cc: recipient list not shown: ; - home | princeton universityadele/papers/papers/cp.doc · web...

43
Cc: recipient list not shown: ; \parskip=2pt \documentstyle[lingmacros,11pt]{article} % Set point size to 11 \pagestyle{plain} % The paper itself starts here. \begin{document} \title{Complex Preds} \section{Farrell Ackerman and Lesourd, Phil ms. Toward a Lexical Representation of Phrasal Predicates. UC San Diego.} (some notes were taken on an earlier draft: double check page numbers. Discussion of Fox is left out) PV V complexes in Hungarian and Fox involve ``functional words" (argument taking predicates), while they are not zero level categories. That is, they are composed in the lexicon. Diachronic development: lexical representation (functional word) is realized by independent syntactic elements, the independent syntactic forms begin to take on morphological status though the pieces may be separable, finally, the morpholigcal entitye begins to exhibit phonological coalescence and syntactic atomicity. Lexical Integrity Hypothesis: (Simpson 1992): "syntactic rules can neither analyze nor change word structure."

Upload: vanhuong

Post on 06-Jul-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

Cc: recipient list not shown: ;

\parskip=2pt\documentstyle[lingmacros,11pt]{article}   % Set point size to 11\pagestyle{plain}

%  The paper itself starts here.

\begin{document}\title{Complex Preds}

\section{Farrell Ackerman and Lesourd, Phil ms. Toward a LexicalRepresentation of Phrasal Predicates. UC San Diego.}(some notes were taken on an earlier draft: double check page numbers.Discussion of Fox is left out)

PV V complexes in Hungarian and Fox involve ``functional words"(argument taking predicates), while they are not zero level categories.That is, they are composed in the lexicon.

Diachronic development: lexical representation (functional word) is realizedby independent syntactic elements, the independent syntactic forms beginto take on morphological status though the pieces may be separable,  finally,the morpholigcal entitye begins to exhibit phonological coalescence and syntacticatomicity.

Lexical Integrity Hypothesis: (Simpson 1992): "syntactic rules can neitheranalyze nor change word structure."

They propose the revised Lexical Integrity Hypothesis: Syntactic rules(e.g. principles of phrasal organization) cannot alter thestructure of syntactic words.

"...if lexical integrity can be claimed to hold overanything, it presumably obtains for X0 level categories.Inasmuch as we are arguing that not all ocmplex morphologicalobjects are integrated (i.e. synthetic) X0 level categories,it would be mistaken to identify lexical integrity withthe morphological status of an object. Synthetic morphologicalexpressions represent a strong diachronic tendency: they do notfollow from a principle of grammar concerning the nature of morphologicalobjects." pg 55

Hungarian PV V exhibit properties which are both morphological (e.g. nominalizations) and phrasal (separability)

PV can be separated by V by: 1) auxiliaries, negative morphemes, "also" morpheme...[CLOSED class elements, like Persian!]

\section{Alsina, Alex. ms A theory of Complex Predicates:Evidence from Causatives in Bantu and Romance. March 1993.}

Causative constructions in both Chichew\^a and Catalan are identialat the level of argument structure, but they differat the level of phrase structure:  the complex predicateis represented by a single verb in Chichewa (and thusis 'formed in the lexicon'), but by two verbs in Catalan ('formed in the syntax').

There is a discussion of AA's mapping theory.His Proto-Role classification is discrete (not gradientlike Dowty).[Lots of oversimplifications here: e.g. all obliques aresaid to be optional, no real semantics is given, noroom for lexical exceptions.]

Two binary features, but only three possibilities:$Obj_{\theta}$ is a type of Obj.[What about the fourth possibility?]

The differences in where the complex predicates areformed is made to explain the differences in:\begin{enumerate}\item coordination (Catalan allows it, Chichewa doesn't)\item separabiltiy (Catalan allows it, Chichewa doesn't)\item nominalization (Chichewa allows it, Catalan doesn't)\end{enumerate}[What about all the languages that fall in between?What does `formed in the syntax' mean?  If the arg-struc is builtup compositionally in the syntax, then isn't thelinking from arg-struc to sytax a little after-the fact?]

\section{Butts, Miriam. The Structure of Complex Predicates in Urdu}(Urdu is basically the same as Hindi)Permissive Complex Predicates

She notes the problem of having two semantic heads, thepermissive ("Let") and the (main) verb, both having up equalsdown arrows.  In footnote she acknowledges the possibilility of Light-Pred andHeavy-Pred features, but notes that there can also be two Light-Preds.

She follows Alsina 1992 and Rosen 1990, insuggesting that the verbs combine by Argument Fusion at a-structure.

Aspectual Complex Predicates

There is split ergativity in Urdu.  Ergative case is assigned when theargument has "conscious choice"\eenumsentence{\item He-nom scream.\\He screamed (despite himself)\item He-erg scream.He screamed (on purpose).}

{\bf Light Verbs taking ergative subject}: le "take", de "give", daal "put", maar "hit" nikaal "pry out"

{\bf Light verbs taking nominative subject}: aa "come" jaa"go", par "fall", mar"die", nikal "emerge"cuk "finish", bait "sit" ut "rise" (see paper for IPA diachritics)

(the light verbs, and not the main verbs determine the case marking of the subject)

If the semantics of the main verb and the light verb are incompatible, then thecomplex is ill-formed:\enumsentence{*He-erg fell put\\He fell on purpose.}[OK in the context of a play?]The light verb cannot impose its semantic contribution on the complexpredicate.

Light predicates have aspectual properties as well.  MB:  "most verbsin Urdu are underspecified as to their [aspectual dimension]." pg 17

MB uses Butt, Isoda and iells (1990)'s idea of a "transparent event":this triggers complex predicate formation at the level of a-structure;a transparent event is an incomplete predicate looking for another argumenttaking predicate.

No thematic tier (the representation is Jackendoff's) is posited for lightverbs; it is this sense in which they are light.

\section{Doostan, Gholamhossein Karimi.  Chapter 4.Chapter 4 Light verb constructions in Persian and Kurdish.University of Essex dissertation.}See to.karimi and to.karimi1 in RESEARCH/Persian

\section{Durie, Mark. ms. Grammatical Structuresin Verb Serialization: some preliminary proposals.University of Melbourne.}

MD discusses two types of serial verb constructions inPamese, an Oceanic language of Vanuatu.

\begin{enumerate}\item a single serial verb complex describes a singleevent\itemthe serial complex has shared tense, aspect, modality and polarity\itemserial verbs "share" at least one and possibily more args

\item the complex takes only one subject/external arg\item there is a very strong diachronic tendency to lexicalization: this can involve treating the whole serial complexas a single lexcical item, or `demotion' of hte meaingand grammatical status of one of the verbsto that of a modifier or casemarker\end{enumerate}

Two types of serial verb complexes in Paamese:

CORE serialization: each verb retains morphologicalmarking for subject agreement.

NUCLEAR serialization: subject \& obj agreementand mood marking only occur once, and one verb stem followsthe other with no intervening morphological material.(This is not compounding (P also has compounding) because verbs are phonologicallytreated as two distinct words.)

MD argues strongly against Baker's account, accusinghim of faking data, ignoring counterexamples, etc.

Some ordering generalizations:\begin{enumerate}\item in instrumental serialization, the verbwhich introduces the instrumental arg alwayscomes first.\enumsentence{`` take axe cut tree."\\``knife take cut"}\item In cause-effect serialization, the verb indicating

a causing deed, e.g. 'hit' comes first.\item In benefactive/goal serialization the main eventverb precedes the verb that contributes the goal orbenefactive role.\end{enumerate}These generalizatioons hold in SVO and SOV languagesshowing serialization in different ways.MD's explanation: the verb sequencing is 'iconic' :the direction of causation will conform to temporalsequencing, where this is applicable, but htedirection of causation does not always allow a temporalordering.  The iconic motivation is often grammaticalized.

Two distinct patterns e.g. in an SVO lang:  SV..V (O) or SV(O)V(O)V(O)

Three generalizatoins:\begin{enumerate}\item verb serialization is univerally characterizedby heavy lexicalization of particular verb combinations(e.g. "burn + make sound" = "start up an engine"--cannot be used to describe a car burning up noisily.;``squeeze+remove = extract"; ``throw+arrive" = spear).\item at the same time, serialization is productive;

One pattern: many serializations in Sranan (Sebba) includeone verb (drawn from a small set of around 40) whoseposition and semantic contribution is fixed (thisseems like the light verbs of Miriam Butts)

Second pattern of productivity:  event schemas, e.g. inKalam (described by Pawley):

\begin{shortex} {5}{I & II & III & IV & V }{movement.to.scene & action.at.scene & movement.to.next.scne & action.at.scene & movement.away.from scene}{}\end{shortex}

\item Non-events are not acceptable:

*She take the fish buy.  (the order is wrong: can'ttake the fish and then buy it)

(cf. She take the fish sell) 

*He ate return (cf. he ate sleep (normal siesta activity))\end{enumerate}

The serial complex as a whole cannot contain duplicateroles.  There cannot be two agts, two patients, twoinstruments, etc.  ``This is one of the most importantways in which verb series act like single verbs." Multiple objects are sometimes allowed, but notif they have the same semantic role.

Sometimes the arg shares two different roles fromtwo different verbs: e.g. `these young women' seemsto be the theme of 'go' but the agent of 'dig.':

woman young these go dig hit carrying come``The young women dig and fetch (these animals)

MD suggests we aim toward an account of argument fusion:a complex like I-take-stick-hit-Bill would have a fused argumentstructure consisting of Agent Instrument and Patient,existing alongside the independent argument structures ofthe two verbs.He reasons for this suggestion:\begin{enumerate}\item the constraint against role-doubling onlymakes sense at the level of a fused argumentsstructure fro the whole serial complex\item the linking requires a fused arg structure\item the sematnic contribution of args canoften only be unerstood in the context of the whole serialcomplex\end{enumerate}

\section{Evans, Nick. ms. Role or Cast? Noun incorporationand complex predicates in Mayali. (do not quote).}

Trivalent verbs derived through benefactive and comitative applicativeconstructions in Mayali: polysynthetic Australian languageof the Gunwinyguan family.

30 incorporable nominals (excluding body parts), theonly human nominal regularly incorporated is "child"

E raises the possibility that some transitive subjectscan incorporate, but is not completely convinced thatthese are truly transitive.

{\bfThematic Role Account:}

Mithun: patients are preferentially incorporated,with some languages also allowing instr and locations.

IO don't incorporate, because they are too high onthe hierarchy.

There is a complementarity of overt codingbetween pronominal affixes and oncorporated nominals: theformer only encode full info for human referents, whilethe vast majority of incorporable nouns refer to inanimates.

{\bf The problem for Thematic role account:Comitative verbs}\enumsentence{see OBJ with COM\\hear OBJ with COM\\feel OBJ for COM (frisk OBJ for COM)\\hit OBJ over/for possission of COM\\follow OBJ inbeing with COM (e.g. when twomen take it in turns to sleep with the same woman)\\clean OBJ out of COM.}NE gives these thematic roles theme or location

\enumsentence{leave OBJ with COM\\put OBJ with COM\\get OBJ from COM, from COM's possession\\hide OBJ with COM}NE gives these all thematic role location[These could be said to have a different thematicrole: possessor]

NE argues that because of these cases, we need tostate the restriction on incorporated nominals:the argument whose prototypical reference is more likelyto be inanimate is able to incorporate

[Doesn't seem to me to be in conflict with thematic role acct: ifroles are assigned differently--one could say that roles lower thanor equal topatient incorporate, because higher roles are prototypicallyanimate.  Would require COM to cover two distinct roles; otherwise,maybe verb specific roles that, since they are associated with framesof the verb, {\em could} capture prototypical information.]

MORAL: only inanimates incorporate (cf. Nunberg, Wasow and Sag stuff)

\section{Ghomeshi, Jila and Diane Massam. 1994. To Appear,{\em Linguistic Analysis}. Lexical/Syntactic Relations without Projection.}

Allows meaning for grammatical constructions.Syntactic canonical object: the DO in the prototypicaltransitive clause (Hopper and Thompson).

Presentational aspect: e.g. English imperfect -ing

Aktionsart: Dowty/Vendler classes (applied to predicatesactually)

Lexical aspect: idiosyncratic properties of verbs

Argues that Persian NI is best explained by allowing thebase-generation of object NPs in more than one position.Cites Mithun's 4 classes of NI.

\begin{enumerate}\item Type 1: Lexical Compounding: V+N denotes a conventionalized activity\begin{enumerate}\item  Type 1a: composition by juxtaposition: V and O form a tight bondbut remain phonologically separate\item  Type 1b: morphological compounding: V and O form a single word\end{enumerate}\item Type II: manipulation of case is possible so that V+N unitcan function as a trasnitive unit with an oblique argument advancingto DO status\item Type III: manipulates discourse strucutre, such that known orincidental information can be backgrounded by means of NI\item Type IV: classificatory NI: N narrows the scope of the verbsuch that a more specific object NP or part thereof canalso appear\end{enumerate}Mithun proposes implicational hierarchy: lng that has Type IV will alsohave all the others, lng that has Type III will have Type I and Type II.

``Persian has only Mithun's type I NI in that it does not allowan oblique arguent ot be advanced intot he case position vacated bythe incorporated noun..." [WRONG!]

(aside: Mohamad and Karimi who treat compounds like {\em dars xAndan}(compositional)to be fundamentally differentthan compounds such as {\em rang zadan} "to paint.")

G and M do not make this claim. Instead they argue thatthere is an aspectual diffrence with CPs being less bounded.

They argue against a movement analysis since a number ofthe compound verbs do not involve DOs, they coudlnot be derived by movement since they would involve a non-properlygoverned trace.  Other args against movement analysis aregiven in Heny and Samiian adn Mohammad and Karimi.

They do not treat CPs as lexical because of hte -eshclitic: "Since compounding is generally assumed to be derivationalas opposed to inflectional, it must occur prior toaffixation of enclitics.  Threfore if the Persian process underdiscussion was an instance of lexical compounding, the encliticcould not occur inside the compound." pg 23Also, compounds can be separated by phrasal material. They cite:

CHECK THESE:\enumsentence{Hasan maSin az in mard xarid \\Hasan car from this man bought+3sg \\'Hasan bought a car from this man.'(Heny and Samiian 195:8)}

\enumsentence{gush be man ne-mi-kon-e\\ear to me NEG+PROG+DO+3sg\\'She doesn't listen to me'\\(Mohammad and Karimi 197:7)}

M \&K reference cited as having othe reasons against a lexicalapproach.

Heny and Samiian propose a Reanalysis soln.

G\& M's solution:V' $--> $ V0 $-->$ X0 V0  (allows base-generated adjunction structures at the X0 level)

``Note that above the V level the structure is identical to that of anintransitive verb. This accounts for the observation made by Mithunthat in Type I NI the N+V unit functions as an intransitive predicate.In Persian this decreased transitivity...has aspectual consequences...the[setences involving juxtaposition] denote processes [not accomplishments]."[WRONG--not all Persian CPs are processes]

\section{ Givon, T. 1991. Isomorphism in the Grammatical Code:Cognitive and Biological Considerations. {\em Studies in Language}15-1. 85-114.}

Quantity Principle

Proximity Principle

Co-leixcalization and event integration: ``The moreintegrated the two events are, the more likely is thecomplement verb to be co-lexicalized--i.e. appearcontiguously with the main verb" pg 95.

Relative proximity of grammatical morphemes to the stem to indicateconceptual {\em scope} relations as in the ordering of tense-aspect-modality morphemes (Givon 1982) or other verbal categories (Bybee 1985)

\section{Hale, Ken and Jay Keyser. On the Complex Nature ofSimple Predicators.}

Unambiguous Projection (i.e. unambiguous paths Kayne 1984: binarybranching) is assumed: I don't see any argument for it.The relation sister holds unambiguously between V and VP,a ndbetween NP and V.

Only V and P take complements

By hypothesis, laugh, sneeze, dance, shelve, corral, box, saddle,blindfold, bandage, clear, narrow, lengthen all involve incorporation.

Unergative verbs: $~[_{V*} ~ V1~  [_{NP}~ Ni~]~ ]~ $, whereNi is e.g. the noun {\em dance}.

What about unergative verbs like {\em speak, grimace, skate, jog,mumble, quack, sleep, cry}: need to posit hypothetical lexical Ns?

ECP: X0 must properly govern any head which incorporates into it.If a noun N is external to V*, e.g. if N is the subject (or adjunct),then incorporation is prevented.

Syntax is projected from the lexicon

Argument Structure = syntactic structures defined in lexical relationalstructure (LRS).

Assumes the UTAH: idential thematic relationships between itemsare represnted by identical structural relationships.

They ask, why are there so few semantic roles? (are there?)Answer: there are no roles: just places in the tree.

Category V is assocated with "event" (e)

\section{Heny, Jeannine and Vida Samiian. 1991. ThreeCase of Restructuring in Modern Persian. In WECOL. K. HuntT. Perry and V. Samiian (eds). Cal State U, Fresno. 191-203.}

Do not specify whether their rule is in morphology orsyntax. ``...the process we propose must be capableof producing `words'. In fact we will claim that Restructuringof the sort we envision can produce output on both theword and phrasal levels, depending on the operation ofother highly general principles which constrain the processas well as its output." pg 193

Argueagainst syntactic movement: is non-productive, donot allow modifier stranding or specific reference of N.Lack of any clear motivation for moving in the grammar.

They focus on cases in which the N-V constructionhas counterparts which clearly involve NP + V syntacticformation:  e.g., {\em harf-ha-yeS-ra bA dustan-eS zad}``he said the words he had to say to his friends."

They note, ``no material may intervene between N and V ifN is unmodified" (examples have adjuncts:\enumsentence{*harf bA dustan-eS zad \\words with friends-his hit}\enumsentence{*Bush tasmin dar in jalese gereft ke...\\Bush decision at this meeting took that}

They note that non-compound cases *do* allowintervening material:\enumsentence{Hasan maSin az in mard xarid \\Hasan bought a car from this man}

Sadock's reanalysis would  propose thatthe N+V is simultaneously treated as a V for the purposesof morphology and a VP for the purposes of syntax.Inflexibility in word order:\begin{quote}...morpheme ordering is generally copmletely rigid andin any case is much stricter thanphrasal ordering.  Thus, where there is a conflictbetween the two, it will be the principles of morphemeordering that win out. (Sadock 1985 pg 407\end{quote}

However, they argue that the CP shows characteristics of V-Bar,not V-0.  They also criticize this type of account because it doesn'tsseem to distinguish between examples like \ex{-1} and \ex{0}.Also, ``it implies a division between sytnax and morphologystronger than we wish to defind atthis stage." pg 198 .

They suggest a restructuring analysis wherein:

V' --> N'' V  is reanlyzed to:  V' --> N V,  if N'' is non-branching.

``The compound retains V' status, although it functions semantically, etc. as a word, possibly by virtue of its lexical specification" (197)(They note the existence of PP-V compounds such as {\em bekar bordan},``to work bring" = ``to use.")

Samiian has worked on the EZAFE construction thatMohammad and Karimi cite. She argues that the restriction is htatEZAFE never occurs within NP preceding a prepositionalphrase which would be regarded as subcategorized.

They do give the example: {\em dadan-e pul be dustan}

\section{Karimi, Essex. Diss chapter 4}

Dear Karimi, Thanks for sending your chapter.  I read it with greatinterest.

  It's good practise to always put your name somewhere onevery paper you give out.  You might also put the titleof your diss, so that people like me will know how to cite it!

 There was some kind of problem with the pages, especiallytoward the end (where it got especially interesting!).Page 31 didn't seem to follow page 30, and 32 didn't seem to follow31.  Page 34 repeated page 31, and page 37 was a repeat ofpage 35.  Is it possible for you to send me a new .ps version?I could print that out here.

 Here are some initial comments:

 I noticed that people's judgments varied quite a lot on whethergapping and conjunction of the PVs was allowed.  Did you also findthat?

 You cite two types of PVs:  ``separable" and ``unseparable."But the ``unseparable" cases are also separable under certainconditions (as you note), and when the ``separable" ones areactually separated, they generally occur with determiners (-i or ra).In that case, as you note later, they are really acting likearguments.  But then it's not clear that they are actually partsof complex predicates in that use. 

 Your take on it seems confusing given your discussion ofqaza: xord vs farib xord, where you want to distinguishLVCs from VPs.

Of course it all depends on how you carve things up, and ifyou can account for more data, that's always better.  But itshould be made clear earlier in the text that these cases areacting differently.  As it is, the glosses do not alwaysnote that the "-i" suffix is an indefinite determiner.

For example you claim correctly that I do not accountfor example 43b: java:b-e xubi be ali da:d.  But it's impossiblefor the reader to know that this may not be something thatI (or others) have considered the same type of thing.  In particular,the "-i" suffix indicates that "java:b-e xub-i" is acting asa regular argument of da:dan.  Therefore, there is arguably nocomplex predicate in the sentence.

The same is true for example 62) Ali raqs-e xaili xub-i kard.

Also, I know you are using the term "unseparable"as a technical term (not to mean literally unseparable),but I found it misleading that you cite me as assumingthat host and light verb are unseparable, since I didn'tuse that terminology and I did try to explain exactly whycertain things *could* intervene (as I think you do mention at some point).

I do agree that if you can count for the complex predicates *and*the corresponding argument+verb cases, then your accountis preferable.

CPs in Iranian lngs is at least as old as middleIranian languages (Karimi 1987; Sheintuch 1973).

There are only 150 simple verbs in common use, but 2000 CPs.In order of frequency:\begin{quote}\item kardan, zadan, da:dan, gereftan, budan, da:stan, ?a:madan, ?a:vardan, xordan, kesidan, ya:ftan, Sodan, bordan, ?ofta:dan, raftan, ?anda:xtan (to fel, to throw), gozastan (to put), didan (to tolerate, to experience), rasa:ndan (cuase to reach), varzidan(to committ, to do), baxsidan (to forgive), rasidan (to reach).\begin{quote}

\section{Klaiman, M.H. The Prehistory of Noun Incorporationin Hindi. 1990. {\em Lingua 81} 327-350.}

Cites Southworth 1971: 129 that {\em kar} (do) is the usualV in newly coined "conjuncts"; these often have relatdintransitive versions with the V stem {\em ho-} (become/be).(Just like Persian).

About a dozen or two Vs are involved.  The following arecited:\eenumsentence{\item aa "come"\item nazar aa "be visible"\item de "give"\item dikhaaii de "show up, appear"\item rakh "put, keep"\item tas\'riif rakh "sit"\item le "take"\item s\~a\~a le "breathe"}[reminiscent of Bowerman and Clark stuff]

Argues that Hindi represents a preliminary stage prior to a potentialfuture development of NI, but is not a true NI language now.

At the same time, it has some properties that Mithun ascribesto later stages of NI.

NOT NI because "at least" the negative morpheme {\em nah\~i\~i}can intervene between N and V: it is not a clitic sinceit can be stressed.

THe V can agree with the N of the CP.

Hindi has cases that are like memory came ({\em yad Amad}) in persian.

\section{Matsumoto, Yo. 1992. On the Wordhood of Complex Predicatesin Japanese. Stanford University dissertation.}

{\em complex predicate}: a predicate that is in some sense one word,but is two words in some other sense.

{\em word} = atom

\begin{enumerate}\item constituent structure: cannot be separated; X0; butone-word status at c-structure does not necssarily entailmono-clausality.  Also a predicate can betwo morphological words in c-structure, but mono-clausal,as in the case of German and Dutch separable complex verbs.\item functional structure: a unit w.r.t. grammatical-functionalproperties: governs its subject and object, is a unit forpassivization.\item argument structure: has one logical subject, can only have oneagent, patient, etc.\item semantic structure: packages meaning in an integrated way:{\em send} is semantically bi-clausal, since there is an resultof something moving. (pg 8) \end{enumerate}

[This leads to 2$^4$ possibilities: are they all exhibited?]

{\bfNoun-incorporated periphrastic verbs}: {\em benkyoo suru}, Lit. `study' + `do'.  Evidence that there are two morphological words (twowords at c-structure):\begin{enumerate}\item focusing particles can intervene (pg 48)\item the first morpheme can be coordinated (pg 48)(and notes that compounds do not allow coordination)\item one morpheme can occur alone in various environments:(48)

\enumsentence{\shortex{3}{sore wa & rakka & shi-mashi-ta ka}{it TOP & fall & do-POL-PAST-Q}{Did it fall?}}

\enumsentence{\shortex{2}{Hai, & shi-mashi-ta }{Yes, & do-POL-PAST-Q}{Yes, it did.}}\end{enumerate}

{\bf T. Mohanan 1990}: Hindi N-V compounds: N+V is a single morphological word, butV agrees with N, which is taken as evidence that N is an OBJ.So, the sublexical unit, N, bears the grammatical relation OBJ.[Matsumoto here seems to assume thrat

Other sublexical functional specifications can include XCOMP as inJapanese morphological causatives (Ishakawa 1985):  V $-->$ V V,where the first V is annotated with (up XCOMP) = down and thesecond is the head (up = down).

The OBL$_{goal}$ phrase is captured by:  (up XCOMP OBL$_{goal}$) = down,which capitalizes on Functional Uncertainty:

{\bfFunctional Uncertainty:}

\smallskip  Kaplan and Zaenen (1988) proposed doing away with the constraint againstfunctional equations like: (up XCOMP OBL$_{goal}$) = down, which hadbeen ruled out by the Functional LOcality Condition, which prohibits more thanone attribute name from appearing in one equation.K\&Z propose introducing a functionally 'uncertain' phrase in a phrase structurerule as below:

\enumsentence{\shortex{4}{ S & $-->$ &  XP* &   V}{& & (up XCOMP*  GF) = down & up = down}{Functional Uncertainty for long distance dependencies}}(Kleene star means 0 or more XCOMPs)

\medskip

[The problem may not arise if you adopt a constructioanl approach.

I should think about that]

{\bf Butts}:Urder permissive constructoin: mono-clausal functional structure,but two morphological verbs: one PRED corresponds to two positions inc-structure  ("let-write") and 2) the two words comprising one PRED atf-structure have different status in arg structure

``It is sometimes assumed that thematic role info...is the onlysemantic info. needed for syntax...but this view is clearly false(Dowty 1991, Matsumoto 1990a, T. Mohanan 1990, Pinker 1989)" pg 6

{\bf Japanese Light Verb construction}

verbal noun$_{acc case}$  + {\em suru} ("do"), analyzed by Grimshaw and Mester 1988as requiring `argument transfer' (wherein args of the verbal nounare transfered to the light verb).

M argues that there are other verbs that allowthe arguemnts of the verbal N to occur outside the transparentNP, that {\em suru} isn'tthe only one:  there are aspectual verbs "begin", "repeat","finish" "continue" and verbs of thinking and wanting: "plan" "forget""decide".  These are raising and equi verbs, respectively. (pg 88)

He argues that {\em suru}, too should be analyzed as a raising or equi verb, wherein {\em suru}or other light verb takes the N as an XCOMP.  As in thecase of other XCOMPs in Japanese, the arguments of the XCOMPare treated as arguments of the main clause.

1) The N must be controlled by the subj: this is expectedfor XCOMP

[{\bf Differencs between Japanese construction and Persian}:

Japanese N is case marked and must be deverbal (has its own args).]

\medskip{\bf Morphological Causatives}

\section{Mohammad, Jan and Simin Karimi.  1992.Light Verbs are Taking Over: Complex Verbs in Persian.{\em Western Conference on Linguistics} Joel Nevis and VidaSamiian (eds). Dept of Linguistics.  Cal State Univerity,Fresno.}

Since 13th century, compound verbs have gradually replacedsimple verbs.

The number ofSimple verbs  is less than 115.  Comparing{\em  geristan} vs {\em gerye kardan} and a few others,the CP is the only one used in colloquial speech.

Cps are productive:  TAyp kardan (to type), telefon kardan (to call).

Claims the N is not not always an X0 on the basis of:

{\bf CHECK}: Can this occur with adverb before verb?

\enumsentence{KimiA ye zamin -e- saxti xord \\KimiA a earth -gen- hard eat \\(her gloss: KimiA a earth -EZ hard collide)\\``KimeA fell badly.''}

Certain things can intervene: future, IO, {\em dige}

Discusses Heny/Samiian's restructuring hypothesis.Reject because the N allows branching as in \ex{0} (so theysay).   Heny/Samiian account for cases wherein theN and V are separated by claiming that in thosecases we have a VP as opposed to a reanalyzed V',cannot account for transitive CPs, since the separated:N would have to be the DO; then what would the other DO be?

Larson's V' Reanalysis is discussed and rejected.

Verbs are `light verbs' (following Jesperson)

They claim that ``The verbal element of the Persiancomplex predicate is semantically empty." Instead theyclaim that the semantic content is based on the nominal element.This claim is supported by the existence of a fewcases wherein different verbs can be used withthe same meaning ({\bf CHECK}: is it really the same meaning?):\enumsentence{ezhAr kardan  vs ezhAr dAStan ``statemnt + V" to state}\enumsentence{majbur kardan vs majbur nemudan(to show) ``forced + V" to force}

[However, this conclusion is clearly unwarrented as a generalstatemetn.  While there are a few cases wherein the change

of V does not drastically alter the meaning of the CP,inthe majority of cases, a change in the V {\em does}result in a predictable change in meaning.  For example,....]

They claim that the light verb does not bear a thematic relationto its nominal element on the basis of the followingsyntactic test:  EZAFE construction, a construction that consistsof a head noun and its thematic arguments:

\enumsentence{\shortex{4}{dAdan-e- ketAb & be rAmin & dorost & nabud}{give-gen-book & to Ramin & right & neg.was}{The giving of the book to Ramin was not right}}

They note that the CP changes its subcat frame.

[{\bf CHECK}:\enumsentence{\shortex{4}{dAdan-e- da?vat & be rAdio & dorost & nabud}{give-gen-invitation & to Ramin & right & neg.was}{The giving of the invitation to the radio was not right}}

Also check if Do in this construction always has a definiteinterpretation: that alone would rule out CPs]

\subsection{Grimshaw and Mesler's (1988) argument transfer}

G\& M discuss Japanese (see Matsumoto for counterarguments).

N is nonreferential and cannot be relativized (the secondfollows from the first, no?)

{\bf CHECK}: Ramin rA Kimea be-heS ketAb dAd

Ramin-ra Kimea to him book gave

Ramin, Kimea gave him books.

They want to claim that there are two structural positionsfor objects in Persian one receiving inherent and onestructural case.  The N and all nonspecific objects are sister to the V (receiving structural case) but specific objectsare in SPEC of VP and receive inherent Case.

So only the specific NP bears the thematic role.

[I don't know what they do when a transitive CP has a non-specificobject: then there are TWo objects receiving structural case?What about binary branching??]

{\bf Conclusion}: ``We have shown ...that the verbal element inPersian copmlex prediates is semantically empty, and thatit coocurs with a transparent NP.  The lattertransfers its thematic roles to the verb.  Thisfact accounts for the differencesbetween the light and the corresponding heavy verb, on theone hand, and the differences observed between complexpredicates that appear with the same verb, on the other hand.We have also shown that the nominal element of the light verb in a complexpredicate construction is a nonspecific NP, thatdoes not bear thematic relationship to the verb.  This fact excludesthe Incorporation hypthesis since incorporation involves an NP that isassigneda thematic role by the verb...." 208-209.

\section{Mohanan, Tara. ms. Multidimensionality of Representation:Noun-Verb Complex Predicates in Hindi. National University of Singapore}

Complex predicate construction is one in which two semanticallypredicative elements jointly determine the structure of a singlesyntactic clause.

TM concentrates on N+LightVerb and complexes (similarto those in Persian).  The noun is the HOST, the verb, a LIGHTVERB (Jespersen 1954).

She argues that the N is a lexical X0 category, and the V is a V1 category(can have auxiliaries as sisters). The N + V1 are taken to form a phrasal unit.

She notes that Lexical Integrity: subconstituents of a lexical unit cannot undergosyntactic reordering, entails that the host and light verb cannot consitute a lexical unit.(Note that while the N+LV cannot be separated by scrambling, the LV can be topicalized).

She notes that the argument structure of the clause is determined conjointly bythe N and the LV. This is then in conflict with theDirect Syntactic Encoding (Bresnan 1978, 1982): function changingphenomena can take place only in the lexicon.

The complex predicate as a whole and the light verb are both phrasal categories."No lexicalist theory would therefore wish to treat the N+V CP in Hindi asbeing formed in the lexicon."

She is with Alsina it seems in treating these complexpreds as being formedin the syntax.

[She and Farrell \& Phil both note a conflict  between Lexical Integrity andDirect Syntactic Encoding; she rejects DSE, F\&P reject Lexical Integrity inits strong form.]

\section{Nash, David George. 1980. Topics in Walpiri Grammar.MIT dissertation.}

The auxiliary clitic (rli: pronominal clitic: we dual inclusive subject) appears in second position afterthe first immediate consituent of hte sentence.  It can appear after the N+V or after the N:\enumsentence{Wuruly(pa)-ya-ni-rli\\seclusion-go-NPast-12 \\Let's go and hide}

\enumsentence{Wuruly-rli ya-ni\\seclusion-12 go-NPast\\Let's go and hide}

Doesn't seem very auxiliary-like, does it:  Also{\em ka} (Present auxiliary) can intrude (1980:53).

\section{Sells, Peter. 1994.  Sub-Phrasal Syntax in Korean.{\em Language Research}}\section{Williams, Edwin. 1993ms. Noteson Lexical and Syntactic Complex Predicates:prepared for Stanford workshop presentationMay 1993}

Argues for two types of complex preds:lexical and phrasal; and that they shouldnot be treated uniformly.

Lexical: can occur: V Pred NP even withoutheavy NPs:\enumsentence{John wiped clean the table.}

That it is lexical is supported by theexistance of nominalizations:\enumsentence{the wiping clean of the table}Pseudopassives:\enumsentence{*John was taken great advantageof.\\John was taken advantage of}Some cases of intrinsic V+Particle (cf below)

{\bf Phrasal}: cannot occur V Pred NP:\eenumsentence{\item *John considers clean the table.\item John considers clean any table witha  reeflectant surface.}Note that this order is out wheneverthe complex predicate is itself complex(Emonds 1969):\eenumsentence{\item *I picked right up the paper\item *I wiped very clean the table.}EW notes that if we take seriously theidea that complex predicate formation isa lexical rule, then it should be incapable ofgenerating "wipe very clean" since thisis not a possible lexical unit.EW concludes that \ex{0} do not involvecomplex predicates (which he takesto be lexical).Further evidence comes from the fact thatconsider+A does not allow nomiinalization:\enumsentence{*the considering silly of Bill}All cases of V+particle(external):

EW suggests that [consider NP AP] structuesare clausal in the sense that there is a subject-predicate structure (I guess, semanticallyclausal), but he rejects the idea that thereis a small clause constituent.Rather, the entire VP is the "clause"The head is the "glue" that binds the subejctand prediate together:\enumsentence{ [consider NP AP]vp \\ glue  subj pred}He observes that the AP cannot be "specificatoinal", but must be "predicational" (cf. Borkin):\enumsentence{*I consider the mayor John}

Two types of verb+particle:\begin{enumerate}

\item intrinsic properties of object \item extrinsic properties of object toother objectsmany verb+part are ambiguous:\eenumsentence{\item John put the planestogether.\\ assemble (intrinsic)\\ side-by-side (extrinsic)\item John kicked the vase over.\\ on its side (intrinsic)\\ over to me (extrinsic)}\end{enumerate}Without heavy NP shift, the V-Part-NP orderis not grammatical with the extrinsic readings.Williams goes on to suggest that althoughEnglish does not have complex predicatesof the form: V --> [V XP]vx, where vx issome kind of argument taking non-maximalunit, such complex predicates do existin French: e.g. {\em faire partir}.

Phrasal result phrases cannot occur before the NP:\enumsentence{*I picked right up the paper.}A lexical rule such as V $\Longrightarrow$ V A could not generate\ex{0} anyway.W claims that:\enumsentence{I wiped the table very clean.}does not involve a complex predicate.

Intrinsic vs extrinsic interpretation of particles:\enumsentence{John put the planes together\\a. assemble (intrinsic)\\b. side-by-side (extrinsic)}The order V prt NP can only have intrinsic interpretation:\enumsentence{I put together the planes.}

Two constructions: one lexical:

a. V $\Longrightarrow$ V P (intrinsic sense)

 (also V $\Longrightarrow$ V A)

The other not:

b. VP $\Longrightarrow$ V NP P(P) (for both the intrinsic and extrinsic cases)

the wiping clean of the table (lexical)

*the considering silly of Bill (non-lexical)

"Small clause constructions" are clausal in the sense that thereis a subject-predicate structure:

\enumsentence{I [consider NP AP$]_{VP}$}

[consider NP AP]

   glue  subj pred

Williams, Edwin. 1983. Against Small Clauses. {\em LI} Vol 14 No 2.pg. 287-308.

Predication theory:

  {\em Subject as External Argument}:  The subject of a predicativephrase XP is the single argument of X that is located outsideof the maximal projection of X. (Williams 1979, 1980, 1981, 1982)

The subject-predicate relation is captured by coindexing.Also, subjects must c-command their predicates.

[NP0 XP0]0, whereXP is a maximal projection of X (so M cannot be a projection of X).

The Predication theory makes the subject-predicate relationprimary, the notion of "clause" derived.  The Small Clausetheory does the opposite:

The small clause theory says, "all subjects must bestructural subjects."

Arguments for Predication theory:

 1)scope

 2) the statement of "thematic independence":  thematicallyindependent predicates cannot occur in argument positions,and they can occur in non-argument positions.

 3) The nature of the small clause (X*) category:  the X* mustbe nonmaximal, since goverment (and casemarking) can applyacross it. But the head of this nonmaximal projection X* turnsout to be maximal:

  John considers [Bill [Bob's friend]  What does John consider [Bill [t]

(Wh Movement moves only maximal projections)That in itself is wierd.  We have:

 (a) X* --> ...XP...

Moreover, both {\em Bill} and {\em Bob}are the subject of the same nominal projection.  Moreover,since X* is nonmaximal, projections such as:

 (b) XP -> ...X*...

are allowed, which would give a single projection of infinitelength, since (a) and (b) are recursive.

This particular argument could be rectified if {\em only}S or S' small clauses were allowed, though.

 4) accounting for the differences between infinitives and predicates of "small clauses".  the infinitive only hasthe property of thematic independence in argument positions,and the property of taking a subject of arbitrary reference.Predication Theory:  only infinitives have a structural PROsubject.  (Williams notes that LFG also does not distinguishbetween infinitives and other predicates, assigning neitherof them a structural subject--this will also failto account for the distinction).

{\em Argument Complex}:

  An arugment complex consists of a predicate, its arguments,its arguments' arguments, and so forth.

{\em The Restricted Theta Criterion} (restricted to argumentcomplexes):

 In an argument complex, each phrase is assigned only one theta role.

BUT:  In {\em John ate the meat raw}, {\em the meat} isassigned two theta roles. "This does not violate the ThetaCriterion, however, because {\em raw} is not an argument of{\em eat}.  Hence there is no argument complex htat includesboth {\em eat} and {\em raw}."[What about resultatives, then?  In particular, what about:{\em render, make} which subcategorize for a resultative?If you don't have small clauses, how can you do these?]

Chomsky (1981) says that the Projection Principle rulesout the possibility of analysing {\em consider} as havingan NP and AP copmlements, since semantically (and at LF structure)it has a single clausal complement.  Williams responds wisely:why do we take the LF structure of a {\em consider}sentence to have a clause? because it is a semantic unit, butthat shouldn't mean it has to be a syntactic unit.  We canlet the NP AP structure be the representation in LF.

  The Predication Theory does run counter to Chomsky'sprinciple (1981) that if a verb subcategorizes for a position itmust theta-mark it, since the postverbal NP of {\em consider}does not get a theta-role from the verb.

\section{Rosen, Sara Thomas. 1989. Two Types of NounIncorporation: A Lexical Analysis. {\em Language 65 2}.}With Mithun and DiSciullo \& WIlliams, she argues that NIis a lexical process (against Baker and Sadock).

Two types of NI:

\begin{enumerate}\item Compound NI: involve simple compounding: one argument of the simple verb is satisfiedwithin the verb.   (Mithun's first 3 classes) (e.g. Engl "meat-eater")\begin{enumerate}\itemthe arg structure of the simple verb ischanged to become intransitive  \item no stranding of modifiers\item no doubling outside verb\end{enumerate}\itemClassifier NI:

can still take an object: the incorporated N is semanticallylike a classifier of the object.  \begin{enumerate}\itemSimple and complex verb's arg structure is the same: can stilltake an object\itemLanguages with this type ofNI are said to allow 'stranding' of modifiers (null-head modifiers).\item doubling outside of verb\end{enumerate}\end{enumerate}Compounding shoudl be invisible to the syntax.

[This typology doesn't seem to allow for transitive CPs which are not classifier type: Mithun's type II]

R notes that there is a subject/non-subject asymetry: withdeep subjects not incorporating: instrumentals incorporatein Nahuatl, means phrases incorporate in Niuean, and some locativesincorporate in Samoan--non of these phrases is ever a DO.

Goals and benefactives never incorporate in any language.

\section{Mohanan, Tara. 1994. Wordhood and Lexicality: NounIncorporation in Hindi. Submitted to NLLT}

NI: noun combining with a verb to form a complex verb:Mithun 1984, Hopper and Thompson 1984, Sadock (1980, 1991),Baker(1988), DiSciullo and WIlliams(1987) and Rosen (1989)are cited as general refs.

Agreement facts argue that the N must be treated syntacticallyas an argument on par with the other args of the verb(I think this is Baker's take, too).  At the same time, though,NI is lexical in various ways.

Most NI sentences are actually ambiguous between an NI reading,and a simple verb reading.  (e.g., "Anil will sell books," OR"Anil will do book-selling.")

\begin{enumerate}\item nothing may intervene between N and V in NI: notsubject, locative adjunct, or negative. (word order isusually free in Hindi)

[certain things can intervene in Persian]

\item N cannot be modified, and modifier cannotbe stranded either: this is not predicted by Baker's accountof head-movement, or by Sadock's coanalysis account.[same is true in Persian]\item neither wh-words nor pronouns can be the N in NI:``*what-selling" ``*this-selling": M attributes this factto the fact that wh-words and pronouns are maximalprojections, and N of NI is necessarily zero-level.[would also seem to be ruled out by the naming pragmaticconstraint, though]\item neither the N or the V can be gapped in a coordinationconstruction: ``*Anil does horse buying and Ram does -selling."``*Anil does hourse-and Ram elephant-selling."This is claimed to follow from the Lexical IntegrityHypothesis: the categorial representation formed inthelexical modeule is not visible to principles of phrasalorganization. [M points out that adopting this principleeffectively rules out a head-movement analysis likeBaker's or a coanalysis like Sadock's as well].\item neither N or V can be conjoined in NI: this is attributedto the fact that lexical categories cannot beconjoined in Hindi.\item N doesn't bare an overt case clitic: but M claims N isin nominative case: is nominative case is unmarked? how dowe know it's there at all? Explanation for this is that clistics apply to phrasalcategories, not zero level categories.\item evidence also comes from phonology: stress contours\item NI can be input to lexical derivational N formation:viaderivational suffix {\em -vaalaa}\end{enumerate}Semantic/Pragmatic constraints on NI:\begin{enumerate}\item N must be generic or ``non-referential": it can onlyrefer to the class of entities denoted by the noun,not to the idividualmembers of the class.  (M cites references that point out thatNi is not necessarily accompanied by genericity or nonspecificity(cf.Mohawk(mithun 1984), Eskimo(Sadock 1980).\item the activity or process refered to by NI must be``salient" or ``name-worthy"in what Hale and Keyser (1991:13)call the ``cultural encyclopedia" of language users.``\#clothes-tearing, \#grass-seeing, \#book-giving, \#book-lifting"[In this section she notes \#kissed bride, vs unkissed bride

in Eng]This also follows from idea that nameworthiness is a propertyof lexical items (Hale \& Keyser 1991: 13)\end{enumerate}\bigskipEVIDENCE FOR PHRASAL STATUS:\begin{enumerate}\item Agreement facts:

1) Verb agrees with the object iff the object is NOM, andthe subject is not NOM.

2) NOM case can only be determined after phrasal concatenation:because phrasally concatenated model can assign DAT case to the subject

3) V can agree with incorporated in NI\end{enumerate}

SOLN to the paradox: separate two dimensions ofsyntactic info: grammatical functions (NI involves PREDand OBJ) and grammatical categories (V0 -> N0 V0).

\section{Mithun, Marianne. 1984. The Evolution of NounIncorporation. {\em Language 60: 4}. pg 847-894.}

{\bf Type I Noun Incorporation (NI)}: Micronesian, Mayan (patients), Lahu (patients, instruments and locatives)Lexical compounding: Compound must designate some institutionalizedactivity.  Noun is not specific, does not refer, is notmarked for definiteness or number, cannot occur with demonstrativesor numerals.Sometimes the formal unity of V N is not obvious,since in most languages be written as separate words, and retain theirindependent stress patterns, but cliticization, aspect suffixes,and the placement of manner adverbs reveal that the V N istreated as a syntactic unit. 

[In Persian, stress is different. ]In Samoan, particles whichnormally cliticize to the right of the Vcliticize to the right of the VN complex under NI (Chung 1978).

In Micronesian languages, suffixes whcih normally follow the

V, follow the VN complex under NI (Mithun 1984):\enumsentence{\shortex{3}{I & kanga-la & wini-o}{I & eat-COMP & medicine-that}{``I took all that medicine."}}\enumsentence{\shortex{2}{I & keng-winih-la}{ & }{``I completed my medicine-taking."}}

V N forms an intransitive verb.``All languages which exhibit such morphological structuresalso have syntactic paraphrases." (pg 848).

Name-worthiness (Zimmer 1971, 1972, Downing 1977, Pawley 1982)

``The term 'incorporation' is generally used to referto a particular type of compounding inwhich a V and N combine to form a new V." (848).

\enumsentence{\shortex{2}{Ngoah & ko oaring}{I & grind coconut}{``I am coconut-grinding." Mokilese (Micronesian, Austronesian;Harrison 1976)}}

Matisoff 1981: Lahu NI often become idiomatic.

Mithun notes that in some languages, the compounds areconsidered single words by speakesr andare subject to all regular word-internal phonological processes(Niska, a Tsimshian language of British Columbia;a number of Australian Aboringial languages).

Gurindji, Australian aboriginal language Hudson 1978:\enumsentence{pina-karri \\ear-stand \\``to listen, hear"}

{\bf Type II NI}: A normally oblique argument appears inthe case position (obj or subj) vacated by the IN.IN's lose their syntactic status as arguments of the clause;they are unmarked for definiteness, number or case, althoughthey are not necessarily indefinite.\eenumsentence{\item his-face I-it-wash (unincorporated)\item I-him-face-wash "I face-washed him" (incorporated)}

(type of possessor raising)\eenumsentence{\item I chop the tree in my cornfield.\item I chop-tree my cornfield.}

{\bf Type III NI}: The manipulation of discourse structure.NI of known, incidental, non-salient nouns. IN is not necessarilynon-specific and indefinite, but is unmarked for those features.\eenumsentence{\item $<$ Where is the knife? I want it now $>$\item He knife-cut bread. "He cut the bread with it."}

{\bf Type III NI}: Classificatory Noun Incorporation.  IN servesto narrow down the possible designation of the V, while a morespecific overt nominal appears as well.\enumsentence{ they-tree-saw cashew.nut \\"They saw a cashew tree"}The accompanying independent NP need not contain a morphologicalN, but a determiner, number or adjective may stand alone.This might lead one to a syntactic account, but in fact theonly languages that do this, always allow a determiner, numbermarker or adjective to stand in the place of a full NP, whetheror not there is NI.

Hopper and Thompson 1980: general tendency for V's to coalesce with indefinite DaDOs. They cite Hungarian in this respect (DOs that are bothreferential and definite follow the V, indefinite, non-referentialObjects precede the V; if hte object is referential but indefinite,no marker appears in the V (and it follows)).

Turkish is cited (seems like Persian): indefinite objects do notappear with an article and adverbs may not intervene betweenN and V.\end{document}