Interpreting Case Hierarchy: an RRG-OT

Typology of Case Systems

Wataru Nakamura

University of Electro-Communications

0 Introduction

Much of functional-typological research on case and case marking in 1980s and 1990s has taken place in the context of goals and assumptions laid down in Silverstein (1976, 1980, 1981).1 Silverstein proposed a number of influential hierarchies in these papers, two of which are the NP content hierarchy [NCH] and case hierarchy [CH]. The NCH has not only received a variety of interpretations, e.g. agentivity (Dixon 1994), empathy (DeLancey 1981), topicality (Wierzbicka 1981), but has guided numerous researches on split-ergative case systems and many other morphosyntactic phenomena. In contrast, there has been little examination of the CH from a typological perspective. This sharp contrast strikes me as odd, since the CH plays a not less important role in Silverstein's case theory than the NCH.2

The purpose of this paper is to investigate what the CH has to offer typological study of case and case-marking systems and provide a starting point for understanding the whole scope of Silverstein (1976, 1980). The paper consists of two parts, as shown in (1):

(1) a. Mapping between case (case feature) and case-marking (case form) 3

b. Typology of case systems

Section 1 is devoted to the first issue and explores how the CH constrains the mapping between case features and case forms, with particular focus on four cases, dative, accusative, ergative, and genitive. This is essentially an investigation into case syncretism among those cases. Sections 2-4 are devoted to the second issue and propose a typology of case systems within the framework of Role and Reference Grammar [RRG] (Van Valin 1993), augmented by Optimality Theory [OT] (Prince and Smolensky 1993; see also Archangeli and Langendoen 1997).

1 Case Syncretism

1.1 Unfolding Case Hierarchy

(2) describes case hierarchy, where the upper part represents implicational relationships among propositional and adnominal case forms (i.e. nominative, dative, accusative, ergative, and genitive), while the lower part of the hierarchy represents those among adverbial case forms:

(2) Case Hierarchy (adapted from Silverstein 1980)

Nom : Dat1 <------ {Acc, Erg} <------ Gen Propositional/Adnominal

Dat2 <------ {Loc, Instr ...} Adverbial/Propositional

I will not handle case syncretism among oblique cases including dative, locative, instrumental, and ablative in this paper (see Blansitt 1988 and Croft 1991: Ch.5 for data and discussion).

The upper portion of the CH states that if a language has two propositional case forms, they are nominative and dative (or "straight" and "oblique"), if a language has three propositional case forms, they are nominative, dative, and accusative or ergative, and that if a language has a case form distinct from these four, it is genitive.4 This means that nominative and dative are the two fundamental case-markings of any case system and that three-way or four-way case-marking systems are elaborations on the fundamental contrast.5 What we are seeing in (2) is a progression of gradually more and more differentiated systems comparable to the color term systems in (3):

(3) Implicational Hierarchy of Color Terms (Berlin and Kay 1969):

black yellow gray


< red < < brown <


white green pink

(2) was originally proposed on the basis of Dyirbal (Pama-Nyungan) and Chinookan (Chinookan family) data, especially case alternations they display under antipassivization, but it turns out to be applicable to a much wider range of languages.

There are two ways of unfolding the upper portion of case hierarchy, depending on whether there is a neutralization between dative and genitive case as in (4a) or between ergative and genitive case as in (4b):

(4) a. Nom. (Abs.) ----------------- Dat.

Acc./Erg. ------------------------ Dat.

Gen. -------------------- Dat.

b. Nom. (Abs.) ----------------- Dat.

Acc./Erg. ------------------------ Dat.

Gen. ---------------------- Erg.

The arrows in (4a,b) represent paths of elaboration of case forms.6 To my knowledge, there seems to be few languages which display a neutralization between accusative and genitive case. I leave it for future research why a neutralization between these two case is very rare.7

Let us take (4a,b) as a model of case syncretism among dative, accusative, ergative, and genitive case.8 (5a)-(5e) represent a range of case syncretism licensed by (4a,b):

(5) a. Two Case System (1): Halkomelem (Salish), Old French, Rumanian

Those case systems with two distinct case forms, nominative and dative, which

mark subjects and direct objects put together and oblique NPs, respectively.

b. Two Case System (2): Kabardian (Northwest Caucasian), Yagnob (Iranian)

Those case systems with two distinct case forms, nominative and dative,

which mark genitive case and accusative and/or ergative case (as case

feature) with the same case form as dative case (as case feature).

c. Three Case System: Bengali (Indo-Aryan)

Those case systems with three distinct case forms, nominative, dative, and

accusative or ergative, which mark genitive case (as case feature) with the same case form as dative case (as case feature).

d. Three/Four Case System: Inuit (Eskimo-Aleut), Tagalog (Austronesia) 9

Those case systems with three or four distinct case forms, nominative, dative,

and accusative or ergative, which mark genitive case (as case feature) with the

same case form as ergative case (as case feature).

e. Four Case System: Hindi (Indic), Georgian (South Caucasian)

Those case systems with three or four distinct case forms, nominative, dative,

ergative, and genitive, which mark accusative case (as case feature) with the

same case form as dative case (as case feature).

The order of elaboration of case forms described in (4a,b) holds not only typologically, but also language-internally. (5a)-(5e) stand in contrast to those case-marking systems, illustrated by Dyirbal, which have distinct case forms for nominative, dative, accusative, ergative, and genitive case (Dixon 1972).

1.2 Illustrations

Let me illustrate these many-to-one correspondences. (5a) is illustrated by Halkomelem (Salish), Old French, Rumanian, and Chemehuevi (Ute-Aztecan) (see Blake 1994 for a survey). This type of two case system is normally restricted to lexical NPs. For example, Halkomelem marks lexical NPs with two case forms, nominative and dative, unlike its pronouns, which have four distinct case forms, nominative, accusative, dative, and genitive (Gerdts 1988):

(6) a. ni ms séni„.

AUX walk DET woman:NOM

'The woman walked'.

b. ni cám kwQ nikw kwQ smént.

AUX go.up DET uncle:NOM DAT DET mountain

'Uncle went up into the mountains'.

c. ni q'wl-t-s Q séni„ tQ scé.tn.

AUX bake-TR-ERG:3 DET woman:NOM DET salmon:NOM

'The woman baked the salmon'.

d. ni cn q'wáqw-t kwQ n„-sápl-.

AUX NOM:1 club-TR DAT DET GEN:2-shovel-PAST

'I hit him with your shovel'.

e. ni cn q'wál kwQ n-tél.

AUX NOM:I speak DAT DET GEN:I-money

'I spoke about my money'.

f. kwQ púkw -í' John

DET book-GEN:3 DAT John

'John's book'

There is no ditransitive verb in Halkomelem, since it allows only two syntactic arguments within a single core (Gerdts 1992). (6a)-(6f) show that the unmarked lexical NPs correspond roughly to subjects and direct objects, while the case form marks all oblique NPs, arguments or adjuncts. Old French also has this nominative-oblique case system in lexical NPs, but it displays a three-way case system in masculine demonstratives (cil/cel/celui 'that', cisti/cest/cestui 'this') and third-person clitic pronouns (Blake 1994: 194).10

(5b) is illustrated by Kabardian (Northwest Caucasian). First, consider (7a)-(7d):

(7) a. '-m s-r fz-m jrjtáhs.

man-DAT horse-NOM woman-DAT (NOM:3)-IO-ACT-gave

'The man gave the horse to the woman'.

b. '-m s-r j-wh'áhs.

man-DAT horse-NOM (NOM:3)-ACT-killed

'The man killed the horse'.

c. ha-r zás-m mabáhna.

dog-NOM night-DAT (NOM:3)-bark

'The dog barks at night'.

d. '-r fz-m náxra naxzs.

man-NOM woman-DAT older (NOM:3)-is

'The man is older than the woman'.

The above examples are from Kuipers (1962). In contrast to the nominative case suffix r, which may appear only once per clause, the case suffix m may appear multiple times (Smith 1992). What is striking about Kabardian is that the case form m marks transitive subjects (6a,b), time adjuncts (7c), and "object of comparison" (7d) (which corresponds to than in English) as well as recipients (7a). The case suffix m also marks possessors (Blake 1994: 158). (8) describes these correspondences between case features and case forms in Kabardian:

(8) Case Feature Case Form





There is no accusative case form in Kabardian.

Another example of (5b) is Yagnob (Iranian), which has a wider range of neutralization than Kabardian; the same case form marks ergative (subjects in ergative constructions), accusative (specific direct objects in nominative-accusative constructions), dative, and genitive case (Comrie 1981). (9) describes these correspondences between case features and case forms:

(9) Case Feature Case Form






(5c) is a case system with three distinct propositional case forms, nominative, dative, and accusative or ergative, which marks genitive and dative case (as case feature) with the same case form. This three-way case-marking system may be illustrated by Bengali (Indo-Aryan). Here are some examples taken from Klaiman (1980, 1981):

(10) a. se ekti sundor meyeke dekhlo.

he:NOM a pretty girl:ACC saw

'He saw a pretty girl'.

b. taar ghum bhaanglo.

him:DAT sleep broke

'His sleep broke (He awakened).'

c. taar asukh holo.

him:DAT illness became

'He became unwell'.

(10b.c) show that Bengali uses the same case form to mark genitive and dative case (as case feature). Analogous neutralizations are found in Australian languages such as Wambaya, Ngandi, Rembarnga, Alawa, and Djaru as well (Blake 1977: 63-65).

(5d) is a case system with three distinct case forms, nominative, accusative or ergative, and dative, which marks genitive case (as case feature) with the same case form as dative case. Inuit examples are given in (11a,b) (Bok-Bennema 1991, Sadock 1994). They show that Inuit uses the same case form, termed relative case, to mark transitive subjects and possessors:

(11) a. Hansi-p inuit tuqup-paa.

Hansi-ERG people:NOM. kill-DEC:3SG:3SG

'Hansi killed the people'.

b. Hansi-p (Aani-mit) ilinniartin-ner-a

Hansi-ERG Anne-ABL teach-NMLZ-DEC:3SG:SG

'the teaching of Hansi (by Anne)'

We have seen in (8) that Kabardian uses the same case form for all oblique NPs, transitive subjects, and possessors. It has also been reported that the same case form is used for marking ergative and genitive case (as case feature) in other languages including Yup'ik Eskimo (Eskimo-Aleut), Zoque (Mexican), Laz (South Caucasian), Lak (Northeast Caucasian), Ubykh (Northwest Caucasian) (see Allen 1964, Blake 1994, and Croft 1991: 206-212).

(5e) may be illustrated by Hindi (Indo-Aryan), a language with four distinct case forms, nominative, dative, ergative, and genitive. (12a)-(12f) come from Mohanan (1990):

(12) a. ilaa-ne ek bacce-ko uthaayaa.

Ila-ERG one child-DAT lift:PERF

'Ila lifted a child'.

b. ilaa-ne ek haar uthaayaa.

Ila-ERG one necklace:NOM lift:PERF

'Ila lifted a necklace'.

c. niinaa-ne raam-ko kitaab-Ø dii.

Nina-ERG Ram-DAT book-NOM give:PERF

'Nina gave Ram a book'.

d. baccaa kamre-se niklaa.

child:NOM room-INSTR emerge:PERF

'The child emerged from the room'.

e. raajaa-kaa hasnaa mantrii-ko buraa lagaa.

king-GEN laugh-NML minister-DAT bad be.struck:PERF

'The king's laughing made the minister feel bad'.

f. ilaa-ne maa-ko yah haar diyaa.

Ila-ERG mother-DAT this:NOM necklace:NOM give:PERF

'Ila gave this necklace to mother'.

(12) shows that Hindi marks animate or definite objects in past tense/perfective clauses with the same forma ko as recipients in ditransitive constructions, leaving inanimate and indefinite objects unmarked. (13) describes the correspondence between case features and case forms in Hindi:

(13) Case Feature Case Form


DAT ko


ERG ne

The fact that Hindi has no case form that represents accusative case (as case feature) alone sets Hindi apart from languages like Dyirbal which have a distinct case form for each of nominative, accusative, ergative, dative, and genitive case feature (Dixon 1972). What is peculiar about Hindi ditransitive constructions is that their theme objects are always nominative even when they are definite, as in (12f). In contrast, the goal objects always receive the case form ko. This contrast suggests that the primary usage of ko is to mark indirect objects, and not direct objects.

Finally, let us consider Tagalog (Austronesia), a language with three distinct case forms. Common nouns are preceded by sa, ng, and ang, while proper nouns are correspondingly marked in the singular and plural by kay/kina, ni/nina, and si/sina. Examples come from Kroeger (1993):

(14) a. Pinutol ng-magsasaka ang-sungay ng-kalabaw.

PERF-OV-cut ERG-farmer NOM-horn ERG-buffalo

'The farmer cut off the buffalo's horn'.

b. Galit kahapon si-Nenette kay-Lito.

angry yesterday NOM-Nenette DAT-Lito

'Nenette was angry at Lito'.

c. Binalutan niya ng-papel ang-libro.

PERF-DV-wrap 3SG:ERG ERG-paper NOM-book

'He covered the book with the paper'.

d. Ibinigay lahat ng-mga-guro sa-mga-bata

IV-PERF-give all ERG-PL-teacher DAT-PL-child



'The teachers gave all the money to the children'.

e. B-um-ili ang-lalake ng-isda sa-tindahan.

PERF.AV-buy NOM-man ERG-fish DAT-store

'The man bought fish at the store'.

f. Bumabaha sa-Maynila.

AV-IMPERF-flood DAT-Manila

'There is a flood in Manila'.

g. Binigyam lahat ng-mga-guro ng-pera

DV-PERF-give all ERG-PL-teacher ERG-money



'The teachers gave money to all the children'.

The case forms sa/kay/kina mark recipients (14d), locations (14f), and definite objects, the case forms ng/ni/nina mark transitive subjects (14a,c,d,g), possessors (14a), instruments (14c), and indefinite objects (14e,g), while the case forms ang/si/sina occur once in a clause and mark transitive subjects (14e), direct objects (14a,b,c,d), and indirect objects (14g), depending on the voice of verbs.

I propose the following correspondences between case features and case forms in Tagalog under the assumption that genitive case (as case feature) may mark non-specific direct objects as well as possessors (see Moravcsik 1978 for Slavic examples):

(15) Case Feature Case Form

NOM ang (common noun)

si/sina (personal name)

DAT sa (common noun)

kay/kina (personal name)


ERG ng (common noun)

ni/nina (personal name)


I regard sa/kay/kina as the dative case form, since the prototypical function of dative case is to mark recipients in ditransitive constructions. The fact that it also marks specific direct objects suggests that Tagalog neutralizes the distinction between accusative and dative case feature.11, 12

1.3 Markedness Hierarchy of Case Features

We are ready to ask what the CH has to offer typological study of case and case marking in addition to the model of case syncretism among dative, accusative, ergative, and genitive case. This subsection shows that the multiple correspondences between case features and case forms, reviewed in Section 1.2, determine the markedness of case features. We have seen the following set of neutralizations in the previous subsection:

(16) a. Case Form: DAT

Case Feature: DAT ACC

e.g. Kabardian, Hindi, Georgian

b. Case Form: DAT

Case Feature: DAT ERG

e.g. Kabardian, Yagnob

c. Case Form: DAT

Case Feature: DAT GEN

e.g. Bengali, Halkomelem, Kabardian

d. Case Form: ERG

Case Feature: ERG GEN

e.g. Inuit, Tagalog, Laz, Kabardian

(16a)-(16d) suggest (17a)-(17d), under the assumption that the unmarked value is realized in neutralized contexts:

(17) a. Dative case (as case feature) is less marked than accusative case

b. Dative case is less marked than ergative case.

c. Dative case is less marked than genitive case.

d. Ergative case is less marked than genitive case.

(16a)-(16d) do not tell us anything about the relative ranking of accusative and genitive case feature, since there seems to be no example of neutralization between those two features. However, we may regard accusative case (as case feature) as less marked than genitive case (as case feature) because of the typological distribution of case forms in (2). Taken together, we may propose the following markedness hierarchy:

(18) DAT > ACC, ERG > GEN

Accusative and ergative case are not ranked with respect to each other in (18), since there are many languages which have accusative and/or ergative case forms in addition to nominative and dative case forms. This means that there is no implicational relationship between accusative and ergative case features. Since nominative and dative case constitute the fundamental contrast, we may add nominative case (as case feature) to (18):

(19) NOM, DAT > ACC, ERG > GEN

The crucial question at this point is whether nominative or dative case is more marked than the other. I follow Gundel et al. (1986) (cf. Croft 1990, Battistella 1990) in assuming that dative case is the less marked, since dative case may involve more elaboration than nominative case, as illustrated in Kabardian and Yagnob. (8) and (9) describe the correspondences between case features and case forms in Kabardian and Yagnob, respectively:

(8) Case Feature Case Form





(9) Case Feature Case Form






(9) shows that the dative case form in Yagnob may cover as many as four case features, while the nominative case form covers only one case feature.

From this, we may propose the following markedness hierarchy:

(20) DAT > NOM > ACC, ERG > GEN

I will show in Sections 2-4 that (20) finds its expression in OT, which incorporates markedness hierarchy and implicational universal under the rubric of dominance hierarchy.

The second part of this paper is organized as follows. Section 2 will provide an overview of RRG and OT, with focus on how RRG handles verbal semantics and transitivity. Silverstein (1980) left values of case features unspecified, but the RRG two-tiered system of semantic roles provides semantic values for the set of case features, nominative, dative, accusative, and ergative. Sections 3 and 4 will propose a universal set of constraints and define the major case systems including split-ergative systems in terms of those constraints. The paper is concluded by a discussion of how the NCH and CH fit into the framework outlined in Section 2.

2 Framework

2.1 Role and Reference Grammar

2.1.1 Projection Grammar

RRG is a version of parallel structure grammar with a multi-tiered lexical representation; it distributes grammatical information over three components, syntax, semantics, and information structure, as shown in Table 1. No component serves as an input for deriving another component. RRG is analogous to Lexical-Functional Grammar [LFG] (Bresnan 1982 and references therein), in that these parallel representations are subject to different organization and governed by different principles (cf. Sadock 1991, Jackendoff 1997):

Table 1: Projection Grammar

Syntax: Constituent Structure

Operator Projection

Semantics: Semantic Structure

Logical Structure (Thematic Relation) Tier

Macrorole Tier

Information Structure: Focus Structure

These representations are termed projections in RRG. Constituent structure carries phrase structure information, semantic structure handles semantics of major lexical categories (i.e. verbs, nouns, adjectives) and serves as essentially the same function as conceptual structure in Jackendoff (1990), while focus structure is concerned mainly with topic and focus assignment (cf. Lambrecht 1994).13 I will focus on constituent structure and semantic structure alone in this paper. I refer you to Van Valin (1993) and Van Valin and LaPolla (1977) for a full account of the framework.

2.1.2 Constituent Structure

Clauses have three layers. The innermost layer is the nucleus, which contains the predicate, the nucleus plus all the argument(s) of its predicate form the core, while the outermost layer is the clause. The periphery is occupied by an adjunct, e.g. a locative/temporal adverbial, which modifies the core within the clause. Figure 1 diagrams the relationship among these three layers, while Figure 2 describes how this scheme would apply to English:

Figure 1: Layered Structure of the Clause [LSC]

Syntactic Unit Semantic Elements

Nucleus Predicate

Core Predicate + Arguments

Periphery Non-arguments (Adjuncts)

Clause (= Core + Periphery) Predicate + Arguments and Non-arguments

Figure 2: LSC in English

2.1.3 Semantic Structure

RRG assumes two tiers of semantic roles, logical structures [LS] and macroroles [MR]. These two tiers and their associations are the basis of the RRG theory of case assignment (see Van Valin 1991, Jolly 1993, and Nakamura to appear, among others). The LS tier is based on the theory of verbal semantics à la Vendler (1967), which classifies verbs into four aspectual classes, state, activity, achievement, and accomplishment, as shown in Table 2. These decompositional representations of verbs are termed logical structures, following Dowty (1979):

Table 2: Aspectual Classification of Verbs

Verb Class Logical Structure (LS)

STATE predicate' (x) or (x, y)

ACTIVITY do' (x, [predicate' (x) or (x, y)]

ACHIEVEMENT INGR predicate' (x) or (x, y)

ACCOMPLISHMENT BECOME predicate' (x) or (x, y)

CAUSATIVE 'P' CAUSE 'Q', where 'P' and 'Q' are LSs of any type.

'INGR' stands for 'ingressive' and encodes instantaneous change, while 'BECOME' represents change over some temporal span. There is no special formal indicator which marks state predicates, while all activity verbs contain the generalized activity predicate do', which serves as the formal marker of membership in this class. Achievement and accomplishment verbs are derived from two primitives, state and activity verbs. Thematic relations are not primitives in RRG, but are only shorthands for particular argument slots in the decompositional analyses of verbs (cf. Jackendoff 1976). Table 3 shows how thematic relations are assigned in RRG:

Table 3: Thematic Relation Assignment


A. Locational be-at' (x, y) x=locative y=theme

B. Non-locational

1. State or condition predicate' (x) x=patient

2. Perception see' (x, y) x=experiencer y=theme

3. Cognition believe' (x, y) x=experiencer y=theme

4. Possession have' (x, y) x=locative y=theme


A. Uncontrolled

1. Single argument do' (x, [predicate' (x)]) x=effector

2. Two arguments do' (x, [predicate' (x, y)]) x=effector y=locus

B. Controlled DO (x, [do' (x,...)]) x=agent

For example, effector is a label for the first argument of do', while experiencer and theme refer to the first and the second argument of a two-place state predicate, respectively. This means that RRG has no separate argument structure tier as assumed in Grimshaw (1990) that carries a list of thematic relations and their relative ranking.

Sample analyses of English verbs are provided in (21a)-(21e):

(21) a. see see' (x, y) x=experiencer y=theme

b. walk do' (x, [walk' (x)]) x=effector

c. die INGR dead' (x) x=patient

d. kill [do' (x, ø)] CAUSE [INGR dead' (y)]

x=effector y=patient

e. give [do' (x, ø)] CAUSE [INGR have' (y, z)]

x=effector y=locative z=theme

(21a)-(21c) respectively illustrate state, activity, and achievement, while (21d)-(21e) are grouped together under the rubric of causative. For example, the verb kill involves a combination of an activity ('[do' (x, ø)]') and achievement ('[INGR dead' (y)]') predicate.

Macroroles consist of actor and undergoer. They are generalized semantic roles each of which subsumes a number of LS arguments for syntactic purposes, e.g. passivization, and serve as the interface between logical structures and grammatical relations. Actor and undergoer correspond to the two major arguments of a transitive verb, either one of which may be the single argument of an intransitive verb. In contrast to Dowty's (1991) proto-roles, macroroles form an independent tier in the linking between syntax and semantics (see Van Valin 1992 for a critical appraisal of proto-roles).

The mapping between these two tiers of semantic roles is captured by the actor-undergoer hierarchy [AUH] (22) and macrorole assignment principles [MAP] (23):

(22) Actor-Undergoer Hierarchy [AUH]

Actor Undergoer



Arg. of Arg. of 1st Arg. of 2nd Arg. of Arg. of state

DO do' (x,...) pred' (x, y) pred' (x, y) pred' (x)

Agent Effector Locative Theme Patient


["----->" = increasing markedness of realization of LS argument as macrorole]

(23) Macrorole Assignment Principles [MAP]

a. Number: the number of macroroles which a verb takes

1. If a verb has two or more arguments in its LS, it will take two macroroles,

actor and undergoer.

2. If a verb has one argument in its LS, it will take one macrorole.

b. Nature: for verbs which take one macrorole,

1. If the verb has an activity predicate in its LS, the macrorole is actor.

2. If the verb has no activity predicate in its LS, the macrorole is undergoer.

The import of (4) is that the LS argument (or thematic relation) which appears leftmost on the hierarchy will be the actor and the LS argument which appears rightmost on it will be the undergoer. The mapping between LS arguments and macroroles exhibits an order isomorphism, i.e. association lines between them do not cross. This means that no actor-undergoer reversal (e.g. association of actor and undergoer with patient and effector) may occur.

(23a) is concerned with the number of macroroles which a verb may take. This is largely predictable from its LS; there are three possibilities: 0, 1, 2. If a verb has two or three arguments in its LS, e.g. [do' (x, ø)] CAUSE [INGR have' (y, z)], admire' (x, y), the unmarked situation is for it to receive two macroroles, actor and undergoer, in accordance with (22). If a verb has only one argument in its LS, e.g. do' (x, [walk' (x)]), BECOME broken' (x), it normally receives one macrorole. If the verb contains an activity predicate in its LS, e.g. do' (x, [walk' (x)]), the macrorole has to be an actor; otherwise, it should be an undergoer. Verbs with no LS argument, e.g. snow', rain', have no macrorole.

It is possible for verbs to receive one less macrorole than stated in (23a). If the number of macroroles does not follow from (23a), it would have to be specified in the lexical entry of the verb. [1MR] means that there is one macrorole, while [0MR] means that there is no macrorole to assign. For example, consider a Japanese inverse verb wakaru 'understand':

(24) Taroo-ni eigo-ga wakat-ta.

Taro-DAT English-NOM understand-PAST

'Taro understood English'.

The standard RRG analysis of the verb wakaru 'understand' is to assume that it has a feature [1MR] in its lexical entry (Nakamura 1996; cf. Van Valin 1991). This lexical feature overrides (23a1). The basic idea is that there may be a discrepancy between the number of core arguments and that of macroroles. (23b2) requires the verb to take an undergoer, since it has no activity predicate in its LS. The question is which LS argument becomes an undergoer. This is where (22) comes into play; it requires the theme argument eigo 'English' to become an undergoer, since it is ranked lower on (22). The remaining argument Taro has no choice but to become a non-macrorole. (25) shows how thematic relations are associated with macroroles in (24):

(25) LS: understand' (Taro, English) [1MR]

Thematic Relation: Experiencer Theme

Macrorole: Non-MR Undergoer

Analogous accounts hold for inverse verbs in a wide variety of languages (see Michaelis 1993 and Yang 1994).

It is very important to keep in mind that just as actor is not equivalent to agent, it is not equivalent to subject. Actor often functions as subject as in (26a,c), but it may be realized as an adjunct in passive constructions such as (26b). Likewise, undergoer is not equivalent to direct object. (26d) illustrates that it may serve as the subject of an unaccusative predicate such as die:

(26) a. John [SUBJ, ACTOR] hit Bill [DOBJ, UNDERGOER].

b. Bill [SUBJ, UNDERGOER] was hit by John [ACTOR].

c. Mary [SUBJ, ACTOR] ran into the classroom.

d. Jane [SUBJ, UNDERGOER] died after the election.

e. John [SUBJ, ACTOR] gave a book [DOBJ, UNDERGOER] to Tom [NON-MR].

Moreover, (26e) illustrates that those LS arguments which do not act as either actor or undergoer become non-macroroles.

2.1.4 Transitivity

The number of macroroles which a verb takes corresponds closely to the characterization of a verb in terms of the traditional notion of transitivity: single macrorole verbs are intransitive, two macrorole verbs are transitive. The traditional notion refers to a number of arguments that appear in the syntax, and this corresponds to the number of core arguments. It is necessary, then, to distinguish semantic transitivity, which refers to the number of macroroles as in (27), from syntactic transitivity, which refers to the number of core arguments:

(27) Transitivity in terms of Macroroles (Semantic Transitivity)

a. Transitive 2 Macroroles

b. Intransitive 1 Macrorole

c. Atransitive 0 Macrorole

Semantic transitivity (i.e. the number of macroroles) does not necessarily coincide with syntactic transitivity (i.e. the number of core arguments), as illustrated by (28a,b):

(28) a. Taroo-ni eigo-ga wakat-ta.

Taro-DAT English-NOM understand-PAST

'Taro understood English'.

Syntactic Transitivity = 2

Semantic Transitivity = 1 [1MR]

b. Taroo-ga kyoosi-ni hankoosi-ta.

Taro-NOM teacher-NOM defy-PAST

'Taro defied a teacher'.

Syntactic Transitivity = 2

Semantic Transitivity = 1 [1MR]

RRG regards (23a) as the default which may be overridden by lexical features such as [1MR], which is responsible for the two case frames in (28a,b).14

A natural question that arises at this juncture is how to accommodate case alternations, illustrated in (29a,b) and (30a,b), to the set of principles given in the MAP (23):

(29) a. Taroo-ga Hanako-o hasir-ase-ta.

Taro-NOM Hanako-ACC run-CAUS-PAST

'Taro made Hanako run'.

b. Taroo-ga Hanako-ni hasir-are-ta.

Taro-NOM Hanako-DAT run-CAUS-PAST

'Taro had Hanako run'.

(30) a. Taroo-ga betu-no mise-o atat-ta.

Taro-NOM another-GEN shop-ACC try-PAST

b. Taroo-ga betu-no mise-ni atat-ta.

Taro-NOM another-GEN shop-DAT try-PAST

'Taro tried another shop'.

The semantic contrast between (29a) and (29b) has been a matter of controversy since Shibatani (1973, 1976). These examples involve variable transitivity, i.e. a situation in which a verb may take one macrorole (intransitive) or two macroroles (transitive). The challenge now is how to derive the two case frames in (29a,b). The easiest solution is to postulate two distinct lexical entries, one with the lexical feature [1MR], and the other with no specification of the number of macroroles. These two lexical entries are given in (31a,b):

(31) a. [do' (Taro, Ø)] CAUSE [do' (Hanako, [run' (Hanako)])] (=29a)

b. [do' (Taro, Ø)] CAUSE [do' (Hanako, [run' (Hanako)])] [1MR] (=29b)

(31a) shows that (29a) has two LS arguments in its LS. (23a1) requires that they receive actor and undergoer. This macrorole assignment brings about the case frame in (29a). On the other hand, (31b) takes only one macrorole. (23b1) dictates that the only macrorole is actor. Since the effector of the superordinate CAUSE outranks the embedded one for actor status (Van Valin 1993), Taro becomes an actor and forces Hanako to become a non-macrorole. (32a,b) describe these macrorole assignments:

(32) a. [do' (Taro, Ø)] CAUSE [do' (Hanako, [run' (Hanako)])] (=29a)

Effector Effector

Actor Undergoer

b. [do' (Taro, Ø)] CAUSE [do' (Hanako, [run' (Hanako)])] [1MR] (=29b)

Effector Effector

Actor Non-MR

This solution is unsatisfactory, however, since it fails to capture the obvious correlation between case marking and semantics, which can be seen from the different behavior of an adverb muriyari 'forcedly' in (33a,b):

(33) a. Taroo-ga Hanako-o muriyari hasir-ase-ta.

Taro-NOM Hanako-ACC forcedly run-CAUS-PAST

'Taro forcedly made Hanako run'. (=29a)

b. ??Taroo-ga Hanako-ni muriyari hasir-ase-ta.

Taro-NOM Hanako-DAT forcedly run-CAUS-PAST

'Taro forcedly had Hanako run'. (=29b)

The contrast between (33a) and (33b) suggests that (29a) involves a stronger coercion than (29b); (29b) de-emphasizes the causee's role in bringing out the event. Given this correlation, we may conclude that it is ad hoc to assume two distinct lexical entries for (29a,b).

The question now is how to derive these two case frames from a single lexical entry in such a way as to capture the correlation between case marking and semantics. A closer look at the MAP (23) provides a key to a principled account of variable transitivity as illustrated in (29) and (30). The previous RRG literature, e.g. Van Valin (1991), Michaelis (1993), Yang (1994), assumes that there are two classes of verbs, those which follow both of (23a) and (23b) and those which follow (23b), but allow lexical features like [1MR] to violate (23a). To postulate two distinct lexical entries (31a,b) for (29a,b) will be the only option under the assumption that the number of macroroles is fully determined in lexical representation. It is important to note that the number of macroroles taken by the complex verb hasir-ase 'cause-run' in (29) is predictable from context, i.e. whether Hanako is directly affected or not. This observation leads us to propose another class of verbs, i.e. those verbs which may take either one or two macroroles:

(34) There is a class of verbs which follows (23b), but underspecifies the number of MRs,

e.g. ataru 'try', intransitive causative verbs.

(34) is a novel proposal in RRG and thus requires an illustration. Suppose the causative verb hasir-ase 'cause-run' in (29a) follows (23b) only. It is possible for the verb to take one or two macroroles, since it has two LS arguments. When the causative verb takes only one macrorole, it should be an actor because of (23b1). The effector Taro takes an actor status, while the remaining argument Hanako has no choice but to become a non-macrorole. On the other hand, when the causative verb hasir-ase 'cause-run' in (29) follows (23a1) as well as (23b), it receives actor and undergoer. It is now easy to see that the semantic contrast between (29a) and (29b) falls out from the macrorole status of Hanako.

This underspecification account, which dispenses with the lexical feature [1MR], holds for analogous case alternations found in other languages as well. Here are a few examples:

(35) a. Die Mutter schlug mich ins Gesicht.

the mother hit me:ACC in.the face

b. Die Mutter schlug mir ins Gesicht.

the mother hit me:DAT in.the face

'The mother hit me in the face'. (German: Wierzbicka 1988)

(36) a. Los perros lo molestan.

the dogs him:ACC harass

'The dogs harass him'.

b. Los perros le molestan.

the dogs him:DAT bother

'The dogs bother him'. (Spanish: Ackerman and Moore 1994)

(37) a. Ça l'a fait récriminer de plus belle.

that him:ACC-has made complain even more

'This made him complain even more'.

b. Ça lui a fait récriminer de plus belle.

that him:DAT has made complain even more

'This had him complain even more'. (Dialects of French: Authier and Reed 1991)

(38) a. La hice entrar.

her:ACC I.made enter

'I made her enter'.

b. Le hice entrar.

her:DAT I.had enter

'I had her enter'. (Spanish: Ackerman and Moore 1994)

(35) means that she did it on purpose and that by doing so she was expressing her attitude toward me or that her blow really hurts when the object NP receives accusative case (Wierzbicka 1988). When the object NP is dative-marked, (35) does not have such implications. Analogous semantic contrasts are found in (36)-(38).

Figure 3 provides the whole picture of the linking schema in RRG:

Figure 3: General Linking Schema in RRG

The mapping between logical structures and macroroles is lexical and universal, while the mapping between macroroles and grammatical relations is syntactic and subject to extensive typological variation.15

2.2 Optimality Theory

OT is a formalism with an emphasis on constraint interaction which regards a grammar as a function which maps each input (e.g. an underlying phonological string) to its correct structural description (e.g. a prosodic parse: Prince and Smolensky 1993) and has been applied mainly, if not exclusively, to generative grammar, phonology or syntax (e.g. Grimshaw 1993, Speas 1997, Pesetsky 1997). However, apart from the question of whether all constraints are universal, there are few reasons to think that OT is not applicable to functionalist frameworks such as RRG, first of all because it provides a systematic means of investigating language typology and acquisition (Demuth 1995; cf. Reiss and Hale 1996), and secondly because the OT concept of dominance hierarchy has a conceptual affinity with a number of key concepts used in the functionalist literature, e.g. markedness, implicational universal (Greenberg 1966), competing motivation (DuBois 1985; see also Haiman 1985).

OT claims that universal grammar [UG] consists of three components in (39):

(39) a. Con: The universal set of constraints out of which grammars are constructed.

b. Gen: A function that defines, for each possible input, the range of candidate

linguistic analyses available to the input.

c. Eval: A function that comparatively evaluates sets of linguistic forms with

respect to a given constraint hierarchy.

The essential idea behind Con is that all constraints at work in particular languages are universal and often conflicting. This entails that all constraints in OT are, in principle, violable. This is a crucial departure from the more common view (Shieber 1986; cf. Bresnan 1996) that constraints in a grammar are mutually consistent. Gen is a very simple generative mechanism which allows ungrammatical as well as grammatical expressions to be created without restriction. GEN produces a possibly infinite set of candidate analyses which are consistent with a given input. These candidate analyses are tested against Eval, a system of ranked constraints.

There are five aspects of Eval that must be highlighted here. First, constraints are non-graded. The relative strengths of constraints are quantified only in ordinal terms. In this respect, OT is in contrast to Variable Rule Model (Labov 1969, Guy 1991), in which constraints are quantified in probabilistic terms.16 Second, a grammar resolves conflicts among constraints by ranking them in a strict dominance hierarchy, in which each constraint has priority over all the lower-ranking constraints in the hierarchy. It is important to keep in mind that in OT, it is possible and quite commonplace for an optimal candidate to violate constraints. Third, constraints may be violated only when doing so allows satisfaction of higher-ranking constraint(s). Fourth, individual grammars are derived from the way universal constraints are ranked. All possible rankings of universal constraints yield a set of possible languages. The procedure of constraint re-ranking lies at the heart of any OT-based typology (see Grimshaw 1993, Legendre et al. 1993, and Prince and Smolensky 1993: Ch.8, among many others). Finally, Eval assesses the various candidate output forms created by GEN and normally yields a single best (optimal) output. The only requirement for a candidate to be optimal is that it is the minimal violator in the given candidate set.

Evaluation of the candidates proceeds recursively. Suppose a set of constraints are ranked in the order: A, B, C, D, E, in which A dominates B, B dominates C, and so on. Assume further a candidate set: X, Y, Z. The operation of an OT grammar is represented as shown in Table 4. The constraints are ranked from left to right (with the most dominant to the left) as columns of the table, while the candidates are listed on separate rows:

Table 4: Competition in OT

Evaluation of these candidates proceeds as follows. Candidates X and Z tie in satisfying constraint A. The decision between X and Z have to be passed on to the subordinate constraints. Candidate Y violates constraint A. Violations of constraints are indicated by asterisks. This is a critical failure for Y and the candidate is eliminated from consideration. An exclamation mark after an asterisk indicates the fatal violation for a non-optimal candidate. Candidates X and Z also tie again, this time by violating B; neither violation is critical. Finally, Z passes constraint C, while X does not. Thus, candidate Z emerges as the winner. Z's optimal status is indicated by the arrow. Although violations after critical decision points are noted for completeness, they are not relevant to the competition. The shaded portions in Table 4 show that they have no bearing on the outcome at all.17

(40) provides a summary of the general methodology used in OT (adapted from Legendre et al. 1993: 465). (40d) and (40c) are concerned with descriptive adequacy and explanatory adequacy (Chomsky 1965), respectively:

(40) a. Hypothesize a universal set of possible structural descriptions (Gen).

b. Hypothesize a universal set of well-formedness constraints (Con)

governing such structures.

c. Consider all possible rankings of the constraints into dominance

hierarchies; these define the predicted set of possible language-

particular grammars (factorial typology).

d. For each possible hierarchy, determine the well-formed structures

of the corresponding language.

(40a), (40b), and (40d) are widely accepted in the OT literature and will be assumed throughout this paper. I will take issue over (40c), whose assumption that constraints may be ranked in any order has already been subjected to criticism (see Demuth 1995 and Itô and Mester 1995).

3 Typology of Case Systems

3.1 Universal Constraints

I propose (41) as a universal set of case marking constraints. (41a)-(41d) apply only to core arguments. This is an alternative to a number of parametric accounts (e.g. Bobaljik 1992, Levin and Massam 1984, Marantz 1991, Murasugi 1992) according to which ergative and absolutive cases are alternate names for nominative and accusative cases, and Legendre et al. (1993), the first OT account of case systems:

(41) Universal Constraints (Con)

a. Non-macroroles take DATIVE case.

b. Core arguments take NOMINATIVE case.

c. Undergoers take ACCUSATIVE case.

d. Actors take ERGATIVE case.

There is no time here for an extended discussion of oblique case assignment (see Jolly 1993 and Van Valin and LaPolla 1997 for details). However, a few sample analyses are given in (42):

(42) a. from: MR: Non-macrorole

Th.Rel.: first argument in LS configuration

'... BECOME/INGR NOT have'/be-LOC' (x, y)'


1. John came from New York.

2. John received the book from Bill.

b. with: MR: Non-macrorole

Th.Rel.: 1. second argument in LS configuration

'... BECOME/INGR have'/be-LOC' (x, y)'

2. first argument in LS configuration

'... do' (x, ....)'


1. John provided the old man with food and water.

2. Mary cut the bread with the knife.

(42) illustrates how the OT grammar licenses oblique cases other than dative case. It has to rank (42a,b) higher than (41a) in order for them to active (Prince and Smolensky 1993: 81-82), since they apply to a subset of inputs to which (41a) may apply, i.e. non-macroroles.

The constraint system (41) is semantic in nature, since it refers crucially to the two levels of semantic roles, and not to grammatical relations or structural configurations. The imports of (41a)-(41d) are straightforward. (41a) states that dative case may mark non-macrorole core arguments and adjuncts (cf. Silverstein 1980). Dative case typically marks recipient or goal NPs in ditransitive constructions, but (42)-(45) illustrate that it may also mark source NPs that are non-macrorole core arguments. The semantic distinction between goal and source is neutralized in (42)-(45):

(42) a. Peter erzählt den Kindern eine Geschichte.

Peter tells the:DAT children:DAT an:ACC story:ACC

'Peter tells a story to the children'.

b. Peter nimmt den Kindern das Buch.

Peter takes the:DAT children:DAT the:ACC book:ACC

'Peter takes the book from the children'. (German)

(43) a. Taroo-ga Jiroo-ni okasi-o age-ta.

Taro-NOM Jiro-DAT cake-ACC give-PAST

'Taro gave a cake to Jiro'.

b. Taroo-ga Jiroo-ni sono uwasa-o kii-ta.

Taro-NOM Jiro-DAT that rumor-ACC hear-PAST

'Taro heard that rumor from Jiro'. (Japanese)

(44) a. Marie lui a donné une pomme.

Marie him:DAT has give:PSTP an:ACC apple:ACC

'Marie gave him an apple'.

b. Marie lui a caché la verité.

Marie him:DAT has hide:PSTP the:ACC truth:ACC

'Marie hid the truth from him'. (French)

(45) a. Ludmila mu dala kytku.

Ludmila:NOM him:DAT give:PAST flower:ACC

'Ludmila gave him a flower'.

b. Ludmila nám utekla.

Ludmila:NOM 1PL:DAT run.away:PAST

'Ludmila ran away from us'. (Czech: Janda 1993)

Similar neutralizations are attested in other languages as well. Moreover, Japanese examples (46a)-(46d) illustrate that dative case may mark non-macrorole adjuncts as well as arguments:

(46) a. Kadan-ni tanpopo-ga sai-tei-ta.

flower.bed-DAT dandelion-NOM bloom-PROG-PAST

'Dandelions were blooming in the flower bed'. (Locative)

b. Taroo-ga Hanako-ni kuruma-o kat-ta.

Taro-NOM Hanako-DAT car-ACC buy-PAST

'Taro bought a car for Hanako'. (Benefactive)

c. Taroo-ga gan-ni taore-ta.

Taro-NOM cancer-DAT die-PAST

'Taro died from cancer'. (Reason)

d. Taroo-ga kanemoti/binboo-ni umare-ta.

Taro-NOM be.born-PAST

'Taro was born rich/poor'. (Attribution)

(46a)-(46e) show that dative case may mark what would be marked with instrumental, locative, benefactive, and ablative case in other languages. For example, many Australian languages have a distinct form for locative (47b) and allative case (47c) (Blake 1987). Finally, (47d) shows that Russian uses instrumental case for qualitative attribution:

(47) a. Mary bought a cake for Susan.

b. Japanangka-rlu nya-ngu marlu pirli-ngka.

Japanangka-ERG see-PAST kangaroo rock-LOC

'Japanangka saw the kangaroo on the rock'. (Warlpiri: Simpson 1991)

c. dhiga-ia ngaba yabarra-miri.

return-TOP 1SG:PRES camp-ALL

'I am returning to camp'. (Baagandji: Blake 1987)

d. Volcata rodilis' slepymi.

wolf.cub:NOM born:REFL blind:INSTR

'The wolf-cubs were born blind'. (Russian: Janda 1993)

The contrast between (46a)-(46d) and (47a)-(47e) demonstrates that dative is the default case for non-macrorole adjuncts as well as non-macrorole core arguments.

(41b) is required by languages such as Japanese and French, which require every clause to have one nominative argument, as shown by (48b)-(50b):

(48) a. Taroo-ga sensei-ni choosensi-ta.

Taro-NOM teacher-DAT defy-PAST

'Taro defied the teacher'.

b. Sensei-ga Taroo-ni choosens-are-ta.

teacher-NOM Taro-DAT defy-PASS-PAST

'The teacher was defied by Taro'.

(49) a. Taroo-ga Hanako-ni yorikakat-ta.

Taro-NOM Hanako-DAT lean.on-PAST

'Taro leaned on Hanako'.

b. Hanako-ga Taroo-ni yorikakar-are-ta.

Hanako-NOM Taro-DAT lean.on-PASS-PAST

'Hanako was leaned on by Taro'.

(50) a. Marie a obéi au capitaine.

Marie has obey:PSTP to-the captain

'Marie obeyed the captain'.

b. Le capitaine a été obéi.

the captain has been obey:PSTP

'The captain was obeyed'.

(41b) states that nominative is the default case for core arguments; it may mark any number of LS arguments. It leaves its own thematic relation and macrorole value underspecified. Another important point to notice in this connection is that (41b) groups nominative case in accusative languages and absolutive case in ergative languages as a single case feature.

(41c) and (41d) are based on the common observation that accusative and ergative case normally mark undergoers and actors, respectively. The next question to be asked is whether or not any possible ranking of (41a)-(41d) actually yields a real case system.

3.2 Deriving Accusative, Ergative, and Active Systems

The RRG-OT grammar of case marking I am proposing is described in (51), where the set of ranked constraints takes as input a set of macrorole values and, in the case of oblique case assignments, a set of thematic relation and macrorole values, and outputs their case frames:

(51) The RRG-OT Grammar of Case Marking

Input: A Set of Thematic Relation and Macrorole Values

A Universal Set of Constraints (41) Ranked in a Particular Way

Output: Case Frame

Two points are worth noting here. First, (51) does not describe a derivational process; it refers to a mapping between two representations which are subject to different organization. Second, (51) describes how speakers generate (correct) case frames. However, the same grammar may be used for comprehension as well if it takes case frames as input.18

The crucial question to ask here is whether (41a)-(41d) are randomly ranked or they have a default ranking. I follow Demuth (1995), Itô and Mester (1995), and others in assuming that markedness is the basis for a default ranking. It is important to recall from Section 1 that there is a markedness hierarchy of case features deduced from the typological distribution of case forms described in (2). The hierarchy is repeated below for convenience:

(2) Case Hierarchy (adapted from Silverstein 1980) 19

Nom : Dat1 <------ {Acc, Erg} <------ Gen Propositional/Adnominal

Dat2 <------ {Loc, Instr ...} Adverbial/Propositional

(20) DAT > NOM > ACC, ERG > GEN

From this, we may assume the following markedness hierarchy:

(52) Markedness Hierarchy: (41a) > (41b) > (41c), (41d)

The most crucial cut-off point in (52) is between (41a,b) and (41c,d). (41a) outranks (41b), since dative case involves more elaboration than nominative case. Finally, (41c) and (541) are not ranked with respect to each other, since languages may have accusative or ergative case as well as nominative and dative case.

(53)-(57) describe accusative, ergative and active case systems. The distinction between accusative and ergative case systems, for example, boils down to the relative ranking of (41c,d). These typological variations are derived from re-ranking the universal constraints (41a)-(41d):

(53) Case Marking Constraints (Accusative 1)

a. Non-macroroles take DATIVE case.

b. Core arguments take NOMINATIVE case.

c. Undergoers take ACCUSATIVE case.

d. Actors take ERGATIVE case.

e.g. Kannada (Bhat 1991), Korean, Icelandic (Van Valin 1991), German

(54) Case Marking Constraints (Accusative 2)

a. Core arguments take NOMINATIVE case.

b. Non-macroroles take DATIVE case.

c. Undergoers take ACCUSATIVE case.

d. Actors take ERGATIVE case.

e.g. Japanese (Nakamura 1995), French

(55) Case Marking Constraints (Ergative) 20

a. Non-macroroles take DATIVE case.

b. Core arguments take NOMINATIVE case.

c. Actors take ERGATIVE case.

d. Undergoers take ACCUSATIVE case.

e.g. Warlpiri (Simpson 1991), Inuit (Bok-Bennema 1991)

(56) Case Marking Constraints (Ergative-Active)

a. Non-macroroles take DATIVE case.

b. Actors take ERGATIVE case.

c. Core arguments take NOMINATIVE case.

d. Undergoers take ACCUSATIVE case.

e.g. Basque (Levin 1983, Ortiz de Urbina 1989)

(57) Case Marking Constraints (Accusative-Active)

a. Non-macroroles take DATIVE case.

b. Undergoers take ACCUSATIVE case.

c. Core arguments take NOMINATIVE case.

d. Actors take ERGATIVE case.

e.g. Acehnese (Durie 1985)

It is important to recall that (41a)-(41d) form the markedness hierarchy (20). (20) is not observed by every language, however, since the active case systems (56) and (57) rank (41c) or (41d) above (41b). In order to incorporate this ranking reversal to (41), I will focus on Dyirbal, which displays a similar, but more complex ranking reversal.21

3.3 Split Case Systems

Dyirbal is one of the best-known split-ergative languages (Pama-Nyungan: Dixon 1972). It displays a person-conditioned split pattern in which 1st and 2nd person pronouns are case-marked on an accusative basis, while 3rd person argument NPs are marked on an ergative basis.22 I will not discuss those case systems whose splits are based on the contrasts in tense, aspect, and mood (see Tsunoda 1981) in this paper. Nouns in Dyirbal are accompanied by noun markers, abbreviated to 'NM' in (58), which show their class and agree with them in case. The subjects of intransitive clauses normally take nominative case whether they are unaccusative or unergative. (58a)-(58d) illustrate a whole range of transitive case frames in Dyirbal:

(58) a. balan _ugumbil-Ø baNgul ya NM:NOM woman-NOM NM:ERG man-ERG see-TNS

'The man sees the woman'. (Actor=ERG, Undergoer=NOM)

b. Na_a bayi ya ISG:NOM NM:NOM man-NOM see-TNS

'I see the man'. (Actor, Undergoer=NOM)

c. Ninda Nayguna bu 2SG:NOM 1SG:ACC see-TNS

'You see me'. (Actor=NOM, Undergoer=ACC)

d. Nayguna baNgul ya ISG:ACC NM:ERG man-ERG see-TNS.

'The man sees me'. (Actor=ERG, Undergoer=ACC)

In (58b), both arguments, actor and undergoer, take nominative case. In (58d), 'the man' receives ergative case, while 'me' receives accusative case. (58b) and (58d) do not fit into either the nominative-accusative (58c) or the ergative-nominative (58a) pattern and thus pose a serious problem for Bobaljik (1992), Levin and Massam (1984), and Murasugi (1992), all of whom hold the mirror-image view of accusative case systems and ergative case systems.

(59) is a constraint hierarchy for Dyirbal, with language-particular information in (59b) and (59c) italicized:

(59) Case Marking Constraints (Dyirbal)

a. Non-macroroles take DATIVE case.

b. Only first and second person transitive undergoers take ACCUSATIVE case.

c. Only third person transitive actors take ERGATIVE case.

d. Core arguments take NOMINATIVE case.

(59a)-(59c) are not crucially ranked, since they are never in conflict in (59). However, they crucially dominate (59d). (59) violates (20), since it ranks (41b) below (41c) and (41d). I will show later in this subsection that the language-particular information in (59b,c) is derived from a general principle of cognition and does not have to be stored in each of the constraints.

Tables 5 and 6 are the constraint tableaus for (58b) and (58d):

Table 5: Competition in (58b)

Table 6: Competition in (58d)

The competitions in Tables 5 and 6 proceed as follows. Both arguments in (58b) must receive nominative case, since there would otherwise be a violation of (59b) and (59c). In contrast, (58d)'s input to (59) is the third person actor 'the man' and the first person undergoer 'me'. This input satisfies (59b) and (59c). The lowest-ranking constraint (59d) has to be violated in order to satisfy those higher-ranking constraints. The same procedure applies to (58a) and (58c) as well.

There are two facts to be explained here: the defective distribution of accusative and ergative case and the ranking reversal. They remain stipulated in (59). The standard explanation along the line of Silverstein (1976) is that ergative case on the 3rd person actor indicates the markedness of its serving as actor, while accusative case on the 1st/2nd person undergoer shows the markedness of its functioning as undergoer with respect to ego-centricity or empathy (cf. DeLancey 1981). (60) describes the form-content alignment in Dyirbal transitive clauses, while (61) is a summary of the functionalist explanation of split-ergative patterns as in (58):

(60) Form-Content Alignment in Dyirbal Transitive Clauses


Case Feature Marked <--------------------------------> Unmarked

(Erg.) (Nom.)

Semantic Content: Marked Unmarked

(3rd person) (1st/2nd person)


Case Feature Marked <--------------------------------> Unmarked

(Acc.) (Nom.)

Semantic Content: Marked Unmarked

(1st/2nd person) (3rd person)

(61) Split Ergativity = Markedness Assimilation

Split-ergative case systems involve markedness assimilation (Andersen 1972)

or markedness reversal, which brings about ranking reversal as shown in (60)

(cf. Silverstein 1976).

(60) is a typical example of markedness assimilation or reversal, since it represents two clusters of the opposite markedness values, marked and unmarked.

What is peculiar about Dyirbal and many other split case systems is that markedness assimilation occurs only in transitive clauses, i.e. when both actor and undergoer occur within the same clause. In intransitive clauses, the default ranking (20) prevails. It is not difficult to see why, since there is no conflict about the perspective-taking in intransitive clauses. We may extend (61) to other semantic/pragmatic parameters as in (62). Actor or undergoer in transitive clauses may receive the unmarked form only when all of their natural correlations (Croft 1988) are present:

(62) Natural Correlations

Actor <---------------> Undergoer

Ego-centricity: 1st, 2nd person 3rd person

Agentiveness: Animate Inanimate

Definiteness: Definite Indefinite

Topicality: Pronouns Lexical NPs

It is important to notice that the NCH may be taken as a collection of natural correlations, e.g. 1st/2nd person vs. 3rd person, animate vs. inanimate.23

The explanation of split-ergative patterns in terms of markedness assimilation has been widely accepted in the functionalist literature. It is, however, tempting to attribute it to a more general motivation. My proposal is to appeal to iconicity (Peirce 1965-66; see also Haley 1988). The intuition behind the concept of iconicity is that linguistic forms often reflect their contents in some way. Few individual linguistic signs are iconic to any degree, but iconicity shows up in the ways combinations of forms are related to their semantic contents. This is termed diagrammatic iconicity (see Bybee 1985a for its application) The simplest linguistic example of this relational iconicity is a pair of the singular and plural forms of nouns. As shown in (63), the singular form of a noun is generally shorter than the corresponding plural form. (64) is my proposal:

(63) Form: 'car' 'car-s'

Content: singular plural

(64) Markedness Assimilation = Diagrammatic ("Relational") Iconicity

Case systems with a split pattern as in Dyirbal involve diagrammatic iconicity,

a type of iconicity which applies to the ways combinations of case forms and

related to the semantic contents of NPs they are attached to (e.g. ego-centricity).

(64) is based on the proposal made by Andersen (1972) and Shapiro (1983) to include a parallelism between markedness relations as in (60) as well as conceptual ones under the rubric of iconicity. The relation of a case feature and its semantic content itself is arbitrary, but their relations may be aligned with each other, as in (65):

(65) Case Feature Semantic Content



It is clear that to realize the parallel alignment in (60), it is necessary to rank both (41c) and (41d) above (41b) in violation of the markedness hierarchy (20).

Ascribing a split pattern displayed by a case system to diagrammatic iconicity obviates the need to refer to language-specific information in (41c,d) and stipulate ranking reversal. There is only one thing which one has to specify in describing a split-ergative case system: what kind of semantic/pragmatic parameter(s) a language selects for markedness reversal or assimilation, animacy, ego-centricity, or topicality. In the case of Dyirbal, one has to specify that it chooses ego-centricity as the parameter.24

4 Toward a Typology of Possible Case Systems

4.1 How to Relate Markedness with Iconicity

We are left with the question of how to relate markedness (20) with iconicity as in (60). My proposal is to regard markedness as a reflex of economy, i.e. the amount of processing cost (Givón 1990: 947). If this equation is reasonable, we may regard the typology of case systems as an example of competing motivation, a competition between economy and iconicity. The role of economy has been emphasized in OT, but I propose that it is violable when it competes with other motivations, in this case iconicity. When economy predominates, the markedness hierarchy (20) remains intact. Then, 'across-the-board' accusative or ergative case systems will emerge. In contrast, if iconicity predominates, the markedness hierarchy (20) is reversed as in (56) and (57):

(56) Case Marking Constraints (Ergative-Active)

a. Non-macroroles take DATIVE case.

b. Actors take ERGATIVE case.

c. Core arguments take NOMINATIVE case.

d. Undergoers take ACCUSATIVE case.

e.g. Basque (Levin 1983, Ortiz de Urbina 1989)

(57) Case Marking Constraints (Accusative-Active)

a. Non-macroroles take DATIVE case.

b. Undergoers take ACCUSATIVE case.

c. Core arguments take NOMINATIVE case.

d. Actors take ERGATIVE case.

e.g. Acehnese (Durie 1985)

Then, active case systems will show up. These two functional motivations show up in grammars through constraint ranking. The form-content alignment in Basque is described in (66):

(66) Form-Content Alignment in Basque

Case Feature: Ergative <-------------> Nominative

Semantic Content: Actor <----------------> Undergoer

Basque displays a much more direct parallelism than (60), since the iconic relation holds throughout the language, not just in transitive clauses. The iconic relationship holds between conceptual relations, actor and undergoer, and not between markedness relations. Split-ergative case systems like Dyirbal fall between 'across-the-board' accusative/ergative and active systems. These functional principles, economy and iconicity, provide a non-vacuous way of restricting the rankings of (41a)-(41d), since they correctly rule out one logically possible type, non-iconic and non-economic rankings, as shown in (67):

(67) Predicted Typology (* indicates an unattested type)

(41a) > (41c) > (41d) > (41b) 1 (41b) > (41c) > (41d) > (41a)*

(41a) > (41d) > (41c) > (41b) (41c) > (41b) > (41d) > (41a)*

(41c) > (41a) > (41d) > (41b) (41b) > (41d) > (41c) > (41a)*

(41c) > (41d) > (41a) > (41b) (41d) > (41b) > (41c) > (41a)*

(41d) > (41a) > (41c) > (41b) (41c) > (41d) > (41b) > (41a)*

(41d) > (41c) > (41a) > (41b) (41d) > (41c) > (41b) > (41a)*

(41a) > (41b) > (41c) > (41d) 2 (41a) > (41b) > (41d) > (41c) 4

(41b) > (41a) > (41c) > (41d)? (41b) > (41a) > (41d) > (41c)*

(41a) > (41c) > (41b) > (41d) 3 (41a) > (41d) > (41b) > (41c) 5

(41c) > (41a) > (41b) > (41d) (41d) > (41a) > (41b) > (41c)

(41b) > (41c) > (41a) > (41d)* (41d) > (41b) > (41a) > (41c)*

(41c) > (41b) > (41a) > (41d)* (41b) > (41d) > (41a) > (41c)*

1. Split-Ergative 4. Ergative

2. Accusative 5. Ergative-Active

3. Accusative-Active

These 24 (=4!) rankings group around only five distinct case systems. The only exception is (54), a small number of accusative case systems including Japanese and French which involve the ranking marked by a question mark.

The foregoing discussion suggests that constraint ranking is an epiphenomenon of the compromise between a number of grammar-external factors. Split-ergative case systems bring to light the interaction between economy and iconicity and enable us to treat the typology of case systems on a par with other examples of competing motivations (see Haiman 1985). This leads us to propose (68) as an alternative to (40c), whose assumption is summarized in (69):

(68) Functionalist Hypothesis

Constraint ranking is determined by a number of functional motivations extrinsic

to grammar. Typological variation arises from the way of solving conflicts among

these functional motivations.

(69) Formalist Hypothesis

The source of typological variation lies in the representation of grammar and

the set of principles which govern the representation.

(69) is shared by the mainstream OT and all parametric accounts. The major merit of (68) is that we may keep the constraint set (41) universal, while leaving the typological variation of case systems determined by the interaction between economy and iconicity.

To summarize this section, OT allows us to attribute the typological variation of case systems to the competition between economy and iconicity. The analysis of split-ergative case systems such as Dyirbal forced us to regard the range of constraint ranking as an epiphenomenon of the conflict between economy and iconicity.25

4.2 Getting The Whole Picture

Reduced to its essentials, I have addressed the following two questions:

(70) a. How the CH constrains the mapping berween case features and case forms

b. A typology of case systems which is based on the markedness hierarchy of

case features, derived from the CH

This paper has revolved around the CH, which represents markedness relationships among case forms. The analysis of split-ergative and active case systems has led us to introduce iconicity as a determinant of the constraint ranking in addition to economy (markedness). This, in turn, has led us to the NCH, which represents markedness relationships among semantic contents of NPs.26

It is important to emphasize here that these hierarchies represent markedness of form and meaning. (71) provides the overall picture of Silverstein's case theory:

(71) Case Hierarchy


markedness patterns


NP Content Hierarchy

Attention to the form/meaning correlation via markedness (which is taken as economy here) and markedness reversal (which has been taken as diagrammatic iconicity here) places Silverstein's (1976, 1980) case theory firmly in the Prague School and Jakobsonian traditions (see Holenstein 1976 and Battistella 1990 for a useful survey).

Although the CH remains to be elaborated with respect to the treatment of adverbial cases (see Blansitt 1988 and Nakamura 1997: Ch.3), we may say that it provides a foundation for a functional-typological theory of case and case marking together with the NCH.


I would like to thank Luigi Burzio, Ken Hale, Kiyoshi Ishikawa, Jean-Pierre Koenig,

James Myers, Bhuvana Narasimhan, Yasuhiro Shirai, Daniel Silverman, and Robert

D. Van Valin, Jr. for questions and comments on earlier versions of the paper. Any

remaining error is my own, of course.

1. Silverstein (1976, 1980) identifies four independent parameters which determine a type of case-marking oppositions: inherent lexical content of NPs, clause-level propositionality, clause-clause logical relations, and discourse reference-maintenance.

2. Van Valin made a sporadic reference to Silverstein (1980) in some of his papers, in particular Van Valin (1991), but he did not go on to explore the typological scope of the CH.

3. The reader should keep the following table in mind. I will use the first and second pairs interchangeably in this paper:

Case Case-marking

Case feature Case form

Case relation Case-marking

4. English has lost its case forms in the following order:

Gen -----> Acc -----> Dat.

This order is also consistent with the CH.

5. Old French [OF] provides a piece of evidence for the claim that case features should be distinguished from case forms (see Comrie 1991, Blake 1994, and Wierzbicka 1988, among others). OF uses a two-way case-marking system for lexical NPs:

Dat. Nom.

OF maintains a three-way system in masculine demonstratives and third person clitic pronouns:

Dat. Nom. Acc.

Finally, OF uses another two-way case-marking system in first/second person clitic pronouns:

Dat. Nom.

This time, the dative case form covers both dative and accusative case (as case feature). If we made no distinction between case features and case forms, we would have to say that OF has as many as three distinct case systems. This is obviously an undesirable result. It makes more sense to claim that OF has a single accusative case system, described in (53), and attribute the variation to morphophonology.

6. Dyirbal displays an alternation between accusative and dative case forms on O (which is a 1st or 2nd person pronoun) under antipassivization, while Chinookan displays an alternation between ergative and dative case forms on A under what Silverstein terms "inverse" antipassivization, which has to apply in relative clauses expressing "ascribed habitual action". These alternations between case forms explain why dative branches into accusative/ergative and dative in (4a,b), under the assumption that case features (or, to use Silverstein's term, case relations) remain constant under antipassivization.

7. See Croft (1991: 206-212) for a cognitive explanation of why neutralizations between accusative and genitive case are rare. However, Blake (1994: 158) notes that Classical Arabic exhibits a partial neutralization between these two case features.

8. Case syncretism arises from a tradeoff between the need for a language to use as few resources, i.e. case forms, as possible, and the need to represent particular semantic concepts as clearly as possible, i.e. to have as many case forms as necessary to be able to distinguish semantic concepts from each other.

9. Silverstein (1980) handles cross-referencing systems (Chinookan) on a par with case systems (Dyirbal). It is interesting to note that Tzotzil (Mayan: Aissen 1987) uses the same agreement marker (Set A affix) for ergatives (transitive subject) and genitives (possessor). This is the same type of neutralization between ergative and genitive case as in Inuit and Kabardian.

10. First and second person clitic pronouns in Old French display a two-way case-marking system in which the same case form is used to mark both accusative and dative case (as case feature).

11. This analysis receives support from the fact that there are some languages, e.g. Turkish, which mark specific direct objects with accusative case (Knecht 1986):

Ali-Ø kutu-Ø/kutu-yu yap-tî.

Ali-NOM boxes-NOM/boxes-ACC make-PAST

'Ali made boxes/the boxes'.

12. Here are some diagnostics for identifying nominative, dative, accusative, and ergative case forms:

Nominative: it is normally unmarked (citation form) and most likely to mark subjects.

Dative: it marks recipients in ditransitive constructions. If a language does not have ditransitive constructions (e.g. Chamorro, Halkomelem), we can identify a case form which marks the widest range of oblique NPs.

Accusative: it marks transitive undergoers (O).

Ergative: it marks transitive actors (A).

Active languages require different criteria for identifying ergative and accusative case forms.

13. Constituent structure and operator projection might be compared to a verb and its VP-internal arguments and functional projections such as NegP and TP in GB, respectively.

14. The RRG notion of transitivity seems incompatible with Hopper and Thompson's (1980) prototypical notion of transitivity (cf. Langacker 1986), but they are compatible, on the assumption that RRG focuses on categorization by schema, while Hopper and Thompson focus on categorization by prototype. They are two sides of the same coin (see Taylor 1989: 65-68).

15. The pivot selection is construction-specific. When actor outranks undergoer for the pivot status in construction A, for example, you can say that A is an accusative construction. When actor and undergoer are not in competition in construction B (e.g. "subject honorification" in Japanese and Korean), B is not accusative or ergative (see Manning 1996 for an analogous proposal within LFG).

16. Strict dominance may be regarded as a special case in VRM (Guy 1995). This means that we may translate OT hierarchies into VRM, where each constraints has a certain weight.

17. Thus, OT is deterministic. There are some attempts to accommodate optionality into OT, e.g. Anttila (1995), Broihier (1995), Hammond (1994), Pesetsky (1997), Zubritskaya (1994). The simplest way would be to relax one of the main assumptions made by OT that constraints are non-graded. However, this would force us to abandon the hallmark of OT: strict dominance.

18. The OT grammar takes a surface case frame as input and outputs a (partly underspecified) combination of thematic relation and macrorole values.

19. Silverstein's (1976) original case hierarchy, given below, is slightly different from the version in Silverstein (1980):

Nom/Abs: Dat1 <---- Acc <----Erg <---- Gen

The above hierarchy states that the presence of ergative case form implies that of accusative case form, which is obviously false.

20. It remains to be explored how to accommodate three-way case systems, e.g. Wangkumara (Pama-Nyungan: Blake 1987), into the constraint set (41a)-(41d).

21. The case system of Kabardian is represented as follows:

Non-macroroles take DATIVE case.

Core arguments take NOMINATIVE case.

Actors take ERGATIVE case.

Undergoers take ACCUSATIVE case.

Case features Case forms




The basic idea here is that (41a)-(41d) assign case features, which are mapped onto their morphophonological forms. An analogous account holds for Old French:

Non-macroroles take DATIVE case.

Core arguments take NOMINATIVE case.

Undergoers take ACCUSATIVE case.

Actors take ERGATIVE case.

In spite of the fact that Old French has three distinct case-marking systems (see footnote 5 above), I claim that this constraint hierarchy holds in the whole language. The following tables show how case features are mapped onto their morphophonological forms, depending on whether NPs are lexical, masculine demonstratives/third person clitic pronouns, or first/second person clitic pronouns:

Lexical NPs First/Second person clitic pronouns

Case features Case forms Case features Case forms




Masculine demonstratives/third person clitic pronouns

Case features Case forms




The above constraint hierarchy holds for the whole language, while leaving surface variations determined by the mappings between case features and case forms.

22. There is no 3rd person pronoun in Dyirbal (Dixon 1972).

23. Gugu-Yimidhir (Pama-Nyungan) has an animacy-based split in its case systems (Dixon 1994).

24. It is worth emphasizing that I am trying to subordinate constraint ranking to the interplay between two language-external principles, economy and iconicity. This EXPLAINS why Dyirbal re-ranks (41a)-(41d), with language-particular information in (41c) and (41d) (cf. Woolford 1995). Thus, there should be no room under (68) for an objection that it is not good to appeal to re-ranking within a single language. This objection is obviously raised under (69), i.e. the assumption that the range of all possible rankings of constraints provide a basis for typology.

25. OT may serve as a bridge between functionalist frameworks (e.g. Croft 1990, Givón 1990, Van Valin 1993) which seek to explain typological variation in terms of grammar-external motivations and formalist frameworks (e.g. Chomsky 1981) which emphasize the issue of learnability (see Hayes to appear for an analogous suggestion).

26. See Silverstein (1987) for a cognitive implication of the NCH.

Selected References

Andersen, Henning. 1972. "Diphthongization." Language 48: 11-50.

Anttila, Arto. 1995. "Deriving Variation from Grammar." Rutgers Optimality Archive [ROA] 63.

Archangeli, Diana and D. Terence Langendoen (eds.) 1997. Optimality Theory. Oxford:


Bhat, D.N.S. 1991. Grammatical Relations. London: Routledge.

Battistella, Edwin L. 1990. Markedness. Albany, NY.: State University of New York Press.

Berlin, Brent and Paul Kay. 1969. Basic Color Terms. Berkeley, CA.: University of California


Blake, Barry J. 1994. Case. Cambridge: Cambridge University Press.

Blansitt, Edward L., Jr. 1988. "Datives and allatives." In Michael Hammond, Edith A.

Moravcsik, and Jessica R Wirth (eds.), Studies in Syntactic Typology. Amsterdam: John


Bok-Bennema, Reineke. 1991. Case and Agreement in Inuit. Dordrecht: Foris.

Bresnan, Joan. 1994. "Locative Inversion and the Architecture of Universal Grammar."

Language 70: 72-131.

Broihier, Kevin. 1995. "Optimality Theoretic Rankings with Tied Constraints." ROA 46.

Bybee, Joan B. 1985. Morphology. Amsterdam: John Benjamins.

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

Croft, William. 1988. "Agreement vs. Case Marking and Direct Objects." In Michael Barlow

and Charles A. Ferguson (eds.), Agreement in Natural Language. Stanford, CA.: CSLI.

Croft, William. 1990. Typology and Universals. Cambridge: Cambridge University Press.

Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago: University of

Chicago Press.

DeLancey, Scott. 1981. "An Interpretation of Split Ergativity and Related Patterns." Language

57: 626-657.

Demuth, Katherine. 1995. "Markedness and the Development of Prosodic Structure."

Proceedings of the 25th Annual Meeting of the North East Linguistics Society

Dixon, R.M.W. 1972. The Dyirbal Language of North Queensland. Cambridge: Cambridge

University Press.

Dixon, R.M.W. 1994. Ergativity. Cambridge: Cambridge University Press.

Dowty, David. 1979. Word meaning and Montague Grammar. Dordrecht: Reidel.

Dowty, David. 1991. "Thematic Proto-Roles and Argument Selection." Language 67: 547-619.

DuBois, John A. 1985. "Competing Motivations." In John Haiman (ed.), Iconicity in Syntax.

Amsterdam: John Benjamins.

Durie, Mark. 1985. A Grammar of Acehnese on the Basis of a Dialect of North Aceh. Dordrecht:


Givón, Talmy. 1990. Syntax Vol. 2. Amsterdam: John Benjamins.

Greenberg, Joseph H. 1966. "Some Universals of Grammar with Particular Reference to the

Order of Meaningful Elements." In Joseph H. Greenberg (ed.), Universals of Grammar.

Cambridge, MA.: MIT Press.

Grimshaw, Jane. 1993. Minimal Projection, Heads, and Optimality. Rutgers University, Center

for Cognitive Science, Technical Report 4.

Gundel, Jeanette K.; Houlihan, Kathleen; and Gerald A. Sanders. 1986. "Markedness and

Distribution in Phonology and Syntax." In Fred R. Eckman et al. (eds.), Markedness. New

York: Plenum Press.

Guy, Gregory R. 1991. "Explanation in Variable Phonology." Language Variation and Change

3: 1-22.

Haiman, John. 1985. Natural Syntax. Cambridge: Cambridge University Press.

Haley, Michael Cabot. 1988. The Semeiosis of Poetic Metaphor. Bloomington, IN.: Indiana

University Press.

Hammond, Michael. 1994. "An OT Account of Variability in Walmatjari Stress." ROA 20.

Harris, Alice. 1981. Georgian Syntax. Cambridge: Cambridge University Press.

Itô, Junko and Armin Mester. 1995. "The Core-Periphery Structure of the Lexicon and

Constraints on Reranking." University of Massachusetts Occasional Papers 18: Papers in

Optimality Theory.

Jake, Janice L. 1985. Grammatical Relations in Imbabura Quechua. New York: Garland.

Jakobson, Roman. 1932. "The Structure of the Russian Verb." Reprinted in Linda R. Waugh and

Morris Halle (eds.), Russian and Slavic Grammar Studies, 1931-1981. Berlin: Mouton.

Jolly, Julia. 1993. "Preposition Assignment in English." In Robert D. Van Valin, Jr. (ed.),

Advances in Role and Reference Grammar. Amsterdam: John Benjamins.

Knecht, Laura. 1986. "Lexical causatives in Turkish." In D. I. Slobin. and Karl Zimmer (eds.),

Studies in Turkish Linguistics. Amsterdam: John Benjamins.

Labov, William. 1969. "Contraction, Deletion, and Inherent Variability of the English Copula."

Language 45: 715-762.

Legendre, Géraldine; Raymond, William; and Paul Smolensky. 1993. "An Optimality-Theoretic

Typology of Case and Grammatical Voice Systems". Proceedings of the 19th Annual Meeting

of the Berkeley Linguistics Society.

Levin, Beth Carol. 1983. On the Nature of Ergativity. Unpublished Ph.D. Dissertation, MIT.

Levin, Juliette and Diane Massam. 1984. "Surface Ergativity." Proceedings of the 15th Annual

Meeting of the North East Linguistic Society.

Mohanan, Tara. 1990. Arguments in Hindi. Unpublished Ph.D. Dissertation. Stanford University.

Moravcsik, Edith A. 1978. "On the Case Marking of Objects." In Joseph H. Greenberg (eds.),

Universals of Human Language 4: Syntax. Stanford, CA.: Stanford University Press.

Murasugi, Kumiko. 1992. Crossing the Nested Paths. Unpublished Ph.D. Dissertation. MIT.

Nakamura, Wataru. 1995. "A Constraint-Based Approach to Japanese Case Marking." Paper

read at the 69th LSA Annual Meeting at New Orleans, Louisiana.

Nakamura, Wataru. 1997. A Constraint-Based Typology of Case Systems. Unpublished Ph.D.

Dissertation. State University of New York at Buffalo.

Nakamuwa, Wataru. to appear. "Functional Optimality Theory". In Michael Darnell et al. (eds.),

Formalism and Functionalism Vol.2. Amsterdam: John Benjamins.

Peirce, Charles Sanders. 1965-66. Collected papers of Charles Sanders Peirce. Cambridge, MA.:

Harvard University Press.

Pesetsky, David. 1997. "Optimality Theory and Syntax." In Diana Archangeli and D. Terence

Langendoen (eds.), Optimality Theory. Oxford: Blackwell.

Prince, Alan and Paul Smolensky. 1993. Optimality Theory. Rutgers University, Center for

Cognitive Science, Technical Report 2.

Sadock, Jerrold M. 1991. Autolexical Syntax. Chicago: University of Chicago Press.

Shapiro, Michael. 1983. The Sense of Grammar. Bloomington, IN.: Indiana University Press.

Shieber, Stuart. 1986. Introduction to Unification-Based Approaches to Grammar. Stanford,


Silverstein, Michael. 1976. "Hierarchy of Features and Ergativity." In R. M. W. Dixon (ed.),

Grammatical Categories in Australian Languages. New Jersey: Humanities Press.

Silverstein, Michael. 1980. "Of Nominatives and Datives". Unpublished. Reprinted in Robert D.

Van Valin, Jr. (ed.), Advances in Role and Reference Grammar. Amsterdam: John Benjamins.

Simpson, Jane. 1991. Warlpiri Morpho-Syntax. Dordrecht: Kluwer Academic Publishers.

Taylor, John R. 1989. Linguistic Categorization. Oxford: Clarendon Press.

Tsunoda, Tasaku. 1981. "Split Case-Marking Patterns in Verb-Types and Tense/Aspect/Mood."

Linguistics 19: 389-438.

Van Valin, Robert D., Jr. 1991. "Another Look at Icelandic Case Marking and Grammatical

Relations." Natural Language and Linguistic Theory 9: 145-194.

Van Valin, Robert D., Jr. 1993. "A Synopsis of Role and Reference Grammar." In Robert D.

Van Valin, Jr. (ed.), Advances in Role and Reference Grammar. Amsterdam: John Benjamins.

Van Valin, Robert D., Jr. and Randy J. LaPolla. 1997. Syntax. Cambridge: Cambridge

University Press.

Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca, NY.: Cornell University Press.

The logic of syncretization is that the unmarked category, perhaps by virtue of its greater familiarity or frequency or syntactic distribution, lends itself to greater differentiation (Battistella 1990: 40).