Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To...

23
Passives and impersonals James P. Blevins Research Centre for English and Applied Linguistics University of Cambridge [email protected] Draft, April 2001 1 Introduction Passive and impersonal constructions have a strikingly different status in current theoretical and descriptive studies. All formal approaches recognize passive constructions and provide some means of relating their properties to those of corresponding actives. Any framework that did not would be considered fundamentally deficient or incomplete. Many descriptive grammars likewise apply a broad notion of ‘passive’ to any alternation that involves subject suppression. In contrast, impersonal constructions often remain implicit in language descriptions, and are neglected altogether in theoretical work. This omission is not accidental, as most frameworks expressly exclude the possibility of subjectless constructions. hpsg is one of the few that appears relatively agnostic on this score. Nearly all other approaches incorporate some subject-legislating constraint, whether expressed as the ‘Extended Projection Principle’ of Chomsky 1981, the ‘Final 1 Law’ of Perlmutter and Postal 1983b, or the ‘Subject Condition’ of Bresnan and Kanerva 1989. Taken together, the near-universal recognition of passives and the corresponding neglect of impersonals introduce a tacit descriptive bias in favour of passives. Constructions that occupy the communicative niche associated with the passive are often treated as passives, even when they differ from passives in formal respects that are noted in the traditional or specialist literature. This paper argues that the neglect of impersonal constructions has had two significant con- sequences. The first is a descriptive misanalysis of individual constructions, illustrated by the ‘passive’treatment of ‘impersonal voice’forms in Finno-Ugric, and ‘autonomous’forms in Celtic. The second consequence is an extended notion of ‘passive’ that subsumes formally distinct sub- constructions and exhibits variation that confounds attempts to impose substantive constraints. The conclusions drawn from ‘impersonal passives’ in Balto-Slavic illustrate the theoretical ef- fects of this misclassification. A passive analysis of synchronically impersonal forms in -no/-to in Ukrainian has fostered the belief that passive constructions may retain structural accusative objects (Sobin 1985). A similar misclassification of -ta forms in Lithuanian underlies claims that passives may be formed from what Perlmutter 1978 calls ‘initially unaccusative’verbs (Timber- lake 1982, Nerbonne 1982). Since these patterns violate various laws proposed within Relational Grammar (rg), they have been interpreted as refuting relational analyses of the passive. In response, a number of accounts, including Baker et al. 1989 and Bresnan and Kanerva 1989, introduce generalized mechanisms that admit, amongst other possibilities, passives of unaccusative verbs. Yet no account that sanctions an expanded class of ‘passive’constructions provides any general strategy for excluding unaccusative passives in the many languages that 1

Transcript of Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To...

Page 1: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

Passives and impersonals

James P. BlevinsResearch Centre for English and Applied Linguistics

University of [email protected]

Draft, April 2001

1 Introduction

Passive and impersonal constructions have a strikingly different status in current theoretical anddescriptive studies. All formal approaches recognize passive constructions and provide somemeans of relating their properties to those of corresponding actives. Any framework that didnot would be considered fundamentally deficient or incomplete. Many descriptive grammarslikewise apply a broad notion of ‘passive’ to any alternation that involves subject suppression.In contrast, impersonal constructions often remain implicit in language descriptions, and areneglected altogether in theoretical work. This omission is not accidental, as most frameworksexpressly exclude the possibility of subjectless constructions.hpsg is one of the few that appearsrelatively agnostic on this score. Nearly all other approaches incorporate some subject-legislatingconstraint, whether expressed as the ‘Extended Projection Principle’of Chomsky 1981, the ‘Final1 Law’of Perlmutter and Postal 1983b, or the ‘Subject Condition’of Bresnan and Kanerva 1989.Taken together, the near-universal recognition of passives and the corresponding neglect ofimpersonals introduce a tacit descriptive bias in favour of passives. Constructions that occupythe communicative niche associated with the passive are often treated as passives, even whenthey differ from passives in formal respects that are noted in the traditional or specialist literature.

This paper argues that the neglect of impersonal constructions has had two significant con-sequences. The first is a descriptive misanalysis of individual constructions, illustrated by the‘passive’treatment of ‘impersonal voice’forms in Finno-Ugric, and ‘autonomous’forms in Celtic.The second consequence is an extended notion of ‘passive’ that subsumes formally distinct sub-constructions and exhibits variation that confounds attempts to impose substantive constraints.The conclusions drawn from ‘impersonal passives’ in Balto-Slavic illustrate the theoretical ef-fects of this misclassification. A passive analysis of synchronically impersonal forms in-no/-toin Ukrainian has fostered the belief that passive constructions may retain structural accusativeobjects (Sobin 1985). A similar misclassification of-ta forms in Lithuanian underlies claims thatpassives may be formed from what Perlmutter 1978 calls ‘initially unaccusative’ verbs (Timber-lake 1982, Nerbonne 1982). Since these patterns violate various laws proposed within RelationalGrammar (rg), they have been interpreted as refuting relational analyses of the passive.

In response, a number of accounts, including Bakeret al. 1989 and Bresnan and Kanerva1989, introduce generalized mechanisms that admit, amongst other possibilities, passives ofunaccusative verbs. Yet no account that sanctions an expanded class of ‘passive’ constructionsprovides any general strategy for excluding unaccusative passives in the many languages that

1

Page 2: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

disallow them. Nor does any such account offer a coherent classification that identifies (non-diacritic) properties that correlate with the possibility of forming unaccusative passives.

This paper argues that there can be no general classification of languages that permit passivesof unaccusatives, because, as claimed inrg accounts, there are no such languages. Passivesof unaccusative verbs do not exist because passivization is arelation-changingprocess that de-motes an ‘initial’ subject to an optional oblique. Whether a passive construction is personal orsubjectless depends entirely on the properties of any remaining arguments. In contrast, imper-sonalization is arelation-preservingprocess that directly suppresses the realization of a ‘final’subject. Impersonalization is thus generally insensitive to the initial argument structure of a verb,and, in particular, is wholly indifferent to the contrast between unaccusatives and unergatives.

The reclassification of impersonal constructions thus preserves therg insight that ‘[n]o imper-sonal Passive clause in any language can be based on an unaccusative predicate’ (Perlmutter andPostal 1984:107). At the same time, the recognition of subjectless constructions violates some ofthe constraints proposed inrg to implement this ban, and undermines the rationale for others. Yetby jettisoning these often theory-bound constraints, it is possible to consolidate the complemen-tary strengths of relational and lexicalist perspectives. Althoughrg accounts provide a highlydeveloped theory of passives, and relation-changing alternations in general, these accounts arealso incomplete in critical respects, as they offer only the most programmatic remarks about mor-phological realization. At the other extreme, lexicalist approaches, such as Lexical FunctionalGrammar (lfg) and Head-driven Phrase Structure Grammar (hpsg), contain precise mechanismsfor expressing changes in form and function, while sacrificing some of the descriptive coverageachieved inrg by abstracting away from issues of morphological implementation.

The following sections suggest that a revealing analysis of passive and impersonal constructionscan be obtained if relational insights are, contrary to their proponents’ original intent, recast asclaims about lexical valence alternations. The main benefits of this synthesis are listed in (1).

(1) a. A lexicalist perspective accommodates the fact that passivization is, morphologi-cally, a derivational operation. Applying relation-changing rules to lexical argumentstructures accounts for the abstractness ofrg analyses, since such structures are lessspecified than syntactic arguments. The lack of a morphological component withinrg also follows ifrg is reinterpretedin totoas a theory of morpholexical alternations.

b. A traditional subject-oriented treatment of passives interacts transparently with theUnaccusative Hypothesis of Perlmutter 1978. On the assumption that only unergativepredicates specify ‘initial’ subjects, it follows that a derivational passive rule thattargets initial subjects will only apply to unergatives. Hence a lexicalist account thattreats subject demotion as the primary effect of passivization can derive the effectsof the 1-Advancement Exclusiveness Law (1aex) of Perlmutter and Postal 1984.

c. Restoring the traditional distinction between passive and impersonal constructionsaccommodates a range of impersonal passives that appear to violate the1aex andother relational laws. A reexamination of the ‘transitive’ passives reported in Welsh(Comrie 1977) or Ukrainian (Sobin 1985) supports the traditional view that theseconstructions are synchronically impersonal, not passive. The forms that occurin ‘unaccusative’ and ‘double’ passives in Lithuanian (Timberlake 1982) likewisehave an evidential meaning that identifies them as part of the mood system of thelanguage. Moreover, the impersonal nature of these forms is arguably a consequenceof a categorial difference, as they retain conservative nominal characteristics.

2

Page 3: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

2 Passive and impersonal constructions

To clarify the basic contrast between passives and impersonals, it will be useful to examine somecanonical examples of each construction. Sections 2.1 and 2.2 identify some of the core propertiesthat distinguish representative passive constructions in German from impersonal constructionsin Estonian and Welsh. Section 2.3 then considers the synchronic status of the historically neuterparticiples that head the various types of ‘impersonal passive’ constructions in Balto-Slavic.

2.1 Passives in German

Personal passives are often regarded as ‘the core case of passive’ (Chomsky 1981), in that theyexhibit all of the properties associated with the passive construction. The German example in(2b) illustrates the canonical profile of a personal passive. The ‘logical’ subject of the active (2a)is demoted to an optional oblique in (2b). The ‘logical’ object in (2a) is obligatorily realized in(2b) as a ‘derived’ subject, which occurs in the nominative and triggers subject agreement.

(2) a. Derthe.nom

Beamteofficial

hathas

denthe.acc

Vorschlagproposal

abgelehnt.rejected

‘The official has rejected the proposal.’

b. Derthe.nom

Vorschlagproposal

wurdewas

(vomby+the.dat

Beamten)official

abgelehnt.rejected

‘The proposal was rejected (by the official).’

Debates about the universal properties of passives turn, for the most part, on which propertiesof personal passives are taken to be definitional, and which are merely characteristic. Impersonalpassives have played a pivotal role in these debates. As stressed by Comrie 1977, impersonalpatterns like (3) suggest that a universal passive rule should not refer to objects or promotion.

(3) a. Vielemany

Leutepeople

rauchensmoke

inin

derthe

Küche.kitchen

‘Many people are smoking in the kitchen.’

b. Inin

derthe

Küchekitchen

wurdewas.3sg

(von(by

vielenmany

Leuten)people)

geraucht.smoked

‘There was smoking (by many people) in the kitchen.’

Comrie’s conclusions are contested in all varieties ofrg, since they entail a treatment ofthe passive in which demotion of the logical subject occurs ‘spontaneously’, rather than as aconsequence of an antecedent promotion. Hence, therg literature develops alternative analysesin which impersonal passives involve advancement of a demotion-inducing ‘dummy’ nominal.Setting aside the issue of dummies, the Motivated Chômage Law, which is preserved inrg

accounts, is one of the most theory-bound and excisable components ofrg. The claim that alldemotions arise as a side-effect of promotions serves to constrain the space of derivations inrg.However, there is no analogue of this sort of restriction in any other approach, and it is unclearwhether the constraint rests on any basic intuition that can be reconstructed outside ofrg.1

1The intuition that objects are somehow essential to the passive construction can be expressed in the conjecturethat no language contains a passive strategy that solely defines impersonal passives.rg accounts do not make thisassumption; Perlmutter and Postal (1984:110), for example, assume that Celtic contains a dedicated impersonalpassive rule. However, the Celtic forms in question are ‘autonomous’ impersonals (cf. §2.2.2), rather than passives.

3

Page 4: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

The present account thus follows Comrie 1977 in treating passivisation as a subject-sensitiveoperation that directly demotes the subjects in personal and impersonal passives. Object ad-vancement in personal passives is then a purely ‘opportunistic’side-effect of passivization. Theseadvancements reflect general constraints on the mapping between lexical argument structure andfinal grammatical functions, constraints that are specified in more detail in section 3.

One of the most striking and theory-neutral hypotheses advanced withinrg is the UnaccusativeHypothesis (uh), which holds that ‘many intransitive clauses have an initial direct object but nosubject’ (Perlmutter and Postal 1983a:69). Such clauses, termed ‘initially unaccusative’, contrastwith ‘initially unergative’ clauses, which do contain an initial subject. In addition to the variouslanguage- and construction-specific ways in which this contrast is manifested,rg identifiesone invariant, putatively universal property, namely that ‘[n]o impersonal Passive clause in anylanguage can be based on an unaccusative predicate’ (Perlmutter and Postal 1984:107).

The claim that unaccusative intransitive verbs may not be passivized is robustly supported bycontrasts in German, and other ContinentalWest Germanic languages. As example (3b) indicates,unergative intransitives likerauchen ‘smoke’may be passivized in German. Unaccusatives likebleiben ‘remain’ cannot, however, as the unacceptability of (4b) shows.

(4) a. Vielemany

Leutepeople

bleibenremain

inin

derthe

Küche.kitchen

‘Many people are remaining in the kitchen.’

b. * Inin

derthe

Küchekitchen

wurdewas.3sg

(von(by

vielenmany

Leuten)people)

geblieben.remained

‘*There was remaining (by many people) in the kitchen.’

The difference between unergative and unaccusative passives is masked in English by a blanketprohibition against impersonal constructions in general. Since the passive of any intransitive verbwill be subjectless, and hence unacceptable, no intransitive verb may passivize. Yet the contrastbetween unergative and unaccusative passives reasserts itself in transitive verbs. The Germanverb dauern and its English counterpartlast in (5) are canonical unaccusatives that fail topassivize. Verbs likewiegen ‘weigh’ andkosten ‘cost’ follow a similar pattern.

(5) a. Diethe

Tagungmeeting

dauertlasts

eineone

Woche.week

‘The meeting lasts a week.’

b. * Eineone

Wocheweek

wurdewas

(von(by

derthe

Tagung)meeting)

gedauert.lasted

‘*A week was lasted (by the meeting).’

It is initially somewhat surprising that the prohibition against passives of unaccusatives doesnot follow directly from the form of the passive rule inrg. However, this is a straightforwardconsequence ofrg commitment to the Motivated Chômage Law and a promotional treatment ofpassives. Since promotional accounts are intrinsically object- rather than subject-oriented, theycannot discriminate between those verbs that specify initial subjects and those that lack subjects.

Accordingly, rg accounts attribute the anomaly of passives of unaccusatives to a constraintthat bars multiple advancements to subject. This constraint is stated informally in (6).

(6) The 1-Advancement Exclusiveness Law (1aex) (cf. Perlmutter and Postal (1984:84))There is at most one advancement to 1 (subject) in a single clause.

4

Page 5: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

The 1aex rests on the assumption that the final subjects of initially unaccusative verbs arepromoted by a object-to-subject advancement. Given the Motivated Chômage Law, these derivedsubjects can only be demoted by the advancement of another argument to subject. However, theadvancement of any other argument violates the1aex. Hence the passive rule cannot apply.

Whatever the merits of this solution withinrg, it is clear that the need for a dedicated constraintlike the1aex must be counted among the descriptive costs of a promotional theory of the passive.A subject-oriented view of the passive can directly exploit the difference in argument structurebetween unergative and unaccusative verbs. If the passive rule demotes initial subjects, it will notapply to verbs that lack initial subjects. Since the absence of an initial subject is precisely whatdefines unaccusatives as a class, it follows that they should never passivize. A subject-orientedrule is thus intrinsically sensitive to the variation in argument structure posited by theuh.

Passivization may yield a subjectless output, if it applies to an intransitive verb, or to a transitiveverb that governs a non-structural or ‘inherent’ case. However, the resulting subjectlessness is aside-effect of subject demotion, and not directly specified by the passive rule itself. Furthermore,although it is never necessary to express the demoted agent in a passive, it is usually possible.

2.2 Impersonal actives

Whereas subjectlessness is a purely contingent property of passive constructions, it is a definingproperty of impersonals. This fundamental difference has formal as well as interpretive reflexes.Formally, there can be no promotion to subject in a subjectless construction. Hence impersonalsof transitive verbs may retain structural objects; depending on the case conventions of a language,these objects may occur in objective cases, such as accusative or partitive. The unrealized subjectof an impersonal construction also receives a distinctive interpretation, sometimes referred to interms such as ‘active indefinite’ or ‘ambipersonal suppressive’ (Tommola 1997).

From a formal and interpretative perspective, impersonals pattern with synthetic verb formsthat incorporate a subject argument, except that the suppressed subject of an impersonal receivesan indefinite interpretation. Functionally, however, impersonals may fill the communicativeniche occupied by passives in other languages. In traditional terms, impersonal constructionsare thus ‘passive in meaning or use’ but not ‘passive in form’. It is clearly the second sense of‘passive’ that is principally relevant for the formal analysis of impersonal constructions.

This is of course not to deny the legitimacy or usefulness of functional classifications. Thepoint is just that theformal constraints on passive constructions must be stated and evaluatedwith reference toformally passive constructions. The formal properties of constructions thatachieve the same communicative goal are not relevant, and subsuming them under a generalizednotion of the passive merely defines a formally heterogeneous macro-construction.

Applying a consistent distinction between formal and functional notions of ‘passive’ clarifiesthe impersonal character of the subjectless forms in a range of languages. The ‘impersonalpassives’ described in Polish, Ukrainian, North Russian and Lithuanian introduce a confoundingfactor, as the form of these constructions betrays a historical connection with the passive system.It is thus helpful to begin with the morphotactically distinctive patterns in Estonian and Welsh.

Furthermore, since the interpretation of impersonal forms often correlates significantly withtheir formal properties, it is useful to review descriptions of their basic meaning and use — fromboth the specialist and pedagogical literatures — in addition to examples that illustrate theirform and distribution. In many cases, one finds a convergence of opinion between specialist andpedagogical sources that contrasts with the position assumed within theoretical studies.

5

Page 6: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

2.2.1 Estonian

The use of impersonal forms or an impersonal ‘voice’is a distinctive feature of Finno-Ugric. Somedescriptions explicitly contrast impersonals with passives, as Viitso 1998 does in distinguishingthe impersonal voice reconstructed for proto-Fennic from the passive innovated in Estonian.

Corresponding to the six forms that make up the personal voice there was anim-personalvoice in proto-Fennic. … Such impersonalization is still possible in alltenses and moods in all Fennic languages except Livonian … Võru Estonian alsohas apassive: a transitive clause with the predicate verb in a personal form of thepresent or imperfect indicative can be made passive by transforming the object into asubject and replacing the active predicate verb with the corresponding passive verb.(Viitso 1998:112)

Other descriptions recognize a broad class of ‘passive’ forms, though the discussion of forms,as well as the glosses assigned to them, often bring out an overt personal/impersonal opposition.This opposition is clearly expressed in the two descriptions of Estonian passives below.

In Estonian, the passive voice (umbiskuline tegumood) is an impersonal verb form.That is, the agent responsible for the action of the verb is usually unknown orgeneralized. It can be translated into English either by using the passive voice orthe indefinite ‘one’ or ‘they’. (Tuldava 1994:272)

In Estonianvoice refers to whether the subject or agent of an action is known orunknown. If the subject is explicit in the context then personal forms of the verb areused. The impersonal forms are most similar to Frenchonas inon chante(someonesings) or Germanmanas inman singt. (Mürk 1997:21)

The references to the impersonal pronounsman, onandtheyhighlight the indefinite and humanreference implied by impersonal forms in Estonian. This interpretation is reflected in the glossesassigned to the Estonian examples in (7). In (7a), the subject is suppressed, and the object occursin the partitive case, which here marks the ‘partial object’ of an uncompleted action.

(7) a. Siinhere

ehitataksebuild.pres.imp

uutnew.part

maja.house.part

‘Here they are building a new house.’

b. Tullaksecome.pres.imp

jaand

minnakse.go.pres.imp

‘They [People] come and go.’ (Tuldava 1994:273)

As example (7b) indicates, impersonals may be freely formed from intransitive verbs in Esto-nian. Furthermore, as Torn 2001 shows, impersonal forms of canonical unaccusative verbs arealso possible, provided that the original verb can be construed as having a human subject.

(8) a. Pärastafter

reisitrip

jäädaksestay.pres.imp

hotelli.hotel.part

‘After the trip one stays at the hotel.’

b. Pärastafter

suurtbig

söömisteating

kaalutakseweigh-pres.imp

niiso

mõnigiseveral

kilokilo

rohkem.more

‘After a big feast, one weighs several kilos more.’ (Torn 2001)

6

Page 7: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

It is highly unnatural to specify the suppressed indefinite subject by means of an obliqueagentive phrase in Estonian or Finnish, even though both languages allow agents to be expressedin the genitive case in other constructions. The restriction in Estonian is stated in a scholarlyidiom by Matthews 1955, and in a more direct pedagogical style by Moseley 1994:

the Estonian construction withpoolt is not normally found associated with a passiveverb, and the presence of the ergative syntagma with this postposition would beimmediately felt as intrusive, because the Estonian verb is impersonal here. It is theimpersonality of the Estonian passive verb which is its most characteristic feature.Once a subject is introduced however, the verb is placed in a position of dependenceand must perforce assume an active and congruent form. (Matthews 1955:370)

But it would be more accurate to call this verb form impersonal rather than passive,as it has a slightly different usage that in English. An important difference is that itcannot take an agent … in Estonian; instead we use the normal active form of theverb and just change the word order … (Moseley 1994:129)

Tuldava (1994:273) concedes that thepoolt-construction ‘is not often used in Estonian’, butthen adds that ‘[o]ccasionally the postpositionpoolt ‘on the part of’ is used to form a construc-tion … with the agent in the genitive case’. However, Torn 2001 concurs with Matthews’ andMoseley’s assessment, suggesting that native speakers do not readily accept examples like (9).

(9) ?TartuTartu

ülikooluniversity

asuta-tifound-past.imp

GustavGustav

IIII

Adolf-iAdolf-gen

poolt.by

‘Tartu University was founded by Gustav II Adolf.’ (Tuldava 1994:273)

The foreign character of ‘poolt-phrases’ is likewise stressed by Nemvalts 1998.2

Passiivnepoolt-tarind on üks indoeuropism, mis üha rohkem ja rohkem laiutab kakodumaises eesti keeles. (Nemvalts 1998:63)

2.2.2 Welsh

Impersonal constructions are also characteristic of the Celtic languages. Personal passives inCeltic are, like periphrastic perfects and progressives, formed with verbal nouns. In contrast, theclass of forms that are often classified as impersonal passives in theoretical studies (e.g., Awbery1976, Comrie 1977, Perlmutter and Postal 1984, Noonan 1994) are synthetic, containing averbal exponent that takes the place of the regular personal endings. Descriptive grammarscharacteristically distinguish the impersonal or ‘autonomous’ forms in Celtic from the passiveconstruction. The following descriptions of Welsh autonomous forms are particularly explicit:

Although these two forms [in-id and-wyd, jpb] are sometimes listed as ‘passive’,they are properly referred to as autonomous or impersonal, since they are not strictlyspeaking passive in meaning (note that they can be formed for all verbs, includingintransitives likecomeor go that have no passive). (King 1993:220)

The impersonal forms that select an object are active grammatically, and since theverb in Welsh has no form corresponding exactly to the passive construction inEnglish they are also used to express the passive voice in Welsh. (Thorn 1993:314)

2‘The passivepoolt-construction is one of the Indo-Europeanisms that is spreading more and more in the homelandof the Estonian language’.

7

Page 8: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

As in Estonian, impersonal forms of transitive verbs in Welsh preserve a single, structuralobject. Since there is no case in Welsh, the object status ofdynis not obvious in (10a). However,as Comrie 1977 shows, a pronominal argument is realized by the object clitici in (10b).

(10) a. Lladdwydkill- aut

dynman

(ganby

draig).dragon

‘A man was killed (by a dragon).’

b. Fe’iPt+him

lladdwydkill- aut

(ganby

draig).dragon

‘He was killed (by a dragon).’

Unlike Estonian, Welsh, like other Celtic languages, allow a suppressed agent to be expressedby the same type of oblique that represents agents in personal passives. In Welsh, such obliquesare realized as prepositional phrases headed bygan, as the examples in (10) illustrate. This patterndemonstrates that the prohibition against oblique agent phrases cannot be regarded as a propertythat defines impersonal constructions, but rather one that is characteristic of impersonals.

2.3 ‘Impersonal passives’ in Balto-Slavic

The difficulties of classification presented by impersonal constructions are particularly acute inBalto-Slavic, in which impersonalno/toforms show a formal affinity with passive participles.

2.3.1 Passive and impersonal voice in Slavic

Polish provides a good point of departure, as it is transitional between the synthetic patterns inEstonian and Welsh and the more participial constructions in Ukrainian and North Russian. Theindefinite subject of ano/to form in Polish is again interpreted as personal but unspecified, andindeed as human. These interpretations are explicitly specified in the descriptions below.

In contemporary Polish, clauses with theno/to participle … are unanimously re-garded by Polish linguists as active impersonal … [with an] implied human subject[that] typically excludes the speaker. (Siewierska 1988:271)

The subject position is also eliminated in a second construction, which Polish sharesonly with Ukrainian. The construction is active (that is, a direct object or othergoverned case is possible) even though the verbal form is related to the passiveparticiple and the most natural English translation is usually a Passive construction… Although not all verbs have this form, it occurs with many verbs that do nototherwise form a passive participle (for example, intransitive and ‘reflexive’ verbs).The construction has the value of past tense and the unspecified subject (singular orplural) is understood to be human and indefinite. (Rothstein 1993:713)

The contrast between the personal passive in (11a) and theno/toconstruction in (11b) highlightthe impersonal character of the latter. Although-o historically marks the neuter singular formof passive participles,no/to forms do not synchronically exhibit the form or distribution ofparticiples. Forms likeczytanoare interpreted as finite, indeed as past, and may not combinewith any form of the passive auxiliarybyc. As in Estonian, the suppressed agent in an impersonal

8

Page 9: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

construction may not be expressed by an oblique phrase. As in Estonian andWelsh, the remainingargument in (11b) is an object, as indicated by the fact that it occurs in the accusative case.3

(11) a. Gazetanewspaper.fem.nom

byławas.fem

czytanaread.fem.sg

przezby

dzieci.children.

‘The paper was read by children.’

b. Gazeteønewspaper.fem.acc

(*była/*było)was.fem.sg/was.neut.sg

czytanoread.past.imp

(*przezby

dzieci).children.

‘One/they read the paper (*by children).’

The Ukrainian examples in (12) exhibit a similar contrast. In the personal passive in (12a),the patient is realized as a nominative subject, which controls agreement on the auxiliary andparticiple. In contrast, the ‘no/to’ construction in (12b) again exhibits apparent subject demotionwithout concomitant promotion. The participlezbudovanoand auxiliarybuloboth occur in whatSobin 1985 glosses as the neuter singular, and are accompanied by the accusative objectcerkvu.In (12c), the suppressed agent of theno/toform is realised as an instrumental oblique.

(12) a. Cerkvachurch.fem.nom

bulawas.fem

zbudovanabuilt.fem.sg

vin

16401640

goc’i.year

‘The church was built in 1640.’

b. Cerkvuchurch.fem.acc

(bulo)was.neut

zbudovanobuilt.imp

vin

16401640

goc’i.year

‘There was built a church in 1640.’

c. Cerkvuchurch.fem.acc

(bulo)was.neut

zbudovanobuilt.imp

Lesevym.Lesiv.inst

‘There was built a church by Lesiv.’ (Sobin 1985:653–8)

Sobin 1985 suggests that the accusative object in (12b) and (12c) presents a problem fortheories of the passive that assume obligatory promotion of structural arguments. Yet the sourceof this problem is ultimately Sobin’s (1985) assumption thatno/toforms are synchronically neuterpassive participles. This claim is problematic in two respects. First, as in Polish, the ‘neuter’treatment ofno/to forms can only be regarded as a historical analysis. In modern Ukrainian,neuter singular participles are marked by-ne/-te(Babby 1989, Shevelov 1993), as (13) illustrates.

(13) Mistotown.neut.nom

bulowas.neut

zbudovanebuilt.neut.sg

vin

16401640

goc’i.year

‘The city was built/founded in 1640.’

The view thatno/toforms express a voice opposition also reflects an ultimately historical per-spective. As Shevelov 1963 points out,no/toforms are, synchronically, dedicated impersonals:

The predicative form ending in-no, -torepresents historically a fossilized, nominalform of the neuter gender of the passive past participle. (Shevelov 1963:139)

Shevelov 1963 identifies the residual formal similarity betweenno/to forms and passive par-ticiples as the basis for the gradual introduction of past auxiliaries and instrumental agents intoimpersonal constructions. Nevertheless, the original past interpretation ofno/toforms is reflected

3The object status of accusativegazetuin (11b) is confirmed by the fact that it alternates with genitive undernegation. See Kibort 2001 for a comprehensive discussion of passive and impersonal constructions in Polish.

9

Page 10: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

in the optionality of past auxiliarybuloand fact thatno/toforms may not occur with a non-pastauxiliary. The future tense of passive constructions is periphrastic, consisting of the present formof buti ‘be’ and a passive participle. This pattern is illustrated in (14a). Yet as (14b) indicates,no present form ofbuti is compatible with impersonal forms likezbudovano.4

(14) a. Cerkvachurch.fem.nom

budeis.3sg

zbudovana.built.fem.sg

‘The church will be built.’

b. * Cerkvuchurch.fem.acc

budeis.3sg

zbudovano.built.imp

‘There will be built a church.’

2.3.2 Voice, mood and category in Lithuanian

Modern Lithuanian contains a class of forms in-ta that are again standardly glossed as neuterpassive participles. Yet the properties of these forms not only contrast with those of passivesbut also vary in significant respects from the impersonal constructions described above. Theexamples in (15) illustrate the basic contrast between passive and impersonal constructions inLithuanian. The simple transitive clause in (15a) corresponds to two alternatives that are oftenglossed as ‘passive’: the personal passive in (15b), and the impersonal construction in (15c).

(15) a. T .evasfather.nom

kvieciainvites.3rd.pres

sveciùs.guests.masc.pl.nom (active)

‘Father invites guests.’

b. Sveciaıguests.masc.pl.nom

yràbe.3rd.pres

(t .evo)father.gen

kvieciamì.invited.masc.pl.nom (passive)

c. Sveciaıguests.masc.pl.nom

yràbe.3rd.pres

(t .evo)father.gen

kvieciama.invited.imp (impersonal)

‘Guests are invited by father.’ (Ambrazas 1997:277)

It is the ‘passive’ analysis of examples like (15c) that underlies Timberlake’s (1982) claimsthat Lithuanian contains ‘unaccusative’ and ‘double’ passives, in clear violation of the1aex.The relevant patterns are summarized in (16) and (17). On first blush, the alternation in (16)would appear to show that unaccusatives can passivize. There is no denying thatb

-uti ‘be’ is a

canonical unaccusative. However, it is another matter entirely whether the formb-uta in (16b) issynchronically neuter, singular, or, indeed, passive. Given that modern Lithuanian lacks a neutergender, the neuter classification ofb-uta in (16b) must again be regarded as a diachronic analysis.

(16) a. Jìshe.masc.sg.nom

bùvobe.3rd.past

kareıvis.soldier.masc.sg.nom

‘He was a soldier.’

b. Jõhe.masc.sg.gen

b-utabeen.imp

kareıvio.soldier.masc.sg.gen

‘(They say) he was a soldier.’ (Ambrazas 1997:662)

The description inAmbrazas 1997 highlights further respects in which this construction differsfrom canonical passives (or impersonals). One point, which sets (16b) apart fromanyattestedpassive construction, is the fact that the genitive oblique in this example cannot be omitted:

4Shevelov 1963 also suggests that instrumental agents are somewhat less felicitous in impersonal constructions.

10

Page 11: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

The agentive genitive is obligatory in these sentences, which distinguishes themfrom other passive constructions. (Ambrazas 1997:662)

More generally, as the sentence glosses in (16) clearly signify, there is no voice oppositionbetweenb-uta and the preteritebùvo. Instead,b-uta conveys a nonfactive or evidentialmood:

Passive constructions with the neuter participle and agentive genitive, especiallywithout an auxiliary, are used in the evidential meaning (of an action not observeddirectly, but inferred from its consequences, assumed or hearsay). In such cases, theagentive genitive is usually preposed to the verb, the word order of the respectiveactive constructions being retained … (Ambrazas 1997:282)

Cases of ‘double passives’, in which an auxiliary and participle both occur in theta form,similarly express a mood interpretation rather than a voice contrast, as the glosses in (17) indicate.

(17) a. Jìshe.masc.sg.nom

bùvobe.3rd.past

ìš.ejeøs.left.masc.sg.nom

‘He has gone out.’

b. Jõhim.masc.sg.gen

b-utabeen.imp

ìšeita.left.imp

‘(They say/Evidently) he has gone out.’ (Ambrazas 1997:284)

Timberlake 1982 classifies (16b) as a ‘double’ passive, on the grounds that both the auxiliaryand participle occur in the ‘passive’ form. At the very least, one could object that the notion ofmultiple passivization is inapplicable here, as the formal similarity merely represents concord.However, the detailed discussion of these constructions in Ambrazas 1997 further clarifies theirrole in Lithuanian. The formb-uta is used to express evidential mood when, as in (16b), itsnominal complement lacks a-ta form. In periphrastic verbal constructions like (17), the concordbetweenb-uta and the-ta form of a participle serves to reinforce an evidential interpretation:

If an active periphrastic verb form undergoes passivization (to express evidentialmeaning explicitly), both the auxiliary and the main verb assume the form of thepassive neuter participle … The active form of the auxiliary can be retained, butin this case the evidential meaning is less clear, cf.:Jõ bùvo ìšeita. (Ambrazas1997:284)

The relevant contrast in (17) thus again involves mood, rather than voice. There is a simpleexplanation for the use ofta forms to express mood distinctions. The evidential mood is princi-pally expressed by using participles without auxiliaries. However, the fact that present indicativeforms of auxiliaries may be omitted in periphrastic formations leads to potential ambiguity:

The main formal difference between the oblique mood and the compound formsfor the indicative mood is the obligatory absence of the auxiliary verb. But sincethe auxiliary verb of the present tense of the indicative mood is sometimes omitted,three forms of the oblique mood [viz. the past, past imperfect passive and past perfectpassive,jpb] may formally coincide with those of the indicative mood and thereforethey may become ambiguous. (Ambrazas 1997:284)

The use ofta forms resolves this ambiguity, as such forms do not occur in periphrastic indicatives.The ta forms in (16b) and (17b) are thus impersonal predicates that express evidential mood.

Although the form of these elements records a historical connection to passive participles, they

11

Page 12: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

express no voice opposition. The dissociation ofta forms from the voice system is confirmed byother formal and interpretive contrasts. In addition to the nonfactive interpretation in (16b) and(17b),ta forms may be used to describe ‘an action causing surprise’ (Ambrazas 1997:281).

The contrast betweenb-uta andbùvoin (18) illustrates the ‘surprise’ interpretation ofta forms.

(18) a. Kàswho.nom

ciàhere

bùvo.be.3rd.past

b. Kienõwho.gen

ciàhere

b-uta.been.pass.neut.sg

‘Who was here?’ (Dambri unaset al.1966:263)

Dambri unaset al.(1966:662) elucidate this contrast by remarking that (18a) ‘has a more generalmeaning, while [(18b)] implies the surprise caused by the facts which only now were perceived’.

Another striking property ofta forms is their productivity. Predicates in-ta can be formednot only for all intransitive verbs, unergative and unaccusative, but also for certainnontransitiveverbs, notably weather verbs. Ambrazas (1997:282) remarks that ‘[s]ome impersonal verbs (e.g.lýti ‘rain’, snìgti ‘snow’ pustýti ‘drift (of snow)’) also have passive forms with the neuter pastparticiple.’ Mathiassen 1996 clarifies the meaning of these formations in the following terms:

Observe, finally, the opposition betweenpalyta‘it has rained’ andpalijo ‘it rained’,where the participle construction denotes aresult (state) as opposed topalijo whichfocusses on theprocess. (Mathiassen 1996:145)

The idea of a ‘passive’ form of a nontransitive predicate is clearly incongruous with the useof ‘passive’ to designate a morphosyntactic class. On the other hand, this notion is perfectlyintelligible if ‘passive’ is instead used to designate aform class. The reference in Ambrazas(1997:284) to a form that ‘undergoes passivization to express evidential meaning’overtly signalsthat Ambrazas 1997 is using the term ‘passive’ in precisely this, purely formal, sense. This usageis analogous to the use of terms like ‘past’ and ‘present’ to designate participles. Just as ‘past’participles are typically tenseless forms that share a base with the preterite series of a language,‘passive’ta forms are impersonals with a residual formal similarity to passive participles.

It is of course possible thatta forms may occur in a number of different subconstructions.However, there is no clear basis for treating any of these constructions formally as passives.Hence the observation thatta forms contravene the1aex does not bear in any direct way on theevaluation of universal claims about passive constructions. The formal passive in Lithuanian isrepresented solely by the personal pattern in (15b). This construction expresses a standard voice— rather than mood — opposition, and is subject to familiar syntactic constraints. For example,Ambrazas 1997 remarks that the class of unaccusative intransitives lack personal passives:

The passive voice of some verbs governing the accusative object of quantity (kainúoti‘cost’, sverti/svérti ‘weigh’, trùkti ‘last’, sukàti ‘turn (about age)’) is formed withneuter passive participles only. (Ambrazas 1997:280)

Thus, unlike the transitive unergative in (15), the transitive unaccusative (19a) has no correspond-ing personal passive, but only the impersonal construction in (19b).

(19) a. Vištàhen.sg.nom

sveriaweigh.3rd.pres

dùtwo

kilogramùs.kilograms.pl.acc

‘The chicken weighs two kilograms.’

12

Page 13: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

b. Vištoshen.sg.gen

sveriamaweigh.imp

dùtwo

kilogrãmai/kilogramùs.kilograms.pl.nom/pl.acc

‘The weight of the chicken is two kilograms.’ (Ambrazas 1997:280–1)

The factors considered above suggest thatta forms are, likeno/toforms in Polish and Ukrainian,not formally passive. However, the productivity and interpretation ofta forms also differssignificantly from those ofno/toforms in Slavic. The conservative position of Lithuanian withinIndo-European may provide a clue to these differences. It must first be acknowledged that theorigins of the Lithuanian impersonal construction are not entirely clear. Matthews (1955:362)considers Leskien’s claim that it reflects Polish influence (lit. anAnlehnung an einem polnischenSprachgebrauch(Leskien 1919:214)), but rejects this suggestion and concludes

that its source is not to be found in any one East European influence, and that theconstruction itself would appear to have developedin situ as a unique expansionof the scope of neuter passive participles in the East European area. (Matthews1955:371).

Schmalstieg (1987:183) likewise entertains Schwyzer’s proposal that the use of the agen-tive genitive with passive participles in Lithuanian derives from an antecedent possessive use(Schwyzer 1942:14). However, Schmalstieg 1987 contests this analysis, and suggests that theagentive genitive is instead a relic of an ancient ergative construction in Indo-European.

Nevertheless, both Matthews and Schmalstieg accept that the properties ofta forms betray anultimately nominal origin. Matthews (1955:356) regards ‘the neuter passive participle qualifiedby a genitive subject’ as ‘a formally possessive construction’. Schmalstieg likewise remarks:

According toAmbrazas 1979, 185–186, the use of participles in the nominal sentenceis based on their nominal characteristics. Nevertheless, the participle differs fromother adjectives by its verbal meaning, which enables it to express action or process innominal constructions. … The deeper one penetrates into the Proto-Indo-Europeanpast, the clearer the nominal morphology and the verbal semantics of the participlebecome. (Schmalstieg 1987:112)

The nominal character attributed tota forms recalls Shevelov’s (1963) observation that theUkrainian form inno/to is ‘historically a fossilized,nominalform’. Schmalstieg’s remarks aremeant to apply to participles as a class in Lithuanian. However they also suggest a basis ofdistinguishingta forms from passive participles and impersonalno/to forms. If ta forms are, asMatthews and Schmalstieg suggest elsewhere, the most conservative participles in Lithuanian,one might expect them to retain more residual nominal traits than the participial forms that havebeen fully integrated into the verbal system. Indeed,ta forms might be analyzed synchronicallyas nominal categories with only a ‘verbal semantics’, somewhat like the class of gerundivenominals in English. Such a nominal analysis accounts for the presence of genitive obliques, aswell as for the observation thatta forms occur with nominal-subcategorizingb-uta in (17b).

More significantly, a nominal analysis also accounts for the subjectless character ofta formsand for the lack of agreement with the logical subject. Gerundive nominals, verbal nouns, andother classes of nominalizations may retain the basic lexical semantics of a verb root. However,reflecting their nominal character, these categories do not preserve the tight syntactic bondthat finite verbs tend to establish with their syntactic subject. In particular, the logical subjectof nominalizations are, like other arguments of nominals, not generally obligatory. Concordwith logical subjects is likewise not typically represented in the form of a nominalization. The

13

Page 14: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

obligatory genitive subjectjõ ‘me’ in (16b) is thus not strictly required by theta form b-uta butsatisfies the demands of the agreement targetkareıvio‘soldier’ for an agreement controller.

In short, one can account for the impersonal character ofta forms if one treats them as residuallynominal forms that differ from true verbal participles in Lithuanian with respect to category, notvoice (or gender). A nominal analysis also accommodates the striking productivity ofta forms.The formation of gerundive nominals or verbal nouns is not, in general, subject to the sorts ofmorphosyntactic constraints that apply to passives or even impersonals. It is hard to make sense ofthe notion of passivizing or impersonalizing a nontransitive predicate, since there is no candidatefor demotion or suppression. However, there is no principled reason why a nontransitive verbshould not have a corresponding nominalization with a characteristic meaning, such as the resultinterpretation described by Mathiassen (1996:145).

3 Valence and valence-changing operations

The preceding section argues for the recognition of classes of impersonal constructions that areformally distinct from passives. Distinguishing impersonal and passive constructions not onlyyields a more accurate description of the construction inventories of individual languages, but alsocontributes to a clearer analysis of formal passive constructions. This is perhaps most obviouslytrue in the case ofrg, where an impersonal reclassification ofno/toandta forms vindicates anumber of relational laws, or at least the basic intuitions that underly them. However, the sameis ultimately true of other approaches. Although it is possible to incorporate a broad definitionof ‘passive’ constructions within a formal framework, this extension does not capture any usefulgeneralizations. Instead, by recasting a distinction between constructions as a contrast within aconstruction, formal accounts must be qualified to accommodate the resulting variation.

The core contrasts between passives and impersonals are summarized in (20) and (21).

(20) Passives

• Passivisation demotes the initial subject argument of a verb to an oblique.

• Only unergative verbs passivize because only unergatives specify initial subjects.

• Advancement is opportunistic in passives. The promotion of initial nonsubjectsis not stipulated by the passive rule, but reflects the same general constraints thatapply to basic predicates and require that some element be realized as the subject.

(21) Impersonals

• Impersonalization suppresses the syntactic realization of the final subject.

• Since final subjects neutralize the contrast between initial subject and nonsubjectterms, impersonalization may apply to unaccusative as well as to unergative verbs.

• Advancement is blocked, since there is of course no final subject to advance to.

The single most important difference concerns the determination of cause and effect. Passiviza-tion directly demotes an initial subject. This demotion indirectly permits the promotion of aneligible initial nonsubject, and otherwise results in a subjectless construction. Impersonalization,in contrast, defines a subjectless construction, irrespective of the initial argument structure.

The minimal prerequisites of the present account are applicable notions of ‘initial’ and ‘final’subjects, and appropriate notions of argument ‘demotion’ and ‘suppression’. Support for thesenotions can be found in most contemporary approaches. Subjects are invariably distinguished

14

Page 15: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

in some way from other subcategorized arguments. Virtually all lexicalist frameworks preservea Bloomfieldian contrast between stems and words. Hence initial subjects can, in principle, beidentified as stem-level subject arguments, and final subjects as word-level subject arguments.The demotion or suppression of subjects can likewise be defined in terms of the relation/function-changing rules ofrg (Perlmutter and Postal 1977) and early models oflfg (Bresnan 1982b), orin terms of the argument-permutation operations proposed inhpsg (Pollard and Sag 1994).

However, a formalism-dependent description of subjects or valence alternations essentiallyprecludes an general treatment of passives and impersonals. For the purposes of the presentstudy, it does not ultimately matter whether final subjects are assigned to a list-valuedsubj

feature, as inhpsg, or represented by a category-valuedsubj feature, as inlfg. Nor does itmatter where exactlysubj attributes occur in the feature geometry of a verb, or what precisestructure is imposed onsubj values. All that is important is that a formalism has some means ofidentifying the subject argument selected by a lexical verb. Similar remarks apply to differencesin rule or constraint inventories. It is again desirable that the characterization of notions like‘demotion’ and ‘suppression’ should not rely on implementation details of any given formalism.

3.1 Lexical Mapping Theory

The relational terminology developed inrg permits a suitably high-level description of valenceand valence-changing operations. Therg distinction between core ortermrelations and periph-eral orobliquedependents is of particular relevance. However, in order to express generalizationsover natural classes of arguments, or describe the morphosyntactic properties of arguments, itis useful to work with feature decompositions of the atomic relations proposed withinrg. Lex-ical Mapping Theory (lmt) fills this gap by specifying a general decomposition of argumentstructure. The feature classification from Bresnan and Kanerva 1989 is repeated in (22a).

(22) Canonical features and argument classifications (Bresnan and Kanerva 1989)a.

[−r] [+r]

[−o] subj oblθ

[+o] obj objθ

b. ag pt/th

| |

[−o] [−r]

c. θ

|

[−r]

Although the feature names [±r(estricted)] and [±o(bjective)] follow lfg conventions,these features impose a standard classification. The features [r] and [o] correspond transparentlyto the ‘semantic’ case featuresmarginal anddirectional in Jakobson 1936, 1958.5 Hence theanalyses in (22a) can be embedded into a general analysis of morphological case that governscompatibility with syntactic arguments. The feature[r] also distinguishesrg terms from obliquerelations, and mirrors the contrast between ‘structural’ and ‘inherent’ cases in transformationalaccounts (Chomsky 1981). The feature[o] cross-cuts terms and obliques, distinguishing ob-jects from non-objects. Subjects are thus canonically non-objective terms, while objects arecanonically objective terms. Although thematically restricted objects play no role in the presentaccount, the agentive obliqueoblag (which instantiatesoblθ ) corresponds to chômeurs inrg.

The features in (22a) are assigned to lexical forms in accordance with intrinsic role classifi-cations. The basicagentandpatient/themeclassifications in (22b) determine the underspecifiedargument structures in (23). The agentive arguments of unergativeski andbite are classified as[−o], while the patient argument ofbiteand the theme argument offreezeare marked as[−r].

5See also the feature systems proposed in Bierwisch 1967 and Wunderlich 1995.

15

Page 16: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

(23) Underspecified unergative and unaccusative argument structures

a. ski 〈 ag 〉

|

[−o]

b. bite 〈 ag pt 〉

| |

[−o] [−r]

c. freeze〈 th 〉

|

[−r]

The mapping from intrinsic features to grammatical functions is determined by a varied classof conditions. The highest orprimaryargument, represented ‘θ ’, has a distinguished status. Thedefault constraint (22c) assigns a[−r] feature to a primary argument. The mapping principle in(24a) also follows Bresnan and Zaenen (1990:51) in associating thesubj function preferentiallywith a primary argument. If none is available, thesubj is mapped by default to the highest term.

(24) Mapping principles (cf. Bresnan and Zaenen (1990:51))a. subj is mapped to a[−o] primary argument; otherwise to an[−r] argument.

b. Other arguments are associated to the lowest compatible function in themh in (25).

The principle in (24b) associates arguments with nonsubject functions. This association isbased on the relative ordering determined in (25) by treating positive values in (22a) as markedand negative values as unmarked. As in argument structures, leftmost elements are ‘highest’.

(25) Markedness hierarchy (mh) (Bresnan 2001:309)subj ≻ obj, oblθ ≻ objθ(Negatively specified features are unmarked.)

The mapping principles, along with the other other rules and constraints that apply to argumentstructures, are subject to a generalmonotonicityrequirement, meaning that they can add but notchange features. This ensures that agentive[−o] arguments can only be mapped to asubj orchômeur (oblag) function, while[−r] patients and themes can only be mapped tosubj or obj

functions. Monotonicity also avoidsrg-style multistratalism, as alternations between intrinsicand surface properties can be described with reference to a single level of argument structure.6

The effect of these conditions is illustrated in (26). The default in (22c) assigns a[−r] valueto the primary arguments ofskisandbites, and applies redundantly tofreeze. The first clause in(24a) associates the primary arguments ofskisandbiteswith thesubj function, while the defaultclause in (24a) applies tofreeze. Principle (24b) associates the[−r] argument ofbitesto theobj.

(26) Final unergative and unaccusative argument structures

intrinsic:default:final:

a. ski 〈 ag 〉

|

[−o]

[−r]

subj

b. bite 〈 ag pt 〉

| |

[−o] [−r]

[−r]

subj obj

c. freeze〈 th 〉

|

[−r]

subj

Since the features associated with objects in (22a) differ from those of subjects, the monotonicdevices provided bylmt do not permit an object to advance to subject in unaccusatives orpassives. However, as Bresnan and Zaenen 1990 argue, it is not necessary to treat the argument ofan unaccusative verb likefreezeas an object at any level of analysis. The unergative-unaccusativesplit ultimately involves a subject-nonsubject contrast: unergatives have an initial argument thatqualifies as an initial subject, whereas unaccusatives do not. This distinction can be captured bycharacterising the nonsubject argument of an unaccusative just as an underspecified[−r] term.

6It is by no means certain that all valence-changing operations are monotonic. However, as Ackerman 1992 notes,this assumption is most plausible for morphosyntactic operations, such as passivization or locative inversion, andmore clearly problematic for ‘morphosemantic’ (Sadler and Spencer 1998) operations like causativization.

16

Page 17: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

3.2 Lexical Argument Structure

The analyses in (26) illustrate how underspecification here avoids the need for function-changingoperations. The argument of unaccusativefreezein (26) receives the same intrinsic analysis asthe nonsubject argument ofbite. Yet the mapping principle in (24a) identifies each of the primaryarguments in (26) as final subjects. This alternation between intrinsic and final classifications in(26) preserves therg intuition that unergatives and unaccusatives exhibit an initial contrast thatis neutralized ‘on the surface’, though without identifying any final subject as an intrinsic object.

It is worth emphasising the essentially generic character of the analyses in (26). The structuresin (26) can be interpreted as classifying semantic roles, as proposed inlfg. However, littlein the present account depends on the semantic content of particular roles, as relative order istheir only significant property. To emphasize the placeholder status of these arguments, thefollowing analyses follow laterlmt accounts in using the neutral variablesx, y andz ratherthan mnemonic role labels likeag, pt or th. The features in (26) can then be interpreted asunderspecified descriptions of syntacticarg-s elements inhpsg, or as feature decompositions ofthe grammatical function inventory inrg. On either of these alternatives, the features[+o] and[+ r] implicitly characterize the marked grammatical relationsobj(ect) andobl(ique) so thatthe property of being thematically restricted is independently represented by thematic subscripts.

This point can be made more clearly by reconstructing therg contrast between terms andobliques in terms of an essentially privative opposition between[+ r]elements and other elements.

(27) a. A[+ r] element is anoblique.

b. All other elements areterms.

By classifying as terms elements that are negatively specified forr, as well as those that areunspecified forr, we avoid the need for redundancy rules that ‘fill in’[−r] values. This revisioninverts the role of markedness in the mapping principle in (28a). Since only[+ r] elements can bemapped to oblique relations, nonsubject terms are mapped to the highest grammatical function.

(28) Revised mapping principles

a. subj is mapped to a[−o] primary term; otherwise to a term argument.

b. Other arguments are associated to thehighestcompatible function in themh in (25).

A virtue of this revision is that it avoids associating a[−r] value with the intrinsic nonsubjectterms that may surface as obliques in intransitivized unaccusative clauses. By leaving[r] under-specified, we permit these elements to be associated to[+ r, −o] oblθ relations in Ukrainian.

In sum, therg constructs ‘initial’and ‘final’ subject correspond, respectively, to the notion of a‘ [−o] primary term’and an ‘argument mapped to thesubj function’. It is important to recognizethat these notions are nonderivational and are defined in terms of the lexical argument structureof a verb or other predicate. In particular, final subjects, like the other grammatical relations inthe analyses above, do not specify surface syntactic relations, but merely specify the interfaceto the syntax. The formal interpretation of these notions depends, in a relatively transparentfashion, on the way that ‘surface’ grammatical relations are represented in a particular approach.

Of course, initial and final subjects map directly onto initial and final ‘1’ relations inrg. Withinhpsg, initial and final subjects correspond to the singleton value of thesubj list of a verb stemand word, respectively. The notion of a final object corresponds to an f-structureobj in lfg, afinal ‘2’ in rg, and the first element of a verb’scomps list in hpsg. The demoted obliqueoblθ

again has a direct interpretation inlfg and corresponds to a ‘Chô(meur)’ relation inrg and toan element of the ‘dependents’ (deps) list in current versions ofhpsg (Boumaet al.2001).

17

Page 18: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

3.3 Demotion and suppression

The difference between input-sensitve passivization and output-sensitive impersonalization isreflected in the rules in (29) and (30). Given that subjects and chômeurs differ solely in theirvalues for[r], the passive lexical rule can, as Kibort 2001 proposes, be formulated as in (29).

(29) Passive lexical rule (plr)

The initial subject is demoted to an oblique relation.(A [+ r] value is added to a[−o] primary argument.)

Restoring a relation-sensitive passive rule suggests a reinterpretation of the argument sup-pressing rule in Bresnan and Kanerva 1989 as a generalization of a passive and impersonal rule.The dedicated impersonal rule in (30) suppresses, or prevents the syntagmatic realization of, thehighest argument. Since this argument will be mapped to the subject function, the ‘final’ subjectis thereby suppressed. There is no standard description of this operation, since most theories donot recognize the existence of subjectless constructions. Thelmt notation in (30) conveys theintended effect, though the precise implementation will again vary across approaches.7

(30) Impersonal lexical rule (ilr)

The subject function is syntagmatically suppressed, notated:(subj)

Sections 3.3.1 and 3.3.2 now concisely illustrate the application of these rules.

3.3.1 Passives

Theplr applies to the unergative English stems in (23), defining the analyses in (31). The analysisof bittenin (31) shows how theplr determines the argument structure of personal passives. Theagentive argument, which would otherwise be realized as the subject, is demoted to an oblique.This patient is the sole remaining term, and consequently realized as the subject.

(31) Passivization of initially unergative verbs in English

initial:plr:final:

skied 〈 x 〉

|

[−o]

[+r]

oblθ

bitten 〈 x y 〉

| |

[−o] [ ]

[+r]

oblθ subj

As it stands, theplr appears to ‘overapply’, as it also defines the unacceptable subjectlesspassive form ‘skied’ in (31). Overapplication could be prevented by restricting the application ofplr to transitive verb stems, as is often proposed in analyses of English. Although this might wellbe the correct solution for some languages, it would merely obscure a significant generalization inEnglish. The prohibition against subjectless constructions in English does not refer particularlyto passives, but applies to all verbs in the language. Presumably as a consequence of inflectionaldecay, English has personalized the various subjectless constructions found in other Germaniclanguages (and in earlier stages of English). It is thus unnecessary and indeed inappropriate torestrict theplr from producing an output that is independently illformed in the language.

7Within initial versions oflfg, suppression could be expressed as an operation deleting thesubj function of alexical form. Inhpsg, it would be expressed by a constraint assigning thesubj an empty list. Within the version ofrg known as Arc Pair Grammar (Johnson and Postal 1980), suppression would correspond to self-erasure.

18

Page 19: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

A general subject constraint is stated in (32). This constraint is again generic, though itsframework-specific interpretation is transparent. Withinlfg the fsc is interpreted as ‘[e]verylexical form must must have a subject (Bresnan and Kanerva 1989:28); withinhpsg it would beinterpreted as ‘a predicate’s feature structure must contain a non-emptysubj list; within rg it isinterpreted as ‘every basic clause contains a final-stratum 1’ (Perlmutter and Postal 1983b:100).

(32) Parochial final subject condition (fsc)

An argument structure must contain asubj in English.

Thefsc is general in the sense that it applies to the argument structures of all verbs, irrespectiveof voice. Thefsc is also, however, parochial, in that it applies to English, but not to languagesthat contain impersonal constructions. In this respect, thefsc differs from the constraints ofBresnan and Kanerva 1989 and Perlmutter and Postal 1983b cited above, as well as from the‘Extended Projection Principle’of Chomsky 1981 and most other subject-legislating constraints.

German differs from English in retaining a variety of impersonal constructions, basic andderived, suggesting that German lacks a counterpart of thefsc. Hence nothing bars theplr fromapplying to the initially unergativerauchen ‘smoke’ in (33) to yield the impersonal participlegeraucht. However, since unaccusativebleiben ‘remain’ lacks an initial subject, theplr doesnot apply, and the participlegebliebencan only receive an active interpretation.

(33) Passivization of intransitive verbs in German

initial:plr:final:

geraucht〈 x 〉

|

[−o]

[+r]

oblθ

geblieben〈 x 〉

|

[ ]

—subj

The contrast between English and German is less pronounced in transitive clauses. Bothlanguages permit personal passives of unergative transitives. However, neither language permitspassives of unaccusative transitives likedauern and its English counterpartlast in (34). Theanalysis of the second argument of verbs likelast or weigh is perhaps somewhat unclear.However, all that matters for present purposes is that it is not an initial subject, as it is surely not.

(34) Passivization of transitive unaccusatives

initial:plr:final:

lasted 〈 x y 〉

| |

[ ] [+ r]

— —subj objθ

gedauert 〈 x y 〉

| |

[ ] [+ r]

— —subj objθ

The analyses in (31)–(34) indicates how theplr discriminates between initially unergative andunaccusative clauses. A rule that demotes initial subjects avoids the need for a separate constraintto bar multiple advancements, analogous to the1aex. Unaccusatives will fail to passivize becausethey lack initial subjects. Passivized stems likewise fail to undergo further passivization, becausethey have been deprived of their initial subject, and any initial nonsubjects will, by definition,not qualify as initial subjects, though they may qualify as final subjects.

3.3.2 Impersonals

The basic analysis of impersonal constructions is likewise straightforward. Suppression of thesubj function suppresses the argument mapped to this function by the mapping principles. How-

19

Page 20: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

ever, suppression makes no reference to the properties of the argument, and, in particular, doesnot discriminate between unaccusative and unergative predicates. This point is illustrated by theanalysis assigned to the impersonal Estonian formstullakse‘one comes’jäädakse‘one stays’ in(35). Althoughtullakseis provisionally treated as unergative in (35a), nothing hinges on this.

(35) Impersonalization of intransitive and transitive verbs in Estonian

initial:final:

tullakse 〈 x 〉

|

[−o]

(subj)

jäädakse 〈 x y 〉

| |

[ ] [+ r]

(subj) objθ

The implied subject of bothtullakseandjäädaksereceives an indefinite human interpretation.Yet whereastullakseis syntactically nontransitive,jäädaksetakes a syntactic object.

The role of oblique agents within a language or construction will then reflect whether theypermit suppressed arguments to be reassociated to a thematically compatible oblique relation.This can be expressed formally by weakening the biuniqueness principle in (36) to (37).

(36) Function-argument biuniqueness (Bresnan and Zaenen 1990:51)

Each a-structure role must be associated with a unique function, and conversely.

The revised principle in (37) maintains a unique mapping from relations to arguments, butpermits variation in the inverse mapping. Without the parenthesized qualification in (37b), theseconditions are a more verbose variant of (36). This strict associate is appropriate for languageswhich, like Estonian, prohibit agentive obliques in impersonal constructions.

(37) Revised biuniqueness principle

a. Each grammatical function is associated to a unique argument.

b. Each argument is associated to a unique (realised) function.

The weaker constraint that is obtained by including the qualification in (37b) describes languageswhich, like Welsh or Ukrainian, permit agentive obliques. This reassociation is depicted in (38).

(38) Impersonalization of transitive verbs in Ukrainian

initial:final:

zbudovano 〈 x y 〉

| |

[−o] [ ]

/\ |

(subj) oblθ objθ

4 Conclusion

This paper suggests that the misanalysis of ‘passive’constructions stems from two general charac-teristics of current theoretical work. The first is a tendency to accord universal status to parochialtraits, particularly traits that are prominent in English. The idea that subjects are obligatory is aobvious case in point: while arguably true for English, this claim is certainly false for the manylanguages that contain impersonal constructions. The second characteristic is a narrow focus onsynchronic description, leading to the neglect or misanalysis of diachronic patterns. The passiveclassification of-no/-toconstructions in Balto-Slavic rests on precisely this sort of confound, asforms in-no/-toare historically passive but synchronically impersonal. More constructively, thispaper also suggests that the reclassification of misanalyzed passives leads to a more coherentconstruction inventory and permits significant improvements to lexicalist accounts.

20

Page 21: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

References

Ackerman, Farrell. 1992. Complex predicates and morpholexical relatedness: Locativeinversion in Hungarian. InLexical matters, ed. by Ivan A. Sag and Anna Szabolcsi, 55–83.Stanford: CSLI.

Ambrazas, Vytautas (ed.) 1997.Lithuanian Grammar. Vilnius: Baltos Lankos.

Awbery, G. M. 1976.The Syntax ofWelsh: A Transformational Study of the Passive. Cambridge:Cambridge University Press.

Babby, Leonard H. 1989. Subjectlessness, external subcategorization and the projectionprinciple. Zbornik za Filologiju i LinguvistikuXXXIII.7–40.

Baker, Mark, Kyle Johnson, andIan Roberts. 1989. Passive arguments raised.LinguisticInquiry 20.219–251.

Bierwisch, Manfred. 1967. Syntactic features in morphology: General problems of so-calledpronominal inflection in German. InTo Honour Roman Jakobson: Essays on Occasion ofhis Seventieth Birthday, 239–70. The Hague: Mouton.

Bouma, Gosse, Rob Malouf, andIvan A. Sag. 2001. Satisfying constraints on extraction andadjunction.Natural Language and Linguistic Theory.

Bresnan, Joan, andJonni M. Kanerva. 1989. Locative inversion in Chichewa: A case studyin factorization in grammar.Linguistic Inquiry20.1–50.

Bresnan, Joan, andAnnie Zaenen. 1990. Deep unaccusativity in LFG. InGrammaticalRelations: A Cross-Theoretical Perspective, ed. by Katarzyna Dziwirek, Patrick Farrell,and Errapel Mejías-Bikandi, 45–57. Stanford: CSLI.

Bresnan, Joan (ed.) 1982a.The Mental Representation of Grammatical Relations. Cambridge:MIT Press.

——. 1982b. The passive in lexical theory. In (Bresnan 1982a), 3–86.

——. 2001.Lexical-Functional Syntax. Oxford: Blackwell.

Chomsky, Noam. 1981.Lectures on Government and Binding. Dordrecht: Foris.

Comrie, Bernard. 1977. In defense of spontaneous demotion: The impersonal passive. InSyntax and Semantics 8: Grammatical Relations, 47–58. New York: Academic Press.

Dambriunas, Leonardas, Antanas Klimas, andWilliam R. Schmalstieg. 1966.Introduc-tion to Modern Lithuanian. Brooklyn, NY: Franciscan Fathers.

Jakobson, Roman. 1936. Beitrag zur allgemeinen Kasuslehre: Gesamtbedeutungen der russis-chen Kasus.Travaux du Cercle Linguistique de Prague VI240–299. Reprinted as ‘Contri-bution to the general theory of case: General Meanings of the Russian cases’ in Waugh andHalle 1984, 59–103.

——. 1958. Morphological observations on Slavic declension: The structure of Russian caseforms. InAmerican Contributions to the Fourth International Conference of Slavicists. TheHague: Mouton. Reprinted in Waugh and Halle 1984, 105–133.

21

Page 22: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

Johnson, David E., andPaul M. Postal. 1980.Arc Pair Grammar. Princeton, NJ: PrincetonUniversity Press.

Kibort, Anna, 2001.The passive and passive-like constructions: A contrastive study of Englishand Polish with particular reference to impersonal constructions. University of Cambridgedissertation.

King, Gareth. 1993.Modern Welsh: A Comprehensive Grammar. London: Routledge.

Leskien, A. 1919.Litauisches Lesebuch. Heidelberg: .

Mathiassen, Terje. 1996.A Short Grammar of Lithuanian. Columbus: Slavica.

Matthews, W. K. 1955. Lithuanian constructions with neuter passive participles.The SlavonicReview33.350–71.

Moseley, Christopher. 1994.Colloquial Estonian. London: Routledge.

Mürk, Harri William. 1997.A Handbook of Estonian: Nouns, Adjectives and Verbs. IndianaUniversity Uralic and Altaic Series, v. 163. Bloomington: Indiana University.

Nemvalts, Peep. 1998. Kas väliseesti keeles on märgata süntaktilist omapära? [Does one noticea syntactic character of Estonian abroad?]. InVäliseesti Keeles [Estonian abroad], 166–78.Tartu: Tartu Ülikooli Kirjastuse trükikoda.

Nerbonne, John A. 1982. Some passives not characterized by universal rules: Subjectlessimpersonals. InGrammatical Relations and Relational Grammar, ed. by Brian D. Joseph,Working Papers in Linguistics, No. 26. Columbus: Ohio State University.

Noonan, Michael. 1994. A tale of two passives in Irish. InVoice: Form and Function, ed. byBarbara Fox and Paul J. Hopper, 275–311. Amsterdam: Benjamins.

Perlmutter, David M., andPaul M. Postal. 1977. Toward a universal characterization ofpassivization. InProceedings of the third annual meeting of the Berkeley Linguistics Society.Reprinted inStudies in Relational Grammar I, pp. 3–29.

——, and ——. 1983a. The relational succession law. InStudies in Relational Grammar 1, ed.by David M. Perlmutter, 30–80. University of Chicago Press.

——, and ——. 1983b. Some proposed laws of basic clause structure. InStudies in RelationalGrammar 1, ed. by David M. Perlmutter, 81–128. University of Chicago Press.

——, and ——. 1984. The 1-Advancement Exclusiveness law. InStudies in Relational Grammar2, ed. by David M. Perlmutter and Carol G. Rosen, 81–125. University of Chicago Press.

Perlmutter, David M. 1978. Impersonal passives and the unaccusative hypothesis. InPro-ceedings of the fourth annual meeting of the Berkeley Linguistics Society.

Pollard, Carl, andIvan A. Sag. 1994. Head-driven Phrase Structure Grammar. Stanford:University of Chicago Press and CSLI.

Rothstein, Robert A. 1993. Polish. InThe Slavonic Languages, ed. by Bernard Comrie andGreville G. Corbett, 666–758. London: Routledge.

22

Page 23: Passives and impersonals - pdfs.semanticscholar.org · 2 Passive and impersonal constructions To clarify the basic contrast between passives and impersonals, it will be useful to

Sadler, Louisa, andAndrew Spencer. 1998. Morphology and argument structure. InHand-book of Morphology, ed. by Andrew Spencer and Arnold M. Zwicky, 206–36. Blackwell.

Schmalstieg, William R. 1987.A Lithuanian Historical Syntax. Columbus: Slavica.

Schwyzer, Eduard. 1942.Zum persönlichen Agens beim Passiv, besonders im Griechischen,volume Nr. 10. Berlin: Akademie Abhandlung.

Shevelov, George Yuri. 1963.The Syntax of Modern Literary Ukrainian. The Hague: Mouton.

——. 1993. Ukrainian. InThe Slavonic Languages, ed. by Bernard Comrie and Greville G.Corbett, 947–98. London: Routledge.

Siewierska, Anna. 1988. The passive in Slavic. InPassive and Voice, ed. by MasayoshiShibatani. Amsterdam: Benjamins.

Sobin, Nicholas. 1985. Case assignment in Ukrainian morphological passive constructions.Linguistic Inquiry16.649–62.

Thorn, David A. 1993.A Comprehensive Welsh Grammar. Oxford: Blackwell.

Timberlake, Alan. 1982. The impersonal passive in Lithuanian. InProceedings of the EighthAnnual Meeting of the Berkeley Linguistics Society.

Tommola, Hannu. 1997. O supresive i ob ambipersonale. InTrudy po russkoj i slavjanskojfilologii, Lingvistika, Novaja serija, 173–87. Tartu: Tartu Ülikooli Kirjastus.

Torn, Reeli. 2001. The status of the passive in English and Estonian. InRCEAL WorkingPapers in Linguistics. Cambridge: RCEAL.

Tuldava, Juhan. 1994.Estonian Textbook, volume 159 ofIndiana University Uralic and AltaicSeries. Bloomington: Indiana University.

Viitso, Tiit-Rein. 1998. Fennic. InThe Uralic Languages, ed. by Daniel Abondolo, 96–114.London: Routledge.

Wunderlich, Dieter. 1995. Minimalist morphology: The role of paradigms.Yearbook ofMorphology93–114.

23