Case and Grammatical Functions: The Icelandic Passive
Author(s): A. Zaenen, J. Maling and H. Thráinsson
Source: Natural Language & Linguistic Theory, Vol. 3, No. 4 (Nov., 1985), pp. 441-483
Published by: Springer
Stable URL: http://www.jstor.org/stable/4047609 .
Accessed: 15/09/2014 12:55
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
Springer is collaborating with JSTOR to digitize, preserve and extend access to Natural Language &Linguistic
Theory.
http://www.jstor.org
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
A. ZAENEN, J. MALING, AND H. THRAINSSON
CASE AND GRAMMATICAL
THE ICELANDIC
FUNCTIONS:
PASSIVE*
0. INTRODUCTION
Recent attempts to capture the universalcharacteristicsof passive have
moved away from relying on word order configurations.Ignoring proposals that have recourse to essentiallysemantic means of capturingthe
active-passive relation (such as Gazdar& Sag 1981, Dowty 1982), there
are two types of proposals that are currentlybeing debated. The first is
that passive morphology inhibits case marking, and hence that the
advancementof the object to subject in passive is essentiallyforced by
the violation of the constraintrequiringthat all NPs have case. This idea
is embodied in slightly different forms in various GB and GB-related
proposals (see e.g., den Besten 1981, Kayne 1981, Chomsky 1981,
Freidin & Babby 1982, 1983, and also Lieber 1979).1 The second
* An earlyversionof this paperwas read at the WinterMeetingof the LinguisticSocietyof
America in December, 1982. Later versions were presented at the Second Annual
Workshopon ScandinavianSyntax at Biskops-Arno (Sweden) and at the University of
Iceland in June, 1983. We are especially grateful to Anna Sigurbardottirand Johann G.
Johannssonfor countless hours of insightfulwork as native speakerinformants,supported
in part by NSF GrantNo. BNS80-14730. We thankthe followingpeople for commentsand
suggestions: J. Bresnan, R. Cooper, A. Harris, R. Jackendoff,J. Kornfilt,L. Levin, J.
Schindler,C. Watkins,and five anonymousreaders.
1 The case-dependentapproachis actually an adaptationof a traditionallymade observation, namely that oblique case is immune to passivization.Whereas the traditional
observation is correct for Icelandic under the proper interpretationof 'oblique', its
adaptationin the current GB literatureis not. To show that the two approachesare not
empiricallyequivalent,we can comparethe followingtwo quotes.
... I want to express in a generative frameworkthe traditionalinsight among studentsof
Germaniclanguagesas well as among grammariansof older phasesof Englishthat oblique
case is immunefrom passivization.(den Besten 1981, p. 67)
From most inflected languages we know that only those verbs which take an accusative
object can form a personalpassive.This seems to be the generalrule, thoughin Old Greek
also non-accusativeobjects can be made the subject of a passive sentence. In the Low
German (...) of my hometown (Krefeld) there is no distinction between dative and
accusative, or rather, there is only one objective case. So verbs which in High German
govem the dative (e.g., helfen, kundigen:help, give notice)come to be construedwith the
objective case in Lower German.In passive accordingly,we have personalconstructions:
Ich werde geholfen, er ist gekundigtworden.(Marchand1974, p. 98)
NaturalLanguage and LinguisticTheory3 (1985)1441-483. 0167-806X/85.10
1985 by D. Reidel PublishingCompany
?
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
442
A. ZAENEN, J. MALING, AND H. THRAINSSON
proposal relies on grammaticalfunctions, and is embodied in both RG
and LFG. In this paper we will argue on the basis of data from Icelandic
that the case dependent approach is not general enough, whereas an
approach based on grammaticalfunctions gives the correct result both
for Icelandic, and also for languages like German for which the casedependent approach was developed. The main evidence for our
argumentwill come from passives of the type illustratedin (1).
(1)
Honum var hjalpab.
him(D) was helped
We will argue that such sentences contain a non-nominativesubject.2
A second theoretical issue that we will address is whether the same
grammaticalfunction can have more than one realization in the same
clause. As we will see, the Icelandic data at first glance seems to lead to
the conclusion that this can happen, but virtually all current syntactic
theories propose a prohibitionagainst such double assignments.It is part
of the theta criterionin GB (Chomsky1981); it has been proposedas the
strataluniquenesslaw in RG (Perlmutter& Postal 1983, p. 92ff), and as
the functionaluniquenesslaw in LFG (Bresnan 1982, p. 163, Grimshaw
1982, p. 91). While there are empirical differences among these formulations,they lie beyond the scope of this paper. As we will see, a more
careful study of the Icelandic data shows that it is compatible with the
For Marchand,the notion 'subject'seems to coincide with that of nominativeNP; hence his
statementcan be given an interpretationthat allows it to apply correctlyto both German
and Icelandic. Den Besten's statement,however, does not hold for Icelandic, as we will
show. There are exceptions to Marchand'sstatement too, as he himself notes: Ancient
Greek, but also Irishuntil it lost its dative case (C. Watkins,personalcommunication)and
withinGermanic,Modem Faroese,as describedby Lockwood(1964), from whomwe quote
the followingsentences:
(i)
Teir fagnabuDepilsmunnum vael.
they receivedthe-depilsmen(D)well
(ii)
voru vel fagnabir.
Depilsmenn
the-Depilsmen(N)were well received(Nom.,pl.)
2 Before proceedingwith our analysishowever, we must remarkthat it would be all too
easy to empty these theories of their distinguishablecontent. For example, if case is used
not to refer to any morphologicallyobservablepropertiesof NPs, but simply to refer to
whatever behaves syntacticallylike an OBJect, then the case-markingtheory of passivizationbecomes a notationalvariant of the function-basedtheory. Similarly,if OBJ is
defined to be "whateverNP is assigned accusative case by the verb", then it will not be
distinguishablefrom the case-assignmenttheory. In whatfollows, then, we will assumethat
"case", and "function" are concepts with distinct morphologically and syntactically
observableproperties.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
443
functional uniquenessprinciple and the assumptionthat only DOs passivize.3
Our attention will be focussed on what we will call the periphrastic
passive, illustratedin a non-controversialinstance in (2):
(2)a.
fasta.
Logreglan t6k Siggu
the-police took Sigga(A) fast(A)
The police arrestedSigga.
b.
af logreglunni.
var tekin fost
Sigga
Sigga(N) was taken fast(N) by the-police(D)
Sigga was arrested by the police.4
Icelandic also has some morphologically'middle'forms in the suffix-st,
some of which have a passive meaning, as illustratedin (3):5
(3)a.
b.
fossinum.
fyrir
Ekkert
heyrist
nothing(N) can-be-heard on-account-of the-waterfall(D)
Urib
tyindist.
the-watch(N) got-lost
3 The most problematic case of double objects discussed in the literature is that of
Kinyarwanda(Gary & Keenan 1977); see also Dryer (1983) for argumentsthat these
are not really double objects. Zaenen (1983) argues that Kikuyu verbs which seemingly
allow the doubleobject constructionare better analyzedas allowingalternativeassignments
of grammaticalfunctions; the argumentis based on the interactionof Passive with the
prefixationof the direct object marker.
4 Agent af-phrasesare much less common in Icelandic passives than by-phrasesare in
English. A sentence like (i) will alwaysbe interpretedas meaning(ii) ratherthan as (iii):
(i)
(ii)
(iii)
Bokin var tekin af J6ni.
The book was taken from John.
The book was taken by John.
However, other sentences,such as (iv), are multiplyambiguousfor at least some speakers.
(iv)
Myndin var tekin af
J6ni.
The picturewas takenfrom/of/by John.
5 Example(3a) is taken from Einarsson(1945). Example(3b) shows that middleformation
does not preserve oblique case; the verb tyna 'to lose' takes dative objects in the active
voice, dative subjectsin the passive voice, but nominativesubjectsin the middlevoice. See
Zaenen & Maling (1984) for discussion of oblique case markingwith respect to other
quasi-productivelexical rules. Unaccusative verbs show that case preservation is not
restrictedto fully productive 'syntactic' rules like passive ('major' rule in the sense of
Wasow 1980). As far as we know, Helgi Bem6dusson(1982) was the firstto point out the
existence of unaccusativeverbs preservingaccusativecase on themes.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
444
A. ZAENEN, J. MALING, AND H. THRAINSSON
These will be ignored in this paper.6We assume that they are derived in
the lexicon and not in the syntax.
The general outline of the study is as follows: after giving some
background information about Icelandic syntax in section 1, we will
quickly review in section 2 the syntactic characteristicsof grammatical
subjects in Icelandic. We then discuss in section 3 the problem of which
NP's that are in post verbal position in unmarkedactive sentences can
show up as the subject of a passive sentence. In section 4, we give an
LFG account of passives in Icelandic, and outline the principles of
casemarking and assignment of grammaticalfunctions that obtain in
Icelandic. We will show that our account does not force us to the
conclusion that Icelandic has double objects. We will then brieflydiscuss
in section 5 whether German constructions like the one illustrated in (4)
should be treated in the same way as their superficially similar Icelandic
translations.
(4)
Ihm
wurde geholfen. (German)
Honum var hjalpab. (Icelandic)
him(D) was helped
We show that the two constructionsare syntacticallyvery differentin the
two languages,and then extend our account of case markingto cover the
German data.
1. SOME RELEVANT FEATURES OF ICELANDIC SYNTAX
Icelandic has the richest inflectional system of any modern Germanic
language. There are four cases (nominative, accusative, dative, genitive)
and three genders (masculine, feminine, neuter). To make it easier to
follow the examples, we give the paradigmfor personal pronounsin (5).
(5)
Singular
masc fem
N hann hun
A hann hana
D honum henni
G
hans
neut
pab
Pab
pvi
hennar Pess
masc
Peir
pa
Peim
Plural
fem
Twr
Oaer
Peim
neut
Pau
Pau
Peim
Peirra peirra Oeirra
6
The use of the -st passive is less widespreadin Icelandicthan it is e.g., in Swedish.This is
not meant to imply that there aren't some uses of the -st forms in Icelandic that are not
found in Swedish,e.g., the one exemplifiedunder(i):
(i)
Deir sogbust
ekki vilja
gera Pab.
they said-themselvesnot to-want to-do that
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
445
The basic word orderof Icelandic is SVO, both in main and in embedded
clauses. The main verb directly follows any auxiliaries,as in English but
unlike German.This is illustratedin (6).
(6)a.
b.
Eg hafbi seb hana.
I had seen her(A)
*Eg hafbi hana seb.
However, unlike English and like the other Germanic languages,
Icelandic has the Verb-Second Constraint.That is, the subject inverts
with the finite verb whenever another constituent is topicalized. This is
illustratedin (7).
(7)a.
b.
c.
d.
e.
i skfiffunni.
Olafur fann peysuna sina
Olaf(N) found sweater(A) his[+REFL] in the-drawer(D)
Peysunasina fsvnnOlafur i skuiffunni.
*Peysunasina Olafurfann i skuiffunni.
I skuffunnifann Olafurpeysuna sina.
*I skuffunniOlafurfann peysuna sina.
Thus far we have consideredonly simple transitiveverbs with accusative
objects. Icelandic, however, also has transitive verbs with dative and
genitive objects as illustratedin (8), and numerousditransitiveverbs (see
section 3.2 below).
(8)a.
b.
Eg hjalpabihonum.
I helped him(D)
DAT
GEN
Eg mun sakna hans.
I will miss him(G)
Icelandic has an impersonalpassive construction,illustratedin (9). When
no topicalizationtakes place in impersonalpassives, a dummy iaboccurs
in sentence-initial position, as shown in (9a). Whenever there is a
topicalized constituent,dummypab cannot occur in the sentence, either
before or after the finite verb. This fact is illustratedin (9b-d).
(9)a.
b.
c.
d.
Dab var dansab i gxer.
There was danced yesterday
I gaervar dansab.
*I gaervar Pab dansab.
*I gxrrPab var dansab.
When the auxiliaryis vera 'to be' or verba 'to be, become', the participle
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
446
A. ZAENEN, J. MALING, AND H. THRAINSSON
agrees in numberand gender with the nominativeNP in the clause.7This
is illustratedin (10).
(10)a.
til Islands.
6lafur var farinn
Olaf(N) was gone(masc.sg.) to Iceland(G)
b.
var farin
Sigga
til Islands.
Sigga(N) was gone(fem.sg.) to Iceland
c.
Barnib var farib
til Islands.
the-child was gone(neut.sg.) to Iceland
2. Two HYPOTHESES: SUBJECT VERSUS ToPic IN ICELANDIC
Given these basic features of Icelandic syntax, there are two possible
analysesof sentences like the ones in (11), where a DAT or GEN is in
the first position, and the tensed verb in second position is 3rd person
neuter, sg.
(11)a.
b.
Deim var hjailpab.
them(D) was helped
Hennarvar saknab.
her(G) was missed
The question is whether these forms are instances of topicalization in an
impersonalpassive construction, or personal passives in which the first
NP is a bona fide subject. These two analysesare sketched in (12a) and
(12b).
S
(12)a.
NP
Aux
VP
V
var
hjailpab
NP
feim
7 Strongverb participleslike fwa end in -inn, -in, -ib; one class of weak verbs (e.g. telja)
also has participlesof this form, but most weak verb participlesend in -abur,-ub, -b.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
b.
AND
GRAMMATICAL
FUNCTIONS
447
S
NP
Aux
VP
Deim
var
hjalpab
Under the first hypothesis, sentences like (11) are simply examples of
impersonal passives combined with the topicalization of the dative or
genitive NP; under this analysis,the sentences are theoreticallyunproblematic and uninteresting.Under the second hypothesis, however, the
dative NP is indeed a grammaticalsubject, and the construction therefore presents difficultiesfor those accounts of passive that are based on
case-marking.In what follows, we will argue that the correct analysisis
(12b).8 Before tackling that problem we will first review the arguments
that have been amassed during the last years to the effect that nominative casemarkingis not a necessaryprerequisite(nor a sufficientone) for
subjecthood in Icelandic.
Recent research has shown that there are several syntactic properties
that distinguishbetween topics and subjects in Icelandic (Andrews 1976,
1982a, 1982b, Thrainsson 1979, Maling 1980, Zaenen 1980). Andrews
(1976) was the first to point out that this distinction does not coincide
with that between nominative and non-nominative NPs, and that contraryto traditionalbelief, a sentence like (13) must be analysedas having
a simple subject-predicatestructurewith a dative subject and a nominative object.
(13)
Henni hefur alltaf
her(D) has
o6tt
Olafur leibinlegur.
always thought Olaf(N) boring(N)
Arguments for that view were further developed in Maling (1980),
Andrews (1982a), and Thrafinsson(1979), who summarizesmost of the
tests. We will refer to these non-nominativepreverbal NPs as "oblique
8
This destroysthe argumentgiven in section 4.2 of Maling & Zaenen (1978) againstthe
applicationof the that t-filterin Icelandic.However, the generalconclusionreachedin that
paperstill holds.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
448
A. ZAENEN, J. MALING, AND H. THRAINSSON
subjects". We illustrate here each property with an example of a
nominative subject and an example of oblique subject, and refer to the
relevant literaturefor furtherdiscussion.
2. 1. Raising
Only subjects can raise, as illustratedby the contrast between (14b and
d). The verb sakna "to miss" takes a nominativesubject and a genitive
object.
(14)a. Gubruin saknarHaraldar.
Gudrun(N)misses Harold(G)
b.
Eg taldi
Gubruinu i barnaskapminum sakna
to-miss
I believed Gudrun(A)in foolishness my
Haraldar.
Harold(G)
c.
Haraldar
saknar Gubruin.
Harold(G) misses Gudrun(N)
d. *EJgtaldi JHaraldar
IHarald J
sakna JGubru'n
IGubrtunu
I believedJHarold(G)l to-miss JGudrun(N)
1Harold(A)J
IGudrun(A)
Since i barnaskap minum 'in my foolishness' in (14b) is an adverbial
belonging to the matrix clause, its presence between the NP Gubruinu
and the infinitive complement sakna Haraldar 'to miss Harold' is good
evidence that the NP has been raised into matrix object position. (See
Thrafinsson1979, pp. 389-393, for furtherdiscussion.)Of course, predicate nominalscannot be raised,despite theirnominativecase marking,as
illustratedin (15).9
9 Non-NPs cannot in general undergoRaising, as shownin (i):
(i)
i barnaskapminumhafa veribmargt f6lk.
*Eg taldi a fundunum
I believed[at the-meeting]ppin foolishnessmy
to-have been many people
However,just when the PP can be analyzedas the grammaticalsubject,then it can undergo
Rtaising,as illustratedin (ii):
(ii)
Eg taldi undir ruiminui bamaskap minumvera gobanfelustab.
I believedunderthe-bed in foolishnessmy
to-be good hiding-place.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
(15)a.
AND GRAMMATICAL
FUNCTIONS
449
Olafur er bondi.
Olaf(N) is a-farmer
b.
Bondi er 1lafur.(Topicalization)
a-farmer is Olaf
c.
Olaf
Eg tel
vera bonda.
I believe Olaf(A) to-be a-farmer(A)
bonda
vera Olaf.
d. *Eg tel
I believe a-farmer(A) to-be Olaf(A)
However, so-called oblique subjects can raise, as shown in (16):
(16)
Eg tel
henni hafa
alltaf p6tt
I believe her(D) to-have always thought
Olafur leibinlegur.
Olaf(N) boring(N)
We can conclude that whatever the surface case-marking,all and only
grammatical subjects can raise. (See Andrews 1982b and Thra"insson
1979 for furtherdiscussion.)
2.2. Reflexivization
A second test is reflexivization.Many speakers of Icelandic allow only
grammatical subjects to be the antecedents of reflexive pronouns, or
more accurately, allow objects to be antecedents only if the reflexive
occurs in a predicate complement predicatedof that object (see Maling
1982 for discussion).For such speakerswe find the same sort of contrasts
with respect to Reflexivizationas we did with respect to Raising. We will
give the judgmentsfor those speakerswho allow only subject-controlled
reflexivization.(For other speakers the contrasts are more subtle: subjects control obligatoryreflexivization,whereasobjects control reflexives
only optionally; see Thraiinsson(1976) and Maling (1982) for further
discussion.)Boldface indicates intended coreference.
(17)a.
b.
Sigga
barbimig met duikkunnisinni/*hennar.
Sigga(N) hit me(A) with doll(D) her(*[-REFL])
Sigga hit me with her doll.
Eg barbiSiggu
meb duikkunnihennar/*sinnm.
I hit Sigga(A) with doll
her(*[+REFL])
I hit Sigga with her doll.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
450
A. ZAENEN, J. MALING, AND H. THRAINSSON
c.
Siggu barbieg meb duikkunnihennar/*simn.
Sigga I hit with her doll.
Again the non-nominativesubject can (in fact, must) control reflexivization, as shown in the following example:
(18)a.
b.
Henni Pykir brobir
sinn/*hennar leibinlegur.
her(D) thinks brother(N)her(*-REFL) boring
fagur. (Proverb)
fugl
Hverjum pykir sinn
everyone(D)thinks his[+REFL] bird(N) beautiful
2.3. Topicalizationand Subject-VerbInversion
Subjects appear immediatelyafter the finite verb if another constituent
has been preposed;whenever an object has been topicalized, no further
Topicalizationcan take place.
(l9)a.
Refinn
skaut Olafur meb pessaribyssu.
the-fox(A) shot Olaf(N) with this shotgun
Olafur.
b. *Meb Pessaribyssu skaut refinn
with this shotgun shot the-fox(A) Olaf(N)
Similarlyin direct questions the tensed verb is immediatelyfollowed by
the subject; hence direct questions and topicalizationsare incompatible
as shown in (20):
hafbi aldrei hjilpab Haraldi.
(20)a. Sigga
Sigga(N) had never helped Harold(D)
b. Hafbi Sigga
aldrei hjilpab Haraldi? (Yes-No-Question)
Had Sigga(N) never helped Harold(D)
c. *HafbiHaraldiSigga aldrei hjalpab?
d. Hvenar hafbi Sigga hjalpabHaraldi? (Wh-question)
When had Sigga helped Harold(D)
e. *Hvenxr hafbi HaraldiSigga hj'alpab?
f.
Haraldihafbi Sigga aldrei hjalpab.(Topicalization)
The same is true of oblique subjects. Unlike topics, oblique subjects can
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND GRAMMATICAL
FUNCTIONS
451
immediately follow the finite verb.10As shown in (21b), topicalization
can apply to sentences like (13), in which case the oblique subject inverts
with the finite verb.
(21)a.
Hefur henni alltaf t0tt
has
b.
Olafur leibinlegur?
she(D) always thought Olaf(N) boring(N)
Olafur hefur henni alltaf frtt
leibinlegur.
Olaf(N) has she(D) always thoughtboring(N)
c. *Hefur Olafur henni alltaf tott
leibinlegur?
has Olaf(N) she(D) always thoughtboring
2.4. Extraction
Most speakers of Icelandic do not generally allow Topicalization in
binding domains; in other words, they allow Topicalizationto occur in
embedded ab 'that'-clauses, but not under indirect questions, comparatives, relatives, etc. This is illustratedby the contrast in (22). (For
furtherdiscussionsee Zaenen 1980, Rognvaldsson 1984.)
(22)a.
i gaer.
Jon telur ab Maria hafi kysst Harald
Jon believes that Mary(N) has kissed Harold(A) yesterday
b.
ab Maria hafi kysst Harald?.
Hvenawrtelur Jon
when believesJon(N) that Mary(N) has kissed Harold
c.
Jon telur ab Haraldhafi Maria kysst i gxer.(Topicalization)
d. *Hvenxr telur Jon ab Harald hafi Maria kysst?
Whendoes Jon believethat Harold, Mary kissed?
Again, oblique subject NPs pattern here with the nominativesubjects as
"' Note that the superficially similar Ist dir kalt geworden? in German does not necessarily
have to be analyzed as an instance of Subject-Verb Inversion. According to German word
order, both (nonsubject) dir and kalt c;anoccur between the finite verb ist and the nonfinite
geworden even without this rule, as illustrated in (i).
(i)
Es ist dir
it
(ii)
kalt geworden.
(German)
is you(D) cold become
* ab er Pjr
kalt orbib.
* Pab er per orbib kalt.
(Icelandic)
It is clear that the Icelandic construction differs significantly from its German counterpart.
See section 5 below for an account of certain syntactic differences between German and
Icelandic.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
452
A. ZAENEN, J. MALING, AND H. THRAINSSON
opposed to topicalizedNPs, as shown in (23):
(23)a.
Jon telur ab henni hafi alltaf frtt
Jon believes that she(D) has always thought
Olafur leibinlegur.
Olaf(N) boring(N)
b.
ab henni hafi p06tt
Hvenar telur Jon
when believesJon(N) that she(D) has thought
Olafur leibinlegur?
Olaf boring
When does Jon believe that she considered Olaf boring?
c.
Jon telur ab Olafur hafi henni alltaf frtt leibinlegur. (Topi-.
calization)
Jon believes that Olaf she has alwaysfound boring.
at Olafur hafi henni
d. *Hvenar telur J6n
when believesJon(N) that Olaf(N) has she(D)
00ott
leibinlegur?
thoughtboring
Note in particularthat in (23d), the object Olafuris not patterninglike a
subject, despite its nominativecase-marking.
2.5. IndefiniteSubjectPostposing
When a subject is indefinite, it can be postposed by a rule of indefinite
NP postposing (plus pab-insertion).This cannot be done when a nonsubject is in first position.
(24)a.
Dab hefur pj6fur
stolib hjolinu
minu.
there has a-thief(N) stolen bicycle(D) mine(D)
b.
hefur jojfurinnstolib.
Hjoli
a-bicycle has the-thief stolen
c. *Dab hefur hjoli
pjofurinn stolib.
there has a-bicycle(D) the-thief(N) stolen
d. * Dabhefur hjoli stolib Ojofurinn.
Again oblique subjects pattern like nominative subjects rather than like
topicalized NPs.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
(25)a.
AND
FUNCTIONS
GRAMMATICAL
453
Olafur leibinlegur.
Pab hefur einhverjum frtt
there has someone(D) thought Olaf(N) boring(N)
Someone found Olaf boring.
b.
Olafur hefur einhverjuml06ttleibinlegur.
leibinlegur.
c. *Dab hefur Olafur einhverjum f6tt
there has Olaf(N) someone(D) thoughtboring
2.6. SubjectEllipsis
Another syntactic rule which distinguishessubjects from non-subjectsin
modern Icelandic is the kind of ellipsis which allows the subject of a
coordinated clause to be deleted under identity with the subject of the
preceding conjunct clause.11(The examples in (26c, d) are grammatical
only if the verb grafa is interpretedintransitivelyas 'to dig' rather than
transitivelyas 'to bury'.)
(26)a.
b.
c.
d.
fluttu llkib
og heir grofu pab.
Deir
they(N) moved the-corpseand they buriedit
Deirfluttu likib og _
grofu tab.
D
likib
Peir fluttu
og Peir grofu
Likib var flutt og peir grofu
e. *Likib hrxddi pa og
-
.
(Subj-Subj)
(Obj-Obj)
.
(Subj-Obj)
grofu pab.
The corpse scared them and
__
(Obj-Subj)
buried it.
Now consider the coordination of sentences, one of which takes a
nominative subject, the other an oblique subject. Consider the contrast
between (27a) and (27b). Oblique subjects can be deleted under identity
with nominativesubjects and vice versa, but objects cannot be deleted in
1' It is well-knownthat Old Norse allowed ratherfree ellipsis of NPs under identitywith
some NP in the previous sentence. Faarlund(1980, p. 70) observes that ellipsis in Old
Norse does not seem to be sensitive to either subjecthoodor nominativecase, but simply
deletes any unstressedrecoverablepronoun;one could argue that the basic conditionwas
not strict identitybut simplycoreference.However,in the modernScandinavianlanguages,
such ellipsis or coordinationis sensitive to grammaticalrelations.The examplesin (26) are
based on Faarlund'sNorwegianexamples.Note furtherthat sentences like (i) where both
subject and object are elided are grammatical,if somewhatstilted:
(i)
Ieir fluttu likib
og
they moved the-body and..
grofu
buried
The syntacticconstraintson this kind of 'conjunct-splitting'merit furtherinvestigation.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
454
A. ZAENEN,
J. MALING,
AND
H. THRAINSSON
this fashion,even when their case is nominative.This is illustratedby the
followingexamples:
(27)a. Hann segist vera duglegur, en
he(N) says-self to-be diligent, but..(D)
finnst
finds
of Oungt. (Subj-Subj)
verkefnib
the-homeworktoo hard.
He says he is diligent, but finds the homeworktoo hard.
b. *Hann segist vera duglegur,en mer
he(N) says-self to-be diligent, but I(D)
latur. (Subj-Adj)
find ..(N) lazy
He says he is diligent, but I find [him] lazy.
finnst
It is clear that ellipsis is sensitive to grammaticalrelationsratherthan to
morphologicalcase. In (27a), ellipsis is perfectly acceptable even though
the cases differ, because both NPs are subjects. However (27b) is
unacceptable, despite the fact that the coreferential NPs are both
nominativecase, because only one of the two is a grammaticalsubject.
(For further discussion see Eirikur Rognvaldsson 1982b, Bresnan &
Thrainsson 1984.)
2.7. InfinitiveComplements
Only subjects can be the target of EQUI-NP-Deletion, or be understood
as 'arbitrary'or anaphoricallycontrolled PROs in Icelandic.
(28)a. Eg vonast til ab fara heim.
I hope for to go home
b. Ab fara heim snemmaer ovenjulegt.
to go home early is unusual
Oblique subjects can also be the target of EQUI or be arbitraryPRO
subjectsof infinitives,as shown in (29). The verb vanta 'to lack' takes an
accusative subject and an accusative object.
(29)a.
Mig vantar peninga.
me(A) lacks money(A)
b. Eg vonast til
I hope for
ab vanta ekki peninga.
(A) to lack not money(A)
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
c.
AND
GRAMMATICAL
FUNCTIONS
455
Ab vanta peninga er alltof algengt.
to lack money is all-too common
In conclusion, there are (at least) seven different syntactic criteria to
distinguishsubjects from topics in active sentences in Icelandic, and we
see that these tests do not necessarilypick out the nominativeNP.
3.
WHICH
POSTVERBAL
NPs
CAN
PASSIVIZE
IN ICELANDIC?
3.1. ObliqueSubjectsin IcelandicPassives
We are now in a position to return to the question posed above: what is
the grammaticalfunction of the preverbalNPs in passive sentences like
those in (11) (repeatedhere for convenience):
(11)a.
b.
var hj'alpab.
Peim
them(D) was helped
Hennarvar saknab.
her(G) was missed
Note that regardlessof the person or numberof the initialNP, the verb is
always in the 3rd person singular. Verbs agree in person and number
with a nominativeargument;if there is no nominativeNP, then the verb
occurs in the 3rd person (neuter) singular, which we take to be the
unmarkedform.
We will now test the syntacticbehaviorof passivizedoblique NPs using
the same tests that we used above in section 2 for active sentences. As
can be seen from the examples in the following sections, the preverbal
dative NP in a sentence like (1 la) behaves in all respects like a grammatical subject.
3.1.1. Raising
A first such test is Subject-to-ObjectRaising.
(30)
hafa
verib hjalpabi profinu.
Eg tel
peim
I believe them(D) to-have been helped in the-exam
The preverbalNP undergoes Raising and keeps its dative case-marking.
3.1.2. Reflexivization
A second test is reflexivization.Many speakers of Icelandic allow only
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
456
A. ZAENEN, J. MALING, AND H. THRAINSSON
subjects to be the antecedents of reflexive pronounsor more accurately,
allow objects to be antecedents only if the reflexive occurs in a predicative complement predicated of that object (Maling 1982). For such
speakers, we find the same sort of contrastswith respect to Reflexivization as we did with respect to Raising. Note that the reflexive is
obligatoryin (31).
(31)
Honum var oft hj'alpabaf foreldrum sinum/*hans.
he(D) was often helped by parents his[+REFL]/
*[-REFL]
3.1.3. Subject-VerbInversion
Third, the preverbal dative can immediatelyfollow the finite verb, in
direct questions,or if another constituenthas been preposed.
(32)a.
b.
Var honum aldrei hjalpabaf foreldrumslnum?
was he(D) never helped by parents his
I pr6finu svar honum vist
hjalpab.
in the-exam was he(D) apparentlyhelped
3.1.4. Extraction
Fourth,many speakersof Icelandic do not generally allow Topicalization
in bindingdomains;in other words, they allow Topicalizationto occur in
embedded ab clauses ('that'-clauses),but not under indirect questions,
comparatives, relatives, etc. However, these oblique NPs can appear
preverballyin these contexts.
(33)a.
b.
Hann telur, ab henni hafi verib hjalpabi gaer.
he
believes that she(D) has been helped yesterday
Hvenaertelur hann ab henni hafi verib hjalpab?
when believeghe that she(D) has been helped
3.1.5. IndefiniteSubjectPostposing
Fifth, the preverbaldatives under considerationpatternlike grammatical
subjects ratherthan like topicalizedNPs with respect to indefinitesubject
postposing.
(34)
Dab hefur morgum studentum verib hjalpabi profinu.
there has many(D) students(D) been helped on the-exam
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
457
FUNCTIONS
3.1.6. SubjectEllipsis
Sixth, the preverbaldative can be deleted by subject ellipsis.
(35)
Hann segist
vera saklaus en
he(N) says-self to-be innocentbut __(D)
hefur vist
has apparently
verib hjalpab?
i profinu.
been helped on the-exam
3.1.7. InfinitiveComplements
Seventh, just like other grammaticalsubjects the passivizeddative object
can be the target of EQUI as illustrated in (36a); it can also be
understoodas an arbitraryPRO subject of an infinitive, as in (36b).
(36)a. Eg vonast til ab ver?a hjailpab.
I hope for to be
helped
b.
Ab vera hjailpabi pr6finu er oleyfilegt.
to be helped on the-exam is un-allowed.
Thus, by the above seven tests, the preverbaldative and genitive NPs in
(11) are clearly grammaticalsubjects and not topicalizedobjects, and the
sentences in (11) must be analyzed as ordinarypersonal passives. It is
clear that for Icelandic, oblique case is not immune to passivization.
3.2. Passive and DitransitiveVerbs
Icelandic has numerousditransitiveverbs. The possibilitiesare illustrated
in (37) (from Thriainsson1979, pp. 21-22).
(37)a.
Deir leyndu Olaf
sannleikanum.
they concealed[from]-Olaf(A) the-truth(D)
ACC-DAT
b. Jon bab mig bonar.
Jon asked me(A) a-favor(G)
ACC-GEN
c.
Eg sagbi per
soguna.
I told you(D) a-story(A)
DAT-ACC
d.
Olafur lofabi Mariu pessum hring.
Olaf(N) promisedMary(D) this(D) ring(D)
DAT-DAT
e.
ails
Maria oskab 61afi
gobs.
Mary wished Olaf(D) everything(G)good(G)
DAT-GEN
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
A. ZAENEN, J. MALING, AND H. THRAINSSON
458
We see that virtually any combination of two case-marked postverbal
NPs is possible, except that the first NP can never be genitive case.
There are several different types of ACC ACC verbs, but it is
debatable whether any of them are truly ditransitive. First, there are
transitiveverbs like kalia 'to name, call', as illustratedin (38).
(38)a. Deir k6llubuhana Kiddu.
they called her Kidda(A)
b. Hun var kollub Kidda.
she(N) was called Kidda(N)
But this type of verb is not a true ditransitiveverb, since the second
postverbalNP is clearly an (objective) complementrather than a verbal
object. This can be seen from the passive version in (38b) where the
'retained complement' becomes nominative, agreeing with the subject
NP it is predicatedof.
Second, there are also simple transitiveverbs like hoggva 'to hit' and
kyssa 'to kiss', which:'cantake an additionalargumentwhich might be
describedas a cognate object.
(39)a. hoggva einhvern banahogg
to-hit someone(A) a-deadly-blow(A)
b. sla' einhvern kinnhest
to-hit someone(A) a-blow-on-the-cheek(A)
c.
kyssa einhvern rembingskoss
to-kiss someone(A) a-deep- kiss(A)
Only the first postverbal NP in such constructions can passivize, as
illustrated in (40), in which case the second object not surprisingly
remainsaccusative.
banahogg.
(40)a. Skarphehinn hjo Draiin
hit
a-deadly-blow(A)
Thrain(A)
Skarphedin(N)
b. Drainn var hoggvinn banahogg.
a-deadly-blow(A)
Thrain(N) was hit
c. *Banahoggvarnh6ggvib Prain.
We will brieflydiscuss the analysisof this type of ACC ACC verb below
in section 4.4.
A third possible type of ACC ACC verb, pointed out to us by an
anonymous reviewer, consists of simple transitive verbs like keyra
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
459
'drive', which can optionally be followed by a second accusative NP
indicatingthe thematicrole of 'path'. Here, too, only the first postverbal
NP can passivize, and the second NP remainsaccusative.
(41)a. Hann keyrbibilinn
Pessa leib.
he drove the-car(A) this route(A)
b. Billinn
var keyrburpessa leib(A).
the-car(N) was driven this route(A)
var keyrb bilinn.
c. *Dessileib
this route(N) was driven the-car(A)
Similar facts obtain for the verb aka 'drive', which takes DAT rather
than ACC case on 'car'. We will discuss the analysisof these verbs, too,
in section 4.4.
To the extent that none of the above-mentioned types are truly
ditransitive,12there seem to be no triadic verbs in Icelandic where both
objects are marked accusative. We will assume this to be an accidental
gap. Nothing in our account of case marking would prevent such a
combination.
Ditransitiveverbs fall into two classes with respect to passive. In the
first class, only the first postverbal NP passivizes, as illustratedby the
following exampleswith the DAT DAT verb skila 'to return':
(42)a. Eg skilabi henni peningunum.
I returnedher(D) the-money(D)
b.
Henni var skilab peningunum.
she(D) was retumedthe-money(D)
c. *Peningunumvar skilabhenni.
d.
Eg tel
henni hafa
verib skilab peningunum.
I believe her(D) to-have been returnedthe-money
e. *Eg tel peningunumhafa verib skilabhenni.
12 The well-known class of
exceptions to passive like weigh includes one superficially
ditransitiveACC ACC verb, namely,kosta 'to cost' illustratedin (i).
(i)
Dabkostabimig eina kronu.
it cost me(A) one kronur(A)
Just as in English, neither postverbalNP passivizes.Our guess is that this is because the
subject is not agentive enough. The classificationof kosta thus depends on whether one
thinksthat a necessarycriterionfor (di)transitivityis having a passivizableobject.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
460
A. ZAENEN, J. MALING, AND H. THRAINSSON
The behavior of these verbs is straightforward,the only noteworthy
detail being that a DAT object passivizes; but this should not be at all
surprisingafter the discussionof the previous section. Examples (42a-c)
contrast nicely with the following, where the verb skila is used as a
simple transitiveverb, taking a DAT theme but with the goal realized as
a PP complementinstead of a DAT NP.
(43)a. Eg skilabi peningunum til hennar.
I retumedthe-money(D) to her(G)
b.
Peningunum var skilab til hennar.
the-money(D) was returnedto her
We see that when. peningunum is the sole bare NP argument, it can
passivize. These facts seem to suggest that the mapping between grammatical functions and phrase structure positions in Icelandic is fairly
fixed: subjects are mapped onto the preverbalNP and direct objects are
mapped onto the immediatelypostverbalNP. However, this mapping is
not the only possibilityfor objects, as shown by the passivizationpattern
of the second class of ditransitives,namely DAT ACC verbs like gefa 'to
give'.
In some sense, the DAT ACC pattern represents the core class of
ditransitiveverbs: only these verbs survive as ditransitivesin English and
the Scandinavian languages other than Icelandic, i.e., in related languages without morphological case marking. All the other ditransitive
verbs have become instead verbs taking NP PP complements (presumably because at least one of the NPs is semanticallyquite predictable,e.g.
a recipient).
For DAT ACC verbs, either postverbalNP can passivize, as illustrated
in (44). Note that in (44a) the retained object is nominative (ratherthan
accusative); this rather surprisingfact will follow from the principlesfor
default case markinggiven in (61d) below.
amb'attir.
(44)a. Konunginumvoru gefnar
the-king(D) were given(f.pl.) maidservants(N.f.pl.)
The king was given female slaves.
b.
var gefin
Amb'attin
konunginum
the-maidservant(N,sg)was given(f.sg.) the-king(D)
As noted above, the verb agrees with the nominative argument even
when this argumentis not the grammaticalsubject, as in (44a).
To show that in both (44a) and (44b), the object has indeed become a
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND GRAMMATICAL
FUNCTIONS
461
subject, we will run the examples through the tests for subjecthood
summarizedpreviously.
(45) Raising
ambattir.
a. Eg tel
konunginumhafa verib gefnar
I believe the-king(D) have been given(f.pl.) slaves(N)
I believe the king to have been given slaves.
b.
ambattina hafa verib gefna
Eg tel
konunginum.
I believe the-slave(A) have been given(A) the-king(D)
I believe the slave to have been given to the king.
(46) Reflexivization
a. Konunginumvoru gefnar amb'attiri holl
the-king(D) were given slaves in palace
sinni/?hans.
his(+REFL/?-REFL)
b.
Ambattin var gefin konunginumvegna
fegurbar
the-slave(N) was given the-king(D) because-of beauty
sinnar/?hennar.
her(+REFL/?-REFL)
(47) Subject-VerbInversion
a. Um veturinn voru konunginumgefnar ambaittir.
in the-winterwere the-king(D) given slaves(N)
In the winter, the king was given (female) slaves.
b.
Um veturinn var ambfittin gefin konunginum.
in the-winterwas the-slave(N) given the-king(N)
In the winter, the slave was given to the king.
(48)a. Voru konunginumgefnar ambaittir?
were the-king(D) given slaves(N)
Was the king given slaves?
b.
Var amb'attin gefin konunginum?
was the-slave(N) given the-king(D)
Was the slave given to the king?
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
462
A. ZAENEN, J. MALING, AND H. T14RAINSSON
(49) Extraction
a.
Hvaba ambaittir heldur pui ab
konunginum verbi
gefnar?
which slaves(N) think you that the-king(D) will-be given
b.
Konunginum held
eg ab
verbi
ambattin
gefin.
the-king(D) believe I that the-slave(N) will-be given
To the king I think that the slave will be given.
(50) IndefiniteSubjectPostposing
a. Dab voru konungi gefnar ambattir i vetur.
there was king(D) given slaves(N) in winter
There was a king given slaves this winter.
b. Pab var ambatt gefin konunginum i vetur.
there was slave(N) gefin konunginum(D)in winter
(51) SubjectEllipsis
a.
Konungarnir f6ru
viba
og
-
voru
the-kings(N) traveledwidely and -(D)
were
oft gefnar ambattir.
often given slaves
b.
Ambattin kom fra
Irlandi og
-
the-slave came from Ireland and -(N)
var
was
gefin konunginum.
given the-king(D)
(52) InfinitiveComplements
a.
Ab vera gefnar ambattir var mikill heibur.
to be
given slaves(N) was great honor
miklumvonbrigbum.
b. Ab vera gefin konunginumolli
to be given the-king(D) caused great disappointment
To be given to the king caused great disappointment.
c.
Ambattin vonast til ab verba gefin konunginum.
the-slave(N) hopes for to be
given the-king(D)
The slave hopes to be given to the king.
These tests show that both the dative object and the accusative object
can be made grammaticalsubjects by passive. When the dative object is
passivized, the postverbal 'retained object' appears in the nominative
case, and the verb agrees with it in number. At first glance, this might
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
463
seem to contradict our conclusion that it is the preverbal dative NP
which is the subject, but as we will show below, the nominative case
markingis predictable.The accusative object of the active becomes the
nominativesubject of the passive, and passes the same subjecthoodtests
with equal success.
4. PASSIVE IN LFG
In lexical-functional grammar, actives and passives are related by a
lexical redundancyrule. Given the existence of impersonal passives in
Icelandic, it seems reasonable to dissociate the part of the rule that
relates the active subject to the passive by-phrasefrom the part of the
rule that associates the active object with the passive subject. Hence we
have the two rules in (53), whose effect is illustratedin (54):
(53)a.
b.
SUBJ~- AF-OBJ/0
OBJ +SUBJ
(54)
agent
theme
taka: V, 'take' (SUBJ, OBJ)
tekinn: V[+part] (AF-OBJ, SUBJ)
We ignore here the fact that the agent is only rarelyexpressedin passive
sentences in Icelandic. A treatmentof optional agents,can be found in
Bresnan (1982a); for the use of 0 in Lexical rules, see Bresnan(1982b, p.
166). The auxiliaryverbs vera and verbawill be treated as 'raising'verbs
taking a SUBJ and an XCOMP. No thematic role is associated with the
subject of these verbs and they get whatever subject their XCOMP gets
(see Kaplan & Bresnan (1982) for furtherexplanationof the formalism
used in LFG). The morphological principles assigning case will be
discussed below.
4.1. GrammaticalFunctions,ThematicRoles and Case-Marking
in Icelandic
Our account of passive in Icelandic is based on the LFG analysis of
passive sketched above. Note in particularthat only NPs bearing the
grammaticalfunction OBJ passivize. To develop a full account of the
passive construction, we need to specify how these rules interact with
other components of the grammar.In this section, we will outline our
view of this interaction, particularlywith respect to thematic roles and
case-markingrules.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
464
A. ZAENEN,
J. MALING,
AND H. THRAINSSON
4.1.1. The Lexical Component
For the sake of clarity, we restate here the distinction made in LFG
between semantically restricted and semantically unrestricted Grammatical Functions (GFs). The same distinction is made in Relational
Grammarbetween 'obliques' and 'terms'. SUBJ, OBJ, and 2OBJ are
semantically unrestrictedfunctions because they can bear any type of
thematic role, dependingon the verb. For example, in (55a) the SUBJ is
an agent, whereas in (55b) it is an instrument.
(55)a. The presidentkissed the baby.
b. The wrench opened the safe.
This characteristic of SUBJs, OBJs and 2OBJs contrasts with the
behavior of GFs such as with-phrasesand to-phrases.While these PPs
may bear more than one function, the functions are restricted by the
form of the PP: there is no verb in English that expresses its goal by
means of a with-phrase,or its instrumentby means of a to-phrase.(This
is necessarily an oversimplification;see Bresnan (1982c), for a more
elaborate discussion, and Levin (1985) for further developments.)Thus
the nonverbal constituents of a sentence can be classified as follows
into two kinds of verbal argumentsplus a category of adjuncts.
(56)
ARG: (i) semanticallyunrestricted(SUBJ, OBJ, 2OBJ)
(ii) semanticallyrestricted
all types of adjuncts
ADJ:
4.1.2. Case Assignment
Case can be assigned in (at least) three different ways,13which we will
call semantic, lexical or idiosyncratic, and functional case assignment.
SEMANTIC case-markingincludes such traditional'adverbial'categories
as accusatives of time or duration, and instrumentaldatives; these are
illustrated in (57).14
(57)a.
b.
Straikurinn
beib allan dagin.
the-boy waited all(A) day(A)
Hann tok vini sinum
opnum ormum.
he
took friend his[+REFL] open(D) arms(D)
He greeted his friend with open arms.
13 Although our terminology may differ, these are the same three types of case distinguished,e.g., in Hjelmslev,and more recently in a GB-frameworkin Babby (1980) and
Freidinand Babby (1983).
14 These examplesare taken from Fribjonsson(1978).
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
465
We will not consider this type of case assignment in any detail here.
Depending on their function, semanticallycase-markedconstituents are
either semanticallyrestricted functions or adjuncts. IDIOSYNCRATIC or
lexical case marking is an idiosyncratic property of a lexical item,
assigned by a verb, preposition or adjective. We assume that idiosyncratic case is associatedwith a particularthematicrole, and that this case
marking is assigned before thematic roles are associated with grammaticalfunctions.FUNCTIONAL case markingis what is widely referredto.
as regular or "default" case marking, which results in nominative subjects and accusative objects. It is sensitive to surface grammatical
relations, and hence applies after all association principles and reassociationrules, includingpassive.
4.1.3. The MappingbetweenThematicRoles and GrammaticalFunctions
Our account of passive relies on the existence of association principles
which establish a mapping between thematic roles and grammatical
relations. The existence of such association principles is assumed in
various current linguistic theories, but has not been fully fleshed out.15
Our account makes use of both universal and language-specific association principles.
We postulate a level of representationat which the valency of a verb is
determinedand its argumentscan be distinguishedin terms of thematic
roles. At this level of thematic structure, verbs like dansa 'dance' and
lesa 'read' have the following representations:
(58)
dansa 'dance' (agent)
lesa 'read' (agent, theme)
The relationship between different verb forms can be represented in
termsof the optional suppressionof an argument.For example,we might
want to express the relation between the causative and non-causative
forms of a verb like faekka 'decrease' as follows:
(59)
fa!kka 'decrease' ((agent), theme)
15 It is clear that this general schema owes
much to the work of Anderson (1977) and
Wasow (1977, 1980), to Ostler (1979) and Marantz(1981), and to Amritavelli (1980),
Rappaport (1983) and Levin (1985). We do not have the time here to spell out the
importanceof these various contributions.These accounts assume the existence of something like a thematiclevel of representation.It is not completelyclear to us that this is a
purely linguisticlevel; it seems ratherto be a level that mediates between cognitive and
linguisticrepresentations.In this paper,however,we will treat it as a linguisticlevel.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
466
A. ZAENEN, J. MALING, AND H. THRAINSSON
Idiosyncratic case-marking takes this level as its input. Thus we are
assuming idiosyncratic case is assigned to thematic roles and not to
grammatical functions. This assumption is motivated by the fact that for
a given verb, the idiosyncratic case-marking associated with a given
thematic role is preserved under rules like passive and raising when the
GF changes. Hence verbs like gata 'take care of' or lofa 'promise'get
the following representationsafter the assignmentof idiosyncraticcase:
(60)
gzeta 'take care of' (agent, theme)
GEN
iofa 'promise'(agent, theme, goal)
DAT DAT
A set of association principles determines which thematic role will be
mapped onto which grammatical function. Some of the association
principles seem to be universal, whereas others are language-specific.
Where the universal and the language-specific principles interact, the
language specific ones take precedence, and the universal ones act as
elsewhere conditions. One universal principle that has often been proposed is that the agent role is associated with the SUBJECT function.
We will assume that this principle is by and large correct.16 We also
assumea hierarchyof grammaticalfunctions.17
Given these background assumptions, we can now formulate the
following set of associationprinciplesfor Icelandic.
(61) IcelandicAssociationPrinciples
a. If there is only one thematic role, it is assigned to SUBJ; if
there are two, they are assignedto SUBJ and OBJ; if there are
three, they are assigned to SUBJ, OBJ, 2OBJ. (Universal)
b. AGENTS are linked to SUBJ. (Universal)
c. Case-markedTHEMES are assigned to the lowest available
GF. (Language Specific)
d. Default Case Marking:18the highest available GF is assigned
NOM case, the next highest ACC. (Universal)
16
We have ignoredthroughoutthe problemposed by the existence of ergative languages.
It is not clear to us whetherergativityshould be thoughtof as a difference in association
principles('deep' ergativity)or in case-markingprinciples('surface'ergativity)or both.
17 The hierarchyof grammaticalrelationsstems from work in Relational Grammar(see
Perlmutter(ed.) 1983, Cole & Sadock 1977).
18 The default case-marking is in a certain sense the equivalent of "structuralcase
marking"in GB, except that we allow for exceptionssuch as nominativeobjects (see 62b).
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND GRAMMATICAL
FUNCTIONS
467
4.2. Applyingthe AssociationPrinciples
These association principles are not specific to passives; they also account for preverbaloblique NPs in active sentences, as in (62).
(62)a.
b.
Mer er kalt.
me(D) is cold.
Henni hefur alltaf pott
Olafur leibinlegur.
her(D) has always thought Olaf(N) boring(N)
We assume that Icelandic does not have any impersonal verbs in the
sense of 'subjectless'verbs except for those with no semantic arguments,
e.g., weather-verbs,or those with prepositionalphrase complementsbut
no argumentsrealized as bare NPs.
Now note that the association principles given here allow for two
different assignmentsof GFs in the case of accusative THEMES which
do not bear idiosyncraticcase. The associationconvention in (61c) does
not say anythingabout them, since it only specifies that idiosyncratically
markedthemes will be associatedwith the lowest available GF. Hence in
the ditransitive cases where we have a non-case-marked theme and
another nonsubject argument,we can assign the GFs in two ways. For a
verb like gefa 'to give', we will get the following two possibilities.
(63)
a.
b.
gefa: V(agent
theme goal)
[+dat]
SUBJ OBJ 2OBJ
SUBJ 2OBJ OBJ
Note that this type of dual association resolves the problem of double
objects. Although we'can associate either the THEME or the GOAL
with the direct object in the lexical form of a verb like give, we will not
be able to associate both argumentswith the direct object at the same
time because the association principles respect functional uniqueness.
Whichever GF the GOAL is linked to will get markedwith the (lexically
assigned) dative case; whichever GF the THEME is linked to, it will get
marked accusative in the active version by virtue of the principle of
default case marking.
This solution is not empiricallyequivalent to a solution in which we
have two direct objects. It predicts that whatever othercharacteristicof
OBJ we might find in the language, it will not apply to the immediately
postverbalNP (the 'retainedobject') of a ditransitiveverb in the passive,
because that NP must have been assigned to 2OBJ. To test this predic-
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
468
A. ZAENEN, J. MALING, AND H. THRAINSSON
tion, other syntactic rules that single out OBJs as opposed to all other
postverbal GFs in Icelandic must be found. Rognvaldsson (1982) discusses some data which might illustrate such a rule, and which support
our analysis of DAT ACC verbs.19 The evidence is based on the
interaction of heavy NP shift and reflexivization (at least for those
speakers who allow nonsubject antecedents). Rognvaldsson shows (pp.
133-135) that the reflexivizationpossibilitiesreflect the underlyingorder
of the two postverbal NPs. This is illustratedfor an ACC DAT verb,
svipta 'deprive', by the examples in (64), taken from Rognvaldsson (his
[74a, b] and [80].)
(64)a.
Sjorinnsvipti hannii [manni
sinumi]
the-sea deprivedher(A) husband(D) her[+REFL]
The sea deprived heri of heri husband.
b. *Sj6rinnsvipti [konu sinai]
manninumi
the-sea deprived wife(A) his[+REFL] the-man(D)
The sea deprived hisi wife of the mani.
c. *Sj6rinnsvipti manninumi [gomlu konuna sinlai... ]NP
the-man(D) old
the-wife his[+REFL]
The sea deprived of the mani hisi old wife.
Sentence (64b) is ungrammatical because the supposed antecedent
doesn't precede the reflexive; however, crucially, applying heavy-NP
shift to the first NP, as in (64c), does not make the reflexive possible.
One possible interpretationof this fact (not Rognvaldsson's)is that an
OBJ but not a 2OBJ may be the antecedent of a reflexive. This
generalization(togetherwith linearprecedence) accounts for the contrast
noted by Rognvaldssonin footnote 25 (pp. 231-2). The DAT ACC class
of ditransitive verbs appear to have not one but two reflexivization
patterns: i.e., either object can be an antecedent. The following sentences are based on Rognvaldsson'sexamples.
(65)a.
Eg gaf konungi [ambaittinasina.]
I gave a-king (D) slave(A) his[+REFL]
19 We would like to thank an anonymousreader for pointing out the relevance of these
data.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
469
b. Eg gaf amb'attina [konungi sinum]Np
I gave the-slave(A) king(D) his[+REFL]
Example(65b) contrastswith (64c) above. [Note that the shifted (Dative)
NP in (65b) need not be especially heavy.] Given our analysis of DAT
ACC verbs as getting dual assignmentsto GFs, these facts receive a very
simple and natural explanation: an OBJ but not a 2OBJ can be the
antecedent of a reflexive. Given the under-determinationof GF assignments embodied in the principles(61a-d) and the alternativeassignments
displayedin (63), this follows naturally.
As pointed out by an anonymousreviewer, further support for alternative assignmentsas in (63) seems to come from the facts discussed in
Bernodusson(1982, pp. 37-8). He observes that for many active DAT
NOM verbs (verbs with DAT subjects and NOM objects) either order
feels natural ("eblileg"); this intuition about 'unmarked' word order
correlateswith the fact that either argumentcan occur between the finite
and non-finiteV, a position reservedfor grammaticalsubjects. The facts
for actives are analogous to the facts for DAT-NOM passives derived
from DAT ACC ditransitiveverbs.
The solution proposed here is preferable to one in which we would
allow 2OBJ to passivize in Icelandic. Under such a proposal, we would
need a different principle to exclude double passivizationin the case of
ditransitiveverbs other than with DAT ACC objects,20a restrictionthat
was illustratedabove in (42).
We conclude this section by presentingChart I, which summarizesthe
case markingand grammaticalfunctionsassignedto the argumentsof the
various classes of ditransitiveverbs.
The default case-markingprinciple stated in (61d) correctly accounts
for a rather surprisingfact already noted in passing. Recall that in the
passive sentence in (44a), the postverbal theme is marked nominative,
even through it is not the grammaticalsubject but rathera 2OBJ. Since
the theme is not assigned idiosyncraticcase by this class of ditransitive
verbs, whetheror not it becomes the subject in the passive version, it will
be the highest GF not alreadyassignedcase. Hence it will get nominative
case by the default case-markingprinciple.
20
Dual assignmentshave been shown to be independentlymotivatedin other languages,
and hence preferableto analyseswhich assumeeither that 2OBJs can passivizeor that the
same sentence can contain two (direct)OBJs. One such languageis Kikuyuas discussedin
Zaenen (1983). Given that this possibility of alternative assignmentsdoes exist in the
world's languages, we can make use of it for Icelandic. Our conclusion, then, is that
Icelandicdoes not violate the functionaluniquenessprinciple,and can be analyzedalong
lines that are knownto be independentlynecessaryfor other languages.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
A. ZAENEN,
470
J. MALING,
AND H. THRAINSSON
CHART I
Verb Type
Thematic
Roles
DAT-ACC
Agent
Theme
Goal
ACC-DAT
Agent
Theme
Source
ACC-GEN
DAT-DAT
DAT-GEN
Agent
Theme
Goal
Lexical
Case marking
Grammatical
Functions
Default
Case-marking
NOM
ACC
DAT
SUBJ (61b)
2OBJ/OBJ (61a)
OBJ/20BJ (61a)
SUBJ (61b)
2OBJ (61c)
OBJ (61a)
NOM
DAT
NOM
GEN
SUBJ (61b)
2OBJ (61c)
OBJ (61a)
ACC
ACC
Agent
Theme
Goal
SUBJ (61b)
2OBJ (61c)
OBJ (61a)
NOM
DAT
DAT
Agent
Theme
Goal
SUBJ (61b)
2OBJ (61c)
OBJ (61a)
NOM
GEN
DAT
4.3. Verbsof VariablePolyadicity
The associationprinciplesmake some furtherpredictionsthat are borne
out by the data. As in English, many verbs can be used with variable
numbersof grammaticalarguments.For example, the verb oska 'to wish'
can be either a simple transitive verb taking a genitive THEME, or a
ditransitivetaking a dative GOAL and a genitive THEME; in other
words,the goal argumentis optional.
(66)a.
Du' hefur Oskab henni Pess.
you have wished her(D) this(G)
Ni hefur Oskabpess.
b.
c. * Pi hefur Oskabhenni.
Given the association principles for Icelandic in (61), the mapping
between theta-rolesand GFs is determinedby the polyadicityof the verb.
Hence we predict that for such verbs, the ability of a NP with a given
thematic role (e.g., the THEME) to passivize will also depend on the
polyadicityof the verb. This can be seen from the different assignments
of the OBJ-functionin (67):
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND GRAMMATICAL
(67) oska: V(agent
FUNCTIONS
471
theme (goal))
[+gen] [+dat]
a.
SUBJ 2OBJ
b.
SUBJ
OBJ
OBJ
Only when a NP is assigned to OBJ will it passivize.7 As an idiosyncratically case-marked theme, bess can only be assigned to the lowest
available GF. When the verb is used as a simple transitive,this will be the
OBJ function, and hence tess will passivize and raise as shown in (68a).
But when oska is used as a ditransitive,pess will be assigned to 2OBJ,
and hence will be unable to become the subject of a passive and
subsequentlybe raised. This is illustratedby the contrastin (68).
(68)a.
Dess var Oskab (*henni).
this(G) was wished(her(D))
b.
Eg tel
Pess hafa
verib Oskab (*henni)
I believe this(G) to-have been wished (her(D))
c.
Henni var Oskabpess.
d.
henni hafa
verib 'skab tess.
Eg tel
I believe her(D) to-have been wished this(G)
Other verbs that work the same way include leyna 'to conceal' ACC
DAT; bibja 'to ask' ACC GEN; skila 'to return' DAT DAT; hota 'to
threaten' DAT DAT; synja 'to deny' DAT GEN.22The judgments are
21 Note furtherthatpassivein Icelandicis not restrictedto NPs that are thematicallylinked
to the passiveverb. We also find passivesin the so-called raisingconstructions,as shownin
(i).
(i)
Hun er talin vera dugleg.
she(N) is believedto-be conscientious(N).
Such examplesindicatethat passive in Icelandicis not a lexical rule in the sense of Wasow
(1977) or Chomsky(1981). We have not taken Raisingand Equi constructionsinto account
here; see Andrews(1982b) for a fuller treatmentof case-assignmentin ModernIcelandic.
22 We can speculateabout the reason for the distributionof DOs in Icelandic:on the one
hand all immediatelypostverbalNPs that are argumentsof the verb are DOs, regardlessof
their case marking;on the other hand all postverbalaccusativesthat are neitherXCOMPs
nor have a semantically restricted function can passivize. It is difficult to resist the
temptationto link these observationsto the fact that Icelandicis a languagethat has at the
same time a ratherrigidwordorderand a rich case markingsystem.It has often been noted
that languages have two major strategies to encode grammaticalrelations,the one being
word order,and the other case marking.Icelandic does both, and it can be assumedthat it
is this convergenceof propertiesthat is illustratedin the notion of direct object that is used
in the language: a DO is an argumentthat either immediatelyfollows the verb or is an
accusativeobject.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
472
A. ZAENEN, J. MALING, AND H. THRAINSSON
extremely clear-cut. These fact show clearly that passive is not sensitive
to thematic roles but must be expressed as operations on grammatical
functions.
4.4. Some FurtherAlternations
A few furtheralternationsremainto be accountedfor. Since we have not
studied these alternationsextensively, our proposals here are very tentative, but we want to give some indication of how such alternations
could be analyzedwithin the general frameworkwe are assuming.
In section 3.2, we noted the existence of a few different types of
ACC-ACC verbs, some of which were arguablyditransitive.The first
type, which was illustrated in (38) and is repeated here for convenience,
is clearly not ditransitive:
(38)a.
Deir k6llubu hana Kiddu.
they called her Kidda(A)
b.
Hun
var kdllub Kidda.
she(N) was called Kidda(N)
In verbs of this type, the second NP is neither an OBJ nor a 2OBJ, but
rather an XCOMP, i.e., an (objective) complement, and hence falls
outside the scope of this paper, since predicative complements never
passivize. The status of Ahe second type, illustrated in (39), isn't so clear:
(39)a.
hoggva einhvern banahogg
to-hit someone(A) a-deadly-blow(A)
b.
slai einhvern kinnhest
to-hit someone(A) a-blow-on-the-cheek(A)
c.
kyssa einhvern
rembingskoss
to-kiss someone(A) a-deep-kiss(A)
Here we have an alternationbetween a simple transitiveform and what
looks like a ditransitiveform of the verb. For ease of exposition, let us
refer to the second accusative, e.g., 'a deadly blow', as a COGNATE
ACCUSATIVE. The cognate accusative might be analyzed as (some kind
of) an object; alternatively,it might be that the cognate accusative is
actually a semanticallyrestricted GF, i.e., one restricted to a particular
thematic role. As such, it would be neither an OBJ nor a 2OBJ, and
hence would not participatein any of the alternationswe are discussing.
The principles we have already formulated give us a way to test the
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
473
correctness of this analysis:if alternatingforms exist in which only the
cognate ACC is present, do these monotransitive forms have passive
counterparts?If the cognate accusative in sentences like (39) is assigned
to a semantically unrestrictedfunction, then its function is simply the
highest available function given the polyadicity of the verb; we would
then expect it to passivizewhen it is assigned to the OBJ-function.If, on
the other hand, it does not passivize when it is the sole postverbal NP,
then we would want to look for aguments that indeed in both the
ditransitiveand monotransitiveforms the cognate accusative is either a
semanticallyrestricted (OBLique) argument, or not an argumentat all.
It turns out that the cognate accusative can at least marginallybe the
sole postverbal argument,and in such examples it can be passivized, as
illustratedin (69).
(69)a.
b.
c.
d.
Njall hj6 aldrei banahogg.
Njal hit never a-deadly-blow(A)
var hoggvib 'a mibnatti.
?Banah6ggib
the-deadly-blow(N) was struck at midnight
J6n kyssir bara rembingskossa.
Jon kisses only deep-kisses(A)
?Rembingskossareru oft kysstiri astars6gum.
deep-kisses(N) are often kissed in love-stories
Crucially,the ACC case markingturnsinto NOMinative,showing that in
these examples at least, it was not idiosyncraticallyassigned. These facts
suggest that the cognate accusative should be analyzed as a 2OBJ in
exampleslike (39), but as an OBJ when it is the sole postverbalNP, as in
(69a, c).23
If this is right, the question becomes: what is its thematic role?
Consider the example in (39a). The relation of hoggva 'hit' to the
animateobject einhvern 'someone' seems to be the same with or without
the cognate accusative; moreover, the thematic role of the cognate
accusative is presumablythe same with or without the 'hittee'. Although
the 'hittee' is arguably a GOAL rather than a THEME, it seems
unreasonableto assign the THEME role to 'a deadly blow' in either type
23
This is not the only possibleway to look at the data. We leave it to the readerto devise
other possibilities,too tedious to spell out at this point. The one we are exploringhere, i.e.,
the analysisas 2OBJ, is the one that seems the most problematicfor our generalaccountof
passive, but also the one that intuitivelyseems to us to be the most likely.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
474
A. ZAENEN, J. MALING, AND H. THRAINSSON
of sentence. For most ditransitiveverbs, the THEME is an obligatory
argument. The cognate accusative, on the other hand, is merely an
optional complement; moreover it certainly does not undergo the
movement or action described by the verb 'hit'. It seems to be a
semanticallyempty argument,one which cannot really be questioned or
focussed. In fact its semantic contributionto the meaning of the verb is
ratherlike a manneradverbial.Just what role it bears is not clear to us.
What is clear is that such cognate accusativesnecessitate the formulation
of anotherassociationprinciple,to make sure that they are alwayslinked
to 2OBJ if anotherNP-argumentis present,since this assignmentwill not
follow automaticallyfrom the principlesin (61) if the cognate accusative
is not an idiosyncraticallycase-markedtheme.
It also appears that these NPs are part of the VP. Consider the
placementof an adverblike sjaldan'rarely'in sentences with an auxiliary
verb, as illustratedin (70).
(70)a.
b.
*Hann hefur gefib barninu
sjaldanpelann.
he
has given the-child(D) rarely the-bottle(A)
*Hann hefur hoggvib ovini
sjaldanbanahogg.
he
has struck enemies(A) rarely a-deadly-blow(A)
Thrainsson(1984) argues that there is a class of adverbsin Icelandic (including sjaldan),that can occur anywherein S as an immediatedaughter
of S (as long as other word order constraints,such as the verb-second
constraintare observed), but not embedded at a lower level, e.g., within
VP. He argues that in sentences with an auxiliary,there is a VP-node
following the (finite) auxiliary and dominating the verb and its (postverbal) arguments. If his arguments are correct, the examples in (70)
indicate that cognate objects like banahogg 'deadly blow' in (70b) are
within the VP just like regular objects such as pelann 'the bottle' in
(70a).
A third type of double accusative can be found with verbs like keyra
'drive'. Here we find the following alternations, pointed out by an
anonymousreviewer:
(71)a. Hann keyrbibilinn
pessa leib.
he drove the-car(A) this route(A)
b. Hann keyrbiPessa leib.
he drove this route(A)
The second accusative in (71a) behaves in some respects like an object,
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND GRAMMATICAL
FUNCTIONS
475
namely, it passivizeswhen it is the sole postverbalNP:
(72)
Iessi leib
hefur aldrei verib keyrb.
this route(N) has never been driven
However, when two postverbalNPs are present, as in (71a), only the first
NP passivizes:
(73)a.
Billinn
var keyrburpessa leib.
the-car(N) was driven this route(A)
b. *lessi leib var keyrb bilinn.
this-route(N)was driven the-car(A)
Here again, we have a pattern in which only the first postverbal NP
passivizes, confirmingprinciple (61a). (Similarlyfor another verb, aka
'drive', but with DAT ratherthan ACC case on 'the-car'.)
We will assume that some transitive verbs allow the PATH-role,
normally realized as a PP, to be assigned to the OBJ function when that
position is otherwise unfilled. It is not entirely clear, however, how to
analyze the accusative Pessa leib 'this route' when it is the second postverbal NP and the OBJ function is filled, as in (7 Ia). Such accusatives are
frequently referred to as adverbials or semantic accusatives, which would
suggest that they are not ARGuments (20BJ) of the verb, but rather
some sort of ADJunct. If that were the case, one might expect them to
contrast with regular verbal arguments with respect to the VP-test
discussed above. While most speakers agree that there is some
difference, the judgments are unfortunately not very clear.
(74)a. ?*Jon hefur gefih barninu
oft
pennan pela.
Jon has given the-child(D) often this
bottle(A)
b. ?Omar hefur ekib bilnum
oft Oessaleib.
Omar has driven the-car(D) often this route(A)
In addition, the placement of manner adverbs gives a similar (weak)
contrast:
(75)a. ??Hannmun gefa barninu
mjog varlega pennan pela.
he
will give the-child(D) very carefully this bottle(A)
b.
Omarmun keyra bilinn
mjog varlega bessa leib.
he
will drive the-car(A) very carefully this route(A)
It is not obvious what to make of these rather subtle differences.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
A. ZAENEN, J. MALING, AND H. THRAINSSON
476
However, within our framework, the fact that Iessa leib remains accusative when the dative object of aka 'to drive' is passivized clearly
dictates treatingit as an adverbialadjunctratherthan a verbal argument.
To summarize,we have proposed the following representationsat the
thematic level for the two classes of verbs discussed in this subsection.
(76)a. hoggva 'strike' (agent, (theme), (cognate-obj))
b. keyra 'drive' (agent, theme)or (agent, path)
c. aka 'drive'
(agent, theme)or (agent, path)
[+DAT]
The possible associationswith GFs are shown in (77).
(77)a. hoggva: (agent (theme) (cognate-obj))
SUBJ OBJ
2OBJ
SUBJ
SUBJ
OBJ
OBJ
b.
keyra: (agent theme)
SUBJ OBJ
or
(agent path)
SUBJ OBJ
c.
aka: (agent theme)
[+DAT]
SUBJ OBJ
or
(agent path)
SUBJ OBJ
If both the theme and the path-role for the verbs keyra or aka are
realized, then the path is an ADJunct and not an ARGument of the verb.
5.
GERMAN
ASSOCIATION
PRINCIPLES
Having demonstrated that in Icelandic oblique NPs can be made grammatical subjects by passivization, we need to address the question of
whether the same is true of German, which has constructions that are
superficially similar. Consider the German equivalent of an Icelandic
sentence like (4), repeated here for convenience.
(78)
Ihm
wurde geholfen.
Honum var
hjailpab
him(D) was
(German)
(Icelandic)
helped.
Although the work on German syntax has not led to as extensive a study
of the syntactic properties of subjects as has been made for Icelandic, it is
clear that the German facts do not parallel the Icelandic ones. There are
no reasons to assume that ihm in (78) above is a grammatical subject.
Thus the German examples, while superficially very similar to the
Icelandic ones, are functionally quite different. Some of the evidence
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE AND GRAMMATICAL FUNCTIONS
477
supportingthis conclusion is illustratedin the following examples,mostly
from Cole et al. (1978).
Subjects of infinitivesin Germancan be controlled (either functionally
or anaphorically),but this is not possible with the passives of verbs taking
oblique case marking. ArbitraryPROs can be found as the understood
subject of German infinitivals,as shown in (79a), but not with idiosyncraticallymarkedNPs, as shown in (79b).
(79)a.
Im Sommerzu reisen ist angenehm.
in summer to travel is agreeable
To travel in the summeris nice.
b. *Geholfen zu werden ist angenehm.
is agreeable
helped to be
To be helped.is nice.
c.
Aufgenommenzu werden ist angenehm.
admiited
to be
is agreeable
To be admittedis nice.
Similarly,EQUI-control is not possible with the passives of verbs taking
oblique objects. German has a rule of EQUI control, illustratedin (80).
(80)
Er hofft weg zu gehen.
he hopes away to go
As shown in (81a), this rule applies to passives; however, as illustrated
(81b), it does not apply to the dative NP of a passive sentence like (78).
(81)a.
Er hofft aufgenommenzu werden.
he hopes admitted
to be
b. *Ihm/*Er
hofft geholfen zu werden.
him(D)/he(N) hopes helped to be
Further examples can be found in Cole et al. (1978); for example, the
preverbal oblique NPs cannot be deleted under identity was a (nominative) subject.
(82)a.
b.
Er
kam und (er) besuchte die Kinder.
he(N) came and (he) visited the children
Er kam und (er) wurde verhaftet.
he came and (he) was arrested
c. *Er kam und
-
he came and __(D)
wurde geholfen.
was
helped
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
478
A. ZAENEN,
J. MALING,
d. *Er sah die Damen und
- he saw the ladies and
AND H. THRAINSSON
gefielen sie.
(D) pleased them(f.)
e. *Er sah die Damen und sie gefielen
he saw the ladies and they pleased
(D)
Nor can the oblique NP be deleted by relative clause reduction.
(83)a. Der das Buch lesende Junge heisst
Wilhelm.
the [the book reading] boy is-named Wilhelm
b. *Das der Junge lesende Buch heisst
Sieben Legenden.
the [the boy reading book] is-named Seven Legends
c.
Das vom Jungen gelesene Buch heisst
the [by-the boy
read] book is-named
Sieben Legenden.
Seven Legends
d. *Der vom Lehrer geholfene Junge bekam eine gute Note.
the by-the teacherhelped] boy got
a good grade
c. *Der das Buch gefallende Junge sitzt da in der Ecke.
the [the book pleasing] boy sits there in the corner
Thus the same type of tests that show that oblique NPs can be grammatical subjects in Icelandic show that the Germananalogues cannot be
analyzedas such.
As in Icelandic, the 'quirky'argumentsof passive verbs do not behave
differently in this respect from the quirky arguments of active forms.
Consideractive constructionslike the one illustratedin (84).
(84)
Mir ist ubel.
me(D) is nasty
I am nauseated.
Here too it can be shown that mir does not behave like a grammatical
subject, as illustrated in the following examples.
(85)a. *Mir hofft ubel
zu sein.
me(D) hopes nauseated to be
*Ich hoffe ubel
zu sein.
I(N) hope nauseated to be
b. *Ubel
zu sein ist unangenehm.
nauseated to be is disagreeable
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE AND GRAMMATICAL FUNCTIONS
479
In spite of the superficialsimilaritiesbetween Icelandic and German,the
analysis of constructions without nominative arguments (be they active
or passive voice) in the two languages must be quite different. We can
account for these differences by means of one language-specific association principle for German,which replaces the language-specificprinciple for Icelandic in (61c):
(86)
Case-markedTHEMATIC ROLES are assigned to 2OBJ.
In other words, idiosyncraticallymarked argumentsare associated with
the function 2OBJ regardlessof the valence of the verb. Thus the entire
set of Association Principlesfor Germanis as given in (87):
(87) GermanAssociationPrinciples
a. If there is only one thematic role, it is assigned to SUBJ; if
there are two, they are assignedto SUBJ and OBJ; if there are
three, they are*assigned to SUBJ, OBJ, 2OBJ. (Universal)
b. AGENTS are linked to SUBJ. (Universal)
c. Case-marked THEMATIC ROLES are assigned to 2OBJ.
(Language Specific)
d. Default Case Marking: the highest available GF is assigned
NOM case, the next highest ACC. (Universal)
Remember that universalprinciplesare interpretedas elsewhere conditions applying after any language-specificprinciplesor assignmentsof
restricted GFs. Hence all idiosyncraticallymarked argumentswill have
the statusof 2OBJs, and thus will not passivizeunder the assumptionthat
German has only the unmarked passive rule according to which only
(direct) OBJs passivize. Thus they will never be subjects in either passive
or active sentences. This is exactly the result that we want for German.
Together with the functionaluniquenesslaw, the associationprinciples
for German make some further predictions: namely that in German
(unlike Icelandic) we cannot get two idiosyncraticallycase-markedNPs
after the verb. This follows from (87c) given functionaluniqueness:two
different idiosyncraticallycase-marked thematic roles cannot both be
assigned to the same GF, namely 2OBJ. Thus, under the assumptionthat
both DAT and GEN case are assigned idiosyncraticallyby the verb in
German, there should be no DAT-DAT, DAT-GEN or GEN-GEN
verbs in that language.24Two accusative argumentsare allowed if one is
24
This same predictionis made within the GB frameworkby HartmutCzepluch (1982);
his modificationsof the GB theory will not extend to Icelandic however, where many of
these combinationsdo occur. There are no GEN-GEN verbs in Icelandiceither. We have
not ruledout this possibility,and treat it as an accidentalgap in that language.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
480
A. ZAENEN, J. MALING, AND H. THRAINSSON
an OBJ and the other is idiosyncraticallymarked,and hence a 2OBJ. In
fact, DAT-GEN combinationsdo not exist in contemporaryGerman,so
the prediction made by our analysis is correct in that respect. Our
predictions are also correct with respect to the behavior of verbs with
two accusatives. The few cases that still exist allow only for the passivization of one of the objects, as shown by the following examples.25
das Gedicht abgehort.
(88)a. Ich habe ihn
I have him(A) the poem(A) heard
I had him recite the poem.
b. Er ist das Gedicht abgehort.
c. *Das Gedicht ist ihn abgehort.
So we can account for the difference between Icelandic and Germanby
means of one language-specific association convention for each language. This account also correctly predicts the non-occurrence of a
certain number of combinations in German. (Some other gaps remain
unexplained, however; for example, nothing in our account would
prevent Icelandic from 'havingGEN GEN or ACC ACC combinations
in postverbal position.26)We have shown that superficiallyvery similar
In support of the claim that verbs assign at most one lexical (i.e., idiosyncratic)case,
Holmberg (1985) suggests that while the case markingon themes is truly idiosyncraticin
Icelandic, the apparent instances of two lexical cases actually reflect predictablecase
marking associated with particularthematic roles: for first objects, dative case marks
[+recipient]and accusativecase marks[-recipient]. Holmbergattributesthe lack of GEN
first objects to the fact that genitive case is not associatedwith any particulartheta role.
The hypothesisthat lexical items can assign at most one lexical case is clearly desirable
from a theoreticalstandpoint,and Holmberg'scharacterizationcomes close to descriptive
adequacy.Unfortunately,there are counterexampleswhich clearly requirethe assignment
of lexical case to two arguments.Among the DAT GEN ditransitivesare at least three
verbs whose first (DAT) objects are not recipients:synja 'to deny', varna 'to deny' and
frfja 'to question'.
25 Of course, one must explain the fact that such forms existed in earlier stages of the
language. Unfortunately,it is difficultto determinewhat their status was. They seem to
have always been rather bookish (see Curme 1922, for discussion). Our guess is the
following:in earlierstages, the dative case was not just an idiosyncraticcase markingon a
2OBJ, but could also encode semanticallyrestricted grammaticalfunctions (e.g., instrumentals).Hence there would have been no violation of the functionaluniquenesslaw.
One might speculateabout why the evidence in the modernstages of both Icelandicand
German points to an analysis in which bare NPs tend to be analyzed as occupying a
semanticallyunrestrictedfunction. But it seems true that even in the Germaniclanguages
that have maintained overt case, semanticallyrestricted grammaticalrelations are in
generalrealizedas PPs.
26 As noted above in sections 3.2 and 4.4, it is debatablewhether
any of the superficially
ACC-ACC verbs are trulyditransitive.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE
AND
GRAMMATICAL
FUNCTIONS
481
sentences in two rather closely related languages must be analyzed in
quite different ways. In doing so, we have shown that a possibilitythat
has been rejected on the basis of German data, namely that of oblique
subjects and non-accusative direct objects, actually exists within Germanic. This leads to an interesting question in historical syntax: was
English in its earlierstages more like Germanor more like Icelandic?We
will not try to answer that question, but merely note that it cannot be
assumedon the basis of universalprinciplesof language that English was
like German,as has been done in some currentwork (see e.g., Lightfoot
1979, 1981).27
REFERENCES
Allen, Cynthia: 1982, 'Dative Subjects in Old and Middle English', unpublishedmanuscript,AustralianNationalUniversity.
Amritavelli,Raghavachari:1980, Aspectsof the Organizationof RedundancyRules in the
Lexicon,unpublishedPh.D. dissertation,SimonFraserUniversity.
Anderson,Stephen: 1977, 'Commentson the Paper by Wasow',in Culicover et al. (eds.),
pp. ;361-377.
Andrews, Avery: 1976, 'The VP ComplementAnalysis in Modem Icelandic', NELS 6,
1-21.
1982a, 'Long Distance Agreement in Modem Icelandic',in P. Jacobson& G. K.
Pullum(eds.), The Natureof SyntacticRepresentation,
Reidel, Dordrecht,pp. 1-31.
: 1982b, 'The Representationof Case in Modem Icelandic',in J. Bresnan(ed.), The
Mental Representation
of GrammaticalRelations,MIT Press, Cambridge,pp. 427-503.
Babby, Leonard:1980, 'The Syntaxof SurfaceCase Marking',CornellWorkingPapersin
Linguistics,No. 1, pp. 1-32.
Bern6dusson,Helgi: 1982, 'Opersonulegarsetningari islensku',('ImpersonalSentencesin
Icelandic')cand. mag. thesis, Universityof Iceland.
den Besten, Hans: 1981, 'A Case Filterfor Passives',in A. Belletti, L. Brandiand L. Rizzi
(eds.), Theoryof Markednessin GenerativeGrammar,Scuola normaleSuperioredi Pisa,
Foris, Dordrecht.
Breckenridge,Janet: 1975, Rules WhichNothing Undergoes,unpublishedBachelor'sthesis,
HarvardUniversity,Cambridge.
Bresnan,Joan: 1982a, 'The Passive in Lexical Theory',in Bresnan(1982d), pp. 3-86.
: 1982b, 'Polyadicity',in Bresnan(1982d), pp. 149-172.
: 1982c, 'Controland Complementation',in Bresnan(1982d), pp. 282-390.
(ed.): 1982d, The Mental Representationof GrammaticalRelations, MIT Press,
Cambridge.
and HoskuldurThrainsson:1984, 'A Note on IcelandicCoordination',unpublished
manuscript,StanfordUniversityand Universityof California,Santa Cruz.
Butler, M.: 1980, GrammaticallyMotivatedSubjectsin Early English, unpublishedPh.D.
dissertation,Universityof Texas at Austin.
27 That such preverbaloblique NPs may have been grammaticalsubjects in earlierstages
of Englishhas been.hintedat in Butler (1980), who gives statisticsshowingthat they do not
behave like normalimpersonalconstructions.See also Allen (1982), who shows that this is
true for some but not all such verbs, and also Harris(1975).
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
482
A. ZAENEN,
J. MALING,
AND H. THRAINSSON
Chomsky,Noam: 1981, Lectureson Governmentand Binding, Foris, Dordrecht.
Cole, P. and J. Sadock (eds.): 1977, GrammaticalRelations, Syntax and Semantics 8,
AcademicPress,New York.
, W. Harbert, G. Hermon, and S. N. Sridhar: 1978, 'On the Acquisition of
Subjecthood',in Studiesin the LinguisticSciences8, 42-71.
Culicover,Wasowand Akmajian(eds.), 1977: FormalSyntax,Academic Press,New York.
Curme,G. O.: 1922, A Grammarof the GermanLanguage, MacMillan,New York.
Czepluch,H.: 1982, 'Case Theory and the Dative Construction',The LinguisticReview 2,
1-38.
Dowty, David: 1982, 'GrammaticalRelationsand MontagueGrammar',in P. Jacobsonand
G. Pullum(eds.), The Natureof SyntacticRepresentation,
Reidel, Dordrecht,pp. 79-130.
Dryer,Matthew:1983, 'IndirectObjectsin KinyarwandaRevisited',in D. Perlmutter(ed.),
Studiesin RelationalGrammar,Universityof Chicago Press, Chicago, pp. 129-140.
Einarsson,Steffin:1945, Icelandic,Grammar,Texts, Glossary,Johns Hopkins,Baltimore.
Faarlund,Jan Terje: 1980, 'Subject and Nominative in Old Norse', ScriptaIslandica:
Islandskasallskapetsarsbok31, 65-73.
Freidin,Robert and LeonardBabby: 1982, 'On the Interactionof Lexical and Structural
Properties',unpublishedpaper, Cornell University,Ithaca.
:1983, 'On the Interactionof Lexical and SyntacticProperties:Case Structurein
Russian',CornellWorkingPapersin Linguistics,No. 6.
Fribjonsson,J6n: 1978, A Coursein ModernIcelandic,TimaritibSkak, Reykjavik.
Gary, J. 0. and E. Keenan: 1977, 'On Collapsing GrammaticalRelations in Universal
Grammar',in P. Cole and J. Sadock(eds.), GrammaticalRelations,Syntaxand Semantics
8, AcademicPress,New York, pp. 83-120.
Gazdar,Geraldand Ivan Sag: 1981, 'Passivesand Reflexives in Phrase StructureGrammar', in J. A. G. Groenendijk,T. N. V. Janssen,and M. B. J. Stokhof (eds.), Formal
Methodsin the Studyof Language, AmsterdamMathematicalCenter Tract No. 135.
Grimshaw,Jane: 1982, 'RomanceReflexive Clitics', in Bresnan(ed.), pp. 87-148.
Harris, Alice: 1975, 'On the Mechanism of Syntactic Rule Loss', unpublishedpaper,
HarvardUniversity,Cambridge.
Holmberg,Anders: 1985, 'Case and Word Order in Icelandic and Swedish',unpublished
manuscript,Academyof Finland;paperpresentedat GLOW,April, 1985.
Kaplan,Ron and Joan Bresnan:1982, 'A FormalSystemof GrammaticalRepresentation',
in J. Bresnan (ed.), The Mental Representationof GrammaticalRelations, MIT Press,
Cambridge,pp. 173-281.
Kayne, Richard,S.: 1981, 'ECP Extensions',LinguisticInquiry12, 93-133.
Keenan,E.: 1976, 'Towardsa UniversalDefinitionof "Subject"',in C. N. Li (ed.), Subject
and Topic,Academic Press, New York, pp. 303-333.
Levin, Lori: 1981, 'Lexical Representationsof Quirky Case in Icelandic', unpublished
paper,MIT, Cambridge.
1985, 'Operationson Lexical Forms: Unaccusative Constructionsin Germanic
Languages',Ph.D. dissertation,MIT, Cambridge.
and Jane Simpson: 1981, 'Quirky Case and Lexical Representationof Icelandic
Verbs', ChicagoLinguisticsSociety17, 185-196.
Lieber,Rochelle: 1979, 'The English Passive:An Argumentfor HistoricalRule Stability',
LinguisticInquiry10, 667-688.
Lightfoot, David: 1979, Principlesof Diachronic Syntax, Cambridge University Press,
Cambridge.
:1981, 'The Historyof Noun PhraseMovement',in Bakerand McCarthy(eds.), The
LogicalProblemof Language Acquisition,MIT Press,Cambridge,pp. 86-119.
Lockwood,W. B.: 1964, An Introductionto ModernFaroese,Munksgaard,Copenhagen.
Maling,Joan: 1980, 'Inversionin EmbeddedClausesin ModernIcelandic',Islensktmdl og
almennmdlfraebi
2, 175-193.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions
CASE AND GRAMMATICAL FUNCTIONS
483
1982, 'Clause-BoundedReflexivesin Modem Icelandic',paper read at the Trondheim Workshopon ScandinavianSyntax, June, 1982; to appear in L. Hellan and K.
Koch-Christensen(eds.), Topicsin ScandinavianSyntax,Reidel, Dordrecht.
- and Annie Zaenen: 1978, 'The Nonuniversalityof a Surface Filter', Linguistic
Inquiry9, 475497.
Marantz,Alec: 1981, On the Natureof GrammaticalRelations, Ph.D. dissertation,MIT;
revised version publishedby MIT Press,Cambridge,1984.
Marchand,H.: 1951, 1974, 'The Syntactic Change from Inflectional to Word Order
Systemsand Some Effects of this Change on the Relation Verb/Object in English. A
Diachronic-SynchronicInterpretation',in Studies in Syntax, Wilhelm Finks Verlag,
Munchen,pp. 98-117.
Ostler, N.: 1979, Case Linking:A Theoryof Case and VerbDiathesis, unpublishedPh.D.
dissertation,MIT, Cambridge.
Perlmutter,David: 1978, 'ImpersonalPassivesand the UnaccusativeHypothesis',BLS 4,
Universityof California,Berkeley.
(ed.): 1983, Studiesin RelationalGrammarI, Universityof Chicago Press,Chicago.
and Paul Postal: 1983, 'Some Proposed Laws of Basic Clause Structure', in
Perlmutter(ed.), pp. 81-128.
Rappaport,Malka: 1983, 'On the Natureof Derived Nominals',in Levin, Rappaportand
Zaenen (eds.), Papersin Lexical-FunctionalGrammar,IULC, pp. 113-142.
Rognvaldsson,Eirikur:1982a,'Orbar6bogf;ersluri islensku',('WordOrderand Movement
in Icelandic')cand. mag. thesis, Universityof Iceland.
1982b, 'We Need (Some Kind of a) Rule of ConjunctionReduction', Linguistic
Inquiry13, 557-561.
:1984, 'IcelandicWord Orderand Pab-Insertion',WorkingPapersin Scandinavian
Syntax8, Universityof Trondheim.
Thrafinsson,Hoskuldur: 1976, 'Some Arguments against the Interpretive Theory of
Pronounsand Reflexives',HarvardStudiesin Syntaxand Semantics2, 573-624.
in Icelandic,Garland,New York.
: 1979, On Complementation
: 1984, 'On Auxiliaries,Aux and VPs in Icelandic',to appearin Hellan and Koch
Christensen(eds.), Topicsin ScandinavianSyntax,Reidel, Dordrecht.
and the Lexicon', in Culicover, et al. (eds.), pp.
Wasow,Thomas: 1977, 'Transformations
327-360.
1980, 'Major and Minor Rules in Lexical Grammar',in Hoekstra et al. (eds.),
LexicalGrammar,Foris, Dordrecht,pp. 285-312.
Zaenen,Annie: 1980, ExtractionRules in Icelandic,Ph.D. dissertation,HarvardUniversity,
Cambridge;to be publishedby Garland,New York.
: 1983, 'Double Objects in Kikuyu?',CornellWorkingPapersin LinguisticsNo. 5,
pp. 199-206.
and Joan Maling: 1984, 'Unaccusative,Passive and QuirkyCase', in Cobler et al.
(eds.), Proceedingsof 3rd West Coast Conferenceon FormalLinguistics,StanfordUniversity,pp. 317-329.
Received 5 December 1984
Revised 7 June 1985
Maling
Dept. of Psychology
BrandeisUniversity
Waltham,MA 02254
U.S.A.
This content downloaded from 137.99.78.63 on Mon, 15 Sep 2014 12:55:50 PM
All use subject to JSTOR Terms and Conditions