What Syntax is For Pre-Talk Andy Barssbarss/andy_mpp_seminar.pdf · 2007-11-03 · Dell; Berwick;...

What Syntax is ForPre-TalkAndy Barss

1

Tom: Why do we even have syntactic capacity left over?

What would language look like without syntax?

2

This is a language without syntax (click on the image):

3

Optimization:MakingtheMostoftheLeast

AndyBarss,Linguistics&CogSciNovember1,2007

4

Suppose Minimalism is true. Then, major aspects of language follow a Minimization pattern: structures/relations/processes/rule applicationswill be shortest/smallest/least/maximally simple for whatever task is at hand.

(Example tasks: given some proposition p, express p, using a set of lexical items L, that involves the least needed computation).

5

A question that we need to address while exploring minimization effects:

How does the system doing the minimizing “know” what the minimal solution to the task at hand is? (Competence/performance distinction on steroids)

Below I’ll:

show some cases where minimizing/following least action doesn’t require a global comparison across alternatives, but “emerges” from a straightforward

show other cases where two analyses differ in their need for explicit global comparison of alternatives.

raise the issue of how interesting LAP becomes if everything works this way.

6

Minimization:ANon-LinguisticExample(withaneasysolution)

Youhaveafullpitcherandanemptygoblet.

Task:Determinetheminimalpourfromthepitcherthatwillfillthegoblettoaparticularpoint.

7

“Minimal”solutionisreachedfirstinaneasyway:youcan’tpourthreeouncesinwithoutpouringtwoouncesinfirst.

Butothercasesaren’tsosimple.BelowI’lldiscussseveralexamplesfromthelinguisticsworld,someofwhicharelikethepitcher/gobletexample,someofwhicharemaybemoremysteriousandpeculiar.

QuotefromFeynman:“Does[theelectron]smellthenearbypaths,andcheckthemagainsteachother?Theansweris,yes,itdoes,inaway."”

8

Chomsky1986:ProposedaLeastEffortPrincipleforsyntacticmovementoperations.

First,somebackground.

Raising=displacementofanNPfromitssemanticargumentpositiontoastructurallyhigheronewhichgivesitCase.

(1)Itseems[heishappy](2)Itseems[hehasreleasedthedoves]

(1’)Heseems[___tobehappy](2’)Heseems[___tohavereleasedthedoves]

(1’’)*Itseems[hetobehappy](2’’)*Itseems[hetohavereleasedthedoves]

9

Twoclassicquestions:A:Whydoesthisdisplacementhappen?(in1’,2’)B:Whyisitobligatory?(1’,2’vs.1’’,2’’)

(1)Itseems[heishappy](2)Itseems[hehasreleasedthedoves]

(1’)Heseems[___tobehappy](2’)Heseems[___tohavereleasedthedoves]

(1’’)*Itseems[hetobehappy](2’’)*Itseems[hetohavereleasedthedoves]

AnswerstoA&B:

AllNPsneedCase.Finiteverbs(seems,is,has)canprovideittotheirsubjects.Infinitiveverbscan’t.hence,(1’’),(2’’)are*,duetoviolationofthis“CaseRequirement”.

10

ClassicGBpicture:freegenerationofstructures,subsequentfiltrationbyrequiredprinciples.ForanysentenceS,Sisgrammaticaliffitdoesnotviolateanygrammaticalprinciple/filter.

(1)Itseems[heishappy](1’)Heseems[___tobehappy](1’’)*Itseems[hetobehappy]

Newpuzzle:Whyis(3)bad?

(3)*Heseems[___ishappy]

Newtypeofanswer:(3)isbadbecause(1)isfine.

11

(1)Itseems[heishappy]

(3)*Heseems[___ishappy]

(1)and(3)compete.Eachisviewedasadifferentwayofcomposingasentencefromthefundamentalpartshe,seem,is,happy.

(1)doesthejobwithoutmovement.(3)involvesmovement.

Because(1)isgrammatical,themovementin(3)isunnecessary.

Thatis...(3)violatesLeastEffort:Moveonlyifnecessary.

ThisseemedaRADICALIDEAatthetime.

12

Almostimmediateobjection:thischangeswhat“grammatical”means,givingitaglobalflavor.

Howdoesthesystem“compute”theclassofrelevantcompetitorsandtheirderivations,andthencountthenumberofmovementineach,toconvergeontheoptimal(=grammatical)solution?

Chomsky2000:Constructthesentencefromthebottomup,oneMergefollowinganother.

(3)isoutnowbecauseitcannotbeconstructedtobeginwith.Itdoesnotdirectlycompetewith(1).

13

Derivationgoeslikethis:⊕=Merges(combinesintoanewconstituentwith)

a)is[NOM]⊕happy

b)he[NOM]⊕[is[NOM]happy]

c)he[NOM✓]⊕[is[NOM✓]happy]

d)seems[NOM]⊕[he[NOM✓][is[NOM✓]happy]]

e)it[NOM]⊕seems[NOM][he[NOM✓][is[NOM✓]happy]]

f)it[NOM✓]⊕seems[NOM✓][he[NOM✓][is[NOM✓]happy]]

Nocompetition,nochoiceatpoint(d).Caseonheisdeactivated,cannotbe“seen”byhigherverbseems.

So,onthisnewview,it’smuchlikethegobletproblemIbeganwith.

14

Sum so far:two cases of least action/minimizationfirst has an apparent “ procedural” acount (goblet)second has two accounts:competition within a class of competitors, all of which need to be computed and compared --> global quality, much criticism

local, procedural account

15

Case2:AttachmentPreferencesinSentenceComprehension.

3-wayAmbiguity:

“Isawthespywiththebinoculars”a)Isawthespy,andIusedthebinocularstodosob)Isawthatspywhohadthebinoculars(nottheotherspy,whodidn’t)c)Isawthespyinpossessionofthebinocs(akinto“Isawherwithacigarette”)

(a)vs.(b)subjecttomanyexperiments.

Classicresult:Subjects(inEnglish)preferthe(a)readingtothe(b)reading.

16

Frazier:ThisisevidenceofaMinimalAttachmentStrategyinparsing:

GivenanewconstituentX,attachXintotheexistingtreeinwhateverwaywillminimizeaddedstructure.

(a)reading:[VPsaw[NPthespy]]-->[VPsaw[NPthespy]PP]

(b)reading:[VPsaw[NPthespy]]-->[VPsaw[NP[NPthespy]PP]]

(b)ismorecomplexthan(a),so(a)ischosen.

17

ConceptualPuzzler:

IfthesystemispowerfulenoughtobuildBOTHstructuresandcomputetheirdifferencestodeterminewhichissimpler...

whyisitburdenedwiththeMAPtobeginwith?Doesn’tthisargumentshootitselfinthefoot?

Analogy:You’retired,needtogettothesummitofthemountaininanenergyefficientway.So,youhikeallthepossiblepaths,thendecideafterthefactthatoneisbest;soyoupickthatoneandgobackuptothesummit.

18

Analternativeaccount

The(a)readingisconsistentwiththerebeingjustonespy.The(b)readingrequiresthattherebeatleasttwospies.

(b)ismorepresuppositionallycomplexthan(a).

Startwithnopresupps.

Addasneeded.

Yougettoonebeforeyougettotwo.

So,youpreferreading(a),becauseit’sgotfewerpresuppositions.

YoudoNOTneedtocompare(a)to(b)--structurallyorpresuppositionally.Youjustgetto(a)FIRST.

Thismakestheoptimizationagainakintothegobletproblem.

19

RecalltheRaisingexample,andthedebateoverGlobalvs.localminimization.Aparallelcase:dognavigation(fetchingsticks).

20

Meet Elvis. He’s a welsh corgi. He also can do calculus.

21

Elviswillchasethestick.Whatpathwaydoeshefollow?

NOTtheshortestdistance(i.e.thestick’strajectory).

22

Whatpathwaydoeshefollow?Crucially:alongthebeachisfaster,butobliquetothestick.Intothewaterallowsadirectapproachtostick,butisslower.Whathedoes...Partwaystraightalongthebeach.Thenintothewaterandstraightouttothestick.

23

Partwaystraightalongthebeach.Thenintothewaterandstraightouttothestick.

There’saninfinitenumberofsuchroutes.Elvispicksthemostefficientone:LeastTime/LeastAction.HOWDOESHEDOTHAT?

24

Originalaccount:Globalcomputation(Pennings2003).Later:Seriesoflocalcomputations,eachoneassessingsppedofapproachtoball.Jumpintowaterwhen

SAbeach=SAwater(Gallego,J.,andP.Perruchet.2006)

StillLater:No,itreallyisglobal(Pennings200x).Backtrackingwhentrajectorybeginsinwater.

25

Case4:LanguageAcquisitionandparametersetting/rulechoice.

Thefundamentalproblemyou’refamiliarwith:languagesvary,sodogrammarschildrenmustfigureouttheiradultgrammarfrominputinputispositiveonly

Puzzle(originsinGold1967,broughtintolinguisticsbyDell1981,widelyadverttisedbyManziniandWexler1987,Wexler1994,andmanyothers):howtoconvergeontherightgrammar/rulesystemwheninputiscompatiblewithMORETHANONEGRAMMARCHOICEinthehypothesisspace?

26

Severeversionofthisproblem:

GrammarsG1andG2bothgenerateformAGrammarG1doesnotgenerateformB,butG2does

ChildhearsformA,neverhears“notB”.

So,G2generatesasupersetoftheformsG1generates.

HowdoesthelearnerconvergeonG1,andavoidG2?

27

Dell; Berwick; Manzini/Wexler: Learner follows a Subset Principle:

Given a choice in that situation, always choose the grammar that generates the smaller language.

So ... learner doesn’t choose the “larger” grammar, because she is required to follow the strategy of not doing that.

Whew!

28

Evenifwesidesteptheapparentcircularityinthis‘explanation’,

howdoesthechilddecidewhatthebiggergrammaris?

29

30

I can tell you, but I’m out of time! Sorry!

30