The Role of Analysis and Recombination in the Evolution of ... · The Role of Analysis and Recombination in the Evolution of Language Structure Mónica Tamariz monica@ling.ed.ac.uk

The Role of Analysis and Recombination in the Evolution

of Language Structure

Mónica Tamariz monica@ling.ed.ac.uk

OVERVIEW I.  Aques)onandananswer

– Whatareanalysisandrecombina.on?

II.  Originofhumanculture–  Theore)calargumenta)on(archaeology,evolu)onarytheory)

–  Theroleofanalysisandrecombina)on

III.  Theevolu)onofcomposi)onalityinlanguage–  Empiricaldata(ar)ficiallanguagelearningstudies)

–  Adapta)onofformstomeaningsthroughanalysisandrecombina)on

IV.  Conclusion

- Whyishumanlanguagestructuresocomplex?

- Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

E.g.Cons)tuency,Redundancy

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

Regularityallows:

• Generalisa)on• Compression

am wasthink

thoughttell toldsee saw

look lookedwalk walkedwork workedplay played

look‐=‘look’walk‐=‘walk’work‐=‘work’play‐=‘play’‐ =‘present’‐ed=‘past’

am=‘am’was=‘was’think=think’ thought=‘thought’tell=‘tell’told=‘told’ see =‘see’saw =‘saw

IRREGULAR REGULAR

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina/onofsymbols

hit

cut

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina/onofsymbols

hit

cut

withhand

moverepeatedly

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina/onofsymbols

hit(liRleeffort)

cut(lotsofeffort)

withhand

moverepeatedly

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina/onofsymbols

hit(liRleeffort)

cut(lotsofeffort)

withhand

moverepeatedly

Cut(liRleeffort)

-  Whyishumanlanguagestructuresocomplex?

-  Becauseonlyhumanscandoanalysisandrecombina2onofsymbols

Dobzhansky(1973):“Nothinginbiologymakessenseexceptinthelightofevolu)on”

…alsoapplicabletolinguis)cbehaviour!

II. Language Origins

•  Apuzzle:Ifcumula)vecultureissoadap)ve,whyisitsorare?(Boyd&Richerson1996)

•  Ananswer:Becausethecogni)vebiasesfavouredbynon‐cumula)vecultureareatoddswiththoserequiredforcumula)veculture(Tamariz2008)

Origins of culture

•  Cultural

1.Emergenceofearlyculture

2.Emergenceofcumula)veculture

(Tamariz2008)Notonebuttwomajortransi)ons

Nonhuman cultural traditions

•  Holis)ctransmission)edtofunc)on•  Brokenchains

‐  Innova)onthrough“localenhancement”‐  Maximumcomplexity:individuallearning

‐  Noaccumula)on(Marshall‐Pescini&Whiten,2008)

1. Transition to early culture

•  Holis)ctransmission)edtofunc)on

•  Unbrokenchains–  Someinnova)on(analysis,recombina)on)–  Someaccumula)onofstructure

Oldowan&Acheuleanstonetools:negligibleinnova2onfor1Millionyears!

Language:holis)cprotolanguage,veryliRleanalysis/recombina)on(Wray2000)

Ifsomecomplexity,innova)ondeleterioustofunc)on

Sonorms/pressuresagainstinnova)on

2. Transition to cumulative culture

•  Analytic transmission •  Accumulation of features

–  Extensive innovation (analysis, recombination)

–  Interaction of different traditions

Language:grammar,composi)onality

Veryadvantageous!(Boyd&Richerson1996)

hit(liRleeffort)

cut(lotsofeffort)

withhand

moverepeatedly

Cut(liRleeffort)

If it is so advantageous, why is cumulative culture so rare?

Anadap)vegap

•  Earlycultureposespressure– Forrigidimita)on– Againstanalysisandrecombina)on

•  Cumula)veculture– Requiresanalysisandrecombina)on

If it is so advantageous, why is cumulative culture so rare?

Bridgingthegap•  Largegroups/groupcontactmanycoexis)ngtradi)ons

•  Culturalenvironmentfavours“localenhancement”•  Theculturalenvironmenttriggersitsownevolu.on!

III. Adaptation of linguistic forms to meanings

Adaptation between form structure and meaning structure

Languageisasystemofsigns(Saussure)

MFFORMS MEANINGS

Adaptation between form structure and meaning structure

Adapta)onofmeaningstoforms

Tracesofformstructureinthemeanings

FFORMS MEANINGS

Adaptation between form structure and meaning structure

Adapta)onofformstomeanings

Tracesofmeaningstructureintheforms Focusonthis

MFORMS MEANINGS

kimako

koni

kanige

kuni

winige

komako

Genera.on0:randomsignals

3featuresx3values

27differentmeanings

Artificial Language Learning Kirby, Cornish & Smith 2008

Genera.on1

•  Acorpusofdiachronic‘linguis)c’data

•  8lineages

•  Examineonelineagewithpressureforexpressivity

Artificial Language Learning

Kirby, Cornish & Smith 2008

•  Oneofthefinallanguages

•  Theirresults–  Learningofforms–  Structureofthelanguage(systema)city)–  Composi)onalityisapparent,butitisnotquan2fied

Artificial Language Learning Kirby, Cornish & Smith 2008

Analysis and recombination

•  Evidenceforanalysis–  Independenttransmissionofcomponentsoftheforms

•  Evidenceforrecombina)on– Aretheyrecombinedindifferentways?

 Evolu)onarydynamics–  Formunitsandtheirfrequency– Adapta)ontothestructureofmeanings

keko 2

keke 1

ki 1

kiki 5

kiko 4

kio 1

ko 2

koku 1

ku 1

ni 1

niki 3

niko 4

nuko 1

kehu 1

keko 1

kewa 1

kike 1

kiki 1

kiko 1

guke 1

ko 2

guki 1

hu 1

huke 1

huki 1

huko 1

hupo 1

mu 2

ni 1

nigu 1

nike 1

nimu 1

niwa 1

po 1

wa 2

wagu 1

wapo 1

keko 3

ki 2

kiki 4

kiko 8

kuki 1

kuko 1

neki 1

neko 5

niki 1

niko 1

kiko 7

kuki 8

neki 2

neko 9

nuki 1

keko 1

kiko 8

kuki 9

neko 8

niko 1

kike 6

kiko 2

kuki 9

neke 2

nike 8

keke 1

keko 1

kike 7

kiko 10

neke 6

nike 2

kike 8

kiko 10

neke 4

nike 5

keke 1

ike 1

kike 6

kiko 8

neke 6

nike 5

kike 2

kiko 9

neke 2

nike 14

kiki 1

kike 10

kiko 8

neke 5

nike 3

Segment 3

1 2 3 4 5678910

Evolutionary dynamics - Variant units

Evolutionary dynamics - Frequency

!"

#"

$!"

$#"

%!"

%#"

&!"

!" $" %" &" '" #" (" )" *" +" $!"

," -"

." /"

0" 1"

2"

!"

#"

$!"

$#"

%!"

%#"

&!"

!" $" %" &" '" #" (" )" *" +" $!"

," -"

." /"

0" 1"

2" 3"

• Likegene)cdrik:intheabsenceofvaria)on,onevariantgoestofixa)on

Last consonant

Second consonant

!"#

"#

$#

%#

&#

'#

""#

"$#

(# "# )# $# *# %# +# &# ,# '# "(#

-# .# /#

0# 1# 2#

First consonant

!"

#"

$"

%"

&"

'!"

'#"

'$"

'%"

!" '" #" (" $" )" %" *" &" +" '!"

,-,." ,-,-" /0,." /-,."

,1,0" ,0,0" /0,-" /-,-"

,0,." ,0,-"

Final segment

• Stabilitysuggestsadapta)on

• Innova)on(genera)on5)

frequ

ency

generation

Evolutionary dynamics – Adaptation of forms to meanings

•  Signalsegmentsbecomeassociatedwithdifferentaspectsofthemeaning

–  Ini)alsegment:colour

– Middlesegment:mo)on?

– Finalsegment:mo)on

•  Quan)ta)veanalysis

0 1 2 3 4 5 6 7 8 9 10

0.0

0.2

0.4

0.6

0.8

1.0

Generation

RegMap

!

!

!!

!!

!

!

!

!

!

! Colour Shape MotionSegment 1

Evolutionary dynamics – Adaptation of forms to meanings

0 1 2 3 4 5 6 7 8 9 10

0.0

0.2

0.4

0.6

0.8

1.0

Generation

RegMap

!

!! ! !

!

!

!

!

!

!

! Colour Shape MotionSegment 2

Evolutionary dynamics – Adaptation of forms to meanings

0 1 2 3 4 5 6 7 8 9 10

0.0

0.2

0.4

0.6

0.8

1.0

Generation

RegMap !

!

!

!

!!

!

!

!

!

!

! Colour Shape MotionSegment 3

Evolutionary dynamics – Adaptation of forms to meanings

Evolutionary dynamics – Adaptation of forms to meanings

•  Adapta)onofformscomponentstomeaningcomponentsresultsinComposi.onality

•  Throughanalysisandrecombina)on

“ponekuki”

po‐=red‐kuki=bouncy

“Themeaningofawholeworddependson

themeaningoftheparts,

andtherulestocombinethem”

Artificial Language Learning Population approach

•  Kirby,Cornish&Smith(2008)1par)cipantpergenera)on.Enormousimpactofindividualvaria)on

•  Tamariz&Smith(2008)10par)cipantsforasinglegenera)onmanipulatelevelofstructureintheinput

tunodi

kinaga

kilobe

+

kinoga

kildige

pelobe+

Artificial Language Learning Population approach

•  Kirby,Cornish&Smith(2008)1par)cipantpergenera)onEnormousimpactofindividualvaria)on

•  Tamariz&Smith(2008)10par)cipantsforasinglegenera)onmanipulatelevelofstructureintheinput

• Sensi)vetoinputstructure(p<0.001)

• Alotofinnova)on(Kirby,Cornish&Smithdatashowthatinnova)ontowardscomposi)onalityisselectedfor)

0 1 2 3

0.0

0.2

0.4

0.6

0.8

1.0

Input Languages

RegMap

Artificial Language Learning Musical language

•  Par)cipantsexpect/mayinten)onallycreatepaRernsinwriRenlanguage–Effectofextensiveencultura)on?

•  Brown&Tamariz(submiYed)amusicallanguage,comparetrainedmusiciansvs.non‐musicians

+ +

Artificial Language Learning Musical language

•  Par)cipantsexpect/mayinten)onallycreatepaRernsinwriRenlanguage–Effectofextensiveencultura)on?

•  Brown&Tamariz(submiYed)amusicallanguage,comparetrainedmusiciansvs.non‐musicians

RESULTSFORMUSICIANS

• ComparabletothoseofTamariz&Smith(2008)

• Sensi)vetoinputstructure(p<0.001)

• LiRleinnova)on1 2 3

0.0

0.2

0.4

0.6

0.8

1.0

Input Languages

RegMap

Artificial Language Learning Musical language

•  Par)cipantsexpect/mayinten)onallycreatepaRernsinwriRenlanguage–Effectofextensiveencultura)on?

•  Brown&Tamariz(submiYed)amusicallanguage,comparetrainedmusiciansvs.non‐musicians

RESULTSFORNON‐MUSICIANS

• VerydifferentfromTamariz&Smith2008

• NOTsensi)vetoinputstructure

• Noinnova)on1 2 3

0.0

0.2

0.4

0.6

0.8

1.0

Input Languages

RegMap

Artificial language learning - Conclusions

•  Composi)onality– Analysisofsignalsandmeaningsintocategories/components

–  Independenttransmissionofsignalcomponents– Recombina)onofsignalcomponents

•  Evolu)onarydynamics– Replica)on,varia)on,selec)on– Descentwithmodifica)on

•  Effectsofencultura)on(prac)ceandliteracy)

IV. Summary and conclusions (1)

•  Otheranimalscancategorizetheirenvironment

•  Onlyhumanscancategorize,analyze;imitateandrecombinecomponentsoftheirownbehaviouralproduc)ons

•  Componentsundergoevolu)onarydynamics

–  Replica)on,varia)on,selec)on

" Cumula)veculture

" Increasinglevelsofcomplexity

“Standingontheshouldersofgiants”

Summary and conclusions (and 2)

•  Roleofanalysisandrecombina2oninlanguageorigin,changeanddiversifica)on

•  Uniquelyhumancapaci)es,lowprobabilityofevolving•  Onceitevolvedgene)cally,allowedawholenewsystem,

culture,toevolveindependentlyofgenes

!

Biologicalevolu)o

n‐gen

es

Cultu

ralevolu)o

n–

‘com

pone

nts’