The Interaction of Transitivity Features in

the Sinhala Involitive1

John Beavers
The University of Texas at Austin
Cala Zubair
Georgetown University

Abstract
The Sinhala volitive/involitive contrast is characterized by verb stem
and subject case marking alternations, and broadly indicates the volitionality/non-volitionality of the subject, plus other co-varying features. While superficially a high/low transitivity split a` la Hopper
and Thompson (1980), we argue that the distinction actually emerges
from the interaction of just two factors: a realis/irrealis mode contrast
relating to expectations of certain event participants and an independent semantic case system (building on Inman 1993). Co-variation of
other semantic features including volitionality follows directly
from their interaction with semantic case and modality. Explaining
this transitivity split through the interaction of language-specific elements, our analysis refines recent Optimality Theoretic approaches to
transitivity (Malchukov 2005, 2006) by partly obviating the need for

separate, transitivity-specific constraints and constraint rankings.

1 Introduction
Prototype theories of transitivity (Hopper and Thompson 1980, Tsunoda
1981, 1985, Lazard 1998, 2003, Malchukov 2005, 2006, inter alia) maintain that transitivity is a gradient notion, wherein a clause may be more
or less transitive based on the degree to which it conforms to a transitive
prototype defined in terms of some set of distinct but co-varying semantic
features. Much work in this area has looked at so-called transitivity splits,
where two clauses that differ in one or more relevant semantic features also
differ by some overt grammatical distinction. However, certain questions regarding the nature of the systematic co-variation remain unexplained in the
literature: why should the semantic features that tend to vary across transitivity splits vary together, and why should they result in specific types of
grammatical splits? Using evidence from Colloquial Sinhala verb classes,
we address these questions, demonstrating that a careful spelling out of the
semantics of the relevant semantic and grammatical transitivity features in
particular languages can sharpen our predictions about co-variation.
Sinhala verbs fall into two stem classes, the volitive and involitive. Subjects of volitives are almost invariably nominative and subjects of involitives
occur in nominative, accusative, dative, or the postpositional case atiN. Volitives generally indicate volitional action and involitives non-volitional action (Gair 1970, Inman 1993). Examples are given in (1), where a transitive
volitive in (1a) contrasts with both a transitive involitive in (1b), where the
subject is marked by atiN (glossed POST), and an inchoative involitive with
2

a nominative or accusative subject in (1c) (Beavers and Zubair 2009).

(1)

a. Aruni Nimal-w@

giluwa.

Aruni Nimal-ACC drown.VOL.PST

Aruni drowned Nimal (deliberately).
b. Aruni atiN

Nimal-w@

giluna.

Aruni POST Nimal-ACC drown.INV.PST

Aruni drowned Nimal (accidentally)/Nimal drowned.
c. Nimal-w@/Nimal

giluna.

Nimal-ACC/Nimal drown.INV.PST
Nimal drowned (accidentally).
Superficially, (1a) vs. (1b,c) represents a transitivity split since volitionality and number of participants (two of Hopper and Thompsons features)
systematically co-vary with the case marking and stem alternations. However, we argue that a careful examination of volitive and involitive semantics
reduces the contrast to just two cross-classifying properties of the language.
First, the stem alternation reflects a type of realis/irrealis contrast relating to
the intentions or expectations of certain participants in the event (building on
Inman 1993). This interacts significantly with other features that reference
intentions, including volitionality and ultimately actorhood, but not other
features, predicting when we see co-variation. Second, case-marking patterns follow from an independently motivated set of semantic (i.e. inherent,
non-structural) cases, with nominative (the default structural case) surfacing
to cover gaps in the paradigm. The categorical use of the nominative in the
volitive follows from the fact that all volitive stems have subjects that fall
into these gaps due to the types of events that volitive verbs describe.
3

Our analysis has ramifications for prototype theories of transitivity. First,

it addresses the nature of co-variation, showing that a careful spelling out of
the semantics of the relevant transitivity features can predict which features
will co-vary and which will not, sometimes in surprising ways. Second, transitivity splits may arise partly due to the specific morphosyntactic resources
available in a language for marking transitivity features, rather than separate
constraints on transitivity. However, this again requires a clear spelling out
of the semantics of the various resources.
In 2 we review prototype theories of transitivity and unanswered questions that remain. In 3 we outline the basic morphosyntactic and semantic
patterns of volitive and involitive stems suggesting a transitivity split. We
look at these factors more deeply in 4 and argue that the volitive/involitive
contrast is fundamentally a realis/irrealis contrast subsuming and thus predicting co-variation with volitionality, among others. In 5 we examine the
distribution of the different subject cases in Sinhala, arguing that each case
is predicted on independent semantic grounds. We conclude in 6.
Before we begin, we note that Spoken Sinhala consists of a diglossic linguistic system with two related but contrastively defined varieties referred
to as Formal and Colloquial Sinhala (Paolillo 1997). Colloquial and Formal
Sinhala are lexically and grammatically distinct, and prescriptively used in
different social domains. While our data draws primarily on Colloquial Sinhala, we note that subject case and the interpretation of involitive verbs is
subject to regional variation. Thus we focus on the semantics underlying
certain cases and interpretations when they arise for speakers, drawing from
previously published and new data. Additional constraints may apply on an
item by item or speaker by speaker basis.
4

2 High and Low Transitivity

The prototype theory of transitivity of Hopper and Thompson (1980) differentiates two types of transitive clauses. A prototypical, or high transitive,
clause is one associated with a certain set of semantic features, while low
transitive clauses are associated with essentially the opposite features (or the
absence of them). These features are given in (2) (p. 252, (1)):
(2)
A. PARTICIPANTS

HIGH TRANSITIVE

LOW TRANSITIVE

2 or more participants

1 participant

(A and O)
B. K INESIS

action

non-action

C. A SPECT

telic

atelic

D. P UNCTUALITY

punctual

non-punctual

E. VOLITIONALITY

volitional

non-volitional

F. A FFIRMATION

affirmative

negative

G. M ODE

realis

irrealis

H. AGENCY

A high in potency

A low in potency

I. A FFECTEDNESS

OF

O totally affected

O not affected

J. I NDIVIDUATION

OF

O highly individuated

O non-individuated

Hopper and Thompson further propose that when a clause overtly marks a
high/low transitivity feature, other features of the clause will also co-vary in
the same direction, their Transitivity Hypothesis (pp. 254-255, (9)):
(3)

T RANSITIVITY H YPOTHESIS : If two clauses (a) and (b) in a language differ in that (a) is higher in Transitivity according to any of
the features [in (2)], then, if a concomitant grammatical or semantic
difference appears elsewhere in the clause, that difference will also
show (a) to be higher in Transitivity.
5

For example, they claim that in ergative languages, the contrast between
ERG-ABS vs. ABS-OBL (i.e. antipassive) case frames co-varies with several transitivity features as follows (Hopper and Thompson 1980: 268, (53)):
(4)

E RGATIVE

A NTIPASSIVE

Verb codes two participants Verb codes only one participant

Perfective aspect

Imperfective aspect

Total involvement of O

Partitive O

Definite O

Indefinite O

Kinetic/volitional V

Stative/involuntary V

Active participation of A

Passive participation of A

While Hopper and Thompson (1980) initiated many invaluable discussions

on transitivity, several components of their work require further theorizing.
First, the Transitivity Hypothesis lacks independent grammatical definitions
of high and low transitivity through which to verify their claims. Tsunoda
(1981, 1985) addresses this by tying high/low transitivity to morphological markedness. He asserts that unmarked NOM-ACC or ERG-ABS case
frames reflect high transitivity, while alternative (more marked) case frames
indicate low transitivity. This provides a testable link between the semantic factors in (2) and some grammatical consequence, yet does not clarify
which marked case frames will occur for any given low transitivity feature.
Malchukovs (2005, 2006) Optimality Theoretic interpretation of transitivity partly addresses these issues. Malchukov (2006) argues that high/low
transitive marking follows from the relative ranking of various functional
constraints on grammatical encoding (see also Wunderlich and Lakamper
2001, de Hoop and Narasimhan 2005). One such constraint is the Relevance Principle (RelP), wherein the element in the clause (A, O, V) that
6

overtly marks high/low transitivity is the one associated semantically with

the relevant feature in (2). Ranking RelP high means the element that overtly
indicates a certain feature is the one that feature pertains to (e.g. agency of
A is marked on A), whereas ranking it low means we may find mismatches
(e.g. agency of A is marked on V or O). In conjunction with this are constraints on case-marking for A/O-related features, including a faithfulness
constraint on thematic roles, FaithRole, and a Transitive Default (TransDef)
constraint. FaithRole requires the grammatical output form be faithful to the
underlying thematic role and TransDef prefers unmarked case frames. The
relative ranking of these constraints determines different patterns. Ranking
TransDef high produces languages that prefer unmarked cases no matter
the underlying semantics, while ranking FaithRole high produces languages
that show more variability based on semantics, even if this produces marked
cases. Malchukovs constraints thus reveal a more regimented process of
transitivity marking than avaliable under Hopper and Thompson (1980).
Malchukov (2005) (building on Tsunoda 1981) furthermore demonstrates
that different constraint rankings have different consequences for case-marking
with different verb-types. He proposes the two-dimensional semantic map
in (5) to explain how verb classes deviate from core high transitivity depending on how much A deviates from a prototypical agent (upper tier) and how
much O deviates from a prototypical patient (lower tier) (cf. p.81, Fig.1):
(5)

contact

pursuit

(motion)

perception/cognition

emotion

(sensation)

effective action

high transitivity

low transitivity

For example, NOM-ACC and ERG-ABS satisfy all constraints for effective
action verbs. But as we move down the O-hierarchy we see more marked
cases on O and conversely for A on the A-hierarchy. In addition, language
type plays a role: an additional Unmarked Case Constraint (UCC) requiring
one unmarked (nominative/absolutive) argument per clause produces different outputs for ergative and accusative languages, generating still more
complex patterns (see Malchukov 2005: 96-107 for a detailed discussion).
Though these updates to Hopper and Thompsons proposals greatly contribute to developing a testable theory of transitivity, a few problems remain.
First, as Tsunoda (1985) and Malchukov (2006) note, while the Transitivity Hypothesis in (3) allows for systematic co-variation, nothing inherent in
the hypothesis explains such variation (or predicts when it will occur). Why
should the features in (2) vary in the same direction when they co-vary?
For example, telicity, affectedness of O, and individuation of O are closely
related (see e.g. Tenny 1994, Krifka 1998, Beavers 2009) and thus intuitively are expected to co-vary (though not categorically; Lazard 2003: 175,
Beavers 2009). However, other features are clearly unrelated (e.g. agentivity
of A and affectedness of O) and intuitively should (and do) vary independently, contra (3). A suitable Transitivity Hypothesis should have a mechanism for calculating which semantic features co-vary and when.
Malchukov (2006: 333-334) partly addresses this by proposing that Hopper and Thompsons features can be ranked along a Transitivity Scale (a
one-dimensional semantic map) from those most likely to pertain to A, to
those most likely to pertain to V, to those most likely to pertain to O.
(6) Transitivity Scale
A-features

V-features

O-features

animacy|volitionality|kinesis|factivity|tense/aspect|affectedness|(O-)individuation

Malchukov argues that adjacency of features predicts where there should

be relationships, e.g. O-individuation and affectedness are adjacent and covary, but not affectedness and volitionality. Furthermore, if non-adjacent
features do co-vary than all features between them must also co-vary. Although a step towards understanding co-variation, Malchukov still does not
fully motivate why the ranking in (6) is the appropriate ranking, or if the
semantic map might be multi-dimensional rather than one dimensional. Yet
presumably a scale like (6) should follow from how the features are defined
and need not be posited independently. Thus a second issue follows directly
from the first, namely giving precise definitions to the features in (2). Nontypological work in formal and lexical semantics has focused on nearly all
of these features (independently or in combination), but unfortunately has
rarely been applied to work in transitivity.
A third and final issue is the grammatical flip-side of the second: although Malchukov regiments semantic features into A, O, and V-related,
thus associating them with A , O, and V-related grammatical markers, the
question still remains which A, O, and V-related features will be exploited
in a language. This is an important question, since the resources available
in a language could have an impact on the shape of a split. For example,
for Malchukov ranking FaithRole high produces more quirky case patterns.
But what about a language with few semantic cases? Could it have FaithRole
ranked high yet still assimilate most verb types to the transitive default?
In the following, we update Malchukovs Transitivity Scale and OT analysis as pertains to the Sinhala volitive/involitive split, in a way that follows
from independently documented semantic patterns of the relevant features.
9

The emergent definitions of each feature allows us to make specific predictions about when we will get systematic co-variation between features and
when we will not. Furthermore, doing so we show that the semantic features
do not form a one-dimensional hierarchy, but rather a multi-dimensional
map. Likewise, the devices independently available for marking low transitivity have semantic content that also plays a role in predicting when we get
high/low transitive marking, regardless of the underlying clausal semantics.

3 Volitive and Involitive Verbs - Syntax and Semantics

The Sinhala volitive/involitive stem contrast is marked by a combination of
morphophonological features (Inman 1993: 23-27). Present tense involitive
stems have front root vowels and a thematic vowel -e-, while present tense
volitive stems have no place restrictions on their root vowels, and have thematic vowels -a- or -i- . Past tense stems for both volitive and involitive
forms have front root vowels, but the thematic vowel is -un- for the involitive and -u(w)- or -i(y)- (with additional final consonant gemination) for
the volitive. The inflected forms are derived by appending -n@wa (simple
present), -a(a) (simple past), or -nn@/n.d.@ (infinitive). Example paradigms
are given in (7) (adopted from Inman 1993: 24-25, (1)-(2)):
(7)

Present Stem Formation

Past Stem Formation

gloss

volitive

involitive

volitive

involitive

wash

hood-a-n@wa

heed-e-n@wa

heed-u(w)-a

heed-un-a

suck

ur-a-n@wa

ir-e-n@wa

ir-u(w)-a

ir-un-a

make

had-a-n@wa

hd-e-n@wa

hd-u(w)-a

hd-un-a

open

ar-i-n@wa

r-e-n@wa

r-i(y)-a

r-un-a

10

Volitives take nominative subjects and indicate volitional action as in

(8a) for an unergative verb and in (8b-e) for (di)transitive verbs. Direct objects may be accusative (8c) or dative (8d), although, since Sinhala is a differential object-marking language, only animates show accusative (8b,c),
and then only optionally (8c). Indirect objects are dative as in (8e).
(8)

a. Malini dalada maaligaw@-t@ diuwwa.

Malini Temple Tooth-DAT run.VOL.PST

Malini (deliberately) ran to the Temple of the Tooth.
b. Aliya

nestomalt(-*w@) biuwwa.

elephant nestomalt

drink.PST

The elephant drank nestomalt (chocolate milk).

c. Aruni Nimal(-w@)

giluwa/mruwa.

Aruni Nimal(-ACC) drown.VOL.PST/kill.VOL.PST

Aruni (deliberately) drowned/killed Nimal.
Joon-t@ ghuwa.

elephant John-DAT hit.VOL.PST

d. Aliya

The elephant hit John with his trunk.

e. Eyaa mat@
mee sumaane slli dunna.

he 1SG.DAT this week.GEN money give.VOL.PST

He gave me this weeks money.

(Gair 1970: 64)

Semantically, volitive verbs do not strictly entail volitionality. At least some

volitives may occur in explicitly non-volitional contexts. In (9), the verb has
a volitive stem and a nominative subject, but can occur with the modifier
hit@la nemeyi without intention, indicating accidental action.

11

(9) Lam@ya piNgaan@y@ kduwa,

child

plate

eet hit@la

nemeyi.

broke.VOL.PST but intention without

The child broke the plate unintentionally. (Inman 1993: 98, (39))
Nonetheless, some volitives require volitionality, including the verb for murder, which is incompatible with a continuation denying volitionality:
(10) #Siri Gunee-w@

minimruwa,

eet hit@la

nemeyi.

Siri Gunee-ACC murder.VOL.PST but intention without

Siri murdered Gunee, but not intentionally.
Therefore while some volitives entail volitionality, others leave it unspecified. We thus assume volitionality can be a property of the verb root, but for
roots that do not entail volitionality, we assume it arises by Gricean implicature (in contrast with the involitive stem), which we discuss further below.
Turning to involitives verbs, as noted in 1 subjects of involitives are
found in a range of different cases. Subjects of transitive verbs are marked
by the postposition atiN or by dative case as in (11a,b) respectively. Subjects
of intransitives may be dative, accusative, or even nominative, as in (11ce) respectively. In addition, the unaccusatives in (11d,e) represent detransitivized versions of (8c). Crucially, unlike volitive stems, most involitives
categorically entail non-volitionality, as also shown in (11).
(11)

a. Siri atiN

(#hit@la)

piNgaan@y@ kduna.

Siri POST (intentionally) plate

break.INV.PST

Siri accidentally broke the plate #intentionally.

b. Lam@ya-t.@ (#hit@la)

sadd@yak huna.

child-DAT (intentionally) noise

hear.INV.PST

The child accidentally heard a noise #intentionally.

12

c. Amma-t.@

(#hit@la)

wewuluna.

Mother-DAT (intentionally) shiver.INV.PST

Mother involuntarily shivered #intentionally.
d. Lam@ya-w@ (#hit@la)

giluna/mruna.

child-ACC (intentionally) drown.INV.PST/die.INV.PST

The child involuntarily drowned/died #intentionally.
e. Aruni (#hit@la)

giluna/mruna.

Aruni (intentionally) drown.INV.PST/die.INV.PST

Aruni involuntarily drowned/died #intentionally.
Conversely, these are acceptable with nohitaa unintentionally (though our
informants note some redundancy). On the surface the volitive and involitive contrast supports the Transitivity Hypothesis: volitive verbs have NOMACC case frames while involitives have frames that deviate from this. This
corresponds to the volitionality feature (E) in (2) as well as the number of
participants feature (A) for unaccusatives. Also, all of the subjects in (8) are
actors, in the sense of performing some action or exerting some internallyderived force; the closest match here with one of Hopper and Thompsons
features would be agency (H) (A is high in potency). Actorhood does not
hold for involitives, which can also have patient and experiencer subjects.
However, this apparent co-variation is as yet unexplained. Furthermore,
the correlation is incomplete: unergatives such as duwann@ run in (8) appear in the volitive on volitional readings but have one argument; thus (A)
and (E) conflict. We argue next that a careful spelling out of the semantics
of the relevant features predicts co-variation, with the addition of another
feature.
13

4 The Irrealis Nature of Involitive Stems in Sinhala

As depicted in previous analyses (Inman 1993, Henadeerage 2002), the
semantics of the involitive are complex beyond a simple correlation with
non-volitionality. Inman (1993) analyzes volitionality intensionally, building on Kratzer (1977, 1981). Kratzer proposes that intensional modal operators consist of at least two core components: (1) the modal force, representing either necessity or possibility of a proposition being true in some
world(s) and (2) the modal base, representing the relevant conversational
background against which the modal force is interpreted. Inman analyzes
the intentions/expectations of individual x as a set of possible worlds, such
that if x intends for P to be true, then xs intentions are a subset of the
worlds at which P is true. Thus if there exists any world in the intention set
such that P is false, P was not intended, yielding the following definition
of unexpectedness/unintendedness (drawing from Inman 1993: 120, though
we also allow for cases where x has no intentions):
(12) P is unintended by/unexpected to x iff there is some possible world
in which xs intentions/expectations are realized in which P is false,
or x has no intentions.
We follow Inman in analyzing the involitives in (11) as reflecting a modal
operator over the proposition P expressed by the predicate indicating unintendedness of P by the subject x. The modal base is xs intentions and the
force is possibility (there is some world in the base at which P is false):
(13) Involitive (version 1): inv P is true iff P is true and P is unexpected
to the subject x.

(cf. Inman 1993: 138, (107))

14

Our updated definition in (12) covers not just cases where the subject did not
intend to make P true, but also cases where the subject had no intentions at
all, thus permitting inanimate subjects with unaccusatives as in (14).
(14) PiNgaan@y@ mes@yen
plate

wtuna.

table.INST fall.INV.PST

The plate fell from the table.

However, Inman notes that the intentions of the subject are not always
at question. Another (less common) use of the involitive is where the event
is unexpected from the speakers perspective (cf. Inman 1993: 100, (41)):
(15) Kell@ atiN

maalu ageet.@

girl POST fish

pihen@wa.

very.well cook.INV.PRS

The girl can cook fish very well. (despite expectations)

Here the reading is that the girl cooked well on purpose, but this was unexpected to the speaker, a reading Inman refers to as doxastic. To capture
this, Inman (1993: 140-152) suggests that the modal base of the involitive
can in some cases be the expectations of the speaker, all else being equal:
(16) Involitive (version 2): inv P is true iff P is true and P is unexpected
to speaker y or subject x.
Thus the involitive seems to have as its semantics (a) that the event occurred
and (b) that it was unexpected from the viewpoint of either the subject or
the speaker, where the modal force is that of possibility (there exists at least
some world in the modal base at which the event did not occur).
Unfortunately, our informants were unable to verify Inmans doxastic
examples. However, we did find an additional ironic use of the involitive that
occurs during smack talk (playful taunting and insulting). An example is
15

given in (17), where B uses the volitive form of a verb for watch to claim
C participated in a certain watching event, and C counters with an involitive
use to deny that the event occurred (cp. sarcastic intonation in English).
(17) ((C is explaining why he left work late the night of his car accident))
A: i

parakku wela

why late

eheng giye?

happen.PST there go.PST

Why did (you) leave there (work) so late?

B: Okkom@la g@d@r@ giyaa-t.@ passe porn bluwa.
everyone home go.PST after porn watch.VOL.PST
when everyone left (he) looked at porn.
C: Ouu machang, okkom@la ged@r@ giyaat.@ passe mam@ atiN
yes dude

everyone home go.PST after 1SG POST

wla blen@wa.
porn watch.INV.PRS
Yeah I look at porn when everyone leaves.
Additional examples are given in (18)-(19).
(18) ((A is surprised that B has cooked for himself))
A: Uyen@we-d@? Oyaa atiN

iwen@wa,

n.

cook.PRES-Q 2SG POST cook.INVOL.PRES NEG

(You) cooked? You cook, no.
(19) ((B shows A pictures of Nuwara Eliya; A asks if B went there))
B: Ehe giye?

N, machang. Mam@ atiN

there go.PST NEG dude

1SG POST street-LOC

hinganna-gen ewa hor@k@n keruna.

beggar-INSTR them steal
16

pare

do.INV.PST

Go there? No, dude. I stole them from a beggar on the street.

Such data have two important consequences for analyzing the involitive.
First, contra Inman, the propositions are not asserted to be true rather,
their truth is denied. Second, the modal force is necessity, since it is not just
that the speaker did not expect for the proposition to be true, but rather the
speaker expects it to be false, i.e. in all worlds in the modal base.
Thus factoring out the semantics common to all involitives, we first see
that assertion of P is not part of its semantics; it obtains only on some uses.
Second, the strongest modal force across all uses is possibility; necessity the
ironic use is a stronger reading, leaving the core semantics in (20).
(20) Involitive (final): inv P is true iff P is unexpected to the speaker y
or subject x.
In the ironic use the modal force is strengthened to entirely unexpected
and the modal base is the speakers intentions. In the non-volitional and doxastic readings P is additionally asserted to be true, though the two readings
differ in the modal base (subjects intentions vs. speaker intentions respectively). Thus Inmans modal analysis (with some modifications) provides a
simple way to capture the range of uses of the involitive, by differences in
modal base, modal force, and truth of the proposition.
Nonetheless, Inman does not give a corresponding definition of the volitive, which we provide in (21b), based on the definition of intendedness in
(21a), where P must be true and (by default) must be intended by the subject
(the modal base is the subjects intention and the modal force is necessity):
(21)

a. P is intended by x iff for all possible worlds in which xs intentions/expectations are realized x made P is true.
17

b. Volitive: vol P is true iff P is true and (by default) P was intended by the subject x.
Note that (21a) is not the negation of (12). First, intendedness (unlike unintendedness) requires not just that x expects P to be true, but expects to make
P true (e.g. if one expects to be forced to do something then one expects to
do it but does not intend to do it). Second, vacuous universal quantification
over an empty intention set allows inanimates, which lack intentions, to occur with volitives, a prediction borne out for natural forces and instruments:
(22)

a. Hul@N@ ham@n@wa.
wind

blow.VOL.PRS

The wind blows.

(Inman 1993: 182, (82b))

b. Mitiy@ piNgaan@y@ kduwa.

hammer plate

break.VOL.PST

The hammer broke the plate

Thus on our analysis, the definitions of (un)intendedness/(un)expectedness
in (12) and (21a) taken together exempt inanimates from the computing of
intentions, explaining their occurrence as subjects of both kinds of verbs.
To explain why volitional readings arise for animates when not entailed, we
propose the pragmatic principle in (23), a variant of the principle of Holisky
(1987: 118-119) for interpreting human effectors as agents in the absence of
evidence to the contrary (see also Van Valin and Wilkins 1996: 309-317):
(23) Principle of Volitional Subject Interpretation: Interpret the subject of a volitive verb as volitional unless contextually denied.
Crucially, now that we have spelled out the semantics of volitive and
involitive stems, it is clear that the contrast does not fundamentally mark
18

volitionality vs. non-volitionality, since doxastic and ironic involitives may

be volitional and some volitives are non-volitional. Thus the grammatical
contrast does not necessarily co-vary with volitionality (E) in Hopper and
Thompsons system in (2). However, there is one constant difference between volitive and involitive stems: the volitive is necessarily realis the
event must have occurred in the real world and in all worlds in the modal
base while in the involitive the event may not have occurred and furthermore the involitive involves calculating other worlds in which it did not
occur. Thus the transitivity feature in (2) that systematically co-varies with
volitivity marking is the (ir)realis mode feature (G).2
However, the fact that this specific modal operator references subject
intentions most often means that it typically boils down to a volitional/nonvolitional contrast, even if it fundamentally is not one. In other words, (E) in
(2) is reducible to a particular subtype of (G), and because of this additional
property of this particular modal operator, the effect is one of systematic
co-variation of (E) and (G). Likewise, recall that subjects of volitives must
be actors (H), another co-varying feature. Why should this be? Here we
tentatively suggest that it has to do with the pragmatic principle in (23):
for someone to act volitionally is for her to intend to make an event occur,
and thus she must be capable of bringing it about, i.e. an actor. Thus action
by the subject is a prerequisite for (23). Conversely, for involitives no such
constraint applies, and both actor and non-actors are possible.
Thus co-variation between mode, volitionality, and actorhood is expected
semantically. Conversely, co-variation with other features is not expected.
For example, number of participants is not inherently a mode distinction,
and thus we predict (and see) no correlation. That being said, there is a
19

partial correlation: since the modality of volitives requires actorhood, patient subjects are precluded, correctly predicting that unaccusatives occur
exclusively in the involitive. Furthermore, our analysis also predicts that covariation should not be categorical when it does occur, although free variation is strictly limited. For example, volitives can be non-volitional, but only
if explicitly canceled in context, something not possible for all roots. Conversely, involitives can be volitional, but only if the speakers expectations
are taken into account instead, meaning that involitives are not a general way
of encoding volitionality. The overall picture of co-variation is as follows:
(24)

volitive stem

inolitive stem

(+realis+actor)

(-realis)

volitional

non-volitional

volitional

(canceled implicature)

(unexpected to subject)

non-volitional

This analysis makes predictions that differ from Malchukovs Transitivity Scale in 2. Malchukov puts mode (factivity), adjacent to tense/aspect
but separated from volitionality. Yet our discussion has shown that volitionality can be effectively reduced to a subtype of the mode once the semantics
of volitionality are better spelled out. This means that mode can be both an
A-related and V-related feature, depending on the specific modal operator
(indeed also supporting the Relevance Principle, since the feature is marked
on both elements via stem form and quirky case). But of course it need not
be; other modal operators may make no reference to A or O participants.
Likewise, the mode to actorhood connection (mediated through volitionality) is also unexpected in general, but predicted in this particular case.
More fundamentally, our analysis demonstrates that the relationships between features are not one-dimensional, but multi-dimensional, contingent
20

on the specific meanings of the specific features in a given language.

Ideally, once the semantics of each feature is spelled out, there should be
no need for a Transitivity Scale to predict co-variation all of the predictions follow from the semantics, recognizing a more fluid set of interactions.
Our analysis has shown that this is possible for three features relevant for
the involitive. With this analysis of the volitive/involitive stem contrast as a
backdrop, we turn next to the case variation we see in involitives.3

5 Subject Case and Verb Type

Recall that Malchukov (2005) proposes that verb type (as determinant of
thematic roles for A and O) interacts with argument realization constraints
to produce different case patterns in different languages. However, the Sinhala data above pose an interesting puzzle for this proposal: volitive verbs,
regardless of verb type, are always coded as high transitives. Why should
this be? One might suggest that realis mood the only core feature for
volitives is necessary for effective action. But why would this manifest
as an unmarked case frame except through the A-related feature of volitionality, which does not hold categorically?4 Alternatively, it could be that the
constraints are ranked differently for volitives and involitives, so that TransDef (preferring unmarked case frames) is ranked more highly for volitives
than involitives. But why should this be?
We argue that the high transitive case frames of volitives are expected
once we make a careful examination of the semantics of the relevant cases.
In particular, the quirky cases found with involitives all have independently
motivated meanings that are all ruled out for volitives, leaving the default
21

case, nominative, as the only possible subject case. Indeed, even among involitives there are gaps in the semantic case paradigm, predicting the presence of nominative and thus high transitive marking there as well.

5.1 Dative vs. atiN Subjects

Subjects marked by atiN occur exclusively with transitive caused changeof-state verbs (e.g. issenn@ lift; kdenn@ break, renn@ open; ellenn@
hang; hellenn@ shake; plenn@ split (as in wood); kr@kenn@ turn,
weelenn@ twine; see Wijayawardhana et al. 1991: 113-122), while dative subjects occur primarily with unergative verbs (e.g. n.d.enn@ cry; hinaawenn@ laugh; wewulenn@ shiver; numak renn@ yawn; nt.enn@
dance; widenn@ walk; Henadeerage 2002: 137-140). We could analyze
the distribution of atiN and dative structurally: atiN only occurs with transitives and dative only with intransitives (i.e. atiN indicates the presence of a
lower argument as in the analysis of ergative case by Wunderlich 1997).
However, this analysis is clearly insufficient, since (as discussed above)
unaccusative verbs are intransitive but do not take dative subjects. Furthermore, we also find dative subjects with some transitive involitive verbs, including performance verbs (kiy@wenn@, read, pwenn@ perform, kiyenn@
speak, say, recite), psych-verbs (dnenn@, feel; peenn@, see; henn@,
hear; teerenn@, understand), and verbs of receiving (lbenn@, receive):
(25)

a. Mat.@/*mam@

atiN

pot@ kiy@wewuna.

1SG.DAT/1SG POST book read.INV.PST

I (accidentally) read the book. (i.e. may eyes just fell on it)

22

b. Lam@ya-t.@/*lam@ya atiN
child-DAT/child

sadd@yak huna.

POST noise

hear.INV.PST

The child hear a noise.

c. Lamayi-t.@/*lamayi

atiN

tgi lbuna.

child.PL-DAT/child.PL POST gift.PL receive.INV.PST

The children received gifts.
Thus we cannot simply associate atiN with transitives and dative with intransitives. An alternative is to take a partly semantic view of the two cases.
Inman (1993: 32-34) argues that atiN represents a generalized ergative
(following the terminology of Kiparsky 1992), wherein its distribution is
determined structurally as the default subject of involitive verbs, superseded only by a more specific semantic case. Dative, he argues, is semantically determined, marking participants that do not intend for the event to
occur (p.161ff). Further support for this comes from the fact that on this use
datives are ruled out for inanimates. Indeed, our speakers find natural force
subjects acceptable with involitive unergatives, but they are nominative:5
(26) Hul@N@ hmen@wa.
wind

blow.INV.PRS

The wind blows.

However, as Inman also notes, sometimes dative and atiN alternate (Inman 1993: 163, (8); glosses from original):
(27)

a. And@ree atiN

wli kwen@wa.

Andare POST sand eat.INV.PRS

Andare happens to be eating sand.

23

b. And@ree-t.@

wli kwen@wa.

Andara-DAT sand eat.INV.PRS

Andare happens to eat sand involuntarily.
Inman argues that in (27a) the subject acted intentionally, perhaps to eat
something, but ended up eating sand, while in (27b) the subject did not
intend to do anything at all, yet found himself eating sand. We return to this
below, but the fact that the verbs in (25) do not allow an alternation while
(27) does suggests that there is a semantic restriction on atiN as well.
We argue that atiN marks involuntary causers while dative marks a host
of thematic roles sharing non-volitionality in common. First, dative generally marks experiencer subjects (as in many Indo-Aryan languages; see the
papers in Verma and Mohanan 1990). For example, experiencer subjects of
non-verbal predicates can also be dative:
(28) Api-ge

lam@ya-t.@ .tikak as@niipay.

1PL-GEN child-DAT a.bit sick-ASN

Our child is a bit sick.

(Gair and Paolillo 1988)

Thus the dative in (25b) can be assimilated under a more general use of the
dative for experiencer subjects. Dative also independently marks possessors,
including both possessor subjects and indirect objects (the stative volitive in
(29a) being a very rare example of a non-volitional, quirky subject volitive,
which we treat as a lexical exception):
(29)

a. Lam@ya-t.@ sellaNbaduwak tiyen@wa.

child-DAT toy

exists.VOL.PRS

The child has a toy.

(Gair and Paolillo 1997: 66)

24

b. Eyaa mat@
mee sumaane salli dunna.

he 1SG.DAT this week.GEN money give.VOL.PST

He gave me this weeks money.

(Gair 1970: 64)

Thus the dative in (25c) can also be subsumed under a more general use
of the dative. Indeed, these two datives may be collapsible, as argued by
Mohanan and Mohanan (1990) based on data from Malayalam, where experiencers may be expressed as the goal argument of a light deictic motion
verb with the stimulus as a theme (cp. English happiness came to him). Thus
a single notion of goal may underlie both uses of the dative: either the goal
of a physical object or the goal of a mental/physical sensation.
However, this analysis does not obviously explain why dative also marks
subjects of unergatives or subjects of performance verbs such as those meaning read, perform, and recite. Before we offer a tentative solution to
this, we first argue that these two uses of the dative are not distinct. As
suggested by Tenny (1994), and then argued (in different ways) by Beavers
(2009) and Rappaport Hovav (2008), performance verbs are not standard
agent/patient transitive verbs on par with verbs meaning break or cool.
Rather, these verbs take direct objects that measure out the event by the subjects progress through the object the subject moves through the book
in (25a), much in the same way that the subject moves along the path object
in John hiked the trail. Evidence for this comes from the fact that performance objects pattern like path objects in terms of measure phrases: in both
cases V DO halfway means roughly V half of DO, an equivalence that does
not maintain for most change-of-state verbs:
(30)

a. John hiked half of the trail John hiked the trail halfway.

25

b. John read half of a book John read a book halfway.

c. John cooled half of the soup = John cooled the soup halfway.
In (30c) the first clause is true if half of the soup becomes any cooler, while
the second clause is true only if there is a contextually specified target temperature for the soup to which John has cooled it halfway. No separate
contexts are possible for the pairs in (30a,b). Furthermore, the subjects in
(30a,b) are in a sense in motion, albeit abstractly in (30b), but not for
(30c). This can be seen in (31), where for (31a,b) the subject is entailed to
be in the middle of the object, but not so for patient objects in (31c).
(31)

a. John is hiking the trail, so he is somewhere in the middle of it.

b. John is reading a book, so he is somewhere in the middle of it.
c. #John is cooling the soup, so he is somewhere in the middle of it.

This is again explained if performance verbs are analyzed as abstract motion verbs, where the subject progresses through the direct object, which in
turn measures out the event (see Beavers 2009 for a detailed analysis). On
this analysis, performance verbs are abstract unergative verbs, and thus the
dative is subsumable under the same analysis as more canonical unergatives.
With this reduction, we argue that the subject role for involitive unergatives is that of involuntary actor. Evidence comes from a comparison with
the semantics of atiN, which we suggest marks involuntary causers. It is a
crucial difference between clauses that take atiN and those that take dative
that the former have causative semantics, i.e. in standard event decompositional terms we can say that causative events have the structure in (32a) and
unergatives have the structure in (32b) (following the notation of Rappaport
Hovav and Levin 1998: 108 for expository convenience).
26

(32)

a. [ x CAUSE [ y BECOME < STATE > ] ]

(x, y individuals)

b. [ x ACT (y) ]

(x, y individuals)

Causation can be probed for by scope ambiguities with modifiers such as

negation (Dowty 1979: Ch.5). Negated causative clauses are ambiguous between a reading where the causing event did not occur, the resultant event
did not occur, or neither occurred, as in (33a). This is not possible with
non-causative verbs as in (33b), there being only one subevent in (32b).
(33)

a. John did not break the vase.

i. The vase broke by itself.

(negated cause)

ii. The vase did not break, though John hit it. (negated result)
iii. John neither acted nor did the vase break.
b. John did not run/read the book.

(both negated)
(no ambiguity)

Crucially, clauses that take atiN subjects show ambiguity under negation, but
those that take dative do not, as in (34), supporting our analysis.
(34)

a. Eyaa atiN

piNgaan@y@ kdune

3SG POST plate

n.

break.INV.PST NEG

She did not accidentally break the plate.

i. The plate broke by itself.

(negated cause)

ii. The plate did not break, though she hit it. (negated result)
iii. She neither acted nor did the plate break.
b. Mat.@

pot@ kiy@wewune

(both negated)

1SG.DAT book read.INV.PST NEG

I did not (accidentally) read the book.

27

(no ambiguity)

Further evidence for the contrast between clauses with atiN subjects and
those with dative subjects comes from verbs that show an alternation. As
noted above, the verb meaning eat in (27) allows either atiN or dative; with
the former the subject intended to do something, though not to eat sand,
while in the latter case the subject had no intentions at all. Crucially, with
eat verbs the subject is both an actor and a causer: the subject works his way
through the sand in (27) like a motion verb (Ramchand 2008) and simultaneously the sand comes to be affected (eaten) by the action of the subject
like an agent/patient verb (Beavers 2009). This event type is schematized
in (35) (again borrowing for convenience the notation for accomplishments
with associated actions of Rappaport Hovav and Levin 1998: 108).
(35) [ [ x ACT ] CAUSE [ y BECOME < EATEN > ] ] (x, y individuals)
If dative marks involuntary actors and atiN marks involuntary causers, the
dual nature of the subjects role explains the alternation and the semantic
contrast. If the action is intentional but the causation is not, the subject is
an involuntary causer, yielding atiN. Alternatively, if the action itself is unintended, the subject is an involuntary actor and dative is licensed. Thus
the semantic contrast reflects the variable scope of the involitive operator,
providing further motivation for the contrast in the two cases.
We now ask what independent evidence there is for these analyses. For
datives, we tentatively suggest that an involuntary actor is essentially an experiencer, albeit an experiencer of a physical sensation of performing an
action rather than a psychological or static physical sensation. The explanation for the lack of dative inanimate involuntary actors may follow from
this fact (since inanimates cannot be experiencers), or from a restriction of
dative to actors that are truly non-volitional, requiring that they have inten28

tions (i.e. unlike the involitive operator more generally, inanimates are not
exempt from computing intentions if dative).
For atiN we propose that its distribution derives from its etymology: it
is synchronically identical to the instrumental form of at@ hand, i.e. it is
superficially identical to the expression by hand. However, there is evidence that it has grammaticalized into a substantive postposition (Gair 1970:
73) that heads a PP that is the subject of the involitive clause. First, the
[ DP atiN ] phrase is a constituent, as can be shown by constituency tests.
For example, while scrambling is possible in Sinhala, as shown in (36a-c),
atiN cannot be scrambled away from the DP it marks, as shown in (36d,e).
(36)

a. Siri atiN

piNgaan@y@ [email protected]

Siri POST plate

binduna.

accidentally broke.INV.PST

Siri broke the plate accidentally.

b. PiNgaan@y@ Siri atiN [email protected] binduna.
c. PiNgaan@y@ [email protected] Siri atiN binduna.
d. *Siri piNgaan@y@ atiN [email protected] binduna.
e. *Siri piNgaan@y@ [email protected] atiN binduna.
Thus atiN forms a constituent with the DP to its left, and is not (for example) an instrumental adverbial. However, the resultant constituent shares an
index with the embedded DP, not the postposition. As Inman (1993: 53-56)
makes clear, these PPs can act as antecedents for various types of anaphors,
including the null subject PRO of adverbial clauses as in (37a) and reflexive
pronouns as in (37b) (Inman 1993: 54-56, (52), (57)).
(37)

a. [ PROi wd@ k@r@n@ gamaN ], Guneei atiN Siri-w@

work do.PRT while

Gunee POST Siri-ACC
29

wiweec@nee keruna.
criticism

do.INV.PST

Guneei criticized Sirij while PROi/j working.

b. Nimali atiN

Siri-w@

taman-gei ged@r@-di tallu

Nimal POST Siri-ACC self-GEN house-at pushing

keruna.
do.INV.PST
Nimali pushed Sirij at selfsi/j house.
These data confirm that atiN phrases are not literally translated as (by) Xs
hand, since otherwise the referent of the phrase would be Xs hand, not
X, and these binding facts would be surprising. Thus atiN is historically an
instrumental, but synchronically a case-marking postposition.
However, we suggest that two (vestigial) aspects of atiNs erstwhile instrumental semantics explain its use marking involuntary causes in transitive
clauses. First, we consider the semantics of instrumentality vis-`a-vis causation. Croft (1991:178) analyzes the English instrumental with as a marker of
causal intermediacy, i.e. an entity intermediate in the events force-dynamic
structure, being acted upon by an external causer but causally prior to some
patient. This general semantics subsumes the two different uses of with
marking canonical instruments as in (38a) and so-called displaced themes
(Rappaport and Levin 1988: 28-31) as in (38b), with the common forcedynamic semantics illustrated in the (simplified) Croft-style diagrams.
(38)

a. John cut the bread with a knife.

John knife bread

b. John filled the glass with water.

John water glass

The Sinhala instrumental case shows the same two uses, suggesting it also
30

encodes causal intermediacy:

(39)

a. Chamara kothu roti atiN

kaewwa.

Chamara kothu roti hand.INST eat.VOL.PST

Chamara ate kothu roti with her hands. C. hands roti
b. Mam@ wiiduru-w@ wature-n

piruwa.

1SG glass-ACC water-INSTR fill.VOL.PST

I filled the glass with water.

I water glass

This in turn explains the restriction that atiN always occur with transitive
causation verbs: the causal intermediacy requires the presence of some other
entity in the clause that is acted upon, which requires the presence of a lower
argument, effectively replicating the Wunderlich (1997) analysis of ergative
case noted above. The only aspect of causal intermediacy not preserved is
the requirement of an external causer, though this may have been lost in
grammaticalization since atiN is not referential on its own.
Likewise, the exclusive use of atiN in involitive clauses may also derive
from its erstwhile instrumental status. Instruments are by definition unwitting, so that we might expect atiN to only occur in non-volitional clauses.
However, as noted above, doxastic and ironic uses with atiN are volitional,
suggesting that non-volitionality is not a synchronic component of its meaning. Nonetheless, it is plausible that this component of instrumentality has
grammaticalized into a restriction that atiN only occur with involitives. Thus,
while postpositional atiN is not attested outside of involitive clauses, its distribution may have an independent diachronic semantic explanation.
We can thus sweep non-nominative subjects of transitive and unergative
verbs under two semantic cases:

31

(40)

a. Dative = goal (goal, recipient, experiencer, involuntary actor)

b. atiN = causally prior participant (for involitives)

We turn next to intransitives, and argue that semantics also determines the
subject cases there, although at least one case may indeed be a default case.

5.2 Nominative vs. Accusative Subjects

Unaccusative verbs occur exclusively in the involitive. There are at least

two types of unaccusatives: externally caused change-of-state (ECOS) (e.g.
mrenn@ die; wt.enn@ fall; gilenn@ drown; hir@wenn@ get stuck; lissenn@ slip) and internally caused change-of-state (ICOS) (pipenn@ bloom)
(Levin and Rappaport Hovav 1995: 90ff.). Subjects of ICOS verbs are necessarily nominative, while subjects of ECOS verbs are either accusative (Inman 1993) or nominative (Henadeerage 2002), as in (41).
(41)

a. Aruni-ge

ratu rosa(-*w@) pipuna.

Aruni-GEN red rose-*ACC bloom.INV.PST

Arunis red rose bloomed.
b. Lam@ya/Lam@ya-w@ giluwa/mruna.
child/child-ACC

drown.INV.PST/die.INV.PST

The child drowned/died (accidentally).

The fact that nominative subjects occur with involitive verbs is surprising
Inman claims that nominative never occurs with involitive verbs, though
Henadeerage gives numerous examples and our informants also find them
acceptable. Beavers and Zubair (2009) propose that the nominative/accusative
contrast reflects ultimate causation, illustrated in (42) and (43). In (42), a
32

nominative subject is compatible with a context in which there was no external causer, but not an accusative subject. Similarly, the emphatic/reflexive
clitic -m@ (which conveys a by himself/herself reading) is only compatible
with nominative subjects as shown in (43).6
(42)

a. Amma lissuna,

eet kowwuruwat eyaa-w@

Mother fall.INV.PST but somebody

lisseuwe

3SG-ACC

n.

push.CAUS.PST NEG
Mother fell, but nobody pushed her.
b. #Amma-w@

lissuna,

eet kowwuruwat eyaa-w@

Mother-ACC fall.INV.PST but somebody

lisseuwe

(No external cause)

3SG-ACC

n.

push.CAUS.PST NEG
Mother fell, but nobody pushed her.
(43)

a. Shameela-m@

(#No external cause)

hir@wuna/lissuna.

Shameela-REFL get.stuck.INV.PST/slip.INV.PST
Shameela got herself stuck/slipped on her own.
b. #Shameela-w@-m@

(reflexive)

hir@wuna/lissuna.

Shameela-ACC-REFL get.stuck.INV.PST/slip.INV.PST
Shameela got herself stuck/slipped on her own.

(#reflexive)

Beavers and Zubair argue that ECOS unaccusatives are derived from transitive causative volitive verbs by a process of causer suppression that may
result in two possible interpretations the suppressed causer is interpreted
reflexively as co-identified with the patient argument, or it is existentially
bound and is interpreted as distinct from the patient. These two meanings
33

are represented in (44b,c) respectively. ICOS verbs we assume are noncausative and reflect only inchoative semantics as in (44a).
(44)

a.

[ y BECOME < STATE > ]

b.

[ x CAUSE [ x BECOME < STATE > ] ]

(ICOS)
(Refl. ECOS)

c. x[ x CAUSE [ y BECOME < STATE > ] ] (-bound ECOS)

What unifies (44a) and (44b) is that in neither case is there an entailment of
external causation, i.e. reflexive ECOS verbs are de facto ICOS verbs. We
thus suggest that accusative in Sinhala has a semantic use that indicates a
patient of an externally caused event (cp. Wunderlich 1997).
(45) ACC = Patient of an externally caused event
Thus any verb that takes a patient argument and entails an external causer
will realize that argument as accusative, including ECOS verbs on the reading in (44b) plus all transitive change-of-state verbs. Note that for such arguments dative and atiN are of course ruled out: these participants are neither
actors nor causers. Unfortunately, this definition of accusative is something
of a stipulation accusative is also a structural case and can mark nonpatient arguments as well. Nonetheless, all externally manipulated patients
are accusative, so that the stipulation represents a unidirectional correlation.
This leaves the distribution of nominative, which realizes (a) actor subjects of volitive verbs (which are by default volitional, but not always),
(b) inherently non-volitional patients of involitive ICOS/reflexivized ECOS
verbs, and (c) non-volitional inanimate subjects of unergatives. Prima facie
there is no semantic unification of these roles since it would have to cover
both volitional and non-volitional participants as well as both actors and patients. The obvious conclusion is simply that nominative is the default case,
34

and arises due to gaps in the paradigm: the arguments that take nominative
are unable to bear any of the semantic cases discussed above.
Among involitives, unaccusative subjects cannot be marked by atiN since
there is no causal precedence, nor dative because they are not involuntary
actors, recipients/goals, or experiencers. If there is external causation accusative is possible, but minus that there is no semantic case left. Likewise,
inanimate actors such as natural forces cannot be marked dative since they
have no intentions, nor atiN or accusative due to lack of causally subsequent/prior entities. In all cases, default nominative is all that is left.
Among volitives, animate subjects of unergatives may not be dative
since its inherent non-volitionality clashes with the interpretative principles
governing volitives. Inanimate subjects are ruled out for dative a priori. Accusative may not mark subjects of volitives since no volitive takes a patient
subject (as discussed in 4), and atiN is (as argued above) restricted to involitive clauses and thus also never occurs with volitives. Thus again default
nominative is the only case left. Note that it is possible to have quirky subjects in the volitive; as shown in (29a) dative possessor subjects are possible.
Thus the only explanation is that nominative fills in the gaps where other
cases fail, and there are more such gaps with volitive verbs than involitives.

5.3 The Interaction of Verb Type, Semantic Case, and Volitivity

To summarize, we have argued for particular semantic definitions for dative,

atiN, and accusative that predict their distributions. Nominative is not associated with any semantics, and shows up when no more specific case applies,
including nearly all volitive verbs and even some involitives. The patterns in
35

3 and 5.1-5.2 emerge through the interaction of three factors: verb class,
the modal operators associated with the two stem types, and the system
of specific semantic cases found in the language. In terms of Malchukov
(2005), FaithRole (overtly indicate thematic roles) is presumably ranked
highly in Sinhala, explaining why we get quirky cases at all. However, were
we to focus just on Malchukovs constraints, we might expect some volitives
with quirky subjects, and likewise no nominative subjects in involitives. We
could overcome this by saying TransDef (prefer unmarked frames) is ranked
more highly just for volitive stems, but this would not explain nominative
subjects with involitives. On our analysis, no reranking is necessary: even
with FaithRole ranked highly across the board, if there is no case for a given
thematic role, then nominative is all that is left (i.e. FaithRole is vacuously
satisfied). However, it is only by examining the semantics of particular cases
in detail that this conclusion emerges, adding an additional level of explanation beyond Malchukovs broader constraints on surface forms.

6 Conclusion
We have argued that the Sinhala volitive/involitive contrast is a high/low
transitivity split that reduces to two factors: a stem alternation indicating
a realis/irrealis distinction that has significant interaction with volitionality
and actorhood, and a system of semantic cases. In contrast to Malchukovs
one-dimensional Transitivity Scale, our analysis indicates that semantic features form a multi-dimensional map. The crucial factor is the way particular
semantic features are independently defined, so that mode not normally
considered an A-related feature happens to be an A-related feature here.
36

Likewise, the resources available in a language for marking low transitivity

are semantically contentful but may contain gaps that result in the emergence of high transitive (default) marking through interaction with semantic
features. This gives Sinhala the appearance of preserving transitive defaults
across verb types in one stem but not the other. Thus examining how the
semantic and grammatical transitivity features in the language interact offers an explanation behind this superficial pattern. This shows the necessity
of a close examination of semantic and grammatical features that play into
transitivity splits, since relying on constraints purely on surface form alone
may not fully capture all of the co-variations seen within a language.

References
Beavers, J. 2009. On affectedness. Unpublished ms., The University of
Texas at Austin.
Beavers, J. and C. Zubair. 2009. Anticausatives in Sinhala: Involitivity and
causer suppression. Unpublished ms., The University of Texas at Austin
and Georgetown University.
Brandt, P. 2006. Receiving and perceiving datives (cipients): a view from
German. In Datives and other cases: between argument structure and
event structure, D. P. Hole, W. Abraham, and A. Meinunger (eds.), 103
139. Amsterdam: John Benjamins.
Croft, W. 1991. Syntactic Categories and Grammatical Relations: The
Cognitive Organization of Information. Chicago: University of Chicago
Press.
Curnow, T. J. 2003. Nonvolitionality expressed through evidentials. Studies
37

in Language 27: 3959.

de Hoop, H. and B. Narasimhan. 2005. Differential case-marking in Hindi.
In Competition and Variation in Natural Languages: The Case for Case,
M. Amberber and H. de Hoop (eds.), 321345. Amsterdam: Elsevier.
Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel.
Gair, J. W. 1970. Colloquial Sinhalese Clause Structures. The Hague:
Mouton.
Gair, J. W. and J. Paolillo. 1988. Sinhala non-verbal sentences and argument
structure. In Cornell Working Papers in Linguistics 8 (Special Issue on
South Asian Linguistics), V. Srivasta, J. W. Gair, and K. Wali (eds.), 39
77. Ithaca: Cornell University.
Gair, J. W. and J. C. Paolillo. 1997. Sinhala. Munchen: Lincom Europa.
Henadeerage, D. K. 2002. Topics in Sinhala Syntax. Ph. D. dissertation,
The Australian National University.
Holisky, D. A. 1987. The case of the intransitive subject in Tsova-Tush
(Batsbi). Lingua 71: 103132.
Hopper, P. J. and S. A. Thompson. 1980. Transitivity in grammar and discourse. Language 56: 251299.
Inman, M. V. 1993. Semantics and Pragmatics of Colloquial Sinhala Involitive Verbs. Ph. D. dissertation, Stanford University.
Kiparsky, P. 1992. Structural case. Unpublished ms., Institute for Advanced
Study, Berlin.
Kratzer, A. 1977. What must and can must and can mean. Linguistics
and Philosophy 1: 337355.
Kratzer, A. 1981. The notional category of modality. In Words, Worlds and
38

Contexts, H.-J. Eikmeyer and H. Rieser (eds.). Berlin: de Gruyter.

Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein (ed.), 197235. Dordrecht: Kluwer.
Lazard, G. 1998. Actancy. Berlin: Mouton de Gruyter.
Lazard, G. 2003. Transitivity revisited as an example of a more strict approach to typological research. Folia Linguistica 36: 141190.
Levin, B. and M. Rappaport Hovav. 1995. Unaccusativity: At the SyntaxLexical Semantics Interface. Cambridge: MIT Press.
Malchukov, A. 2005. Case pattern splits, verb types and construction competition. In Competition and Variation in Natural Languages: The Case
for Case, M. Amberber and H. de Hoop (eds.), 73118. Amsterdam: Elsevier.
Malchukov, A. 2006. Transitivity parameters and transitivity alternations:
Constraining co-variation. In Studies on Case, Valency and Transitivity,
L. Kulikov, A. Malchukov, and P. de Swart (eds.). Amsterdam: John Benjamins.
Mohanan, K. P. and T. Mohanan. 1990. Dative subjects in Malayalam: Semantic information in syntax. In Experiencer Subjects in South Asian
Languages, M. K. Verma and K. P. Mohanan (eds.), 4357. Stanford:
CSLI Publications.
Paolillo, J. C. 1997. Sinhala diglossic variation: Continuous or discrete.
Language in Society 26: 26996.
Ramchand, G. 2008. Verb Meaning and the Lexicon: A First Phase Syntax.
Cambridge: Cambridge University Press.
Rappaport, M. and B. Levin. 1988. What to do with -roles. In Thematic
Relations, W. Wilkins (ed.), 736. San Diego: Academic Press.
39

Rappaport Hovav, M. 2008. Lexicalized meaning and the internal structure

of events. In Theoretical and Crosslinguistic Approaches to the Semantics
of Aspect, S. Rothstein (ed.). Amsterdam: John Benjamins.
Rappaport Hovav, M. and B. Levin. 1998. Building verb meanings. In The
Projection of Arguments: Lexical and Compositional Factors, M. Butt
and W. Geuder (eds.), 97133. Stanford: CSLI Publications.
Tenny, C. 1994. Aspectual Roles and the Syntax-Semantic Interface. Dordrecht: Kluwer.
Tsunoda, T. 1981. Split case-marking in verb-types and tense/aspect/mood.
Linguistics 19: 389438.
Tsunoda, T. 1985. Remarks on transitivity. Journal of Linguistics 21: 385
396.
Van Valin, R. D. and D. P. Wilkins. 1996. The case for effector: Case
roles, agents, and agency revisited. In Grammatical Constructions: Their
Form and Meaning, M. Shibatani and S. A. Thompson (eds.), 289322.
Oxford: Oxford University Press.
Verma, M. K. and K. Mohanan. (eds.) 1990. Experiencer Subjects in South
Asian Languages. Stanford: CSLI Publications.
Wijayawardhana, G. D., D. Wickramasinghe, and T. Bynon. 1991. Passiverelated constructions in colloquial Sinhala. In Subject, Voice and Ergativity, D. C. Bennet, T. Bynon, and B. G. Hewitt (eds.), 105141. London:
Routledge.
Wunderlich, D. 1997. Cause and the structure of verbs. Linguistic Inquiry 28: 2768.
Wunderlich, D. and R. Lakamper. 2001. On the interaction of structural and
semantic case. Lingua 11: 277418.
40

Endnotes
1

We would like to thank Patrick Brandt, Marco Garcia, Andrej Malchukov, and an

anonymous reviewer for their very useful comments on an earlier draft. Thanks also to
Andrew Koontz-Garboden, John Peterson, Farzan Zaheed, Bettina Zeisler, and audiences
at the Transitivity Workshop at Cologne and the 2008 LSA Annual Meeting for their feedback. We would also like to thank our informants. Any errors or omissions are our own.
2

There are of course other modalities in Sinhala (e.g. future tense) that vary indepen-

dently; we refer here just to the specific modality associated with (in)volitive-marking.
3

Alternatively, volitive-marking could be a type of evidential-marking, which is inde-

pendently known to express non-volitionality (Curnow 2003). However, while evidentials

and involitivites clearly overlap in functionality, as Curnow (p.40) notes, the key feature
of evidentiality is reference to the source of the information conveyed by the proposition, something not encoded by volitivity-marking. Thus volitive-marking is not a type of
evidential-marking, though there is clearly a semantic link; we leave this for future work.
4

There is some evidence of this in other languages as well. Brandt (2006) notes a paral-

lel situation with German cipient-datives in too-comparatives similar to The soup is too hot
(to me). Brandt argues that the predicate expressing too hotness generates a presupposition
that there is some other index (world) in which this state of affairs does not obtain, and this
index is associated with the dative-marked participant, which he analyzes as a predicate
internal subject of sorts. Thus again computing possible worlds results in A-marking.
5

The difference between (26) and the volitive variant in (22a) is that the involitive typ-

ically conveys a meaning that the blowing is unexpected/unexplained to the speaker, i.e. a
doxastic use. To simply say that the wind blows, the volitive is the preferred form. This is as
expected on our approach. Since the subject has an empty intention set, the non-volitional
reading of the involitive means the same as the volitive. In this case we would expect the
unmarked volitive to be preferred, leaving the involitive for uses that do not code subject
intentions, such as the doxastic. Thanks to an anonymous reviewer for pointing this out us.
6

As noted in 3 Sinhala is a differential object-marking language, so that accusative

never occurs with inanimate objects and only optionally occurs with animates. However,
(42)-(43) show that overtly accusative subjects have a different reading from overtly nominative ones, suggesting that when marking subjects these cases are not in free variation.

41