Overview Complexity Trade‐offs in Language do not Imply an ...
Transcript of Overview Complexity Trade‐offs in Language do not Imply an ...
07.10.2011
1
ComplexityTrade‐offsinLanguagedonotImplyanEqual
OverallComplexity
GertraudFenk‐Oczlon&AugustFenk
Alps‐Adria8cUniversityofKlagenfurtAustria
XIII International Conference "Cognitive Modeling in Linguistics-2011“ Corfu, Greece, September 22-29, 2011
Overview
• Equaloverallcomplexity?– Complexityavaluablefeature?
– Twoconflic8ngassump8ons
• Somegeneraldefini8onsofcomplexity
• Resultsofaseriesofstudiesshowing– Complexitytrade‐offsbetweenthesubsystemsoflanguage
• Whycomplexitytrade‐offsdonotindicateequaloverallcomplexity
Equaloverallcomplexity?
• Intheperennialdebatewhetherlanguagesdifferintheircomplexity,quiteanumberoflinguistsholdtheopinionthatalllanguagesareequallycomplex.
Twoconflic@ngassump@ons
• Thismightbeinteraliaanunderstandablereac8onto19thcenturytypologists:
• Schleicher(1850)forinstanceclassifiedmorphologicallylesscomplexlanguagesasinferiorandhighlyinflec8onallanguages“asthepinnacleoflinguis8cevolu8on”
07.10.2011
2
Twoconflic@ngassump@ons
• Butthisassessmentof“morecomplex”asafavorablequalitywasalreadychallengedbyJespersen(1894)whoconsiders,usingEnglishasanexample,lesscomplexityintheinflec8onalsystemandatendencytoisola8ngmorphologyasaprogressinlanguage.
Complexity:avaluablefeature?
• Complexityperseisnotnecessarilyvaluable:Ifalanguagecanexpresswhatisrequiredwithlessgramma8calorphonologicalcomplexitythanthiscouldbeseenratheranadvantageorasmoreefficient.
Parkvall(2008)trea8ngtheequalexpressivenessoflanguagesasaxioma8c,arguesthat“asimplelanguagecouldratherbeseenasamore‘efficient’one,inthesensethatitdoesthesamejobatalessercost”.
Complexityandhierarchy
H.ASimon,inhisfamousar8cleon“TheArchitectureofComplexity”(1996),viewshierarchyasanuniversalprincipleofcomplexstructures,
andbyacomplexsystemhemeans“onemadeupofalargenumberofpartsthathavemanyinterac8ons.”(183f)
“Complexity”ingeneral
Ahierarchyofcriteriaforcomplexity:• numberofcomponents• complexityofcomponents• numberofdifferentcomponenttypes• numberofpossibleinterac@onsbetweencompounds(Simon)
• numberofdifferentrulesdeterminingtheseinterac@ons(Gell‐Mann)
07.10.2011
3
Ahierarchyofcomplexityinlanguage:numberofcomponents,complexityofcomponents
Complexityof
• syllables:numberofphonemes(1)• words:(1)&numberofsyllables(2)
• clauses:(1)&(2)&numberofwords(3)• sentences:(1)&(2)&(3)&numerofclauses(4)
Remarks:
• con8nua8ons:complexityofphonemesandoftexts• mono‐phonemicsyllables;monosyllabicwords;“mono‐clausal
sentences”
Assigningfeaturesofcomplexitytosubsystemsoflanguage
• Phonology:sizeofphonemicinventory,ar8culatorycomplexityofphonemes,numberofphonemespersyllable,...
• Morphology:wordforma8on(numberofmorphemsandnumberofsyllablesperword),numberofcases,genderdis8nc8ons;transparencyvsopaquenessofmorphologicalforms,..
• Syntax:rigidvsfreewordorder,SVOorothers,paratac8cversushypotac8cconstruc8ons,....
• Seman@cs:homophony,polysemy,idioms
1stseriesofstudies:7plusminus2syllablesperclause?
General assump@on: Natural languages have developed inadapta8ontogeneralcogni8vefunc8onsandconstraints:e.g.workingmemorylimita8ons,psychologicalpresent
• Hypothesis(Fenk‐Oczlon1983):Thenumberofsyllablesperclause(proposi8on)willvary within the range of Miller’s magical numbersevenplusorminustwo.
•
7plusminus2syllablesperclause?
• Method:
Na8vespeakersof33(17Indo‐European,16non‐Indo‐European)languageswereaskedtogiveawrilentransla8onof22German/Englishclausesintotheirmothertongueandtodeterminethelengthoftheirtransla8onsinnumberofsyllablesandnumberofwords.
07.10.2011
4
Sentences
(1)Thechildiswai@ngforitsmeal.
(2)Thesunisshining.
(3)Bloodisred.
(4)Ithinkofyou.(5)Ourneighbourisafarmer.
(6)Shetrustsherfriend.
(7)Shesings.
(8)AfatherlooksaYerhisfamily.
(9)Thegirlisindustrious. (10)Ithanktheteacher.
(11)Thespringisontheright.
(12)Mygirlfriendishelpingme.
(13)Mybrotherisahunter.
(14)Thewateriscold.
(15)Thedogisoutside.(16)Myfatherisafisherman.
(17)Grandfatherissleeping.
(18)Amotherlovesherson.
(19)Auntyisathome.
(20)Mysisteriscollec@ngwood.(21)Heisbuildingahut.
(22)It´sraining.
Results • The mean number of syllables per clause
was found to be located almost exactly in the range of Miller’s 7 plus minus 2 elements
The lowest size was a mean of 5.05 syllables (Dutch), and only Japanese with 10.2 syllables per clause was located outside the hypothesized range of 5-9 syllables; the overall mean was 6.8 syll/clause
• The mean number of words was about 4, ranging from 2.5 in Arabic to 4.4 in English (→ Cowan’s number 4 plus minus 1)
Themeannumberofsyllablesperclausein34languages
Syllable complexity as decisive factor (Fenk-Oczlon & Fenk 1985)
Dutch is known for its complex syllables (e.g. CCVCC), Japanese for its simple syllable structures (CV, V). Hypothesis: The more syllables per clause, the fewer phonemes per syllable?
Result: A statistically significant negative correlation between the number of phonemes per syllable and the number of syllables per clause.
r = – 0.77, p < .01
07.10.2011
5
Furthercrosslinguis@ccorrela@ons
• The more syllables per word, the fewer phonemespersyllable.r=–0.54(p<0.1%)
• Themoresyllablesperclause,themoresyllablesperword.r=+0.47(p<1%)
• Themorewords per clause, the fewer syllables perword.r=–0.66(p<0.1%)
• High syllable complexity is significantly associatedwithVOorder.
Fenk&Oczlon&Fenk1999
Anextendedsampleof51languagesFenk‐Oczlon&Fenk2010
• The applica8on of this method in a meanwhileextended sample of 51 languages corroborated thepreviously found nega8ve correla8on betweennumber of syllables per clause and number ofphonemespersyllable.
Themoresyllables,thefewerphonemespersyllable
• The nega8ve correla8on between number ofsyllablesandnumberofphonemesproved tobeveryrobust:Acoefficientof–0.77inasampleof34predominantlyIndo‐Europeanlanguagesand–0.73inthesampleof51predominantlyNon‐Indo‐Europeanlanguages.
Themeannumberandtherangeofthenumberofsyllablesperclause
• Themeannumberofsyllablesperclauseacrosslanguagesshisedfrom6.48intheprevioussampleto7.02intheextendedsample.
• Therangealsoincreased:Ameanof4.64inThaiandameanof10.96inTelugu.
07.10.2011
6
The mean number of syllables per clause in 51 languages
from Fenk-Oczlon & Fenk (2010:1538)
Monosyllabismandsyllabiccomplexity
Across‐linguis8chypothesis:
Thelargerthepropor8onofmonosyllablesintextualmaterial,thehigherthemeansyllablecomplexity.
Method
32languageswerecomparedwithrespecttotheirpropor8onofmonosyllablesintextualmaterial.The“text”wasourcontrolledsetof22simplesentencesthatweretranslatedbyna8ve‐speakersintotheirmothertongue.
Languagesample• 13Indo‐European:Armenian,Bulgarian,Dutch,English,German,
Icelandic,Latvian,Norwegian,Polish,Romanian,Russian,Tajik,Welsh
• 4Austronesian:Indonesian,Javanese,KaroBatak,Malagasy
• 3NigerKongo:Bambara,Kirundi,Yoruba
• 2Uralic:Hungarian,Finnish,• 2Sino‐Tibetan:Mandarin,Cantonese
• 1Austroasia8c:Vietnamese
• 1Kartvelian:Georgian
• 1Tai‐Kadai:Thai
• 1Dravidian:Telugu• 1Uto‐Aztecan:Hopi
• Navaho
• 1Macro‐Ge:Chiquitano
• 1Australian:Marranju
07.10.2011
8
Results:propor@onofmonosyllables
Thehighestpropor8onofmonosyllableswasfoundinDutch(67outof88words,i.e.,76.14%)andinEnglish(73outof96words,i.e.,76.04%),thelowestintheAboriginallanguageMarranju(1outof83words,i.e.,1.2%).
Themeanvaluewas20.56%.Languageswithmorethan65%monosyllablescouldbelocalizedinnorthwesternEurope:Dutch,English,Welsh,GermanandFrenchshowedhighervaluesthanthetradi8onaltextbookexamplesVietnamese(63%)andMandarin(62%).
Results:monosyllablesandsyllabiccomplexity
Thehypothesispredic8ngaposi8vecorrela8onbetweenmonosyllabismandsyllabiccomplexityturnedouttobesignificant:
Themoremonosyllablesintextualmaterial,themorephonemespersyllable.r=+.62(p<.001)
Complexitytrade‐offsbetweenmorphologicalandphonologicalcomplexity
Thenega8vecorrela8ons“thefewersyllablesperword,themorephonemespersyllable”and/or“themoremonosyllablespertextualmaterial,thehigherthemeansyllablecomplexity”canbeinterpretedasacomplexitytrade‐offbetweenafacetofphonologicalcomplexityandafacetofmorphologicalcomplexity.Thistrade‐offbetweensyllablecomplexityandwordlengthseemstobeuniversalandcanalsobeobservedinpidginlanguagesthataremostcommonlysupposedtobelanguageswithlowcomplexity.
Complexitytrade‐offsinPidginlanguages
PidginlanguagestendtosimplifysyllablestructureandaccordingtoHall(1966)thepidginlanguagesNeo‐MelanesianandChinesePidgin‐Englishhave,ascomparedwiththeirlexifierlanguageEnglish,amuchhigherpropor8onofbisyllabicwords.
Assyllablestructuresaresimplified,thenumberofsyllablesperwordincreases.
e.g.: Chinese‐Pidgin: piece→piecee wife → wifoo
Kamtok: what→we8
07.10.2011
9
Complexitytrade‐offsdonotindicateanequaloverallcomplexity
Butthistrade‐offbetweensyllablecomplexityandnumberofsyllablesperword(ornumberofsyllablesperclause)doesnotimplyanoverallequalcomplexityoflanguages.Wewillreinforceourargumentsagainstsuchaviewontheoccasionofsomemisinterpreta8onsofourpreviousstudies.
Althoughweexplicitlystatedin(2008)thatthetrade‐offwehavefoundbetweenfacetsofphonologicalcomplexityandmorphologicalcomplexity“bynomeanssupports/…/theideaofanequaloverallcomplexityinnaturallanguages”itise.g.arguedin:
Complexitytrade‐offsdonotindicateanequaloverallcomplexity
• Maddieson(2009):Fenk‐Oczlon&Fenk(2008)answertotheques8onwhetheritistruethatalllanguagesareequallycomplex“intheaffirma8ve”.Orin
• Glaudert2009:“Firstofall,Ishallargue,followingFenk‐OczlonandFenk(2008),thatthesubsystemsofanylanguagearegovernedbycomplexitytrade‐offs,whichleadsustopostulatethatalllanguagesare,roughlyspeaking,equallycomplex.”
Complexitytrade‐offsdonotindicateanequaloverallcomplexity
• Somearguments,whycomplexitytrade‐offsdonotindicateanequaloverallcomplexityoflanguages:– Firstofallforthegeneralreasonthatsyllablestructureandwordstructureareonlytwofacetsoutofanunknownnumberoffacetsthatmaycontributetoalanguage’stotalcomplexity.
– Morespecificallybecausewordlengthitselfisdeterminedbydifferentmorphologicalcomplexityfactors.e.g.:
Complexitytrade‐offsdonotindicateanequaloverallcomplexity
Wordlengthitselfisdeterminedbydifferentmorphologicalcomplexityfactors.e.g.:Monosyllabicwordrootsandfusionaloragglu8na8ve
morpholoy
Engl:sg.hand/pl.hands
tosit/shesits/shesat
Disyllabicwordrootsandisola8ngmorphologyIndonesian:sg.orang‘person’/pl.orang(orang)
duduk‘tosit’,pergi‘togo’,7dur‘tosleep’
(verbsdonotchangeaccordingtosubjectandtense)
07.10.2011
10
Complexitytrade‐offs:anon‐linguis@cexample
Anon‐linguis8cexample:
Alowbudgetuniversitydepartmentmayrecordastrongnega8vecorrela8onbetweenthenumberofprintoutsandthenumberofcopiespersinglemember:themorecopies,thefewerprintouts,andviceversa.Butthiscorrela8onisfullycompa8blewithmembersproducingbothfarmoreprintoutsANDfarmorecopiesthanothersorproducingatanyrateahighersumofprintoutsandcopies.
Someforthertrade‐offs:Englishasanexample
• Englishhasahighnumberofmonosyllabicwords(roughly8000accordingtoJespersen)
• ahighnumberofhomonyms,homophones,lexicalandgramma8calpolysemy
• ahighnumberofidioms
– e.g.phrasalverbs:theyareidioma8c,becausetheirmeaningcannotbederivedfromthemeaningofeachwordseparately.Theverbandthepreposi8on/adverbformingthephrasalverbareosenpolysemousandMONOSYLLABIC:inananalysisof1406phrasalverbswefoundthat1367or97%weremonosyllabic:
EnglishascomparedtoRussian(Polikarpov1997)
• Wordlength:Russianwordsareontheaverage1.48meslongerthanEnglishwords
• Polysemy:Englishwordshaveontheaverage2.7meanings,Russianwordsonly1.7
• HomonymsinEnglish:atleast2000HomonymsinRussian:~500
• IdiomsinEnglish:~30000IdiomsinRussian:~10000
highphonologicalcomplexity:e.g.largephonemicinventoryhighnumberofsyllabletypescomplexsyllablestructures
lowmorphologicalcomplexity:e.g.monosyllabism
lownumberofmorphemesperword
highwordordercomplexity:e.g.rigidwordorder
highnumberofwordorderruleshighnumberof(lexical)colloca@onrules
formulaicspeech,idioms
part‐of‐speechambiguitygramma@calpolysemy
lexicalambiguityhomonymy,polysemy
from Fenk-Oczlon & Fenk (2008: 61)
e.g. English:
07.10.2011
11
Equaloverallcomplexityoflanguages?
• Weassumethatlanguagesmaywelldifferintheiroverallcomplexity.Butaslongasitisimpossibletoquan8fytheoverallcomplexityofasinglelanguageitisalsoimpossibletocomparedifferentlanguageswithrespecttothatquan8ty.
Equaloverallcomplexityoflanguages?
• Shosted(2006)concludes:“thedictum‘Alllanguagesareequallycomplex’isdogma8c.Statementsofthissortshouldbeusedwithgreatercau8on–ifnotdiscardedaltogether–un8lsuch8measfalsifiable,quan8ta8veevidenceofcorrelatedcomplexityisbroughtforward.”Weagreeandareevenpessimis8cfortheremotefuture:
Equaloverallcomplexityoflanguages?
• Weagreeandareevenpessimis8cfortheremotefuture:
• Thenumberofcross‐linguis8ccorrela8onsindica8ngcomplexitytrade‐offsmaywellincrease.Suchcorrela8onsareinteres8ngforourunderstandingofsinglenaturallanguagesasself‐organizingsystems.
• Suchinferenceswereuntenableevenifallrelevantparameterscontribu8ngtocomplexitywereascertainable,asillustratedinouraboveexamplewithprintoutsversuscopies.Thisisratherunpromisinginfaceoflanguagesystems:Howshouldweeverknowifthelistavailableatagiven8meisreallyexhaus8ve?
Equaloverallcomplexityoflanguages?
• The“equaloverallcomplexity”‐hypothesisisa0‐hypothesis,i.e.,ahypothe8calnega8onofpossiblecomplexity‐differencesdependingonthespecificlanguage.Anda0‐hypothesiscannotbereallycorroborated;itcanonlyberefuted,andthisthroughasta8s8calcorrobora8onof“posi8ve”differences.
07.10.2011
12
Conclusion
Complexitytrade‐offsbetweenthesubsystemsofalanguagefavorarela8vely“constant”flowofinforma8onwithinthatlanguageaswellasatendencytothemeanoftheoverallcomplexitybetweenlanguages.Buttheydonotatallexcludelanguage‐specificdifferencesinthatoverallcomplexity.
Thankyouforyouraien@on!
SelectedReferences
• Fenk‐Oczlon,G.(1983).BedeutungseinheitenundsprachlicheSegmen8erung.EinesprachvergleichendeUntersuchungüberkogni8veDeterminantenderKernsatzlänge.Tübingen:Narr
• Fenk‐Oczlon,G.&Fenk,A.(1985).
• Fenk,A.&Fenk‐Oczlon,G.(1993).Menzerath’slawandtheconstantflowoflinguis8cinforma8on.InR.Köhler&B.Rieger(Eds.)Contribu7onstoQuan7ta7veLinguis7cs.Dordrecht:KluwerAcademicPublishers,11–31.
• Fenk‐Oczlon,G.,&Fenk,A.(2008).Complexitytrade‐offsbetweenthesubsystemsoflanguage.InM.Miestamo,K.Sinnemäki&F.Karlsson(Eds.)LanguageComplexity:Typology,Contact,Change.Amsterdam/Philadelphia:JohnBenjamins,43–65.
• Fenk‐Oczlon,G.&Fenk,A.(2010).Measuringbasictempoacrosslanguagesandsomeimplica8onsforspeechrhythm.Proceedingsofthe11thAnnualConferenceoftheInterna8onalSpeechCommunica8onAssocia8on(INTERSPEECH2010),Makuhari,Japan,1537‐1540.
• Glaudert,N.(2009).Thetheoryofmarkednessandforeignlanguageteaching.InA.Shafari&M.Neja8(Eds.)AnnalsofLanguageandLearning.Proceedingsofthe2009Interna7onalOnlineconference(IOLC2009).
SelectedReferences
• Jespersen,O.(1894).ProgressinLanguage:withSpecialReferencetoEnglish.London:SwanSonnenschein&Co.
• Maddieson,I.(2009).Calcula8ngmorphologicalcomplexity.InF.Pellegrino,E.Marsico,J.Chitoran&C.Coupé(Eds.)ApproachestoPhonologicalComplexity.Berlin/NewYork:MoutondeGruyter,85‐109.
• Parkvall,M.(2008).Thesimplicityofcreolesinacross‐linguis8cperspec8ve.InM.Miestamo,K.Sinnemäki&F.Karlsson(eds.)LanguageComplexity:Typology,Contact,Change.Amsterdam/Philadelphia:JohnBenjamins,265‐285.
• Shosted,R.K.(2006).Correla8ngcomplexity:Atypologicalapproach.Linguis7cTypology10‐1,1‐40.