Why is there NOM - NOM but
no ERG - ERG?∗
Yining Nie
New York University
This paper examines a typological gap in double case arrays, where arguments of a transitive verb exhibit the same
case marking. While languages permit NOM - NOM, ACC ACC and ABS - ABS case arrays, ERG - ERG is unattested. I
call this gap the Single Ergative Generalisation (SEG). I
show that existing approaches to ergative case fail to capture the SEG and then provide an account of the SEG that
requires verbal arguments to be introduced by syntactic
heads which are interpreted based on their syntactic context. I propose that ergative case is strictly configurational:
the assignment of ergative case to an argument depends on
the content of its sister constituent.
1
Introduction
This paper investigates a typological gap in what I will refer to as double case arrays, in which the
internal and external arguments of a transitive clause exhibit the same case morphology. Various languages
permit NOM - NOM, ACC - ACC and ABS - ABS case arrays, but ERG - ERG constructions are entirely unattested.
The aim of the paper is to show that this typological gap should be taken seriously in our theory of ergativity
and provide a mechanism for ergative case assignment that is able to derive this gap.
Let us begin with some attested double case arrays. Double nominatives, for example, are permitted
in some stative transitive constructions in Japanese, as shown in (1). Multiple nominatives similarly occur
in the Korean passive construction in (2b); these arguments are accusative in its active counterpart (2a).
(1)
(2)
∗
a.
Hiromi-ga syuwa-ga
deki-ru.
H-NOM sign.language-NOM capable-PRES
‘Hiromi can use a sign language.’
b.
John-ga Mary-ga suki-da.
J-NOM M-NOM fond-PRES
‘John is fond of Mary.’
a.
Cheli-ka Mary-lul panci-lul senmwul-ul hay-ss-ta.
C-NOM M-ACC ring-ACC gift-ACC
do-PAST- DECL
‘Cheli presented Mary a ring.’
b.
Mary-ka panci-ka senmwul-i toy-ess-ta.
M-NOM ring-NOM gift-NOM do.PASS - PAST- DECL
‘Mary was presented with a ring.’
(Koizumi, 2008: 141, 144)
(Wechsler & Lee, 1996: 635)
Many thanks to Bronwyn Bjorkman, Jessica Coon, Stephanie Harves, Richard Kayne, Itziar Laka, Sabine Laszakovits, Julie Anne Legate, Kuo-Chiao Lin, Shih-Yueh Jeff Lin, Alec Marantz, Diane Massam, Jim Wood and the
participants of NYU Morph Beer and MOTH 2016 for their valuable discussion.
Toronto Working Papers in Linguistics (TWPL), Volume 36
c 2016 Yining Nie
Y INING N IE
In addition to having double nominatives, Icelandic exhibits double accusatives as a result of quirky case
assignment, as shown in (3). Icelandic ACC - ACC verbs, such as vanta ‘need, lack’, do not inflect for person
or number.
(3)
Mig
vanta-r
pening-a.
Me.ACC need-PRES money-PL . ACC . INDEF
‘I need money.’
(Jónsson, 2003: 140)
We find double case arrays in ergative languages as well. In Shipibo, for instance, transitive subjects normally receive overt ergative marking, as shown in (4). However, desiderative (5) and reciprocal (6) constructions are ABS - ABS, despite being transitive; absolutive arguments in Shipibo are morphologically unmarked
(Baker, 2014).
(4)
Ochibi-baon-ra bake
natex-kan-ke
dog-PL . ERG - PRT child(. ABS ) bit-PL . SUBJ - PRF
‘The dogs bit the child.’
(5)
José-ra
yapa
keen-ai
J(. ABS )- PRT fish(. ABS ) want-IMPF
‘José wants some fish.’
(6)
Ja-bo-ra
piti
meni-anan-ke
they-PL (. ABS )- PRT fish(. ABS ) give-RECIP - PRF
‘They gave fish to each other.’
(Baker, 2014: 344)
(Baker, 2014: 346–347)
Given that double nominative, accusative and absolutive case arrays are all attested, we might expect double
ergative constructions to be possible as well. However, I have not found a single instance of ERG - ERG in
any language. This result is confirmed by Jessica Coon (p.c.), Julie Anne Legate (p.c.) and Diane Massam
(p.c.), among others. No language marks ergative case on more than one argument of the same verb.
The wholesale lack of ERG - ERG constructions cross-linguistically leads me to posit the following
descriptive generalisation:
(7)
Single Ergative Generalisation (SEG)
A syntactic clause may contain at most one ergative argument.
Being thus far exceptionless, the SEG can be used as a criterion for evaluating the adequacy of existing
analyses of ergativity: our theory of ergative case assignment must exclude the possibility of producing
ERG - ERG arrays. I set aside the definition of ‘clause’ in (7) for now but will note here that the SEG may not
apply in cases involving, for example, predicates with clausal complements.
Current approaches to ergative case assignment generally fall into one of two camps: inherent or
structural. Inherent ergative case tends to be idiosyncratic and depends on the properties of particular predicates (Mahajan, 1990; Woolford, 1997, 2006; Laka, 2006; Massam, 2006; Legate, 2008; Coon & Preminger,
2013). The other camp takes ergative to be a purely structural case, assigned to an argument in a particular
structural configuration with its case-assigning head (Marantz, 1991; Bittner & Hale, 1996; Béjar & Massam, 1999; Bobaljik & Branigan, 2006; Bruening, 2007; Rezac et al., 2014). In this paper, I will show that
both approaches fail to account for the lack of ERG - ERG constructions cross-linguistically.
The paper is organised as follows. In §2, I argue that an inherent case view of the ergative predicts
the possibility of ERG - ERG in causatives of transitives. If, on the other hand, ergative case is structural, then
we would expect it to be subject to the same syntactic operations that multiply assign other structural cases,
leading to multiple ergatives (§3). A dependent case analysis also fails with regards to the SEG if we allow
dependent case to be calculated in more than one case domain (§4). In view of these undesirable predictions,
2
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
in §5 I sketch a novel account of ergative case assignment that is capable of deriving the SEG. Adapting
Wood & Marantz’s (2015) configurational approach to the thematic interpretation of external arguments, I
propose that ergative case is assigned to an argument on the basis of the content of the constituent it merges
with. §6 concludes.
2
Ergative case is not inherent
An inherent case is assigned by particular predicates and typically coincides with a specific set of
theta-roles, as captured by the Inherent Ergative Hypothesis (adapted from Rezac et al., 2014: 1275):
(8)
Inherent Ergative Hypothesis
Ergative case/agreement reflects inherent Case: it depends on the relationship(s) between an argument
and its lexical predicate.
According to the inherent case approach, these lexical predicates license an ergative-assigning functional
head, normally assumed to be Voice (Massam, 2006; Legate, 2012; Coon & Preminger, 2013). Split-S
languages such as Basque and Hindi show the relevance of particular lexical predicates in the assignment
of ergative case. As shown in these examples from Basque, unaccusative subjects in split-S languages are
absolutive (9) but unergative subjects are marked ergative (10).
(9)
Txalupa
hondora-tu da.
boat.DEF - ABS sink-PERF is
‘The boat sank.’
(10) Gizon-a-k
aharrausi egi-n
du.
man-DEF - ERG yawn
do-PERF has
‘The man yawned.’
(Laka, 2006: 376–377)
The crucial prediction made by the inherent case view is that ergative-assigning predicates should consistently assign ergative case (Mahajan, 1990, 2012; Woolford, 1997, 2006; Laka, 2006; Massam, 2006, 2009;
Legate, 2008, 2012). In this section, however, I show that ergative case assignment is not exhaustively determined by the lexical properties of the predicate.
2.1
Derived ergative subjects
The inherent case intuition of the rigidity case patterns was also captured by Marantz (1991/2000: 14),
who argues for a different approach to ergative case but nonetheless observes that only particular types of
subjects may be ergative:
(11) Marantz’s Ergative Generalisation
Ergative case does not occur on an argument moved into a non-thematic subject position.
The generalisation in (11) states that derived subjects cannot be ergative. This includes, for example, the
derived internal argument subjects of unaccusative verbs; these are predicted to never receive ergative case
marking in any language. Similarly, we would expect the inability of inherent ergative case to be removed
from arguments in derived positions.
However, evidence from Shipibo suggests that derived subjects may indeed be marked ergative (Baker,
2014). For example, while the subject of the unaccusative verb joshin ‘ripen’ is absolutive in (12a) as expected, it is promoted to ergative with the addition of an applied argument, as shown in (12b).
3
Y INING N IE
(12) a.
b.
Kokoti-ra
joshin-ke.
fruit(. ABS )- PRT ripen-PRF
‘The fruit ripened.’
Bimi-n-ra
Rosa
joshin-xon-ke.
fruit-ERG - PL Rosa(. ABS ) ripen-APPL - PRF
‘The fruit ripened for Rosa.’
(Baker, 2014: 345, 346)
The appearance of ergative case on a theme argument in (12b) therefore appears to be in direct contradiction
to Marantz’s Generalisation as well as the predictions made by the inherent case approach.
The same case contrast is exhibited by the simple unaccusative in (13a) and its adversative (13b).
(13) a.
b.
Maria-ra
mawa-ke.
Maria(. ABS )- PRT die-PRF
‘Maria died.’
Nokon shino-n-ra
e-a
mawa-xon-ke.
my.GEN monkey-ERG - PRT me-ABS die-APPL - PRF
‘My monkey died on me.’
(Baker, 2014: 345, 366)
Again, the addition of applicative structure results in ergative marking on the unaccusative subject. These
Shipibo examples demonstrate that the assignment of ergative case is dependent not on theta-roles or lexical
properties of predicates but on the transitivity of the clause. The intransitive unaccusative constructions in
(12a) and (13a) have absolutive subjects while their more ‘transitive’ counterparts in (12b) and (13b) have
ergative subjects. An inherent case approach to the ergative has no way to account for these case alternations.
2.2
Non-ergative agents
Another case alternation of sorts can be found between transitive constructions and their causatives.
In Niuean, for instance, the agent of the transitive predicate totō ‘hold’ is normally marked ergative, as in
(14a). However, this agent surfaces as absolutive when embedded in a causative construction (14b), and the
causer then becomes the ergative argument. (15) provides another example of this causative case alternation.
The morpheme e marks ergative on pronouns and absolutive on common nouns in Niuean (Massam, 2009).
(14) a.
Ne
totō e
ia e
kapiniu.
hold ERG . P 3 SG ABS . C cup
‘He held the cup [in his hand].’
PAST
b.
(15) a.
b.
Kua faka-totō aki
e
ia e
kato e
tama haaku.
PERF FAKA -hold INSTR ERG . P 3 SG ABS . C basket ABS . C child 1 SG . GEN
‘She made my child hold the basket.’
Kua teka e
ia e
polo.
PAST bowl ERG . P 3 SG ABS . C ball
‘He bowled the ball.’
Kua faka-teka nı̄
e
ia haana
a tama he
tā
kilikiki.
PERF FAKA -bowl EMPH ERG . P 3 SG 3 SG . GEN LIG child LOC . C match cricket
‘He made his own child bowl in the cricket game.’
(Massam, 2009: 127)
4
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
If ergative case were inherent, then we would expect it to be assigned consistently to the agentive subjects
of predicates like ‘hold’ and ‘bowl’, independently of whether it is embedded in a causative construction.
Ergative case would also appear on the causer, predicting ERG - ERG in the causative of a transitive. However,
(14b) and (15b) show that the agent of the lower transitive clause (the causee) cannot be marked ergative in
the presence of an ergative causer.
As predicted by the SEG, therefore, we see no recursion of ergative structure in the causative constructions, regardless of what predicate appears in the lower clause. If inherent cases are assigned consistently to
the arguments of particular predicates, then ergative case cannot be inherent.
3
Ergative case is not structural
The main alternative to the inherent case view of the ergative is the structural case approach, where
ergative case is either assigned via an Agree relation to a functional head or computed post-syntactically
(Marantz, 1991; Johns, 1992; Bobaljik, 1993; Bittner & Hale, 1996; Béjar & Massam, 1999; Bobaljik
& Branigan, 2006; Bruening, 2007; Baker, 2014; Rezac et al., 2014; Coon & Preminger, 2015; Johns &
Kučerova, 2015). In this section, I discuss some syntactic operations which we would expect to apply to
ergative case if it were indeed structural, and show that they predict the possibility of ERG - ERG.
3.1
Multiple Agree
The nominative case is generally taken to be the canonical example of a structural case. Recall that
Japanese has a range of double nominative constructions, like that repeated in (16).
(16) John-ga Mary-ga suki-da.
J-NOM M-NOM fond-PRES
‘John is fond of Mary.’
(Koizumi, 2008: 144)
In order to capture the assignment of nominative case to both of the arguments in the transitive clause,
Hiraiwa (2001) suggests that in some languages, a single probe can Agree with multiple matched goals
simultaneously in the derivation. Multiple Agree operates when a probe feature specified as [+multiple]
enters into an Agree relation with all of the matching goals within an ‘accessible’ domain simultaneously.
In Japanese NOM - NOM constructions, then, Hiraiwa proposes that probe φ-feature on finite T can
enter into a Multiple Agree relation with the goal φ-features of the internal and external arguments simultaneously. The tree in (17) shows how Multiple Agree would operate in the derivation of example (16).
(17) Multiple Agree in the derivation of (16)
TP
T
[ PRES ]
[uφ]
VoiceP
DP1
John
[iφ]
Voice’
vP
Voice{D}
v
√
FOND
v
MULTIPLE AGREE
5
DP2
Mary
[iφ]
Y INING N IE
Multiple Agree then permits the assignment of nominative case to both Agreeing goals. There is presumably
also an EPP feature on finite T which allows both the internal and the external argument to raise to produce
the SOV word order characteristic of Japanese.1
There are numerous gaps to the Multiple Agree reasoning, such as why the internal argument does
not receive accusative case before T merges in the derivation, or why only particular constructions have a
probe φ-feature that is specified as [+multiple]. The issue that is most relevant for us, however, is whether
Multiple Agree can apply to ergative case. Hiraiwa (2001) assumes that nominative case is a structural case
assigned by finite T. Under the Multiple Agree approach, the Korean double accusative array in example
(2a) is presumably assigned structurally as well. If ergative case were similarly structural, as many have
argued, nothing in the application of Multiple Agree would prevent the assignment of double ergative case,
in parallel to the double nominative and accusative arrays found in Japanese and Korean. If structural cases
participate in Multiple Agree, then ergative case cannot be structural given the SEG.
3.2
Multiple case checking
Like Hiraiwa (2001), Bobaljik & Branigan (2006) pursue a structural case analysis and propose that
finite T may check features on multiple goals. Bobaljik & Branigan in fact take this approach even further,
arguing that in Chukchi, T must be able to assign two different cases.
Like other ergative languages, Chukchi differentiates intransitive (18a) and transitive (18b) subjects.
(18) a.
b.
G@m
t@-k@tG@ntat-GPak
I(. ABS ) 1 SG . SUBJ -run-1 SG . SUBJ
‘I ran.’
PaatSek-a kimitP-@n ne-nìPetet-@n
youth-ERG load-ABS 3 PL . SUBJ -carry-3 SG . OBJ
‘(The) young men carried away the load.’
(Skorik, 1977: 19; Kozinsky et al., 1988: 652)
The intransitive clause in (18a) marks its subject as absolutive, and the verb agrees with the subject. In (18b),
the transitive subject is marked ergative, and the verb agrees with both arguments. As is typical of ergative
languages, Chukchi also has an antipassive voice alternation, marked with the antipassive morpheme -ine.
Compare, for example, the active transitive clause (18b) with its antipassive counterpart in (19).
(19) PaatSek-@t ine-nìPetet-GP@t
kimitP-e
youth(. ABS ) ANTI -carry-3 PL . SUBJ load-INSTR
‘(The) young men carried away a load.’
(Kozinsky et al., 1988: 652)
In the antipassive, the object appears in an oblique case, but the clause looks otherwise intransitive: the verb
displays agreement only with the subject, which is marked absolutive. Furthermore, transitive clauses are
SOV, but the antipassive is generally SVO.
Finally, Chukchi also has a construction known as the spurious antipassive, used for clauses with
“inverse” agreement, where the object outranks the subject on a person-number hierarchy. The spurious
antipassive is so called because it has antipassive morphology (-ine) and intransitive agreement yet displays
transitive case marking. (20) provides an example.
1
SOV word order cannot be attributed solely to the effects of Multiple Agree, as it is also the default word order
canonical NOM - ACC constructions, in which Multiple Agree does not operate.
6
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
(20) @-nan G@m
ine-ìPu-GPi
he-ERG I(. ABS ) ANTI -see-3 SG . SUBJ
‘He saw me.’
(Skorik, 1977: 44)
In (20), the verb is in the antipassive form, agreeing only with the subject, but arguments are case-marked
as in regular transitive clauses.
According to Bobaljik & Branigan (2006), the ergative case system in Chukchi differs from nominativeaccusative systems in that the Voice head in Chukchi cannot check accusative case. In the active transitive
and spurious antipassive constructions, therefore, both the subject and object must raise to the TP domain
for checking. T then checks and assigns ergative case to the subject and absolutive to the object, under the
assumption that ergative case tracks markedness, which apparently correlates with height in their approach
(see Marantz, 1991). Bobaljik & Branigan’s multiple case checking account is schematised in (21), which
can be a representation of either the active transitive or the spurious antipassive.
(21) Multiple case checking
TP
DP1
He
ERG
T’
DP2
I
ABS
T’
T
[ PAST ]
VoiceP
<DP1 >
Voice’
vP
Voice
MULTIPLE CHECKING
v
√
SEE
<DP2 >
v
Bobaljik & Branigan argue for a similar analysis for French faire-causatives in which the faire head
can check both dative and accusative case. Their multiple checking approach is therefore even less constrained than Hiraiwa’s (2001) Multiple Agree, as two (or perhaps more) different cases can be assigned at
a single step of the derivation. Again, if structural cases can be multiply checked, then ergative case cannot
be a structural case.
The existence of double case arrays suggests that mechanisms like Multiple Agree and multiple case
checking may be needed in the grammar. However, if we allow (i) the multiple assignment of structural
nominative and accusative case and (ii) the differential assignment of structural ergative and absolutive
case, we should also expect to be able to multiply assign structural ergative case. However, ERG - ERG
constructions are unattested. The SEG therefore indicates that ergative case cannot be structural.
4
Ergative case is not dependent
Another approach to ergative case along similar lines to the structural approach is the dependent case
view. A dependent case is one that is assigned to an argument in the correct syntactic configuration relative
to another argument (Marantz, 1991/2000). Proponents of the dependent case view assume that, in at least
some languages, the assignment of ergative and accusative case on one DP is dependent upon the existence
of another DP in the clause (Baker & Vinokurova, 2010; Baker, 2014; Coon & Preminger, 2015). Marantz
7
Y INING N IE
(2000: 24) suggests that different types of cases are ranked in a case hierarchy, listed in descending rank
order in (22).
(22) Case realisation disjunctive hierarchy
1. Lexically governed case
2. Dependent case (accusative and ergative)
3. Unmarked case (environment-sensitive, e.g. genitive)
4. Default case
Lexically-assigned cases, such as those assigned by prepositions or quirky case in Icelandic, take precedence. Dependent case and the unmarked cases follow. (23) slightly modifies Baker’s (2014: 342) summary
of Marantz’s mechanism for dependent case assignment.
(23) If there are two distinct DPs in the same domain, then:
1. Mark the lower one with dependent case (accusative) and/or
2. Mark the higher one with dependent case (ergative).
If at this point any DP remains unspecified for case, then it surfaces as nominative or absolutive by default.
4.1
Soft and hard phases
Given just the heuristics in (23), however, a number of questions arise, namely (i) how is ‘height’
determined and (ii) within what domain? A sensible way to characterise the relative syntactic height of
arguments in terms of c-command (Baker & Vinokurova, 2010); argument A is higher than argument B if
A c-commands B. The domain of evaluation of the argument height, meanwhile, could conceivably be the
whole clause, i.e. CP or TP.
At first glance, the dependent case approach seems to be good news for the SEG: in a monotransitive
construction, only one argument receives ergative case. However, it is unclear what happens in ditransitive constructions such as the double object construction sketched in (24); the goal DP2 is assumed to be
introduced by a low applicative head (Pylkkänen, 2008). In this tree, what case should DP2 be assigned?
(24) Clausal dependent case in double object constructions
VoiceP
DP1
VoiceP
ERG
vP
Voice
v
ApplP
DP2
?
ApplP
Appl
DP3
ABS
While DP2 c-commands DP3 , suggesting that DP2 should receive ergative marking, DP2 is also c-commanded
by DP1 , which suggests absolutive case. If DP2 is marked as ergative, however, then we would be left with
an ERG - ERG - ABS case array, which is unattested according to the SEG. We can avoid this prediction if
we assume that ergative case may only be assigned once and only when all argument structural material is
merged. Thus the dependent approach to ergative case requires the assumption of a single case assignment
domain per clause (Baker & Vinokurova, 2010; Coon & Preminger, 2015).
8
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
However, Baker & Vinokurova (2010) and Baker (2014) propose that in different languages, VoiceP
may be either a ‘hard phase’ whose vP complement is invisible for later operations, or a ‘soft phase’ whose
complement remains visible in the next stage of the derivation. In languages for which VoiceP is a soft
phase, ergative case is not assigned before the complement of Voice is spelled-out but during the CP phase,
which is always a hard phase. Thus ergative case assignment is “keyed” to the spell-out of the TP domain but
not the vP domain, so that ergative case may be assigned maximally (but not minimally, as will be shown)
once per clause, despite multiple spell-outs.
Baker (2014) claims that VoiceP must be a soft phase in Shipibo since the language has a class of
double absolutive verbs such as keen ‘want’, as shown in (25).
(25) José-ra
yapa
keen-ai
J(. ABS )- PRT fish(. ABS ) want-IMPF
‘José wants some fish.’
(Baker, 2014: 347)
According to Baker, these kinds of verbs take two vP-internal arguments: an experiencer or source subject
and a theme object. In order to ensure that the experiencer is not marked ergative despite c-commanding
the theme, Baker suggests that no structural case (Shipibo does have not a structural dative case) is assigned
when vP is spelled-out as the complement of Voice. Furthermore, the higher argument crucially does not
receive ergative case even after it moves to Spec-TP because movement does not create any new c-command
relations; since the higher argument already c-commanded the lower one within the vP and this c-command
relation is identical after movement, no ergative case is assigned.
Shipibo does assign ergative case to the highest argument if it is generated outside the VP domain
and/or creates a new c-command relation with the another argument. The double object construction in (26)
shows that ergative case is assigned to the highest argument (in the TP domain) and ABS - ABS to the lower
arguments (within the vP domain). Absolutive, being a default case, is only assigned after the more marked
cases have been realised.
(26) Maria-nin-ra José
piti
meni-ke.
M-ERG - PRT J(. ABS ) fish(. ABS ) give-PRF
‘Maria gave José fish.’
(Baker, 2014: 352)
From these Shipibo examples, we can see that keying ergative case assignment to only one phase per clause
satisfies the SEG.
Baker and colleagues make the crucial assumption that VoiceP is a soft phase in Shipibo. Baker
(2014) suggests that the permeability of the VoiceP phase might be parameterised across languages, so that
VoiceP is a soft phase in Shipibo but could be a hard phase in another language. Yet if VoiceP is a hard
phase, then we would predict the possibility of ERG - ERG - ABS in double object constructions, for example.
Indeed, all of the existing dependent case analyses that I have come across posit a single case assignment
domain for ergative case.
4.2
Possessives
In fact, multiple case domains might be desirable for ergative/genitive case syncretisms, which are
frequently found across ergative languages, such as in Kaqchikel (27) and Nez Perce (28).
(27) R-ixjayil
nu-xb’al
x-∅-pe.
ERG .3 SG -wife GEN / ERG .1 SG -brother PERF - ABS .3 SG -come
‘My brother’s wife came.’
9
(Imanishi, 2014: 35)
Y INING N IE
(28) ’ip-nim
lawtiwaa-ma-pim pe-’pt’e-six-∅
haacwal-a.
3 SG - GEN / ERG friend-PL - ERG .1 SG 3/3-hit. IMPERF. PL - PRES boy-OBJ
‘His friends are hitting the boy.’
(Deal, 2013: 410)
Languages which exhibit case marking on possessors are usually said to have a genitive case. In both (27)
and (28), the external argument of the clause and the possessor embedded within the external argument are
marked with the same morphology. The result is that the possessor argument features case marking that
is syncretic with ergative case morphology in the language. It is therefore possible that the ergative and
genitive cases in Kaqchikel and Nez Perce may be assigned in the same or similar ways. If we take the
ergative/genitive case to be dependent, then there must be an additional cyclic domain in which ergative
case may be assigned.
Based on this and similar evidence, Imanishi (2014) proposes that ergative case in Kaqchikel is assigned upon the spell-out of each of the CP, VoiceP and DP phases. However, if multiple case domains are
permitted, then we would expect ERG - ERG arrays, in violation of the SEG. It is possible to stipulate that
ergative case cannot be keyed to both CP and VoiceP. Yet limiting ergative case assignment to a single phase
seems to weaken the dependent case approach to the point where it is no longer in practice any different
from an inherent case approach in which a particular functional head assigns ergative case.
So far I have discussed the SEG only in relation to verbal arguments. If ergative/genitive case syncretisms are to be taken seriously as true ergatives, then this apparent ERG - ERG pattern must be addressed.
I return to the status of possessors in §5.5.
This section has shown that nothing in the dependent case view of ergative case prevents the assignment of multiple ergatives. This approach therefore fails with respect to the SEG.
5
Configurational case
My analysis of the ergative case dovetails with Wood & Marantz’s (2015) configurational approach
to the thematic interpretation of external arguments. In the Government and Binding/Minimalism tradition,
theta-roles are assigned by a collection of functional heads (e.g. Voice, Appl, p) at varying places in the
syntactic structure. Wood & Marantz attempt to unify these functional heads into a single external argumentintroducing head i*, which “generally assigns to the second constituent it merges with the θ-role implied by
the first constituent it merges with” (2015: 33). That is, XP in (29) is assigned the theta-role implied by YP.
(29) Configurational theta-role assignment
i*P
XP
i*P
i*
YP
If XP = DP and YP is a vP, for example, the DP would be assigned an agent theta-role, as implied by the
properties of the vP. The functional head i* would be therefore be interpreted as Voice, which typically
introduces an agent. Other configurations result in different interpretations of i* and its external argument,
as shown in (30).
(30) Function of i* as determined by its syntactic context
1. Voice (agents): bare i* that merges with vP
2. High appl (affectees): i* with adjoined prepositional root
3. Low appl (possessors): bare i* that merges with a DP
10
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
Wood & Marantz suggest that theta-roles are not directly assigned by i* in the narrow syntax but computed
post-syntactically at the semantic interface, so that the function of i* is subject to contextual allosemy.
In a similar fashion, I propose that some cases are determined based on syntactic context. The ergative
case is the prime example of what I will call a configurational case:
(31) Configurational case
A case is configurational if its assignment on an argument XP depends on the content of the complement (YP) of the functional head that introduces XP (i*).
Like how XP in (29) is interpreted as an agent if YP contains a vP, ergative case in this approach is assigned
if YP contains a particular category. In fact, I propose very similar syntactic conditions on the spell-out of
ergative case in (32) as those assumed for agent theta-role assignment.
(32) Basic configurational requirement for ergative case
Ergative case is assigned to an argument introduced by a bare i* with a vP complement.
(32) ensures that only a subset of external arguments receive ergative marking: non-applied arguments
whose sister constituent contains a vP. For clarity, I use the label vP to refer only to the projection within
which the internal argument of the verb may be introduced.2
A configurational approach to ergative case is able to derive the SEG because the function of each
argument-introducing head (i* with or without an adjoined root) is determined by its syntactic context. Since
the syntactic context changes each time a new argument is merged, ergative structure cannot ‘recurse’. That
is, ergative case is assigned to an argument whose sister constituent is composed of bare i* plus vP. The
resulting constituent is Voice*P rather than vP, and so no additional material that merges would be in the
correct configuration to receive ergative case. Thus the SEG emerges because the configuration for ergative
case assignment may only arise once in a clause.
The rest of the paper fleshes out the details of this account of ergative case assignment, showing how
it plays out in various argument structural configurations.
5.1
Transitives
Let us see the configurational assignment of ergative case in action, starting with a basic transitive
predicate such as the one sketched in (33). The external argument ‘Rosa’ merges with a bare i* which has a
vP within its complement, and is therefore marked as ergative.
(33) Transitive predicates in ergative languages
Voice*P
DP1
Rosa
ERG
Voice*P
vP
i*=Voice*
vP
√
EAT
v
DP2
fruit
ABS
As shown in (33), the internal argument is assigned absolutive case. Given the possibility of ABS - ABS
constructions in ergative languages, I suggest that the absolutive is a structural case, assigned by an Agree
2
My labels therefore differ from those used in Wood & Marantz (2015), where every external argument introduced by
i* becomes a specifier in an extended vP projection.
11
Y INING N IE
relation to a functional head or by default, viz Marantz’s (1991/2000) case hierarchy as discussed in §4.
This approach accounts for the case pattern of the Chukchi spurious antipassive discussed in §3.2,
where a transitive clause may surface with ERG - ABS case marking but antipassive morphology and intransitive agreement. As Chukchi is a non-split-S language, ergative case assignment on the external argument is
the result of merging its argument-introducing head i* with a vP containing another DP. This is represented
in tree form in (34).
(34) Spurious antipassive in Chukchi
Voice*P
DP1
he
ERG
Voice*P
vP
i*
vP
√
SEE
v
DP2
me
ABS
Despite the antipassive morphology and intransitive agreement pattern of the spurious antipassive, then,
ergative case is assigned as in a regular transitive predicate. A configurational approach to ergativity in fact
predicts the possibility of exactly this kind of system, in which case assignment tracks argument structure
and nothing else.
Because of their similar configurational requirements, ergative marking in this system often appears
on the agentive argument, capturing the frequent association of ergative case with a particular set of thetaroles. However, their conditions must be different. We saw in §2.2, for example, ergative cannot appear
on the subject of the lower transitive clause of a causative, even if the subject is agentive. Furthermore, as
discussed in the next section, some languages assign absolutive case to the agents of unergatives.
5.2
Intransitives
The treatment of unaccusative predicates is straightforward in the configurational system. In unaccusative predicates, the internal argument is introduced directly by v, as shown in (35).3 Given that the
internal argument is not introduced by i* but generated within vP, this unaccusative structure does not fulfill
the configurational conditions for ergative case assignment.
(35) Unaccusative predicates in Basque
vP
v
√
RIPEN
v
DP1
fruit
ABS
The unaccusative subject thus cannot be assigned ergative case but instead receives absolutive case.
The behaviour of unergative predicates differ across ergative languages. While unaccusative subjects
are always absolutive, unergative subjects are ergative in split-S languages like Basque and Hindi but absolutive in non-split-S languages like Shipibo and Inuktitut. The tree in (36) shows that the configurational
3
Wood & Marantz (2015) assume that internal arguments are introduced by v. Their theory could in principle be
modified so that internal arguments are also introduced by i*, but our configurational requirements for ergative case
would then also have to be adjusted.
12
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
requirements for ergative case is satisfied in split-S unergatives, as the external argument is introduced by i*
which has already merged with a vP.
(36) Unergative predicates in Basque
Voice*P
DP1
Rosa
ERG
Voice*P
vP
i*
√
SING
v
For split-S languages like Basque, any argument whose sister constituent meets the basic requirement (bare
i* merged with a vP) receives ergative marking.
In languages like Shipibo, however, intransitive subjects are always absolutive, whether the internal or
external argument of the verb. Therefore non-split-S ergative languages such as Shipibo have an additional
requirement for the assignment of ergative case: i* must merge with a vP containing an internal argument.
(37) Unergative predicates in Shipibo
Voice*P
DP1
Rosa
ABS
Voice*P
vP
i*
√
SING
v
I propose that non-split-S languages require that the features of the internal argument be passed up through
vP in order to be ‘visible’ for evaluating the configurational requirements for assigning ergative case to the
external argument. This feature visibility mechanism will also be needed to capture Shipibo applicatives of
unaccusatives, discussed in §5.4.
The difference between split-S and non-split-S languages therefore lies in their configurational requirements on the assignment of ergative case. In split-S languages like Basque, ergative case is marked
on any external argument introduced by i* and which satisfies the compositional requirements of a vP; this
includes unergative subjects. In languages like Shipibo, on the other hand, the vP must also contain an
internal argument, so that no intransitive predicates qualify for ergative case.
5.3
Causatives
In my approach, the causee in the causative of a transitive clause does not receive ergative marking
because it is not introduced by regular Voice (bare i*). The evidence for this comes from the optionality
of the causee: leaving a causee unexpressed yields existential closure of the argument, while leaving the
external argument of regular Voice unexpressed yields a specific null subject interpretation (Folli & Harley,
2007). I therefore take the causee to be a high applicative, introduced by High Appl (i* with a prepositional
root) and an adjunct to vP such that the resulting constituent is still vP.
Therefore, when the causer is introduced by Voice (bare i*), we have the configuration of a transitive
clause, which satisfies the requirements for ergative case assignment, as seen in the sketch of a Niuean
causative in (38):
13
Y INING N IE
(38) Niuean causative of transitive
Voice*P
DP1
I
ERG
Voice*P
vP
i*=Voice*
vP
DP2
Rosa
vP
i=Appl
ABS
√
v
i*
FAKA
√
HOLD
v
DP3
the basket
The subject ‘I’ in (38) is assigned ergative case because its sister constituent satisfies the ergative configurational requirements. This captures how agents of simple transitive clauses and causers pattern together with
regards to ergative case and how they are introduced syntactically. The lower agent ‘Rosa’, by contrast, is
crucially an applied argument introduced by i* with a prepositional root rather than a bare i* and is therefore
ineligible for ergative case assignment. Thus the configurational approach is able to derive the SEG even in
causatives, which proved problematic for the inherent case approach.
5.4
Applicatives
This account also extends to derived ergative subjects in Shipibo, which we discussed in §2.1. In
Shipibo applicatives of unaccusatives, the theme receives ergative case marking. As shown in (39), I assume
that the theme raises to the specifier of Voice, whose complement is vP, placing it in the correct configuration
for ergative case assignment.
(39) Shipibo applicative of unaccusative
Voice*P
DP2
fruit
ERG
Voice*P
vP
i*=Voice*
vP
DP1
Rosa
vP
i=Appl
ABS
√
FOR
v
i*
√
RIPEN
<DP>
v
In (39), the benefactor ‘Rosa’ is introduced by High Appl (i* with adjoined prepositional root) and is thus
ineligible for ergative case. As in causatives, I assume that applicative structure is adjoined to vP so that the
resulting constituent is also vP. I propose that the features of this applied argument are passed up through
vP—a mechanism that we determined in §5.2 to be independently required to capture unergative case patterns in non-split-S languages. This establishes the necessary configuration for ergative case assignment on
14
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
‘fruit’, which surfaces with ergative morphology.
5.5
Possessors
The proposed configurational approach to ergative case also makes predictions for possessives. I take
possessors to be external arguments introduced by a bare i* that merges with a DP, interpreted as Low Appl,
which relates two DPs to each other.
(40) Ergative possessors
Appl*P
DP1
his
ERG
Appl*P
i*=Appl*
DP2
friend
ABS
Viewing possessors as external arguments provides an explanation for why genitive and ergative case are
often syncretic in ergative languages: ergative case reflects the sensitivity of a language to ‘transitivity’ in
both the verbal and nominal domains.
One concern that arises given frequent ergative/genitive syncretisms in ergative languages is whether
the SEG should also apply within the nominal domain and what is means for there to be only one ergative
argument per ‘clause’. For this reason, Imanishi (2014) appeals to a particular theory of phases, but it is not
clear whether this is necessary or desirable. This ‘clause’ issue is important but is left for further study.
6
Conclusion
In this paper, I argued that, on the empirical basis of the SEG, ergative case cannot be inherent or
structural. I then proposed a new analysis of the ergative, arguing that it is assigned configurationally and
that languages may vary (to a limited extent) as to what these configurational requirements are. My approach
is able to derive the SEG because the function of i* is determined by its syntactic context. Since the syntactic
context necessarily changes each time a new argument is merged, ergative structure cannot ‘recurse’. Thus
the SEG emerges because the configuration for ergative case assignment may only arise once in a clause.
The SEG seems to indicate that ergative case is special, though it may turn out that other cases
are configurational in some languages; genitive case, for example, may be subject to a similar uniqueness
requirement within the DP (Sabine Laszakovits, p.c.). Continuing to explore the parallels between verbal
and nominal ‘transitivity’ in future research should prove fruitful.
References
Baker, M. (1988). Incorporation: a theory of grammatical function changing. Chicago: University of
Chicago Press.
Baker, M. (2014). On dependent ergative case (in Shipibo) and its derivation by phase. Linguistic Inquiry
45, 341–379.
Baker, M.C. & N. Vinokurova. (2010). Two modalities of case assignment in Sakha. Natural Language and
Linguistic Theory 28: 593–642.
Béjar, S. & D. Massam. (1999). Multiple Case Checking. Syntax 2: 65–79.
15
Y INING N IE
Bittner, M. & K. Hale. (1996). The structural determination of Case and agreement. Linguistic Inquiry
27: 1–68.
Bobaljik, J. (1993). Ergativity and ergative unergatives. MITWPL 19: 45–88.
Bobaljik, J.D. & P. Branigan. (2006). Eccentric agreement and multiple case checking. In A. Johns, D.
Massam & J. Ndayiragije (eds.), Ergativity: emerging issues. Amsterdam: Springer, 47–77.
Bruening, B. (2007). On the diagnostics of structural case and the nature of ergative case: a reply to
Woolford 2006. University of Delaware, ms.
Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin, D. Michaels & J. Uriagereka
(eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT
Press, 89–155.
Chomsky, N. (2001). Derivation by phase. In M. Kenstowicz (ed.), Ken Hale: A life in language. Cambridge, MA: MIT Press, 1–52.
Comrie, B. (1978). Ergativity. In W.P. Lehmann (ed.), Syntactic typology: Studies in the phenomenology of
language. Austin, TX: University of Texas Press, 329–394.
Coon, J., P.M. Pedro & O. Preminger. (2012). The role of case in A-bar extraction asymmetries: Evidence
from Mayan. McGill University & Harvard University, ms.
Coon, J. & O. Preminger. (2013). Transitivity in Chol: A new argument for the Split VP Hypothesis.
Proceedings of NELS 46, 1–14.
Coon, J. & O. Preminger. (2015). Split ergativity is not about ergativity. lingbuzz/002509
Deal, A.R. (2013). Possessor raising. Linguistic Inquiry 44: 391–432.
Folli, R. & H. Harley. (2007). Causation, obligation, and argument structure: On the nature of little v.
Linguistic Inquiry 38: 197–238.
Hiraiwa, K. (2001). Multiple Agree and the Defective Intervention Constraint in Japanese. MITWPL, 67–80.
Imanishi, Y. (2014). Default ergative. MIT: PhD dissertation.
Johns, A. (1992). Deriving ergativity. Linguistic Inquiry 23: 57–88.
Johns, A. & I. Kučerova. (2015). Towards an information structure analysis of ergative patterning in the
Inuit language. ling.auf.net/lingbuzz/002430
Jónsson, J.G. (2003). Not so quirky: on subject case in Icelandic. In E. Brandner & H. Zinsmeister (eds.),
New Perspectives in Case Theory. Stanford: CSLI Publications, 129–165.
Koizumi, M. (2008). Nominative object. In S. Miyagawa & M. Saito (eds.), The Oxford Handbook of
Japanese Linguistics, 141–164.
Kozinsky, I.S., V. Nedjalkov & M.S. Polinskaja. (1988). Antipassive in Chuckchee: oblique object, object
incorporation, zero object. In M. Shibatani (ed.), Passive and Voice. Amsterdam: John Benjamins.
Laka, I. (2006). On the nature of case in Basque: structural or inherent? In H. Broekhuis, N. Corver, J.
Koster, R. Huybregts & U. Kleinhenz (eds.), Organizing Grammar: Linguistic Studies in Honor of
Henk van Riemsdijk. Berlin/New York: Mouton de Gruyter, 374–382.
Legate, J.A. (2008). Morphological and abstract case. Linguistic Inquiry 39: 55–102.
Legate, J.A. (2012). Types of ergativity. Lingua 122: 181–191.
Legate, J.A. to appear. The locus of ergative case. In J. Coon, D. Massam & L. Travis (eds.), The Oxford
Handbook of Ergativity. New York: Oxford University Press.
Mahajan, A.K. (1990). The A/A-Bar Distinction and movement theory. MIT: PhD dissertation.
Mahajan, A.K. (2012). Ergatives, antipassives and the overt light v in Hindi. Lingua 122: 204–214.
Marantz, A. (1991/2000). Case and licensing. In Eric J. Reuland (ed.), Arguments and Case: Explaining
Burzio’s Generalization. Amsterdam: John Benjamins, 11–30.
Massam, D. (2006). Neither absolutive nor ergative is nominative or accusative. In A. Johns, D. Massam &
J. Ndayiragije (eds.), Ergativity: Emerging issues. Dordrecht: Springer, 27–46.
Massam, D. (2006). Neither absolutive nor ergative is nominative or accusative. In A. Johns, D. Massam &
J. Ndayiragije (eds.), Ergativity: Emerging issues. Dordrecht: Springer, 27–46.
16
W HY IS THERE NOM - NOM BUT NO ERG - ERG ?
Odriozola Pereira, J.C. (2003). Verb-deriving processes in Basque. ASJU Geh 46, 185–222.
Plank, F. (1979). Ergativity, syntactic typology and Universal Grammar: Some past and present viewpoints.
In F. Plank (ed.), Ergativity: Toward a Theory of Grammatical Relations. New York: Academic Press,
3–36.
Pylkkänen, L. (2008). Introducing arguments. Cambridge, MA: MIT Press.
Rezac, M., P. Albizu & R. Etxepare. (2014). The structural ergative of Basque and the theory of Case.
Natural Language & Linguistic Theory 32: 1273–1330.
Skorik, P.J. (1977). Grammar of Chukchi, vol II: verb, adverb and auxiliary words. Leningrad: Nauka.
Wechsler, S. & Y.-S. Lee. (1996). The domain of direct case assignment. Natural Language and Linguistic
Theory 14: 629–664.
Wood, J. & A. Marantz. (2015). On the interpretation of external arguments. lingbuzz/002487
Woolford, E. (1997). Four-way case systems: Ergative, nominative, objective, and accusative. Natural
Language and Linguistic Theory 15: 181–227.
Woolford, E. (2006). Lexical case, inherent case, and argument structure. Linguistic Inquiry 37: 111–130.
17