Nominalizations and aspect

by
Andrés Pablo Salanova
Submitted to the Department of Linguistics and Philosophy
in partial fulfillment of the requirements for the degree of
Doctor of Philosophy in Linguistics
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
September 2007
c Andrés Pablo Salanova, MMVII. All rights reserved.
The author hereby grants to MIT permission to reproduce and
distribute publicly paper and electronic copies of this thesis document
in whole or in part.

Author . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Department of Linguistics and Philosophy
August 30th, 2007

Certified by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sabine Iatridou
Professor of Linguistics
Thesis Supervisor

Accepted by . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Irene Heim, Professor of Linguistics
Head, Linguistics Program
2
 Nominalizations and aspect
by
Andrés Pablo Salanova

Submitted to the Department of Linguistics and Philosophy

on August 30th, 2007, in partial fulfillment of the
requirements for the degree of
Doctor of Philosophy in Linguistics

Abstract
Languages that have aspectually-conditioned ergativity splits generally oppose a “per-
fect” tense (often called perfective or aorist), with ergative-absolutive case pattern,
to an imperfective where case marking follows the nominative-accusative pattern.
The split exists in main clauses in several northern Jê languages, among which
Mẽbengokre, though in a slightly different form. Mẽbengokre opposes two verbal
forms that roughly express an aspectual opposition between a “perfect”, and a per-
fective or unmarked aspect. Rather than being two forms of the verb that differ
simply in an aspectual feature, however, these forms (herein referred to as A and B,
respectively) differ in many important respects:

1. Form A:

(a) has a wide range of temporally stative interpretations when not embedded;
(b) heads ergative-absolutive clauses;
(c) is the only verbal form that can be embedded;
(d) when embedded, its temporal and aspectual interpretation depend on that
of the main clause;

2. Form B:

(a) has a perfective interpretation; advances narrative time;

(b) heads nominative-accusative clauses;
(c) can’t be embedded.

In this dissertation, I propose that the opposition between the A and the B form
boils down to an opposition between a truly verbal form (the B form) and a nominal
form of the verb (the A form), and that the change in category explains both the
ergative marking and the perfect interpretation associated with the A form. I argue
that nominalization underlies many aspectually-conditioned splits described in the
literature, as well as being at the core of the perfect construction in languages such

3
as French and Italian. For the analysis to go through, two propositions have to be
worked out: (i) that ergativity is a given when there is nominalization, and (ii) that
the interpretation of a nominalization used as a main clause is in fact that of the
perfect.
To work out (ii), matrix clauses constructed with nominal forms of the verb are
treated as a special case of existential sentences, which in Mẽbengokre are verbless
clauses of the form [[Location]y [NP]x ].
I propose that the interpretation of nominalizations as main clauses, like the in-
terpretation of nominal clauses, is effected by the existential frame “There is an x
in y”, i.e., one where the main “predicate” is the nuclear scope x of an existential,
which requires a locative restriction y. In existentials constructed with plain nomi-
nals, this restrictor is provided by the locative, dative or possessive PP. In existentials
constructed with a nominalization, the restrictor is a time span. This span, which
is distinct from topic time, is what gives nominal clauses their “subject-oriented” or
“background” interpretation, as opposed to truly verbal clauses, which get linked to
topic time and are interpreted perfectively by default.

Thesis Supervisor: Sabine Iatridou

Title: Professor of Linguistics

4
 To Kwỳrkrô, Bepmajti and Ngrejre

5
6
Acknowledgments

The first I wish to thank are the Xikrin and Kayapó that have shared their language
with me: my consultants Nory Kayapó, prematurely deceased, Bep Kamrêk Kayapó,
and Ikrô Kayapó. Before I dared to subject any of them to formal elicitation sessions, I
spent several trips to the field walking around with a notebook taking down everything
I heard, and asking anybody around me for clarification; so I have to thank the Kayapó
communities of Baú, Mẽtyktire, Kapôt and Kubẽkàkre, and the Xikrin communities
of Pykatingrà (Cateté) and Djudjêkô as a whole for putting up with me. It has been
an immense privilege, one afforded to very few researchers working with minority
languages in times of globalization, to work with a people as proud and vigorous
as the Mẽbengokre, who, despite the aggressive encroachment of settlers and the
sustained effort of missionaries and other agents, are still at the center of their own
universe.

I wish to thank the elders Ropkrori Xikrin, Poropot Xikrin and Katàmti Xikrin,
and Bepmotire Mẽkrãknõti, for recording texts, Bepnhı̃, Kôkôdjwỳrỳti and Pãnhkê
Mẽkrãknõti, among many others, for help in transcribing some of them.

I consider myself privileged to have had the dissertation committee I had: Sabine
Iatridou, Irene Heim, Maria Polinsky and Lisa Matthewson. Their influence on the
maturing of my ideas can be seen in every aspect of this dissertation. I thank Sabine
for guiding me through the difficult last few months of dissertation-writing, and Lisa
for our frequent meetings while I was in Vancouver.

I am very grateful to the linguistics faculty at MIT. In particular, I would like to

thank Cheryl Zoll and Morris Halle for their encouragement early on. Alec Marantz
was very influential in the planning stages of the dissertation, as were Léa Nash and
Rajesh Bhatt, during their semesters as visiting faculty.

For discussion of the ideas in this thesis, I would especially like to thank Lynn
Gowan (the Nicolas Bourbaki of modern linguistics), David Pesetsky, Norvin Richards,
Michel DeGraff, Danny Fox, Dave Embick, Victor Manfredi, Michael Wagner, An-
drew Nevins, Maria Amélia Reis Silva, Pranav Anand, Valentine Hacquard, Rajesh

7
Bhatt, Fernanda Pratas, Márta Abrusán, Julie Legate, Justin Fitzpatrick, as well
as the following people, who have given of their time to meet or correspond with
me: Mark Baker, Cristina Schmitt, Hakyung Jung, Rose-Marie Déchaine, Martina
Wiltschko, Vaneeta Dayal, Kai von Fintel, Elena Anagnostopoulou, Tom Roeper,
Tamina Stephenson, Artemis Alexiadou, Ivona Kučerová, Roni Katzir, Jessica Coon,
Asaf Bachrach, Ken Hiraiwa, Henry Davis, Tyler Peterson, Marcelo Ferreira, Flávia
Alves, Eduardo Rivail Ribeiro, Sérgio Meira, Francisco Queixalòs, Filomena Sandalo
and Bruna Franchetto. Thanks also to Flávia Alves and Christiane de Oliveira for
sharing unpublished materials about related languages that have been useful here.
I thank my friends at the department for discussions and encouragement: Cristina,
Daniel, Marcelo, Michael, Justin, MaryAnn, Allison, Shoichi, Pranav, Valentine,
Cristina Ximenes, Giorgio, Márta, Tamina, Maria, and Hye-Sun. I would also like
to thank the department headquarters staff — Chrissy, Jen, Stephanie and Mary —
who often went beyond the call of duty to be of help.
Going to the field would have been an ordeal if I hadn’t counted with the support
of Denny Moore and Geiva Picanço, from the Museu Goeldi, of Eimar Araújo and
many others at FUNAI, Karàngre, Sr. Francisco, Luana and Flávio at Kàkàrekre,
and of Clésia, Bela and Jacaré in the villages. I’m very thankful also to my friends
Wilmar, Juracilda, and family, to Carmen, Miguel and family, and to Martha, all of
whom have supported me from a distance, as well as hosting me in my visits to Brazil.
For the latter I would also like to thank Letícia and family, in Belém. Thanks go also
to the linguists in Buenos Aires and General Roca, in particular Laura, Andrés, Lucía
and Pablo, whose commitment to knowledge is an important source of inspiration for
me.
I am really thankful to the people that received me in their house in the Boston
area during these past few years: Susi, Roberta, Pranav, Nikki, Victor and Ivor.
Their company was most gratifying.
Lastly, but most importantly, I would like to thank my family — Tomás, Rafael,
Amélia, Pablo, Juani, Héctor, Gloria, Victoria, y la Nona — who have supported me
at every stage, in every possible way.

8
 Part of the field research conducted for this dissertation was funded by National
Science Foundation dissertation improvement grant number 0639870.

9
10
Contents

Introduction 15
0.1 The puzzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
0.2 Road map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
0.3 A note on methodology . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1 Mẽbengokre verbal morphology, clause structure and case 21

1.1 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.2 The structure of main clauses . . . . . . . . . . . . . . . . . . . . . . 22
1.3 Verbal morphology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.1 Category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.3.2 Number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.4 Case in Mẽbengokre . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.1 Absolutive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.2 Accusative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.3 On the referential nature of object prefixes . . . . . . . . . . . 33
1.4.4 Nominative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
1.4.5 Case in non-pronominal noun phrases . . . . . . . . . . . . . . 36

2 Nominalizations and ergativity 39

2.1 Ergativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 A brief survey of ergative splits . . . . . . . . . . . . . . . . . . . . . 43
2.2.1 Person splits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.2.2 Aspect splits . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

11
 2.2.3 Split S systems . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.2.4 Ergativity in nominalizations . . . . . . . . . . . . . . . . . . 51
2.3 Ergativity in Mẽbengokre . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.1 Ergative is the alignment of embedded clauses . . . . . . . . . 56
2.3.2 The progressive construction . . . . . . . . . . . . . . . . . . . 59
2.4 Summing up: nominalizations and ergativity . . . . . . . . . . . . . . 63
2.5 A formal analysis of case . . . . . . . . . . . . . . . . . . . . . . . . . 65
2.5.1 Case in Mẽbengokre noun phrases . . . . . . . . . . . . . . . . 67
2.5.2 Case in nominative-accusative clauses . . . . . . . . . . . . . . 69
2.5.3 Case in embedded clauses . . . . . . . . . . . . . . . . . . . . 72
2.6 The source of main clause ergativity . . . . . . . . . . . . . . . . . . . 75

3 The structure and interpretation of nominalizations 77

3.1 Internally-headed relative clauses . . . . . . . . . . . . . . . . . . . . 78
3.2 The structure of IHRCs . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3 The basic semantic analysis . . . . . . . . . . . . . . . . . . . . . . . 83
3.4 Eventive complement clauses . . . . . . . . . . . . . . . . . . . . . . . 85
3.5 The semantics of negation, manner modification, and short nominal-
izations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.5.1 Short nominalizations . . . . . . . . . . . . . . . . . . . . . . 87

4 The interpretation of matrix nominals 89

4.1 The interpretation of verbs . . . . . . . . . . . . . . . . . . . . . . . . 91
4.1.1 The event argument . . . . . . . . . . . . . . . . . . . . . . . 91
4.1.2 What are verbs? . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.1.3 Excursus on viewpoint aspect and tense . . . . . . . . . . . . 97
4.2 Verbal number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.3 The interpretation of matrix nominal forms . . . . . . . . . . . . . . . 102
4.3.1 Stativity as a property of nominal forms . . . . . . . . . . . . 104
4.3.2 Stativity as a property of the construction . . . . . . . . . . . 107
4.4 The parallel with existential constructions . . . . . . . . . . . . . . . 108

12
 4.4.1 Nominal forms of verbs in existential constructions . . . . . . 115
4.4.2 The location . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.4.3 Provisional summary . . . . . . . . . . . . . . . . . . . . . . . 120
4.5 Obtaining the experiential perfect . . . . . . . . . . . . . . . . . . . . 121
4.5.1 The experiential perfect . . . . . . . . . . . . . . . . . . . . . 122
4.5.2 Excursus on perfect and perfective . . . . . . . . . . . . . . . 124
4.6 Speculations on generics and habituals . . . . . . . . . . . . . . . . . 127
4.6.1 The meaning of plural . . . . . . . . . . . . . . . . . . . . . . 127
4.6.2 Projection into the future . . . . . . . . . . . . . . . . . . . . 129
4.6.3 Excursus on perfects and generics . . . . . . . . . . . . . . . . 130
4.6.4 More on the modality in imperfectives . . . . . . . . . . . . . 133
4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

Envoi 139

Bibliography 143

13
14
Introduction

Hace diez años bastaba cualquier simetría con apariencia de orden — el

materialismo dialéctico, el antisemitismo, el nazismo — para embelesar
a los hombres. ¿Cómo no someterse a Tlön, a la minuciosa y vasta
evidencia de un planeta ordenado? Inútil responder que la realidad
también está ordenada. Quizá lo esté, pero de acuerdo a leyes divinas
— traduzco: a leyes inhumanas — que no acabamos nunca de percibir.

J. L. Borges, Tlön, Uqbar, Orbis Tertius

When one studies any language in detail, one often finds several traits that stand
out as the grammatical system’s identity, so to speak. This study concerns primarily
the Mẽbengokre language, spoken by the Xikrin and Kayapó nations in central Brazil.1
In the case of Mẽbengokre, we identify the following traits as being central to its
grammar: (a) the ambiguity of noun phrases between referential and clausal readings,
(b) the pervasiveness of nominalizations, and (c) the ergative split that accompanies
a contrast in the aspectual value of verbal predicates. These are exemplified in order
in examples (1-3):

(1) kubẽ ñõ k2

barbarian poss canoe
a. “The white man has a canoe.”

1
The Mẽbengokre language belongs to the northern branch of the Jê family, a language family that
is wholly contained within the boundaries of Brazil (though for a period in the 19th and early 20th
century the Kaingang, from the family’s southern branch, established villages in the northeastern
tip of Argentina). For a family tree of the Jê languages, see Rodrigues (1999).

15
 b. “The white man’s canoe.”

(2) a. ba [kutE tEp krẽn] pumũ

1nom 3erg fish eat.n see.v
“I saw him eating fish.”

b. ba [kutE tEp krẽn] m7r7ri ipej

1nom 3erg fish eat.n during do.v
“I did it while he was eating fish.”

c. kutE tEp krẽn mEj

3erg fish eat.n good
“He eats fish properly.”

d. kutE tEp krẽn ket

3erg fish eat.n neg
“I haven’t eaten the fish.”

(3) a. ba tẽ
1nom go.sg.v
“I went.”

b. ba ku-kwW̃r
1nom 3acc-break.sg.v
“I broke it.”

c. i-tẽm
1-go.sg.n
“I’ve gone.”

d. ijE kwW̃ñ
3erg 3.break.sg.n
“I’ve broken it.”

Example (1) straightforwardly shows that a noun phrase can stand by itself, and
have a clausal meaning. The examples under (2) show that nominalizations are called
for in various contexts, including negation, modification by a manner predicate, and
eventive complements of perception verbs. That these are nominalizations has of
course to be taken on faith for now.
Finally, the examples under (3) show that two alignments coexist in the language,
and are related to some aspect of verbal meaning (which we have translated rather
vaguely as a contrast between a simple past and a perfect).

16
 In this dissertation, we have endeavored to explain this last trait of Mẽbengokre
by recourse to the other two. Hence, for expository reasons, we begin by advancing
a puzzle that pertains to split ergativity, and work our way back to the other traits,
which we consider more fundamental. The puzzle is as follows.

0.1 The puzzle

The initial motivation for our inquiry is a contrast between the form and interpre-
tation of the two forms in a verb’s paradigm in Mẽbengokre. The distribution and
interpretation of these forms, which we temporarily label A and B,2 can be summa-
rized as follows:

1. Form A:

(a) has a range of temporally stative interpretations, among which the perfect
described above, when not embedded;

(b) heads ergative-absolutive clauses;

(c) is the only verbal form that can be embedded;

(d) when embedded, its temporal and aspectual interpretation depend on that
of the main clause;

2. Form B:

(a) has a perfective3 simple past interpretation; advances narrative time;

(b) heads nominative-accusative clauses;

(c) can’t be embedded.

The question is simply whether there is reason to believe that such disparate traits
as characterize the two forms of the verb come together for a principled reason. Our
answer is of course yes, and the reason we advance is based on the following claims:
2
Form A is exemplified above in (3c) and (3d), and all of the embedded sentences in (2). Form
B can be seen in (3a) and (3b), and the matrix verbs in (2a-b).
3
This aspectual label should be considered provisional.

17
 • Form A of verbs is nominal.

• Ergativity in nominalizations is a given.

• Embedded clauses and matrix clauses with Form A have the same structure
(and denotation) at their core.

• The interpretation of matrix clauses with Form A arises from independent prin-
ciples governing the interpretation of all nominal matrix clauses.

An obvious objection could be raised at this point: wouldn’t it be simpler to say

that Forms A and B just differ in a particular feature value, i.e., [+/–perfective]?
Such an analysis would be unsatisfying for the following reasons:

1. No connection would made between all the contexts where Form A is used.

2. It would say nothing about the contrast in aspectual meaning between main
clause and embedded uses of Form A.

3. The ergative pattern that accompanies A Forms would remain unexplained.

In addition, with our work on Mẽbengokre we wish to show that nominalization

can be plausibly argued to be at the core of aspect-driven ergativity splits in a wide
array of languages.

0.2 Road map

In chapter 1 of the dissertation, we give a general outline of the verbal morphology
of Mẽbengokre, and of the morphological manifestation of different case categories in
the language. In chapter 2, we introduce ergativity and ergativity splits, and propose
a theory of how nominative-accusative and ergative-absolutive alignments arise which
gives a central role to the category label of the main predicate’s extended projection.
In chapter 3, we propose a general structure and semantics for nominal projections in
Mẽbengokre. Finally, in chapter 4, we show how the structures introduced in chapter

18
3 are put together with higher clausal projections to yield sentences, and in particular
how the aspectual meaning of such nominal clauses gets to be what it is.

0.3 A note on methodology

The majority of the examples presented in this dissertation correspond to structures
which we have heard as spontaneous utterances in our eleven years of work with
Mẽbengokre speakers. Establishing their precise meaning is however a tricky prob-
lem. For the most part, we have based ourselves on the following methodology: we
presented consultants with a discourse context in Mẽbengokre, constructed by us,
and culminating in the sentence whose semantics we were interested in. We asked
consultants for judgements on the felicity of sentences in the given contexts, and for
a translation of the narrative (and the culminating sentence) into Portuguese. Where
relevant, we have given indication of the context in which particular sentences can
be uttered. Though often for simplicity we just provide translations, it should be
borne in mind that these are translations that one arrives at by considering what
would be felicitous in a particular context, rather than by taking down a consultant’s
translation ipsis verbis.
The method was repeated with certain variants, to confirm our results: (a) pre-
senting the context in Portuguese (with the culminating sentence still being presented
in Mẽbengokre), and (b), presenting the context and the culminating sentence in Por-
tuguese, and asking for the most adequate translation of the culminating sentence into
Mẽbengokre.

19
20
Chapter 1

Mẽbengokre verbal morphology,

clause structure and case

The goal of this chapter is to provide a description of some core aspects of Mẽbengokre
morphosyntax and define some notions that will be employed in later chapters. The
assumptions and conclusions in this chapter are generally in line with those of previous
work on Mẽbengokre and other Jê languages. A brief survey of this work is provided
in §1.1.
The description is divided as follows: §1.2 deals with the basic structure of main
clauses and the classification of predicate types; §1.3 examines the morphological
makeup of verbal stems; §1.4 discusses the manifestation of morphological case in
Mẽbengokre, a prerequisite for the discussion of its split ergative system, which is the
subject of chapter 2.

1.1 Previous work

Though previous work on Mẽbengokre that is relevant to our present purposes is
limited to Reis Silva and Salanova (2000) and Reis Silva (2001), several languages
of the northern branch of the Jê family are close enough to Mẽbengokre that their
descriptions bear on our discussion, and vice-versa. Apinayé, the closest among these,
is described by de Oliveira (2005); a previous description by Callow (1962) gives great

21
detail about its morphology and phonology, but relatively little about the topics that
interest us here. Timbira, also quite closely related to Mẽbengokre, has been described
by Alves (2004); older published descriptions can be found in Popjes and Popjes
(1986) and Shell (1952). Suyá has been described by Santos (1997), and Panará,
slightly more distant from the rest, is the subject of Dourado (2001). Neither of the
latter have been given consideration in this study.
In the southern branch of the Jê family, one finds the work of Urban (1985)
discussing ergativity in Xokleng, and Wiesemann’s (1972) description of Kaingang.
Wiesemann (1986) is a survey of the pronoun systems of several Jê languages. We
didn’t have access to any usable source on the syntax of central Jê languages.
In recent times, a fair amount of work has been produced on the subject of erga-
tivity in Jê languages, partly in response to the new synthesis of the facts put forth
by Reis Silva (2001). Most of this work, known to us solely from conference pro-
grams, remains unpublished, and wasn’t accessible to us during the preparation of
this dissertation.

1.2 The structure of main clauses

Word order in Mẽbengokre is fairly rigid. The following diagram shows the position of
constituents in a matrix clause. Tense and Aspect are the positions for two separate
classes of particles, with roughly temporal and (viewpoint) aspectual meaning:1

(4) left field middle field

Focus Tense Subject Aspect Adjuncts Object Predicate

Most grammatical descriptions include early on a section devoted to establishing

the lexical categories found in the language under study. The distinction between the
class of nouns and the class of verbs is of great importance for all the discussion that
is to follow, but we will not be too concerned about the remaining categories. We

1
It’s not clear whether there is a “right field” for extraposed constituents in Mẽbengokre or
if these only appear exceptionally as “afterthoughts”, but in any case this aspect of Mẽbengokre
syntax doesn’t enter consideration in the present work.

22
introduce a couple of notions that will only be fully explained in §1.3 and §1.4, namely
the opposition between two forms of verbs, and the opposition between different case
forms of pronominal elements, as the main criteria to establish the opposition in
lexical category between nouns and verbs.
Predicates can be classified descriptively on the basis of the number of arguments
they take and the morphological case that they assign to them. The following table
summarizes the different predicate types found in Mẽbengokre matrix clauses:

(5) Type I Type II Type III Type IV Type V

Form A erg–abs erg–abs abs abs dat–abs
Form B nom–acc nom–abs nom — —
Examples of each of these types will be given in §1.3. In a move that at this
point will sound irremediably circular, types I, II and III will be called verbs, based
on the criteria mentioned above, i.e., that they oppose two forms (“A” and “B”) in
their paradigm, and that this correlates with the fact that they exhibit two different
patterns of case-marking their arguments. The circularity of this definition will be
gradually mitigated, as we discuss other morphological properties that line up with
this definition of the categorial opposition.

1.3 Verbal morphology

While non-verbal predicates are generally unanalyzable,2 verbal stems are often mor-
phologically complex. In this section, we provide a quick overview of the morpho-
logical categories expressed on verbs. We will concentrate initially on form and on a
relatively superficial characterization of their function. In later chapters, and partic-
ularly in chapter 4, we will be much more precise about their semantics.
The maximal template of the verb in this language is given in (6).3 The morphol-
ogy that is most important for the purposes of this dissertation is what in the template
2
Except for compounding, which is quite productive in Mẽbengokre, and the “nominal applicative”
(cf. fn. 29, on p. 68).
3
The transcription used throughout the thesis is quite abstract, and is chosen to highlight mor-

23
is called “category”, and number. The former was already introduced above as the
contrast between the two verbal forms. Hereinafter we will call Form A nominal, and
Form B verbal.

(6) a. person – voice/valence – number/class – root – category

b. i-bi-ku-nO-r
1-intr-class-lose-n
“I’ve gotten lost.”

As far as morphological structure goes, category seems to be the affix closest to the
root: class and voice prefixes are sensitive to whether the stem they are attaching to is
verbal or nominal. Choice of category, on the other hand, can trigger root suppletion,
but never suppletion of a constituent larger than the root, whereas number can trigger
suppletion of the whole stem.

1.3.1 Category

In previous versions of our work, we identified the opposition between the two forms
of verbs as an opposition between “finite” (form B or the verbal form) and “non-finite”
(form A or the nominal form) forms of the verb. Though the intuition behind these
labels seems to us to be correct, as form A has many of the properties of participles
(though not infinitives) in better-known languages, we wish to highlight the parallel
between the interpretation of nominal forms of verbs and underived nouns, and to
avoid confusion with what it means to be finite or non-finite in the grammar of more
familiar languages. In particular, we wish to exclude the possibility of implying that
there are non-finite forms of verbs in addition to their nominal(ized) form. No such
distinction exists.
The semantics of “category” is at the core of what will be discussed in chapters 2
and 4. As far as form goes, verbal and nominal forms of verbs contrast in that the

phological structure rather than to approximate pronunciation; for a discussion of Mẽbengokre

phonology and morphology, cf. Salanova (2001), Salanova (2004), and references therein.

24
latter normally have an extra final consonant that is idiosyncratically determined by
the root:

(7) a. tẽ, tẽm

go.sg.v, go.sg.n

b. mõ, mõr
go.pl.v, go.pl.n

c. rw7, rw7k
go.down.sg.v, go.down.sg.n

d. mrã, mrãñ
walk.v, walk.n

There are some cases with complicated morphophonology, and also suppletion of
the root (ka- is a class/number prefix compatible with both nominal and verbal forms
of verbs):

(8) a. ka-tE, ka-Pek

class-break.v, class-break.n

b. ka-ba, ka-dZ2r
class-take.out.sg.v, class-take.out.sg.n

c. Nõr, ñõt
sleep.v, sleep.n

In Salanova (2001), I suggested that the nominal form should be considered basic,
since the shape of the verbal form could be predicted from it, modulo suppletion, but
not vice-versa. More detailed examination of the regularities found across all known
verbs suggests that the shape of neither form can be predicted straightforwardly from
the other in the general case. We omit consideration of the relevant examples here;
the interested reader may consult Salanova (2004). In this dissertation we will be
agnostic about segmentation, and simply gloss nominal forms as root.n, and verbal
forms as root.v.

25
 As we stated above, category correlates with patterns of case-marking of depen-
dents: with verbal forms of verbs, subjects are nominative and objects accusative;4
with nominal forms, transitive subjects are marked ergative, and intransitive subjects
and objects are marked absolutive:5,6

(9) a. ba ar7m ku-ma

1nom already 3acc-hear.v
“I already heard it.”

b. ijE a-kabẽn mar ket

1erg 2-speech hear.n neg
“I haven’t heard you speak.”

(10) a. ba kãm prW kot tẽ

1nom then path along go.v
“I then went on the (animal’s) path.”

b. b2 kam i-tẽm ket

forest in 1-go.n neg
“I haven’t gone into the woods.”

Predicates of Types IV and V have a single form, that patterns like the nomi-
nal form of true verbs, in that the first argument gets absolutive, while the second
argument (of Type V) is expressed by means of an oblique:7

(11) a. i-NrWk
1-angry
“I’m angry.”

4
In Type I transitive verbs. Another class of transitive verbs (Type II) marks its objects in the
absolutive, regardless of category. We return to this below.
5
In chapter 2, we will see that this simple generalization seems to be contradicted by the progres-
sive construction; however, the progressive will be analyzed as bi-clausal, preserving the correlation
made here.
6
Note that absolutive case is not indicated in the glosses (only the person is glossed).
7
In the case of experiencer predicates such as (11b), the subject is dative, rather than ergative
as with the nominal forms of predicates of Type I and II.

26
 b. i-m2̃ piPok jã kı̃ñ
1-dat book this pleasant
“I like this book.”

Clauses involving predicates of Types IV and V are morphosyntactically parallel

to possessive sentences, in which the predicate slot is occupied by a noun:8

(12) i-kra
1-child
“I have a son or daughter.”

It could thus be said that nominal sentences are a special case of predicates of
Types IV or V, or vice versa. In chapter 4, we will delve into the analogy between
possessive (and, more generally, “existential”) constructions and sentences involving
predicates of Types IV and V, and nominal forms of predicates of Types I, II and III.

1.3.2 Number

Many, but certainly not all, Mẽbengokre verbs present a contrast for “number”, which
on the surface seems to be agreement with the subject of intransitive verbs, and
the object of transitive verbs. Morphologically, verbal number is often manifested in
class/number prefixes,9 but also through stem suppletion in verbs that have no such

8
If the noun involved in the possessive sentence is “alienably possessed”, its possessor is expressed
by means of a postposition, rather than by an absolutive pronoun:

kubẽ ñõ k2
barbarian poss canoe
“The white man has a canoe.”

9
The class/number prefixes are for the most part idiosyncratically selected by particular roots.
In the cases where two verb stems contrast only in the choice of prefix, the semantic contrast is often
one of number (involving ku- ‘singular’ vs. ja- ‘plural’), though a few examples exhibit a contrast
that has to do with other properties of the object or the action. Here we are not concerned with the
non-number components of the meaning of these prefixes.

27
prefixes. The contrast between (13a) and (13b) is an example of number being coded
in the prefix:

(13) a. ku-oj
“light a fire” (often used intransitively)

b. ja-oj
“light several items on fire”

An example where the suppletion targets the whole stem is given by the follow-
ing example, where a morphologically simple stem is in a suppletive relation with a
morphologically complex one (for a simpler example, see the contrast between (7a)
and (7b) above):

(14) a. rw7k
“go down (sg.n)”

b. bi-ja-dZw7r
intr-class-put.down.n
“go down (pl.n)”

It is fair to ask why such widely divergent verbal stems are considered to be part
of the same paradigm. The criterion to classify two (often suppletive) verbal stems as
instantiating a number opposition is simply the obligatory substitution of one for the
other in constructions where the number of the object or intransitive subject is ex-
plicitly marked as singular or plural, so here we are subordinating morphological form
to a paradigmatic opposition in the syntax. Also, as we will see in §4.2, the number
contrast is relevant for the determination of the aspectual value of predicates. For
these reasons, we consider this opposition to be systematic, despite its morphological
quirks and the fact that not all verbs exhibit it.

We will not discuss argument structure alternations here. Our description of

verbal morphology therefore ends here.

28
1.4 Case in Mẽbengokre

The present section discusses the case forms of pronouns and extends the notion of
case to non-pronominal noun phrases, even though case isn’t manifested morpho-
logically in them at all in Mẽbengokre. The discussion follows and expands on the
exposition in Reis Silva and Salanova (2000) and Reis Silva (2001). The purpose of
this discussion is to provide the morphological facts that ground the discussion of
split ergativity in chapter 2.

Mẽbengokre distinguishes between four case forms in pronouns: ergative, absolu-

tive, nominative and accusative. The forms for accusative and absolutive are identical
in all persons but the third. The following table, based on Wiesemann (1986), gives
the different forms for the singular pronouns. Number (singular vs. paucal vs. plural)
is normally marked with a freestanding particle, but there is some fusion in the first
person inclusive which need not concern us.10

(15) Nominative Ergative Absolutive Accusative

1 excl. ba ijE i- i-
1 incl. gu (gu) bajE (gu) ba- (gu) ba-
2 ga ajE a- a-
3 ta/∅ kutE ∅ ku-

As the table indicates, absolutive and accusative pronouns are prefixes, while
ergative and nominative forms are freestanding. Below, we establish that all of the
case forms are pronominal.

10
First person inclusive means “inclusive of the hearer”, and likewise for first person exclusive. The
singular of the first person inclusive, of course, refers to two people (you and I ). The fact that there
is fusion of the number particle in the first person inclusive explains the slightly different distribution
of gu, which is optionally present with absolutive and accusative inflection, contrary to what occurs
with ba and ga. In what follows, we will describe the properties of pronouns based on the behavior
of first person exclusive and second person, on the one hand, and third person, on the other.

29
1.4.1 Absolutive

The absolutive forms occur in all branches of the Jê family, and in the Southern
branch, which consists of the Kaingang and Xokleng languages, they seem to be free-
standing pronouns that occur in all functions, without case accidence. It is plausible
to suppose that this is the case in previous stages of other present-day Jê languages.
Since it is the form with widest distribution in Mẽbengokre, we will gloss it simply
as 1, 2, 3, without an explicit case marker.
A peculiarity of third person absolutive inflection is that, while it is zero on most
stems, it is realized as the truncation of an initial consonant on stems beginning with
/j/, /dZ/, /ñ/ and /pW/:

(16) stem 3+stem

jaka aka ‘to be white’
dZir ir ‘to put’
pWma uma ‘to be frightening’

In Salanova (2004) we suggest that this peculiar behavior can plausibly be traced
back to a prefix in a previous stage of the language that provoked the truncation as
an onset fix. In this dissertation, we gloss the third person absolutive pronominal as
an element fused to the stem (i.e., 3.stem) whether it has a morphological reflex or
not.
Absolutive inflection is also used for possessors of inalienably-possessed nouns
(i.e., a structural genitive case). In fact, when we address the connection between
case-marking inside noun phrases and case marking in the clause in chapter 2, we will
conclude that the absolutive should be called genitive throughout.

1.4.2 Accusative

Accusative is minimally distinct from absolutive, and it is employed in only a subset of

transitive verbs. This has prompted efforts to merge the two, and explain the surface
contrast as based on prosodic demands, as in de Oliveira (2005), where the accusative
ku- is considered to be an allomorph of the absolutive ∅ used with monosyllabic

30
verbal stems, or on semantics, as in Reis Silva and Salanova (2000), where ku- is
said to indicate a particular noun class or specificity of the object (more individuated
animate entities).

While both approaches have a certain plausibility, neither completely fits the syn-
chronic facts. For one thing, ku- only occurs on transitive verbs, while ∅ is the only
one to occur on nominal forms of transitives and intransitives, as well as on inalien-
ably possessed nouns, regardless of their phonology or semantics. It is precisely for
this reason that we employ the labels “accusative” versus “absolutive”. Furthermore,
postpositions are arbitrarily divided into those that govern the accusative (m2̃ ‘to’,
be ‘in’, and jE ‘erg’) and those that govern the absolutive (O ‘with’, kot ‘along’, P2̃
‘on’, among others). The following table summarizes the facts.11

(17) a. break (tr. v.) ku-kwW̃r * ∅-kwW̃r

b. break (tr. n.) * ku-kwW̃ñ ∅-kwW̃ñ
c. go (intr. n.) * ku-tẽm ∅-tẽm
d. son (inal. n.) * ku-kra ∅-kra
e. with (abs. p.) * ku-kot ∅-kot
f. to (acc. p.) ku-m2̃ * ∅-m2̃

One is therefore forced to admit the existence of an accusative inflection distinct

from the absolutive.12 Another peculiar fact about ku- is that it’s substituted by the
second person prefix when the subject of a transitive verb is in the second person.

11
The abbreviations should be interpreted as follows: tr. v., transitive verb; tr. n., nominal form
of a transitive verb; intr. n., nominalized form of an intransitive verb; inal. n., inalienably possessed
underived noun; abs. p., postposition governing the absolutive; acc. p., postposition governing the
accusative.
12
The fact that the morphological contrast is only visible in the third person shouldn’t make
us hesitate: this is precisely what happens with the dative versus accusative distinction in the
pronominal clitics of most Romance languages.

31
(18) a. ba ku-bı̃ ‘I killed it’
1nom 3acc-kill
b. ga a-bı̃ ‘You killed it’
2nom 2-kill
c. * ga ku-bı̃
d. ba a-bı̃ ‘I killed you’
1nom 2-kill

As to how to explain the peculiarity that absolutive is used for the objects of
some verbs, we adopt a proposal by Reis Silva (in progress), who sets off from the
generalization that all verbs that assign absolutive to their objects are characterized
by extra morphology (a class/number prefix), which could be plausibly characterized
as applicative. Neither underived intransitives nor accusative-assigning transitives
have class/number prefixes:

(19) Accusative-assigning transitive: bı̃ ‘kill’, ma ‘hear’, etc.

VP
HH
DPacc V

(20) Absolutive-assigning transitive: ka-ba ‘take out’, ja-ko ‘blow’, etc.

VP
H
 HH
PP Pi +V
HH
DPabs ti

Objects introduced by the applicative morpheme get absolutive case because they
are the objects of the applicative morpheme, rather than of the verb.13 Direct objects
of inherently transitive verbs get accusative. This might have semantic consequences
which are out of the purview of this dissertation.

13
The apparent circularity of this statement will be attenuated once we discuss the formal mech-
anisms for case assignment in §2.5.

32
1.4.3 On the referential nature of object prefixes

Noun phrases are in complementary distribution with third person pronominal object
forms (in both the accusative and the absolutive):

(21) a. ku-bı̃ ‘(he) killed him’

3acc-kill
b. mrW bı̃ ‘(he) killed an animal’
animal kill
c. mrW ku-bı̃ ‘the animal killed him’ (not: ‘killed the animal’)
animal 3acc-kill

The pattern is not unfamiliar, as it is attested in object markers in Bantu (Bres-

nan and Mchombo 1987), in Romance dialects which don’t permit clitic doubling
(Cinque 1990), and claimed to be the case in several Amerindian languages from dif-
ferent families (Jelinek 1984). What does seem peculiar about the pattern found in
Mẽbengokre, as opposed to the languages mentioned (but is also characteristic at least
of a few other Jê, Tupian and Carib languages), is that the object markers, though
apparently attached to the stems, can be replaced by a non-pronominal noun phrase.
That is, in those words that require it, the absolutive argument is obligatory,14 but
it can be either a pronominal prefix or a non-pronominal noun phrase. We take this
to mean that the pronominal prefixes are referential (i.e., not agreement);15 objects
are considered to be clitic left dislocated whenever the object pronominal forms are

14
Anticipating our discussion in chapter 2, even in constructions involving control.
15
One shouldn’t read too much into the claim that the person markers are referential. In fact, first
and second person markers are literally referential, but third person markers can also be variables,
as will be seen in chapter 3, and as is already evident in (22). Syntactically, they are pronominal
arguments of the predicate that selects them.
On a related note, an alternative approach would say that the object markers themselves aren’t
referential, but rather are licensers of pro. We don’t have sufficient data to distinguish between these
two positions, which in the literature on Romance languages have generated a long debate. Note
that in this literature it has been remarked that clitic left-dislocation is incompatible with focus, so
examples such as (22) are prima facie problematic for the approach we adopt here.

33
present. Such situations arise when the object is in focus position, before the tense
markers nẽ or dZa (cf. (4)). Object wh- questions also require that there be an overt
object pronominal, with the wh- word occupying the same focus position:

(22) a. mrW nẽ ku-bı̃

animal nfut 3acc-kill.sg.v
“He killed an animal (focus)”

b. m7j nẽ ku-bı̃

what nfut 3acc-kill.sg.v
“what did he kill?”

Despite the caveats expressed in fn. 15, we will often, for simplicity, refer to all
of the forms in table (15) as pronominal forms, rather than distinguishing between
freestanding pronouns and person prefixes. The distinction between these is assumed
to be simply a morphological or prosodic one.

1.4.4 Nominative

Nominative forms of pronouns are likely to have originated from inflected auxiliaries,
i.e., from the fusion between an absolutive pronominal form and a left-peripheral
particle.16 Though synchronically it’s implausible to consider them auxiliaries in
Mẽbengokre, some aspects of their behavior are atypical in garden variety pronouns.
The unusual characteristic of nominative pronouns is that, in main clauses, they
can duplicate a subject that is already expressed lower in the clause by an erga-
tive, dative or absolutive pronominal form. These pleonastic nominative pronouns,
unavailable in embedded clauses, seem not to indicate any emphasis.17

(23) a. ba i-tẽm
1nom 1-go.n
“I go.”

16
Auxiliaries that agree in person and number with the subject subsist in Xokleng, a language of
the southern branch of the Jê family (cf. Urban 1985, p. 166-7).
17
In fact, they are topics, as subjects often are. With special prosody much more is possible, of
course, but this shouldn’t surprise us.

34
 b. ba ijE ir
1nom 1erg 3.put.down.n
“I put it down.”

We take the ability to be “duplicated” by a nominative pronoun in the position

between tense and aspectual particles as the primary diagnostic for subjecthood in
Mẽbengokre main clauses. In the following example one can see that, while a dative
subject can be doubled by a nominative pronoun, other datives, such as ethical datives
or indirect objects, can’t:18

(24) a. (ba) i-m2̃ jã kı̃ñ

1nom 1-dat this please

18
All of the sentences considered above can be further augmented with an emphatic pronoun, a
fact first described by Borges (1995).

a. ba nẽ ba tẽ
1nom nfut 1nom go.v
“I go.”

b. ba nẽ ba ku-bW
1nom nfut 1nom 3acc-put.down.v
“I put it down.”

c. ba nẽ ba i-tẽm
1nom nfut 1nom 1-go.n
“I go.”

d. ba nẽ ba ijE ir
1nom nfut 1nom 1erg 3.put.down.n
“I put it down.”

Pronouns in this position also take the nominative form. The difference between this position,
which we call Focus, and the former, is that while Focus can be coindexed with any argument lower
in the clause, the Nominative subject position can only be coindexed with the subject:

a. gai nẽ baj ai -pumũ

2nom nfut 1nom 2-see
“I saw you.”

b. * baj nẽ gai ai -pumũ

1nom nfut 2nom 2-see

There is also a difference in information structure, as is implied by the labels. The lower position
is topical, whereas the higher one is either presentational or contrastive focus.

35
 “I like this.”

b. ga (*ba) i-m2̃ jã N2̃

2nom 1nom 1-dat this give.v
“You give this to me.”

Similar examples could be produced showing a contrast between absolutive sub-

jects of intransitive verbs and absolutive objects of transitive ones.

1.4.5 Case in non-pronominal noun phrases

Non-pronominal noun phrases do not show any case accidence, except in the ergative,
where they are followed by kutE or tE. As we suggested in (15) above, kutE is in fact a
pronoun, and thus noun phrases followed by kutE are clitic left-dislocated. From the
morphology, though this is inconclusive, it would seem that tE is a reduced form of the
pronoun, rather than a case marker or postposition directly on the noun phrase, so
even in this case one would be dealing with a clitic left-dislocated construction. If non-
pronominal noun phrases followed by an ergative pronoun are clitic-left dislocated,
they plausibly occupy the same slot as nominative pronouns in constructions where
they appear pleonastically. This is a matter for future research.
Throughout the thesis, we will talk of noun phrases (or arguments) in Mẽbengokre
as absolutive, nominative, etc., to refer to noun phrases licensed in the same positions
as pronouns that receive absolutive, nominative, etc.19 Given that pronouns are
generally in complementary distribution with full noun phrases, this is not an unrea-
sonable move, but by it we don’t mean to imply that “abstract case” has to be a part
of our grammar. The question is again discussed in §2.5.
This concludes our description of the case facts of Mẽbengokre. We should make a
brief remark about the absence of any reference to agreement in the dissertation. The
assumption that we will make here is that all of the phenomena that are relevant to
our analysis of ergativity fall under the purview of case theory, rather than involving

19
With the caveat advanced in the previous paragraph about non-pronominal noun phrases in the
ergative.

36
agreement. It is for this reason that we have argued that the bound person forms
should be considered pronouns, rather than agreement markers. There is, however,
a clear case of agreement in Mẽbengokre, namely the “eccentric” agreement of the
object pronominal with a second person subject, shown in (18b). This case won’t be
taken into consideration, and should therefore be put aside for consideration in later
research.

37
38
Chapter 2

Nominalizations and ergativity

In the previous chapter we offered a description of the morphosyntax of Mẽbengokre,

with particular attention, in §1.4, to the manifestation of case in pronominal elements.
In this chapter, we relate these morphological cases described in the abstract to
the intricate split-ergative alignment found in Mẽbengokre. This chapter takes as a
starting point the findings of Reis Silva (2001), which is devoted to the subject of
case marking in Mẽbengokre and other Jê languages.
The alignments to be discussed are summarized in the following three pairs of
examples, all involving verbal predicates:

(25) a. ba tẽ
1nom go.sg.v
“I went.”

b. ba ku-kwW̃r
1nom 3acc-break.sg.v
“I broke it.”

(26) a. i-tẽm
1-go.sg.n
“I’ve gone.”

b. ijE kwW̃ñ
1erg 3.break.sg.n
“I’ve broken it.”

39
(27) a. ba i-tẽm O=mõ
1nom 1-go.sg.n instr=go.pl.v
“I’m going.”

b. ba kwW̃ñ O=mõ
1nom 3.break.sg.n instr=go.pl.v
“I’m breaking it.”

In example (25), subjects of both transitive and intransitive verbs are expressed
by a pronoun in the nominative, while the object of a transitive is expressed by
an accusative pronoun. In (26), both intransitive subjects and transitive objects
are expressed by an absolutive pronoun, while the transitive subject is expressed by
an ergative pronoun. In (27), the transitive object and the intransitive subject are
expressed by an absolutive pronoun, while subjects, both transitive and intransitive,
are expressed by a nominative pronoun.
The pattern in (25) can be described straightforwardly as a nominative-accusative
alignment. The pattern in (26) can be described equally straightforwardly as ergative-
absolutive. The pattern in (27) is not a “case alignment pattern” at all, since one of
the core arguments (the intransitive subject) is coded twice, once as absolutive and
once as nominative.
The different case alignments found in Mẽbengokre correspond to constructions
that could be characterized as differing in aspectual meaning. Before we can go deeper
into what conditions the different case patterns, however, we give a brief introduction
to ergativity and ergative splits.

2.1 Ergativity
Case, agreement, and word order treat different core participants of the clause in
different ways cross-linguistically. In English, the pre-verbal position and the agree-
ment of the inflected verb or auxiliary are reserved for the subject, whether it be of
transitive or intransitive verbs. In addition, subjecthood is indicated on a subset of
the pronouns of English by nominative case, which in main clauses is opposed to the
accusative case on pronouns in object function:

40
(28) a. I see you.

b. You see me.

c. I shout.

Marking direct objects as distinct from subjects, of course, is not the only conceiv-
able way of distinguishing participants in a clause. A priori, if one distinguishes three
primitive grammatical relations, A, S and O, where S stands for the single argument
of an intransitive verb, A for the transitive subject and O for the transitive object,1
the following five logical possibilities exist (cf. Comrie 1978, p. 332):

S S S S S

A O A O A O A O A O
neutral accusative ergative tripartite unattested

Figure 2-1: Logically possible alignments, given grammatical functions A, S and O.

With the exception of the last, all of these possibilities for case marking core
participants in the sentence exist in natural languages. In this dissertation we will
concern ourselves with the opposition between the most common of these case sys-
tems, i.e., the one seen in the English examples above, called nominative-accusative
or simply accusative, and the ergative-absolutive, or simply ergative, where the case
mark on the sole participant of an intransitive verb is identical to the case mark on the
object of the transitive verb.2 The contrast between the ergative and the accusative
systems is exemplified below with Dyirbal, which presents a clear cut ergative system,
and Quechua, which presents a straightforward accusative system, where all objects
(and not only pronouns, as in English) are distinguished by their marking from an
unmarked class that encompasses transitive and intransitive subjects:

1
We will assume that these primitives can be defined in a language-particular way on the basis of
semantic (position within a thematic hierarchy) or syntactic (ability to control and bind anaphors,
etc.) criteria.
2
The assumption is that the other two attested systems can be reduced to ergative or accusative,
or a combination of the two.

41
(29) Dyirbal (Dixon 1994): (30) Quechua:

a. yabu banaga-ny u a. mama jamu-n

mother return-nonfut mother come-3
“Mother returned.” “Mother comes.”

b. yabu Numa-Ngu bura-n b. tata mama-ta riku-n

mother father-erg see-nonfut father mother-acc see-3
“Father saw mother.” “Father sees mother.”
Though initially defined in terms of marking on dependents, the designations
ergative and accusative can be easily extended to encompass other means by which
participants’ functions are coded in a sentence’s syntax. As we mentioned above,
agreement and word order facts in English oppose objects to all subjects, transitive
and intransitive, and thus they could be said to display an accusative alignment
system, even though case is not marked in non-pronominal noun phrases. An example
of an agreement system with ergative alignment is argued by Urban (1985) to exist
in Xokleng.3 In Xokleng, verbs display agreement (through various morphological
means, which include suppletion) with the absolutive argument:

(31) a. tã wũ tẼ mũ

he 3nom go.sg active
“He went.”

b. ON wũ mũ mũ

they 3nom go.pl active
“They went.”

c. ON wũ ti pEnũ mũ

they 3nom he shoot.sg active
“They shot him.”

d. tã wũ mẼ ON pin mũ

he 3nom distrib they shoot.pl active

3
In §4.2 we will analyze the “number agreement” of Mẽbengokre as an interpretable marker of
event plurality, an analysis which presumably extends to the roughly commensurate facts of Xokleng.
For now, we accept Urban’s analysis as a reasonable description of the surface facts. Note that since
noun phrases in Xokleng aren’t obligatorily marked for number, the number on the verb is usually
the main indication of the number of a particular argument.

42
 “He shot them.”

In English and other SVO languages, word order could be said to follow a nominative-
accusative pattern, with the accusative noun phrase following the verb, and the nom-
inative one preceding it. We don’t know of any clear-cut example of ergativity man-
ifesting itself solely in word order.
In what follows, we will address ergativity as it manifests itself in case marking
on noun phrases and agreement in heads. In this way, we will skirt the difficult
question raised by so-called syntactically ergative languages, in which the absolutive
argument functions as the syntactic subject (or “pivot”) for the purposes of control
or correference in coordinated sentences. According to the typological literature,4
all syntactically ergative languages are also morphologically ergative, while in many
morphologically ergative languages a notion of subject conflating the A and S cate-
gories (like in nominative-accusative languages) is the relevant pivot for control and
correference in coordination. We won’t have the opportunity here of fully addressing
the issue of syntactic versus morphological ergativity, but it again comes up at the
end of this chapter.

2.2 A brief survey of ergative splits

It is often the case that a single language will have parts of its grammar that dis-
play ergative alignment, while others display accusative alignment. These “splits” in
case-marking are conditioned by several factors. According to the typological and
functional literature, such factors include tense, aspect, mood, person, the semantics
of the main predicate or of the arguments, and several others.
We will examine briefly three types of splits as traditionally defined: person splits,
splits conditioned by tense, aspect or mood, and splits conditioned by the semantics
of the predicate. The types of split are ordered from most to least frequent.5 Though

4
References to the typological literature herein are primarily to the surveys in Comrie (1978) and
Dixon (1994).
5
This section has benefitted extensively from notes to two lectures offered at MIT by M. Polinsky

43
our main concern in the dissertation is with aspectually-conditioned splits, surveying
the common types of ergativity splits serves the twofold purpose of introducing an
empirical domain that is relevant to the description of Mẽbengokre,6 and presenting
evidence that variation in this domain is much more constrained than one is led to
believe from superficial descriptions.

2.2.1 Person splits

In some split-ergative languages, a certain subset of nominal expressions receive case

marks in the ergative-absolutive system, whereas the remainder are case-marked in
the nominative-accusative system. The descriptive generalization that seems to hold
crosslinguistically is that if a person split obtains, it is preferentially first and second
person pronouns that are case-marked in the nominative-accusative system, whereas
inanimate nouns are case-marked in the ergative-absolutive system. There is room
for variation as to how the remainder of nominal expressions is divided between these
two classes, but in any case a nominal “animacy” (Silverstein 1976) or “empathy”
(Lehmann 1998) hierarchy seems to be respected, such that if a certain class of nouns
is marked in the accusative system, all nouns with greater or equal “animacy” or
“empathy to the speaker or hearer” will also be graced with nominative-accusative
marking. A not particularly fine-grained version of this hierarchy would be as follows:

1st and 2nd person ≫ 3rd person pronouns ≫ humans ≫ animates ≫ inananimates

The cutting point between ergative and accusative alignment is set arbitrarily in
a particular language at a certain point along this scale, with those noun phrases
to the left of the cutting point showing accusative alignment, and those to the right
ergative.

in March 2007.
6
Both aspect splits and splits conditioned by the semantics of the predicate occur in a superficial
description of Mẽbengokre. Though person splits don’t occur in Mẽbengokre, we consider them here
given their cross-linguistic pervasiveness.

44
 An example of a person-based ergativity split is given by the Panoan language
Wariapano, discussed by Valenzuela (2000), where pronouns pattern as nominative-
accusative, while other noun phrases pattern as ergative-absolutive:7

(32) Wariapano

a. Jabon-bi-ra ka-iní-kain
3pl-nom-evid go-incompl-3pl
“They are going.”

b. Ja-bi-ra ano pi-ini

3-nom-evid paca eat-incompl
“He is eating pacas (Agouti paca).”

c. Nojkon pajpa-n-ra kajpe tsajka-ke

1poss father-erg-evid lizard poke-compl
“My father poked the lizard.”

d. Nojkon koka-ra manish-no ka-ki

1poss uncle-evid forest-dir go-compl
“My uncle went to the forest.”

As a special case of person splits, one might include systems where case marking
on noun phrases exhibits one type of alignment, while agreement (or pronominal
inflection) exhibits a different one. The following Warlpiri data from Hale (1973)
(apud Jelinek 1984, p. 45) exemplify such a split:

(33) a. ngajulu-rlu ka-rna-ngku nyuntu-∅ nya-nyi

I-erg pres-1sg.nom-2sg.acc you-abs see-nonpast
“I see you.”

b. nyuntulu-rlu ka-npa-ju ngaju-∅ nya-nyi

you-erg pres-2sg.nom-1sg.acc me-abs see-nonpast
“You see me.”
7
Valenzuela complements the discussion of Wariapano with a discussion of Shipibo-Conibo, where
all noun phrases are marked in the ergative-absolutive, and of Yaminawa, where only first and second
person pronouns get nominative-accusative marking, while third person pronouns and all other
noun phrases get ergative-absolutive. The Yaminawa data presented in the article are nevertheless
inconclusive.

45
 c. nyuntu-∅ ka-npa purla-mi
you-abs pres-2sg.nom shout-nonpast
“You are shouting; you shout.”

As can be seen here, while noun phrases are marked as ergative or absolutive,
pronominal clitics on the auxiliary are nominative or accusative. According to Corbett
(2006), if there is a discrepancy between case marking and agreement, then agreement
is nominative-accusative, while case marking is ergative-absolutive, but never the
other way around,8 thus in some sense respecting the “empathy hierarchy” that applies
to pronouns and other noun phrases.
There are no person splits in Mẽbengokre; person splits won’t be considered any
further in this dissertation.

2.2.2 Aspect splits

A second type of split ergative system is often described as being conditioned by

tense, aspect and mood. Examples of languages with such splits include Georgian,
Burushaski, and many of the Indo-Aryan languages. The following Georgian examples
are from Nash (1995):

(34) Georgian: Ergative in the aorist

8
Recent data from Kutchi-Gujarati in Patel (2007), however, exemplify a situation (the past
perfective tense) in which agreement is ergative-absolutive, while case marking on noun phrases is
nominative-accusative:

a. tu aav-i
you.sg come-f.sg
“You (fem.) came.”

b. tu chokra-ne mar-ya
you.sg boys-acc hit-pfv.m.pl
“You hit the boys.”

Note that the same split pattern holds in (31), if in fact the number alternation is to be treated
as agreement. In addition, participial agreement in languages such as French and Italian is with
the absolutive argument, while all noun phrases in these languages follow nominative-accusative
alignment.

46
 a. Nino-m surat-i da=xat.-a
Nino-erg picture-abs prev=draw-aor.sg
“Nino has drawn a picture.”

b. Bavšv-eb-i ga=braz-d-nen
child-pl-abs prev=angry-inch-aor.pl
“The children have gotten angry.”

(35) Nominative in the imperfective

a. Nino surat-s xat.-av-s

Nino.nom picture-obj draw-th-sg
“Nino draws a picture.”

b. Bavšv-i t.ir-i-s
child-nom cry-th-3sg
“The child cries.”

The cross-linguistic generalization capturing this type of ergative split seems to

be that “tenses” often designated by the terms perfect, perfective or aorist tend to
occur with ergative-absolutive alignment, whereas the remainder, which includes at
least imperfectives, occur with nominative-accusative alignment. As far as we can
ascertain, then, so-called tense-aspect-mood splits essentially boil down to aspectually
conditioned splits; in no case known to us are splits based unequivocally on tense (i.e.,
situating the proposition with respect to utterance time) or mood. Furthermore, it is
also not clear (M. Polinsky, p.c., 3/2007) that the aspectual value usually described
as a perfective is not in fact better described as a perfect, with perfective simple
past as one of its many readings. There are no cases known to us of languages with
distinct forms for the perfect and for the perfective, and an aspectually-conditioned
split in which it is the perfective, and not the perfect, which is associated with ergative
alignment. We will return to the semantics of these distinctions in chapter 4.
The Mẽbengokre facts introduced in the opening paragraphs of the chapter consti-
tute an aspectually conditioned ergativity split, found also in several other languages
of the Jê family, and described variously in the sources enumerated in §1.1. The
precise semantics of this split in Mẽbengokre will be the subject of chapter 4. The
contrast between ergative and accusative main clauses has been described in the

47
northern Jê language Timbira as one of tense in Shell (1952), Popjes and Popjes
(1986), and other sources. The following Timbira examples are from Shell, op. cit.
(apud Urban 1985):

(36) Ergative in the “past tense”

a. wa i-te a-py-w
already 1-erg 2-grab-past
“I grabbed you.”

b. wa i-wy-k
already 1-descend-past
“I descended.”

(37) Accusative in the “present tense”

a. wa a-py
1nom 2-grab
“I grab you.”

b. wa wy
1nom descend
“I descend.”

The pattern in Mẽbengokre is superficially identical. In the analysis that we

develop in chapter 4, the split in Mẽbengokre main clauses, is characterized as a
contrast between a perfect with ergative alignment and an unmarked (perfective or
imperfective) aspect with accusative alignment. Our translations differ accordingly
from the Timbira:

(38) Ergative pattern in the perfect:

a. ijE a-bWr
1erg 2-grab.n
‘I have grabbed you.’

b. i-rw7k
1-go.down.n
‘I have gone down.’

(39) Accusative pattern in perfectives:

48
 a. ba a-bW
1nom 2-grab.v
‘I grabbed you.’

b. ba rw7
1nom go.down.v
‘I go down.’

If we wish to make an analogy with the hierarchy that subsumes possible person
splits, the following would be a reasonable way to summarize what we’ve said in this
section:
progressive and habitual ≫ perfective ≫ perfect

As with the nominal hierarchy, a language sets a cutting point between ergatively
and accusatively aligned constructions somewhere along this hierarchy; all aspectual
values to the left of the cutting point show ergative alignment, while aspectual values
to the right of the cutting point show accusative alignment. Using a hierarchy here
is somewhat tricky, since we are referring to constructions, rather than directly to
aspectual meaning. Elucidation of these issues will be postponed until chapter 4.

2.2.3 Split S systems

A third common type of split ergativity is often called split intransitivity, split S
system, or “active-stative” system. It arises when some intransitive subjects (of “ac-
tive” predicates) align with transitive subjects, while the rest (subjects of “stative”
predicates) align with direct objects. Schematically, if we distinguish between the
two types of intransitive subjects, we could represent this system as in figure 2-2.

SA SO

A O

split S

Figure 2-2: Split S system, considered by some authors (Klimov 1974) as an alignment
on par with those of figure 2-1.

49
 To illustrate split intransitivity, consider the Guaraní data, from Velázquez-Castillo
(1996) in figure 2-3.

Stative Active
che-yta ‘I can swim’ a-yta ‘I swim’
che-monda ‘I’m a thief’ a-monda ‘I steal’
che-karu ‘I’m a big eater’ a-karu ‘I eat’
che-ka’u ‘I’m a drunk’ a-ka’u ‘I get drunk’
che-kakuaa ‘I’m big’ a-kakuaa ‘I grow’
che-guata ‘I’m a fast walker’ a-guata ‘I walk’
che-kirirı̃ ‘I’m a quiet person’ a-kirirı̃ ‘I stop talking’
che-tyarõ ‘I’m mature’ a-tyarõ ‘I mature’
che-vevui ‘I’m light’ a-vevui ‘I float’
che-poi ‘I lose grip’ a-poi ‘I drop’
Figure 2-3: Guarani verbs that can be both “active” and “stative”

The predicates given constitute a small set of predicates in Guaraní that can be
both “active” and “stative”,9 stativity or activity being revealed by the choice of the
pronominal form for the subject; the majority of intransitive predicates in Guaraní
belong to one class or the other, but not both.
In Mẽbengokre, a similar split could be argued to exist on the basis of the con-
trast between Type III and Type IV predicates from table (5), repeated here for
convenience:

(40) Type I Type II Type III Type IV Type V

Nominal erg–abs erg–abs abs abs dat–abs
Verbal nom–acc nom–abs nom — —

While regular intransitive verbs (Type III predicates) mark their subjects as nom-
inative when in their verbal (or “finite”) form, a class of nominal predicates (Type IV,
as well as the nominal form of Type III predicates) which translate certain stative
notions (cf. §1.3) mark their subjects in the absolutive, i.e., making them identical to
the direct objects of Type II predicates.
So far, we have established that Mẽbengokre has two “splits” in its surface syntax:
an aspect split, and a predicate-type split (“split S”) in intransitive predicates. In

9
Possibly in the sense of Vendler (1967); cf. the discussion in Mithun (1991).

50
§§2.3 and 2.5 we will unify the aspectual split and the predicate-type split found in
Mẽbengokre under a single conditioning factor.

2.2.4 Ergativity in nominalizations

To conclude our discussion of ergativity splits commonly found cross-linguistically,

we now address the opposition between nominalizations (or noun phrases in general),
which tend to have an “ergative” alignment,10 and nominative-accusative main clauses.
The discussion of ergativity in nominalizations is usually excluded from surveys of
ergative splits, as these are generally restricted to clausal constructions. This omis-
sion, however, has obscured the fact, to which we return in §2.3, that nominalizations
are at the heart of many ergative constructions in the clausal domain.
Many types of nominalizations exist. We concentrate on process nominals, (also
known as action nominalizations) which denote properties of eventualities, i.e., “name
the event” denoted by the verb (destruction, growing). They thus differ from par-
ticipant nominalizations (grower, payee), result nominalizations (payment, vomit),
and “fact-that” nominalizations (so-called acc-ing and poss-ing nominalizations in
English).11 Ergativity is strongly correlated with process nominalizations, even in lan-
guages that are otherwise not ergative (cf. the survey in Koptjevskaja-Tamm 1993,
and the discussion in Alexiadou 2001):

(41) a. the capture of Rome (by the Wisigoths)

la prise de Rome (par les Wisigoths)

b. the arrival of the Wisigoths

l’arrivée des Wisigoths

In these noun phrases, the O participant of the transitive construction, and the
S participant of the intransitive construction bear the same case, genitive. The A

10
The reason for the quotes is that structural cases within noun phrases are often different from
those found in the clausal domain. We speak of ergativity in nominalizations whenever a participant
corresponding to an intransitive subject of the non-nominalized predicate receives a case identical
to that assigned to participants corresponding to direct objects.
11
We will go into much greater detail on the structure of nominalizations in chapter 3.

51
participant of the transitive construction is introduced by an adjoined PP, like in
passives. This is an ergative alignment, which in languages like English obtains only
in the nominal domain. Such an alignment also obtains in nominalizations in Greek,
French, Spanish, Italian, and several other well-described languages.12
Note that English has a complicating factor that the other languages mentioned
above lack. English noun phrases can have up to two genitives: the post-nominal
of -genitive exemplified in (41), and the so-called Saxon genitive, which occurs pre-
nominally:

(42) a. The city’s destruction.

b. John’s destruction of the city.

c. John’s arrival.

If the pre-nominal position is excluded, as was done in the examples in (41), the
English pattern is the same as that of Greek and other languages without a prenominal
genitive:

(43) a. The destruction of the city.

b. The arrival of John.

c. * The destruction of John. (with agent reading)

12
This alignment accounts for over a quarter of the constructions in Koptjevskaja-Tamm’s (1993)
sample. Three other types are also common: possessive–accusative, double possessive, and sentential.
The first and last of these (as well as other minor types described by Koptjevskaja-Tamm) will be
excluded from our consideration here, as they seem to involve more structure than strictly nominal
constructions. Cf., for instance, the contrast between the following two types of nominalizations in
English, the second being an example of the possessive–accusative pattern:

a. John’s destruction of the city.

b. John’s having destroyed the city.

The differences are many and have been pointed out many times before (note in particular that
(b) can’t be used to describe an event, but rather “the fact that”); an analysis of different types of
nominalizations as being effected at different levels of an essentially verbal extended projection is
advanced by Abney (1987).

52
 d. * The destruction of the city of John.

e. The destruction of the city by John.

It’s not clear to us what the correct analysis of the English pre-nominal genitive
should be. The analogy with (nominative-accusative) active sentences should nev-
ertheless be constrained by the fact that the pre-nominal genitive’s relation to the
nominalized predicate is much more tenuous than that of subjects to verbal predicates
(i.e., John’s destruction can refer not only to the destruction John caused, but also
to the one he discussed, or was negatively affected by, and so on), suggesting that
the “subject” is not the subject of the nominalization itself, but of a higher possessive
predicate. We will not delve into this issue here.
Mẽbengokre event nominalizations, which are employed in a variety of construc-
tions in the language, display an ergative-absolutive pattern:13

(44) a. b2 kam i-mõr kuni

forest in 1-go.pl.n all
“all my goings into the woods”

b. ijE 2ktirE krõr jã

1erg hawk.people make.peace.n this
“this (occasion in which) I was making peace with the Àktire”

In the case of Mẽbengokre, contrary to what happens in English, there is morpho-

logical identity between the cases employed in nominalizations and those employed
in ergative main clauses. This identity will play an important role in §2.3, as we
attempt to reduce all the ergativity found in Mẽbengokre to a single source, that is,
to nominalization.
13
The arguments for constructions such as those in (44) to be considered nominalizations rather
than complement clauses are presented in chapter 3. For now, note that they refer to events, occur
with determiners and cardinality expressions used with nouns, and head constituents that have
the distribution of regular noun phrases (i.e., they can appear in left-peripheral focus positions,
can be wh-expressions when merged with which, and so on), in addition to their translations as
nominalizations (which are much less natural in the metalanguage than, e.g., equivalent relative or
complement clauses) offered spontaneously by consultants.

53
 More generally, nominalizations have been postulated to be at the core of certain
clause-level ergative constructions in language families such as Inuit (discussed below,
in §2.4) and Mayan. An important fact to note is that, contrary to other ergativity
splits, which languages may or may not have, action nominalizations are normally
ergative.14 Given an adequate theory of ergativity in nominalizations, the attempts
to reduce main clause ergativity to nominal ergativity can constitute a significant
advance in our understanding of ergative splits. We will return to this at the end of
§2.3, and in §2.5.

2.3 Ergativity in Mẽbengokre

So far, we have mentioned that ergativity arises in Mẽbengokre nominalizations, in
main clauses with certain aspectual values, and obligatorily with some intransitive
predicates. In chapter 1 we related ergative-absolutive case marking to a particular
form of the verb, which we’ve called nominal, without further justification. In this
section we will be more precise about the contexts where ergativity arises. To be-
gin, consider again examples (25–27), which opened the chapter, repeated here for
convenience:

(45) a. ba tẽ
1nom go.sg.v
“I went.”

b. ba ku-kwW̃r
1nom 3acc-break.sg.v
“I broke it.”

(46) a. i-tẽm
1-go.sg.n

14
We don’t pretend at this point to have explained away all of the non-ergative nominal construc-
tions found in Koptjevskaja-Tamm’s survey, but simply hinted above that controlling for more than
is usually controlled in broad typological studies, the preponderance of ergativity in nominalizations
is even more overwhelming than what the survey suggests. Of course, the truth of this claim cannot
be deduced from anything else, but rather requires detailed cross-linguistic analysis.

54
 “I’ve gone.”

b. ijE kwW̃ñ
3erg 3.break.sg.n
“I’ve broken it.”

(47) a. ba i-tẽm O=mõ

1nom 1-go.sg.n instr=go.pl.v
“I’m going.”

b. ba kwW̃ñ O=mõ
1nom 3.break.sg.n instr=go.pl.v
“I’m breaking it.”

The constructions in (45), which were described above as displaying a nominative-

accusative alignment, arise in matrix clauses that have a perfective interpretation.
Those in (46), described as ergative-absolutive, arise in matrix clauses that have a
perfect interpretation.15 Finally, the “mixed alignment” displayed by (47) occurs in
clauses that receive a progressive interpretation.
The data could be summarized by the following table, which is based on the one
presented in (5) above:16

(48) Type I Type II Type III Type IV Type V

Perfect erg–abs erg–abs abs abs dat–abs
Progressive nom–abs nom–abs nom+abs — —
Perfective nom–acc nom–abs nom — —

Our ultimate goal in this chapter is to explain the split ergativity exhibited by
this data set. This involves not only reducing all manifestations of ergativity to a

15
The meaning of this will be clarified in chapter 4. In fact, as we will see in that chapter, the range
of interpretation of this type of main clause is much wider than just the perfect which translates this
example. This fact should be borne in mind, but it doesn’t affect the argumentation in this chapter.
16
As can be seen, the split in case alignment only applies to verbal predicates (Types I, II and
III), not to nominal ones (incidentally, this gap in the table will be explained once we reduce
ergativity in Mẽbengokre to nominalization). The label “perfect” applied to Type IV and V predicates
is misleading, of course, since in these predicates no contrast in aspectual value is instantiated
morphologically.

55
single component of the structure, something which we’ll do within the confines of
this chapter, but also imbuing the labels “perfect” and “perfective” with meaning,
something which will only be fully accomplished in chapter 4.

2.3.1 Ergative is the alignment of embedded clauses

Let’s temporarily step aside from considering the case alignment found in matrix
clauses, and consider embedded clauses.
The literature dealing with Jê languages has several times put forth the generaliza-
tion that nominal forms occur in embedded contexts, while verbal forms occur in main
clauses. Notably, Wiesemann (1972) considers the verbal form (which she calls “short
form”) and the nominal form (“long form”) of the verb to be positionally-determined
allomorphs of the verb: the verbal form would occur solely in matrix clauses, while
the nominal form occurs in all embedded clauses.17 Nothing is said by Wiesemann
about the correlation between the two forms of the predicate and the case marking of
arguments in the clause. From what we’ve said since the table in (5) in chapter 1, we
know we should expect an ergative alignment in embedded clauses in Mẽbengokre.
This is in fact what happens in the following two clear cases of subordination:

(49) a. ba [kutE tEp krẽn] pumũ

1nom 3erg fish eat.n see.v
“I saw him eating fish.”

b. ba [kutE tEp krẽn] m7r7ri ipej

1nom 3erg fish eat.n during do.v
“I did it while he was eating fish.”

An assortment of other constructions seem to involve clause subordination. Con-

sider the following:

(50) a. ar7m kutE tEp krẽn m2̃

already 3erg fish eat.n to
“He’s already about to eat fish.”

17
Wiesemann retracts from this position in a later paper (Wiesemann 1986), in which she recog-
nizes the use of nominal forms in main clauses, contrasting in aspectual value with verbal forms.

56
 b. kutE tEp krẽn mEj
3erg fish eat.n good
“He eats fish properly.”

c. kutE tEp krẽn ket

3erg fish eat.n neg
“I haven’t eaten the fish.”

A superficial inspection of these data might lead us to the descriptive generaliza-

tion that ergativity is associated with negation, as in (50c), with adverbial modifica-
tion, (50b), and with a particular aspectual value (“prospective aspect”), (50a), which
are in themselves unanalyzable into more basic constituent parts (cf. Urban 1985, p.
186). This is clearly the wrong path to take, however.
As indicated in the gloss to (50a), the marker following the verb is nothing other
than a postposition, which in fact subordinates the semantically main verb. Several
postpositions can appear in this position, though only three, or possibly four of them
can appear forming matrix clauses, as in the example discussed.18
That is, (50a) is structurally identical to the subordinated clause in (49b), and
we propose that it be analyzed as a subordinate clause as well: i.e., a nominal clause
is subordinated to a main predicate which is a postposition. Suggesting an analogy
with the construction that translates it in English (“is about to”) can serve to give
plausibility to this proposal, but at this point the nature of the construction might
need some more elucidation; i.e., is there a subject to the postposition, or is the
postposition some sort of raising predicate? We return to this in §2.5 below.
The case of (50b) falls into the same category. The construction that translates
predicates with a manner modifier is actually one where the clause is embedded as
an argument of a manner predicate that is syntactically the main predicate of the
clause.

18
Gildea (1992) discusses the relation (which he fancies purely diachronic) between postpositions
and case endings and aspectual or temporal values in the Carib language family. This discussion
is excluded from the published version of his work (Gildea 1998), and therefore unavailable to us
during the preparation of this dissertation.

57
 A similar expression of manner occurs in St’át’imcets, and is described in Arregui
and Matthewson (2001). The following are some relevant data extracted from this
paper:

(51) St’át’imcets

a. skenkín ti n-s-xát’-em-a ta sqwém-a

slow det 1sg.poss-nom-hard-intr-det det mountain-det
“I walked up the hill slowly (lit., my walking up the hill was slow)”

b. áma ti s-nik’-in-ás-a ti sts’úqwaz’-a s-Mary

good det nom-cut-tr-3erg-det det fish-det nom-Mary
“Mary cut the fish nicely (lit., Mary’s cutting of the fish was good)”

In these St’át’imcets examples, the manner modifier is a main predicate, while

what is semantically the main predicate appears as a nominalized complement of the
former, as is evidenced both by the nominalizing morphology and the determiners.
We are now left to deal with (50c). We will argue that, like manner predicates,
negation is the main predicate of the clause, and takes the negated clause as its
complement.
There are a couple of simple arguments to consider negation a main predicate that
takes nominal (or nominalized complements); the same arguments could be extended,
mutatis mutandis, to manner predicates:19

(52) Negation can be used with a noun phrase argument to negate existence:

a. tEp ket
fish neg
“There is no fish.”

b. ar7m wa ket
already 3.teeth neg
“He no longer has any teeth.”

(53) Negation takes person inflection, with a negative existential meaning:

19
To this effect, see the discussion around ex. (140) below.

58
 i-ket=ri
1-neg=at
“When I didn’t exist.”

We return to the semantics of all of these constructions in chapter 3, once we have

given a semantics for nominalizations.
What is then the connection between embedding, the category of the predicate,
and ergativity? We contend that all embedded clauses in Mẽbengokre are nominal
(i.e., are headed by nominal forms of verbs), something for which we’ll argue more
extensively in chapter 3. We saw that nominal clauses in Mẽbengokre are ergative
(we take this to be a primitive for now; the fact is related to more general principles
in §2.5). The intent of this section has been to reduce as many of the ergative con-
structions found in Mẽbengokre as possible to constructions involving subordination.
We have done this with negative clauses, clauses with manner modifiers, and clauses
with post-verbal postpositions used to convey particular aspectual meanings. We still
have to explain the facts of the progressive, and explain ergativity in main clauses.
In the following section, we deal with the progressive. Main clause ergativity is the
subject of chapter 4.

2.3.2 The progressive construction

We have reduced all but one of the cases where ergativity arises in Mẽbengokre
to instances of clause subordination. Clause subordination requires the use of the
nominal form of the verb, which, in turn, implies ergative alignment in the case
marking of its arguments. There seems to be one case, though, where the use of such
a form does not imply ergative case marking; this is the progressive construction,
introduced above (cf. 27). We turn to it now.
The construction consists of a clause subordinated to the postposition O (instru-
mental), followed by an “auxiliary” verb. The latter is generally chosen from intran-
sitive verbs denoting positions or motion; the most common are dZa “stand up”, ñW̃
“sit down”, nõ “lie down” and mõ “go (pl.)”. The nuances in the meaning of the con-
struction according to the choice of verb are interesting in their own right, though

59
irrelevant for our present purposes. The following set of examples illustrates this:

(54) a. ba tEp krẽn O=ñW̃

1nom fish eat.n instr=sit.sg.v
“I’m eating fish (sitting down).”

b. ba tEp krẽn O=mõ

1nom fish eat.n instr=go.pl.v
“I’m eating the fish (gradually).”

c. ba tEp krẽn O=dZa

1nom fish eat.n instr=stand.sg.v
“I’m eating fish (standing).”

In such constructions, it can be seen that the case of the transitive subject is
nominative rather than ergative. In intransitive clauses, a nominative duplicates the
absolutive subject:

(55) ba i-rw7k O=mõ

1nom 1-go.down.n instr=go.pl.v
“I’m going down.”

As Reis Silva (2006) has shown, the nominativity of the subject depends on the
“auxiliary” being verbal. When the auxiliary is in its nominal form, the normal
ergative pattern arises:

(56) a. ga tEp krẽn O dZa

2nom fish eat.n instr stand.sg.v
“You are eating fish.”

b. ajE tEp krẽn O a-dZãm

2erg fish eat.n instr 2-stand.sg.n
“You are eating fish.”

Our proposal for this construction is that it should be analyzed as its outer syntax
suggests it should: i.e., the “auxiliary” should be treated as a main verb, while the
semantically main verb should be considered to be subordinated to an instrumental

60
postposition, forming an adjunct to the clause headed by the auxiliary.20 Nominative
subjects are the subjects of the “auxiliary”, when it is in its verbal form. The ergative
subject of the subordinated predicate gets deleted because there is a coreferential
nominative subject above it, and so only arises if there is no nominative subject, i.e.,
when the “auxiliary” is in nominal form, as in (56b).
Thus the construction in question is in effect biclausal, and its lack of ergativity
despite the nominal character of the predicate is epiphenomenal. A peculiarity of the
construction is that while ergative subjects can be deleted, absolutive subjects don’t:

(57) a. ba i-tOr O=dZa

1nom 1-dance.n instr=stand.v

20
There are a couple of facts which are prima facie problematic for this approach, however. First
is that the “auxiliary” becomes destressed and cliticizes to the main verb, whereas we’d expect it to
keep word level stress if it was the main predicate. In addition, the clause under the instrumental
postposition O behaves differently from true instrumentals, as shown by Reis Silva (2006), in that it
can’t be clefted together with the O, as the latter can:

a. ba karatSu O ku-krẽ
1nom spoon instr 3acc-eat.v
“I ate with a spoon.”

b. ba tEp krẽn O ñW̃

1nom fish eat.n instr sit.v
“I am eating fish.”

c. karatSu O nẽ ba ku-krẽ

spoon instr nfut 1nom 3acc-eat.v
“I ate with a spoon.”

d. karatSu nẽ ba O ku-krẽ

spoon nfut 1nom instr 3acc-eat.v
“I ate with a spoon.”

e. * tEp krẽn O nẽ ba ñW̃

fish eat.n instr nfut 1nom sit.v

f. tEp krẽn nẽ ba O ñW̃

fish eat.n nfut 1nom instr sit.v

The significance of these facts is not completely clear. Further consideration of them will be left
for future research. Both Reis Silva (op. cit.) and de Oliveira (1998) consider O to be a light verb
rather than an instrumental postposition.

61
 “I’m dancing.”

b. * ba tOr O=dZa
1nom dance.n instr=stand.v

As we saw in chapter 1, however, the obligatoriness of absolutive arguments is an

independent fact about Mẽbengokre morphosyntax, and the above pattern could be
ascribed to this fact.

Laka (2006) proposes something similar to account for the nominative subjects
found in the progressive construction in some Basque dialects. The relevant data are
the following:

(58) a. emakume-a-k ogi-a jan du

woman-det-erg bread-det eat has
‘The woman has eaten (the) bread’

b. emakume-a-k ogi-a jaten du

woman-det-erg bread-det eating has
‘The woman eats (the) bread’

c. emakume-a ogi-a jaten ari du

woman-det bread-det eating prog is
‘The woman is eating (the) bread’

Rather than accepting “split ergativity” in the Basque progressive, Laka proposes
that (58c) has a biclausal structure. In Laka’s approach, ari, which was considered
to be an antipassive in some previous accounts, or a progressive auxiliary in others,
is taken to be a main verb (meaning ‘to be engaged’) that takes a nominalization
as a complement: jaten is analyzed as ja-te-n ‘eat-nom-loc’. The subject of the
progressive construction receives the case and thematic role from the verb ari. In
dialects where ari has been reanalyzed as a grammatical marker (Asp), ergative
marking appears in the construction. This is essentially what we propose here for the
Mẽbengokre progressive.

62
2.4 Summing up: nominalizations and ergativity
In the previous section, we have considered several of the ergative constructions found
in Mẽbengokre. We have seen that they all involve some form of subordination, hence
“nominalization”.
The connection between subordination and nominalization comes from an arbi-
trary fact about Mẽbengokre, which we will not attempt to explain: that there is no
embedding of finite verbal clauses in Mẽbengokre, only of nominal constructions.21
We saw that embedded nominalizations subdivide into those that are subordinated
to a one place predicate (as in negation, manner modification, and other constructions
considered in §2.3.1), and those where the subordinating predicate has a (nomina-
tive) subject, causing the ergative subject of the embedded clause to be deleted (the
progressive construction, considered in §2.3.2). In any case, we were able to maintain
that even in the latter case, the embedded nominalization was ergative. Furthermore,
we claimed at the end of §2.2.3 that the “split S” of Mẽbengokre, like the ergative splits
triggered by subordination, also boils down to the opposition between the category
of nouns and verbs.
What is, then, the connection between nominalization, or rather nounness, and
ergativity? Before giving our own account of this, we need to consider one previ-
ous approach in which nominalization is made to be at the heart of clausal ergativ-
ity: Johns’s (1992) analysis of Inuktitut. Johns proposes that sentence (59c), which
translates “The man stabbed the bear”, is derived through the two intermediate con-
structions that precede it.

(59) a. kapi-jaq
stab-pass.part
“The stabbed one.”

b. anguti-up japi-ja-a
man-erg stab-pass.part-3s

21
This is an altogether not uncommon pattern, however (cf. Polinsky 2007 on Adyghe). It should
also be pointed out that Mẽbengokre possesses other constructions, such as paratactic constructions,
that translate embedded propositions.

63
 “The man’s stabbed one”

c. anguti-up nanuq kapi-ja-a

man-erg bear-abs stab-pass.part-3s/3s
“The bear is the man’s stabbed one.”

That is, the passive participle in (59a), which is essentially a patient nominaliza-
tion, is at the core of a passive free relative, (59b). This free relative in turn will
become the object of a copular construction, yielding (59c). The structure is thus
more or less22 as in Figure 2-4.

CopulaP
H
 H
 HH
 HH
 HH
 H
NP Copula’
H
H
nanuq  HH
 HH
the bear  H
AgrN P Copula
H
H
 HH
∅
 HH
NP AgrN ’
HH
 H
anguti-up N AgrN
the man-rel
kapi-jaq -a
stabbed one

Figure 2-4: Structure of an Inuktitut transitive clause

There is an obvious circularity in this proposal, from the point of view of relating
nominalization and ergativity. Nominalization here is being used as an abbreviation
for patient nominalization; these nominalizations are “passive” by definition. Thus
Inuktitut ergativity essentially boils down to the ergativity that results from the
obligatory passivization of all transitive clauses. There is no real need to invoke
nominalization as an intermediate step.23

22
Under what we here call Copula, Johns puts agreement with “the bear”.
23
Of course, the possibility exists that nominalization is more basic than passivization; i.e., passives

64
 This is crucially not the way that we propose to relate nominalizations and erga-
tivity in Mẽbengokre. For one thing, the nominalizations involved in Mẽbengokre
are not passive participant nominalizations, but rather action nominalizations which
are presumably active. The structural differences between action nominals and the
parallel verbal clauses in Mẽbengokre are therefore much smaller than that between
the structure proposed by Johns for (59c) and a regular finite clause in any other
language. They boil down to the case-theoretic properties of the functional struc-
ture above the lexical projection. The latter is common to both verbal and nominal
constructions.
In the following section, we will attempt to build the essentially ergative character
of nominal projections into the syntax, and examine the conditions that have to be
met for a nominative-accusative alignment to arise.

2.5 A formal analysis of case

Case is a morphological category found in natural language that in the descriptive

and typological tradition is traditionally assumed to have the functional motivation
of keeping participants of a clause distinct. Though it is often the case that there is
a correlation between a participant’s case and its position in the thematic hierarchy,
case cannot be reduced to semantics, as can be seen in cases of raising to object, such
as the following:24

(60) a. We believe that he is qualified.

b. We believe him to be qualified.

cross-linguistically (at least those that are constructed analytically, like the one found in English)
would embed a nominalization, rather than the other way around, as is usually thought. This is a
possibility that, to our knowledge, hasn’t been discussed in the literature, and one that we are not
capable of pursuing here.
24
Taken from Merchant (2006), on whose overview of standard case theory we base the discussion
surrounding (60-61).

65
 Though in neither of these sentences the pronoun bears a thematic relation with
the predicate believe, and in both cases the thematic relation with the lower predi-
cate be qualified is the same, the pronoun receives nominative case in the first, and
accusative in the second. A similar situation arises with valency-changing operations
such as passive and antipassive, among others:

(61) a. Everyone trusted him.

b. He was trusted by everybody.

In these examples, the pronoun bearing the θ-role of theme receives the accusative
in the active sentence, and the nominative in the passive one. Thus the mapping from
thematic roles to case is both one to many and many to one.
A common view since Vergnaud (1977) is that case is a formal mechanism to license
noun phrases in the syntax, and that there is a one-to-one mapping between cases (or
case positions) and case assigners. Thus, accusative case is normally assumed to be
assigned by the verb, or by some head in its immediate projection, while nominative
case is assigned by tense. In the nominal domain, one could say, in a similar vein,
that a structural case such as the postnominal genitive is assigned by some nouns,25
while the English prenominal genitive is assigned by the determiner.
The relation between thematic licensing and case is most straightforward in situa-
tions described as “inherent case assignment”. In these cases the head and the maximal
projection involved in thematic licensing and in the determination of morphological
case are the same.26

25
For arguments that the postnominal genitive is structural, see Alexiadou (2001).
26
Work by Fraga (2006) suggests that this relation might not be as straightforward as is tra-
ditionally assumed. Fraga decomposes prepositions into lexical roots, presumably responsible for
relating their themes to an associated locus of points in space, and a category-assigning head that is
responsible for case assignment, and possibly for the dynamic versus static contrast which correlates
with the case governed by P in languages such as Greek and German. In addition, and discussed
explicitly by Fraga, several uses of prepositions are purely functional, such as last-resort for -insertion
in English, or marking of specific animate direct objects in Spanish with a.

66
 case

(62) John walked to school.

θ-role

In the case of lexical roots that are to become verbs or nouns the situation is
slightly more complicated. If we accept that pairs such as destroy and destruction
are composed of an identical lexical root which attaches to two distinct category
labels,27 then the thematic role of the theme and its case are assigned by different
heads. This is a good result, since we want the thematic relation between the lexical
head and its complement to be the same in principle regardless of category; case, on
the other hand, depends on whether a noun or a verb is formed.

2.5.1 Case in Mẽbengokre noun phrases

There are two types of underived nouns in Mẽbengokre: those that are inalienably
possessed (“relational nouns”), and those that are either alienably possessed or gen-
erally not possessed at all. The first class is exemplified in (63); the second in (64).

We will not consider prepositions further in this dissertation, so these complications arising in
their syntax will be put aside.
27
The proposal that lexemes always decompose into a category-less root and a category-assigning
functional head is discussed in Pesetsky (1995) and Marantz (1997), and has been adopted in much
work within Distributed Morphology. We will not present any empirical arguments in favor of that
position, as opposed to the traditional view in which words come with their categories from the
lexicon. It is probably not difficult to translate the syntactic part of our proposal into the tradi-
tional approach to nominalization, which involves a nominalizing head over a verbal projection; see
Alexiadou (2001), who interprets Grimshaw’s (1990) distinction between action and result nominals
as a distinction between embedding or not embedding a verbal projection.
Nevertheless, when we move on to describe the function of category-assigning heads in the seman-
tics, we believe that the traditional approach would become unwieldy. In later parts of the thesis
we will imbue the category-forming heads n and v with content. In particular, cf. the discussion in
§4.1.2.

67
Inalienably possessed nouns require their “possessor” argument to be expressed.28 The
case of this argument is the genitive, like that of themes in English nominalizations.
A few nouns, i.e. (65), can be in both classes, sometimes with a slight (or not so
slight) semantic difference.

(63) a. i-kra ‘my son’

b. i-pa ‘my arm’
c. i-ñũrkwã ‘my home’

(64) a. i-ñõ kikrE ‘my house’

b. i-ñõ krit ‘my pet’
c. (mẽ i-ñõ) b2 ‘(our) forest’

(65) a. i-dZudZe ‘my weapon’ a’. i-ñõ dZudZe ‘my bow’

b. i-bikwa ‘my relative’ b’. i-ñõ bikwa ‘my relative or friend’

We will assume that the difference between the alienably and inalienably possessed
nouns is one of argument structure. While inalienably possessed nouns project their
possessors as sisters of the lexical root, the possessor of alienably possessed nouns is
expressed by means of a postpositional phrase adjoined to the nominal projection.29
√
In the approach to case sketched above, the argument of doesn’t get case from
the head selecting it (unlike what happens with arguments of prepositions) and is

28
Inalienable possessors are usually in part-whole or possessor-possessed relation with the lexical
root, though the latter only in the sense in which ‘my brother’ can be considered possessive.
29
This explains the fact that the thematic interpretation of alienable possessors is fixed, only
expressing literal possession; other types of relations (benefactive, locative, etc.) are expressed by
means of different postpositions.
There is a limited form of alternation in argument structure in the nominal domain, in what one
might call the “nominal applicative alternation” of Mẽbengokre, which allows certain alienable nouns
to become relational, i.e., take an argument to their left, which would otherwise be expressed by
means of a locative postpositional phrase, and assign genitive to it:

a. No ‘water’ a’. pidZo ka-No ‘fruit water (i.e. juice)’

b. ko ‘carved wood’ b’. ak2 ka-ko ‘lower lip wood (i.e., lip-disk)’

68
forced to get case elsewhere: it gets structural genitive case from n, the category
assigning head. Thus, if we represent the two relations of thematic selection and case
assignment by arrows, we have the representation in Figure 2-5.

nP
√HH
P n
H
 H√
DP

Figure 2-5: Thematic selection and case in a simple nominal projection.

We may assume that structural genitive case is simply not assigned if there is no
√
argument projected as a sister of ; it also can’t be assigned to constituents adjoined
higher in the structure, such as the alienable possessors.
It seems that in this case, even though morphological case and thematic licensing
occur in separate heads, the relationship between the two is straightforward. The real
challenge comes from the behavior of verbs.

2.5.2 Case in nominative-accusative clauses

Getting the case facts right when there is one structural case to be assigned is straight-
forward enough. Finite verbal clauses (in Mẽbengokre and more generally in languages
with nominative-accusative alignment) pose a particular challenge. In them there are
two structural cases to be assigned, nominative and accusative. The usual assumption
in the government and binding tradition is that the latter is assigned by the verb,
while the former is assigned by tense, as in the structure in Figure 2-6.30
Yet nominative-accusative case marking requires a special type of dependency
between the two noun phrases or case-assigning heads: assignment of accusative case
depends on what goes on higher in the structure; it is not assigned, even to noun
√
phrases generally assumed to be generated as sisters of the verb (or ), if a higher

30
It is often assumed that nominative case is assigned upwards to a subject that has moved to
[Spec,TP], rather than downwards to [Spec,vP], as we have it here. The question is orthogonal to
our discussion.

69
 TP
H
H
 H
T vP
H
 H
DP v′
H√
H
v P
H
 √
H
DP

Figure 2-6: Assignment of nominative and accusative in a transitive finite clause.

argument (chain) isn’t present. A common approach to capture this generalization

descriptively is synthesized by the following generalization:

(66) Burzio’s generalization31

Accusative case is only assigned if (and only if) an external argument is
present.

Burzio’s generalization has several exceptions,32 in addition to suffering from the

purely deductive anomaly of bringing together argument structure (i.e., the presence
of an external argument) and case, after we’ve argued that they should be kept dis-
tinct. The following examples show some of the exceptions to Burzio’s generalization:

(67) a. Greek

Mou aresei i thalassa.

1sg.dat please.3sg the.f.sg.nom sea.f.sg.nom
“I like the sea.”

b. It struck me that all these counterexamples are irrelevant.

These examples fall into two classes. In one case, if the external argument has
lexically determined case, as in (67a), no accusative is assigned; conversely, if an
expletive receives nominative, as in (67b), accusative is available even without the
verb having a thematic external argument. That is, accusative seems to depend

31
Cf. Burzio (1986), Reuland (2000).
32
Cf. discussion in Marantz (1991).

70
on whether structural case is assigned to an external argument, not on whether an
external argument is projected.
This is suggestive of a second approach to the case dependency: that case is “oppo-
sitional”, i.e., that it serves to keep different noun phrase participants in the sentence
morphologically distinct. The intuition is that two noun phrases in a particular do-
main of case assignment will get their cases not locally, as the single argument of a
preposition or noun, but rather “relationally”.
In this dissertation, we will maintain the traditional view of case as being assigned
by heads, but will adopt an element from one recent formalization of the relational
approach to case, namely the idea of dependent case found in Marantz (1991).
To give a concrete example of how dependent case functions (again, without stray-
ing too far from received ideas about case), consider what happens in a regular finite
transitive clause.33

TP
H
H
 HH
DPj T’
HH
 H
T VP
H
 H HH
Vi T tj V’
H
 H
ti DP

Figure 2-7: Configuration in which dependent case is assigned to one DP chain.

In the configuration depicted in Figure 2-7, both subject and object DPs are in
the government domain of V+T (in addition, the base — or possibly all — of both
chains is in the government domain of the trace of the verb, but this is not relevant
here). Dependent case, i.e. accusative, is assigned by the following postulate:

(68) Dependent case is assigned by V+T to a position governed by V+T when a

distinct position governed by V+T is:
√
33
For the sake of simplicity, and v are merged into a single head V in this structure.

71
 a. not part of a chain governed by a lexical case assigner

b. distinct from the chain being assigned dependent case

Dependent case assigned up to the subject: ergative

Dependent case assigned down to the object: accusative

That is, accusative, which is the dependent case in this situation, is assigned
to the object noun phrase in case there is another DP chain (i.e., the subject’s) in
the government domain of V+T. The remaining DP in the government domain will
receive the unmarked case, i.e., nominative. The clause governing the assignment of
dependent case in effect replaces Burzio’s generalization with a homologue which is
based solely on case, rather than on thematic licensing of the external argument.
Why is dependent case assigned down to the object, rather than up to the subject,
yielding an ergative alignment in finite verbal clauses? We attempt to answer this in
the following section.

2.5.3 Case in embedded clauses

Let us now turn to clauses of the type considered in §2.3.1. We concluded that these
ergative constructions are headed by a nominal form of the verb. Though we will go
into much greater detail about their structure in chapter 3, let us assume a lexical
projection similar to that of verbs, but with a different category label, as in Figure
2-8.

nP
HH
 H
DPsubj n′
H
 H√
H
n P
HH
√
DPobj

Figure 2-8: Structure of a nominal projection.

In Marantz’s (1991) approach, (morphologically) ergative and accusative lan-

guages are mirror images of each other. Though not made explicit by Marantz,

72
it’s plausible to suppose the distinction is coded as a construction-specific parameter
governing the direction of assignment of dependent case by V+T; that is, whether
a particular construction has ergative or accusative alignment rests on a parameter
that is orthogonal to the rest of its structure. We suggest a slightly different ap-
proach: nominative-accusative case marking arises when two case-assigning heads
(v and T) are found within the domain where two arguments are to receive case;
ergative-absolutive arises when both structural cases are assigned by a single head (n
or v not linked to T). We will have the opportunity to return to this after we explore,
in chapter 4, what the link between v and T, which is lacking in nominal predicates,
consists of.

To implement our approach to case in the nominal domain so as to yield the case
marking pattern of Mẽbengokre nominalizations, all we have to do is make explicit
that dependent case in the situation in which there is a single case governor (in this
case n) is what we’ve been calling ergative, while unmarked case in the same domain
is the genitive of §2.5.1, which is what we’ve been calling absolutive until now. Case
marking in all embedded clauses, including those where the embedding predicate is
negation or a manner predicate, and in intransitive predicates of Type IV, is subsumed
under this mechanism.

To summarize this, consider the following two sentences:

(69) a. ba ku-krẽ
1nom 3acc-eat.sg.v
“I ate.”

b. ijE krẽn ket

1erg eat.sg.n neg
“I didn’t eat.”

Though we haven’t fully argued for the structure that corresponds to these sen-
tences, the following are first approximations that take heed of what we’ve said so
far:

73
 TP TP
H
H  HH
 HH  HH
DP T’ T ketP
H
(70) a.  HH b. HH
 H
ba T vP nP ket
P
 PP P
 PP
ku-krẽ ijE krẽn

In the first situation, whether v moves to T or not, we contend that v has a

privileged link with T that creates a domain with two case governors, which is the
context for accusative alignment to arise. In the second case, the negation ket is the
highest predicate in the clause, and so is the one that has the potential of linking
with T, to the detriment of any lower predicates, which are left to fend for themselves
as far as case assignment goes. Since ket has no arguments other than the nominal
clause below it, we don’t see any reflexes of dependent case assignment.

At this point one could raise the following question: why don’t any of the func-
tional categories that intervene between v and T (in Mẽbengokre, if there are any, or
in any other language) function like ket, and break the link between them? Intuitively,
of course, the reason is that ket, differently from, e.g., negation in English, is a main
predicate, whose argument, the nominal clause, is closed off to any operation that
would extend the government domain for case assignment. Any attempt at formal-
ization of this intuition at this point risks circularity; cf., for instance, Hunt (1993),
who deals with contexts for ergativity in Gitxsan; to her, what determines whether
a clause will be ergative or accusative is whether it’s embedded under a lexical or a
functional head, respectively. In the case at hand, Mẽbengokre negation would be
a lexical predicate, whereas negation in English would be functional. In chapter 4
we skirt this problem altogether by building the link with higher temporal functional
categories into the lexical entry for v; n lacks such a link. That is, the ability to link
with higher functional structure is a property of the category label merged with the
main predicate.

74
 The reader might have observed a parallel between what we are proposing and the-
ories of ergativity such as those found in Nash (1995). Putting it somewhat vaguely,
for Nash, and for several others since then, ergativity arises in constructions that lack
certain higher functional projections. What we maintain here, howevel, is not that
higher functional structure is lacking, but rather that the link between the lexical
projection and functional structure cannot be established directly. That is, while in
nominative-accusative constructions v and T are directly linked, they (or rather n
and T) are prevented from linking in the constructions that surface as ergative. The
nature of the link between v and T is only made fully explicit in §4.1.2.

2.6 The source of main clause ergativity

So far, we’ve addressed case marking in both verbal main clauses, and embedded
nominal clauses. If they denote entities, it is not clear how constructions headed by
nominal forms of verbs (or any other noun phrase) can in and of themselves become
main clauses; so the ergativity found in main clauses with the perfect aspect is still
mysterious. We will address this problem in chapter 4; before we get to that, we need
to understand nominalizations both in their structure and in their meaning. That is
the subject of chapter 3.

75
76
Chapter 3

The structure and interpretation of

nominalizations

In §2.2.4, we stated that Mẽbengokre nominalizations display an ergative alignment.

The relevant examples are repeated here for convenience:

(71) a. b2 kam i-mõr kuni

forest in 1-go.pl.n all
“all my goings into the woods”

b. ijE 2ktirE krõr jã

1erg hawk.people make.peace.n this
“this (occasion in which) I was making peace with the Àktire”

Subsequent parts of chapter 2 reduced other ergative constructions found in the

language to nominalizations. In particular, it was claimed that all embedded clauses
are nominal, and that several apparently matrix constructions that display ergativ-
ity, such as sentences with manner modifiers and negation, should be analyzed as
“biclausal” constructions embedding a nominalization.
The aim of this chapter is to examine embedded nominalizations, and to come
up with a precise enough characterization of their structure and meaning so that
the constructions in which they appear may be understood. The constructions to be
considered are broadly two: those that denote actions, such as (71) above, which one

77
could call “action nominalizations”, and those that denote participants, such as (72),
which could descriptively be labeled “internally-headed relative clauses”.

(72) a. kubẽ kutE 2ktirE krõr nẽ jã

barbarian 3erg hawk.people make.peace.n nfut this
“This is the white man that made peace with the Àktire.”

b. i-m2̃ ajE amũ=jã=P2̃ kubẽk2 jadZ2r=jã N2̃

1-dat 2erg farther=this=at clothes put.on.sg.n=det give.v
“Give me the clothes that you put on yesterday.”

In this chapter, we will claim that both of these types of constructions share a
similar structure, and differ minimally in their semantics. We begin by providing a
description of internally-headed relative clauses, and later show that action nominal-
izations represent a simple extension of the expressive possibilities of relative clauses.

3.1 Internally-headed relative clauses

Word order in Mẽbengokre relative clauses is identical to that found in matrix clauses.
In particular, the noun that is modified by the relative clause, i.e., the head, appears
inside it, in the position it would occupy as an argument of a regular matrix clause.

(73) kubẽ kutE mẽ i-m2̃ mẽkrı̃dZ2 ñõr=ja

barbarian 3erg pl 1-dat chair give.n=the
‘The chair(s) that a/the white man gave us’, or
‘The white man/men that gave us a/some/the chair(s).’

No special marking appears on the head. This results in the ambiguity observed
in the preceding example, which was pointed out also for Lakhota by Williamson
(1987), among other languages,1 and seems to be an essential characteristic of the
construction:

(74) a. Wi˛yą wą owi˛ža wą kaǧe ki

woman a quilt a make the

1
See the surveys in Culy (1992) and Basilico (1996).

78
 ‘The quilt that a woman made’, or
‘The woman that made a quilt.’

There is no restriction as to the grammatical function within the relative clause of

the noun phrase that serves as head: noun phrases in adjunct roles are freely allowed,
as attested in (75). Null third-person pronominals can also be interpreted as heads,
in a construction that could be considered the equivalent of a free relative (cf. 76):

(75) kubẽ kot i-tẽm nẽ ijE aNro bı̃-n nẽ jã
barbarian with 1-go.sg.n and 1erg peccary kill.sg.n nfut this
‘This is the white man with whom I went and killed peccaries.’

(76) mẽ tũm kutE arẽñ nẽ jã

pl old 3erg 3.say.n nfut this
‘This is what the ancients told.’

Nevertheless, a relative clause cannot be headed by an adjunct that is not overtly

present. We take this to mean that the heads of relative clauses in Mẽbengokre can
only be noun phrases, i.e., there are no relative clauses headed by how, when, etc.:

(77) a. akati kam i-tẽm ja

day in 1-go.sg.n the
‘the day I go’

b. * i-tẽm ja
(impossible with a when translation)

In this they seem to differ from a very similar construction discussed by Larson
(1982), Warlpiri adjoined relative clauses. The latter, in addition to the ambiguity as
to which participant is the head of the relative clause, display an ambiguity between
a participant reading and a temporal one:

(78) ngajulu-rlu-rna wawiri nyangu, kuja-npa pantu-rnu nyuntulu-rlu

I-erg-1sg kangaroo see-past, comp/aux-2sg spear-past you-erg
a. ‘I saw the kangaroo which you speared’

b. ‘I saw the kangaroo when you speared it’

79
 It should be kept in mind that we are explicitly ruling out this freedom of inter-
pretation in Mẽbengokre relatives, as we later include eventive readings among their
expressive possibilities. We will argue that eventive readings are permitted precisely
because, like nominal participant readings, they arise from binding of a variable that
is projected within the nP, i.e., the variable that saturates the lexical root’s referential
argument when it is event-denoting.

3.2 The structure of IHRCs

We will take internally-headed relative clauses such as those given in the previous
section as the prototypical example of a nominalization in Mẽbengokre. In this section
we will propose a structure for them, and in §3.3 we propose a way to derive their
semantics compositionally. The relevance of looking at internally-headed relative
clauses is that other meanings of nominalizations, and in particular the eventive
readings which we will need in chapter 4, will follow as small extensions of the simple
syntax and semantics for IHRCs developed here.
We’ll be minimalistic, and assign the structure in Figure 3-1 to internally-headed
relative clauses such as that in (73).2
The key elements of this structure are the following: (a) the relative clause itself
consists solely of a lexical root, all of its adjuncts and arguments, and an external
argument introduced by the category-assigning head;3 (b) this structure is selected
√
directly by a determiner, and (c) arguments of (and of n) can be either DPs or
determinerless NPs. The point of the latter two assumptions will be made clear below.
Before we move on to that, we need to discard a few alternative structures.

A common analysis of internally-headed relative clauses, advanced, e.g., by Cole

(1987) in his analysis of Quechua, holds that they have a structure more or less as in

2
The dative adjunct of (73) is omitted here.
3
Certain aspectual projections, such as that containing ar7m, might have to be added to this
structure, but we claim that neither tense nor a complementizer are part of IHRCs.

80
 DP
H
 HH
 HH
 H
nP D
H
H
 HH


HH ja

 H the
NP n′
P
PP H
 P √ HH
kubẽ kutE P n
white man H
H
 √
H
NP

mẽkrı̃dZ2 ñõr
chair give

Figure 3-1: Structure of Mẽbengokre internally-headed relative clauses

the following tree, i.e., they are headed by a null external head, to which the visible
part of the relative clause is adjoined.

(79)
NP
HH
 H
TP/IP NP
P
 P
∅
...

We contend that internally-headed relative clauses in Mẽbengokre never have an

external head. In Mẽbengokre this is so for a very simple deductive reason: there is
no adjunction inside noun phrases (or elsewhere, in fact) other than of postpositional
phrases. The following are forbidden:

(80) a. * tũm kikrE

old house

b. * bo kikrE
hay house

c. bo=O kikrE
hay=instr house

81
 “House made of hay.”

The only apparent exception to this is a type of dvandva compound discussed by

de Oliveira (2005), where two nouns are juxtaposed. This is actually only found in a
handful of names of imaginary creatures, and should be considered marginal:

(81) a. kubẽ=rOp
barbarian=jaguar
“Jaguar people.”

b. kubẽ=ñep
barbarian=bat
“Bat people.”

“Adjectives” inside a noun phrase are always the syntactic head of the construction;
i.e., they seem to constitute a special case of internally-headed relative clause, rather
than standing as a class of their own. This question will be touched on again in §4.4.
We conclude that a structure such as that proposed by Cole for Quechua is quite
unlikely for relative clauses in Mẽbengokre, since such adjunction structures don’t
occur with anything other than postpositional phrases in the language.

As for how much structure Mẽbengokre IHRCs contain, the following contrast
between matrix clauses and IHRCs can be adduced to show that certain left-peripheral
positions are absent:

(82) kukrWt nẽ ba ar7m ku-bı̃

tapir (foc) nfut 1nom already 3acc-kill.sg.v
“I killed tapir.”

(83) (*kukrWt) (*nẽ) (*ijE) ar7m ijE bı̃n

tapir (foc) nfut 1erg already 1erg 3.kill.sg.n
“What I already killed.”

The left periphery of matrix clauses such as (82) is constituted by a focus position,
that can contain at most one dislocated XP, a delimiting particle that indicates future
versus nonfuture tense, and a position reserved for nominative subjects, which is

82
higher than that of any oblique subject. This latter position, in particular, was
discussed in §1.4.
None of these left peripheral positions are available in the relative clause in (83).
The ergative subject can appear only after the particle ar7m, which appears just after
the nominative in (82). This puts whatever projection ar7m is in as the upper bound
of structure in IHRCs, effectively excluding TP and CP.

3.3 The basic semantic analysis

Let us take as a starting point the structure that we established in Figure 3-1, and
derive the translations we observed in (73) by working our way through one example.
Consider the following:

(84) kubẽ kutE rOpkrOri bı̃n=ja

barbarian 3erg jaguar kill.n=the
a. ‘The white man who killed the/a jaguar.’

b. ‘The jaguar that the/a white man killed.’

The following denotations are straightforward:4

(85) a. Jbı̃K = λxe .λev .kill′ (e, x)

b. JrOpkrOriK = λxe .jaguar′ (x)

c. JkubẽK = λxe .barbarian′ (x)

How do these parts come together to give the correct denotation to the subcon-
stituent rOpkrOri bı̃? Clearly not by Functional Application.
Given that indefinite noun phrases in Mẽbengokre have no overt determiners, we
will consider them to be determinerless NPs, of type he, ti. They come together with
the main predicate by the compositional rule of Predicate Restriction, introduced by
Chung and Ladusaw (2004). The rule can be summarized as:

4
The semantic types used are: individuals (e), eventualities (v), which can be considered just a
special type of the former, and truth-values (t).

83
 Predicate Restriction (Chung and Ladusaw 2004, p. 5)
λy.λx.P (y, x) ∧ Q(x)
H
H
 HH

λx.λy.P (y, x) λx.Q(x)

If we apply this composition rule to our example, we get:

(86) JrOpkrOri bı̃nK = λe.λx.kill′ (e, x) ∧ jaguar′ (x).

We won’t go into how the external argument is introduced,5 and we will assume
that kutE is vacuous. So the denotation we get for the core of the relative clause (84)
is:

(87) Jkubẽ kutE rOpkrOri bı̃nK =

λy.λe.λx.kill′ (e, x) ∧ jaguar′ (x) ∧ barbarian′ (y) ∧ Agent(e, y)

As can be seen in the representations above, we assume, with Heim (1982) and
much subsequent work, that indefinite noun phrases lack quantificational force of their
own. In (87), this results in that whenever the syntactic arguments are indefinite (i.e.,
determinerless) noun phrases, a verbal projection is as unsaturated semantically as
just a verb by itself. The denotation of such nPs is an n-place property of individuals.
This nP combines with D to form internally-headed relative clauses.6
What is the determiner ja, then? We will claim that it is an unselective binder.
This means that it binds a variable contained in its sister constituent, but which
variable is bound (if the constituent contains more than one) is not determined by
structure. Any one variable is bound by ja, while all other variables that are free at
this point in the structure are bound by existential closure:

(88) JjaK = λP et.ιxP (x)

(89) kubẽ kutE rOpkrOri bı̃n ja

barbarian 3erg jaguar kill.n the

5
We return to this question in §4.3.
6
For a somewhat similar approach to relative clauses in Salish, see Jelinek (1995).

84
 a. ‘the jaguar that a white man killed’
ιx∃e∃y: kill′ (e, x) ∧ jaguar′ (x) ∧ barbarian′ (y) ∧ Agent(e, y)

b. ‘the white man that killed a jaguar’

ιy∃e∃x: kill′ (e, x) ∧ jaguar′ (x) ∧ barbarian′ (y) ∧ Agent(e, y)

This is pretty straightforward. Note that in addition to these readings, there

are readings of this sentence where the non-head noun phrase is definite, as definite
determiners are optionally null.

To repeat what we said at the beginning of §3.2, we considered internally-headed

relative clauses in this and the preceding section because we believe them to be the
prototypical example of a nominalization in Mẽbengokre. We will assume that the
structure of all nominalizations we deal with in the dissertation is what we have
proposed for IHRCs. In the following section, we show that the semantics of eventive
nominalizations is a simple extension of the semantics which we developed for IHRCs
in this section.

3.4 Eventive complement clauses

Complement clauses that are formally identical to internally-headed relative clauses
can get eventive interpretations, as opposed to participant, or fully clausal, interpre-
tations. This can be seen in direct perception constructions:

(90) ba 2k k2r ma
1nom fowl coo.n hear
‘I heard the bird calling.’

Arguably this is also the interpretation they get when they are complements of
manner predicates:

(91) a-dZu-jarẽñ mEj

2-antipass-say.n good
‘You spoke well.’ (lit.: ‘Your saying was good.’)

85
 The semantics that we have developed for relative clauses extends without signif-
icant modification to get the senses in (90–91):

(92) ba bẽnjadZw7r kutE bẽn dZir ma

1nom chief 3erg speech place.n hear
a. ‘[I heard] the reciting of the/a ritual speech by the/a chief’
ιe∃x∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

b. ‘[I heard] the ritual speech that the/a chief recited’

ιx∃e∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

Remember from the discussion surrounding example (77) that for a particular
participant to be the head of the relative clause, it has to be present in the structure,
i.e., whatever variable gets bound by the determiner outside the internally-headed
relative clause to become the semantic head, has to be projected in the structure.
This argues for a variable e ∈ D v to be present, presumably as an argument of the
verbal root.

This is all we need to proceed to our analysis of matrix nominal clauses in chapter
4. The next section is simply an addendum showing how the semantics we’ve given
to nominal constructions7 merges with various things to yield negative sentences,
sentences with a manner modifier, and “short nominalizations.”

3.5 The semantics of negation, manner modification,

and short nominalizations
We observed above that in Mẽbengokre sentences that have manner modifiers, such as
(91), the semantically main clause appears syntactically embedded under the manner
modifier. Arregui and Matthewson (2001) discuss manner modification in Salish,
which seems to function in a similar way:
7
Throughout this dissertation we’ve been avoiding the term nominalization, as we believe these
nominal structures to be basic, rather than derived from verbs, but of course the structures under
discussion are nominalizations under most different approaches.

86
(93) Mẽbengokre

mẽ tOr mEj

pl dance.n good
“They danced well (lit., their dancing was good)”

(94) St’át’imcets

skenkín ti n-s-xát’-em-a ta sqwém-a

slow det 1sg.poss-nom-hard-intr-det det mountain-det
“I walked up the hill slowly (lit., my walking up the hill was slow)”

As Arregui and Matthewson point out, this structure is amenable to an analysis

where the nominalized clause denotes a definite description of an event that saturates
the argument position of the manner predicate:

(95) Jmẽ tOr mEjK = JmejK(ιe.Jmẽ tOrK(e))

Note that the meaning of the embedded nominalization is crucially not that of
a proposition, but rather is a description of an event, so (93) cannot mean “It was
good that they danced.” A formalization of this distinction, which is recognized since
Vendler (1967), is advanced by Zucchi (1993).
Something similar could be applied to negated sentences (cf. Davis 2005 for a
treatment along those lines of one type of negation in Salish.)
Why does manner modification work like this in Mẽbengokre? I suspect that it’s
because the language drastically restricts adjunction: there are no open classes of
adjectives or adverbs, and, as we saw above, relative clauses aren’t adjoined either.
This, coupled with the fact that finite clauses can’t be embedded, is the reason why
nominalizations are so pervasive in the language. The settings of these two parameters
are what make Mẽbengokre look superficially different from other languages, like
French and English, that also have ergativity in nominalizations.

3.5.1 Short nominalizations

Mẽbengokre has two morphemes, dZ2 and dZw7j, that are used to create a large
repertoire of what could be intuitively called “lexical” nominalizations, such as the

87
following:

(96) a. piPok-jarẽ-ñ-dZw7j
writing-say-nlzr-master
‘teacher’

b. i-dZ2-ku-r-dZ2
1-ap-eat.n-container
‘My eating utensils’, but also: ‘my eating place’, ‘my food’, etc.

In the literature on other Jê languages (cf., e.g., de Oliveira 2005), these have
been considered to be an instrument and an agent nominalizer, respectively. Our
contention is that what the “nominalizers” attach to is already nominal (i.e., it’s an
eventive complement clause, as described above), and they themselves are no more
than the semantically bleached nouns dZ2 ‘container’ and dZw7j ‘master’.
What is the relation between these nouns and the nominalized clause? The nouns
cannot be external heads: as we saw above, what is interpreted as the head of an
internally-headed relative clause has to be a null pronoun or a determinerless noun
phrase in a governed position. This is not the case in either (96a) or (96b). In
addition, dZ2 and dZw7j are compatible with both an overt internal head,8 and an
external head, which in any case appears to the left of the relative clause.
Instead, we propose that the structure of these “short nominalizations” is just
what the morphology leads us to believe: they are full nominalized clauses that are
(genitive) complements to the bleached nouns.9 How exactly “the master of saying
writing” comes to mean “teacher” will have to be worked out on another occasion,
but the path to follow should be clear.

8
For instance, dZw7j can co-occur with an overt agentive subject, though the sense of these
constructions is not clear to us at this point.
9
Alternatively, we could take these bleached nouns to sit in n, and have a “classificatory role”,
i.e., restricting the interpretations of the lexical projection to those that are compatible with the
classifier’s feature set.

88
Chapter 4

The interpretation of matrix nominals

In chapter 2, we explored the relation between nominalizations and ergative syntax.

In chapter 3, we provided a semantics for nominalizations in embedded contexts, and
derived the meaning of constructions employed for manner modification, negation,
and progressive aspect, all of which embed a nominalization.
In the present chapter, we address the original puzzle presented in the introduc-
tion. The puzzle can be restated as follows: The opposition between nominalized
and non-nominalized verbs, which so far only signaled a difference in context, has
nevertheless a clear effect on aspectual interpretation in main clauses; this can be
seen in the following paradigm:1

(97) a. krw7j jã nẽ mop krẽ

parakeet dem nfut malanga eat.v.sg
“This parakeet ate the malanga.”

b. krw7j jã nẽ mop ku

parakeet dem nfut malanga eat.v.pl
“This parakeet ate the malangas.”

(98) a. krw7j jã nẽ kutE mop krẽn

parakeet dem nfut 3erg malanga eat.n.sg

1
As in previous chapters, v indicates the verbal form of the verb, n the nominal form of the
verb; though in principle identifiable with pieces of the verb’s morphology, we opt for being agnostic
about segmentation. sg and pl indicate verbal number; when not marked, the particular verb
doesn’t display an opposition between singular and plural.

89
 “This parakeet has eaten malanga (at least once in his life).”

b. krw7j jã nẽ kutE mop kur

parakeet dem nfut 3erg malanga eat.n.pl
“This parakeet eats malanga (often).”

The translations given should be considered first approximations, and, as we will

see later in the chapter, reflect only some of the possible meanings of each construc-
tion. For now, we can summarize the effect of nominalization and verbal number on
interpretation in non-future main clauses as follows:

(99)
Verbal Nominal

Singular V Perfective; singular object Experiential perfect, relative

to the subject’s life-span
Plural V Perfective; plural object Habitual or generic

The meaning of (97) can be characterized as positioning the event with respect to
a topic time that is set by narrative context.
In (98), on the other hand, the event is, like in (97), contained in an interval,
though one that is not anaphoric but rather coterminous with the subject’s life-span
(mutatis mutandis for inanimate subjects). The interpretation of these sentences
containing nominal forms of verbs has been variously described as “stative” or “subject-
oriented.”
Our project in this chapter is to provide a semantics for the different verb forms of
Mẽbengokre that is compositional, in the sense that it respects facts about Mẽbengokre
clause structure that we have established in previous chapters. In particular, we wish
to derive the stativity of the constructions in (98) from the fact that the verb forms
are nominal, and are therefore forced to be interpreted in a particular way.

A priori, there are at least two ways to explain the aspectual opposition between
nominal and verbal forms of verbs: on the one hand, one could say that Mẽbengokre
nominals inherently denote states, not unlike the participles of better-known lan-

90
guages; on the other, one could say that something about the construction is respon-
sible for the stativity, but not in itself the fact that the forms employed are nominal.
Though we aim to give both possibilities a fair hearing, we will contend that
the latter is the correct approach, based primarily on the fact that nominal forms
of verbs are not necessarily stative in embedded contexts, but rather are so only in
main clauses. Furthermore, we will contend that the device that renders nominal
forms of verbs stative in matrix clauses is the same that allows underived nouns to
be interpreted as existential clauses. Our arguments are expanded in §4.3. Before
turning to that, however, we develop a semantics for regular (non-nominal) verbs, as
this will remain the same regardless of how we choose to analyze nominal ones.

4.1 The interpretation of verbs

The simplest position we can take as to the meaning of a sentence such as (100), which
is consistent with previous discussion, is one where the logical form is as follows:

(100) bẽnjadZw7r bẽn dZi

chief speech put.v.sg
‘The chief recited (put down) a ritual speech.’
∃e∃x: recite′ (e, the-chief′ , x) ∧ speech′ (x)

Of course, to this we need to add viewpoint aspect and tense. We will assume
that viewpoint aspect, i.e., perfective versus imperfective, and tense, are projections
above a structure such as (100). That is, viewpoint aspect and tense are operators
that take propositions as arguments, and restrict their semantics in particular ways.
The basis of their interaction is explained in the following few sections.

4.1.1 The event argument

In chapter 3, we argued for the existence of an event argument in predicates because

of the need to refer to definite descriptions of eventualities in embedded nominaliza-
tions. Summarizing our discussion from that chapter, we propose, following Davidson

91
(1967), that certain types of predicates have, in addition to their arguments of type
e, an argument of type v (event).2
It is not important here to give a precise definition of what an element of D v is;
for our present purposes, it can be thought of as a special type of individual with
specific properties (in particular, a duration). Informally, we can say that it refers to
a situation that is described by a verb, corresponding to the intuitive meaning of a
noun phrase with a gerund or nominalization as its head.
In Davidson’s proposal, the event argument is simply a device to hold together
the various components of action sentences, allowing certain entailments to be easily
derived. The event argument is shared by the main predicate and all secondary
predicates (temporal and other adjuncts) in a sentence, as represented in the following
example:

(101) a. John played the violin yesterday at Symphony Hall.

b. ∃e: violin-playing′ (e) ∧ Ag(John′ , e) ∧ yesterday′ (e) ∧ at-Symphony-Hall′ (e)

In chapter 3, we made a case for the usefulness of a definite description of events

such as (102) to identify the semantics of action nominalizations with that of other
nominal expressions, and give a semantics to manner modification constructions in
Mẽbengokre.

(102) a. John’s playing of the violin yesterday at Symphony Hall

b. ιe.violin-playing′ (e) ∧ Ag(John′ , e) ∧ yesterday′ (e) ∧ at-Symphony-Hall′ (e)

The event argument is also the key for assigning an interpretation for time-
delimited propositions. Intuitively, a simple past sentence such as “John saw Mary”
could be given the following interpretation:3

2
Note that in the traditional nomenclature, which we employ here, variables in the domain
of individuals are generally represented by x, y, etc., whereas variables in the domain of events
are represented by e, e′ , etc. This e shouldn’t be confused with the e that represents the type of
individuals.
3
In this dissertation, we will use t to refer to the utterance time.

92
(103) a. John saw Mary.

b. ∃t < t : John see Mary at t

Given what we said so far, the translation could be recast as follows:

(104) ∃e: τ (e) < t : Ag(John′ , e) ∧ seeing-Mary′ (e)

The function τ is a surjection from the domain of eventualities to the domain of

time intervals, mapping each eventuality to the time interval it occupies. τ (e) = t is
introduced as a presupposition, rather than as an assertion.
How can we make this more compositional?

4.1.2 What are verbs?

As we saw in chapter 3, “nominal forms of verbs” in embedded contexts can head

noun phrases that denote definite descriptions of events. We proposed the following
semantics for an embedded noun phrase containing a nominal form of a verb:

(105) ba bẽnjadZw7r kutE bẽn dZir ma

1nom chief 3erg speech put.n.sg hear
‘[I heard] the reciting of the/a ritual speech by the/a chief’
ιe∃x∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

It seems reasonable to assume that the meaning of the embedded nominal form
of the verb, up to but not including D, is shared by all forms of verbs. The difference
between the verbal and the nominal form of the verb is introduced above this level:
while nominal forms may be selected by a determiner, which can bind either an event
variable or an individual, true verbs obligatorily become propositions, that is, their
event variable (as well as any other free variables) is bound by an existential operator.
This existential operator might just be “thrown in” (i.e., be a syncategorematic rule
of existential closure) when a vP merges with a higher sentential functional category,
or it might be introduced as part of the denotation of some lexical item. Given the
discussion in chapter 3, we propose that the existential semantics is part of the lexical
entry of v:

93
(106) JvK = λP vt .∃eP (e)

This functional category has a morphological reflex on Mẽbengokre verbs, namely

the truncation of the final consonant. That is, we propose that the nominal forms of
verbs have to be considered morphologically basic, while the properly verbal forms are
derived by merging the root with v. v is also associated with nominative-accusative
marking on the arguments, as opposed to the ergative-absolutive employed in clauses
headed by non-finite verbs.
The denotation of n is, for our present purposes, vacuous, and it has no morpho-
logical expression. Its only effect is in the ergative case marking, which, as we said
in §2.5, is a sort of default. Thus, it might seem that the only reason why we keep
a nominal head n in our syntax is for symmetry. The reader should nevertheless be
reminded of the short note on the “classificatory” role of n which was made at the
end of chapter 3. It is possible that the noun-forming head might have other con-
tent related to the fact that its extended projection is often a referential expression,
much in the way that the content of v is related to the fact that it normally heads
propositions.
We can restate what we have said in this section as follows: there is no lexical
distinction between nouns and verbs in Mẽbengokre, except in the argument structure.
If the referential argument of a lexical root (i.e., the event argument, or the entity
argument to which the root “refers”) is of the right type, i.e., is an event,4 it can become
both a noun if nothing happens, or a verb if its event argument gets bound by the
verbalizing head v. As a verb, its destiny is to become the head of a proposition; as a
noun, it can become a referential expression, or appear in an existential construction.
Lexical roots that don’t have an event argument, i.e., underived nouns, don’t have
the option of becoming verbs.

4
Of course, there is a certain circularity in putting things that way, as there might be certain
entity-type variables (cf. 132d) that are a priori difficult to distinguish from eventuality-type ones.
This problem is discussed briefly in chapter 3. Here, we will simply assume that the distinction can
be established on the basis of dynamicity features on the referential argument of the lexical root,
something which is largely true.

94
 Let’s move on to describe some of the properties of verbs.

How does the vP get linked to tense, to yield time-delimited propositions such
as (104)? We need to revise the lexical entry for v slightly. Let us assume that
denotations are always relative to time intervals, but that only v (so far) is assignment-
dependent. We thus have the following revised definition of v:

(107) JvKt = λP vt .∃e: τ (e) = t.P (e)

At this point we are ready to give closure to the discussion of case in §2.5. It
should be recalled from that section that we considered the necessary condition for
nominative-accusative case marking a configuration in which two case-assigning heads
were within the same domain of case assignment, as opposed to the situation that gives
rise to ergative-absolutive, in which only one case-assigning head is responsible for
both cases. We wish to propose that the forced link with a higher functional category
introduced in the denotation of v is what enlarges the domain of case assignment to
eventually engulf T and yield the desired configuration. No such link is introduced
by n, whose denotation is not assignment-dependent.

Many alternatives to the way we build structure within our proposal are imag-
inable. We will not explore any alternatives here, but rather only highlight two im-
portant elements of our particular implementation: (a) eventualities are kept distinct
from time intervals, and (b) the event variable is bound by v, rather than higher in the
structure. These assumptions, while probably not crucial, are meant to capture two
important facts about Mẽbengokre clauses, which have been recurrent in this thesis;
they are, respectively: that nominal clauses can refer to the eventualities themselves
(e.g., when embedded under perception verbs), and that verbal projections are always
propositional. Other characteristics of the particular formalization chosen are merely
technical choices, and can be easily recast in other frameworks.
What we have said so far amounts to saying that “verbness” (which in previous
versions of our work we referred to as “finiteness”) is what links the event description
to a higher functional category that binds the evaluation time.

95
 The tree in Figure 4-1 sums up our proposal so far. In this tree, events and time
intervals are kept distinct, and are related by v in the way that was described above.
Tense is a relation between the utterance time t and the evaluation time for v.

TP
H
 HH
 H
T vP
HH
 H√
∃t: tRt0 v P
P
 PP
∃e λe. . . .

Figure 4-1: Minimal structure of a verbal clause

We are now in a position to return to (104), still simplifying slightly.

(108) a. JJohn seeing MaryK ∈ D vt

b. JvKt (JJohn seeing MaryK) = ∃e: τ (e) = t.JJohn seeing MaryK(e)

c. JPast φK = ∃t: t < t : JφKt

(109) JJohn saw MaryK = ∃t: t < t : ∃e: τ (e) = t.JJohn see MaryK(e)

This decomposes as indicated in the tree given in Figure 4-2.

TP
H
H
 HH
 HH
 H
Past vP
H
 HH
∃t: t < t0  HH
 H√
v P
PPP
 PP
∃e: τ (e) = t  PP
λe.John-seeing-Mary(e)

Figure 4-2: Structure of (109).

96
4.1.3 Excursus on viewpoint aspect and tense

It has long been observed that in many cases the tense specification doesn’t relate
the time of an event directly to utterance time, but rather intermediates between the
utterance time and another, intermediate time interval. This can be observed in the
following set of sentences, in which sentence (110a) sets a temporal context, while
(110b) relates the event times to such a context, rather than to utterance time.5

(110) a. What was going on when you got home?

b. John had met Mary and was talking to Peter about it.

(111) a. ∃t∗ : t∗ < t : ∃t: t < t∗ : John meet Mary at t

b. ∃t∗ : t∗ < t : ∃t: t ⊇ t∗ : John talk to Peter at t

(111) are first approximations to a translation of the propositions in (110b). The

intermediate time interval t∗ , called Reference Time in Reichenbachian parlance, and
Topic Time by Klein (1994), the name that we adopt herein, like the event time (t in
the formulas above), seems to be set by the narrative context.
Viewpoint aspect presupposes a type of connection between event time and this
topic time. We assume, with Klein (1994) and others, that viewpoint aspect has
essentially two values, perfective and imperfective, for which we adopt the following
definitions, from von Fintel and Iatridou (2005):6

′
(112) a. JPfv φKt ⇔ ∃t′ : t′ ⊆ t: JφKt
′
b. JImp φKt ⇔ ∃t′ : t′ ⊇ t: JφKt

These formalizations capture the intuition that the main contrast between the
perfective and the imperfective is that the former claims that the event is fully con-

5
We will continue with the practice of treating the relations between time intervals effected by
Tense and Asp to be presuppositions. Whether this is as correct for Asp as for Tense is an issue
into which we won’t enter.
6
The pluperfect is useful here for expository purposes in this section, but will play no role in our
discussion of Mẽbengokre, so we don’t define it here.

97
tained within the topic time, whereas the latter makes no claim to this effect; cf., for
instance, the following sequences in Spanish:

(113) ¿Viste a Juan?

Did you see Juan?

(114) a. Estaba jugando en el jardín recién.

was.3s.Imp playing in the yard recently

. . . maybe he’s still there.

b. Estuvo jugando en el jardín recién.

was.3s.Pfv playing in the yard recently

# . . . maybe he’s still there.

. . . he can’t be too far.

Given these facts, we have to redefine the function of tense as making a particular
connection between utterance time and this same topic time, rather than the event
time directly. The denotation for the different tenses doesn’t have to be any different
from what we previously conceived, but the time variable that will be bound by Tense
will be Asp’s, rather than v’s. We can assume the following definitions for now:
′
(115) a. JPast φKt ⇔ ∃t′ < t: JφKt

b. JPres φKt ⇔ JφKt

′
c. JFut φKt ⇔ ∃t′ > t: JφKt

Tense is therefore higher in the tree than viewpoint aspect, yielding the partial
tree in Figure 4-3 for the first half of (110b).
Neither tense nor viewpoint aspect are directly encoded in Mẽbengokre verbs. For
the purposes of this dissertation, we will assume that tense and viewpoint aspect are
not what is involved in producing the aspectual contrasts that introduced the chapter.
Though we have stated this since the introduction,7 we are now in a position where
we can be more precise about our claims.
7
I.e., in rejecting the possibility that the forms might differ solely in an aspectual feature, and
adopting the solution that they differ in category.

98
 TP
H
 HH
 HH
 HH
T AspP
H
 HH
∃t : t < t0
∗ ∗
 HH
 HH
Asp vP
HH
∃t: t < t∗  HH
 HH
 √
v P
P
 PP
∃e: τ (e) = t  PP
PP

λe.John-meet-Mary(e)

Figure 4-3: Minimal structure which accomodates viewpoint

Smith (1997) argues that aspect should be divided into two domains, which she
calls “upper” and “lower” aspect. The latter is identified with Aktionsart, i.e., the
inherent aspectual structure of lexical predicates, whereas the former is straightfor-
wardly what is usually called “viewpoint aspect” (i.e., most prominently, the distinc-
tion between perfective and imperfective). Our claim is that, on the one hand, though
not crucially, viewpoint aspect should be limited to perfective and imperfective. On
the other hand, we claim that aspectual values such as those that nominal forms
of verbs take, that is, all of the aspect that is discussed in this chapter, should be
characterized as “lower aspect”, rather than viewpoint. It should be borne in mind
that this is a non-standard extension of the notion of “lower aspect” that, if applied
to better-known languages, would encompass participial formation, considering it to
be the “lexical” creation of a stative predicate out of an eventive one, thus making it
completely independent from viewpoint, which only determines how this state is to
be related to topic time.8
Tense being by semantic necessity a higher projection than viewpoint aspect, it
follows that it is also independent from the “lexical aspectual operations” discussed in

8
The discussion of “compound tenses” in English in §4.3 is relevant in this regard. In such tenses,
we would contend that viewpoint is a property of the inflected auxiliary, and is completely separate
from the stativity introduced by the participial form of the verb.

99
this dissertation. Working out how left-peripheral particles instantiate the different
categories of viewpoint and tense, and how default values of both viewpoint and
tense are assigned to different types of predicates, are matters which are left for
future research.

4.2 Verbal number

As we said in chapter 1, and again in the introduction to this chapter, Mẽbengokre
verbs display an alternation for number, most often manifested by root suppletion,
or by a change in the “classifier” prefix of verbal stems.9 The purpose of this section
is to establish, though with caveats, the semantics of verbal number. The semantics
of number will be relevant from §4.3 on, and in particular in §4.5 as we attempt to
reduce the contrast between (98a) and (98b) to the visible pieces of verbal accidence,
i.e., number and lexical category.
The following sentences exemplify the number contrast found in Mẽbengokre:

(116) a. Nra nẽ ba saku kam ku-dZ2

paca nfut 1nom bag in 3acc-put.v.sg
“I put a paca (Agouti paca) in the bag.”

b. Nra kumEj nẽ ba saku kam ku-Nje

paca many nfut 1nom bag in 3acc-put.v.pl
“I put many pacas in the bag.”

(117) a. ku-beNet nẽ k2jm2̃ dZa

sg-old.man nfut up stand.v.pl
“The old man stood up.”

b. mẽ-beNet nẽ mẽ k2jm2̃ kuPe

pl-old.man nfut pl up stand.pl
“The old men stood up.”

There has been some discussion about what verbal number, which is quite widespread
in Amazonian languages, is about, with some authors claiming that it’s number agree-
ment with the absolutive argument and others claiming that it’s exclusively a marker
9
For more details on the morphology of number, see chapter 1.

100
of plural action.10 That in Mẽbengokre the plural mark on a verb can stand primarily
for plurality of the action can be seen if we contrast (116b) with the following:

(118) aN2 kumEj nẽ ba saku kam ku-dZ2

bead many nfut 1nom bag in 3acc-put.v.sg
“I put many glass beads in a bag (all at once).”

We will assume for the purpose of this dissertation that the number marked on
Mẽbengokre verbs always refers to the cardinality of the event.11 According to consul-
tants, the plural is used when referring to large (and undefined) quantities, whereas
the singular can be used for a plurality, as long as it consists of relatively few indi-
viduals (“up to ten”, according to one consultant).
The question is thornier than this discussion might suggest. Cf., v.g., the following:

(119) a. ar7m nẽ ba i-ñõ puru kam mop kuni kaba

already nfut 1nom 1-poss garden in malanga all uproot.v.sg
“I already uprooted all of the malanga from my garden.”

b. ar7m nẽ ba i-ñõ puru kam mop kuni krw7

already nfut 1nom 1-poss garden in malanga all uproot.v.pl
“I already uprooted a lot of the malanga from my garden (but there might
still be some left).”

Examples like (119) suggest more a contrast in definiteness of the object than
of cardinality, with “singular” making the object definite and exhaustive (whatever
the real meaning of kuni is), and “plural” standing for a large quantity that isn’t
necessarily exhaustive.12 (120) is even thornier. Here the plural seems to have an

10
For Jê, cf. Urban (1985), who holds the former position, and D’Angelis (2004), who holds the
latter. Queixalós (1998), describing an unrelated language of the Orinoco basin, defines a category
of its own, “distensivité”, which is fuzzily related to aspect, agentivity, effectiveness of the action,
and so on.
11
In fact, it suffices for our purposes to say that the number marker on verbs may refer to plurality
of events, in addition to being agreement with some core argument. For discussion of verbal number
(“pluractionality”) as event plurality, cf. Lasersohn (2005) and Cusic (1981).
12
Cardinality might be an inherently tricky notion in Mẽbengokre, given that there are no native
expressions to refer precisely to quantities over two.

101
evidential value, yielding an interpretation where the event either happened in the
past, or in a location far away from both speaker and hearer.

(120) k2mrãñt7j nẽ pidZo b2ri P2̃ adZ2. . .

car nfut fruit tree on run.into
“The car ran into the fruit tree. . . ”

a. . . . nẽ o kuni ñikota

nfut 3.fruit all throw.down.sg
“. . . and made all the fruit fall.”

b. . . . nẽ o kuni kaPu

nfut 3.fruit all throw.down.pl
“. . . and made all the fruit fall (not present).”

Though in what follows we control for these effects by avoiding cardinality ex-
pressions13 on internal arguments, and maintain the claim that verbal plurality is
event plurality, examples such as these should be borne in mind when we return to
the puzzle that opened this chapter. In §4.5, we ascribe the modal component of
generics to a phonetically unrealized morpheme (which in §4.6.3 we identify with the
-n- that distinguishes present from past participles in Romance languages). It might
be the case, nevertheless, that we are completely wrong about this, and the modality
of habituals really resides in the “plural” morphology on the verb root. Such would
be a matter for further research.

4.3 The interpretation of matrix nominal forms

As we said in the introduction to the chapter, nominal forms of verbs have “stative”
or “subject-oriented” interpretations in matrix clauses. What is responsible for such
interpretations?
The alternatives are essentially two: (a) ascribing the stativity of the construction
involving non-finite (or nominalized) verbs to the nominalizing morphology itself, or

13
In fact, we try to avoid things like kuni, whose meanings aren’t clear, but since bare noun
phrases are always ambiguous between definite and indefinite meanings, this might not be enough
to exclude unwanted elements.

102
(b) ascribing the stativity of the construction to something in its syntax, rather than
in the morphology of the nominalized predicate. To explain what we mean, we will
discuss what each approach would involve if applied to the “compound tenses” in
better-known languages. Consider the following:

(121) a. John is eating.

b. John has eaten.

The question of how to divide up the meaning of the construction between the
participial form of verbs and the auxiliaries in cases such as (121) is parallel to the
question we are asking about the source of stativity in main clauses with nominal
forms of verbs in Mẽbengokre. While it might seem clear that in (121a) the copula
is semantically vacuous, and the present participle is an adjectival form, and is thus
stative in the required way (i.e., has the sub-interval property), (121b) is not as clear
cut.14 On the one hand, one could adopt the position that the participial denotes
a state (“the state of having had the experience of eating”), while the auxiliary is
vacuous; on the other hand, one could believe that the participial is eventive (i.e., has
as a denotation whatever the verb’s denotation is, without any stativization), and the
stativity comes from the auxiliary, which in this account would be a type of raising
predicate with a meaning approximately equivalent to “to have the experience of. . . ”

The first of these approaches corresponds to option (a) above; the second, to option
(b). In Mẽbengokre, however, the question has to be framed in slightly different terms,
as there is no overt auxiliary to ascribe any meaning to. For this reason, we talk of
“stativity as a property of the construction.” In fact, we will eventually identify this
component of the construction’s meaning with a particular covert element.

14
Cf. the appendix of Iatridou et al. (2001), where the issue of how the meaning of the perfect
is distributed between the participial and the auxiliary is discussed, without reaching a definitive
conclusion.

103
4.3.1 Stativity as a property of nominal forms

It is common for languages to have resources to create derived stative predicates from
verbs; the following are examples from English:

(122) a. The gates are closed.

b. This is a flowering tree.

In both of the preceding examples, the emphasized words are derived from verbs,
and seem to be inherently stative: in (122a), the predicate refers to a state result-
ing from an event that would satisfy the verb’s description, whereas in (122b), the
predicate refers to an ability or habit.
Would it be possible to say that Mẽbengokre non-finite verbs are precisely like
English closed and flowering? Certainly this seems to have been in the air in previous
work on Mẽbengokre, where the non-finite forms are called “stative” or “adjectival”.
The idea to implement would be that non-finite forms of verbs contain stativizing
morphology that take the eventive meaning of the verb root and yield a word that
denotes a set of stative eventualities that is related in some way to the original mean-
ing.
Basing ourselves in what we know about participial forms in the better-described
languages, we could list the following types of temporally stative eventualities derived
from (or simply related to) an eventive predicate:

(123) a. The property (i-level state) of having experienced a particular event (the
“existential perfect”).

b. The ability or habit to perform eventualities that are described by the

eventive predicate (the “generic” or “habitual”).

c. The target-state of an event (the “adjectival passive”, or “perfect of present

relevance”, depending on its valence).

d. The state leading to the completion of an eventuality described by the

eventive predicate (the “progressive”).15

15
The progressive is stative only in the sense of having the “sub-interval property”; for a discussion

104
 To exemplify the approach, we give a formal implementation of “target-state”
statives (i.e., 123c), which is advanced by Kratzer (2000).16 This is how one would
derive “closed” from the verb “to close”:

(124) a. JcloseK = λxλsλe.closing′ (e) ∧ event′ (e) ∧ closed′ (x)(s) ∧ cause′ (s)(e)

b. J-dK = λRλs.∃eR(s)(e)

c. JclosedK = λsλx.∃e closing′ (e) ∧ event′ (e) ∧ closed′ (x)(s) ∧ cause′ (s)(e)

That is, a verb that denotes a change of state, such as “to close”, would have to
include in its lexical meaning both the change and the end state. The job of the
stativizing morphology is to pick that state as the denotation of the derived stative
predicate.17 In this approach, contrary to what we’ve contemplated so far, the verb
would be basic (both morphologically and semantically), and stativizing morphology
would apply to it.
To account for the Mẽbengokre data, we need to propose two types of stativizer,
one to derive the existential perfect (98a) from (97a),18 repeated below with a partial
LF translation, and another to derive the habitual (98b) from (97b). For purposes of
illustration, we offer an implementation of the first one:

(125) a. krw7j jã nẽ mop krẽ

parakeet dem nfut malanga eat.v.sg
“This parakeet ate the malanga.”
λe.∃x: eating′ (e, x) ∧ malanga′ (x) ∧ Ag(e, this-parakeet′ )

of stativity, see Jackson (2005).

16
An application of such an approach to Pima is found in Jackson (2005). Pima derived statives
seem to have much in common with Mẽbengokre non-finite forms, and Jackson’s analysis could be
considered a counterpoint to the analysis that we develop here.
17
The semantic type of states (s) can be thought of as a special type of eventuality with the
characteristic of being temporally stative (i.e., having the sub-interval property) and being non-
dynamic. Though for events we insist that they be kept distinct from time intervals, it’s harder to
argue to keep such a distinction in the case of states.
18
Though in principle we might be able to find a “general purpose” stativizer that yields the whole
range of interpretation of the non-finite (or nominalized) forms.

105
 b. krw7j jã nẽ kutE mop krẽn
parakeet dem nfut 3erg malanga eat.n.sg

“This parakeet has eaten malanga (once in his life).”

λs.∃e: malanga-eating′ (e)∧Ag(e, this-parakeet′ )∧Result(e, s)∧Attr(s, this-parakeet′ )

(126) λxλP λs.∃eP (e) ∧ Ag(e, x) ∧ Result(e, s) ∧ Attr(s, x)

The Mẽbengokre stativizer could be given an approximate LF translation as in

(126).19 Of course, this semantics only pushes the problem “further back”, as it makes
use of two new relations. The first of these is one that identifies any event with its
“result state”, in the sense of Parsons (1990), i.e., the state that obtains after any
event is completed; this is not to be confused with the target state that appeared
in (124c), which is a state that is in principle reversible. The second relation is the
“attribution” relation, which could be thought of as the functional homologue for
states of the “agent-of” relation. We won’t pursue a definition of these relations any
further here, though they will appear in a different, and perhaps more transparently
motivated, form in the solution we adopt later.

An issue that didn’t arise when we defined v is that the external argument has
to be an argument of the stativizer in this case, as it is explicitly referred to in the
expansion of its denotation. In fact, it would not be unreasonable to identify the sta-
tivizing head with n in Mẽbengokre, given the fact that non-finite forms of verbs are
morphosyntactically identical to nouns. If so, a revision of our lexical entry for v as in
(127) is consistent both with the definition of n and with the now standard assump-
tion that category-assigning heads are responsible for the introduction of external
arguments.20

(127) JvKt = λx.λP vt .∃e: τ (e) = t.Ag(e, x) ∧ P (e)

19
/-n/ is only one of the phonological reflexes of the putative stativizer.
20
See Marantz (1984), Kratzer (1996), Hale and Keyser (1993), Chomsky (1993), among others.

106
4.3.2 Stativity as a property of the construction

The analysis sketched in the preceding section is initially plausible (and in fact we will
use certain elements of it for our definitive proposal), but it suffers from one important
shortcoming: nominalized forms aren’t stative when in embedded contexts; stativity
is only evidenced when they are used in matrix clauses.
It would be possible to fix the analysis by identifying the stativizer with a category-
assigning head other than n, i.e., a (adjective), and proposing that embedded nomi-
nalizations are really composed with n, which would have non-stative semantics, while
matrix ones are composed with a. It would, however, be desirable to identify both
constructions, and ascribe the stativity of main clause nominalizations to something
about the construction involved in matrix nominalizations. As we will see here, there
are good empirical grounds to create statives “in the syntax”, as it were.

As a point of departure note that English nominalizations are generally not stative:

(128) The opening of the doors occurred at exactly 10.

The nominal in this example doesn’t refer to a “post-state” or to any of the states
in (123), but rather to the opening event, which is itself a change of state. The verb
“occur” takes such a denotation (for discussion of event denotations, see chapter 3)
and yields a proposition that is roughly synonymous with “the doors were opened at
exactly 10”, a plainly eventive meaning. Nothing of this sort is possible with statives;
i.e., once a state, forever a state.
Like English nominalizations, Mẽbengokre nominalizations, when merged with a
definite determiner, denote definite descriptions of eventualities, as in example (105),
repeated here:21

(129) ba bẽnjadZw7r kutE bẽn dZir ja ma

1nom chief 3erg speech put.n.sg det hear

21
As we saw in chapter 3, embedded nominalizations don’t have space for tense and aspect parti-
cles, unlike main clauses. We will assume that such embedded nominalizations are at most definite
descriptions of events, without ever projecting Asp and T. Our discussion of tense and viewpoint
aspect above is therefore only relevant for main clauses.

107
 ‘[I heard] the reciting of the/a ritual speech by the/a chief’
ιe∃x∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

That is, the nominalization itself has among its meanings a set of eventualities,
which can combine with an optional determiner to yield a definite description of an
event. In chapter 3, we explored how this semantics was a natural extension of the
semantics we developed for internally-headed relative clauses.
As we said above, n, though fulfilling a morphosyntactic function (licensing struc-
tural genitive case, cf. chapter 2), is semantically vacuous. In embedded contexts,
binding by a definite determiner or an existential operator (as in internally-headed
relative clauses) yields a non-stative interpretation. In matrix clauses, properties
of the construction as a whole embed the event described by the projection of the
predicate in another eventuality, which is itself stative in the required way.
In what follows, we will pursue an analysis where the stativity of matrix nominal-
izations stems from the fact that matrix sentences headed by nouns, be they underived
or “deverbal”, are always interpreted as existential or possessive constructions.

4.4 The parallel with existential constructions

Regular noun phrases have a property that is very relevant for our analysis: they show
an ambiguity between referential interpretations, as in the (ii) readings, and proposi-
tional interpretations, as in the (i) readings, with no obvious distinction between the
two interpretations in the morphosyntax:22

(130) a. tEp kam tSaw

fish in salt

22
There is further ambiguity in noun phrases such as the above, namely that either of the bare
nouns in the complex noun phrase can be the head of the referential expression, yielding the readings
“the fish on which there is salt” and “the white man who has a canoe”, in addition to those given. This
ambiguity also has parallels in the domain of constructions with non-finite (or nominalized) verbs,
but is not directly relevant to the present discussion. This is a common cross-linguistic property of
internally-headed relative clauses that is discussed in chapter 3.

108
 i. “There’s salt on the/a fish.”

ii. “The/a salt that there is on the/a fish.”

b. kubẽ ñõ k2
barbarian poss canoe
i. “The/a white man has the/a canoe.”

ii. “The/a canoe that the/a white man has.”

This ambiguity has parallels in the domain of constructions involving nominal

forms of verbs. Compare one such case, (131), with the examples above. The am-
biguity doesn’t arise in (verbal forms of) verbs, which always receive a propositional
interpretation.23

(131) kutE arẽñ

3erg hear.n.sg
i. “(S)he has said.”

ii. “(The event) of her saying it.”

What should we make of the ambiguity between “nominal” and “sentential” read-
ings of all noun phrases? Are nouns always ambiguous between being “predicative”
and being “referential”?
We contend that this is precisely what is not the case. Nouns, contrary to finite
verbs, never predicate directly. To show this, observe the following examples, which
are more or less representative of the full range of nominal clauses:

(132) a. b2 kam mrW

woods in game
“There is/are (an) animal(s) in the woods.”

b. kubẽ ñõ k2
barbarian poss canoe
“The white man has a canoe.”

23
In line with what we said above about the multiple readings of internally-headed relative clauses,
the construction in (131) also has the reading “what she said.”

109
 c. i-kra
1-child
“I have a child.”

d. ij-2̃ ları̃ZitSi
1-on laryngitis
“I have laryngitis.’

In none of these cases do we have a regular subject that is identified with the noun’s
referential argument. Instead, the “subjects” of matrix clauses headed by nominal
predicates are locative postpositional phrases, or, in the case of (132c), a noun phrase
in the genitive that is assigned as a structural case by inalienably possessed nouns.24
The constructions in which they appear can be described as existential, in a way to
be made precise shortly.
The most straightforward example of an existential construction is represented by
(132a). Existential constructions simply state that there are individuals that fit the
description of the predicate in a particular location. Several scholars (cf. Benveniste

24
There is one construction that looks like predication where it is required that the “subject” be
in focus position:

tEp nẽ jã

fish nfut this
“This is fish.”

This construction, which looks like an equative copula construction, is quite limited, being possible
only with the demonstratives jã and wã as “predicates”. The following, for instance, is not permitted:

* i-kra nẽ kw7r7kro

1-son nfut Kwỳrkrô
Intended: “Kwỳrkrô is my son.”

But to predicate of a locatum focalization seems to be required:

mrW *(nẽ) b2 kam

game nfut woods in
“The animal is in the woods.”

In any case, all of these involve the pre-nẽ position, which is used by clefted or contrastively
focalized constituents, about which we haven’t said much in this dissertation.

110
1971 and Freeze 1992, among others) have noted the parallels between possessive
and existential constructions.25 To Freeze (1992), possessive sentences are a special
case of existential constructions with dative or genitive “locations”. In this spirit,
we consider possessive constructions such as (132b) and (132c), and “affected theme”
constructions such as (132d), to be part of the same phenomenon.

More specifically, we contend that while verbal predication (where α is the sub-
ject) is just [α P (x)] → P (α), predication in nominal sentences is indirect, i.e.,
[α P (x)] → ∃x P (x) ∧ Q(x, α), where Q represents a relation expressed by a postposi-
tion.26 The relation can be locative or possessive, something which, as we said above,
we consider a special type of locative relation. One might nevertheless ask whether,
giving enough latitude to what Q can be, “indirect” predication doesn’t mimic the way
in which external arguments are introduced in a proposal such as Kratzer’s (1996),
i.e., ∃eP (e) ∧ Q(e, α). Of course, this is something we wish to avoid, and for this
reason we will characterize Q more precisely below.

A first approximation of the translation of examples (132a-d) is, respectively, the

following:

(133) a. ∃x: animal(x) ∧ in(the woods)(x)

b. ∃x: canoe(x) ∧ to(barbarian)(x)

c. ∃x: child(x) ∧ of(me)(x)

d. ∃x: laryngitis(x) ∧ on(me)(x)

25
Strictly speaking, Benveniste (1971) notes that have-constructions historically replace existential
be-to-constructions, but no claim is made about a synchronic relation between the two.
26
Note the counter-intuitive postulation that in “there are animals in the woods”, the subject is
“woods”. This nevertheless accords with the cross-linguistic generalization established by Freeze
(1992), where locations in existential constructions pattern distributionally with subjects of verbal
predicates.

111
 Here we are not concerned with the fact that different postpositions are used to
express slightly different relations between the “subject” and “predicate”.27
This approach highlights the essential unity between the “sentential” and “refer-
ential” interpretations of nominal constructions. Note the parallel with one of the
“nominal” interpretations of such constructions:28

(134) a. ιx: animal(x) ∧ in(the woods)(x) (i.e., the animal in the woods)
b. ιx: canoe(x) ∧ to(barbarian)(x) (i.e., the barbarian’s canoe)
c. ιx: child(x) ∧ of(me)(x) (i.e., my child)
d. ιx: laryngitis(x) ∧ on(me)(x) (i.e., the laryngitis I have)

There are other examples of nominal sentences that might seem prima facie slightly
thornier to reduce to existential constructions. Let us consider them now. The first
case is the equative copular construction.

(135) a. i-be a-ñõ bikwa

1-at 2-poss friend
“I’m your friend.”

b. ∃x: friend(x) ∧ of(you)(x) ∧ in(me)(x)

In an analysis of the closely related language Apinayé, de Oliveira (2005) simply

calls be a copula, and ascribes to the construction the expected syntax. The problem
with such an approach is that be has the syntax of a postposition (i.e., it appears to
the left of the main predicate), and would be homophonous to a locative postposition
that appears elsewhere. But, given the semantics of the above examples, can we
maintain that it is a postpostion?

27
Note in particular the opposition between inalienably possessed nouns (132c), which express their
possessor as genitive inflection, and alienably possessed ones, in which the possessor is expressed by
an ad hoc postposition.
28
A not particularly careful reader will have noted that the relation with the other nominal reading
is not as direct. We return to this later. Also, the definiteness comes not from the expressions
themselves, but from the particular determiner that is merged. We assume ι, as elsewhere.

112
 In fact, one can practically give a literal translation of (135a) in English as “in me
you have a friend”, so this particular example doesn’t present much of a problem, in
our view, and can also be translated as an existential, as in (135b).29
Another case that is worth discussing is represented by the examples in (136), not
only because of their translation as adjectives into English, but because in de Oliveira
(2003) they are treated as part of a class of adjectives (“descriptives”), distinct from
nouns:

(136) a. i-m2̃ krW

1-dat cold
“I’m cold.”

b. i-NrWk
1-angry
“I’m angry.”

We are not a priori committed to asserting that the heads of dative subject con-
structions like (136a) or adjective-like predicates like (136b) are actually nouns, but
especially in the latter case it is desirable to assimilate them to the morphologically
identical (132c), repeated below as (137), which is straightforwardly nominal (cf.
138):30

29
An obviously related, yet slightly different construction is the following:

i-be kajtirE
1-at Kajtire
“I am Kajtire.”

Here the “predicate” is a definite description. We could follow Dixon (2004), who on p. 564
discusses a similar construction in the unrelated language Jarawara and glosses it as the equivalent
of “(the spirit of) Kajtire is in me”. This makes it a locative, rather than an existential construction,
putting it out of the purview of this chapter. To be absolutely fair, however, the example is also
different from other locatives, since in these the locatum normally appears in the pre-nẽ position, as
in the last example of footnote 24.
For the relation between existentials and locatives, see Freeze (1992).
30
For a more thorough description of the morphosyntactic properties of lexical categories in
Mẽbengokre, see chapter 1.

113
(137) i-kra
1-child
“my child”

(138) a. i-kra-rE boj

1-child-dim arrive
“My child arrived.” (also “I who have a child arrived”)

b. mẽ kra-rE boj

pl child-dim arrive
“Those with children arrived.” (also “the people’s children arrived”)

We contend that the cases in (136) are no different from possessive expressions
such as the following, in Spanish and Portuguese:

(139) a. Tengo hambre. (I-have hunger — Spanish)

b. Estou com raiva. (I’m with anger — Portuguese)

Arguing in favor of this (in addition to the identity of agreement patterns) is the
fact that modification of “I’m angry” in Mẽbengokre is identical to modification of
“my head”:

(140) a. i-NrWk t7j

1-anger hard
“I’m really angry (I have a strong anger)”

b. i-kr2̃ t7j
1-head hard
“I have a tough head.”

Thus we have it that practically everything that morphosyntactically looks like a

noun in Mẽbengokre is a noun, and nouns have the peculiarity of not being able to
predicate directly of a subject, but rather being forced to be construed with a locative
postpositional phrase (or a genitive noun phrase) in an existential construction.
A small class of words like nW “recent”, kumrẽtS “authentic”, kaP2k “ersatz”, tũm
“old, former”, dZw7j “real”, pWdZi “one”, and a few others, might pose some problems.
Morphologically, they are identical to the nouns discussed above, yet their semantics
is non-intersective, and thus can’t be reduced to an existential construction. In fact,

114
such a class has to be admitted, and admitting it somewhat weakens our case for
treating non-finite verbs as nouns rather than adjectives, because, after all, there
seem to exist adjectives in the language. Yet this class of lexical items is a small
closed class, and there seems to be no derivational process to form members of this
class out of other words. Another word which belongs to this class is ket, the sentential
negator. Recall from our discussion of negation and manner modification in chapter
3 that negation is a main predicate that takes a nominalized clause or an underived
nominal as its complement. neg has, like the other words discussed in the text,
the agreement pattern of a noun, rather than that of a verb. It might be possible to
assimilate this class to the class of postpositions, which do appear as main predicates,
as in the construction exemplified in (50a), in §2.3.1. We will not pursue the matter
further in this dissertation.

4.4.1 Nominal forms of verbs in existential constructions

As we saw in chapter 3, and again in §4.1.2, one of the readings of a non-finite (or
nominalized) verb is just λe.∃x , . . . , xn P (e)(x ) . . . (xn ), or, after merging with a
(possibly null) determiner, ιe.∃x , . . . , xn P (e)(x ) . . . (xn ). By analogy to what was
described for underived nouns, predication involving a nominalized verb will be done
“indirectly”, i.e., what we represented above as [Q(x, α) P (x)] → ∃x P (x) ∧ Q(x, α).
Let’s examine how one gets from the embedded reading of the nominalization, which
we have already worked out, to the matrix interpretation, if we apply the reasoning
applied to underived nouns:

(141) ba bẽnjadZw7r7rE kutE bẽn dZir ma

1nom chief 3erg speech put.n.sg hear.v.sg
‘[I heard] a chief reciting a ritual speech’
ιe∃x∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

(142) bẽnjadZw7r7rE kutE bẽn dZir

chief 3erg speech put.n.sg
“There is a reciting of a ceremonial speech by a chief.”
∃e∃x∃y: recite′ (e, y, x) ∧ chief′ (y) ∧ speech′ (x)

115
 That is, matrix clauses headed by a nominalized verb are interpreted as “there is
a V-ing” or “there is X V-ing”. Yet this seems to give us no leads into the particular
aspectual interpretation that matrix clauses with a non-finite form get. Let’s see if
we can derive this.

4.4.2 The location

One important fact about existential sentences such as those in (132) is that they
have a “location”, as it were. Nominal constructions without a location are weird out
of the blue as clauses in Mẽbengokre (though obviously not as noun phrases):

(143) a. ?? tSaw
salt
“There is salt.”

b. ?? k2
canoe
“There is a canoe.”

Why might this be the case? Not differently from what we might say about
the English translations, one could maintain that a location is always independently
required. An overt location can be dispensed with if one is salient in the discourse
context, and perhaps, like in English, in special cases such as “there is a God”, “there
are unicorns”, and so on. Nevertheless, whether for pragmatic reasons or, as we will
argue, because of the syntax of the construction, a location restricting the existential
claim is always implicit.
In clauses formed with underived nouns, such as those in (132), the location is
straightforwardly a locative phrase, that can be a possessor, a location, and possibly
other things. In the case of nominalizations, there are a few options as to what the
location can be:

(144) a. There could be no location, just ∃e P (e).

b. The location could be the ergative subject.31

31
For simplicity, we will at first only examine transitive sentences.

116
 c. The location could be a (phonologically null) spatial location.

d. The location could be a (phonologically null) time interval.

The choice that makes the most of the analogy with existential sentences is su-
perficially (b), as can be seen by comparing a plain existential clause formed with an
underived noun with a clause headed by a nominal form of a verb:32

(145) a. [b2 kam]S [mrW]P

woods in game
“There is/are (an) animal(s) in the woods.”
∃x[in′ (the-woods′ , x)]S [animal′ (x)]P

b. [bẽnjadZw7r7rE kutE]S [bẽn dZir]P

chief 3erg speech put.n.sg
“There is a reciting of a ceremonial speech by a/the chief.”
∃e[erg′ (the-chief′ , e)]S [speech-reciting′ (e)]P

The logical form in (145b), however, as we anticipated at the beginning of §4.4,

is no different from the way external arguments are introduced in proposals such
as Kratzer’s (1996), with no effect on aspectual interpretation. So, if we want to
account for the stativity of ergative clauses by recourse to the parallel with existential
constructions, the story can’t end here.
Recall that in chapter 2 we established that ergative case was not θ-dependent, like
inherent cases, and instead was a structural case assigned to external arguments inside
a nominal projection. erg therefore doesn’t have an inherently locative semantics,
as the “locations” in other existential constructions do. That is, (145b) isn’t really
parallel to (145a).
It seems to be the case that the “locations” in locative constructions necessarily
get a “location” θ-role. While those relations expressed by the locative postpositions
in (132) fit the bill, and possibly make the locative relation precise (in addition to
assigning case to their complements), erg arguments by themselves are not enough

32
In these examples, subscript S stands for the “location”, and subscript P for the “locatum”,
something that we will expand on below.

117
to be “locations” or “subjects” of existential constructions. This seems to be the case
also in English, given examples such as the following:

(146) a. # There was a performance by Marta Argerich.

b. There was a performance last night.

c. There was a performance at the amphitheater.

Under our assumptions, the perceived incompleteness in (146a) is due to the fact
that the existential predicate requires a locative argument, and there is no way of
getting it from an agent in English. That is, only spatial or temporal locations satisfy
the “thematic” requirements of the external argument (“location”) of the locative
construction.33
We propose that the difference between English and Mẽbengokre existential con-
structions is that in Mẽbengokre a noun phrase that doesn’t fit the θ-role assigned by
the existential construction to its “subject” or “location” is interpreted twice: once as
whatever theta role it gets from the embedded clause, and once more as a location.
The equivalent of (146a) in Mẽbengokre is therefore interpreted as “there was a per-
formance by Marta Argerich to Marta Argerich”, or (given that “there is X to Y” in
English is spelled out as “Y has X” 34 ) “Marta Argerich has performances by herself”,
or, as we ultimately wish to argue, “Marta Argerich has performed”.
How does a single participant come to be interpreted twice in the structure? For
purely illustative purposes, we could make an analogy with the following construction,
described by Freeze (1992):

(147) This flour has weevils in it.

33
The essentially locative nature of that argument is evidenced in English by the etymology of the
expletive used in existential constructions.
34
Cf. also the following:

There is a message for you ≡ you have a message.

118
 Freeze characterizes (147) as involving inalienable possession, which might suggest
that even in the case where the location is literally locative, it is interpreted twice
in an existential construction, once as a pure location, once as the subject of an
“inalienable property”.35
The particular problem posed by Mẽbengokre is therefore not whether a “double
thematic interpretation” of the subject is plausible, but rather how one obtains it.
No overt pronominal, as in (147), marks the position where the locative θ-role would
be transmitted. What is, then, the structure that yields the required interpretation?
We should be careful to distinguish the case of (145b) from the control construction
we found in the progressive, discussed in §2.3.2.36 The crucial property relating the
subject of the locative predicate and the subject of the lower nominal clause, aside
from their obligatory referential identity, is that its case, and presumably its scope
possibilities, are given solely by the lower clause. This seems to discard raising, even
if we admit an approach such as Hornstein’s (1999), where picking up two theta

35
A further analogy could be made with Spanish datives (cf. the description in Cuervo 2003).
In Spanish, clitic-doubled datives can only be recipients, and therefore animate, contrasting with
non-clitic-doubled datives, which are destinations. If an inanimate dative is doubled by a clitic, as
in the third example, the reference is interpreted as disjoint:

a. Envié un paquete a Francia.

Sent a package to France
“I sent a package to France.”

b. Le envié un paquete a Juan.

3dat sent a package to Juan
“I sent Juan a package.”

c. Le envié un paquete a Francia.

3dat sent a package to France
“I sent him a package to France.”

If the interpretation of these cases is similar to what we maintain for the location in the locative
construction, there are two θ-roles in (c), one associated with the P, the other with the clitic, and
they can both be absorbed by a single referent if the relevant noun phrase has the right features.
36
It should perhaps also be kept distinct from what happens in negation and other forms of
subordination, where no higher subject is thematically interpreted.

119
roles by movement is allowed. For the purposes of this dissertation, we will assume
that the locative predicate’s subject is saturated by a pronominal element that is
correferential with the highest argument in the lower clause, that is, something like
“backward control” (cf. Polinsky and Potsdam 2002 for discussion). The matter is of
course open for future research.
At this point we could ask where the locative predicate in nominal clauses comes
from. For the purposes of this dissertation we bite the bullet and admit that it is
a predicate that exists in the lexicon, though one that is independently needed to
interpret clauses “headed” by nouns. Why it is required will become clear in the
following sections, as we discuss linking of the eventualities to topic time through
higher functional projections.
The predicate in question could be considered to be a sort of (phonologically-null)
positive counterpart to the negation ket, discussed in chapter 3, with the caveat that
the latter doesn’t seem to require a locative subject.37 We will be more precise about
the decomposition of this predicate in §4.5.

4.4.3 Provisional summary

We have taken the position that the projection of lexical predicates is category-
independent. Lexical predicates that project a referential argument with the right
features, i.e., an argument e ∈ D v , can become both nouns and verbs. If they merge
with v, the e variable gets existentially bound and is linked to topic time. If they
merge with n, they can head referential expressions, or become propositions by fur-
ther merging with a higher predicate, which, unlike the existential closure effected by

37
Essentially, this is the device used by Reis Silva (2001) to justify main clause ergativity in
Mẽbengokre and Timbira, respectively; i.e., both depart from the assumption that embedded-clause
ergativity is a given, and propose that there is a null predicate embedding the ergative clauses that
seem to be matrix clauses. In neither of those works is there an independent justification for such
a predicate. Cf. also a similar approach to ergativity in Gitxsan by Hunt (1993). Also relevant
here, though limited only to possessive constructions, is Vieira’s (2001) discussion of bahuvrihi
constructions in Guaraní.

120
v, requires a locative subject.
Let us call this predicate loc. In the following section, we endeavor to derive the
aspectual interpretation of main clauses headed by nominal forms of verbs from the
semantics of existential constructions, that is, from the way loc relates “subjects”
and “predicates” in nominal clauses.
One might ask if loc isn’t just another name for a stativizer, bringing our solution
very close to the solution in §4.3.1, which we discarded. The answer is that while the
semantics resulting from merging loc might be like the semantics of a stativizer,38
separating the stativizing element from the category-assigning head allows us to ac-
count for the fact that nominal constructions are not only stative clauses, but also
non-stative clauses (in embedded contexts) or referential expressions. Furthermore,
as we will see in §4.7, this move will allow us to decompose the notion of derived
stativity in an interesting way.

4.5 Obtaining the experiential perfect

So far, we have established an equivalence between the Mẽbengokre sentences in

(148a) and (148b) and the English sentences (149a) and (149b), respectively.

(148) a. krw7j jã nẽ kutE mop krẽn

parakeet dem nfut 3erg malanga eat.n.sg
“This parakeet has eaten malanga.”

b. krw7j jã nẽ kutE mop kur

parakeet dem nfut 3erg malanga eat.n.pl
“This parakeet eats malanga.”

(149) a. There is an eating of malanga to this parakeet.

b. There are eatings of malanga to this parakeet.

(150) a. ∃e: loc′ (e, parakeet′ ) ∧ eating-malanga′ (e) ∧ Ag′ (e, parakeet′ ) ∧ sg′ (e)

b. ∃e: loc′ (e, parakeet′ ) ∧ eating-malanga′ (e) ∧ Ag′ (e, parakeet′ ) ∧ pl′ (e)

38
The prototypical states then being, at some deep level, “having”, or “existing in a location.”

121
 Of course, the parallel is only structural. The English sentences in (149) are
meaningless for independent reasons. We will assume the translations in (150), which
already incorporate the notion that subjects are interpreted twice in existential con-
structions, once in the role which relates them to the predicate, and once as locations.
This makes loc the locus of our discussion.39
Above we claimed that (148a) is interpreted as an experiential perfect, while
(148b) is interpreted as a generic or habitual. Our task is to show that the logical
forms in (150), which are composed of the morphological categories that are apparent
in the accidence of Mẽbengokre verbs, are equivalent to these interpretations. In
what remains of the chapter, we contend to have completely derived (150a). (150b)
presents us with a series of interesting complications that we haven’t been able to
fully address so far. We nevertheless sketch what we believe needs to be done to
proceed.

4.5.1 The experiential perfect

Iatridou et al. (2001) propose a semantics for the perfect broadly in accordance with
the “extended now” theory of McCoard (1978). In such a theory of perfect meaning,
the perfect consists of an interval, the “perfect time span”, whose right boundary
(RB) is the evaluation time, and whose left boundary (LB) is set by a special type of
adverbial. The semantics are formalized by von Fintel and Iatridou (2005) as follows:

′
(151) a. JPerf φKt ⇔ ∃t′ : RB(t, t′ ) ∧ JφKt

b. RB(t, t′ ) ⇔ t ∩ t′ 6= ∅ ∧ ∀t′′ ⊆ t′ : t′′  t

The claim is that the proposition φ is true at some interval that goes up to
the evaluation time. For the existential perfect, the definition needs to be adapted
somewhat, namely to a claim that the proposition φ is true at some point in the
interval. That is:

39
Note also that the formulas assume that the cardinality of the eventuality is affirmed, rather
than presupposed or implicated. We will return to this issue below.

122
 ′′
(152) JPerf φKt ⇔ ∃t′ : RB(t, t′ ): ∃t′′ ⊆ t′ : JφKt

It’s relatively trivial to arrive at this meaning starting from the translation given
in (150a). Informally, we could propose a lexical entry for loc as follows:


 λy.λx.x is in space in y at t, if x ∈ D e
(153) JlocKt =
λy.λe.e is in time in the experience of y at t, if e ∈ D v



It seems counterintuitive to set the endpoints of an individual’s life-span (“expe-

rience”) in the semantics, since, after all, if P (α) at some interval t, then the interval
has to be contained in the time span during which α exists. Nevertheless this is one
plausible way to assign an interpretation to an existential perfect that lacks an ad-
verbial phrase to specify the left boundary of the perfect time span within von Fintel
and Iatridou’s (2005) proposal. Thus:40

(154) a. JlocKt = λx.λe.τ (e) ⊆ τ (x)

b. τ (x)De →Di := λx.ιt′ : RB(t′ ) = t ∧ LB(t′ ) = the birth of x

This is nothing other than the meaning of the experiential perfect that we ex-
panded in (152) above.
We are now in a position to understand why loc is required: as we saw before,
the denotation of nouns is not relativized to times. We stipulate that being linked
to topic time41 is a sine qua non condition for the interpretation of a proposition.
An additional (time-dependent) predicate is therefore necessary in order to interpret
nouns. What Mẽbengokre has in its lexicon that can satisfy this requirement is
the locative relation loc, which is employed to interpret both “underived nouns”
(i.e., those whose referential argument is an entity) and “verbal nouns” (i.e., those in
which the referential argument is an eventuality). Though Mẽbengokre has only this
resource, it seems that Universal Grammar provides languages with another option
to resolve the mismatch between noun denotation and higher functional structure,

40
In the definition of τ , t stands for the evaluation time applied to loc. Di is a domain containing
all time intervals.
41
By Asp; cf. discussion in §4.1.3.

123
namely the equative copula that we know from many better-studied Indo-European
languages. A discussion of the differences between these two “auxiliary predicates”,
their acquisition, and other questions that could be raised here would take us too far
afield, and is therefore left for later research.42

4.5.2 Excursus on perfect and perfective

We should now summarize our thoughts on the distinction between perfect and perfec-
tive, which so far have been scattered. For a more complete contemporary discussion,
including a description of the perfect’s formal properties, which we are not directly
interested in, the reader is referred to Iatridou et al. (2001) and Katz (2003).
Consider the following minimal pair:

(155) Bill arrived at seven last night.

a. He read the paper.

b. He has read the paper.

The two continuations differ in many respects; what we wish to call the reader’s
attention to is that, while (155a) links the event time directly to the topic time (which
is set by the previous discourse context, and then advanced), no such direct link exists
in (155b). That is, in narrative, (155a) means that Bill read the paper at some time
sufficiently soon after arriving. No such relation between arriving and reading the
paper is implicated if the continuation is (155b). That is, that sentence could be used
for a reading of the paper that took place before or after Bill’s arrival.43
In the perfect, i.e., (155b), what seems to be linked to the topic time is not the
time interval corresponding to the event of reading the paper, but rather some other
time interval. The eventuality’s time (i.e., the reading of the paper) is contained in
this interval.

42
Of relevance here is the cross-linguistic survey of the verb “to be” compiled in Verhaar (1968).
43
Before, as in: “Bill arrived last night. I found out from him then that he has read the paper.”
After, as in: “Bill arrived last night. He has read the paper since then.”

124
 What is this interval? In Iatridou et alii’s work this is what is called the “perfect
time span”. In English, the right boundary of this time span is set by evaluation time
(i.e., the present in the present perfect), while the left boundary may be set by a
prepositional phrase headed by specialized adverbials such as since.
The Mẽbengokre “perfect” which we have discussed here is more restricted than the
English perfect, in that it only allows what we’ve called experiential (earlier “subject-
oriented” or i-level) reading, that is, the left boundary is arbitrarily set to coincide with
the birth of the subject, rather than being fixed by an overt prepositional phrase.44
The point to be made is the same, however: the event’s time is not linked directly to
topic time; it’s the “experience” of the subject that is, as it is “the experience of the
subject up to evaluation time”, as we saw in (154b). Note that “experience”, as we
introduced it above, means the whole of the subject’s lifespan up to evaluation time,
and crucially not the timespan from the moment of “experiencing something” to the
present; i.e., “x is in my experience” should be understood as “x is in the domain of
my life experience”.
The difference in representation between the perfect and the perfective, which
we summarize in the two semi-formal LFs below, is what accounts for many of the
properties of the nominal form of the verb; in particular, it should be clear why such
forms are employed to give background information: topic time is side-stepped by
them, so to speak.

(156) Perfective
′
JPfv φKt ⇔ ∃t′ : t′ ⊆ t: JφKt

(157) Perfect (experiential)

′′
JPerf φKt ⇔ ∃t′ : RB(t′ ) = t ∧ LB(t′ ) = the birth of a: ∃t′′ : t′′ ⊆ t′ : JφKt

At this point, one could raise the following objection: any event in which x
is involved has to have taken place in x’s experience, whether it be described as
perfective, perfect, or imperfective. This is absolutely true; the point, however, is

44
In fact, this happens in English if no specialized adverbial is present, as in “I’ve read Annu
Kareninu five times (i.e., in my life).”

125
that the (experiential) perfect claims no more than this, whereas the perfective and
the imperfective further claim (or implicate) that the event in question took place
relative to a more restricted interval that is manipulated by surrounding discourse.
This, we claim, is the main defining trait of the perfect.45 In particular, linking or
not the event time to topic time is the essential point of contrast between perfects
and perfectives.
Other properties of the perfect are often taken to define it, as opposed to the
perfective. To take an example, consider the “perfect paradox” (cf. Klein 1992 and
Pancheva and von Stechow (2004)):46

(158) a. John arrived last night.

b. * John has arrived last night.

It is surprising that (158b) should be ungrammatical, if, like (158a), it describes

an event in the past. Providing a full account of the ungrammaticality of (158b)
requires us to spell out several assumptions about its structure, and would take us
too far afield.
In Mẽbengokre, though there is no paradox, in the sense of an unexpected un-
grammaticality, there is a contrast in the interpretation of time adverbials depending
on whether they occur with nominal or verbal forms of the verb. Whereas with the
latter the adverbial straightforwardly modifies the time of the event, as in (159a), an
adverbial occuring in a nominal sentence doesn’t.

(159) a. amrẽbe nẽ ba arẽ

long ago nfut 1nom say.v
“I said it a long time ago.”

b. amrẽbe nẽ ijE arẽñ

long ago nfut 3erg say.n

45
Of course, other aspectual values share this trait, in particular habituals or generics, so we will
need to propose criteria to distinguish between these and the perfect.
46
In not all languages that have it is the perfect subject to the perfect paradox. Cf. Giorgi and
Pianesi 1998. This in itself would be sufficient to cast doubt on the paradox being a good defining
trait of the perfect. We won’t go into the merit of the question here.

126
 “It’s been a while since this has been said by me.”

The exact interpretation of (159b) is not clear to us; it either has an anteriority
component, or it forces an interpretation where a present situation extends back to
the specified time, as in “this is the law since a long time ago; I’ve said it”, which is
the way that underived statives get interpreted:

(160) amrẽbe kamrek

long ago red
“It’s been red since a long time ago.”

We thus have evidence that Mẽbengokre main clauses with nominal forms satisfy
two defining criteria for perfects. Of course, like any term with a long tradition
of use, “perfect” has many additional associations, which probably don’t extend to
Mẽbengokre nominal forms. We nevertheless believe that the choice of criteria in our
definition of perfect is quite promising in terms of cross-linguistic comparison.

4.6 Speculations on generics and habituals

Can the reasoning sketched in the previous section be extended to the generic or
habitual? Our starting point is the meaning in (150b), repeated here:

(161) ∃e: loc′ (e, parakeet′ ) ∧ eating-malanga′ (e) ∧ Ag(parakeet′ , e) ∧ pl′ (e)

That is, “eatings of malanga are in the experience of this parakeet”. Two problems
arise: one is how a simple plurality of events is interpreted to mean a frequent event;
a second problem is whether having an event repeated frequently in the subject’s
past experience really amounts to the meaning that a habitual or a generic has. We
will see that this latter problem further subdivides into two, which will be treated in
ß4.6.2 and 4.6.4.

4.6.1 The meaning of plural

As we anticipated in §4.2, the indication of number of number on Mẽbengokre verbs

should properly be analyzed as indication of event plurality, at least in the cases that

127
are relevant to us here. We therefore have the logical form in (161).47 But simple
event plurality is plainly not equivalent to the frequent or habitual repetition of the
event.
The problem seems to arise with pluractionals cross-linguistically: as we already
anticipated in §4.2, rather than being a simple plurality, pl seems to denote a large
number of eventualities, more than the number that could be readily described as a
well-defined quantity with one of the cardinality expressions available in Mẽbengokre.
In fact, this might be a property not only of pluractionals, but more generally also of
bare plurals. Consider the first two sentences from below:

(162) John has a horse.

(163) John has horses.

∃x: to′ (x, John′ ) ∧ horse′ (x) ∧ pl′ (x)

(164) John raises horses.

Though (163) is true even if the cardinality of the individual x is 2, a bare plural
would be considered uninformative in most contexts if the fact that |x| = 2 were
known by the speaker. In fact, (163) has a flavor not unlike (164). What is the
meaning of the bare plural in (164)? Intuitively, it seems to implicate that John
raises enough horses to keep him busy, or to make it his primary occupation.
If we make a parallel between the interpretation of the plural marked on verbs and
the bare plural on nouns, we can analogously state that the former has the implicature
that the number of events is enough to fill an interval in a contextually salient way.
That the “filling of the interval completely” is an implicature can be seen by the
possibility of canceling it, given certain left-peripheral particles and the availability
of a pragmatically plausible reading, as in the following example:

(165) rOprE jã ar7m kutE krarE jadZw7r

dog this already 3erg puppy put.pl.n
“This dog has already given birth to puppies.”

47
A further possibility, which we cannot address here, is that event plurality not be part of the
truth conditions of plural verbal phrases, but rather part of their presuppositional content.

128
 In this case, the event of giving birth, even though it’s plural, is interpreted as
contained within some past time interval in the dog’s life-span, rather than filling the
latter completely. The proposition as a whole is interpreted as an existential perfect,
with the cardinality of the eventuality being greater than one.
We can now move to the second mismatch between (161) and habitual or generic
meaning: the modal component of the latter, not expressed in the former.

4.6.2 Projection into the future

A much more complex problem than the one discussed in the previous section is
presented by what has been called the modal component of habituals or generics.48
So far, we have established that the meaning of (161) is such that the cardinality
of the plural eventuality is enough to fill the previous experience of the parakeet in
some relevant way; i.e., what is intuitively translated by the English (149b). This,
however, is not what a habitual or generic means.
Habituals or generics “project modally”, so to speak. This is exemplified by the
following minimal pair:

(166) a. John has raised horses.

b. John raises horses.

These two sentences differ in more than one respect, but one clear contrast they
show is that in (166a), by evaluation time all of the horse-raising events are past, and
no commitment is made as to the continuation of horse-raising events by John beyond
evaluation time. (166b), on the other hand, entails that, under normal circumstances,
there are more horse-raising events by John to come after evaluation time.
Before we can address this, we need to backtrack somewhat. Example (165) forces
us to revise our empirical generalizations in a way that affects our analysis.

48
In this and the following sections, our discussion is partly based on Ferreira’s (2005) treatment
of modality in progressives and habituals.

129
4.6.3 Excursus on perfects and generics

At the beginning of the chapter, we synopsized the relationship between the morpho-
logical categories of number and lexical category and the aspectual interpretation of
predicates with the following table:

(167)
Verbal Nominalized

Singular V Perfective; singular object Experiential perfect, relative

to the subject’s life-span
Plural V Perfective; plural object Habitual or generic

In §4.6.1, we showed that the relationship between plural number and habituality
is not direct: plural doesn’t automatically mean habitual; rather, it means habitual
through an implicature that can be cancelled, as was shown in example (165). In
§4.6.2, we have hinted that the habitual is one more step removed from the plural:
the habitual has a modal component that is unexplained by whatever apparatus we
have introduced so far.
In fact, example (165) also shows that the habitual’s modality can be dissociated
from plural marking on the verb. The “plural perfect” given in that example has as
its most readily available translation a “plural existential”, i.e., one where more than
one event occurred, and possibly many did, but no modal projection into the future
is implied.49
If we wish to translate the dissociation between number and modality in the
habitual into a matrix comparable to (167), we would have to add one more dimension
to the matrix, albeit one that is only reflected in the n.pl cell of the matrix. We can
call this dimension the “modal” dimension. If we add it to the matrix, we get the
following:

49
We leave open the possibility that this interpretation compares with the so-called universal
perfect, i.e., “this dog has been giving birth to puppies”, which also lacks modal projection.

130
(168)
Non-Modal Modal
Verbal Nominal

Singular V Perfective; singu- Singular experiential —

lar object perfect
Plural V Perfective; plural Plural experiential Habitual
object perfect or universal
perfect

Why doesn’t the modal contrast apply in the other cells of the old matrix? In the
case of the verbal cells, we stipulated that the lexical entry of v contains ∃e, so modal
readings are excluded.50

The case of the n.sg is rather interesting. Ferreira (2005) has argued that two
types of imperfectives differ solely in the cardinality of the eventuality described.
While habituals are plurals, progressives are singular. Both have the same modal
component. We would therefore expect n.sg to be interpreted as a progressive.51
Why doesn’t this happen?

The answer seems to be that while all of the nominal aspects in the matrix above
are subject-oriented (i.e., i-level) and stative, the progressive is s-level and dynamic.
For this reason, the progressive construction in Mẽbengokre always requires an overt
auxiliary,52 which is actually an activity-denoting verb which takes a nominalized

50
There is again modality higher in the tree: the particle dZa, that is used most often to indicate
the future, also has certain other uses (v.g., in yes-no questions) which suggest that it is a sort of
irrealis, the particle r2̃ñ, which is used to express doubt and in counterfactual constructions, the
evidential particle wE, and possibly others. Throughout this dissertation we’ve fixed the higher
structure to exclude any modality other than what is introduced low in the tree.
51
There is in fact another possibility, which is actually attested (cf. Thomson 1974): nominal forms
of verbs are used in antecedents of counterfactual expressions, regardless of the number marking.
We don’t discuss counterfactuals in this dissertation.
52
See §2.3.2.

131
complement (activities are always s-level; see Fernald 2000).53
So, pending a better solution, non-finite verbs in Mẽbengokre display an opposition
in modality which has no phonological reflex. The source of this modality is unclear to
us at this point; it might, like the interpretation of the plural as “sufficient to occupy
the subject completely”, be supplied by the pragmatics as a plausible inference. For
purely speculative purposes, we propose that it actually be a phonologically null
morpheme.
What we have in mind is the morphological category that distinguishes between
present and past participles in certain modern Indo-European languages, in particular
those of the Romance family.In Spanish, for example, the exponent of this morpho-
logical category is the -n- that distinguishes amando ‘loving’ (or amante ‘lover’) from
amado ‘loved’. The present participle, used in progressives, and the adjectival form
amante, which has habitual meaning, are the modal counterpart of the past partici-
ple, which is always non-modal. The -n- in both the adjectival form and the present
participle could thus be argued to encode the modality that characterizes these two
forms.
Let’s call the phonologically abstract modal element found in Mẽbengokre N.
How does N fit into our analysis of Mẽbengokre? We propose that it merges with the
existential raising predicate loc to yield a slightly different relation, locn , which we
could informally translate as follows:

(169) λy.λe.∀w: wRw ∗ : e is in time in the experience of y in w

53
Note that the progressive construction in English doesn’t exclude an “s-level habitual” reading
(or, as the second example shows, generic):

a. John is eating at the trucks these days.

b. John is answering the mail from Antarctica now, but when Bill gets back from vacation
it will be his job again.

This seems to indicate that what’s essential to the meaning of the English progressive construction
is the s-levelness, not the singular number. There are many limbs like this that it’s tempting to go
off on, but we won’t.

132
 That is, in all worlds related to the actual world (w ∗ ) by a certain accessibility
relation (in the case of the progressive, the relation that selects those worlds where
the expected consequences of present conditions in the actual world hold). It is not
clear to us how this particular lexical item would be tied to evaluation time.

4.6.4 More on the modality in imperfectives

The second point about a habitual or generic’s modality is that it doesn’t need veri-
fying instances to be true. Or rather, a distinction is often made between habituals,
which require verifying instances, and generics, which do not.54 Observe the following
examples:

(170) a. This machine crushes oranges.

b. This elevator carries eight passengers.

c. John answers the mail from Antarctica.

In none of these sentences is it necessary for an eventuality of the type described

by the clause to have actually occurred; that is, the machine in question might not
have been used yet, the elevator might have never been used to full capacity, and
there might be no mail from Antarctica, but the sentences will still be judged to be
true. This is the generic reading.

In Mẽbengokre we find that such “potential” readings are also characteristic of

nominal forms of verbs in main clauses, whenever the verb is either plural or unmarked
in number. The following examples illustrate this:

(171) a. i-kra nẽ ar7m 2bir

1-son nfut already 3.climb.n
“My son is already capable of climbing on things.”

b. ar7m ajtE 2kur

already again 3.feed.n.pl
“He’s able to eat again (after having been sick).”

54
We thank Jürgen Bohnemeyer for bringing this question to our attention.

133
(172) bEp kutE ları̃ZitSi kanE
Bep 3erg laryngitis treat
“Bep is the one that treats people for laryngitis.”

Example (172), elicited in a context where it was made explicit that nobody had
had to be treated for laryngitis yet, is meant to be parallel to the English examples
in (170).
Is this just a special case of the modality introduced above? This is something
that we cannot answer at this point. In any case, whatever the exact formulation of
this modal element’s semantics is, there are consequences to the way we proposed to
introduce it into the structure: if N merges with loc, then we expect that existential
clauses with underived nouns in Mẽbengokre might have modal readings. This in fact
appears to be the case, though the following examples, which were collected by us as
spontaneous utterances in the field, are somewhat hard to reproduce in elicitation:

(173) a. wajaNa nẽ udZW O ku-bı̃

shaman nfut 3.spell instr 3-kill.v.sg
“The shaman killed him with a spell.”

b. i-dZudZW
1-spell
“I can do witchcraft (lit., I have spells)”

c. i-ñimrW kumEj
1-prey much
“I’m a good hunter (lit., I have much prey)”

d. dZOri kra r2̃P2̃

int child still
“Can she still bear children? (lit., does she still have children?)”

Example (173), though easier to reproduce in elicitation than the others, is not a
very clear-cut case, given that it’s hard to establish that dZudZW doesn’t really mean
“power”; (173d) has an inchoative meaning which might cast doubt as to whether it
really is a plain existential construction, but, like (173c), it is nominal and has the
ability and/or projection into the future that is characteristic of nominalized verbs.

134
 The fact that all existential constructions, whether involving deverbal nouns or
underived ones, may have this modal element is another point in favor of creating
stativity “in the syntax”, rather than internal to the deverbal nominal word.

4.7 Conclusion

In this chapter, we have derived the various subject-oriented readings of matrix clauses
headed by nominal forms of verbs by analogy with existential constructions. The tree
√
that we arrive at for nominal projections, when is event-denoting, looks as shown
in Figure 4-4.

H
H
 HH
∃e locP
H
H
 HH
 H
proi loc’
H
 HH
 H
loc nP
H H
 H  HH
loc N Subji n′
H
 HH
n NumP
H
 H√H
Num P
√HH
Obj

Figure 4-4: Structure of a nominal matrix clause.

This can be contrasted with verbal projections, shown in Figure 4-5, which link
the event directly to topic time, without the mediation of loc.
Above, we sketched how the verbal projection would be linked to viewpoint aspect
and tense. Recall that in the case of verbs, it was v itself which had a time-dependent
denotation. In the case of nouns, the required time-dependency is introduced by loc.
Contrary to v, we didn’t stipulate that loc binds a variable existentially, though
nothing crucial depends on whether we do or not, as far as we know. Presumably,

135
 vP
HH
 H
Subj v′
H
H
 H
v NumP
HH
√
∃e Num P

P (e)

Figure 4-5: Structure of a verbal matrix clause.

as we have indicated in the first tree, existential closure takes care of binding all free
variables, before the “bare proposition” is merged with the functional categories that
will bind t.

Many issues remain open, of course. Among the most important, we could cite
the linking of these structures to topic time through Asp. The issue is interesting
because there isn’t full freedom as to what value of viewpoint is attached to different
types of structure: while verbal clauses often receive perfective viewpoint (and this
of course depends on the particles present in the left periphery), from what we’ve
said it should be obvious that the default viewpoint assigned to nominal clauses is
imperfective, i.e., the states described extend beyond topic time.

Another important question that remains open is whether, in addition to the

experiential perfect interpretation and the generic or habitual readings which we
have discussed, the constructions involving nominal forms of the verb have other
readings which are not “subject-oriented”. If the construction indeed has parallels with
the perfect of better-known languages, we are led to expect other temporally-stative
readings to exist, such as the “perfect of present relevance”, and existential perfects
that aren’t bounded by the subject’s lifespan, but rather by an arbitrary timespan
extending to the evaluation time. The data collected by us are still inconclusive in
this regard.

Finally, our hasty discussion of modality in §§4.6.2 and 4.6.4 didn’t consider the
possibility that the contrast between the modal and non-modal readings might have a

136
source other than the abstract morpheme that we identified with the -n- of Romance
present participles. As we suggested in §4.2, the number contrast might do more work
in this regard than what it does in our analysis, and a series of implicatures might be
responsible for the modal extensions that are observed.

137
138
Envoi

In this dissertation, we have endeavored to derive all the ergative constructions in

Mẽbengokre from a single source, namely nounness. We have argued that all embed-
ded clauses are nominal, hence their ergativity, and that the cases in which ergativity
arises in matrix contexts, though a priori unexpected, can also be reduced to a nom-
inal construction embedded under a phonologically null existential copula that is
independently required.

Though we didn’t aggressively pursue applications of our analysis to other, better-

known languages, the direction to proceed should be clear. As we discussed in §4.3,
the division of labour in the composition of perfects and other “compound tenses” is
not the same in all languages, but the differences are minimal: on the one hand one
has languages where the stativizing morphology is combined with the verbal word,
yielding a participial; on the other, one has those where stativity is in the auxiliary
(or, as we said back then, “in the construction”). It’s not unreasonable to suppose
that the same constituent parts underlie both options for constructing the perfect.

Mahajan (1997) pointed out for the first time that ergativity could be found in the
perfects55 of certain Romance languages. Mahajan’s insight consisted in identifying
auxiliary selection in, e.g., the French (and Italian) perfect, with ergative marking
in the Hindi perfective. For Mahajan, these two constructions share an underlying
structure, and differ solely in that a particular element merges with the external
argument in one case, yielding ergative case marking, and with the auxiliary in the

55
By this we mean the construction that is structurally parallel to the English perfect, which in
most contemporary dialects of French and Italian has the meaning of a simple perfective past.

139
other, yielding have, rather than be, as an auxiliary.

(174) a. Raam-ne vah kitaabẽ parh ı̃ı̃ hẼ

˙
Raam-erg those booksi read-perf:agr i be-pres:agri

“Raam has read those books.”

b. Ram a lu ces livres.

c. Raam P be read those books

[Raam P] be → Raam-erg (Hindi)
Raam [P be] → Raam have (French)

If we are right in claiming that Mẽbengokre nominal clauses are essentially compa-
rable to the perfects found in certain Indo-European languages, then our contention
that ergativity in Mẽbengokre is always tied to nounness might also explain ergativity
in these cases as well. The perfect construction in all of the cases discussed involves
a participle, which is the functional equivalent of the nominal form of the verb found
in Mẽbengokre. Like the nominal forms in Mẽbengokre, the participial cannot link
directly to higher layers of inflection, and so the “smaller” domain of case assignment
that triggers ergative alignment is created.
The problem with this position is that meaning and form don’t always match in
the expected ways: on the one hand, we have that the French and Italian “perfects”,
for lack of a commonly used simple past form in the verbal paradigms of these lan-
guages, have taken up the functions of perfective past. On the other hand, simple
perfective tenses can mimic the semantics of perfects with the help of adverbs such
as already. Perfects and perfectives seem to be in a tug of war to divide up semantic
space, often overlapping. This “pragmatic residue” seems to be what functional and
typological approaches to aspect splits try to capture through implicational hierar-
chies, without taking note, however, of the possibility that there exist “prototypical”
structures for perfects and perfectives, in which the emergence of ergativity is deter-
mined structurally, rather than by discourse function.

Incidentally, the discussion of Mẽbengokre nominalizations offers an unexpected

answer to the problem of the opposition between perfects and perfectives, which could

140
be paraphrased through the following question: why should languages have recourse
to two distinct tense forms that have often overlapping truth conditions (i.e., the
existential perfect and the simple past)? The answer I propose is the following: such
a distinction exists precisely because languages can exploit the categorial distinction
between nouns and verbs to produce nuances in aspectual interpretation.

141
142
Bibliography

Abney, Steven Paul. 1987. The English noun phrase in its sentential aspect. Doctoral
Dissertation, Massachusetts Institute of Technology. Available through MITWPL.

Alexiadou, Artemis. 2001. Functional structure in nominals: nominalization and

ergativity. Amsterdam/Philadelphia: John Benjamins.

Alves, Flávia de Castro. 2004. O Timbira falado pelos Canela Apãnjekrá. Doctoral
Dissertation, State University of Campinas, SP, Brasil.

Arregui, Ana, and Lisa Matthewson. 2001. A cross-linguistic perspective on the

expression of manner. In Proceedings of the 36th ICSNL, 1–24. UBCWPL.

Basilico, David. 1996. Head positions and internally-headed relative clauses. Language
72:498–532.

Benveniste, Émile. 1971. The linguistic functions of “to be” and “to have”. In Problems
in general linguistics. University of Miami Press.

Borges, Marília. 1995. Aspectos da morfossintaxe do sintagma nominal na língua

Kayapó. Master’s thesis, Universidade de Brasília.

Bresnan, J., and Sam A. Mchombo. 1987. Topic, pronoun, and agreement in
Chicheŵa. Language 63:741–782.

Burzio, Luigi. 1986. Italian syntax . Dordrecht: Reidel.

Callow, John. 1962. The Apinayé language. Doctoral Dissertation, London University.

143
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from
building 20: essays in linguistics in honor of Sylvain Bromberger , ed. K. Hale and
S. J. Keyser. Cambridge, MA: MIT Press.

Chung, Sandra, and William Ladusaw. 2004. Restriction and saturation. Cambridge,
MA: MIT Press.

Cinque, Guglielmo. 1990. Types of A′ -dependencies. Cambridge, MA: MIT Press.

Cole, Peter. 1987. The structure of internally headed relative clauses. Natural Lan-
guage and Linguistic Theory 5:277–302.

Comrie, Bernard. 1978. Ergativity. In Syntactic typology: studies in the phenomenol-

ogy of languages, ed. W. P. Lehmann, chapter 7, 329–394. Austin: University of
Texas Press.

Corbett, Greville. 2006. Agreement. Cambridge University Press.

Cuervo, Cristina. 2003. Datives at large. Doctoral Dissertation, Massachusetts Insti-

tute of Technology.

Culy, Christopher. 1992. The syntax and semantics of internally headed relative
clauses. Doctoral Dissertation, Stanford University.

Cusic, David Dowell. 1981. Verbal plurality and aspect. Doctoral Dissertation, Stan-
ford University.

D’Angelis, Wilmar da Rocha. 2004. Concordância verbal de número em Kaingáng:

algumas pistas. LIAMES 4:71–81.

Davidson, Donald. 1967. The logical form of action sentences. In The logic of decision
and action, ed. N. Rescher. Pittsburgh: University of Pittsburgh Press. Reprinted
in P. Ludlow (ed.), Readings in the philosophy of language.

Davis, Henry. 2005. On the syntax and semantics of negation in Salish. International
Journal of American Linguistics 71:1–55.

144
Dixon, Robert M. W. 1994. Ergativity. Cambridge University Press.

Dixon, Robert M. W. 2004. The Jarawara language of southern Amazonia. Oxford

University Press.

Dourado, Luciana. 2001. Aspectos morfossintáticos da língua Panará (Jê). Doctoral

Dissertation, IEL/Unicamp, Campinas.

Fernald, Theodore. 2000. Predicates and temporal arguments. New York: Oxford
University Press.

Ferreira, Marcelo. 2005. Event quantification and plurality. Doctoral Dissertation,

MIT.

von Fintel, Kai, and Sabine Iatridou. 2005. Since since. MIT ms., March 2005.

Fraga, Carolina. 2006. Las preposiciones y la morfología distribuida. Presented at III

South American Summer School in Formal Linguistics.

Freeze, Ray. 1992. Existentials and other locatives. Language 68:553–595.

Gildea, Spike. 1992. Comparative cariban morphosyntax: on the genesis of ergativity

in independent clauses. Doctoral Dissertation, University of Oregon.

Gildea, Spike. 1998. On reconstructing grammar: comparative cariban morphosyntax .

New York and Oxford: Oxford University Press.

Giorgi, Alessandra, and Fabio Pianesi. 1998. Tense and aspect: from semantics to
morphosyntax . Oxford University Press.

Grimshaw, Jane. 1990. Argument structure. Cambridge, MA: MIT Press.

Hale, Ken, and Samuel J. Keyser. 1993. On argument structure and the lexical
expression of syntactic relations. In The view from building 20: essays in linguistics
in honor of Sylvain Bromberger , ed. Ken Hale and S. J. Keyser, 53–109. Cambridge,
USA: MIT Press.

145
Hale, Kenneth. 1973. Person marking in Walbiri. In A Festschrift for Morris Halle,
ed. Stephen R. Anderson and Paul Kiparsky. Holt, Rinehart and Winston.

Heim, Irene. 1982. The semantics of definite and indefinite noun phrases. Doctoral
Dissertation, University of Massachusetts, Amherst.

Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry 30:69–96.

Hunt, Katharine. 1993. Clause structure, agreement and case in Gitksan. Doctoral
Dissertation, University of British Columbia.

Iatridou, Sabine, Elena Anagnostopoulou, and Roumyana Izvorski. 2001. Observa-

tions about the form and meaning of the perfect. In Ken Hale: a life in language,
ed. Michael Kenstowicz, 189–238. Cambridge, MA: MIT Press.

Jackson, Eric. 2005. Resultatives, derived statives, and lexical semantic structure.
Doctoral Dissertation, UCLA.

Jelinek, Eloise. 1984. Empty categories, case and configurationality. Natural Language
and Linguistic Theory 2:39–76.

Jelinek, Eloise. 1995. Quantification in straits salish. In Quantification in natural

language, ed. Emmon Bach, Eloise Jelinek, Angelika Kratzer, and Barbara Partee.
Kluwer.

Johns, Alana. 1992. Deriving ergativity. Linguistic Inquiry 27:57–87.

Katz, Graham. 2003. On the stativity of the English perfect. In Perfect explorations,
ed. Artemis Alexiadou, Monika Rathert, and Arnim von Stechow, 205–234. Berlin:
Mouton de Gruyter.

Klein, Wolfgang. 1992. The present perfect puzzle. Language 68:525–551.

Klein, Wolfgang. 1994. Time in language. London, New York: Routledge.

Klimov, Georgij A. 1974. On the character of languages of active typology. Linguistics

131:11–25.

146
Koptjevskaja-Tamm, Maria. 1993. Nominalizations. London: Routledge.

Kratzer, Angelika. 1996. Severing the external argument from its verb. In Phrase
structure and the lexicon, ed. J. Rooryck and L. Zaring, 109–137. Dordrecht:
Kluwer.

Kratzer, Angelika. 2000. Building statives. In Proceedins of the Berkeley Linguistic

Society, volume 26. Available at semanticsarchive.net.

Laka, Itziar. 2006. Deriving split ergativity in the progressive: the case of Basque. In
Ergativity: emerging issues, ed. Alana Johns, Diane Massam, and Juvénal Ndayi-
ragije, 173–196. Dordrecht: Springer.

Larson, Richard. 1982. A note on the interpretation of adjoined relative clauses.

Linguistics and philosophy 4:473–482.

Lasersohn, Peter. 2005. Plurality, conjunction, and events. Dordrecht: Kluwer.

Lehmann, Christian. 1998. Possession in Yucatec Maya. München: Lincom Europa.

Mahajan, Anoop. 1997. Universal grammar and the typology of ergative languages.
In Studies on universal grammar and typological variation, ed. A. Alexiadou and
T. Alan Hall, 35–57. Amsterdam: Johns Benjamins.

Marantz, Alec. 1984. On the nature of grammatical relations. Cambridge, MA: MIT
Press.

Marantz, Alec. 1991. Case and licensing. In Proceedings of the eighth Eastern States
Conference in Linguistics, 234–253.

Marantz, Alec. 1997. No escape from syntax: don’t try morphological analysis in the
privacy of your own lexicon. In Proceedings of the 21st annual Penn Linguistics
Colloquium, ed. Alexis Dimitriadis et al., 201–225. Philadelphia: UPenn WPL.

McCoard, Robert W. 1978. The English perfect: tense choice and pragmatic infer-
ences. Amsterdam: North-Holland.

147
Merchant, Jason. 2006. Polyvalent case, geometric hierarchies, and split ergativ-
ity. In Proceedings of the 42nd annual meeting of the Chicago Linguistics Society,
ed. Jackie Bunting, Sapna Desai, Robert Peachley, Chris Straughn, and Zuzana
Tomkova.

Mithun, Marianne. 1991. Active/agentive case marking and its motivations. Interna-
tional Journal of American Linguistics 67:510–546.

Nash, Lea. 1995. Portée argumentale et marquage casuel dans les langues SOV et
dans les langues ergatives: l’exemple du géorgien. Doctoral Dissertation, Université
de Paris VIII.

de Oliveira, Christiane Cunha. 1998. Some outcomes of the grammaticalization of

the verb o ‘do’ in Apinajé. In Santa Barbara Papers in Linguistics 8 .

de Oliveira, Christiane Cunha. 2003. Lexical categories and the status of descriptives
in Apinajé. International Journal of American Linguistics 69:243–274.

de Oliveira, Christiane Cunha. 2005. The language of the Apinajé people of Central
Brazil. Doctoral Dissertation, University of Oregon.

Pancheva, Roumyana, and Arnim von Stechow. 2004. On the present perfect puzzle.
In Proceedings of NELS 34 .

Parsons, Terence. 1990. Events in the semantics of English. Cambridge, MA: MIT
Press.

Patel, Pritty. 2007. Split ergativity and the nature of agreement. Master’s thesis,
University College London.

Pesetsky, David. 1995. Zero syntax . MIT Press.

Polinsky, Maria. 2007. Complementation without complement clauses. Handout of

talk delivered at “Unusual events fest”, Stanford University., January 2007.

Polinsky, Maria, and Eric Potsdam. 2002. Backward control. Linguistic Inquiry
33:24–282.

148
Popjes, Jack, and Jo Popjes. 1986. Canela-Krahô. In Handbook of Amazonian lan-
guages, ed. D. C. Deryshire and G. Pullum, volume 1, 128–199. Berlin: Mounton
de Gruyter.

Queixalós, Francisco. 1998. Nom, verbe et prédicat en Sikuani (Colombie). Paris:

Peeters.

Reis Silva, Maria Amélia. 2001. Pronomes, ordem e ergatividade em Mẽbengokre.

Master’s thesis, State University of Campinas, SP, Brazil.

Reis Silva, Maria Amélia. 2006. Aspect and split ergativity in Mẽbengokre. Talk
delivered at NWLC 22, Simon Fraser University, February 2006.

Reis Silva, Maria Amélia. in progress. (on Mẽbengokre argument structure). UBC
ms.

Reis Silva, Maria Amélia, and Andrés Pablo Salanova. 2000. Verbo y ergatividad
escindida en Mẽbengokre. In Indigenous languages of lowland South America, ed.
Hein van der Voort and Simon van de Kerke. Leiden, Netherlands: Research School
of Asian, African, and Amerindian Studies (CNWS), Leiden University.

Reuland, Eric, ed. 2000. Arguments and case: explaining Burzio’s generalization.
Amsterdam/Philadelphia: John Benjamins.

Rodrigues, Aryon. 1999. Macro-Jê. In Amazonian languages, ed. R. M. W. Dixon

and A. Aikhenvald. Cambridge University Press.

Salanova, Andrés Pablo. 2001. A nasalidade em Mẽbengokre e Apinayé: o limite do

vozeamento soante. Master’s thesis, Universidade Estadual de Campinas.

Salanova, Andrés Pablo. 2004. Subtractive truncation in Mẽbengokre. MIT ms.

Santos, Ludoviko dos. 1997. Descrição da morfossintaxe da língua Suyá/Kı̃sêdjê (Jê).

Doctoral Dissertation, Universidade Federal de Santa Catarina, Florianópolis.

Shell, Olive. 1952. Grammatical outline of Krahô (after Buell Quain). International
Journal of American Linguistics 18:115–29.

149
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Grammatical cate-
gories in australian languages, ed. R.M.W. Dixon, 112–171. Canberra: Australian
Institute of Aboriginal Studies.

Smith, Carlota. 1997. The parameter of aspect. Kluwer, 2nd edition.

Thomson, Ruth. 1974. Contrafactuals in Kayapó. Arquivo Lingüístico 059. Brasilia:

SIL.

Urban, Greg. 1985. Ergativity and accusativity in Shokleng (Gê). International

Journal of American Linguistics 51:164–87.

Valenzuela, Pilar. 2000. Ergatividad escindida en wariapano, yaminawa y shipibo-

konibo. In Indigenous languages of lowland South America, ed. Hein van der Voort
and Simon van de Kerke. Research School of Asian, African, and Amerindian Stud-
ies (CNWS), Leiden University.

Velázquez-Castillo, Maura. 1996. The grammar of possession. Amster-

dam/Philadelphia: John Benjamins.

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

Vergnaud, Jean-Roger. 1977. Letter to Noam Chomsky and Howard Lasnik. Available
at lingBuzz/000461, April 1977.

Verhaar, John W. M., ed. 1968. The verb ‘be’ and its synonyms. Foundations of
language, supplementary series. Dordrecht: Reidel.

Vieira, Márcia Dámaso. 2001. A natureza transitiva das sentenças possessivas em

Mbyá-Guarani. In Des noms et des verbes en tupi-guarani: état de la question, ed.
F. Queixalós, 67–85. Lincom Europa.

Wiesemann, Ursula. 1972. Die phonologische und grammatische Struktur der

Kaingang-Sprache. Number 90 in Janua Linguarum, Series Practica. The Hague:
Mouton.

150
Wiesemann, Ursula. 1986. Aspect and mood as a nine-cell matrix. In Language in
global perspective, ed. Benjamin Elson, 471–506. Dallas: SIL.

Williamson, Janis S. 1987. An indefiniteness restriction for relative clauses in Lakhota.

In The representation of (in)definiteness, ed. Eric Reuland and Alice ter Meulen,
168–190. Cambridge, MA: MIT Press.

Zucchi, Alessandro. 1993. The language of propositions and events. Dordrecht:

Kluwer.

151