�eGrammar of AutonomyIn Irish

James McCloskey

December

In Natural Language and Linguistic �eory : –. A shorter and slightly revised version willappear inHypothesis A/Hypothesis B: Linguistic Explorations in Honor of David M. Perlmutter, DonnaB. Gerdts, John Moore, and Maria Polinsky, eds.

Research for this paper was completed while I was a visitor in the School of Celtic Studies of theDublin Institute for Advanced Studies, and I am very grateful indeed for the ideal working conditionsthat I enjoyed there. I am also grateful to Sandy Chung for many different kinds of help. �e paperhas benefitted a great deal from presentations and discussion at the University of Toronto (September), at University College Dublin (November ), at (December ) and at (April).

Introduction

T I a form of the finite verb known as the briathar saor or ‘free(form of the) verb’. �e English term usually used for this inflectional class is

the ‘autonomous’ form, and that is the term I will use here. Autonomous forms arederived by adding a distinctive suffix, one for each tense, to the verbal stem. Fromthe verb cuir (‘put, send, bury’), for example, five autonomous forms can be built.

cuir-tear Present Tensecuir-eadh Past Tensecuir-fear Future Tensechuir-fí Conditional Moodchuir-tí Past Habitual

In origin, these forms are passives, and their functional range in the contem-porary language continues to be close to that of the agentless passive in English. Itis thus that they are most naturally translated:

() a. Tógadhraise [PAST-AUT]

suasup

anthe

corpánbody

aron

bharrtop

nathe

haillecliff [GEN]

‘�e body was li ed to the top of the cliff ’b. scaoileadh

release [PAST-AUT]

amachout

nathe

líontanets

‘�e nets were let out’c. Cuirtear

put [PRES-AUT]

iin

mboscaíboxes

iadthem

‘�ey are put in boxes.’

A broad array of intransitive verbs also accept autonomous inflection, as shown in(), the resultant forms o en difficult to render naturally in English.

() a. H-éirigheadhbecome [PAST-AUT]

cleachtuightheaccustomed

lewith

daoinepeople

a bheithbe [−FIN]

ag teachtcome [PROG]

‘One became accustomed to people coming.’ b. Do

[PAST]

chreidtíbelieve [PAST-HABIT-AUT]

insnain-the

seanscéaltaold-stories

sinDEMON

go léirall

fadlong

ó shinago‘People used to believe in all those old stories long ago.’

Even in the oldest recorded forms of the language, however (ClassicalOld Irish of the th century),the ancestor of the autonomous form shows a very strange mix of passive and active characteristics.See �urneysen (, p. , §)

c. hItheadh,eat [PAST-AUT]

hóladh,drink [PAST-AUT]

ceoladhsing [PAST-AUT]

agusand

ansinthen

chuathasgo [PAST-AUT]

a sheanchasstorytelling [−FIN]

‘�erewas eating, drinking, singing, and then the storytelling began.’ d. Tostadh

be-silent [PAST-AUT]

seala-while

leiswith

anthe

iongantassurprise

‘People went silent for a time in surprise.’

Preliminaries

Despite its origin, and despite the fact that it fulfillsmany of the samediscourse func-tions as short passives in English, the autonomous construction is not a passive—ornot at least if by a passive form we mean one in which the underlying object ofa transitive verb is rendered as a surface subject. �e internal argument of an au-tonomous form derived from a transitive verb is indistinguishable in its behaviorfrom any other direct object (see McCloskey, ; Stenson, , ). �ere areat least three kinds of arguments which establish this conclusion.�e first is that theinternal argument appears in accusative rather than nominative case:

() a. Cuirfearbury [FUT-AUT]

éhim [ACC]

sain-the

reilggraveyard

áitiúil.local

‘He will be buried in the local graveyard.’b. *Cuirfear

bury [FUT-AUT]

séhe [NOM]

sain-the

reilggraveyard

áitiúil.local

‘He will be buried in the local graveyard.’

Secondly, if the internal argument is a light pronominal, it may be postposed—an option permitted freely to direct objects but absolutely forbidden to subjects(Stenson (, –), Chung andMcCloskey (), Ó Siadhail (, –),Duffield (, -), Adger (), McCloskey ()):

() a. Cuirfearbury [FUT-AUT]

sain-the

reilggraveyard

áitiúillocal

amárachtomorrow

é.him [ACC]

‘He will be buried in the local graveyard tomorrow.’b. *Cuirfidh

bury [FUT]

éhim

sain-the

reilggraveyard

áitiúillocal

siad.they

‘�ey will bury him in the local graveyard.’

Finally, the internal argument may be a resumptive pronoun—again an option per-mitted to direct objects but forbidden to subjects (see McCloskey, , and refer-

ences cited there).

() a. fearman

gurC-[PAST]

bualadhstrike [PAST-AUT]

lewith

camánhurley-stick

éhim

‘a man that was struck with a hurley-stick’ b. *fear

mangurC-[PAST]

bhuailstruck

séhe

lewith

camánhurley-stick

méme

‘a man that (he) struck me with a hurley-stick’

What these observations in sum indicate is that the autonomous inflection is notassociated with promotion of a direct object to subject status.

�e puzzle of understanding the autonomous form, then, is the puzzle of un-derstanding what becomes of the subject argument of the verb to which the inflec-tion applies. Or to be more precise: what becomes of the most prominent of theverb’s arguments? �at in turn becomes a puzzle at the syntax-morphology inter-face. What is it about this set of inflectional endings which licenses silence wherethe most prominent of the verb’s arguments ought to be?

To facilitate discussion, I will use the term ‘autonomous argument’ in what fol-lows for the argument corresponding to this silence—the external argument of atransitive verb, the internal argument of an unaccusative (as in (a)), the experi-encer argument of a psych-predicate and so on.

A Structural Subject?

Given the data presented so far, it is tempting to understand the autonomous formin terms of event quantification (in the sense, say, of Parsons () andmuch otherwork growing out of Davidson ()) and the complete absence of themost promi-nent argument. �at is, one could conclude that what the autonomous inflectiondoes is to license total elimination of the external (or most prominent) argument.On that conception, one might understand () in terms of a semantics like that in():

Leaving open the possibility that the missing subject in such cases is, in some sense, like the im-plicit agent of a short passive. We will see below that there is something deeply correct about thisidentification, in that we will identify the silent subject of an autonomous verb with arbitrary sub-jects such as French on or Germanman, whose properties, in turn, are very close to those of implicitagents in passive constructions (see Koenig and Mauner () especially). As we will also see below,however, the silent subject of an autonomous verb is like an arbitrary subject pronoun, but unlike animplicit agent, in being syntactically active (in binding anaphors, for example). Overt agents with au-tonomous forms were common in earlier stages of the language but not in modern varieties, as shownin a very careful recent study by Ó Sé ().

() Buaileadhstrike [PAST-AUT]

lewith

clochstone

é.him

‘He was hit with a stone.’

() ∃e [strike (e) ∧�eme (e, pro) ∧ Instr (e, stone) ∧ Past (e)]

On this understanding, () claims that a striking event took place (in the past) whose�eme was the referent of the pronoun, and whose Instrument was a stone. No in-formation whatever is given about the striker, because the external argument of theverb (which would have provided such information) is simply absent.

Tempting as this approach may seem, Nancy Stenson (, –, drawingin part on earlier unpublished work by Dónal Ó Baoill) has developed a suite ofdecisive arguments which establish that it is incorrect. More specifically, Stenson’sarguments establish that verbs to which the autonomous inflection has been addedhave the same number of arguments as their counterparts without the inflection,and that (in the case of transitive verbs in the autonomous form) the presence of anexternal argument is detectable in standard ways.�e presence of such an argumentis demonstrated, for instance, by its ability to act as a controller (as in (a)), and byits ability to support subject-oriented adverbials (as in (b)):

() a. Socraíodhsettle [PAST-AUT]

aron

ionsaíattack

a dhéanamhmake [−FIN]

orthu.on-them

‘It was agreed to mount an attack on them.’b. Glacadh

take [PAST-AUT]

go fonnmhareagerly

leiswith

anthe

ainmniúchán.nomination

‘�e nomination was eagerly accepted.’

Stenson also documents a systematic set of contrasts between the behavior of tran-sitive verbs in the autonomous form and the behavior of unaccusative verbs (forwhich a semantics along the lines of () might well be appropriate). �ese argu-ments jointly establish very clearly that transitive verbs to which the autonomousinflection has been attached differ from unaccusative verbs exactly in their abilityto license an external argument.

Such observations establish that an autonomous form licenses an autonomousargument—corresponding to themost prominent argument of the base verb towhichthe suffix attaches. It is a different question whether or not they also establish thatthat argument is syntactically realized. For it remains unclear whether phenomenasuch as those in () test for the presence of a structural subject as opposed to thepresence of an argument—an argument which might remain implicit or syntacti-cally unrealized. An English example like (), for instance, will, on many accounts,involve a controller (the implicit agent) which has no syntactic realization.

() It was decided to go public.

A way to further probe the issue would be to ask if the silent subject of an au-tonomous form can bind reflexive or reciprocal pronouns. For it is widely acceptedthat such elements require syntactically realized antecedents:

() a. �ey arranged for each other to be on the committee.b. *�ey talked about it for days. It was finally arranged for each other to be

on the committee.

Initial investigation of this territory might suggest that the autonomous argumenthas no syntactic realization. As Stenson observes (, ), reflexive pronounsmay not have the autonomous argument as their antecedent:

() *Gortaíodhhurt [PAST-AUT]

éhim

féin[REFL]

‘People hurt themselves.’

�ere is, however, an independent reason why examples such as () might be im-possible. Reflexive pronouns are formed in Irish by adding the suffix féin to a per-sonal pronoun. In (), for example, féin is added to the third person singular mas-culine pronoun é, to make the corresponding reflexive pronoun. When these com-posite pronouns enter into binding relations, the base pronoun must agree in per-son, number (and for third person singular pronouns, also gender) with the binder.If there is a null argument in (), that element might well lack the necessary personand number features which would allow it to bind the reflexive in (). It is hard tobe sure, then, whether () is impossible because it contains no syntactic antecedentat all for the reflexive pronoun, or because the only available syntactic antecedent(the autonomous argument) lacks a crucial property.

No such issues arise for the reciprocal pronoun, which has a single invariantform (a chéile) no matter what form its binder takes:

() a. Chonaicsaw

muidwe

a chéile.each-other

‘We saw each other.’b. Chonaic

sawsibhyou [PL]

a chéile.each-other

‘You saw each other. ’c. Chonaic

sawsiadthey

a chéile.each-other

‘�ey saw each other.’

Among the kinds of arguments thatmay function as an antecedent to the reciprocal,if the conditions are right, is the autonomous argument, as shown in ():

() a. chuirtíput [PAST-HABIT-AUT]

geallbet

len-with

a chéileeach-other

‘People used to place bets with each other.’ b. Tógadh

raise [PAST-AUT]

suasup

anthe

corpánbody

aron

bharrtop

nathe

haillecliff [GEN]

ansanthen

lewith

cabhairhelp

a chéileeach-other

‘�e body was raised to the top of the cliff then with each other’s help’

c. Tátharbe [PRES-AUT]

a’ strócadhtear [PROG]

a chéile.each-other

‘People are tearing each other apart.’ d. �éití

go [PAST-HABIT-AUT]

ag itheeat [PROG]

béilemeal

lewith

chéileeach other

‘People used to go for a meal with each other.’

�e hedge above (‘if the conditions are right’) is necessary because not all examplesof this type are acceptable. Nor are the attested examples always judgedwell-formed,out of context, by consultants. Wewill have more to say about this variability whenwe have gone farther in investigating semantic properties of the autonomous argu-ment. �e idea, in brief, will be that the referential properties of this element willnot always support the kinds of part-whole relations which the logic of reciprocitydemands (see McCloskey () and references cited there for some discussion ofthese requirements as they hold in Irish). For present purposes, though, the impor-tant observation is that it is not difficult to find examples like () in ordinary usage,a fact which in turn indicates that they are not excluded in principle. From this, inturn, it seems reasonable to conclude that the reciprocals in () have syntacticallyrealized antecedents and also to conclude in turn that the autonomous argumentcorresponds to a structural subject of some kind.

We are brought, then, to the interim conclusion (following in the foot-steps ofStenson () for Irish and of Anderson () for the corresponding construc-tion in Breton) that the autonomous inflection involves no re-arrangement of argu-ment structure and no associated syntactic movement. Rather, that inflection, whenattached to a finite verb, licenses the appearance of a silent argument with very par-

Note the ungrammaticality of the English rendering of (b), reinforcing again the contrast be-tween the autonomous form and agentless passives.

Stenson (, p. ) cites one example of this type as ungrammatical.

ticular semantic properties—close to those of elements usually called ‘arbitrary’ or‘impersonal’. Viewed in this light, the Irish autonomous inflection is close kin to theimpersonal or arbitrary pronoun constructions commonly found in European lan-guages (Italian si, French on, Germanman, Swedishman, Icelandicmaður , Yiddishme(n) and so on), as in the Italian andGerman examples of () (fromD’Alessandroand Alexiadou () and Malamud () respectively):

() a. Inin

quelthat

ristoranterestaurant

siArb

mangiavaeat [PAST-HABIT]

benewell

‘People used to eat well in that restaurant.’b. Man

Arbwäschtwash [PRES]

diethe

Händehands

vorbefore

demthe

Essenmeal

‘One washes one’s hands before meals.’

If this identification (recognized already by David Greene , ) is correct, theextensive and rich literature on such impersonal constructions (see, among manyothers, Jaeggli (), Cinque (), Rizzi (), Authier (), Condoravdi(), Chierchia (), Kratzer (), Koenig (), Koenig andMauner (),D’Alessandro and Alexiadou (), Egerland (), D’Alessandro (), Mala-mud ()) can be brought to bear on our understanding of the Irish construction,and reciprocally the Irish construction can make its own contribution to the studyof arbitrary subject constructions more generally.

Parallels

A standard description of the arbitrary subjects of French, Italian or German is thatthey are used ‘when the intention of the speaker is to remain vague about the ex-act identity of the subject’ (D’Alessandro () cited in Malamud ()). �isdescription corresponds closely to descriptions of the autonomous form found inIrish grammars such as that of the Christian Brothers (, §, p. ):

Úsáidtear iad nuair nach mian nó gach gá nó nach féidir an gníomhaía lua.

[�ey are used when it is not desirable, not necessary, or not possibleto specify the agent.]

() can now be seen as parallel to the German () (Kratzer ()):

() ManArb

redetespeak [PAST]

mitwith

einandereach other

‘People talked to each other.’

or the Italian () (Cinque ()):

() SiArb

erabe [PAST]

parlatospoken

l’unothe-one

conwith

l’altrothe-other

‘People talked to each other.’

And German () (Kratzer ()) is possible (in contrast with ()) because thereflexive pronoun in German is invariant sich—like the Irish reciprocal, but unlikethe Irish reflexive:

() ManArb

erkundigteinquire [PAST]

sich[REFL]

nacha er

mir.me

‘�ey/one inquired about me. ’

If we push beyond this very general characterization, the range of interpretationsopen to the autonomous argument (and some of the restrictions on its use) can beseen to parallel closely the range of interpretations that have been documented forarbitrary subjects.

In the context of habitual aspects, the interpretation can be quasi-universal orgnomic, as in ():

() a. éinneanyone

goC

bhfeicfísee [COND-AUT]

breoiteachtsickness

farraigesea

ag teachtcome [PROG]

air,on-him

déarfaísay [COND-AUT]

leiswith-him

…

‘anyone who you would see getting sea-sick, you would say to them… ’

b. tugtargive [PRES-AUT]

‘madadhdog

uisce’water

(go minic)(o en)

aron

anthe

dobharchúotter

‘�e otter is o en called a water-dog.’c. Gaeilge

IrishaC

labhartarspeak [PRES-AUT]

anseo.here

‘It’s Irish that people speak here.’

In the context of an episodic tense or aspect, however, the quantificational force isusually closer to that of an existential:

() a. léiríodhproduce [PAST-AUT]

drámaíplays

leisby-him

sanin-the

Abbey

‘Plays of his were produced at the Abbey.’ b. Labhradh

speak [PAST-AUT]

go hiongantach,wonderfully

go buadhach,victoriously

go feargachangrily

‘People spoke wonderfully, victoriously, angrily’ c. Tógadh

raise [PAST-AUT]

scoilschool

úrnew

bliainyear

ina dhiaidh sina er that

‘A new school was built a year later.’

Finally, the autonomous argument frequently has a pseudo-specific use. By this, Imean that it can occur in a narrative in the course of which the reference of the au-tonomous argument has been clearly and unambiguously established before the au-tonomous form itself is used. In (a), for example, the reference of the autonomousargument is established in the preceding temporal clause. (b), which is typical of awide range of such uses, comes towards the end of a fairly long narrative in which itis established in an unambiguous, exhaustive, and specific way who the participantsare in the events being described.

() a. Nuairwhen

aC

bhímiswe-were

ag dulgo [PROG]

thairis siúdby-this-guy

arísagain

chaitíthrow [PAST-HABIT-AUT]

clochastones

leatceannroof

anthe

tíhouse

‘When we’d be going by this guy again, stones would be thrown at theroof of the house’

b. Bhíwas

séit

an-deireanachvery-late

faoinby-the

amtime

arC

fágadhleave [PAST-AUT]

anthe

Castle agusand

aC

ndeachthasgo [PAST-AUT]

abhailehome

‘Itwas very late by the timepeople le theCastle andwent home.’

Anyone familiar with recent work on arbitrary pronoun subjects will recognize thatthe kind of variability just documented is typical of arbitrary subject constructionsgenerally. It is exactly this property which has been at the heart of discussions ofwhat the semantics of such pronouns might be. Kria et al. (, ) summarizethat body of work in the following terms:

Recent research …points towards the view that arbitrary interpretationsare essentially like a general indefinite referring to persons; if the sentenceshave a generic flavor, then this is due to additional generic operators inthem.

Chierchia () in particular (see also Authier (), Koenig (), Koenig andMauner (),Malamud ()) argues extensively that arbitrary pronouns shouldbe understood as indefinites in the sense of dynamic semantics, or Discourse Repre-sentation�eory (Heim (), Kamp (), Kamp andReyle ())—essentiallyas restricted variables whose ultimate contribution to the larger structures of which

they are a part depends on the operators with which they happen to interact. �eappeal of this line of analysis is exactly that it provides a unified understanding of theapparently disparate range of interpretations shown by arbitrary subject pronouns.�ey tend to have quasi-universal force in habitual contexts like (), but existentialforce in episodic contexts such as ().

() a. ManArb

sprichtspeak [PRES]

EnglischEnglish

inin

Amerika.America

‘People in America speak English.’ German, Malamud ()b. In

InItaliaItaly

siArb

bevedrink [PRES]

moltomuch

vino.wine

‘In Italy, everybody/people drink wine.’ Italian, Chierchia ()c. Man

Arbmastemust

arbetawork [−FIN]

tilluntil

.

‘People have to work until the age of .’ Swedish, Egerland ()

() a. ManArb

tanztedance [PAST]

aufat

derthe

Party.party

‘People danced at the party.’ German, Malamud ()b. In

InItaliaItaly

ieriyesterday

siArb

èhave [PRES]

giocatoplayed

malebadly

‘In Italy yesterday people/they played badly. ’ Italian, Chierchia ()c. Man

Arbarbetadeworked

ifor

tvatwo

manadermonths

för attto

lösasolve

problemet.the problem

‘Peopleworked for twomonths to solve the problem.’ Swedish, Egerland()

Notice that these possibilities parallel closely the interpretive possibilities just docu-mented for the autonomous argument in Irish. Arbitrary subjects also have pseudo-specific uses much like the Irish patterns of () (see especially Kratzer (),Chierchia ()).

An additional characteristic of arbitrary subjects that has been the focus of the-oretical attention is the fact that they exhibit a very curious and very distinctive setof anaphoric properties. It is possible (as we have seen) for an arbitrary pronoun tobind a reflexive or a reciprocal, and also to enter into anaphoric relations with otherarbitrary pronouns, as in () (Chierchia (, (b), p. )):

() Ieri,yesterday

siArb

èisgiocatoplayed

malebadly

eand

siArb

èisperso.lost

‘Yesterday, people played badly and they/people lost.’

However, the pattern in (), in which the arbitrary subject enters into a forwardanaphoric relation with a personal pronoun, is firmly excluded. Italian () is fromChierchia (), French () and German () from Koenig and Mauner ().

() *sijArb

èisdettosaid

chethat

loroj

theyhannohave

sbagliatoerred

‘Peoplej said that theyj were wrong.’

() *Onj

Arbahas

assassinékilled

lathe

présidente.president

Iljhe

étaitwas

dufrom-the

Berry,Barry

paraît-il.seems-it

‘Someonej murdered the (woman) president. Hej comes from the Berry, itseems.’

() *Manj

Arbhathas

diethe

Präsidentinpresident

erschossen.shot

Erjhe

kamcame

ausfrom

Bayern.Bavaria

‘Someone shot the (woman) president. He comes from Bavaria.

�e corresponding patterns in Irish are exactly parallel, as seen in ():

() a. do[PAST]

stadadhstop [PAST-AUT]

agusand

scaoileadhrelease [PAST-AUT]

amachout

nathe

líontanets

‘One stopped and let out the nets’ b. *Dúradh

say [PAST-AUT]

goC

rabhadarbe-[PAST] -[P]

bocht.poor

‘Peoplej said that theyj were poor.’

To express the intended meaning of (b), one would have rather ():

() Dúradhsay [PAST-AUT]

goC

rabhthasbe-[PAST-AUT]

bocht.poor

‘Peoplej said that theyj were poor.’

Our general conclusion, then, is that the properties of the autonomous argumentparallel point for point the established properties (interpretive and anaphoric) ofarbitrary subject pronouns.

Consider a final parallel. One of the threads which runs all through the liter-ature on arbitrary pronouns is the intuition that such pronouns are similar to, oridentical with, the ‘arbitrary’ understanding of —the silent subject of controlledinfinitival clauses. It is striking, then, that examples such as that in () are possible

Chierchia () argues that the impossibility of (a) derives from the sortal restrictions whichlimit the possible reference values of the arbitrary subject. Given the observations of the next section,however, this proposal seems insufficiently general, since it will not extend to the Irish (b), eventhough the facts seem to be largely parallel and to deserve parallel explanation.

in Irish.

() D’iarrasked

síshe

peannpen

agusand

páipéarpaper

a thabhairtbring [−FIN]

chuici.to-her

Tugadh.bring [PAST-AUT]

‘She asked that pen and paper be brought to her. �ey were.’

() exemplifies an ellipsis construction in Irish which mimics point for point theproperties of ellipsis in English. It has been standardly analyzed as ellipsis ofthe complement of one of the functional heads to which the finite verb raises (fordetailed discussion see McCloskey (), Goldberg () and references citedthere). A crucial property of this ellipsis, and one that is useful for our presentpurposes, is that (by contrast with English) the post-verbal subject forms part ofthe elided material. Hence only the raised verb (along with any adverbial elementswhich attach high enough) survives to pronunciation.

�e importance of () in the present context, however, lies in the fact that herethe verbwhich has survived ellipsis (tugadh) is in the autonomous form. From that itfollows, on our present assumptions, that the elidedmaterial contains an occurrenceof the autonomous argument.�e elided material, including the autonomous argu-ment, must then meet the requirement of identity with an antecedent by which theellipsis is licensed.�e antecedent in () is a nonfinite clause whose subject is ‘arbi-trary ’. And it is arbitrary which corresponds to the autonomous argumentwithin the ellipsis-site. From this in turn it follows that the autonomous argumentmust be similar enough in relevant respects to ‘arbitrary ’ that it counts as beingidentical to it in whatever sense is necessary for the licensing of ellipsis.

Such observations provide support for the position of Stenson (), Anderson() and Harley (), all of whom identify the null subject of a verb in theautonomous form with the null subject of Control structures (i.e. with so-called

�e claim that the nonfinite clause in () has an arbitrary subject probably requires somecomment. Such a possibility would not be allowed in English in the complement of a verb such as ‘ask’.However it is allowed freely in Irish (as also in Polish, see Bondaruk ()), as can be seen in (i)–(iii).

(i) ar mhaith leatwould-you-like

anthe

carrcar

beagsmall

a ghléasadhget-ready [−FIN]

duit?for-you

‘Would you like the small car to be readied for you?’

(ii) Ní maith ledoes-not-like

haoinneanyone

éhim

a náiriúshame [−FIN]

‘Nobody likes to be shamed’

(iii) goC

raibhwas

glanchuimhnepure-memory

aigeat-him

aron

PhucslaíPuxley

a mharúkill [−FIN]

‘that he had the clearest memory of Puxley being killed’

). If the autonomous argument just is arbitrary , then obviously it will countas ‘similar enough’ to arbitrary to allow the kind of ellipsis exemplifed by ().

We will return to the issues raised by these observations. For the moment, andin the larger frame, they serve as a final strand of confirmation for the general thesisthat the autonomous argument of Irish shares the essential properties of the arbi-trary pronoun subjects of Romance and Germanic.

Contrasts

�ere is also, however, one important respect in which the Irish autonomous argu-ment goes its own way. Arbitrary pronouns in Germanic and Romance are subjectto a sortal restriction that they are semantically plural (refer to groups) and referonly to humans. No such restriction holds of the autonomous argument in Irish.() occurs as part of a narrative about the murder of Robert Kennedy in which itis clear that only one attacker is involved:

() Siúladhwalk [PAST-AUT]

suasup

go dtíto

Robert Kennedy …

‘Somebody walked up to Robert Kennedy …’ Ó Curnáin (, Vol.:)

In the case of (), the immediate linguistic context also makes it clear that the in-tended referent of the autonomous argument is singular.

() Scríobhfadwrite [FUT] [S]

chuigto

luchtpeople

stiúrthadirect [GEN]

ChonradhLeague

nathe [GEN]

hÉireann.Ireland [GEN]

Scríobhadhwrite [PAST-AUT]

chuigto

ConradhLeague

naof

GaeilgeIrish language

iin

mBaile Átha Cliath.Dublin‘I will write to those who run the Irish League.�e Gaelic League in Dublinwere written to.’

Such observations indicate that there is no requirement of plurality.�e examples in() indicate that the autonomous argument is further not restricted in its referenceto human, or even animate, individuals. Inanimate causes appear routinely:

() a. níorNEG–PAST

dóghadhburn-[PAST-HABIT]

nathe

nótaínotes

‘�e notes were not burned. b. Raiceáladh

wreck [PAST-AUT]

aron

chóstacoast

nathe [GEN]

SíneChina [GEN]

éhim

tráthtime

‘He was wrecked on the coast of China once.’ c. Nuair

whenaC

dhearcaimidwe-look

aron

anthe

méidquantity

léinn,learning

litríochta,literature

agusand

ceoilmusic

aC

tháinigcame

asfrom

áitplace

chomhas

beagsmall

leis,as-it

cuirtearput [PRES-AUT]

iontaswonder

orainnon-us

‘When we look at the quantity of learning, literature, and music thatcame from such a small place, we are amazed’

d. tháinigcome [PAST]

láday

millteanachterrible

gaoithewind [GEN]

móiregreat [GEN]

agusand

rinneadhmake [PAST-AUT]

smionagairlittle-pieces

denof-the

choláistecollege

adhmaidwood [GEN]

‘�ere came aday of terrible storms and thewooden collegewas smashedto pieces.’

What these observations cumulatively indicate is that there is no intrinsic require-ment that arbitrary pronouns be subject to sortal restrictions on possible referents.�is should hardly be a surprising conclusion, given the arbitrary character of suchrestrictions and given the historical origins of the lexical material out of which theRomance and Germanic pronouns have mostly been constructed (based on wordswhich in origin mean ‘human being’ or ‘man’ or else, as in Italian, on reflexive pro-nouns). It would surely be a strange thing if arbitrary pronouns were universallysubject to such strange and arbitrary restrictions.

Interim Summary

We have seen so far, then, that the autonomous inflectional ending on a finite verblicenses the appearance of a syntactically active but phonologically null element inthe position of the most prominent of the verb’s arguments. We have also seen thatthis element is close in its interpretive and anaphoric properties to the impersonalsubject construction of Germanic and Romance languages. Being syntactically ex-pressed, the null subject can bind anaphoric elements such as reciprocals. However,given the particularities of its interpretation (in particular, its failure to introducea discourse marker, in the analysis of Koenig and Mauner ()) it may not func-tion as an antecedent for subsequent definite pronouns. Such arbitrary subjects wein turn take to be indefinites in the sense of dynamic semantics or Discourse Repre-sentation �eory (‘ultra-indefinites’ or ‘a-indefinites’ in the terminology of Koenig(), Koenig andMauner ()).�e autonomous argument in Irish differs fromits European kin only in not being associated with sortal restrictions of any kind.Wehave seen, in addition, that the element licensed in this position is either identicalto in one of its uses (the so-called ‘arbitrary’ use), or else is similar enough in

relevant respects to arbitrary to make ellipsis of a containing constituent possi-ble.

�ere is much here that deserves closer investigation. For now, my limited goalis to flesh out empirically the widely held intuition that the autonomous form rep-resents a species of arbitrary subject construction, and further to provide reason tobelieve that at the heart of this construction is the licensing of a null element similarin its essentials to the arbitrary pronoun subjects of Romance and Germanic.

Licensing the Autonomous Form

Given this much, the analytical task is to construct an understanding of the licens-ing of a null pronominal element (with interpretive properties similar to, or iden-tical with, those of arbitrary ) by a set of verbal endings which are associatedwith finite tenses. Framed in these terms, the task is simplified both by theoreticalcommitment and by knowledge of the larger patterns governing such licensing inthe language.

�e general syntactic configuration underlying agreement in Irish is that seenin ():

()

[

αNumβPers

]

pro[

αNumβPers

]

Here, the relation between and pro is that of local command (i.e. pro is the mostprominent nominal in the domain of and is not contained within a phase whichexcludes ). itself is a member of one of the four or so functional (closed-class)categories of the language which may bear person and number marking morphol-ogy. Specifically, can be any of:

• Finite Tense (giving rise to subject-verb agreement)

• (giving rise to possessor agreement)

• v (giving rise to various species of object agreement)

• (giving rise to agreement between a preposition and its object)

�e first of these sub-cases is the one that is principally relevant to our present con-cerns.�is is the case seen in () (inwhich Tense is the closed-class categorywhichencodes finiteness and to which the verb raises in finite clauses).

() TenseP

Tense[

αNumβPers

]

pro[

αNumβPers

]

Schematic structures such as () surface in the form of examples like ():

() a. Labhradarspeak-[PAST] -[P]

pro leiswith

nathe

ceoltóirí.musicians

‘�ey spoke to the musicians.’b. Ní

NEG

abraimsay-[PRES] [S]

pro a dhath.anything

‘I don’t say anything.’

inwhich the inflected verb occupies either the position of Tense or a position furtherto the le than Tense.

�is overall view of the syntax of agreement represents a natural updating ofMcCloskey and Hale () and is very much in harmony with current thinkingabout the generalized syntax of agreement relations. It is defended and amplifiedin McCloskey (). Since the autonomous argument is a null pronoun licensedonly in the domain of a finite verb bearing the appropriate morphology, we are al-most required to extend () to the current case, by assuming that features of personand number are not the only ones which can figure in the relationship of (). Letus assume, more specifically, that the autonomous argument is a null pronominalwhich agrees with a finite Tense bearing the feature Arb. �e null pronominal willin turn bear an occurrence of that same feature. �e feature Arb is clearly uninter-pretable on Tense, and equally clearly interpretable on pro, triggering, as it does, theinterpretation in terms of a pure Heimian indefinite:

() TenseP

Tense[

FinArb

]

pro[ Arb]

Routine as this proposal is (almost forced by current theoretical commitments), itbrings with it some real analytical gains. In particular:

• It allows us to understand why autonomous forms are restricted to ‘subject’

position. �is restriction is a reflection of the general requirement of localityon the syntax of the relation . Once the feature Arb appears on Tense,it is determined that there must be a null pronominal within its domain withrespect to which it can act as a Probe (in the sense of Chomsky ()). �atnull pronominal must, in addition, be the most prominent nominal in thedomain of Tense; otherwise the crucial relationship of ()will be blocked andthe uninterpretable Arb feature on Tense will not be eliminated as required.�e restriction to subjecthood thus follows from very basic design principlesof language (as does the same requirement on agreement more generally).

• In addition, we understand the requirement that the autonomous argumentbe null. Or at any rate we understand that requirement as being one facet ofa larger pattern, since it is a general fact about this language that only nullpronominals may participate in the agreement relation (see McCloskey andHale () and much subsequent work, especially Andrews (), Legate(), Ackema and Neeleman (), McCloskey ()).

• It provides the right level of generalization to state the distribution of au-tonomous forms. �at is, these patterns are available for all finite verbs—exactly the level of generality which is achieved by associating the crucial li-censing feature with finite Tense.

• We understand the impossibility of (), since the complex reflexive formedby combining a personal pronoun with féin requires an antecedent with per-son andnumber featureswhichmatch those of the pronoun—somethingwhichthe autonomous argument lacks, given the analysis of (). We also under-stand the possibility of (), since arbitrary pronouns in general are knownto be able to bind reciprocals and the Irish reciprocal imposes no additionalfeatural requirements on its antecedent. What remains then is the more deli-cate empirical challenge of understanding the somewhat marginal characterof the phenomenon represented by (). For this, I think that we need to as-sume that the autonomous argument may be (but is not required to be) se-

As noted by Stenson (), some unaccusative verbs resist the autonomous inflection. For someof these (the so-called ‘salient unaccusatives’ whose only visible argument is prepositional), the pres-ence of a null expletive pronoun would block the crucial relationship of (). Others, for instance(i),

(i) a. *Taitnítearplease [PRES-AUT]

liom.with-me

‘�ings in general please me/I am easily pleased.’

I suspect are to be understood in the same context as Cherchia’s , observation that arbitrarysi in Italian is incompatible with kind-denoting predicates.

mantically plural. �e task for a speaker asked to judge the well-formednessof (), then, is first to determine (fromwhatever context is supplied or can beconjured up) whether or not the (variable of the) autonomous argument hassingular or plural reference. If plural, the example will be judged well-formed;if singular, the example will be judged ill-formed.

• Finally, these proposals allow us tomake a link with a larger typological prop-erty of Irish. Likemost verb-initial languages, Irish is amassively head-markinglanguage, in the sense of Nichols ().�at is, in grammatical relationshipswhich link a headwith a phrasal dependent (a probe with a goal inminimalistterms), the language marks the head (the probe) rather than the dependent(the goal). �e contrast between Romance and Germanic on the one handand Irish on the other now emerges as a contrast between languages whichmark the licensing of the arbitrary pronoun on the licensed pronoun (Ro-mance and Germanic languages) and those which mark it on the licensinghead (Irish). Put another way, Irish emerges as the head-marking counter-part of the dependent-marking pattern found in other European languages,and this trait emerges as but one aspect of a much larger typological pattern.

�ere is at least one large issue le open in the partial analysis developed here.�is issue emerges as a seeming contradiction at the heart of the proposal. On theone hand, it treats the autonomous argument as being licensed by the same kindsof mechanisms as license null personal pronouns in finite clauses (the mechanismsof pro-drop in the classic sense). On the other hand, it points to evidence that theautonomous argument is deeply similar to, or identical with, the null subject of ar-bitrary control infinitivals–the evidence of () in particular. Put differently, the dis-cussion to date seems to identify the autonomous argument both with pro and with, something which indeed was, in classic Government and Binding theory, acontradiction.

It is unclear to me whether or not this seeming contradiction is real. In thechanged theoretical context, it is very unclear what the difference between pro and might consist of, and the classic treatment of the limited distribution of (that it can appear only in un-governed positions) is not available.�e correct char-acterization of the element called is at present, as far as I know, an open theo-retical issue.

�at being so, it is reasonable, it seems tome, to interpret facts like those consid-ered here as revealing part of the truth about . �at truth I take to be something

If this optionality has a syntactic correlate, we will asssume that the autonomous argument bears aspecification for number, as well asArb.�e crucially missing feature that makes () impossible mustthen be person. Compare Kratzer () onGermanman. Note, however, that there is no requirementthat the binder of a reciprocal pronoun in Irish be syntactically plural, as shown in McCloskey ()

like this: among the very limited range of items which nonfinite Tense can licensein its domain is a null pronominal whose interpretive properties mirror those of theautonomous argument in Irish and the arbitrary subject pronouns of Romance andGermanic. �at is the fact that we need to understand.

�e Lexically Restricted Cases

In the final part of the paper, I try to show that the simple proposal developed inthe previous section lets us understand an aspect of the syntax of the autonomousconstruction that seems initially very strange.

�e initial observation here is that there are lexically restricted uses of the au-tonomous inflection. �ese are cases in which verbs exhibit the form of the au-tonomous but not its particular interpretation. Rather, the meaning of these struc-tures is unpredictable from their apparent subparts. �e verb caill, for instance,means ‘lose’. But in its autonomous form, it means ‘die’ as in (a).�e basemeaningof the verb cas is ‘turn’ or ‘twist’ (transitive and intransitive). But in the autonomousform, it means ‘to meet’ (accidentally), as in (b,c).

() a. Cailleadhlose [PAST-AUT]

dhátwo

bhliainyear

ó shoinago

é.him

‘He died two years ago.’b. Casadh

turn [PAST-AUT]

ormon-me

éhim

aréir.last-night

‘I met him last night.’c. Casadh

turn [PAST-AUT]

aron

a chéileeach other

iadthem

aron

anthe

aonach.fair

Stenson () (like Anderson () and Harley ()) analyses the autonomous argumentas an instance of —a proposal which I take to be consistent with the discussion of this section.Stenson also, however, presents arguments that the autonomous argument is distinct from the nullpronominals licensed by agreement morphology. �e core argument is that the various tests whichindicate the presence of a null pronominal (those identified in McCloskey and Hale ()) fail withrespect to the autonomous argument.�ese diagnostic tests are of two kinds. One shows that null procan be coordinated with other (overt) nominals. A second class shows that certain suffixes (demon-strative, reflexive, contrastive) which attach to overt pronouns also attach to null pro. None of thesediagnostics identify pro in the position of the autonomous argument.I think, though, that we can understand this failure within the context of the current proposals.

For the coordination facts, I believe it to be systematically true even of the overt arbitrary pronounsof Romance and Germanic that they cannot be coordinated with other nominals. For the suffixes, Ithink that in all cases there is a semantic incompatibility between the requirements of the suffix andthe semantic character of the autonomous argument. Demonstratives attach only to definite nominals.Contrastive suffixes attach only to focused pronouns. And so on. If this interpretation is maintainable,the failure of such diagnostics is understandable even in the context of ().

‘�ey met (each other) at the fair.’

Similar, though in a less obvious way, is the case in ():

() Báitheadhdrown [PAST-AUT]

anuraidhlast year

é.him

‘He drowned last year.’

�ere is a transitive verb báigh meaning ‘drown,’ of which the verb in () is itspast autonomous form. �us, a possible interpretation of () is: ‘He was drownedlast year by individuals or forces unspecified.’ However, this is not the only possi-bility. �e more common interpretation for () is that suggested by the translationoffered, one corresponding to an English sentence with intransitive unaccusativedrown—a verb with a single (internal) argument and one which as a consequencesupports no entailment about the existence of an agent or cause of the drowning.

�ere is a semi-productive extension of this phenomenon. Verbs denoting psy-chological states may appear in the autonomous form, in which case their experi-encer argument is expressed not as a nominative subject, but rather as a dative—marked with the preposition do. �is is illustrated for various pairs in ()–().

() a. Chonaicsee [PAST]

méI

goC

raibhbe [PAST]

séhe

seoDEMON

iontachvery

contúirteach.dangerous

‘I saw that this guy was very dangerous.’b. Títhear

see [PRES-AUT]

domhto-me

goC

bhfuilbe [PRES]

séhe

seoDEMON

contúirteach.dangerous

‘It seems to me that this guy is dangerous.’

() a. ónfrom-the

uiscewater

aC

samhlaighimagine [PAST]

méI

ba cheartshould

a bheithbe [−FIN]

glanclean

‘from the water that I imagined ought to be clean’b. ón

from-theuiscewater

aC

samhlaíodhimagine [PAST-AUT]

domto-me

ba cheartshould

a bheithbe [−FIN]

glanclean

‘from the water that I imagined ought to be clean’

() a. Cheapthink [PAST]

méI

goC

raibhbe [PAST]

cumalook

ghruamagloomy

orthu.on-them

‘I thought that they looked gloomy.’b. ceapadh

think [PAST-AUT]

domto-me

goC

raibhbe [PAST]

cumalook

ghruamagloomy

orthuon-them

‘It appeared to me that they looked gloomy.’

() a. �uigunderstand [PAST]

méI

náC NEG–PAST

raibhbe [PAST]

anthe

geimhreadhwinter

fósyet

ann.in-it

‘I understood that it wasn’t the winter yet.’b. tuigeadh

understand [PAST-AUT]

domto-me

náC NEG–PAST

raibhbe [PAST]

anthe

geimhreadhwinter

fósyet

annin-it‘I gathered that it wasn’t the winter yet.’

() a. anthe

t-athrúchange

aC

mheasthink [PAST]

siadthey

aC

bheadhbe [COND]

acuat-them

‘the change that they thought they would have’b. an

thet-athrúchange

aC

measadhthink [PAST-AUT]

dóibhto-them

aC

bheadhbe [COND]

acuat-them

‘the change that it seemed to them they would have’

Not all of the autonomous uses in ()–() are available in every dialect or forevery speaker, this being an area in which there is considerable lexical variation,it seems.But the general pattern is that illustrated in (). Informally speaking, apsychological verb which can appear in the syntactic frame (a) may also appearin the frame (b).

() a. [ V DP[Nom]

CP ]

b. [ V[Aut]

[PP do DP ] CP ]

�e range of verbs which actually participate in the alternation in () (those whichallow both variants) must be listed in an idiosyncratic way, it seems, for differentvarieties of the language. Here too the appearance of autonomous inflection is asso-ciated with a shi in the base meaning of the verb to which the inflection attaches.Although it is hard to be exact about the matter, the variants in (a) and (b) (thepairs in ()–()) are not completely synonymous. �e difference is subtle, but Isuspect that the key notion in explicating it is evidentiality—the speaker has less

�eb-examples of () and (), though, are, I think, available in every variety.�ere is in additionat least one impersonal psych-predicate which participates in the alternation illustrated in ()–():

(i) a. �aibhsighappear [PAST]

séit

domto-me

goC

gcualahear [PAST] [S]

daoinepeople

ag caint.talk [PROG]

‘It seemed to me that I heard people talking.’b. Taibhsíodh

appear [PAST-AUT]

domto-me

goC

gcualahear [PAST] [S]

daoinepeople

ag caint.talk [PROG]

‘It seemed to me that I heard people talking. ’

warrant for the truth of the embedded proposition in the b-examples than in thea-examples.

Besides understanding the morphosyntax of the alternation in (), then, wemust in addition link a semantic difference with the syntactic alternation. In thissense, the task for these cases is the same as the task for ().

Concerning these special uses of the autonomous inflection, there are two puz-zles to be solved.�e first is how we link the special meaning that they express withthe appearance of the autonomous inflection. �e second is a little more challeng-ing, at least initially. Notice that in the examples of ()–(), there are no silentarguments. �at is, all of the required arguments of the predicates in question areexpressed overtly by way of the normal morphosyntactic means that the languagemakes available (accusative case-marking in particular, and by way of the dativepreposition do).�ere is every reason to believe, that is, that examples such as ()–() contain no instance of the autonomous argument, even though the ‘normal’function of the autonomous inflection is exactly to license such an argument.

�ese two puzzles are linked in a way that is worth stressing since it is not ob-vious. For every case that I know of in which the autonomous inflection is used toexpress idiomatic meanings of this kind, it is also true that in the idiomatic use, itdoes not license a null arbitrary pronominal within its domain—even though it doesso in all more routine uses. �is linking is a property which any adequate analysismust aim to capture.

Let us examine how these puzzles might be understood.�e obvious (and I think also correct) analysis within our frame of reference is

to assume that in such cases also there are two instances of the featureArb (just as inthe regular uses of the autonomous inflection).One instance of the feature resides onfinite Tense and gives rise ultimately to the appearance of autonomous morphologyon the finite verb. Being an uninterpretable feature it is obliged to enter into anagreement relation with another instance of the same feature. �at second instanceresides on the lexical verbs of ()–() and it is this feature which, by way of the relation, eliminates the Arb feature on Tense. It is also this feature which actsas a diacritic to trigger construction of the special meanings of ()–().�at is, wetreat these special cases in much the same way that deponent verbs in Latin (passivein theirmorphology, active in their syntax and interpretation) are treated by Embickand Noyer (). �e second (lower) occurence of Arb in cases such as ()–()is uninterpretable in the technical sense, since in this position, it is dissociated fromits normal semantic content (that of introducing a ‘pure indefinite’, if the discussionearlier is on the right track). Its function is entirely diacritic, in that it serves only toformally mark in the lexicon the special meaning in question.

�is line of analysis too is almost forced given the framework as developed so far.To provide for the morphology, finite Tense in ()–() must bear the feature we

have called Arb. But that means that there must in turn be another instance of thesame feature in its command-domain with which it can interact so as to be elim-inated before the structures of which it is a part are transferred to the semantic-pragmatic interface. But as we have seen, there is no occurrence of arbitrary pro inexamples ()–() which could perform this crucial function.

But there must also be some feature on the verb which forces the special inter-pretations exemplified in ()–() and which distinguishes such uses from moreroutine uses. Considerations of analytical parsimony will now lead us to identifythese two features (the uninterpretableArb of Tense and the diacritic feature of ()–()) if possible.�e problem of the uninterpretableArb feature on Tense is therebyresolved.

We now hold, then, that there are two distinct interactions which the definingfeature of the autonomous morphology may enter in to—the one we saw earlier in() and the one seen in ().

() TenseP

Tense[

FinArb

]

[ Arb ]

It is the structure of () which gives rise to ()–().We now have a way of understanding the puzzle laid out earlier. �e Arb fea-

ture of Tense must interact with some element in its domain (in order to ensure itsown elimination). But it can interact with only one such feature. Once it has enteredinto an agreement relation with one element or the other, it is checked (valued, as inChomsky () or as in Pesetsky and Torrego (, )) and will be inactive—unavailable for further interaction. Hence, it can have within its domain either arbi-trary pro or one of the special verbs in ()–(), but never both.�is is the linkingthat we had hoped to ensure.

�e results, then, are what we want—in accounting for the core facts and alsoin accounting for the various puzzles connected with the idiosyncratic uses of theautonomous inflection.

Itmay seem, however, that toomuch is beingmade of this.Would it not bemuchmore straightforward to simply assume that the feature Arb occurs on suitably in-flected forms of the verbs in ()–() to indicate their special interpretation? Anyfurther talk of an agreement relation between a disembodied Tense element and afeature on those verbs might seem to represent a spurious and un-necessary com-plication. My final effort here will be to try to show that such scepticism is unwar-ranted and that the interpretation offered earlier is in fact the right one. �ere are

cases in which the autonomous inflection which (ultimately) licenses the special in-terpretations of ()–() does not in fact appear on the verb, but rather in a higherinflectional position. Nonetheless, the special interpretations of ()–() persist.

Two Periphrastic Aspects—Progressive and Perfect

�e crucial observations involve two periphrastic aspects—the progressive and theperfect. For the progressive in ()we have the particle spelt ag preceding a nonfiniteform of the verb.

() Tábe [PRES]

siadthey

ag tógáilraise [PROG]

tithehouses

aron

anthe

Mhullach Dubh.

‘�ey’re building houses in Mullaghduff.’

Historically this particle is the preposition meaning ‘at’ but in the modern languageit has been grammaticized (brought into the closed class vocabulary) as an aspectmarker (for extended discussion, see McCloskey ()). For such structures, wecan assume a structure something like ().

() TenseP

Tense

v

v AspP

Aspv

[Subj]

v

[Obj]

tá

ag

In () is seen an instance of the perfective aspect.

() Bhíbe [PAST]

siadthey

(díreach)(just)

i ndiaidha er

anthe

bailehome

a �ágáil.leave [−FIN]

‘�ey had (just) le home.’

Here we have the preposition (ungrammaticized, it seems) meaning ‘a er’, a prepo-sition which in turn takes a nonfinite (Raising) complement. I will therefore assumea syntax along the lines of ().

() TenseP

Tense

v

v

TenseP[-Fin]

[Subj]

[Obj]

[-Fin]

tá

i ndiaidh

�e interaction between such periphrastic structures and the autonomous form isinitially unremarkable. For the progressive, we have (a) and for the perfect (b).

() a. Tátharbe [PRES-AUT]

ag iarraidhtry [PROG]

airgeadmoney

a bhailiú.gather [−FIN]

‘�ere is an attempt to raise money.’b. Bhíothas

be [PAST-AUT]

i ndiaidha er

airgeadmoney

a bhailiú.gather [−FIN]

‘Money had been raised.’

In such cases, the feature Arb on finite Tense licenses the appearance of the au-tonomous argument in the most prominent argument position within its domain(the subject of the complement of the aspectual marker in each case) and the mor-phology is ultimately realized on the finite verb a er it has fused with Tense.

Amore interesting interaction, though, is visiblewhenwe ask how the periphrasticaspects interact with autonomous forms of the second kind—idiomatic verbs such

�e specifics here are by no means crucial for the larger point under discussion. Both () and() presupppose raising of the subject to a position in the inflectional layer to the right of the surfaceposition of the finite verb.�e label F does duty forwhatever structure beyondTense is to be postulatedfor the inflectional layer.

as we saw in ()–(). �e relevant examples are given in ().

() a. tátharbe [PRES-AUT]

a’ mo[PROG] –[S]

chailleadhlose [−FIN]

‘I’m dying.’ b. Bhíothas

be [PAST-AUT]

i ndiaidha er

anthe

bheirttwo

bhanwomen

a chastáilturn [−FIN]

aron

a chéile.each other

‘�e two women had just met (each other).’ c. nach

NEG C

rabhthasbe [PAST-AUT]

ag casachtáilturn [PROG]

aonany

duineperson

deof

nathe

buachaillíboys

ógayoung

airon-him‘that he wasn’t meeting any of the young boys’

d. nathe

créatúircreatures

aC

bhíthearbe [PRES-AUT]

a bháitheadhdrown [PROG]

‘the poor creatures who were drowning’

Similar patterns are found with the psych-predicates, as shown in ():

() a. toisigheadhbegin [PAST-AUT]

a thaidhbhsiughadhseem [−FIN]

rudaíthings

marlike

sinthat

domh-sato-me

‘I began to imagine things like that.’ b. go

C

rabhthasbe [PAST-AUT]

ag samhladhimagine [PROG]

anthe

amatime

aC

bhíbe [PAST]

le theachtto-come

dítheto-her

‘that she was imagining the time that was to come’ c. an

therudthing

aC

bhíthearbe [PAST-AUT]

a shamhailtimagine [PROG]

damhto-me

‘the thing that I was imagining’ d. bhíthidhe

be [PAST-HABIT-AUT]

ag taidhbhreamhseem [PROG]

damhto-me

inin

momy

shuansleep

goC

…

‘it used to seem to me in my sleep that … ’

�e structures of () and () are very easy to document in published sources fromthe s, s and s and earlier, but seem to be largely unavailable to contempo-rary speakers. �e analytical task, then, is twofold—the theory must be such as toprovide for the possibility of () and (), and it must also provide a way of under-standing the mechanisms of variation and change which lie behind the diachronicfragility of the structure behind () and ().

For more detailed discussion and a larger collection of examples, see Mac Cana and Ó Baoill(), McCloskey ().

�ere are speakers who have () but not ().

Consider (a) as a typical instance of the phenomenon. We have here an in-stance of the special interpretation of the verb ‘lose’ in which it expresses the mean-ing ‘die’ (just as in ())—an intransitive verb with a single internal argument. �eexpression of this meaning, as we have seen, is normally linked with the appear-ance of autonomous morphology. And in (a) that licensing morphology is in-deed present. However, it does not appear on the verb itself (it is morphologicallyimpossible to have a nonfinite autonomous form).�emorphology appears, rather,exactly where we would expect it to appear given ()—on the Tense element, real-ized ultimately on the raised auxiliary tá (‘be’). �e structure is as in ().

() TenseP

Tense[

FinArb

]

Fv

vAspP

Aspv

v[

stSing

]

[ Arb ]

pro[

stSing

]

tá

a’

chailleadh

mo

-thar

�e featural interactions in () are routine. �e Arb feature of Tense agreeswith the corresponding feature on the verbal root caill (since this latter is the mostprominent occurrence ofArb in the domain of Tense and the unaccusative v is not aphase-defining head).Arb is thus eliminated fromTense, as required (just as in ()).�e occurrence of the diacriticArb on the verbal root forces construction of the spe-cial meaning ‘die’.�e verb, then, has just one argument—a single internal argumentsince it is unaccusative. �at argument in turn is realized as pro, which enters intoan agreement relation with the governing v , as required.�at is, the interactions arewhat we would expect given (). In each case, the inflectional morphology whichindicates presence of the feature Arb (a presence required ultimately by the special

meanings of () and ()) appears not on the lexical verb itself but rather on ahigher (but still accessible) Tense element—just as we would expect given () andthe theoretical context associated with it.

We understand, then, why () and () are possible in principle. What is thesource of the variation reflected in the loss of these structures in very recent times?A notable feature of these structures is that the single argument of the unaccusativeverb is realized as a direct object (see McCloskey () for more discussion). In(a), for instance, the morphosyntax which licenses the internal argument of ‘die’is the morphosyntax of object agreement, and in (b) the internal argument is inthe pre-verbal Object Shi position typical of direct objects in transitive nonfiniteclauses. �is is, in a sense, the only option open to the language for the licensing ofsuch structures. Finite Tense (whose responsibility it would normally be to licensethe internal argument of an unaccusative verb, given Burzio’s Generalization) is un-available. It must bear the feature Arb in order to license the special meaning of themain verb. But that in turn renders it incapable of licensing an argument (by wayof the interaction of nominative case and subject verb agreement). If such struc-tures are to be possible at all, then, some other licensing mechanism for the internalargument must be called in to play. �e mechanisms appealed to are those typicalof the licensing of direct objects in nonfinite clauses—object agreement, accusativecase licensing, genitive case licensing, and Object Shi (McCloskey (), Mc-Closkey (), McCloskey (), Chung and McCloskey (), McCloskey andSells (), Guilfoyle (), Noonan (), Duffield (), Carnie and Harley(), Bondaruk (), McCloskey () ).

�e licensing of () and (), then, depends on the availability of an unac-cusative v which bears the features responsible for the licensing of direct objects innonfinite clauses. Put another way, the possibility of () and () depends on thepresence, in a given variety of the language, of a set of exceptional items among theclosed class vocabulary of the language—members of the class v which introduceno second (external) argument but which, in violation of Burzio’s Generalization,still have the morphosyntactic wherewithall to license an internal argument.

In this way, we can meet the second of our two challenges—that of understand-ing the possibility of variation and the diachronic fragility of structures such as ()and (). �eir survival depends on speakers’ having continued access to sufficientpositive evidence for the postulation of marked and exceptional lexical items.

Note that the logic above applies also to the more routine idiosyncratic uses documented in (),(), and ()–(). �at is, they too depend on the availability of an instance of v with exceptionalcase-licensing properties—note the accusative case of the internal argument of (a), for example.�is must be a different member of the class than figures in () and (), however, since the mor-phosyntax of objecthood is quite different in finite and in nonfinite clauses. �e relative stability of(), () and ()–(), then, must reflect their relative frequency of occurrence—there is sufficient

Conclusion

Much work remains to be done on all of these matters. In particular, the quantifica-tional, referential, and anaphoric properties of the autonomous argument need to beprobed in a deeper and more serious way than has been possible here. In addition,the implications of these phenomena for the theory of Control and for the nature ofthe silent element called remain to be properly thought through. But I hope tohave shown here at least that the proper typological context for those investigationsis the context of arbitrary subject constructions generally. I also hope to have shownthat a morphosyntactic analysis of the autonomous construction which grows verynaturally (if not inevitably) out of current theoretical commitments achieves a rea-sonable level of descriptive success.

In the context of this volume in particular, it is worth noting that, in the ex-amples of () and of (), we have direct evidence for the existence of a class ofintransitive verbs which are non-agentive and which take only direct object argu-ments.We have, that is, direct evidence for one of the central contributionsmade byDavid Perlmutter to syntactic theory—theUnaccusativeHypothesis, which predictsthe existence of exactly such a class of verbs.

positive evidence available to learners to overcomewhatever default pressure lies behind Burzio’s Gen-eralization. �e variability within the class ()–() presumably reflects that same pressure. See alsofootnote .

If it is true that nonfinite Tense can also license an occurence of the feature Arb (in line with thediscussion of Control and arbitrary which grows out of () above), then we expect the specialmeanings of ()–() to be licensed in the domain of nonfinite Tense as well (i.e. in nonfinite clauses).In fact they are:

(i) a. Féadanncan [PRES]

túyou

carbadcart

nóor

dhótwo

deof

nathe

rudaíthings

sean-aimseardhaold-fashioned

a chastáilturn [−FIN]

duiton-you

‘You can come across one or two of the old-fashioned kinds of carts.’ b. Bhí

be [PAST]lúcháirjoy

aron

anthe

�earman

bhochtpoor

a leithéidsuch-a-one

a chastáilturn [−FIN]

airon-him

‘�e poor man was delighted to meet such a one.’ c. trí

threeseachtainíweeks

comhthromaeven

i ndiaidha er

momy

dhearbhráthairbrother

a chailleadhlose [−FIN]

‘three weeks exactly a er my brother had died’

�ese patterns depend on the same instance of v as () and () (unaccusative v bearing the mor-phosyntactic features which license objects in nonfinite clauses), and so we expect them to show thesame diachronic fragility as () and (). �is is, I believe, exactly the case.

Appendix—Sources of Attested Examples

Many of the examples used in this paper have been taken from published sourcesof one kind or another. When this is the case, it is indicated by way of a tag whichconsists of an abbreviation of the title of the publication followed by a page number.�e title-abbreviations are explained here.

AII Allagar II, Tomás Ó Criomhthain, ed. Pádraig Ua MaoileoinAM An Mhiorbhailt, translated Niall Mac SuibhneAT A�ig Ná Tit Orm, Maidhc Dainín Ó SéCCC Cnuasach Céad Conlach, Seán Bán Mac MeanmanCDC Castar na Daoine ar a Chéile, Scríbhinní Mháire , Séamus Ó

Grianna, ed. Nollaig Mac CongailCFC Céad Fáilte go Cléire, ed. Marion GunnDCA Dith-Chéille Almayer, translated Seosamh Mac GriannaDII Desiderius a Dó, Pádraig Ó CíobháinEMIT Eader Muir is Tír, translated Niall Ó DomhnaillFBF Fiche Blian ag Fás, Muiris Ó SúilleabháinGSA An Giorria San Aer, Ger Ó CíobháinI Ise, translated Niall Ó DomhnaillIA Iomramh Aonair, Liam Mac Con IomaireIAE In Aimsir Emmet, translated Colm Ó GaoraLDS Lá Dár Saol, Seán Ó CriomhthainLG Le Gealaigh, Pádraig Ó CíobháinMD An Mairnéalach Dubh, translated Seosamh Mac GriannaNCN Nár Chlos Ár Namhaid, Ger Ó CíobháinOT Ó�uaidh, Pádraig Ua MaoleoinPMB Pádhraic Mháire Bhán, Seán Ó RuadháinPNG Pobal na Gaeltachta, ed. Gearóid Ó Tuathaigh, Liam Lillis Ó

Laoire, Seán Ua SúilleabháinSAT Seanchas an Táilliúra, Aindrias Ó MuimhneacháinSS Scéalta Sealgaire, translated Maighréad Nic MhaicínSSOTC Sí-Scéalta ó �ír Chonaill, ed. Seán Ó Heochaidh, Máire Ní

Néill and Séamas Ó Catháin.U Unaga, translated Eoghan Ó NeachtainUMI Uaill-Mhian Iúdaigh, translated Tadhg Ó Rabhartaigh

References

Ackema, Peter, and Ad Neeleman. . Context-sensitive spell-out. Natural Lan-guage and Linguistic �oery :–.

Adger, David. . order and weak pronouns in Goidelic Celtic. CanadianJournal of Linguistics :–.

Anderson, Stephen. . Where’s morphology? Linguistic Inquiry :–.

Andrews, Avery. . Unification and morphological blocking. Natural Languageand Linguistic �eory :–.

Authier, J.-Marc P. . Arbitary null objects and unselective binding. in �enull subject parameter, ed. Osvaldo Jaeggli and Kenneth J. Safir, –. Dordrecht,Boston and London: Kluwer Academic Publishers.

Bondaruk, Anna. . PRO and control in English, Irish and Polish—a Minimalistanalysis. Lublin, Poland: Wydawnictwo .

Carnie, Andrew, andHeidiHarley. . Clausal architecture: the licensing ofmajorconstituents in a verb initial language. , University of Arizona.

Chierchia, Gennaro. . �e variability of impersonal subjects. inQuantificationin natural languages, volume , ed. Emmon Bach, Eloise Jelinek, Angelika Kratzer,and Barbara Partee, –. Dordrecht, Boston and London: Kluwer AcademicPublishers.

Chomsky, Noam. . Derivation by phase. in Ken Hale: A life in language, ed.Michael Kenstowicz, –. Cambridge, : Press.

Christian Brothers, (�e). . Graiméar Gaeilge na mBráithre Críostaí. Dublin,Ireland: Mac an Ghoill.

Chung, Sandra, and James McCloskey. . Government, barriers and smallclauses in Modern Irish. Linguistic Inquiry :–.

Cinque, Guglielmo. . On Si constructions and the theory of Arb. LinguisticInquiry :–.

Condoravdi, Cleo. . Indefinite and generic pronouns. in Proceedings of theEighth Annual West Coast Conference on Formal Linguistics, ed. E. J. Fee andK. Hunt, –. Stanford, California: CSLI.

D’Alessandro, Roberta. . Impersonal si constructions: Agreement and inter-pretation. Doctoral Dissertation, University of Stuttgart.

D’Alessandro, Roberta, and Artemis Alexiadou. . Inclusive and exclusive im-personal pronouns: a feature-geometrical analysis. Paper presented at the XXIVIncontro di Grammatica Generativa, Urbino, Italy, February .

Davidson,Donald. .�e logical formof action sentences. in�e logic of decisionand action, ed. N. Rescher, –. Pittsburgh, PA: University of Pittsburgh Press.

Duffield, Nigel. . Particles and projections in Irish syntax. Dordrecht, Boston,and London: Kluwer Academic Publishers.

Egerland,Verner. . Impersonal pronouns in Scandinavian andRomance.Work-ing Papers in Scandinavian Syntax :–.

Embick, David, and Rolf Noyer. . Movement operations a er syntax. LinguisticInquiry :–.

Goldberg, Lotus. . Verb-stranding ellipsis: A cross-linguistic study. DoctoralDissertation, McGill University, Montreal, Canada.

Greene, David. . �e Irish language/an Ghaeilge. Dublin, Ireland: Publishedfor the Cultural Relations Committee of Ireland at the �ree Candles, .

Guilfoyle, Eithne. . Functional categories and phrase structure parameters.Doctoral Dissertation, McGill University, Montreal, Canada.

Harley, Heidi. . Irish, the , and . University of Arizona, Tucson.

Heim, Irene. . �e semantics of definite and indefinite noun phrases. DoctoralDissertation, University of Massachussets, Amherst.

Jaeggli, Osvaldo. . On arbitrary plural pronominals. Natural Language andLinguistic �eory :–.

Kamp, Hans. . A theory of truth and discourse representation. in Formal meth-ods in the study of language, ed. Jeroen Groenendijk,�eoM. V. Janssen, andMar-tin Stockhof, –. Amsterdam: Mathematical Centre.

Kamp, Hans, and Uwe Reyle. . From discourse to logic. introduction to mod-eltheoretic semantics of natural language, formal logic and discourse representationtheory. Dordrecht, Boston and London: Kluwer Academic Publishers.

Koenig, Jean-Pierre. . On a tué le président! the nature of passives and ultra-definites. in Cognition and function in language, ed. B. Fox, D. Jurafsky, andL. Michaelis, –. Stanford, California: Publications.

Koenig, Jean-Pierre, and Gail Mauner. . A-definites and the discourse status ofimplicit arguments. Journal of Semantics :–.

Kratzer, Angelika. . German impersonal pronouns and logophoric-ity. Presented to Sinn und Bedeutung, Berlin , Handout available athttp://semanticsarchive.net.

Kria,Manfred, Francis Jeffry Pelletier, GregoryN. Carlson, Alice terMeulen, Gen-naro Chierchia, and Godehard Link. . Genericity: An introduction. in �egeneric book, ed. Gregory N. Carlson and Francis Jeffry Pelletier, –. Chicagoand London: �e University of Chicago Press.

Legate, Julie. . �emorphosyntax of Irish agreement. in : Papers onmorphology and syntax, cycle one, ed. Karlos Arregi, Benjamin Bruening, CorneliaKrause, andVivian Lin, –. Cambridge,: Department of Linguistics,.

Mac Cana, Proinsias, and Dónal P. Ó Baoill. . Gnéithe den chéasta sa Nua-Ghaeilge. in Dán do oide, essays in memory of Conn R. Ó Cléirigh, ed. AndersAhlqvist and Vera Capkova, –. Dublin, Ireland: Institiúid TeangeolaíochtaÉireann.

Malamud, Sophia A. . (In)definiteness-driven typology of arbitrary items. Toappear in Passives and Impersonals in European Languages, ed. S. Manninen, K.Hietaam, E. Keiser & V. Vihman. John Benjamins.

McCloskey, James. . Transformational syntax and model theoretic semantics: Acase-study in Modern Irish. Dordrecht: Reidel.

McCloskey, James. . Is there raising in Modern Irish? Ériu :–.

McCloskey, James. . A in a language. inOrder, concord and constituency,ed. Ewan Klein Gerald Gazdar and Geoffrey K. Pullum, –. Dordrecht: ForisPublications.

McCloskey, James. . Raising, subcategorization, and selection inModern Irish.Natural Language and Linguistic �eory :–.

McCloskey, James. . Case, movement and raising in Modern Irish. in WC-CFL IV: Proceedings of the fourth west coast conference on formal linguistics, ed.S. MacKaye J. Goldberg and M. Wescoat, –. Stanford, CA: Stanford Lin-guistics Association.

McCloskey, James. . Resumptive pronouns, A-binding and levels of represen-tation in Irish. in Syntax of the modern Celtic languages, ed. Randall Hendrick,volume of Syntax and Semantics, –. NewYork and SanDiego: AcademicPress.

McCloskey, James. . Clause structure, ellipsis and proper government in Irish.Lingua :–.

McCloskey, James. . Trí nóta comhréire. Ériu :–.

McCloskey, James. . On the right edge in Irish. Syntax :–.

McCloskey, James. . Reciprocals, parts, and wholes. inWebFest for Jorge Han-kamer, ed. Sandra Chung, James McCloskey, and Nathan Sanders. Department ofLinguistics, University of California, Santa Cruz. Http://ling.ucsc.edu/jorge.

McCloskey, James. . Class materials from A Course on Irish Syntax, LSA Insti-tute, Harvard andMIT. Summer .

McCloskey, James, and KennethHale. . On the syntax of person numbermark-ing in Modern Irish. Natural Language and Linguistic �eory :–.

McCloskey, James, and Peter Sells. . Control and A-chains in Modern Irish.Natural Language and Linguistic �eory :–.

Nichols, Johanna. . Head-marking and dependent-marking grammar. Lan-guage :–.

Noonan, Máire. . Case and syntactic geometry. Doctoral Dissertation, McGillUniversity, Montreal, Canada.

Ó Curnáin, Brian. . Aspects of the Irish of Iorras Aithneach, County Galway.Doctoral Dissertation, University College Dublin, Dublin, Ireland.

Ó Sé, Diarmuid. . Agent phrases with the autonomous verb in Modern Irish.To appear in Ériu.

Ó Siadhail, Mícheál. . Modern Irish: Grammatical structure and dialectal vari-ation. Cambridge, England: Cambridge University Press.

Parsons, Terence. . Events in the semantics of English. A study in subatomicsemantics. Cambridge,MA:MITPress.

Pesetsky, David, and Esther Torrego. . T-to-C movement: Causes and conse-quences. in Ken Hale: A life in language, ed. Michael Kenstowicz, –. Cam-bridge, : Press.

Pesetsky, David, and Esther Torrego. . Tense, case, and the nature of syntacticcategories. in�e syntax of time, ed. Jacqueline Guéron and Jacqueline Lecarme,–. Cambridge, : Press.

Rizzi, Luigi. . Null objects in Italian and the theory of pro. Linguistic Inquiry:–.

Stenson, Nancy. . Studies in Irish syntax. Tübingen: Max Niemeyer Verlag.

Stenson, Nancy. . Irish autonomous impersonals. Natural Language and Lin-guistic �eory :–.

�urneysen, Rudolf. . A grammar of Old Irish. Dublin, Ireland: Dublin Institutefor Advanced Studies.