunction.Inthematerialconsideredherethereisnoproblemtoestablishameaning‐markerrelationshipinthe‘want’‐domaininvirtuallyalllanguagesconsidered.

Page 32: Morphosemantics, constructions, algorithmic typology and parallel

32

WANT=SAY,butinverydifferentwaysKobon(Mark10:51)“Yɨp nɨhöng‐aŋ, a gɨ‐mön, au‐ab‐ön?” ö g‐a...1SG.OBJ what do‐IMP3SG, QUOT do‐SS2SG come‐PRS‐2SG QUOT do‐RMPST3SG“Amgöu kauyaŋ nɨŋ‐nam, a g‐em, au‐ab‐in,” a g‐a.eye that again see‐PRESCR1SG QUOT do‐SS1SG come‐PRS1SG QUOT do‐RMPST3SG[Jesusansweredhim,]"Whatdoyouwantmetodoforyou?"[Theblindmansaidtohim,"Rhabboni,]thatImayseeagain."Literally:Helike:“Youcomelike‘Domewhat’”...Bukiyip(Mark10:51)“Nyak ny‐a‐kli i‐nek‐um‐enyu moneken?”...2SG 2SG‐REA‐say 1SG:IRR‐do‐BENEF‐2SG.OBJ what “...yek y‐a‐kli i‐na‐tulugun.”1SG 1SG‐IRR‐say 1SG:IRR‐REFL‐look/see[Jesusansweredhim,]“Whatdoyouwantmetodoforyou?”[Theblindmansaidtohim,“Rhabboni,]thatImayseeagain.”

Page 33: Morphosemantics, constructions, algorithmic typology and parallel

33

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Want - Samoan

manao

fia

nagalo

mananaoloto

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Want - Bukiyip

yakli

Considerableparadigmaticpolymorphy.Samoan<manao>/mana‘o/(verb)‘want,desire’,fia(particlepreposedtoverb)expressesawish,finagalo(nounhonorific)‘wish’,<mānana‘o>/mana‘o/(verb)pluralofmana‘o‘want,desire’,loto(noun)‘heart,will’.

Firstpersonsingularsubjectisageneralcollocationof‘want’

Page 34: Morphosemantics, constructions, algorithmic typology and parallel

34

Alocalapproachtosemanticdecomposition Everyutterance(orpartofutterance)hasanexemplarycontextualmeaningofitsownthatisunique(see,e.g.,WälchliandCysouw2012fordiscussion).

Thisexemplarymeaningisusuallyhighlysimilartothatofmanyotherutterances,whichiswhyitispossibletoidentifyclustersofutteranceswithhighlysimilarmeaning:thesesharerangesofmeaning.

Foreveryrangeofmeaningthereisalocaldecompositionofexemplarycontextualmeaningintotwocomponents:thegeneralmeaningoftherangeversuseverythingelse.

Ifthemeaningrangeislexical,theexemplarswillmostobviouslydifferintheirgrammaticalmeanings.Ifthemeaningrangeisgrammatical,thevariableelementswillmostobviouslybelexical.Thisyieldsanappearanceofaglobaldivisionofmeaningintolexicalandgrammaticalmeaning.However,thisdivisionisnotinanywayrigid.

Grammaticalmeaningsandlexicalmeaningsaretreatedalike.Nodifferencebetweenlexicalandgrammaticaltypology.

Page 35: Morphosemantics, constructions, algorithmic typology and parallel

35

Futuretense (Indonesianakan)

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - French

er

a#

ont#

0 1 2 3 40.

00.

20.

40.

60.

81.

0

Future - Haitian

v a

pral

p'ap

y 'an'a

Page 36: Morphosemantics, constructions, algorithmic typology and parallel

36

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - English

shall

will

should

be

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - English2

will

believ e

be

Page 37: Morphosemantics, constructions, algorithmic typology and parallel

37

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - Hungarian

majd

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - Finnish

AccordingtoDahl(1985:105)thenumberoflanguageswithoutfuturetensecategoryisfairlysmall.

Page 38: Morphosemantics, constructions, algorithmic typology and parallel

38

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - Papago

wo

s-wohochs-ha

at

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Future - Yanesha

cha'# teruerr amach

InPapagothepotentialmodalmarkerwoisextractedalongwithatconsistingofa‐non‐imperativemoodand‐tcontemporarytense(usedinpast,presentandfuturecontextsopposedtozeropre‐experientalandd remotepast).

Page 39: Morphosemantics, constructions, algorithmic typology and parallel

39

PredicativePossession

Stassen(2009):fourbasictypesofpredicativepossession:Type Definingmarker

Locational ‘At/topossessor[there]is/existsapossessee Possessor With ‘Possessoris/existswithapossessee’ Possessee Topic ‘[Asfor]Possessor,possesseeis/exists’ None Have ‘Possessorhasapossessee’ PredicatePredicateNegationPossessorIndefinitePossesseeNegationandindefinite(indefinitearticleofpossessum)aregeneralcollocationsofpredicativepossession

Page 40: Morphosemantics, constructions, algorithmic typology and parallel

40

Have‐possessivesareeasiesttoextractiftheyarerecent(highdedication)whentheyhavenotmanagedyettogrammaticalizeintosomethingelse

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Spanish

tiene

tenía

teniendo

tienen

tenemos

tenían

tengotenéis

tienestenga

no

oigasiete

#necesi

0 1 2 3 40.

00.

20.

40.

60.

81.

0

Pred. Possession - Lithuanian

tur

reikia

#ne

ka

Spanish: [tiene|tenía|teniendo|tienen|tenemos|tenían|....]1[no|oiga|siete]2[>#necesi<]3Lithuanian: [>tur<|reikia]1[>#ne<|ką]2

Page 41: Morphosemantics, constructions, algorithmic typology and parallel

41

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - German

haben

hatte

hat

habe

habt

hattenhast

einen

macht

zu

0 1 2 3 40.

00.

20.

40.

60.

81.

0

Pred. Possession - French

n'

un

v ie

av ons

besoin

ay ant

av ait

ont

asav ez

aiav aient

entendeelle

aune

pas

oreilles

German: [haben|hatte|hat|habe|habt|hatten|hast]1[einen|macht]2[zu]3French: [n'|un|vie]1[avons|besoin|ayant|avait|ont|as|avez|ai...]2[a]3[une|pas|oreilles]4

Page 42: Morphosemantics, constructions, algorithmic typology and parallel

42

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Vietnamese

cầncầmquỉ

không

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Indonesian

ada

menaruh

berolehbertelinga

tiada

memegangsakitperempuan

#ber

padany amempuny ai

padamu

Vietnamese: [có|cần|cầm|quỉ]1[không]2Indonesian: [ada|menaruh|beroleh|bertelinga|tiada|...]1[>#ber<|padanya...]2[padamu]3

Page 43: Morphosemantics, constructions, algorithmic typology and parallel

43

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Finnish

on

oli

olisitarv itse

jolla

ei

meilläteilläminulla

sinulla

olehänellä

heillä

0 1 2 3 40.

00.

20.

40.

60.

81.

0

Pred. Possession - Latvian

kam

nav

mums

man

ir

bija

butuausis

ta

tev

Finnish: [on|oli|olisi|tarvitse]1[jolla|ei|meillä|teillä|minulla|sinulla]2[ole]3[hänellä|heillä]4Latvian: [kam|nav|mums|man]1[ir|bija|būtu|ausis]2[tā|tev]3

Page 44: Morphosemantics, constructions, algorithmic typology and parallel

44

Withthe‘with’possessiveextractionoftenfails:

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Hausa

ba

bakwaibiy ukunnen

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Hixkaryana

Hausa: [ba|bakwai|biyu|kunnen]1‘not/seven/two/ear’

Page 45: Morphosemantics, constructions, algorithmic typology and parallel

45

Stassen(2009): Predicativization:reanalysisofthecategorialandsyntacticstatusofthephrasewhichcontainsthepossessee

Transitivization/‘Have’‐drift:ifalanguagestartstoreanalyseitspossessiveinthedirectionofamajortype,theintendedoutputwillalwaysbeaHave‐possessive

0 1 2 3 4

0.0

0.2

0.4

0.6

0.8

1.0

Pred. Possession - Somali

#hay sa

leey ahay

lahay nlahaaleennahay

umahay sto

baahan

WhyHavedrift?“Itishardtoseewhythereshouldbeashiftfromspatialcontacttowardscontrol,butnottheotherwayaround”(Stassen2009:242)Whenevertheextractionissuccessfulitmostlycontainsapredicatemarkerinoneoftheslots(notnecessarilyinthefirstslot)

leh‐‘have’relatedtola‘with’

Page 46: Morphosemantics, constructions, algorithmic typology and parallel

46

Enter

Talmy(1991,2000)Satellite‐framinglanguages Verb‐framinglanguagesPathexpressedinadposition/case,verbalaffixoradverbialparticle

Pathexpressedinverb

Frenchentr‐,Turkishgir‐ Englishin,Russianv/v‐

AccordingtoBerthele(2006:235)theprepositionmakesacontributiontotheencodingofthepath,eventhoughtoalesserextentthantheverb.ThissuggestsforFrenchthatwewouldgettheverbentr‐inthefirstslotandtheprepositiondansinthesecondslot.

Sinha&Kuteva(1995)DistributedSpatialSemantics

Talmy(1972):(Spanish)[aPOINT] MOVE<‐TOIN TO(IN) [aSPHERE] entrar athe“motiveverb”(themotioncomponentlocatedintheverb)conflateswithacopyfromthedirectional(thepreposition)

Page 47: Morphosemantics, constructions, algorithmic typology and parallel

47

DomaindefinedbyClassicalGreeklemmaeiserchomai‘enter’

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Enter - French

entrdans

maison

roy aume

0 1 2 3 4 50.

00.

20.

40.

60.

81.

0

Enter - Alemannic

goht

kumme

in

ins

inä

hus

gehn

kummt

Page 48: Morphosemantics, constructions, algorithmic typology and parallel

48

‘House’aspartofthemeaningof‘Enter’

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Enter - Samoan

uluf ale

ulu

sao

o

atu

f ale

malo

aai

0 1 2 3 4 50.

00.

20.

40.

60.

81.

0

Enter - Dinka

la

baai

Inaconveniencesampleof51languagesthereisanaverageof2.4slotsperlanguage.In36languages(42%)thereisatleastonenounextracted(mostly‘house’)

Page 49: Morphosemantics, constructions, algorithmic typology and parallel

49

Inthevastmajorityoflanguagesbothverbal(V)andadnominal/adverbal(AN/AV)componentscontributetotheencodingof‘enter’.V Somali,Maltese,Hausa,FulAdamawa,Vietnamese,Tagalog,Mandarin,

Burarra,YineVAN/AV Basque,Kannada,Albanian,Alemannic,Greek(Modern),Hindi,French,

Italian,Latin,Portuguese,Romanian,Spanish,Korean,Buriat,Kalmyk,Tatar,Turkish,Finnish,Komi,Mari,Mordvin(Erzya),Swahili,Zulu,Ewe,Wolof,Bambara,Moore,Yoruba,Dinka,Zarma,HaitianCreole,Saramaccan,Maori,Samoan,Indonesian,Malagasy,Lahu,Tobelo,Kuot,WikMungkan,Greenlandic(West),Mixe(Coatlán)Otomí(Mezquital),Trique,Paumarí,Quechua(Cajamarca),Aymara

AN/AVV Avar,Welsh,Danish,German,English,Icelandic,LowSaxon,Norwegian,Swedish,Greek(Classical),Saami(Northern),TokPisin,Yanesha’

AN/AV Latvian,Lithuanian,Croatian,Hungarian,Cakchiquel,Bribri ANAV&V Afrikaans,Dutch,Czech,Polish,Russian,OsseticV&AV Mapudungunboldface:onlyoneslot,allotherdoculectshavetwoormoreslots

Page 50: Morphosemantics, constructions, algorithmic typology and parallel

50

0 1 2 3 4 5

0.0

0.2

0.4

0.6

0.8

1.0

Enter - Latin

#intr

#ingre

domum

regnum

in

Page 51: Morphosemantics, constructions, algorithmic typology and parallel

51

Page 52: Morphosemantics, constructions, algorithmic typology and parallel

52

0 1 2 3 4 50.

00.

20.

40.

60.

81.

0

Enter - Yanesha

osbe't

allcha'

all

pa'pacllo allña

Page 53: Morphosemantics, constructions, algorithmic typology and parallel

53

SomemarkercomplexesforEnter:Verbal,Adverbal,AdnominalAlemannic: [goht|kumme]1[in|ins]2[inä|hus]3[gehn|kummt]4English: [into|in]1[entered|enter]2[house|kingdom|came]3GreekModern: [>#μπ<]1[σπίτι|μέσα|στη]2[στο|βασιλεία]3GreekClassical:[>#εισ<]1[>λθ<]2[εις]3[την]4[βασιλειαν|οικιαν]5Italian: [>#entr<]1[casa|nel]2[nella|in]3Latin: [>#intr<|>#ingre<]1[domum|regnum]2[in]3Spanish: [>#entr<]1[en]2[casa|reino]3Russian: [в]1[вошел|войти|войдя|вошли]2[дом]3Turkish: [>#gir<]1[evine|içeri|eve]2Finnish: [meni|mennä]1[sisään|sisälle]2[siellä|valtakuntaan]3[tuli|sinne]4Hungarian: [>#be<]1[>ba#<]2Maltese: [daħal|jidħol|daħlu]1Ewe: [>ge<]1[me]2[ɖe]3Haitian: [antre]1[kay|wa]2[kote|lakay]3[nan]4TokPisin: [insait]1[go]2[haus|kingdom]3[taun]4Vietnamese: [vào]1[nhà]2Tagalog: [>asok#<]1[bahay|kaharian]2Mapudungun: [>#konp<]1[ruka|mülewe]2

Page 54: Morphosemantics, constructions, algorithmic typology and parallel

54

Middlevoice(“Reflexive”):triggerCroatianse>areal/genealogicaleffectCroatian se 1.0Polish się 0.5718Czech se 0.55836Russian >ся#< 0.52197Russian >сь#< 0.51321Romanian se 0.42018Icelandic >st#< 0.41464Spanish se 0.38316Latvian >ies#< 0.3582Afrikaans word 0.35431Romanian s' 0.35357Italian si 0.34838Latin >ur#< 0.34398Portuguese se 0.3165Alemannic sich 0.31625Latvian >ās#< 0.3152LowSaxon sich 0.31472Danish sig 0.30998Swedish sig 0.30734German sich 0.30669French se 0.30658

GreekClassical >θη< 0.30607Dutch worden 0.29763English3 be 0.29591English be 0.29451English2 be 0.29373Norwegian sig 0.28876Portuguese >‐se#< 0.273SaamiNorthernn >uvv< 0.27032Albanian >ohe< 0.27002Danish >es#< 0.2656GreekModern >ηκ< 0.2586Welsh >ir#< 0.24938Welsh >#ym< 0.24607Lithuanian >si< 0.24275Dutch zich 0.23754Norwegian >es#< 0.23708Yoruba a 0.23662Malayalam >pped< 0.23651Mari >алт< 0.23095Swedish >as#< 0.22728Hausa yi 0.22458

Page 55: Morphosemantics, constructions, algorithmic typology and parallel

55

Spanish >se#< 0.22393Romanian vă 0.22252Komi >öдч< 0.21995Albanian u 0.21817Dutch wordt 0.21636Papago e 0.21098LowSaxon woare 0.21081English2 >ed< 0.21043GreekClassical >αι#< 0.20949Finnish >ty< 0.20837Lahu la 0.20774Hungarian >ék#< 0.20648SaamiNorthern >oj< 0.20549Alemannic wird 0.20413Zulu >wa#< 0.20181Tatar >ыл< 0.20094English3 was 0.19769QuechuaCajamarca >aka< 0.19682TokPisin kamap 0.19448Kalmyk >гд< 0.1938English were 0.19252Icelandic sig 0.19147Norwegian blev 0.19144

Bribri e' 0.19057Dutch werd 0.19035French s' 0.18798Tagalog >aŋag< 0.18769English3 >#re< 0.1859Yine >tka#< 0.18529Alemannic wäre 0.18476Turkish >nm< 0.18391Norwegian bli 0.18309Swedish bliva 0.18194Czech >no#< 0.18086Wolof >iku< 0.17981Kannada >iko< 0.17874Romanian te 0.17851Hixkaryana >os< 0.17817Kuot >#te< 0.17683Yoruba nigbati 0.17679Dutch >ver< 0.17521English2 were 0.17355German werden 0.17214Somali la 0.17183GreenlandicWest >neqa< 0.17171Romanian de 0.17127

Page 56: Morphosemantics, constructions, algorithmic typology and parallel

56

Middlevoice(“Reflexive”)Croatian se 1.0Polish się 0.5718Czech se 0.55836Russian >ся#< 0.52197Romanian se 0.42018Icelandic >st#< 0.41464Spanish se 0.38316Latvian >ies#< 0.3582Afrikaans word 0.35431Italian si 0.34838Latin >ur#< 0.34398Portuguese se 0.3165Alemannic sich 0.31625LowSaxon sich 0.31472Danish sig 0.30998Swedish sig 0.30734German sich 0.30669French se 0.30658GreekClassical >θη< 0.30607Dutch worden 0.29763English3 be 0.29591

English be 0.29451English2 be 0.29373Norwegian sig 0.28876SaamiNorthernn >uvv< 0.27032Albanian >ohe< 0.27002GreekModern >ηκ< 0.2586Welsh >ir#< 0.24938Lithuanian >si< 0.24275Yoruba a 0.23662Malayalam >pped< 0.23651Mari >алт< 0.23095Hausa yi 0.22458Komi >öдч< 0.21995Papago e 0.21098Finnish >ty< 0.20837Lahu la 0.20774Hungarian >ék#< 0.20648Zulu >wa#< 0.20181Tatar >ыл< 0.20094QuechuaCajamarca >aka< 0.19682TokPisin kamap 0.19448

Page 57: Morphosemantics, constructions, algorithmic typology and parallel

57

Can all lexical and grammatical meanings be addressed in this way? No,probablynot.Gramswithextremelyhightextfrequency(“inflectionalcategories”),suchasplural,adnominalpossession(‘genitive’),present,imperfectivearedifficulttoaddressinmostlanguages.Thepresentversionoftheapproachisverycrude,possibleimprovements: Lexemesandgramsasmarkercandidatesinsteadofwordformsandmorphs Cross‐linguisticsemanticprototypesassemantictriggersratherthanwordformsfromparticularlanguagesinstantiatingameaning(Dahl1985)

Therearemanypracticalproblems: Accidentalcollocationsinaparalleltext Lexicalorgrammaticalmeaningsnotattestedinaparalleltext

Page 58: Morphosemantics, constructions, algorithmic typology and parallel

58

Isthismoderntypology?

“Moderntypologyisadisciplinethatdevelopsvariablesforcapturingsimilaritiesanddifferencesofstructuresbothwithinandacrosslanguages(qualitativetypology),exploresclustersandskewingsinthedistributionofthesevariables(quantitativetypology),andproposestheoriesthatexplaintheclustersandskewings(theoreticaltypology)”(Bickel2007:248) qualitative>quantitative>theoreticalHereweratheruseaninverseprocessingchaintheoretical>quantitative>qualitativeTheoreticalconsiderationandquantitativeanalysiscomefirst;theoutcomeisadescriptivemeasurementwhichmustbeevaluatedqualitativelyCross‐linguisticdescriptionhasbeenstronglyneglectedintypology.Descriptionintypologyshouldnotbefullyoutsourcedtofieldlinguists,anditshouldbeindependentfromexplanation(datacollectionintypologyshouldnotbeexplanation‐driven)

Page 59: Morphosemantics, constructions, algorithmic typology and parallel

59

ReferencesAnsaldo, Umberto. 1999. Comparative constructions in Sinitic. Areal typology and patterns of grammaticalization. PhD

Thesis. University of Stockholm. Berthele, Raphael. 2006. Ort und Weg. Die sprachliche Raumreferenz in Varietäten des Deutschen, Rätoromanischen

und Französischen. Berlin: De Gruyter. Bickel, Balthasar. 2007 Typology in the 21st century: major current developments. Linguistic Typology 11 (1): 239–251. Croft, William. 2001. Radical Construction Grammar: Syntactic theory in typological perspective. Oxford: Oxford

University Press. Dahl, Östen. 1985. Tense and Aspect Systems. Oxford: Blackwell. Davies, John. 1981. Kobon. (Lingua Descriptive Studies, 3.) Amsterdam: North-Holland. Goddard, Cliff & Wierzbicka, Anna. 2010. ‘Want’ is a lexical and conceptual universal. Studies in Language 34(1): 108–

123 Goldberg, Adele E. 2006. Constructions at Work. The Nature of Generalization in Language. Oxford: Oxford University

Press. Grice, Herbert Paul. 1957. Meaning. Philosophical Review 66(3): 377-388 Haspelmath, Martin. 2005. ‘Want’ complement subjects. In Haspelmath, Martin & Dryer, Matthew & Gil, David &

Comrie, Bernard (eds.) 2005. The World Atlas of Language Structures. (Book with interactive CD-ROM). Chapter 124. Oxford: Oxford University Press.

Khanina, Olesya. 2008. How universal is ‘wanting’? Studies in Language 32(4): 818–865. Khanina, Olesya. 2010. Reply to Goddard and Wierzbicka. Studies in Language 34.1: 124-130 Manning, Christopher D. & Schütze, Hinrich. 1999. Foundations of Statistical Natural Language Processing.

Cambridge, Mass: MIT Press. Miestamo, Matti. 2005. Standard Negation: The negation of declarative verbal main clauses in a typological

perspective. (Empirical Approaches to Language Typology 31.) Berlin/New York: Mouton de Gruyter. Saussure, Ferdinand de. 1968. Cours de linguistique générale. Édition critique par Rudolf Engler. Tome 1. Wiesbaden:

Harrassowitz

Page 60: Morphosemantics, constructions, algorithmic typology and parallel

60

Sinha, Chris & Kuteva, Tanja. 1995. Distributed spatial semantics. Nordic Journal of Linguistics 18: 167-199. Sölling, Arnd. 2011. Bewegungsverben in Nordamerika - Semantische Elemente in narrativen Texten. Diss. phil. hist.,

Universität Bern, Stassen, Leon. 1985. Comparison and Universal Grammar. Oxford: Blackwell. Stassen, Leon. 2009. Predicative Possession. Oxford: Oxford University Press. Stolz, Christel, Stolz, Thomas. 1996. Funktionswortentlehnung in Mesoamerika. Spanisch-amerindischer Sprachkontakt

(Hispanoindiana II). Sprachtypologie und Universalienforschung (STUF) 49: 86-123. Talmy, Leonard. 1972. Semantic structures in English and Atsugewi. Ph.D. dissertation, University of California at

Berkeley. Talmy, Leonard. 1991. Path to realization: a typology of event conflation. Proceedings of the Seventeenth Annual

Meeting of the Berkeley Linguistics Society, February 15-18, 1991, 480-519. Talmy, Leonard. 2000. Toward a Cognitive Semantics. Vol. II: Typology and Process in Concept Structuring.

Cambridge, MA: MIT Press. von Waldenfels, Ruprecht. Forthcoming. Explorations into variation across Slavic: taking a bottom-up approach. In

Szmrecsanyi, Benedikt & Wälchli, Bernhard (eds.), Linguistic variation in text and speech, within and across languages. To be published in Walter de Gruyter’s Linguae et Litterae series.

Wälchli, Bernhard. Forthcoming. Algorithmic typology and going from known to similar unknown categories within and across languages. In Szmrecsanyi, Benedikt & Wälchli, Bernhard (eds.), Linguistic variation in text and speech, within and across languages. To be published in Walter de Gruyter’s Linguae et Litterae series.

Wälchli, Bernhard & Cysouw, Michael. 2012. Lexical typology through similarity semantics: Toward a semantic map of motion verbs. Linguistics 50.3: 671-710. (Theme issue edited by Koptjevskaja-Tamm, M. & Vanhove, M. (eds.), New Directions in Lexical Typology).

Wiemer, Björn & Wälchli, Bernhard. 2012. Contact-induced grammatical change: Diverse phenomena, diverse perspectives. In Wiemer, B. & Wälchli, B. & Hansen, B. (eds.), Grammatical Replication and Borrowability in Language Contact, 3-64. Berlin: Mouton de Gruyter.

Page 61: Morphosemantics, constructions, algorithmic typology and parallel

61

Appendix1:Extractfromthedatabase(TriggerVietnameseđã,2545tokens)

No Domain Doculect Slot Type Marker Amplitude Dedication Extraction21 Perfect German 1 W hat 0.229971724788 0.648936170213 0.3852622 Perfect German 1 W habe 0.0890669180019 0.549418604651 0.2556423 Perfect German 1 W haben 0.0947219604147 0.463133640553 0.2554124 Perfect German 1 W ist 0.213477851084 0.276556776557 0.2386425 Perfect German 1 W hatte 0.0508953817154 0.421875 0.2651726 Perfect German 1 W war 0.062205466541 0.236135957066 0.2247227 Perfect German 1 W hast 0.0263901979265 0.427480916031 0.2285628 Perfect German 1 W habt 0.0311027332705 0.308411214953 0.2318229 Perfect German 1 W hatten 0.0254476908577 0.409090909091 0.2449930 Perfect German 1 W sind 0.0433553251649 0.159169550173 0.2109731 Perfect German 2 M ge 0.8821866164 0.320877613987 0.2166932 Perfect English 1 W hath 0.123939679548 0.57423580786 0.2571633 Perfect English 1 W had 0.116399622997 0.505112474438 0.2496434 Perfect English 1 W have 0.144203581527 0.398956975228 0.2514835 Perfect English 2 W which 0.16918001885 0.468057366362 0.2413736 Perfect English 2 M ed# 0.538171536287 0.296931877275 0.2223337 Perfect English 2 W made 0.0400565504241 0.291095890411 0.2345638 Perfect English 2 W sent 0.0268614514609 0.322033898305 0.2195539 Perfect English 3 W been 0.0725730442978 0.709677419355 0.22933163 Perfect Vietnamese 1 W đã 1.0 1.0 1.0164 Perfect Vietnamese 2 W cho 0.395852968897 0.356234096692 0.21582...

Page 62: Morphosemantics, constructions, algorithmic typology and parallel

62

Appendix2:R‐codewrittenbythePythonprogramgeneratingthevisualizationofmarkercomplexesplot(c(0,5),c(0,1),col="white",main="Perfect‐German",xlab="",ylab="")slot=0;par=0ing=0.648936170213;ingg=0.648936170213;ed=0.229971724788;edd=0.3;str="hat"rect(slot,par,slot+ing,par+ed,col="green")text(slot+ingg/2,par+ed/2,str,cex=si*edd)par=par+eding=0.549418604651;ingg=0.549418604651;ed=0.0890669180019;edd=0.3;str="habe"rect(slot,par,slot+ing,par+ed,col="green")text(slot+ingg/2,par+ed/2,str,cex=si*edd)par=par+eding=0.463133640553;ingg=0.463133640553;ed=0.0947219604147;edd=0.3;str="haben"rect(slot,par,slot+ing,par+ed,col="green")text(slot+ingg/2,par+ed/2,str,cex=si*edd)par=par+eding=0.276556776557;ingg=0.3;ed=0.213477851084;edd=0.3;str="ist"rect(slot,par,slot+ing,par+ed,col="green")text(slot+ingg/2,par+ed/2,str,cex=si*edd)...