(Potsdam Linguistic Investigations _ Potsdamer Linguistische Untersuchungen _ Recherches Linguistiques à Potsdam 6) Peter Kosta, Lilia Schürcks-Formalization of Grammar in Slavic Languages_ Contributi

6 Potsdam Linguistic Investigations
This book assembles the contributions of the Eighth European Conference on

Formal Description of Slavic Languages (FDSL VIII) which took place from 2 nd
Potsdamer Linguistische Untersuchungen
to 5 th December 2009 at the University of Potsdam. The concern was to bring
together excellent experienced but also young scholars who work in the field of Recherches Linguistiques Potsdam
formal description of Slavic languages. Besides that two workshops on typology
of Slavic languages and on the structure of DP/NP in Slavic were organized.
Edited by Peter Kosta
Herausgegeben von Gerda Haler
Edit par Lilia Schrcks
Nadine Thielemann
Peter Kosta / Lilia Schrcks (eds.) Formalization of Grammar in Slavic Languages

Peter Kosta / Lilia Schrcks (eds.)
Formalization of Grammar
in Slavic Languages
Peter Kosta is Professor of Slavic Linguistics and Chair at the Slavic Department
at Potsdam University. Focus of research: Biolinguistics, Theory of Language,
Comparative Slavic Syntax and the Typology of Languages.
Lilia Schrcks, Ph.D., is since 2005 Associate Professor of Slavic linguistics
at the Institute of Slavic Studies, University of Potsdam. Her fields of research
include Slavic Syntax, Generative Syntax, Binding Theory, DP/NP, Identity
and Language.
PETERLANG
LANG
www.peterlang.de ISBN 978-3-631-61869-1

Internationaler Verlag der Wissenschaften
PLI 06-Kosta-Schurks-261869-HCA5-TP.indd 1 16.05.11 11:15:00 Uhr

Formalization of Grammar in Slavic Languages
Potsdam Linguistic Investigations
Recherches Linguistiques Potsdam
Edited by/Herausgegeben von/Edit par
Peter Kosta, Gerda Haler, Lilia Schrcks
and/und/et Nadine Thielemann
Vol./Bd. 6
PETERLANG
Frankfurt am MainBerlinBernBruxellesNew YorkOxfordWien
in Slavic Languages
Contributions of the Eighth International
Conference on Formal Description
of Slavic Languages FDSL VIII 2009
University of Potsdam, December 25, 2009
PETERLANG
Bibliographic Information published by the Deutsche
Nationalbibliothek
The Deutsche Nationalbibliothek lists this publication in the
Deutsche Nationalbibliografie; detailed bibliographic data is
available in the internet at http://dnb.d-nb.de.
Cover Design:
Olaf Gloeckler, Atelier Platen, Friedberg
ISBN 978-3-653-01040-4 (eBook)

ISSN 1862-524X
ISBN 978-3-631-61869-1
Peter Lang GmbH
Frankfurt am Main 2011
All rights reserved.
All parts of this publication are protected by copyright. Any
utilisation outside the strict limits of the copyright law, without
the permission of the publisher, is forbidden and liable to
prosecution. This applies in particular to reproductions,
translations, microfilming, and storage and processing in
electronic retrieval systems.
www.peterlang.de
Editorial
The series Potsdam Linguistic Investigations Potsdamer linguistische Untersuchungen
Recherches linguistiques Potsdam presents cutting-edge fundamental linguistics research
carried out at the University of Potsdam. Its major goal is to publish collection of articles,
conference proceedings and monographs on contemporary issues in the fields of Slavic lan-
guages and literature, Romance studies, English and American studies, German studies and
general linguistics. A special focus of study is the formal, functional and cognitive description
of language. The following areas of linguistics will seek to develop their own profile: phonol-
ogy, morphology, syntax (with special attention to generative syntax), semantics, pragmatics
(discourse analysis, speech act theory), sociolinguistics and language contact.
We do not set any theoretical, methodological or geographical boundaries. The series will
serve greatly as a forum for young scholars as well as other researchers working in various
linguistic fields and frameworks in Potsdam or elsewhere. The indication of Potsdam stands
for the crucial importance and outstanding quality of linguistics research at the University of
Potsdam. On the other hand, researchers from other Universities with proven excellence of
their work are most welcome to publish their doctoral dissertations, habilitation monographs
or conference proceedings in this series. The languages of publication are German, English
and French.
Editorial
Die Reihe Potsdam Linguistic Investigations Potsdamer linguistische Untersuchungen

Recherches linguistiques Potsdam ist eine Plattform fr linguistische Forschungen an der
Universitt Potsdam. Sie publiziert Sammelbnde und Monographien zu aktuellen Fragen der
zeitgenssischen internationalen Linguistik aus den Disziplinen Slavistik, Romanistik, Ang-
listik/Amerikanistik, Germanistik und Allgemeine Linguistik. Ein besonderer Schwerpunkt
liegt in der formalen, funktionalen und kognitiven Sprachbeschreibung. Darin bilden vor al-
lem die Bereiche Phonologie, Morphologie, Syntax (unter besonderer Bercksichtigung der
generativen Syntax), Semantik, Pragmatik (Diskursanalyse, Sprechhandlungstheorie, Ge-
schlechterforschung), Soziolinguistik und Sprachkontakt ihre eigenen Profile.
Wir wollen keine theoretischen, methodischen oder lokalen Grenzen setzen. Deshalb richtet
sich die Reihe sowohl an Nachwuchswissenschaftler als auch an Kollegen in Potsdam und
auerhalb Potsdams, die in verschiedenen Richtungen, Modellen und theoretischen Anstzen
der modernen Linguistik arbeiten. Der Hinweis auf den Standort Potsdam soll zum einen die
herausragende Bedeutung der linguistischen Forschung an dieser Universitt signalisieren.
Andererseits bedeutet die Nennung nicht, dass ausschlielich Forschungsergebnisse (ein-
schlielich Dissertationen, Habilitationen und Konferenzsammelbnde) verffentlicht werden,
die von Linguistinnen und Linguisten an der Universitt Potsdam stammen. Die drei Publika-
tionssprachen sind Deutsch, Englisch und Franzsisch.
Editorial
La serie Potsdam Linguistic Investigations Potsdamer linguistische Untersuchungen

Recherches linguistiques Potsdam reprsente une plate-forme dtudes linguistiques
luniversit de Potsdam. Elle publie des recueils et des monographies sur les questions ac-
tuelles de la linguistique contemporaine internationale dans les domaines des tudes des
langues slaves et romanes, anglaise et amricaine, des langues germaniques et de la linguis-
tique gnrale. Un point principal de recherche est pos sur la description formelle, fonction-
nelle et cognitive des ces langues. Dans ces domaines, on met laccent sur les profils de la
phonologie, morphologie, syntaxe (en tenant compte de la syntaxe gnrative), smantique,
pragmatique (lanalyse du discours, la thorie des actes de la parole, la recherche sur le
genre), la sociolinguistique o la linguistique de contact.
Nous ne voulons pas poser des limites dans la thorie, la mthode et le lieu de recherche.
Cest pourquoi la srie invite les jeunes chercheurs ainsi que les collgues de Potsdam et des
autres universits qui travaillent dans les secteurs de la linguistique moderne. Le titre de la s-
rie veut dmontrer dun cot lexcellente qualit de la recherche linguistique Potsdam sans
toutefois exclure les autres. Cela veut dire que nous acceptons et nous invitons les linguistes
de Potsdam et de lextrieur (inclus les thses de doctorat et dhabilitation et les actes de col-
loques). Les trois langues de publication sont : lallemand, langlais et le franais.
Table of Contents
Preface 9
I. Phonetics & Phonology 11
ALE BIAN: Structure of Syllables in Czech 13

MAGORZATA AVAR: Merger of the Place Contrast in the
Posterior Sibilants in Croatian 29
ONDEJ EVK: Features of Common Slavic Ablaut Alternation 43
II. Machine Translation 55

NATALYA KLYUEVA, PETR HOMOLA & ONDEJ BOJAR: Towards a Rule-Based
Machine Translation System between Czech and Russian 57
III. Semantics 65
ZHANNA GLUSHAN: On Animacy and Unaccusativity in Russian 67
ELENA GORISHNEVA: Inductive vs. Non-Inductive Generics in Russian and
Bulgarian 81
BEATA TRAWISKI: A Compositional Semantics for Comitative Constructions 93
IV. Syntax 109

ANDREI ANTONENKO: Binding by Phases: Principle A in Russian 111
STEVEN FRANKS: Dynamic Spell-Out as Interface Optimization 127
ELENA GORISHNEVA & ILSE ZIMMERMANN: Wh-Words and the Indefinite
Particle -to in Russian 165
HANA GRUET-SKRABALOVA: Czech questions with two wh-words 179
GAPER ILC: Optionality of the Genitive (of Negation) in Slovene 193
KATARZYNA JANIC: On development of antipassive function: what do Australian
and Slavonic languages have in common? 207
SLAVICA KOCHOVSKA: Dislocated Direct Objects in Macedonian 221
PETER KOSTA: Causatives and Anti-Causatives, Unaccusatives and Unergatives:
Or how big is the contribution of the lexicon to syntax? 235
ALEXANDER LETUCHIY: Reciprocity and similar meanings in Slavic languages
and SAE 297
NINA RADKEVICH: PPs of Different Sizes 315
TANYA SCOTT: Multiple Sluicing: A purely syntactic account 329
JOANNA MIECISKA: Wh-scope marking strategies in Polish 339
NATASHA TODOROVICH: How many da(s) are there in Serbian? 351
HANNU TOMMOLA: On Slavic and Finno-Ugric vs. Standard Average European 365
ROK AUCER: Some multiply prefixed verbs as covert serial verb construction 391
Preface
The University of Potsdam was after 1997, 2001 and 2005 from 2 to 5
December 2009 already for the fourth time the host of one of the most
internationally renowned Slavic linguistic conferences. The 8th European Con-
ference on Formal Description of Slavic languages has gathered more than 100
participants from 25 countries in Europe, Israel, Asia and the USA.
We are most delighted to present the contributions of the Conference Formal
Description of Slavic Languages (FDSL VIII) in this issue. After the formal
linguistics has experienced an enormous boom in recent decades worldwide, it is
now the Slavic languages which shift more and more into the focus of the formal
language description. The current conference, hosted by the University of Pots-
dam and the Institute for Slavic Studies, provided participants the opportunity to
talk about the latest state of research on formalization of grammars and thus to
develop new joint projects. The program included a total of 75 lectures in three
sections. During the three-day events, more than 100 participants from various
parts of the world took part in this international venue. They devoted interesting
models and methods of the so-called Principles-Parameters framework and other
formal frameworks. The focus has been put on the accuracy of this theory in the
field of phonology, morphology, machine translation, semantics and syntax of
Slavic and other languages. The participating scientists were here to include also
relevant directions of the computational and corpus linguistics.
Among the highlights of the meeting, five panels with keynote speakers
from the U.S., Italy, Norway and France and two workshops on language
typology and the structure of DP/NP in Slavic languages were conducted.
In addition to the traditional three sections on three mornings the organizers
of the FDSL VIII conference decided to conduct two workshops, which took
place on the two afternoons in succession. The first workshop Perspectives of
Language Typology: Slavic and Standard Average European was organized by
Anton Zimmerling and Peter Kosta. The basic idea of this workshop was to
check out whether Slavic languages as a group of closely related Indo-European
idioms constitute a separate (sub)-type and to establish the position of Slavic
languages within the construct known as Standard Average European. Thus,
typologically valid descriptions of Slavic languages may contribute to verifying
or reconsidering this construct.
The second workshop The Structure of NP/DP and its Implications for QP
was organized by Lilia Schrcks and was dedicated to the following goals: 1) to
inspire discussions whether the DP projection exists in articleless languages, 2)
to address recent approaches in the area of the structure of determiner phrases,
quantificational phrases, and the linking element D, and 3) to establish and to
evaluate the role of cross-linguistic variation.
10 Preface
While most articles from the three sections and from the first workshop on lan-
guage typology are published in this volume, the contributions of the second
workshop will be published in a special issue under the title The Structure of
NP and Beyond in the series Studies in Generative Grammar (eds. Lilia
Schrcks, Anastasia Giannakidou, Urtzi Etxeberria, and Peter Kosta,) in de
Gruyter Verlag Berlin New York (to appear).
We would like to thank the German Research Foundation (DFG) and the
University Society Potsdam for the generous financial aid! Finally, we would
like to thank the many helpers in the organization and implementation of FDSL
VIII Conference warmly. Without their selfless efforts, the conference would
not have become what it is. We would finally like to thank the two bands
Jazzmissia and Phoenixband and the theater group AP Chekhov for their musi-
cal and professional performance. Last but not least, it was Monika Kruschinski,
our secretary and good ghost of the institute, whom we owe a warm thank you
for her effective support before, during and after the conference.
The Editors Potsdam, April 2011

I.
Phonetics & Phonology

Structure of syllables in Czech
Ale Bian
In the following paper we will describe the syllable in present standard Czech.
In particular, we will focus on syllable-initial and syllable-final consonantal
combinations and the way their structure can be predicted from the model called
distributional unit, originally developed by JAN MULDER (1968, 1989). The sa-
me model was used by EL-SHAKFEH (1987) for English, and it was applied on
other languages, too. Something similar to the tree schemes we present for
Czech at the end of this paper was offered for English e.g. by WHORF (1940;
reproduced in GOLDSMITH 2009) and FUDGE (1969). They, however, used a dif-
ferent theoretical background.
1. Preliminaries
Every description must be based upon a certain theoretical framework. We have
chosen to follow the theory of functional phonology originally conceived by the
Prague School and NIKOLAI TRUBETZKOY (1939), later developed by ANDR
Martinet and his school (MARTINET 1991, AKAMATSU 1992) and further
expanded and formalized by Jan Mulder and his followers (MULDER 1968, 1989,
EL-SHAKFEH 1987).
Functional phonology views phonemes as unordered bundles of distinctive
features. Phonemes enter into mutual paradigmatic and syntagmatic relations,
oppositions and contrasts, respectively. Oppositions are valid if there is some
relevant paradigmatic difference between two or more phonological entities.
Similarly, contrasts are valid if there is some relevant syntagmatic difference
between phonological entities. However, both oppositions and contrasts may
become invalid if some differences become irrelevant, i.e. if they are canceled
under clearly defined circumstances. In that case we speak about neutralization
of an opposition and about neutralization of a contrast. The outcome of the for-
mer are archiphonemes; of the latter archi-positions. We will return to archi-
positions below.
In Czech, neutralization of voicing is operative. It means that the phono-
logical difference between voiceless and voiced consonants is canceled and not
valid in some contexts. In effect, the realizational voicing of consonants is
predictable from the context they appear in. And as wholly predictable features
cannot be part of phonological representation, it follows that phonological enti-
ties occurring in the context of neutralization of voicing cannot be either voice-
less or voiced. Consequently, they cannot be identified with either voiceless or
voiced phonemes, and the notion archiphoneme has to be introduced. Archi-
14 Ale Bian
phonemes may be viewed as phonemes in a subsystem representing two or more

phonemes of the overall system.
There is not enough space discuss all subtleties of the Czech phonematic
systems, so we will resort to mention just a handful of necessary points and di-
rect the readers elsewhere for details (BIAN 2008). The overall system of Czech
consonants is given in the table (1). The table does not contain so-called
affricates [ ] and [
], because these are analyzed as corresponding to
combinations of two consonants /Ts/ and /T/ (the capital letters standing for
archiphonemes). There are good reasons for it. For one thing, their structure
reflects that of attested combinations /Ps/, /P/, /Ks/ and /K/ (cf. pst to write,
penice wheat, xylofon xylophone, kiltovka peaked cap), which fact
makes the subsequent phonotactic analysis simpler.
(1) occlusive fricative

nasal
voiceless voiced voiceless voiced
labial p b f v m
alveolar t d s z n
palatal
velar k g x h
Phonemes outside the system of proportions: /j/ approximant, // spirant.
The system of archiphonemes resulting from the neutralization of voicing is in
the next table (2). They occur in the contexts summarized in the formulas in (3)
(i.e. in these contexts the voicing of consonants is wholly predictable).1
(2)
occlusive constrictive nasal
labial P F m
alveolar T S n
palatal n
velar K X
__________
1 _ = the context, Cv-less = any voiceless Consonant, Cv-ed = any voiced C, # = end of word.
Examples for (3a): /Stan/ [stan] stan tent, /FStT/ [fstat] vstt to get up, /Kin/
[krcn] ktin christening part (gen. pl.). Examples for (3b): /Kdo/ [do] kdo who,
/FSbuiT/ [vzbut] vzbudit to get up, /XbeT/ [rbt] hbet spine but /svjeT/
[svjt] svt world and /zvjeT/ [zvjt] zvd spy. Examples for (3c): /ploT/ [plot] plod
fruit, /peP/ [ppr] pep pepper. The stipulation before pause is necessary, because
if the form is followed by a word beginning with a voiced consonant, the archiphoneme
will likewise be realized as voiced, cf. /ploT#bil/ plod byl the fruit was realized as
[plodbl].
Structure of syllables in Czech 15
(3)
(a) _Cv-less, _Cv-less archiphoneme always realized as voiceless
(b) _Cv-ed, _Cv-ed except for /v/ archiphoneme always realized as
voiced
(c) _#, _# archiphoneme always realized voiceless before pause
The vowels of Czech are the following ones; we do not give here any table, be-
cause the constituency of vowels as to their distinctive features is not imme-
diately relevant for our paper. Short vowels: /i/, /e/, /a/, /u/, /o/; long vowels: //,
//, //, //, //; diphthongal vowels: //, //, //. It should be noted that
diphthongal vowels are here viewed as single phonemes (there are again good
reasons for it, see BIAN 2008); they are realized as [u], [] and [],
respectively. Vowels are phonemes only occurring in the syllable nucleus
whereas consonants are phonemes occurring only in syllable margins, i.e. in
non-nuclear contexts. In addition, Czech has a third class of phonemes, so-called
semiconsonants. Semiconsonants are phonemes that can occur both in the
nuclear as non-nuclear context. Czech has two such phonemes: /r/ and /l/, cf.
/prST/ prst finger /prase/ prase pig. For the sake of simplicity, we will
include /r/ and /l/ under the term vowels if they occur in the nuclear context,
and under the term consonant if they occur in the non-nuclear context.
2. Distributional unit
Now that we have the phonemes of Czech, we can turn our attention to the main
topic of our discussion, which is the structure of syllables in Czech. However,
because the space allotted to this paper is limited, we cannot describe the exact
procedure and the way we have arrived at our model. For that see BIAN (ms.).
Distributional unit is meant to be a model of phonotactic distribution of pho-
nemes. That is to say, in this particular case, it is meant to be a model for syl-
lables in Czech, as it accounts for the distribution and combinations of pho-
nemes within syllables in this language. It is defined as a bundle of positions. A
position may be viewed as a syntagmatic slot where a phoneme can occur and
where it can be replaced here by other phonemes. A position can also be empty.
The set of phonemes capable of occurring in a given position forms a paradig-
matic position class.
One of the positions is nuclear and it is that upon which other positions are
functionally dependent. This position can never be empty, because phonemes
occurring there function as identity elements of syllables without which they
could not be well-formed.
The distributional unit in Czech has ten positions. One of them is nuclear.
Five positions are pre-nuclear, that is, they appear before the nuclear position.
Four positions are post-nuclear. The five pre-nuclear positions correspond to the
maximum number of consonants capable of occurring at the beginning of a
16 Ale Bian
syllable in Czech. Similarly, the four post-nuclear positions correspond to the

maximum number of consonants capable of occurring at the end of a syllable.
This is deduced from the forms /FSkvjeT/ [t] vzkvt prosperity and
/borT/ [
] bor borsch. The way they are mapped onto the distributio-
nal unit is shown in (4).
(4)
F S k v j e

pre2 pre1 e3 e2 e1 n i1 i2 i3 i4

o r T
It must be stressed right away that the notion position used here differs from
traditional conceptions which generally view it as a place of possible occurrence
of an object relative to another object. In the case of distributional unit, however,
positions are not viewed as relative placements but as absolute and constant
ones. This is to say that the number of positions in the distributional unit is given
once and for all, and what actually changes is only the way positions are filled
with phonemes or whether they are empty. This can be conveniently illustrated
on the syllables /ta/, /va/ and /ma/. The phonemes /t/, /v/ and /m/ occur in diffe-
rent positions, even though they are the first phonemes of those syllables. It is
because their phonotactic properties are different: They show dissimilar ways of
how they combine with other phonemes, e.g. as to how many consonants can
precede or follow them. The reason why it is so will hopefully become obvious
in the due course. See (5) for the way they are mapped onto the distributional
unit. The zeros represent positions not filled with a phoneme.
(5)
pre2 pre1 e3 e2 e1 n i1 i2 i3 i4 syllable
t a ta
v a va
m a ma
Every position may be filled with a phoneme from a certain class. These classes
together with the structure of the distributional unit represent an underlying mo-
del for the phonotactic structure of Czech, that is, for all syllables in Czech. The
model is given in the table (6). The positions marked with e are explosive posi-
tions or pre-nuclear positions. The positions marked as pre1 and pre2 are so-
called pre-explosive positions; they are filled with a phoneme only if at least of
the previous pre-nuclear positions has already been filled with a phonemes.
Otherwise, they are empty. The positions marked with i are implosive or post-
nuclear positions. Finally, n stands for the nuclear position.
The capital /M/ given in the positions e2 and pre2 is the nasal
archiphoneme. It is a result of neutralization of the place of articulation for na-
sals, as that becomes irrelevant in certain contexts. Let us say it in other words.
The only nasal that can occur before a consonant or semiconsonant in the pre-
nuclear context is the bilabial nasal. However, as the nasal is always bilabial in
this situation, it is obvious the bilabialness cannot be a distinctive feature here,
because it is wholly predictable from the given context. Consequently, the nasal
archiphoneme /M/ must be introduced.
(6)
pre2 pre1 e3 e2 e1 n i1 i2 i3 i4
kgx rl PTK TS
PTK h v
T mn iea KT
FS td M mn
S rl ou
X sz rlj
j
rljM FX

pbf

mn
The figure is a result of our previous analysis, but as already mentioned, we will
not explain in detail how we arrived at it, because it would lengthen our paper
considerably. Moreover, we have done it elsewhere (BIAN forthcoming and
BIAN ms.). Yet the basic principle of the model can be summarized as follows:
a phoneme belongs to a certain position if and only if it possesses phonotactic
properties of the position in question.
Let us demonstrate it on the position which is labeled as e3. It is the third
pre-nuclear position. It can be occupied by velars /k/, /g/, /x/, /h/, alveolars /t/,
/d/, /s/, /z/ and by palatal fricatives // and //. They belong to this position
because they all share the same distributional characteristics. Or at least the po-
tential for them, as they need not always be actually manifested, for instance, in
the case of /g/ which has rather limited distribution due to historical reason. Two
major distributional characteristics are summarized in (7)2.
__________
2 Examples for (7a) are: /FTkT/ vtkt to weave into, /FSxo-/ vzchopit se to brace up,
/FSh-/ vzhru upward, /jSte/ jste you are (pl.), /FSdT/ vzdt (se) to give up,
/XTse/ chce (he) wants, /FzT/ vzt to take, /rTe-/ ren saying, /li/ li lies. Ex-
amples for (7b): /kvjeT/ kvt flower, /gla-/ glazura glaze, /xvjeT/ chvt (se) to trem-
ble, /hvjeST/ hvzd star (gen. pl.), /tM e/ tm dark (loc. sg.), /dvje/ dv two,
/svjeT/ svt world, /zMra-/ zmrazit to freeze, /rm/ rm wound, /lu-/ lut yel-
low. Examples for (7c): /FSkvjeT/ vzkvt properity or /PtroS/ ptros ostrich.
18 Ale Bian
(7)
(a) In the pre-nuclear context, the phonemes /k/, /g/, /x/, /h/, /t/, /d/, /s/,
/z/, // and // can be preceded by one or two phonemes belonging
to the positions pre1 and pre2 and by no others.3
(b) In the pre-nuclear context, the mentioned phonemes can be fol-
lowed by one or two phonemes from the positions e2 and e1 and
no other.
(c) The properties (a) and (b) can be realized at the same time.
In other words, the membership of these phonemes in the position e3 reflects
two facts. First, it holds that at the beginning of a syllable, all of these
consonants can appear right next to a vowel and can be, at the same time,
preceded by one or two other consonants. Second, it holds that at the beginning
of the syllable, one or two other consonants can appear between of any of these
consonants and the vowel.
A similar reasoning lies behind all other positions. Once again: a phoneme
belongs to some position if and only if it exhibits properties of that position. Po-
sitions are thus representations of phonotactic properties of phonemes. It is
possible that some phoneme belongs to more than one position. In that case its
distributional properties are the sum of the properties of the individual positions.
For example, the semiconsonants /r/ and /l/ belong to four positions and
therefore they have properties of all these positions.
3. Archi-positions
We should now explain why several phonemes in the table (6) extend over two
or more positions. It is because they occur in a so-called archi-position. An
archi-position is a position is a subsystem representing two or more positions in
the overall system. Its purpose is to account for special distribution of some
phonemes. It is a result of neutralization of a certain syntagmatic contrast. Its
practicability can be demonstrated on the palatal occlusives // and //. They
have distributional properties summarized in (8). Examples demonstrating them
are: /Kin/ ktin christening party (gen. pl.), /lSi-/ lstiv deceitful, /MS/
mst (se) (he) avenges, /FSe-/ vzdlan educated, /MSe/ mzd salary
(dat. sg.).
(8)
(a) In the pre-nuclear context, the phonemes // and // can be pre-
ceded by up to two phonemes belonging to the positions pre1 and
pre2 and by no others.
__________
3 This is important to realize, because it really holds that these phonemes cannot be pre-
ceded by any other consonant or consonantal combination.
(b) In the pre-nuclear context, the mentioned phonemes must be imme-

diately followed by a vowel, never by a consonant or semiconso-
nant.
If these properties are taken into account, we have to conclude that the differ-
ence between the positions e1, e2 and e3 is in fact canceled. They no longer
serve the purpose they were established for. For instance, the position e1 has
been established to account for the fact that some phonemes occurring right next
to a vowel can be preceded by up to four consonants. All phonemes belonging to
the position e1 share this capacity. However, as (8) shows, the phonemes //
and // are never capable of this. They can be preceded only by one or two
consonants, and hence the position e1 is actually of no use for them. And so
are the positions e2 and e3. Yet // and // have still to be assigned to some
position, and for that reason, an archi-position has to be postulated.
During the course of the analysis we have found necessary to postulate four
archi-positions for phonemes with special distribution. First of all, there is the
already mentioned archi-position for // and //; we label it E3 as it results
from neutralization of three explosive positions. Then there is another explosive
archi-position E2 where the phonemes /p/, /b/ and /f/ can occur. For the implo-
sive/post-nuclear context, we have postulated two archi-positions: I2 for the
archiphonemes /F/ and /X/ and I3 for the phonemes /m/, /n/, / /. These nasals
occur here in forms like /jilm/ jilm elm, /fajn/ fajn fine or /Ter / ern
blackness; they cannot be followed by any other phoneme in that case.
Finally, there is an archi-position we have marked as N; it is where the
semiconsonants /r/ and /l/ occur if they are syllabic. Simply said, syllabic /r/ and
/l/ can never be preceded by /m/, /n/, / /, /r/, /l/, /j/ or //, i.e. by any phoneme
belonging to the position e1. In fact, the only nasal they can be preceded with
is /M/ which, it is to be remembered, is an archiphoneme whose place of
articulation is predictable from the context (actually, it is always realized as bi-
labial nasal).
To conclude this section, we give here the table (9) where several examples
of syllable in Czech are given. They are all derivable from the distributional unit
as given in the table (6). In the next section we will outline how this is achieved.
20 Ale Bian
(9)
pre2 pre1 e3 e2 e1 n i1 i2 i3 i4 word
a a
n a na
u u
F S k v j e T vzkvt
b o r T bor
K t u ktu
K i n ktin
n e X nech
j i l m jilm
r t i rty
P S t r u X pstruh
v o j S K vojsk
p o m S T pomst
T p i T tpyt
h v j e S T hvzd
l S t n lstn
K i l T kilt
X T chtt
s M r smr
S p l spl!
T t v r T tvrt
4. Structure of consonantal combinations

Now that we have the distributional unit, we can redraw it to a graphical repre-
sentation in the schemes (10) and (11) for the pre-nuclear and post-nuclear con-
texts. They are given at the end of this paper.
The distributional unit may be viewed as a series of points where a choice
can be made. The points are the ten positions. At each of these points you can
choose from a limited set of phonemes which have previously been recognized
as belonging to the respective position (see (6) for the sets). In other words, the
distributional unit is a series of points where, at each of these points, you can
choose one phoneme which will appear there. It is also possible to leave the po-
sition empty. In fact, all positions except for the nuclear position and archi-
positions can be empty. The distributional unit can be likened above to an onion
with positions being like its peels clustered around the nucleus. The nuclear po-
sition is the central and essential part of the distributional unit to which non-
nuclear positions are glued one by one according to their degree of
peripherality. When deriving the structure of syllables, we thus start from the
nuclear position which is filled with a certain vowel. To this vowel several
phonemes may be added, either before or after it or both. A vowel may be

preceded by up to five phonemes and followed by up to four phonemes. We go
through the paths as indicated in the schemes. Most paths are branching which
means there are several directions to take. Every path has several points at which
we choose one phoneme from a certain class or leave that point empty.
Let us start with the pre-nuclear context first, with the scheme (10). Once
the nuclear position is filled with a vowel, there are two possibilities how it can
be expanded; this is indicated by the first branching of the scheme (viewed from
the top to bottom). Of course, the vowel may be not be expanded at all, because
onset-less and code-less syllables are possible in Czech (cf. a and). First, a
vowel can be preceded by a phoneme from the position e1, i.e. by /m/, /n/, / /,
/r/, /l/, // or /j/4, though the position can also be empty, as is indicated by the
next branching. Second, it may be expanded by // or // from the archi-position
E3 in which case some steps are skipped, as these phonemes can in turn be
preceded only the phonemes from the positions pre1 and pre2. We will return
to them. Note that archi-positions are never empty5 and thus there is no subordi-
nate branching for them like in the case of other positions for which there is
always a possibility to remain empty.
So the position e1 is either filled with a phoneme or left empty. The subse-
quent branching of the scheme may look a little bit complicated, but it is not. In
fact, our next step is either to fill the position e2 with a phoneme or leave it
empty or to fill the archi-position E2 with a phoneme in which case we again
skip some steps and get right next to the position pre1. The archi-position E2
can be filled with /p/, /b/ or /f/. These phonemes exhibit special combinability:
they cannot combine with the phonemes from the positions e2 and e3 and are
thus mutually exclusive with them. The position e2 can be filled with /v/ or
/M/ or it can, of course, be left empty. At this point we encounter a restriction on
the occurrence of a phoneme in e2, which explains the complicated branching:
although the phonemes /v/ and /M/ belong to this position, the latter can occur
there only if the position e1 is non-empty (or in case /r/ or /l/ occurs in N).
Simply said, /M/ cannot occur before a vowel because it is a product of
neutralization of the opposition between /m/, /n/ and / / before a consonant or a
semiconsonant.
It is now obvious that if the position e1 is filled with a phoneme and so is
e2, we get a combination of two phonemes. However, it should be noted that
not all combinations are attested: /Mm/ and /Mj/ are not, though the latter occurs
marginally in a surname Mjachk. Similarly unattested are combinations /bm/,
__________
4 However, not all vowels may be expanded by these phonemes: for instance, the combina-
tion / / is not possible.
5 Archi-positions were established for phonemes with special distribution; an empty archi-
position would be thus useless.
22 Ale Bian
/bn/, /b / or /pm/ in which case phonemes from the archi-position E2 are

combined with those of e1.
With the position e2 filled with a phoneme or left empty, we step up to
another position, the position e3, which may be filled with /t/, /d/, /s/, /z/, //,
//, /k/, /g/, /x/ or /h/ or again left empty. If the positions e2 and e1 are also
filled with a phoneme, we get a pre-nuclear combination of three consonants. Or
we can get a combination of two consonants if either e3 and e2, or e3 and
e1, or E2 and e1 are filled with a phoneme. Again, there are, however,
restrictions as to which combinations are attested and possible. Some
combinations are missing by an accident, i.e. no structural rule can be
introduced to explain their non-occurrence, whereas other combinations are
structurally impossible. We can, for instance, mention a constraint of the co-
occurrence of the consonants // and // or //: these phonemes cannot occur in
one and the same combination: although pre-nuclear /s-/, /z-/ are attested,
similar combinations /-/ or // are not. Likewise, pre-nuclear /St-/, /Sd-/ are
attested but not /t-/ or /d-/, i.e. the impossibility of the palatal fricatives to
co-occur with // is not limited to the immediate adjacency, but applies to
combinations as wholes.
Once we have gone through the positions e1, e2 and e3 or through e1
and the archi-position E2 instead of e2 and e3 or just though E3, we get at
a junction where all the paths unite and continue together toward the positions
pre1 and pre2. They may be viewed as expansions because there is one
important condition under which they are filled with a phoneme. They may be
filled with a phoneme only if at least of the positions e1, e2, e3 and/or their
archi-positions E2 and E3 have already been filled with a phoneme. In other
words, they are never filled with a phoneme if the positions e1, e2 and e3
are all empty. That is why they are expansions: they expand the pre-nuclear
phonemes or their combinations.
The position pre1 can be occupied by /T/, /S/, // or // and the position
pre2 by /P/, /T/, /K/, /F/, /S/, //, /X/, /r/, /l/, /j/ or /M/. These positions
represent contexts for neutralization of voicing. What is more, we can see that
both positions can be filled with /T/, /S/ and // but it holds that these
archiphonemes occur in pre2 only if the position pre1 is already filled;
otherwise they belong to pre1, cf. /FSkaS/ vzkaz message (/S/ in pre1)
/STkT/ stkt to weave together (/S/ in pre2 because /T/ is in pre1). There
are additional restrictions as to which the mentioned phonemes may occur in the
positions pre1 and pre2 but we cannot deal with them here. Some of these
constraints are directly derivable from the structure of the distributional unit.
Particularly interesting is the occurrence of // in pre1. This phoneme can also
occur in the position e1. It means that it has phonotactic properties similar to
two classes of phonemes: on the one hand, to sonants /r/, /l/, /m/, /n/, / / and
/j/ (cf. /PtroS/ ptros /PSte / pste (a kind of mushroom)), and on the other,
to /T/, /S/ and // (cf. /KiTse/ ktice fell (of hair) /Kini/ ktiny
christening party). Also noteworthy is the occurrence of /r/, /l/, /j/ and /M/: if
they occur there, they form so-called side-syllables (pobon slabiky), cf. /rTi/
rty lips, /lStn/ lstn deceitful.
With the two pre-explosive positions we have exhausted pre-nuclear
context, and by having filled each particular position with a phoneme or by
leaving it empty, we have generated one possible pre-nuclear consonantal
combination. We might have either left all the pre-nuclear position empty, in
which case there is of course no consonantal combination, or filled exactly one
position with a phoneme, in which case the nuclear vowel is preceded by one
phoneme. Or we could have filled two, three, four or five positions in which ca-
se we have gotten a combination of two, three, four or five consonants. In our
database we have compiled there are nearly 400 pre-nuclear consonantal
combinations each of which is attested in at least one Czech word (cf. BIAN
forthcoming). All of these combinations are describable with the scheme (10).
Similarly, the database lists over 80 post-nuclear consonantal combinations
which are describable with the help of the scheme (11) to which we now get.
Once the nuclear position has been filled with a vowel, it may be expanded
by a phoneme from the post-nuclear position i1, i.e. by /m/, /n/, / /, /r/, /l/ or
//, though that position may also be left empty. It is noteworthy that the mem-
bers of the position i1 are the same phonemes that belong to the position e1
except for //, which belongs to the post-nuclear position i3.
The branching after this first post-nuclear position might again look
complicated at the first sight. We have three possibilities. Having gone through
the position i1, we can, first, step to the position i2, second, to the archi-
position I2, or third, to the archi-position I3. However, only if the position
i1 is filled with a phoneme (i.e. not empty), the third step is feasible. This is
simply to say that the archi-position I3, where /m/, /n/ and / / occur, is filled
with a phoneme only if the position i1 is filled with a phoneme. On the other
hand, the first two steps are not dependent on i1 being filled or not. If we
choose the path of i2, we can fill it with /P/, /T/, /K/ or // or we can leave it
empty. We then proceed to i3. If we choose the path of the archi-position I2,
we skip the positions i2 and i3 and get right to the last post-nuclear position
i4. By various ways these positions are filled with a phoneme we get particular
post-nuclear consonantal combinations attested in Czech.
The archi-position I2 has been introduced for the archiphonemes /F/ and
/X/. It reflects the fact that these archiphonemes are mutually exclusive to the
phonemes belonging to i2 and i3 and never combine with them. It is most
obvious in the case of // which never combines with /F/ and /X/ in the post-
nuclear context. In fact, it always combine with an occlusive, cf. /peP/ pep
pepper, /-T/ from dovnit inside, /buT/ but sausage. In addition, it holds
that /F/ and /X/ can, in the post-nuclear context, be preceded by one phoneme
24 Ale Bian
from i1 or one phoneme from i4. Interesting is the fact that they are not
capable of both at the same time, i.e. they are either preceded or followed by a
phoneme. Cf. /harF/ harf harp (gen. pl.) u /naFT/ naft petrol (gen. pl.).
The position i3 can be occupied by the phonemes /T/, /S/ or //. We have
already seen that the position i2 could be filled with /T/. The conditions for its
occurrence in i3 are similar as those of /T/, /S/ and // in the pre-nuclear posi-
tions pre1 and pre2: /T/ occurs in i3 only if the position i2 has already
been filled with a phoneme, which means that it cannot thus be mapped onto
that position. The archiphoneme can also belong to the position i4, but once
again this is conditioned by whether the position i3 has already been filled
with a phoneme or not. If it was not, it belongs to i3, but if it was, it must
belong to i4. Cf. /peTS/ pec oven (/T/ in i2) /kumT/ kumt art (/T/ in
i3) /teKST/ text text (/T/ in i4).
The last post-nuclear position i4 can be filled with /K/, /T/, // or // or it
left empty. The circumstances allowing the occurrence of /T/ here have already
been mentioned. In fact, the only phoneme belonging to the position i4 not
dependent on whether the previous positions have been filled with a phoneme or
not is //6. It is because the remaining two archiphonemes /K/ and // belong
primarily to the position i2. They can appear in the position i4 only if one of
the previous positions has been already filled with a phoneme, in particular if
either the position i2 or i3 or the archi-position I2 has been filled with a
phoneme. Cf. /teKST/ text text (/K/ in i2) /vojSK/ vojsk army (gen. pl.)
(/K/ in i4).
By filling the position i4 with a phoneme or leaving it empty, we have
exhausted all possibilities for the post-nuclear context. As we already gone
through the scheme (10) for the pre-nuclear context and as we have already
filled the nuclear position with a vowel, we get a full-fledged syllable. That is to
say, the trees (10) and (11) are models for all well-formed syllables in present
standard Czech.7 It should be mentioned that there are several restrictions as to
which phonemes can combine with which in particular contexts, but these
restrictions could not have been incorporated into the trees in (10) and (11)
without making it a little clumsy and less clear. They are well mapped, though.
5. Conclusion
The distributional unit as we described here for Czech is a model of the phono-
tactics of Czech, that is to say, it is a model from which statements about the dis-
tribution of phonemes and their combinations can be derived. In short, it is a
__________
6 From this follows that if // occurs, it is always the last consonant of a syllable.
7 There is one exception, though: the syllables containing syllabic /r/ and /l/. They would
require a separate scheme for the pre-nuclear context involving the archi-position N,
but we have not discussed it here for the lack of space.
model for all well-formed syllables in Czech. Two particular types of statements
can be derived from the model as given in (6) and their tree-like visualizations in
(10) and (11). It is widely acknowledged that any phonotactic description should
be able to account for both statements (cf. GOLDSMITH 2009).
First, our model is capable of describing all consonantal combinations in
Czech that are attested in Czech. In our database we have nearly 400 pre-nuclear
consonantal combinations and over 80 post-nuclear combinations (see BIAN
forthcoming) are they are derivable from the distributional unit. Consequently,
any attested syllable in Czech is describable with these models.
Second, our model is capable of calculating (predicting) consonantal
combinations which are not actually attested in the data, but which are still
structurally possible. That is to say, combinations which are not used in any
phonological form of Czech words but which have a structure that they could
function as such, should there ever appear a word containing them. For example,
there is a combination /dvj/ occurring in the word dv but a structurally similar
combination /tvj/ is not attested. However, there is no reason why Czech could
not have a word which would begin with the combination /tvj/. There are other
combinations of this kind, and the distributional unit is capable of predicting
them.
We claim that all those consonantal combinations (and in turn syllables) that
are derivable from our model are well-formed in the Czech phonological system.
This is not to say that each and every syllable our model predicts is well-formed.
There are additional restrictions to be considered, some of them mentioned in
our paper, others not discussed to the lack of space. What we claim is this: a
combination that cannot be derived from the distributional unit is not well-
formed in Czech. So far we have not been able to refute this hypothesis.
References
AKAMATSU, T. (1992) Essentials of Functional Phonology. Peeters: Louvain-La-Neuve.
BIAN, A. (2008) Phonematics of Czech: An axiomatic-functionalist view. Masaryk Universi-
ty. Unpublished PhDr. thesis, available online:
<http://www.phil.muni.cz/linguistica/art/bican/bic-rigo.pdf>.
BIAN, A. forthcoming. Distribution and combinations of the Czech consonants. Zeitschrift
fr Slawistik.
BIAN, A. ms. Distributional unit: hypothesis and testing. Manuscript available online:
<http://www.phil.muni.cz/af/files/bican-du-12-04-10.pdf>.
EL-SHAKFEH, F. (1987) The phonematics, phonotactics and para-phonotactics of southern
Standard British English. University of St. Andrews. Unpublished PhD. thesis,
available online via the EThOS service: <http://ethos.bl.uk>.
FUDGE, E. C. (1969) Syllables. Journal of Linguistics 5.25386.
GOLDSMITH, J. (2009) The syllable. Manuscript, to be published in the second volume of The
Handbook of Phonological Theory. Available online
<http://hum.uchicago.edu/~jagoldsm//Papers/syllables.pdf>
MARTINET, A. (1991) lements de linguistique gnrale. Paris: Armand Colin.
26 Ale Bian
MULDER, J. (1968) Sets and Relations in Phonology. Oxford: Clarendon Press.

MULDER, J. (1989) Foundations of Axiomatic Linguistics. Berlin New York: Mouton de
Gruyter.
WHORF, B. L. (1940) Linguistics as an exact science. Technology Review 41.80-3.
TRUBETZKOY, N. S. (1939) Grundzge der Phonologie. Prague: Jednota eskoslovenskch
matematiku a fysik.
Ale Bian, Masaryk University, Department of Linguistics and Baltic Lan-

guages, Arna Novka 1/1, 60200 Brno, Czech Republic, bican@phil.muni.cz
n vowels
e1 E3
mn jrl

e2 M v E2
pbf
e3 t d s z k g x h
pre1 TS
pre2 PTKFSXrljM
(10)
28 Ale Bian
n vowels
i1 mn rlj
i2 PTK
I2 I3
FX
i3 TS
mn
i4 KT
(11)
Merger of the place contrast in the posterior sibilants in
Croatian
Magorzata E. avar
Abstract
The paper investigates the reasons behind the merger of the place contrast in posterior sibi-
lants in Croatian, i.e. of // and //, and on the other hand, of /d/ and //.1 It is argued that
systemic factors such as inventory density are not a sufficient trigger for merger. On the other
hand, acoustic variation in the realization of the categories may lead to merger. This approach
is formalized in terms of Functional Phonology (BOERSMA 1998).
1. Introduction
Whereas prescriptive grammars of standard Croatian list two series of posterior
affricates including post-alveolar /, d/ on the one hand, and prepalatal (al-
veopalatal) /, / on the other hand, see (1), it is clear that in every-day speech of
the substantial proportion of speakers the contrast between post-alveolars and
prepalatals simply does not exist.
(1) Standard Croatian coronal inventory
affricates affricates fricatives sonorants
voiceless voiced
post-alveolar d /
prepalatal/palatal - lj (IPA ), nj (IPA )
In this paper the reasons for the merger are scrutinized. One potential area of
explanation refers to the density of inventories (Dispersion Theory, cf. LILJENK-
RANTZ & LINDBLOM 1972, FLEMMING 1995/2002, PADGETT 2001, PADGETT &
TABAIN 2005, PAGETT & ZYGIS 2007, CAVAR 2004, etc). In the inventories with
a high number of contrasts, the same perceptual space is by necessity shared by
a bigger number of phonemes and particular sounds are perceptually closer (thus
more prone to misperception and merger) than in the systems with a smaller
amount of contrasts. This approach alone, however, is not sufficient to explain
the tendency to merger in Croatian, as it becomes clear when we compare the
__________
1 Throughout the paper, the Croatian orthographic symbols are used instead of IPA.
Whereas Croatian// and // are clearly IPA alveopalatal affricates /t"/ and /d$/ respec-
tively, the classification of // and /d/ is problematic even without considering the varia-
tion discussed in the paper, the symbols used in literature ranging between IPA /t/ and
/d%/ over less-specific American symbols /t/ and /d/ to IPA retroflex symbols /t'/ and
/d*/. Since the goal of the paper is the discussion of the category internal phonetic varia-
tion, the orthographic symbols are used as least qualifying.
30 Ma+gorzata E. avar
Croatian sibilant system with the inventory of sibilants in Polish (section 2). In
section 3, we will look at the possibility of the systemic avoidance of palatal and
prepalatal consonants as the driving force of merger. Then, in section 4, another,
traditional explanation of the merger in the standard Croatian will be discussed.
The merger might be connected with the dialectal background of the speakers
(cf. KARI 2009), in particular, the fact that in some Croatian dialects the con-
trast does not exist and this simplified system is transferred to the standard. It is
clear that the regional dialectal realizations of the respective sibilants have deci-
sive impact on the realizations of the contrast in the standard pronunciation. The
assumption made here is that speakers with the different dialectal background
learn the standard to some extent as if it were a foreign language, using the same
strategies as in the acquisition of a second language. In section 5, the predictions
of this assumption are tested in a formal analysis coined in terms of Functional
OT (BOERSMA 1995, ESCUDERO 2005).
2. Perceptual density as a trigger of merger
One can look at the phonological processes in language from the functional per-
spective (PASSY 1891, MARTINET 1955), that is, assume that in language two
contradictory drives are present to reach some sort of equilibrium between the
needs of the participants in communication: the sequence of speech sounds has
to be maximally easy to produce for speakers, and the same sequence of sounds
must be maximally easy to perceive and interpret for listeners. These two drives
lead potentially to contradictory processes in languages. For example, sounds
such as retroflexes, which are perceptually very distinct, with very distinct pat-
terns of acoustic cues, and thus optimal from the perspective of the listener, in-
volve also more complex articulatory side, that is, they are sub-optimal from the
perspective of the speaker. And the other way round, sounds which are produced
with less difficulty, such as palatoalveolars, which are produced with a less ex-
treme displacement of the tongue from the neutral position and with less com-
plex combination of gestures than retroflexes, they are optimal from the point of
view of the speaker but also, since e.g. the formant transition values are less dis-
tinct, they are also less attractive for the listener. If some two sounds are not
perceptually distinct enough, the two categories will merge. Modelling of vowel
inventories were done first by LILJENKRANTZ & LINDBLOM (1972), modelling of
the development of sibilant inventories were shown later by BOERSMA & HA-
MANN (2007), and for obstruent systems by BOERSMA & HAMANN (2010).
BOERSMA & HAMANN (2007) have demonstrated that given idealized condi-
tions with one perceptual dimension of the center of gravity systems with two
sibilant categories will always end up in the simulation with the unmarked ar-
ticulatorily-perceptually balanced /s-/ system, as in English. In their model, a
system with three sibilants will develop into a Polish-like system with retroflex-
like /'/, a prepalatal /"/ and dental /s/, with respective spectral means of roughly
Merger of the place contrast in the posterior sibilants in Croatian 31
3500Hz, 5500Hz and 7500Hz, relatively equally dispersed in the perceptual

space. After reaching this state, the inventory will not develop any further, hav-
ing reached the optimal dispersion between the members of the contrast. Inter-
estingly, in their simulations, /s/ and /sj/, sounds which are relatively close per-
ceptually, instead of merging, they develop in a three-way system into a more
distinct contrast between dental /s/ and prepalatal /"/, yielding an optimal and
stable system as in Polish (2007:25). This is what actually happened in the his-
tory of the Polish language (KLEMENSIEWICZ 1985, cf. PADGETT & ZYGIS 2007
and references therein). The question arises why the Polish three-way sibilant
system turns out stable whereas in Croatian affricates tend to merge to one post-
alveolar place of articulation.
3. On the track of the alternative explanation: Systemic avoidance
of palatal consonants
One answer might point to the obvious lack of the distinction between retroflex
and prepalatal fricatives in the Croatian system. Whereas the Polish system is
completely symmetric between fricatives and affricates, in Croatian this symme-
try does not exist, as shown in (2).
(2) Standard Croatian versus standard Polish
Croatian Polish2
post-alveolar affricates /d cz (= Cro. )/d\ (=Cro. d)
post-alveolar fricatives / sz(=Cro.)/\ (=Cro.)
prepalatal affricates / / d^ (=Cro. )
prepalatal fricatives `/^
sonorants lj/nj/j /{ (= Cro. nj) /j
One might stipulate that Croatian is in the process of elimination of the
(pre)palatal sounds, although this does not seem convincing in the light of the
fact that Croatian has more soft palatal sounds in the class of sonorants than
Polish. Consequently, one should rather exclude some systemic removal of a
particular contrast as a driving force of the merger.
4. On the track of the alternative explanation: variation in realiza-
tion
One striking difference between Polish and Croatian is the amount of variation
in the realization of the sibilant segments in the standard versions of the lan-
guage. In Polish impressively there is very little variation in the phonetic realiza-
tion of sibilants depending on the regional background of the speaker. The reali-
zations that are deviating from the norm are not considered standard and are ei-
ther simply absent from or stigmatized in the media and public discourse.
__________
2 Polish orthographic convention is used.
4.1. Polish
I am not aware of studies that would directly target the question of the regional
phonetic variation in sibilants in standard Polish but AVAR & HAMANN (2003)
recorded for the purpose of their analysis three speakers of Polish from geo-
graphically (and dialectally) very distinct areas, i.e. Warsaw, Czestochowa, and
Szczecin, and found them homogenous.
4.2. Croatian
For Croatian, the variation in the realization and the tendency to the merger of
the two affricate series is discussed in KARI (2009). He distinguishes three
registers of standard Croatian which differ in the realization of the contrast: (a)
the classical pronunciation with the clear distinction, which supposedly is no
longer in common use, (b) the received pronunciation as spoken in the media by
trained speakers such as actors and radio presenters, where the realization is
somewhere between the classical and the merger, (c) and the accepted pronun-
ciation which merges the two series, that is spoken by everybody else. Below
the spectra from KARI (2009) are reproduced. The method on how the spectra
were produced, is not described, thus, one cannot rely on the absolute frequency
ranges, however, one can compare the rough shape of the envelope and the rela-
tive concentration of energy. Thus, the classical type has two distinct shapes of
the envelope: // has one peak between 4 and 5 kHz, // has the concentration of
energy slightly above // with two peaks, one overlapping with the peak in ,
and the other at around 7 kHz. In the received pronunciation both series have
two peaks, the peaks located relatively close to the peaks of the other series,
with // slightly more compact in the middle ranges and // slightly more diffuse.
In the accepted pronunciation, the two series do not differ in realization.
(3) Spectra of the two fricative series from kari 2009:75

a. classical type b. received pronunciation c. accepted type
KARI (2009) assumes that in the general use, all speakers assume the accepted
type of articulation, irrespective of the dialectal background.
HAMANN & CAVAR (in progress) have also investigated the realization of
sibilants in standard Croatian: 18 educated speakers (university students or
young faculty members) have been recorded, each speaker has produced 30 re-
alizations of /, , d/ and 35 of //. In this study we found an overwhelming in-

tra- and inter-speaker variation in the realization of affricates, contrary to the
claim in KARI (2009). No overall results are available at the moment, so for
the purpose of this current paper four representative speakers have been selected
impressionistically, and for each speaker two realizations of each voiceless af-
fricate were chosen for the analysis3. For each sound, 5 FFT slices were created
with the help of Praat (BOERSMA & WEENINK 2009). The slices were taken at 5
ms intervals. The spectral peaks were read off the FFT slices directly. Then the
mean values for the peaks in 5 slices were recorded.
4.3. Results of the acoustic analysis of Croatian sibilants
Each of the selected speakers represented a different type of realization of the
contrast between hard and soft affricates. Speaker A produced two very distinct
categories, with soft series peaks at 4 and a much weaker at 8-9 kHz and, on the
other hand, the hard series peaks at 3 and a weak one 6-7 kHz. The second
speaker, speaker B, showed a huge intra-speaker variation in the realizations.
The realizations of the soft series tend though to have two peaks at 4 and 6 kHz,
and the hard series had the first peak only slightly lower at approximately 3.5
kHz and the second peak between 6-7 kHz, that is, more-less overlapping with
the soft series. Speaker C displayed a complete merger of the two categories,
both series realized with a broad energy plateau between 4 and 6 kHz. The result
of the merger is realized as a segment acoustically between the soft and hard se-
ries. Finally, speaker D also merges the two categories, yet the resulting sound is
realized differently than for speaker C, i.e. it sounds softer with the two peaks
located at 5 and (a weak one) at 8 kHz. The peak location is summarized for the
four speakers in (4).
(4) Location of the first and second peak for the analyzed speakers
a. in the four speakers
soft series hard series
Speaker A: 4 & 8-9 kHz 3 & 6-7 kHz
Speaker B: 4 & 6 kHz 3,5 & 6-7 kHz
Speaker C: 4 & 6 kHz
Speaker D: 5 & 8 kHz
__________
3 Voiced affricates were neglected in the study. It seems that their realizations are not nec-
essarily parallel to the realizations of voiceless affricates.
b. First (higher balks) and second peak (lower balks) across the four
speakers
As displayed in (4b), if the first peak is located between 4 and 5 kHz, the sound
might be a realization of both hard and soft series. Both series may be realized
with the concentration of energy around 6 kHz.
4.4. Discussion
If such realizations appear all in an official context of the university-internal in-
teraction, listeners must face a difficulty deciding what phoneme they actually
hear on a particular occasion. Since the contrast between the hard and soft series
of affricates does not have a high functional load in Croatian, it seems plausible
that on the basis of the acoustic information listeners can only roughly identify
the category, i.e. narrow the identification to posterior sibilants, and the actual
decision about what they have heard is made on the lexical level.
5. Formal functional analysis
Functional Optimality Theory (Functional OT) is an approach using the standard
OT mechanisms (PRINCE & SMOLENSKY 1993, MCCARTHY & PRINCE
1986/1996, MCCARTHY 2002), assuming the evaluation of plausible candidates
by ranked constraints and the elimination of the suboptimal candidate represen-
tations violating the constraints which are highest ranked. Functional OT differs
from the classical OT with respect to the assumption of the articulatory and
auditory grounding of constraints and structures, as well as with respect to the
overall model of the phonological grammar: phonology is in this approach a bi-
directional model for both production and perception. In the perception gram-
mar, we have three levels of representation (BOERSMA 1998, ESCUDERO 2005):
the non-discrete auditory input is run through the perception grammar with
ranked constraints, which produces discrete perceptual input, the phonological
representation. The perceptual input, in turn, is subject to the evaluation by the
recognition grammar with its ranked constraints, bringing the listener to the
lexical representation, see (5).
(5) Perception grammar in Functional Optimality Theory (based on ESCUDERO
2005:43)
Under this approach, a child acquiring a language first has to learn the categories
and the constraint rankings of the language, using two strategies: discovering the
sound categories from the distribution and repeating patterns in the acoustic sig-
nal, and then the fine-tuning of the system, adjusting the category boundaries,
using the gradually built lexical knowledge (Gradual Learning Algorithm:
BOERSMA & HAYES 2001, ESCUDERO & BOERSMA 2003). For L2 learning, it is
assumed that the same strategies, i.e. both distribution- and lexicon-driven learn-
ing are used (ESCUDERO 2005), with the assumption of the Full Transfer Hy-
pothesis (SCHWARZ & SPROUSE 1996) and Full Copying Hypothesis for lan-
guage perception (ESCUDERO & BOERSMA 2004).
My proposal to the Croatian data is that standard Croatian is learned as a
second dialect (D2), i.e. similarly like a second language, the only difference be-
ing the full access to lexical representations transferred from the first dialect
(D1) already at the onset of the learning. Thus, the starting point for the learner
of the standard dialect is the firm knowledge about the membership of some
acoustic event in a given phonological category. Then, the learner of the second
dialect will fine-tune the phonetic boundaries of the categories, i.e. the learner
will adjust the respective ranking of the perception grammar constraints to en-
compass maximal number of phonetic events into the appropriate categories. In
the following, the two scenarios for learning standard Croatian are discussed.
5.1. Speakers with the contrast in their native dialect
Speakers with a clear distinction between soft and hard posterior affricates in
their D1, like e.g. speaker A in section 4.3, will produce the typical // with the
fist peak at 3 kHz and the typical // with the first peak at 4 kHz4. Their D1 per-
__________
4 For the sake of clarity of exposition the concentration of energy in the higher frequency
ranges are neglected. In reality, the concentration of energy in higher frequencies may in-
fluence the perception of the energy distribution in the lower area (e.g. BLADDON 1986).
ception might look as in (6) and (7), where the constraint against categorizing
the input signal with the concentration of energy at 3 kHz as // must be higher
ranked than the constraint against categorizing such a signal as //, because the
signal with the peak at 3 kHz is more likely to be the realization of // than //.
The input signal with the peak at 3 kHz will be interpreted as // and this infor-
mation is then passed over to the recognition grammar that interprets the signal
as a part of the realization of some lexical entry.
(6)
Peak 1=3kHz 3 kHz not 4 kHz not 4 kHz not 3 kHz not
*!
*
On the other hand, the input with the peak at 4 kHz, given the same ranking, will
be interpreted as //, because, crucially, the constraint 4 kHz not is ranked
higher than 4 kHz not .
(7)
Peak 1=4 kHz 3 kHz not 4 kHz not 4 kHz not 3 kHz not
*
*!
At this point, if the user of this grammar is confronted with the data of the stan-
dard, where the lexical item containing // might be realized with the first peak
at 4 kHz, the mechanism of the Gradual Learning Algorithm (lexical-driven
learning) would cause the re-ranking of the perception grammar constraints and
broadening of the boundaries of category //.
(8) Listeners confronted with standard Croatian data
a. before re-ranking
Peak 1=4kHz intended as 4 kHz not 4 kHz not
*!
*
b. after re-ranking
Peak 1=4kHz intended as 4 kHz not 4 kHz not
) *
*!
Under normal circumstances in the first and second language acquisition the re-
ranking of constraints would result in the permanent change in the category
boundaries. In standard Croatian, however, the user of the grammar will be con-
fronted with as many realizations of // with the peak at 4 kHz as with the peak
at 3 kHz, driving an endless re-ranking of the constraints. One way out would be
the modification of the lexical representation in that the two categories merge.
This, however, does not happen because the standard Croatian system with two
categories is firmly coded by written grammars and taught at schools. The solu-
tion for the listener is the strategy as in the perception in noisy conditions: the
rough identification in terms of a voiceless posterior sibilant is made by the per-
ception grammar and the final decision is handed over to the recognition gram-
mar, compare the tableau in (9). Recognition grammar evaluates the fitting of
candidate lexical entries into semantic context. In the example below, it is more
likely that the input /plae/ (with non-distinct realization of the sibilant) that
appears in the context of lexeme spend is the realization of //plae// meaning
salaries (candidate b) rather than of lexeme //hlae// meaning trousers (can-
didate c) or s/he cries (candidate a), as illustrated in (9)5.
(9)
/plae/ *LEX plae *LEX hlae *LEX plae

Context=spend cry/spend FAITH trousers/
spend
salaries/
spend
(a) //plae// *!
cry
(b) //plae// *
salaries
(c) //hlae// *!
trousers
To sum up, the number of the perceptual categories for the speakers with the
first dialect background of two separate categories does not have to permanently
change when acquiring standard Croatian as a second dialect: listeners will con-
tinue perceiving two separate categories.
5.2. Speakers with no contrast in their native dialect
For the speakers who have just one posterior sibilant category in their native dia-
lect D1, when faced with standard D2 inputs, the task is potentially to
split/redistribute the native category into two separate categories. This corre-
sponds to ESCUDEROs (2005) SUBSET scenario in the learning of a foreign
language. Escudero gives an example of Spanish native speakers with just one
high front vowel category (/i/) learning Dutch with its two categories in the
same perceptual space (/i/ and /}/). It might be achieved through the interaction
of recognition grammar with the distribution-driven learning of categories (cate-
gory boundaries) and it should result in the change in the underlying representa-
__________
5 Additionally, in Croatian the agreement of morphosyntactic features of words will play a
role in the recognition. This aspect has not been taken into account.
tion of words. For example, minimal pairs differing only in the presence of /i/
versus /}/ will have for the learners in the early stages of learning Dutch the
same underlying representation, and only later two separate underlying represen-
tations will develop. However, in the case of second dialect learning, when
learners have clear underlying representations for all lexical items, the first step
will consist in stretching the native D1 category to encompass all kinds of in-
put from the standard D2, see tableau (10a). For comparison, the perception of
an input with a typical for D1 frequency peak at 4 kHz is shown in (10b).
(10) a. Perception of the D2 input
Peak 1=3 kHz PERCEIVE 3 kHz not 5 kHz not 4 kHz not
' *
-- *!
b. Perception of the D1 input
Peak 1=4kHz PERCEIVE 3kHz not 5kHz not 4 kHz not
' *
For the language user with D1 with no posterior sibilant contrast (just one cate-
gory typically realized with the first peak in the spectrum at around 4 kHz) no
changes in the perception grammar, or more precisely, in the ranking of percep-
tual constraints are necessary in the process of D2 learning. In other words, no
re-ranking of constraints is necessary in order to perceive D2 sibilant with a
slightly different energy distribution in the spectrum as a member of the D1
category. This is because of the relatively higher ranking of constraint PER-
CEIVE (cf. BOERSMA 1998: 163; ESCUDERO 2005:72 ff): even if the input has
some properties beyond the normal category borders, we want to categorize it.
Thus, in the lack of other, better-fitted categories, the input with the first peak of
energy at 3 kHz will be categorized as a member of a category whose members
usually have the energy peak in D1 at 4 kHz. Listeners will categorize inputs
from D2 with the peaks anywhere between 3 and 5 kHz as the member of their
D1 category, and this without any change in grammar, only with more variation
in the input.
The question arises whether the listener under this scenario can develop two
separate categories. This seems unlikely because of the inconsistency of the in-
put, as demonstrated in section 4.3. The fact that they do not develop two sepa-
rate categories in their perception of D2 is also not very problematic from the
point of view of communication because of the low semantic load of //-// (and
/d/-//) contrast in Croatian. The low type (and token) frequency of the contrast
hinders the distribution-driven learning mechanism; consequently, no changes in
grammar for D2 speakers are expected.
5.3. Standard as L1
Whereas for the speakers of standard Croatian using it as their second, non-
native dialect, the exposure to the varied input does not trigger change in either
constraint ranking of perceptual grammar or in the modification of the inventory
of categories, the situation for a native speaker of standard Croatian as their
D1/L1 is slightly different. On the basis of the acoustically inconsistent input,
they cannot in a natural way develop two separate categories. The process of
merger might be, however, slowed down or temporarily halted by sociolinguistic
factors represented by the normative tendencies and language policies. Yet, if
these are rendered less important, the natural consequence of the acoustic varia-
tion is the complete merger of the two categories.
6. Conclusions
In this paper the reasons for the merger of posterior sibilant categories in Croa-
tian are scrutinized; in particular, the question of acoustic variation versus struc-
tural factors such as the density of perceptual space are discussed. It is quite
unquestionable that the category density of sibilants is higher in standard Croa-
tian than for other European languages, which in itself might be a factor leading
to merger of the contrasting categories. However, structural factors obviously
are not sufficient to trigger merger, which can be concluded from the compari-
son of Croatian and Polish systems. The difference between Polish and Croatian
systems is in the absence versus presence of the variation in the realization of
the categories. In Croatian, due to varied dialectal background of speakers, the
listener is exposed to an immense acoustic variation between different realiza-
tions of the same category. Two scenarios have been investigated. For speakers
who in their native dialect have two separate categories, no new ranking of rele-
vant perceptual constraints in standard Croatian can be established, and the lexi-
cal decisions are made on the basis of semantic information rather than bottom-
up perception. On the other hand, for speakers who do not have a sibilant place
contrast in their native dialect, no such contrast can be learned from the data of
standard Croatian because the data is inconsistent, thus, no change in grammar
results either.
Acknowledgements
This research was supported by grant No. 269-2120920-0896 by the Croatian
Ministry of Science and Education. The analysis is based on the recordings done
in cooperation with Silke Hamann in spring 2008 and with Antonio Otari in
summer 2009: I would like to thank them here for their contribution to this pa-
per.
References
BLADDON, A. (1986) Phonetics for hearers. In: MCGREGOR, G. (ed.) Language for hearers.
Oxford: Pergamon Press.
BOERSMA, P. (1998) Functional Phonology. Formalizing the interaction between articulatory
and perceptual drives. The Hague: Holland Academic Graphics.
BOERSMA, P. & S. HAMANN (2007) The evolution of the auditory contrast. Rutgers Optimality
Archive 909.
BOERSMA, P. & S. HAMANN (2010) Simulating eternal optimization: Grimms Law. Lecture at
colloquium at the University of Zadar, 3.03.2010.
BOERSMA, P. & B. HAYES (2001) Empirical tests of the Gradual Learning Algorithm. Linguis-
tic Inquiry 32: 45-86.
BOERSMA, P. & D. WEENINK (2009) Praat: doing phonetics by computer (version 5.1.04)
[Computer program].
AVAR, M. E. (2004) Palatalization in Polish. An Interaction of Articulatory and Perceptual
Factors. University of Potsdam: PhD diss.
AVAR , M.E. & S. HAMANN (2003) Polish Velar and Coronal Palatalization its Perceptual
Background. In: KOSTA, P. et al. (eds.) Investigations into Formal Slavic Linguistics.
Contributions of the Fourth European Conference on Formal Description of Slavic
Languages (FDSL IV) held at Potsdam University, November 28-30, 2001. Frankfurt
am Main: Lang Verlag.
ESCUDERO, P. (2005) Linguistic Perception and Second Language Acquisition. The Hague:
Holland Academic Graphics.
ESCUDERO, P. & P. BOERSMA (2003) Modeling the perceptual development of the phonologi-
cal contrasts with OT and Gradual Learning Algorithm. In: ARUNACHALAM, S., E.
KEISER & WILLIAMS, A. (eds.) Proceedings of the 25th Annual Penn Linguistics Collo-
quium. Penn Working Papers in Linguistics 8. 71-85.
ESCUDERO, P. & P. BOERSMA, P (2004) Bridging the gap between l2 speech and perception re-
search and phonological theory. Studies in Second Language Acquisition 26. 551-585.
FLEMMING, E. (2002) Auditory representations in phonology. New York: Routledge.
HAMANN, S. & M. AVAR (in progress) Acoustic variation in posterior sibilants in Croatian.
KLEMENSIEWICZ, Z. (1985) Historia jzyka polskiego. Warszawa: Pa{stwowe Wydawnictwa
Naukowe.
LINLJENKRANTZ, J. & B. LINDBLOM (1972) Numerical simulation of vowel quality systems:
the role of perceptual constrast. Language 48. 839-862.
MARTINET, A. (1955) Economie des Changements Phontique. Berne: Francke.
MCCARTHY, J. (2002) A Thematic Guide to OT. Cambridge: Cambridge University Press.
MCCARTHY, J. & A. PRINCE (1986/1996) Prosodic Morphology. Tech. report in Rutgers Uni-
versity Center for Cognitive Science, New Brunswick, NJ. Report no RuCCS-TR-32.
PADGETT, J. (2001) Contrast dispersion and Russian palatalization. In: HUME, E. & K. JOHN-
SON (eds.) The role of speech perception in phonology. 187-218. San Diego,
CA:Academic Press.
PADGETT, J. & M. TABAIN (2005) Adaptive Dispersion Theory and Phonological Vowel Re-
duction in Russian. Phonetica 62: 14-54.
PADGETT; J. & M. ZYGIS (2007) The evolution of sibilants in Polish and Russian. Journal of
Slavic Linguistics 15(2).
PASSY, P. (1891) Etudes sur les changements phontiques et leurs characteures gnreaux.
Paris: Librairie Firmin Didot.
PRINCE, A. & P. SMOLENSKY (1993) Optimality Theory: Constraint Interaction in Generative
Grammar. Tech. report in Rutgers University Center for Cognitive Science, New
Brunswick, NJ. Report no RuCCS-TR-2.
SCHWARTZ, B. & R. SPROUSE (1996) L2 cognitive states and the Full Transfer/Full Access
model. Second Language Research 12. 40-72.
KARI (2009) Hrvatski izgovor. Zagreb: Globus.
Ma+gorzata E. avar, University of Zadar, Linguistics Department, Obala kralja

Petra Kreimira IV. 2, 23000 Zadar, Croatia, mcavar@me.com
Features of Common Slavic Ablaut Alternation
Ondej efk
1. Introduction
The issue of the my paper is, first, that the system of Old Church Slavonic ablaut
could be described with the use of the set of ablaut values and features derived
from them, and second, that such ablaut features could be taken for vectors;
third, that such features form a vector and a metric space in which all differences
between ablaut grades could be exactly stated.
The analytic approach to the features which is here presented is derived
from that of Marcus (MARCUS 1967). There are some similarities to the methods
of Dependency Phonology (see ANDERSON & EWEN 1987 or DURAND 1996) and
closely to the method of phonemic description by Hubey (HUBEY 1999), who
reworked a rather intuitive approach of DP using the algebraic form. The basic
idea of Hubeys revisited DP-method is to found and to define basic orthogonal
vectors and to describe the whole given system as a vector space. Third, our at-
tention is more focused on metric spaces than vector spaces, which is the shift of
focus, not a different approach at all.
The solutions of the method proposed here could be applied not only on the
OCS ablaut system, but is relevant for the reconstructed Common Slavic ablaut
system, too. The data from OCS are used purely as raw material for the
presentation of the method, although the method itself is not determined by the
data.
2. Ablaut grades
The OCS ablaut is such an alternation between morphs of one identified mor-
pheme (most often a root-morpheme) which indicates grammatical information.
Or in other words, the ablaut is the grammatical derivation of the morpheme.
The OCS ablaut is then a non-linear alternation, i.e. such an alternation not
triggered by the phonemic context of a given morph which is different from the
linear alternation such as palatalization of velars neighboring palatal vowels and
diphthongs.
(1): Ablaut grades of different verbal roots1:
r_k: rekD, rok
, -rkati, rci, -ricati
t_k: teet, tok
, -tkati, -taati, tci
g_n: enD, goniti, -gaati, g
nati
__________
1 All prefixes are separated. All grades are not ordered due to proportionality.
44 Ondej efk
sl_v: slovo, slaviti, (sl

ati2), slyati
b_r: berD, -bor
, brati, -birati
v_d: vedD, voditi, vs
m_r: mrti, moriti, -mariti, -mrD, -mirati

d_x: dux
, d
xnDti, dyxati
Each of the ablaut variants of the given morpheme is termed the ablaut grade.
All ablaut grades form the set of ablaut grades.
In this phase of analysis the set of ablaut grades is an unordered set, i.e.
there is no additional information on relations between different ablaut grades.
Such an unordered set is not a system, because any system requires an
organization, or better, the structure over it.
We will try to develop such a structure using sets of values of all ablaut gra-
des to postulate ablaut features.
An ablaut value is then such a property which could be related to an ablaut
grade and which makes the difference between grades in at least one example.
For all ablaut grades we need to define at least a minimal set of such values.
This is equivalent to the relation between phonemes of any phonemic system
and the sets of values of its phonemes, for both systems consist of a set of ele-
ments (phonemes, ablaut grades) and the organization itself (system of phonem-
ic oppositions, system of morphonemic oppositions). Other similarities will be
clear from the following lines.
3. Values of ablaut
Each ablaut grade has then its own set of such values, or in other words, any
given ablaut grade will be attached to the one and only one set of values (or bet-
ter there is a bijective3 relation between each ablaut grade and its set of va-
lues).
Any value of a given ablaut grade is here symbolized V{x} in general or in
the case of a concrete value simply as {value}.
The set of values of a given ablaut grade is symbolized as V{x} (i.e. V{x} =
v1{x} v2{x} vi{x}).
The set of values of the whole ablaut system is symbolized simply as V,
which is the set union of particular values of concrete ablaut grades, hence V =
V{a} V{b} V{z}i).
__________
2 Probably such a form is only a variant of slyati, but definitely of a different ablaut value.
3 Bijection is such a relation between sets A and B so that if every element of A is related
to exactly one element in B and if then every element in B is related to only exactly one
element of A.
Features of Common Slavic Ablaut Alternation 45
Any value could be classified with the help of three criteria, every one
expressing some properties of the given value in a language system. Each
criterion consists of a pair of possible incompatible properties.
The first criterion is homogeneity. Two values are homogeneous if they are
members of the same subset Vi, otherwise such values are heterogeneous
(MARCUS 1967: 46-47, KORTLANDT 1972: 57; cf. EFK 2008: 6, EFK 2009:
186-187).
(2a) The values {e-grade} and {o-grade} belong to the same subset of values,
hence they are homogeneous.
(2b) The values {reduced-grade} and {o-grade} do not belong to the same
subset of values, hence they are heterogeneous.
The second criterion is compatibility. Two values are compatible if they are
attached to the same ablaut grade {x} (members of the same set of ablaut values
V{x}). If such values are not members of the same set of ablaut values (they are
not attached to the one ablaut grade), they are incompatible values. It should be
mentioned that all compatible values are heterogeneous, but not the other way
round; thus, some heterogeneous values are not compatible (MARCUS 1967: 47-
48, KORTLANDT 1972: 57 cf. EFK 2008: 6, EFK 2009: 187).
(3a) The values {e-grade} and {reduced-grade} are attached to the same ablaut
grade; hence they are compatible.
(3b) The values {e-grade} and {o-grade} are not attached to the same ablaut
grade; hence they are not compatible.
The third criterion is contrastivity. Any values vi and vj are contrastive if there
are two ablaut grades such that V{x} V{y} = vi and V{y} V{x} = vj. In other
words, if replacement of one value of a given ablaut grade by another value
results in another ablaut grade, the values are contrastive. Otherwise both values
are incontrastive. All contrastive values are homogeneous and therefore
incompatible, but not the other way round (MARCUS 1967: 48-49, KORTLANDT
1972: 58 cf. EFK 2008: 6-7, EFK 2009: 187). Curiously, the ablaut system
of OCS is in fact structured in a way, so that any pair of homogeneous values is
necessary contrastive.
(4a) The values {e-grade} and {o-grade} are contrastive, because the absence of
the value {e-grade} necessarily means that the grade is {o-grade}.
(4b) The values {e-grade} and {reduced-grade} are not contrastive, because the
absence of one value does not necessarily mean that the second value is present.
Every value which is contrastive, homogeneous and incompatible we will term
the pertinent value (MARCUS, 1967: 46-47).
46 Ondej efk
For OCS (and CSl, too) we deal with the following set of ablaut pertinent
values: {e-grade}, {o-grade}, {reduced-grade}, {non-reduced-grade},
{lengthened-grade}, {non-lengthened-grade}.
The above mentioned ablaut grades of different roots could hence be arranged in
the following table4:
(5) eG oG G eG eG oG G e G
gubiti *g
nDti gybati
g_b 5
g
nuti
g_r erav
gorti *ar
6 garati g
rno7
vr_t vrm vratiti vrtit
b_d bljusti buditi b

dti
r_k rekD rok
rkati raiti8 rci ricati
t_k teet tok
tkati taati tci
g_n enD goniti gaati g
nati
slux
, slaviti sl
ati9 slyati
sl_v(s)
slovo
b_r berD bor
brati birati
v_d vedD voditi vs
m_r mrti moriti mariti mrD mirati

d_x dux
d
xnDti dyxati
Note: Marking of ablaut grades:
{reduced non-lengthened e-grade} (represented by , marked eG);
{reduced non-lengthened o-grade} (represented commonly by
, marke-
doG);
{non-reduced non-lengthened e-grade} (represented commonly by e,
marked eG);
__________
4 All prefixes are removed.
5 Only in Russian Church Slavonic, besides g
bnuti, see DERKSEN (2008: 197)
6 The Proto-Slavic form is reconstructed on Russian ar, Czech r, Slovak iar, P. ar,
Serbo-Croatian, Slovene r, Bulgarian ar, cf. MACHEK (1971: 722), VASMER (1953: I:
410), DERKSEN (2008: 554).
7 Only in Russian Church Slavonic, see DERKSEN (2008: 199).
8 Cf. LIV (4578).
9 Although sl
ati we consider as a variant of slyati, formally it is {reduced non-
lengthened o-grade} = oG.
{non-reduced non-lengthened o-grade} (represented commonly by o,

marked oG);
{non-reduced lengthened e-grade} (represented commonly by , marked
G);
{non-reduced lengthened o-grade} (represented commonly by a, marked
eG);
{reduced lengthened e-grade} (represented by i, marked G);
{reduced lengthened o-grade} (represented by y, marked eG).
Two mutually contrastive pertinent values form an ablaut feature. Hence, any
ablaut feature could have two values, one negative, the second marked posi-
tive. The values are attached arbitrarily (EFK 2008: 7, EFK 2009: 187).
Between both pertinent values of each given feature there is then a binary
opposition. For simplicity we will mark each positive pertinent value in the
features as 1, each negative pertinent value as 0, and hence each ablaut grade
could be expressed in a binary code.
When organizing the above enumerated pertinent values of OCS, we face
three ablaut features (AFS):
AF1: {non-reduced-grade} {reduced-grade}
AF2: {non-lengthened-grade} {lengthened-grade}
AF3: {e-grade} {o-grade}
As we can see, all three AFs are mutually independent, i.e. no one is dependent
on the other AF. Such relations between AFs can be expressed in the following
graph.
(6) Systems of ablaut features of OCS:
AF2
0
AF1
AF3
48 Ondej efk
In such a system there are basic vectors attached to basic ablaut grades: eG is in
the zero position, eG is the basic value on AF1 vector, G is basic on AF2
vector, oG is basic on AF3 vector.
4. Vector space
All three features then could be interpreted as vectors and the ablaut system as a
vector space. Because AFs are mutually independent, and their dot product is
zero, they form an orthogonal vector space.
For simplicity we will mark the AF1 vector as R, the AF2 vector as L and
the AF3 vector as O.
Any concrete ablaut grade V{x} has hence its unique set of values v1-v2-v3 of
the features of the same ordering (RLO). The commas will be omitted in the
following lines for simplicity. Values (and also features) for each ablaut grade
are then ordered and both sets of ablaut grades and the sets of values could then
be considered as an ordered sets due to the equivalence between both sets.
(7) AF vectors
R L O
{non-reduced non-lengthened e- = eG 0 0 0
grade}
{reduced lengthened e-grade} = G 0 1 0
{reduced lengthened e-grade} = G 1 1 0
Ablaut grades
{non-reduced non-lengthened e- = eG 1 0 0
grade}
{reduced non-lengthened o-grade} = oG 1 0 1
{reduced lengthened o-grade} = eG 1 1 1
{non-reduced lengthened o-grade} = eG 0 1 1
{non-reduced non-lengthened o- = oG 0 0 1
grade}
The AF vector space is also considered as unit vector, with the size of vector
equal to 1, for the distance between two values of one feature is equal to 1 (more
on distances see below).
The unit vectors, which are orthogonal, too, are termed orthonormal
vectors. Hence our vector space of ablaut grades is an orthonormal vector space.
Any ablaut grade (i.e. V{x}) could be then described as a linear combination
of orthonormal unit vectors which could be written as (vx = value of a given ab-
laut feature, i.e. 0 for the unmarked value, 1 for the marked value):
V{x} = v1R +v2L + v3O

As a consequences of arbitrary ordering, symbols for vector could be omitted
and any set of values of a given ablaut grade V{x} could be expressed in a three
digit code, as above in the table 8.
Any ablaut grade V{x} could then be taken for a vertex of a unit cube. The
set of such vertices is then marked Bn, and that set is a set of ordered n-tuples,
containing numbers 0 and 1, or in other words, as Bn={0, 1}n.
In that sense the elements of the set Bn form dyadic (or Boolean) vectors
and the structure of the ablaut system of OCS is a Boolean algebra, consisting
two elements 0 and 1 (i.e. binary two-valued Boolean algebra) (see
KURATOWSKI 1977: 3435).
The distance between any pair of ablaut grades is then given by the sum of
differences between codes of given ablaut grades:
(8a) Example of minimal distance:

V{non-reduced non-lengthened e-grade} 000
V{non-reduced non-lengthened o-grade} 001
V{non-reduced non-lengthened e-grade} 001 =1
V{non-reduced non-lengthened o-grade}
(8b) Example of zero distance:

V{non-reduced non-lengthened e-grade}
(8c) Example of more than minimal distance:

V{reduced non-lengthened o-grade} 101
V{reduced non-lengthened o-grade}
50 Ondej efk
Such distance between codes is known as Hamming distance10. Hence, any

difference between set of values, considered as codes, could be expressed as a
Hamming distance.
The ablaut vector space which we now face could be considered as a metric
space.
And indeed, we can consider the difference between two values of a same
feature of two ablaut grades as a minimal distance sui generis in such a vector
space and the identity between two values of the same feature of two ablaut gra-
des as zero distance; then we can metricize such a vector space easily.
5. Metric space
Any metric space is generally a tuple (A, U) where A is a set of elements and is
the distance between them given by the mapping of the set A on itself (i.e as a
Cartesian product A A).
The metric space complies with the properties:
1) if two elements have null distance between them, then we are dealing with
one element (axiom of identity, i.e. (x, y) = 0, if x = y);
2) the distance between an element x and an element y is equal to the distance
between y and x (axiom of symmetry, i.e. (x, y) = (y, x));
3) the distance between x and z is equal or smaller than the sum of the distances
between elements x and y and y and z (axiom of the triangle inequality, i. e.
(x, y) + (y, z) (x, z)). See (MARCUS 1967: 34-35, KURATOWSKI 1977:
115).
For a more obvious image of the relations inside the system of the OCS ablaut
we can draw the graph, expressing the system of OCS, in the form of 3-cube
(dyadic cube), with the length of each edge equal to 1. All grades are expressed
by their sets of values, as described in the table 7. In the following examples the
metric space of OCS ablaut is presented both with codes due to the Hamming
distance, both with enumerated ablaut grades, in the form of two mutually pro-
portional 3-D cubes, compare with AF vectors above:
(9a) oG oG
eG eG
eG e G
G G
__________
10 First described by R. W. Hamming in HAMMING (1950).
(9b) 001 101
000 100
011 111
010 110
The distance between any pair of vertices is then equal to the number of edges
between them. Beside the null distance between a given ablaut grade and the
same ablaut grade (axiom of identity) we face the following examples of
distances, see example 10:
(10a) eG oG = 000 100 = eG eG = 000 010 = eG G = 000 001 = 1
etc.
(10b) eG eG = 000 101 = eG oG = 000 110 = oG G = 000 011 = 2
etc.
(10c) eG eG = 000 111 = oG G = 100 011 = 3 etc.
The distance between any pair of ablaut grades is the same without regard to the
order of vertices (axiom of symmetry), see example 11:
(11a) eG oG = 000 100 = oG eG = 1 etc.
(11b) eG eG = 000 101 = eG eG = 2 etc.
Such a metric space we term the fine metric space, because the distances could
be precisely stated and enumerated.
6. Conclusion
In the present paper we have demonstrated that the Old Church Slavonic (and
supposedly the Proto-Slavic) ablaut system could be expressed in the terms of
vector and/or metric space. Such spaces offer strictly formal description of the
whole system and the precise description of the relations between concrete ab-
laut grades. The proposed description using vector or metric spaces is not a
completely new approach to the ablaut, but rather an improvement of the stan-
dard knowledge, using formal instruments.
The main difference between our (and Hubeys) approach and the method of
DP is that DP vectors are not commutative, i.e. the linear combination of
such vectors differs according to the order of vectors. That is the reason why
we use the term vector for DP in the quotations marks, because vectors are
generally considered as commutative. On the contrary, our ablaut features form
commutable vectors in the full sense of the word.
52 Ondej efk
It should be kept in mind that our solution offers a model not for one solitary
root, but for the complex model of the system.
The reader could see in table 5 that the distribution of reduced grades is
uneven. Some roots have only e-colored grades ({reduced non-lengthened} and
{reduced lengthened}), some roots only o-colored grades (again {reduced non-
lengthened} and {reduced lengthened}). It is in a strong contrast to distribution
of full grades that are present for both possibilities of vowel color, either e-
colored or o-colored.
Hence, any individual root has only six grades: {non-reduced non-
lengthened e-grade}, {non-reduced non-lengthened o-grade}, {non-reduced
lengthened e-grade}, {non-reduced lengthened o-grade}, {reduced non-
lengthened} and {reduced lengthened}. This model fits to any given root
(although many roots are not attested for all those grades), but has its
disadvantages: first four grades are described using three values but two with
two values, second reduced grades of different roots could easily differ in the
vocalism (compare roots n_s- and g_n- in the table 5).
The set of all roots could then be split in two disjoint (on the level of
reduced grades) subsets: e-roots and o-roots. Each subset has its own ablaut
pattern which is a sub-pattern of the general ablaut system, as described in the
present paper. This opens new possibilities for the describing of root-
morphology in Old Church Slavonic.
References
ANDERSON, J. M. C. J. EWEN (1987) Principles of Dependency Phonology. Cambridge:
Cambridge University Press.
ARUMAA, P. (1964) Urslavische Grammatik. I. Band Einleitung, Lautlehre. Heidelberg:
Carl Winter.
BRAINERD, B. (1971) Introduction to the Mathematics of Language Study. New York: Ameri-
can Elsevier.
DERKSEN, R. (2008) Etymological Dictionary of the Slavic Inherited Lexicon. Leiden: Brill.
DURAND, J. (1996) Generative and Non-linear Phonology. London New York: Longman.
HAMMING, R.. W. (1950) Error Detecting and Error Correcting Codes. Bell System Technical
Journal 26(2): 147160.
HUBEY, H. M. (1999) Mathematical and Computational Linguistics. Mnchen: Lincom Euro-
pa 1999.
KORTLANDT, F. H. H. (1972) Modelling the Phoneme New Trends in East European
Phonemic Theory. The Hague: Mouton.
KOSMK, L. R. POTEK (2004) Metrick prostory (Metric Spaces). Praha: Academia.
KURATOWSKI, K. (1977) Introduction to Set Theory and Topology. Warszawa: Pergamon
Press.
LIV = RIX, H. (ed.) (1998) Lexikon der indogermanischen Verben. Wiesbaden: Dr. Ludwig
Reichert Verlag.
MACHEK, V. (1971) Etymologick slovnk jazyka eskho. Praha: Academia.
MARCUS, S. (1967) Introduction mathmatique la linguistique structurale. Paris: Dunod.
MEILLET, A. (1934) Le Slave Commune. Paris: Libraire ancienne Honor Champion.
EFK, O. (2008) Values, Features, Fine Metrics and Oppositions. Linguistica Brunensia 56:
514.
EFK, O. (2009) Preliminary Description of the Czech Phonemic System Using Feature
Geometry. In: DOEKAL, M. M. ZIKOV (eds.) Czech in Formal Grammar 183-196.
Muenchen: Lincom Europa.
VAILLANT, A. (1950) Grammaire compare des langues slaves. Tome I. Phontique. Lyon
Paris: IAC.
VASMER, M. (19531958) Russisches etymologisches Wrterbuch I-III. Heidelberg: Carl
Winter.
Ondej efk, Masaryk University,Department of Linguistics and Baltic Lan-

guages, Arna Novka 1/1, 60200 Brno, Czech Republic, sefcik@phil.muni.cz
II.
Machine Translation
Towards a Rule-Based Machine Translation System
Between Czech and Russian*
Natalia Klyueva, Petr Homola and Ondej Bojar
1. Introduction
In this paper we describe the ongoing work on developing a Machine
Translation (MT) system between Czech and Russian. This system is to be
implemented within the project eslko an MT system between closely-related
languages, in which the relatedness of Slavic languages is exploited (HAJI et al.
2003). Here we will present the dictionary used in the system and the initial
steps in writing transfer rules.
Attention will be focused mainly on syntactic differences between the two
languages. We empirically maintain most frequent incorrespondences and
therefore check the statistics of such a phenomena in a parallel corpus. As it is
virtually impossible to describe all constructions that differ in two languages, we
pay attention only at those most frequent according to the usage in the corpus.
The differences then form the set of rules for a transfer module between Czech
and Russian, and we believe that they will be useful for a comparative study of
the languages as well.
2. Machine Translation between Czech and Russian
There are different approaches to Machine Translation: rule-based, example-
based and statistical MT (SMT). The last one explits large corpora of texts to
extract translation patterns, and is nowadays one of the most successful methods.
For closely-related languages the rule-based approach was stated to suit more, as
it can capture some of the features that are alike in the languages.
Our experimental system will be implemented within the rule-based system
eslko. eslko already includes Czech-to-Slovak, Czech-to-Polish and Czech-
to-Lithuanian language pairs. The first one was based merely on the direct word-
to-word translation, and as the languages are very closely related and share most
of their syntactic properties, the results were more than satisfactory. Czech and
Russian are more distant languages, so the additional step of transfer on a
shallow syntax level is needed.
__________
* The work in this paper was supported by the grant No MSM0021620838,
EuroMatrixPlus (FP7-ICT-2007-3-231720 of the EU and 7E09003 of the Czech
Republic)
58 Natalia Klyueva, Petr Homola and Ondej Bojar
The hypothesis that a rule-based system is better for related languages can
be approved or dissaproved by comparing the output results of the rule-based
MT system, that we are going to built and some of the statistic-based ones, for
example Google.
The main components of our system are:
x dictionary,
x morphological analysis and synthesis,
x set of transfer syntactic rules.
Morphological analysis and synthesis are provided by the tools freely available
on-line, namely Haji's tagger for Czech
(http://ufal.mff.cuni.cz/pdt/Morphology_and_Tagging/Tagging/EXP_tagger/ind
ex.html) and tools for morphological analysis and synthesis from AOT
(http://www.aot.ru/) and Tree Tagger (http://www.ims.uni-
stuttgart.de/projekte/corplex/TreeTagger). In the next sections we will discuss
the components we implemented dictionary and syntactic rules.
3. Czech-Russian Dictionary
It was not so easy to find a Czech-Russian dictionary in a plain text format,
because almost all the dictionaries available were of a so-called box search
format. We could not convert the commercial dictionaries into a machine-
readable format either. The only way was to induce it with the help of the
available resources. First we intended to make use of Ruslan Czech-Russian
dictionary (OLIVA 1989), but it contains only 6000 words and is adjusted to the
special format of that MT system.
The way that we chose was to extract the dictionary from the freely
available parallel Czech-Russian corpus UMC 0.1 (KLYUEVA & BOJAR 2008).
We used sentences, that were aligned 1-to-1 because they are most reliable.
Over 88,000 sentences parallel in Czech and Russian were processed by
taggers. For Russian we used TreeTagger and for Czech the Hajic's tagger in
order to get lemmas. We use lemmas instead of word forms in subsequent steps.
The output of the parallelly lemmatized text looks like the following:
(1) v zoufal snaha udret se u moc hodit Parvz Muaraf pkistnsk stavn
rmec za hlava a vyhlsit v zem vjimen stav . ||
"#$ , %& '"/6 7" 9";
<9 >
# & <$ .
Then we ran GIZA++ word alignment (OCH & NEY 2003) on the lemmatized
corpus and got the table of word correspondences, sorted them by frequency.
The output was exactly the machine-readable dictionary we needed.
Towards a Rule-Based Machine Translation System Between Czech and Russian 59
The nouns on the Russian side were enriched with the morphological feature
of gender and animatedness ensure the proper generation of a word forms
(zem||gender=fem). For example, due to this feature, krasn tramvaj (cz:
beautiful-fem tram) will be translated into Russian as 9 (ru:
beautiful-masc tram).
The number of entries that we got was over than 400,000 words, and we
faced the question how much we should trust the word alignment. During
manually processing we found lots of incorrespondences so we decided to delete
the double entries on the Czech side even at a price of non-resolved ambiguity.
4. Syntactic Transfer Module
As we mentioned earlier, the syntax of Czech and Russian differs in a number of
constructions, though not so radically as for example Czech and English. So our
next step will be to develop a set of transfer rules, that can capture the syntactic
differences between the two languages. In this section we will demonstrate the
most frequent differences that will form the base of the transfer rules.
4.1 Pro-dropness in Czech and copula drop in Russian
One of the most evident differences in sentence structure concern such language
phenomena as regular pro-drop in Czech a drop of a copula > (to be) in
present tense in Russian.
Czech tends to incorporate a person morpheme into verb and leave out the
personal pronoun, whereas Russian (as English) rather uses the pronoun. To
illustrate this fact we have calculated the statistics of pronoun usage in the
parallel Czech-Russian corpus described in Section 3.
In the following table we show that for the same sentences the usage of
personal pronouns in Russian language is approximately more frequent than in
Czech.
1st person 2nd person 3rd person sg&pl
Sg Pl Sg Pl Sg Pl
Czech 143 462 8 18 264 167
Russian 5433 2361 24 334 5102 4131
Next we will demonstrate concrete examples of different types of dropness in

two languages.
Copula constructions
Following is the example of the copula construction for the languages:

(2) Jsem student (cz) -> (ru)
be-1SgPrae student-Nom (cz) -> I-1Sg student-Nom(ru)
I am a student.
In this construction one more variant of translation into Russian exists. A
construction with a hyphen as a substitution for the verb 'to be' is also very
widespread: ? - "# (ru) (I student). According to the corpus statistics,
the construction with the hyphen is less frequent, so we prefered the first variant
of translation, without the hyphen. There is also the possibility to translate (2) as
'? @ "#', but it sounds very formally.
The transfer rule for this difference can be schematized like this:
(#1) 'to be'(person X) -> pronoun(person X)
This rule means that the tool will remember the person of the verb to be from
the input Czech sentence. The morphological synthesis will transfer this
information so that on the Russian side the pronoun will appear in the respective
person.
Past tense of the verb 'to be'
(3) Vera jsem byl doma (cz) -> Y > # (ru)
Yesterday Be-1SgAux Be-PastParcipleSg at_home (cz) ->Yesterday I-1Sg be-
SgPast at_home (ru)
'I was at home.'
Past Tense
We also observe the difference in syntactic constructions in the case of the past
tense of the 1st and 2nd person. In order to express the person of the noun-actor,
the Russian language uses the pronoun (, , , ) while the Czech
language uses the corresponding form of the auxilary verb 'to be' in addition to
the main verb (jsem, jsi, jsme, jste).
(4) Oslavila jsem Silvestr s rodinou (cz) ->
.
celebrate-PastParcipleSg be-1SgAux (cz)-> I-1Sg celebrate-PastSg...(ru)
'I celebrated New Year with my family'
Here is the generalization of the rule for the past tense:
(#2a) PastParciple + 'to be'(person X) ? pronoun(person X) + verb
(#2b) 'to be'(person X) + PastParciple ? pronoun(person X) + verb
The rule (#2b) reflects the cases when the auxiliary verb appears in the
Wackernagel's position second syntactic element in the sentence, see example
3).
When a pronoun is used in the Czech sentence, the auxiliary 'to be' will be
omitted on the Russian side. The rule works for the copular sentences as well.
(5) My jsme sndli vechno (cz) -> (ru)
we-1Pl be-1PlAux eat-PastParticiplePl everything -> We-1Pl eat-PastPl
everything
'We ate everything.'
4.2 Conditional mood
Conditional mood in Czech and Russian differs with respect to the general
syntactic characteristics of the languages. These are: pro-drop in Czech, where
person is expressed in the auxiliary 'to be'. In Russian on the contrary the
pronoun presents the person, but the verb 'to be' has only one form for all
persons the third person ().
(6) Chtla bych-1_person -> -1_person
'I would like'
4.3 The verb 'to have'
(7) Mm koku (cz) -> (ru)

have-1Sg cat-Acc -> To-Prep me-Gen to_be-1Sg cat-Nom
'I have got a cat.'
Here the system remembers the person of the possessor and inserts the
appropriate pronoun into Russian be-possessive construction.
4.4 Other differences
Some minor transfer rules were written to capture the following differences
between the languages:
Negative particle in Russian vs. Negative prefix in Czech
(8) Nepracuje (cz) -> [ > (ru)
work-NEG (cz) -> Not work (ru)
'He does not work.'
(#3) verb(feature-NEG) -> + verb
The rule (#3) gets a negative feature from the morphological analysis and
transforms it into the negative particle for Russian.
Reflexive suffixes -/- in Russian vs. reflexive particles in Czech

The situation here is opposite to the previous one. Reflexive particles in Czech
should be transfered into suffixes incorporated into the russian verb:
(9) Chlapci se smj.refl nad obrzky (cz) ? '9 @.refl #
99 (ru)
'Boys are laughing at the images.'
(#4a) verb + se/si -> Verb_refl
(#4b) se/si + verb -> verb-> refl (for Wackernagel second position)
In case when a Czech verb is reflexive and a Russian is non-reflexive, the
translation will be provided by the dictionary:
(10) Slon se prochzel.refl po parku (cz) ? ^ 7".non-refl < <9" (ru)
'The elephant was walking in the park.'
In the future we plan to enlarge the list of transfer rules, that is going to capture
the most frequent and evident incorrespondences.
5. First Results
We tested our system on the experimental test set of 1000 sentences from
Project Syndicate mainly news on political and economic events. Those
sentences are to be translated in other experiments, for example while testing the
statistical machine translation system, which will provide the rich material for
comparison.
The test set for Czech-Russian translation is already equipped with reference
translations. MT quality can be thus estimated by comparing MT output with the
reference.
5.1 BLEU
BLEU score is an automatic metric for evaluation of MT. The sentences

translated by our MT system were compared automatically to those from parallel
corpus. For example, if we have the sentence and its reference translation:
(cz) Tento detailn pln stanovuje kosovskou dozorovanou nezvislost,
maximln ochranu Srb a dalch menin a dozor lohu EU.
(ref) %9 >< & <# #&, 9"@
&`" > #"7~ /, <#&" <" [
9 >@#.
(eslko translation) # < " 9
dozorovanou & , 9"@ &`" ^> #"7
/ >@#"@ ^ .
The reference (ref) translation is far from literate and in some parts even the
sense is sacrificed to the well-soundness of the text. So the reference translation
and the output translation are very distant, they do not really reflect the
translation quality (7 of 17 words are translated properly).
5.2 Manual evaluation
Because of the drawbacks of the automatic metrics of evaluation described

above, we made the manual evaluation based on linguistic error classification.
The errors are classified into several classes with the respect to the language pair
and MT system described. They are: morphology errors: (mor::), case errors
(case::), agreement errors (agr::), wrong synonym choice (syn::), wrong word
choice (dic::) as it is mainly caused by poor dictionary quality, wrong word in
a multiword expression (mult::). The errors are therefore marked with the
appropriate abbreviation from the set described above.
(11) Udusit nezvisl soudnictv a svobodn mdia.
syn::#~" & mor::"#> >#
dic::6; .
(12) Mohou - li nm bt vodtkem pkistnsk djiny, pak se jeho rozhodnut
zavst vjimen stav me ukzat jako psloven posledn kapka.
**'7" - > "9# mor::<9 ,
case:: case::/ # case::& mult::
$ 9& 99 mult::/ mult::<# mult::9<.
A preliminary evaluation has shown that we have to improve all modules of the
system. Our next steps will be to:
x Improve the dictionary or get a new one,
x Improve morphology modules,
x Add even more syntactic rules,
x Introduce semantics, namely word sense disambiguation module.
6. Conclusion
In the paper we have described an MT system between Czech and Russian. The
results of the first run of the system are shown, and the steps on how to improve
the quality are suggested. We focused mainly on the development of the transfer
rules that are supposed to capture the differences between syntactic
constructions in Czech and Russian. Now we have only a small set of rules that
reflects only the most frequent incorrespondences. To improve the quality of the
system we are going to add more syntactic rules.
References
KLYUEVA, N. & O. BOJAR (2008) UMC 0.1: Czech-Russian-English Multilingual Corpus.
Proceedings of the Conference 'Corpora 2008'. St.Petersburg.
HAJI, J., P. HOMOLA & V. KUBO (2003) A simple multilingual machine translation system.
In: Proceedings of the MT Summit IX, New Orleans.
OCH F.J. & H. NEY (2003) 'A Systematic Comparison of Various Statistical Alignment
Models', Computational Linguistics, volume 29, number 1, March 2003.
OLIVA, K. (1989) A Parser for Czech Implemented in Systems Q. In: Explizite Beschreibung
der Sprache und automatische Textbearbeitung, MFF UK Prague.
Natalia Klyueva, Charles University in Prague, Institute of Formal and Applied

Linguistics (FAL), Malostransk nmst 25, 11800 Praha, Czech Republic,
kljueva@ufal.mff.cuni.cz
Petr Homola, Charles University in Prague, Institute of Formal and Applied
homola@ufal.mff.cuni.cz
Ondej Bojar, Charles University in Prague, Institute of Formal and Applied
Ondrej.Bojar@mff.cuni.cz
III.
Semantics
On Animacy and Unaccusativity in Russian*
Zhanna Glushan
1. Introduction
Unaccusativity as a linguistic phenomenon has been an important milestone of
linguistic theory for a long time. Starting from PERLMUTTERs (1978)
Unaccusativity Hypothesis, it has been assumed that two types of intransitive
verbs need to be distinguished: unaccusative and unergative. In structural terms,
the intransitive verb distinction has been understood as a deep structure position
of a single argument with respect to the VP: unaccusative predicates select a
single argument internal to the VP, while unergative verbs select a single
argument external to the VP.
In this paper, I will argue for a finer-grained distinction, recognizing four
distinct theta roles that intransitive subjects may have, and correspondingly four
distinct base positions. The evidence for this view comes from the interaction of
animacy and unaccusativity diagnostics in Russian.
Animacy and/or volitionality have long been known to interact with
unaccusativity diagnostics. For example, so-called variable behaviour verbs
are compatible with both volitional and non-volitional interpretations, but
behave as unergatives when the subject is volitional, and unaccusatives when
not.
For Russian, the most famous example of a variable behavior predicate is
the verb plavat which can have the interpretation float (non-volitional) or
swim (volitional). Crucially the verb passes unaccusativity tests, such as the
Genitive of Negation (Gen of Neg) only on the float reading (PESETSKY
(1982)).
(1) v basseine nikakogo malika ne plavaet
in pool no boyGen Sgl not floats
No boy is floating/??swimming in the pool
Intransitive predicates with variable behavior have also been noted for English
(PERMUTTER & POSTAL (1984), ZAENEN (1993), HOEKSTRA & MULDER (1990)).
As demonstrated by (2) and (3) below, while the verb slide can combine with
both animate and inanimate subjects, only an agentive (animate) counterpart of
__________
*
Acknowledgements: This paper appeared as a part of the general examination
requirement and a result of a helpful collaboration of my overseers: eljko Bokovi,
Jonathan Bobaljik and Susanne Wurmbrand to whom I am deeply thankful. This work
was partly supported by NSF grant BCS 0616339 An integrated morphosemantics of
agreement PI: Jonathan David Bobaljik. All errors are mine.
68 Zhanna Glushan
slide passes the unergativity diagnostic (pseudo-passive) (PERMUTTER &

POSTAL, 1984:102).
(2) a. Ted slid into the closet. (agentive or nonagentive)
b. The soap slid into the closet. (only nonagentive)
(3) a. The closet was slid into by Ted. (only agentive)
b. *The closet was slid into by the soap.
In this paper, I will argue that animacy is a more important component of
unaccusativity than has been generally assumed in the literature before. In
particular, I will show that animacy of the NP matters for all valid
unaccusativity diagnostics in Russian and that this generalization extends
beyond classical examples of variable behavior predicates.
To illustrate my key argument, consider the minimal pairs below. The data
in (4) illustrates an unaccusativity test (only internal arguments may be
expressed as po-NPs): the animate counterpart of the fall predicate in (4)b fails
the test independently of the volitionality of the verb, thus (4) is distinct from
standard variable behavior effects.
(4) a. Po jabloku upalo s kadogo dereva
Po appleDat fell from each tree
An apple fell from each of the trees
b. *Po sportsmenu upalo s kadogo trenaera
po sportmanDat fell from each machine
Another piece of the puzzle is that the animacy effect in (4) has different effects
for different unaccusativity diagnostics. Thus, while unergative subjects
uniformly fail the po-test, regardless of animacy (see below), an animacy effect
does arise among unergative subjects with other diagnostics. In (5), inanimate,
but not animate, subjects pass the Genitive of Negation unaccusativity
diagnostic.
(5) a. na kuhne nikakih/ni odnogo holodilnikov/a ne rabotalo
in kitchen no/not single fridgeGen not works
b. *na zavode nikakih enin ne rabotaet
at factory no womenGen not works
I will offer an account to these generalizations along the lines of REINHARTs
(2002) theta role decomposition analysis. I will argue that animacy effects are a
result of an indirect interaction between animacy and deep structure positions. In
my analysis, animacy effects with unaccusativity tests follow from distinct theta
role specifications assigned to animate and inanimate arguments. The full range
of data are accounted for by positing that animacy and verb meaning interact in
the determination of theta-roles. A more fine-grained array of theta-roles than
the simple binary division of VP internal and external is proposed. I conclude
that a standard two-way division of intransitive predicates does not suffice.
On Animacy and Unaccusativity in Russian 69
Russian unaccusativity data makes a clean cut of four types of intransitive

predicates listed in (6) below:
(6) I. True unaccusatives. Sole argument is Theme. Verbs of existence (e.g.
be, appear ...), die, and the existential reading of otherwise
unergative predicates (see below).
II. Experiencer unaccusatives. Sole argument is Experiencer if animate,
otherwise Theme (e.g.arrive, fall, melt burn, freeze, soak,
redden drown).
III. Unergatives. Sole argument is Agent if animate, and Instrument if
Inanimate (e.g. play, run, work, whistle, jump, sing, eat).
IV. Variable behaviour. Sole argument is Agent if volitional, and Theme
if not (regardless of animacy) (e.g. plavat(swim/float),
stojat(stand/be) in Russian).
It follows from my analysis that it is more appropriate to treat agentivity effects
with unaccusativity not as special property of listed verbs, but as contrasts
resulting from distinct theta specifications of the subjects. Russian
unaccusativity data allow us to show the full range of theta role specification
possibilities and the distinctions between them.
2. Summary of the data
There are five unaccusativity diagnostics for Russian that have been discussed in
the literature: distributive po-phrases, verb prefixation, Gen of Neg, Locative
Inversion (LI) and first conjunct agreement (FCA)2. In the following
subsections, I will be looking at each of the diagnostics and investigating
whether/how animacy of the argument interacts with the test outcome. The tests
are applied to several verb types: apart from intransitive and transitive
predicates, the results are compared to the group of existential predicates and
variable behavior verbs. The results obtained for each of the tests in the
following subsections are summarized in table (7) below3.
__________
2 BABYONYSHEV (1996) and HARVES (2002) include First Conjunct Agreement (FCA) into
the list of effective unaccusativity diagnostics in Russian. However, animacy matters in
FCA for all verb classes including transitive predicates. Therefore, in GLUSHAN (2008), I
argue that the animacy effects with FCA lies outside of unaccusativity. I thus exclude
FCA from the table (7).
3 The table is read as follows: yes animacy of the subject influences the test outcome;
no animacy of the subject does not influence the test outcome. Blank slot the
construction is unavailable with this type of predicate for an independent reason.
70 Zhanna Glushan
(7) Effects of Animacy by Verbs Class
Verb types
Diagnostics Unaccus Unerg Trans Exist Variable

behavior
Distributive yes no No no no
po- phrase
Verb yes no No no
prefixaton
Gen of Neg yes yes No no no
Locative yes yes no no

Inversion
The results summarized in table (7) appear to present a challenge to the standard
theory of unaccusativity. In particular, the following questions require an
explanation:
(8) (i) Why does animacy of the subject play a role in unaccusativity
diagnostics?
(ii) Why do animacy effects show different distribution with different
diagnostics?
(iii) Why do animacy effects arise with intransitive subjects but not with
transitive objects?
The analysis laid out in section 3 of this paper provides an account of the data in
table (7) as well as answers to the questions set out in (8).
3. Animacy and theta-roles
In this paper, I offer an account of animacy effects with unaccusativity in terms
of theta decomposition analysis proposed in REINHART (2000, 2002). REINHART
(2000) suggests a system of formal features that compose theta roles. Reinhart
defines causal relationships in terms of two binary features: [c] = cause change,
and [m] = mental state. The role of Agent and Cause share the feature value
[+c]. Agency involves such properties as volition and intention, which Reinhart
defines as [+m]. Note that [m] is to be understood as whether the mental state
of the NP is relevant to the predicate. In this way, the positive value of [m]
entails animacy, but not conversely. By means of features [+/-c] and [+/-m]
feature clusters are derived which directly correspond to theta-roles (REINHART

(2002: 232))4.
(9) a. [+c +m] agent (REINHART, 2002: 232)
b. [+c-m] instrument
c. [-c+m] experiencer
d. [-c-m] patient
I suggest an extended version of REINHARTs (2002) theta decomposition where
theta role specifications in (9) applied to intransitive subjects result in the
following distinctions: animate subjects of unergative verbs bear a theta role of
Agent [+c+m]; inanimate subjects of unergative verbs receive a theta role of
Instrument [+c-m]; inanimate subjects of unaccusative verbs are Themes [-c-m];
while animate subjects of unaccusative verbs are Experiencers [-c+m] wherever
they are affected by the change of state/location denoted by the predicate. Thus,
fall, arrive take Experiencer arguments, but existence predicates select for
Themes, regardless of animacy (see (10) below).
In my analysis, I will stick to the general view that the distributive po-
phrases test, verb prefixation, and Gen of Neg diagnose the LF position of an
argument while LI is a surface diagnostic which reveals VP internal arguments.
In (10) I also indicate the scope of each of these diagnostics.
(10) vP
ru
Agent v
[+c+m] ru
v0 ApplP
ru
Experiencer NegP Gen of Neg
[-c+m] Neg0ei Locative Inversion
VP
ru verb prefixation
Instrument V po-phrases
[+c-m] ru
na- V Theme
distr [-c-m]
Given the structure in (10), the animacy conflicts observed with unaccusative
predicates for all diagnostics (see table (7)) are instances of the
Experiencer/Theme role interaction: these tests give a grammatical result only if
a single argument of the verb is a Theme [-c-m].
__________
4 See REINHART (2000), (2002) for more discussion on the empirical motivation for the
feature specifications. The list in (9) excludes the theta roles which are not relevant for
the purposes of my analysis.
72 Zhanna Glushan
The absence of animacy effects with subjects of unergative verbs for some
(distributive po-phrase, verb prefixation test) but not other diagnostics (Gen of
Neg, LI) is explained by distinct domains of licensing/scopes for the diagnostics.
The domain of licensing for Gen of Neg and LI is higher, thus it includes
Instruments and Themes (see (10)) creating an additional contrast
Agent/Instrument detected by these diagnostics. The domain of licensing for
distributive po-phrases and verb prefixation tests excludes both Agent [+c+m]
and Instrument [+c-m], thus no animacy conflicts arise with respect to these
diagnostics.
Variable behavior verbs show no Experiencer/Theme interaction since
these verbs allow animate Themes on non-agentive reading. Similarly, due to
their semantics, verbs of existence can not take Experiencer [-c+m] subjects,
thus they select for Theme [-c-m] arguments regardless of animacy.
For reasons of space, the discussion in this paper will be limited to the
behavior of intransitives. In GLUSHAN (2008) I argue that the difference between
objects of transitives and unaccusatives is the result of an additional restriction
that there be maximally one VP-internal and one VP-external argument for a
transitive predicate (which could be stated as one theta-role per domain). This
distinctness condition on theta roles entails that the object of a transitive
predicate is always VP-internal (a Theme) regardless of animacy.
4. Data
4.1 Distributive po-phrase, Verb prefixation tests
PESETSKY (1982) argued that the acceptability of distributive po-phrases
followed a classic unaccusative distribution. For transitive verbs, a po-phrase
can be the object, but not the subject. For intransitives, a po-phrase may occur as
the subject of an unaccusative predicate (see (11)), but cannot be the subject of
an unergative one (see (12)).
(11) po jabloku upalo s kadogo dereva
po appleDat fell from each tree
An ( different) apple fell from each tree. Unaccusative
(12) ??po sobake kusaetsja v kadoj kletke
podog bites in every cage
A (different) dog bites in each cage. Unergative
The same distribution arises with quantificational subjects of verbs with certain
quantificational prefixes, such as na-, pere- (BORIK (1995), SCHOORLEMMER
(1995), HARVES (2002)). The data in (13) demonstrates the relevant contrast: the
verb prefixed with a quantificational prefix na- cannot occur with unergative
predicates.
(13) a. Mnogo travy naroslo v parke Unaccusative

A lot grass grew in park
A lot of grass grew in the park.
b. *Mnogo detej naigralo v parke
A lot of children played in park
A lot of children played in park. Unergative
The standard account of these diagnostics is that they show a scope effect: the
po-phrase or quantificational NP must be in the scope of an operator (a prefix, or
a distributive operator in the case of the po-phrase) at the relevant level. If we
assume that quantificational licensing obtains at LF, then these diagnostics
indicate the LF position of the quantified arguments. If, moreover, quantified
expressions can occur at LF no lower than their theta position (whether by
reconstruction or by remaining in situ throughout the derivation), then these tests
serve indirectly to diagnose the base positions of different arguments. The
classic unaccusative behavior in (11)-(12) and (13) is thus explained, if the
relevant operators are positioned somewhere between the position of internal
and external arguments.
My observation is that for a range of verbs typically listed as unaccusative,
the animacy of the subject matters for the outcome of the PO-phrase and the verb
prefixation test. While an inanimate subject is indeed possible as a complement
in a po-phrase (quantifier mnogo in the case of verb prefixation), an animate
subject in the same frame is disallowed. The relevant contrast is represented for
the distributive PO-phrase and verb prefixation test in (14)-(15).
(14) a. Po jabloku upalo s kadogo dereva Unaccusatives
Po appleDat fell from each tree po-phrase
An apple fell from each of the trees.
b. *Po sportsmenu upalo s kadogo trenaera
po sportmanDat fell from each machine
A sportsman fell from each of the machines.
(15) a. Mnogo travy naroslo za vesnu Unaccusative
A lot of grass grew in spring na-prefix
A lot of grass has grown over the spring.
b. *Mnogo detei naroslo za vesnu
Many children grew in spring
A lot of children have grown over the spring.
Animacy plays no role for unergative predicates (see (16)-(17) below). As
expected, distributed po-phrases as well as verb prefixation are
ungrammatical/degraded with unergative predicates5.
__________
5 Animacy also plays no role for animate/inanimate objects for both diagnostics. See
GLUSHAN (2008) for the full range of data and motivation of the analysis.
74 Zhanna Glushan
(16) a. *Po mysli probealo v kadoi golove Unergatives

pothoughtDat ran in each head po-phrase
A thought ran through each head.
b. *Po maliku probealo v kadoj komnate.
Po boy ran in every room
A boy ran in each room.
(17) a. *Mnogo plastinok naigralo v naem parke Unergative
Many records played in our park na-prefix
Many discs have played in our park.
b. *Mnogo detei naigralo v parke
Many children played in park
Many children have played in our park.
Puzzling on standard accounts of unaccusativity in Russian, animacy effects
with unaccusative predicates for both of the diagnostics follow from the analysis
in (10). Given that unaccusative verbs assign a Theme [-c-m] role to inanimate
arguments, but Experiencer [-c+m] to animate ones, the tests detect the
Theme/Experiencer interaction by virtue of the intermediate position of the
licensing domain. For both tests this domain includes the Theme arguments, but
excludes Experiencer arguments.
Variable behavior predicates such as swim, roll, stand6 show no
animacy constrast on the surface as the tests are applied, i.e. they give a
grammatical result with animate subjects. However, the relevant contrast is
observed in terms of a shift in meaning of the predicate (non-agentive reading)
(compare (14), (15) to (18)-(20))7.
(18) a. po maliku plavalo v kadom bassejne
po boy swim in each pool
A boy was swimming/floating in each pool.
swim/float po-phrase
b. po lodke plavalo v kadom prudu
po boat float in each pond
A boat was floating/sailing in each pond.
__________
6 See PESETSKY (1982), BABYONYSHEV (1996), HARVES (2002) on discussion of variable
behavior predicates in Russian.
7 Not all variable behavior verbs allow the attachment of the na- prefix. The prefix itself
contributes special semantics that makes it incompatible with some verbs independently
of transitivity. Roughly, the na- prefix requires reaching a result state in some sort of a
countable, material form. Such a situation can not occur with verbs like plavat
(float/swim), verbs of existence and stative verbs. This is an interfering although an
advantageous property of this test: it rules out an existential/stative interpretation of
unergative predicates, which can interfere with the result of po-phrase and Gen of Neg
test (see also fn 8).
(19) a. po milicioneru stojalo na kadom vyhode

po officer stood on each exit
An officer was standing at each of the exits.
stand po-phrase
b. poznaku stop stojalo na kadom perekrestke
postop sign stood oneach intersection
A stop sign was on each intersection.
(20) a. Mnogo milicionerov perestojalo na etom postu
many policemen stood at this post
Many policemen guarded (standing) this area.
stand pere-prefix
b. Mnogo znakov perestojalo na etom perekrestke
Many signs stood on this intersection
Many signs have been changed at this intersection.
The absence of animacy effects with variable behavior predicates in (18)-(20)
above, as well as a puzzling contrast between these data and the rest of
intransitive predicates (see (14)-(17) above) can be explained by the analysis in
(10). Unlike other intransitive predicates, variable behavior verbs can take
animate Theme arguments, i.e. these predicates are special in the sense that
they can treat their sole animate argument as inanimate. The shift in the
interpretation of a sole argument is reflected by means of a non-agentive
interpretation of such predicates.
Apart from the variable behavior verbs, verbs of existence and the verb
die also pass the po-phrase test regardless of animacy of the subject. The Verb
prefixation test is not available for this type of predicate (see fn 7).
(21) a. v kadoj gruppe bylo po-uastniku
in each group were po-participant
b. v kadoj korobke bylo po knige
in each box was po book Dat
(22) a. po ilcu umerlo v kadoj kvartire
po tenant Dat died in every apartment po-phrase
b. Mnogo muikov poumiralo v derevnjah.
Many men died in villages verb prefix
Many men have died in the villages.
The absence of animacy effects with verbs of existence and die can be
explained if these verbs can not take Experiencer subjects, but allow animate
Theme arguments and create no animacy conflicts with respect to unaccusativity
diagnostics. A similar explanation extends to the verb die: this verb being a
special type unaccusative predicate which does not take Experiencer subjects.
76 Zhanna Glushan
4.2 Gen of Neg, Locative Inversion

Genitive of Negation (Gen of Neg) test is a standard unaccusativity diagnostic
for Russian. PESETSKY (1982) observed that the distribution of this construction
in Russian is limited to transitive objects and subjects of unaccusative
predicates.
(23) Otveta iz polka ne prilo
answer Gen from regiment not come
The answer from regiment did not arrive. Unaccusative
(PESETSKY, 1982:43)
(24) *Na zavode nikakih enin ne rabotaet
Women dont work at a factory. Unergative
Various analyses of Gen of Neg have been suggested in the literature (PESETSKY
(1982), PERELTSVAIG (1999), BABYONYSHEV (1996), BABBY (2001), PARTEE
(2008), PARTEE & BORSCHEV (2007) among many others). For many of the
analyses the Gen NP is required to occupy a position within the scope of Neg0 at
LF (or another level). The NPs where Gen of Neg is disallowed fall outside of
the scope of Neg0.
Locative Inversion has been argued to be an effective unaccusativity
diagnostic in Russian (BABYONYSHEV (1996)). BABYONYSHEV (1996) observed
that under a discourse neutral interpretation, the locative inversion construction
in Russian is allowed only with unaccusative predicates. LI is thus a surface
unaccusativity diagnostic which is sensitive to VP internal arguments.
(25) V sadu rosli tri rozy Unaccusatives
In garden grew three roses
There grew three roses in the garden. (BABYONYSHEV, 1996:40)
(26) *Sebe pod nos svistit Petja Unergatives
Self under nose whistles Petja
To himself sang Petja.
My observation is that, similar to the distr. po-phrases and the verb prefixation
diagnostics discussed above, animacy effects arise with unaccusative verbs for
both Gen of Neg and LI diagnostics.
(27) a. nikakih gribov zdes ne rastet
no kind mushrooms Gen here not grow Gen of Neg
b. *Nikakogo/*ni odnogo rebenka ne roslo/rastet
no kind not single child not grow
(28) a. Na dereve krasnelo jabloko

on tree reddened apple Loc Inv
b. *V saune krasnel muik
in sauna got red (from steam) man
The surprising part is that, unlike with the previous tests, animacy effects are
observed with these diagnostics for unergative predicates as well. In other
words, when intransitive verbs are tested in animate/inanimate subject pairs, no
clear distinction between unaccusative/ unergative predicates can be made with
Gen of Neg or LI8.
(29) a. na kuhne nikakih/ni odnogo holodilnika/a ne rabotalo
in kitchen no/not single fridgeGen not works
Gen of Neg
b. *na zavode nikakih enin ne rabotaet
(30) a. V krane beit voda Loc Inv
In tap runs water
b. ?? V parke beit sportsmen
in park runs sportman
On the theta decomposition analysis outlined above (see (10) above) animacy
distinction with the Gen of Neg and LI can be captured in the following way.
Given the contrast of Gen of Neg and LI data with the two diagnostics discussed
above, I assume that the licensor for Gen of Neg and LI is higher than the
licensor of the distributive po-phrases and verb prefixation. It is on the basis of
the Gen of Neg facts that I establish the relative order between Experiencer and
Instrument: Experiencer arguments are disallowed with Gen of Neg, unlike
Instruments, thus Experiencers are higher. Inanimate unaccusative subjects and
inanimate unergative subjects are allowed with the Gen of Neg since they bear a
__________
8 BABBY (2001) argues against the standard analysis of Gen of Neg on the basis of the data
where unergative subjects can receive Gen of Neg (see (i)).
Context: Na zabroenom zavode upal i razbilsja Saa. (BABBY, 2001: 50-51)
Sasha fell and was badly hurt at the abandoned factory.
(i) Tam bole ne igraet nikakih detej (ii)* detej ne igralo nabajane
there more not play no children childrenGen not played bajan
There are no longer any children playing there.
Importantly, BABBYs (2001) analysis relies on the assumption that Gen of Neg can be
assigned only when the sentence is existential. The special context provided in (i) forces
the existential reading of an unergative predicate. Only on the existential reading,
animate subjects are allowed with Gen of Neg (compare (i) and (ii)). In my account, I
treat such unergative predicates as verbs of existence (see also CALABRESE & MALING
(2009) on similar effects in Italian).
78 Zhanna Glushan
Theme and an Instrument theta role respectively. The same reasoning applies to
inanimate subjects of intransitive verbs in LI structures. Animate subjects of
unaccusatives (Experiencers), animate transitive subjects (Agents) are correctly
excluded since they fall out of the domain of licensing for both diagnostics.
Verbs of existence create no animacy effects with Gen of Neg and LI as
demonstrated in (31), (32).
(31) a. zdes horoih ljudei ne suestvuet
here good peopleGen not exist Gen of Neg
(PESETSKY, 1982:43)
b. nikakih dokladikov ne pojavilos
no speakersGen not appeared
c. rebenka na skamejke ne bylo
childGen on bench not was
(32) a. v derevne byl vra Loc Inv
in village was doctor
b. na katke pojavilsja sudja
on skating ring appeared referee
Gen of Neg and LI are allowed with animate subjects of ambiguous verbs
swim, stand roll on the non-agentive reading, as well as the verb die.
(33) a. v basseine nikakogo malika ne plavaet
in pool no boyGen Sgl not floats
No boy is floating/??swimming in the pool. Gen of Neg
b. v basseine plaval malik
in pool ??swim/float boy
There was a boy floating/??swimming in the pool. Loc Inv
(34) a. Nikakogo/ ni odnogo starika ne umerlo v naem dome
no not single old man not dies in our house
Gen of Neg
b. V derevne umer vra Loc Inv
in the village died doctor
Following the line of analysis laid out in (10), the exceptional behavior with
ambiguous predicates, existential verbs as well as the verb die is explained in
the way analogous to the po-phrase and verb prefixation diagnostics. These
verbs are special in that they can take animate Theme arguments on non-
agentive reading (variable behavior) or in general they take a Theme argument
regardless of animacy.
5. Conclusions and Implications

In this paper, I have shown that animacy plays a crucial role in unaccusativity.
In particular, on the basis of Russian data I have argued that when animacy is
taken into consideration standard unaccusativity tests do not give a consistent
result. I have proposed an analysis of animacy effects observed with
unaccusativity diagnostics in terms of distinct theta role specifications along the
lines of REINHART (2002) theta decomposition analysis. The data contributes an
important missing piece to the theory of unaccusativity in Russian and cross-
linguistically. In particular, in the light of the Russian data, the standard two-
way division of intransitive predicates is not sufficient. Russian unaccusativity
data makes a clean cut of four types of intransitive predicates based on the type
of theta role assigned to a single argument.
References
BABYONYSHEV, M. (1996) Structural connections in Syntax and Processing: Studies in
Russian and Japanese. PhD dissertation, MIT.
BABBY, L. (2001) The genitive of negation: a unified analysis. In: FRANKS, S., T. KING & M.
YADROFF (eds.) (2000) Formal Approaches to Slavic Linguistics, the Bloomington
Meeting 39-55. Ann Arbor, MI: Michigan Slavic Publications.
BAKER, M. (1988) Incorporation: A theory of Grammatical Function Changing. University of
Chicago Press, Chicago, Ilinois.
BEGHELLI, F. & T. STOWELL (1997) Distributivity and negation: the syntax of each and every.
In: A. SZABOLCSI (ed.) Ways of Scope Taking 71-107. London: Kluwer.
BORIK, O. (1995) Syntakticheskij priznak neakkuzativnosti glagola (na materiale russkogo
jazyka) MA Thesis, Dept. of Theoretical and Applied Linguistics, MGU.
CALABRESE, A. & J. MALING (2009). Ne-cliticization and auxiliary selection: agentivity
effects in Italian. Ms. University of Connecticut, Storrs.
GLUSHAN, ZH. (2008) On animacy and unaccusativity in Russian. Ms. University of
Connecticut, Storrs.
HARVES, S. (2002) Unaccusative syntax in Russian. Princeton University: PhD dissertation.
HOEKSTRA, T. & R.MULDER (1990) Unergatives as copular verbs: Locational and existential
predication. The Linguistic Review 7: 179.
KIRSNER, R. (1977) On the Passive of Sensory Verb Complement Sentences, LI vol. 8.1, 173-
179.
LEVIN, B. & M. RAPPAPORT HOVAV (1995) Unaccusativity. MIT Press, Cambridge, MA.
PARTEE, B. & V. BORSCHEV (2007) Existential sentences, BE, and the Genitive of Negation in
Russian. In: COMOROVSKI I. & K. VON HEUSINGER (eds) Existence: Semantics and
Syntax 147-190. Dordrecht: Springer.
PARTEE, B. (2008) Negation, intensionality, and aspect: Interaction with NP semantics. In:
Rothstein, S. (ed) Theoretical and Crosslinguistic Approaches to the Semantics of
Aspect 291-317. Amsterdam: John Benjamins.
PERELTSVAIG, A. (1999) The Genitive of Negation and Aspect in Russian. Ms. McGill
University.
Perlmutter, D. (1978) Impersonal passive and the Unaccusative Hypothesis. In: JAEGER, J. et
al (eds.), Proceedings of the fourth annual meeting of the Berkeley Linguistics Society
159-189. University of California at Berkeley.
80 Zhanna Glushan
PERLMUTTER, D. & P. POSTAL (1984) The 1-Advancement Exclusiveness Law. In: DAVID, M.
& C. ROSEN (eds) Studies in Relational Grammar 2 81-125. Chicago: University of
Chicago Press.
PESETSKY, D.(1982) Paths and Categories. PhD Dissertation, MIT.
PESETSKY, D. (1995) Zero Syntax. Cambridge, MA: MIT Press.
REINHART, T. (2000) The Theta system: Syntactic Realization of Verbal Concepts. OTS
working papers in linguistics. to appear in the LI Monographs Series, MIT Press.
REINHART, T. (2002) The theta system-an overview. Theoretical Linguistics. 28, pp.229-290
SCHOORLEMMER, M. (1993) Dative subjects in Russian. In: TOMAN, J. (ed.) Formal
Approaches to Slavic Linguistics, The Ann Arbor Meeting, 129-172.
SCHOORLEMMER, M. (1995) Participial Passive and Aspect in Russian. Ph.D. Dissertation,
Utrecht University.
TORREGO, E. (1998) The dependencies of objects. Cambridge, Mass.: MIT Press.
ZAENEN, A. (1993) Unaccusativity in Dutch: integrating syntax and lexical semantics. In:
PUSTEJOVSKY, J. (ed.) Semantics and the lexicon 129-161. Dordrecht: Kluwer.
Zhanna Glushan, University of Connecticut, Department of Linguistics, 337

Mansfield Road, Storrs, CT 06269-1145, USA, Zhanna.glushan@uconn.edu
Inductive vs. Non-Inductive Generics in Russian and
Bulgarian
Elena Gorishneva
1. Introduction
This paper addresses the issue of the expression of inductive and non-inductive
genericity in two Slavic languages, Bulgarian and Russian. The topic has been
widely discussed in the literature, mainly for such languages as English and
French, in which the type of generic statements correlates with the use of bare
plurals and indefinite singulars. The main goal of the paper is to show that both
kinds of genericity, inductive and non-inductive, can be encoded in Bulgarian
and Russian not only in the nominal phrase, but also by means of aspectual fea-
tures of the verb. I will discuss the parallels between the verbal and nominal
domains related to the type of generic sentences.
1.1 Theoretic background

The difference between two types of generics inductive vs. non-inductive
was originally noticed by LAWLER (1973) whereby he distinguishes between ac-
cidental generalizations which are based on a set of observations (cf. 1) and
statements which express essential properties of entities (cf. 2).
(1) a. *A madrigal is popular. b. Madrigals are popular.
(2) a. A madrigal is polyphonic. b. Madrigals are polyphonic.
As can be seen in the examples above, these two kinds of generics show distri-
butional difference. Whereas bare plurals can be used to describe essential and
accidental properties, the appearance of indefinite singulars is restricted to es-
sential generalizations.
Besides the formal distinction, generics with bare plurals and indefinite sin-
gulars also yield semantically different interpretations. The first kind represents
descriptive generalizations which are based on a sufficient number of observed
instances. The second kind of generalizations labelled in the literature as
definitional, accidental or in virtue of generics cannot be inferred induc-
tively but is supported by some causal relation (cf. CARLSON 1995, COHEN 2001,
GREENBERG 2003, DOBROVIE-SORIN 2003, MARI 2008, KRIFKA 2009, among
others). Different accounts have been proposed in order to capture the formal
and semantic distinction between these types of generic statements. In the next
sections we briefly introduce two of the leading proposals.
82 Elena Gorishneva
1.2 GREENBERGs proposal (2002, 2003)

The core idea by GREENBERG (2002, 2003) is that the distinction between bare
plural generics and indefinite singular generics in English can be captured by as-
suming different accessibility relations for these types of generalizations, which
have the same semantics involving modal quantification. Greenberg contends
with her enough-presupposition that descriptive generalizations (e.g. Boys
dont cry) hold in all worlds that are maximally similar to our world and are
based on the conclusion that we can draw from a sufficient number of instances
of individuals/events in the actual world, cf. (3):
(9) w[w Inrmax (w.I) x,s [[P(x,w) C(s,x,w)] Q(s,x,w)]]],
where Inrmax (w,I) is the set of inertia worlds that continue w and interval I
in a normal way, i.e. worlds in which the course of events develops in a
way most compatible with the course of events in w.
By way of contrast, in virtue of generalizations (e.g. A boy doesnt cry) are not
supported by a sufficient number of realizations in the actual world. They are in-
ferred in virtue of some specific property associated with the property denoted
by the indefinite singular, that is, the generalization holds in virtue of this prop-
erty. In this kind of generic statements, worlds w are restricted by some modal
base (e.g. epistemic, deontic, stereotypical).
(10) Sw[x,s [[P(x,w) C(s,x,w)] [S(x,w) S is associated with P]]
x[[P(x,w)] Q(s,x,w)]]], where C is a set of contextually relevant
situations which includes x in w.
The representation of the sentence A boy doesnt cry is in (5):
(5) Sw[x,s [[BOY(x,w) C(s,x,w)] [S(x,w) S is associated with
BOY (x)]] x[[BOY (x,w)] nCRY(s,x,w)]]], where S is the prop-
erty being tough.
However, the proposed account is not entirely unproblematic. MARI (2008) ob-
jects that the assumption of an associated property does not always yield correct
results, cf. (6):
(6) Un leader violent est dangereux.
A violent leader is dangerous.
Following GREENBERG (2003), we have to assume a property which is associ-
ated with a violent leader and due to which a violent leader should be danger-
ous. This is not an intended reading of (6) in which the explicitly expressed
property of being violent is a cause of danger. The second option, that is, to con-
sider the property violent as an associated property in virtue of which the gener-
alization holds, is also ruled out: the derived interpretation would be that all
leaders are violent by nature.
Inductive vs. Non-Inductive Generics in Russian and Bulgarian 83
Krifka (2009) points out another shortcoming: the representation of in vir-

tue of generics is still insufficient and the domain of worlds w has to be further
restricted. In (5) it is still logically possible that a boy that satisfies the associ-
ated property S actually does cry in an appropriate situation.
1.3 KRIFKAs proposal (2009)

In his proposal, Krifka considers non-inductive generalizations (i.e. in virtue
of generics in terms of GREENBERG (2002, 2003)) as second order predications
about the meaning of sentences. The main idea of his account is that descriptive
(inductive) generics and definitional (non-inductive) generics have fundamen-
tally different representations. Whereas descriptive generalizations are based on
modal quantification, definitional generics make statements about the meaning
of expressions and how they can be used (KRIFKA 2009:3). In contrast to de-
scriptive statements, they are not quantificational.
The representation of the definitional statement (7) is given in (8):
(7) A greenhorn takes a raccoon for an opossum.
(8) ntake_raccoon_for_opossum (â_greenhorn), where n stands for the lifting
of predicate to a second order predicate on properties.
As we can see, definitional generics are predications made about an intension. In
sentences with an overt quantificational adverb, we deal with the quantification
over contextually relevant situations but not over individuals. See (9) and its
representation in (10):
(9) A boy never cries.
(10) nx[always(s: s is a potential crying situation x is in s; [x cries in
s])] (â_boy)
According to KRIFKA (2009), the adverb doesnt quantify over boys as in de-
scriptive generics but over potential crying situations.
The proposed account explains why indefinite singulars are preferred in
definitional generics. Since definitional sentences provide criteria for the identi-
fication of x as a P, the singular is an optimal option because in the prototypical
case x is a singular object. The categorizing use of definitional generics is as
predicted as they express essential properties of an object.
2. Present study
2.1 Goals and choice of languages
The goal of the present study is to examine the expression of inductive vs. non-
inductive genericity in Bulgarian and Russian. The choice of these languages
has been motivated by the following facts. Firstly, the verbal systems of both
84 Elena Gorishneva
languages manifest the imperfective vs. perfective aspect opposition which en-
ables us to investigate the contribution of aspectual properties of verbs to the in-
terpretation of generic sentences. Secondly, Russian and Bulgarian display dif-
ferent patterns in the nominal domain: the article-less language is opposed to the
language with an incomplete article system (with the postponed definite article).
In the literature about aspect in Slavic languages, it has been observed that
the imperfective aspect is associated with habitual readings (i.e. real iterative,
based on repetition of events) and the perfective aspect with gnomic and generic
events (i.e. unreal and modalized iterative) (cf. LYONS 1977, FIELDER 1993,
STAMBOLIEVA 2008). Habitual aspect is an appropriate option when the validity
of generalizations is based on our knowledge of what is usually the case, that is,
on facts, whereas gnomic sentences tend to be concerned with matters of opin-
ion, rather than fact (LYONS 1977:681). When an event is conceptualized as
real, it is likely to be quantified, and explicit quantification is consistent with the
imperfective aspect, whereas explicit qualification is compatible with the perfec-
tive aspect.
The main question which is the focus in the present investigation is whether
there is a correlation between the form of the NP and the aspectual characteris-
tics of the verb in inductive and non-inductive generic sentences.
2.2 The data

Before discussing inductive vs. non-inductive generalizations in Bulgarian, it
should be mentioned that Bulgarian posits the definiteness requirement for the
expression of genericity, which is well known for Romance languages, Hungar-
ian or Greek. The definite postponed article has to be used with kind predicates
and in characterizing sentences. This requirement holds for count as well for
mass nouns, cf. (11), (12):
(11) Mamutite / *mamuti sa izmrjali.
mammoths-DEF / *mammoths are extinct
Mammoths are extinct.
(12) Zlatoto / *Zlato e cenen metal.
gold-DEF / gold is valuable metal
Gold is a valuable metal.
In descriptive generalizations, which are inferred inductively, definite plural
NPs and the imperfective aspect is an appropriate option (13):
(13) enite gotvjat vseki den.
Woman-DEF.PL cook-PRES.IMPF.3PL every day
Women cook every day.
Interestingly, the use of perfective verbs which occur in the future tense with an
auxiliary te be-FUT indicates the non-inductive reading of generalizations:
(14) Dentlment vinagi te otvori vratata na one woman

gentleman-DEF always be-FUT open-PF. door-DEFfor edna ena.
A gentleman will always open the door for a woman.
Moreover, the non-inductive generics allow the substitution of the definite arti-
cle by the NP headed by edin one used as an indefinite article, cf. (15):
(15) Istinskijat prijatel/edin istinski prijatel vinagi te pomogne.
true-DEF friend/one true friend always be-FUT help-PF.3SG
A true friend will always help.
Although the use of edin one in the function of an article is rather restricted in
Bulgarian, it occurs felicitously in generic statements which are not based on the
inductive reference and express rules, stereotypes etc. (cf. (16), (17)):
(16) Edna ena vinagi te nameri vreme za decata si.
one woman always be-FUT find-PF.3SG time for children REFL
A woman will always find time for her children.
(17) Edin m trjabva da izchranva semejstvoto si.
one man must COMP feed family-DEF REFL
A man must be able to feed his family.
As for generalizations in Russian, it has been found that expressions with bare
plural NPs and imperfective verbs receive preferably an inductive (descriptive)
reading, while singular perfective sentences express non-inductive generaliza-
tions. See (18) and (19) below:
(18) Druzja vsegda pomogajut.
friends always help-IMPF
Friends always help.
(19) (Nastojaij) drug vsegda pomoet.
(real) friend always helps-PF
A (real) friend will always help.
The correlation between inductive vs. non-inductive generalizations and the
nominal and aspectual domains has been found in both languages under discus-
sion. In Bulgarian as well in Russian, the imperfective aspect is appropriate in
descriptive generics which are based on a sufficient number of actualized in-
stances, whereas the perfective aspect can be felicitously used in non-inductive
generics which do not need to be supported by the actualization of entities and
have another underlying causal relation. In both languages, plural NPs usually
occur in descriptive statements and the singular form oa bare noun in Russian
and an indefinite NP headed by edin one in Bulgarian o is preferred non-
inductive (or definitional) generics. Before discussing the observed correlation,
we should take a closer look at the properties of the imperfective and perfective
generics in Russian and Bulgarian.
86 Elena Gorishneva
2.3 Properties of generics

It has been determined that perfective and imperfective generics display differ-
ent properties.
First, they differ with regard to tolerating exceptions. In contrast to imper-
fective generics, perfective generics are not compatible with such markers of
habituality like often or usually, which indicate the possibility of exceptions.
Perfective generics are generalizations about essential properties and do not tol-
erate exceptions. See Russian examples (20) and (21) below. The same holds for
Bulgarian.
(20) *(Nastojaij) drug asto / obyno pomoet.
(real) friend often / usually helps-PF
(21) Druzja asto / obyno pomogajut.
Friends often / usually help-IMPF
Second, another piece of evidence for the claim that perfective generics cannot
express inductive truths is provided by the observation that absurd or weird gen-
eralizations which by nature cannot be inferred from any associated property re-
quire the plurality of NPs and the imperfective aspect in the verbal domain. Ab-
surd generalizations can only receive a descriptive interpretation, that is, the
conclusion is based on observed regularities (cf. (22), (23) in Russian and Bul-
garian).
(22) Norveskie student s familiej na T nosjat
Norwegian students with name PREP T wear-IMPF.3PL
zelenye galstuki.
green ties
Norwegian students whose names begin with T wear green ties.
(23) Studenti(te), iito imena zapova s T, nosjat vrtovrzki.
Students-(DEF) whose names begin with T wear-IMPF.3PL ties
Students whose names begin with T wear ties.
The next distinction between perfective and imperfective generalizations con-
cerns different interpretations of expressions and connected causal relation. We
can clarify this issue by the following examples. In (24a) the sentence contains
imperfective verbs and the derived reading is a habitual statement about Maria.
This generalization in turn gives rise to the conclusion in (24b).
(24) a. Maria dom ubiraet-IMPF, vkusno gotovit-IMPF, gostej kormit-IMPF.
Maria cleans her house, cooks well, serves guests food.
b. Maria choroaja chozjajka.
Maria is a good housewife.
The substitution of imperfective verbs by perfective ones leads to a different in-

terpretation. Whereas in (24a) we deal with assigning properties to a subject NP
(i.e. a descriptive reading), (25b) describes potential properties of an entity. The
statement in (25b) is not based on observed facts but is related to some back-
ground information which enables the speaker to make a prediction about the
behaviour of Maria. In other words, the background knowledge in (25a) makes it
possible to speak about potencies, cf. (25a, b):
(25) a. Maria choroaja chozjajka.
Maria is a good housewife.
b. Ona i dom uberet-PF, i vkusno prigotovit-PF, i gostej nakormit-PF.
She cleans her house, cooks well, serves guests food.
In short, the imperfective sentence (24a) expresses a generalization about actual
properties of an entity and that is what causes the conclusion in (24b). The rela-
tion between the perfective sentence (25b) expressing potential properties of an
entity and the utterance (25a) is different: the statement in (25a) provides a basis
on which the prediction in (25b) can be made.
Although the order of sentences in (24) and (25) exemplifies the most natu-
ral sequence of expressions, this order has been chosen in the paper mainly for
the simplified illustration of the distinction discussed. The ordering of sentences
can vary in a natural language; the different causal connection, however, re-
mains. This can be proved by a simple test adding an explicit causal connector
potomu to because, cf. (26a, b):
(26) Maria choroaja chozjajka, potomu to...
a. ona ubiraet-IMPF, vkusno gotovit-IMPF, gostej kormit-IMPF.
b. *ona uberet-PF, vkusno prigotovit-PF, gostej nakormit-PF.
Maria is a good housewife because she cleans her house, cooks well,
serves guests food.
While the imperfective expression can be felicitously introduced by potomu to
because, the use of the perfective sentence shows the incompatibility with the
causal marker.
Another issue related to different readings of perfective and imperfective
generics should be noticed here. In descriptive habitual statements, the imperfec-
tive verb is used in the present tense. In utterances expressing predictions about
potential properties, the perfective verb in Russian is future-oriented. I will re-
frain here from discussing different views on the Russian tense paradigm of per-
fective verbs. The issue concerning the lack of a present temporal interpretation
of perfective verbs should be further investigated. However, the data from other
languages can be taken in favour of the claim that perfective generalizations are
related to a future reference (cf. RUDNICKA-MOSIDZ (2004) and KLIMEK-
JANKOWSKA (2008) for the perfective aspect in Polish generics, HANY BABU
88 Elena Gorishneva
(2006) for the use of the future tense marking modal particle -um in non-
inductive statements in Malayalam). The striking example is provided by Bul-
garian, in which perfective verbs are explicitly used in the future tense marked
by the auxiliary te be-FUT in non-inductive generalizations. Consider exam-
ples (14), (15) and (16) (repeated as (27) here) from the previous section:
(27) Edna ena vinagi te nameri vreme za decata si.
one woman always be-FUT find-PF.3SG time for children REFL
A woman will always find time for her children.
Another context in which the use of the perfective aspect is appropriate is pro-
vided by utterances expressing dispositions like (28):
(28) Sachar rastvoritsja v vode.
sugar dissolves-PF in water
Sugar will dissolve in water.
It should be pointed out that all expressions in which perfective verbs can occur
have in common that they do not allow exceptions and are not based on quanti-
fication. The question that arises with respect to this observation is the way in
which the semantics of the perfective aspect contributes for this use.
3. Perfective aspect in generics

In comparison with the imperfective aspect, the perfective aspect represents a
situation in its entirety, that is, a whole (completed) indivisible situation. The
imperfective aspect as an unmarked form of the aspectual opposition is neutral
to the feature completeness.
The use of the perfective aspect signals the conviction of the speaker. This
certainty is based on some background information which enables the speaker to
use perfective verb forms. Let us consider two episodic sentences in the follow-
ing examples:
(29) Peter napisal pismo.
Peter wrote-PF letter
Peter wrote a letter.
(30) Peter pridet segodnja na veerinku.
Peter comes-PF today PREP party
Peter will come to the party today.
In order to choose the perfective past form in (29), the speaker has to be sure
that the event described has been finished, that is, that the letter is written; oth-
erwise the imperfective form should be used. The use of the perfective form in
(30) signals that the speaker has evidence that Peter will come to the party, for
instance, in the case that Peter confirmed his attendance at the party.
In (31) below, the sentence predicts that Peter will easily pass his exams.
The prediction should be grounded in some background knowledge of the
speaker, for example, when the speaker knows that Peter prepared well for his
exams or that Peter is an intelligent student that always passes exams or has
passed so far.
(31) Peter sdast exameny bez problem.
Peter passes-PF exams without problems
Peter will easily pass exams.
Perhaps, another piece of evidence for the claim that the perfective aspect is
linked to the speakers conviction can be taken from negative imperatives. In
Slavic languages, the use of perfective aspect is restricted to preventive impera-
tives, whereas the imperfective aspect occurs in prohibitive imperatives. Con-
sider (32) and (33):
(32) Smotri, ne opozdaj na lekciju!
Look not come-IMPF late PREP lecture
Be careful, dont come late to the lecture!
(33) Ne opazdyvaj na lekcii!
not come-PF late PREP lectures
Dont come late to lectures!
In order to prevent someone from doing something, the speaker must have a rea-
son, that is, she/he knows the relationship between the action from which the
hearer will be prevented and its consequences. The prohibitives do not posit
such a requirement: one can prohibit someone doing something without a certain
reason, in other words, the speaker does not have to explain or justify a prohibi-
tion.
Similar considerations about the link between the use of the perfective as-
pect and the conviction of the speaker expressed in episodic sentences hold for
generalizations. In perfective habituals the speakers certainty relies on back-
ground information; in generics like (33) the conviction about the truth of the
statement expressed is linked to the definition of a real friend. That means that a
true friend is defined by the essential property helpfulness. Thus, perfective ge-
nerics are compatible with always but not with usually or often.
(33) Nastojaij drug vsegda pomoet.
real friend always helps-PF
A real friend will always help.
In statements about dispositions, the conviction of the speaker is based on
knowledge about essential properties of entities.
All the generic contexts in which the perfective aspect can be successfully
used have in common that the generalizations are not inferred inductively, that
is, not concluded from a series of observed instances rather in a deductive way:
90 Elena Gorishneva
based on the knowledge about the meaning of entities. For instance, the sentence
in (33) relies on the speakers knowledge about the meaning of an idiom a real
friend. In order to produce perfective statements like in (25b), the speaker draws
the conclusion from information about Maria (she belongs to the class of good
housewives) and the definition of a good housewife. Thus, perfective generics
are related to definitions of entities, which include information of essential prop-
erties of these entities. Such statements do not add any new information to our
knowledge about real friends, good housewives or gentlemen; they are not
meaning-expanding in comparison to inductive generalizations, which introduce
new information about objects. In non-inductive expressions, the meaning of a
subject NP (e.g. a real friend) includes the meaning of a predicate (e.g. to be
helpful).
Generics in which perfective verb forms can be used also display properties
observed for definitional statements in well-investigated languages like English
or French. KRIFKA (2009) remarks on the observation of BURTON-ROBERTS
(1976) that non-inductive generalizations but not descriptive ones allow a to
be-paraphrase, cf. (34a, b):
(34) a. A madrigal is polyphonic. To be a madrigal is to be polyphonic.
b. Madrigals are popular. *To be a madrigal is to be popular.
This transformation is explained by KRIFKA (2009:5) as follows: The second-
order predication fixes a meaning of an expression such that whenever
something falls under , it also falls under .
The to be-paraphrase holds for noun phrases like a well-behaved child, a
true friend and so forth, which are idioms with a non-compositional meaning in
contrast to modified nouns (e.g. a sick child, a long-time friend), with a meaning
derived compositionally. The former are felicitous under a definitional reading
while the latter do not have this interpretation, as illustrated in (35a, b):
(35) a. A well-behaved child always says please and thank you.
To be a well-behaved child is to say please and thank you.
b. A sick child is moody.
*To be a sick child is to be moody.
Whereas in (35a) the predicate meaning is included in the meaning of a subject
NP a well-behaved child, it does not hold for (35b), in which the VP predicate
does not express an essential property of the predicate a sick child and adds new
information about it.
4. Conclusion
The following tendencies have been investigated so far in Bulgarian and Rus-
sian. Firstly, imperfective aspect and plural noun phrases constitute an appropri-
ate and preferred option in inductive (or descriptive) generalizations which are
based on a series of observed instances. Secondly, non-inductive (or defini-

tional) generics provide a suitable context for the use of the perfective aspect
and singular NPs. The most interesting observation concerns the nominal do-
main in Bulgarian in which non-inductive generics can overrule the definiteness
requirement and allow the use of an indefinite NP headed by edin one.
From the cross-linguistic view, inductive vs. non-inductive generics can be
encoded in a different way in languages, which is shown in the graph below.
(36) The expression of inductive vs. non-inductive genericity
NOMINAL DOMAIN VERBAL DOMAIN
English Russian
Bulgarian
French Polish
In article languages like English and French, there are differences in the nominal
domain which reflect the distinction between two kinds of generalizations.
Therefore, whereas bare plurals receive a descriptive reading, indefinite singu-
lars appear in definitional generic statements. In article-less languages like Rus-
sian and Polish (cf. RUDNICKA-MOSIDZ 2004, KLIMEK-JANKOWSKA 2008), the
observed distinction can be expressed by means of the aspectual value of verbs.
Thus, the imperfective aspect is felicitous in both kinds of generics and obtains
preferably a descriptive interpretation, while the use of the perfective aspect in-
dicates a non-inductive inference, in other words, it fulfils a function similar to
that of indefinite singulars in article languages. Bulgarian occupies the position
in between to the extent that it uses both options for the expression of different
kinds of generics: the distinction can be marked by means of the form of a noun
phrase, similar to English and French as well as with aspectual properties of
verbs, parallel to Russian or Polish.
References:
Burton-ROBERTS, N. (1976) On the generic indefinite article. Language 52. 427-448.
CARLSON, G. (1995) Truth conditions of generic sentences: two contrasting views. In: CARL-
SON, G. & F. PELLETIER (eds.) The generic book 224-237. Chicago, London: The Uni-
versity of Chicago Press.
COHEN, A. (2001) On the generic use of indefinite singulars. Journal of Semantics 18. 183-
209.
DOBROVIE-SORIN, C. (2003) Generic indefinites and (un)selective binding. In: STARK, E.& U.
WANRUSZKA (eds.) Syntaxtheorien. Modelle, Methoden, Motive 35-52. Tbingen.
FIELDER, G. (1993) The semantics and pragmatics of verbal categories in Bulgarian. Lewis-
ton: Edwin Mellen Press.
GREENBERG, Y. (2002) Two types of quantificational modalized genericity and the interpreta-
tion of bare plurals and indefnite singulars NPs. In: JACKSON B. (ed.) Proceedings of
SALT XII. 104-123.
92 Elena Gorishneva
GREENBERG, Y. (2003) Manifestations of genericity. London: Routledge.

HANY BABU, M. T. (2006) Genericity, quantification, and modality: The many faces of um
and unnu in Malayalam. Semantics Archive.
KLIMEK-JANKOWSKA, D. (2008) On analogy between English nominal and Polish aspectual
manifestations of genericity. In: A. Grnn (ed.) Proceedings of SuB12 318-336. Oslo.
KRIFKA, M. et al. (1995) Genericity: an introduction. In: CARLSON C. & F. PELLETIER (eds.)
The Generic Book 1-124. Chicago, London: The University of Chicago Press.
KRIFKA, M. (2009) Definitional generics as second-order predications. Talk given at the
GENIUS conference. Paris. May 11-13, 2009.
LYONS, J. (1977) Semantics. 1&2. Cambridge: Cambridge University Press.
MARI, A. (2008) Analyticity under perspective: indefinite generics in French. In: A. GRNN
(ed.) Proceedings of SuB12 414-429. Oslo.
RUDNICKA-MOSIDZ, E. (2004) The role of the perfective aspect in the interpretation of NPs.
In: Pozna Studies in Contemporary Linguistics 39: 143-153. Pozna{ Adam
Mickiewicz University School of English.
STAMBOLIEVA, M. (2008) Building up aspect. A study of aspect and related categories in Bul-
garian, with parallels in English and French. In: DAVIS G. & K. A. BERNHARDT (eds.)
Contemporary studies in descriptive linguistics, vol. 6. Bern: Peter Lang.
Elena Gorishneva, Humboldt University Berlin, Department of Slavic Studies,

Unter den Linden 6, 10117 Berlin, Germany, jelenagor@gmx.de
A Compositional Semantics for Comitative Constructions
Beata Trawiski
1. Introduction
Comitative Constructions (CCs) are expressions consisting of two NPs con-
nected by a comitative preposition (CP) such as the English with, German mit,
French avec, Portuguese com, Spanish con, Russian s, or Polish z, and are at-
tested for Slavic and many other languages of the world. It has been observed
that there are several types of CCs, and different criteria have been proposed in
the literature to distinguish between them. Previous approaches to CCs mainly
focus on syntactic aspects and attempt to explain the differences between the
particular CC types by dint of syntactic means (partially, with a semantic or
pragmatic component).
In this paper, I argue that the differences between the particular types of
CCs are primarily semantic in nature, and I present a purely semantic typology
of CCs, which draws on the typology in TRAWISKI (2009). Based on this typol-
ogy, I propose a compositional semantic analysis which uniformly accounts for
all interpretations available for CCs. Since CCs have a particularly high fre-
quency in Slavic languages such as Polish, where all types of CCs previously
discussed in the literature appear, Polish data are taken into consideration.
In Section 2, some basic properties of Polish CCs are discussed and the semantic
typology of these expressions is presented. Section 3 provides empirical evi-
dence for this typology. Section 4 addresses some morphosyntactic issues that
have been used in the literature as the motivation for assuming different syntac-
tic structures for CCs. I show that, in fact, there is no purely syntactic evidence
for distinguishing between the particular types of CCs, at least not in Polish. In
Section 5, I discuss some of the most prominent approaches to CCs and show
their advantages and shortcomings. In Section 6, I present my semantic analysis.
Finally, Section 7 sums up the discussion.
2. Basic Properties and the Semantic Typology of CCs
The Polish sentence in (1) contains a typical CC, which consists of the NP Jan
Jan and the NP Mari Maria connected by the comitative preposition z
with.
(1) Jan (razem) z Mari wyjecha+ / wyjechali.
Jan together with Maria.INSTR left.SG / left.PL
Jan left (together) with Maria. / Jan and Maria left (together).
In contrast to FELDMAN (2002), who analyzes the Russian expression s in cor-
responding constructions as a transitive noun, I analyze the Polish z as a preposi-
tion. There is strong evidence for this assumption. Firstly, z undergoes a vocalic
94 Beata Trawi{ski
alternation, which is a phenomenon typical of prepositions, but not of nouns.

Secondly, expressions headed by z can be modified by collectivizing adverbs
such as razem together, which is possible for PPs but not for NPs. This ability
is illustrated in (1). This example further demonstrates that CCs can combine
with singular or plural predicates and that the second NP is assigned instrumen-
tal case. Finally, a CC introduces comitative content, which is understood in a
very broad sense as a joint participation of the individuals referred to by the first
NP (NP1) and the second NP (NP2) in the event referred to by the predicate.
Based on the lexicosemantic relationship between NP1 and NP2, I argued in
TRAWISKI (2009) that there are three such relationships. Consider the sentences
in (2), (3) and (4).
(2) Jan z Mari wyjecha+.
Jan with Maria.INSTR left.SG
Jan left with Maria.
(3) Jan z Mari wyjechali.
Jan with Maria.INSTR left.PL
Jan and Maria left.
(4) My z Mari wyjechali`my (jako para ma+\e{ska).
We with Maria.INSTR left.1st.PL as couple married
Maria and I left (as a married couple).
The relationship between NP1 and NP2 in (2) is clearly such that the individual
referred to by NP2 accompanies the individual referred to by NP1 in the event
referred to by the predicate. By contrast, in (3), the individuals referred to by
NP1 and NP2 are equal participants involved in the event of leaving. So, they
function as conjuncts here and bear the same thematic roles. Finally, the rela-
tionship between NP1 and NP2 in (4) is such that the referent of NP2 is included
in the set of referents of NP1. Thus, the meaning of the first person plural pro-
noun my we includes the referent of Maria and the speaker, and, thus, carries
the meaning Maria and I.
On the basis of the data in (2), (3) and (4), I define three classes of CCs: ac-
companitive comitative constructions (ACCs), conjunctive comitative construc-
tions (CCCs) and inclusive comitative constructions (ICCs). This distinction be-
tween the particular interpretations of CCs can be made more precise using sets,
which is done in (5), (6) and (7).
(5) [[ACC]] A, B, where A B =
(6) [[CCC]] = (A B)
(7) [[ICC]] = A, where A B
According to (5), the denotation of the ACC where NP1 denotes a (possibly sin-
gleton) set A of individual entities and NP2 denotes a (possibly singleton) set B
A Compositional Semantics for Comitative Constructions 95
of individual entities includes A and B, where A and B are disjoint sets. (6) indi-
cates that the denotation of the CCC, where NP1 denotes a (possibly singleton)
set A of individual entities and NP2 denotes a (possibly singleton) set B of indi-
vidual entities, is the union of A and B. Finally, (7) says that the denotation of
the ICC where NP1 denotes a set A of individual entities and NP2 denotes a
(possibly singleton) set B of individual entities equals A, where A is a proper
superset of B.
In the next section, I provide empirical evidence for this typology by ex-
amining the particular CC classes with regard to presuppositional effects, con-
trastive focus assignment, plural semantics and coreferential properties.
3. Empirical Evidence for the Typology of CCs
Consider the sentence in (8) uttered before and after May 2004. Recall that until
May 2004 the European Union (EU) involved 15 member states: Belgium,
France, Germany, Italy, Luxembourg, the Netherlands, Denmark, the Republic
of Ireland, the United Kingdom, Greece, Portugal, Spain, Austria, Finland and
Sweden. A major enlargement took place on May 1, 2004, when 10 further
states joined the EU, namely Cyprus, the Czech Republic, Estonia, Hungary,
Latvia, Lithuania, Malta, Poland, Slovakia and Slovenia.1
(8) Ca+a UE z Polsk I S+oweni skorzysta na rozszerzeniu
Whole EU with Poland and Slovenia will profit.SG on enlargement
strefy euro.
zone euro
T1: The whole EU will profit together with Poland and Slovenia from the
enlargement of the euro zone.
P1: 17 countries will profit from the enlargement of the euro zone.
(uttered before May 1, 2004)
T2: The whole EU, including Poland and Slovenia, will profit from the
enlargement of the euro zone.
P2: 25 countries will profit from the enlargement of the euro zone.
(uttered after May 1, 2004)
As the translations in T1 and T2 indicate, the sentence in (8) is ambiguous be-
tween the inclusive and the accompanitive readings. Furthermore, this sentence
triggers two different presuppositions with respect to the number of states predi-

1 On January 1, 2007, the most recent enlargement of the EU took place, involving Bulga-
ria and Romania. Hence, there are currently 27 member states in the EU. In this regard,
the sentence in (8) can trigger a third presupposition concerning the number of countries
predicated, i.e. that 27 countries will profit from the enlargement of the euro zone. For
the present discussion, this presupposition does not play a decisive role. In fact, it corres-
ponds to the presupposition given in P2. I, therefore, ignore the third presupposition in
the discussion below.
96 Beata Trawi{ski
cated in this sentence, which is indicated in (8) by means of P1 and P2. Thus, in
the situation before May 1, 2004, i.e. at the time when the EU consisted of 15
member states, and did not include Poland and Slovenia, the predication in (8)
applies to 17 states, i.e. to all member states of the EU at that time, as well as to
Poland and Slovenia. Under this interpretation, the sentence triggers the presup-
position indicated by P1. By contrast, after May 1, 2004, since the EU has 25
members, including Poland and Slovenia, the sentence in (8) refers to 25 coun-
tries, and carries the presupposition indicated in P2. Given the presuppositions
in P1 and P2, two possible denotations can be assumed to be available for the
expression caa UE z Polsk i Soweni. These two denotations clearly corres-
pond to ACCs and ICCs.
Denotational differences between ACCs, on the one hand, and CCCs and
ICCs, on the other hand, can be demonstrated by contrastive focus assignment.
DYA (1988) has observed that in ACCs, NP1 and the PP can bear contrastive
stress independently of each other, as in (9), and in CCCs, NP1 and the PP must
have the same information status, which is demonstrated by the examples in
(10). DYA & FELDMAN (to appear) also discuss ICCs with this respect and ob-
served that ICCs behave similarly to CCCs. This is illustrated in (11).
(9) Contrastive Focus Assignment in ACC
a. JANEK z Ew poszed+ na spacer.
Janek with Ewa went.SG for walk
It was Janek who went for a walk with Ewa.
b. Janek Z EW poszed+ na spacer.
Janek with Ewa went.SG for walk
It was with Ewa that Janek went for a walk.
(10) Contrastive Focus Assignment in CCC
a. *JANEK z Ew poszli na spacer.
Janek with Ewa went.PL for walk
b. *Janek Z EW poszli na spacer.
c. JANEK Z EW poszli na spacer.
It was Janek and Ewa who went for a walk.
d. Janek z Ew poszli na spacer.
Janek and Ewa went for a walk.
(11) Contrastive Focus Assignment in ICC

a. *MY z Mari wyjechali`my.
we with Maria left.1st.PL
b. *My Z MARI wyjechali`my.
c. MY Z MARI wyjechali`my.
It was Maria and I who left.
d. My z Mari wyjechali`my.
Maria and I left.
These facts show that while NP1 and the PP may differ in information status in
ACCs, they must have the same information status in CCCs and ICCs. The issue
of contrastive focus assignment has also been discussed in IONIN & MATU-
SHANSKY (2003) for Russian.
In the examples in (12) and (13), I examine the three types of CCs with respect
to their ability to occur in collective and distributive contexts. In (12) a collec-
tive context is considered, provided by the collective predicate spotyka si
meet, which only contains groups of individuals in its denotation.
(12) a. *Jan z Mari spotyka si u rodzicw. ACC
Jan with Maria meets RM at parents
b. Jan z Mari spotykaj si u rodzicw. CCC
Jan with Maria meet.PL RM at parents
Jan and Maria meet at their parents.
c. My z Mari spotykamy si u rodzicw. ICC
we with Maria meet.1st.PL RM at parents
Maria and I meet at our parents.
As these examples show, CCCs and ICCs can occur in collective contexts. By
contrast, the sentence involving the ACC is ungrammatical. The same observa-
tions can also be made for the sentences involving other collective predicates,
such as form a circle / line, be numerous, or surround.
The examples in (13) illustrate the occurrence of ACCs, CCCs and ICCs in
sentences providing distributive contexts. The distributive context is introduced
here by the distributive predicate by dumnym to be proud, which can only
have atomic individuals in its denotation and cannot apply to groups.
(13) a. *Jan z Mari jest dumny ze swoich dzieci . ACC
Jan with Maria is proud of PRN children
b. Jan z Mari s dumni ze swoich dzieci. CCC
Jan with Maria are proud of PRN children
Jan and Maria are proud of their children.
98 Beata Trawi{ski
c. My z Mari jeste`my dumni ze swoich dzieci. ICC

we with Maria are.1st proud of PRN children
Maria and I are proud of our children.
As these sentences demonstrate, only CCCs and ICCs can occur in distributive
contexts. Sentences involving other distributive predicates, such as believe, or
have blue eyes, distributive predicates beginning with the prefix roz-, such as
rozjecha si to go different ways, and other distributive expressions yield the
same results with respect to grammaticality.
The proposed semantic typology of CCs is further supported by coreference
phenomena. As the examples in (14) show, in ACCs, only NP1 can control ref-
lexive pronouns, while in CCCs and ICCs, entire subjects act as antecedents.
Note that the reference of the plural pronoun in (14c) is identified with the refer-
ence of the entire ICC, which is indicated by the identity of the indices.
(14) a. *[Jani z Marij]k kupi+ sobiei/*j/*k psa. ACC
Jan with Maria bought.SG PRN dog
Jan bought with Maria a dog for himself.
b. [Jani z Marij]k kupili sobie*i/*j/k psa. CCC
Jan with Maria bought.PL PRN dog
Jan and Maria bought themselves a dog.
c. [Myi z Marij]i kupili`my sobiei/*j psa. ICC
we with Maria bought.1st.PL PRN dog
Maria and I bought ourselves a dog.
Finally, ACCs differ from CCCs and ICCs by the ability of NP1 z NP2 se-
quences to act as controllers of PRO subjects of infinitival clauses. While these
sequences in CCCs and ICCs can act as antecedents of PRO subjects, they can-
not in ACCs.
The following table summarizes all the semantic properties that support evi-
dence for the proposed semantic typology of CCs.
Semantic Property ACC CCC ICC

ability of NP1 z NP2s to trigger inclusive pre- no no yes
supposition
ability of NP1 z NP2s to be assigned focus no yes yes
availability of plural denotation no yes yes
ability of NP1 z NP2s to act as controllers no yes yes
ability of NP1s to act of controllers yes no yes
In the next section, I address some morphosyntactic issues discussed in the lite-
rature, which have often been used as arguments for assuming different syntac-
tic structures for CCs.
4. Morphosyntactic Properties
Three phenomena have essentially motivated previous analyses of CCs in terms
of different syntactic structures: adjacency requirements, extraction and agree-
ment. DYA (1988) claims that, in contrast to ACCs, CCCs do not allow inter-
vening clitics and parentheticals between NP1s and z NP2s. He also claims that
while Polish ACCs allow extraction of comitative PPs out of entire subjects and
extraction of NP1s out of NP1 z NP2 strings, CCCs do not.
The problem with these claims is that in Polish accompanitive PPs can be at-
tached to both NP1s and VPs. This means that the parentheticals and the clitics
in ACCs can be analyzed as intervening either between the NP1 and the PP
within the entire subject or between the subject NP1 and the PP modifying the
VP. Since neither analysis can be excluded, one cannot argue that NP1 z NP2
strings in ACCs allow intervening elements. The same observation holds for in-
clusive PPs in Polish. Another argument against the adjacency requirement in
CCC is the fact that collectivizing adverbs can appear between NP1 and the PP.
Similarly, since comitative PPs in ACCs and ICCs can combine both with NP1s
and with VPs, it cannot be clearly determined whether PPs and NP1s are ex-
tracted out of the NP1 z NP2 expression or out of the VP.
Another issue discussed in the literature on CCs is person, number and
gender resolution, discussed for Polish in DYA (1988), MCNALLY (1993) and
TRAWISKI (2005). The observation is that in the ACC and ICC, only the NP1
triggers person, number and gender agreement with the predicate, the NP2 is not
involved in agreement and resolution. In the CCC, however, both the NP1 and
the NP2 participate in person, number and gender resolution and in agreement
with the predicate. In this respect, CCCs correspond to ordinary nominal coordi-
nation. However, agreement and resolution are not purely morphosyntactic phe-
nomena in Polish, but are strongly context-driven. Moreover, CZUBA & PRZE-
PIRKOWSKI (1995), PRZEPIRKOWSKI et al. (2002) and TRAWISKI (2005) have
shown that subject-predicate agreement in Polish is semantic in nature, in con-
trast to adjective-noun agreement, which is morphosyntactic. Accordingly, I ar-
gue against using agreement and resolution phenomena in CCs as empirical ba-
sis for syntactic stipulations.
To conclude, there is no explicit syntactic or morphosyntactic evidence that
would justify different syntactic structures for ACCs, CCC and ICCs. However,
this is what the majority of the previous approaches to CCs propose. In the fol-
lowing section, I give an outline of them.
100 Beata Trawi{ski
5. Previous Approaches to CCs

Among numerous approaches to CCs, there are four that attempt to account for
all three CC classes, i.e. VASSILIEVA & LARSON (2005), FELDMAN (2002), SKRA-
BALOVA (2003) and IONIN & MATUSHANSKY (2003). VASSILIEVA & LARSON
(2005) propose for Russian CCs with the CCC reading a syntactic structure that
corresponds to the structure of ordinary (flat) coordination, where the two NPs
and the comitative preposition each forms a separate constituent. CCs with the
ACC reading are assigned two structures: one where the comitative PP attaches
to a VP as its left adjunct, and one where it attaches to a VP or to NP1 as its
right adjunct. CCs with the ICC reading are syntactically analyzed as NPs head-
ed by the plural pronoun. This analysis also includes a semantic component:
plural pronouns are treated as derived from the reference of the corresponding
singular pronouns by the addition of another individual. The comitative PP is a
syntactic complement of the pronoun head and supplies the unspecified individ-
ual in the plural pronoun meaning.
FELDMAN (2002) provides a comprehensive analysis of Russian CCs in the
framework of Head-Driven Phrase Structure Grammar. She proposes treating
Russian ACCs in terms of adjunction, CCCs and ICCs in terms of complementa-
tion. The comitative preposition in CCs with the CCC reading is analyzed as a
transitive noun that selects for an instrumental NP (NP2) and a subject NP
(NP1). For CCs with the ICC interpretation, FELDMAN (2002) proposes a com-
plementation-based analysis as well. Similarly to VASSILIEVA & LARSON (2005),
she assumes that plural pronouns occurring in ICCs are heads selecting comita-
tive PPs as their complements.
SKRABALOVA (2003), who analyzes CCs in Czech, argues that there are at
least two syntactically distinct comitative structures in Czech: a comitative
coordination-based structure and an adjunction-based structure. For CCs with
the ACC reading, she proposes a syntactic structure according to which the com-
itative PP adjoins to the VP and is semantically associated with the subject NP,
which is either overt or covert. In ICCs involving plural pronouns, the comita-
tive PP is assumed to be adjoined to the subject NP. The plural pronoun can ei-
ther be overt or covert. Finally, for CCs with the CCC reading, SKRABALOVA
(2003) proposes a Conjunction Phrase (ConjP) analysis. She assumes that the
comitative morpheme within CCCs functions as a conjunction rather than a pre-
position and acts as the head of the entire ConjP. An alternative analysis sug-
gests the CCC is a ConjP involving a zero-conjunction. The comitative PP func-
tions here as its complement.
Finally, IONIN & MATUSHANSKY (2003), who discuss Russian CCs, propose
that the three types of CCs have the same base structure: the PP always origi-
nates as an NP-adjunct. They further assume that NP1 moves out to [Spec, IP].
The PP can either move with it, licensing CCCs, or be stranded / extraposed, li-
censing ACCs. Besides stranding / extraposition, a number of other factors, e.g.
information structure, are considered to play a role in licensing the different CC

readings.
To conclude, previous approaches which deal with different types of CCs
try to explain the differences between them by means of syntactic stipulations
(the analysis of IONIN & MATUSHANSKY (2003) is to some extent an exception).
I have argued above, however, that these constructions and the differences be-
tween them are semantic in nature. Moreover, some of the previous analyses
make wrong predictions. Particularly problematic are the coordination-based
analyses of CCCs proposed in VASSILIEVA & LARSON (2005) and SKRABALOVA
(2003)2. Firstly, coordination-based analyses do not explain how the case as-
signment to NP2 works. Secondly, inversion of NP1 and NP2, which is possible
in an ordinary coordination and not possible in CCCs, cannot be ruled out.
Thirdly, these analyses do not account for sentences where in addition to the
comitative preposition, an alleged conjunction, a proper conjunction is present.
Fourthly, these analyses do not explain why, in contrast to ordinary coordina-
tion, CCCs only allow nominal categories. Also, CCCs allow pro-drop, which is
not possible in ordinary coordination. CCCs also differ from ordinary coordina-
tion by not allowing Across-the-Board extraction, which is possible in Polish
coordination. Finally, by assuming a coordination-based analysis for CCCs, the
contrast in the distribution of possessive and possessive reflexive pronouns in
CCCs and ordinary coordination cannot be explained.
Taking this into account, I assume a single syntactic structure for ACCs,
CCCs and ICCs, and propose that the differences between them follow from
their semantic properties. Since coordination- and complementation-based ana-
lyses are problematic for a number of empirical and theoretical reasons, I adopt
the adjunction-based syntactic structure for CCs. This allows me to treat all
types of CCs in a uniform fashion with regard to their syntactic structure and,
thus, to explain all of the syntactic properties they share. In the following sec-
tion, I provide details of my analysis.
6. The Semantic Analysis
I assume that ACCs, CCCs and ICCs have the same syntactic structure accord-
ing to which each CC is an NP headed by an NP1 which is modified by the z-
PP. The PP is headed by the preposition z with which, in turn, selects for the
instrumental NP2. I further assume that CCs with this internal structure combine
with VPs as their subjects, as presented in Figure 1.3

2 Note that coordination-based analyses have also been proposed in DYA (1988) for
Polish CCCs and in COMACHO (1994) for Spanish CCCs.
3 For lack of space, I omit cases where comitative PPs combine with VPs.
102 Beata Trawi{ski
Figure 1: The syntactic structure of sentences involving CCs

I argued above that the differences between the particular types of CCs are se-
mantic in nature, and I assume that this semantic ambiguity is triggered by the
preposition z, which is semantically ambiguous in its comitative usage between
the accompanitive, conjunctive and inclusive readings. This ambiguity is li-
censed by three different translations of z, each of which provides different truth
conditions. I further assume that the remaining lexical components of CCs, in-
cluding plural pronouns, bear their ordinary meanings. To illustrate how this
analysis works, I apply it to the sentences in (2), (3) and (4), involving an ACC,
a CCC and an ICC respectively. Below, I provide the proposed basic transla-
tions for all expressions of our fragment of Polish.
(15) Jan {j}<e,t>
(16) Maria {m}<e, t>
(17) my `we` {i, }<e, t>
(18) wyjecha leave X<e, t>. E leave(E, X)
(19) wyjecha leave X<e, t>. E<v, t>. leave(E, X)
(20) z with (ACC) Y<e, t> X<e, t> P2<<e, t>, <<v, t>, t>>. E (P2(E, X) (X Y
) E ((E E` E` z ) o (P2(E`, X) E`àccompany (E``, Y,
X))))
(21) z with (CCC) Y<e, t> X<e, t> P1<<e, t>, t>.P1(X Y)
(22) z with (ICC) Y<e, t> X<e, t> P1<<e, t>, t>.Z (P1(Z) Z = X Z Y)
As the translations in (15) and (16) indicate, I analyze individual terms as denot-
ing subsets of the domain of individuals. Consequently, singular individual
terms denote singleton sets and plural individual terms denote sets containing
more than one individual. The denotation of conjoined singular terms is made up
of singleton sets by combining them via the set union operation. In (17), I pro-
vide the translation of the first person plural pronoun my, which is assumed to
denote a set of individual entities including the speaker (represented as i) and at
least one further individual. In (18), I give the translation of the predicate
wyjecha, which provides an obligatory argument position for events in addition
to the subject-argument. Similarly to individual arguments, I propose that event
arguments are sets of atomic events. The verb wyjecha receives an additional
translation in (19), according to which the event variable is bound by the -
operator. This translation is used in combinations with expressions which oper-
ate on its event argument. Finally, the three basic translations of the preposition
z are given in (20), (21) and (22), where X and Y are variables over sets of indi-
vidual entities, E is a variable over a set of event entities, and P is a predicate va-
riable. According to the semantic representation in (20), the accompanitive pre-
position z combines with predicates of the same type as (19). The truth condi-
tions provide an existentially bound event variable E and guarantee that the rela-
tion P (associated with the predicate) holds between the set of events
represented by this variable and the set of individuals denoted by the modified
NP (represented by X). Further, the truth conditions ensure that for each non-
empty subset of the set of events to which the predicate P applies, there is a set
of events such that the relation of accompaniment holds between this set of
events, the set of individuals denoted by the prepositional object, and the set of
individuals denoted by the modified NP. Also, the translation warrants that the
prepositional object and the modified NP have disjoint denotations. Putting it
more intuitively, each situation denoted by the predicate P is associated with an
event of accompaniment (applying to the referents of the PP-object and to the
referents of the modified NP respectively). Due to the disjoint denotation of the
nominal arguments and the fact that there is no single event in which the refe-
rents of these two NPs are involved, neither collective nor distributive interpre-
tation is possible.
The semantic representation of the conjunctive preposition z is provided in
(21). The truth conditions state that a sentence involving a CCC is true in a situ-
ation where the prepositional object and the modified NP have a conjoint deno-
tation. Due to the conjoint denotation, which implies plurality, and the fact that
no restrictions are specified on the cardinality of the set of events, both distribu-
tive and collective interpretations are possible.
Finally, the basic translation of the inclusive preposition z is presented in
(22). The truth conditions state that a sentence involving an ICC is true in a situ-
ation where the denotation of the prepositional object (represented by Y) is in-
cluded in the denotation of the modified NP (represented by X). Since the logi-
cal representation of inclusive z entails a plural entity and there are no restric-
tions on the event variable, both distributive and collective readings can be li-
censed.
The derivations of the meanings of the sentences (2), (3) and (4) using these
basic translations are illustrated in Figure 2, Figure 3 and Figure 4 respectively.
104 Beata Trawi{ski
Figure 2: The representation of sentence (2), including an ACC
Figure 3: The representation of sentence (3), including a CCC

Figure 4: The representation of sentence (4), including an ICC

The structure in Figure 2 shows that the translation of the entire subject is com-
puted by first combining the basic translation of the preposition z with the basic
translation of the NP Mari, and then with the basic translation of the NP Jan by
functional application. The translation of the sentence results from combining of
the translation of the subject NP and the basic translation of the verb wyjecha
(according to (19)), where the subject NP acts as the functor and the predicate as
its argument. According to this semantic representation, the sentence in (2) is
true if there is a set of events of leaving in which Jan participates, and for each
non-empty subset of the events of Jan's leaving, there is a set of events such that
the relation of accompaniment holds between this set of events, Maria and Jan.
In short, the sentence is true if Jan left and Maria accompanied him.
Figure 3 presents the semantic derivation of (3). The basic translation of the
conjunctive preposition z requires that the subcategorized NP and the modified
NP have conjoint denotations. Here also, the subject acts as a functor and the
predicate as an argument. The semantic representation of this sentence entails a
plural entity, composed of the denotation of the two NPs, and no restrictions on
the event variable. As a result, both the collective and the distributive interpreta-
tion can be licensed. The sentence is true if there is a set of events such that the
relation of leaving holds between this set of events and the set including Jan and
Maria. In short, the sentence is true if Jan and Maria left.
Finally, the semantic derivation of the sentence in (4) is given in Figure 4.
This derivation shows that the basic translation of the inclusive preposition z
combines first with the basic translation of the complement NP and then with
106 Beata Trawi{ski
the basic translation of the modified plural pronoun. The truth conditions pro-
vided by the logical representation of the preposition require that the denotation
of the complement NP be a proper subset of the denotation of the modified NP;
thus, the inclusive interpretation is licensed. The translation of the subject NP
then applies to the translation of the predicate (as defined in (18)), resulting in
the semantic representation of the sentence. Note that the set of individuals in
the denotation of the subject includes the speaker (represented by i), the individ-
ual denoted by the complement NP (represented by the constant m'), and possi-
bly further individuals. The sentence is true if there is a set of events such that
the relation of leaving holds between this set of events and the set of individuals.
If the sentence is true of a set of individuals only including the speaker and Ma-
ria in a set of events of leaving, the ICC is interpreted as closed. If the sentence
is true of a set of individuals including the speaker, Maria and further individu-
als, the ICC is interpreted as open.
7. Summary
In this paper, I have argued that the differences between the particular types of
CCs are semantic in nature, and discussed a number of phenomena that support
this assumption. Such an analysis stands in contrast to the majority of previous
approaches to CCs, which try to explain the differences between them by dint of
syntactic means. Some of these approaches are problematic for a number of em-
pirical and theoretical reasons.
I propose an alternative, a fully compositional semantic analysis, whose core
assumptions are that the three types of CCs have a uniform syntactic structure
and that the differences between them follow from three different semantic re-
presentations of the comitative preposition. The polysemy of the comitative pre-
position which my analysis draws on is one of the most typical properties of
prepositions generally and has been largely discussed within the field of cogni-
tive linguistics. In this respect, the semantic analysis of CCs I propose in this
paper seems more natural than analyses which draw on purely syntactic stipula-
tions.
References
COMACHO, J. (2000) Structural Restrictions on Comitative Coordination. Linguistic Inquiry
31, 366375.
CZUBA, K. & A. PRZEPIRKOWSKI (1995) Agreement and Case Assignment in Polish: An At-
tempt at a Unied Account. Technical Report 783, Institute of Computer Science,
Polish Academy of Sciences.
DALRYMPLE, M., I. HAYRAPETIAN & T. H. KING (1998) The Semantics of the Russian Comita-
tive Construction. Natural Language and Linguistic Theory 16, 597631.
DYA, S. (1988) Quasi-Comitative Coordination in Polish. Linguistics 26, 383414.
DYA, S. & A. FELDMAN (to appear) On Comitative Constructions in Polish and Russian. In:
Proceedings of the Fifth European Conference on Formal Description of Slavic Lan-
guages. Leipzig, Germany.
FELDMAN, A. (2002) On NP-Coordination. In: BAAUW, S., M. HUISKES & M. SCHOORLEMMER
(eds.), Yearbook 2002 3967. Utrecht Institute of Linguistics OTS.
IONIN, T. & O. MATUSHANSKY (2003) DPs with a Twist: A Unied Analysis of Russian Com-
itatives. In: BROWNE, W., J.-Y. KIM, B. H. PARTEE & R. A. ROTHSTEIN (eds.) (2002)
Formal Approaches to Slavic Linguistics 11: The Amherst Meeting 2002 255-274.
Ann Arbor: Michigan Slavic Publications.
LADUSAW, W. A. (1989) Group Reference and the Plural Pronoun Construction. In: Papers on
the Plural Pronoun Construction and Comitative Coordination 1-7. UCSC Syntax Re-
search Center Report SRC-89-02.
MCNALLY, L. (1989) Comitative Coordination in Russian. In: Papers on the Plural Pronoun
Construction and Comitative Coordination 9-15. UCSC Syntax Research Center Re-
port SRC-89-02.
MCNALLY, L. (1993) Comitative Coordination: A Case Study in Group Formation. Natural
Language and Linguistic Theory 11 347379.
PRZEPIRKOWSKI, A., A. KUP, M. MARCINIAK & A. MYKOWIECKA (2002) Formalny opis
j zyka polskiego: Teoria i implementacja [A Formal Description of the Polish Lan-
guage: Theory and Implementation]. Warszawa: Akademicka Ocyna Wydawnicza
EXIT.
SKRABALOVA, H. (2003) Comitative Constructions in Czech. In: KOSTA, P., J. BASZCZAK, J.
FRASEK, L. GEIST & M. ZYGIS (eds.) Investigations into Formal Slavic Linguistics.
Contributions of the Fourth European Conference on Formal Description of Slavic
Languages (FDSL IV) held at Potsdam University, November 28.-30. 2001 Linguistik
International, No. 10/1.2 685.696. Frankfurt am Main: Lang.
TRAWISKI, B. (2005) Plural Comitative Constructions in Polish. In: S. MLLER (ed.) The
Proceedings of the 12th International Conference on Head-Driven Phrase Structure
Grammar, Department of Informatics, University of Lisbon 375395. Stanford: CSLI
Publications.
TRAWISKI, B. (2009) Towards a Semantics-Based Typology of Polish Comitative construc-
tions. In: BIRZER, S., M. FINKELSTEIN & I. MENDOZA (eds.) Proceedings of the Second
International Perspectives on Slavistics Conference, Regensburg 2006 volume 36
of Die Welt der Slaven 138 148. Mnchen: Verlag Otto Sagner.
VASSILIEVA, M. B. & L. K. RICHARD (2005) The Semantics of the Plural Pronoun Construc-
tion. Natural Language Semantics 13 101124.
Beata Trawi{ski, University of Vienna, Department of Slavic Studies, Spital-

gasse 2, Hof 3, A-1090 Wien, Austria, beata.trawinski@univie.ac.at
IV.
Syntax
Binding by Phases: Principle A in Russian
Andrei Antonenko*
1. Introduction
The question about levels of application of binding theory principles has gener-
ated a large body of literature, starting from the early days of generative gram-
mar. It has been argued that binding theory must apply at levels such as LF or
SS, now obsolete within the current minimalist framework. More recent propos-
als claim that binding theory applies derivationally. Another question is what
constitutes a local domain for Principles A and B. There are two main goals of
this paper. The first major goal is to provide a minimalist account of subject ob-
viation phenomenon in subjunctive clauses. The second major goal is to argue
(based on Russian data) that binding principle A applies cyclically at the phase
levels, and propose a revision of Phase Theory (CHOMSKY 2001) in order to be
able to treat defective domains, such as embedded subjunctive clauses in Rus-
sian.
The organization of this paper is the following. Section 2 provides data on
Russian subjunctives as opposed to indicatives, and introduces the phenomenon
of obviation. In section 3 I outline the theoretical framework assumed in this pa-
per. Section 4 contains an analysis of (Russian) subjunctive clauses, and pro-
vides an account of obviation effects. In section 5 I argue for phasal nature of
Principle A, based on the interaction of scrambling and binding from Russian
indicative, subjunctive, and infinitival clauses. Section 6 concludes the paper.
2. Russian subjunctive clauses
2.1 Preliminary data on Russian subjunctives
Russian subjunctive clauses are introduced by the complementizer toby. The
verb in the subjunctive clause is morphologically in the past tense, and no other
verbal forms are allowed, as shown in the example (1):
(1) a. Ivan xoet toby Maa proitala/itala LGB
I. wants that.SUBJ M. read.PST.PERF/.PST.IMPERF LGB
Ivan wants Masha to read Lectures on Government and Binding
b. *Ivan xoet toby Maa itaet/proitaet/budet itat' LGB
I. wants that.SUBJ M. read.PRES/.FUT.PERF/.FUT.IMPERF LGB
__________
* I am grateful to John F. Bailyn and Daniel Finer for discussions about the ideas presented
in this paper. I would also like to thank the audiences of FDSL 8 in Potsdam, NWLC
2009 at the University of British Columbia, and PLC 33 at the University of Pennsylva-
nia, and all my colleagues at the Linguistics Department of Stony Brook University for
helpful comments on and discussion of various parts of this paper. All errors of course
remain my own.
112 Andrei Antonenko
On the contrary, Russian indicative clauses are introduced by the complementiz-

er to. The restriction on the morphology/tense of the verb is not present in in-
dicative clauses, as illustrated by the example (2):
(2) a. Ivan skazal to Maa proitala/itala LGB
I. said that M. read.PST.PERF/.PST.IMPERF LGB
Ivan said that Masha have read/was reading LGB
b. Ivan skazal to Maa itaet/ budet itat' LGB
I. said that M. read.PRES/.FUT.IMPERF LGB
Ivan said that Masha is reading/will be reading LGB
Despite the fact that the verb in the embedded subjunctive clause is
morphologically past, the event denoted by the embedded clause is not situated
in the past, either with respect to the event in the matrix clause, or with respect
to the speech act. On the contrary, the event described in the embedded clause (a
reading of LGB in (1a)) is irrealis and might happen in the future with respect to
the time of the event described in the matrix clause (the volition act in (1a)).
2.2 The subject obviation phenomenon in Russian
In this section I illustrate some syntactic differences between subjunctive and
indicative clauses in Russian related to the well known phenomenon of obvia-
tion discussed in detail in AVRUTIN & BABYONYSHEV (1997). This phenomenon
is illustrated in the examples in (3) (ibid.):
(3) Subject obviation1
a. Volodjai xoet toby on*i/j potseloval Nadju
V. wants that.SUBJ he kissed N.
Volodjai wants himj to kiss Nadja.
b. Volodjai skazal to oni/j potseloval Nadju
V. said that he kissed N.
Volodjai said that hei/j kissed Nadja.
In example (3a), where the embedded clause is subjunctive, the pronominal sub-
ject of the embedded clause cannot be coreferential with the matrix subject.
However, when the embedded clause is indicative as in example (3b),
coreference between the matrix and embedded subjects is possible.
__________
1 Russian is not a pro-drop language, therefore examples like in (i) are ungrammatical:
(i) * Volodja xoet toby / skazal to pro potseloval Nadju
V. wants that.SUBJ said that kissed N.
Binding by Phases: Principle A in Russian 113
3. Theoretical framework
3.1 Feature-sharing Agree
In my analysis of the indicative/subjunctive distinction in Russian, I follow the
framework outlined in PESETSKY & TORREGO (2007) which I will briefly sum-
marize below. It is based on the possibility of feature sharing, and allows a fea-
ture to have several instances in various locations within the syntactic tree. The
crucial operation is the following version of Agree stated in (4).
(4) Agree: Feature Sharing Version (from PESETSKY & TORREGO 2007)
a. An unvalued feature F (a probe) on a head H at syntactic location (F)
scans its c-command domain for another instance of F (a goal) at location
(F) with which to agree.
b. Replace F with F, so that the same feature is present in both locations.
This Feature-Sharing version of Agree operation may create multiple instances
of a single feature in various syntactic locations within the structure. After prob-
ing by a head with an unvalued feature, the features of a goal and a probe enter
into an Agree relation, and both become instances of the same feature.
Another crucial assumption which is needed to maintain feature sharing is
the elimination of Chomskys Valuation/Interpretability Biconditional that al-
lows only uninterpretable and unvalued <uF -val> and interpretable and valued
<iF +val> features. In the new system by PESETSKY & TORREGO (2007), two
more types of features are allowed: uninterpretable and valued <uF +val> and
interpretable and unvalued <iF -val>. In PESETSKY & TORREGO (2007) frame-
work, unvalued features (<uF +val> or <uF -val>) act as probes.
For instance, the T-feature on T is interpretable, but unvalued, and that al-
lows it to be a probe. On the contrary, the T-feature on a finite verb is uninter-
pretable (no semantic interpretation happens within the verb itself), but valued,
since the verb comes from the lexicon with morphologically specified tense.
3.2 Move-F and feature approach to binding
The operation of covert feature movement, Move-F, was considered in
CHOMSKY (1995). A set of formal features (FF) of a head can adjoin to another
head, forming a complex, consisting of features of both heads. For instance, fea-
tures of an object of a transitive verb can adjoin to the complex v+V, which is
formed by raising of the main verb V and adjoining it to the v. The result of this
operation is a complex v+V+FF (object). In a similar fashion, the formal fea-
tures of the subject under certain circumstances can adjoin to T, resulting in the
complex T+FF (subject). Adopting the framework of PESETSKY & TORREGO
(2007), I propose (similar to WATANABE 2000) that Move-F happens after prob-
ing by an unvalued feature, and as a result the set of formal features of the goal
adjoins to the probe. The phonological movement, accompanying Move-F takes
place only if there is a relevant EPP feature present on the probe.
WATANABE (2000) compares the approach of CHOMSKY (1998) with the ap-
proach of CHOMSKY (2000). CHOMSKY (1998) argues that feature checking al-
ways involves the adjunction of the features of the goal to the probing head. Un-
der the latter approach by CHOMSKY (2000), the idea of obligatory feature rais-
ing under Agree relation is abandoned: the Agree relation takes place without
feature displacement. Under these two approaches, subject raising to T from the
initial configuration in (5a) gives rise to either the configuration in (5b) or in
(5c).
(5) a. T [vP Subj ... ]
b. [TP Subj [T [T FF(T)+FF(Subj)] [vP tsubj ... ]]] (CHOMSKY 1998)
c. [TP Subj [T FF(T)] [vP tsubj ... ] (CHOMSKY 2000)
Based on complementizer agreement facts from Dutch, following ZWART (1997)
Watanabe argues that the correct approach is the one resulting in the configura-
tion in (5b). He proposes that the agreement morphology on the complementizer
in Dutch comes from the features of the subject itself.
BRANIGAN (2000) argues that binding theory is sensitive not only to the
overt movement of the constituents, but that also movement of the formal fea-
tures can influence the binding relations. His arguments are based on considera-
tion of the English ECM constructions, such as the one shown in (6a).
(6) English ECM constructions (BRANIGAN 2000):
a. Perry proved [[Jill and Tony]i to have lied] during each otheris trials.
b. Perry [vP proved+FF(Jill and Tony)i [TP [Jill and Tony]i to have lied]
during each otheris trials]
c. Perry proved [[Jill and Tony]i [vP e [TP ti to have lied] during each
otheris trials]]
In (6a), the reciprocal each other is located in the matrix clause, while its ante-
cedent is in the embedded clause. However, no violation of Principle A occurs,
and the sentence is grammatical. Two possibilities demonstrated in (6b) and (6c)
have been proposed in the literature. According to (6c), the embedded subject is
in fact located in the matrix clause after undergoing raising to object, and there-
fore Principle A is satisfied. The alternative analysis (6b) involves the raising of
the formal features of the embedded subject to the matrix clause, and it is the
formal features of the subject that serve as an antecedent to the reciprocal. In or-
der to choose between two possible solutions, Branigan combines the ECM con-
structions like the ones in (6) with the locative inversion.
(7) English ECM, locative inversion (BRANIGAN 2000):
a. The photos [VP showed [TP behind this very hedge had been hiding [Jill
and Tony]i] during each otheris trials].
b. the photos [VP FF(Jill and Tony)i-showed [TP behind this very hedge to
have been hiding [Jill and Tony]i] during each otheris trials]
(7a) is similar to (6a), but the locative phrase behind this very hedge has under-
gone locative inversion. Locative inversion is commonly assumed to be the dis-
location of the locative phrase to the TP-peripheral position. Under this assump-
tion, it is clear that the embedded subject stays within the embedded clause, and
there is no raising to object. Therefore the only possible analysis of the sentence
in (7a) involves feature raising of the embedded subject to the matrix clause, as
demonstrated in (7b) and these formal features serve as a binder for the reci-
procal. The featural approach to binding is summarized in (8).
(8) Featural approach to binding:
a. A set of formal features of a nominal is indistinguishable from a no-
minal element itself from the point of view of the computational sys-
tem.
b. Binding theory operates on sets of formal features, even if their dis-
placement is not accompanied by pied-piping of phonological material.
4. An analysis of indicative/subjunctive distinction in Russian
In this section I apply the theoretical framework outlined in section 3 above to
embedded clauses in Russian. I propose that the fact that the subjunctive form of
the verb is identical to the past tense form in Russian is just an idiosyncrasy. I
assume based on the semantic interpretation of the subjunctive sentences that
even though the verb in the subjunctive clauses is morphologically past, it bears
an unvalued T feature, unlike verbs in other finite.
(9) a. Ivan xoet toby Maa proitala LGB
I. wants that.SUBJ M. read.SUBJ LGB
Ivan wants for Masha to read LGB [Subj]
b. Ivan skazal to Maa proitala LGB
I. said that M. read.PST.PERF LGB
Ivan said that Masha have read/was reading LGB [Ind]
That means that in the sentences in (9) the verb proitala read comes from the
lexicon with different features: in sentence (9a) it bears an unvalued <uT -val>,
whereas in the sentence (9b) it bears a valued T feature <uT +val>.
This proposal is summarized below in (10):
(10) The Subjunctive parameter:
a. The subjunctive form of the verb bears a <uT -val> feature;
b. Finite forms of the verb bear <uT +val> feature.
I claim that the adoption of the Subjunctive Parameter in (10), along with the
proposal about the featural approach to binding (WATANABE 2000) within the
PESETSKY & TORREGO (2001, 2007) framework allows us to account for the
asymmetries between indicative and subjunctive clauses illustrated in section 2.
4.1 Indicative clauses

Recall that by the Subjunctive parameter (10), indicative verbs have <uT +val>
T-feature. The derivation proceeds in a standard bottom-up way. The verbal pro-
jection vP is built in a standard manner with V adjoining to v. If the embedded
clause of a sentence is indicative (as in (9b)), after T is merged into the tree
structure, its interpretable but unvalued feature <iT -val> probes to find its goal,
<uT -val> feature on the subject DPemb (assuming that Nominative case is an in-
stantiation of the T-feature on D, as in PESETSKY & TORREGO 2001). After the
Agree operation takes place, the features on T and the subject D are linked, and
become instances of the same feature. However, since the subject DPembs T-
feature is unvalued, the shared T-feature also remains unvalued. The EPP sub-
feature of T-feature on T is active, and the featural complex of the embedded
subject attaches to T forming a complex T+FF(DPemb). Further, T probes further
again in order to find a value. The second probing finds a goal <uT +val> on the
finite verb within the vP projection. After the Agree operation, all three T-
features those on T, the subject DP and v/V become instances of the same fea-
ture, and the valuation of the <iT> on T takes place, resulting in the valuation of
<uT> on subject DPemb also. After this step, all T-features in the embedded
clause are valued. The subject EPP, being a phonological condition, will be sa-
tisfied by further raising of the embedded subject to Spec,TP. The embedded CP
can then be sent off to interpretation.
4.2 Subjunctive clauses
In the case of Russian subjunctive clauses applying the analysis proposed above
gives surprisingly different results. Following the Subjunctive Parameter (10),
the subjunctive verb comes from the lexicon with the unvalued T feature <uT -
val>. Also, I would assume the presence of toby in the numeration for selec-
tional purposes (I would claim that volitional predicates, such as xotet to
want, select CPs headed by toby). I assume that toby also comes from the lex-
icon endowed with uninterpretable unvalued <uT -val> feature.
The embedded vP is built in standard fashion. Further, the embedded T
probes and agrees first with the subject DPemb, and then with the verb (to be
more precise, v+V complex), resulting in feature sharing among all these ele-
ments, making the T-features on T, the subject DPemb and v+V all being in-
stances of the same feature. The formal feature bundle of the embedded subject
adjoins to T, forming a complex T+FF(DPemb). However, unlike in the case of
indicative clauses, no valuation can occur at this point, since the T-feature on the
embedded subjunctive verb is not valued. The derivation proceeds by merging
of toby in the C-head position.
The T-feature of toby is unvalued, and therefore must probe down to find
its goal. The first goal it finds is a T+FF(DPemb) complex with unvalued T-
feature. Feature sharing Agree takes place, and the instances of the T-feature on
toby, on T, on DPemb, and on the embedded verbal complex become instances

of the same feature. Further, the featural bundle created in T adjoins to toby
and the resulting configuration by the completion of the embedded CP-phase is
given in (11), where the index [1] shows which T-features are instances of the
same feature, and DPemb is the subject of the embedded clause.
(11) [CP to by<uT -val>[1]+T<iT -val>[1]+FF(DPemb) [TP DPemb v+V<uT -val>[1] ... ]]
The unvalued T-feature was able to move to the edge of CP-phase to the C-head
position (bolded in (11)), and therefore will remain accessible for further Agree
relations with the probe from the higher domain.
Next, the elements of the matrix clause are merged in the structure: V/v with
the <uT +val> (since the matrix verb is finite), and matrix subject DP with the
instance of <uT -val>. Recall that V in subjunctive constructions selects a CP
headed by toby. This selectional property would result in movement of the fea-
tural complex, present on toby, to the matrix V. By the time the vP of the ma-
trix clause is completed, the featural bundle raised from the head of embedded
CP and adjoined to V, and further to v, still does not have a value for its T-
feature. The configuration at this stage of the derivation is given in (12)2. Notice
that here the featural complex in the head of vP position has two different types
of T-features: ones marked with [1], indicating that they came from the embed-
ded clause, and all other which did not yet enter the Agree relationship.
(12) [vP DPmatr<uT -val> v+V<uT +val>+C<uT -val>[1]+T<iT val>[1]+FF(DPemb) ... [CP
toby... ]]
At the next stage, the matrix T, endowed with <iT -val> feature, is merged into
the structure. Since it is an interpretable feature, it probes down, finding the T-
feature of the matrix subject and agrees with it, resulting in a shared feature be-
tween it and the matrix subject DPmatr. As before, the formal feature bundle of
the matrix subject adjoins to T. Further, since the T-feature of the matrix T is
still unvalued (as none of the elements with which it has agreed have provided it
with a value), it probes down one more time and finds the matrix
v+V+C+T+FF(DPemb) complex as a goal. The Agree operation at this stage
makes all the T-features on the matrix and embedded Vs, and the T-features in
the featural complex located in the matrix v-head position instances of the same
feature, and values them, acquiring the value from the <uT +val> matrix verb.
After this crucial step, all T-features introduced so far in both matrix and
embedded clauses are instances of the same T-feature, and all of them become
__________
2 I will not go into details of how and why toby gets pronounced in the lower clause, and
why its phonological features do not raise along with the formal featural complex to the
position in the matrix clause. For more details on this I refer the reader to ANTONENKO
(2010) where I propose that toby consists of two parts: to, located in Spec,CP, and by,
which is a complementizer C.
valued. The resulting structure is shown in (13). Features raised raised from the
embedded clause are bolded.
(13) [TP T<iT +val>[1]+FF(DPmatr) [vP DPmatr<uT +val>[1] v+V<uT +val>[1]+C<uT +val>[1]+T<iT
+val>[1]+FF(DPemb) ... [CP to by...
4.3 Russian obviation explained

I propose that the obviation constitutes a violation of Principle B on featural lev-
el. As I mentioned earlier in my consideration of Move-F, the featural bundle of
the nominal is indistinguishable from the nominal itself from the point of view
of the computational system, and therefore the formal feature complex can enter
into binding relations. Principle B here is violated if the bundle of formal fea-
tures FF of the pronominal element is locally bound by its antecedent or the set
of formal features of its antecedent.
This analysis of the indicative embedded clauses allows an account of lack
of obviation facts with indicative embedded clause from example (3b) in section
2, repeated here in (14a):
(14) a. Volodjai skazal to oni/j poceloval Nadju
V. said that he kissed N.
Volodjai said that hei/j kissed Nadja.
b. *Volodjai xoet toby oni poceloval Nadju
V. wants that.SUBJ he kissed N.
Volodja wants to kiss Nadja
In (14a) the embedded subject is in the T-domain, and therefore cannot be
bound by Volodja, since the matrix subject cannot see inside the lower CP-
phase.
Now I will consider the obviation phenomenon in the case of subjunctive
embedded clauses, (14b). By the time the matrix vP phase is completed the con-
figuration is in (15):
(15) [vP Volodjai v+V+...+FF(hei) [CP ... [TP he ...
The formal features of the embedded pronominal subject end up adjoined to the
matrix v+V complex, which is c-commanded by the matrix subject. As one can
see, the Principle B is violated at this configuration on featural level.
5. Binding by Phase: Principle A

5.1 Derivational theory of binding
Binding theory governs the use of anaphors/reciprocals, pronouns and R-
expressions. The standard versions of binding principles are given in (16).
(16) Binding theory principles:
a. Principle A: Anaphors and reciprocals must be bound in a local domain
b. Principle B: Pronouns must be free in local domain
c. Principle C: R-expressions must be free anywhere
What constitutes a binding domain and the levels at which the binding principles
must be evaluated is extensively discussed in the literature. There are two major
approaches to binding principles: derivational and representational. Representa-
tional theory of binding means that the principles of the binding theory apply at
certain levels of representation: Deep Structure, Surface Structure, or Logical
Form being the possible choices (PICA 1991, HESTVIK 1992, LEBEAUX 2009
a.o.). Derivational theory of binding means that the certain conditions on ana-
phors, pronouns and R-expressions must be satisfied during the course of deri-
vation, and it is not sufficient to look at a certain level alone (BELLETTI & RIZZI
1988, EPSTEIN et al 1998, SAITO 2003, BAILYN 2007 a.o.). The (strong) deriva-
tional version of Principle A, according to which the anaphor/reciprocal must be
bound at any point in derivation in order for derivation to converge, is given in
(17).
(17) Binding Principle A: (strong) derivational version
Anaphor/reciprocal must be locally bound at any point in the derivation.
In what follows in this section I advocate for a phasal approach to binding, ac-
cording to which binding relations are evaluated derivationally after completion
of the phases.
5.2 Why is strong derivationality undesirable: Russian puzzles
In this section I present the interaction of binding with scrambling out of indica-
tive, subjunctive, and infinitive embedded clauses in Russian.
Example (18) presents the binding data from Russian when the embedded
clause is indicative. In this case, the long-distance binding is unavailable as
shown in (18a). Here, the anaphor sebja self is an object of the embedded
clause and can only be coreferential with the embedded subject. Thus, non-local
binding by matrix subject is unavailable. Long-distance scrambling of the ana-
phor as in (18b) does not change the meaning of the sentence, as the matrix
binder is still unavailable.
(18) a. Ivani skazal to Borisj narisoval sebja*i/j
I. said that B. draw self
Ivani said that Borisj drew himselfj/*himi
b. Ivani sebja*i/j skazal to Borisj narisoval t

I. self said that B. drew
Ivani said that Borisj drew himselfj/*himi
Example (19) illustrates the binding possibilities when the embedded clause is
subjunctive. With respect to long-distance binding, the situation is no different
from the indicative embedded clauses. The long-distance binding of the object
anaphor in the embedded clause is not allowed, as shown in (19a). However, the
sentence (19b) is different. The anaphor, which has undergone long-distance
scrambling to the matrix clause, can now also be bound by the matrix subject,
and not just the embedded subject.
(19) a. Ivani xoet toby Borisj narisoval sebja*i/j
I. wants that.SUBJ B. draw self
Ivani wants Borisj to draw himselfj/*himi
b. Ivani sebjai/j xoet toby Borisj narisoval t
I. self wants that.SUBJ B. draw
Ivani wants Borisj to draw himselfj/himi
Finally, example (20) demonstrates binding abilities when the embedded clause
is infinitival. The most striking difference between this situation and the situa-
tion described in the two previous sections is the availability of long-distance
binding. As example (20a) shows, the anaphoric object of the embedded clause
can now be bound by both the embedded PRO-subject and the matrix subject,
resulting in ambiguity of the corresponding sentence. Also, this binding possi-
bility (predictably) does not disappear when the embedded anaphoric object is
scrambled to the matrix clause, as in (20b).
(20) a. Ivani skazal Borisuj PROj narisovat sebjai/j
I. told B. draw.INF self
Ivani told Borisj to draw himselfj/himi
b. Ivani sebjai/j skazal Borisuj PROj narisovat t
I. self told B. draw.INF
Ivani told Borisj to draw himselfi/himi
The table in (21) presents the summary of data discussed in this section.
(21)
LD binding Binding/ Scrambling Obviation
Indicative No No No
Subjunctive No Yes Yes
Infinitive Yes Yes N/A
The data presented above leaves several questions unanswered if one assumes
the strong derivational theory of Principle A, as described in (17). First, it is un-
clear why the anaphor allows long-distance reading when inside the infinitival
clause, and allows only local antecedent when inside the indicative and subjunc-
tive clause. Second, the difference between anaphors scrambled out of indicative
vs. subjunctive clauses is puzzling. Under strong derivational theory of binding
we would expect that the matrix antecedent in these two cases must be equally
allowed. In what follows I will provide answers to these questions based on the
idea that binding principle A operates phase-by-phase.
5.3 Subject orientation: how high do anaphors raise?
It is traditionally assumed that the subject oriented anaphors raise to T (Pica
1987, Hestvik 1992). In this section I would like to propose that there is no need
for assuming this raising. It suffices for an anaphor to raise to v-head. If we as-
sume that the Principle A applies on the level of the phase, and the subjects ori-
ginate in the Spec,vP position, this raising of the anaphors will give us precisely
the same effects, and will be able to account for the subject orientation of the
anaphors.
As for anaphors originating in the subject positions, I claim that they dont
need to raise independently at all (they only raise together with the subject to the
Spec,TP position).
The condition on the anaphor raising can be formulated as in (22).
(22) Anaphors must be at the edge of the phase.
Object anaphors satisfy this condition by head-raising to v, anaphors within the
subject originate at the edge of the vP-phase, and therefore satisfy this condition
vacuously.
5.4 Extending phase theory: Principle A by phase
In this section I will propose a general theoretical framework which allows us to
solve Russian binding puzzles introduced above. My explanation of the asym-
metries builds on the following assumptions:
(23) An anaphor has a referential feature which must be valued by its antecedent
in syntax (=Principle A); if the feature is not checked, that constitutes a vi-
olation of Principle A.
The assumption (23) was pursued in a range of previous works, such as SAITO
(2003), HICKS (2009) among others. Notice, that unlike REULAND (2005), I as-
sume that the feature participating in binding relation is a referential feature
<Ref>, and is different from -features. For the reasons of space, I refer the
reader to HICKS (2009) for some of the arguments against assumption that bind-
ing operates on -features.
(24) Principle A operates cyclically at the phase level.
(25) When all features within a phase are valued, the phase is sent off to inter-
pretation. I will refer to such phases as complete (CHOMSKY 2001).
(26) If all features within a phase are either valued or shared with some outside
element (PESETSKY & TORREGO 2007), the phase will be closed for syntac-
tic operations (even though it may not yet be complete, and it may be im-
possible to send it off to interpretation).
Principle (25) is standard within the current phase theory. However, in order to
deal with clausal domains such as subjunctive clauses, where the features do not
get valued before the merge of some element in the higher clause, I introduce
the principle in (26) and the notion of the complete phase. Complete phases are
either closed, and in this case they can be sent off to interpretation, or have all
features within them shared with some outside element. As usually assumed, on-
ly elements at the edge may participate in further syntactic computation. There-
fore, this feature-sharing allows unvalued features buried within the phase to be-
come valued at a later stage without violating the Phase Impenetrability Condi-
tion, since they have an instance outside of the phase (as in PESETSKY & TORRE-
GO 2007).
Notice, that complete phases are always closed, therefore complete phases
constitute a subset of the set of closed phases. Further, if one assumes the struc-
ture of subjunctives as discussed above in the paper in the section 4.2, it can be
seen that the embedded vP phase of subjunctive clauses gets closed as soon as
embedded T probes the embedded v, and their T-features become shared. How-
ever, since valuation of embedded T-features does not occur before merge of the
matrix T, embedded vP cannot be sent off to interpretation, and therefore is not
complete. Similar arguments can be made for the embedded subjunctive CP-
phase. Upon merge of the matrix v/V, T-features get shared between the embed-
ded T and matrix v/V, and therefore the embedded subjunctive CP becomes
closed. Similarly to vP phase, it does not become complete until the merge of
the matrix T.
The last assumption I introduce is the one in (27).
(27) Phases need to be sent off to interpretation as early as possible to minimize
computational complexity.
Apart from the general theoretical assumptions (23) (27), the table (28) sum-
marizes the facts about Russian embedded clauses.
(28) Featural difference between verb forms:
CP presence T-feature on the verb

Indicative Present <uT +val>
Subjunctive Present <uT val>
Infinitive Absent <uT val>
5.5 Solving Russian binding puzzles

5.5.1 Long-distance binding without scrambling
First, I consider Russian examples without scrambling. In a case when the em-
bedded clause is indicative, the embedded CP is complete, since all features, in-
cluding all T-features within the embedded CP, are checked by the time the em-
bedded TP is merged. Therefore, under the condition (27), the anaphor must get
bound within this phase, so that the entire embedded CP can be send off to in-
terpretation. That explains why only the embedded subject can serve as an ante-
cedent for the anaphor.
If the embedded clause is subjunctive, T-feature on T and T-feature on v/V
are shared by the time CPemb is merged. Therefore, vPemb is closed, as defined in
(26) (though not complete), since all the features present within vPemb are also
present outside of vPemb, and they will be able to get checked without violating
the Phase Impenetrability Condition. Thus, the binding theory must apply to the
vPemb, which would result in unavailability of long-distance binding, and again
the local subject is the only possible antecedent for the anaphor.
Now, if the embedded clause is infinitive, I would assume that there is no
embedded CP projection, see (28), so the vPemb is closed only at the time vPmatr
is completed. At that level, the possible binder for the anaphor can be a matrix
subject, which is on the edge of vPmatr.
5.5.2 Interaction of long-distance binding and scrambling
Now, lets consider the interaction of binding and scrambling. In the case of an
indicative embedded clause, at the level of the lower CP phase, the tense-feature
of the embedded verb is valued, and the structure of the lower CP phase for the
sentence in (18b) is given in (29a).
(29) a. [CP sebja to [TP Boris T<iT +val> [vP sebja
b. [CP sebja toby [TP Boris T<iT -val> [vP sebja
Notice, that in this configuration, the anaphoric feature is the only feature, which
is left unvalued within the embedded CP phase. Following principle (27), it
must be satisfied at this level, since it is the only feature preventing the phase
from being send off to interpretation. The only way to satisfy it is to bind the
anaphor by the embedded subject. Therefore, the only possibility for an anaphor
is to be bound by an embedded subject.
If the embedded clause is subjunctive, the tense-feature in the embedded
clause of (19b) is unvalued at the level of the lower CP phase (29b); T-feature
within the CP-phase cannot be valued without getting the value from the matrix
clause. Therefore the embedded CP phase cannot be complete; condition (27)
does not apply at this level, since T-feature within this phase cannot be valued at
this level anyway. Thus, at this level the anaphor is allowed to be left unbound
within this phase. In this case, the anaphor will be allowed to get bound at the
level of the matrix vP phase, and in this case the matrix subject will serve as its
binder. Notice, that nothing precludes the anaphor from getting bound in the
embedded clause, and thus the embedded subject is still a possible binder for the
anaphor in case the embedded clause is subjunctive.
In the case of the infinitive embedded clause, assuming that there is no em-
bedded CP projection present, the binding relations will only be evaluated at the
level of the matrix vP phase (similar to the case without scrambling), and there-
fore the matrix subject is still a possible antecedent.
6. Conclusion
In this paper I explored the structure of subjunctive and indicative clauses in
Russian, and argued that subjunctive verbs even though they carry tense mor-
phology, have unvalued T-feature. I explored a featural approach to binding,
showing that the feature displacement can alter binding relations. These assump-
tions allowed me to reduce subject obviation to the violation of Principle B on
featural level.
In the second part of the paper I considered Russian data showing the inte-
raction of long-distance scrambling and binding. I demonstrated that the asym-
metries observed are hard to account under the standard view of Principle A as
derivational, and proposed that Principle A must be evaluated on a phase-by-
phase basis.
References
ANTONENKO, A. (2010) Puzzles of Russian Subjunctives. University of Pennsylvania Working
Papers in Linguistics Vol. 16: Iss. 1, Article 2.
AVRUTIN, S. & M. BABYONYSHEV (1997) Obviation in Subjunctive Clauses and AGR: Evi-
dence from Russian. Natural Language and Linguistic Theory 15:229-262.
BAILYN, J. F. (2004) Generalized Inversion. Natural Language and Linguistic Theory 22:1-
49.
BAILYN, J. F. (2007). A derivational approach to microvariation in Slavic Binding. In: Pro-
ceedings of FASL 15. 25-41. Michigan Slavic Publications.
BELLETTI, A. & L. RIZZI (1988) Psych-verbs and Theta-theory. Natural Language and Lin-
guistic Theory 6:291-352.
BRANIGAN, P. (2000) Binding Effects with Covert Movement. Linguistic Inquiry 31:553-557.
CHOMSKY, N. (1995) The Minimalist Program. MIT Press, Cambridge, MA.
CHOMSKY, N. (1998) Some observations on economy in generative grammar. In: BARBOSA, P.
et al. (ed.) Is the best good enough 115-127. Cambridge, Mass.: MIT Press.
CHOMSKY, N. (2000) Minimalist inquiries. In: MARTIN, R., D. MICHAELS & J. URIAGEREKA
(ed.) Step by step: Essays on minimalist syntax in honor of Howard Lasnik. Cam-
bridge, Mass.: MIT Press.
CHOMSKY, N. (2001) Derivation by Phase. In: KENSTOWICZ, MICHAEL J. (ed.) Ken Hale: A
life in language 1-52. Cambridge, Mass.: MIT Press.
EPSTEIN, S. D., E. M GROAT, R. KAWASHIMA, & H. KITAHARA (1998) A Derivational Ap-
proach to Syntactic Relations. Oxford University Press.
HESTVIK, A. (1992) LF Movement of Pronouns and Antisubject Orientation. Linguistic In-
quiry 23:557-594.
HICKS, G. (2009) The derivation of anaphoric relations. John Benjamins.
LEBEAUX, D. (2009) Where does binding theory apply? Cambridge, Mass.: MIT Press.
PESETSKY, D. & E. TORREGO (2001) T-to-C Movement: Causes and Consequences. In: KENS-
TOWICZ, MICHAEL J. (ed.) Ken Hale: A life in language 355-426. Cambridge, Mass.:
MIT Press.
PESETSKY, D. & E. TORREGO (2007) The Syntax of Valuation and the Interpretability of Fea-
tures. In: KARIMI, S., V. SAMIIAN AND W. K. WILKINS (ed.) Phrasal and Clausal Ar-
chitecture 262-294. John Benjamins.
PICA, P. (1987) On the nature of the reflexivization cycle. In: McDonough, J. & B. Plunkett
(ed.) Proceedings of NELS 17 483-499. UMass Amherst.
PICA, P. (1991) On the Interaction between Antecedent-Government and Binding: the Case of
Long Distance Reflexives. In: Jan Koster, J. & E. Reuland (ed.) Long-Distance Ana-
phora 119-136. Cambridge University Press.
REULAND, E. (2005) Agreeing to Bind. In: BROEKHUIS, H. et al. (ed.) Organizing Grammar:
Studies in Honor of Henk van Riemsdijk 505-513. Mouton de Gruyter.
SAITO, M. (2003) A Derivational Approach to the Interpretation of Scrambling Chains. Lin-
gua 113:481-518.
WATANABE, A. (2000) Feature copying and Binding: Evidence from Complementizer
Agreement and Switch Reference. Syntax 3:159-181.
ZWART, C. J-W. (1997) Morphosyntax of Verb-movement. Dordrecht: Kluwer.
Andrei Antonenko, Stony Brook University, Department of Linguistics, SUNY

at Stony Brook, NY 11794-4376, USA, andant@gmail.com
Dynamic SpellOut as Interface Optimization*
Steven Franks
SpellOut, although traditionally described as a rule stripping off from a
syntactic representation structure relevant for phonetic interpretation from
structure relevant for semantic interpretation and sending that information off to
the appropriate PF or LF interface component, in fact subsumes a complex array
of interacting processes. These processes must take the abstract hierarchical
arrangement of feature bundles we take to be syntax into a linearized sequence
of articulatory instructions we take to be phonetics. This paper explores the
mechanics of SpellOut on the PF side, bringing to bear diverse Slavic language
data. Within the highly derivational minimalist architecture these mechanisms
run the gamut from those which are primarily syntactic in nature, and
presumably apply very early in the mapping from syntax to PF, to those which
are primarily phonological in nature, and presumably apply very late. My goal
in this paper is to provide an overview of the kinds of processes arguably
subsumed under the SpellOut rubric and then to see if a clearer picture of the
overall system might emerge. I will argue that the chief concern of SpellOut is
the optimization of PF, and that optimization proceeds in a derivational fashion.
Minimalist reasoning dictates that we submit to scrutiny the generally
accepted arsenal of analytic devices available under the standard GB model. One
important consequence of this scrutiny is the careful reevaluation of the division
of labor between syntax and PF. In general, it now appears that much which has
traditionally been regarded as part of the syntax proper can be seen as a response
to PF demands, decisions imposed on syntactic structures by the need for
pronunciation. This shifting of the burden to PF is manifested in several ways.
First, the syntax creates structure through the concatenation operation of
Merge, but leaves unspecified the linear order of the concatenated elements.
That is, in the spirit of BOBALJIK (2002), NUNES (2004), and much other recent
work, syntactic representations only express hierarchical information. PF
considerations are thus fundamental in determining word order. Linearization in
this view is a property imposed on language by virtue of the temporal exigencies
of articulation, an essential aspect of SpellOut. It will be argued that
linearization must be done on-line and that it applies multiply in the course of
__________
* I dedicate this paper to the memory of Mariana Lambova, whose work has inspired much
of my own and whose untimely death was a great loss. This paper expands on ideas and
material published as FRANKS (2010), as adapted and expanded for FDSL 8. I thank Peter
Kosta and Lilia Schrcks for their repeated invitations and for putting this volume
together. I am also very grateful to many individuals for their feedback and discussion of
the material represented herein, and especially to eljko Bokovi for his continued
support and advice over the years.
128 Steven Franks
the derivation, making use of diverse kinds of information. In the strongly

derivational approach I adopt, linearization is construed as dynamic rather than
static. I agree with Cedric Boeckx, who contends that none of the conceptual
motivations for phases is compelling andconsider his proposal that we should
adopt (quasi)-uniform SpellOut paths, as well as an alternative that takes
SpellOut domains to be prosodic units instead. I then show some advantages of
a highly iterative derivation in optimizing SpellOut. This optimization,
however, does not posit single access to a set of ranked constraints, as in
traditional Optimality Theoretic models. Instead, SpellOut involves the
continual filtering of structures through a mesh of PFside constraints. In this
vein, assuming the Linear Correspondence Axiom (LCA) of KAYNE (1994),
initial linearization exploits asymmetric c-command, but linearization later in
the mapping to PF can exploit prosodic properties of specific lexical items.1
Other SpellOut effects concern decisions such as copy pronunciation,
lexification, prosodification, and ellipsis, all of which I contend are dynamic in
character.
The paper proceeds as follows. The first section reviews several ideas in the
literature about copy pronunciation and concludes that decisions about copy
pronunciation must, like linearization, be made derivationally. The interactions
are however complex: copy deletion can depend on linear adjacency but
linearization requires a prior decision about which copy is being pronounced,
ellipsis sometimes depends on specific morphology, prosody requires
morphological material to host it but prosodic considerations might also impact
on choice of copy, and so forth. The next section examines pronunciation of
lower copies with wh-movment and argues against successive cyclic movement
in favor of chain formation at the end of the derivation. The third section with
positioning and linearizing clitics in the South Slavic languages, demonstrating
the essential roles play both by prosody and the LCA. Section four subjects
linearization to more careful scrutiny and it is shown that the positioning of the
Bulgarian interrogative clitic li requires linearization to apply cyclically. Section
five examines different kinds of ellipsis, and argues that that there are (at least)
two kinds of PF ellipsis, an early one that saves certain derivations by deleting
structure containing offending features and a late one that ignores syntactic
constituency altogether. It also considers the curious role of morphological
syncretism in letting otherwise infelicitous constructions slip by as well as
__________
1 Interestingly, the same general claim has been recently argued for in LPEZ (2009), on
the basis of a completely different set of data. He convincingly accounts for rightward
dislocation (in Spanish and Catalan) in terms of the LCA as a soft constraint, violable
in order to satisfy prosodic requirements. Like me, Lpez advocates the view that
prosodification and linearization are simultaneous operations.
Dynamic SpellOut as Interface Optimization 129
how choice of allomorph can depend on very superficial phonological factors.2

The last section elaborates on the system of SpellOut an considers alternative
approaches to cyclicity that are not phase-based. An attempt is made to reconcile
optimality theoretic and derivational generative approaches by imposing
constraints to optimize PF desiderata on the output of each SpellOut domain,
and it is shown how the Tobler-Mussafia effect can be handled in this light.
More generally, I argue that the rejection of punctuated resolves puzzles about
the ordering of the various SpellOut processes and trivializes distinctions
between OT and derivational conceptions of SpellOut.
1. Aspects of copy pronunciation
I now turn to some aspects of copy pronunciation that any adequate model of the
syntaxPF interface ought to accommodate. I begin with reviewing two fairly
well-established accounts of lower copy pronunciation in the literature.
1.1 Delayed clitic placement
The first phenomenon can be called delayed clitic placement. The basic idea,
which I put forward years ago (cf. e.g. FRANKS 1998/2010, 2000) and has since
been adopted by BOKOVI (2001a) and many others, is that lower copies of
clitics are pronounced just in case the highest copy is not prosodically viable.
Consider second position (2P) clitics in Bosnian/Croatian/Serbian (BCS), which
move to the highest head position in the clause. These are enclitic, meaning that
they require a prosodic word host to their left; I suggest that we take this to be a
prohibition against being initial in their Intonational phrase (I-phrase).
Interestingly, whenever BCS clitics are left by the syntax at the beginning of
their I-phrase, we encounter them in lower than second position. In the minimal
pair in (1) this effect arises only if the fronted constituent tvome prijatelu your
friend is treated as a (topicalized) separate I-phrase, as in (1b):3
(1) a. #Tvome prijatelju su prodali knjigu.#
your.DAT frend.DAT aux.3PL gave aux.3PL book.ACC
It was to your friend that they sold the book.
b. #Tvome [prijatelju] # # [[prodali] su] knjigu.#
To your friend, they sold the book.
Since the 3rd plural auxiliary clitic su in (1b) cannot find prosodic support when
preceded by an intonational break, the next lower copy down is pronounced.
__________
2 The selection of a particular lexical item with underspecified morphosyntactic features in
one SpellOut domain affects acceptability when that item is shared with another
domain.
3 Note that here and elswhere pronounced copies are in boldface, non-pronounced ones are
struck through and in outline font. Prosodic words are indicated by a subscripted and
I-phrase boundaries are demarcated with #.
130 Steven Franks
This approach correctly predicts this behavior precisely when there is no way
for the highest copy of the clitics to avoid being initial in its I-phrase. Similarly,
since parentheticals are necessarily flanked by I-phrase boundaries, these induce
obligatory pronunciation of lower copies, as in BCS (2b), where tvoja mama
your mother is an appositive:
(2) a. #[[Ja] sam ti] obeala igraku#.
I aux.1SG you.DAT promised aux.1SG you.DAT toy.ACC
I promised you a toy.
b. #Ja#, #tvoja [mama] #, # [[obeala] sam ti] igraku#.
I, your mother, promised you a toy.
In this way, the syntax provides the correct structure, but the phonology filters
out the illicit copies. Prosodic considerations are irrelevant to the syntax, but in
the mapping to PF these play a critical role. In such a PF-filtering system, the
syntax proposes and the phonology disposes.
Evidence in support of this kind of account is based on closely related
Slovenian (Slvn). In this language, the clitics are not necessarily dependent on a
prosodic word to their left to be pronounced (i.e., they are not subject to a
prohibition against being initial in their I-phrase); cf. FRANKS & KING (2000:
3148) or BOKOVI (2001: 151168). Hence, in the Slvn translation of (2b),
cited by GOLDEN & M. SHEPPARD (2000), it is the highest copy of the clitics
which is retained. This is shown in (3).
(3) Jaz#, #tvoja mama#, [sem ti [obljubila] ] igrako.
I, your mother, promised you a toy.
Since it is possible to pronounce the highest copy of sem ti in Slovenian (3), this
copy rather than the lower one must be the one which is ultimately shipped off
to SpellOut.
It is tempting to think of these facts in Optimality Theoretic (OT) terms.
Enclitic here is a lexical property which means the item is subject to
something like (4):
(4) NonInitial(I-phrase)
SpellOut seeks to target the highest copy but compromises when that copy
lacks prosodic viability. And, as we have seen, this desideratum prevails in
BCS, but Slovenian brooks no such compromise, pronouncing the highest copy
regardless. The idea expressed in (4) reflects the standard OT approach to
clitics, as presented for example in Legendre (2003). She exploits the idea of a
competition between EDGEMOST and NONINITIAL constraints, stating that
EDGEMOST(X, LEFT) = E(X) means a feature [X] is left-aligned with the edge
of a projection of the head [X] is associated with and NONINITIAL(X) means
[X] is not realized in Intonational phrase-initial position. Under such a view,
these clitics in BCS prefer being non-initial over being edgemost, whereas their
Slovenian counterparts have the opposite preference.
Unfortunately, the OT conception of the facts characterizes clitic properties
purely in terms of linear order. It does not treat them as syntactic entities within
a hierarchical structure, but instead as phrasal morphology, as articulated for
example in ANDERSON (2005 and elsewhere). Such a purely PF linear account
discards the many structural aspects of clitic placement that have been widely
discussed in the literature.4 Indeed, under the standard OT morphological
account the mere fact that it is the left rather than right edge which is targeted
(that is, Edgemost (X, Left) and NonInitial) becomes an accident rather than a
principled consequence of the assumption that clitics move in the syntax.
Movement, on the other hand, always extends the tree, which means it will
always be upwards, hence, insofar as the LCA pertains, leftwards.
An alternative perspective, developed in FRANKS (2000), is that syntax is
fundamentally generative but optimality-like considerations police the
interfaces. Clitics move in the syntax, but the syntax leaves unspecified which
copy is pronounced. It is up to SpellOut to resolve this, selecting from what the
syntax offers the optimal PF instantiation. SpellOut thus picks the highest
prosodically viable copy, in keeping with the violable (i.e soft) constraint in (5)
and assuming the general principle (i.e undominated or hard constraint)
articulated in (6). It is the interaction of (5) and (6) which forces the higher copy
of the clitics to be pronounced in Slvn (3) but the lower copy in BCS (2b).
(5) PRONOUNCE HIGHEST: Lower identical copies are silent.
(6) PROSODIC SUPPORT: Everything pronounced must be prosodically parsed
In this way, the syntax provides the correct structure, but the phonology filters
out illicit copies. Prosodic considerations are irrelevant to the syntax, but in the
mapping to PF these play a critical role. Under this view, rather than ranking
an Iphrase non-initiality constraint lower in Slvn than in BCS, there would
simply be no reason to invoke a prohibition against Iphrase initiality in Slvn at
all. This prohibition, moreover, is just a property of specific lexical items; it is
not part of the syntax per se of the languages to which these items belong. That
is, different lexical items can be subject to different prosodic requirements, and
PF seeks to meet those requirements as efficient as possible in spelling out what
has been generated by the syntax.
1.2 Avoiding homophonous sequences
A second phenomenon that invokes lower pronunciation of copies to consider
involves multiple whfronting (MWF). As observed in BILLINGS & RUDIN
__________
4 I refer the reader to BOKOVI (2004) for a short summary of reasons why South Slavic
clitics should be analyzed as hierarchically arranged independent syntactic entities.
132 Steven Franks
(1996), there is a PF constraint against consecutive homophonous whwords in

MWF languages such as Bulgarian (Bg), stated roughly as follows:5
(7) Avoid Homophonous Sequences
Given this desideratum, in a multiple whquestion in Bg, although ordinarily the
highest copy is pronounced, when the two whwords are identical, it is the lower
copy of the second one which must be pronounced. Consider the minimal pair in
(8) and (9):
(8) a. Koj kakvo kupi? b. *Koj kupi kakvo?
who what bought
Who bought what?
(9) a. Kakvo obuslavja kakvo? b. *Kakvo kakvo obuslavja?
What conditions what?
Since pronunciation of the higher copy is blocked, (9) looks roughly like (10) in
PF:6
(10) Kakvo obuslavja [ kakvo]? (Bg)
This gives the ordering in (9a) rather than (9b).
Evidence for the strong PF adjacency nature of this constraint is shown by
BCS (11a), from BOKOVI (2002b). Here, the intervening adverb neprestano
constantly obviates the effect:
(11) a. ta neprestano ta uslovljava ?
What constantly conditions what?
b. ?*ta neprestano uslovljava ta?
As (11b) reiterates, pronunciation of the lower copy is only possible where
required.
These facts are also highlighted by the ungrammaticality of the order *na
kogo kogo to whom whom in Bg (12).
(12) a. *Koj na kogo kogo e pokazal?
who to whom whom aux.3SG pointed-out
Who pointed out who to whom?
__________
5 The constraint is part of a larger family of familiar PF constraints against sequences of
homophonous elements; I discuss the one dispreferring two sequences of je in BCS later
on.
6 I assume there is also Vmovement, so that the higher copy of the verb obuslavja
conditions is pronounced.
b. Koj na kogo e pokazal kogo?

c. Koj na koj kogo e pokazal?
In (12a), the first kogo is part of the dative phrase na kogo to whom, whereas
the second kogo is accusative. To avoid adjacent occurrences of kogo, a lower
copy of the second kogo is pronounced, as in (12b). However, as BILLINGS &
RUDIN (1996) note, when na kogo is replaced by the colloquial form na koj, as
in (12c), this disrupts the homophony, so that both AVOID HOMOPHONOUS
SEQUENCES and PRONOUNCE HIGHEST can be satisfied. The dative > accusative
sequence becomes once again felicitous, with the relevant structures given in
(13) and (14).
(13) a. Koj na kogo e pokazal kogo?
b. *Koj na kogo kogo e pokazal ?
(14) a. Koj na koj kogo e pokazal ?
b. *Koj na koj e pokazal kogo?
Ungrammatical (14b) again demonstrates that the lower copy cannot be
pronounced unless motivated by the need to satisfy a higher constraint. It is also
worth noting that there are two valid outputs, (13a) and (14a), depending on how
the indirect object is expressed (na kogo or na koj). Copy deletion can thus be
dependent on lexical access, in this case choice of the adjacent allomorph.
1.3 Intermediate summary
What do facts such as these suggest about SpellOut? For one thing, it seems to
me that, since linear adjacency and morphological identity are crucial in
deciding which copy of the whword to pronounce, linearization and choice of
lexical item to insert should precede copy deletion. Also, since prosodic
information is needed to determine which copy of the clitics to pronounce, (at
least some) prosodic phrasing must precede copy deletion as well. And it surely
makes no sense to impose prosodic structure until linear order has been
established either. So, very crudely, the facts so far suggest the following
ordering of SpellOut processes:
(15) lexical insertion > linearization > prosodification > copy deletion
On the other hand, one might expect copy deletion to precede linearization,
since otherwise the grammar would not know which copy to linearize.
Moreover, it has been argued by MORO (2000), NUNES (2004), and others that
copy deletion serves to render c-command unambiguously asymmetric,
something that is required if Kaynes LCA has anything to do with determining
linear order. Indeed, in FRANKS (2010), I concluded on the basis of the same
data that copy deletion should come first rather than last, just that it would be
vacuous before any movement/remerge, roughly as in (15).
(15) copy deletion > lexical insertion > linearization > prosodification
134 Steven Franks
It now however seems to me that the question of how to order these various
aspects of SpellOut is probably moot, in that they must be taken into
consideration all more or less at once; cf. also LPEZ (2009) for the interaction
of the LCA and prosody. In fact, the ordering of these processes is extrinsically
determined, since the structure cannot be linearized until multiple copies have
been resolved, nor can it be prosodified until linear order is determined, and
copies cannot be deleted until adjacency to a preceding prosodic impasse or
phonologically identical form has been established. The characterization in (15)
thus raises serious questions and may be problematic for a strictly phasebased
approach. Moreover, infelicitous adjacencies that force copy deletion can be
local in the sense of involving only a single phase. On the one hand, this
suggests that deletion can apply at any time, i.e., both before and after
prosodification and/or linearization, and that the traditional phase is not the
appropriate unit. On the other hand, as I will try to show below, there are clear
and much needed effects of making SpellOut decisions derivationally.
Perhaps, then, a better solution would be, as suggested above, to regard
interface conditions as constraints rather than rules per se. However, we would
need to apply them iteratively in order to preserve the derivational nature of the
mapping embodied in MSO. Under this view, it makes no sense to order them,
since they are not really processes applying to transform one representation into
another, but rather criteria for the comparison of candidates generated by the
syntax. SpellOut would then mean that morphological instantiation, selection
of copy, imposition of linear order, and appropriate prosodic structure are all
evaluated simultaneously, with convergence on the maximally efficient choice.
At the end of this paper I return to this idea and suggest that, properly construed,
there may not be any substantive difference between an OT and generative way
of implementing these insights.
2. A puzzle
The characterization in (15) of what happens when begs other questions. For
example, does the entire structure need to be evaluated at once or can an MSO
system work, in which chunks of structure are manipulated piecemeal?
For clitics, when the highest copy cannot be pronounced for PF reasons it is
the next highest prosodically viable copy that must be retained; blind
pronunciation of the lowest (tail) copy is never an option. This suggests a
cyclic algorithm for copy evaluation and, I will claim, for clitic linearization.7
For whphrases, on the other hand, it is the lowest copy that is the by far
preferred alternative. BOKOVI (2002b) cites the judgments for Bg in (16a),
__________
7 In FRANKS & HERRING (2010) as well as in FRANKS (in progress) I explore instead a
single top-down algorithm for the mapping from syntax to PF.
although speakers I have consulted find kakvo what before e that impossible
and before the verb obuslavja conditions degraded (16b).
(16) a. Kakvo (*kakvo) misli (*kakvo) Ivan (%kakvo) e (kakvo) obuslavja
(kakvo)?
What does Ivan think conditions what?
b. Kakvo (*kakvo) misli (*kakvo) Ivan (*kakvo) e (??kakvo) obuslavja
(kakvo)?
In Romanian, the judgments on comparable structures are also as in (16a)
hypothetical intermediate landing sites simply cannot be pronounced, only the
tail can.8 How should we interpret these facts?
Note that the wh feature driving the movement is at the top of the tree, as is
the sequence of homophonous elements which determines that the tail wins. I
see no way of implementing this unless the intermediate sites are not taken as
involving literal (re)merge of a copy at all. This conclusion in fact seems
inescapable, given that it is the tail which must be pronounced when putative
successive-cyclic whmovement fails in English, as in (17):
(17) [CP Who [TP thinks [CP {? ?} that [TP Mary bought what]]]]?
If what moves to the intermediate SpecCP, regardless of how this movement is
motivated, and if MSO is assumed, then how can the lowest copy be activated
just in case the top landing site (being occupied by who) is unavailable? This is a
classic problem, considered inter alia by myself in FRANKS (2006a) as well as
CHOMSKY (2001), BOKOVI (2007), BOECKX (2008 and other works).
Everything points to the conclusion that the intermediate copy of what in (17) is
really not there at all. Thus it cannot be bound or license parasitic gaps, as
shown by (18a) and (18b), respectively.
(18) a. *Who said that Mary thinks {? ?} that Bill
likes which pictures of herself?
b. *Who said that you claimed {? ?}that you loved which
painting [in order to get me to look at pg]?
The fact of the matter is that LF movement invariably behaves like no
movement, which is why the higher unpronounced copies are inside curly
brackets flanked by question marks. The problem here is how does a bottom-up
system know that overt movement is not going to succeed and that LF
movement will eventually be required? In contrast, real, overt, successive-cyclic
__________
8 A curious puzzle I address in FRANKS (in progress) is why, in varieties of German which
allow pronunciation of whwords in intermediate positions, this is never possible for the
tail:
(i) [CP Wen denkst du [CP wen sie meint [CP wen Harald (*wen) liebt]]]?
Who do you think that she believes that Harald loves?
136 Steven Franks
movement does have empirical effects which target the intermediate landing
site. Compare for example (19) to (18):
(19) a. [CP Which pictures of herself does [TP Mary [VP think [CP
(that) [TP Bill [VP likes
]]]]]]]]]?
b. Which painting does Mary think [CP that [TP you
claimed that you love ] [in order to get me to look at
pg]]]?
In (19a) the intermediate copy of herself is bound by Mary, in (19b) the
intermediate copy of which painting licences the parasitic gap.
Another illustration of the absence of expected intermediate PF effects
under hypothetical LF whmovement can be constructed on the basis of V-
Preposing inversion in Spanish (20), based on TORREGO (1984):
(20) a. [CP Qu pensaba Juan [CP que le haba dicho Pedro [CP que
haba publicado la revista]]]?
What did Juan think that Pedro had told him that the journal had
published?
b. [CP Quin pensaba [CP {? ?} que Pedro dijo [CP {? ?}que Juan
haba publicado qu]]]?
Who thinks that Pedro said that Juan published what?
In (20b) the whphrase qu what cannot occupy the matrix SpecCP because it
is filled by quien who. Concomitantly, the inversion which arises with overt
movement in (20a) is also missing. As before, not only is the tail pronounced,
but all intermediate effects also disappear:
One solution that makes sense to me takes advantage of TAKAHASHIs
(1994) Form Chain approach. According to this account, it is not until the
attracting head has been merged and provides a targeta [+wh] C in these
instancesthat a chain is formed to create the intermediate sites. These are
introduced in accordance with the Minimal Link Condition, replicating the
effects of successive cyclic movement, but in one fell swoop and without the
look-ahead problem. The advantages to a system such as Takahashis have been
argued for in various works by Boeckx and Bokovi, inter alia. For present
purposes, the crucial point is that this may provide a way of distinguishing lower
copy pronunciation of clitics (where the next one down wins) from lower copy
pronunciation of wh-phrases (where one in the most embedded clause wins) by
distinguishing clitic movement from whmovement. The latter goes across
phases and is driven by a [+wh] feature in the highest C0. The former is typically
phase internal and does not seem to be feature driven at all. Moreover, much
evidence shows that clitics need not end up in a consistent position; hence it is
hard to imagine their movement as any sort of instantiation of Attract.9
3. Clitics as Formal Feature Bundles
I take clitics to be pure instantiations of Formal Features (FF). That is, they are
devoid of all lexical content features. In this sense, clitics can be regarded as the
SpellOut of functional heads. For example, sam in BCS (2) is just 1st singular
verbal agreement and ti is just 2nd singular dative case. One persistent view
of 2P clitics, as in BCS and Slvn, is that they move to the highest head position
in some appropriate functional domain. The obvious question is then How do
they get there? and a reasonable answergiven that verbs are canonical hosts
for special clitics10is that they somehow piggy-back on the verb. Since I have
attempted to flesh this out in previous work, my intention here is simply to
sketch out the general idea of how this piggy-backing might work.11
3.1 How to get clitics higher?
A compelling reason for the verb to move up through its extended projection is
that its FF need in some way to match those of all associated functional heads.
Of course, it is just the formal features of V (rather than the entire V) which
need to move. So in a verb second language, such as German, the semantic
features of the particular verb are pied-piped, whereas in other languages there is
just FF movement.12 Putting aside the various technical scenarios one could
imagine to make this all work (cf. FRANKS 2000), my basic claim is that the FF
of the verb are copied up the tree and that this happens in a stepwise fashion,
with successive head adjunctions, so that at the end the verbs FF plus those of
all the clitics are situated on the highest head. The question this then raises is
how to exploit the insight that the verb provides a syntactic host for the clitics
even when the clitics are not actually pronounced adjacent to the verb.13
Consider for example the BCS Croatian variants in (21):
__________
9 I put these issues aside in this paper, returning to them in FRANKS (in progress), which
explores a multiattachment solution to top-down chain formation.
10 Adopting Zwickys terminology, special clitics are those reflecting paradigmatic
functional categories such as case, tense, or agreement.
11 I also avoid specifics about which functional heads the various clitics might instantiate or
the kind of cartography involved.
12 Pied-piping is presumably what the diacritic feature strong means, following the
insight in e.g. GROAT & ONEIL (1996) or RICHARDS (1997) that strong amounts to
nothing more than an instruction to PF to pronounce an element in that position.
13 In languages like Bg and Macedonian (Mac) most clitics are positioned with respect to
verb, but in 2P languages, such as BCS and Slvn, the surface position of the verb is
irrelevant.
138 Steven Franks
(21) a. Sestra ih rado poklanja kolskoj knjinici.

sister.NOM them.ACC gladly gives school.DAT library.DAT
Sister gladly gives them to the school library.
b. Sestra ih kolskoj knjinici rado poklanja.
c. kolskoj knjinici ih rado poklanja sestra.
d. Rado ih sestra poklanja kolskoj knjinici.
The key lies in moving the verb overtly in some way that does not require it to
be pronounced in its target position. In essence, although the verb must move to
2P and take the clitics along with it, the clitics need to be realized in that
position while the verb is realized lower down. The question is thus: What
exactly differentiates the clitics from the verb?
My answer is that clitics are pure FF bundles whereas the verb contains
substantive semantic features as well. Thus, if only FF are copied, and if there is
a desideratum as in (5) that the highest copy that can be spelled out must be,
then the clitics are going to pronounced high but the verb cannot be. The verb is,
instead, pronounced, in the highest position to which a copy of its semantic
features have been pied-piped.14 Although the idea is hardly new that 2P for
verbs and clitics is intimately relatedstemming as it does from WACKERNAGEL
(1892) and recently exploited in non-derivational frameworks such as
ANDERSON (2005)the reason why clitics are pronounced higher than the verb
finds a conceptually well grounded explanation only in the context of a model
which formally distinguishes clitics from verbs.
In fact, the scenario of pronouncing elements that are exhaustively FF
bundles higher than their fully lexical counterparts is widespread. A hallmark of
clitics themselves is that they typically appear higher than full noun phrases. The
reason is simple: although FF movement applies equally to clitics and lexical
NPs, (in the absence of pied-piping) this scatters the various features of lexical
NPs but leaves clitics, as pure FF bundles, whole and intact. ROBERTSs (1998)
account of why English auxiliaries raise to T whereas main verbs do not
expresses the same kind of ideaand indeed, he comments that another
obvious place to look ... is the area of clitics. For Roberts too, FF is always
and only overt; strong features in addition cause the entire category to be pied-
piped. Taking verb features in English to be weak, lexical verbs are as expected
pronounced in situ. Auxiliaries, however, only have FF, so that when Move F
moves all features of the element it moves checking the weak feature of the
V node causes the entire auxiliary to move (ROBERTS 1998: 119).
__________
14 This view of movement is very much akin to that put forward in ZWART (1997), for
whom overt movement involves both FF and semantic features (which he calls Lexico-
Categorial), whereas movement traditionally analyzed as covert is in fact overt but
only involves FF.
3.2 Clitics and linearization

This section endorses an approach to clitic movement, due originally to
BOKOVI (2002a), that is compatible with Kaynes LCA.
3.2.1 On why clitic + verb order is more basic
Because pronunciation of verbs and clitics is as we have seen divorced for 2P
clitics as in (21), one cannot tell what the order between them and the verb
really is. However, when one considers verb-adjacent clitics, as are typical in Bg
and Mac, one finds that the clitics always precede the verb if they can. Within a
clause, such clitics ordinarily appear immediately before the verb, regardless of
how much material precedes them, as in Bg (22).
(22) a. Ti minulata godina si mu gi pokazvala.
youlast year aux.2SG him.DAT them.ACC shown.FEM
You have shown them to him last year.
b. Vera v gradinata Mila sigurno mu dade knigite.
yesterday in garden.DEF Mila surely him.DAT gave books.DEF
Yesterday, in the garden, Mila surely gave him the books.
The exception is what in the Romance linguistics tradition is known as the
ToblerMussafia (TM) effect, according to which, if the clitics would end up
in absolute initial position, they must then follow the verb instead:
(23) a. Pokazvala si mu gi. (*Si mu gi pokazvala.)
shown.FEM aux.2SG him.DAT them.ACC
You have shown them to him.
b. Dade mu knigite vera. (*Mu dade knigite vera.)
gave him.DAT books.DEF yesterday
She/he/you gave him the books yesterday.
These Bg clitics are subject to a PF prohibition against being initial. However,
the domain of this non-initiality restriction, although prosodic, is not the I-
phrase, as it is for 2P clitics, but rather the relevant domain is NESPOR &
VOGELs (1986) Utterance. That is, the Bg clitics are subject to (24), as
opposed to the BCS ones subject to (4):
(24) NONINITIAL(UTTERANCE)
The relevance of this larger prosodic domain is further demonstrated by (25),
with Utterance boundaries indicated by .
(25) a. #I si mu gi [pokazvala] #
And you have shown them to him.
b. #Mila# #edna moja prijatelka# #[mi go [dade] ]#
Mila one my friend me.DAT him. ACC gave
Mila, a friend of mine, gave it to me.
140 Steven Franks
The conjunction i, although not a prosodic word or viable host by itself, is

sufficient to render si mu gi non-initial in (25a), and in (25b) the clitics mi go are
oblivious to the intonational phrase boundary necessarily introduced by the
parenthetical edna moja prijatelka a friend of mine; neither of these would of
course be good in BCS. In sum, since the generalization is that the clitics are
pre-verbal unless this would leave them in absolute initial position within the
clause, I contend that linearization first makes them preverbal and then this
ordering is adjusted as needed. Further indication that linearization should
abstract away from the TM effect is the fact that Mac clitics, although
syntactically very similar to Bg, do not display the PF restriction in (24): its
clitics can be initial in the Utterance. Mac is thus the best indicator of what is
really going on between the clitics and the verb, since it involves minimal
confounding effects. And indeed, a comparison of Mac (26) with Bg (23) shows
that the word order facts here are the opposite:
(26) a. Si mu gi. pokazvala.
aux.2SG him.DAT them.ACC shown.FEM
You have shown them to him.
b. Mu dade knigite vera.
him.DAT gave books.DEF yesterday
She/he/you gave him the books yesterday.
I would thus argue that a promising way of understanding Bg (23) is to derive
TM on the PFside of the grammar. I will return to the specific derivation
below: my point here is simply that the syntax would produce an output in
which the clitics precede the verb and that this order would be adjusted on the
PFside to comply with the prohibition in Bgbut not Macagainst Utterance
initial clitics.
3.2.2 Bokovis system for deriving clitic + verb
Kaynes LCA has the effect of left-adjunction: when a head A moves to another
head B, it adjoins to Bs left, producing the schema in (27).
(27) [[ A + B ] ... ]
Taking the LCA to be a principle of linearization means, however, that the
clitics must move to adjoin to the verb, and not vice versa, since, everything else
being equal, the clitics precede the verb. If, however, the clitics instantiate
functional heads above the verb, such as Tense or Agreement, this implies that
the verb must move past the clitics, thereby providing a target to which the
clitics can then adjoin (and linearize to the left, as per the LCA). In Bokovis
system the syntax provides something like (28) for the clitic-verb order mi go
dade in (25b). First, dade moves past go, as in (28a). Next, go adjoins to dade,
as in (28b). Then, mi is merged and the go + dade complex later moves past
that, as in (28c). Finally, mi adjoins to go + dade producing the order in (28d),

with copies pronounced as indicated.
(28) a. [[dade] go ... dade ...]
b. [[go + dade] go ... dade ...]]
c. [[go + dade] [mi [[go + dade ] go] ... dade ...]]
d. [[mi + [go + dade ]] [ [[ + ] ] ... ...]]
In sum, once the verb dade gave has moved past the clitic go, go can adjoin to
it. Then, once the go + dade complex head moves past the clitic mi, mi can
itself adjoin to that. The result is iterated left adjunction, with all the clitics
preceding the verb.
4. Further aspects of linearization
The syntax per se makes no statements whatsoever about linear order, which is
only introduced on the PFside as part of the SpellOut process. So the question
at hand is: How are hierarchical syntactic structures ultimately mapped into flat
ones? I now explore some of the mechanics of this process, focusing on Bg
clitics. I will argue that linearization is iterative and that, contra the system
espoused for example by FOX & PESETSKY (2005), there is no order
preservation. Rather, linearization is an on-going processnot the compilation
of a set of immutable statements about precedenceand linear order can be
manipulated by resubjecting lexical material to it as needed. As stated, however,
the first and most potent linearization principle is the LCA, Kaynes claim that
asymmetric c-command maps into precedence. This applies cyclically, on-line,
with the effect that heads precede their complements and adjunction is linearized
to the left. Under this view, as just outlined, the Slavic special clitics precede
their hosts.
4.1 Invoking prosodic considerations
In many of the Slavic languages, however, there is one simple clitic that displays
very different behavior. This is the lexical item li, which as described in FRANKS
(2006b) and references therein, is always unequivocally enclitic and has
interrogative and focus functions of various types in languages such as Russian,
BCS, Czech, Mac and Bg. Here I concentrate on Bg, which exhibits some of the
more spectacular linearization effects.
The clitic li instantiates Yes/No and focus features in the highest head
position in the tree, which for the sake of discussion will be taken to be C0. As a
clitic, li projects no prosodic structure itself and, as an enclitic, li is always
pronounced at the right edge of an adjacent host prosodic word.15 In terms of
non-initiality, this means that, as a lexical property, li cannot be the initial in its
prosodic word. The relevant constraint is stated in (29).
__________
15 I take this to mean the domain of NONINITIAL for li is the prosodic word.
142 Steven Franks
(29) NONINITIAL(PROSODIC WORD)

If preceded by a possible host, such as grad in Bg (30), which is part of the
phrase in SpecCP v tozi grad, then li is linearized at the right edge of that host:
(30) [V tozi] [[grad] li] [si [xodil] ]?
in this city Q aux.2SG went
Was it to this city that you went?
Sometimes, however, SpecCP is empty so that there is nothing higher than li.
When this happens, li is linearized at the right edge of the prosodic word to its
right. Thus, li differs from the special clitics in Bg in following the verb. Recall
that these latter precede the verb if at all possible: in (31a) the conjunction i
and renders the clitics mu gi non-initial, in (31b) the future proclitic te, does,
and in (31c) the sentential negation element ne does:16
(31) a. [I mu gi DAde] vera.
and him.DAT them.ACC gave yesterday
And she/he/you gave them to him yesterday.
b. [te mu gi predaDE].
will him.DAT them.ACC hand-over.2SG
You will hand them over to him.
c. [Ne MU] [DAde] knigite.
neg him.DAT gave books.DEF
She/he/you didnt give him the books.
We see in (32), however, that under the exact same conditions li necessarily
follows the verb:
(32) a. [[I DAde] li] knigite vera na Ivan?
and gave Q books.DEF yesterday to Ivan
And did she/he/you give the books to Ivan yesterday?
b. [[te predaDE] li] knigite na Ivan?
will hand-over.2SG Q books.DEF to Ivan
Will you hand the books over to Ivan?
c. [[Ne DAde] li] knigite na Ivan?
neg gave Q books.DEF to Ivan
Didnt she/he/you give the books to Ivan?
That is because lis enclitic nature causes it to be linearized after the adjacent
prosodic word, which, in (32), is i dade, te predade, and ne dade.
With post-stressing ne, li appears between the clitic and the verb, as in (33),
which is exactly what one expects if ne mu in (31c) is indeed an independent
prosodic word:
__________
16 Note that in standard Bg ne is post-stressing, so the dative clitic mu in (31c) is actually
stressed (word stress is indicated by upper case).
(33) [[Ne MU] li] [DAde] knigite.

neg him.DAT Q gave books.DEF
Didnt she/he/you give him the books?
Such examples, where li fails to respect the LCA, show that this is not a hard
and fast principle of linearization. Rather, as LPEZ (2009) also concludes, the
LCS is soft constraint:
(34) LCA: C-COMMAND LINEARIZES AS PRECEDENCE
The linearization of li can override (34) in order to satisfy PROSODIC SUPPORT in
(6), in keeping with the lexical properties of li.
Even more interesting is the question of what happens if there is second
special clitic, such as accusative gi them. As seen in (35a), gi is most naturally
proclitic on the verb following it, with ne mu independently a viable prosodic
word. Consequently, li is still prosodified after adjacent ne mu, except that now
li appears between the two pronominal clitics. This is shown in (35b):
(35) a. [[Ne MU] [gi [DAde] ] [Ana] (merger of li)
neg him.DAT them.ACC gave Ana
Ana didnt give them to him.
b. [[Ne MU] li] [gi [DAde] ] [Ana]?
neg him.DAT Q them.ACC gave Ana
Didnt Ana give them to him?
LCA-driven linearization first produces the order in (35a), which is prosodified
as indicated. Next, li is merged in C0. If SpecCP is filled, as in (30), li can be
linearized to satisfy both the LCA and its prosodic requirement of seeking
support to its left. However, if nothing is in SpecCP, li is linearized as
prosodically adjoined to the right edge of the adjacent minimal prosodic word,
which in (35b) is ne mu. This is the best example of prosodically conditioned
linearization that I know of.
Splitting of mu + gi by li in (34b) is particularly striking since nothing else
can ever interrupt such a sequence of two pronominal clitics. This is not
generally true of the clitic + verb cluster in Bg, which as I have discussed in
FRANKS (2005, 2006b), tolerates some intervention by various tonic aspectual
adverbials, such as vse ote still in (36) or vee already in (37), or destressed
emotive particles like u in (38).
(36) a. Ti ne SI mu go vse ote dal.
youneg aux.2SG him.DAT him.ACC still gave
You still have not given it to him.
b. ?Ti ne SI vse ote mu go dal.
c. *Ti ne SI mu vse ote go dal.
(37) a. Az sm ti ja vee dala.
I aux.1SG you.DAT it.ACC already gave
144 Steven Franks
I already gave it to you.

b. ?Az sm vee ti ja dala.
c. *Az sm ti vee ja dala.
(38) a. Az sm ti go u dala.
I aux.1SG you.DAT it.ACC already gave
I have maybe already given it to you.
b. Az sm u ti go dala.
c. *Az sm ti u go dala.
Judgments about adverb interpolation are somewhat uncertain, since, although
produced by the syntax, they require appropriate prosodification, which in turn
requires appopriate discourse context. So, given this prosodic filtering effect,
something like (37b) implies a particular context and intonation; particle
splitting, as in (38b), is more natural. But the point about adverb interpolation is
that the pronominal subcluster cannot ever be split. Except by li.
4.2 On the cyclicity of linearization
I now show why the linearization process must be iterative. While this is hardly
a heretical view, what has I think gone underappreciated is the degree to which
linear order is labile.
Recall the TM effect, which in Bg adjusts clitics left by the syntax in initial
position, as in (22) versus (24). More accurately, of course, the syntax does not
leave them with any order, since all that the syntax expresses is hierarchical
relations. But taking asymmetric c-command to map into precedence, Bg (22a)
and (24a) would have the initial SpellOut orders in (39).17
(39) a. Si mu gi pokazvala.
aux.2SG him.DAT them.ACC shown.FEM
b. I si mu gi pokazvala.
and aux.2SG him.DAT them.ACC shown.FEM
And you have shown them to him.
Once prosodic structure is imposed, the clitics in (39a) but not those in (39b)
end up being initial in their Utterance, hence in Bg the Clitic Group si mu gi,
demarcated CG,18 must be relinearized, as in (40):
(40) a. #[[si mu gi]CG [pokazvala] ]# # [[pokazvala] [si mu gi]CG ]#
b. #i [si mu gi]CG [pokazvala] #
I understand this process as one which erases the precedence relation between
the clitic group and its adjacent prosodic word and redefines it in a way that is
__________
17 Note that the representations in (39) are before TM, so that (39a) is acceptable in Mac
but not in Bg; (39b) on the other hand does not lead to TM, so it is acceptable in both.
18 At this point I adopt Nespor and Vogels CG domain, but later suggest an account
without it.
consistent with the non-initiality constraint pertaining to the Utterance level of

prosodic organization.
One might wonder why I bother first to linearize the clitics in front of the
verb and subsequently to relinearize them to follow the verb. This is probably
my most radical claim, since it implies a strongly derivational model. It also
raises the important question of when initial linearization takes place, a question
to which I will return shortly. In any event, given these facts, one should ask
how li linearization, which is sensitive to the prosodic word, interacts with TM
linearization, which is sensitive to the Utterance. My claim is that the
interrogative version of Bg (39a)/(40a) begins, after LCA linearization, as in
(41a) but eventually emerges as in (41b):
(41) a. li si mu gi pokazvala
Q aux.2SG him.DAT them.ACC shown.FEM
b. Pokazvala li si mu gi?
Have you shown them to him?
How is the order in (41b) derived? I propose a derivational scenario which
roughly proceeds through the steps sketched out in (42):19
(42) a. [[si mu gi]CG [pokazvala] ] (merger of li)
b. [[[si mu gi]CG [pokazvala] ] li] (SpellOut as Utterance)
c. #[[[si mu gi]CG [pokazvala] ] li ]# (TM)
d. #[[[pokazvala] li ] [si mu gi]CG ]#
The idea, as before, is that all heads are first linearized to the left, following the
LCA, but, unlike in the simple case of (39a), the special clitics cannot be
considered Utterance initial until after li has been merged in C0 and is itself
linearized, as in (42b). The entire CP is then sent to SpellOut and prosodified
as an Utterance, indicated in (42c). At this point the clitic group violates the
prosodic prohibition against Utterance initiality. Hence, its linearization with
respect to its host is erased and reevaluated to follow pokazvala li, with the
effect in (42d). It is this kind of double right wrap derivation that gives rise to
the appearance of li splitting the verbal participle pokazvala shown from the
clitics si mu gi.
It is worth observing that addition of the conjunction i and has no effect on
these word order facts, something that is expected only if CP must be
prosodified as an utterance. The reason i is of interest is that this element is
sufficient to render the special clitics not Utterance initial, as shown by (39b),
but is not itself tonic and so cannot host prosodic enclitics, such as li. What I
have noticed and reported in number of places, most accessibly FRANKS &
BOKOVI (2001), is that, although i saves the clitics in (39), it fails to in (41).
Apparently, by the time li is sent to SpellOut it is too late. That is, the
__________
19 Once again, (42c) is the final version in Mac, but in Bg TM applies to produce (42d).
146 Steven Franks
determination that the clitics are Utterance initial and the resolution of this
problem that gives rise to the TM effect must precede merger of the conjunction
i above the interrogative li. This is however exactly what one would expect if i
is a higher head, necessarily outside of CP. That is, CP is sent to SpellOut, it is
prosodified as an Utterance, linearization is reevaluated to respect the clitics
non-initiality requirement, and by the time i is merged the clitics have already
been relinearized so that they follow li. The result, in short, is (43), which is just
(41b) plus i.
(43) I pokazvala li si mu gi?
And have you shown them to him?
When li is present, i is irrelevant to linearization.
There are several crucial assumptions implicit in my account of these data
that can tell us a lot about the timing of SpellOut. In the derivation of (39b) the
string si mu gi pokazvala cannot be prosodified as an Utterance, it must wait
until i has been merged, otherwise the wrong order would obtain. In the
derivation of (43), on the other hand, waiting is impossible: li defines a CP and
this must be prosodified as an Utterance. To accomplish this I assume SELKIRKs
(to appear) Match Theory, which exploits canonical correspondences in
mapping syntactic phrasing into prosodic phrasing (as opposed to more
traditional edge or alignment based approaches). Her system is given in (44):
(44) Selkirks Match Theory
a. Match Clause: Syntactic clause Intonational phrase ()
b. Match Phrase: Syntactic phrase (XP) Phonological Phrase (I)
c. Match Word: Syntactic word Prosodic Word ()
One source of possible confusion is that I have adopted Nespor and Vogels
terminology, following most others who discuss Slavic clitics, so that what I
have called an Utterance is Selkirks I-phrase and what I have called an I-phrase
is her Phonological Phrase. The basic idea is however the same: a clause, here a
CP, maps into a large unit of prosodic organization, a word maps into a small
one, and a phrase is in between.20 My point is simply this: a CP is spelled out as
an Utterance, which means that whatever is in SpecCP necessarily counts, hence
if anything has moved to SpecCP, li is able to satisfy the LCA in (30). But
anything outside of CP, such as the i in (43), is not considered. Hence the li
facts, taken together, require spelling out of CP.21
__________
20 A more careful account might be couched in Ito and Mesters new prosodic adjunction
model.
21 In FRANKS & BOKOVI (2001), we took this TM paradigm as evidence for CHOMSKYs
(2001) phase theory. However, in Chomskys system, where CP is a phase but its
complement IP is not, it is actually the complement to a phase head that is sent to Spell
Out, not the phase itself. Clearly, this does not work for the Bg data, which require CP
itself to be spelled out. Moreover, all other post-LCA linearization facts refer to prosodic
5. Aspects of ellipsis
This section discusses several aspects of ellipsis that a successful model of the
SpellOut process should accommodate.
5.1 Ellipsis and resolution of PF offense
There is a vast literature on ellipsis of diverse types. I follow MERCHANT (2001)
among others in taking ellipsis to be PF deletion than LF copy. One
implementation is to mark a recoverable constituent for flat intonation and
then delete it in the mapping to PF. This is tantamount to non-pronunciation of a
syntactic constituent, i.e., mapping to silence in PF. Standard ellipsis must be
fairly early in the SpellOut process, since it crucially targets syntactic
constituents. Indeed, ellipsis is generally taken as a textbook standard for
diagnosing constituency.
In this context, the capacity for ellipsis to rescue derivations syntactically
illicit structures is of particular interest; cf. especially MERCHANT (2001) for
copious examples and discussion. To illustrate, LASNIK (1999) considers why V
fails to raise when there is pseudogapping (VP-ellipsis), as in (45a), and why I-
to-C movement fails when there is matrix sluicing (IP-ellipsis), as in (45b).
(44) a. You might not believe me but you will {*believe*} Bobi [VP believe ti ]
b. A:Mary will see someone.
B: Whoi {*will*} [IP Mary will see ti ]?
Lasniks insight is that ellipsis can save these derivations by deleting structure
containing offending features.22 The puzzle is that, in order for ellipsis to have a
salutary effect, the offending feature should be within the elided constituent and
the violation it creates is not evaluated until PF. In the end, Lasnik comes up
with a complicated solution that exploits FF movement and appeals to the idea
that ellipsis and pied-piping movement both provide ways of circumventing PF
crash caused by a strong feature on the target. Regardless of the specifics of his
analysis, it seems to me that the general idea could easily apply to PF clitic
violations as well.
Consider a well-known (if controversial) paradigm that seems to militate
against a syntactic account of clitic cluster formation. It is based on the BCS
ellipsis data in (46), discovered by STJEPANOVI (1998a, 1998b) and also
discussed in BOKOVI (2001a), FRANKS (1998/2010), RICHARDS (1997) and
PROGOVAC (2000):
_________
rather than syntactic domains. Since linearization is essentially a PF matter, this is as
expected.
22 LASNIK (1999) offers various ways of implementing this failure of believe or will to
move just in case ellipsis of VP or IP will occur, depending on how strength is construed.
The details do not concern us here.
148 Steven Franks
(46) a. Ona mu ga je dala, a i

she him.DAT it.ACC aux.3SG gave and also
ja sam mu ga [ ].
I aux.1SG him.DAT it.ACC gave
She gave it to him, and I did, too.
b. Ona mu ga je dala, a i ja sam mu [ ].
c. Ona mu ga je dala, a i ja sam [ ].
d. ?*Ona mu ga je dala, a i ja sam [ ] ga [ ].
If ellipsis here always targets a phrasefor the sake of concreteness assume
AgrIOP dominates AgrOP which in turn dominates AuxPthen Stjepanovis
paradigm implies AuxP-ellipsis in (46a), AgrOP-ellipsis in (46b), and AgrIOP-
ellipsis in (46c). The ungrammatical (46d), on the other hand, could only be
obtained by deleting the AgrIO dative head mu independently of AuxP-ellipsis.
On the one hand, these facts provide additional support for my contention that,
wherever they end up, clitics are introduced as separate functional heads. On the
other hand, they also raise a problem for any strictly syntactic approach to clitic
placement: ellipsis must target the phrase containing the clitic before the clitic
continues its upwards move. This strikes me as exactly like Lasniks problem
with verb and auxiliary movement, so that it should have the same kind of
solution, one in which ellipsis trumps movement.
5.2 Ellipsis and prosodic incompatibilities
Turning now to a different sort of prosodic motivation for ellipsis, I briefly
mention the possibility that the ellipsis sometimes can serve to resolve prosodic
paradoxes. In her dissertation on Bg, for example, LAMBOVA (2004) makes
considerable use of intonation clash, which is encountered if one tries to impose
both topic and focus intonations on the same material. She argues that fronted
topic and focus both freely adjoin to a discourse-related P projection, but
that intonation clash filters out the order in which focus precedes topic,
schematically as in (47).
(47) a. TOPIC > FOCUS: no intonation clash
b. *FOCUS > TOPIC: intonation clash!
Lambova also argues that if pronunciation of a lower copy can avoid the
potential clash, then that lower copy is pronounced. One of her examples is
reproduced in (48):
(48) a. Tortata koj kde te pora?
cake.DEF who where will order
The cake, who will order where?
b. [SpecCP [C0 [P tortata [SpecP [koj [kde]] [0 [IP te pora ...
If so, as with the AVOID HOMOPHONOUS SEQUENCES facts discussed above, this
is another place where PF considerations prevent the highest copy of a wh-
phrase from being pronounced. Once again, prosodification needs to feed copy
deletion; moreover, the prosodic clash can surely only be identified after initial
LCA-driven linearization.
5.3 Ellipsis and focus
I have recently encountered a different, far more superficial type of ellipsis that
does not recognize syntactic constituency. The relevant data are from colloquial
Croatian and have to do with focus-driven ellipsis in that language. Although
traditional ellipsis involves deletion of entire copies, there is also evidence for
scattered deletion of pieces of copies, the clearest examples of which are
focus-driven. For example, alongside Bg (30), repeated in (49a), one can also
have (49b).
(49) a. [V tozi] [[grad] li] [si [xodil] ]?
in this city Q aux.2SG went
Was it to this city that you went?
b. [[V tozi] li] grad [si [xodil] ]?
Was it to THIS city that you went?
The difference is that in (49a) the focus is on CITY, but in (49b) the focus is on
THIS. I would argue that (49b) requires there to be copies of v tozi grad
immediately above and below li, which is in C0 and which forces the
discontinuous deletion because li demarcates the focus. The basic idea is that
non-focus material cannot follow the focused element in the fronted phrase. This
is schematized in (50), making use of Lambovas P phrase structure system:
(50) [CP [V [tozi ]] li [P [ [ grad]] si xodil]]?
[+Foc]
A similar example splitting a whphrase is given in Bg (51a), with deletions as
in (51b):
(51) a. Koja li kniga te mi podari?!
which Q book will me.DAT give.2SG
Which(ever) book will you give me?!
b. [CP [Koja [ ]] li [P [ kniga] te mi podari]]?
[+Foc]
These splits are striking because Bg, as a DP language,23 does not otherwise
permit Left-Branch Extraction (LBE). Consequently, this split cannot be derived
through movement of koja. However, because no non-focus material can follow
the locus of the focus feature in SpecCP, li induces scattered deletion, hence
superficial splitting. It is worth noting that this phenomenon is different from the
kind of li splitting displayed by Russian, since, as discussed by RUDNICKAYA
__________
23 For Bg as a DP language and the associated impossibility of LBE see BOKOVI (2005).
150 Steven Franks
(2000) among others, in Russian splitting is more or less obligatory, even with
traditional fortresses as in (52):
(52) a. ?Anna li i Petr priezajut vo vtornik?
Anna Q and Peter arrive.3PL on Tuesday
Is it ANNA AND PETER who are coming on Tuesday?
b. *Anna i Petr li priezajut vo vtornik?
Similarly, the Russian version of (49) requires splitting regardless of which part
of the PP is focused; only (53b) is good:
(53) a. *V tot gorod li ty ezdil? b. V tot li gorod ty ezdil?
The reason for this is that in Russian there is simply no copy above li for
scattered deletion to exploit, hence the only option ever viable for Russian li is
to be pronounced at the right edge of the prosodic word to its right.24
I now turn to an even more superficial type of scattered ellipsis, leading to
sometimes dramatic surface splits, which apparently operates in colloquial
varieties of Croatian. The judgments are due to University of Zagreb Professor
Anita Peti-Stanti; here I just point out some of the highlights, since the facts are
also reported in FRANKS (2010). The basic problem is that, for Peti-Stanti,
splitting can leave non-constituents on both sides of the splitter. Consider
Croatian (54) with the splitting as before effected through ellipsis of string
adjacent material, as indicated in (55):
(54) U izuzetno sam veliku sobu uao.
in exceptionally aux.1SG large room entered
It was into an EXCEPTIONALLY large room that I entered.
(55) [PP U [NP [AP izuzetno ] ]] [sam

[+Foc]
[PP [NP [AP veliku] sobu]] [uao ...
The operative principle is once again that there can be no material within the
fronted phrase that is to the right of the element bearing the [+Foc] feature. All
material following the [+Foc] element is deleted, resulting in pronunciation of
the next highest copy of that material. Crucially, in this system material on
neither side of the [+Foc] element is required to be a constituent; instead, all that
matters is linear adjacency. Note also that for Peti-Stanti non-clitics can also
intervene, a possibility which gives rise to more complex structures than in (55).
Now for the dramatic part. PETI-STANTI (2007), in examining whether spoken
Croatian really adheres to second position for clitics, offers the examples in
(56), in which the clitic u follows whatever word is focused:
(56) a. Svakog u lijepog dana putovati vlakom svojoj kui.
__________
24 This fact correlates with the general lack of superiority effects for Russian whphrases.
every fut.1SG nice day travel train selfs house.

Every nice day I will go to my house by train.
b. Svakog lijepog u dana putovati vlakom svojoj kui.
c. Svakog lijepog dana u putovati vlakom svojoj kui.
d. Svakog lijepog dana putovat u vlakom svojoj kui.
e. Svakog lijepog dana putovati vlakom u svojoj kui.
f. Svakog lijepog dana putovati vlakom svojoj u kui.
Particularly problematic to derive are (56e, f), in which the clitic is lower than
immediately after the verb, since this is generally impossible. Crucially, there
cannot even be a copy of future clitic u that low in (56e) or (56f). These data
thus reveal that what is pronounced in front of the clitics in colloquial Croatian
must actually be higher. I therefore conclude that they must derive from a
syntactic structure as in (57):
(57) [Svakog dana putovati vlakom svojoj kui] u [svakog dana putovati
vlakom svojoj kui].
Essentially, there is a copy of the entire clause both above and below u; ellipsis
then applies to produce the various possibilities in (56), depending upon where
the focus is.
PP-internal splitting provides another interesting illustration of the same
general principles. Consider the split in (58b), again provided by Peti-Stanti:
(58) a. Od juer ga prodaje za velike novce.
from yesterday him.ACC sells for large money
Since yesterday (s)hes selling it for big bucks.
b. Od juer prodaje za velike ga novce.
c. *Od juer prodaje za ga velike novce.
Although (58b) is quite unexpected under familiar models of clitic placement, if
the entire clause is fronted past ga, it can be derived as in (59):
(59) [Od juer prodaje za velike ] ga
[+Foc]
[ novce].
Consider now (58c), in which the clitic cannot go immediately after the
preposition. The proper formulation of the largely universal cohesion between
preposition and complement is a widely researched topic, about which I have
nothing to contribute. However, if I am correct that whatever is going on here is
a matter of PFside deletion, one wonders why syntax should be relevant at all.
And indeed, it seems as though on some level (58c) must be rejected out of PF
considerations. Some support for this conclusion can be found in the interesting
fact that accidental properties of the particular lexical item can sometimes rescue
splittings. Specifically, when such splittings involve a preposition that has an
intransitive (i.e. adverbial) variant, then the splitting is not in fact filtered out by
152 Steven Franks
PF. Two such prepositions in BCS are genitive governing ispred in front of
and pored alongside:
(60) a. Ispred ga je ulaza doekala policija.
in-front-of him.ACC aux.3SG entrance.GEN waited police
The police were waiting for him in front of the entrance.
b. Pored je tog ovjeka sjela.
alongside aux.3SG that. GEN man.GEN sat
She sat alongside that man.
(61) a. On je sjedio ispred/pored.
he aux.3SG sat in-front/alongside
He was sitting in front/alongside.
b. Ispred/Pored je sjedio.
(He) was sitting in front/alongside.
However, prepositions such as prema towards which do not admit intransitive
usage, invariably block this kind of splitting. Compare (62b) with (60b):
(62) a. *Iao je prema.
went aux.3SG towards
Intended: He was going towards.
b. *Prema je tom ovjeku doao.
towards aux.3SG that.DAT man.DAT came
Intended: He was coming towards that person.
[cf. Prema tom ovjeku je doao. Prema tom je ovjeku doao.]
Both scattered deletions in (63) are valid operations, for example if pored or
prema were contrastively focused, but the sequence prema je in (63b) is
subsequently filtered out as deviant whereas pored je in (63a) is not:
(63) a. [Pored ] je [ tog ovjeka] ...
b. *[Prema ] je [ tom ovjeku] ...
The reason (63a) survives is because, from the perspective of the lexical item
pored, the adverbial and prepositional variants are non-distinct, hence pored can
function as a genitive assigning preposition for morphosyntactic case checking
purposes but as an intransitive adverbial for PF purposes. The kind of splitting
in (58c) is thus expected to elude the PF filter with a preposition like ispred and
indeed, according to A. PETI-STANTI (p. c.), (64) is much better than (58c):
(64) Policija doekala ispred ga je ulaza
police waited in-front-of him.ACC aux.3SG entrance.GEN
(... and not alongside)
(... a ne pored).
The police were waiting for him in front of the entrance (... and not
alongside).
In sum, although prepositions need some piece of their complement to follow

them, the existence of an intransitive variant somehow allows such prepositions
to dodge this requirement. It is as if PF cannot tell whether any particular
instance of pored alongside has a case assigning feature or not, nor can it
remember that, before ellipsis in (63a), the case governing variant had been
invoked to make tog ovjeka that man genitive.
6. Superficial lexical choice
In the phenomenon just discussed, syncretism is crucial. An informal survey of
BCS prepositions shows that they fall into two classes along precisely these
lines: splittability is enabled by potential intransitivity. This correlation supports
the superficiality of this kind of ellipsis, but raises the far from trivial question
of how to exploit the correlation. The general structure of the solution however
is clear: at one point in the derivation a lexical item satisfies some requirement
and then, at a later point, when a contradictory requirement is imposed on it, that
item can also satisfy the second requirement. This is I believe a consequence of
non-distinctness: lexical insertion requires that the form selected from the
lexicon for SpellOut of some morphosyntactic feature set be featurally non-
distinct and lexical syncretism is expressed in the lexicon in terms of non-
distinctness (i.e. underspecification) of features.
6.1 More accidental syncretism effects
Language is replete with examples of syncretic morphological forms which
manage to satisfy the contradictory requirements of two different structures.
This section presents three relevant constructions, albeit quite different from the
PP-split phenomenon just discussed.
In the across-the-board (ATB) whmovement construction in Russian (65),
the feminine relative pronoun kotoroj which is syncretic instrumental and
dative, whereas the masculine forms kotorym/kotoromu are distinct:
(65) devuka, kotoroj [ja byl uvleen tINST i
girl (with/to) which.INST/DAT I was carried-away and
asto daval tDAT dengi]
often gave money
the girl who I was carried away with and often gave money to ...
[cf. malik, *kotorym/*kotoromu ... (the) boy *who.INST/*who.DAT ...]
In FRANKS (1995), I assimilated examples like (65) to the parasitic gap
construction, which displays the same syncretism effect, and argued that a null
operator was involved. The features of the null operator were determined by
those of overt kotoroj, which for instrumental and dative are non-distinct.
In German free relatives, as in (66), inanimate was is syncretic nominative
and accusative, whereas the animate forms wer and wen are distinct:
154 Steven Franks
(66) Ich werde was da steht mitbringen.

I will what.NOM/ACC there stands with-bring
I will bring along what is standing there.
[cf. Ich werde ??wer/*wen ... I will ??who.NOM/*who.ACC ...]
In Hungarian, which distinguishes verb conjugations for definite and indefinite
direct objects, coordination of objects of different types creates a problem for
lexical insertion, but syncretism evades it. As shown in (67), the past tense first
singular lttam saw is syncretic definite and indefinite conjugation, whereas
the present tense first singular forms ltom/ltok are morphologically distinct:
(67) Lttam a kt szp kutyt s egy csunya cict.
saw.1SG-DEF/INDEF the twobeautiful dogs and one ugly cat
I saw the two beautiful dogs and one ugly cat.
[cf. *Ltom /*Ltok ... (I) *see.1SG-DEF/*see.1SG-INDEF ...]
In each of these, the various structures are created in the syntax and the
sycretism of particular lexical items allows those forms to be evaluated in two
different ways from two different perspectives. If the particular morphological
form does not happen to be syncretic, then there is no appropriate lexical item
available and the conflict results in ineffability. But if there is a form that can do
double duty, SpellOut remains oblivious to the syntactic feature conflicts.
6.2 Matching and clitic drop
GRAANIN-YKSEK (2009) observes that accusative pronominal clitics can be
dropped in to-relative clauses in BCS just in case the form of the nominal head
of the relative clause is appropriate to the gap in the relative clause. This means
that a nominative noun can license an accusative gap so long as its form is
syncretic with the accusative. Some of her examples are given in (68)(70):
(68) a. ovjek [to sam ga/* vidio] opljakao je banku.
man.NOM that aux.1SG him.ACC saw robbed aux.3SG bank
The man that I saw has robbed a bank.
b. Upoznao sam ovjeka [to ga/ je Jan doveo
met aux.1SG man.ACC that him.ACC aux.3SG Jan brought
na zabavu].
on party
I met the man that Jan brought to the party.
c. Dijete [to sam ga/ vidio] ima dugu kosu.
child.NEUT.NOM/(ACC) that aux.1SG him.ACC seen has long hair
The child that I saw has long hair.
(69) a. elja [to sam je/* osjetio] bila je jaka.
desire.FEM.NOM thataux.1SG her. ACC felt been aux.3SG strong
The desire that I felt was strong.
b. Ljubav [to sam je/ osjetio] bila je jaka.
love.FEM.NOM/(ACC) thataux.1SG her.ACC felt been aux.3SG strong

The love that I felt was strong.
What is going on here? While it is not particularly unexpected that matrix
accusative ovjeka should be able license an accusative gap in (68b), the fact
that matrix nominative dijete in (68c) also can is surprising. The reason is
because nominative dijete is featurally non-distinct from the accusative, so
somehow it is able to function as accusative in the embedded clausedespite
the fact that it is actually nominative. The constrast between distinctly
nominative elja and syncretic ljubav in (69) is similar.
In FRANKS (1995), I suggested a movement account when there is a gap and
base generation when there is a clitic (i.e. regarding the clitic as a resumptive
pronoun). Although Graanin-Yksek makes the same set of assumptions, I
suggest here that the clitic can also arise through movement and that only for
some speakers does it have a resumptive quality. Hence Graanin-Ykseks
(70), which involves an adjunct island, is ungrammatical with a gap and
acceptable only for some speakers with a clitic:
(70) Vidio sam ovjeka [to Sanja plae [jer %ga/* mrzi]].
seen aux.1SG man.ACC that Sanja.NOM cries because him.ACC hates
I saw the man that Sanja cries because she hates (him).
It is clear that the origin of the clitic relates to the fact that the relativizer to
does not mark case. Regardless of how this connection is implemented, the point
remains that the same phenomenon is taking place here as with ATB and
parasitic gap constructions: a null operator moves to SpecCP in the relative
clause and its morphological case features are matched against those of its
antecedent. This is illustrated in (71) for (68c):
(71) Dijete [ [to sam t vidio] ima dugu kosu.
child.NOM-ACC ACC that aux.1SG seen has long hair
The child that I saw has long hair.
If the case features of the null operator are compatible with that of the nominal
head of the relative clause, here dijete, it is licit. If not, the case marked relative
pronoun koji must be used.
6.3 Homophony and allomorph selection in Croatian
I now turn to a demonstration of the superficiality of the factors impacting on
allomorph selection. A famous alternation in the BCS clitic system is between
the standard 3sg feminine accusative pronoun je and its variant ju. Here I rely
partly on data presented in WITCOMBE (2008), and note also that my informants
for these data were Croatian speakers.
The most familiar situation in which ju supplants accusative je is when it
precedes the 3sg singular auxiliary, also je, as in (72).
156 Steven Franks
(72) Konobar mi ju/*je je dao.

waiter me.DAT her.ACC aux.3SG gave
The waiter gave it to me.
This is clearly an instance of constraint against homophonous sequences in (7),
although it pertains to allomorph selection rather than copy deletion. Much less
well known is the fact that ju is selected over je when it precedes a word
begining in je-, as in (73), or when it follows a word ending in -je, as in (74).
(73) a. Ve ju/*je jedem. b. Dao sam ju/*je Jeleni.
already her.ACC eat.1SG gave aux.1SG her.ACC Jelena.DAT
I am already eating it. I gave it to Jelena.
(74) Pije ju/*je
drinks it.ACC
S/he is drinking it.
Unlike accusative je, 3sg singular auxiliary je has no special allomorph to
exploit. Instead, it deletes after another clitic ending in -e, obligatorily after
reflexive se but only optionally after 1st person me or 2nd person te:
(75) a. Predstavila mu se (*je).
introduced him.DAT REFL aux.3SG
She introduced herself to him.
b. Vidio me (je).
saw me.ACC aux.3SG
He saw me.
c. Pozvao te (je).
called you.ACC aux.3SG
He invited you.
There is variation among speakers, dialects, and styles;25 such vicissitudes of
allomorph selection show it to be a relatively superficial process, an interplay
among competing desiderata. WITCOMBE (2008) in fact develops an OT account
of these sorts of phenomena.
7. More speculations about SpellOut
The facts of SpellOut discussed in this paper indicate to my mind that we need
a model in which (i) contradictory factors compete (OT-like competition in
SpellOut), but structures are built derivationally (cyclicity of SpellOut), and
(ii) (at least some) decisions are local (uniformity of SpellOut paths). These
tentative conclusions are listed in (75):
(75) a. Contradictory factors compete. (OT-like competition in SpellOut)
b. Structures are built derivationally. (cyclicity of SpellOut)
__________
25 Note also that Slvn, unlike BCS, does not drop je.
c. (At least some) decisions are local. (uniformity of SpellOut paths).

In this final substantive section I elaborate on these conclusions and examine
associated conceptual considerations. I speculate on alternative approaches to
cyclicity and explore the role of prosodification in SpellOut.
7.1 SpellOut as PF optimalization: uniform paths or prosodic domains?
Recall, first, that we encountered problems in deciding in which order to apply
SpellOut processes such as selection of lexical item, its linearization, decisions
about copy deletion and prosodification, and so forth. My solution was to see
these factors as competing desiderata, along OT-like lines. However, to the
extent that SpellOut is iterative, these desiderata have the effect of processes
rather than constraints. And while the standard embodiment of iterativity is
cyclicity, I have adduced evidence that the path to PF is not necessarily
punctuated along traditional cyclic nodes lines. Rather, linearization needs to be
able to take place repeatedly within the same phase, for example, each time a
clitic adjoins to its host. BOECKX (2008) refers to the standard model, in which
cyclic phasal domains are sent to PF, as punctuated SpellOut paths, and to
alternatives in which every phrase is a SpellOut domain as involving a
(quasi)-uniform SpellOut path. BOECKX (2008: 52) concludes that
Chomskys characterization of C and v as phases does not receive independent
support from interface diagnostics, and boils down to a stipulation. While I
agree with this general conclusion, it is not obvious to me that a uniform
approach is the best solution once traditional phases have been rejected.
Spelling out every phrase does provide a way of making sense of (75). One
interesting consequence is that cyclic consideration of constraints, as in such
derivational OT phonology work as RUBACH (2004) or KIPARSKY (2000),
becomes indistinguishable from pure generative approaches as the cycles
become closer and closer in size. That is, the smaller SpellOut domains are, the
tighter they are nested, hence the less distinguishable contraints become from
processes. Optimally theoretic and derivational generative approaches might
therefore be reconciled by sending every phrase produced by the syntax to
SpellOut. OT-like constraints would be iteratively imposed, optimizing PF
desiderata as each phrase is constructed. In such a model, as the paths approach
uniformity, ordering puzzles disappear and distinctions between OT and
derivational conceptions trivialize.
There is however a provocative alternative that produces cyclic effects but
does not rely on syntactic units per se. A curious fact about the effects discussed
in this paper is that they pertain to prosodic domains of different sizes. If
prosodic rather than syntactic structuring is what is actually relevant to Spell
Out domains, then perhaps all ostensible cyclic effects could be recast in these
158 Steven Franks
terms.26 This sort of approach makes sense to my mind, since SpellOut is really
a PF rather than syntactic matter.
Note that under Boeckxs version the relevant unit must be a phrase, which
is what is meant by quasi-uniform; a perfectly uniform SpellOut path would
apply after every application of merge. But clearly li must wait until CP is
completed, since it needs to violate the LCA only once it is has been determined
that there is no prosodic word to host it in SpecCP. Crucially, SpellOut could
not be allowed to apply at the bar-level, otherwise examples such as (49) would
not be derivable. Similarly under the prosodic domain story, the fact that enclitic
li is not prosodically supported can only be determined after no viable host
merges in SpecCP. Not that, even though li is a syntactic word, as a clitic
Selkirks (44c) fails to apply to it, so that it is only prosodified at the next
prosodic level up, by which time it will already be clear whether or not there is a
host to its left. Either way, it is worth noting that iterative OT addresses one of
the common objections to standard OT: the problem of ineffability. The problem
is this: If there is always a highest ranked candidateno matter how lowthen
how can a derivation ever crash? But in an iterative system, even if there always
is an optimal candidate, that candidate may end up being useless as input to the
next derivational step and the derivation stops. Similarly, on the PF and LF
sides, if the winning SpellOut candidate is something that cannot be seen, i.e.,
that is illegible to PF or LF, then this too results in ineffability.
7.2 Tobler-Mussafia linearization revisited
This section offers a quick demonstration of what this might mean for TM
linearization. Recall the account presented in section 4.2, with the derivation in
(42), repeated in (77), of the Bg sentence in (76):
(76) Pokazvala li si mu gi?
shown.FEM Q aux.2SG him.DAT them.ACC
Have you shown them to him?
(77) a. [[si mu gi]CG [pokazvala] ] (merger of li)
b. [[[si mu gi]CG [pokazvala] ] li] (SpellOut as Utterance)
c. #[[[si mu gi]CG [pokazvala] ] li ]# (TM)
d. #[[[pokazvala] li ] [si mu gi]CG ]#
I had to restructure the clitics to form a clitic group consisting of the cluster si
mu gi and then relinearize them, in order to go from (77c) to (77d) in accordance
with NONINITIAL(UTTERANCE). The reason for the restructuring was that, under
Bokovis system of successive clitic adjunction, the most straightforward
internal order before merger of li would be as shown in (78).
(78) [si [mu [gi [pokazvala]]]]
__________
26 I explore such an account in FRANKS (in progress).
Now, what if the clitics were not subjected to restructuring, but rather the
mapping to SpellOut was optimized in an iterative fashion for each element
successively? The result, interestingly enough, is the same, as shown in (79):
(79) a. [si [mu [gi [pokazvala]]]]
b. [[si [mu [gi [pokazvala]]]] li]
c. [[[mu [gi [pokazvala]]] li] si]
d. [[[[gi [pokazvala]] li] si] mu]
e. [[[[[pokazvala] li] si] mu] gi]
(79a) is just (78), reflecting the successive LCA-compatible linearization of each
clitic adjunction. In (79b) li has been added and linearized to the right of the
prosodic word si mu gi pokazvala, in comformity with its enclitic status, which
outranks the LCA. This is the final linearization in Mac. In Bg, on the other
hand, si now violates NONINITIAL(UTTERANCE), so it is linearized at the right
edge of mu gi pokazvala li, as in (79c). At this point, however, mu violates
NONINITIAL(UTTERANCE), so it is linearized at the right edge of gi pokazvala li
si, as in (79d). But now gi violates NONINITIAL(UTTERANCE), so it is linearized
at the right edge of pokazvala li si mu. This produces the final output, with the
correct Bg order of pokazvala li si mu gi, in (79e).
7.3 Clitic constraints revisited
In the system just described, when each prosodic unit (or phrase, if the
appropriate domain turns out to be syntactic) is sent to SpellOut, it continually
runs the gauntlet of constraints until it is truly optimal. This requires some
reevaluation of the constraints introduced earlier. In particular, the question of
how the BCS special clitics differ from li needs to be examined more carefully.
Whereas the traditional approach to 2P clitics is that they are positioned second
in their I-phrase, as reflected in my use of NONINITIAL(I-PHRASE) in (4), this
cannot actually be correct.27 The problem is that the BCS version in (80b) of Bg
(25a), repeated as (80a), is not possible, despite the fact that si mu ih is indeed
non-initial in its I-phrase:
(80) a. I si mu gi pokazvala.
And aux.2SG him.DAT them.ACC showed.
And you showed them to him.
b. *I si mu ih pokazala. [correct BCS order: I pokazala si mu ih.]
I therefore conclude that the operative constraint here actually pertains to the
prosodic word, just like with li (in all Slavic languages in which it is found).
(81) NONINITIAL(PROSODIC WORD)
__________
27 At the very least, additional clitic parameters would need to be assumed, such as the
traditional three parameter system of KLAVANS (1985), updated in ANDERSON (1995) and
also adopted by BOKOVI (2001a: 82).
160 Steven Franks
The prosodic difference between the special clitics in BCS and Bg is thus that
the domain of non-initiality has expanded to the Utterance in the latter
language;28 in Mac (and Slvn) it is disregarded altogether.29 However, if (81)
pertains both to the special clitics (in BCS) and to li, then an alternative account
of any differences between these two kinds of items is necessitated. I believe
that the answer is straightforward: the special clitics, being paradigmatic and
introduced lower in the tree, move, whereas li does not. In short, there are no
lower copies of li that can be pronounced when (81) is in danger of being
violated.30 This implies that PRONOUNCE HIGHEST is ranked below
NONINITIAL(PROSODIC WORD), so that lower copies are accessed if possible.
This explains the absence of prosodic inversion in BCS.31
The I-phrase nontheless still plays an important role in prosodification in
these languages. The reason is that clitics remain unprosodified until they are
contained within a prosodic domain sent to SpellOut. Hence, it is only when an
I-phrase containing a clitic is prosodified that the clitic is evaluated with respect
to NONINITIAL(PROSODIC WORD). It is at that point that the special clitics (in
BCS) and li (in all the languages) need to be adjoined to an adjacent prosodic
word. Here I am assuming the Strict Layer Hypothesis of SELKIRK (1984) and
NESPOR & VOGEL (1986), which states that speech is organized into a set of
prosodic domains that form a hierarchy of containment, with each non-terminal
constituent made up of a sequence of constituents at the next level down (ITO &
MESTER, in press). Hence, parsing of I-phrases into prosodic words must be
exhaustive.32
__________
28 It must be the Utterance rather than I-phrase because of examples such as Bg (25b).
29 Of course, the other important difference has to do with clitic placement. The simplest
account would be that in 2P languages the verb is pronounced low whereas in the verb-
adjacent clitic languages it is pronounced high, hence adjacent to the clitics.
30 A small caveat: it is likely that, in some languages, li can be introduced as the head of a
FocusPhrase (Lambovas P) just below CP and raise to C. See FRANKS 2006b) for
consideration of this possibility.
31 The issue remains of whytaking the arguments in BOKOVI (2001b) to be
compellingBCS li is never reprosodified at the right edge of the prosodic word to is
right in order to respect NONINITIAL(PROSODIC WORD). My guess is that the situation
simply never arises because li is merged as a Focus head in this language (see fn. 30);
hence there is always prospective host material to its left, in SpecFocusP.
32 Unfortunately, BCS (80b) remains problematic. Although I would prefer to avoid a
directionality of prosodic adjunction parameter (cf. fn. 27), it may well be that a lexical
enclitic/proclitic distinction is unescapable. If BCS clitics require support to their left,
then the sequence i si mu ih will straightforwardly violate the hard constraint PROSODIC
SUPPORT.
8. Conclusion
By way of summary, I began by arguing for a syntax which eschews phase-
induced successive cyclic movement and considered when hypothetical
intermediate whphrase sites are pronounceable or interpretable. I advocated
TAKAHASHIs (1994) proposal that an element does not move until its final
landing site has been introduced into the tree, and attraction takes place
(BOECKX 2008: 68), so that chains are formed only subsequent to movement. I
then identified a variety of likely PF effects on the output of syntax and
considered how factors such as linearization, lexicalization, prosodification, and
deletion might interact. I argued that these are probably best understood as
competing constraints, interacting to select an optimal realization. I then argued
for a highly derivational model of SpellOut, applying from the bottom up to
larger and larger units. However, I suggested that these units may not be
traditional phases, but rather increasingly exhaustive prosodic domains. I also
suggested that every SpellOut domain is optimized and that this optimization
itself iterates until it cannot be further improved upon. Most of the data were
meant to show how small scale tinkering could happen on a local level. I also
discussed situations in which one piece of a syntactic structure needed to made
reference to another piece, and showed that it was oblivious to the internal
syntax of that other piece, relying instead on the actual choice of lexical item.
Returning to my point of departure the division of labor between syntax
and PF it seems that the burden has shifted decidedly over to PF. This PF rich
reconception of the grammar, although it makes syntax simpler, is going to
make the PFside a lot more complex. Solving the puzzles of PF-driven Spell
Out presents a challenge I plan to address in future work, and urge readers to do
the same.
References
ANDERSON, S. R. (2005) Aspects of the Theory of Clitics. Oxford: Oxford University Press.
BILLINGS, L. AND C. RUDIN (1996) Optimality and Superiority: A new approach to multiple
wh ordering. In: TOMAN, J. (ed.) (1994) Formal Approaches to Slavic Linguistics: the
College Park Meeting 3560. Ann Arbor: Michigan Slavic Publications.
BOBALJIK, J. (2002) A-chains A-chains at the PF-interface: Copies and covert movement.
Natural Language and Linguistic Theory 20: 197267.
BOECKX, C. (2008) Understanding Minimalist Syntax: lessons from Locality in Long-
Distance Dependencies. Malden, MA: Blackwell.
BOKOVI, . (2001a) On the Nature of the SyntaxPhonology Interface. Amsterdam:
Elsevier.
BOKOVI, . (2001b) Li without PF movement. In: FRANKS, S., T. H. KING & M. YADROFF
(eds.) (2000) Formal Approaches to Slavic Linguistics: the Bloomington Meeting 57
76. Ann Arbor, MI: Michigan Slavic Materials.
BOKOVI, . (2002a) Clitics as nonbranching elements and the linear correspondence axiom.
Linguistic Inquiry 33.2: 32940.
BOKOVI, . (2002b) On multiple Wh-fronting. Linguistic Inquiry 33: 351383.
162 Steven Franks
BOKOVI, . (2004) Clitic placement in South Slavic. Journal of Slavic Linguistics 12.12:
3790.
BOKOVI, . (2005) On the locality of left branch extraction and the structure of NP. Studia
Linguistica 59: 145.
BOKOVI, . (2007) On the locality and motivation of move and agree: An even more
minimal theory. Linguistic Inquiry 38.4: 589644.
CHOMSKY, N. (1995) The Minimalist Program. Cambridge, MA: The MIT Press.
CHOMSKY, N. (2001) Derivation by phase. In: Ken Hale: a Life in Language. Cambridge,
MA: The MIT Press.
FOX, D. & D. PESETSKY (2005) Cyclic linearization of syntactic structure. Theoretical
Linguistics 31.12: 146.
FRANKS, S. (1995) Parameters of Slavic Morphosyntax. Oxford: Oxford University Press.
FRANKS, S. (1998/2010) Clitics in Slavic. Presented at the Comparative Slavic Morphosyntax
Workshop, Spencer, Indiana and published in Glossos 10. [Available on-line at
<http://www.seelrc.org/glossos/issues/10/>]
FRANKS, S. (2000) Clitics at the interface. In: BEUKEMA, F. & M. DEN DIKKEN (eds.) Clitic
Phenomena in European Languages 146. Amsterdam: John Benjamins.
FRANKS, S. (2005) Adverb interpolation in the Bulgarian clitic cluster. In: ROTHSTEIN, R., E.
SCATTON & CHARLES TOWNSEND (eds.) A Festschrift for Charles Gribble.
Bloomington, IN: Slavica.
FRANKS, S. (2006a) Agnostic Movement. In: DAVIS, C., A. R. DEAL & Y. ZABBAI (eds.),
Proceedings of NELS XXXVI vol 1, 267278.
FRANKS, S. (2006b) Another look at li placement in Bulgarian. The Linguistic Review 23:
161211.
FRANKS, S. (2010) On the mechanics of SpellOut. In: ERTESCHIK-SHIR, N. & L. ROCHMAN
(eds.) The Sound Patterns of Syntax 110139. Oxford: Oxford University Press.
FRANKS, S. (in progress) How multiattachment chains resolve copy puzzles. Unpublished ms.,
Indiana University.
FRANKS, S. AND . BOKOVI (2001) An argument for Multiple SpellOut. Linguistic Inquiry
32: 174183.
FRANKS, S. AND J. HERRING (2010). Against copies: Steps toward a call-by-reference syntax.
Paper presented at Syntaxfest 2010, June 2010.
FRANKS, S. AND T. H. KING (2000). A Handbook of Slavic Clitics. Oxford: Oxford University
Press.
GOLDEN, M. & M. MILOJEVI SHEPPARD (2000) Slovene pronominal clitics. In: BEUKEMA, F.
& MARCEL DEN DIKKEN (eds.) Clitic Phenomena in European Languages 191207.
Amsterdam: John Benjamins.
GROAT, E. & J. ONEIL (1996) Spellout at the LF interface. In:, Werner ABRAHAM, S., D.
EPSTEIN, H. THRINSSON AND J.-W. ZWART (eds.) Minimal Ideas 113139.
Amsterdam: John Benjamins.
KAYNE, R.(1994) The Antisymmetry of Syntax. Cambridge, MA: The MIT Press
KIPARSKY, P. (2000) Opacity and cyclicity. The Linguistic Review 17: 351367.
KLAVANS, J. (1985) The independence of syntax and phonology in cliticization. Language 61:
95120.
LAMBOVA, M. (2004) On Information Structure and Clausal Architecture: evidence from
Bulgarian. Doctoral dissertation, University of Connecticut.
LASNIK, H. (1999) On feature strength: Three minimalist approaches to overt movement.
Linguistic Inquiry 30: 197217.
LPEZ, L. (2009) Ranking the linear correspondence axiom. Linguistic Inquiry 40: 239276.
MERCHANT, J. (2001) The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis.
Oxford: Oxford University Press.
MORO, A. (2000) Dynamic Antisymmetry. Linguistic Inquiry Monograph 38. Cambridge, MA:
The MIT Press.
NUNES, J. (2004) Linearization of Chains and Sideward Movement. Cambridge, MA: The
MIT Press.
NESPOR, M. & I. VOGEL (1986) Prosodic Phonology. Dordrecht: Foris.
PETI-STANTI, A. (2007) Wackernagelovo pravilo norma ili mogunost. In: KUNA, B. (ed.)
Sintaktike kategorije 173187. Zagreb: Institut za hrvatski jezik i jezikoslovlje.
PROGOVAC, L. (2000) Where do clitics cluster? In: BEUKEMA, F. & M. DEN DIKKEN (eds.)
Clitic Phenomena in European Languages 249258. Amsterdam: John Benjamins.
RICHARDS, N. (1997) What Moves Where When in Which Language? Doctoral dissertation,
MIT.
ROBERTS, I. (1998) Have/Be raising, Move F, and procrastinate. Linguistic Inquiry 29: 113
125.
RUBACH, J. (2004) Derivation in Optimality Theory: a reply to Burzio Linguistic Inquiry 35:
656670.
RUDNICKAYA, E. (2000) The derivation of YesNo Li questions in Russian: Syntax and/or
Phonology? In: KING, T. H. & I. SEKERINA (eds.) (1999) Formal Approaches to Slavic
Linguistics: the Philadelphia Meeting 347362. Ann Arbor, MI: Michigan Slavic
Materials.
SELKIRK, E. (1984) Phonology and Syntax: the Relation between Sound and Structure.
Cambridge, MA: MIT Press.
SELKIRK, E. (to appear) The syntax-phonology interface. In: GOLDSMITH, J., J. RIGGLE &
ALAN YU (eds.) The Handbook of Phonological Theory, 2nd edition.
STJEPANOVI, S. (1998a) On the placement of Serbo-Croatian clitics: Evidence from clitic
climbing and VP ellipsis. In: BOKOVI, . ET AL (eds.) Formal Approaches to Slavic
Linguistics: the Connecticut Meeting. Ann Arbor, MI.: Michigan Slavic Materials.
STJEPANOVI, S. (1998b) A note on placement of Serbo-Croatian clitics. Linguistic Inquiry
29: 527537.
STJEPANOVI, S. (2007) Free word order and copy theory of movement. In: CORVER, N. & J.
NUNES (eds.) The Copy Theory of Movement 219248. Amsterdam: John Benjamins.
TAKAHASHI, D. (1994) Minimality of Movement. Doctoral dissertation, University of
Connecticut.
TORREGO, E. (1984) On inversion in Spanish and some of its effects. Linguistic Inquiry
15:10329.
URIAGEREKA, J. (1999) Multiple SpellOut In: EPSTEIN, S. & N. HORNSTEIN Working
Minimalism 251282. Cambridge, MA: MIT Press.
WACKERNAGEL, J. (1892) ber ein Gesetz der indogermanischen Wortstellung.
Indogermanische Forschungen 1: 333.436.
ZWART, J.-W. (1997) Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax
of Dutch. Kluwer Academic Publishers, Dortrecht.
ZWICKY, A. (1977) On clitics. Bloomington: Indiana University Linguistics Club
Steven Franks, Indiana University, Department of Slavic Languages and

Literatures, 1020 E. Kirkwood Avenue, Ballantine Hall 502, Bloomington, IN
47405-7103, USA, franks@indiana.edu
Wh-Words and the Indefinite Particle -to in Russian
Elena Gorishneva & Ilse Zimmermann
1. Introduction
Within a minimalist framework of sound-meaning correlation with a central role
of the lexicon, the morphosyntax, semantics and pragmatics of Russian wh-
words (= k-words) and the indefinite particle -to will be considered.
The contribution is based on ZIMMERMANNs (2000, 2008) assumption that wh-
words basically constitute predicate expressions (NPs or APs), which can be
embedded in DPs and PPs with zero-heads and corresponding meaning.
Our analysis argues for a unified combination of indefinite particles like -to
with wh-words regardless of their function.
1.1 The issue
The referential status of indefinite pronominal series in Russian has been
extensively discussed in the literature (cf. GEIST 2008, KAGAN 2007, TATEVOSOV
2002, ZIMMERMANN 2000, among others). These series consist morphologically
of wh-words which are combined with the particles koe-, -to and -nibud.
Whereas the koe-items can only denote specific known referents, the wh-words
with -nibud indicate the non-specificity of the referent. The semantics of the -
to-words is more complicated,which makes them a challenging topic of
linguistic investigation.
Wh-words with the attached particle -to indicate that the pertinent referent is
specific but can not be identified by the speaker. However, it has been observed
that in their peripheral use -to-words can appear in contexts which are primarily
preserved for wh-words with koe- and -nibud. As shown in the examples below,
-to can replace koe- as a marker of the specific known referent as well as occur
in non-specific contexts instead of -nibud.
(1) Ja znaju, koe-kto / kto-to zdes xoet neprijatnostej, i sejas tot koe-kto /
kto-to ix poluit. (TATEVOSOV 2002: 141)
I know that someone here is looking for trouble and now he is going to
find it.
(2) Esli kto-to / kto-nibud moet to sdelat, pust delaet.
If someone is able to do it, he must do it.
Nevertheless, it should be said that the occurrence of wh-words with -to in such
contexts is not typical and is more an alternative but not preferred option that is
rather characteristic for colloquial speech.
In the present paper we will disregard the marginal use of -to-items and
focus on their main functions. We will consider the various -to-items like in (3)-
166 Elena Gorishneva & Ilse Zimmermann
(6) with pronominal, adverbial and adjectival wh-words as the host of the
attached indefinite particle:
(3) Priexal [ kto-to iz Moskvy].
arrived someone-NOM from Moscow
Someone from Moscow arrived.
(4) Nine xoetsja [ego-to kislen'kogo].
Nina-DAT wants-REFL something-GEN sour-GEN
Nina wants something sour.
(5) Orkestr vystupil [gde-to v Moskve].
orchestra-NOM played somewhere in Moscow
The orchestra played somewhere in Moscow.
(6) V koridore stojal [ej-to emodan].
in corridor stood-3SG someones suitcase
There was someones suitcase in the corridor.
Furthermore, we take into consideration different positions of the corresponding
NP with particle -to. As evident from the examples below, wh-words with -to
can appear in the argumental or predicative positions, which yields different
interpretations. Whereas in (7) the non-identifiable referent is an individual, (8)
allows for only kind reading, that is, we deal with the non-identifiability of a
property by the speaker.
(7) Priexal [kakoj-to znamenityj detskij vra].
came-3SG some famous nursery doctor
Some famous paediatrician arrived.
(8) Petr stal [kakim-to znamenitym vraom ].
Peter became-3SG some famous doctor
Peter became some famous doctor.
(9) Petr xoet stat [kem-to znamenitym].
Peter wants become-INF someone famous
Peter wants to become someone famous.
1.2 The occurrence of -to in different contexts
Before we turn to the analysis of wh-words with -to, we should consider in
which environments they can appear and in which cases they should be
substituted with -nibud -items.
As we can see from the examples in the previous section, -to-series can be
felicitously used in assertions, but their occurrence in interrogatives is rather
restricted, cf. (10), (11):
Wh-Words and the Indefinite Particle -to in Russian 167
(10) Dal li on *komu-to /komu-nibud *to-to/to-nibud?

gave-3SG QUEST he someone-DAT something-ACC
Did he give something to someone?
(11) Kto da *komu-to /komu-nibud *to-to / to-nibud?
who gave-3SG someone-DAT something-ACC
Who gave something to someone?
The use of -to-items is a marginally possible, if not preferred option, only in
questions without wh-words, mostly in colloquial speech, cf. (12), (13):
?
(12) Zvonil li kto-to / kto-nibud?
called-3SG QUEST someone
Did somebody call?
(13) Kto-to zvonil?
someone called-3SG
Did somebody call?
As shown in (13), the acceptance of wh-words with -to improves in
interrogatives with the omitted question particle li, which do not serve as
primary questions but rather as questions of confirmation in which the goal of
the speaker is to ascertain whether the information is true. For instance, Ona
xoet ego-to kislenkogo? Does she want something sour? is suitable in
contexts in which the speaker intends to receive confirmation of whether he
understood the information correctly.
As for imperatives and jussives, they prohibit the use of wh-words with the
particle -to. Thus, the -nibud-items are the only appropriate option, see (14) and
(15):
(14) Zvonite *komu-to / komu-nibud!
call-2PL someone
Call someone!
(15) Pust Boris stanet *kakim-to / kakim-nibud vraom!
JUSS Boris becomes some doctor
Boris should become some doctor!
This fact can be accounted for as follows: The speaker producing an imperative
intends to induce someone to perform a specific action which is clearly
identifiable by the speaker. For example, it would be contradictory if the speaker
asked someone to call a person whom he cannot identify. That is why only
definite noun phrases like proper names or free choice items like -nibud -words
are allowed in imperatives.
Considering debitatives like (16), the use of wh-words with the indefinite
particle -to signals the reported speech, in contrast to sentences with -nibud in
which the source of the command is the speaker him- or herself (16b):
(16) a. Ty dolen pozvonit komu-to. vs.b. Ty dolen pozvonit komu-nibud.

You must call someone
You should call somebody.
The wh-words with attached -to-particle can serve as markers of epistemic
specificity:
(17) V dome gorit svet, znait tam kto-to est.
in house glows light means there someone be
The light is on in the house, which means that someone should be there.
Wh-words with -to can not serve as free choice items, cf. (18), (19):
(18) On podxodit na tu rol lue em *kto-to/ kto-nibud/kto-libo
he suits for this role better than someone anyone
drugoj.
other
He is suited for this role better than anyone else.
(19) Lue vse delat srazu, em *kogda-to/kogda-nibud potom.
better all do-INF immediately than sometime later
It is better to do a job immediately rather than later.
In the protasis of conditionals, wh-words with -to can substitute -nibud-items
and indicate a non-specific referent, at least in colloquial speech. Nevertheless,
the occurrence of -to in such environments appears more complicated and needs
further investigation. Comparing (20) with (21), it seems that -to-words are
more appropriate in episodic contexts in which a referent under question is a
member of the contextually determined set of individuals who are identifiable
by the speaker, whereas the preferred option in generalizations are -nibud -
words.
(20) Obyno, esli (?)kto-to / kto-nibud opasdyvaet, ego dut.
usually if someone is late him-ACC wait-3PL
Usually, if someone is late, one waits for him.
(21) Esli kto-to (iz nas) opazdaet, my pododem.
if someone (from us) comes later we wait
If someone comes later, we will wait for him.
The future-oriented contexts and generalizing assertions posit restrictions on the
use of the wh-words with the indefinite particle -to due to the non-specificity
requirement which is not satisfied by the latter (22):
(22) Ona budet zabotitsja o *kom-to / kom-nibud.
She will take care about someone-PREP
She will take care of someone.
(23) Boris ne otkroet dver, poka *kto-to/kto-nibud ne postuit
Boris NEG opens door till someone NEG knocks
three times
tri raza.
Boris wont open the door till someone knocks three times.
In contrast to generalizing expressions, in episodic sentences with specific
referents only -to-items are felicitous:
(24) Ona spala v to vremja,kak kto-to/*kto-nibud postual v dver.
she slept-3SG in this time how someone knocked in door
She was sleeping while someone knocked at the door.
1.3 Summing up
To summarize, wh-words with the indefinite particle -to serve to denote a
specific non-identifiable referent that can be regarded as their primary function,
which can not be fulfilled by other pronominal series like wh-words with koe-
and -nibud. The use of koe- and -nibud-items is reserved for indicating the
specificity and identifiability of referents and the non-specificity and free choice
respectively. None of these series can replace -to-words to mark specific known
referents, whereas the opposite case is possible. As the most unmarked
pronominal series, wh-words with -to can substitute koe- and -nibud, which can
be treated as their peripherical functions.
In the present study we aim to ask the following questions:
- Why does the wh-word with -to always occur at the left periphery of its
embedding phrase?
- How are adjectival or prepositional modifiers combined with the wh-word with
attached -to?
- Where is the indefinite particle-to added to the wh-word?
- What is the semantic and/or pragmatic contribution of -to?
- How does -to contrast with the particles koe-and -nibud?
1.4 Hypotheses
In S-structure, wh-words with the indefinite particles -to, -nibud, koe- and
others occur as adjuncts of NP, in SpecDP or in SpecPP. Basically, the wh-
words are inserted as NP- or AP-heads. They are pronouns, not determiners. As
referring expressions, they are accompanied by a zero-determiner. The indefinite
particles are added to the wh-words in the lexicon.
Semantically, these particles bring in a predicate variable Q and certain
qualifications on alternatives of Q related to the context set of the respective
modal subject. Q and correspondingly its alternatives could be specified by
various types of predicates like identification with individuals, properties like
znamenityj famous or relative qualifications like detskij nursery. We offer a
unified treatment of wh-words with indefinite particles, regardless of their role

as arguments or as modifiers.
2. The analysis
The following analysis of Russian wh-words with the indefinite particle -to
concentrates on their word structure, syntax and semantics. In the sound-
meaning correlation of these pronouns, the lexicon plays a central role.
2.1 Lexical information
Every lexical entry contains the phonetic characterization, the morphosyntactic
categorization and the semantic form of the pertinent lexical item.
The indefinite particles -to, -nibud and koe- considered here are clitics
which combine with a +wh-host in word-structure. The host is fully inflected,
marked by the feature +max. The host occurs before the clitics -to and -nibud
and after the clitic koe-. Categorically, the clitics add to the host the features
def(inite) and +spec(ific) for -to and koe- and respectively spec for -nibud.
Thus, -to has the lexical information in (8). Its semantics will be added later.
(25) / to /, +wh+max___; def+spec; ...
The wh-words kto who, ej whose, to what, gde where, kakoj which
have the following lexical entries (cf. ZIMMERMANN 2000, 2008):
(26) / kto /; +NV+wh; x [ANIMATE x]
(27) / ej /; +N+V+wh; x [ ANIMATE x]
(28) / to /; +NV+wh; x [ ~ANIMATE x]
(29) / gde /;.+NV+adverbial+wh; x [ PLACE x]
(30) / kakoj /; +N+V+wh; (P) x [ P x]
Basically, these wh-pronouns are interpreted as one-place predicates. In
addition, kakoj can be an identity function. The semantic form of the indefinite
particles combines with these predicates. This will be shown below.
2.2 The syntactic derivation
First, the syntactic derivation of the phrases with wh-words of the examples (3)-
(9) will be considered. It is shown how the wh-words with an indefinite particle
move to the left position of the embedding phrases and how modifiers combine
with them.
(31) [DP[NP kto-to]i [D [NPti [PPiz Moskvy]]]]
(32) [DP[NPego-to]i [D [NP [APkislenkogo] ti]]]
(33) [PP[NPgde-to]i [P [DPti [D [NPti [PPv Moskve ]]]]]]
(34) [DP[APej-to]i [D[NPti [NPemodan]]]]
(35) [DP[AP kakoj-to]i [D [NP ti [NP[AP znamenityj ][NP[AP detskij ] [NPvra]]]]]]

(36) [NP[APkakoj-to]i [NP[APznamenityj][NPti [NPvra]]]]
(37) [NP[NPkem-to]i [NP[APznamenitym] ti]]
As a result of movement, the wh-word occurs on the left of all phrases, in
SpecDP, SpecPP or as adjunct of NP. ej-to and kakoj-to are modifiers, whereas
all other wh-words in the examples above are NP-heads. Prepositional modifiers
are right adjuncts of their head; adjectival modifiers occur on their left.
Arguments are analysed as DPs with a zero-determiner. We assume that the
empty determiner agrees in its features def Dspec with the wh-pronoun in
SpecDP. For adverbial wh-phrases like gde-to, there is a further zero-head,
which delivers the prepositional syntactic and semantic status of the phrase (see
ZIMMERMANN 2000). The respective rightmost occurrence of the trace ti in the
derivations indicates the base position of the wh-words and their corresponding
interpretative possibilities. (38)-(40) are the representations of the trace, of the
indefinite determiner and of the prepositional head.
(38) / ti /; +NDVdef; Q<e,t>
(39) / /; +Ddef Dspec; OP1OP2x [[ P1 x ] [ P2 x ]]
(40) / /; NV+adverbial; OzOy [ y R<e,t> z ]
2.3 The meaning of the indefinite particles
From the analysis given so far it is evident that we differentiate between
determiners and pronominal wh-words. Whereas determiners like (39) bring
about generalized quantifiers, wh-words are predicate expressions. Their
combination with an indefinite particle does not alter this status. The particle
characterizes specificity by adding a modifying predicate which ascribes to the
pertinent individual entity x an unspecified property out of a contextually given
set of alternatives. This set belongs to a certain modal subjects z mental model
M of some world. M can be epistemic, deontic, desiderative, etc. In nature; z can
be the speaker or some other rational being. Shortening, we will use ALTz (Q)
meaning an alternative of the property Q in zs mental model M. We propose
the following semantic form for koe-, -to and -nibud:
(41) || koe- || = OxQ [[ Q x ] Q[[ Q = ALTsp(Q) ] o [ a Q = Q ]]]
(42) || -to || =OxQ [[ Q x ] ~Q[[ Q = ALTsp (Q) ] o [ a Q= Q ]]]
(43) || -nibud|| = OxQ [[ Q x ] Q[[ Q = ALTz(Q) ] o [ Q = Q ]]]
It does not seem misleading to consider wh-words with -nibud as an unspecific
free choice predicate ascribed to an individual x and -to and koe- as
characterizing specificity, as do RUIKA 1973 and HASPELMATH 1993. We
assume that specific entities are presupposed to exist in the pertinent modal
subjects mental model.
Comparing the three particles, we agree with GEIST (2008) that -to is the
most unmarked one. In contrast to koe-, it does not signal identifiability of the
referent x by the speaker. In contrast to -nibud, -to does not have free choice
character. It is normally the case that the speaker uses -to for newly introduced
referents which he is going to describe more concretely in the following
discourse. Koe- and -to occur also in reported speech or thought. In all non-
negated non-veridical contexts, -nibud-pronouns are used. They do not
presuppose the existence of the pertinent referent. Moreover, any member of the
set of contextually determined alternatives can be chosen.
Four aspects of this semantic analysis are to be stressed:
Firstly, the proposal takes into consideration not only argumental phrases
with wh-words and indefinite particles, but also adverbials and predicative NPs.
Secondly, we treat these indefinite pronouns in a unified way, regardless of
their different functions.
Thirdly, we relate the semantics of the indefinite particles exclusively to
modal subjects and by this to mental models (cf. KAGAN 2007). It does not seem
convincing to use the free individual variable z also for co-variation in the scope
of a universal quantifier as do GEIST 2008 and GEIST/ONEA 2007 (see below).
Fourthly, in order to capture the contexts selecting -nibud-pronouns we
propose that the feature spec is chosen by veridical contexts. This will take
place in interrogative, imperative and optative sentences; in future-oriented and
generalizing assertions; in the scope of deontic, desiderative or dubitative
embedding functors; in the scope of the temporal conjunction poka till or in
irreal conditionals, after the negating preposition bez without and in
comparative phrases like em gde-nibud than anywhere. All these contexts
have to do with non-veridicality (see CHENG & GIANNAKIDOU 2006,
GIANNAKIDOU 2009).
2.4 Composed meaning representations
In the following section, some composed meaning representations will be
considered (cf. the derived syntactic structures (31)-(37)). Generally, we follow
the idea of phasewise sound-meaning correlation (cf. CHOMSKY 2000, 2001,
KRATZER & SELKIRK 2007). Furthermore, we assume that in the semantic
composition in addition to the meaning of the morphosyntactic constituents
semantic templates like (44)-(46) are at work.
(44) TSAS: Oz<e>Oy [ y R z] o O<<e,t>t> Oy [ Oz[ y R z]]
(45) TSMOD: OQ1<e,t>OQ2<e,t>Ox [[Q1 x ] [Q2 x ]]
(46) TSLA: OpOX [ p]
The type shift (44) accommodates the argument structure of functors to
generalized quantifiers as arguments. (45) enriches the argument structure of a
one-place predicate by a position for a modifier (cf. ZIMMERMANN 1992). (46) is
lambda abstraction. (47) and (48) illustrate the application of these templates to
the syntactic derivations (31) and (33), repeated here.
(31) [DP[NP kto-to ]i[D [NP ti [PP iz Moskvy ]]]]
(47) OP2x [[[[ ANIMATE x ] Q [[ Q x ] ~Q [[ Q = ALTsp(Q) ] o
[a Q = Q ]]]] [ x FROM MOSKOW]] P2 x ]]
= ((TSLA (|| -def ||(TSMOD (|| ti ||)(|| iz Moskvy ||))))(TSMOD (|| kto ||)
(|| -to ||)))
(33) [PP[NP gde-to ]i[P [DP ti[D [NP ti [PP v Moskve ]]]]]]
(48) Oyx [[[[ PLACE x ] Q [[ Q x ] ~Q [[ Q = ALTsp(Q) ] o
[a Q = Q ]]]] [ x IN MOSKOW ]] [ y R x ]]
= ((TSLA ((TSAS (|| -adverbial||))((TSLA (|| -def ||(TSMOD ||)(|| ti ||)
(|| v Moskve ||))) (|| ti ||)))) (TSMOD (|| gde ||)(|| -to ||)))
Because of several occurrences of traces, the respective semantic amalgamation
appears complicated. Could one possibly dispense with the movements? The
answer is negative. The position of the wh-word with the indefinite particle in
SpecXP guarantees that XP inherits the selectional feature Dspec. Now
we will turn to examples with the pronoun kakoj-to.
2.5 A special lesson with kakoj-to
What aim does the wh-word kakoj serve, in contrast to kto, to, ej and gde?
According to ZIMMERMANN (2008), it can function as a modifier asking for a
property, or it functions simply as a wh-maker of descriptive NPs and is an
identity function.
On the basis of the lexical entry (30) for kakoj (see section 2.1.), we get the
NPs (49) and (50) with the head noun kniga book.
(49) Ox [ BOOK x ]
= || kakoj<<e,t>,<e,t>> || (|| kniga ||)
(50) Ox [[ BOOK x ] [ P x ]]
= TSMOD (|| kniga ||) (|| kakoj<e,t>||)
In (49) x can be questioned and in (50) P can (for the semantics of questions see
ZIMMERMANN 2008, 2009).
In the argumental DP of example (7) with its derived syntactic structure
(35), repeated here, kakoj-to is the highest modifier, while in (8) with its
structure (36) it is the lowest modifier, in its respective base position. In these
positions different classes of modifiers can be realized. The semantic
representation of (35) is (51).
(35) [DP[AP kakoj-to ]i[D [NP ti [NP[AP znamenityj ][NP[AP detskij ][NP vra ]]]]]]
(51) OP2x [[[[[ DOCTOR x ] [ x FOR CHILDREN ]] [ FAMOUS x ]]

Q [[ Q x ] ~Q [[ Q = ALTsp (Q) ] o [ a Q = Q ]]]] [ P2 x ]]
= ((TSLA (|| || (TSMOD (TSMOD (TSMOD (|| vra ||)(|| detskij ||))
(|| znamenityj ||))(|| ti ||)))) (|| kakoj || ( ||to ||)))
Here, the predicate variable Q induced by the indefinite particle could be
specified as an identity function of x with a specific individual.
In examples like (8), to whose syntactic structure (36) we will turn now, the
wh-word with an indefinite particle refers to a specific kind. The composed
semantic form of this example is represented in (52).
(36) [NP[AP kakoj-to ]i [NP[AP znamenityj ][NP ti [NP vra ]]]]
(52) Ox [[[ DOCTOR x ] Q [[ Q x ] a Q [[ Q = ALTsp(Q) ] o [ a Q =
Q ]]]] [ FAMOUS x ]]
= ((TSLA (TSMOD (TSMOD (|| vra ||)( ||ti ||))(|| znamenityj ||)))
(|| kakoj<<e,t>,<e,t>> || (|| to ||)))
In both semantic representations, (51) and (52), the pronoun kakoj as identity
function applies to the indefinite particle -to.
While (35) is an argument expression, (36) functions as a predicative NP.
The same is possible with the NP in (9) from section 1.1. These examples
clearly demonstrate the non-determiner character of the pertinent wh-words with
indefinite particles.
2.6 Determiners and binding
With determiners one has to do in argument expressions as in (3) or (7) and in
prepositional phrases as in (5). Here, we reckon with determiners as generalized
quantifiers and, in addition, with wh-pronouns which are marked for specificity.
According to our analysis, the same would be true for English noun phrases like
[a certain athlete]i. Von HEUSINGER (2007) relates the index i by a contextually
determined function to a referential anchor, but without giving a compositional
semantic analysis for this phrase, its index and its constituents. We propose to
introduce the index with the determiner semantics and to assign to the adjective
certain the specificity characterization and to the noun the descriptive content of
the phrase.
In a sentence with an universal quantifier in subject position as in (53), von
HEUSINGER relates the index i to the index j of the quantified subject phrase to
represent the referential anchoring of the indefinite phrase, for its narrow scope
interpretation. The pragmatically induced correlating function could be spelled
out as his favourite athlete.
(53) [Every trainer ]j will train [a certain athlete]i
For wide scope interpretation, the index i is related to the speaker as the
referential anchor of the indefinite object phrase.
GEIST (2008) and GEIST & ONEA (2007) in parallel sentences instead of
indices use choice functions fx and let the individual variable x co-vary with the
variable of an universal quantifier in cases like (53) to represent quantificational
narrow scope of the indefinite noun phrase. For the wide scope interpretation, x
would be specified as speaker.
In the system of our proposal, there would be no co-variation of variables in
examples like (53). Only if the specificity marker certain is explicated by an
apposition like [his favourite athlete]j there will be co-variation of variables
with respect to the co-indexed phrases.
But this is a different phenomenon compared with the relation of certain to
the speaker as a modal subject. In (53) plus the apposition, every trainer, certain
and his are all in the scope of the assertion and thereby of the speaker. Wide
quantificational scope of the indefinite phrase we would represent by raising.
We restrict the individual variable z in the semantics of the specificity markers
to modal subjects.
PADUEVAs (1985: 220) example (54) illustrates the necessity to take into
consideration modal subjects and corresponding possibly different world models
and different scope properties.
(54) Mne xoetsja ego-to / ego-nibud kislenkogo.
Whereas ego-nibud relates to the mental subject ja I and the desiderative
mental state expressed by xotetsja, ego-to relates to the epistemic state of the
speaker expressed by the assertion.
3. Conclusion
Decisive for the proposed analysis is the assumption that indefinite particles like
-to can be treated in a general way irrespective of the category and function of
the wh-word as its host and of the embedding constituent.
Parallel to pronominal elements, adjectives or case forms in other languages
(see VON HEUSINGER 2007 and GEIST & ONEA 2007), the Russian indefinite
particles considered here serve as specificity markers and according to our
analysis bring in unspecified modifiers. Depending on the function of the
modifiee, the indefinite particles signal specificity or its lack of predicative NPs,
argumental DPs or adverbials.
The wh-words with indefinite particles are NP heads or modifiers, not
determiners and characterize specificity as a contextually determined property of
a set of alternatives in a mental model of a modal subject. In its relation to
other indefinite particles, -to contrasts with -koe in signalling the non-
identifiability of the referent by the speaker and it contrasts with -nibud in
being +specific and without free choice character.
We have disregarded uses of -to with definite pronouns and adverbials as in
(55):
(55) Staruxi osvedomilis, iz tex li on Rajskix, kotorye proisxodili togda-to, ot

tex-to i ili tam-to (RG 1980, I, 1039: I. GONAROV).
Old women inquired whether he is one of the Rajskies family which came
from such and such a family at such and such time and lived in such and
such a place.
Furthermore, various types of restrictive modifiers of NPs have been considered
with respect to their combination with pronominal and adverbial wh-words with
-to. The adjectival modifier kakoj-to can occur at various positions in NP
depending on its interpretative possibilities.
It is evident that the base and surface positions of wh-words with the
indefinite particle -to are very different. It has been shown that P- and D-zero-
heads attract these wh-words to their respective Spec-positions. We assume that
there is a certain selective affinity between specificity of NPs, DPs and PPs and
various factors of veridicality. The details must be worked out.
We are convinced that our analysis is valid not only for Russian, but for
other Slavonic and non-Slavonic languages as well.
References
CHENG, L.-S. & A. GIANNAKIDOU (2006) (In)Definiteness, polarity and the role of wh-
morphology. Journal of Semantics 23, 135-183.
CHOMSKY, N. (2000) Minimalist inquiries: The framework. In: MARTIN, R., D. MICHALS & J.
URIAGEREKA (eds.) Step by Step: Essays in minimalist syntax in honor to Howard
Lasnik 89-115. Cambridge, MA: MIT Press.
CHOMSKY, N. (2001) Derivation by phase. In: KENSTOWICZ, M. (ed.) Ken Hale: A life in
language 1-50. Cambridge, MA: MIT Press.
GEIST, L. (2008) Specificity as referential anchoring. Evidence from Russian. In: GRONN, A.
(ed.) Proceedings of SUB 12 151-164. Oslo: ILOS.
GEIST, L. & E. ONEA GSPR (2007) Specificity and implicatures. In: ALONI, M., P. DEKKER
& F. ROELOFSON (eds.) Proceedings of the sixteenth Amsterdam Colloquium.
December 17-19, 2007 109-116. ILLS/Dept. of Philosophy, University of Amsterdam:
Palteam.
GIANNAKIDOU, A. (2009) (Non)Veridicality and mood choice: subjunctive, polarity, and time.
To appear in: MUSAN, R. & M. RATHERT (eds.) Tense across languages.
HASPELMATH, M. (1993) A typological study on indefinite pronouns. Ph.D. dissertation. Freie
Universitt Berlin. [ Published 1997: Indefinite pronouns. Oxford: OUP.]
VON HEUSINGER, K. (2007) Referentially anchored indefinites. In: COMOROVSKI, I. & K. VON
HEUSINGER (eds.) Existence. Syntax and semantics, 273-292. Dordrecht: Springer.
KAGAN, O. (2007) Specificity and the speakers beliefs. Manuscript. Hebrew University of
Jerusalem.
KRATZER, A. & E. SELKIRK (2007) Phase theory and prosodic spellout: The case of verbs. The
Linguistic Review 24.2/3: 93-135.
PADUEVA, E. V. (1985) Vyskazyvanie i ego sootnesennost s dejvitelnostju
(Referencialnye aspekty semantiki mestoimenij). Moskva: Nauka.
RIKA, R. (1973) Kto-to und kto-nibud. Zeitschrift fur Slawistik 18: 705-736.
TATEVOSOV, S. (2002) Semantika sostavljajuix imennoj gruppy: kvantornye slova. Moskva:
IMLI RAN.
ZIMMERMANN, I. (1992) Der Skopus von Modifikatoren. In: Zimmermann , I. & A. Strigin
(eds.) Fgungspotenzen. Zum 60. Geburtstag von Manfred Bierwisch (= Studia
grammatica 34) 251-279. Berlin: Akademie Verlag.
ZIMMERMANN, I. (2000) Die Analysierbarkeit von Pronomen und Proadverbialia. In: Bittner,
A., D. Bittner & K.-M. Kopcke (eds.) Angemessene Strukturen: Systemorganisation in
Phonologie, Morphologie und Syntax 261-282. Hildesheim, Zurich, New York: Olms.
ZIMMERMANN, I. (2008) On the syntax and semantics of kakoj and to za in Russian. Journal
of Slavic Linguistics 16.2: 289-305.
ZIMMERMANN, I. (2009) Satzmodus. In: Kempgen, S., P. Kosta, T. Berger & K. Gutschmidt
(eds.) Die slavischen Sprachen. Ein internationales Handbuch zu ihrer Struktur, ihrer
Geschichte und ihrer Erforschung Band 1 484-505. Berlin, New York: de Gruyter.
Elena Gorishneva, Humboldt University Berlin, Institute of Slavic Languages,

10099 Berlin, Unter den Linden 6, jelenagor@gmx.de
Ilse Zimmermann, Gontardstrasse 126, 14471 Potsdam, Germany,
ilsezimmermann@arcor.de
Czech questions with two wh-words
Hana Gruet-Skrabalova
1. Introduction
This paper discusses three question forms the Czech language uses in questions
containing two wh-words. The three question forms are exemplified in (1). In
(1a), both wh-words are fronted. In (1b), both wh-words are fronted and the con-
junction a (and) appears between them. In (1c), one wh-word is fronted and
the other one is introduced by the conjunction a in clause-final position.
(1) a. Kdo komu koupil knihu? (multiple fronting)
who-NOM who-DAT bought book
Who bought a book for whom?
b. Kdo a komu koupil knihu? (fronting and conjunction)
who-NOM and who-DAT bought book
Who bought a book, and for whom?
c. Kdo koupil knihu a komu? (conjoined final wh-word)
who-NOM bought book and who-DAT
Who bought a book, and for whom?
Assuming that languages are economical, the fact that a language uses three
question forms with the same linguistic material as in (1) implies that we should
find differences in their interpretation and in their syntactic structure. Moreover,
the syntax of these questions should tell us something about the structure of the
CP domain in Czech.
The paper is organized as follows. In section 2, 3 and 4, I discuss semantic and
syntactic properties of the question like in (1). The section 5 discusses some hy-
potheses about the structure of CP and IP domains in Czech and proposes an
analysis of multiple fronting. The section 6 deals with conjoined wh-words. The
section 7 deals with the questions containing a final wh-word.
2. Semantic interpretation
2.1 Questions with multiple fronting
Questions like in (1) do not receive the same interpretation. It has already been
noted for other Slavic languages (BOKOVI 2002, PRZEPIORKOWKI 1994, RUDIN
1988) and for Hungarian (LIPTAK 2001) that questions with multiple fronting
trigger pair-list reading, as in (2).
180 Hana Gruet-Skrabalova
(2) a. Kdo koho pozval na verek?1

who-NOM who-ACC invited to party-ACC
Who invited whom to the party?
b. for each x and y, which x invited which y to the party?
c. Petr pozval Marii, Pavel Alenu a Jan Evu.
Petr-NOM invited Marie-ACC, Pavel-NOM Alena-ACC and Jan-NOM Eva-ACC
Peter invited Mary, Paul (invited) Alena, and John (invited) Eva.
However, specific reading is also possible for multiple fronting in cases there is
a specific pair x and y given in the context, and the question simply asks to iden-
tify the role of both x and y in the event denoted by the verb, as in (4). The two
wh-words here contrast with each other, so that we can speak about a contrastive
specific reading.
(4) (speaking about Can and Abel)
a. Kdo koho zabil?2
who-NOM killed who-ACC?
Who killed whom?
b. given x and y, is it x who killed y or is it y who killed x?
c. Kain zabil Abela, (a ne Abel Kaina).
Can-NOM killed Abel-ACC (and not Abel-NOM Can-ACC)
It is Can who killed Abel.
2.2 Questions with conjunction
Questions with fronted wh-items and conjunction only receive specific reading,
as we can see in (5). Contrary to multiple fronting, however, the single pair is
not presupposed here, so that the specific reading is not a contrastive one.
(5) a. Kdo a koho pozval na verek?
who-NOM and who-ACC invited to party-ACC
Who invited whom to the party?
b. for which x and y is it true that x invited y to the party?
c. Petr pozval na verek Marii.
Petr-NOM invited to party-ACC Marie-ACC
Peter invited Mary to the party.
The question in (5) cannot be paraphrased by a sentential coordination, nor the
conjunction a (and) replaced by the conjunction nebo (or). This also shows
that both wh-words here are involved in a single event denoted by the verb.
__________
1 Czech questions with multiple fronting do not show any superiority effects, although
some order preferences have been reported by MEYER (2004). The order of wh-words
simply indicates how the demand of information is structured.
2 Moreover, superiority effects hold in these questions, as noted by MEYER (2002).
Czech questions with two wh-words 181
(6) a. *Kdo pozval na verek a koho pozval na verek?

who-NOM invited to party-ACC and who-ACC invited to party-ACC
b. *Kdo nebo koho pozval na verek?
who-NOM or who-ACC invited to party-ACC
On the contrary, questions with final conjoined wh-word are always interpret-
able as two independent questions. In (7), for instance, the first question asks to
identify the x who came. The second one presupposes that x has been identified
and only asks when he/she/they came.
(7) a. Kdo piel a kdy (piel)?3
who-NOM came and when (came)
Who came, and when?
b. for which x is it true that x came and when did he/she come?
c. Piel Martin, a to v ter.
Came Martin, and that on Tuesday
Martin came on Tuesday.
We may thus conclude that questions in (1) are not semantically equivalent,
since questions with multiple fronting trigger a pair-list or a contrastive specific
reading, questions with fronting and conjunction a specific non-contrastive read-
ing, and questions with final conjoined wh-word a sentential reading.
3. Constituency of wh-words
3.1 Questions with multiple fronting
The questions in (1) also differ with respect to the constituency of their wh-
words. In questions with multiple fronting, the two wh-words are syntactically
independent constituents. Indeed, second position clitics, which must immedi-
ately follow the first phrasal constituent of the clause, normally follow the initial
wh-word (LENERTOVA 2001).
(8) a. Komu jsi co koupil?
who-DAT CL:AUX-2SG what-ACC bought
'what did you buy for whom?'
b. *Komu co jsi koupil?
LENERTOVA (2001) has also noted that, sometimes, both wh-words may precede
the clitics, as in (9b). In such case, however, only a specific contrastive reading
is available, as in (4) above. On the contrary, the question (9a) triggers a pair-list
reading. I will propose in section 5 that wh-words in (9a) and (9b) respectively
do not occupy the same positions both at surface and at LF.
__________
3 Questions in (7) are only possible with non-argumental final wh-word (see section 4).
(9) a. Kdo si koho v vc? (LENERTOV 2001)

who-NOM CL:REFL who-ACC appreciated more
= for every x and for every y, which x appreciated more which y?
b. Kdo koho si v vc?
who-NOM who-ACC CL:REFL appreciated more
= given x and y, is it x who appreciates more y or is it y who appreciates
more x?
Another evidence for independency of the fronted wh-words comes from the
placement of the particle e, which must follow the first wh-words and precede
the clitics, as in (10). Note that the question in (10a) is not interpreted as an in-
formation question, but rather as an echo-question, i.e. question asking what
proposition was actually asserted or intended to be asserted (ARNSTEIN 2002).
(10) a. Kdo e mu co koupil?
who-NOM that-PART CL:he-DAT what-ACC bought
(did he actually say that) the person A bought the thing B to him?
b. *Kdo co e mu koupil?
3.2 Questions with conjunction
Turning to questions with conjunction, we observe that clitics as well as the par-
ticle e must follow both wh-words. This implies that wh-words and conjunction
form a single constituent. Again, the presence of the particle e forces to inter-
pret the question (11a) as an echo-question.
(11) a. Kdo a co (e) mu koupil?
who-NOM and what-ACC that-PART CL:he-DAT bought
without e: who bought what to him?
with e: (did he actually say that) the person A bought the thing B to
him?
b. *Kdo e mu a co koupil?
We may thus conclude that questions in (1) also differ with respect to their con-
stituency, since wh-words do form distinct constituents in questions with multi-
ple fronting, while they do form a single constituent in questions with fronting
and conjunction. In questions with conjoined final wh-item, the wh-words
clearly do not form a constituent.
4. Argumental properties
Finally, wh-words are not equally acceptable in questions like in (1), depending
whether they are arguments or adjuncts (see PRZEPIORKOWSKI (1994) for asym-
metry between arguments and adjuncts in other Slavic languages).
4.1 Argumental wh-words

Starting with argumental wh-words (subject and verb complements), we observe
that both questions with multiple fronting and question with conjunction are fe-
licitous. A slight preference for order subject > complement can be observed
when both wh-words refer to the same semantic type, probably due to some
processing difficulties, as suggested by MEYER (2004).
(12) a. Kdo koho /?Koho kdo doporuil komisi?(MEYER 2004)
who-NOM who-ACC/who-ACC who-NOM recommended committee-DAT
b. Komu co / Co komu ekl?
who-DAT what-ACC / what-ACC who-DAT (he) said
(13) a. Kdo a koho / Koho a kdo doporuil komisi?
who-NOM and who-ACC / who-ACC and who-NOM recommended committee-DAT
b. Komu a co / Co a komu ekl?
who-DAT and what-ACC / what-ACC and who-DAT (he) said
On the contrary, questions with conjoined final wh-word are ruled out, except
for those where the final wh-word is an optional complement, as in (14b). This
should not be not surprising, since these questions are interpreted as independent
questions.
(14) a. *Kdo doporuil komisi a koho?
who-NOM recommended committee-DAT and who-ACC
a.*Koho doporuil komisi a kdo?
who-ACC recommended committee-DAT and who-NOM
b. *Komu ekl a co?
who-DAT said and what-ACC
b.Co ekl a komu?
what-ACC said and who-DAT
4.2 Non-argumental wh-words
Turning to non-argumental wh-words, we observe that questions with multiple
fronting are ruled out, while question involving conjunctions are all felicitous.
(15) * Kdy jak / *Jak kdy skonila stolet vlka?
When how / how when finished hundred-years' war
(16) Kdy a jak / Jak a kdy skonila stolet vlka?
When and how / how and when finished hundred-years' war
(17) a. Kdy skonila stolet vlka a jak?
When finished hundred-years' war and how
b. Jak skonila stolet vlka a kdy?
How finished hundred-years' war and when
4.3 Mixed wh-words

Finally, when the wh-words are of different types, questions with multiple front-
ing are good, except for those containing pro (why) and jak (how). Questions
with jak (how) are however judged acceptable by some speakers4.
(18) a. Koho kde / Kde koho vidl?
who-ACC where / where who-ACC (he) saw
b. ??Kdo jak / ?Jak kdo cestoval na konferenci?
who-NOM how / how who-NOM went to conference-ACC
c. *Kdo pro / *Pro kdo piel?
who-DAT why / why who-NOM came
Questions with conjoined wh-items are mainly good, although the order adjunct
> argument seems a little degraded:
(19) a. Koho a kde / Kde a koho vidl?
who-ACC and where / where and who-ACC (he) saw
b. Kdo a jak / ?Jak a kdo cestoval na konferenci?
who-NOM and how / how and who-NOM went to conference-ACC
c. Komu a pro / ?Pro a komu to dal?
who-DAT and why / why and who-DAT it (he) gave
It is plausible that the weak contrast in acceptability in (19b, c) is again due to
some processing difficulties, since it seems easier to obtain a construal in which
a manner is assigned to an individual, than a construal in which an individual is
assigned to a manner.
Finally, questions with conjoined final wh-word are excluded when an
obligatory argumental wh-word is in clause-final position, as in (14) above.
(20) a. Koho vidl a kde?
who-ACC (he) saw and where
a.*Kde vidl a koho?
where (he) saw and who-ACC
b. Kdo hodnotil student a jak?
who-NOM evaluated students-ACC and how
b.*Jak hodnotil student a kdo?
how evaluated students-ACC and who-NOM
The table below gives an overview of the properties of the questions in (1):
__________
4 Acceptable judgements for examples like (18b) can also be found in Meyer (2004).
Table 1: properties of questions with two wh-words
(1a) Wh1 Wh2 (1b) Wh1 Conj Wh2 (1c) Wh1 ... and Wh2
Reading a) Pair-list Specific Sentential
b) Specific contrastive
Constituency a) Wh1 e Cl Wh2 Wh1 Conj Wh2 e Cl Wh1 e Cl.... Wh2
b) Wh1 Wh2 e Cl
Wh-arguments ok ok * / ok5
Wh-adjuncts * ok ok
Wh-mixed ?? jak /* pro ok * / ok5
5. Multiple fronting
Evidence provided in previous sections to show that questions in (1) have differ-
ent semantic and syntactic properties suggests that these questions also involve
different constructions. Before turning to their analysis, I would like to introduce
some preliminary hypotheses about the clause structure in Czech.
5.1 The CP and IP domain
RIZZI (1997, 2002) proposes that the CP domain of the clause contains several
syntactic positions each dedicated to a particular element with respect to the type
of the clause and its information structure, as indicated in (21):
(21) [ForceP [TopP* [IntP [TopP* [FocP [TopP* [FinP [IP]]]]]]]]
Contrary to Italian, the CP domain in Czech is less articulated, as already shown
by LENERTOVA (2001). I follow Lenertova in assuming that clitics are hosted by
the head of FinP, functioning as a boundary between the IP and CP domain. I
also assume that Czech CP contains three other projections (SKRABALOVA
2008): ForceP whose head hosts complementizers, IntP whose specifier hosts a
wh-word in single questions, and a projection between ForceP and FinP, which I
call ContrastP and whose specifier may hosts a focussed or a topicalized XP that
occurs between the complementizer/or the wh-word and the clitics, as in (22):
__________
5 If Wh2 is an optional argument.
(22) a. Myslel, [ForceP e [ContrastP TY [FinP jsi mu [IP to auto u vrtil]]]]

thought that you CL:AUX-2SG CL:he-DAT this car-ACC already gave-back
'He believed that YOU had already given him back this car.'
b. A chtl bys vdt [ForceP [IntP co [ContrastP MN [FinP se [IP stalo]]]]]?
and wanted CL:COND-2SG know what me-DAT CL:REFL happened
'And would you like to know what happened to ME?'
Another evidence in favour of such Contrast projection comes from the particle
e, which may follow a focussed XP or a wh-word in independent interrogative
clauses. These clauses are however not information questions, but echo-
questions (see section 3). I argued elsewhere (GRUET-SKRABALOVA 2010) that
e in these questions is a focus particle, i.e. the head of ContrastP. The focused
XP and the wh-word preceding e would then occupy the Spec(ifier) of Con-
trastP:
(23) a. [ForceP [ContrastP ON [Contrast' e [FinP mi [IP zatelefonuje]]]]]?
he-FOC PART CL:I-DAT will-call
b. [ForceP [ContrastP Kam [Contrast' e [IP el]]]]?
where PART (he) went
Finally, it has been argued by BELETTI (2004) that the architecture of the domain
below IP and above little vP parallels that of the domain CP. It seems plausible
to postulate in Czech an inner Topic projection between IP and vP which would
host topics as the NP [to auto] in (22a):
(22) a.Myslel, [ForceP e [ContrastP TY [FinP jsi mu [IP [TopP to auto [vP u vrtil]]]]]]
thought that you CL:AUX-2SG CL:he-DAT this car-ACC already gave-back
5.2 The analysis of Multiple fronting
Assuming that a topic position occurs both in the CP and in the IP domain, I will
propose that there is also an Interrogative projection in the IP domain. In ques-
tions with multiple fronting, the inital wh-word preceding the clitics will move
to the Spec of IntP in the CP domain, while the lower one following the clitics
would move to the Spec of IntP in the IP domain:
(24) [ForceP [IntP Kdo [FinP si [IP [IntP co [ContrastP [vP peetl]]]]]]]?
who CL:REFL what read
The higher wh-word in (24) takes scope over the lower wh-word. In their papers
on distributive (i.e. pair-list) reading of quantifiers, BEGHELLI & STOWELL
(1994, 1997) define distributivity as a binary relation requiring the simultaneous
presence of a distributor and a distributee. They propose that distributive reading
follows from interaction between two specific positions at L(ogical) F(orm),
Dist(ributive) Phrase and Share Phrase, as in (25). The DistP hosts the distribu-
tor, i.e. the quantifier higher in the clause, and its head selects ShareP, which
hosts the share of distribution, i.e. the quantifier lower in the structure.
(25) [RefP [CP [AgrSP [DistP [ShareP [NegP [AgrOP [VP]]]]]]]]

I claim that wh-words triggering a pair-list reading occupy DistP and ShareP po-
sitions at LF, see (26b). The ShareP can be identified with the lower IntP at sur-
face, see (26a). The DistP occurs immediately above ShareP and below IP. The
higher quantifier is reconstructed to the Spec of DistP:
(26) a. [ForceP [IntP Kdo [FinP se [IP [IntP komu [vP omluvil]]]]]]?
who-NOM CL:REFL who-DAT apologized
'Who apologized to whom?'
b. [ForceP [IntP ti [FinP se [IP ti [DistP Kdoi [ShareP komu [vP ti omluvil t]]]]]]]?
Assuming this analysis, we may now turn to questions in which both wh-words
appear before clitics. I claim that in these questions, the lower wh-word does not
occupy the lower Spec of IntP, but rather moves directly to the Spec of Con-
trastP, as in (27a). Consequently, a distributive relation between the two wh-
words cannot be established, see (27b). On the contrary, activating of the head
Contrast forces a contrastive specific reading:
(27) a. [ForceP [IntP Kdo [ContrastP komu [Fin se [IP [vP omluvil]]]]]]?
who-NOM who-DAT CL:REFL apologized
'Who apologized, and to whom?'
b. [ForceP [IntP Kdo [ContrastP komu [FinP se [IP t [DistP [ShareP [vP t omluvil t]]]]]]]]?
As for examples like (28), I suggest that the presence of the particle e activates
the head of ContrastP, which attracts the lower wh-word at LF and forces again
a contrastive specific reading of wh-words in these questions:
(28) a. [ForceP [IntP Kdo [ContrastP e [FinP se [IP [IntP komu [vP omluvil]]]]]]]?
who-NOM FOC CL:REFL who-DAT apologized
(Did he say that) the person A apologized to the person B?
b. [ForceP [IntP Kdo [ContrastP komui e [FinP se [DistP [ShareP ti [FinP [IP t omluvil
t i]]]]]]]]?
5.4 Argument-adjunct asymmetry
Let us turn now to questions with non-argumental wh-words. I claim that their
unacceptability follows from the adverbial character of how and why, which
prevents them to enter both distributive and contrastive relation. To do that, I
propose to distinguish quantified and adverbial wh-words with respect to their
capacity of being individualized and of being referential, as shown in table 2:
Table 2: Quantified vs. adverbial wh-words
Q-wh [+Ind, +Ref] > Adv-wh [Ind, -Ref]
kdo (who), co (what) > kde, kam (where), kdy (when) > jak (how) > pro (why)
For the distributivity relation to be satisfied, both the Spec of DistP and the Spec
of Share be must be filled with appropriate elements at LF. Namely, the operator
in Dist is defined to apply only to individuals, meaning that only QPs that can
be individualized can access the Spec of DistP. The Spec of Share is required to
be semantically a QP that can co-vary with the distributor. It follows that only
quantified wh-words endowed with the feature [+Ind(ividual)] may enter dis-
tributive relation, i.e. occupy DistP and ShareP.
Furthermore, contrastive reading will only apply to XP endowed with the feature
[+Ref(erential)]6, which is compatible with the presupposition of a specific pair.
Since why and how are adverbials, thus [-Ind] and [-Ref], they may enter neither
distributive nor contrastive relation with another wh-word:
(29) a. *Kde jak Petr uhodil Pavla?
Where how Petr-NOM hit Pavel-ACC
b. *Pro kam Petr poslal Pavla?
Why where Petr-NOM sent Pavel-ACC
As for examples with jak in (18b) above, which were not judged degraded by all
speakers, I suggest that their acceptability depends on speakers capacity for in-
terpreting jak as a quantified, rather than adverbial wh-word.
This analysis correctly predicts that the only way to combine two adverbial
wh-words or different wh-word is to use the question form with conjunction, as
in (30), since these questions trigger neither distributive nor contrastive reading.
The next section will deal with the syntax of these questions.
(30) a. Kde a jak Petr uhodil Pavla?
Where and how Petr-NOM hit Pavel-AC
b. Pro a kam Petr poslal Pavla?
why and where Peter-NOM sent Paul-ACC
6. Multiple fronting with conjunction

6.1 Conjunction as a focus particle
PENN (1999) claims that the morpheme i (and) that occur between wh-words in
Serbo-Croatian is not a conjunction, but a focus particle (also), as in (31ab).
He thus proposes to consider questions in (31c) as questions with multiple front-
ing in which wh-words are simply focused.
(31) a. Ivan je i danas sreo Mariju.7 (PENN 1999)
Ivan CL:AUX-3SG also today met Mary-ACC
Ivan also met Mary today (not only yesterday).
__________
6 Moreover, their referent must be of the same type, otherwise it is not relevant to contrast
them.
7 The glosses and the translations in the example (31) are taken from PENN (1999).
b. Knjigu i Mariji odnesi.

book-ACC and Mary-DAT bring-IMP
Bring the book to Mary.
c. Ko i kome je kupio auto?
who-NOM and who-DAT CL:AUX-3SG bought car
Who bought the car for whom?
There are however three pieces of evidence against Penn's analysis in Czech.
First, fronted wh-words and conjunction form a single constituent (see section
3). Second, they do not trigger list-pair reading (see section 2). Finally, the con-
junction a cannot be analyzed as a focus particle, contrary to the conjunction i
(and), which may actually function as a focus particle (also, even), see
(32ab). However, i cannot appear between wh-words, as shown in (32c).
(32) a. Jan potkal Marii vera a /i dneska.
Jan-NOM met Marie-ACC yesterday and / and today
with a : John met Mary yesterday and today.
with i : John met Mary both yesterday and today.
b. Jan potkal Marii *a / i dneska.
Jan-NOM met Marie-ACC and / even today
John met Mary even today (not only yesterday).
c. Kdo a / *i komu koupil auto?
who-NOM and / and who-DAT bought car-ACC
Who bought the car to whom?
I conclude thus that questions with conjunction do not involve multiple fronting,
but rather coordination of wh-words.
6.2 Wh-coordination
Coordination of wh-words may seem problematic, since constituents with dif-
ferent syntactic functions cannot normally be coordinated (PETERSON 2004), as
shown in (33a). However, coordinations with conjuncts bearing different func-
tions become felicitous with conjuncts being focussed (LIPTAK 2001), as in
(33b). Note that the example (33b) is a possible answer to the question in (33c).
(33) a. *Jan by chtl pozvat [ConjP [NP Marii] a [PP do kina]].
Jan-NOM CL:COND wanted invite Marie-ACC and to cinema-GEN
b. Jan by chtl pozvat [ConjP[+Foc] MARII a DO KINA].
John CL:COND wanted invite Mary-ACC and to cinema
John would like to invite Mary to the movies.
c. [ConjWhP[+Foc] Koho a kam] by chtl Jan pozvat?
who-ACC and where CL:COND wanted Jan-NOM invite
Whom would John like to invite, and where?
Assuming that focusing licenses coordination of unlike categories, I suggest that

coordinate wh-phrases are licensed because wh-words share two features: [+wh]
and [+focus] (LIPTAK 2001). A coordinate wh-phrase, forming a single focused
constituent, would move to a single position in the left periphery of the clause,
the Spec of ContrastP. Embedding of the two words within a coordinate phrase
would block their mutual scope, allowing only for their specific reading:
(34) [ForceP [ContrastP [ConjWhP Koho a kam] [Contrast' [FinP by [IP chtl
who-ACC and where CL:COND wanted
Jan pozvat ]]]]]]?
Jan-NOM invite
7. Questions with conjoined final wh-word

In questions with clause-final wh-word introduced by the conjunction a, the
clause including the initial wh-word and the final wh-word are interpreted as
two independent single questions. I argue thus that these questions involve
clausal coordination with one elliptical conjunct. Clausal coordination prevents
the argument of the verb from occurring in the clause-final position and explains
the sentential interpretation of these questions. It also predicts that such ques-
tions may combine with multiple fronting and wh-coordination:
(35) a. [Kdo komu koupil knihu] a [kdy]?
who-NOM who-DAT bought book and when
Who bought a book to whom, and when?
b. [Kdo a kam el] a [pro] ?
who-NOM and who-DAT went and why
Who went where, and why?
It seems however difficult to analyze the elliptic conjunct as a clause with de-
leted material. The deletion analysis would be indeed plausible if syntactic re-
construction with identity were always possible, which is not the case. The syn-
tactic reconstruction is only possible when the initial wh-word is an adjunct; if it
is an argument, an NP or a pronoun must appear in the second clause, as in
(36c):
(36) a. [Kdy jsi potkal Jana] a [kde (jsi potkal Jana)]?
when CL:AUX-2SG met Jan-ACC and where CL:AUX-2SG met Jan-ACC
When did you meet John and where (did you meet him)?
b. *[Koho jsi potkal] a [kde (jsi potkal)]?
who-ACC CL:AUX-2SG met and where CL:AUX-2SG met
(*Who did you meet John and where did you meet?)
c. [Koho jsi potkal a [kde (jsi ho potkal)]?
who-ACC CL:AUX-2SG met and where CL:AUX-2SG CL:him met
Where did you meet John and where did you meet him?
According to GINZBURG & SAG (2001), elliptical clauses as in (35) and (36) are
syntactically clausal fragments, which are only interpreted as complete clauses.
An analysis in terms of semantic reconstruction seems thus to be needed for this
kind of question. Such an analysis goes however beyond the limits of this paper.
8. Conclusion
In this paper, I have discussed and analyzed three types of questions with two
wh-words that occur in Czech. It was shown that these questions involve differ-
ent syntactic constructions with different semantic representations. In questions
with multiple fronting, wh-words either activate DistributiveP and ShareP pro-
jections at Logical Form, which leads to their pair-list reading, or the second one
moves to ContrastP, which leads to their contrastive specific reading. In ques-
tions with fronting and conjunction, the coordinate wh-phrase occupies the Spec
of ContrastP, which leads to a specific reading of conjoined wh-words. Ques-
tions with a final conjoined wh-word involve conjoined single questions.
References
ARNSTEIN, R. (2002) A focus semantics for echo questions, in BENDE-FARKAS, A. & A.
RIESTER (eds) Proceedings of the Workshop on Information Structure in Context, 98
107.
BEGHELLI, F. (1994) Distributivity and Pair-list reading. In: SZABOLCSI, A. (ed.) Ways of
Scope Taking 349-408. Dordrecht: Kluwer.
BEGHELLI, F. & T. STOWELL (1997) Distributivity and Negation: the Syntax of each and every.
In: SZABOLCSI, A. (ed.) Ways of Scope Taking 71-109. Dordrecht: Kluwer.
BELLETI, A. (2004) Aspects of the low IP area. In: RIZZI, L. (ed.) The Structure of CP and IP.
The Cartography of Syntactic Structures vol. 2. New York: OUP.
BOKOVI, . (1999) On multiple feature-checking: Multiple wh-fronting and multiple head-
movement. In: EPSTEIN, S. & N. HORNSTEIN (eds.) Working Minimalism 159-187.
Cambridge MA: MIT Press.
BOKOVI, . (2002) On multiple wh-fronting, Linguistic Inquiry 33. 351-383.
GINZBURG, J. & I. SAG (2000) Interrogative Investigations, CSLI Publications.
GREPL, M. & P. KARLK (1998) Skladba etiny. Praha: Votobia.
GRUET-SKRABALOVA, H. (2010) Le statut syntaxique de e (que) dans les phrases indpen-
dentes en tchque, to appear in Cahiers du Laboratoire de recherche sur le langage,
Presses Universitaires de Clermont-Ferrand.
LENERTOV D. (2001) On Clitic Placement, Topicalization and CP-Structure in Czech. In:
KOSTA, P. & J. FRASEK (eds.) Current Approaches to Formal Slavic Linguistics, 294-
305, Frankfurt a/Main: P. Lang.
LIPTK, A. (2001) On the Syntax of Wh-items in Hungarian, PhD Dissertation: Utrecht, LOT.
MEYER, R. (2002) On multiple wh-fronting and wh-clustering in Czech, Formal Approaches
to Slavic Linguistics 11: Amherst meeting.
MEYER, R. (2004) Superiority effects in Russian, Polish, and Czech, Cahiers linguistiques
d'Ottawa, 32: 44-65.
PENN, G. (1999) Linearization and WH-Extraction in HPSG: Evidence from Serbo-Croatian.
In: BORSLEY, R. & A. PRZEPIORKOWSKI (eds.) Slavic in HPSG, CSLI Publications.
PETERSON, P. (2004) Coordination: consequences of a lexical-functional account Natural

Language and Linguistic Theory 22.3, 643-679.
PRZEPIORKOWSKI, A. (1994) Critical Review of Approaches to multiple Wh-Movement, Re-
search paper, Centre for Cognitive Science. University of Edinburgh.
RIZZI L. (1997) The Fine Structure of the Left Periphery. In: HAEGEMAN, L. (ed) Elements of
grammar 281-337. Dordrecht: Kluwer.
RIZZI, L. (2002) Locality and Left Periphery. In: Belletti, A. (ed) Structures and Beyond. The
Cartography of Syntactic Structures vol. 3. Oxford: OUP.
RUDIN, C. (1988) On Multiple questions and Multiple Wh-fronting, NLLT 6 445-501.
SKRABALOVA, H. (2007) Conjoined Wh-items. In: KARLK, P. et al. (eds.) Czech in Generati-
ve Grammar 161-174. Munique: Lincom Europe.
SKRABALOVA, H. (2008) Dtachement gauche en tchque. In: APOTHLOZ, D. et al. (eds.)
Les linguistiques du dtachement 535-549. Bern: Peter Lang.
Hana Gruet-Skrabalova, Clermont Universit, Universit Blaise Pascal EA 999,

Laboratoire de recherche sur le langage, BP 10448, F-63000 CLERMONT-
FERRAND, hana.gruet-skrabalova@univ-bpclermont.fr
Optionality of the Genitive (of Negation) in Slovene1
Gaper Ilc
1. Introduction
The paper explores the Genitive of negation (henceforth: GoN) in Slovene from
a cross-linguistic perspective and proposes an account of its observed properties
in terms of minimalist Case checking theory (Pesetsky & TORREGO 2001, 2004,
BAILYN 2004, MATUSHANSKY 2008 a.o.).
It is a well-established fact that Slavic languages fall into three typological
groups with regard to the (non-)occurrence of the GoN: (i) languages without
the GoN (Serbian/Croatian), (ii) languages with the optional GoN (Russian) and
(iii) languages with the obligatory GoN (Polish). Traditionally, Slovene has been
assumed to be a language with the obligatory GoN, however, examples from
(spoken) Slovene show that the optionality of the GoN is also available. In such
cases, the Gen alternates with the Acc(usative). The analysis of the data shows
that Slovene cannot be straightforwardly categorised into any of the three typo-
logical groups, since it displays characteristics of languages with the obligatory
GoN as well as languages with the optional GoN.
I argue that the optionality of the GoN can be best understood in terms of
feature selection and probing (BAILYN 2004). Thus, the Gen/Acc variation under
the scope of negation is determined by the type of the variant feature on the no-
minal. In the case of the GoN, the variant feature on the nominal is the quantifi-
cation feature [q], which requires to agree with the Q head selected by the Neg.
On the other hand, in the case of the Acc, the variant feature on the nominal is
the unvalued [telic] feature, which in turn Agrees with the [telic] in AspP as is
the case in non-negative sentences.
The paper is organised as follows. Section 2 presents the GoN in Slavic and
its previous syntactic accounts. In 2.1 relevant Slovene data are presented with a
special focus on the syntactic properties of the GoN, the optionality of the GoN
and the multiplicity of the GoN in Slovene. In section 3 a syntactic account of
the GoN in Slovene is proposed. Section 4 concludes the paper.
2. The Genitive of Negation
The GoN is frequently associated with the negation in Slavic languages, which
can be classified into three groups with respect to the (non-)occurrence of the
GoN:
__________
1 I am grateful to the audience at the Formal Description of Slavic Languages-8 confer-
ence, Potsdam, December 2-5, 2009 for valuable comments and suggestions.
194 Gaper Ilc
Group 1. The GoN is almost completely absent (Serbian/Croatian)2 or archaic

(Czech):
(1) Jovan nije itao asopis/*asopisa.
Jovan not-is read newspaper-ACC/newspaper-GEN
Jovan didnt read the newspaper.
Group 2. The GoN is optional and alternates with the Acc(usative). The most
studied language belonging to this group is Russian, for which BAILYN (1997)
claims that the Gen/Acc alternation under the scope of negation pertains to dif-
ferent pragmatic and semantic dimensions: the Acc yields an individual, topical
or definite interpretation, while the Gen is associated with an existential or inde-
finite interpretation.3 In Russian, the GoN is not restricted only to underlying
objects under negation (2a), but may, though need not, appear with the unaccus-
ative subjects (2b) and derived subjects of passives (2c):
(2) a. Ja ne itaju urnaly/urnalov. BROWN (1999)
I not read magazines-ACC/ magazines-GEN
I dont read magazines.
b. Otveta ne prilo.
answer-GEN not cameNEUT,SG
No answer came.
c. Ne bylo polueno gazet.
Not was received-NEUT newspapersGEN, PL
No newspapers were received.
Group 3. In this group, the GoN is completely grammaticalised and obligatory
(Polish (3a)). In contrast to Russian, in Polish the GoN cannot apply to the un-
accusative subjects (3b) and derived subjects of passives (3c):
(3) a. Ewa nie czyta *gazety/gazet. ABELS (2005)
Ewa not reads newspapers-ACC/newspapers-GEN
Ewa doesnt read newspapers.
b. *nie przyso adnej/ (ani) jednej kobiety FRANKS (1995)
NEG arrived not-any(gen)/not-even one(gen) woman(gen)
c. *nie byo budowano katedry FRANKS (1995)
NEG was(n sg) built(n sg) cathedral(gen)
In group 1 and group 2 languages, the GoN is assigned in two syntactic envi-
ronments: (i) the local GoN assigned by a clausemate negation to the comple-
ment that is otherwise assigned the Acc (see (2a), (3a)), and (ii) the long-
__________
2 The only exceptions are negated existentials (see HARTMANN & MILIEVI to appear)
and complements to the negated form of the verb imati ('to have') (see FRANKS &
DZIWIREK 1993).
3 Not all authors agree that there is a semantic difference between the Gen/Acc under the
scope of negation. cf. BROWN (1999)
Optionality of the Genitive (of Negation) in Slovene 195
distance GoN assigned to the object of the non-negative embedded infinitival

clause by the matrix negation. The local GoN affects not only argument objects
of V but also non-argument objects of V. In Russian as well as in Polish such
occurrences of GoN are optional:
(4) a. Ivan ne ital ni minuty. FRANKS (1995)
(nom) NEG read not-even minute(gen)
b. nie spa+am godzin/godziny
NEG slept(f 1sg) hour(acc)/(gen)
In Russian, the Long-distance GoN occurs only in complements of subject con-
trolled verbs (5a) but not object controlled verbs (5b); whereas in Polish, it in-
volves complements of object and subject controlled verbs (5c,d), as well as
subject-to-subject raising structures (5e).
(5) a. Nataa ne xotela itat knig. ABELS (2005)
Nataa not wanted read-INF books-GEN
Nataa didnt want to read books.
b. *Ja ne ugovorila Natau itat knig.
I not persuaded N-ACC to-read books-GEN
I didnt persuade Natasha to read.
c. Nie chcia+em pisa listw.
NEG wanted write letters-GEN
I didnt want to write letters.
d. Nie kaza+em Marii pisa listw. PRZEPIRKOWSKI (2000)
NEG order M write letters-GEN
I didnt order Mary to write letters.
e. Nie wydawa+ si pisa listw.
NEG seem REFL WRITE letters-GEN
He didnt seem to be writing letters.
In Polish, the standard assumption is that both the local and the long-distance
GoN are obligatory regardless of the interpretation of the nominal(s) involved.
However, PRZEPIRKOWSKI (2000: 156-158) argues that the occurrences of the
long-distance GoN differ from the local GoN in the fact that the long-distance
GoN is optional (i.e. alternating with the Acc), and concludes that L[ong-
]D[istance] GoN is in principle always optional, and that additional pragmatic,
lexical, etc., factors may influence the actual preference for the accusative or the
genitive in various ways.
196 Gaper Ilc
PRZEPIRKOWSKI (2000) also shows that there can be more than one GoN
per single negation in the case of long-distance GoN, and refers to this pheno-
menon as the multiplicity of the GoN (6):4
(6) Janek nie uczy+ Marii lepi gankw. PRZEPIRKOWSKI (2000)
John not taught Mary-GEN mold pots-GEN
John didnt teach Mary how to make pottery.
2.1. The Genitive (of Negation) in Slovene
It is noteworthy that in Slovene with some masculine singular nouns there are
two distinct genitive endings: -a/-u (7). The -u marking is not productive; when
both endings are possible, they are used interchangeably.5
(7) Ne vidim mosta/mostu.
Not see bridge-GEN/bridge-GEN
I do not see the bridge.
In Slovene, the local GoN is typically assigned to direct objects (assigned the
Acc otherwise) under the scope of negation and in existential copula construc-
tions ((8a) and (8b) respectively). Under negation, subjects of the unaccusatives6
and derived subjects of passives cannot take the genitive (8c,d). The local GoN
can but need not be assigned to non-argument objects of V if introduced by the
negative particle niti not even (8e):
(8) a. Janez ni bral asopisa.
Janez not-is read newspaper-GEN
Janez didnt read the newspaper.
b. Knjige ni.
book-GEN not-is
There is no book.
c. *Odgovora ni prispel/prispelo.
AnswerGEN,MASC not-is arrivedSG,MASC/arrivedSG,NEUT
The answer did not arrive.
d. *Ni bilo zgrajeno hie. //*Ni bila zgrajena hie.
not-is beenNEUT builtNEUT houseFEM,GEN //not-is beenFEM builtFEM houseFEM,GEN
The house was not built.
__________
4 The same phenomenon seems non-existent in Russian (cf. ABELS 2005).
5 In contrast to Russian genitive1 and genitive2 ending with different distributions. cf.
FRANKS (1995)
6 In Russian, the status of unaccusatives is determined by the possibility of its subject to be
genitive marked. Since this test is inapplicable in Slovene, I use the test of secondary im-
perfective as proposed by SCHOORLEMMER (1995): if an intransitive perfective verb
derives a Secondary Imperfective (SI) it is an unaccusative verb.
(i) prispeti prispevati
arrive-pf. arrive-imp.
e. Janez ni delal niti minute/minuto.

Janez not-is worked not-even minute-GEN/minute-ACC
He didnt work for even a minute.
Slovene also displays the long-distance GoN with objects of the non-negative
embedded infinitival clause, including complements of subject (9a) and object
(9b) control verbs. The assignment of the long-distance GoN to object of the
non-negative wh-infinitival clauses (9c), and non-negative finite clauses is unac-
ceptable (9d). As (9e) shows, in Slovene, the multiplicity of the GoN is accepta-
ble, since a single negation may trigger the GoN over a series of infinitival com-
plements.
(9) a. Nataa ni hotela itati knjige.
Nataa not-is wanted read-INF book-FEM-GEN
Nataa didnt want to read a book.
b. Uitelj ne sili tudentov reevati taknih problemov.
teacher not forces students-GEN solve-INF such problems-GEN
The teacher does not force the students to solve such problems.
c. *Ne vem zakaj napisati pisma.
not know why write letter-GEN
I dont know why to write a letter.
d. *Janez ne pravi, da bere asopisov.
Janez not says that reads newspapers-GEN
Janez doesnt say that he reads newspapers.
e. Ne elim siliti tudentov reevati taknih problemov.
not want force students-GEN solve-INF such problems-GEN
I dont want to force the students to solve such problems.
Table 1 summarizes the Russian, Polish and Slovene data with regard to the
(non-)occurrences of the GoN. At this point, it could be concluded that Slovene
GoN displays the same pattern as Polish.
Russian Polish Slovene

argument objects of V 9 9 9
non- argument objects of V 9 9 9
Unaccusative 9 8 8
derived subjects of the passives 9 8 8
long-distance (subject control) 9 9 9
long-distance (object control) 8 9 9
Multiplicity 8 9 9
Table 1: Occurrence of the GoN in Russian, Polish and Slovene.
198 Gaper Ilc
In every-day, colloquial Slovene, the local and the long-distance GoN are fre-
quently replaced by the Acc, as exemplified by the Gen/Acc pairs in (10).7 Since
Slovene allows both the optionality and the multiplicity of the GoN, all, some or
no nominal(s) under the scope of negation may be assigned the GoN (10c):
(10) a. Janez ni bral asopis/asopisa.
Janez not-is read newspaper-ACC/newspaper-GEN
Janez didnt read the newspaper.
b. Nataa ni hotela itati knjigo/knjige.
Nataa not-is wanted read-INF book-ACC/book-GEN
c. tudentov/tudente ne elim siliti reevati problemov/probleme.
students-GEN/-ACC not want force solve-INF problems-GEN/-ACC
I dont want to force students to solve problems.
Occurrences of the Acc replacing the GoN, be it in the local or the long-distance
environment, are quite frequent, especially in spoken Slovene and varieties of
Slovene spoken in the Eastern parts of Slovenia. The corpus data also reveal that
the Gen/Acc alternation under the scope of negation is relatively free, and thus
does not seem to be governed by pragmatic and semantic factors as claimed for
Russian. Most examples with the Acc in lieu of the local GoN involve the rela-
tive position of the object with respect to the negation: (i) more accusative NPs
can be found in non-adjacent (11b) than in the adjacent position (11a), and (ii)
the majority of cases involve the pronominal clitic jo (3rd person, sg., feminine)
as in (11c).
(11) a. Zagotavljali so, da ne bodo prikrajali nobeno obino.
assured are that not will deprived no-ACC municipality-FEM-ACC
They assured that they wouldnt deprive any municipality.
b. Maribor v nobenem oziru nismo zoperstavili.
Maribor-MAS-ACC in no respect not-are neglected
We didnt neglect Maribor in any respect.
c. Janez mu jo ni dal.
Janez him-DAT her-ACC not-is given
Janez didnt give her to him.
In the case of the Long-distance GoN, the number of examples in which the Acc
replaces the GoN rises. In fact, in contrast to Polish where the majority of
speakers still prefer the Long-distance GoN to the Acc (cf. PRZEPIRKOWSKI
__________
7 It is noteworthy that in Slovene the GoN is still the prescribed form, which bears conse-
quences for any corpus based analysis, since the majority of printed texts in Slovenia are
still officially edited for grammatical correctness.
2000), many Slovene speakers frequently use the Acc in lieu of the long-
distance GoN:
(12) a. Ni elel izzivati jezo.
not+is wanted provoke-INF anger-ACC
He didnt want to provoke anger.
b. Nima pravice zahtevati demarkacijsko rto.
not+has right-GEN demand demarcation-ACC line-ACC
He doesnt have the right to demand the demarcation line.
Another interesting issue to explore is whether there is any kind of interdepen-
dency between the optionality and multiplicity of the GoN. For the multiplicity
of the GoN in Polish, PRZEPIRKOWSKI (2000) argues that the optionality affects
only distant objects and when the Acc is selected instead of the Gen, the next
object that could be assigned the GoN is also assigned the Acc. Hence, for the
majority of Polish speakers, patterns such as GoN+GoN+Acc and
GoN+Acc+Acc are possible, whereas patterns like GoN+Acc+GoN are highly
questionable. It seems that in Slovene the selection is less rigid allowing both
GoN+Acc and Acc+GoN (see also (10c)):
(13) Nisem imel pravice/pravico dajati navodil/navodila.
not-am had right-GEN/-ACC give-INF instructions-GEN/-ACC
I didnt have the right to give instructions.
2.1.1 Other cases of the optionality
Apart from the GoN, the use of the Gen is also optional when denoting the parti-
tive meaning (14a). The partitive genitive is felt unnatural and forced, and its
Acc counterpart (14b) as normal or typical. The question that the two examples
in (14) open is whether they are really identical in meaning, i.e. whether the par-
titive reading can still be obtained with the Acc. Most speakers use the Acc re-
gardless of the whole vs. part reference, and when the partitive meaning is in-
volved, they resort to the quantificational modification (e.g. a piece/loaf/slice
of).
(14) a. Daj mi kruha. b) Daj mi kruh.
give me bread-GEN give me bread-ACC
Give me some bread. Give me (some) bread.
In contrast to the partitive genitive, it seems that the s.c. do-in-quantity geni-
tive is more meaningful, since the Acc does not automatically trigger the mean-
ing of great quantity:
(15) Natovorili smo peska/?pesek. (do-in-quantity genitive)
loaded are sand-GEN/sand-ACC
We (have) loaded lots of sand.
200 Gaper Ilc
In other occurrences of the genitive case that are not associated with lexical se-
lection, such as adnominal genitive and quantificational genitive, there is no free
Gen/Acc variation:
(16) rojstvo Micke Kovaeve/*Micko Kovaevo (adnominal genitive)
birth Micka-GEN Kova-GEN/Micka-ACC Kova-ACC
The birth of Micka Kova.
(17) veliko/pet problemov/*probleme (quantificational genitive)
many/five problems-GEN/problems-ACC
Many/five problems.
When the Gen is selected lexically (e.g. reflexiva tantum), it is virtually obliga-
tory (18a) with perhaps the only exception of the 3rd sg. fem. pronominal clitic
je,8 which is occasionally replaced by the Acc, mostly in colloquial speech and
in non-adjacent positions (18b).
(18) a. Bojim se Mete/*Meto. b) Bojim se je/jo.
fear REFL. Meta-GEN/-ACC fear REFL. her-CL.GEN/ACC
Im afraid of Meta. Im afraid of her.
Table 2 summarizes the occurrences of the Gen in Slovene with respect to their
(non-)optional status.
type of genitive Optionality

partitive 999
long-distance GoN 99
local GoN 9
negated existentials 8
adnominal 8
quantificational 8
Lexical 8
Table 2.
3. Analysis
Within the Government and Binding theory (CHOMSKY 1981, 1986, 1993) the
structural case was understood in terms of the Case Filter, requiring that each
overt NP be assigned Case, and Case Criterion, according to which every NP
receives only one case and each case is assigned to only NP. However, it can be
observed cross-linguistically that both criteria are violated: (i) case can be as-
signed to nodes other than NPs, and (ii) case can be assigned to multiple heads
__________
8 This is the only pronominal clitic form that has a distinct genitive form.
by a single head. An example of the latter is (10c), in which a single negation

head assigned the GoN to two nodes tudentov and problemov.
In the recent theory, stemming from PESETSKY & TORREGOs (2001, 2004)
seminal works, case is understood as uninterpretable counterparts of the inter-
pretable features of functional heads. Thus, the following proposals have been
launched: (i) nominative case is uT on D (PESETSKY AND TORREGO 2001), (ii)
accusative is uninterpretable Inner Aspect (SVENONIUS 2001) or telic aspect
(RICHARDSON 2007), (iii) genitive is uQ on N/D (BAILYN 2004), etc.
From this perspective, BAILYN (2004) resolves the genitive case assignment
in terms of selection and probing: Neg head selects the VP with a [q] feature
which in turn selects a QP. The null head Q checks/probes the uninterpretable
[q] feature on the internal argument. According to this proposal, Slavic languag-
es differ with respect to the selectional properties of the Neg head: only if Neg
selects a VP with a [q] feature, then the underlying nominal can be assigned the
GoN (Russian/Polish vs. Serbian/Croatian).
This proposal, however, cannot be straightforwardly adopted for Slovene,
since there is a problem of the free Gen/Acc variation under the scope of nega-
tion. Adopting this theory, we would need to claim that Slovene Neg head may
but need not select [q], the latter resulting in Acc, the former in Gen. This ex-
plains the free Gen/Acc variation but it faces at least one insurmountable prob-
lem, namely, how to account for the combined cases of the local and the long-
distance GoN in which there is a Gen/Acc or Acc/Gen selection under one sen-
tential negation. In (13), for example, [q] needs to be selected if the matrix or
embedded object is case-marked for the Gen. But if [q] has been selected, it re-
mains open why it may but need not trigger off the Gen on the remaining object.
In contrast to one-feature-one-case proposals, MATUSHANSKY (2008, 2009)
claims that a DP can be assigned more than one case during derivation, the re-
sult of which varies cross-linguistically. The author (2009) argues that all for-
mal feature of a head are copied onto its complement [] and percolate down to
[] all leaves. As a result, a terminal node can be marked for a number of unin-
terpretable features, some of which may be spelled out at PF. Case is thus
viewed as a property of a domain rather than of an NP. The PF spell out of the
bundle of Case features is resolved by language specific vocabulary insertion
rules. To exemplify, in the standard TP>vP>VP hierarchy, every argument be-
low T is marked the nominative if a case-assigning vo is absent; when a case-
assigning vo is present, then every argument below it receives the Acc and can-
not receive the nominative. When negation is present, MATUSHANSKY (2009)
assumes that the hierarchical order of functional projections is
TP>vP>NegP>VP,9 and that the Gen is assigned by [q] feature on the Neg. The
__________
9 As pointed out by MATUSHANSKY (2009, f. 10) the relatively low position of the senten-
tial negation is an incorrect assumption, but still pursued in view of two facts: (a) the
correct generalisation has not so far been established, and (b) [her] purpose [t]here is
202 Gaper Ilc
PF spell-out of the feature bundle in the Neg domain is determined by vocabu-

lary insertion rules:
(19) Vocabulary insertion rules for case-marking under negation in Russian:
[Q] /_ [specific][V]
[Q] GENITIVE10
[acc] ACCUSATIVE
[nom] NOMINATIVE MATUSHANSKY (2009)
In view of (19), the feature [q] is realized as the Gen, except on semantically
specific underlying objects, where it is deleted, and consequently such object is
marked with the Acc. In the absence of [q], both objects and subjects receive the
Acc and the Nom respectively. For Slovene data presented in 2.1. this proposal
again seems inapplicable. In Slovene, there is a semantically unrestricted
Gen/Acc selection under the scope of negation, hence the proposal of the [q]
feature deletion as in Russian seems untenable. If the [q] remains undeleted in
the case of the Acc under the scope of negation, then what we are left to claim is
that [q] may be realized either as the Gen or the Acc, which is clearly a concep-
tually unwanted result.
A similar approach to multiple case assignment can be found in RICHARDS
(2007). Within this proposal, the Gen should be understood as an instance of
multiple case assignment in which a nominal first assigned the structural Acc
case is assigned the Gen. Consider the derivation of multiple case assignment in
(20). The DP is assigned, checked off and deleted the structural case by v.11 In
the next step, Neg assigns the Gen to the DP, and finally the DP undergoes PF
spell-out. The final result of the derivation is that the DP is not morphologically
marked for the structural case, which is deleted prior to PF spell-out, but for the
Gen.
(20) 3
Neg
3
v [structural]
3
DP
The author identifies three possible scenarios of multiple Case assignment, de-
pending on the type of the case that has been assigned first:
__________
mostly to demonstrate how multiple Case assignment can be handled under [the] pro-
posal.
10 Labels GENITIVE, ACCUSATIVE, etc., refer to actual lexical entries.
11 I use the functional category v for the probe that assigns the accusative. This probe could
also be Asp or any other proposed in the relevant literature.
(21) Multiple Case Assignment (RICHARDS 2007):

a) In all languages, if Case is assigned to a DP which already has structural
case, the structural case should vanish, replaced by new Case.
b) In some languages, a DP with Inherent case cannot be assigned another
case.
c) In other languages, a DP with Inherent case can be assigned another
case; the new case will be attached outside of the Inherent Case mor-
pheme.
As the GoN is concerned, (21a) claims that it is obligatory: structural case is al-
ways overridden. The question why the structural case needs to be overridden
(21a) yet the inherent case (21b,c) may but need not be deleted, however, re-
mains open in RICHARDSs (2007) proposal. Examining Slovene instances of
optionality (10)-(12), it could be argued for Slovene that the structural Acc case
may but need not be overridden by another structural case (here: the Gen). As
such Slovene could provide evidence that there is no asymmetry between the
structural and inherent case in terms of multiple case assignment as stated in
(21). However, such a proposal would have at least one flaw. (21b,c) assume
that in those languages in which a DP with the inherent case cannot be assigned
another case, it cannot be assigned in all instances; and similarly, in those lan-
guages in which it can be assigned another case, it is assigned always, leaving
no room for free variation.12
3.1 Proposal
Let us start building the proposal on the assumption that Case is understood as
uninterpretable features of interpretable features of functional categories in lines
of PESETSKY & TORREGO (2001, 2004). In addition, let us follow BAILYNs
(2004) claim that the feature make up of a lexical item contains two types of fea-
tures: (i) inherent features (including phonological, semantic and formal fea-
tures) and (ii) variant features (comprising of -features and features associated
with functional categories). From this perspective, the option between the Acc
and the Gen is to be understood in terms of variant features: if a nominal con-
tains a variant feature associated with [q], then is occurs in the Gen, if it contains
[telic], then it appears in the Acc. Therefore, all the derivation needs to establish
is a configuration in which the unvalued variant feature of the nominal can
Agree with the relevant feature/head. In case of the Slovene Acc/Gen variation
under the scope of negation, if the relevant nominal contains the variant feature
[uq], then it must be probed by the [q] node selected by NegP, resulting in the
__________
12 A similar example is also provided by RICHARDS (2007): [t]here are reports of lan-
guages in which accusative subjects of passives are possible, but for the examples I have
seen, accusative subjects and nominative subjects are both options, which makes the
situation somewhat less clear (Ukrainian).
204 Gaper Ilc
GoN. On the other hand, if the nominal bears the variant feature [utelic], then it
is probed by the [telic] node in AspP, resulting in the Acc.13
This proposal explains why the Acc is more frequent in the pre-negation and
non-adjacent positions (see (11)): the variant [telic] feature on the nominal re-
quires that the nominal be raised to a position in which it Agrees with the higher
lying AspP.
A similar claim can be put forward in the case of the long distance GoN in
control structures. Following MARUI (2003) that Slovene control non-finite
infinitival clauses are not CPs, and hence involve a single phase for both the
embedded as well as the matrix clause, it can be proposed that the [q] feature
percolates from the matrix Neg to the embedded clause. Thus, the underlying
object can get either its [uq] checked by matrix [q] selected by Neg, or its [utel-
ic] checked by the [telic] node in the embedded Asp:
(22) a. Nataa ni hotela itati knjigo/knjige.
Nataa not-is wanted read-INF book-ACC/book-GEN
b. matrix [q]

[telic]
Asp
NP/DP(knjigo[ACC]/knjige[GEN])
If this proposal proves to be correct, what remains open for further investigation
is the question why the movement of the internal argument out of the existential
closure in sense of BAILYN (1997) does not trigger off the non-existential read-
ing of the moved argument as is the case in Russian.
4. Conclusion
The paper has examined the properties of the GoN in Slovene against the back-
ground of recent minimalist case checking theories. Special attention has been
paid to the question of the optionality of the GoN. I argued that the Gen/Acc al-
ternation under the scope of negation can be best understood in terms of feature
__________
This proposal resembles in a way the analysis of the GoN in Russian by BROWN (1999),
who argues contrary to BAILYN (1997) that the Acc/Gen variation under the scope of ne-
gation is relatively free. In her view (op. cit., 74-75), case assignment properties are de-
termined in the lexicon, and the derivation creates case assignment domains which check
relevant case features on the nominals. According to Brown (op. cit.), the order of func-
tional projections is TP>AspP>NegP, where TP is the potential checking domain for
Nominative, Asp for Accusative and NegP for Genitive. The cases are assigned by the
s.c. feature complexes, Nom by [T, ], Acc by [Asp, +Pred, + VMAX] and Gen by [Neg,
+ VMAX].
selection and probing. In particular, the derivation of the GoN and the Acc under
the scope of negation is determined by the type of the variant feature on the no-
minal and the syntactic configuration in which the unvalued variant feature of
the nominal can be probed. The relevant variant features involved are [q] result-
ing in the Gen and [telic] resulting in the Acc.
References
ABELS, K. (2005) Expletive negation in Russian: A conspiracy theory. Journal of Slavic Linguis-
tics 13: 5-74.
BAILYN, J. F. (1997). Genitive of Negation is Obligatory. Proceedings of FASL 5. Michigan Slav-
ic Publications.
BAILYN, J. F. (2004) The case of Q. Proceedings of FASL 12. Michigan Slavic Publications.
BROWN, S. (1999) The Syntax of Negation in Russian. A Minimalist Approach. Stanford, CA:
CSLI Publications.
CHOMSKY, N. (1981) Lectures on Government and Binding. Dordrecht: Floris.
CHOMSKY, N. (1986) Knowledge of Lanugage. New York: Praeger.
CHOMSKY, N. (1993) A Minimalist Program for Lingustic Theory. In: K. HALE & S. J. KEYSER
(eds.) The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger.
Cambridge, MA: MIT Press.
FRANKS, S. & K. DZIWIREK. (1993) Negated Adjunct Phrases are Really Partitive. Journal of Slav-
ic Linguistics 1 (2): 208-305.
FRANKS, S. (1995) Parameters of Slavic Sintax. New York, Oxford: OUP.
HARTMANN, M. J. & N. MILIEVI (to appear) Case Alterations in Serbian Existentials. Proceed-
ings of FDSL-7.
MARUI, F. (2003) CP under control. In: ZYBATOW, G., L. SZUCSICH, U. JUNGHANNS AND R.
MEYER (eds.) Formal Description of Slavic Languages. Frankfurt am Main: Peter Lang.
408-422.
MATUSHANSKY, O. (2008) A Case Study of Predication. In: MARUI, F. & R. AUCER (eds.) Stu-
dies in Formal Slavic Linguistics. Contributions from Formal Description of Slavic Lan-
guages 6.5. Frankfurt am Main: Peter Lang. 213-239.
MATUSHANSKY, O. (2009) Some Cases of Russian. Proceedings of FDSL 7.5.
PESETSKY, D. & E. TORREGO (2001) T-to-C Movement: Causes and Consequences. In: M.
KENSTOVITCZ, (ed.) Ken Hale, a Life in Language. Cambridge, MA: MIT Press. 355-426.
PESETSKY, D. & E. TORREGO (2004) Tense, Case, and the Nature of Syntactic categories. In:
GURON, J. & J. LECARME, (eds.) The Syntax of Time. Cambridge, MA: MIT Press.
PRZEPIRKOWSKI, A. (2000) Long Distance Genitive of Negation in Polish. Journal of Slavic Lin-
guistics 8: 151-189.
RICHARDS, N. (2007) Lardil Case Stacking and the Structural/Inherent Case Distinction.
http://ling.auf.net/lingBuzz/000405.
RICHARDSON, K. (2007) Case and Aspect in Slavic. New York: OUP.
SCHOORLEMMER, M. (1995) Participial passive and aspect in Russian. Ph.D. diss. Utrecht.
SVENONIUS, P. (2001) Case and Event Structure. In: N. Zhang. (ed.) Syntax of Predication. ZAS
Papers in Lingusitics 26. Berlin: ZAS.
Gaper Ilc, University of Ljubljana, Faculty of Arts, Akereva 2, SI-1000

Ljubljana, Slovenia, gasper.ilc@ff.uni-lj.si
On development of antipassive function: what do
Australian and Slavonic languages have in common?
Katarzyna Janic
1. Introduction
This article deals with a type of a construction known in the literature under the
name antipassive. Defined as a syntactically derived intransitive construction in
which object argument loses its properties of a core argument, antipassive is of-
ten said to be a syntactic counterpart of the passive. While the passive is habi-
tually identified in accusative languages, the antipassive constitutes a main sub-
ject in the syntactic description of ergative languages. Recently, however, this
type of construction has also started to be recognized in languages of accusative
alignment (POLINSKY 2005). To draw the attention of scholars to the existence
of antipassives in accusative languages, this paper investigates Slavonic family,
in particular Polish.
The cross-linguistic studies reveal that in some ergative languages antipas-
sives are morphologically related to reflexive constructions. This means that
along the whole variation observed with respect to antipassive derivation, in cer-
tain instances antipassive verbs are derived by the affix that is normally used to
mark reflexive and/or reciprocal constructions. The cognate morphology of anti-
passive and reflexive constructions is particularly observed in Australian lan-
guages. What is, however, more important to observe is that the same morpho-
logical coincidence is also observed in some other genetically unrelated lan-
guages across various geographical areas, namely in Slavonic languages. This
raises an intriguing question on the possible relation of the respective construc-
tions and the extent to which one has developed to another. Building on the his-
tory of the reflexive based-marker in Indo-European languages, the Slavonic
languages of which are the example, it is a well-known fact that all the SE-forms
in Indo-European languages go back to one Indo-European SE-morpheme the
function of which was primarily restricted to the notion of reflexivity. In com-
parison to the extensive literature on the reflexive-based marker in Indo-
European languages, the sources of reflexive development in Australian lan-
guages have not been studied much. Thus, the possible explanations of the cog-
nate morphology of antipassives and reflexives are based mainly on hypothesis
and assumptions.
In comparison to the studies on the synchronic properties of antipassives and
reflexives and the cross-linguistic variations that they exhibit, a question on their
possible diachronic relation has received relatively little attention. Apart from a
few attempts (TSUNODA 1988, TERILL 1997 POLINSKY 2005), little is known
about the nature of their cognate morphology. As a result, many questions still
208 Katarzyna Janic
remain open. Chief among them is whether the shared morphology results from
a pure coincidence of two morphologically unrelated forms or whether it is the
same morphological form that depending on the lexical meaning of the verbal
stem and pragmatic factors persists to its polysemous status. Nevertheless, given
that the syncretism between reflexive and antipassive construction is not re-
stricted to a one language family but it is widely attested crosslinguistically, par-
ticularly in Slavonic and Australian family, it is then legitimate to claim that the
morphological overlap between two respective constructions is by no means ac-
cidental in nature but comes rather from the polysemous properties of the reflex-
ive marker. As a result, a question pertaining to the development of antipassives
from reflexivity remains open.
In this paper I will explore in detail the extent to which canonical reflexive
constructions could have possibly developed into antipassive in Slavonic lan-
guages. To outline the possible diachronic pathway, in section 2, first, I will
provide some theoretical background to the antipassive notion. The discussion
will focus on the structural properties indispensable to the formation of antipas-
sives in ergative and accusative languages. Then, in section 3 I will analyse the
syntactic and semantic properties of the antipassive marker. I will propose in
section 4 a possible diachronic scenario that will capture a possible development
pathway of antipassives from reflexivity in Slavonic languages. Some conclud-
ing remarks will be offered in section 5.
2. Theoretical background
2.1 Problem with definition
Given that antipassives constitute a heterogeneous class of constructions, it
seems to be rather difficult to offer a unified definition that restrictive in nature
would capture a different degree of variability in the form and function of anti-
passives. Thus, to investigate the antipassive phenomenon, two solutions are
provided. The linguists either proceed to a general structural definition of anti-
passives that far from being absolute encompasses a possible range of different
constructions (POLINSKY 2005) or they propose a definition with fairly strict cri-
teria according to which only a limited number of syntactic derivation are rec-
ognized as antipassives (DIXON 1994). Consequently, certain constructions in-
cluded under the term antipassive by one author, fall outside the scope of an-
tipassive investigation by some others.
Another problem related to the antipassive definition pertains to the theoret-
ical framework within which the phenomenon is subscribed. The analysis of a
given construction does not have to be exclusively limited to a syntactic descrip-
tion. In recent time, there has been some pressure of the functionally oriented
contributions to give the priority to the functional definition of the syntactic con-
struction. As documented by GIVN (2007: 19) there are some good reasons
why the purely structural definition of syntactic constructions is, in general, un-
On development of antipassive function: Australian and Slavonic languages 209
tenable. Rather, a typology of syntactic constructions requires a function a defi-

nition. Consequently, certain constructions may be considered to be antipassive
under the functional but not syntactic analysis. Even though a functional ap-
proach appears to be very tempting and useful, I would refrain from introducing
the antipassive construction, at least in the beginning, from a functional descrip-
tion and instead I would propose a loose structural definition that would encom-
pass a common major class of antipassives in ergative and accusative languages.
2.2 Antipassive: structural definition
Many linguists working on the syntactic description of ergative languages tend
to define antipassives as a type of construction that typical for ergative lan-
guages serves as a morphosyntactic alternation for the same transitive proposi-
tion, (COOREMAN 1993:50). Notice that the antipassive such defined does not
necessarily imply an obligatory markedness on the verb. This means that a sim-
ple reorganisation of the agreement pattern on the verb or a change in the case
marking may be viewed as a sole indication of the antipassivization process. Se-
condly, Cooremans characterisation of antipassive is restricted only to ergative
languages. Some linguists consider it to be inadequate in a sense that it does not
include the parallel constructions from accusative languages.
Contrary to COOREMAN (1993), some linguists i.e. POLINSKY (2005) do not re-
strain the antipassive analysis to ergative languages only. To avoid a clear-cut
dichotomy between antipassives of ergative languages and the corresponding
constructions from accusative languages, they propose a general definition that
argues in favour of the existence of the antipassive in both systems. Obviously,
such definition is less restrictive in nature and above of all it does not pertain to
any particular system. Nevertheless, the constructions from accusative languages
that were traditionally excluded from the antipassive investigation are now sub-
jects for debate. As such, POLINSKY (2005: 438) defines antipassives as a de-
rived detransitivized construction with a two-place predicate, related to a corre-
sponding transitive construction whose predicate is the same lexical item. In the
basic transitive construction, the patient-like argument is realized as a direct ob-
ject, in the antipassive constructions, that argument is either suppressed (left im-
plicit) or realized as an oblique complement. As such, antipassive is viewed as
a simple valency changing mechanism that regardless of the grammatical system
of a language affects the object core argument of a transitive clause. To investi-
gate antipassives in accusative languages, I will adopt Polinskys definition.
3. Property of antipassive marker
Another intriguing question pertains to the formal markedness of antipassive
constructions. A typological investigation on transitive/antipassive alternations
reveals in the first place that most antipassives are derived from the correspond-
ing transitive clause with the help of the overt morphology i.e. the antipassive
210 Katarzyna Janic
marker the function of which can be purely syntactic or syntactic and semantic
in nature. In the former case, the role of antipassive marker boils down to a
simply syntactic detransitivisation of a transitive construction; whereas in the
latter case, the antipassive marker in addition to syntactic function has some se-
mantic impact on the verb interpretation. Indeed, the syncretism of antipassive
markers with other categories i.e. aspect/modality is widely observed crosslin-
guistically.
In addition to the aspect/modality syncretism, the antipassive marker can al-
so be diachronically related to the notion of reflexivity. It is an extremely well-
known fact that languages with both reflexive and antipassive verbal morpholo-
gy use the same morpheme in both cases. This phenomenon is particularly ob-
served in Australian Pama-Nyungan family and Slavonic languages, wherein the
morphology of antipassives coincides systematically with the one of reflexives.
Examples (1) and (2) illustrate the antipassive and reflexive derivation from
Yidiny, a language spoken in the upper Herbert River region of north Queen-
sland in Australia in which the respective constructions are derived by the same
-: morpheme:
Yidiny (FOLEY & VAN VALIN 1984 :172)
(1) a. Yidu bua - mayi buga-
this-ERG woman-ERG vegetables.ABS eat-PRS
This woman is eating vegetables.
b. Yiu bua buga-:di-
this.ABS woman.ABS eat-AP-PRS
This woman is eating [something].
Yidiny (Foley & VAN VALIN 1984 :173)
(2) Wagu:da bambi-:di-nu
man.ABS cover-REFL-PST
The man has covered himself.
The analogous morphological correlation is attested in Slavonic languages, in
particular in Polish. Example (3b) and (4b) illustrate the antipassive and the ref-
lexive constructions respectively, both derived by the same morpheme si :
(3) a, Janek pcha dzieci na przerwach
Janek.NOM push.PRS.3SG children.ACC on breaks
Janek pushes children on school breaks.
b. Janek pcha si na przerwach

Janek.NOM push .PRS.3SG SI on breaks
Janek pushes [children] on school breaks.
(4) a. Matka czesze dziecko
mother.NOM comb.PRS.3SG child.ACC
The mother combs the child.
b. Matka czesze si
mother.NOM comb.PRS.3SG SI
The mother combs herself.
Like Yidiny, the Polish antipassive and reflexive construction are derived with
the help of the same morpheme that depending on the lexical meaning of the
verb either performs the reflexive or antipassive function. Note that in (3b) the
verbal valency is reduced by the si morpheme, due to which the object argu-
ment is omitted from the syntax of the clause.
A typological investigation on antipassive derivations shows that in fact
there are relatively few languages the antipassive marker of which restrains to a
purely syntactic function only. In most cases, the antipassive marker is multi-
functional in nature being either related to the category of aspect/modality or
reflexivity. In this paper, I will be particularly concerned with those antipassive
derivations the antipassive marker of which is syncretic with reflexivity.
4. Possible diachronic evolution
Having established the structural properties of antipassives, we can now proceed
to the chief question of our discussion, namely, how antipassive could have
possible developed from reflexivity. To determine the evolutionary pathway of
the antipassive from the source construction and its degree of grammaticaliza-
tion, it will be convenient in the first place, to establish the extent to which re-
spective constructions resemble and differ.
4.1 Semantic detransitivisation of antipassive
The semantic detransitivisation is often said to be one of the main functions of
antipassives. It indicates that the action expressed by the verb in certain ways,
less directly affects the Patient. Building on the THOMPSON & HOPPERs (1980)
parameters pertaining to the prototypical transitive event, antipassive events tend
to be semantically lower in transitivity for two main reasons. Firstly, the action
expressed by the verb is prototypically less transitive. Unlike the prototypical
transitive clause wherein the verb codes completed, real and cognitively salient
event (GIVN 2001), antipassive verbs tend to express ongoing, repetitive, un-
completed activities that are without a particular onset or extended over a period
of time. Of the multiple properties a decreased telicity or perfectivity is central
to the antipassive predicate.
212 Katarzyna Janic
Secondly, antipassive derivation affects the semantic properties of the pa-

tient argument. With respect to the prototypically transitive event, wherein the
patient is considered to be a highly individual, non-volitional, non-controlling
participant entirely affected by the Agent action, in the antipassive its the se-
mantic properties i.e. full affectedness and high individuation are called into
question.
Example (5) illustrates the instance of antipassive derivation from Polish,
the Patient of which highly correlates with a low degree of identifiability. Re-
moved from the syntax though not from the semantics, its referent denotes
loosely specified human collective children:
(5) a. Jak by+ ma+y, Marek bi+ dzieci
When was little Marek.NOM beat.up.PST.3SG child.PL.ACC
When [he] was little, Marek used to beat up children.
b. Jak by may, Marek si bi
when was little Marek.NOM SI beat.up.PST.3SG
When [he] was little, Marek used to beat up [other children].
Russian
In regard to the semantic nature of the subject argument, it retains in a certain
way its agentive properties. It means that it is still viewed as a volitional, con-
trolling participant who initiates the action. Depending on the example, howev-
er, it may display a different degree of agentivity. In (6), for instance, the subject
argument koshki cat presents a decreased agentive property i.e. it is not pre-
sented as being involved in a particular type of activity but rather as having cer-
tain predisposition to perform it. The clause is primarily understood as a state-
ment about the agent that attributes to him a certain property:
(6) a. Koshki tsarapaj-ut malchika
Cat.NOM.PL scratch.PRS.3PL boy.ACC.SG
The cats scratch the boy.
b. Koshki tsarapajut-sja
cat.NOM.PL scratch.PRS.3PL-SJA
The cats are in the habit of scratching [people].
The verb to scratch in (6b) is derived from the corresponding transitive clause
with the help of the reflexive based-marker -sja that in the given example per-
forms the antipassive function i.e. it refers to the indefinite group of individuals
people.
4.2 Syntax and semantics of reflexivity
To determine to which extent reflexives resemble antipassives, lets have a look
at their structural properties first. Example (7) illustrates a reflexive derivation in
Polish. The active clause in (7a) that expresses a transitive action involving two
distinct participants i.e. subject and object contrasts with the reflexive one in
(7b), wherein the respective arguments co-refer:
(7) a. Matka czesze Dziecko
Mother.NOM comb.PRS3SG child.ACC
The mother combs the child.
b. Matka czesze Si
mother.NOM comb.PRS.3SG SI
The mother combs herself.
Due to the coreference requirement imposed on the reflexivity, the object argu-
ment is removed from the syntax of the clause. Its omission is coded by the
presence of the si morpheme that in the given example performs the reflexive
function. As such the clause is objectless i.e. syntactically intransitive. Adopting
LANGACKERs terminology (1976), it is defined as a non-distinct argument that
is zeroed out in the syntax.
Note that similar to antipassives, the syntactic structure of reflexives also re-
quires a two-place argument predicate, thus potentially transitive. In both cases a
morphosyntactic derivation triggered by the si morpheme boils down to the
syntactic detransitivisation of the syntactically transitive construction the object
argument of which is removed from the clause. Thus, in terms of the argument,
structure the relevant constructions tend to affect the same argument i.e. the ob-
ject which for the convenience it is referred here as the patient. The only differ-
ence pertains to the functional ground that triggers object omission.
The syntactic parallelism between reflexive and antipassive clauses strongly cor-
relates with the semantic factors that rank the respective constructions lower on
the semantic transitivity scale. In terms of reflexives, they are said to be seman-
tically lower in transitivity for one main reason i.e. they deviate from the proto-
typical transitive event in terms of the number of participants involved in the ac-
tivity. A semantically transitive event requires two distinct arguments i.e. the
agent and the patient. In reflexives, however, the agent and the patient of the
event are coreferent. The coreference condition requires that the same semantic
argument acts upon itself. Consequently, there is only one participant involved
in the action.
In relation to the semantic role of the subject, it assigns simultaneously both
the agent and the patient role. Nevertheless, it keeps, in certain ways, its agen-
tive properties. Indeed, in (7b) the subject argument mother is still viewed as a
volitional and controlling participant who initiates the action; however, instead
of performing the activity on some distinct entity, she acts upon herself. Thus,
the coreference condition renders the subject argument less agentive. Note that
in comparison to the semantic properties of the subject argument of antipassives,
the ones of reflexives do not differ much. In both circumstances, the semantic
agentivity of the given argument is reduced. Unlike in reflexives, however, a de-
214 Katarzyna Janic
creased agentivity of the subject in antipassives is not triggered by the corefe-

rence strategy but by the object omission.
4.3 Nature of the verbs
In terms of the nature of the verbs that may intervene in reflexives and antipas-
sives, both types of constructions tend to impose some constraints, though anti-
passives are considerably more restrictive than reflexives. As observed by
GIVN (2001:96), the coreference requirement imposed on reflexivity causes
that only verbs that can take the same semantic participant type as both subject
and object can be reflexivized. As a result a number of verbs that are prototypi-
cally transitive cannot undergo a reflexivisation process i.e. to build, to cook,
to fold, etc. Unlike antipassives, reflexives are not necessarily based on the
prototypical transitive verbs. In addition to agentive verbs, they can be easily
formed from stative or non-agentive verbs.
With respect to antipassive verbs, as stated above, they tend to be much
more restrictive than reflexive ones. First of all, they are very limited in number
and meaning. My investigation of the Slavonic literature, in particular of Polish
contributions, has revealed a handful of potential verbs that are plausible to oc-
cur in antipassives. Only the verbs that express the antagonistic actions such as
kopa to kick, pcha to push, bi to beat/to fight, drapa to scratch, gry
to bite, plu to spit, chlapa to splash, szczypa to pinch, przezywa to
nickname or affectionate actions obejmowa to hug, caowa to kiss can in-
tervene in the antipassive derivation in Slavonic languages. Then, unlike reflex-
ives, antipassives tend to be derived from prototypical transitive verbs. Even if
antipassive verbs do not exhibit any transitive restrictions, note that they are still
strongly constrained in terms of their lexical-semantic properties.
4.4 Reflexive / antipassive polysemy
The analysis of reflexive and antipassive construction necessitates also a discus-
sion on the reflexive/antipassive polysemy of reflexive-based marker. Building
on NEDJALKOV (2007), we can distinguish two patterns of polysemy i.e. the one
in which the reflexive and antipassive meanings of a reflexive-based marker are
actualized on different verb stems, and the one in which the respective meanings
are expressed on the same derivative.
A study of the reflexive/antipassive polysemy of grammatical reflexive-
based marker in Slavonic languages reveals that it is more common to come
across with the circumstances the reflexive/antipassive polysemy is realised on
the same derivative rather than on different verbal stems. In other words, the
verbs in which reflexive and antipassive meanings coincide to a significant de-
gree are more frequently attested than the ones which accept the antipassive
reading only:
(8) Prosz pani, a on si drapie

Excuse me, Madam but 3SG.NOM SI scratch.PRS.3SG
Madam, he is scratching himself. reflexive reading
Madam, he is scratching [other children]. antipassive reading
In (8) the Polish derivative drapa si to scratch si accepts two readings i.e.
the reflexive and the antipassive one. As a result, a given derivative becomes
ambiguous vacillating between two meanings. Warrungu, an Australian lan-
guage, provides another example of reflexive/antipassive polysemy actualized
on the same derivative. In the example (9), a verb bangaa accompanied by the -
gali suffix is interpreted as a reflexive form, (9a), or as the antipassive one, (9b),
(TSUNODA 2006: 309):
(9) a. Bama bangga-gali-n jurba-nggu
Man.NOM paint-REFL-NONFUT white.ochre-ERG
The man is painting himself with white ochre.
b. Bama jurba-nggu bangga-gali-n
man.NOM white.ochre-ERG paint-AP-NONFUT
The man is painting [someone else] with white ochre.
Note that verbs that accept both the reflexive and antipassive readings differ to
the extent to which a respective meaning can be realized. Alongside the verbs
that equally accept both interpretations, there are some for which either the ref-
lexive or antipassive reading is more or less strongly preferred. Nevertheless, the
second interpretation is still plausible, though, in the very restricted context. In
(10) for example, the verb beat up accompanied by the reflexive-based marker
admits primarily the antipassive reading; however, in certain though pragmati-
cally awkward situation the reflexive reading is not entirely excluded:
(10) Jak by+ ma+y Jan si bi+
when was little Jan.NOM SI beat.up.PST.3SG
When [he] was little, Jan used to beat up [other children].
?When [he] was little, Jan used to beat up himself
Crosslinguistic studies reveal that when the polysemy of a given derivative is at
play, the meaning may be determined by different factors. To explain for in-
stance the reflexive/antipassive polysemy of a derivative in (8) or in (9), I will
follow NEDJALKOV (2007) proposing that the semantic overlap between the ref-
lexive and antipassive reading is primarily driven by two parameters i.e. the
pragmatic one and/or the lexical one. The former pertains to the related dis-
course context, whereas the latter to the lexical properties of the base verb. De-
pending on the given verb either both parameters intervene or only one of them.
For example, the lexical meaning of the verb drapa si to scratch si in (8)
favours both the anaphoric interpretation in which the reflexive-based marker
means itself and the antipassive one in which the same marker implies some-
216 Katarzyna Janic
body. This means that the Agent can equally perform the act of scratching him-
self or on some other participant and no special pragmatic context is required to
actualize one of the meanings. Thus, an equal acceptance of both interpretations
is a subject to a similar semantic condition. To disambiguate a derivative the
lexical meaning of which is neither permanently antipassive nor reflexive, a
broader context or special modifiers can be introduced to support one or the oth-
er interpretation.
It happens also that a given derivative with a polysemous marker may have
the reflexive or antipassive meaning as a permanent feature. Depending on the
language, this may refer to an individual derivative or a class of derivatives. In
Slavonic languages, in particular in Polish, a number of derivatives that accepts,
for example, the antipassive and not the reflexive reading, is limited to one in-
stance only. Example (11) illustrates a case of a derivative in which the reflexive
and antipassive meanings never overlap. Accompanied by the reflexive-based
morpheme, the verb always admits the antipassive reading i.e. to push [some-
body]:
(11) Nie Pchaj si bo spadne
NEG push.IMP.2SG SI because fall.down.FUT.1SG
Do not push [me], because I will fall down.
*Do not push [yourself], because I will fall down.
In regard to the verbs the reflexive-based marker of which exhibits the reflex-
ive/antipassive polysemy, it is important to note that in Slavonic languages most
of them are to a large extent ambiguous in nature. Even though their dominant
interpretation is antipassive, the reflexive one is also possible, though in a li-
mited context.
4.5 Givns functional approach1
To explain the antipassive evolution from reflexivity in Slavonic languages, I
will follow TERRILL (cf. 1997 for Australian languages) and I will adopt
GIVNs (2001:93) functional approach in which he claims that de-transitive
constructions () commonly arise via re-grammaticalization of some function-
ally-related construction. This means that the functional extension of the source
construction into a target one takes place under the condition that both construc-
tions share at the beginning some functional features. The functional similarity
or partial functional overlap is considered to be one of the motivating factors
triggering the diachronic extension of the syntactic construction (GIVN 2007).
Building on Givns functional approach, I propose that in Slavonic lan-
guages the reflexive i.e. a source construction evolved into the antipassive i.e.
the target one through the functional extension and that the function extension
__________
1 Note that the term functional approach employed by GIVN (2001) is polysemous and it
pertains to the discursive aspects of the linguistic analysis.
was primarily triggered by the functional similarities. This assumption is based

on the fact that both constructions share to a large extent some functional fea-
tures that pertain particularly to semantic and/or pragmatic domain. Thus, if I
manage to show that antipassives and reflexives in Slavonic languages function-
ally overlap to a large extent, then it will be legitimate to argue for a diachronic
link between those two and to claim that the antipassive developed from the ref-
lexivity through the functional extension. Note that determining the functional
overlap between respective constructions will help us also to estimate a degree
of grammaticalization of the given construction and to assess whether it evolved
into a canonical construction or whether it is still in the evolutionary process
characterised by functional ambiguity.
4.5.1 Functional comparison of reflexives and antipassives
With regard to the functional properties of antipassives and reflexives discussed
in section 4.1 and 4.2 respectively, it is clear that the constructions in question
share some common features in at least one of the relevant functional semantic
and/or pragmatic detransitive sub-domain.
The functional overlap between reflexives and antipassives pertains mainly
to the semantic parameters that temper with transitivity. More precisely, both
constructions are ranked lower on the semantic transitive scale because of the
decreased agentivity of the subject argument, the decreased telicity of the verb,
and the decreased affectedness and/or individuation of the patient argument. In
the reflexive construction the subject argument is less agentive because of the
coreference requirement imposed on the reflexivity. It is still considered to be a
volitional controlling participant who initiates the action; however, its reduced
agentivity is due to the fact that it does not transmit the action on some distinct
entity. In antipassives a decreased agentivity of the subject argument results
from the fact that Agent is often presented as not being engaged in a particular
type of activity but rather as having certain predisposition to perform it. In addi-
tion there is always a certain difficulty with which an effect stemming from an
activity by Agent on an identifiable object can be recognized, (COOREMAN
1993:51).
Then, in comparison to the corresponding transitive prototype, reflexives
and antipassives express the events that are semantically less transitive. As for
reflexives, their events are considered to be semantically lower on the transitivi-
ty scale due to the reduced number of participants involved in the activity. A
semantic definition of a prototypical transitive event requires two highly distinct
participants, whereas the coreference condition of reflexives imposes that the
same argument performs simultaneously the agent and patient role. In terms of
antipassive events, they often denote progressive, iterative, habitual actions that
are cognitively non-salient without a particular onset.
218 Katarzyna Janic
Finally, both constructions tend to have non-topical patient argument that in

reflexive constructions is considered to be non-distinct and it is left completely
unmarked on the syntactic ground. Like in reflexives, in antipassives the Patient
argument is also omitted, however, on different terms. In reflexives the argu-
ment at issue is removed from the syntax due to the coreference condition, whe-
reas in antipassives its omission is motivated on the pragmatic grounds. It is re-
moved from the clause because it is considered to be unimportant to the dis-
course context. Its referent is often ranked very low on the identifiability scale
denoting indefinite or generic individual or class of individuals.
The only difference that occurs between the respective constructions is ma-
nifested in terms of the motivation on the basis of which the object argument is
omitted. In both cases, however, the removed object argument is of low promi-
nence. It means that it is less distinct, less focused and pragmatically less promi-
nent.
4.5.2 Antipassive in early stage of grammaticalization
In this section I will propose a possible scenario of antipassive development in
Slavonic languages that is based particularly on Givns functional approach. In
addition, it will be claimed that contrary to Australian languages in which the
antipassive evolution from reflexivity is in its late stage of the grammaticaliza-
tion process, the diachronic change in Slavonic languages was only a two-step
process resulting in ambiguous patterns. It will be shown that unlike in Australi-
an languages, in Slavonic languages, in particular in Polish, reflexives did not
developed into a canonical structural antipassive of what may be interpreted to
be the target construction but they are still in the early stage of the grammatica-
lization process characterized by functional ambiguity.
Building on JOHNSON & POSTAL (1980), who claim that functional ambigui-
ty is the earliest step in diachronic change, we can imagine a possible pathway
of change of reflexives into antipassives in the following way. In the first stage
of the development, speakers used reflexive constructions to code a reflexive
event, wherein agent and patient were coreferent. Consequently, the action de-
noted by a verb was prototypically less transitive, the agent was less agentive
and that patient that was removed from the syntax due to the coreference strate-
gy was pragmatically less focussed, less salient and functioned as a non-distinct
argument. Such reflexive event was morphosyntactically coded by the object de-
letion and by the special morphology on the verb. Now being confronted with
the similar pragmatic situation, in which the action expressed by the verb was
also less prototypically transitive, the subject participant was prototypically less
agentive, and the patient was less focussed, less distinct but not coreferent with
the agent argument, the speakers may have employed the source construction to
code this new though pragmatically very similar situation. In other words, they
have extended the use of the old reflexive construction to the new context in
which the patient argument was not coreferent with the subject argument, but all
other functions of the situation remain unchanged. Like the reflexive event, the
antipassive one was morphosyntactically coded by the object omission and by
the si morphology on the verb. Thus, we pass from one construction the main
function of which was to express the reflexive event to another one that per-
forms now two functions i.e. the antipassive one that viewed as the extension of
the reflexive one. Note that this functional re-analysis i.e. the functional exten-
sion of old constructions to novel contexts took place instantaneously, and was
considered to be spontaneous activity employed by individual speakers during
the communication process (GIVN 2007).
Note that unlike the Australian languages, in the Slavonic ones the complete
separation of the reflexive and antipassive functions has not taken place yet. The
Slavonic languages still use the same syntactic construction to express either the
reflexive or antipassive meaning. In Australian languages, the functional exten-
sion of the given constructions was additionally followed by the structural ad-
justment that gave rise to the iconic coding of the new and older function now
viewed as two distinct constructions. The structural reanalysis is considered to
be the last stage of the grammaticalization processes (GIVN 2007). This means
that the diachronic process of reflexive extension into the antipassive in Austral-
ian languages has been entirely completed resulting in the target canonical con-
struction that differs from the source construction in terms of structural proper-
ties. Given that the Slavonic antipassive verbs illustrate the possibility of ambi-
guity i.e. depending on the lexical meaning of the verb, either both interpreta-
tions are equally acceptable with the possible dominance of the reflexive read-
ing, or the antipassive reading is more or less strongly preferred and that the
structural re-analysis of the given construction has not taken place yet, we can
conclude that in Slavonic languages antipassives did not evolve in the canonical
target construction and that they are still in the early stage of the grammaticali-
zation process.
5. Concluding remarks
The aim of this paper was to show that in Slavonic languages antipassives de-
veloped diachronically from reflexivity through the functional extension and
that the functional extension was primarily driven by the functional similarities.
In relation to this issue, we have seen that antipassives originating in Slavonic
languages from reflexive constructions did indeed undergo a grammaticalization
process, though, the distance the antipassive traversed from the postulated
source construction to what may be interpreted to be a target construction has
not been yet completely covered. As a result, the same syntactic structure per-
forms two similar but not identical functions: the old and the new one. This is
due to the fact that in Slavonic languages antipassives are still in the early stage
of grammaticalization.
220 Katarzyna Janic
References
COOREMAN, A. (1993) A Functional Typology of Antipassive. In: FOX, B. & HOPPER, P.J.
(eds.) Voice: Form and Function 49-87. Amsterdam: John Benjamins Publishing
Company.
DIXON, R. M. W. (1994) Ergativity. Cambridge: Cambridge University Press.
FOLEY, W. A. & R. D. VAN VALIN (1984) Functional Syntax and Universal Grammar. Cam-
bridge : Cambridge University Press.
GIVN, T. (2001) Syntax: An Introduction. II. Amsterdam/Philadelphia: John Benjamins Pub-
lishing Company.
GIVN, T. (2007) On the relational properties of passive clauses: A diachronic perspective. In:
FERNNDEZ, Z. E., S. WICHMANN, C. CHAMOREAU & A. A. GONZLEZ (eds.) Studies in
voice and transitivity (Estudios de vos y transitividas). Mnchen: Lincom Europa.
HOPPER, P. & S. A. THOMPSON (1980) Transitivity in grammar discourse 251-299. Language
56.
JOHNSON, D. E. & P. M. POSTAL (1980) Arc pair grammar. Princeton: Princeton University
Press.
LANGACKER, R. (1976) Non-Distinct Arguments in Uto-Aztecan. Berkeley: UC Press.
NEDJALKOV, V. P. (2007) Polysemy of reciprocal markers. In: NEDJALKOV, V. P. (ed.) Recip-
rocal Constructions 231-334. Amsterdam/Philadelphia: John Benjamins Publishing
Company.
POLINSKY, M. (2005) Antipassive Constructions. In: HASPELMATH M., M. S. DRYER, D. GIL &
B. COMRIE (eds.) The World Atlas Of Language Structure 438-439. Oxford University
Press.
TERRILL, A. (1997) The Development of Antipassive Constructions in Australian Languages.
In: Australian Journal of Linguistics 71-88. Australia.
TSUNODA, T. (2006) Reflexive and middle constructions of Warrungu (Australia). In: TSUNO-
DA, T. & T. KAGEYAMA (eds.) Voice and Grammatical Relations 299-333. Amster-
dam/Philadelphia: John Benjamins Publishing Company.
TSUNODA, T. (1988) Antipassives in Warrungu and other Australian languages. In: SHIBATA-
NI, M. (ed.) Passive and Voice 595-649. Amsterdam/Philadelphia: John Benjamins
Publishing Company.
Abbreviations
abs Absolutive acc. Accusative ap. Antipassive
erg. Ergative imp. Imperative fut. Future
neg. Negation nom. Nominative nonfut. Non-future
pl. Plural prs. Present pst. Past
sg. Singular refl. Reflexive
Katarzyna Janic, University of Lyon, Laboratoire Dynamique Du Langage, Ins-

titut des Sciences de l'Homme, 14 Avenue Berthelot, 69363 Lyon Cedex 07,
France, katarzyna.janic@univ-lyon2.fr
Dislocated Direct Objects in Macedonian
Slavica Kochovska
1. Introduction
The main goal of this paper is to address a peculiarity relating to the dislocation
of direct object DPs in Macedonian and their co-occurrence with clitic pronouns.
Consider the data in (1a-c).
(1) a. Sekoja knigai, Ana *(jai) proita.
every book Ana it(f.sg.) read
Every book, Ana read it.
b. Mnogu knigi, Ana (*gi) proita.
many books Ana them read
Many books, Ana read.
c. Dve knigii, Ana (gii) proita.
two books Ana them read
Two books, Ana read.
In (1a), the dislocated universal obligatorily co-occurs with a clitic. In contrast,
the clitic with many is not allowed, as shown in (1b). Finally, the clitic appears
to be optional with the dislocated numeral in (1c). The question is what
determines the distribution of the clitic in such cases? Before we answer it,
consider also the patterning of these DPs when in positions in IP, as in (2a-c).1
(2) a. Ana *(jai) proita sekoja kniga.
Ana it(f.sg.) read every book
Ana read every book.
b. Ana (*gi) proita mnogu knigi.
Ana them read many books
Ana read many books.
c. Ana (*gii) proita dve knigi
Ana them read two books
Ana read two books.
As can be seen from (2a-b), the distribution of the clitics with the universal and
many remains the same. The distribution with the numerals, however, is diffe-
rent. The numeral in (2c) now pattern with many in that the clitic is no longer
__________
1 The same distribution holds when the direct objects are in preverbal position, as in (i):
(i) Ana sekoja kniga/mnogu knigi/dve knigi *(jai)/ (*gi)/ (*gi) proita.
Ana every book/many books/two books it(f.sg.)/them/them read
Ana read every book/many books/two books.
222 Slavica Kochovska
allowed. Thus, we see that the behavior of the numeral with respect to the clitic
changes depending on whether it occurs within IP or outside IP (cf. (1c) vs.
(2c)). The same, however, does not hold for the universals and many.
The paper offers an analysis for the facts in (1)-(2). The explanation it puts
forth relies on the idea that the distribution of the clitics is determined by the
features for strength of the DP with which it co-occurs, where strength is
understood to correlate with the semantic properties of the DPs.
The paper is organized as follows. In section 2, I outline the main proposal
regarding the feature specification of the DPs in Macedonian. In section 3, I
address the issue of the derivation of the constructions with dislocated DPs. Sec-
tion 4 is the conclusion.
2. The feature specification of Macedonian DPs
Given the distribution of the clitic in (1)-(2), it is natural to assume that its pres-
ence/absence relates to the strength of the DP (in the sense of BARWISE & COOP-
2
ER (1981)). In fact, looking at the data in (2a-c), we see a clear correspondence
between the strength of the DP and the presence of the clitic: the clitic is present
when the direct object is a strong DP (see (2a)), but absent when it is a weak DP
(see (2b-c)). Given this, I claim that the distribution of the clitic relates to the
strength of the DPs with which it co-occurs. More specifically, I propose that the
distribution is determined by the DPs feature specification for strength (see also
KOCHOVSKA 2010, to appear). The underlying assumption here is that strong
DPs are characterized by a [+strong] feature, which gets checked in the course
of derivation. This means that the clitic is a feature marker (following
SPORTICHE 1998) and that the doubling constructions in Macedonian arise as a
result of a feature checking operation. In other words, I propose that the clitic is
a licensor of a [+strong] DP; whereby the licensing is carried out through a spec-
head relation (CHOMSKY 1995).
Following SPORTICHE (1998), ANAGNOSTOPOULOU (1999, 2006) and
others, I assume that the doubled DP is the argument of the verb, and that the
clitic is base-generated in its surface position. Following SPORTICHE (1998), I
take clitics to be heads of their own functional projections within IP. The clitic
licenses a particular property in the DP it doubles, which in the case of
Macedonian is [+strong].3 The licensing of the feature must be carried out in an
__________
2 FRANKS & KING (2000) correlate the presence of the clitic in Macedonian with the speci-
ficity of the DP. For arguments that the clitic corresponds to the strength rather than
specificity or referentiality of the DP, see KOCHOVSKA (2010, to appear).
3 See also KALLULLI (1999, 2008), ALEXIADOU & ANAGNOSTOPOULOU (1997), ANAGNOS-
TOPOULOU (1999) for analyses of the clitic as head of ClP. Kallulli argues that the clitic
in Greek and Albanian licenses a [-focus] feature. ALEXIADOU & ANAGNOSTOPOULOU
(1997) and ANAGNOSTOPOULOU (1999) maintain that the clitic is as an agreement mar-
ker, following SUER (1988). SPORTICHE (1998) takes the view that clitics license
specificity in the doubled DP.
Dislocated Direct Objects in Macedonian 223
appropriate agreement relationship, which is achieved in a spec-head configura-

tion. This entails movement of the doubled DP (by LF) to the specifier position
of the clitic phrase. The structure of the clitic phrase (and its relative location
within the clause in Macedonian) is given in (3).
(3) ClP
2
Cl
2
0
Cl VP
2
DP
Following Sportiche, I take the licensing of the feature by the clitic, i.e. the
movement of the DP to SpecClP, to be governed by the Clitic Criterion in (4).
(4) Clitic Criterion (SPORTICHE 1998:267)
a. A clitic must be in a spec-head relationship with a [+F] XP at LF.
b. A [+F] XP must be in a spec-head relationship with a clitic at LF.
This means that when the clitic is related to a particular DP, the DP itself has to
move to the specifier position of the ClP, in order to satisfy the Clitic Criterion.
With this in mind, let us look at the data in (2a-c). The simple strong/weak
distinction in the feature specification of the DPs we have just posited works
well in this case. If we assume that a semantically strong DP is marked as
[+strong] but a semantically weak DP is marked as [-strong], the distribution of
the clitic follows straightforwardly. The clitic, being the licensor of the [+strong]
feature, must be present when a strong DP enters the derivation. The clitic will
predictably be absent when a weak DP is introduced. Thus, we see that the Clitic
Criterion, combined with the proposed feature specification of the DPs, can
successfully account for the distribution of the clitic with the DPs in (2a-c). For
illustration, the derivation of a clitic-doubled strong DP is given in (5).
(5) a. Ana ja proita sekoja kniga.
Ana it(f.sg.) read every book
b. [IP [DP Ana] [ClP jai + proitaj [VP tj [DP sekoja knigai ]]]]
Given our proposal, a [+strong] object DP without a clitic, would be ungram-
matical because a lack of clitic means that the relevant feature of the DP cannot
be checked. As a result, the derivation cannot converge, as shown in (6).
(6) a. *Ana proita sekoja kniga.
Ana read every book

b. *[IP [DP Ana] [VP proita [DP sekoja knigai ]]]]
A weak DP does not trigger the clitic because no features are being checked in
this case. Due to principles of economy, the ClP would not be projected, as
shown in (7).
(7) a. Ana proita dve knigi.
Ana read two books
Ana read two books.
b. [IP [DP Ivan] [VP proita [DP dve knigii ]]]
A weak DP cannot license the clitic because the conditions needed for licensing
the [+strong] feature do not obtain. In other words, the DP cannot enter into a
spec-head relation with the clitic, as required by the Clitic Criterion. Hence, (8)
is ungrammatical.
(8) a. *Ana gi proita dve knigi.
Ana themread two books
Ana read two books.
b. *[IP [DP Ivan] [ClP gii + proitaj [VP tj [DP dve knigii ]]]]
We have just seen that the doubling structures in (2a-c) can be regulated by the
Clitic Criterion. A simple extension of the proposal to the dislocated structures
in (1a-c), however, proves problematic. Although the obligatoriness of the clitic
with the universal in (1a) and its absence with many in (1b) still follows from
the Clitic Criterion, the optionality of the clitic with the numeral in (1c) remains
unexplained. In fact, given its feature specification as [-strong], we would
predict that the clitic should not be allowed with the dislocated numerals. This is
clearly not the case.
There could potentially be different reasons for the patterning of the numer-
als, but if we maintain that the clitic in Macedonian corresponds to the feature
specification of the DP, one obvious reason for the unexpected distribution pat-
tern in (1c), would be the specification of the numerals as [-strong]. Assuming
that this is the determining factor (see KOCHOVSKA 2010), following a suggesti-
on by VENEETA DAYAL (p.c.), I would like to propose a slightly different feature
specification of the DPs in Macedonian to the one above. In particular, I would
like to claim that the weak DPs, as far as their features for strength are
concerned, split into two classes: [-strong] and unspecified for strength. This
gives rise to a three-way feature specification of the DPs in the language as
[+strong], [-strong], and unspecified for strength. On this view, universals come
out as [+strong], weak DPs headed with many are [-strong], and weak DPs like
the numerals are unspecified for strength. As I show below, this characterization
of the DPs accounts for the apparent optionality of the clitic with the dislocated
numerals in (1c), as well as the facts in (2a-c).
Before we look at the dislocated structures in (1a-c), I outline the syntactic

assumptions on which the analysis of these constructions is based. First, I take
that, in principle, direct objects can occur at the left periphery either by move-
ment or by base-generation. At present, the data in (1a-c) is consistent with
either of these possibilities. In section 3.1, I give arguments that decide between
the two options (see also KOCHOVSKA 2010, to appear).
Next, I adopt CINQUEs (1990) view that base-generated constructions in-
stantiate binding chains (where the dislocated element enters into a binding rela-
tion with its coindexed clitic and pro in object position) and that constructions
derived by movement instantiate government chains, which as Cinque argues,
following RIZZI (1990), are defined in relation to the absence vs. presence of
wh-move to their clause initial position, they leave a trace in the argument posi-
tion in IP. Thus, the derivation of dislocated strong and weak DPs would be as
in (9a) and (9b), respectively.
(9) a. TopP
ei
DP TopP
6 3
sekoja knigai Top CP
every book 3
IP
3
DP ClP
5 3
Ana ti Cl
Ana 3
Cl VP
g 6
jai proita ti
it read
b. TopP
ei
DP TopP
6 3
dve knigii Top CP
two books 3
IP
ei
DP VP
5 6
Ana proita ti
Ana read
Under a movement analysis, the difference between the constructions in (9a) and
(9b) comes out as a result of the Clitic Criterion. In the case of strong DPs, as in
(9a), the ClP is projected, enabling the DP to check its [+strong] feature in the
specifier position, as it moves to the left periphery. In the case of weak DPs, as
in (9b), the lack of [+strong] feature means that the ClP is absent. (As mentioned
previously, I assume that the ClP in such cases is not projected due to general
principles of economy.) Note that this holds for both [-strong] and unspecified
for strength weak DPs.
I now turn to the second option, that of base-generation. A base-generated
DP in clause initial position would be connected to a null pro in the argument
position in IP. The dislocated DP in such case would enter into a chain relation
with pro. The derivation of a construction involving a dislocated strong DP will
be as in (10).
(10) TopP
wo
DP TopP
6 3
sekoja knigai Top CP
every book 3
IP
3
DP ClP
5 3
Ana proi Cl
Ana 3
Cl VP
g 6
jai proita ti
it read
The presence of the clitic in (10) is explained as follows. The null pro, being a
pronoun, is [+strong] (following BARWISE & Coopers (1981) treatment of
pronouns; see KOCHOVSKA 2010). Given that the [+strong] feature must be
licensed for the derivation to converge, the clitic must be present. Thus, the ex-
planation here is consistent with the analysis of clitic-doubled [+strong] DP in
(2a).
The same general principles apply when a weak DP is base-generated in the
left periphery. In this case, too, the dislocated DP connects to pro, forming a
binding chain. But there are differences and these stem from the feature
specification of the weak DP. Suppose that a weak DPs is marked as [-strong]
(as in (1b)).4 The configuration that involves a [-strong] DP, and a [+strong] pro
(and clitic) would be ungrammatical because the features of the weak DP clash-
es with the [+strong] feature of pro (as well as that of the clitic). I take this to
follow from the requirement that syntactic chains share the same feature
specification (see SUER 1988; see also CINQUE 1990, BAKER 1996). As pointed
out by SUER (1988:394) (and references cited therein), a chain is well-formed
only when there is no clash in features between its elements. Suer explains that
__________
4 The assumption would in no way alter our explanation of their derivation by movement.
The ClP will not be projected with an unspecified DP or a [-strong] one because its pre-
sence is only triggered by a [+strong] feature.
a clitic may form a chain with a constituent only if it fulfills this requirement.5 In
our case, we can extend this to include the dislocated DP as well. The
ungrammaticality of constructions involving a left-dislocated, [-strong] weak DP
and a clitic in the IP thus follows from general conditions on chains. The deriva-
tion of constructions involving a left-dislocated many as it co-occurs with a clitic
is given in (11) below.
(11) * TopP
qo
DP TopP
6 3
mnogu knigii Top CP
many books 3
IP
ei
DP ClP
5 3
Ana proi Cl
Ana 3
Cl VP
g 6
jai proita ti
it read
Thus, a strong DP is acceptable in a binding chain because its features are com-
patible to those of pro in the argument position. A [-strong] weak DP in a bin-
ding chain yields an ungrammatical sentence because its features clash with the
features of pro. Now, let us look at the unspecified for strength weak DPs.
In this case, too, the clitic is generated because of the [+strong] pro, thus sa-
tisfying the Clitic Criterion. Remember that a [-strong] weak DP is not to be
allowed to enter into a binding relation with pro because of clash of features in
the chain it creates. A weak DP unspecified for strength, on the other hand, can
enter into a binding chain because there is no clash with the features of pro.
Because chains have to share features, elements participating in them will either
have the features inherently (as in the case of strong DPs) or they will be able to
acquire them from the chain (as in the case of numerals, for example). Crucially,
__________
5 This condition is fulfilled in all cases of doubled DPs in Macedonian: both the clitic and
the doubled element are marked as [+strong]; see KOCHOVSKA (2010) for details.
though, the numerals are not [-strong]. They either remain unspecified or change
to [+strong] through the binding relation with pro. The derivation of left-
dislocated numeral, coindexed with a clitic, is given in (12) below.
(12) TopP
qo
DP TopP
6 3
dve knigii Top CP
two books 3
IP
ei
DP ClP
5 3
Ana proi Cl
Ana 3
Cl VP
g 6
jai proita ti
it read
To summarize, by positing a three-way distinction among the DPs in terms of
feature specification, we are able to account for the full set of data in (1)-(2). By
invoking the general condition that elements in a chain must not have clashing
features, we were able to explain the divergent behavior of the two types of
weak DPs. Crucially, the proposed analysis leaves it open that an unspecified
weak DP can co-occur with a clitic when left-dislocated. In the next section I
show that both these options are available and argue that the constructions
without a clitic are derived by movement, while constructions with a clitic are
base-generated.
3. Deriving the DPs at the left periphery
I maintained in the preceding section that left-dislocated direct objects can be
generated in the CP domain either by movement or by base-generation. In this
section, I show that Macedonian uses both options. This, in turn, gives rise to
two types of dislocated constructions in the language. The strongest evidence for
the existence of two left-dislocated constructions comes from the weak DPs
unspecified for strength.
Recall that the derivation of dislocated DPs could, in principle, rely on

movement or base-generation. While strong DPs, for example, are compatible
with both, the unspecified for strength weak DPs, like the numerals, provide
evidence that separates the two options. In particular, it shows that the clitic-
doubled DPs and the non-clitic-doubled ones are derived in two different ways.
The evidence I consider here is the sensitivity of the two types of phrases to
syntactic islands.6
3.1 Syntactic islands
I have maintained that DPs co-indexed with clitics enter into a binding chain
while DPs not co-indexed with clitics enter into a government chain, following
CINQUE (1990).7 As is well-known, both types of chains are sensitive to strong
islands, but only government chains are sensitive to weak islands. The same
holds for Macedonian, as illustrated by the examples in (13a-b).
(13) a. Sekoj student, se prauvam kako kje gonajde.
every students refl. wonder-I how will him find
Every student, I wonder how you will find him.
b. *Nekogo, se prauvam kako kje najde.
someone refl. wonder-I how will find
Someone, I wonder how you will find.
The ungrammaticality of (13b) can be explained if we assume that DPs without
clitics in Macedonian can only connect to the trace in object position via a chain
of antecedent government.8 Since the dislocation of the universal in (13a) does
not result in a weak island violation, we can conclude that the relation between
the clause initial element and pro in object position is one of a binding chain.
The weak island violation in (13b), on the other hand, shows that a fronted ele-
ment and the trace it leaves in object position are part of a government chain. It
should be noted that (13b), which has a DP unspecified for strength, can occur
with a clitic in which case the grammaticality is repaired.
__________
6 Additional evidence comes from their behavior with respect to Weak crossover effects;
see KOCHOVSKA (2010, to appear).
7 CINQUE (1990) maintains that the properties of the chains relate to the (non)referential
properties of the elements that participate in them. Binding chains are formed by
referential phrases only because they are able to connect to the gap in argument position
through a binding relation (the latter is achieved through coindexation of the EC with
their antecedent). Nonreferential phrases, Cinque maintains, identify the EC in object po-
sition as a variable, and connect to it only through a government chain. Referential
phrases for CINQUE (1990) are those which are D-linked, in the sense of PESETSKY
(1987).
8 For evidence that the trace in such cases is a variable was proved by the presence of
WCO effects in such constructions, see KOCHOVSKA (2010).
The same pattern emerges in (14a-b). A chain created by a clitic-doubled

numeral as in (14a) does not create a weak island violation, but a chain relation
between a fronted numeral not coindexed with a clitic, as in (14b), results in a
weak island violation.
(14) a. Dve knigi, se prauvam kako kje gi najde.
two books refl. wonder-I how will themfind
Two books, I wonder how you will find them.
b. ??/* Dve knigi, se prauvam kako kje najde.
two books refl. wonder-I how will find
Two books, I wonder how you will find them.
On the proposed analysis, [-strong] DPs are predicted to form government
chains only. The prediction is borne out as shown by the ungrammaticality of
(15), where the fronting of the DP headed by many results in a weak island vio-
lation.
(15) *Mnogu knigi, se prauvam kako kje najde.
many books refl. wonder-I how will find
Many books, I wonder how you will find them.
In contrast to this, both binding chains and government chains obey strong isl-
ands, as is shown in (16a-b)9:
(16) a. *Site maki, plaev zatoa to Petar gi istepa.
all cats cried-I because that Petar them beat
All cats, I cried because Petar beat them.
b. *Dve maki, plaev zatoa to Petar gi istepa.
two cats cried-I because that Petar thembeat
Two cats, I cried because Petar beat them.
In (16a-b) the adjunct modifier clause creates a strong island. The
ungrammaticality of (16a-b) is then due to the fact that the dislocated phrases,
strong as in (16a) or weak as in (16b), cannot enter into a chain with the pronoun
inside the adjunct clause. The same effects are found in relative clauses. Here,
too, the binding relation between the dislocated DP and the pronoun inside the
relative clause is disrupted. This explains the ungrammaticality of (17a) for
strong DPs and (17b) for weak DPs.
(17) a. *Konicava, gopoznavam ovekot to ja isplete.
basket-this him know-I man-the that it(f.sg.) wove
The basket, I know the man who wove it.
b. *Dve konici, gopoznavam ovekot to gi isplete.
two baskets him know-I man-the that themwove
__________
9 The examples in (16a-b) and (17a-b) are fashioned after BAKER (1996:104).
Two baskets, I know the man who wove them.

The examples in (13a), (14a), (16)-(17) show that the relation between a clitic-
doubled phrase at the left periphery and a null pro in object position is sensitive
to the presence of strong islands, but not weak ones. Aside from the fact that
such examples show that binding and government chains pattern differently with
respect to syntactic islands (cf. (13a), (14a) vs. (13b), (14b), (15)), such
examples are in fact, as BAKER (1996) points out, evidence that a clitic-doubled
dislocated phrase and pro are part of a chain to begin with.
To summarize, the behavior of clause initial DPs with respect to syntactic
islands shows that constructions with non-doubled direct objects are derived by
movement of a DP from a position within IP to the left periphery, while con-
structions with doubled direct objects instantiate binding chains between a base-
generated DP at the left periphery and a pro inside IP.
4. Conclusion
The paper argued that the distribution of the clitics with direct objects in Mace-
donian is governed by the feature specification for strength in the DPs with
which they co-occur, where strength is understood to correlate with the semantic
properties of the DPs as defined in BARWISE & COOPER (1981). The main
proposal is that the clitics that double direct object DPs are licensors of a
[+strong] feature of such DPs. As such, the doubling of the DPs in Macedonain
is regulated by the Clitic Criterion (SPORTICHE 1998). The analysis incorporates
a novel, three-way distinction between the DPs in Macedonian, with strong DPs
being marked as [+strong] and weak DPs being either [-strong] or unspecified
for strength.
The paper further shows that dislocated DPs can be derived either by
movement or by base-generation. Evidence for the claim comes from island ef-
fects which show that clause-initial DPs coindexed with a clitic are base-
generated in their surface position, while those that do not co-occur with clitics
are derived by movement from positions within the IP.
Acknowledgments
This paper is based on my doctoral dissertation. I would like to thank Veneeta
Dayal, Mark Baker, Roger Schwarzschild and Barbara Partee for their insightful
comments and generous feedback. Some of the issues have also been discussed
at the 2nd Budapest Generative Syntax Workshop. I thank the participants there
for useful comments and discussions.
References
ALEXIADOU, A. & E. ANAGNOSTOPOULOU (1997) Toward a Uniform Account of Scrambling
and Clitic Doubling. In: ABRAHAM, W. & E. VAN GELDEREN (eds.) German: Syntactic
Problems Problematic Syntax 142-161. Tbingen: Max Niemeyer.
ANAGNOSTOPOULOU, E. (1999) Conditions on Clitic Doubling in Greek. Clitics in the
Languages in Europe 3: 762-798.
ANAGNOSTOPOULOU, E. (2006) Clitic Doubling. In: EVERAERT, M. & H. VAN RIEMSDIJK (eds.)
The Blackwell Companion to Syntax. 519-581. Oxford: Blackwell.
BAKER, M. (1996) The Polysynthesis Parameter. Oxford/New York: Oxford University Press.
BARWISE, J. & R. COOPER (1981) Generalized Quantifiers in Natural Languages. Linguistics
and Philosophy 4: 159-219.
CHOMSKY, N. (1995) The Minimalist Program. Cambridge: MIT Press.
CINQUE, G. (1990) Types of -Dependencies. Cambridge: MIT Press.
FRANKS, S. & T. H. KING (2000) Clitics in Slavic. Oxford: Oxford University Press.
KALLULLI, D. (1999) The Comparative Syntax of Albanian: On the Contribution of Syntactic
Types to Propositional Interpretation. Ph.D. thesis, University of Durham.
KALLULLI, D. (2008) Clitic Doubling, Agreement and Information Structure: The Case of
Albanian. In: KALLULLI, D. & L. TASMOWSKI (eds.) Clitic Doubling in the Balkan
Languages 227-256. Amsterdam: John Benjamins.
KOCHOVSKA, S. (2010) Macedonian direct objects, clitics and the left periphery. Ph.D. thesis,
Rutgers University.
KOCHOVSKA, S. (to appear) Clitics and Direct Objects in Macedonian.
PESETSKY, D. (1987) Wh-in-Situ: Movement and Unselective Binding. In: Reuland, E. & A.
ter Meulen (eds.), The Representation of (In)definiteness 98-129. Cambridge: MIT
Press.
RIZZI, L. (1990) Relativized Minimality. Cambridge: MIT Press.
RIZZI, L. (1997) The Fine Structure of the Left Periphery. In: L. HAEGEMAN (ed.) Elements of
Grammar: Handbook in Generative Syntax 281-337. Dordrecht: Kluwer.
SPORTICHE, D. (1998) Partitions and Atoms of Clause Structure: Subjects, Agreement, Case
and Clitics. London/New York: Routledge.
SUER, M. (1988) The Role of Agreement in Clitic-Doubled Constructions. Natural Langua-
ge and Linguistic Theory 6: 391-434.
Slavica Kochovska, University of Nova Gorica, Institute for Cultural Studies,

Language and Cognitive Science, Vipavska 13, SI-5000 Nova Gorica, Slovenia,
slavica.kochovska@p-ng.si
Causatives and Anti-Causatives, Unaccusatives and
Unergatives: Or how big is the contribution of the lexicon to
syntax?
Peter Kosta
1. Introduction and overview over the types of causative construc-

tion (CC) in various languages of the world
During the last 40 years research of causativity belonged to the central themes of
the general and comparative resp. typological linguistics. In this respect it is as-
tonishing, that in my opinion from the Slavic side this subject was treated if at
all very marginally in the past. My interest was motivated by the fact that CC
require an analysis, which touches an interface of morphology, semantics, lex-
icon and syntax. Therefore it is also replicable by the Minimalistic Program
(with the inclusion of Distributive Morphology). Furthermore, the theme com-
prises important observations concerning questions of language change and
grammaticalization, which motivated me to choose it for the present volume.
In this article the features of externally and internally caused verbs will be de-
scribed and put into context of the phenomenon of unaccusativity. My approach
will be the following: I will try to characterize the relationship between lexicon
and syntax, namely including the concepts of Distributive Morphology and of
ROOT-Semantics of verbs, which participate resp. do not participate in Causa-
tive Alternation and unaccusativity.
The Causative Alternation (CAL) will serve as criteria to distinguish be-
tween externally and internally caused causation; with help of the CAL the un-
accusative Verbs will be divided into two subgroups: alternating unaccusative
(AU-) verbs and non-alternating unaccusative (NAU-) verbs. In the following an
alternate distinction between AU- and NAU-verbs will be developed, namely
the presence/absence of information about how the process to be treated was
caused.
The universal concept of the encyclopedic lexicon in the English, German
and Czech languages seems to assume four different ROOTS of verbs at base to
classify the Anti-Causativity-Opposition: agentive (murder, assasinate, cut),
internally caused (blossom, wilt, grow), externally caused (destroy, kill, slay)
and cause unspecified (break, open, melt).
Moreover, it will be shown that unergative/causative pairs depict an inde-
pendent phenomenon, which does not affect considerations about CAL (corres-
pondent to ALEXIADOU et al. 2006a, b and MARANTZ 1997, but dissenting LEVIN
& RAPPAPORT HOVAV 1995 and REINHART 2000). In sections 5 and 6 I will pro-
vide an exact analysis of the roots and the syntactic projections which derive
from them.
236 Peter Kosta
2. General syntactic, morphologic and semantic characteristics of

CC
Causative Constructions (hereafter CC) are grammatical expressions, describing
a complex situation which consists of two components (COMRIE 1989: 165-166;
SONG 2001: 256-259): (i) the causing event (CAUSER-EVENT), where the
CAUSER initiates or causes something and (ii) the caused event (CAUSEE-
EVENT/STATE), where the CAUSEE is doing an action or is subject to a
change of state, as a result of the initiated or caused action of the CAUSER. The
following Japanese sentence describes such a situation of causativation:
(1) Japanese
Kanako ga Ziroo o ik- ase -ta
Kanako nom Ziro acc go- caus-pst
Kanako made Ziro go.
In example (1) the CAUSER (Kanako) is the cause that induces the CAUSEE
(Ziro) to do the action of walking. In the case of Japanese we talk about a so-
called non-periphrastic construction. A distinction is made between non-
periphrastic and periphrastic (analytical) CC. The non-periphrastic CC can be
realized in one sentence (monoclausal) by the morphologic procedure of affixa-
tion or composition; the periphrastic CC can be realized in two different clauses
by two different verbs, the basis or causative verb and the main verb.
2.1 Non-periphrastic CC
Non-periphrastic CC are causative expressions with the following three charac-
teristics: Firstly, the expression of the action initiated by the causer (i.e. ase in
(1)) and the expression of the effect (i.e. ik- in (1)) must be comprised in the
same predicate. It can consist of two or more verbs (i.e. ik-ase in (1) or two
verbs me and in example (6) below). In other words, in non-periphrastic CC
CAUSER (causative verb) and CAUSEE (effect verb) must be contained within
one and the same clause. Secondly, the NP of the CAUSER must take a gram-
matically more prominent position (i.e. subject in (1)) than the NP of the CAU-
SEE (object in (1)). Thirdly, the expression of the CAUSERs action whether
realized as an affix or as a separate verb may not contain a specifically lexical
meaning. In (1) the suffix ase has no lexical meaning as distinct from the verb
ik- go. The affix ase has a purely grammatical meaning of causation. There-
fore the periphrastic CC will be left unconsidered in this class of non-
periphrastic CC. Like the Papua language Manam (Papua, New Guinea), the ac-
tion of the CAUSER and the action of the effect can be expressed biclausally
with two different predicates, i.e. they are treated as predicates of two different
clauses.
Causatives and Anti-Causatives, Unaccusatives and Unergatives 237
(2) Manam (LICHTENBERK 1983: 447)

wsa si i-ema--be i-moasu
wind cloth 3sg.real-cause-3sg.obj-and 3sg.real-move
The wind made the loin cloth move.
In addition to the three conditions of the definition of non-periphrastic CC there
is the overt marking of the CAUSER element and of the CAUSEE element (ef-
fect). The case of Tuvaluan (Polynesia) therefore, cannot be considered a CC,
because the action caused by a CAUSER is not marked overtly by a predicate or
affix, while the one of the effect is:
(3) Tuvaluan (BESNIER 2000: 325)
te paala e tasi ne sao ia Aaifoou
the kingfish npst one pst escape because.of Aifou
The kingfish got away because of Aifou.
In general two subtypes of non-periphrastic CC are dinstinguised: 1) the mor-
phologic and 2) the compound type.
Firstly, the morphologic type refers to the main verb + causative affix. This
includes example (1) of the Japanese causative suffix ase, which is affixed to
the causative main verb (cf. COMRIE 1989: 167f.; SONG 1996: 21-26; DIXON
2000: 33-34). The anti-causative shows a zero-affix morpheme, cf. (4):
(4) Japanese
Ziroo ga ik-u
Ziro nom go- pres
Ziro goes.
It is vital for our analysis to stress the circumstance that causative elements can
occur not only as suffix, but also as prefix, as in Abkhazian, or as infix, as in
Lepcha, or as circumfix, as in Georgian (cf. SONG 1996: 21-28). Furthermore,
the morphologic processes can involve vocalic or consonantal introflections, as
in Lahu d drink vs. t (give, i.e. cause to drink), reduplication of the main
verbs consonant (cf. Egyptian Arabic mawat die vs. mawwit kill), internal
vocal stretching (cf. Kashmiri marun die vs. m run kill), reduplication of the
main verb (cf. Korana xa learn vs. xa-xa teach) and even change of pitch (cf.
Lahu c eat vs. c feed).
The Slavic languages, represented here by Russian, depict a different ver-
sion of the morphological type according to Comrie. In the Russian language it
is often the non-causative (anti-causative) form of the verb, which contains an
additional morpheme, namely the anti-causative suffix sja, corresponding to
the Czech free unbound anti-causative clitic reflexive pronoun se (cf. Russian
lomatsja/lmat se to break). For the purpose of a coherent typological treat-
ment we will treat the Russian and Czech causative verbs of the mentioned type
also as an entity of the morphological type for the moment.
238 Peter Kosta
In comparison to anti-causative verbs we could assume a causative zero-

morpheme for the causative verbs. Globally, the morphological type of the non-
periphrastic causative construction constitutes a rather small group (cf. Lahu,
Egyptian Arabic, Kashmiri, Korana, Russian, Czech, et al.).
Secondly, with the composed type the action of the CAUSER is expressed
by a single separate verb instead of a morpheme. However, the causative verb
must be adjacent to the main verb, in order that no further elements are to be
found between them. Thus, both verbs (the causative and the lexical verb) are
regarded as a single predicate (i.e. as verbal compound), similar to type 1) where
the causative suffix and the main verb form a single predicate. The Romance
languages French, Spanish and Italian belong to the languages of the second,
composed type. In French both verbs (except for few lexical elements e.g. the
clitics, several adverbs and the negation pas) must be adjacent, in order that the
CAUSEE-NP (and the object-NP of the main verb) cannot be split, as demon-
strated in (5):
(5) French
Je le lui ferai lire.
I it. acc him/her. dat make. fut read
Ill make her/him read it.
In the Tibeto-Burman language, the Eastern Kaya Li (Myanmar), the verb of
cause [VCaus] and the verb of effect [Veffect] must be in immediate adjacency.
The NP of the CAUSER (a) and the NP of the CAUSEE (phc) appear before
and after the sequence of both verbs:
(6) Eastern Kaya Li (SOLNIT 1997: 69)
a me phc
he do laugh child
He made the child laugh.
It is not only the overt adjacency of the verbs of the cause and the effect, which
allow us to analyze them as compound, but especially their ability to form a
grammatical unit. In WALS Comrie discerns this difference with the compari-
son of Tamil (Dravidian language in South India and North Sri Lanka) and the
Kobon (Madang, Papua New Guinea). Example (7) shows that verbs of cause
and of effect are collocated:
(7) Tamil (ASHER 1985: 155)
naan avane veekamaa oo-a vacceen
I he. acc quickly run- purp cause. pst.1sg
I made him run quickly.
The reason why (7) cannot count as an example for the compound type, is the
circumstance, that the verb of effect, oo-a , contains an additional final element
(purpose-a). The presence of this final verb element induces, that both verbs do
not form a unit, but that oo-a is part of a subordinate final clause. Therefore
Tamil in (7) should be considered as an example of a periphrastic CC. In reality

both verbs in (7) are only collocated in final position because Tamil is a lan-
guage of verb final word order. The Kobon is also a verb final language and the
verb of the causer and the verb of the effect are immediately adjacent. However,
different from Tamil, there is evidence that both verbs actually form a grammat-
ical unit, characteristic for the compound type, cf. (8):
(8) Kobon (DAVIES 1981b: 164)
a. mab dudu.g-p
tree be.bent- perf.3sg
The tree is bent.
b. yad mab dudu.g yu -bin
1 sg tree be.bent throw- perf.1sg
I bent the tree.
(8a) is an anti-causative sentence, while (8b) is the corresponding causative sen-
tence (the verb in 8b yu- lost its original meaning to throw; this matches our
definition, whereupon the expression of the causing action of the CAUSER
ought to be without specific meaning). In (8b) the verb of cause [VCaus] yu- is
immediately adjacent to the verb of effect [VEffect] dudu.g-p. The only aspect
marker of the [VCaus] has its scope over the complete unit, supporting the opi-
nion that this is a case of compounding.
2.2 Periphrastic CC
Periphrastic CC are expressions with the following three characteristics: Firstly,
the expressions of the causing action [VCaus] and the effected action or change
of state [VEffect] must appear in two different clauses, i.e. they must be bisen-
tential, cf. (9):
(9) Kinyarwanda (KIMENYI 1980: 172)
umukobwa y- a- tum- ye n--andik-a
girl she- pst -cause- asp I- pst -write- asp
ambrwa menshi
letters many
The girl caused me to write many letters.
In (9) the causer the girl causes me to the action of writing a letter. The German
language with the construction of lassen + infinitive pertains thereof.
The term biclausal (zweisatzwertig) indeed seems to be understood as conti-
nuum.
Secondly, the NP of the causer and the predicate of cause (i.e. umukobwa
y-a-tum-ye in (9)) must be forgrounded, while the NP of the effected and the
predicate of effect must be backgrounded (i.e. n--andik-a in (9)). In (9) this
contouring of Information Structure (IS) is accomplished by the placement of
the CAUSER sentence before the CAUSEE sentence. Thirdly, the expression of
the causing action should not contain a specific lexical meaning. In (9) the pre-
240 Peter Kosta
dicate of CAUSE tum is semantically faded in comparison to the predicate

andik , i.e. it has no other specific meaning than the grammatical meaning of
CAUSE.
3. The morphosyntactic basics of CC and anti-causatives (AC) in
Czech and German
The present article aims to illustrate the correlation of syntax and semantics of
causative and anti-causative verbs and constructions in Czech and German; the-
reby contributing to the working hypothesis of the Montague grammar and its
later development (the categorical grammar), which can be described in two par-
tial hypotheses: the principle of compositionality and the principle of homomor-
phism. Some of the first problems raised concern the relation of causativity to
transitivity, as well as anti-causativity to reflexivity, resultativity (telicity) and
the difference of intransitive ergative/unaccusative and unergative verbs.
The first part of the present article wants to define the group of construc-
tions, in which the German verb lassen + infinitive construction permits the
causative in Czech. Semantic correlates of many types of situations, in German
expressed by the causative meaning of the sentence (proofs (10a)-(24a)), is ex-
pressed in Czech by constructions of partly the same, partly different structural
types, cf. (10b)-(24b) (I tried to translate the German proof examples from NED-
JALKOV 1976: 7 into Czech).
(10) a. Karl lie den Brief fallen.

Charles dropped the letter.
b. Karel upustil dopis. / * Nechal/*dal upustit dopis.
Charles dropped the letter. /* Let / * gave the letter fall.
(11) a. Karl lie sich zu Boden fallen.
Charles dropped/let himself fall to the ground.
b. Karel upadl na zem/Karel se svalil na zem.
Charles fell to the ground. Charles let himself fall to the ground
(12) a. Karl lie das Pferd galoppieren.
Charles let the horse gallop.
b. Karel vybdl kon do cvalu.
Charles urged the horse into a gallop.
(13) a. Karl lie seine Tochter den Brief abtippen.
Charles had his daughter type out the letter
b. Karel dal sv dcei /nechal svou dceru naklepat dopis na stroji.
Charles gave his daughter / let his daughter had run off a letter to the
machine.
(14) a. Er lie seinen Sohn die Jacke ausziehen.
He let his son take off his jacket.
b. Vybdl svho syna, aby si sundal bundu./
*Nechal/*dal synovi si sundat bundu.
He urged his son to remove his jacket. / *He let / *gave (to) his son
take off his jacket.
(15) a. Er lie mir von seinem Sohn eine Tasse Tee geben.
He let to give me from his son a cup of tea.
b. Poprosil svho syna, aby mi pinesl lek aje.
He asked his son to bring a cup of tea to me.
*Nechal/*dal mi pinst lek aje / svm synem / od svho syna.
*He let / *gave meDAT bring a cup of tea / (by) his sonINSTR / from his son
He asked his son to bring a cup of tea to me.
(16) a. Er lie sich von seinem Sohn das Geld geben.
He asked (for) himselfDAT from his son the money to give.
He asked from his son to give the money.
b. Podal syna, aby mu dal penze.
He asked his son to give him money.
*Nechal si/ *dal si dt penze
Let (himself) gave (himself) give money
svm synem / od svho syna.
by his son / from his son.
He * let the money to give
by his son / from his son.
(17) a. Karl lie sich (durch sie/durch ihren Besuch) nicht stren.
Charles was not bothered/disturbed (by them/by their visit).
b. Ona/ jej nvtva Karla neruila. /
She / her visit Karel did not disturb
Her visit did not disturb Karl.
(18) a. Die Tr lie sich leicht ffnen.
b. Dvee se lehce otevrali
The door was easily openable.
(i) Die Tr ffnete sich leicht, die Tr stand ein spaltbreit offen.
The door opened up slightly, the door stood ajar.
(ii) Die Tr war leicht zu ffnen.
The door was easily openable.
(19) Dvee se otevraj LEHce
(i) Die Tr ffnet sich gerade leicht, die Tr steht ein spaltbreit offen.
The door is just opening slighty, the door is standing ajar.
(ii) Die Tr ist leicht zu ffnen.
The door is easily openable.
(20) a. Sie lieen es sich schmecken.
b. Chutnalo jim / *Nechali /*Dali si chutnat.
They enjoyed the meal
242 Peter Kosta
(21) a. Er lie sich bei diesem Schneider einen Anzug anfertigen.

b. Dal/nechal si ut oblek u tohoto krejho.
He had a suit made at this tailor.
(22) a. Mit ihm / hier lsst sich gut arbeiten.
b. S nm se mi / tady se dobe pracuje. / ??Tady se d dobe pracovat.
Working with him / here is good.
(23) a. Die Freude lie ihn erbleichen.
b. Radost zbledl. / *Radost ho nechala/*dala zblednout.
He blanched with joy.
(24) a. Er lie sich im Lift hinauftragen.
b. Vtah ho vyvezl nahoru.1
He had the lift carry him upstairs.
4. (Anti-)causatives in general
Short discription of syntax:
Causative constructions of the German analytical periphrastic type with lassen +
infinitive construction are evidently expressed in Czech either analytically with
the verb nechat, dt (lassen or geben, resp. let or give) (cf. examples (13b),
(21b)), or with a resultative transitive verb by means of a causative prefix (cf.
10b), or with an embedded infinitive clause; that is, a construction in which the
verb of the matrix clause possesses the illocution of a direct speech act (wish,
request, demand = Vcaus); and the embedded verb exhibits the lexical semantics
of the verb action to be performed (Veffect), containing a final subordinate
clause with the final embedding2 by the conjuction aby + conditional (cf. (14b),
(15b), (16b)).
Anti-causative constructions in Czech however, are described by an intransi-
tive unaccusative (less often unergative) verb, which either takes an overt reflex-
ive pronoun ((18b), (22b)) or stays morphologically unmarked ((11b), (23b)).
In addition, there are more or less phraseologized resp. lexicalized solutions.
Therefore, Czech can be classified morphologically as well as syntactically as
language of both the first language type (non-periphrastic CC with morphologic
marker for Anticausatives se) and the second language type (periphrastic con-
structions with analytical expression of causativity).
__________
1 For all translations into Czech there are variations, imitating the German pattern. Howev-
er, they do not sound Czech, but in Bohemistic literature are clearly declared as loan
translations; e.g. (20) Nechali si chutnat, (18) Dvee se daj lehce otevrat or (24) Ne-
chal se vyvzt vtahem a nahoru are possible translations, but stylistically distinctly
marked as Germanisms.
2 cf. JUNGHANNS (1994) for the Russian language.
Short characterization of semantics:

It is an intesting fact that the meaning of the causative comprises both transitive
and intransitive verbs. Thereby the latter class seems to contain both unaccusa-
tive inchoative verbs (examples (11b) padat, (18b) otevratse, (19b) aufgehen
(Tr), (23b) zblednout) and unergative verbs (example (22b) pracovat). In any
case, inchoative verbs must be involved semantically, inevitably expressing the
change of an action into a state or the change of one state to another state as is
the case with unaccusative verbs.
Restrictions:
The quite strict syntactic and morphologic restrictions are striking. Apparently,
they are established on a semantic basis, which will be my task to detect. I the-
reby think about examples of the anti-causatives of type (18a): Die Tr lie sich
leicht ffnen. The combination with the Czech verb nechat (lassen, let) is un-
grammatical, while with dt it can only be combined with the perfective aspect
of the main verb. The information structure and the kind of adverbs apparently
seem to play an important role for the interpretation. Namely, if the sentence
(18b) is realized with the focus on the adverb of manner and right dislocation
(19), only the interpretation of habitual modal reading can be licensed, while the
existing syntactic realization shows a certain ambiguity:
(18) b. Dvee se lehce otevraj
(i) Die Tr ffnet sich gerade leicht, die Tr steht ein spaltbreit offen
The door is just opening slightly, the door is standing ajar.
(ii) Die Tr ist leicht zu ffnen
(19) c. Dvee se otevraj LEHce
(i) #Die Tr ffnet sich gerade leicht, die Tr steht ein spaltbreit offen
#The door is just opening slightly, the door is standing ajar.
(ii) Die Tr ist leicht zu ffnen
The alternation between inchoative intransitive verbs and transitive causative
verbs by means of affixation is indeed a productive word formation process and
occurs in Czech as well as in other Slavic languages. This results from the fol-
lowing evidences: varit to cook ob-varit scald, zelenit to make green, to
color green, o-zelenit to plant with trees and bushes, Czech. zelent, ernt,
modrat, blet to shimmer green, black, blue, white (na) zelenit, naernit,
namodit, nablit to blacken, to color blue, to bleach, viset to hang povsit
to hang (up), zavsit to hang up, to hook in, navsit distributive: to hang
everywhere, to distribute.
244 Peter Kosta
Evidence of introflexion is unproductive; however, they are proven in Proto-

Slavic and in the basic vocabulary of the modern Slavic languages: Czech. pt
to drink (na)pojit to water, sedt to sit (po)sadit to set, to put, tct to
flow (na)toit to pour (in).
A further word formation process is the type pemlet (o nem) to think
about something, to ponder promyslit nco to think something through.
These types also allow a transitive, however, not necessarily a causative, mean-
ing.
An inverted word formation process, productive in modern Indoeuropean
languages, whereupon a simplex with causative-transitive meaning by means of
reflexiveness changes into an intransitive reflexive verb, can be seen in the fol-
lowing examples: serdit kogo-to to annoy someone / serditsja to be an-
noyed, vernut to-nibud to give back sth., to return sth. / vernutsja to come
back, to return, Czech. zlobit (maminku) to annoy (the mother) / zlobit se to
be annoyed, vrtit knihu to return the book / vrtit se to return, oblkat
nkoho to dress s.o. / oblkat se to dress.
My working hypothesis assumes that in the Czech example (10b) Karel is
the causer [Scause], the prefix u- [Vcause], because it fulfills the same function
as the lexical verb lassen in the German example (10a) and the verb pustit is the
basic form of the derivation [Veffect], i.e. the state resulting from the action of
falling. Yet, how to analyze unaccusative verbs like fallen, to fall, in (11b)?
The German example (11b) suggests that the zu Boden fallen, to fall to the
ground, may happened deliberately and controlled, so that Karl can be inter-
preted both as [Scause] and [Seffect], i.e. he was both the cause (CAUSER) and
the victim (CAUSEE). While in the Czech example (11b) the property [ inten-
tional] is not expressed. If the verb upadl na zem is substituted by the verb svalil
se na zem, plcnul sebou na zem, to let oneself fall, to chuck oneself on the
floor, the causing element and Karel as causer become evident. Thereby, in my
opinion, to start with as an assumption and pre-theoretically, the following cha-
racteristics play an important part: the lexical semantics of the verb [ con-
trolled, intentional action], the telicity (i.e. limitation, the direction of the ob-
jects falling is expressed characteristically with a prepositional phrase) and the
resulting meaning of the verb, which together with the PP signals the inner limi-
tation of the objects with the limit na zem [Veffect].
b. Karel upadl na zem. / Karel se svalil na zem.
(25) a. Hans zerbrach die Vase.
Hans broke the vase.
b. Die Vase zerbrach (*sich).
The vase broke.
(26) a. Hans ffnete die Tr.

Hans opened the door.
b. Die Tr ffnete *(sich)
The door opened.
Examples (25a) and (26a) present transitive, causative verbs. Examples (25b)
and (26b) present the intransitive, anti-causative counterparts to these causative
verbs. The examples show that in the German language only two classes of anti-
causative verbs are available. One class, demonstrated by the verb sich ffnen in
(26b), must take the reflexive pronoun sich, together with the DP and its themat-
ic role [Theme]. The other class, demonstrated by the verb zerbrechen in (25b),
must not take a reflexive pronoun together with the DP [Theme]. The transitive
versions of both of these types of anti-causatives do not differ from each other;
they are each combined with a subject-DP and an object-DP.
In Czech there are equivalents in which Causative Alternation can be no-
ticed; however, most verbs that are able to form a causative do form an intransi-
tive anti-causative, marked by a reflexive pronoun (in the following referred to
as marked anti-causatives). Cf.:
(27) a. Petr posadil dvtko na idli.
Peter setzte das Mdchen auf den Stuhl.
Peter seated the girl on the chair.
b. Dvtko *(se) posadilo na idli (*Petrem)
Das Mdchen setzte *(sich) auf den Stuhl (*durch Peter)
The girl sat down/seated herself on the chair (*by Peter).
(28) a. Karel unavil studenty pednkou.
Karl ermdete die Studenten mit seinem Vortrag.
Charles wearied his students with his lecture.
b. Studenti *(se) unavili pednkou *Petrem/ Petra (o.k).
Die Studenten (*sich) ermdeten durch den Vortrag * durch Peter / von
Peter (o.k.).
The students became weary by the lecture * by Peter / of Peter (o.k.).
(29) a. Marie uspala miminko. / Marie vzbudila miminko.
Maria brachte das Baby zum schlafen / Maria weckte das Baby.
Mary put the baby to sleep. / Mary woke the baby.
b. Miminko (*se) usnulo (*Mari). / Miminko *(se) vzbudilo (*Mari).
Das Baby (*sich) schlief ein / Das Baby wachte (*sich) (*durch Maria) /
(mit Hilfe von Maria, o.k.) auf.
The baby fell (*himself) asleep./The baby woke up (*himself) (*by
Mary) / (owing to Mary, with the help of Mary, o.k.)
(30) a. Slunce oteplilo vodu.
Die Sonne erwrmte das Wasser
246 Peter Kosta
The sun warmed (up) the water.

b. Voda (*se) oteplala (*sluncem)
Das Wasser (*sich) erwrmte (*durch die Sonne) / (durch die Einwirkung
von Sonne)
The water warmed up (*himself) (*by the sun) / (under the influence of
the sun).
(31) a. Slunce, vlha a teplo nechaly dozrt / shnt ito a obil.
Die Sonne, Flssigkeit und Wrme lieen den Roggen und das Getreide
reifen / verderben.
The sun, the liquid, and the warmth ripened/ruined the rye and the
crops.
b. Obil a ito (*se) dozrly. Obil a ito (*se) shnily.
Das Getreide und der Roggen reiften (*sich) / verdarben (*sich).
The crops and the rye ripened (*themselves) / were ruined (*by them-
selves).
(32) a. Emil a detektvov rozesmli chlapce.
Emil und die Detektive brachten den Jungen zum Lachen.
Emil and the Detectives made the boy laugh.
b. Chlapec *(se) rozesml.
Der Junge fing (*sich) an zu lachen.
The boy started (*himself) to laugh.
(33) a. Vichr potrhal plachtu.
Der Wirbelsturm zerriss die Segel.
The hurrican ripped the sails.
b. Plachta *(se) potrhala (*vichrem).
Die Segel (*sich) zerrissen.
The sails ripped (*themselves).
(34) a. Petr zmnil strategii.
Peter nderte die Strategie
Peter changed the strategy.
b. Strategie *(se) zmnila.
Die Strategie nderte *(sich).
The strategy changed (*itself).
It is remarkable that the above b-examples prove clearly, that unaccusatives and
impersonal passives show the same syntactic structure, i.e. the inner object be-
comes the subject and maintains its theta role [Theme] and the theta role [Agent]
of the external argument as well as the argument itself are suppressed. The theta
role [Agent] of the causative version may not be continued in the anti-causative
version by means of an instrumental NP with the theta role [Agent], but at the
most by means of a PP (skrz, pes)/NP (bleskem, hromem, pednkou) with the
theta role [Instrument], [Cause] or by an adnominal genitive with the theta role
of possession (cf. (28b) vs. (28a)), where the instrumental NP pednkou is [In-
strument] or [Cause], but not [Agent]. This fact proves that the theta role
[Agent] in fact must be suppressed (similar to the impersonal se-passive).
The Causative Alternation between causative verb and anti-causative verb is
well-proven cross-linguistically and relates to inchoative verbs only (also to
verbs of change of state or change of level, like tauen to thaw, dick werden to
thicken, ndern to alter etc.). Usually the Causative Alternation is taken as a
test for unaccusativity, i.e. unaccusative verbs take part in it (vgl. KOS-
TA/FRASEK 2004; LEVIN & RAPPAPORT HOVAV, 1995, hereafter L&RH). It fol-
lows a short recollection of the differentiation of unaccusative and unergative
verbs in connection with the theory of anti-causatives.
4.1 (Anti-) causatives and the unaccusative-unergative-opposition according to
PERLMUTTER (1978)
Since PERLMUTTERs formulation (1978) of the Unaccusative Hypothesis (UH)
most theories classified intransitive verbs as either unaccusatives or unergatives.
The terminology of argument structure of the UH proceeds from the assumption
that unaccusative predicates select a single internal argument, while unergative
predicates select a single external argument (HARVES 2009: 415). In syntax un-
accusative predicates project their subjects VP-intern, in the position of the di-
rect object (DO), while unergative predicates project their subjects VP-extern,
similar to transitive verbs projecting basis-generated subjects. Our understand-
ing of the syntactically different entity of unaccusative predicates goes back to
various syntactic tests, which can be considered as means of diagnosis for the
differentiation between both verb classes of intransitive verbs globally and
cross-linguistically. PERMUTTER (1978) e.g. demonstrated that the impersonal
passive in the Dutch language can only derive from unergative predicates; re-
sulting from the contrast of (35) vs. (36):
(35) Dutch impersonal passives: unergative predicates
a. Er wordt hier door de jonge lui veel gedanst.
Es ist hier von den jungen Menschen viel getanzt
It is here from the young people much danced
Es wird hier viel von den jungen Menschen getanzt.
b. Er wordt in deze kamer vaak geslapen.
Es ist in diesem Zimmer oft geschlafen
It is in this room often slept
Es wird oft in diesem Zimmer geschlafen. (PERLMUTTER 1978: 168)
248 Peter Kosta
(36) Dutch impersonal passives: *unaccusative predicates

a. *Door de lijken wird al ontbonden.
Von den Leichen ist schon zersetzt
From the corps is already decomposed
Es ist schon zersetzt von den Krpern.
b. *In dit ziekenhuis wordt (er) door de patienten dikwijls
In diesem Krankenhaus ist es von den Patienten oft
In this hospital is it from the patients often
gestorven
gestorben
died.
In diesen Krankenhusern wird von den Patienten oft gestorben.
(PERLMUTTER 1978: 169)
BURZIO (1986) was possibly the most important researcher, who clarified our
understanding of the characteristics of intransitive predicates within the scope of
the GB theory. He observed for the Italian language that unaccusatives differ
from unergatives in at least three points: Unaccusatives select essere to be as
auxiliary in past tense (vs. avere to have), they demonstrate participle agree-
ment with their subject and they allow ne-clitisation: Proven by the examples
(37)-(39) for the Italian and (39, 40) for the French language.
(37) Auxiliary selection
a. Maria / *ha arrivata (unaccusative)
Maria ist / *hat angekommen
Mary has / *is arrived
Mary has arrived.
b. Maria * / ha telefonato (unergative)
Maria *ist / hat telefoniert
Mary *has / is phoned
Mary has phoned.
(38) Past-participle-agreement
a. Maria arrivat-a / *-o. (unaccusative)
Marie ist angekommen-FEM / *MASC
Mary has arrived-FEM / *MASC
Mary has arrived.
b. Maria ha telefonat-o / *-a. (unergative)
Maria hat angerufen-MASC / *FEM
Mary has phoned-MASC / *FEM
Mary has phoned.
(39) ne-clitisation (extraction of DO)

a. Giovanni nei inviter molti ti
Giovanni von-ihnen wird-einladen viele
Giovanni of them will-invite many
Giovanni wird viele von ihnen einladen.
Giovanni will invite many of them.
b. Nei arrivano molti ti
von-ihnen werden-kommen viele
of-them will-come many
Viele von ihnen werden kommen.
Many of them will come.
c. *Nei telefonano molti ti
von-ihnen werden-anrufen viele
of-them will-call many
Viele von ihnen werden anrufen.
Many of them will call.
d. Je nen ai pas beaucoup (= du lait). (transitive)
I do not have much (= milk).
e. *Il nen pas tlphon aujourdhui beaucoup. (unergative)
(40) Past-participle-agreement in French
a. Anne la aime. (transitive)
Anne sie-hat geliebt-FEM (e.g. Maria)
Anne her-has loved-FEM (e.g. Maria)
Anne loved her.
b. Anne est arrive. (unaccusative)
Anne ist angekommen-FEM
Anne has arrived-FEM
Anne has arrived.
c. Anne a tlphon (unergative)
Anne hat telefoniert
Anne has phoned
Anne has phoned.
The examples for ne-raising in Italian in (39) and for past-participle-agreement
with the direct object in French (40) prove that the syntactic subjects of unac-
cusatives agree with DO, i.e. subjects of unaccusatives are deep structure DO. In
(39b) the ne von ihnen (of them) from the position of the DO molti (ne)
(many of them) is clitisised onto the verb arrivano. This movement is not legi-
timate with the unergative verb telefonano (they phone) in (39c). In the French
language subjects of unaccusative verbs are likewise associated with DO. As
demonstrated in (40a) the French language shows object agreement in the case
of transitive verbs. The object agreement becomes evident for unaccusative
250 Peter Kosta
verbs, since the subject Anne acts like the deep structure DO in the case of unac-
cusative verbs ((40b) arriver). When Maria is pronominalized (la her), it pro-
clitisises onto the auxiliary avoir, crosses the past participle and results in object
agreement between the part-FEM and the pronoun-FEM. In the case of unerga-
tive verbs (40c) there is no such object agreement, for the form PPP tlphon is
not the agreeing (default) form.
4.2 Diagnosis for unaccusativity in Russian, Czech and Polish (according to
HARVES 2009 and KOSTA/FRASEK 2004)
Like the Romance and Germanic languages, there is enough evidence in Slavic
languages to differentiate between unergative and unaccusative predicates. Since
this differentiation will be of importance for the correlation between transitivity,
intransitivity, and (anti-)causativity, I will shortly elaborate on the most impor-
tant diagnosis tests for unaccusativity in the Slavic languages (Russian, Czech,
Polish). They have been discussed in several papers: PESETSKY (1982),
SCHOORLEMER (1995, 2004), BABYONYSHEV (1996) as well as HARVES (2002,
2003, 2009) for the Russian language, CETNAROWSKA (2000, 2002) and KOS-
TA/FRASEK (2004) for Polish and KOSTA/FRASEK (2004) for Czech.
We assume that unaccusative verbs fulfill one of either semantic property:
(i) they belong to the group of verbs that are subject to change-of-state, or (ii)
they belong to the group of verbs that mark action-change-of-state:
(i) achievement ii) accomplishment
event event
state state process state

The syntactic structure of unaccusative verbs corresponds with the chart (41a):
Thereby, unaccusative verbs project only internal arguments with the theta role
[Theme] and have no external argument with the theta role [Agent], cf. (41a):
(41) a. unaccusatives (ergatives)
_______
VP V NP
[ AGENT] Predicate [+ THEME]

Conversely, unergative verbs demonstrate a syntactic structure, whose external
argument contains the thematic role [Agent] or another thematic role and the in-
ternal argument with corresponding thematic role [Theme] is absent, resp. not
projected, cf. (41b):
(41) b. unergatives
SpecvP vP v NP
[+ AGENT] Predicate [ THEME]

Accordingly, one has to assume that unaccusative verbs only form a lexical pro-
jection (VP) with a single internal argument (complement) (cf. 41a):
(41) a. VP
V Compl
Transitive verbs demonstrate a light vP-projection, which the nominative case
(in Spec) is assigned to, as well as a lexical projection, which the thematic role
[Theme] is assigned onto the internal argument; cf. (41b):
(41) b. vP
Spec v
V VP
Spec V
V Compl
When assuming a coherent structure in all of the three verb classes, in which the
issue of projection is decided by the process of MERGE and MOVE, namely in
dependence of the lexical structure of the root (of which more later, cf. 1.4-1.6),
the result is one of the below possibilities of MERGE and MOVE:
(41) c. vP
DP v
Petr VP
V DO
dlal prci
Petr pracoval = Petr dlal prci
Peter worked = Peter made (the/a) work
252 Peter Kosta
(41) d. vP
v AP
BECOME DP A
Dvee se otevely = Dvee se stly otevenmi

(41) e. vP
DP v
v AP
DP A
dvee oteven
Proceeding gradually our attempt is now to differentiate the class of ergative
(unaccusative) verbs from the class of unergative verbs. There are five syntactic
tests available. The fact that subjects of unaccusative verbs are deep structure
objects (i.e. internal argument of the lexical verb), allows the following predic-
tions:
(42) a. Petr rozbil sklenice
PeterNOMSG zerschlug die GlserAKKPL
PeterNOMSG broke the glassesAKKPL
Peter broke the glasses.
b. Piotr zgubi+ rkawiczki.
PeterNOMSG verlor die HandschuheAKKPL
PeterNOMSG lost the glovesAKKPL
Peter lost the gloves.
(43) a. Petrovi se rozbily sklenice
PeterDATSG Refl zerschlugenPrt-3PS.PLFEM GlserNOMPLFEM
PeterATSG Refl. brokePast-3PS.PLFEM glassesNOMPLFEMBI
b. Piotrowi (zgubi+y si) zgin+y rkawiczki.
PeterDATSG Refl verlorenPrt-3PS.PLFEM HandschuheNOMPLFEM
PeterDATSG Refl lostPast-3PS.PLFEM glovesNOMPLFEM
1. Test for ergativity/unergativity: anaphora vs. pronoun
Only real subjects can take reflexive pronouns (anaphora) according to principle
A of the binding theory. The fact that in (44b) and (44d) anaphora cannot be
bound, implies a position below the anaphora.
Ergative verbs *anaphora / AplPronoun

(44) a. Sklenice se mu (= Petrovi) rozbily
Die GlserNOMPLFEMBI CLREFLDAT3PSSG CLAPLDAT3PSSG zerschlugenPrt-3PS.PLFEM
The glassesNOMPLFEMBICLREFLDAT3PSSGCLAPLDAT3PSSG brokePast-3PS.PLFEM
The glasses broke.
b. Sklenice se (*si/sob) (= Petrovi) rozbily
Die GlserNOMPLFEMBI CLREFLCLREFLDAT3PSSG zerschlugenPrt-3PS.PLFEM
the glassesNOMPLFEMBI CLREFLCLREFLDAT3PSSG brokePast-3PS.PLFEM
c. Rkawiczki mu zgubi+y/zgin+y si
HandschuheNOMPLFEMBI CLAPLDAT3PSSG verlorenPrt-3PS.PLFEM CLREFLDAT3PSSG
glovesNOMPLFEMBI CLAPLDAT3PSSG lostPast-3PS.PLFEM CLREFLDAT3PSSG
The gloves got lost.
d. Rkawiczki (*sobie/si) zgubi+y/zgin+y si
HandschuheNOMPLFEMBI CLREFLDAT3PSSGverlorenPrt-3PS.PLFEM CLREFLDAT3PSSG
GlovesNOMPLFEMBI CLREFLDAT3PSSGlostPast-3PS.PLFEM CLREFLDAT3PSSG
Conversely, unergative verbs can bind anaphora, for here the subject is base-
generated as antecedent.
Unergative verbs anaphora / *AplPronoun
(45) a. Petri sii le/pracuje/stoj/zpv
Peter sichCLREFLDAT3PSSG liegt/arbeitet/steht/singt
Peter himselfCLREFLDAT3PSSG lies/works/stands/sings
b. Petri (*mui) le/pracuje/stoj/zpv
Peter ihmCLAPLDAT3PSSG liegt/arbeitet/steht/singt
Peter himCLAPLDAT3PSSG lies/works/stands/sings
c. Piotri sobiei le\y/pracuje/stoi/`piewa
Peter sichCLREFLDAT3PSSG liegt/arbeitet/steht/singt
Peter himselfCLREFLDAT3PSSG lies/works/stands/sings
d. Piotri (*mui) le\y/pracuje/stoi/`piewa
Peter himCLAPLDAT3PSSG lies/works/stands/sings
2. Test for energativity/unergativity (according to ABRAHAM 2004): Past passive
participle (PPP) of ergative verbs is grammicalized as adjunct of the deep struc-
ture object; for unergative verbs the result is ungrammaticality (*):
(46) a. Svalen ze svahu kmen Kmen se svalil ze svahu (eV)
Gerollter vom Abhang Stein Stein rollte den Abhang hinunter
Rolled from slope stone stone rolled down the slope
b. Rozbit vtrem sklenice Sklenice se rozbily (eV)
Zerschlagene durch den Wind Glser Die Glser zerschlugen
Broken by the wind glasses The glasses broke.
254 Peter Kosta
c. Zwalone na ziemi rzeczy Rzeczy zwali+y si na ziemi

Gerollte auf die Erde Sachen Die Sachen rollten auf die Erde
Rolled onto the floor things The things rolled onto the floor.
(47) a. *Pracovan Petr pracoval (uV)
b. upracovan Petr Petr se upracoval (eV?)
3. Test for ergativity (according to ABRAHAM 2004): eV cannot form agent
nouns (nomina agentis) where as uV can:
(48) a. Vtr se utiil *Utiitel, *utink (eV)
Der Wind beruhigte sich *der Beruhiger
The wind calmed.down. *the calmer
b. Petr pracoval pracovnk Arbeiter (worker)
Petr arbeitete der Arbeiter (uV)
Peter worked. the worker (uV)
c. Lano se napnulo *Napnatel, napnek (lana)
The line stretched *the stretcher (of the line)
Die Leine spannte sich. *Der Spanner, *die Spannerin (der Leine)
The line stretched. *the stretcher (of the line)
d. Petr stavil dm Stavitel, stavebnk (tV) (domu)
Peter baute ein Haus der Bauherr (des Gebudes)
Peter built a house. the (home-) builder (of the house)
4. Test for ergativity (according to ABRAHAM 2004): eV cannot be passivized by
an impersonal se-passive, uV and tV can be passivized:
(49) a. * Vtr byl / bude / je uten Petrem
*The wind was / will be / is calmed by Peter
b. Vtr se util (*Petrem)
The wind-REFL calmed down (*by Peter)
c. Vtr se util.
The wind-REFL calmed down
d. Lano bylo napnuto Petrem
The line has been stretched by Peter
e. Lano se napnulo (*Petrem)
The line-REFL stretched (*by Peter).
f. Kmen byl svalen Sisyfem
The stone was tumbled and rolled down by Sisyphus
g. Kmen se vykutlel na svah (* Sisyfem)
The stone is rolled in on the rise (*by Sisyphus)
h. Kov se tav ohnm
The metal is melted by fire
i. Led taje *sluncem (pomoc slunce, skrz slunce = o.k.)
Ice thaws *by the sun // from the sun, through the sun = o.k.
(50) a. Tam se stav, stanuje, hraje, zpv, bh, le, stoj

*There REFL built_3Ps.Sg, played, sung, run, layed, stood (in English, imper-
sonal passives of this kind are forbidden)
5. Test: distributive po-phrases (according to HARVES 2009) in the Czech lan-
guage not relevant (for Russian cf. also GLUSHAN in this volume):
(51) a. Ztratilo se mi po jedn knce z kadho reglu.
It has been lost to me one book eachDISTRIBUTIVE from each shelf
b. Zpvalo tam po jednom pvci v kadm chru.
It has sung one singer eachDISTRIBUTIVE in each choir
c. Za druh svtov vlky umral jeden po druhm.
During the Second World War died one after another
d. Za druh svtov vlky pracovalo v muninch
fabrikch po pti ench u jednoho bcho psu
During the Second World War, there worked in munitions factories five
women each in every treadmill.
6. Test for negation (GenNeg)
In Polish, the direct object of a transitive verb must be assigned genitive of ne-
gation if it is in the scope of clausal negation. Only a real object (deep structural
object) can be assigned genitive of negation, not a deep structure subject, cf. the
difference between ergatives r kawiczek mu nie zgubio si vs. unergatives
*Piotra nie piewa // Piotr nie piewa. In Czech, the genitive of negation is not
productive and this is the reason why the accusative case is assigned to the direct
object NP.
4.3 Unergative and unaccusative verbs according to L&RH (1995)
L&RH incorporate Permutters twofold suggestion. On the one hand he post-
ulates that certain phenomena in the behavior of intransitive verbs can be ex-
plained when assuming that unaccusative and unergative verbs possess funda-
mentally different syntactic structures. On the other hand he assumes that both
verb classes can be differentiated based on their meaning.
L&RH find a single theory of rules for the connection of verbs with both
semantics and syntax. With these linking rules they describe which semantic
characteristics are determining for a unergative/unaccusative structure. They
come to the conclusion that unaccusativity is represented syntactically but de-
fined semantically.
A semantic approach, which describes the observations satisfactorily with-
out the assumption of specific syntactic structures, is unknown by Levin and
Rappaport Hovav (L&RH); however, they do not exclude the option on prin-
ciple.
The following part of the present paper reconstructs the steps that resulted in
the conclusion of L&RH. Starting point is the fundamental definition:
256 Peter Kosta
An unergative verb possesses a subject and not an object in deep structure.

An unaccusative verb possesses an object and not a subject in deep structure.
This definition is recited by L&RH in two further notations:
1) They describe argument structures:
An unaccusative verb possesses a direct internal, but no external argument.
An unergative verb possesses an external, but no direct internal argument.
2) They describe the syntactic configuration of the deep structures, as demon-
strated in (52a) vs. (52b):
(52) a) Unaccusative Verb: ___ [VP V NP]
b) Unergative Verb: NP [VP V]
Examples (52a) and (52b) illustrate this definition:
(52) a. Der Mann lacht. Mu se smje.
The man laughs.
b. Der Mann fllt. Mu pad.
The man falls.
The noun phrase [verb phrase LACHEN] becomes a sentence, by filling the NP
by der Mann and eliminating the brackets of lacht.For the unaccusative verb the
NP is put in brackets.
This clarifies that the syntactic subject basically, in deep structure, acts to the
verb as an object: The fall happens to the man and the blank (___) means: The
causer (deep structure subject) remains unidentified.
The deep structure of unaccusative and impersonal passive verbs are identical in
this approach, i.e. (52b) applies for a sentence like:
(52) c. Das Haus wird gebaut.
The house is built.
d. Stav se dm.
They are building a house.
L&RH use the above described syntactic differences to discuss the usefullness
of tests for the differentiation of verb classes (unaccusative diagnostics). Among
other things they want to clear up the cases when different tests result in differ-
ent conclusions (unaccusative mismatches). They only accept tests as actual un-
accusative diagnostic, which explain the behavior of verbs by means of the
above mentioned syntactic differences.
Two tests are presented in detail below:
1) Resultative constructions, used by L&RH to demonstrate that unaccusativity
is represented syntactically (cf. 4.3.1).
2) Causative Alternation, by means of which they derive linking rules and there-
by demonstrate that unaccusativity is semantically determined (cf. 4.3.2).
4.3.1 Result testing for the syntactic representation of unaccusativity

Resultative phrases (RPs) are assigned to NPs. They describe a state of the refe-
rent of the NP, who enters the process of the verb mentioned in the phrase:
(53) a. Er schlug den Nagel flach. / Zatloukl/zabil hebk do stny.
He hit the nail flat. / (He) knocked / killed a nail into the wall.
b. Er strzte zu Tode. / Letadlo se ztilo. / On se ztil.
He fell to his death. / The aircraft crashed. He crashed.
c. Usmrtil se/zabil se. (non-intentional meaning, e.g. non-reflexive)
d. (He) was killed. (e.g. by someone else)
L&RH confine themselves to the English languages; the German and Czech ex-
amples come from me. They are to prove temporarily that L&RHs observations
also apply to the German and Czech language. Noticably, the German RP often
corresponds surface-syntactically and morphologically to a combination of affi-
gation in the Czech language, corresponding to a perfectivatizing prefix and a
clitic-like affix morpheme, cf. On za-tloukl hebk: Er schlug den Nagel flach.
He hit the nail flat.
For (53) we can observe the following: Resultative phrases in German, resp.
the reflexive pronoun se in Czech3 can depending on the predicate be com-
bined with either the (surface-)subject (54b) or the direct object (54a); thereby
the predicates are fixed to either of those options, as demonstrated by the follow-
ing examples:
(54) a. *Er schlug flach. / *He beat flat.
b. *Hebk se zatloukl / * Nail is knocked
L&RH assume that the lexical representation of a verb and its argument struc-
ture does not change in resultative constructions. Therefore they interpret resul-
tative constructions according to the semantic principle of composition from
Montague: The whole construction bears the meaning of the verb plus RP in
German and the meaning of the verb plus AA-affix morphology in Czech. In the
Czech language the prefix morphology is responsible for the partial meaning
<telic, resultative, perfective> and the se-particle is responsible for the partial
meaning <anti-causative>. Only the causative relation between the action of the
verb and the state described by the RP is not syntactically represented.
In the German language RPs are only connected with the (surface-)subjects
for passive and unaccusative verbs. As said before, this subject is fundamentally
an object. Thus, in the Czech language the morphologically syntactically corres-
ponding (German) RPs are expressed by the prefixes and affixes, resp. the ref-
lexive pronoun se.
__________
3 In the Czech language German resultative phrases are expressed by various prefixes of
perfectivation or Aktionsarten(AA), while in German the resultativity is expressed
phrasal or by means of AA-prefixes.
258 Peter Kosta
L&RH consider the following questions:

a) Why is unaccusativity (resp. impersonal passive) the crucial condition for the
connection of RP and the subject?
b) Which additional assumptions are necessary to explain the phenomena in
connection with the RPs?
Nine steps present how L&RH conduct this study.
First step: Resultative constructions indicate, in which state the NP-referent en-
ters by the process of the verb. Therefore, they act as limiters, for the process is
assumed to be finished when they are situated in the new state. Hence, resulta-
tive construction in Czech and German are always telic; in Czech furthermore
perfective. The telicity is expressed by the perfective aspect in Czech and by RP
with the meaning resultativity in German; confirmed by the fact that RPs are in-
compatible with durative phrases, which in turn can only be connected with atel-
ic predicates:
(55) a. istil si dv minuty zuby.
Er putzte sich zwei Minuten lang die Zhne (durative)
He cleaned his teeth for two minutes.(durative)
b. Vyistil si zuby za dv minuty.
He has brushed his teeth in two minutes.
Er putzte sich die Zhne sauber in zwei Minuten. (resultative, telic)
He brushed his teeth clean. (resultative, telic)
c. *(On) si vyistil dv minuty zuby.
* He has been brushing his teeth for two minutes.
*Er putzte sich zwei Minuten lang die Zhne sauber.
*He brushed two minutes clean the teeth.
d. Opravoval si hodinky ti hodiny.
Er reparierte seine Uhr ber drei Stunden.
He has been repairing his clock for three hours.
e. Opravil/zpravil si hodinky za ti hodiny.
Er reparierte seine Uhr in drei Stunden.
He repaired his clock in three hours.
f. *Er strzte Klaus zu Tode
He rushed Klaus to death.
g. Usmrtil (*se) Klause
He killed (*himself) Klaus.
When a verb is already telic, the RP in German and the prefix-morphology in
Czech serves as a further specification of the final state, see also point 8. Thus,
(55f) cannot take simultaneously the RP and the DO and the Czech example
(55g) cannot take both the reflexive pronoun (se) and the DO (clause) simulta-
neously.
(56) a. Lhev se rozbila.

Die Flasche brach.
The bottle broke.
b. Lhev se rozbila na kousky.
Die Flasche brach in Stcke.
The bottle broke into pieces.
Second step: The case of RPs is subject to a syntactic restriction, the Direct Ob-
ject Restriction (DOR) according to L&RH at least in English, with restric-
tions also in German and Czech.
(57) Direct Object Restriction (DOR):
A resultative phrase can only be dependent on an immediate postverbal NP,
not on the subject of the verb or on oblique arguments.
Examples:
(58) a. *Karel natloukl na hebk do stny.
*Karl schlug auf den Nagel in die Wand. (oblique)
*Charles drove onto the nail into the wall (oblique)
b. Karel zatloukl hebk do stny.
Karl schlug den Nagel in die Wand. (non-oblique)
Charles drove the nail into the wall.
(59) a. *Dora nasytila pee // Dora se nasytila peen (o.k.).
b. *Dora isst satt den Braten
(Nicht akzeptabel in dem Sinn, dass Dora durch das Essen des Bratens
satt wird.)
*Dora eats a roast fed
(Not acceptable in the sense, that Dora is sated from eating the roast)
Third step: For transitive verbs in German RPs can only be attached to the di-
rect object (60b):
(60) a. Dora smazala tabuli doista. (neutral word order)
Dora wiped the blackboard clean.
b. Dora wischte die Tafel sauber.
Dora wiped the blackboard clean.
c. Dora smazala doista TAbuli Dora wiped clean the BLACKboard.
(was wischte Dora sauber?) (what did Dora wipe clean?).
260 Peter Kosta
d. Dora tabuli(= DO-Scrambling) smazala DOISTA

(wie wischte Dora die Tafel?) Dora the board (= DO-Scrambling) com-
pletely obliterated
e. Dora doista SMAzala tabuli (verum Focus)
Dora completely clean | WIped the blackboard
In Czech this rule corresponds to the condition of natural word order alongside
the hierarchy of the theta-roles: AGENS > THEMA > INSTRUMENT | AD-
VERB OF MANNER (cf. 60a). Another order of the thematic hierarchy would
result in another information structure (cf. (60c)-(60e)).
Transitive verbs, whose direct object remains implicit (as in e.g. essen, ko-
chen, fahren to eat, to cook, to drive), are syntactically intransitive and
behave like unergatives (ef. KOSTA 1992, chapter 5).
Unergatives have no direct object, to which an RP can be attached. Howev-
er, unergative resultative constructions are possible:
(61) a. Dora se najedla dosyta.
Dora a sich satt.
Dora ate her fill.
b. Pes zatkal na Fritze.
Der Hund bellte Fritz an.
The dog barked at Fritz.
(62) a. *Dora se j.
*Dora isst sich.
*Dora eats herself
b. *Pes zatkal Fritze.
*Der Hund bellte Fritz.
*The dog barked Fritz
(62) shows, that in (61) se/sich, resp. Fritz is no direct object of the verb. This
holds true generally: Since RP referents unergatives occupy an NP that is not di-
rectly dependant on the verb. Only through the RP is the NP connected to the ac-
tion of the verb. If the NP is a reflexive pronoun se, as in example (61a), the re-
sult will refer to the subject, as is the case of unaccusative verbs (see below).
Fourth step: Passive and unaccusative verbs seem to contradict the DOR. In
their case, a resultative phrase is attached to the subject:
(63) a. Zuby se vyistily a jsou te ist.
Die Zhne wurden sauber geputzt und sind jetzt geputzt.
The teeth were cleaned and are now clean.
b. Dvee se zavraj a jsou te zaven.
Die Tre fllt zu und ist jetzt geschlossen.
The door slams shut and is now closed.
However, the surface subjects are direct object in deep structure, therefore the
DOR (57) remains valid.
Phrases that are attached to the subject of transitive or unergative verbs are
not resultative, but descriptive: (61a) can be interpreted that Dora is full, while
she is eating. Constructions like these are not subject to the DOR in (57).
Fifth step: In the case of unaccusatives, reflexive pronouns or independent NPs
cannot function as an anchor for resultative constructions, as they do in the case
of unergatives (see above, third step)):
(64) a. *Snh roztl ulici na blto / do blta.
*The snow melted the road into mud
*Der Schnee schmolz die Strae matschig.
*The snow melted the road muddy
b. Dv se splilo *na popel.
*Das Holz brannte sich zu Asche.
The wood burned (*itself) to ashes.
The rules of L&RH explain the ungrammaticality of the examples: The verbs in
these sentences are unaccusative. In accord with the formula (52a) they do not
possess external arguments. Verbs like these do not self-motivatedly assign the
case to any NP (Burzios Generalization); the only possibility for an NP to re-
ceive a case is by means of surface object function. There is no grammatical
space for a second NP (die Strae, resp. sich).
Sixth step: The usefulness of DOR in (57) is also dependent on its deducability
from general grammatical principles. L&RH reflect upon this matter thoroughly.
Starting from the syntactic structures of resultative constructions and intransitive
verbs they deduce the so-called Change-of-State Linking Rule, and show that
this rule is encompassing the DOR in (57), cf. (65).
(65) Change-of-State Linking Rule
An NP, whose referent is subject to a change of state that is based on the
process described in the VP, has to be governed by the verb of the VP.
Nominal definition: The word governed is chosen to incorporate not only the
direct object, but also the non-dependent NP in unergative resultative construc-
tions. If the frame theory unlike this paper counts such NPs as direct objects,
the rule will be: An NP, whose , has to be the direct object of the verb of the
VP.
The existence of such a general Linking Rule would permit the assumption
that all RPs are combinable with all unaccusative verbs. The Change-of-State-
Linking Rule, however, only constitutes an essential and not a sufficient condi-
tion. There are additional independent, semantic restrictions, eliminating certain
combinations.
262 Peter Kosta
Seventh step: Several authors criticize the Unaccusative Hypothesis the follow-
ing way: Not all the unaccusative verbs yield the same results to the tests of un-
accusativity. This can e.g. be observed for the resultative construction; not all
the unaccusative verbs can occur in relustative constructions. These verbs be-
long to two subclasses. This involves
x Static verbs, like bleiben (to stay) and
x Verbs of directional movement, like kommen (to come, to arrive)
Those two verb classes are the objects of research concerning the next two sub-
items. First we will provide two examples:
(66) a. *Petr zstal doma hladov
b. *Peter blieb zuhause hungrig.
*Peter stayed at home hungry.
c. *Anna vela bez dechu do pokoje
d. *Anna kam atemlos im Zimmer an.
*Anna came breathlessly to the room
(Not acceptable in the sense that the action of the verb ankommen im
Zimmer results in the state of atemlos.)
The examples can be understood descriptively; in this case, however, they do
not present the same phenomena.
Eighth step: The term verbs of directional movement can be specified to verbs
of inherent directional movement. The end, to which the action of the verb is di-
rected, is a spatial restriction. These verbs cannot occur in resultative construc-
tions due to the fact that in their case the principle of Tenny applies:
(67) Tennys principle:
Only one limiter can be connected to a verbal phrase.
Since RPs depict a distinct restriction of state, an applicable verbal phrase would
be connected to two restrictions. Exceptionally, this is permitted in order to spe-
cify the other, as in (56b), repeated here as (68):
(68) a. Lhev se rozbila na kousky.
b. Die Flasche brach in Stcke.
c. The bottle broke into pieces.
The restriction to one limiter is of mere grammatical nature. Semantically, there
is no objection to a declarative statement like Annas Kommen lie sie atemlos
werden The arrival of Anna did made her breathless.Still,
(66) c. *Anna vela bez dechu do pokoje
d. *Anna kam atemlos im Zimmer an
*Anna arrived in the room breathlessly
can only be interpreted descriptively as Anna war atemlos, als sie ankam. Anna
was breathless when she arrived in the room.
Tennys principle is also essential in order to explain the phenomena that
atelic verbs of movement cannot occur together with RPs, when a further phrase
acts spatially limiting:
(69) a. Wir liefen uns die Fe wund.
We walked our feet sore.
b. *Wir liefen uns die Fe wund in die Stadt.
*We walked into the town our feet sore
Ninth step: L&RH trace the incompatibility of static verbs and RPs, cf. (66a),
back to the fact that there is no linguistic concept of a restricted state. Since stat-
ic verbs describe a state and RPs describe a restriction, they are incompatible.
Thus, in order to exclude various possible resultative constructions correct-
ly, static verbs and verbs of directional movement have to draw on semantic in-
formation. Those are the semantic aspects in L&RHs approach to resultative
constructions.
Summary: Phenomena related to resultative constructions can be explained
by means of the Unaccusative Hypothesis, the Linking Rule (65) and various
additional rules for special cases, provided that one assumes an unaccusative,
resp. an unergative syntactic structure for intransitive verbs. L&RH see no alter-
native to attain an equally useful description without the assumption of specific
syntactic structures. For that reason the deliberations about resultative construc-
tions prove the first part of the Unaccusative Hypothesis in the form used by
L&RH: Unaccusativity is represented syntactically.
4.3.2. Causative Alternation (CAL) as means to determine semantic components
of unaccusativity (Levin and Rappaport Hovav)
Hoekstras approach to resultative construction also takes the grammatical dif-
ferences between unaccusative and unergative verbs as a starting point. He does
not contribute new insights for the discussion of unaccusativity; however, he
serves to support the approach of L&RH by direct comparison to it. Hoekstra
brings together RPs and the NPs that predicate them in resultative constructions
into Small Clauses:
(70) Terry wischt den Tisch sauber.
Terry wipes the table clean.
Terry wischt [SC den Tisch sauber].
Terry wipes [SC the table clean].
Most of the syntactic properties of resultative constructions he can explain with
the properties of the Small Clauses; semantic restrictions are traced back to the
properties of aspect as do L&RH (see point 8 above).
264 Peter Kosta
In contrast to the approach of L&RH, for Hoekstra the argument structure of

a verb does not implicitly remain identical, when adding an RP:
(71) a. [Terry]SUBJEKT wischt [den Tisch]OBJEKT D. VERBS.
a. [Terry]SUBJECT wipes [the table]OBJEKT OF THE VERB
b. [Terry]SUBJEKT wischt [(den Tisch)OBJEKT D. SC sauber]SC.
b. [Terry]SUBJECT wipes [(the table)OBJEKT OF SC clean]SC.
Den Tisch in the resultative construction does not constitute the argument of the
verb, but the argument of the RP sauber. This, in turn, is subordinated to the
verb, but is no direct object; for the reason that unaccusative verbs can occur
with Small Clauses; however, not with direct objects.
For Hoekstra transitive verbs in resultative constructions are syntactically
unergative, since the direct object goes missing. Therefore, Hoekstras ap-
proach is especially elegant for verbs that can be used both transitive and intran-
sitive, because the direct object does not have to be specified:
(72) a. Non-specified object:
Wir trinken. Wir trinken die Kanne restlos leer.
We drink. We empty the jug completely.
b. Specified object:
Ich esse das Ei nicht, ich trinke es.
I do not eat the egg, I drink it.
Ich esse das Ei nicht, ich trinke es leer.
I do not eat the egg I drink it up.
Namely in this case Hoekstra does not have to pay attention as to whether die
Kanne could be an object of the verb or not, because he treats both cases equal-
ly. L&RH regard such verbs as verbs able to be classified multiply, having a
transitive and a unergative form. Therefore, these verbs present a problem for
their theory.
Hoekstra remarks that for verbs in resultative constructions not only the ar-
gument structure but also the manner of action changes: Most transitive and un-
ergative verbs that occur in resultative constructions are Activity Verbs, i.e.
verbs of an unrestricted action such as Tisch-wischen. If the argument structure
of Activity Verbs is changed, by a subordinated Small Clause that describes a
state of result rather than by a subordinated direct object, the Activity Verbs turn
into Accomplishments, i.e. into verbs that aim at a final state such as sauber-
wischen.
The comparison of Hoekstra with L&RH is interesting: For Hoekstra a syn-
tactic process influences the semantics of the verb, of the manner of action,
while for L&RH the syntax is determined by the semantics.
Hoekstra assigns a complete Small Clause to unergative and transitive verbs:
(73) a. Transitive: Terry [VP wischte [SC den Tisch sauber]].

a. Transitive: Terry [VP wiped [SC the table clean]].
b. Unergative: Der Hund [VP bellte [SC ihn wach]].
b. ?Unergative: The dog [VP barked [SC him awake]].
On the other side, in unaccusative and passive resultative constructions, the ar-
gument of RP is an NP-trace, i.e. a pronoun invisible on the surface, which re-
fers to the already mentioned noun:
(74) a. Unaccusative: [Karl]i [VP strzte [SC ti zu Tode]].
a. Unaccusative: [Charles]i [VP fell [SC ti to death]].
b. Passive: [The table]i [VP was wiped [SC ti clean]].
Signs and symbols: t trace (NP-trace), i index, marking the reference of NP
and t.
The example (74a) corresponds in my opinion with the examples (11a) and
(11b) here repeated as (11) a, b:
b. Karel upadl na zem. / Karel se svalil na zem.
Charles fell to the ground. Charles let himself fall to the ground
Both cases deal with unaccusative, anti-causative verbs, whereas Hoekstras
analysis can be adopted:
(74) Unaccusative: (a) [Karl]i [VP strzte [SC ti zu Boden]].
Unaccusative: (a) [Charles]i [VP fell [SC ti to the ground]].
Unaccusative: (b) [Karel]i [VP sei svalil [SC ti na zem]].
L&RH use the phenomena of the so-called Causative Alternation (CAL); in or-
der to investigate which semantic components denote a verb as unaccusative
resp. unergative. The CAL appears partially different in German and Czech than
in the English language. In the following L&RHs examples, as far as possible,
will be transferred to German and Czech. By this means the validity of their
thoughts for these languages is to be examined temporarily.
CAL means the ability of certain intransitive verbs to also occur in a causa-
tive form; this form bears the meaning that corresponds approximately with the
below formula (75):
(75)
The Causative Alternation:
External Argument VP Internal Argument
b. Agent V-transitive Theme (causative)
c. Theme V-intransitive (anticausative)
x causes [intransitive verb]
266 Peter Kosta
In English as in German the two formulas are mostly morphologically identical;

in Czech a reflexive marker and/or a prefix marker as CAUSE affix is added:
The CAL is considered as test for unaccusativity by various authors. How-
ever, L&RH prove that not all unaccusative verbs take part, and that on the other
hand several unergatives partake in a CAL-like process. The latter, though, is
clearly distinguishable from the real CAL, as explained below.
With the criterion takes part, resp. does not take part in CAL L&RH sub-
divide the unaccusative verbs into two subclasses: alternating unaccusative
(AU)- verbs and non-alternating unaccusative (NAU-) verbs:
(76) AU-Verb:
a. causative: Pat zerbrach die Stange. Pat zlomila ty.
a. Pat broke the rod.
b. intransitive: Die Stange zerbrach. Ty se zlomila.
b. The rod broke.
(77) NAU-Verb:
a. causative: *John blieb mich zuhause. John mne zstal doma.
(In the sense of: John verursachte, dass ich zuhause blieb.)
a. *John stayed me at home.
(In the sense of: I stayed home because of John./John caused me to stay
home.)
b. intransitive: Ich blieb zuhause. Zstal jsem doma.
b. I stayed home.
Excursus:
The distinction between brechen and zerbrechen is not a distinction of causativi-
ty, as one might be tempted to assume:
(78) a. Ich brach den Stock entzwei. / Der Stock brach entzwei.
Zlomil jsem ty vejpl. / Ty se zlomila vejpl.
I broke the rod in two. / The rod broke in two.
b. Ich zerbrach den Stock. / Der Stock zerbrach.
Zlomil jsem ty. / Ty se rozlomila.
I broke the rod. / The rod broke.
It is much rather the case that the particle zer- substitutes for the RP, demanding
a verb such as brechen as an argument:
(79) a. Verb without RP: ?Ich brach den Stock.
I broke the rod.
b. Verb with RP: Ich brach den Stock entzwei.
I broke the rod in two.
c. Particle verb: Ich zerbrach den Stock.
I broke the rod.
d. Particle verb with RP: *Ich zerbrach den Stock entzwei.

I broke the rod in two.
In the above example Tennys principle applies again. Particle verbs (and also
prefix verbs such as absgen, rozlomit) at least follow several of the up to now
established rules. A study of whether and how prefixes and particles in German
and Czech influence the phenomena in reference to unaccusativity would be
worthwhile, however, would go beyond the scope of the present paper. In this
context we refer to the book of MARCEL DEN DIKKEN (1995).
Since the here mentioned authors do not elaborate on such verbs, I will try
to avoid using them in further examples, since I do not know, whether and
which additional phenomena could occur in their case. If such a verb appears in
the following examples, there will exist at least in English a normal verb with
the same characteristics. Several English AU-verbs correspond in German to
pairs of a causative and a reflexive verb. These pairs behave like AU-verbs in
many respects:
(80) Die Tre ffnete sich. / Katja ffnete die Tre.
The door opened. / Katja opened the door.
Also for: sich rasieren, setzen, waschen, (to shave,to sit,to wash)
At least at the surface both forms of verbs are transitive. Whether and to which
extent the reflexive form behaves unaccusatively, shall be illustrated by means
of four tests (see 4.1-4.2):
(81) Reflexive Verbs choose haben as perfect auxiliary:
Die Tre hat sich geffnet.
The door opened.
(82) Reflexive verbs are available to the formation of the impersonal passive:
(?)Erst wurde gefrhstckt, dann wurde sich rasiert...
(?)Firstly, it was eating, and then it was shaving
(83) The participe II (past participle) can be assigned attributively to the subject
of the reflexive sentence. It is not necessarily understood as passive:
Unendlich langsam ffnete sich die Tre. Und wieder eine Ewigkeit spter
erst kam durch die geffnete Tre...
Infinitely slowly the door opened. And only after another eternity entered
through the opened door
(84) Reflexive verbs in contrast to transitive verbs are not availabe to the -er
nominalisation:
??Der Rasierer dort (hardly acceptable in the reflexive sense)
??The shaver over there
From my current point of view this behavior can be explained by the following:
Perfect auxiliary and impersonal passive are tests for surface characteristics. On
268 Peter Kosta
the surface those verbs are transitive, that is why haben is chosen and the imper-
sonal passive is (restrictedly) possible. However, the attributive participle II
(past participle) die geffnete Tre and the impossibility of er nominalization
are test results that are easiest explained by the underlying unaccusativity. As
evidence that the participle-II-test does not check verbs for surface-
unaccusativity, I will therefore choose example (80). Nevertheless, as shown
above, this test is equally not sufficient to prove underlying unaccusativity, es-
pecially not in Czech.
The formation of the participle II (past participle) thus takes place some-
where in the middle, after the composition of verbs to complex predicates, but
before the transitive surface structure is assigned to reflexive verbs such as sich
ffnen. The exact localization depicts another point of further research.
L&RH demonstrate that in the case of a linguistic rule of derivation a re-
striction to the area of validity was to be explained more conclusively than an
extension. This means for the CAL: The intransitive variant of the verb is de-
rived from the causative, since the number of possible subjects of the former de-
picts a subset of the objects of the latter.
(85) a. Antonia brach den Stock entzwei.
Antonia broke the rod in two.
Der Stock brach entzwei.
The rod broke in two.
b. Antonia brach ihr Versprechen/den Weltrekord.
Antonia broke her promise / the world record.
*Ihr Versprechen/der Weltrekord brach.
*Her promise/ the world record broke.
The rule of derivation therefore works in the direction causative intransitive
and has to provide both: to suppress the appearance of the causer-argument for
the intransitive variant and to restrict the area of validity:
(86)
Causative form:
Antonia broke a) the vase b) her promise
Restriction:
Antonia broke a) the vase b) - - - - - -
Suppressing the causer:
------ broke the vase
Intransitive form: The vase broke.
L&RH assume the following semantic structure for the AU-verbs (here using
the example of break):
(87) break: [[x DO-SOMETHING] CAUSE [y BECOME broken]]
In the intransitive form the causer-argument is not expressed; however, the un-
derlying structure remains the same. If, however, in the causer-brackets [x DO-
SOMETHING] a causing action is specified, no component in brackets can be
omitted.
Next L&RH examine which components of verb meaning enable, resp.
hinder a verb, to partake in CAL.
(88) A possible thesis:
All verbs expressing change of state take part in the CAL.
L&RH object by pointing out that indeed there is a close relation between
AU-verbs and verbs expressing change of state; however,
Not all intransitive verbs expressing change of state have transitive causative va-
riants:
(89) aufblhen, errten, sterben... (to flourish, to blush, to die) e.g.:
a. Ich errtete. / *Karin errtete mich.
I blushed. / *Karen blushed me.
b. Valerys zervenala / se zaervenala /
Valerys blushed / * Valerys blushed herself
c. *Petr z(a)ervenal Valerys
*Peter made Valerys blush
Not all transitive causative verbs expressing change of state have an intransitive
variant:
(90) schneiden, tten, bauen... (to cut,to kill,to build) e.g.:
a. Der Bcker schneidet das Brot. / *Das Brot schneidet.
b. Peka krj chlb. / *Chlb se krj.
The baker cuts the bread. / *The bread cuts.
c. Vrah zabil mladou dvku./ *Mlad dvka *(*se) zabila4.
(in the sense of byla zabita vrahem)
d. Inenr stav dm. / *Dm se stav inenrem5.
There are intransitive AU-verbs, which do not express change of state.
(91) luten: Er lutete die Glocke. / Die Glocke lutete.
Zvonil na zvony. / Zvony zvonily.
to ring:He rang the bell. / The bell rang.
L&RH continue the analysis of the so far mentioned data and complete it with
additional; here we will not go into further details of the individual steps of the
analysis. L&RH conclude that for the characterization of phenomena concerning
__________
4 This sentence is only grammatical in the meaning of suicide, which in this case is not the
intention.
5 This sentence is not the reading variant of the impersonal passive, which however depicts
a possibility: dm se stav.
270 Peter Kosta
causation internal and external cause are determining. The below example of
the verbs brennen and schneiden demonstrates this characterization:
(92) a. Das Feuer brannte. / Ohe /tbork hoel.
The fire burned.
*Die Camper brannten das Feuer.
*Tbornci hoely ohe /tbork.
*The camper burned the fire.
b. Die Bltter brannten zu Asche. /Listy shoely a na popel.
The leaves burned to ashes.
Die Camper brannten die Bltter zu Asche. / Tbornci shoeli listy a
na popel.
The camper burned the leaves to ashes.
c. *Das Brot schnitt in Scheiben. / Chlb *(se) nakrjel na krajky (as
impersonal passive o.k.)
*The bread cut into slices.
Der Bcker schnitt das Brot in Scheiben.
Peka nakrjel chlb na krajky.
The baker cut the bread into slices.
In (92a) brennen signifies to emit light and warmth; this process cannot be
conceived as caused externally, as the following sentence demonstrates: ??Ohne
meine Hilfe htte das Feuer nicht Licht und Wrme abgestrahlt. Without my
help the fire would not have emited light and warmth. brennen here behaves
purely intransitive. In (92b) brennen signifies to convert into another state.
This process can be caused externally as well as internally. Ich berfhrte die
Bltter in einen anderen Zustand. I transfered the leaves into another state. Or:
Die Bltter nderten beim Brennen ganz von selbst ihren Zustand. During the
course of burning the leaves changed their state all by themselves. brennen
therefore can behave both transitive and intransitive. Finally, in (92c) one can
not picture that the bread itself by means of a sharp object falls into slices. One
can only assume an externally caused process. schneiden therefore only occurs
transitive. The consequence: In the case of an external cause, the verb is transi-
tive (92b) and (92c); if it additionally can be conceived as internally caused, it
has an additional intransitive form (92a), and is thus and AU-verb. In the case of
an only internally cause the verb is only intransitive (92a).
However, L&RH describe the situation slightly differently: Only-transitive
verbs depict situations that are not spontaneous, without the intentional action of
an Agent. Only-intransitive verbs in contrast are verbs that only occur sponta-
neously. Possibly, this distinction can be illustrated by the following pair of ex-
amples (93a) vs. (93b) and (93c) vs. (93d):
(93) a. Devorubec pokcel strom. /Der Holzfller fllte den Baum

The woodcutter cut down the tree.
b. Vichice vyvrtila strom. /Der Sturm entwurzelte den Baum.
The storm uprooted the tree.
c. *Strom se pokcel. / skcel. / Der Baum strzte um.
The tree fell down.
d. Strom se vyvrtil./ Der Baum entwurzelte.
The tree was uprooted.
Apparently the paradigm Bume fllen has a suppletive paradigm available, in
the meaning of CAU: pokcet, fllen, to fell, to cut down vs. skcet se,
umstrzen durch fallen, to fall; a direct anticausative equivalent to the causa-
tive verb pokcet therefore can only be expressed by the intransitive verb skcet
se. Naturally, the question is whether similar to the criteriological determina-
tion of Aspect pairs one can still refer to a CAU-pair, especially when there is
neither complete formal nor lexical conformity. In my opinion it is an issue of
personal preference whether or not to introduce spontaneity as an additional cha-
racteristic the results remain unaffected.
Finally, I want to point out the following distinctions: Internal cause is
closely connected to agentivity and Kaufmanns internal control. However, it is
not identical with either of them, as demonstrated by verbs of internal cause,
such as errten or zittern, which can neither be termed internally controlled nor
agentive. This analysis does not completely satisfy L&RH, since according to
their opinion, this means the following sentence would be impossible:
(94) Ich warf das Glas an die Wand, und es zerbrach.
I threw the glass against the wall, and it broke.
Namely here, the breaking (zerbrechen) is clearly caused by the action of an
Agent, and therefore zerbrechen would not be allowed to be used intransitively.
This seems to be problematic, since in my opinion zerbrechen describes a
process that occurs independently of the throwing (werfen). Also the following
situation is easily conceivable: Ich warf das Glas an die Wand, und es zerbrach
nicht. However, the analysis that L&RH enclose to the example, applies not-
withstanding; it is an important part to a complete description of causativation.
In order to explain (94) L&RH point out a characteristic that is more fundamen-
tal than the internal resp. external cause. By means of this characteristic all of
the above mentioned cases can be classified correctly: Causative predicates in-
herently consist of a causing and a caused event. If now a causative verb does
not reveal anything about the causing event, then exactly does it belong to the
AU-verbs. The de-transitivation takes place by lexical detaching of the causer
position for the projection of Lexical Semantic Representation (LSR) to the ar-
gument structure. This detaching however is blocked by the indication about the
272 Peter Kosta
type of causing event. This facts can be demonstrated by the following illustra-
tions:
(95) Transitive: break
LSR [[x DO-SOMETHING] CAUSE [y BECOME BROKEN]]
Detaching not necessary
Linking
Arg.str. <x> <y>
(96) Intransitive: break
LSR [[x DO-SOMETHING] CAUSE [y BECOME BROKEN]]
Detaching x is bound
Linking
Arg.str <y>
(97) cut can only be used transitively:
LSR [[x USE-A-KNIFE] CAUSE [y BECOME CUT]]
Detaching is blocked
Linking
Arg.str. <x> <y>
The argumentation will be concluded with a suggestion for the semantic struc-
tures of verbs in (92a) vs. (92b); with them and the above considerations the be-
havior in (92) can just as well be explained as with the characteristic [internal
cause]:
(92) a. [[ ] [DIE BLTTER STRAHLENLICHT & WRME]]
[[ ] [THE LEAVES RAYS OF LIGHT & WARMTH]]
b. [[x DO-SOMETHING] CAUSE [DIE BLTTER BECOME ASCHE]]
[[x DO-SOMETHING] CAUSE [THE LEAVES BECOME ASH]
c. [[x BRENNEN-MACHEN] CAUSE [DIE BLTTER BECOME
[[x BURN-MAKE] CAUSE [THE LEAVES BECOME
ASCHE]]
ASH]
Two questions remain unsolved: Firstly, why is there a class of unaccusative
verbs that do not participate in the CAL?
This class consists of verbs of existence, of appearance (in the meaning of
auftauchen, to appear) and of disappearance. L&RH assume that these NAU-
verbs have two underlying arguments: a first one of entity that exists, appears or
disappears; and a second one that describes the spatial or temporal informa-
tion of appearance. Although these verbs have two arguments like the AU-verbs,
they do not occur causative, since the event described is not interpreted as
caused. Secondly, what is the case for unergative/causative verb pairs? L&RH
reserve the term Causative Alternation for cases that include an unaccusative
verb after the pattern of break. However, there are also causative pairs with a
unergative verb (in the following referred to as causative pairs in contrast to

AU-verbs).
An example for a causative pair is (98):
(98) a. The horse jumped over the fence.
b. The rider jumped the horse over the fence.
In the same way, most agentive verbs in the manner of motion can be used: run,
march, However, this does not apply to verbs of inherently directed motion
such as come and go. In German, as far as I can see, all verbs of motion are cau-
sativated by means of adding the auxiliary verb lassen: Der Reiter lie das
Pferd ber den Zaun springen. This construction displays other characteristics
than the English causative pairs, because ich lie ihn kommen is by all means
possible, while *I came him is not. However, the English language also displays
constructions similar to German. In contrast to AU-verbs causative pairs are se-
verely limited, e.g. through the intentional cause of an animate Agens: cf.
(99) a. CAL: The vase shattered into little pieces.
The lightning shattered the vase into little pieces.
b. causative pair:
The horse jumped over the fence.
*The lightning jumped the horse over the fence.
Causative pairs also show a semantically different structure than AU-verbs:
(100) a. causative form of a causative pair:
a causes b to act, so that something is happening to b.
b. causative form of a AU-verb:
a causes, so that something is happening to b.
Because of this distinction L&RH regard the causative pairs as a separate phe-
nomenon and not as an counter-example for the fact that only externally caused
verbs partake in the CAL.
There is a further class of unergative verbs with causative variants/by-forms,
for which the same considerations about the verbs of manner of motion apply:
the verbs of emission (ring, bleed, sneeze, ). Both classes are more severely
limited in their causative variant than in their intransitive; for that reason L&RH
consider the intransitive variant as underlying in contrast to the AU-verbs.
5. Causative Alternation and argument reduction for transitive
causatives (and the syntax of marked anti-causatives)
In secondary literature to Causative Alternation parallels between three-figure
(triadic) verbs and transitive CC have been determined in a vast amount of lan-
guages (see DEN DIKKEN 1995: 239).
In this section I will go into detail about two cases that are connected to the
assignment of Case and to the problem of deduction in CC and AC.
274 Peter Kosta
5.1. Dative-/accusative alternation of the embedded subject (Causee) in transi-

tive causatives and ditransitive sentences
In several languages the Case of the embedded subject (the Causee) of a tran-
sitive CC is identical with the Case of the Goal Argument in a datival preposi-
tional phrase, in other languages the Case alternates between dative and accusa-
tive. Among others the French (101) and the Sanuma (102) belong to those lan-
guages that do not permit a double accusative object in ditransitive structures:
(101) a. Jean offrira des bonbons ses enfants.
John will give candies to his children
John will give his children candies.
b. Jean fera manger des bonbons ses enfants.
John will make eat candies to his children.
John will let his children eat candies.
(102) a. Ipa hao-n hama te niha masulu kk toto- ki kite
My father-AG visiters 3SG to pearls 3DL give-FOCFUT
My father will give pearls to the visitors.
b. Kamisa n setenapi te niha manasi sa
1SG-AG non-Indians 3SG to guan- bird 1SG
ta-ma-na-ni ke.
see-CAUS-EXT-FOC PAST
I let the non-Indian see the guan-bird.
I showed the guan-bird to the non-Indian.
As becomes apparent, the German and Czech equivalent of the examples (101)
differs depending on the verb. The triadic verbs geben/dt and anbie-
ten/nabdnout require for their transitive variant in any case a chronological or-
der of IO (DAT) DO (ACC) (cf. (103a), (104a)); while the causative transitive
equivalent to essen lassen in German (103b) and Czech (104b) not only permits,
but requires the double accusative.
(103) a. Jan bot seinen Kindern Bonbons an.
Jan offered candies to his children.
Jan bot Kindern-IO(DAT) Bonbons-DO(AKK) an
Jan offered children-IO(DAT) candies-DO(ACC)
b. Jan lie seine Kinder Bonbons essen.
Jan let his children candies eat
Jan-NOM KAUS AKK AKK AKK essen
Jan-NOM CAUS ACC ACC ACC eat
c. Jan lie *seinen Kindern Bonbons essen
Jan let *his children candies eat
Jan-NOM KAUS DAT DAT AKK essen
Jan-NOM CAUS DAT DAT ACC eat
d. Jan gab seinen Kindern Bonbons (zum) essen

Jan gave his children candies (to) eat.
Jan-NOM KAUS DAT DAT AKK PrpDAT essenVNOM
Jan-NOM CAUS DAT DAT ACC PrepDAT eat VNOM
(104) a. Jan nabdl/dal svm dtem bonbony.
Jan-NOM an-bot/gab Kindern-IO(DAT) Bonbons-DO(AKK)
Jan-NOM offered/gavechildren-IO(DAT) candies-DO(ACC)
Jan bot seinen Kindern Bonbons an
Jan offered his children candies.
b. Jan nechal sv dti jst bonbony.
Jan lie seine Kinder essen Bonbons
Jan let his children eat candies
NOM KAUS AKK AKK essen AKK
NOM CAUS ACC ACC eat ACC
Jan lie seine Kinder Bonbons essen.
Jan let his children eat candies.
c. *Jan nechal svm dtem jst bonbony.
Jan-NOM KAUS DAT DAT essen AKK
Jan-NOM CAUS DAT DAT eat ACC
d. Jan dal svm dtem jst bonbony.
Jan-NOM gab seinen Kindern-IO(DAT) zu essen Bonbons-
Jan-NOM gave his children-IO(DAT) to eat candies-
DO(AKK)
DO(ACC)
MARANTZ (1984) and BAKER (1988) explained the relation between triadic and
transitive causative consructions the following: They explain the dative-PP in
triadic structure in the CC for the CAUSEE as inserted CASE-marker; while
they do not explain triadic verbs the same way, cf. (105):
(105) Cross-linguistic generalisation I (no case alternation):
The Case form of the causee of a transitive causative
Construction is identical with the Case form of the Goal
of a prepositional triadic construction.
REED (1989) analyses the preposition that precedes the CAUSEE as a dummy
element, which is employed for the purpose of Case assignment only. By the
assumption of a void preposition that is employed, the correlation (150) is in a
certain syntactic context and thematically compatible with this context (i.e. the
GOAL and the CAUSEE have to receive the same Theta-role THEMA, cf. also
REED 1992: 171).
According to DEN DIKKEN (1995: 241) this notion is problematic because
case-assigning prepositions usually do not assign Theta-roles:
276 Peter Kosta
If, however, Case-marking prepositions do not play a thematic role (i.e. do not have
a -role of their own to assign), it is difficult to imagine how this requirement of
thematic compatibility can be formulated. A constraint on -role matching is inap-
plicable: since the dummy preposition is thematically inert, there is nothing to match
the causees -role with. It seems that a requirement of thematic compatibility can
be operative only if one inflates the lexical entry of dummy Case-markers with a di-
acritic label specifying the thematic context(s) with which the non--assigning
dummy is compatible. Clearly, this is an undesirable complication. (DEN DIKKEN
1995:241)
For triadic structures of ditransitive verbs den Dikken assumes a similar struc-
ture as for the causative variant, i.e. a DATIVE-Small Clause in D-structure, cf.
(106) vs. (107):
(106) Ditransitiv Verbs:
[VP Vtriadic [SC1 Spec [VP V [SC2 Spec [XP X[SC3 embedded object [PP
causee]]]]]]]
(107) Causative Verbs:
[VP Vcaus [SC1 Spec [VP V [SC2 Spec [XP X [SC3 embedded object [PP
cause]]]]]]]
Thus, the causatives in French and Sanuma result in the following structure
(108):
(108) VP
V SC
Spec VP
V SC
Spec XP
X SC
NP PP
P NP
French faire manger des bonbons ses enfants

Sanuma ta ma manasi niha setenapi
make eat candies to his children
It remains to explain the structures of periphrastic causative constructions in
Czech and German. As is clearly identifiable, the Case changes according to the
matrix verb. If lassen/nechat is the causative verb of the matrix sentence, this
verb apparently assigns the ECM-Case Accusative to the CAUSEE modula
ECM-Case Assignment (thus, like an ECM-verb or AcI-Verb) and the structural
Case Accusative to the embedded verb [Veffect]. If the triadic verb geben/dt is
the causative verb of the matrix sentence, this verb assigns the Case Dative and
the Case Accusative to the embedded verb. The validity of both the analysis of
den Dikken and ours, is among others proven by the data of the Clitic Climbing
and the anaphorical binding resp. control, cf. (109) for French and (110) for
Czech, resp. (111) for German:
(109) a. Elle fera manger des bonbons aux enfants.
She will-make eat candies-ACC to the children-DAT
b. Elle les fera manger aux enfants.
She-NOM they-ACC will-make eat to the children-DAT
c. Elle leur les fera manger.
She-NOM them-DAT the-ACC will-make eat
d. *Elle les leur fera manger.
She-NOM they-ACC them-DAT will-make eat
In French the clitic pronoun of the CAUSEE (IO) has to climb higher than the
weak pronoun with the theta-role THEMA (DO); therefore the word order IO-
DO has priority. The explanation is purely syntactical: The clitica receive their
case in the basis-generated case position, i.e. in the respective case projection of
the DPs of the category they are governed by. They can easily be raised across
the causative composed by the Baker Incorporation, because neither the causa-
tive nor the lexical verb constitutes a barrier of minimality.
The mechanism of Clitic Climbing for the Czech examples is slightly more
intricate:
(110) a. Maminkai nechala svi dti jst
Mother-NOM let-past her children-ACC eat
svi/j bonbony.
her candies-ACC
b. *Maminka je je nechala jst.
Mother-NOM theyCL-ACC theyCL-ACC let eat
The mother let they-(the children) they (the candies) eat.
c. Maminka je nechala je jst.
Mutter-NOM sieCL-ACC lie sieCL-ACC essen
Mother-NOM theyCL-ACC let theyCL-ACC eat
d. Maminka je nechala je jst.
Mother-NOM theyCL-ACC let theyCL-ACC eat
e. *Maminkai sv bonbonyj nechala jst sv dtii.
Although the Czech language allows Clitic Climbing across certain bridge
verbs, up to here restrictions of the type (110b) have not, or in my opinion not
278 Peter Kosta
correctly, been explained. DENISA LENERTOV (2004: 154) mentions a phono-

logic constraint of the 3rd Ps. Acc/Gen masculine ho which cannot appear in-
itially in the cluster. Whereas the co-occurrence of genitive and accusative ar-
guments requires the order ACC-GEN (38), if the accusative argument is ho
In this context she also mentions a causative verb that moves the mentioned
pronoun as DO of the 3rd Ps. Acc/Gen masculine ho above the female counter-
part; and therefore, independently of whether it is governed by the same verb or
by two different verbs, the result would be ungrammatically:
([38]) a. *Stejn ho ji nenechali dokonit.
Anyway itACC.M itACC.FEM not-letPL finish
b. Stejn ji ho nenechali t dokonit t.
Anyway itACC.FEM itACC.M not-letPL finish
LENERTOV (2004:154)
However, I claim that the strict order of clitics in these examples results from
the hierarchy of the sentence structure of causative constructions. The lower
verb dokonit assigns the case to the clitic ho, while the causative verb nenechali
assigns the case to the upper clitic ji by means of ECM-marking. The upper cop-
ies of the clitics on the PF are spelt out by means of prosodic rules postsyntacti-
cally.
Examples of the type (110b) prove on the one hand, that the restrictions of
Clitic Climbing not only affect the clitic ho in the 3rd Ps. Sg. Acc, but also two
clitics in the same (homophone) form, when they are to come to stand next to
each other directly. We assume that the restrictions are only concerned with the
Case assignment, and claim that both clitics in (110b) are subject to the *Case
filter, i.e. no case was assigned to them, since they are not located in the corres-
ponding governed case position, because their trace cannot be governed strictly.
The reason is that the causative verb, i.e. the matrix verb [VCaus], has to assign
its case to the CAUSEE with the Theta-role GOAL or ADRESSEE while the
caused verb [VEffect] has to turn back its case to the DO of the verb with the
Theta-role THEME. The validity of this explanation is proven by the binding-
and control verb data in (110a) or (110e). In (110a) the subject of the matrix sen-
tence (maminka) can bind both the first anaphora (sv dti) and the DO of the
embedded Small Clause (svi/j bonbony) by means of long distance binding (i.e.
both the children and the bonbons are in possession of the mother).
Another interpretation or analysis, even more obvious for reasons of control,
states that the causative verb functions like an object-control-verb, so that the
children as object of the matrix sentence and subject of the infinitive sentence
can be coreferential, i.e. the anaphora sv bonbony is in the possession of the
children and not of the mother.
The examples in (110e) provide further prove for the analysis that both
verbs with their arguments project a strict hierarchy of the sentence structure;
whereby both the matrix verb lassen per ECM to its argument and the infinitive
verb of the Small Clause per structural case assignment to its internal DO-
argument assign the case accusative. For this reason we reject the analysis of
causativation by GREWENDORT (1994) as incorporation following BAKER (1988)
for both German and Czech, but rather argue for a bi-sentential structure (as in
(111):
(111) [VoiceP DP [Maminka]l nechalaj ECP [vPCAUS [svi dti] vCAUSj
Mutter-NOM lassen-Past-FEM
Mother-NOM let-Past-FEM
[RootP [sv bonbony]k jst tk]]]].
ihre Kinder-ACC essen ihrei/j Bonbons-ACC
her children-ACC eat theiri/j candies-ACC
5.2 Motivation of the derivation of marked anti-causatives with non-thematic
reflexive pronouns and applicative datives
The following section discusses the marked anti-causatives from the perspective
of non-thematic reflexive pronouns and the applicative datives.
Chomsky assumes that next to CP there are only transitive vPs in sentence
structure. (Below we refer to this phase as VOICE/DIATHESES-phase accord-
ing to SCHFER 2008: 263, because as functional phase it has importance for the
causative-anticausative-dichotomy). In contrast, the passive and unaccusative
vPs are no phases according to Chomsky. Since the marked anti-causatives of
the type auto se rozjelo setrvanost, das Auto fuhr durch Trgheitskraft los
(caused by inertia force the car started driving), lod se potopila, das Boot
sank, (the boat sank), but also examples of type (18a) die Tr lie sich leicht
ffnen, (18b) dvee se lehce otevrali, can be classified semantically as unaccus-
atives, with the simultaneous assumption that the reflexive marker refers to tran-
sitivity (for more detail see SCHFER 2008: 263), we are confronted with the
theoretical problem of how to solve the phase-theoretical status of such anti-
causatives. LEGATE (2003) on the other hand assumes that not only transitive but
all verbal phases are phases, including anti-causative and semantic unaccusative
phases. If this is correct, we have to explain how TENSE checks its characteris-
tics against the Theta-role THEME within the vP phase in unmarked or marked
unaccusatives. This refers to both the marked anticausatives (with reflexive pro-
nouns, type otevt se, rozbt se) and unmarked anticausatives of the type fallen,
schmelzen; padat, tt (to fall, to melt).
5.2.1 The theory of (anti-)causative and passive according to Alexiadou (2006a,
b)
We already discussed the most important problems of Causative Alternation. At
this point finally I want to go into detail about two recent theories: ALEXIADOU
et al. (2006a, b) and SCHFER (2008). The discussion of the Causative Alterna-
280 Peter Kosta
tion in 1.3 was mainly concerned with the semantic analysis of both classes of
unaccusative verbs (AU and NAU). It remains to clarify how anti-accusatives
behave towards the passive. As repeatedly noted in respective literature anti-
causative and passive differ in the following aspects (cf. MANZINI 1983, MA-
RANTZ 1984, JAEGGLI 1986, ROEPER 1987, BAKER, JOHNSON & ROBERTS 1989,
KOSTA 1992, LEVIN & RAPPAPORT HOVAV 1995, REIHART 2000 and SCHFER
2008): (i) modification and control, (ii) verbal restrictions in the case of passiva-
tion.
(i) Modification and control
Passives, but not anticausatives can be modified by an Agens by-phrase (in
Czech Instrumental-NP), by Agens-oriented adverbs (subject adverbs) and by
control in embedded final sentences:
Passive-agens vs. *anti-causative-agens:
(112) a. The boat was sunk by Bill.
b. *The boat sank by Bill.
(113) a. Das Boot wurde durch Bill versunken.
b. *Das Boot sank durch Bill.
(114) a. Lo byla potopena Billem.
b. *Lo se potopila Billem.
Agens-oriented adverbs
(115) a. The boat was sunk by purpose.
b. *The boat sank on purpose.
(116) a. Das Boot wurde absichtlich versenkt.
b. *Das Boot sank absichtlich.
(117) a. Lo byla potopena naschvl.
b. *Lo se (mu) potopila naschvl.
Control in embedded final sentences
(118) a. The boat was sunk to collect the insurance.
b. *The boat sank to collect the insurance.
(119) a. Das Boot wurde versenkt, um eine Versicherungsprmie zu kassieren.
b. * Das Boot sank, um die Versicherungsprmie zu kassieren.
(120) a. Lo byla potopena, aby dostali penze od pojiovny.

b. *Lo se potopila, aby dostala/dostali penze od pojiovny.
c. *Lo se potopila pro penze pojiovny.
(ii) Verbal Restrictions
Theoretically, every transitive verb can be passivated. However, a minor sub-
groub of transitive verbs can also form anti-causatives. This will be demonstrat-
ed with the example of three verb groups: the verbs brechen, break, rozbt per-
mit both passivation and causativation; schneiden, cut, krjet, sthat, ezat per-
mit passives but no anti-causatives, although as verbs of change of state they
likewise fulfill the semantic conditions for Causative Alternation. Finally, transi-
tive verbs such as lesen, read, st are indeed capable to be passivized; but a
verb that does not change a state can never form an anti-causative:
(121) a. Bill broke the glass.
b. Bill zerschlug das Glas.
c. Bil rozbil sklo.
(122) a. The glass was broken by Bill.
b. Das Glas wurde von Bill zerschlagen.
c . Sklo bylo rozbito Billem.
(123) a. The glass broke.
b. Das Glas zerschlug.
c. Sklo se rozbilo.
(124) a. The baker cut the bread.
b. Der Bcker schnitt das Brot ab.
c. Peka krjel chlb.
(125) a. The bread was cut by the baker.
b. Das Brot wurde vom Bcker abgeschnitten.
c. Chlb byl krjen pekaem.
(126) a. *The bread cut.
b. *Das Brot (zer)schnitt.
c. *Chlb se ukrojil.
(127) a. John read the book yesterday.
b. John las das Buch gestern.
c. John etl knihu vera.
(128) a. The book was read yesterday by John.
b. Das Buch wurde von John gestern gelesen.
c. Kniha byla tena vera Johnem.
282 Peter Kosta
(129) a. *The book read yesterday.

b. *Das Buch las gestern.
c. *Kniha se vera etla.
(The reading as impersonal passive is possible in this case)
Concerning the two differences (modification and control)6, it is generally
agreed that the reason is the presence vs. absence of implicit external arguments
in passives vs. anti-causatives. The fact that passives contain an implicit external
argument, can according to the theory of B/J/R (1989) be explained in the
way that n-/t-passives contain an implicit external argument, which through the
modification (by a by-phrase or an Agens-oriented adverb) can again be recon-
structed and control a final sentence; this is not an option for anti-causatives, be-
cause an anti-causative does not contain an external argument (remember that
most anti-causatives are derived from unaccusative verbs).
While these assumptions are consensus, the question in which form this im-
plicit external argument in the passive-diathesis is to be represented poses a
problem.7 SCHFER (2008: 117) assumes a configurational Theta-Theory, which
predicts that an implicit argument has to be represented syntactically, e.g. as a
thematic characteristic in Voice (cf. EMBICK 2004) or as covert pronoun in the
specificator position of a passive-voice-phrase (cf. STERNEFELD 1995, BORER
1998, RAMCHAND & SVENONIUS 2004). Furthermore, there is disagreement
about why anti-causatives lack an implicit external argument. SCHFER (2008:
117) discusses two logically possible approaches: (i) According to one approach
anti-causatives lack an implicit external argument, because they are basically
monadic. The causative alternant derives from an anti-causative/inchoative al-
ternant via the operation of causativation (causativization), i.e. via adding a
causative predicate (CAUSE) to the semantic decomposition. This operation is
demonstrated by (130):
(130) a. breakincho : x [BECOME broken(x)]
b. breakcaus : yx [(y) CAUSE [BECOME broken (x)]]
According to the (ii) second approach alternating verbs are inherently dyadic
predicates. Anti-causatives lack an implicit argument because of the lexical
process of detransitivation, which leads to an intransitive entry of the transitive
verb.
__________
6 Cf. the above examples in (132) ff.
7 Cf. LEVIN & RAPPAPORT HOVEV (1995) see this problem in the lexical syntactic repre-
sentation of verbs, i.e. in their argument structure; others in the semantic representation
(REINHART 2002) and others even in syntax (BAKER, JOHNSON & ROBERTS 1989 and
KRATZER 1996).
(131) LEVIN & RAPPAPORT HOVAV (1995: 83, 108) suggest a bi-eventive anal-
ysis of causative verbs. Their lexical-semantic representation (LSR) of
such verbs includes the predicate cause, which takes two arguments:
the causing sub-event and the central sub-event (which causes the
change of state of the verb). The CAUSE-argument is associated with
the causing sub-event and the THEME with the central sub-event.
As for the transitive verb break the CAUSE and the THEME are projected from
the LSR into the argument structure (AS) and further into the syntax, as demon-
strated by (132):
(132) Transitive break:
LSR [[x do-something] cause [y become BROKEN]]
Linking rules
AS x y
For the intransitive break the CAUSE-argument is lexically detached in the
mapping from LSR to AS and can thus not be projected into the syntax, cf. (133)
(133) Intransitive break:
LSR [[x do-something] cause [y become BROKEN]]
Lexical binding
Linking rules
AS y
REINHART (2000, 2002) suggests to regard causation as its lexical operation,
which is coded by the lexical feature cause, formalized as [+c], being part of
the definition of a Theta-set that can cause a change, namely Cause, Agent and
Instrument. Agents are to be positive for the feature [+m] (mental state), i.e.
agents are defined by the combination [+c+m]. Instruments are defined by the
feature-cluster [+c-m]. However, their presence is to implicate the existence of
an Agent due to a lexical generalization. Cause is characterized by the feature
[+c], which is consistent with the [+c+m] and [+c-m] specification (whereby the
generalization is captured that verbs selecting cause arguments can also select
instruments and agents). Themes are defined by the feature cluster [-c-m]. Ac-
cording to Reinhart alternating verbs are inherently transitive. They select an
[+c] external argument (which can be realized as agens, cause or instrument)
and a [-c-m] internal argument (i.e. Theme). Anti-causatives are lexically de-
rived from a transitive dictionary entry by a reduction operation, called expleti-
vization, by which the external [+c] Theta-role is reduced. The output of this
expletivization is a monadic/one-place predicate, i.e. an intransitive verb. This
lexical operation is illustrated by (134):
284 Peter Kosta
(134) Expletivization: Reduction of an external [+c] role

a. Vacc (1 [+c], 2) Re (V) (2)
b. Re (V) (2) = V (2)
5.2.2. The theory of (anti-)causatives according to SCHFER (2008) and ALEXIA-
DOU et al. (2006a, b)
ALEXIADOU et al. (2006a, b) proceed from the assumption that agentivity and
causativity are represented by various heads of the decomposition of causatives,
and that (according to KRATZER 2005) the eventive event-head in causatives is
identical with the ones in anti-causatives. Their arguments are the following:
(i) The by-phrase (in Czech instrumental) licenses the same type of exter-
nal arguments as it is licensed being active.
(ii) On the other hand, we do know that anti-causatives do not licence by-
phrases:
(135) a. John/the explosion/Wills banging broke the window
b. The window was broken by John/by the explosion/by Wills bang-
ing.
c. John/die Explosion/Wills Schlag zerschlug das Fenster
d. Das Fenster wurde durch John/durch die Explosion/durch Wills
Schlag zerschlagen
e. John/exploze/rna Willa rozbil(a) okno
f. Okno bylo rozbito Johnem/exploz/rnou Willa.
(136) a. *The window broke by John / by the explosion / by Wills banging
b. *Das Fenster zerbrach durch John/durch die Explosion/durch Wills
Schlagen
c. *Okno se rozbilo Johnem/exploz/ rnou Willa
The contrast between (135) and (136) is considered as an indication that actives
and passives have the potential to assign an external theta-role, while anti-
causatives do not have this potential. A possible explanation is the analysis that
passives and actives have a projection voice, which anti-causatives are lacking.
However, anti-causatives can license CAUSERS and CAUSING EVENTS
cross-linguistically, though no agents, namely by means of a prepositional
phrase, cf.:
(137) a. The window opened from the explosion / Wills banging.
b. *The window opened from Mary
c. Okno se rozbilo od exploze / skrz explozi
d. *Okno se rozbilo od Marie
e. Das Fenster ffnete sich durch die Explosion
f. *Das Fenster ffnete sich von Maria
This fact allows us to assume that the anti-causatives contain a causative seman-
tic, which licens a CAUSER-PP.
6. The morphologic analysis of the lexicon: the four ROOTS and
the encyclopedic lexicon
Therefore, we can decompose the ANTI-CAUSATIVES according to ALEXIA-
DOU et al. (2006 a,b) from a [ Root + Theme] complex, which expresses a re-
sultative state and contains a verbal event-head CAUS, which takes the resulta-
tive state as a complement. CAUS engenders a causal relation between the
CAUSING EVENT (the implicit argument of CAUS) and the resultative state,
which is denoted by [ Root + Theme]. Additionally, causatives show (active
and passive) a VOICE-projection at TOP of CAUS, which is responsible for the
introduction of an external argument. That way there is no directional relation
between the causative-anti-causative-alternation, since no construction is direct-
ly derived from the other. Instead, both derive from the same root:
(138) The abstract decomposition of anticausatives
[CAUS [ Root + DPTheme]
(139) The abstract decomposition of causatives
[DP external Argument VOICE [CAUS [ Root + DPTheme]]]
The decomposition according to SCHFER (2008: 140) is illustrated by the fol-
lowing examples:
(140) a. Peter opened the door.
b. Petr otevel dvee.
c. Peter ffnete die Tr.
d. [Peter VOICE [CAUS [the door open]]]
(141) a. The door opened.
b. Dvee se otevely.
c. Die Tr ffnete sich.
d. [CAUS [the door open]]
The important restriction: VOICE does not introduce an event itself, but only
expresses the relation between the element in a specificator and the event in the
complement position (CAUS). This is also the original conception of voice as
introduced into categorial grammar and the conception of KRATZER (1996).
VOICE contains characteristics, which specify the thematic roles of the ex-
ternal arguments and the corresponding manner: The presence of a feature [
agentive] is responsible that in active and passive constructions AGENTS and
CAUSER are licensed as external arguments. Agentive VOICE (VOICE [+AG])
licenses agentive NPs and instrumental PPs in active and passive; non-agentive
VOICE (VOICE [-AG]) licenses CAUSERS in active and passive. When a
VOICE-head is active, the argument with the corresponding thematic role is rea-
286 Peter Kosta
lized as specificator; when it is passive, the argument with the corresponding

thematic role is implicit.
Concerning the existence of PPs in passives and anti-causatives, Alexiadou
et al. assume that adjunct-PPs are licensed by a structural layer, which contains
the relevant semantic characteristics. The decomposition in (138) and (139)
shows two types of licensing heads, VOICE and CAUS for the PPs. Passive
voice with the feature [+AG] licenses agents while passive voice with the fea-
ture [-AG] licenses only CAUSER by-phrases. Those causative PPs that occur in
anti-causatives in English (from), in German (durch), in Czech (od, skrz), or in
Greek (apo-, me-), are licensed thematically by CAUS.
The final component in the analysis of decomposition is the neutral category
ROOT. ALEXIADOU et al. (2006a, b) and SCHFER (2008: 141) assume that the
roots function as residue of lexical entries and contain information about
whether or not a verb enters Causative Alternation. Thereby, at least the follow-
ing verb types (or verbal classes) are to be divided into four different ROOTS:
(142)
1. A big class of verbs, which limit their external arguments to the theta-role
AGENS: cut, krjet, schneiden... cf. (143). This is the class, taken by Le-
vin, Rappaport & Hovav (1995) and Reinhart (2000) as the crucial criteria
for Causative Alternation.
2. A smaller group of verbs, which allows both AGENS and CAUSER as sub-
jects; however does not form anti-causatives, e.g. destroy, zerstren, zniit
(cf. 144).
3. ROOTS/verbs, which form both causatives and anti-causatives, e.g. break,
zerschlagen, rozbt. (cf. 145).
4. ROOTS/verbs, which only form anti-causatives/inchoatives, but no causa-
tives, e.g. blossom, blhen, kvst (cf. 146).
(143) a. The baker cut the bread. (agent subject)
b. Der Bcker schnitt das Brot
c. Peka krjel chlb
d. *The lightning cut the clothesline (*causer subject)
e. *Der Blitz (zer)schnitt die Wscheleine
f. *Bouka krjela drt
g. *The bread cut (*anticausative)
h. *Das Brot (zer)schnitt.
i. *Chlb se krjel
(144) a. John destroyed the parcel. (agent subject)
b. The explosion destroyed the parcel. (causer subject)
c. *The parcel destroyed (*anticausative)
d. John zerstrte das Grundstck
e. Die Explosion zerstrte das Grundstck
f. *Das Grundstck zerstrte

g. John zniil parcelu.
h. Exploze zniila parcelu.
i. *Parcela se zniila.
(145) a. The vandals broke the window. (agent subject)
b. The storm broke the window. (causer subject)
c. The window broke (anticausative)
d. Die Vandalen (zer)brachen das Fenster
e. Der Sturm (zer)brach das Fenster
f. Das Fenster (zer)brach
g. Vandalov rozbili okno
h. Bouka rozbila okno
i. Okno se robilo
(146) a. *The gardener blossomed the flower (*agent subject)
b. *The warm weather blossomed the flower (*causer subject)
c. The flower blossomed (anticausative)
d. *Der Grtner (er)blhte die Blumen
e. *Das warme Wetter erblhte die Blumen
f. Die Blumen erblhten
g. *Zahradnk rozkvetl kvtiny
h. *Tepl poas rozkvetlo kvtiny
i. Kvtiny rozkvetly
ALEXIADOU et al. (2006a, b) propose, that ROOTS are classified differently ac-
cording to their encyclopedic semantics, also cf. MARANTZ (1997), HARLEY &
NOYER (2000), BHATT & EMBICK (in progress). In accord with the terminology
of LEVIN & RAPPAPORT HOVAV (1995), ROOTS are refered to as agentive if
they form verbs that have AGENS as subject. ROOTS forming verbs, which do
not restrict the external argument to a specific semantic role, but all the same do
not form anti-causatives, are referred to as externally caused. ROOTS forming
verbs which agree to Causative Alternation are referred to as cause unspecified
and ROOTS forming verbs which form only inchoatives/anti-causatives, but no
causatives are referred to as internally caused. The index (147) illustrates this
classification:
(147) agentive (murder, assassinate, cut)
internally caused (blossom, wilt, grow)
externally caused (destroy, kill, slay)
cause unspecified (break, open, melt)
(according to SCHFER 2008: 142)
Furthermore, SCHFER (2008: 161) specifies a spontaneity scale, which de-
scribes the syntactic frame of a ROOT. ROOTS of the types agentive and
288 Peter Kosta
externally caused are associated with events of low spontaneity and a transitive
syntax. ROOTS of the type internally caused express events with high spon-
taneity and require an intransitive syntax. ROOTS of the type cause unspeci-
fied are located in between and can express both syntactic frames. Cf. (148), in-
dex 2:
(148) Index 2: Spontaneity scale (all types of ROOTS)
agentive <externally caused <cause unspecified <internally caused
-spontaneous < ... ... ... ... ... < +spontaneous
... ... ... ... < -transitiv... alternate... intransitive
7. The syntactic projection of the lexical ROOTS
(149) The structure of agentive causatives:
Rosa/the sun melted the ice. (intentionally/non-intentionally)
vP
Spec: Agent v
vo VP
<[+intent] >
[+cause] Spec V
V Compl
(150) The structure of non-agentive causatives:
The sun melted the ice.
vP
Spec: Causer v
vo VP
<[+cause]>
Spec V
V Compl
(151) The structure of non-intentional cause-construction:

The ice melted.
vP
Spec: Unint. v
Causer
vo VP
<[+intent] >
[+cause]
Spec: Affected V
Vo Compl
Syntactic analysis (first approach):
Intransitive anti-causatives are derived syntactically by the addition of non-
active morphology: in Czech and German this is the non-active, semantically
faded reflexive morpheme sich/se or its empty counterpart, the reflexive mor-
pheme reflpron as a further empty pronominal category to pro and PRO (with the
function of [+anticaus] anti-causation resp. intransivation). Since the head of the
light v only shows the single and first feature [+cause], it is suppressed and re-
sults in the structure (152). The morphology and syntax are engendered by a fea-
ture-suppression operation in (153), homomorph to the semantics.
(152) The structure of intransitive (non-agentive) anti-causatives:
The door opened.
vP
Spec: Agent v
vo VP
<[+cause]>
se/sich/reflpron
Spec V
V Compl
(153) Definition of non-active/unaccusative morphology:
Non-active (and/or unaccusative) morphology suppresses the first fea-
ture in the predicate structure (cf. KALLULI 2006: 289; SCHFER 2008:
91)
290 Peter Kosta
How does this operation (153) proceed on the level of syntax? We need a syn-
tactic explanation for the suppression (reduction) of the external arguments. We
assume that anti-causatives indeed are semantically intransitive, however syn-
tactically transitive; i.e. there are two case-marked DPs but only one semantic
argument of the verb. This assumption automatically leads to the following
question: If marked anti-causatives involve two syntactic arguments, the
THEME and the reflexive pronoun, which of those two is internally and which
is externally connected through MERGING? At first glance different modules of
grammar provide different solutions:
Case Theory: the thematic role THEME has the case nominative and the
reflexive pronoun the case accusative. German and Czech are accusative lan-
guages. Therefore, the nominative is to be assigned to the external argument and
the nominative to the internal argument. Thus, the THEME appears to be the ex-
ternal argument.
Binding Principle A: An anaphor has to have a c-commanded antecedent.
This suggests that the THEME is merged externally and the reflexive pronoun
internally.
Configurational Theta-Theory (i): THEMES are always internally com-
bined through MERGING. Thus, the external argument position would be empty
for the reflexive pronoun.
Configurational Theta-Theory (ii): Both internal and external MERGING is
to entail. However, sich/se has no Theta-role.
The fact that many languages distinguish morphologically between un-
marked and marked anti-causatives (with reflexive pronouns) is solved in lexi-
calist theories by means of assuming a dyadic structure for Causative Alterna-
tion and an operation of detransitivation in the lexicon, which concerns the ex-
ternal argument. This operation was already introduced in (153) as lexical bind-
ing (following Levin & Rappaport Hovav), resp. as expletivization (following
REINHART (2000, 2002)). According to these theories the specific morphology in
the case of marked anti-causatives that a detransitivation occurred.
In the paragraphs 3.3.1 and 3.3.2 I examined marked anti-causatives in rela-
tion to unaccusativity and Causative Alternation, in section 4 the case alternation
and argument reduction for transitive anti-causatives in Czech and German. The
result was that several tests did not prove pertinent, and others showed that they
are conform with unaccusatives. We saw that marked anti-causatives are syntac-
tically transitive. How to test whether the only argument in anti-causative con-
structions is an internal one? One test is depicted by the test with the so-called
affectedness datives, which can only be licensed by unaccusative verbs, not by
ergative verbs.
8. Conclusion
In this paper the features of externally and internally caused verbs have been de-
scribed, and connected with the phenomenon of unaccusativity. It was my ap-
proach to elaborate on the relation between lexicon and syntax, including the
concept of distributive morphology and the ROOT-semantics of verbs that do,
resp. do not partake in Causative Alternation and unaccusativity.
The Causative Alternation served as criteria to distinguish between external
and internal causation; with its help the unaccusative verbs have been divided
into two subclasses: alternating unaccusative (AU-) verbs and non-alternating
unaccusative (NAU-) verbs. An alternative distinction between AU- and NAU-
verbs has been found, namely the presence/absence of information about how
the considered process was caused. Thereby the universal concept of encyclo-
pedic lexica in the languages English, German and Czech seems to assume at
least four different ROOTS of verb as basis for classification of the anti-
causativity-opposition: agentive (murder, assassinate, cut), internally caused
(blossom, wilt, grow), externally caused (destroy, kill, slay) and cause unspe-
cified (break, open, melt).
Furthermore, it has been shown that unergative/causative pairs depict an in-
dependent phenomenon and do not effect the considerations about CAL (in ac-
cordance with ALEXIADOU et al. 2006a, b and MARANTZ 1997; however not in
accordance with LEVIN & RAPPAPORT HOVAV 1995 and REINHART 2000).
References
ALEXIADOU, A., E. AGNASTOPOULOU & F. SCHFER (2006a) The properties of anticausatives
crosslinguistically. In: FRASCARELLI, M. (ed.) Phases of Interpretation 187-211. Ber-
lin: Mouton de Gruyter,.
ALEXIADOU, A., E. AGNASTOPOULOU & F. SCHFER (2006b) The fine structure of (anti-)
causatives. In: DAVIS, C., A.-R. DEAL & Y. ZABBAL (eds.) Proceedings NELS 36 115-
128. Amherst MA: GLSA.
ALEXIADOU, A. & F. SCHFER (2007) On the realization of external arguments in nominaliza-
tions. Ms. Ljgs Konstanz:
http://ifla.uni-stuttgart.de/institut/mitarbeiter/handouts/GGS07.pdf
ASHER, R. E. (1985) Tamil. London: Croom Helm.
BABBY, L. H. (1980) Existential Sentences and Negation in Russian. Ann Arbor.
BABBY, L. H. (1987) Case, prequantifiers, and discontinuous agreement in Russian. Natural
Language and Linguistic Theory 5. 91-138.
BABYONYSHEV, M. (1996) Structural connections in syntax and processing: Studies in Rus-
sian and Japanese. Ph.D. dissertation. Massachusetts Institute of Technology.
BAKER, M. C. (1988) Incorporation. A Theory of Grammatical Function Changing. Chica-
go/London: The University of Chicago Press.
BAKER, M., K. JOHNSON & I. ROBERTS (1989) Passives arguments raised. // LI 20:219-252. =
B/J/R (1989).
BESNIER, N. (2000) Turaluan. A Polynesian Language of the Central Pacific. London. Rout-
ledge.
292 Peter Kosta
BORER, H. (1998) Passive without theta grids. In: S. LAPOINTE, P. FARRELL & D. BRENTARI.
Morphology and its Relations to Phonology and Syntax. Stanford: CSL1.
BORER, H. (2000) Exo-skeletal vs. Endo-skeletal explanations: Syntactic projections and the
lexicon. Explorat.
BORER, H. (2005) Structuring sense 2: The normal course of events. Oxford: Oxford Universi-
ty Press.
BORER, H. (2007) Structuring sense 1: In name only. Oxford: Oxford University Press.
BURZIO, L. (1986): Italian Syntax: A Government-Binding approach. Dordrecht.
CETNAROWSKA, B. (2000) The Unaccusative/Unergative Split and the derivation of resultative
adjectives in Polish. In: KING, T. H., I. A. SEKERINA (eds.) Formal Approaches to
Slavic Linguistics. Ann Arbor. 78-96.
CETNAROWSKA, B. (2002) Unaccusative Mismatches and Unaccusative diagnostics from deri-
vational morphology. In: Baucher, P. (ed.) Many Morphologies. Somerville, M. A. 48-
81.
COMRIE, B. (1976) The Syntax of Causative Constructions: Cross-Language Similarities and
Divergences. In: SHIBATANI, M. (ed.) Syntax and Semantics Vol. 6: The Grammar of
Causative Constructions. New York: Academic Press.
COMRIE, B. (1989) Language Universals and Linguistic Typology. 2nd ed. Oxford: Blackwell.
COMRIE, B. & M. POLINSKY (1993) Causatives and Transitivity. Amsterdam/Philadelphia:
John Benjamins.
DAVIES, J. (1981) Kobon. Amsterdam: North Holland.
DEN DIKKEN, M. (1995) Particles. On the Syntax of Verb-Particle, Triadic, and Causative
Constructions. New York, Oxford: Oxford Studies in Comparative Syntax.
DIXON. R. M. W. (2000) The rise and fall of languages. Cambridge: Cambridge Univ. Press.
(reprint).
EMBICK, D. (2004) On the Structure of Resultative Participles in English. In: LI 35; 3. 355-
392.
GREWENDORF, G. (1989) Ergativity in German. Dordrecht Holland / Providence USA
(SGG 35).
GREWENDORF, G. (1994) Kohrente Infinitive und Inkorporation. In: STEUBE, A., G.
ZYBATOW (eds.) Zur Satzwertigkeit von Infinitiven und Small Clauses. Tbingen:
Niemeyer. 31-50.
GUNKEL, L. (2003) Infinitheit, Passiv und Kausativkonstruktionen im Deutschen. Tbingen (=
Diss. Universitt Potsdam 2001). (Studien zur deutschen Grammatik 67).
HALE, K. & J. KEYSER (1993) On argument structure and the lexical expression of syntactic
relations. In: Hale, K./Keyser, S.J. (eds.), The View from building 20. Cambridge MA:
The MIT Press, 53-109.
HALE, K. & KEYSER, J. (2002) Prolegomena to a theory of argument structure. Cambridge
MA: The MIT Press.
HARLEY, H. & NOYER, R. (2000) Formal versus encyclopedic properties of vocabulary: Evi-
dence from nominalizations. In: PETERS, B. (eds.) The lexicon-encyclopedia interface.
Amsterdam: Elsevier, 349-374.
HARVES, STEPHANIE (2002) Unaccusative syntax in Russian. Ph.D. Dissertation. Princeton
University.
HARVES, STEPHANIE (2003) Getting Impersonal: Case, agreement, and distributive po-phrases
in Russian. In: BROWNE, W., J.-Y. KIM, B. PARTEE & R. ROTHSTEIN (eds.) Formal Ap-
proaches to Slavic Linguistics 11. Ann Arbor. 235-254.
HARVES, S. (2009) Unaccusativity. In: KEMPGEN, S., P. KOSTA, T. BERGER & K. GUTSCHMIDT
(eds.) Die Slavischen Sprachen/The Slavic Lanuages. Ein internationales Handbuch zu
ihrer Struktur, ihrer Geschichte und ihrer Erforschung / An International Handbook of
their History, their Structure and their Investigation. Berlin, New York. 1. Teilband
(HSK 32.1). Mouton & deGruyter. 415-430.
HASPELMATH, M. (1993) More on the typology of inchoative/causative verb alternations. In:
COMRIE, B., M. POLINSKY (eds.) Causatives and Transitivity. Amsterdam/Philadelphia:
John Benjamins, 87-111.
HOEKSTRA, T. (1992) Aspect and Theory. In: ROCA, I. M. (1992) Thematic Structure: Its
Role in Grammar. - Berlin: Mouton de Gruyter. S.145-174.
JAEGGLI, O. A. (1986) Passive. In: Linguistic Inquiry 17.4. 588-622.
JUNGHANNS, U. (1994) Satz und doch nicht Satz: ber die (doppelte) Einbettung finaler Infini-
tive des Russischen. In: STEUBE, A., G. ZYBATOW (eds.) Zur Satzwertigkeit von Infini-
tiven und Small Clauses. Tbingen: Niemeyer. 107-139.
KALLULLI, D. (2006a) Argument demotion as feature suppression. In: LYNGFELDT, B. & T.
SOLSTAD (eds.): Demoting the agent. Amsterdam/Philadelphia : John Benjamins, 143-
166.
KALLULLI, D. (2006b) A unified analysis of passives, anticausatives and reflexives. In: BO-
NAMI, O., P. CABREDO-HOFHERR (eds.) Empirical issues in formal syntax and seman-
tics 6. Published online: http://www.cssp.snrs.fr/eiss6/
KALLULLI, D. (2006c) Unaccusatives with dative causer and experiencers: A unified account.
In: HOLE, D., A. MEINUNGER & W. ABRAHAM (eds.): Datives and other cases. Amster-
dam/Philadelphia: John Benjamins, 271-301.
KALLULLI, D. (2006d) Passive as feature suppression operation. In: ABRAHAM, W. & L.
LEISI (eds.) Passivization and typology. Amsterdam/Philadelphia: John Benjamins,
442-460.
KALLULLI, D. (2006e) A unified analysis of passives and anticausatives. In: EBERT, C. & C.
ENDRISS (eds.): Proceedings of the Sinn und Bedeutung 10 [ZAS Papers in Linguistics
44] Berlin: ZAS, 171-182.
KAUFMANN, I. (1995) Konzeptuelle Grundlagen semantischer Dekompositionsstrukturen: die
Kombinatorik lokaler Verben und prdikativer Komplemente. - Tbingen: Niemeyer,
1995.
KIMENYI, A. (1980) A relational grammar of Kinyarwanda. Berkeley: University of
California Press.
KOSTA, P. (1992) Leere Kategorien in den nordslavischen Sprachen. Zur Analyse leerer Sub-
jekte und Objekte in der Rektions-Bindungs-Theorie. Habilschrift Frankfurt am Main
(online: http://www.uni-potsdam.de/u/slavistik/wsw/habil/habil.htm.)
KOSTA, P. & J. FRASEK (2004) Neakuzativita (ergativita) vs. neergativita v etin, poltin a
jinch slovanskch jazycch na rozhran morfologie a syntaxe. Unaccusativity (erga-
tivity) in Czech, Polish and some other Slavic languages at the Morpho-Syntactic In-
terface. In: KARLK, P. & Z. HLADK (ed.). etina univerzlia a specifika 5. Praha.
172-194.
KOSTA, P. (2009): Satzglieder und impersonale Konstruktionen im Slavischen. In: KEMPGEN,
S., P. KOSTA, T. BERGER & K. GUTSCHMIDT (eds.) Die Slavischen Sprachen. The
Slavic Languages. Ein internationales Handbuch zu ihrer Struktur, ihrer Geschichte
und ihrer Erforschung. An international Handbook of their Structure, their History and
their Investigation. Berlin, New York: Mouton, de Gruyter. Teilband 1 (HSK 32.1),
391-415.
294 Peter Kosta
KRATZER, A. (1996) Severing the external argument from its verb. In: ROORYCK, J. & L. ZAR-
ING (eds.) Phrase structure and the lexicon. Dordrecht: Kluwer, 109-137.
KRATZER, A. (2005) Building resultatives. In: MAIENBORN, C. & A. WLLSTEIN-LEISTEN
(eds.) Event arguments in syntax, semantics and discourse. Tbingen: Niemeyer, 178-
212.
LEVIN, B. & M. RAPPAPORT HOVAV (1995) Unaccusativity: at the syntax-lexical semantics in-
terface. Cambridge (Massachussetts): MIT Press.
LENERETOV, D. (2004) Czech pronominal clitics. In: JSL 12, 1-2 (2004) 135-171.
LICHTENBERK, F. (1983) Manam Language; Grammar. Univ. of Hawaii. Honolulu.
MANZINI, M. R. (1983) Restructuring and reanalysis. Ph. D. Thesis. Ph. D. Massachusetts In-
stitute of Technology. Dept. of Linguistics and Philosophy. 8. Cambridge, Mass.
MARANTZ, A. (1984) On the Nature of Grammatical Relations in Syntax. Cambridge, MA:
MIT Press.
MARANTZ, A. (1987) No escape from syntax: Dont try morphological analysis in the privacy
of your own lexicon. In: Dimitriadis, A., L. Siegel, C. Surek-Clark & A. Williams.
Proceedings of the 21st annual Penn linguistics colloquium [Penn Working Papers in
Linguistics]. Philadelphia PA: University of Pennsylvania, 201-225.
NEDJALKOV, V. P. (1976) Kausativkonstruktionen. Tbingen: Verlag Gunter Narr.
OHLER, U. G. (1997) Syntaktische und semantische Eigenschaften unakkusativer Verben.
Staatsexamensarbeit Universitt Konstanz. (pdf-file).
PERLMUTTER, D. M. (1978) Impersonal passives and the Unaccusative Hypothesis. In: JAEG-
ER, J. et al. (Hg.) (1978) Proceedings of the Fourth Annual Meeting of the Berkeley
Linguistic Society. Berkeley: University of California. 157-189.
PERLMUTTER, D. M. & P. M. POSTAL (1983) The Relational Succession Law. In: Perlmutter,
D. (Hg.) Studies in Relational Grammar 1. Chigago (Illinois): University of Chigago
Press, 1983, 30-80.
PESETSKY, D. (1982) Paths and Categories. Doctoral dissertation. Massachussetts Institute of
Technology. Dept. of Linguistics and Philosophy.
PUSTEJOVSKY, J. (1995) The generative lexicon. Cambridge MA: The MIT Press.
RAMCHAND, M. (2006) Verb meaning and the lexicon: A first phase syntax. Cambridge: CUP.
RAMCHAND, M. & P. SVENONIUS (2004) Prepositions and external argument demotion. In:
SOLSTAD, T. & B. LYNGFELT (eds.): Demoting the agent: Passive and other voice-
related phenomena. Amsterdam/Philadelphia: John Benjamins, 93-99.
REED, L. (1989) A Barriers Approach to the Issues of Complementation and Word Order in
French Causatives. Ms. Ottawa.
REED, L. (1992a) Remarks on Word Order in Causative Constructions. In: LI 23. 164-172.
REED, L. (1992b) French Clitic Case Alternations in a Parametric Theory of Grammar. Ms.
University of Otawa.
REINHART, T. (2000) The theta system: Syntactic realization of verbal concepts. OTS working
papers. TL-00.002. Utrecht University.
REINHART, T. (2002) The theta system An overview. Theoretical Linguistics 28(3): 229-
290.
ROEPER, T. (1987) Implicit Arguments and the Head-Complement Relation. In: LI 18. 267-
310.
SCHFER, F. (2008) The Syntax of (Anti-) Causatives. External arguments in change-of-state
contexts. Amsterdam/Philadelphia: John Benjamins Publishing Company.
SCHOORLEMER, M. (1995) Participial Passive and Aspect in Russian. Ph. D. Dissertation,
OTS, Utrecht University.
SCHOORLEMER, M. (2004) Syntactic Unaccusativitiy in Russian. In. ALEXIADOU, A. E. ANAG-
NOSTOPOLOU & M. EVERAERT. The Uniaccusativity Puzzle. Explorations of the Syn-
tax-Lexicon Interface. New York, 207-242.
SHIBATANI, M. (2002) Some basic issues in the grammar of causation. In: SHIBATANI, M. (ed.)
The Grammar of Causation and interpersonal manipulation. Amsterdam/Philadelphia
(Typological Studies in Language 48) 1-22.
SHIBATANI, M. & P. PARDESHI (2002) The causative continuum. In: SHIBATANI, M. (ed.) The
Grammar of Causation and interpersonal manipulation. Amsterdam/Philadelphia (Ty-
pological Studies in Language 48) 85-126.
SOLNIT, D. B. (1997) Eastern Kayah Li. Grammar, Texts, Glossary. Univ. of Hawaii Press.
Honolulu.
SONG, J. J. (1996) Causatives and Causation. A Universal-Typological Perspective. London
and New York.
SONG, J. J. (2001) Toward a Typology of Causative Constructions. Lincom Europa. Languag-
es of the World 23.
SYBESMA, R. (1992): Causatives and Accomplishments: The case of Chinese ba. Ph.D. Diss.
Holland Institute of Generative Linguistics (HIL).
SZUSCICH, L. (2007) Nothing wrong with finite T: Non-Agreeing Accusative Impersonal Sen-
tences. In: GOLDZINOWSKA, M. et al. (eds.) Formal Approaches to Slavic Linguistics
15. Ann Arbor.
SZUCSICH, L. (2008) Evidenz fr syntaktische Nullen aus dem Burgenlandkroatischen, Polni-
schen, Russischen und Slovenischen. Merkmalausstatung, Merkmalshierarchien und
morphologische Defaults. Beitrag auf dem 15. JungslavistInnen-Treffen in Bochum
27.-30. September 2006. In: Zeitschrift fr Slawistik 53(2) (2008) 160-177.
Peter Kosta, University of Potsdam, Department of Slavic Languages and Lite-

ratures, Chair Slavic Linguistics, Am Neuen Palais 10, 14469 Potsdam, Germa-
ny, peter.kosta@uni-potsdam.de
Reciprocity and similar meanings in Slavic languages and
SAE
Alexander Letuchiy
1. Introduction
In this article, I will sketch some typologically relevant features of reciprocal
constructions in Slavic languages in comparison with some other European
languages. My aim is to find out whether expression of reciprocity (and similar
meanings) can draw a borderline between Slavic languages and a language type
which HASPELMATH (2001), following Boas, calls Standard Average European
(SAE) and which mainly includes languages of Central and Western Europe.
Let me recall some relevant features of this language type. First, the SAE lan-
guages are detransitivizing in terms of NICHOLS et al. (2004). In other words,
they possess productive grammatical mechanisms which decrease valency or
transitivity of the base verb (passivization, anticausativization) but lack produc-
tive mechanisms of valency increase for instance, causativization.
Second, these languages are reference-dominated in terms of HASPELMATH
(2001). The coding of the verb arguments is not directly related to their semantic
features. For instance, in modern SAE languages canonical marking (nominati-
ve) of grammatical subjects is by far more frequent than any other one.
In the domain of reciprocity and reflexivity, SAE languages, as well as Slavic
languages, are characterized by a big set of common properties.
First of all, all of them possess productive means of expressing reciprocity,
and in all of these languages one of the reciprocal markers goes back to the re-
flexive pronoun: cf. German sich, French se, Swedish -s and so on in SAE
languages, and Russian -sja, Bulgarian se, Polish sebe and se in Slavic
languages.
Then, all of the languages under analysis have multiple reciprocal markers:
in none of them reciprocity can only be expressed with a grammatical marker. In
the Slavic group, Russian has also the pronouns drug druga each other and
odin drugogo one another. Bulgarian has the pronoun edin drug each other,
in Polish we find edin drugego each other, and so on. The same picture can be
observed in SAE languages of Western Europe: for instance, German has ei-
nander one another, in French lun lautre one another is used.
Usually one of these strategies of expressing reciprocity prevails over the
other one(s) but which one can be different for different languages.
Another feature, very relevant for the present paper is that all languages un-
der analysis have one or several markers which combine the reciprocal meaning
with other ones, for instance, middle, passive, reflexive and so on. Sich in Ger-
298 Alexander Letuchiy
man is very productive in the anticausative meaning, -sja in Russian is often

used in the reflexive and anticausative meanings, and so on.
However, the two groups are not homogenous inside themselves and differ sig-
nificantly from each other. Below I analyze common and differential features of
reciprocal markers.
In Section 1, the general features of reflexive / reciprocal markers, such as
French se or Russian -sja are examined. I show that these general features are
closely related to expression of reciprocity and reflexivity in the languages un-
der analysis.
In Section 2, I analyze the properties of different mechanisms of expressing
reciprocity. I consider not only purely grammatical means, but also lexical mar-
kers, such as Russian vzaimnyj mutual, and intermediate cases, for instance,
the French prefix entre between.
A particular consideration will be given to combinations of reciprocal mar-
kers of different types. I take French, German and Spanish to illustrate the SAE
type; the Slavic group is illustrated by Russian, Bulgarian and Polish.
1. Properties of reflexive / reciprocal marker
1.1. Relations between meanings
As I have said, all languages under analysis have polysemous reciprocal mark-
ers. More precisely, all se-like markers in SAE and Slavic languages are
polysemous below I list some of possible meanings:
x passive
x anticausative
x reciprocal
x reflexive etc.
A natural question is whether the same derivative can have several meanings.
The question is different for different languages.
In all languages of our sample, except Russian: Polish, Bulgarian, Serbian,
French, German and so on, the meanings of the reflexive marker interfere in
many cases. Even one lexeme, containing this marker, can have different
readings. For instance, in Bulgarian obiam se has at least two meanings: refle-
xive (love oneself) and reciprocal (love each other). The latter variant is
much more natural for the situation love, but the former is also possible.
The same is true for French marker se. Although the verb saimer has the
main reading love each other (reciprocal), the derivative has also a special rea-
ding feel well (somewhere, in some consequences):
Reciprocity and similar meanings in Slavic languages and SAE 299
(1) Ils saim-ent depuis longtemps.

they REC-love-PRS.3PL since long
They love each other for a long time.
(2) Il s'aim-e la campagne.
he REFL-love-PRS.3SG in DEF village
He feels well in the village.
In Russian, most derivatives have one meaning. The passive meaning is an
exception: it combines with most of other readings; for instance, the verb
mytsja means wash (oneself); bath in its main meaning, but the passive
meaning to be washed is also possible:
Russian:
(3) Ja moj-u-s v vann-e.
I.NOM wash-PRS.1SG-REFL in bath-SG.LOC
I bath in the bath.
(4) Posud-a moj-et-sja v rakovin-e.
dishes-PL.NOM wash-PRS.3SG-REFL in sink-SG.LOC
The dishes are being / must be washed in the sink.
This parameter correlates to another one: namely, productivity of reciprocal
and reflexive meanings. Data from NEDJALKOV (2007) and NEDJALKOV (ed.)
(2007) show that in most SAE languages the reflexive and reciprocal uses of the
polysemous reflexive marker, such as se in French, are productive and are not
restricted with narrow lexical classes. This seems to be a common feature of
most SAE languages.
Slavic languages do not behave uniformly in this aspect. In some of them,
such as Bulgarian and Polish, reflexive and reciprocal uses are productive. For
instance, the following Bulgarian and Polish reflexive verbs are polysemous and
can denote both reflexivity and reciprocity:
Bulgarian:
(5) obiam se love each other, love oneself;
vidja se see each other, see oneself (GUENTCHVA, RIVIRE 2007);
Polish:
(6) instruowa se instruct each other, instruct oneself
wynagrodzi se reward each other, reward oneself (WIEMER 2007).
In Russian the reflexive and the reciprocal use are not highly productive. Refle-
xive is restricted with so-called body-care verbs (cf. myt-sja wash (oneself),
brit-sja shave (oneself)). It is not possible to derive sja-derivatives with refle-
xive meaning from verbs like ljubit love, videt see, bit beat and so on.
The verb ljubitsja exists only in colloquial Russian, and it has a special type of
reciprocal meaning: it means fuck; videtsja and bitsja also have the
reciprocal meaning correspondingly, meet (see each other) and fight.
In turn, the reciprocal meaning is compatible mainly with inherent
reciprocals, in terms of KEMMER 1993 (obnimat-sja hug (each other),
celovat-sja kiss (each other), see other examples in KNJAZEV 2007). Of
course, there are some inherent reciprocals which do not denote any type of
physical contact: for instance, vstreat-sja meet (from vstreat meet (transiti-
ve)) and videt-sja meet (from videt see) belong here.
What is really productive in Russian is the anticausative and passive
readings of -sja (razbit-sja break (intr), zakryvat-sja close (intr), be closed,
etc.).
Another parameter characterizing the relation between meanings of
reciprocal / reflexive marker is presence / absence of reflexively marked passi-
ves. This feature will be discussed in the following section.
1.2. Reflexively marked passives
In all of the languages under analysis, there are special markers of passives
namely, constructions with participles and verbs to be / to become:
(7) Larbre a t abattu.
DEF-tree cut.down.PF.PASS
The tree was cut down.
However, many of these languages can also code the passive meaning with se-
like markers. In general, the passive meaning, according to HASPELMATH 2001,
is characteristic for reflexive and reciprocal markers. However, not all SAE and
Slavic languages show this meaning in equal degree.
In Slavic languages, such as Russian and Bulgarian, sja-marked passive is
productive. In Russian, it is possible only in imperfective forms, in Bulgarian no
aspectual restrictions apply:
Russian:
(8) a. Dom- stroi-t-sja rabo-imi.
house-NOM build-PRS.3SG-PASS worker-PL.INS
The house is being built by the workers.
b. *Dom- po-stroi-t-sja rabo-imi.
house-NOM PF-build-PRS.3SG-PASS worker-PL.INS
The house was built by the workers.
Bulgarian:
(9) Kak mo-e da se po-stroi- detska-ta plotadka?
How can-PRS.3SGF TO PASS PF-build-AOR.3SG children-DE area
How can be build a children area?
In SAE languages, according to WIEMER 2007 (German) and GUENTCHVA,

RIVIRE 2007 (French) the sja-marked passive is not productive. For instance,
(10) does not mean that the cup was broken by someone it only means that the
cup broke by itself:
French:
(10) La tasse sest casse.
i. The cup broke. (anticausative).
ii. #The cup was broken (passive).
The absence or low productivity of reflexively-marked passives can be
considered as a common feature of SAE languages. However, the productivity
of this phenomenon in Slavic languages is too different to regard it as a
borderline between Slavic and SAE languages.
1.3 Decreasing derivations and transitivity
Se-like markers usually either make the base verb syntactically intransitive or
decrease the number of verbal arguments. However, the relation between
transitivity and valency change can be different in different languages.
In most SAE and Slavic languages, se-like derivatives are intransitive,
independently of their semantics.
Note that this rule is valid for derivatives build with circumfixes where
semantic motivation of detransitivization is weakened:
Russian:
(11) On na-pi-l--sja vod-y.
he.NOMPREF-DRINK-PST-SG.M-REFL WATER-SG.GEN
He drank enough water.
Bulgarian:
(12) Toj na-pi- se ot voda-ta.
he.NOMPREF-drink-AOR.3SG REFL from water-DEF
He drank enough water.
Both in SAE and Slavic languages there are cases when se-derivatives are transi-
tive. However, these cases are different, and this difference corresponds to more
general differences between the two groups of languages.
In SAE, the group of syntactically transitive reflexive verbs includes
indirect reciprocals, in terms of NEDJALKOV (2007) in other words, these are
reflexive verbs which denote co-reference between the subject and the indirect
object. For instance, the French verb se donner give to each other is
syntactically transitive. It takes as a direct object the same argument which was
the direct object of the base verb: cf. se donner la main hold each others hand,
lit. give each other hand.
In Slavic languages, the transitive group of reflexive verbs includes some

verbs formed by means of some circumfixes including -sja (see JANKO-
TRINICKAJA 1962):
Russian:
(13) Malik- do-da-l--sja mam-u.
boy-SG.NOM PREF-wait-PST-SG.M-REFL mother-SG.ACC
The boy waited until his mother came.
The relation between valency-changing derivation and transitivity can serve as a
borderline between SAE and Slavic languages. In SAE, valency-changing deri-
vations are not directly correlated to transitivity, which is the case in Slavic
languages.
2. Expression of reciprocal meaning
2.1 Semantics of reciprocals
In this section, I will describe the semantics of reciprocal construction which are
the main topic of our article. Let me first describe briefly the ways to analyze
semantics of reciprocal constructions in order to show the difference between
lexical and grammatical reciprocal markers. Note that reciprocity is analyzed in
different ways in formal semantics and generative linguistics.
In formal syntax, beginning from Chomsky , reciprocal markers, just as
reflexive markers, are analyzed as a special type of co-reference marker
anaphors (just as reflexives). In other words, most reflexive / reciprocal markers
require the antecedent to be expressed in the same clause. In (3), each other has
an antecedent (John and James) in the same clause. For (4), only the reading i. is
plausible where the antecedent (John and James) is in the same clause; the rea-
ding ii. where the antecedent is we, located in the main clause, is inapplicable:
(14) John and James criticized each other.
(15) We asked John and James to criticize each other.
i. John and James must criticize each other.
ii. *I asked John to criticize someone, and someone asked John to critic-
ize me.
However, this purely syntactic point of view has some shortcomings: it does not
account for semantic and pragmatic effects of reciprocals. For instance, the
syntactic point of view does not explain why the reciprocal from Russian videt
see means meet, and not just see each other, because the requirement of an
antecedent in the same clause holds both for see each other and meet.
In descriptive linguistics, and partly in formal semantics: reciprocal
constructions are considered from the semantic point of view. The main compo-
nent of the reciprocal meaning seems to be the component of bi-directionality:
the given situation occurs symmetrically in two directions.
A special point of view is given in BECK 2001: the author thinks that
reciprocals are definite expressions with the meaning roughly all the other
ones. Reciprocity proper is not a part of meaning of markers like each other or
drug druga.
We adopt a view which is close to NEDJALKOV 2007: two components of
reciprocity:
x bi(multi)directionality: We criticize each other = A criticizes B, B criticizes
A
x closed relation: We criticize each other = there are no C other than B and A
which is included into the reciprocal relation.
Most reciprocal markers, such as each other, one another in English, -sja and
drug druga in Russian, se and lun lautre in French denote closed relation. In
other words, they denote that the given situation is symmetrical and does not co-
ver other subjects, besides the given set of subjects.
This component becomes evident in constructions like (16) where the
reciprocal pronoun is in the scope of negation and is contrasted to a non-
reciprocal object.
Russian:
(16) On-i ljub-at ne drug drug-a,
they.NOM love-PRS.3SG not other other-ACC
[a drug-ix ljudej].
but other-PL.ACC people.ACC
They love not each other, but other people. (each of them loves another
person).
The sentence means that the relation is not restricted by the given set of people P
each of them loves a person which does not belong to P. Note that the negation
of drug druga can only mean that the relation is not restricted by the given set.
Another reading is impossible: (16) cannot mean that the relation is
unidirectional they do not love each other, only A loves B, but B does not
love A..
On the other hand, in many languages under analysis there exist markers which
can be called peripheral reciprocal markers. I distinguish this class of units
based on their morphological and syntactic status. The grammatical markers,
such as se or each other, either occupy the syntactic position of direct object, or
form a small clause where one of the components is in the object, another one in
the subject position, or, finally, they can be pure detransitivizers, which is the
case of Russian -sja and other highly grammaticalized markers.
In contrast, peripheral markers do not occupy either the subject or the object
position, nor they demote the syntactic transitivity of the base verb. They are
either adjectives, as Russian vzaimnyj mutual, German gegenseitig or English
mutual, or adverbs, as Russian vzaimno. Finally, there is a small class of units,

including French entre- and Russian vzaimo- which are morphological prefixes,
which, in contrast to suffixes like -sja, do not make the verb intransitive and
usually cannot be used as the sole marker of reciprocity (see for details).
The core component of meaning of peripheral reciprocal markers is
bidirectionality, rather than closeness of relation. The contrastive context yields
us negation of bidirectionality, and not that of closeness.
Russian:
(17) Et-o ne vzaimn-aja ljubov.
this-SG.N not mutual-NOM.SG.F love-SG.NOM
It is not mutual love (only one of them loves the other).
This tendency seems to hold both in Slavic and in SAE languages. For instance,
see the German phrase Diese Liebe ist nicht gegenseitig This love is not mu-
tual means that the relation is not symmetric A loves B, but B does not love
A. However, the phrase Peter und Julia lieben nicht einander aber does not
mean that only Peter loves Julia. It means that Peter and Julia love other people.
2.2 Interaction of autonomous markers and grammatical markers
Above I analyzed mainly grammatical means of marking reciprocity. However,
both SAE and Slavic languages have a very special type of construction: the one
where both a grammatical and a lexical reciprocal marker occur. Below I call
them constructions with double marking of reciprocity.
In general, it has been long pointed out that valency derivation can be
marked with two markes in the same time. KULIKOV (1993), as well as many
authors of descriptive grammars, points out that the causative derivation can be
doubled in many languages of the world: for instance, causativization of the
verb tur- stand up; stand in Khakas yields a transitive causative tur-s put,
which, in turn, can be causativized to yield tur-s-tr cause to put.
However, this is not the type of doubling which we see in the domain of
reciprocity. First of all, being formally double, reciprocity in the constructions
under analysis cannot be semantically double. It is hardly possible to imagine an
example like Khakas tur- for reciprocals: if a verb is reciprocalized (the
reciprocal relation is built between two of its participants) it is impossible to
reciprocalize it for the second time.1
When the reciprocal meaning is expressed with two markers in the same
sentence, the construction is usually synonymous with a simpler one, where only
one lexical or grammatical item expresses reciprocity.
I will draw an example from Russian. In Russian, the double reciprocal
construction denotes multiplicity of events (see LETUCHIY 2009 for details and
__________
1 In fact there are some special situations when
(23) below). A single event between two participants is usually not denoted by
this construction:
???
(18) Dva elovek-a celova-l-i-s drug s drug-om.
two.M person-SG.GEN kiss-PST-PL-REC other with other-SG.INS
People kissed (with) each other.
This effect, according to BEHRENS (2008), is absent in German. Languages with
transitive doubling do not show this effect.
Along with semantic differences, constructions with double expression of
reciprocity are syntactically different in different languages of Europe. I
distinguish here two types of doubling: transitive and intransitive doubling.
2.2.1 Transitive reciprocal doubling
In the transitive doubling type, represented, for instance, by French and
Bulgarian, verbs with the reciprocal/reflexive pronoun se can simultaneously
take another reciprocal marker (the pronoun lun lautre (French) and edin drug
in Bulgarian) without a preposition.
Bulgarian:
(19) Te se obi-at edin drug.
they.NOM REC love-PRS.3PL one another
They love each other. (PENCHEV 2007)
French:
(20) Ils saiment lun lautre.
They love each other. (GUENTCHVA 2007)
In (19) and (20), the reciprocal pronoun is in the same form in which it would be
if the verb was transitive without a reciprocal marker se. Nothing in the marking
of edin drug and lun lautre points to the fact that the verb became intransitive.
In other words, the structure in both examples is not characteristic of a European
language: it seems that the direct object position of the verbs obiam love and
aimer love is occupied simultaneously by the grammatical marker se / s and by
the pronoun edin drug and lun lautre, respectively.
Note, however, that (21) is ungrammatical in Bulgarian both with and
without se:
Bulgarian:
(21) *Kogo (se) obi-at? Edin drug.
who.ACC REC love-PRS.3PL one another
Who do they love? Each other.
Therefore, edin drug is not just an object NP in constructions with transitive
doubling, as it is in constructions like They love each other in English, where we
can impress a dialog like Who do they love? Each other.
Note that all se-verbs in Bulgarian are intransitive. Therefore, edin drug can
hardly be a direct object in this case we would have to postulate an object po-
sition in the VP, which is reserved only for reciprocal markers.
Two alternative analyses can be chosen to explain this situation:
x small clause: in languages like Bulgarian, marker edin drug functions as a
small clause: They love [one love other]. Therefore, edin drug is not an
object of the matrix verb.
x complex marker: in constructions with transitive doubling, we observe
non-fixed order of derivations: edin drug and se attach to the verb
simultaneously, as one complex marker:
se-love-edin drug
An interesting fact is that in constructions with samo only, both se and edin
drug occur:
(22) Te se obi-at samo edin drug.
they.NOM REC love-PRS.3PL only one another
They love only each other.
If edin drug were to any extent autonomous from se (for instance, if it
represented a small clause), we would not expect se to occur in this construction.
Semantically, se is not in place here, because the situation se obiam love
each other can by definition be only reciprocal, and the construction with samo
becomes awkward at first glance (the operator only is applicable only if there
are other possibilities, besides the real one).
The only option is to propose that se and edin drug compose a sort of single
marker where one part (edin drug) is autonomous and can be emphasized, but
2.2.2 Intransitive reciprocal doubling
In contrast to Bulgarian, in Russian transitive doubling is impossible: verbs with
-sja can only take the reciprocal pronoun in combination with the preposition s
with:
Russian
(23) Ludi celova-l-i-s drug s drug-om / *drug drug-a.
people kiss-PST-PL-REC other with other-SG.INS other other-SG.ACC
People kissed (with) each other.
In this case, the order of derivations is fixed: first, sja-like marker detransitivizes
the verb and makes it impossible to attach anything in the direct object position.
Even a reciprocal modifier cannot occupy the object slot. Afterwards, the mar-
ker drug druga can only be in the comitative form, since the verb is intransitive.
The type of doubling does not either distinguish between Slavic and SAE lan-
guages. In both groups, there are languages with transitive doubling (Bulgarian
in Slavic group, French among SAE languages), as well as languages with in-
transitive doubling (Russian in Slavic group, German among SAE languages).
The type of doubling is determined by the syntactic properties of reciprocal
markers, rather than by the general language type. It seems that the following
characteristics of markers make it possible that the given language will have a
transitive, and not intransitive type of doubling:
x morphological / non-autonomous marker is not highly grammaticalized
x non-morphological autonomous marker cannot be used as the sole
reciprocal marker
On the one hand, if the non-morphological marker of the type each other cannot
be the sole marker, this means that it does not have all properties of the direct
object and requires the support of another marker. On the other hand, therefore,
there must be both markers in the reciprocal construction.
Note that the first characteristic is not obligatory:
x German sich is more autonomous than French se. However, transitive
doubling exists in French and is very restricted in English, because Ger-
man einander, and not French lun lautre can be used as the sole marker.
Outside Europe, both types of doubling are represented:
Adyghe: only transitive doubling:
(24) TezE-m zE-r z-jE-wE{a-ZE-R
we one-OBL one-ABS REC-3SG.A-wound -INTF-PAST
We wounded each other. (canonical reciprocal of a transitive verb).
Arabic: only intransitive reciprocal doubling
(25) Y-ahubb-u baD-u-hum baD-an.
3M-love-SG some-NOM-3PL some-ACC
They love each other.
One language can combine different types of doubling: in Bulgarian, we mostly
find transitive doubling, but also reciprocal intransitive doubling:
(26) Vseki p
t se sretn-at edin s drug.
each time REC meet-PRS.3PL one with other
They meet each other every time.
2.2.3 Types of minor markers
As I have said, in many of SAE and Slavic languages, there are peripheral re-
ciprocal markers which occupy an intermediate position between grammatical
and lexical markers. Let me list some of them:
x Russian: vzaimo-, vzaimno, obojudo-, obojudno etc.
x Bulgarian: vzaimno
x English: reciprocally, mutually

x French: rciproquement, mutuellement
NEDJALKOV (2007) calls units of this type reciprocal modifiers. Their principal
property is that they cannot be the sole markers of reciprocity in the clause
they must be supported by another (purely grammatical) productive marker.
Syntactically, they are often adverbs however, there are some exceptions when
peripheral markers become morphological prefixes, such as Russian vzaimo- or
French entre-.
As I will show below, the presence of peripheral also has some semantic ef-
fects constructions with PM do not always mean the same thing as construc-
tions with purely grammatical markers.
Therefore, in the reciprocal domain, a specific situation exists: in most languag-
es of Europe, there are three levels of reciprocal expressions:
x morphological / clitical
x pronominal (anaphorical)
x minor markers (adverbial)
Note that this situation is rather rare not only for European languages, but also
for the languages of the world. Existence of three levels of synonymous markers
which can occur simultaneously in many combinations is far from being a
typologically frequent situation.
Each of them has its special function:
x Morphological / clitical markers not only denote reciprocity, but also
serve for valency increase.
x Pronominal markers denote reciprocity. They are often used for intransiti-
ve verbs.
Bulgarian:
(27) Te vjarv-at edin v drug.
they.NOM believe-PRS.3PL one in other
They believe in each other.
x minor markers denote that the situation are symmetrical.
2.2.4 Minor marker + morphological marker vs. minor marker + lexical marker
Bulgarian: minor markers can only co-occur with se (non-autonomous marker).
Russian: minor markers can co-occur both with -sja (morphological marker) and
drug druga (lexical marker).
German:
French: the minor marker rciproquement can co-occur with se; moreover,
triple reciprocal construction, as in (29), exists:
(28) Ainsi, Matre et disciples se respectent rciproquement.

Therefore, the Teacher and his pupils respect each other (mutually).
(29) Ils se parlent l'un l'autre mutuellement.
They talk to each other (mutually).
Existence of complex combinations proves that the three levels of markers carry
out different functions. The combinations of clitics and pronominal markers are
pleonastic in that their semantic content is the same. However, as I have said,
their degree of grammaticalization and syntactic properties are different. The
function of peripheral markers
2.2.5 Morphological status of minor marker
In most SAE languages, minor markers are autonomous (adverbs): cf. French,
English, Spanish mutuo, recproco, mutuamente, recprocamente; German ge-
genseitig, wechselseitig:
German:
(30) sich gegenseitig berhren touch each other
French: both autonomous and morphological markers. However, the
morphological marker entre- is not productive:
(31) s'entredvorer eat each other
(32) se flatter rciproquement praise each other
Russian: both autonomous and morphological markers
(33) vzaimn-aja zavisimost- = vzaimo-zavisimost-
mutual-N.SG.F dependence-SG.NOM REC-dependence-SG.NOM
mutual dependence
Bulgarian: both autonomous and morphological markers:
(34) vsek-i etap e vaen- i vzaimo-sv
rzan- s ostanal-i-te.
each-SG.M stage be.3SG important-SG.M and REC-concerned SG.M
with rest-PL-DEF.PL
Each stage is important and related to the rest.
2.2.6 Order of derivations
In languages with morphological minor markers, the problem of order of deriva-
tions occurs (main marker vs. minor marker).
Russian:
(35) Et-i kultur-y vzaimo-obogaaj-ut drug drug-a.
this-NOM.PL culture-NOM.PL VZAIMO-enrich-3PL.PRS each.other-ACC
These cultures mutually enrich each other.
x if drug druga is attached first, this is strange because the morphological

marker vzaimo- is attached after the syntactic marker drug druga;
x if vzaimo- is attached first, this is strange because verbs like
vzaimoobogaat in (35) do not exist without drug druga.
Only in Russian can the prefix yield the reciprocal meaning to verbs which are
otherwise non-reciprocal:
(36) astic-y unitoaj-ut-sja.
particle-PL.NOM annihilate-3PL.PRS-SJA
i. The particles annihilate / are annihilated.
ii. *The particles annihilate each other.
(37) astic-y vzaimo-unitoaj-ut-sja.
particle-PL.NOM VZAIMO-annihilate-3PL.PRS-SJA
The particles (in physics) annihilate each other.
In all languages, deverbal nouns can also be modified by drug druga-like mar-
kers.
2.2.6 Sociative/reciprocal polysemy
Comitative / sociative / reciprocal polysemy is considered to be one of the most
frequent types of polysemy (see NEDJALKOV (ed.) 2007), cf. Turkic languages:
Khakas:
(38) Olar sayn-(y)s-a-lar.
(s)he.PL think-REC-PRS-PL
They think of each other. (reciprocal)
(39) Praj-zy olar xoryx-(y)s-a-lar direktor-da.
all.3SG.POSS (s)he.PL fear-REC-PRS-PL director-INS.SG
All of them are afraid of their director. (associative)
In SAE and Slavic languages, it does not exist for main (lexical and
grammatical) markers. However, it occurs in the domain of minor markers. In
(40) in Russian, the adjective vzaimnyj bears the reciprocal meaning, whereas in
(41), it apparently shows the sociative meaning. In (42) in Bulgarian, the
adjective also has the sociative meaning, while in (43) it is used reciprocally:
Russian:
(40) vzaimn-aja simpatij-a
mutual-N.SG.F sympathy-SG.NOM
mutual sympathy;
(41) vzaimn-aja vygod-a
mutual-F.SG.NOM advantage-SG.NOM
mutual advantage (= advantage of all participants).
Bulgarian:
(42) vzaimn-a-ta radost koj-a-to dostavj-at
mutual-SG.F-DEF.F joy which-SG.F-SUFF give-3PL
zanjatij-a-ta nadete-to i naroditel-i-te
seminar-PL-DEF to child-DEF.N and to parent-PL-DEF.PL
mutual joy which these seminars give to the child and to the parents (not
reciprocal (joy from each other); rather sociative (joy of all participants).
(43) vzaimn-a-ta ljubov
mutual-SG.F-DEF.F love
mutual love
Spanish:
(46) utilidad reciproca mutual advantage (advantage of all parties sociative).
This polysemy seems to be characteristic for the whole Europe. Note that purely
grammatical markers do not demonstrate it. However, it is important to take into
account that the sociative / comitative meaning demonstrated by reciprocal mar-
kers is restricted with a small set of contexts. For instance in (37), the adjective
vzaimnyj mutual cannot denote sociativity.
Russian:
(47) *vzaimn-aja igra
mutual-NOM.SG.F play
mutual play
I suppose that the sociative meaning is possible only if in the context there is a
reciprocal component:
mutual advantage = advantage caused by success of a process by the two
participants together with each other
The reciprocal component must be present in the semantic structure: A gets ad-
vantage because of some Bs (and As) actions; B gets advantage because of
some As (and Bs) actions. This type of sociative can be an intermediate stage
between sociative proper and reciprocal.
A question arises: Why doesnt this polysemy exist among the grammatical
markers? I suppose that there are two reasons:
x se-like markers are derived from reflexive markers (see HASPELMATH
for details) which are not characterized by this type of polysemy
x drug druga-like markers are case-marked in accordance with valency
structure of the verb. Therefore, they must only denote two different ar-
guments.
So far we have mainly analyzed cases where the reciprocal relation connects the
subject and the direct object:
A beats B A and B beat each other.

However, in all languages under analysis it is also possible to reciprocalize ar-
guments with other syntactic status, e.g. S and IO.
A gives B money A and B give each other money.
A relies on B A and B rely on each other.
Languages under analysis can behave in three ways in this respect.
First of all, the same marker can denote reciprocity between S and DO and
reciprocity between S and IO / Oblique. This is the case in SAE languages. In
French, the expression se donner un main give a hand to each other includes
the same marker se as in transitive reciprocals. Here it is used to mark S / DO
and S / IO reciprocity.
The second variant is that the indirect type of reciprocals is marked with
another marker. For instance, in Bulgarian transitive reciprocals are marked with
the clitic se in the accusative case. In indirect reciprocals, the same clitic takes
the dative case form:
Bulgarian:
(49) Te si pomogna-xa.
they.NOM REFL.DAT help-AOR.3PL
They helped each other.
Finally, in Russian the S / IO reciprocity cannot be marked inside the verb form.
Only in one verb, namely, (po)sovetovat, indirect reciprocity can be marked
with -sja:
Russian:
(50) a. Vasj-a posovetova-l- Petj-e ujexa-t.
Vasja-SG.NOM advise-PST-SG.M Petja-SG.DAT leave-INF
Vasja advised Petja to leave.
b. Vasj-a i Petj-a posovetova-l-i-s.
Vasja-SG.NOM and Petja-SG.NOM advise-PST-PL-REFL
Vasja and Petja discussed the situation with each other.
Even this example is not a pure example of indirect reciprocal. Strictly speaking,
the verb sovetovatsja does not means advise (something) to each other. It
rather means to discuss something.
Conclusions
We can conclude that reciprocal and reflexive constructions in SAE (Romance,
Germanic) and in Slavic languages do not allow us to distinguish two clear-cut
types: SAE reflexives / reciprocals and Slavic reflexives / reciprocals. As in
many other domains, we deal with a continuum, rather then a binary opposition.
For instance, the nature of polysemy of se-like markers does not yield us two
clear-cut types. It rather distinguishes Russian from all other languages. In Rus-
sian, the main uses are the passive and anticausative ones. The uses denoting
different types of co-reference of arguments (reflexive and reciprocal) are cha-
racteristic only for a small group of verbs. In other languages, these uses are
productive. Perhaps, the only property of reflexive / reciprocal marker allowing
us to distinguish SAE and Slavic languages is high (Slavic) vs. low (SAE) fre-
quency of se-marked passives.
Let us consider now the means of marking the reciprocal meaning. First, all
languages under analysis, whether they are Slavic or represent the SAE type,
have three types of reciprocal markers: (i) se-like markers which demote
transitivity of the base verb; (ii) each other-like markers which do not demote
transitivity but denote reciprocity and coreference between arguments and (iii)
markers which we called peripheral which denote that the situation is
symmetrical and, thus, reciprocal. Ways of combining the (i) and (ii) types are
different, but the type of doubling the reciprocal markers cannot serve as a
borderline between the two types: the transitive type of doubling is observed in
French (SAE) and Bulgarian (Slavic), the intransitive type can be found in
German (SAE) and Russian (Slavic). As I have shown, this parameter is strongly
correlated to another one: syntactic properties of the (ii)-type markers. In
languages with transitive doubling they cannot serve the sole marker of
reciprocity and seem to build a small clause or be a part of a complex marker
together with a se-like marker.
One more parameter does not distinguish any groups at all: namely,
meaning of peripheral markers. In all languages under analysis, peripheral
markers get a sociative use.
In contrast, the parameter reciprocalization of intransitive predicates
distinguishes more than two types. There are languages where this type of
reciprocalization is (nearly) impossible (Russian); languages, where it is
expressed in the same way as reciprocalization of intransitive predicates
(French, German) and languages where it is expressed in another way
(Bulgarian).
Therefore, we cannot speak of existence of two distinct types of reciprocal
formation: one for SAE, the other one for Slavic languages. However, it is
interesting that some systemic principles underlying the system of reciprocal
markers are the same for European languages: namely, presence of three levels
of markers and reciprocal-sociative polysemy only in the sphere of peripheral
markers.
References
BECK, S. (2001) Reciprocals are definites. Natural language semantics 9.1.
BEHRENS, L. (2008) Amy Winehouse schlug sich mit seinem Ehemann Gedanken zur Re-
ziprozitt und (In-)transitivitt. Handout of the talk at Transitivity Workshop. Kln.
GUENTCHVA, Z. & N. RIVIRE (2007) Reciprocal and reflexive construction in French. In:
NEDJALKOV V.P., GUENTCHVA, Z. & GENIUIEN, E.. (eds.) (2007) 561-608.
HASPELMATH, M. (2007) Further remarks on reciprocal constructions. In: NEDJALKOV V.P.,
GUENTCHVA, Z. & GENIUIEN, E.. (eds.) (2007) 2087-2115.
HASPELMATH, M. (2001) Non-canonical marking of core arguments in European languages.
In: AIKHENVALD, A., R.M.V. DIXON & M. ONISHI (eds) Non-canonical marking of
subjects and objects. Amsterdam: Benjamins. 53-83.
KEMMER, S. (1993) The middle voice. Amsterdam / Philadelphia: Benjamins.
KNJAZEV, J. P. (2007) Leksieskaja tipologija. Russkij jazyk v tipologieskom osvesenii.
Moscow: Yazyki Russkoj Kultury.
LETUCHIY, A.B. (2009) Russian double reciprocals. In: Press in Proceedings of the confe-
rence FDSL 7.5 (Moscow, 2008).
NEDJALKOV V.P., GUENTCHVA, Z. & GENIUIEN, E.. (eds). 2007. Reciprocal
Constructions. Typological Studies in Language, 71. Amsterdam / Philadelphia: John
Benjamins.
NEDJALKOV, V.P. Questionnaire on reciprocals. In: NEDJALKOV V.P., GUENTCHEVA, Z. & GE-
NIUSIENE, E.. (2007) 379-434.
NEDJALKOV, V.P., & B. WIEMER (2007) Reciprocal and reflexive constructions in German. In:
NEDJALKOV V.P., GUENTCHVA, Z. & GENIUIEN, E.. (eds.) (2007) 455-512.
WIEMER, B. (2007) Reciprocal and reflexive constructions in Polish. In: NEDJALKOV V.P.,
GUENTCHVA, Z. & GENIUIEN, E.. (eds.) (2007) 513-560.
Alexander Letuchiy, Russian Academy of Science, Vinograd Russian Language

Institute, Department for Corpus Linguistics and Linguistic Poetics, Volkhonka
str 18/2, 119019 Moscow, Russia, alexander.letuchiy@gmail.com
PPs of Different Sizes1
Nina Radkevich
1. Introduction
The syntactic structure of PPs has been debated for three decades. Since VAN
RIEMSDIJK (1978), it has been recognized that spatial PPs have a complex
internal structure, beyond simply [PP P [NP]]. Although the many current
proposals (see references below) differ in many respects, there is a consensus
that at a minimum, there are at least two layers of functional structure within
spatial PPs: an inner layer denoting location, and an outer layer denoting
direction/motion (with authors disagreeing as to whether the outer layer is
present for static/locational expressions, such as on the table), as in (1)2. This
complexity is transparent in some constructions (e.g., Russian: iz-pod doma
from under the house, English: into the house) but is generally posited even
where Ps are not visibly complex. Ranging from the minimum structure in (1),
many proposals (especially BOKOVI 2004, DEN DIKKEN 2006, SVENONIUS to
appear) posit more articulated functional structure within PPs.
(1) Path
3
Place
3
on DP
the table
It has been previously suggested that PPs have a full clausal structure similar to
CP (BOKOVI 2004b, NOONAN 2004, DEN DIKKEN 2006). In this paper I
suggest on the basis of a comparative study of Slavic and Romance languages
that PP functional structure is not uniform across languages. More specifically, I
propose that there may be cross-linguistic variation in the amount (but not the
ordering of) functional projections in the PP, in line with similar proposals about
variation in functional inventories in the inflectional domain (BOBALJIK 2002,
BOBALJIK & THRINSSON 1998), in the size of infinitives (WURMBRAND 2001,
etc.), and the DP/NP domain (CORVER 2003, BOKOVI 2008).
__________
1 I am especially grateful to my language informants: Miloje Despi, Neda Todorovi
(both Serbo-Croatian), Magdalena Mullek (Slovak), Pavel Caha (Czech), Krzysztof
Migdalski (Polish), Simona Herdan (Romanian), Benjamin Girard-Bond (French), Carlos
Buesa Garca (Spanish), Maria del Carmen Parafita Couto (Galician).
2 However, there are proposals arguing for poor functional structure of PPs, e.g., ABELS
(2003).
316 Nina Radkevich
I examine three diagnostics for PP-internal functional structure across Slavic

(Russian, Polish, Czech, Slovak, and Serbo-Croatian) and Romance languages
(French, Spanish, and Galician): first, I discuss availability of measure phrases
(MP) in PPs in these languages, then, I examine binding properties in PPs, and,
finally, I compare languages with respect to the possibility of quantifier float
within PPs. Based on the three diagnostics listed above, I propose that languages
differ with respect to the amount of functional structure in PPs, as in (2);
moreover, the amount of functional structure can vary even within a single
language. Some assumptions are necessary to spell out about the structure in (2).
First, I assume that all languages have the minimum structure of [DIRPP Dir [LOCPP
Loc [NP/DP]]]. Second, there are implicational relations between functional
projections in the structure in (2), which go top-down, i.e. the presence of a
higher projection (XP) implies the presence of a lower projection (MP), but the
presence of a lower projection (MP) does not imply the presence of a higher
projection (XP).
(2) YP Serbo-Croatian, Spanish, Galician
3
3
Y XP Slovak, Romanian
3
3
X MP Polish, Czech
3
3
M DirPP Russian, French
3
3
P LocP
3
3
P NP/DP
2. Degree modification in PPs
The possibilities for degree modification in PPs have been discussed by
KOOPMAN (2000) and DEN DIKKEN (2006) with respect to Dutch and English.
Den Dikken observes that PPs with measure phrases are ambiguous. Consider
the following example in (3). The sentence in (3) is at least two-way ambiguous:
it can mean that the ball was flying at distance of 10 meters above the fence and
it can also mean that the ball landed 10 meters behind the fence, as illustrated in
(4). Den Dikken calls these two readings Place and Path respectively. He
interprets these facts about ambiguity of sentences with degree modifiers in PPs
as evidence for two structural positions for measure phrases in PP, yielding two
PPs of Different Sizes 317
types of interpretations: one is in the locational component and the other one is
in the directional component of PP.
(3) John threw the ball 10 meters over the fence. (ambiguous)
(4)
10 feet
10 feet
Place Path
Interestingly, languages vary with respect to the availability of measure phrase

modification in PPs. The Slavic languages under discussion do not behave
uniformly. First, one language, namely Russian, does not have measure phrases
with PPs at all, as in (5-6).
(5) *Ivan brosil mja 10 metrov za zabor. (Russian)
Ivan threw ball 10 meters behind fence
Ivan threw the ball 10 meters over the fence.
(6) *Ivan brosil mja 10metrov nad zaborom.
Ivan threw ball 10 meters above fence.
The only way to have degree modifiers in Russian is to use another PP to
introduce them, as in (7-8).
(7) Ivan brosil mja na 10 metrov za zabor. (Russian)
Ivan threw ball on 10 meters over fence
Ivan threw the ball 10 meters over the fence. (Path)
(8) Ivan brosil mja na 10 metrov nad zaborom.
Ivan threw ball on 10 meters above fence. (Place)
Other Slavic languages covered in this small survey do not pattern with Russian
but with English and Dutch in allowing measure phrases within PPs. Consider
the following example from Serbo-Croatian (9), which allows measure phrases
in PPs, like English and Dutch. However, Serbo-Croatian is different from these
two languages: it does not show the ambiguity found in English and Dutch, but
it uses two different prepositions to express Place and Path meanings.
318 Nina Radkevich
(9) a. Jovan je bacio loptu 10m iznad ograde. (Serbo-Croatian)

Jovan is threw ball 10m over fence (Place)
b. Jovan je bacio loptu 10m preko ograde. (Path)
Jovan is threw ball 10m over fence
Other Slavic languages (Czech, Slovak, and Polish) pattern with Serbo-Croatian:
first, they allow degree modification in PPs; second, they do not show
ambiguity, as in (3), but use two individual prepositions for the two meanings
(Path and Place). The examples from Czech, Slovak, and Polish illustrating this
point are given in (10-12).
(10) a. Honza hodil ten baln 10 metr pes plot. (Czech)
Honza threw this ball 10 meters over fence (Place)
b. Honza hodil ten baln 10 metr za plot. (Path)
Honza threw this ball 10 meters behind fence
(11) a. Jn prehodil loptu p metrov nado hradou. (Slovak)
Jan threw ball five meters above fence (Place)
b. Jn prehodil loptu p metrov za ohradu. (Path)
Jan threw ball five meters behind fence
(12) a. Jan rzuci+ pi+kdziesi metrw nad p+otem. (Polish)
Jan threw ball ten meters above fence (Place)
b. Jan rzuci+ pi+kdziesi metrw za p+ot. (Path)
Jan threw ball ten meters behind fence
Interestingly, the Romance languages that I surveyed behave like Slavic
languages in some aspects. First, French, like Russian, does not allow measure
phases with PPs, as shown in (13). The only way to make the sentence
grammatical is to use another preposition, as in (14). French, however, is still
different from Russian: the sentence in (14) has two readings Place and Path.
(13) *Jean a lance la balle 10 mtres par-dessus la clture. (French)
John has thrown the ball 10 meters over the fence
(14) Jean a lance la balle 10mtres par-dessus la clture. (French)
John has thrown the ball at 10meters over the fence
All other Romance languages (Galician, Spanish, and Romanian) allow degree
modification in PPs. Nevertheless, these languages differ with respect to the
Path-Place ambiguity. Spanish and Galician are the Romance counterparts of the
Slavic languages of the paper (besides Russian): Spanish and Galician have
measure phrases in PPs but they do not show the ambiguity found in English and
Dutch, as shown in (15) and (16).
(15) a. El chico golpe le pelota 10meters sobre le valla. (Spanish)
The boy kicked the ball 10meters over the fence (Place)
b. El chico golpe le pelota 10meters ms all de le valla.

The boy kicked the ball 10meters over the fence. (Path)
(16) a. Jon tirou a pilota 10metros sobre a cancela. (Galician)
Jon threw the ball 10meters over the fence (Place)
b. Jon tirou a pilota 10metros mais alo da cancela.
Jon threw the ball 10meters overof.the fence (Path)
The last language of the present study is Romanian. Romanian behaves like
English and Dutch: Romanian allows degree modification of PP and shows the
Path-Place ambiguity.
(17) Ion a aruncat mingea zece metri peste gard. (Romanian)
John has thrown ballthe ten meters over fence
As has been shown above, Slavic and Romance languages do not behave
uniformly with respect to measure phrases in PPs: some languages do not allow
them at all (Russian, French), some languages (Serbo-Croatian, Czech, Slovak,
Polish, Spanish and Galician) allow measure phrases in PP but do not show the
Path-Place ambiguity with one preposition, whereas Romanian exhibits
syntactic behavior similar to English and Dutch, i.e. it allows degree
modification of PPs and such expressions are ambiguous between Place and
Path interpretations. The differences in the behavior among Slavic and Romance
languages with respect to degree modification in PPs can be explained the
following way: not all languages have a measure phrase (MP) projection
(Russian, French); what is more, I suggest that there is only one MP in
languages that allow them and the Path-Place ambiguity can be attributed to
differences in lexical meanings of prepositions across languages, i.e. in some
languages (Romanian, English) a preposition is polysemous, whereas in other
languages (Spanish, Serbo-Croatian, etc.) it is not.
3. Binding in PPs
It is well-known that complementarity between pronouns and anaphors breaks
down in PPs (HESTVIK 1991, REINHART & REULAND 1993), as in (18).
(18) Maryi put the gun near heri/ herself.
Interestingly, pronouns co-referential with a clause-mate subject are not allowed
in non-spatial PPs.
(19) Maryi does not like talking about *heri / herself.
It has been proposed that co-referential pronouns are possible in spatial PPs
because spatial PPs constitute a binding domain for pronouns, therefore, they
can be co-indexed with subjects. HESTVIK (1991) suggests that there is a null
subject in spatial PPs, which is obligatorily controlled by the object of the verb,
as shown in (20).
320 Nina Radkevich
(20) Maryi put the gunj [PROj near heri].

In the case of non-spatial PP, co-referential pronouns are impossible because
they form a syntactic unit with VP (HESTVIK 1991, REINHART & REULAND
1993).
There are two possible ways to address the binding facts in PPs: to adopt
domain-based theories (HESTVIK 1991, STURGEON 2003, among others) or to
assume that the possibility of co-referential pronouns in PPs can be attributed to
the rich functional structure of PPs. The choice between these two options will
be made at the end of the paper.
Not all languages show complementarity between anaphors and co-
referential pronouns in PPs. The Slavic languages under discussion are not
uniform: some languages do not have option of using co-referential pronouns in
PPs, whereas others allow them. Russian, Czech, and Polish show a similar
pattern with respect to the binding facts: pronouns co-referential with a clause-
mate subject are not possible even in PPs. I illustrate this fact with data from the
three languages in (21-23).
(21) Ivani poloil rujo vozle sebjai/*negoi. (Russian)
Ivan put gun near self/ him
Ivan put the gun near himself.
(22) Mariei poloila ten revolver vedle sebei/*niii. (Czech)
Maria put this gun near self/ her
Maria put the gun near herself.
(23) Mariai po+o\y+a bro{ ko+o siebiei/*nieji. (Polish)
Maria put gun near self her
Maria put the gun near herself.
Unlike the three languages above, Serbo-Croatian and Slovak have option of co-
referential pronouns in PPs, as shown in (24-25)3.
(24) Jovani je video psa blizu sebei/?njegai. (Serbo-Croatian)
Jovan is saw dog near self/ him
John saw a dog near him
(25) Mriai poloila zbra blzko seba/?neji. (Slovak)
Maria put gun near self her
The Romance languages are more uniform in their behavior with respect to
binding in PPs. Unfortunately, French does not have a SELF-anaphor
comparable to SELF-anaphors in other Romance languages under discussion:
__________
3 It is important to point out that judgments of acceptability of co-referential pronouns vary
across speakers. Miloje Despi (p.c.) and Magdalena Mullek (p.c.) note that for Serbo-
Croatian and Slovak the anaphor is more preferable for some speakers.
the French anaphor lui-mme has emphatic and contrastive meanings (Labelle
2008). Therefore, it is impossible to test if a SELF-anaphor is the only variant,
since French is forced to have a co-referential pronoun in PPs due to the lack of
the SELF-anaphor. However, in all other Romance languages discussed in the
paper both options (SELF-anaphor and co-referential pronouns) are available in
spatial PPs, as illustrated in (26-28).
(26) Mariai puso la pistola cerca de ellai/ s. (Spanish)
Maria put the gun near of her self
(27) Mariai a pus pistotul lng eai./sine. (Romanian)
Maria has put gun-the near her self
(28) Mariai puxo a pistola preto delai./se mema (Galician)
Maria put the gun near of.her self
In this section of the paper I have shown that not all languages show non-
complementarity between anaphors and co-referential pronouns in spatial PPs:
two Slavic languages (Serbo-Croatian and Slovak) and all Romance utilize both
options, while the rest of Slavic languages only allow anaphors in these
contexts. In the next section, I will use the third test, the availability of quantifier
float in PPs, to determine the amount of functional structure in PPs.
4. Quantifier float in PPs
The next diagnostic for PP-internal functional structure has to do with the
availability of quantifier float in PPs. BOKOVI (2004b) argues for a PP/clause
parallelism based on certain facts concerning quantifier float in Germanic
languages. He notes that quantifier float in PPs is only possible with definite
NPs, which is also true for object shift in Icelandic: only definite NPs can
undergo object shift, as in (29-30). BOKOVI (2004b) points out a correlation
between these two phenomena concluding that object shift is responsible for
quantifier float in PPs.
(29) a. *g talai vi stdentai alla ti. (BOKOVI 2004b: 104)
I talked with students all (Icelandic)
b. g talai vi alla stdenta
I talked with all students
c. ?g talai vi stdentanai alla ti.
I talked with the-students all
(30) a. Halldr las bkurnari allar ti.
Halldr read the-books all
b. *Halldr las bkuri allar ti.

Halldr read books all
c. cf. Halldr las allar bkur.
322 Nina Radkevich
BOKOVI (2004b) shows that quantifier float of definite NPs is possible under
assumption that the functional structure of VP must be rich enough to allow
movement of NPs. The same is true for the movement of definite NPs in PPs.
BOKOVI (2004a) shows that quantifier float is disallowed in -positions. Since
this, (30a) is derived as follows. NP first undergoes movement to a non-
position; the quantifier, which is an adjunct, gets inserted acyclically, then, NP
moves to another functional projection, leaving the quantifier behind in a non-
position, as in (31).
(31) Halldr lasj [OP bkurnari tj [ekki [AgroP [allar ti] tj [VP tj ti]]]].
Halldr read the-books not all (BOKOVI 2004b: 107)
The assumptions underlying the analysis of quantifier float in VP (BOKOVI
2004a) are extended to the analysis of this phenomenon in PPs. Due to the ban
on quantifier float in -positions, the relevant NP must move to a non--position
before the acyclic insertion of the quantifier itself, then, NP moves to the next
functional projection, whereas the quantifier remains in its original (non-)
position. In other words, we need more structure than in (1) to have floating
quantifier constructions4. (32) shows the minimal structure needed, under
Bokovis analysis, for quantifier float in a locative PP.
(32) *under windows all
YP
3
3
Y XP
underj 3
windowsi 3
X MP
3
all ti 3
M DirPP
3
3
P LocP
3
3
tj ti
In the light of the discussion in this section, the possibility of quantifier float is
thus a good test for additional structure in PPs. Languages which, by the test
above, lack sufficient functional structure in PPs should therefore also disallow
__________
4
I assume that -domain in PPs includes both DirP and LocP.
quantifier float in PPs. First, I will discuss the availability of this phenomenon in
the Slavic languages. Then, I will compare them with the Romance languages.
The Slavic languages discussed in the paper are not uniform in their behavior
with respect to the availability of quantifier float in PPs. Russian, Czech, Polish,
and Slovak disallow quantifier float in PP, as in (33-36).
(33) *Ivan prool vdol rek vsex. (Russian)
Ivan walked along rivers all
Ivan traveled along all rivers.
(34) *Honza se podval do bar vech. (Czech)
John REFL looked into bars all
John looked into all bars.
(35) *Zagldneli`my do ksi\ek wszystkich. (Polish)
looked-1pl in books all
We looked in all books.
(36) *Jn sa plavil po riekach vsetkch. (Slovak)
Jan has swam on rivers all
Jan sailed along rivers all.
Interestingly, Serbo-Croatian does not pattern with the rest of the Slavic
languages in that it does allow quantifier float in PPs.
(37) Plovili smo niz reke sve. (Serbo-Croatian)
swam are down rivers all
We traveled along all rivers.
The Romance languages do not behave uniformly with respect to this test.
Romanian and French disallow quantifier float in PPs, as shown in (38) and
(39).
(38) *Jean est all dans les bars tous (cherchant Jeanne) (French)
John is gone in the bars all (looking for Jane)
John went into all bars looking for Jane.
(39) *Ion s- a uitat n barurule toate cutnd-o pe Ioana. (Roman)
Ion refl-has looked in bars-the all searching-her.cl PE Ioana
John looked into all bars, looking for Jane.
However, Spanish and Galician can have floating quantifiers in PP, as illustrated
in (40-41).
(40) Juan busc en los bares todos (mientras buscaba a Ben) (Spanish)
John looked into the bars all (while looking for Ben)
John looked into all bars while looking for Ben.
324 Nina Radkevich
(41) Jon mirou en bares todolos. (Galician)

John looked into bars all
Following BOKOVI (2004b), we can conclude that Spanish, Galician, and
Serbo-Croatian have more structure in PPs than Romanian, French, Russian,
Czech, Polish, and Slovak. The results of the three tests discussed so far are
presented in the table below.
Table 1: summary of the properties of PPs cross-linguistically
Language MP Pronouns in PP QF
Russian * * *
French * n/a *
Polish * *
Czech * *
Slovak *
Romanian *
Serbo-Croatian
Spanish
Galician
The languages in the survey can be divided into four groups: 1) languages that
do now allow any of the three phenomena (Russian, French); 2) languages that
allow only MP in PP but not the other two phenomena (Polish, Czech); 3)
languages that allow MP and co-referential pronouns in PPs, but disallow
quantifier float (Romanian, Slovak); 4) languages that allow all the three
phenomena (Serbo-Croatian, Spanish, and Galician).
I hypothesize that the differences reflected in the table above can be
attributed to differences in the amount of structure in PPs, i.e. Russian and
French have the smallest amount of functional structure, whereas Serbo-
Croatian, Spanish, and Galician have the biggest amount of functional structure
in PPs, as in (42).
(42) Russian, French >>Polish, Czech>>Romanian, Slovak>>Serbo-Croatian,
Spanish, Galician
The next step is to confirm that the structural hierarchy in (42) is preserved
when the tests are combined and to see whether their interaction can reveal
further insights into the functional structure of PPs. I will focus only on the
languages that allow MP, co-referential pronouns, and floating quantifiers in PPs
(Serbo-Croatian, Spanish, and Galician).
5. Interaction between the tests
In this section I will combine several tests to see if the hypothesis about
rich/poor functional structure of PPs holds. First, I will test an interaction
between quantifier float and binding in PPs. Second, I will look at an interaction
between measure phrases and quantifier float.
5.1 Quantifier float and binding in PPs
In the previous section I showed that the availability of quantifier float in PPs
presupposes a certain amount of structure. If there is more structure, then the
binding facts may change since PP may be big enough to constitute a separate
binding domain. Unfortunately, Spanish and Galician do not have possessive
SELF-anaphors; therefore, I will not be able to use them here. On the other
hand, Serbo-Croatian has both a possessive SELF-anaphor and pronouns.
Consider the following sentence from Serbo-Croatian. The co-referential
pronoun is still slightly degraded.
(43) Marijai je zasadila rue ispod svih svojihi/njenihi prozora. (SC)
Maria is planted roses under all selfs ?her windows.
Maria planted roses under all her windows.
However, when the quantifier svih is floated, the acceptability of the co-
referential pronoun increases. Moreover, the possessive pronoun is preferred to
the possessive SELF-anaphor5.
(44) ?Marijai je zasadila rue ispod ?(*)svojih/njenihi prozora svih. (SC)
Maria is planted roses under selfs her windows all.
The increased acceptability of co-referential pronoun in PPs in sentences with
floating quantifiers may be due to the fact that PPs with floating quantifiers have
more (richer) structure which is sufficient to constitute a binding domain. What
is more, the sentence in (44) is an argument for some degree of optionality in the
structure of PPs: whenever a language allows pronoun to be co-referential with a
clause-mate subject, it has the full structure (up to YP), while language has
smaller structure (up to MP), if it does not allow co-referential pronouns in PPs.
5.2 Interaction between quantifier float and measure phrases
There are three languages that allow both quantifier float and degree
modification in PPs: Serbo-Croatian, Spanish, and Galician. The sentences
which have quantifiers and measure phrases in PPs are slightly degraded in all
three languages, as shown in (45-47).
(45) ? Marija je zasadila rue 10m ispod svih prozora. (Serbo-Croatian)
Maria is planted roses 10m under all windows.
Maria planted roses 10 meters under all windows.
(46) ? Maria plantou rosas 10metros debaixo de todalas fiestras. (Galician)
Maria planted roses 10metros under of all windows
__________
5 For one of my consultants the pronominal option is actually ungrammatical in (44).
326 Nina Radkevich
(47) ? Maria planto rosas 10metros debajo de toda ventanas. (Spanish)

Maria planted roses 10meters under of all windows
However, the situation changes when we try to float a quantifier in these
sentences: they become ungrammatical, as in (48-50).
(48) *Marija je zasadila rue 10m ispod prozora svih. (Serbo-Croatian)
Maria is planted roses 10m under windows all.
(49) *Maria plantou rosas 10metros debaixo de fiestras todalas. (Galician)
Maria planted roses 10metros under of windows all
(50) *Maria planto rosas 10metros debajo de ventanas toda . (Spanish)
Maria planted roses 10meters under of windows all
The impossibility of quantifier float in PPs with measure phrases can be
straightforwardly accounted for under BOKOVIs (2004a) analysis of quantifier
float discussed in the previous section. Recall that there is ban on quantifier float
in -positions. First, NP must move to the closest non--position, i.e. MP, and
the quantifier gets inserted acyclically. Then, NP moves up to the next
projection (XP). However, in the cases with measure phrases this is impossible
because NP cannot move to the specifier of the first functional projection (MP),
since it is filled with the measure phrase, as shown in (51). Quantifier float in
PPs is only possible when languages have more functional structure.
(51) YP
3
3
Y XP
3
3
X MP
3
10m 3
M DirPP
3
3
P LocP
3
3
P NP/DP
under all windows
6. Conclusion
In this paper I compared the syntactic behavior of languages from two language
groups, namely Slavic and Romance, with respect to the three diagnostics of the
functional structure of PPs: the possibility of degree modification, co-referential
pronouns, and quantifier float in PPs. The results of the tests indicate that
languages can have different amount of functional structure in PPs: Russian and
French have the smallest PPs, whereas Serbo-Croatian, Spanish, and Galician
can have the largest PPs, since they allow quantifier float PP which requires
more functional structure (BOKOVI 2004a, 2004b) than in other languages of
the survey. I also showed that the amount of functional structure can vary even
in the same language, as in the case of Serbo-Croatian.
References
ABELS, K. (2003) Successive cyclicity, anti-locality, and adposition stranding. Ph.D
dissertation, University of Connecticut.
BOBALJIK, J. (2002) A-Chains at the PF-Interface: Copies and "Covert" Movement. Natural
Language and Linguistic Theory 20.2, 197-267.
BOBALJIK, J. & H. THRINSSON (1998) Two heads aren't always better than one. Syntax 1.1,
37-71
BOKOVI, . (2004a) Be careful where you float your quantifiers. Natural Language and
Linguistic Theory 22:681-742.
BOKOVI, . (2004b) Object shift and the clause/PP parallelism hypothesis. Proceedings of
WCCFL 23: 99-112.
BOKOVI, . 2008. What will you have, DP or NP. In Proceedings of NELS 37.
CORVER, N. (2003) On three types of movement within the Dutch nominal domain. In:
COENE, M & Y. DHULST (eds.). From NP to DP. v.1: Syntax and semantics of noun
phrases 297-328. Amsterdam: John Benjamins.
DEN DIKKEN, M. (2006) On the functional structure of locative and directional PPs. Ms.,
CUNY.
HESTVIK, A. (1991) Subjectless binding domains. Natural Language and Linguistic Theory 9:
455-497.
KOOPMAN, H. (2000) Prepositions, postpositions, circumpositions and particles: The structure
of Dutch PPs. In: H. KOOPMAN (ed.) The syntax of specifiers and heads pp. 20460.
London: Routledge.
LABELLE, M. (2008) The French reflexive and reciprocal se. Natural Language and Linguistic
Theory 26: 833-876.
NOONAN, M. (2004) German shadows: revealing the structure of spatial PPs. Ms., McGill
University.
REINHART, T. & E. REULAND (1993) Reflexivity. Linguistic Inquiry 24: 657-720.
STURGEON, A. (2003) Two-tiered binding domain formation: evidence from Czech. In:
Proceedings of FASL 11: 495-514.
SVENONIUS, P. (to appear) Spacial P in English. In: CINQUE, G. & L. RIZZI (eds.) The
Cartography of Syntactic Structures, vol. 6. Oxford: Oxford University Press.
VAN RIEMSDIJK, H. C. (1978) A case study in syntactic markedness: the binding nature of
prepositional phrase. The Peter de Ridder Press, Lisse.
328 Nina Radkevich
VANRIEMSDIJK, H. C. (2007) Case in spatial adpositional phrases: the dative-accusative

alternation in German. In: ALBIOU, G., A. AVRAM, L. AVRAM, & D. ISAC Festschrift
for Alexandra Cornilescu. Bucharest University Press: Bucharest.
WURMBRAND, S. (2001) Infinitives: restructuring and clause structure. Mouton de Gruyter.
Nina Radkevich, Harvard University, Linguistics Department, Boylston Hall,

Third Floor, Cambridge, MA 02138, USA, nina-radkevich@yahoo.com
Multiple Sluicing: A purely syntactic account
Tanya Scott
1. Introduction
This paper provides an account of multiple sluicing in Russian, which heavily
relies on the analysis of the multiple wh-behavior in Russian given in SCOTT
(2000). Its advantage is that, in essence, it reduces the account of Russian
multiple sluicing to the MERCHANT-style (2001) analysis of IP-elision, and
allows us to curtail additional mechanisms and explanatory devices necessary
when Russian is believed to be a non-wh movement language, a trend started by
STEPANOV (1998).
Traditionally the definition of sluicing comes from ROSS (1969) (followed
by MERCHANT 2001) given below in (1):
(1) Sluicing is a phenomenon of IP-ellipsis, where a wh-phrase moves out of IP,
and IP is deleted at PF (ROSS, 1969; MERCHANT, 2001). It represents a
construction where there is an interrogative clause with only a wh-element
(wh-elements) pronounced. It occurs in embedded clauses, (2), as well as in
main clauses, (3).
(2) Carl devoured somebody but I dont know who[Carl devoured]. (NG)
(2) Max proglotil kogo-to, no ja ne znaju kogo [Max proglotil] (RU)
Max devoured someone but I not know whoACC Max devoured
Max devoured someone but I dont know who(m).
(3) A: Carl devoured somebody. (ENG)
B: Who [Carl devoured] ?
(3) A: Max proglotil kogo-to. (RU)
Max devoured someone
B: Kogo [Max proglotil] ?
whoACC Max devoured
ROSS (1969) claimed that in sluicing the elided part forms a non-constituent.
Thus, his solution was to anlyze sluicing as regular wh-fronting with ellipsis of
the remaining part of the wh-phrase (question). Rosss ideas were further
detailed by MERCHANT (2001). Thus, the constructions in (1) and (2) are
analyzed as shown in (4)
330 Tanya Scott
(4) Standard Account (ROSS 1969, LASNIK 1999, MERCHANT 2001)
The syntax of (4) is based on the syntactic properties of ellipsis (following

LOBECK 1991, SAITO & MURASUGI, 1997 among others). Thus, the ellipsis head
is an empty category, and it is important to understand what licenses the
processes of ellipsis. It is assumed (MERCHANT, 2001; STJEPANOVI, 2003,
TAKAHASHI 1993, and others) that sluicing is related to the properties of the wh-
behavior in a particular language. Thus, one should assume that if wh-movement
is possible, then sluicing is possible. Subsequently, if wh-movement is not
possible, then, the question arises whether sluicing is possible (MERCHANT,
1998). For instance, languages that dont have multiple wh-movement dont
have multiple sluicing English, and languages that have multiple wh-
movement should have multiple sluicing, i.e. Bulgarian. Then, the following
questions necessarily arise:
(i) What positions do wh-phrases occupy in Russian sluices?
(ii) What functional projection(s) license(s) Sluicing?
We have already seen that Russian exhibits sluicing behavior, but so does
English in cases where wh-fronting occurs. It also appears that sluicing occurs in
Russian in multiple wh-contexts, where multiple wh-remnants are possible
(following TAKAHASHIs 1984 definition). However, it has been widely claimed
that Russian does not exhibit multiple wh-movement (STEPANOV 1997,
STRAHOV 2001, GREBENYOVA 2004), at best wh-fronting happens. In the next
section we take a closer look at what happens in multiple sluices.
2. Multiple Sluicing Behavior
There are two general observations about the word order in Russian multiple
sluiced constructions (GREBENYOVA 2004, 2006; SCOTT 2004): (i) in the matrix
clauses the order of wh-remnants is somewhat free; (ii) in the embedded clauses,
the order of the wh-remnants is fixed. This is shown in examples (5) (8) for
the embedded and (9) (10) for the matrix clauses respectively.
(5) Kadyj priglasil kogo-to na tanec, no ja ne pomnju kto kogo
everyoneN invited someoneA to dance but I not remember whoNwhomA
Everyone invited someone to a dance but I dont remember who (invited)
whom.
Multiple Sluicing: A purely syntactic account 331
When in the matrix clause the nominative quantifier precedes the accusatively
marked quantifier, and the nominative wh precedes the accusative wh-word in
the embedded clause, the sentence is fine. However, in (6) where the matrix
clause looks the same as in (5), but the order of the wh-words in the embedded
clause is reversed the sentence crashes.
(6) *Kadyj priglasil kogo-to na tanec, no ja ne pomnju kogo kto
everyoneN invited someoneA to dance but I not remember whom whoN
whom.
(7) Kogo-to kadyj priglasil na tanec no ja ne pomnju kto kogo
someoneA everyoneN invited to dance but I not remember whoN whomA
whom.
In (7) the main clause is identical to the one in (8), but the wh-word in the
nominative case precedes the wh-word in the accusative case and the sentence is
ok. Conversely, the sentence in (8) have both the quantifier in the accusative and
the wh-word in the accusative case precedes the nominative quantifier and the
nominative wh-word in the matrix and the embedded clause respectively, and it
results in an ungrammatical sentence.
(8) *Kogo-to kadyj priglasil na tanec no ja ne pomnju kogo kto
someone everyone invited to dance but I not remember whomA who
whom
Now, if we look at the matrix clause sluices only, given in (9A-10A), we notice
that the order of the statement (the part that corresponds to the matrix clause in
the previous examples) is free, and the orders of the question responses (what
corresponds to the embedded clauses in the previous examples (9B, B 10B,
B) are also free with respect to Superiority.
(9) A: Kadyj priglasil kogo-to na tanec .
everyoneN invited someoneA to dance
B: Kto kogo?
whoN whomA
B: Kogo kto?
whomA whoN
(10) A: Kogo-to kadyj priglasil na tanec.
someoneA everyoneN invited to dance
B: Kogo ktoN?
whomA whoN
B: Kto kogo?
whoN whomA
332 Tanya Scott
The table below summarizes all the possibilities of word orders presented in the
examples above and their combinations.
Clause Embedded Matrix

Matrix Nom > Acc Acc > Nom Nom > Acc Acc > Nom
Nom >
Acc 9 8!!! 9 9!!!
Acc >
Nom 9!!! 8!!! 9!!! 9
Table 1. Summary of NOM>ACC matrix/matrix; matrix/embedded
combinations
We have observed that multiple sluicing exists and there is Superiority or
something like it in the embedded clauses, and a lack thereof in the matrix
multiple sluices. GREBENYOVA (2005, 2006, 2007) gives an account of multiple
wh-sluicing in Russian. She uses STEPANOVs (1998) widely accepted analysis
of wh-behavior as the basis for her analysis of the sluiced constructions. For her
then, the asymmetry in Superiority requirements in the matrix and the embedded
clauses are problematic, because Superiority is usually triggered in the CP-
domain. Thus, for her, the licensing IP deletion (Merchant-style) cannot be
accounted for; there should be another way of accounting for multiple wh-
remnant behavior (i.e. superiority effects or lack thereof). Additionally, we
cannot assume wh-fronting into CP, and an account of either some other licensor
of the elision is needed. Since for Grebenyova no wh-fronting into the CP
domain is available, GREBENYOVA (2005) is forced to propose an alternative
account of sluicing: a semantic approach to Superiority in Russian. She is forced
to admit that Superiority effect emerges in sluicing in the embedded clauses, and
the explanation for it is semantic. Namely, she uses the notion of Semantic
Parallelism of FIENGO & MAY (1994) defined as variables in the elided and
antecedent clauses are bound from parallel positions. But as clearly seen from
the table above such explanation fails to explain the situations marked with
triple exclamation marks (see Table 1). I show such examples here using NOM
vs. DAT case marked pronouns and wh words in (11) and (12)1
(Komu-to kto-to zvonit s utra, no ja ne znaju kto komu.
someoneDAT smbd.NOM called since morning but I not know whoN whoDAT
Somebody calls someone in the morning, but I dont know who [calls] whom.
__________
1 Even though some case hierarchies other than NOM ACC- DAT exist for Russian, it is
rather difficult to account for any other Superiority effects that do not involve nominative
case.
(?*Komu-to kto-to zvonit s utra, no ja ne znaju komu kto.

someoneDAT smbd.N called sincemorning but I not know whoDAT whoDAT
Somebody calls someone in the morning, but I dont know who [calls] whom.
Thus, it is evident that an existing account is not satisfactory, and a categorically
different, but perhaps not an entirely new approach to Russian sluicing is called
for. It is not new in a sense that a similar account for sluicing environments has
been offered for Serbian/Croatian by STJEPANOVI (2003). STJEPANOVI (2003)
considers sluicing in SC as PF deletion (+E feature) of the constituent after the
wh-phrase, which is interpreted at LF. The technical implementation of this
proposal is that placement of [+E] feature triggers deletion at PF. STJEPANOVI
(2003) states that since wh-phrases must be in Spec, CP in sluicing constructions
(a special environment in which wh-phrases are in CP-domain), C0 must be
present in overt syntax in order to license the IP (or possibly TP) deletion.
Strong [+wh] feature of C0 triggers Superiority effects. It is radically new in a
sense that calls for a new look at the Russian wh-behavior in general.
3. Russian Multiple Wh
I claim that Russian is a multiple wh-movement language where wh-phrases
move into CP, counter STEPANOV 1998, STRAHOV 2001, and following SCOTT
2000, and LIAKIN 2004. The problem that has always existed in understanding
Russian wh-behavior is accounting for some optional Superiority facts, and
more specifically for an asymmetry in Superiority facts between the matrix and
the embedded clauses. In fact, the same asymmetry, we have observed in the
sluicing environments above. Consider the following asymmetry (13 a, b vs. 14
a, b):
(13) a. Kto to posovetoval Darii?
whoNOM whatACC advised DariaDAT
Who advised what to Daria?
b. to kto posovetoval Darii?
whatACC whoNOM advised DariaDAT
Who advised what to Daria?
! SUPERIORITY DOES NOT HAVE TO HOLD IN MATRIX CLAUSES
(14) a. Maria ne znaet kto to posovetoval Darii?
Maria not knows whoNOM whatACC advised DariaDAT
Maria doesnt know who advised what to Daria?
b.?*Maria ne znaet to kto posovetoval Darii?
Maria not knows whatACC whoNOM advised DariaDAT
Maria doesnt know who advised what to Daria?
! SUPERIORITY DOES HOLD IN THE EMBEDDED CLAUSES
334 Tanya Scott
However, I go further and claim that in essence, Superiority holds in both matrix
and the embedded clauses. It is the availability of an extra position that is
located higher than CP: a Higher Operator Phrase (and I have evidence to argue
that it is a Topic Phrase TopP2) of the matrix clauses that allows an escape
path from Superiority. This position is not available in the embedded clauses;
hence, the asymmetry between the clauses. The structure then, is given in (15)
and the tree structure for both (14 a and b) is given in (16) with respective
arrows.
(15) [TopP [SpecCP wh wh][IP]
(16) Kto to posovetoval Darii?
who what advised Daria
to kto posovetoval Darii?
what who advised Daria
To take this point even further and to give some brief additional evidence for
this position, I state that Russian exhibits a pattern in which when this position is
occupied by an overt Topic, the Superiority between the wh-elements must hold,
because the escape path to the higher domain is now blocked. While I am not
going to go into too much detail of the mechanisms here, I present the contrast
in (17) vs. (18).
(17) Mae kto toi podaril ti?
MashaDAT whoNOM whatACC gave-a-gift
Who gave what to Masha?
(18)3 *? Mae toi kto podaril ti?
MashaDAT whatACC whoNO gave-a-gift
Who gave what to Masha?
__________
2 In the spirit of RIZZI (1997)
3 There might be a certain contradiction in judgments between speakers. The discussion of
the availability of registers is beyond the scope of this paper. Only the data relevant here
is presented for more discussion see SCOTT (2000), SCOTT (2010) (forthcoming).
When TopP is occupied by an overt topic-phrase (in this case Maa Masha, but
it can be pretty much any constituent), the movement of a wh-phrase is blocked
(19).
(19)
Thus, I have briefly established that Russian exhibits multiple wh-behavior,

where all wh-phrases end up in the C-domain, i.e. true wh-movement. Thus, it
could be sufficient to state that Russian wh-phrases in sluices are also located in
CP.
4. Back to Sluicing
If all wh-phrases are located in CP, then an account of sluicing becomes rather
trivial. The existence of an extra position in matrix clauses allows us to dispose
of any additional explanations, such as Semantic Parallelism for Superiority to
hold. Such analysis also does not require the invention of any additional elision
licensors, such as FocP (suggested Hungarian by Liptk and van Craenenbroek
2008). For instance, assigning such an additional licensor would suggest that we
should expect utterances that allow both a wh-element and an overt Focus at the
edge of the elision to be grammatical, but they are not (20, 21).
(20) *Prepodavateli vydavali studentam diplomy no ja ne zametil kto
professors gave students diplomas but I Neg noticed who
STEPANU.
STEPAND-FOC
The professors gave out diplomas to students, but I didnt notice who
[gave a diploma] to STEPAN.
336 Tanya Scott
(21) *Ispolniteli predstavljali raznye pesni na konkurs no ja ne

singers performed different songs at competition but I Neg
remember who GorodokFOC
pomnju kto GORODOK
Singers performed different songs at the competition, but I dont
remember who (performed a song) LITTLE TOWN.
Therefore, it appears that analysis of Russian multiple sluicing becomes rather
straightforward if multiple wh-behavior in Russian is approached accurately,
which I believe this account achieves. No additional mechanisms become
necessary, i.e. the search for extra licensors or semantic justification for
syntactic requirements such as Superiority.
5. Conclusions
In this article, I present an account of Russian multiple sluicing on the basis of
the re-analysis of the Russian wh-behavior, based on the idea that sluicing is
traditionally believed to be dependent on the wh-behavior of a given language.
The incompatible assumptions about Russian being a non-wh-movement
language led to the analyses of Russian sluicing to become creative and to apply
Semantic mechanisms to account for Superiority asymmetries. The account
given here claims that a true wh-movement takes place in Russian where
[+wh] feature is strong, and that Superiority must hold in wh-contexts in
Russian. The existence of an external functional category TopP which is
available in the matrix clause only allows violation of the Superiority. Under
such analysis multiple sluicing is merely reduced to IP deletion where all wh-
remnants are in CP- domain, which essentially supports the la MERCHANT
(2001) analysis of sluicing. The approach suggested here has far reaching
consequences, of course, to the other wh-related and unrelated phenomena, such
as coordinated wh-constructions, particle usage, and the structure of the left
periphery in general.
References
BOKOVI Z. (1996) Fronting Wh-phrases in Serbo-Croatian. In: LINDSETH, M. & S. FRANKS
(ed.). Proceedings of Formal Approaches to Slavic Linguistics V 86-107. Ann Arbor,
Michigan Slavic Publications.
BOKOVI Z. (1998) Wh-phrases and Wh-Movement in Slavic. Comparative Slavic
Morphosyntax.
BOKOVI, Z. (1994) ECP, Spec-Head Agreement, and Multiple Wh-Movement in Overt
Syntax. In: AVRUTIN, S., S. FRANKS AND L. PROGOVA (ed.) Proceedings of Formal
Approaches to Slavic Linguistics II 119-143. Ann Arbor, Michigan.
BOKOVI, Z. (1999) On Multiple Feature Checking: Multiple WH Fronting and Multiple
Head Movement. In: Epstein, D. S. & N. Hornstein (eds.) Working Minimalism Ch. 6.
159-188. MIT Press, Cambridge, MA.
CHOMSKY, N. (1995) The minimalist program. Cambridge, Mass.: MITPress.
CRAENENBROECK, J. & A. LIPTK (2008) On the Interaction between Verb Movement and
Ellipsis: New Evidence from Hungarian. In: C. B. CHANG AND H. J. HAYNIE (eds.)
Proceedings of the 26th West Coast Conference on Formal Linguistics 138-146.
Cascadilla Proceedings Project, Somerville, MA.
FIENGO, R. & R. MAY (1994) Indices and Identity. MIT Press: MA
GREBENYOVA, L. (2005) Sluicing and Left-branch Extraction out of Islands. In: ALDERETE, J.,
C. HAN & A. KOCHETOV (eds.) Proceedings of the West Coast Conference on Formal
Linguistics (WCCFL) XXIV 164-172.
Grebenyova, L. (2006) Sluicing puzzles in Russian. In: LAVINE, J., S. FRANKS, M. TASSEVA-
KURKTCHIEVA & H. FILIP (ed.) Formal Approaches to Slavic Linguistics 14: The
Princeton Meeting 157171. Ann Arbor: University of Michigan, Michigan Slavic
Publications.
GREBENYOVA, L. (2007) Sluicing in Slavic. Journal of Slavic Linguistics 15(1): 49-80.
MERCHANT, J. (2001) The syntax of silence: Sluicing, islands, and the theory of ellipsis.
Oxford University Press: Oxford.
LASNIK, H. (2000) When can you save the structure by destroying it? NELS meeting, 2000.
Liakin, D. (2004) On Wh-movement and Wh-in-situ in Russian. In: FRANKS, S., F. GLADNEY
& M. TASSEVA-KURKTCHIEVA (ed.) Proceedings of Formal Approaches to Slavic
Linguistics the South Carolina Meeting 218-229. Ann-Arbor, Michigan.
Ross, J.R. (1969) Guess who? In: BINNICK, R. I., A. DAVISON, G. M. GREEN & J. L. MORGAN
(ed.) Papers from the Fifth Regional Meeting of the Chicago Linguistic Society 252-
286.
SCOTT, T. (2001) Russian questions and the question of Topics. Ms. SUNYat Stony Brook.
SCOTT, T. (2002) CP-Topic Parameter and Russian wh. Paper presented at the 79th Annual
Meeting of the Linguistic Society of America, Atlanta, GA
STEPANOV, A. (1998) On Wh-fronting in Russian. In: TAMAJI, P. N. & K. KUSOSMOTO (ed.)
Proceedings of NELS 28 453-467.
STJEPANOVI, S. (2003) Multiple wh-fronting in Serbo-Croatian matrix questions and the
matrix sluicing construction.
STRAHOV, N. (2001) A Scrambling Analysis of Russian WH-questions. In: FRANKS, S., T.
HOLLOWAY KING & M. YADROFF (ed.) Proceedings of Formal Approaches to Slavic
Linguistics IX 293-310. Ann Arbor, Michigan Slavic Publications.
Tanya Scott, Stony Brook University, Department of Linguistics, SUNY at

Stony Brook, NY 11794-4376, USA, tane.scott@gmail.com
Wh-scope marking strategies in Polish
Joanna mieciska
Abstract
The paper presents an analysis of two strategies for long wh-scope marking in
Polish. The jak w construction is treated in the recent literature (STEPANOV 2001,
LUBANSKA 2004) in line with the Indirect Dependency Approach (DAYAL 2000).
A competing construction co, e w is, to the best of my knowledge, absent from
the literature. Because of the numerous restrictions of jak w as opposed to the
classical Hindi IDA examples (Dayal 2000), I argue that at the syntactic level
this construction is best analysed as a paratactic sequence of two independent
questions, and that, since semantically it doesnt let itself to lambda conversion
analysis of DAYAL (2000: 163), as jak in Polish is not an argument but an
adjunct, equivalent to viewpoint markers (cf. WILLIM 1989:113), it does not fit
in the IDA long wh-scope marking. The co, ze w construction, on the other hand,
takes an argument wh in the matrix clause (case marking of what depends on the
verb). An indirect semantic dependency between co and the embedded wh can,
thus, be postulated. Under the analysis pursued here, syntactically, co originates
as an object of the matrix clause with an appositive propositional modification
of the second clause, and is then moved to the clause initial position. Co is, thus,
treated as a classical quantifier over propositions and parallels the Hindi IDA
constructions discussed by DAYAL (2000:160-164. The claims are made on the
basis of comparison of the two constructions and an empirical survey on 66
Polish native speakers. It follows that jak w is bound to the immediate situation
by means of tense (present only), choice of addressee (deictic you only) and
choice of verbs (a few verbs of saying, thinking only). Co, e w is flexible as far
as the tense, the choice of verbs and the addressee are concerned, it is however
restricted to a highly d-linked or echo reading context only.
1. Introduction
Polish is described in the literature as involving Indirect Dependency Approach
(DAYAL 1994, 1996, 2000) in sequential scope marking (STEPANOV 2001,
LUBASKA 2004). Both authors base their assumptions on the apparent analogy
between the jak, w construction (1) and corresponding examples from Hindi (2).
The paper challenges this conjecture.
(1) Jak my`lisz, co powinni`my zrobi?
How think2sing pres whatacc should1stpl. doinf.
What do you think we should do?
340 Joanna mieci{ska
(2) Jaun kyaa soctaa hai ki merii kis-se baat karegii?

J. what think-PR that Mary who-INS talk do-F
Who does John think Mary will talk to? (DAYAL 2000:190)
The two main proposals are that; first, contra Stepanov and Luba{ska, jak w
involves a paratactic sequence of a viewpoint adjunct and a genuine question
rather than either a direct or an indirect semantic dependency between jak and
the wh-expression; and second, that a competing construction (co, e wh) does
involve IDA. These are proposed on the basis of an analysis of the differences
between the two constructions and an empirical survey concerning
grammaticality judgments of long distance dependencies among young educated
native speakers of Polish.
Section 2 of the paper presents the main qualities of the jak w construction
in Polish and its alleged Hindi counterpart. In section 3 the lexical, syntactic,
semantic and prosodic differences between Polish jak w and co, e w
constructions are discussed. Section 4 reports on an empirical survey concerning
the choice of the addressee and tense in the two constructions, as well as their
grammaticality judgments. In the final section the results of the survey and their
implications are discussed.
2. Polish long dependencies and the DDA/ IDA distinction.
Long distance questions are impossible in Polish. A common Polish analogue of
the English long wh-movement (3) or the German wasw construction (4) is the
jak w construction (5).
(3) Who do you think [we should ask t]?
(4) Was glaubst du [wen wir fragen sollen t]?
(5) Jak my`lisz, kogo powinni`my zapyta?
How think2sing pres whoacc should1stpl. askinf.
While (3) and (4) involve syntactic subordination and can be subsumed under
the so called Direct Dependency Approach (MCDANIEL (1989), MLLER (1997),
BECK & BERMAN (2000) and CHENG (2000)) as there is a direct syntactic and
semantic dependency between the upper clause wh-element and its trace (in the
case of the German construction was is standardly treated as an expletive
replaced at LF by its contentful counterpart (wem)), Polish (2) does not allow for
the same analysis. Jak w does not involve syntactic embedding (the
complementiser e that in corresponding indicative complex sentences is
obligatory (6)).
(6) My`l, */\e powinni`my zapyta Henia.
think1sing that should1stpl. askinf H.
I think we should ask Henio.
Wh-scope marking strategies in Polish 341
Accordning to LUBASKA (2004) and STEPANOV (2001), with whom I share the
view on non applicability of DDA for Polish, jak w (2) is Polish scope marker
analogous to Hindi examples discussed by Dayal and should be subsumed under
the Indirect Dependency Approach, as in (7).
(7) Jaun kyaa soctaa hai [ki merii kis-se baat karegii?]
J. what think-PR that Mary who-INS talk do-F
Who does John think Mary will talk to? DAYAL (2000:190)
2.1 Dayals IDA for Hindi
Under the IDA the relation between the interrogative phrases in the relevant
constructions is considered to be indirect. The wh element in the matrix clause is
assumed to be an argument of the matrix verb which qualifies over propositions
rather than individuals, and which is moved higher up. The status of the
embedded CP under IDA is that of an appositive clause interpreted as a
restriction of the wh-object. According to DAYAL (2000), the wh-expressions
kyaa and kis-se in (7) do not enter into a direct relationship with each other.
Rather, they belong to two local dependencies, indirectly linked by coindexation
of the trace (lowest copy) of kyaa (wh1) and the CP that dominates wh2...whn.
The IDA does not attribute matrix scope to the second wh-expression at any
level of syntactic representation. Nevertheless, the set of answers to questions
like (7) includes the same possibilities as would be allowed for if one
maintained that there is a direct dependency between the argument position and
the matrix spec position in (7) holding, for instance, at transparent LF, as
claimed by MAHAJAN (1990). That is why, Dayal provides a semantic analysis
of long wh-dependencies in Hindi, which assumes indirect relation but in effect
allows for the same set of possible answers as the direct dependency. The set of
possible answers under Dayals approach is determined by the matrix wh-
phrase. The kyaa in the example determines that the quantification will be over
propositions to which John stands in the think relation. Dayals claim is that
there is a logical possibility of obtaining the same set of answers to the long wh-
dependencies in Hindi under the assumption that the kyaa element is not an
expletive but a contentful wh-expression quantifying over propositions. Thus, in
Hindi the scope marker kyaa is assumed to originate in argument position and to
be coindexed with a CP in an adjoined position.
2.2 Polish jak w does not involve IDA
The assertion that Polish jak w (1) is parallel to Hindi kyaa w (2,7) and that it
involves IDA is, however, problematic for a number of reasons.
First, unlike Hindi kyaa Polish jak is not an argument, but an adjunct:
(8) A: Co o nim my`lisz?
Whatacc about him think2sing pres
What do you think about him?
B: (My`l), \e si zmieni+.
think1st.sing, that himself changed
(I think) that he has changed.
(9) A: Jak o nim my`lisz?
How about him think2sing pres
How are you thinking about him?
B: (My`l o nim) ciep+o.
(think1st.sing,about him) warmly
Warmly.
Any use of jak in Polish be it in a direct or in an embedded question is always
adverbial and never nominal. Thus, sematically jak w does not let itself to
lambda conversion analysis of DAYAL (2000: 163), which is at the heart of
Dayals proposal, where the set of possible answers is determined by the matrix
wh-prase. Jak should be a contentful wh-element quantifying over propositions.
It cannot be an expletive, let alone an adjunct, if it is to fulfill this condition.
Next, aside jak w, at least 3 additional options exist in Polish and could be
used as translations of English long wh-extraction (10).
(10) Who do you think John will bring along?
(11) Twoim zdaniem, kogo Janek przyprowadzi?
Yourinstr opinioninstr whoacc Janek bring 3singfut
Who will J. bring along in your opinion?
(12) Co my`lisz, \e kogo Janek przyprowadzi?
what think2sing pres that whoacc Janek bring 3singfut
Who do you think J. will bring along?
(13) My`lisz, \e kogo Janek przyprowadzi?
think2sing pres that whoacc Janek bring 3singfut
Who do you think J. will bring along?
None of the options involves a DDA; still, subsuming them all under IDA would
be a major oversimplification. The availability of these options is not random,
but, as will be shown, closely related to pragmatic and semantic factors.
3. Differences between jak w and co, e w
Structurally the most strikingly different counterpart of (10) is the co, e w
construction (12). To the best of my knowledge it is absent from the literature,
but is used in present day Polish1. Its syntactic structure makes it a good
analogue of Hindi IDA; (12) takes an argument wh in the matrix clause (case
__________
1 Based on observations of spontaneous speech and google search, see appendix 1 for
examples from internet blogs and forums.
marking of what depends on the verb, see (14) below). An indirect semantic
dependency between co and the embedded wh can, thus, be postulated.
(14) a. Co/*Jak o tym my`lisz?
What/How about it think2sing pres
What do you think about it?
b. Czego si dowiedzia+e`, (\e kogo Janek przyprowadzi)?
Whatgen refl. find out2sing past that whoacc Janek bring 3singfut
What did you find out?
(Who did you find out J. will bring along?)
Table 1. presents syntactic, lexical, semantic and pragmatic differences between
jak w and co,ze w.
Table 1. Differences between Polish jak w as in (1) and co, e w, as in (5).
SYNTACTIC DIFFERENCES
jak w tense in jak clause: present only

(viewpoint *Jak (po)mylaa, kto przyjdzie?
adjunction) how think2sing past whonom come3sing fut
Who did you think would come?
Jak mylisz, kto przyjdzie?
how think2sing past whonom come3sing fut
Who do you think would come?
co, e w tense in co clause: present, future, past

(wh-scope Co (po)mylaa, e kto przyjdzie?
marking)
what think2sing past that who come3sing fut
Who did you think would come?
jak w direct yes-no questions in the second clause possible

Jak mylisz, (czy) ja jestem naiwny?
how think2sing pres, whether I am nave
Do you think I am nave?
co, e w direct yes-no questions in the second clause not likely, unless
paratactic, with a long pause between the two parts
*Co mylisz, ( czy) ja jestem naiwny?
what you think, yes/no q operator I am naive?
jak w complementation with that-clause impossible

*Jak mylisz, e kto przyjdzie?
how think2sing pres that who come3sing fut
Who do you think will come?
co, e w complementation with that-clause obligatory
Co mylisz, e kto przyjdzie?
whatacc think2sing pres that who come3sing fut
Who do you think will come?
jak w multiple embedding impossible as no 3rd person is allowed
*Jak mylisz, jak on sdzi, jak ona uwaa...
How think2sing pres how he thinks, how she thinks
co, e w multiple embedding possible

Co mylisz, e co on uwaa, e kogo powinnimy
zapyta?
what think2sing pr that whatacche thinks that whoacc should1pl ask.
Who do you think he thinks we should ask?
jak w order of clauses reversible

Kogo Janek przyprowadzi, jak mylisz?
co, e w order of clauses irreversible

*e kogo Janek przyprowadzi, co mylisz?
LEXICAL DIFFERENCES
jak w A few verbs of saying and thinking: myle, sdzi, uwaa

(verbs of thinking), wydawa si (seem), mwi (say),
przypuszcza (suppose)
verbs: my`le, sdzi, uwa\a, mwi, twierdzi (claim),

co, e w utrzymywa (maintain), dowiedzie si czegogen (find out
what), odkry (discover) etc., all bridge verbs, i.e. non-factive
verbs taking that-clause complementation in declarative
sentences, but also factive verbs.
PRAGMATIC DIFFERENCES
jak w verb forms in jak clause: 2nd person singular and plural, 3rd
person singular honorific
(addressee of the whole sentence necessarily equal to the
subject of the jak clause)
Jak Marysia myli, gdzie Piotr poszed
(Marysia as the addressee)
verb forms in co clause: 2nd, 3rd sing. and pl.

co, e w
Co Marysia myli, e gdzie Piotr poszed?
(mostly addressee from outside of the sentence)
SEMANTIC DIFFERENCES
jak w Jak adjunct
co, e w Co argument
jak w Relatively week d-linking, fewer presuppositions.
Jak mylisz, gdzie on poszed?
co, e w a. Relatively strong d-linking, more pressupositions

A ty co mylisz, e po co ja te walizki nosz ?
And why do you think I am carrying these suitcases along?
b. Narrow scope, asking for repetition context
Co on powiedzia, e gdzie ona posza?
He said that she went where?
PROSODIC DIFFERENCES
jak w Information seeking, falling intonation
500
Pitch (Hz)
0
0 5.5
Time (s)
Jak mylisz, gdzie on poszed

co, e w b. echo-reading, rising intonation
500
Pitch (Hz)
0
0 2.99
Time (s)
Co mylisz, e kogo on lubi?
4. The survey
What are believed to be the most conspicuous differences between the two
constructions, i.e., tense and addressee choice (boldfaced in Table 1), were
posited on the basis of a pilot survey with 20 participants, and then verified with
a bigger study on 66 Polish native speakers.
4.1 Procedure
A sample of 5 questions (Appendix 2) were fed through Ivona text-to-speech
software, to avoid intonation cues, and played to the participants. The task was
to indicate the addressee of each question. The choice was between grandma,
Marysia and Szymon; the participants were encouraged to designate more than
one addressee per question whenever possible in an order from most to least
obvious. Next, the participants were given the written sentences and were asked
to rate the sentences in the written form as either acceptable or not acceptable.
4.2 Results
Note: Using a proper first name addressee as honorific 3rd person direct
addressee sounds very archaic, if not ridiculous, in Polish many participants
pointed it out, still choosing the direct addressee (Szymon in 2.ex1).
The survey shows a considerable bias (84,8% of 66 tokens) towards
analyzing the subject of the jak clause as the most likely addressee, compared to
39,4% of the subject of the co clause (example 2s vs 1s respectively). However,
the jak w with past tense (example 4s), was treated as most likely referring to
third party in 83,3% of 66 tokens. What should be noticed is that, at the same
time, this sentence was rated as unacceptable by 94% of participants. On this
basis one could conclude that jak w does not tolerate past tense. The past tense
with co, e w (example 6s) was treated as most likely referring to third party in
69,7% of cases and had a 93,4% acceptability.
Table 2. Addressee specification and grammaticality judgements of jak,w, co, ze
w and jak, ze w.
all 3 choices
first choice
unacceptable
unacceptable
all 3 choices
first choice
acceptable.
participants
acceptable
(total 66)
%
%
0s. Jak sdzisz, co c.66 c. 66 0 100 0
Marysia zjad+a? B. 25+41
S. B. 25
S. 41
1s. Co Szymon a.38 38 S a. 26 39,4 58 8 87,8 12,1
twierdzi, \e kogo b. 28 28 B b. 0 0
babcia lubi?
c. 34 34 M c. 40 60,6
2s. Jak Szymon sdzi, a. 86 75,5 S a.56 84,8 52 14 78,8 21,2
co Marysia zjad+a? b.0 0 M b. 0 0
c. 28 24,5 B c. 10 15,5
3s. Jak Marysia my`li a. 36 34,6 M a. 28 42,4 56 10 84,8 15,1
\e co babcia je? b.16 15,4 B b. 4 6,1
c. 52 50 S c. 34 51,5
4s. Jak babcia uwa\a+a a. 14 14,9 B a.7 10,6 4 62 6 94
co Szymon zjad+? b. 10 10,6 S b. 4 6,1
c. 70 74,5 M c. 55 83,3
5s. Co Marysia my`la+a a. 65 64,3 M a. 46 69,7 62 4 93,4 6,6
\e co Szymon zjad+? b. 6 5,9 S b. 4 6,1
c. 30 29,7 B c. 16 24,2
In each case (1 through 6) the (a) option stands for the first participant to appear
in the text, (b) the second one, and (c), a person from the outside of the text.
Thus, in 1 (a) stands for Szymon (S.), (b) stands for babcia (B.) and (c) stands
for Marysia (M.).
5. Conclusions
Jak w (1) is bound to the immediate situation by means of tense (present only),
choice of addressee (deictic you only) choice of verbs (a few verbs of saying,
thinking only).
The co, e w (4) is flexible as far as the tense, the choice verbs and the
addressee are concerned; it tends, however, to be more linked to echo reading.
The qualities of the co, e w construction make it a perfect analogue of the
Hindi IDA examples. Syntactically, co originates as an object of the matrix
clause with an appositive propositional modification of the second clause, and is
then moved to the clause initial position. Co can be, thus, treated as a classical
quantifier over propositions parallel to the Hindi IDA constructions discussed by
DAYAL (2000:160-164).
The jak phrase, on the other hand, due to its defective status, is in fact more
equivalent to a viewpoint marker (11) (as was also suggested by WILLIM
1989:113). At the syntactic level this construction should be analyzed as a
paratactic sequence of two independent questions or a view point adjunct
followed by an interrogative clause.
Thus, while DAYAL (2000) shows that there is a continuum between IDA
and DDA cross linguistically, and REIS (2000) has proven that that same
continuum holds intra-linguistically for German, one could also claim, on the
basis of Polish, that there is a continuum between a pure view point adjunct (in
your opinion) and IDA (with different levels of semantic and syntactic
integration of the two elements and differing pragmatic and prosodic qualities).
Appendix.1
Google serach results for co my`lisz, \e co
A ty co mylisz, e co ja w tej walizce nios?!?!
http://www.emotoryzacja.net/hyde-park-f13/dowcipy-t9.html
co mylisz e co dadz takie transy
http://www.kibice.net/forum/viewtopic.php?f=1&t=4485&sid=a0cd04aeb37903
fa9854dd519ca42241&start=480
No a co mylisz, e co ja na przerwach mam do roboty?
http://www.tombraiderforum.tcz.pl/printview.php?t=278&start=0
Appendix .2
0s. Jak sdzisz, co Marysia zjad+a?
how think2sing pres, what M. ate
What do you think Mary ate?
1s. Co Szymon twierdzi, \e kogo babcia lubi?

what S. thinks, that whom grandma likes
a. Who does S. think grandma likes?
b. Simon, who do you think grandma likes?
2s. Jak Szymon sdzi, co Marysia zjad+a?
how S. judges what M. ate
a. What does S. think M. ate?
b. S, what do you think M. ate?
?3s. Jak Marysia my`li, \e co babcia je?
how M. thinks, that what grandma is eating
4s. Jak babcia uwa\a+a, co Szymon zjad+?
How grandma thought, what S. ate?
5s. Co Marysia my`la+a, \e co Szymon zjad+?
what M. thought that what S. ate
a. What did M. think Simon ate?
b. M, what do you think Simon ate?
References
CHENG, L. (2000) Moving just the feature In: LUTZ, U., G. MLLER & A. VON STECHOW (eds.)
77-99.
DAYAL, V. (1994) Scope marking as indirect wh-dependency Natural Language Semantics 2:
137-170.
DAYAL, V. (1996) Locality in WH quantification: Questions and relative clauses in Hindi.
Dordrecht: Kluwer.
DAYAL, V. (2000) Scope Marking: Cross-linguistic Variation in Indirect Dependency In:
LUTZ, U., G. MLLER & A. VON STECHOW (eds.)Wh-Scope Marking 157-194. John
Benjamins.
LUBANSKA, M. (2004) Wh-scope marking in Polish In: Pozna Studies in Contemporary
Linguistics 39,73-88.
MCDANIEL, D. (1989) Partial and Multiple Wh-Movement Natural Language and Linguistic
Theory 7 565-604.
MLLER, G. (1997) Partial wh-movement and Optimality Theory Linguistic Review 14: 249-
306.
BECK, S. & S. BERMAN (2000) Wh-scope marking: direct vs. indirect dependency In: LUTZ,
U., G. MLLER & A. VON STECHOW (eds.) 17-44.
STEPANOV, A. (2001) Cyclic domains in syntactic theory. MITWPiL
WILLIM, E. (1989) On word order: A Government-Binding Study of English and Polish.
Zeszyty Naukowe Uniwersytetu Jagiello{skiego. Prace Jzykoznawcze 100, Krakow.
Joanna mieci{ska, Adam Mickiewicz University, al. Niepodleg+o`ci 4, 61-874

Pozna{, Poland, smiecjo@ifa.amu.edu.pl
How many da(s) are there in Serbian?
Natasha Todorovich
1. Introduction1
As noted by VRZI (1996) da is used in optatives, secondary imperatives and in-
terrogatives, in both matrix and embedded contexts. It is also used in condition-
als and the future tense constructions. What is known in traditional grammars of
Serbian as da + present is actually a complement construction very often se-
lected by a number of verbs, and can replace the infinitive in some instances. Al-
though the infinitive is still relatively productive in Serbian, it is less preferred to
da + present while in some dialects, such as Torlak, it is almost entirely absent
(JOSEPH (1983)). The loss of the infinitive is one of the characteristics of other
Balkan languages, for example Greek.
In the following discussion, I only focus on da-constructions that serve as
complements to indicative and subjunctive verbs. According to PROGOVAC
(2005), Serbian verbs can be classified as subjunctive selecting verbs and indica-
tive selecting verbs. Indicative verbs are verbs of saying, believing, claiming or
ordering while subjunctive verbs are verbs of wishing, requesting or wanting.
Similarly, GIANNAKIDOU (1998) classifies verbs in Greek as indicative and sub-
junctive verbs. The indicative group consists of assertive (to say, to read, to
claim), fiction verbs (to dream, to imagine), epistemic (to believe, to think), fac-
tives (to be glad, to know, to regret), and semifactives (to discover, to remem-
ber). GIANNAKIDOU (1998) further divides the subjunctive group of verbs into
two subgroups. I would label them here as a subjunctive subgroup 1which in-
cludes volitionals (to want, to hope, to plan), directives (to order, to advise, to
suggest), modals (must, it is possible), permissives (to allow), negative (to avoid,
to refuse, to forbid), and verbs of fear (to be afraid) and a subjunctive subgroup
2 which includes aspectual (to start, to continue), perception (to see, to hear),
commissive (to be forced to, to promise), and implicative verbs (to manage). I
adopt GIANNAKIDOUs (1998) classification of verbs for Greek and adapt it for
classification of Serbian verbs. While in Greek there is a number of different
particles and complementizers (some indicative such as oti and pu some sub-
junctive such as na and as) that are restrictively used after indicative or subjunc-
tive verbs in embedded clauses (GIANNAKIDOU 2009), in Serbian one particle da
or the same da-constructions are used after all of the indicative and subjunctive
verbs listed above. I examine the properties of da-constructions that serve as
__________
1 I would like to thank Anastasia Giannakidou, Lilia Schrcks, Steven Franks, Iliyana
Krapova, Guglielmo Cinque and many other participants at FDSL 8 for valuable feed-
back, comments, and suggestions.
352 Natasha Todorovich
complements to indicative and subjunctive verbs in relation to aspect, tense, ne-

gation, licensing of negative polarity items, and clitics.
2. Aspect
2.1 Aspect in da-complements of epistemics
In (1) the indicative, epistemic verb verovati/to believe allows da + imperfective
non-past (INP adopted from GIANNAKIDOU 2009) as a complement, but it does
not allow da+ perfective non-past (PNP) (the construction with na prefix). Per-
haps, the aspect restriction in the indicative complements is rather semantic than
syntactic. Following GINNAKIDOU (2009), veridical verb verujem/to believe in
sentence (1) is true in the moment of utterance and da + PNP cannot be used as
its complement to express the truth in the present since something that is perfec-
tive cannot be present too. More evidence for semantic restrictions of the aspect
in the complement comes from examples in (4) with aspectuals.
(1) a. Verujem da (*na) pie zadau
believe-1sg that (* perf pfx.) write-2sg.impf. homework
I believe that you are writing the homework.
2.2 Aspect in da-complements of volitionals
In (2a) the subjunctive, volitional verb, eleti/to want, selects either da+ INP or
da + PNP (with na prefix). The construction with the INP has the habitual/ pro-
gressive interpretation while the construction with the PNP has the future inter-
pretation. As proposed by GIANNAKIDOU (2009) subjunctive verbs are non-
veridical and are not true in the moment of utterance expressed 'now' in the
present; therefore, Serbian volitionals allow da + PNP which manifests the ac-
tion that cannot be completed 'now' in the moment of utterance.
While in (2a) the embedded subject is controlled by the matrix subject, in
(2b) the subject of the complement agrees in number and person with the com-
plement verb and is not controlled by the matrix subject. Subjects can separate
verbs from complementizers, and this behavior is observed in (2b).
(2) a. elim da (na)piem pismo.

want-1p.sg to (perf. pfx) write-1sg.impf/perf letter-acc.sg.
I want to write a letter.
b. elim da Jelena (na) pie pismo.
want-1sg that Helen (pref) write-3sg.perf/impf letter-acc.sg
I want that Helen writes a letter.
2.3 Aspect in da-complements of raising verbs
On the other hand, da +INP is the only possible option in (3a) after the verb iz-
gledati (to seem). Unlike the English verb to seem that is a raising verb, Serbian
izgledati, as noted by RADII (2006), does not allow raising or control. In this
How many da(s) are there in Serbian? 353
case, the subject must stay in the embedded clause usually positioned between
da and the embedded verb.
(3) a. Izgleda da Jelena (*na)pie pismo.
seem-3sg that Helen (perf.pfx)write-3sg.imperf. letter-acc.sg
It seems that Helen is writing a letter.
b. (*Jelena) izgleda da pie pismo.
Helen seem-3sg to write-3sg.imperf. letter-acc.sg.
Helen seems to write a letter.
2.4 Aspect in da-complements of aspectuals
For constructions as in (4a-b) BROWNE (1986) states that the action of the com-
plement verb overlaps in at least one point with that of the main clause, or it
lasts/continues until or after the action of the matrix clause is complete, so the
aspect of the complement verb must be imperfective and for this reason the pos-
sibilities with the perfective are illicit. BROWNE (1986) claims that da used in the
complements of aspectuals is the same da used after volitionals but different
from da used after epistemic and assertive verbs.
(4) a. Marko poinje da (*na)pie.
Mark start-3sg.impf. to (*perf.pfx)write-3sg.impf.
Mark starts to work.
b. Marko prestaje da (*na)pie.
Mark stop-3sg.impf to (*perf. pfx.)write-3sg impf.
Mark stops to work.
3. Tense
3.1 Tense in da-complements of assertives and epistemics
Indicative verbs select da-complements that can host their own tense which
could but does not have to be disassociated from the matrix tense. In (5a) the
past tense of assertive rei (to say) selects a da-complement with a verb in the
future tense. In (5b), the present tense of epistemic misliti (to think) selects a da-
complement with a verb in the past tense.
(5) a. Rekla sam da e Marko da doe.
said aux-1sg that aux-3sg. Mark to arrive-3sg.
I said that Mark will arrive.
b. Mislim da je Ivan napisao pismo.
think-1sg that aux-3sg Ivan wrote letter-acc.
I think that Ivan wrote a letter.
3.2 Tense in da-complements of volitionals

The common characteristic of the subjunctive constructions is that they cannot
host their independent tense. In (6a) the only acceptable da-construction after
the present tense of a volitional is da+ present. In (6b) the volitional in the past
or future tense selects only da+ present again. As GIANNAKIDOU (2009) propos-
es for Greek, the tense of the verbal dependent is anaphoric and picks up the
tense of the higher clause. While the aspect selection is optional in volitional da-
complements where da + PNP or da + INP are both accepted, the tense is not.
As BULATOVI (2008) notes, da is seen as a binder of a dependent tense and a
tense of the matrix, subjunctive verb.
(6) a. elim da idem /(*u ii2) /(*sam otiao).
want-1sg to go-1sg /(will go-inf.) /(*aux-1sg left)
b. Marija je htela/e hteti da proita knjigu.
Mary aux 3sg. wanted/will want-inf. to read-3sg.perf book-acc.
Mary wanted /will want to read a book.
3.3 Tense in da-complements of modals
The impersonal form treba (it is necessary) derived from the verb trebati (to
need) also selects da-complements that behave much like the da-complements
of the volitional verb eleti (to want.) Although da-complements of either aspect
can complement treba, only the present tense can be used after treba as exam-
ples in (7) indicate.
(7) a. Treba da idem.
need-3sg to go-1sg.
I need to go.
b. Marija je trebalo da proita knjigu.
Mary aux-3sg. needed to read-3sg.perf. book-acc.sg
It was necessary for Mary to read a book.
c. *Treba da u da idem.
need-3sg to aux-1sg. to go-1sg.
d. *Treba da sam otiao.
treba-3sg. to aux-1sg. left
4. Negation
Serbian is a language that exhibits negative concord which means that, in addi-
tion to the negative particle that must be present for the interpretation of nega-
tion, negation is also expressed (but not interpreted), on multiple words in a
__________
2 The future tense in Serbian could be formulated in several ways. We can have Aux + in-
finitive or infinitive +Aux, or Aux +da+ present.
clause/sentence (GIANNAKIDOU 2000). The following data shows that the indica-
tive and subjunctive da-complements provide different environments for posi-
tion of negation.
4.1 Negation in da-complements of epistemics
Negation must precede the embedded verb in non-control da-complements of
assertives as in (8a), and cannot precede the matrix assertive verb as in (8b).
(8) a. Mislim da me ne vidi.
think-1sg that me-acc not see-3sg.impf.
I think that he/she doesnt see me.
b. (*Ne) mislim da me vidi.
neg think-1sg.impf. that me see-3sg.impf.
4.2 Negation in da-complements of volitionals
Unlike in the da-complements of assertives, the negative particle ne cannot pre-
cede the embedded verb in non-control da-complements of volitionals as in (9a),
and it can only precede the matrix volitional verb as in (9b).
(9) a. eli da Jovan (*ne) pie zadau.
want-2sg that Jovan not write-2sg.impf homework-acc.sg
b. Ne eli da Jovan pie zadau.
not want-2sg that Jovan write-2sg.impf homework-acc.sg
You do not want that Jovan writes the homework.
4.3 Negation in da-complements of modals
While in (9a) negation cannot precede the embedded verb in the da-
complements of volitionals, this is possible in (10a) when the matrix verb is
moi (can). Although the negative particle ne can also precede the matrix modal
verb, it is important to notice the difference in interpretation between (10a) and
(10b). However, not all modals behave in the same way in Serbian; for example,
impersonal treba (it is necessary) and morati (must) allow only the order as in
(10b).
(10) a. Mogu da ne piem zadau.
can-1sg that not write-1sg.impf. homework-acc.sg.
It is possible that I do not write the homework.
b. Ne mogu da piem zadau.
not can-1sg to write-1sg.impf. homework-acc.sg.
I cannot write the homework.
The infinitive can replace da + present in the subjunctive control complements,
but the replacement of da + present in constructions where the embedded sub-
ject is not controlled by the matrix subject is impossible. MOSKOVLJEVI (1936)
suggests that there is a difference between da-complements and infinitives that

follow modal verbs, and that difference is even more obvious when a modal
verb is negated. Moskovljevi states that if in (11a) the da-complement is re-
placed with the infinitive, the meaning is changed and (11a) has the future inter-
pretation while with the da-complement, as in (11b), the modal interpretation of
not having desire or not wanting to come is more emphasized. As
Moskovljevi observes, some speakers would not clearly see the difference be-
tween (11a) and (11b) as a future vs. modal interpretation, but the same speakers
would agree that the difference between the future interpretation (with the infini-
tive) and the modal interpretation (with the da-complement) is more obvious
with (11c) and (11d) in the presence of negation.
(11) a. On moe doi.
he can-3sg. come-inf.
b. On moe da doe.
he can-3sg to come-3sg.
He can come.
c. On ne moe doi.
he not can-3sg. come-inf.
He cannot come.
d. On ne moe da doe.
he not can-3sg to come-3sg.
5. Licensing of Negative Polarity Items
As noted by PROGOVAC (1994), there are two groups of negative polarity items:
NI-NPIs (those which begin with the negative prefix ni such as niko (nobody)
and nita (nothing)) and I-NPIs (those which begin with the prefix i such as iko
(anybody) and ita (anything)). She notes that NPI must be licensed by clause-
mate negation if they are objects, subjects or adjuncts, and unlike English, even
the subject NPIs can be licensed by clausemate negation (e.g.: Niko me ne poz-
naje/ Nobody knows me). On the other hand, I-NPIs cannot be licensed by clau-
semate negation and are licensed by matrix (superordinate) negation.
GIANNAKIDOU (1998) proposes that I-NPIs are licensed by non-verdicality and
anti-licensed by negation.
5.1 NI-NPIs in da-complemets of assertive
Licensing of NI-NPIs is possible within the da-complements of assertive only if
negation is present within the complement as in (12a-b); otherwise, the construc-
tions are illicit as in (12c-e) regardless of the subject/object position of NI-NPIs.
(12) a. Tvrdi da ne vidi nikoga.
claim-3sg that not see-3sg nobody-acc.
She/he claims that she/he does not see anybody.
b. Tvrdi da niko ne vidi Mariju.

claim-3sg that nobody-nom not see-3sg Maria
She claims that nobody sees Maria.
c. *Ne tvrdi da niko vidi Mariju.
not claim-3sg that nobody-nom. see-3sg. Maria-acc
d. *Ne tvrdi da Marija vidi nikoga.
not claim-3sg that Maria-nom see-3.sg nobody-acc.
e. *Nikoga ne tvrdi da vidi.
Nobody-acc. not claim-3sg to see-3sg.
5.2 NI-NPIs in da-complements of volitionals
If NI-NPIs can only be licensed by clausemate negation, then it is clear why
(13c) is acceptable, but the question is why the licensing of NI-NPI is possible
by superordinate negation in (13a). Another problem or asymmetry is observed
with (13a) and (13b) which show that NI-NPIs exhibit different licensing prop-
erties in the da-complements of volitionals if in a subject or object position; al-
though this should not be accepted as noted by PROGOVAC (1994), who also ad-
dresses this problem of asymmetry. Moreover, in case when the embedded sub-
ject is not controlled by the matrix subject, clausemate licensing of NI-NPIs is
possible after the volitional verb as in (13d). Another important behavior of NI-
NPIs in the da-complements of volitional verbs is captured in (13e) where niko-
ga is fronted and precedes negation while this is impossible with the da-
complements of assertive or epistemic verbs.
(13) a. Ne elim da vidim nikoga.

not want-1sg that see-1sg nobody-acc.
I do not want to see anybody.
b. Ne elim da?*niko vidi Jovanu.
not want-1sg that nobody-nom. see-3sg Jovana
I do not want anybody to see Jovana.
c. elim da ne vidim nikoga.
want-1sg that not see-1sg nobody-acc.
d. elim da niko ne vidi Jovanu.
want-1sg that nobody not see-3sg Jovana-acc.
e. Nikoga ne elim da vidim.
nobody-acc. not want-1sg. to see-1sg.
5.3 I-NPIs in da-complements of assertives

As with licensing of NI-NPIs, the licensing of I-NPIs is uniform in the da-
complements of assertive verbs. I-NPIs must be licensed by superordinate nega-
tion regardless if in the subject or object position as in(14a) and (14b), cannot be
licensed if negation is not present as in (14c), and are not licensed by clausemate
negation as in (14d).
(14) a. Ne tvrdi da vidi ikoga.
not claim-3sg that see-3sg. anybody-acc.
She/he does not claim to see anybody.
b. Ne tvrdi da iko vidi Mariju.
not claim-3sg that anybody-nom see-3sg. Mary-acc.
She/he does not claim that anybody sees Mary.
c. *Tvrdi da vidi ikoga.
claim-3sg that see-3sg. anybody-acc.
d. *Tvrdi da ne vidi ikoga.
claim-3sg that not see-3sg. anybody-acc.
5.4 I-NPIs in da-complements of volitionals
I-NPIs in the da-complements of volitional verbs are licensed by superordinate
negation as in (15a) and are also illicit like the da-complements of assertives
without presence of negation as in (15c), or with the presence of clausemate ne-
gation as in (15d). However, I-NPIs do not exhibit the subject-object asymmetry
with da-complements of volitionals as NI-NPIs do, and both (15a) and (15b) are
licit.
(15) a. Ne elim da vidim ikoga.
not want-1sg to see-1sg anybody-acc.
b. Ne elim da me iko vidi.
not want-1sg that me anybody-nom. see-3sg
I do not want anybody to see me.
c. *elim da me iko vidi.
want-1sg to me anybody-nom. see-3sg
d. *elim da ne vidim ikoga.
want-1sg to not see-1sg anybody-acc.
6. Clitics
6.1 Different approaches to the position of clitics in Serbian
Clitic placement in Serbian is sensitive to the syntactic/semantic consideration
as noted by PROGOVAC (2005). RIVERO (1994) states that Slavic clitics trigger
Long Head Movement since pronoun or auxiliary clitics cannot be clause initial;
therefore, they require support that is provided by a verb that moves to a high-
er position preceding and supporting the clitic. RIVERO & TERZI (1995) state that
C is the only licensing head in W-languages second position clitics. AVAR &
WILDER (1994) propose that clitics in Croatian are syntactically enclitics, occu-
pying a canonical position right-adjoined to C, and not syntactically proclitics,
left adjoining to some head in IP. They also add that the cliticization is clause-
bound and cannot cross a complementizer. BOKOVI (2004, 2008) states that
clitics occur in the second position of their intonational phrase (I-phrase) which
does not necessarily correspond to C, and that clitic placement is phonological
in nature. He also observes that Aux and object clitics do not occupy the same
position. Within the pronominal clitics, there is also a difference, and even the
pronominal clitics do not cluster in the same position as BOKOVI notes. He
adds that only elements that can undergo syntactic movement can precede and
host clitics and that clitics can be projected low in the structure which is much
lower than CP.
6.2 Clitics in da-complements of factives
In the da-complements of factives, clitics must be positioned within the da-
complement, and they usually follow da. Clitics cannot climb out of the embed-
ded construction as (16c) is illicit.
(16) . Znam da gai voli ti.
know-1sg that him-acc. love-2sg.
I know that you love him.
b. Znam da si mu ga dala.
know-1sg that aux-2sg him-dat. it-acc. gave
I know that you gave it to him.
c. *Znam gai da voli ti.
know-1sg him-acc that love-2sg
6.3 Clitics in da-complements of volitionals
In (17a) and (17c) the pronominal clitics occur in the embedded phrases in the
second position following da. In the presence of AUX je, the pronominal clitic
ga still remains in the embedded da-construction supported by da, and it does
not raise to the second position of the matrix clause as in (17c). However, in
(17b) the clitic form ga climbs to the second place of the matrix clause sup-
ported by the volitional eleti (to want).
(17) a. Tanja eli da gai vidi ti.

Tanja want-3sg to him-acc see-3sg.impf.
Tanja wants to see him.
b. ?Tanja gai eli da vidi ti.
Tanya him-acc want-3sg to see-3sg.impf.
c. Tanja je htela da ga vidi.
Tanja aux-3sg. wanted to him-acc. see-3sg.
Tanja wanted to see him.
MARKOVI (1955) pays attention to instances similar to those in (17b) and ana-
lyzes examples under (18). He points out that in such cases where instead of da
+present there is the infinitive serving as a complement, clitics usually follow
the first word in a sentence and precede the infinitive. On the other hand, when
the infinitival constructions are replaced by da +present, clitics should follow
da, but as Markovi points out, clitics may remain in the second position preced-
ing da +present. Markovi states that this phenomenon might be a result of the
mechanical replacement of the infinitive by da + present leaving the word or-
der the same as with the infinitival complements.
(18)3 a. Niko ga nije mogao da zaustavi.
nobody-nom him-acc not could to stop-3sg
Nobody could stop him.
b. Nije se mogao das jeti nijedne od tih

stranih
not self could to remember-3sg not one of those-gen horri-
ble noi.
nights
He could not remember any of those horrible nights.
7. Conclusion
Based on the tests of aspect, tense, negation, licensing of negative polarity items
and clitics, it is possible to conclude that da-complements do not exhibit the
same behavior after the indicative (assertive, epistemics, and factives) and after
the subjunctive (volitionals, modals, aspectuals) verbs . While any tense is al-
lowed in the da-complements of the indicative verbs, with the restriction of only
the imperfective aspect for the present tense, after the subjunctive verbs the only
tense allowed is the present of either the perfective or imperfective aspect. The
negative particle ne in non-control da-constructions of the subjunctive verbs is
placed before the matrix verb; however, this is impossible with the da-
__________
3 Examples in (18) are from MARKOVI (1955).
complements of the indicative verbs where ne must precede the embedded verb
in non-control constructions. While NI-NPIs are licensed only by clausemate
negation in da-complements of the indicative verbs, after the subjunctive verbs
(volitionals) NI-NPIs can be licensed either by clausemate or superordinate ne-
gation. As clitic climbing is widely unacceptable out of CP in Serbian and out of
da-complements that follow the indicative verbs, this seems to be somewhat ac-
ceptable out of da-complements of volitionals.
Finally, to better understand the situation of da in Serbian, it is very useful
to compare the similar sentence constructions to other Balkan languages, in this
case Greek. While (19a) is a Serbian example, (19b) is a Greek example. As ex-
amples clearly indicate, Greek employs different particles/complementizers in
the instances, where in Serbian we find the same and one da. But, is this in-
deed the same and one da? The further research and clear analysis is neces-
sary in order to properly address this question.
(19) a. Marija misli da sam rekla da u da napiem
Mary think-3sg that aux-1sg said that aux-1sg to write-
1sg.perf.
knjigu da bih postala slavna.
book-acc to aux-1sg became famous-fem.
b. I Maria pistevi oti ipa oti tha grapso en vivlio ja
na
The Mari believes that said that fut write-perf. one book for
na jino dniasimi
become famous
Mary thinks that I said that I will write a book to become famous.
References:
BOKOVI, . (2004) Clitic placement in South Slavic. Journal of Slavic Linguistics. 12:37-
90.
BOKOVI, . (2008) On the clausal and NP structure of Serbo-Croatian. Proceedings of For-
mal Approaches to Slavic Linguistics 15.
BROWNE, W. (1986) Relative Clause in Serbo-Croatian in Comparison with English. Institute
of Linguistics. University of Zagreb.
GIANNAKIDOU, A. (1998) Polarity Sensitivity as (Non)veridical Dependency. Amsterdam and
Philadelphia. John Benjamins.
GIANNAKIDOU, A. (2000) NegativeConcord? Natural Language and Linguistic Theory. 18:
457-523
GIANNAKIDOU, A. (2009) On the temporal properties of mood: the subjunctive revisited. Lin-
gua. 119: 18831908
GOLB, Z. (1964) The problem of verbal moods in Slavic languages. International Journal of
Slavic Linguistics and Poetics 8:1-36
GRIVII, T. (2004) Functional domain of the Croatian complementizer da. Colorado Re-
search in Linguistics. 17 (1).
ISAC, D. & E. JAKAB (2004) Mood and force features in the languages of Balkans. In: TOMI,
O. M. (ed.) Balkan Syntax and Semantics 1-59. Amsterdam and Philadelphia. John
Benjamins.
IVI, M. (1970) O upotrebi glagolskih vremena u zavisnoj recenici: Prezent u recenici sa vez-
nikom DA. Zbornik za filologiju i lingvistiku 13: 43-53.
IVI, M. (1972) Problematika srpskohrvatskog infinitiva. Zbornik za filologiju i lingvistiku
15/2:115-138.
IVI, M. (1973) Neka pitanja reda reci u srpskohrvatskoj zavisnoj recenici s veznikom da.
Zbornik za filologiju i lingvistiku 16/1:187-95.
JAKAB, E. (1999) Is PRO really Necessary? A minimalist approach to infinitival and subjunc-
tive-(like) constructions in Serbo-Croatian and Hungaria. Formal Approaches to Slav-
ic Linguistics. 7:206-224.
MARKOVI, S. (1955) Poloaj zamjenike enklitike u vezi sa naporednom upotrebom infiniti-
va i prezenta sa svezicom da. Na Jezik 6/1-2: 33-40.
PHILIPPAKI-WARBURTON, I. (1994) The subjunctive mood and the syntactic status of the par-
ticle na in Modern Greek. Folia Linguistica XXVIII (34): 297326.
PROGOVAC, L. (1993a) Locality and subjunctive-like complements in Serbo-Croatian. Journal
of Slavic Linguistics 1:116-44.
PROGOVAC, L. (1993b) Negative polarity: entailment and binding. Linguistics and Philosophy
16 (2): 149-80.
PROGOVAC, L. (1993c) Subjunctive: The (mis)behavior of anaphora and negative polarity.
The Linguistic Review 10:37-59.
PROGOVAC, L. (1994) Negative and Positive Polarity: A binding approach. Cambridge: Uni-
versity Press.
PROGOVAC, L. (2005) A Syntax of Serbian-Clausal Architecture. Bloomington: Slavica.
RADII, M. (2006) Subject Raising and Obligatory Control. Proceedings of the 2006 annual
conference of the Canadian Linguistic Association.
RIVERO, M.-L. (1994a) Clausal Structure and V-movement in the Language of Balkans. Natu-
ral Language and Linguistic Theory 12. 63-120.
RIVERO, M.-L. (1994b) Negation, Imperatives and Wackernagel Effects. Rivista di
Linguistica. 6.1. 39-66.
RIVERO, M.-L. & A. TERZI (1995) Imperatives, V-movement and Logical Mood. Journal of
Linguistics 31. 310-332.
RIZZI, L. (1990) Relativized Minimality. Cambridge, Mass.:MIT Press.
RIZZI, L. (1997) The Fine Structure of the Left Periphery. In: HAEGEMAN, L. (ed.) Elements of
Grammar 281-337. Dordrecht. Kluwer.
ROUSSOU, A. (2000) On the Left periphery; modal particles and complementizer. Journal of
Greek Linguistics 1: 63-93.
ROUSSOU, A. (2001) Control and raising in and out of subjunctive complements. In: RIVERO,
M. L. & A. RALLI (eds.) Comparative Syntax of Balkan Languages 74-104. New York.
Oxford University Press.
STEVANOVI, M. (1935) Infinitiv i svezica da sa prezentom. Na Jezik 3/7: 282-88.
SOBOLEV, A. N. (2004) Spread of the Balkan sprachbund morpho-syntactic properties. In:
TOMI, O. M. (ed.) Balkan Syntax and Semantics 1-59. Amsterdam and Philadelphia.
John Benjamins.
STJEPANOVI, S. (2004) Clitic climbing and restructuring with finite clause and infinitive
complements. Journalof Slavic Linguistics 12(1):173-212.
SZCZEGIELNIAK, A. (2005) Clitic positions within the Left Periphery: evidence for a phono-
logical buffer. In: HEGGIE, L. & P. ORDOEZ (eds.) Clitic and Affix Combinations:
Theoretical perspectives 283299. Amsterdam and Philadelphia. John Benjamins.
TOMI-MIESKA, O. (2003) The Syntax of the Balkan Slavic Future tenses. Lingua 114: 517-
549.
TOMI-MIESKA, O. (2004) The Balkan Sprachbund properties: An introduction. In: TOMI,
O. M. (ed.) Balkan Syntax and Semantics 1-59. Amsterdam and Philadelphia. John
Benjamins.
TOMI-MIESKA, O. (2004) Genesis of the Balkan Slavic future tenses. Formal Approaches to
Slavic Linguistics. 12:395-417.
TOMI-MIESKA, O. (2005) The Syntax of Negated Imperatives in Balkan Slavic. Formal Ap-
proaches to Slavic Linguistics. 13: 365-375.
VRZI, Z. (1994) Categorial status of the Serbo-Croatian modal da. In: TOMAN, J. (ed.) An-
nual Workshop on Formal Approaches to Slavic Linguistics: The college Park Meet-
ing 1994 291-312 Ann Arbor: Michicagan Slavic Publications.
WERKMAN, V. (2007) Subjunctive complements of modal verbs in Bulgarian and Macedo-
nian. Formal Approaches to Slavic Linguistics. 15:458-476.
WILDER C. & D. AVAR (1994a) Long head movement? Verb Movement and Cliticzation in
Croatian. Lingua 93: 1-58
Natasha Todorovich, University of Illinois at Chicago, Department of Slavic and

Baltic Languages and Literature, 601 South Morgan Street, 1722 University
Hall(m/c 306), Chicago, Illinois 60607-7117, USA, ntodor1@uic.edu
On Slavic and Finno-Ugric vs. Standard Average
European
Hannu Tommola
1. Introduction
This paper is devoted to testing the concept of Standard Average European
(SAE), coined by BENJAMIN WHORF (1941), and the main features identified as
typical of SAE by MARTIN HASPELMATH (2001) and others. All the main
branches of the Indo-European languages spoken in Europe are by Haspelmath
considered SAE, i.e. the Germanic, Romance, Baltic and Slavic languages, as
well as Albanian and Greek. Among the Germanic and Romance languages
some are obviously more SAE than others; at the core are the Western
Germanic and Romance languages of the Charlemagnian Europe. The Baltic
and Slavic branches as a whole do not belong to the core of SAE. At least two of
the twelve main features typical of Europe are more or less restricted to the
Western European languages (notably German and French). Such are the
sentence negation expressed by a negative indefinite and the periphrastic perfect
formed with the verb to have with the tendency of the perfect to be
reinterpreted as a general or a perfective past.
The set of features considered as shared by the European languages is not
necessarily an optimal choice in contrastive studies of the European languages.
While comparative constructions composed with a particle probably can be
regarded as restricted to Europe, most of the rest of the features acknowledged
as SAE are distributed more evenly, over large areas outside Europe, as well.
For example, languages with definite articles are common also in central Africa,
in Mesoamerica, in New Guinea and the Pacific (DRYER 2008a: 1213).
Languages in which there is some kind of indefinite article are widely
scattered, and even languages with an indefinite article distinct from the
numeral one are not restricted to Europe, either (DRYER 2008b: 12).
In what follows I shall present data from Slavic, Finno-Ugric and other (mainly
Germanic) European languages. The criteria used to define the features typical
of the European languages are to some extent contradictory, and the application
of the set of features to identify the languages pertaining to SAE is too
straightforward. Namely, there is considerable variation in the use of alternative
means to express these meanings, for example, in different Slavic languages. We
should not forget the remark made in COMRIE & KUTEVA (2008: 13) on the
strategy of feature assignment in WALS:
Note that the assignment of a particular feature value to a particular language does
not mean that this feature value is the only one that has been attested in that
366 Hannu Tommola
particular language. It only means that this particular feature value is considered to
be the most frequent, or the canonical one in non-marked contexts.
2. Definite articles
For my purposes, I will identify the marking of definite noun phrases only, and
exclude a discussion of the availability of indefinite articles and their behaviour
in different languages as too complicated in this connection. As to the definite
article, there are at least three types of marking: a prepositive article,
exemplified here by Hungarian (see ex. (1)), a postpositive (or enclitic) article as
in Bulgarian (2), and a mixed type of the Scandinavian languages (see the
example (3b) in the next section for Norwegian). However, in counting features
shared by the languages investigated, I confine myself to the parameters of the
presence or the absence of a definite article, and its location in preposition vs.
postposition (see the tables in the Appendix).
(1) Hungarian
A trja-i hbor fegyveres konfliktus volt
DEF Trojan war armed conflict be.PRET
az kor-ban az akhj-ok s a
DEF ancient-INESS DEF Achaean-PL and DEF
kiszsia-i Trja kztt.
Asia_Minor-ADJ Troy Postp.between
The Troyan War was an armed conflict in the Ancient times between the
Achaeans and the Asia Minor Troy.
(2) Bulgarian
-- , -- ,
her-N-DEF escape Prep Paris according legend-PL-DEF
-- -
.
become-PRET-N cause Prep Trojan-DEF war
According to the legends, her [Helens] escape with Paris brought about the
Trojan War.
Bulgarian and Macedonian are clearly more SEA than the rest of the Slavic
languages because of the availability of the articles. In the same way, Hungarian
differs from the Finnic languages that, like the other Uralic languages, lack
articles.
3. Relativization strategies
The relativizing strategy based on an inflected relative pronoun is usual in
almost all European languages (the exceptions being Basque and the Celtic
languages). This feature is almost exclusively restricted to Europe,1 present in
all the languages investigated here (English, German, Swedish, Russian, Polish,
Czech, Slovakian, Serbian/Croatian, Bulgarian, Macedonian, Hungarian, Finnish
__________
1 In WALS the only exception here is Acoma, an American Indian language.
On Slavic and Finno-Ugric vs. Standard Average European 367
and Estonian). In English and in Scandinavian languages there is another, a kind

of gap strategy, which may be used (3), unless the relative pronoun is the
subject of the relative clause:
(3) Gap
a. Swedish
Hon var inte lngre den flicka-n jag
she be.PRET NEG longer DET girl-DEF I
trffa-de i sandlda-n.
meet-PRET Prep sandbox-DEF
She was no more the girl I met in the sandbox
b. Norwegian (bokml)
Den trojanske krig var, iflge legend-en, en
DEF Troyan war be.PRET according legend-DEF INDEF
krig de greske kongen-e fr-te mot
war DEF Greek king-PL conduct-PRET Prep.against
by-en Troja i Lilleasia etter at prins Paris av
town-DEF Troy Prep Asia_Minor after that prince Paris Prep
Troja bortfr-te Helena av Sparta.
Troy take-away-PRET Helena Prap Sparta
The Trojan War was, according to the legend, a war the Greek kings
conducted against the city of Troy in Asia Minor after Helen of Sparta was
taken away by prince Paris of Troy.
Postnominal relative clauses are common in all Slavic, and also in Finnic
languages. However, the pattern typical of Uralic languages is to embed the
relative clause in the nominal phrase, and this strategy is still quite usual in
Finnic (4a); moreover, similar participle clauses are not rare in Slavic, either
(4b).
(4) Premodifying participle construction
a. Estonian
Iga-s olukorra-s ausalt kitu-v
every-INESS situation-INESS honestly behave-PTCP
mees on eeskuju-ks noorte-le.
man is exemplar-TRNSL young.PL-ALL
A man behaving / who behaves under any circumstances honestly is an
exemplar for the young
b. Russian
%9&-/-- -,
show-PTCP.PRET-PL-Refl you-DAT threat-PL it
<9&-/-- -.
only show-PTCP.PRET-PL-Refl you-DAT threat-PL
Those threats that seemed to you to be threats are just something that
seemed to you to be threats
368 Hannu Tommola
Even though postnominal relative clauses are common in Finnic and in

Hungarian, the original pattern, allowing prepositive participle constructions of
considerable length, are still quite usual in Finnish, less so in Estonian and in
Hungarian.
The examples from Turkish quoted after COMRIE (1998: 82) in (5a) and (6a)
show different constructions that are used depending on which position is
relativised. These examples have corresponding counterparts in Finnish (5b, 6b),
Estonian (5c, 6c) and Russian (5d, 6d). The relativized subject position is
expressed with an active participle (5), and the object position requires a passive
participle (6).
(5) Subject position relativized (the student who bought the book)
a. Turkish kitab- al-an renci
book-ACC buy-PTCP student
b. Finnish kirja-n ostan-ut opiskelija
book-GEN buy-PTCP.PRET student
c. Estonian raamatu ostn-ud tudeng
book.GEN buy-PTCP.PRET student
d. Russian , -- -
student buy-PTCP.PRET-M book-ACC
(6) Object position relativized (the book which the student bought)
a. Turkish renci-nin al-d- kitap
student-GEN buy-NMLZ-3SG book
b. Finnish opiskelija-n osta-ma kirja
student-GEN buy-NMLZ2 book
c. Estonian tudeng-i ost-et-ud raamat
student-GEN buy-PASS-PTCP.PRET book
d. Russian -- -
buy-PTCP.PRET-F student-INSTR book
As we can see in the alternative translation of (4a) above, a participial modifier
is possible to use in the place of a relative clause in English (and in some other
Germanic languages), too. This strategy (which is also in a way a gap) is made
more acceptable, if the participle construction follows the head noun. An
example of such a postmodification we see also in (5d) for Russian, and the
construction in (6d) could be postmodifying, too. Some formal varieties of
Russian allow quite extensive participle constructions, which would be too
__________
2 The form glossed here as a nominalizer (NMLZ) in Finnish grammar tradition called
agent participle always requires an explicit agent marking, either as a noun in the
genitive case (as in (6b)), or as a possessive suffix.
heavy in preposition, e.g. in Finnish, where postmodification is not possible.

Among the Slavic languages there seems to be considerable variation as to what
degree they make use of the participial relativizing construction, cf. the
strategies employed in Russian (with an embedded participle) and in Slovak
(with a relative pronoun) in (7):
(7) SVETLK (1972: 250)
a. Russian -- -- --
plant-PTCL-PL 1SG.INSTR birch-DIM-PL become-PRET-PL
- -.
big-INSTR.PL tree-INSTR.PL
b. Slovak Z brez-i-iek, ktor- som zasadi-l,
Prep birch-DIM-PL REL 1SG plant-PRET
vyrst-l-i ve
k- strom-y.
grow-PRET-PL big-PL tree-PL
The birches that I planted have grown to big trees.
4. The perfect
According to the rough classification of perfect markers in WALS, the
possessive perfect is restricted to the Western European languages (Germanic
and Romance), plus Modern Greek. However, also in Albanian, not included in
the sample in the WALS perfect map, there is a perfect formed with the verb to
have (8a), while a periphrastic perfect with the auxiliary to be is used when
the meaning is reflexive (8b).
(8) Albanian (JNTREJ 1982: 90)
a. Ka-m lar faqt, duart.
have-1SG wash.PTCP face.ACC hand.ACC.PL
Ive washed my face and hands.
b. Ja-m lar
be-1SG wash.PTCP
Ive washed myself.
The Finnic languages have no possessive perfect, while in Hungarian there is no
perfect at all.3 However, both in Finnish and Estonian the perfect gram (even
though not a have perfect, but an other perfect as identified in WALS) is in its
functions quite similar to the prototypical perfect found, for instance, in English
and Swedish. At the same time, in some West Germanic and Romance
languages the possessive perfect has partly lost its specific meaning and appears
as a general (German and Dutch) or perfective (French) past.
__________
3 It could be added that this is what you expect to get, if the language does not encode
predicative possession with transitive have verb, but locationally, as do the Finno-Ugric
languages.
370 Hannu Tommola
In all Slavic languages there has been a periphrastic perfect (even if not a
habere but an essere perfect, exactly as in Finnic), which in East Slavic has
totally lost the auxiliary, while rests of the auxiliary are in other groups to
various degrees preserved, and along with the decline of the simple past tenses
(aorist and imperfect) the perfect has become a general past. However,
Bulgarian and Macedonian have retained the old tense system with the aorist,
the imperfect and a functional perfect. As a new development, in some West
Slavic languages new resultative constructions (based on expressions of after,
already etc.) to some extent compensate the lost perfect meaning in discourse
(see TOMMOLA 2000). New possessive resultatives also arise, like the Czech
mm construction (HAVRNEK & JEDLIKA 1981, TOMMOLA 2000, GIGER 2003:
126159) and the Macedonian imam perfect (GRAVES 2000) which tend to
replace the functional gap left by the former perfect (for a discussion of Russian
resultatives and possessive perfects see in GIGER 2003: 492493, and for other
Slavic languages GIGER 2003: 478494; for the North Slavic and Baltic
languages: WIEMER & GIGER 2005).
5. Experiencers in nominative case
The preponderance of generalizing predicates to encode experiencers is seen as
typical of SAE, i.e. in these languages experiencers tend to appear as surface
subjects in nominative case. The English way to say I love/like music seems to
be possible all over Europe. It is questionable whether the experiencer in
nominative is prevalent in the Slavic languages; at least an alternative syntactic
structure with a dative experiencer and a 3rd person predicate verb is very
frequent, see (9b) and (10b).
(9) Czech: I like this song.
a. M-m rd tuto pse.
have-1SG glad this.ACC song
b. Tahle pse se mi lb
this song Refl 1SG.DAT please
(10) Bulgarian: I like this song.
a. &>- & <.
1SG love-1SG this song
b. ' ~ & <.
1SG.DAT please this song
This type of an alternative pattern is common not only in Slavic, but also in a
number of other European languages, in German (11c) and in Baltic languages
(12b, c), and in Estonian (13b) as well.
(11) German
a. Ich lieb-e dies-es Lied
I love-1SG this-N.ACC song
I love this song.
b. Mag-st du Erdbeere-n und Eis?
like-2SG you.SG strawberry-ACC.PL and ice-cream
Do you like strawberries and ice-cream?
c. Mir gefll-t dies-es Lied.
I.DAT please-3SG this-N.ACC song
I like this song.
(12) Latvian
a. mt dzimten-i
love.INF fatherland-ACC
to love the fatherland
b. Vi-am filma patik-a.
3SG-DAT film please-PST
He liked the picture.
c. Vi-am gar-o stipra kafija
3SG-DAT please-3SG strong coffee
He likes strong coffee.
(13) Estonian: armastama (transitive) love vs. meeldima (with indirect object)
like
a. Ma armasta-n se-da laul-u
1SG love-1SG this-PRTV song-PRTV
I love this song.
b. Mu-lle meeldi-b see laul
1SG.ALL please-3SG this song
I like this song.
In addition, some deviance from the transitive pattern is represented by Swedish
(14b) and Finnish (15b). These languages employ verbs that govern an indirect
complement in this function.
(14) Swedish: lska (transitive) love vs. tycka (with a prepositional object)
like
a. Jag lsk-ar detta land
1SG love-1SG this country
I love this country.
b. Jag tyck-er om glass
1SG like-1SG Prep ice-cream
I like ice-cream.
372 Hannu Tommola
(15) Finnish: rakastaa (transitive) love vs. pit / tykt (with the elative case)
like
a. Rakasta-n t-t maa-ta
love-1SG this-PRTV country-PRTV
I love this country.
b. Pid-n / Tykk-n jtel-st
like-1SG ice-cream-ELAT
I like ice-cream.
6. Passive formed with a copula verb and the past participle
The copula + past participle construction is common in Germanic, Romance
and Slavic languages, but the reflexive passive is also frequent in all these
language groups, notably in Romance and Slavic languages spoken in southern
Europe, and in the Scandinavian languages (for the latter, see LAANEMETS 2004,
2009).
(16) Serbian
- - -
therefore REFL ask-3SG all member-PL commission-GEN CONJ
- - - -.
all PronREFL-PL material-PL hand_over-3PL chairperson-DAT
That is why all the members of the Commission are asked to hand all
materials over to the chair person.
In North Germanic, the (actional) passive is most often, notably in Swedish,
realized with the suffix -s which has its origin in the reflexive marker. In Slavic,
the reflexive passive is a systemic device with imperfective verbs, but in West
and South Slavic it is used also with perfective verbs (20b).
(17) North Germanic: Participial passive
a. Swedish
Han var knd verallt i Europa.
3SG.M Cop.PST known everywhere Prep Europe
He was known everywhere in Europe.
b. Danish (LAANEMETS 2009: 147)
Min-e billed-er bliv-er ofte brug-t (af andr-e).
my-PL picture-PL become-PRS often use-PTCP.PRET Prep other-PL
My pictures are often used by others.
(18) North Germanic: Reflexive passive

a. Swedish
Seger-n jubla-de-s i hel-a den muslimsk-a
victory-Def celebrate-PST-PASS Prep whole-Def ArtDef muslim-Def
vrld-en
world-Def
The victory was celebrated in all the Muslim world.
b. Norwegian (LAANEMETS 2009: 146)
Facebook bruke-s (av mange mennesk-er) over hel-e verd-en.
Facebook use-REFL Prep many people-PL over all-DEF world-
DEF
Facebook is used by many people all over the world.
(19) South Slavic: Participial Passive
a. Croatian
Narukvica je lijepo izra-en-a od
bracelet be.3SG beautifully create-PTCP.PASS.PRET-F Prep
sam-oga majstor-a.
self-GEN master-GEN
The bracelet is beautifully made by the master himself.
b. Bulgarian
-- -- -
numerous-PL-DEF gather.PFV-PTCP.PRET-PL Refl citizen-PL
a - -
be.AOR.3PL evacuate-PL Prep police-DEF Prep
- -.
place-DEF Prep incident-DEF
Several citizens gathered at the place of the incident were evacuated by the
police.
(20) Reflexive passive
a. Croatian
Ovdje se ne pu-i.
here REFL NEG smoke-3SG
No smoking
b. Bulgarian
- - ( -)
lesson-Def Refl teach.PFV-3SG Prep teacherDef
The lesson is taught by the teacher.
SAAVEDRA (2007), who has compared passive constructions in the South Slavic
languages, shows that the participial passive may be more acceptable in
Bulgarian, and less so in Croatian. In (21) the Croatian sentence is in active,
while the corresponding Bulgarian is passive.
374 Hannu Tommola
(21) (SAAVEDRA 2007: 228)

a. Croatian
Izvedb-u pjes-ama i ocjene iri-ja
performance-ACC song-GEN.PL and assessment jury-GEN
burno je ispraa-l-a oduevlj-en-a .
stormily be.3SG accompany-PRET-F inspire-PTCP.PASS.PRET-F
publika
public
b. Bulgarian
- --
performance-DEF Prep song-PL-DEF and
-- -
assessment-PL-DEF Prep jury-DEF be.AOR.PL stormily
-- ---
meet-PTCP.PASS.PRET-PL Prep inspire-PTCP.PASS.PRET-F-DEF
.
public
The performance of the songs and the assessments of the jury were stormily
celebrated by the enthusiastic audience.
In Finno-Ugric languages neither of these types of passive are common. They
may occur sporadically in Estonian but are almost unknown in Finnish and in
Hungarian. In Finnic, an agent NP is possible exclusively in state passive
constructions, exemplified in (3a) and (6b) above (and see fn. 2). Thus the
Finnic languages lack actional passive constructions, whereas the impersonal
passive, which more adequately should be called indefinite-personal, is denoted
by a finite (active) verb form, traditionally called passive.4 In Karelian, under the
influence of the corresponding usage in Russian, it has been replaced by the 3rd
person plural verb form. The same applies to Hungarian, where the old reflexive
passive is almost extinct today.5
(22) Indefinite-personal
a. Finnish Hn-et tunne-taan kaikkialla.
3SG-ACC know-AMB everywhere
He is known everywhere.
b. Russian - .
3SG.ACC know-3PL everywhere
He is known everywhere.
__________
4 I have termed this form ambipersonal (TOMMOLA 1997, 1998), and gloss it here as
AMB. This form is identified as passive in WALS, although it is questionable, whether
the criterium of a nominative subject position of the patient (see SIEWIERSKA 2008) is
fulfilled.
5 In Karelian the ambipersonal and the 3rd person plural forms have mutually replaced each
others functions, as the ambiperson form has taken over the role of the 3rd person plural.
c. Hungarian Az-t beszl-ik, hogy

that-ACC talk-3PL that
It is told that/They say that
7. Anticausative strategies
In English, the intransitive verbs are often derived semantically from the
transitive ones, e.g., in the pair The ice is breaking vs. They are breaking the ice,
or in German Der Teller zerbrach The plate broke vs. Er zerbrach den Stock
He broke the stick. This does not hold for most Germanic languages, at least
not as the most common strategy. For example, in German schmelzen and tauen
have both meanings; but the usual anticausative strategy is to add the reflexive
pronoun (sich) to the transitive verb.6 In some cases, the tendency to
differentiate both meanings is still left in the conjugation, cf. Swd smlta melt
with an optional strong inflection in the inchoative meaning only (in the preterite
smlte vs. smalt).
In Romance and in many Slavic languages a pseudo-reflexive construction
is the normal strategy (cf. Croatian Staklo se razbilo The glass shattered). In
East Slavic this appears as a postfix, added to the end of the (inflected) verb
(Russian (<) vs. (<)). The same applies also to the Baltic
and Scandinavian languages.
Also, for example, in Russian the causative is frequently derived with the
help of an affix from the inchoative verb, cf. the ichoative ipf. simplex
which has a prefixed pf. counterpart with a secondary ipf. partner
; whereas only the derived pair / has a
causative meaning.
The typologically commonest type of anticausative formation is an
inchoative affix changing the stem into intransitive. This is typical of the Finno-
Ugric and Turkic languages, but it is not rare in Slavic either. In Polish, for
example, we have the transitive topi melt, and the corresponding intransitive
verb is either the reflexive topi si or topnie:
(23) Inchoative affix
a. Finnish
Vaattee-t-kin kuiv-u-vat nopeasti, kun tuuli ja
clothe-PL-also dry-INCH-3PL quickly when wind and
lmp kuiva-tta-vat ilma-a.
warmth dry-CAUS-3PL air-PRTV
Clothes also dry quickly when the wind and warmth make the air dry.
__________
6 Occasionally also periphrastic expressions are used, consider verlieren vs. verloren
gehen, and lose vs. get lost in English.
376 Hannu Tommola
b. Polish
Wieczna zmarzlina na p+nocy Kanad-y zaczyn-a
eternal ice Prep north Canada-GEN begin-3SG
top-n-ie
melt-INCH-INF
Eternal ice in North Canada begins to melt.
In Finnish, the derivational mechanism is quite complex and causatives are often
derived from inchoatives (e.g. sula-a melt, itr. > sula-tt-aa melt, tr.). There is
also an elaborated system of diathese suffixes for automative inchoatives -U-
(hukku-a be/get lost; drown) and reflexives -UtU- (huku-tta-utu-a drown
oneself, derived from the causative huku-tta-a make drown< huk-ata lose;
miss).
8. External possessors in dative
Dative external possessors seem to be common in all SAE languages, also in
Slavic ((25) and (26b,c)), but not really in Finno-Ugric (26d). However, in some
variations of Finnish a similar use of the ablative case is found (27); but this
kind of separative construction, either with the elative or ablative, is mainly met
with in negative contexts (27b). In archaic proverbs the dative case still occurs
even in Bulgarian, which has otherwise totally lost the nominal inflection (26b).
(24) German
Die Mutter wusch dem Kind die Haar-e.
DEF mother wash.PRET DEF.DAT child DEF hair-PL
The mother washed the child's hair.
(25) Czech
Prevence proti prase- chipc-e? Mt si ruce,
prevention against pig-DAT.F flu-DAT wash REFL.DAT hand.PL
nosit rouk-y ale hlavn ne-jst vepov
bear mask-PL but mainly NEG-eat pork
Protection against pig flu? Wash your hands, wear masks, and above all,
dont eat pork...
(26) The raven to raven will not tear out the eye.
a. Latin Corvus ocul-um corv-i non erue-t.
raven eye-ACC raven-DAT NEG dig_out-3SG
b. Bulgarian - -.
raven raven-DAT eye NEG pull-3SG
c. Russian Y -" 7& -9@-.
raven raven-DAT eye NEG out-peck-3SG
d. Finnish Ei korppi korpi-n silm- noki.
NEG raven raven-GEN eye-PRTV peck.STEM
(27) Finnish (Ablative)

a. Pese ensin lapse-lta hampaa-t.
wash.IMP first child-ABL tooth-PL
Brush first the childs teeth.
b. Hne-lt kuol-i viime vuon-na vaimo.
he-ABL die-PRET last year-ESS wife
His wife died last year.
9. Negation with indefinites
In some European languages a sentence can be negated with a negative
indefinite pronoun without a separate predicate negation, e.g. Nobody listened,
German Niemand/Keiner hat zugehrt, Swedish Ingen lyssnade etc. In German
and Dutch an additional predicate negation is not allowed at all, in some other
languages (e.g. Scandinavian languages, English) there is a mixed usage, and
still in others a double negation is the rule (all Slavic languages). In Finnish,
Estonian and Hungarian the predicate negation is always present, but actually
the strategy is similar to the mixed pattern, because there are, for example in
Finnish (29), no negative indefinites; the sentence negation is expressed with the
negative auxiliary verb.
(28) Double negation
a. Russian Ni-kto ne slua-l.
NEG-who NEG listen-PRET
Nobody listened.
b. Czech Nen to dn ert.
NEG.3SG that no joke
It is no joke.
c. Polish Nikt nic nie wie.
nobody nothing NEG know.3SG
Nobody knows anything.
(29) Predicate negation
a. Finnish Kukaan ei kuunnel-lut.
who.Indef NEG.3SG listen-PRET.PCTP.SG
Nobody listened.
b. Hungarian Nem tesz semmit.
NEG do.3SG anything
It doesnt matter.
10. Comparative constructions
With the comparative, a particle (or conjunction) of inequality (than) is
involved to mark the standard of comparison in almost all European languages,
378 Hannu Tommola
e.g. in Swedish n, in German als, and in Slavic: Bulgarian 99, Polish

ni, Russian , $, Serbian 7 (#), etc.
(30) Better late than never.
a. Bulgarian - .
COMPV-good late than never
b. Italian Meglio tardi che mai.
better late than never
c. Finnish Parempi myhn kuin ei milloinkaan.
better late than NEG ever
(31) Serbian
a. - -.
abstract NEG need PRTCL be.T longer Prep one-GEN page-GEN
The abstract need not be longer than one page.
b. - -
reception at vizier-GEN be.PTCP.PRET is some shorter
- - -.
and cold-COMPV than reception French-GEN consul-GEN
(Ivo Andri)
The reception of the vizier was somewhat shorter than the reception of the
French consul.
However, the locational strategy which is as a rule used in the Uralic
languages, is also found in Slavic, along with the analytic particle comparison
(e.g., Russian bole slona or bole, em slon bigger than an elephant). In
Uralic languages the standard of comparison is originally marked by a
separative case ending ((32) through (36)) or postposition denoting separation
(36a). We find this pattern still in Finnish and Hungarian, and more seldom in
Estonian.
(32) Finno-Ugric, Permic: Udmurt
a. :- -
hare fox-ABL smart-COMPV
The hare is smarter than the fox.
b. -
bear woolf-ABL strong
The bear is stronger than the wolf.
(33) Finno-Ugric, Volgaic: Erza Mordvin
a. -
apple cherry-ABL big
An apple is bigger than a cherry.
b. -
cherry apple-ABL small
A cherry is smaller than an apple.
(34) Finno-Ugric, Volgaic: Moksa Mordvin
a. -
tree-ABL high
higher than a tree
b. -
sugar-ABL sweet
sweeter than sugar
c. -
cherry apple-ABL small
smaller than an apple
(35) Finno-Ugric: Finnic
a. Vepsian Tmbei om rk-emb, mi eglai
today be.3SG hot-COMPV than yesterday
Today it is hotter than yesterday.
b. Estonian Ants on Mardi-st pik-em.
Ants is Mart-ELAT long-COMPV
Ants is taller than Mart.
(36) Finno-Ugric, Ugric
a. Ob-Ugric: Khanty
.
this tree Postp that tree high
This tree is higher than that (tree).
b. Hungarian
sem rossz-abb ms-ok-nl.
s/he NEG bad-COMPV other-PL-ADESS
S/he is not worse than the others.
(37) Fixed case comparative
a. Russian
- /
one time can PREF-run early.COMPV
#"7-, --.
other-GEN.PL but health already break-PASS.PART.PRET-N
You can once reach the finish line faster (lit. earlier) than the others, but
youve already ruined your health.
380 Hannu Tommola
b. Finnish
Oli-n selvsti nopea-mpi kaikk-i-a mu-i-ta
be.PRET-GEN clearly fast-COMPV all-PL-PRTV other-PL-PRTV
uusi-en talli-en kuljettaj-i-a.
new-GEN.PL team-GEN.PL driver-PL-PRTV
I was clearly faster than all the other drivers from the new teams.
In Czech, the comparison is most frequently expressed with the particle ne (38),
but, according to HAVRNEK & JEDLIKA (1981: 188) the genitive case is used
in formal, and in sport journalism. The preposition nad with the locative case
is also a marked device (cf. the idiomatic nad slunce jasnj brighter than the
sun). However, in the spoken obecn etina also jak is used (39) that is the
normal particle in the equative constructions.
(38) Czech (esk nrodn korpus)
Snad vce ne kdekoliv jinde zde plat-, e as

probably more than anywhere elsewhere here apply-3SG that time
jsou penze.
be.3PL money.PL
It is probably more than anywhere else true here that time is money.
(39) Czech (esk nrodn korpus)
Prmrn plocha byt-u je v R
average surface flat-GEN be.3SG Prep Czech Republic
jen o nco vt jak polovina prmr-u
only Prep something larger as half average-GEN
byt-u v Nmeck-u.
flat-GEN Prep Germany-LOC
The average habitable surface of a flat in the Czech Republic is only some
more than half of an average flat in Germany.
The ambiguity of the fixed-case comparative has probably enhanced the rise of
the particle comparative construction where the standard NP derives its case
assignment from the case of the comparee NP (STASSEN 2008: 2). In fact, the
instances of a fixed-case comparative in Russian with the Genitive, and in
Finnish with the Partitive appear equally ambiguous as Classical Latin with the
Ablative in (40).
(40) I love Brutus no less than you (love Brutus) or I love Brutus no less than
(I love) you.
a. Classical Latin (KHNER & STEGMANN 1955: 466 < STASSEN 2008)
Brut-um ego non minus te am-o
Brutus-ACC I NEG less 2SG.ABL love-1SG
b. Finnish
E-n rakasta Brutus-ta yhtn sinu-a vhemmn.
NEG-1SG love-STEM Brutus-PRTV any PRON2SG-PRTV less
c. Russian
- - .
Brutus-GEN I love-1SG NEG less 2SG.GEN
11. Equative constructions
In European languages equative comparison is marked either by interrogative
words meaning 'how'7 which also may function in similative expressions (as
German wie, Russian 99, Estonian kui) or words used in relative clauses and
similative expressions (Scandinavian som as, that).
(41) Bulgarian
,
-.
he is also so clever as and fisherwoman-Def
He is as clever as the fisherwoman.
(42) Serbian
Generalni sekretar UN je pozva-o zemlj-e
general secretary UN is appeal-PTCP.PRET country-PL
svet-a da prema ile-u bude isto toliko
world-GEN PRTCL against Chile-GEN be.T as such
dareljiv-e kao to je ta zemlja bi-l-a dareljiva
generous-PL as what is that country be-PTCP.PRET-F
u januar-u prema Haitij-u, ...
generous-F Prep January-LOC against Haiti-GEN
The Secretary General of UNO appealed to the countries of the world to be
as generous towards Chile as they were in January towards Haiti,
A noticeable discrepancy can be attested between Finno-Ugric, on the one hand,
and the Indo-European languages of Europe, on the other hand, in comparative
constructions. For the comparative, i.e. the cases of inequality there is one
particle, for the equative constructions another (e.g. English bigger than an
elephant, but as big as an elephant), Bulgarian 99 vs. 99, Czech
ne(li) vs. jak, Slovak ne vs. ako, Polish ni(eli), anieli vs. jak, Russian ,
$ vs. 99, Serbian 7 (#), vs. (/) 9 /. However, at least
in some parts of Germany, in spoken casual speech the same particle (wie) is
used in both instances. The same applies to Slovak, moreover, the use of aka
as is accepted as a norm together with ne than (45). This is the Finno-Ugric
pattern. In Finnic, and in Hungarian, the particle used is the same as with
comparisons of equality ((43) and (44)); whereas in all Slavic standard
languages the distribution is complementary ((30a), (31b) and (38)).
(43) Equality: as big as an elephant
a. Finnish yht suuri kuin elefantti
__________
7 This applies actually also to Finnish kuin, although its use in the meaning of kuinka
how is restricted to regional spoken Finnish. The same can be said of the Estonian kui
vs. kuidas how.
382 Hannu Tommola
b. Estonian niisama suur kui elevant

equally big as elephant
c. Hungarian olyan nagy mint egy elefnt
equally big as INDEF elephant
(44) Inequality: bigger than an elephant
a. Finnish suur-empikuin elefantti
b. Estonian suur-em kui elevant
big-COMPV as/than elephant
c. Hungarian nagy-obb mint egy elefnt
big-COMPV as/than INDEF elephant
(45) Slovak
Brat je star ne/ako som ja.
brother is older than/as be.1SG I
My brother is older than I (am).
12. Subject person affixes as strict agreement markers
The subject is obligatory in many languages considered SAE, also when there is
an unambiguous person affix in the verb, but by no means in all. In Slavic, this
applies actually only to the East Slavic languages, which in the past tense do not
have person conjugation, whereas in South and West Slavic the subject pronoun
often may be dropped even in the 3rd person. The same situation prevails in
several Romance languages, and from the Finno-Ugric languages notably in
Hungarian. In Finnic, only the 1st and 2nd person pronouns are optional.
(46) Hungarian
Ki fog neki segteni, ha Berlin-be rkez-ik?
who will Pron3SG help when Berlin-ILL arrive-3SG
Who was going to help him when hed arrive in Berlin?
13. Differentiation between intensifiers and reflexive pronouns
Although most languages in Europe keep intensifiers and reflexives apart, this
characteristic is not at all restricted to Europe alone, and the clustering here can
be attributed to genetic factors (KNIG & SIEMUND 2008: 5).
Differentiation between intensifiers and reflexive pronouns is found in all
Slavic languages (e.g. Slovak sam himself, sama herself but si/sa as reflexive
pronouns). In German the intensifier (selbst) and the reflexive pronoun (sich) are
differentiated, while in English (myself, yourself, him/herself, etc.) the intensifier
also takes the tasks of the reflexive pronouns. In fact, in Germanic languages the
situation varies, and is not unambiguous. Thus, in WALS Scandinavian
languages are along with Dutch (zich(zelf)) classified as showing identity,
which seems dubious (cf. Swedish jag, du, han/hon sjlv I myself, you
yourself, he/she him/herself vs. mig myself, dig yourself, sig him/herself).
(47) Matthew 27:5

a. English (King James Bible)
And he cast down the pieces of silver in the temple, and departed, and went
and hanged himself.
b. Spanish (Modern)
[Entonces l, arrojando las piezas de plata dentro del santuario,]
se apart-, se fue y se ahorc-.
REFL go-AOR REFL go.AOR and REFL hang-AOR
c. German (Luther-Bibel 1912)
[Und er warf die Silberlinge in den Tempel, hob sich davon,]
ging hin und erhngte sich selbst.
go.PRET there and hang-PRET REFL self
d. German (Luther-Bibel 1984)
[Und er warf die Silberlinge in den Tempel,]
ging fort und erhng-te sich.
go.PRET and hang-PRET REFL
e. Slovenian
[Tedaj je vrgel srebrnike v tempelj,]
se umakn-il ter el in se obes-il.
REFL depart-PRET and go.PRET and REFL hang-PRET
f. Estonian
[Ja visanud hberahad templisse maha, Juudas eemaldus,]
lks ra ja poos ene-se les.
go.PRET away and hang-PRET self-GEN up
g. Hungarian
[ pedig eldobvn az ezst pnzeket a templomban, eltvozk;]
s elmen-vn felakasz-t mag-t.
and go-CONV hang-PRET self-ACC
In Finnish itse self is used in all needed positions with proper cases followed
by the enclitic possessive suffix for each person respectively. Estonian differs
from its closest relative only minimally; while there is a corresponding pronoun-
intensifier ise used in the nominative, in the oblique cases it is replaced with a
suppletive stem (e.g. GEN enese, PRTV ennast; see (47f)). Hungarian maga
behaves in a similar way (1st person magam, 2nd person magad etc.), see
(47g). The use of the intensifier in the function where many languages have
separate reflexive pronouns is explained also by the fact that anticausatives are
in Finno-Ugric derived by the suffix, see (48b).
(48) Judas hanged himself
a. Old English < KNIG & SIEMUND (2008)
Judas ahng hine selfne.
Judas hang-PRET 3SG.M self.ACC
384 Hannu Tommola
b. Finnish
Juudas hirtt-yty-i. = Juudas hirtt-i itse-ns.
Judas hang-REFL-PRET Judas hang-PRET self-POSS3
14. Conclusion
There seems to be a great variance in the distribution of most features, and in
many cases the distribution is uneven, sometimes clearly depending of areal
contingency. A few of the features are uncontroversial. Between the different
Slavic languages, however, there seems to be considerable variance. Moreover,
there are some features that clearly separate the Finno-Ugric languages from
SAE.
Quite interestingly, it is claimed that among the languages acknowledged as
SAE there is one non-Indo-European, namely Hungarian. Of course, the affinity
of Hungarian with the Indo-European languages can reasonably be explained by
the fact that the Magyars have long lived in regions largely surrounded by
German and Slavic population, in recent history within the multicultural
Danubian Dual Monarchy of the Austrian Empire and the Kingdom of Hungary.
The availability of the definite and indefinite articles is without any doubt a
European feature; yet most of the Slavic languages lack articles as well. One
feature where Hungarian is farther from SAE than the Finnic languages is the
explicit marking of the subject. Whereas Finnish and Estonian here are mixed in
their behaviour with the mandatory 3rd person pronoun, Hungarian shows pro-
drop in all persons.
The calculated data for the parameters concerning the features discussed
above are demonstrated in the tables and figures in the Appendix. Table 1
presents the values for the parameters (SAE features 1-12) in each language; in
Table 2 the values are cross-tabulated, so that all languages of the sample are
compared with each other, and the mutual correlations are displayed. In Figures
1, 2 and 3, the mutual distance (based on the elaborated SAE features) from all
the other languages of the sample is shown for one Germanic (German), one
Slavic (Russian), and one Finnic (Finnish) language.
As a conclusion a few comments upon the correlations between the
languages can be made. First, what is immediately clear from the charts (Figures
1, 2 and 3) is that the average distance in each genetic group of the languages to
the members of the own group is considerably shorter than to the languages of
the other groups. Second, in the Germanic group German stands out showing
more similarity with the Slavic languages, especially East Slavic and Polish (see
Fig. 1 and 2) than the others. Third, not unexpectedly, Bulgarian and
Macedonian are not only distant from both Germanic and Finno-Ugrian
languages, but also from Russian. Their distance from this relative language is at
the level of Estonian, which in this measuring has more similarity to Russian
than Macedonian.
References
COMRIE, B. (1998) Rethinking the typology of relative clauses. Language Design 1: 5986.
COMRIE, B. & T. KUTEVA (2008) Relativization Strategies. In: HASPELMATH et al. (eds.)
(2008) Ch. 122123.
DAHL, . (ed.) (2000) Tense and aspect in the languages of Europe. Empirical Approaches to
Language Typology. EUROTYP 20-6. Berlin & New York: Mouton de Gruyter.
DRYER, M. (2008a) Definite articles. In: HASPELMATH et al. (eds) (2008) Ch. 37.
DRYER, M. (2008b) Indefinite articles. In: HASPELMATH et al. (eds) (2008) Ch. 38.
JNTREJ (1982) = , .., >9 &9 (
). .: - -.
GIGER, M. (2003) Resultativa im modernen Tschechischen. Unter Bercksichtigung der
Sprachgeschichte und der brigen slavischen Sprachen. Slavica Helvetica, 69. Bern:
Peter Lang.
GRAVES, N. (2000) Macedonian a language with three perfects. In: DAHL, . (ed.) (2000):
479494.
HASPELMATH, M., M. S. DRYER, D. GIL & B. COMRIE (eds) (2008) WALS (The World Atlas of
Language Structures) Online. Munich: Max Planck Digital Library.
HASPELMATH, M. (2001) The European linguistic area: Standard Average European.
Handbuch der Sprach- und Kommunikationswissenschaft vol. 20.2, pp. 14921510.
HAVRNEK, B. & A. JEDLIKA ([1959]1981)4 esk mluvnice. Praha: Sttn pedagogick
nakladatelstv.
KNIG, E. & P. SIEMUND (with S. TPPER) (2008) Intensifiers and reflexive pronouns. In:
HASPELMATH et al. (eds) (2008) Ch. 47.
LAANEMETS, A. (2004) Dannelse og anvendelse af passiv i dansk, norsk og svensk. Nordistica
Tartuensia 11. Tartu: Tartu University Press.
LAANEMETS, A. (2009) The passive voice in written and spoken Scandinavian. In: FRYD, M.
(ed.) The passive in Germanic languages. GAGL 49. 144166.
LINDSTEDT, J. (2000) The perfect aspectual, temporal and evidential. In: DAHL, . (ed.)
(2000) 365383.
SAAVEDRA, D. (2007) Pasivne konstrukcije u hrvatskome i bugarskome jeziku. LAHOR:
journal for Croatian as mother, second and foreign language 2, 4: 217241.
SIEWIERSKA, A. (2008) Passive constructions. In: HASPELMATH et al. (eds) (2008) Ch. 107.
STASSEN, L. (2008) Comparative constructions. In: HASPELMATH et al. (eds) (2008) Ch. 121.
SVETLK, J. (1972) Najcharakteristickejie rozdiely medzi sloveninou a rutinou. In:
MISTRK, J. (ed.) Studia Academica Slovaca, 1. Prednky VIII. Letnho seminra
slovenskho jazyka a kultry. Bratislava: stav kolskch informaci. 245254.
TOMMOLA, H. (1997) = , , . "# <
"9 9 67. 79, [ , <. 1. Tartu: Tartu
likooli Kirjastus, 173187.
TOMMOLA, H. (1998) = , , . In: ..
& .. (eds), <7. 9. ^9. -
: , 4157.
TOMMOLA, H. (2000) On the perfect in North Slavic. In: DAHL, . (ed.) (2000) 441478.
WHORF, B. L. (1941) The relation of habitual thought and behaviour to language. In: SPIER, L.
(ed.) Language, culture, and personality: essays in memory of Edward Sapir.
WIEMER, B. & M. GIGER (2005) Resultativa in den nordslavischen und baltischen Sprachen.
Bestandsaufnahme unter arealen und grammatikalisierungstheoretischen
386 Hannu Tommola
Gesichtspunkten. LINCOM Studies in Language Typology, 10. Mnchen: LINCOM

EUROPA.
Hannu Tommola, University of Tampere, Department of Translation Studies,

Pinni B, 4th floor, FI-33014 Tampere, Finland, hannu.tommola@uta.fi
Appendix
Table 1. Values for the parameters (SAE features 0112)

Feature / Eng Grm Dan Nor Swd Bel Rus Cze Pol Slk Blg Mac SCr Hun Est Fin
Language
011 Article 1 1 1 1 1 0 0 0 0 0 1 1 0 1 0 0
012 Prep 1 1 1 1 1 0 0 0 0 0 0 0 0 1 0 0
DefArt
013 Encl 0 0 1 1 1 0 0 0 0 0 1 1 0 0 0 0
DefArt
021 RelPron 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
022 Gap 1 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0
023 EmbPart 0 1 0 0 0 0 1 0 0 0 0 0 0 1 1 1
031 Perfect 1 0 1 1 1 0 0 0 0 0 1 1 0 0 1 1
032 habeo-PRF 1 1 1 1 1 0 0 0 0 0 0 1 0 0 0 0
033 essere-PRF 0 1 1 0 0 0 0 0 0 0 1 1 0 0 1 1
041 NomExp 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
042 IndObj 0 1 0 0 1 0 0 0 0 0 0 0 0 1 0 1
043 DativeExp 0 1 0 0 0 1 1 1 1 1 1 1 1 0 1 0
051 PRTPPass 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0
052 1 1 0 0 0 1 1 0 1 0 1 0 0 0 0 0
PassActional
053 PassState 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 1
054 PassRefl 0 0 1 1 1 1 1 1 1 1 1 1 1 0 0 0
055 Ambi 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1
061 AntiCaus 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
062 CausDeriv 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
071 DatExtPoss 0 1 0 0 0 1 1 1 1 1 1 1 1 0 0 0
081 NegIndef 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0
082 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
PredicateNeg
083 Mixed 1 0 1 1 1 0 0 0 0 0 0 0 0 1 1 1
091 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
CompParticle
092 FixedCase 0 0 0 0 0 0 1 1 0 0 0 0 1 1 1 1
101 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
EquatParticle
102 0 1 0 0 0 0 0 1 0 1 0 0 0 1 1 1
Equat=Comp
111 OblSubj 1 1 1 1 1 0 1 0 0 0 0 0 0 0 0 0
112 3rdPers 1 1 1 1 1 1 1 0 1 0 0 0 0 0 1 1
113 ProDrop 0 0 0 0 0 0 0 1 0 1 1 1 1 1 0 0
121 Intens/Refl 0 1 0 0 0 1 1 1 1 1 1 1 1 0 1 0
388 Hannu Tommola
Table 2. Correlations between the languages of the sample according to the specified features
Grm
Mac
Swd
Hun
Dan
Language
Nor
Rus
Cze
SCr
Blg
Bel
Pol
Slk
Fin
Est
English 0,34 0,67 0,74 0,67 0,04 - - 0,04 - - - - - - -
0,04 0,29 0,22 0,11 0,11 0,22 0,09 0,20 0,10
German 0,11 0,07 0,11 0,21 0,21 - 0,21 0,07 - - - - - -
0,02 0,03 0,03 0,07 0,07 0,10 0,12
Danish 0,94 0,89 - - - - - 0,07 0,21 - - - -
0,02 0,11 0,22 0 0,15 0,15 0,15 0,10 0,04
Norwegian 0,94 0,04 - - 0,04 - 0,02 0,15 - - - -
0,04 0,16 0,09 0,09 0,09 0,20 0,10
Swedish - - - - - - 0,07 - - - -
0,02 0,11 0,22 0 0,15 0,06 0,15 0,02 0,20 0,04
Belorusian 0,82 0,68 1 0,74 0,64 0,51 0,74 - 0,30 0,10
0,04
Russian 0,62 0,82 0,56 0,41 0,28 0,69 0,04 0,35 0,16
Czech 0,68 0,94 0,56 0,56 0,94 0,23 0,36 0,16
Polish 0,74 0,64 0,51 0,74 - 0,30 0,10
0,04
Slovak 0,64 0,64 0,87 0,22 0,30 0,10
Bulgarian 0,87 0,64 - 0,15 -
0,02 0,04
Macedonian 0,64 - 0,15 -
0,02 0,04
Serbian 0,22 0,30 0,10
/Croatian
Hungarian 0,30 0,49
Estonian 0,81
German
0,4
0,3
0,2
0,1
-0,1
-0,2 Eng Dan Nor Swd Bel Rus Cze Pol Slk Blg Mac SCr Hun Est Fin
German 0,344 0,113 0,067 0,113 0,21 0,206 -0,022 0,21 0,072 -0,03 -0,027 -0,07 -0,067 -0,1 -0,12
Figure 1. Distance of German from the other languages of the sample

390 Hannu Tommola
Russian
0,9
0,8
0,7
0,6
0,5
0,4
0,3
0,2
0,1
-0,1 Eng Grm Dan Nor Swd Bel Cze Pol Slk Blg Mac SCr Hun Est Fin
Russian -0,042 0,21 -0,02 0,042 -0,02 0,824 0,619 0,82 0,563 0,41 0,28 0,69 0,042 0,35 0,16
Figure 2. Distance of Russian from the other languages of the sample
Finnish
0,8
0,6
0,4
0,2
-0,2
Eng Grm Dan Nor Swd Bel Rus Cze Pol Slk Blg Mac SCr Hun Est
Finnish -0,1 -0,12 -0,04 -0,1 -0,04 0,1 0,16 0,16 0,1 0,1 -0,04 -0,04 0,1 0,49 0,81
Figure 3. Distance of Finnish from the other languages of the sample

Some multiply prefixed verbs as covert serial verb
constructions
Rok aucer
This paper takes a new perspective on certain multiply prefixed Slavic verbs,
such as (1): it links them to seemingly unrelated serial verb constructions
(SVCs), which are combinations of two verbs with one tense value and no
marker of coordination or subordination1, such as (2).
(1) Konduktor ue na-ot-ryvala biletikov. (Russian)
ticket-seller already on-off-tore ticketsGEN (ROMANOVA 2007: 273)
The ticket-seller has prepared a lot of tickets by tearing them off the roll.
(2) z gh gb w khin. (Edo)
Ozo FUT hit goat sell
Ozo will kill the goat and sell it. (BAKER & STEWART 2002)
Multiple prefix constructions like (1) are typically analyzed as combining a verb
with a prefix that originates in the VP (ot-) and a prefix that originates above the
VP in the Infl-domain (na-). Serial verb constructions like (2), on the other hand,
are typically analyzed as combinations of multiple VPs under a single TP.
In this paper, I will argue that the recently advanced analysis of (1), whereby
the stem-adjacent prefix originates as a VP-internal resultative predicate and the
left-hand prefix originates in the Infl-domain of the same VP, cannot be correct,
since the left-hand prefix (just like the stem-adjacent prefix) tests positive on
diagnostics of resultativity. I will suggest that multiply prefixed strings of the
type in (1) can instead be fruitfully analyzed as covertly combining two VPs
under a single TP, much in the spirit of various analyses of SVCs; in addition to
the V that is phonologically realized in (1) by the stem tear, which embeds the
stem-adjacent prefix as a resultative secondary predicate, I will posit a second V,
realized by a phonologically null verb which embeds the left-hand prefix as a
resultative secondary predicate. This analysis reconciles our multiply prefixed
cases with a widespread assumption in the theory of resultatives, i.e. that there
can be only one independent resultative secondary predicate per VP.2
__________
1 There is some disagreement about the set of defining features of SVCs (cf. AIKHENVALD
2006).
2 It is important to note that not every stacked prefix is claimed to come with its own VP.
In addition to the accumulative use of na- discussed here, AUCER (2009) proposes a
two-VP structure for two other prefix uses, but also mentions cases which do not contain
two VPs but rather a resultative prefix and a result-modifying prefix.
392 Rok aucer
1. Introduction
It is often noted that only languages with poor tense, person and number
morphology on verbs allow SVCs of the type found in, say, Edo (BAKER &
STEWART 2002, MUYSKEN & VEENSTRA 2006: 263-5, etc.), (3).
(3) z gh gb w khin. (Edo)
Ozo FUT hit goat sell
Ozo will kill the goat and sell it. (BAKER & STEWART 2002)
This correlation covers Slavic languages, in which the verb (unless packed
inside a nominalization, etc.) must occur inflected up to subject agreement, and
which do not show SVCs of the type in (3).
At the same time, syntactic theory has posited a number of null verbs, i.e.
verbs which are present in the syntax/interpretation but are phonologically null
(ROSS 1979, MCCAWLEY 1979, RIEMSDIJK 2002, MARUI & AUCER 2006a,
etc.). The best-known is the null HAVE posited for cases like (4a), which are
assigned the structure in (4b) (ROSS 1979, MCCAWLEY 1979, MARUI &
AUCER 2006B).
(4) a. John wants a new bike.
b. [John wants [TO-HAVE a new bike]]
Now, given that SVCs of the type in (3) are typically analyzed as combining two
VPs or AspPs under a single TP/AgrSP (i.e. two non-fully inflected verbs), the
following question arises: could we find SVCs in rich-verbal-inflection
languages when one of the serialized verbs is a null verb? After all, if the
requirement that verbs be inflected for tense/subject agreement is some sort of a
PF condition on the verb, one could imagine that it need not apply if we are
dealing with a phonologically null verb; and if the requirement stems from a PF
condition on the element that realizes tense/subject agreement (i.e. from its
affixal rather than free-word specification), a single overt verb should be enough
to satisfy it given that the structure has a single TP/AgrSP (cf. also MUYSKEN &
VEENSTRA 2006: 263-5, BAKER & STEWART 2002).
In this paper, I propose that some multiply prefixed Slavic verbs, such as
the one in (1) (repeated below), in which the stacked prefix is one type of the
measure na-, exhibit just such a situation: an SVC in which one of two
resultative VPs is headed by a null verb, and the two VPs occur under a single
TP and AgrSP, and share their internal argument, as in the simplified (5).
Some multiply prefixed verbs as covert serial verb constructions 393
(5) TP
ei
VP
wo
VP VP
ru to
V SC V SC
r- tu tu
tear biletikovi ot- biletikovi na-
tickets off tickets on
2. Accumulative/cumulative/vague-measure prefixation with na-

2.1 Background
Slavic prefixes are widely assumed to split into resultative and non-resultative
prefixes (a.k.a. lexical/internal/inner vs. superlexical/external/outer). Resultative
prefixes are characterized by having argument structure changing effects on
their base verb, by contributing idiosyncratic or spatial meanings to their base
verb, by attaching directly to the verbal stem rather than stacking over other
prefixes, and by the fact that there can be only one such prefix per verb. Non-
resultative prefixes, on the other hand, are characterized by contributing adverb-
like, measure or aspectual meanings to their base, by being able to stack over
other prefixes, and by having no argument structure changing effects.
It is typically assumed that the difference between the two classes is
captured by assigning them different structures. Resultatively prefixed verbs are
assigned the structure in (6a), which is also the structure widely assumed for
resultative particle verbs in Germanic languages; non-resultative prefixes are
claimed to be introduced in the clausal structure above the verb phrase, (6b).3
(6a) VP (6b)
ru ru
V SC/ResultP PExtPrf(P) VP
ru 6
DP PIntPrf(P)
Seeing the prefix in (6a) as an argument-introducing element, and having its
small clause (rather than a DP) act as the complement of the verb captures the
argument-structure effect of resultative prefixes (i.e. loss of selection restrictions
of the verb stem over the object after the addition of the prefix). On the other
__________
3 (6a)-(6b) are generalizations that simplify many details, which are not relevant for the
discussion (see SVENONIUS 2004, ROMANOVA 2007, TATEVOSOV 2008, AUCER 2009,
etc.).
394 Rok aucer
hand, introducing the prefix above the VP in (6b) captures the lack of such
effects of non-resultative prefixes, as well as their adverb-/aspect-/measure-like
effects and their ability to stack over resultative prefixes.
Despite the clearly different behavior that the two structural options in (6)
lead one to expect from a prefix, accumulative na-, exemplified in (1) above and
(7) below, has been seen both as a resultative (PIN 1994, BABKO-MALAYA
1999, BISKUP 2007) and a non-resultative prefix (PERELSTVAIG 2006,
ROMANOVA 2007, TATEVOSOV 2007, 2008, etc.).
(7) Dti na-rvli cvty/ cvetv na lug. (Russian)
children on-plucked flowersACC flowersGEN in meadow
The children picked a lot of flowers in the meadow. (FILIP 2000: 49)
The first position was based on the observation that na- changes the base verbs
selection restrictions, but the accounts did not mention cases of stacked na-. The
second position was based on the observation that na- can stack, and tried to
explain its selection-changing effects by treating it as some sort of
quantificational element/measure expression.
In what follows, I will show that na- should be analyzed as a resultative,
argument-introducing prefix with the structure in (6a). Moreover, I will suggest
that when na- is found stacked over another resultative prefix, as in (1) above,
we have an SVC-like structure with two resultative VPs; each of the resultative
prefixes belongs to its own VP, though one of these Vs is phonologically null.4
2.2 Accumulative na- is resultative
The hallmark of resultative secondary predication is the change in the selection
properties of the resultative verb when compared to those of the base verb. The
most radical and best known selection change is often discussed under the label
unselected objects: a nominal that cannot function as the internal argument of a
verb due to a semantic-selection violation (#write the car) is acceptable with the
same verb in the presence of a resultative predicate (9write the car off) (see
MCINTYRE 2007). Indeed, the most widely adopted syntactic analysis of
resultatives, outlined in (6a) above, is designed to take care of precisely these
cases: since the internal argument is introduced by the resultative secondary
predicate, it need not respect the s-selection properties of the base verb.
Turning to verbs prefixed with the accumulative na-, we see that they test
positive on the diagnostic of changed selection properties. The pattern in (8)
__________
4 I do not discuss another measure-like use of na- here, i.e. the reflexive-introducing use in
(i). See AUCER (2009, 2010) for a detailed discussion; although cases where this na-
stacks over another prefix are also analyzed there as containing two resultative VPs, the
two uses also show important differences.
(i) Tone se je na-govoril. (Slovenian)
Tone self is on-talked
Tone got his fill of talking.
shows, firstly, that the addition of na- blocks an internal argument that is a
perfectly good internal argument in the absence of na-, (8a-b), and secondly,
that the addition of na- licenses an internal argument which is not supported by
the base verb in the absence of na-, (8c-d).5
(8) a. molsti veliko krav
milkINF a-lot cowsGEN
b. #na-molsti veliko krav (Slovenian)
on-milkINF a-lot cowsGEN
milk a lot of cows
c. #molsti veliko mleka
milkINF a-lot milkGEN
d. na-molsti veliko mleka
on-milkINF a-lot cowsGEN
obtain a lot of milk by milking
Therefore, the conclusion must be that the internal argument of na-verbs is
introduced by the prefix rather the verb, and that na- must thus be a resultative
prefix, as in (6a), rather than a non-resultative prefix above the VP, as in (6b).
In addition to such obvious cases, there also exist cases which are often
assumed not to exhibit na-licensed unselected objects but which, on closer
scrutiny, also turn out to exhibit just that. Measure expressions like the one in
(9), for example, are often assumed to be adjuncts (cf. ROMANOVA 2007 for the
Russian equivalent).
(9) na-letati 300 ur (Slovenian)
on-fly 300 hours
accumulate 300 hours of flying
However, when predicates like (9) are put under sentential negation, the
measure expression turns genitive, i.e. it undergoes the genitive of negation,
which affects direct internal arguments and not adjuncts. To be able to
appreciate this case change, the numeral 300 from (8) must first be replaced with
something like 303. This is because unlike numerals such as 300, which prevent
the noun from getting the usual case of the direct object and assign genitive case
to the noun, Slovenian numerals from 1 to 4, from 101 to 104, etc., function like
simple adjectives and do not affect the case of the noun. In other words, the
numeral 300 in (8) will make the noun hours genitive regardless of sentential
negation, and thus prevent using the genitive of negation to test its
argumenthood; but switching 300 for 303 makes this test applicable. As shown
__________
5 See ROMANOVA (2007: 202) for a parallel pattern from Russian based on the stem grab-
(grabitj 9proxoix/#denjgi rob passers-by/money vs. na-grabitj #proxoix/ 9denjgi
steal a lot of passers-by/money).
396 Rok aucer
in (10), the measure expression with na-verbs becomes genitive under sentential
negation, showing that it is not an adjunct but rather a direct internal argument.
(10) Ta pilot letos e ni na-letal
9tristotreh ur/*tristotri ure. (Svn)
this pilot this-year still not-is on-flown
303GEN hrsGEN 303ACC hrsACC
This year, this pilot has not accumulated 303 hours yet.
Given the intransitive base verb, the measure expression must thus be introduced
as an argument of the prefix, suggesting that the prefix is a resultative one, as in
(6a). Not surprisingly, (10) thus contrasts clearly with the minimally different
prefixless(11), where the measure expression cannot turn genitive: with an
intransitive verb and in the absence of the prefix, it can only be introduced as an
adjunct, unaffected by sentential negation.6
(11) Ta pilot letos e ni letal/letel
*tristotreh ur/9tristotri ure.
this pilot this-year still not-is flown
303GEN hrsGEN 303ACC hrsACC
This year, our pilot has not flown for 303 hours yet.
Moreover, the internal-argumenthood of the measure expression in (10) can be
supported with the do so constituency test. When the verb is prefixed with na-
and the expression do this is used to refer back to the first clause, the measure
expression must be part of the constituent replaced by do this, so that trying to
leave it out results in ungrammaticality, (12)-(13). This is unexpected if the
measure expression is an adjunct, but not if it is an internal argument.
(12) Ju je na-laufal 300 km lani, rt pa je to naredil letos.
Ju is on-run 300 km last-year rt ptcl is this done this-year
Ju accumulated 300 km last year, and rt did so this year. (Svn)
(13) *Ju je na-laufal 300 km (lani), rt pa je to naredil 200 km
Ju is on-run 300 km last-year rt ptcl is this done 200 km
(letos).
this-year
In addition to the genitive-of-negation facts in (9)-(11) above, the do so facts
present another piece of evidence that the measure expression of na-verbs is an
argument, not an adjunct. Given an intransitive base verb such as fly, it is thus
reasonable to assume that the measure expression is introduced by the prefix na-
, especially in view of the fact that an argument-introducing character must
__________
6 See AUCER (2009: 87-9, 152-4) for more discussion and a clarification of some
confounding factors with respect to the genitive-of-negation test.
independently be posited for na- for cases with non-measure-expression

unselected objects, such as milk in (8) above. Na-, then, must be a resultative
prefix. Several further selection-based arguments for a resultative status of na-
can be found in PIN (1994), BABKO-MALAYA (1999), FILIP (2000,2005),
PERELTSVAIG (2006), BISKUP (2007) and AUCER (2009).7,8
2.3 Measure/quantifier-like properties of na-
It has been claimed that unless on the kind reading, the internal argument of
na-verbs cannot be a singular count noun (PIN 1994, FILIP 2005, PERELTSVAIG
2006), (14a-b), and it must be nonspecific indefinite and as such not preceded by
demonstratives or quantifiers like all (FILIP 2005, PERELTSVAIG 2006), (14c).
(14) a. Ju je na-kradel 50 koles.
Ju is on-stolen 50 bikes
Ju amassed 50 bikes by stealing. (Slovenian)
b. #Ju je na-kradel (eno) kolo.
Ju is on-stolen one bike
c. #Ju je na-kradel tistih biciklov /tiste bicikle.
Ju is on-stole those bikesGEN /those bikesACC
It has been proposed that these quantity and definiteness selection restrictions of
na-verbs can be explained if na- is treated as a VP-external functional element
which introduces the direct object but whose c-selection features restrict the
possible objects to small nominals/NumPs (PERELTSVAIG 2006), or if na- is
treated as a VP-external quantification-at-a-distance-like event quantifier
(ROMANOVA 2007), but not if it is a resultative secondary predicate.
Note, however, that such quantity and definiteness restrictions hold also of
internal arguments of some rather plain-looking resultative verbs for which no
VP-external material has been posited, such as amass or accumulate, (15).
__________
7 If na- is resultative, we expect that unless unaccusative, na-verbs will normally have to
occur with a direct object. While this is generally claimed to be the case (e.g. PIN
1994, FILIP 2005, PERELTSVAIG 2006, BORIK 2008), ROMANOVA (2007) challenges this
position; however, her counterexamples are explained away in AUCER (2009).
8 Space restrictions prevent me from reviewing PERELTSVAIGs (2006) and TATEVOSOVs
(2007) analysis of na-, where the latter is also seen as an argument-introducing prefix but
still as located above the VP. See AUCER (2009: 130-3) for a detailed discussion and
refutation of that analysis, though one problem for this analysis will be indirectly
presented in the following section, along with showing that the main set of data that
motivated this analysis does not require a VP-external analysis in the first place.
398 Rok aucer
(15) a. Ju amassed 50 bikes.

b. #Ju amassed a/one bike.
c. #Ju amassed those bikes.
In fact, these restrictions on the internal arguments of both na-verbs and amass
are context-sensitive, that is, they depend on information packaging and are
cancelable, as shown in (16) (see translation lines for English amass).
(16) a. Tistih 50 roic, ki jih je Ju na-trgal veraj je
those 50 flowers that them is Ju on-plucked yesterday is
Those/The 50 flowers that Ju plucked/amassed yesterday are
(Svn)
b. [context: And how many did Ju manage to amass?]
Hja, Juu je ratal pa na-trgat eno samo roco.
well Ju is managed ptcl on-pluck 1 only flower
Well, J. managed to pluck/amass one single flower.
Firstly, this cancelability establishes a further parallel between na-verbs and
some ordinary resultative verbs such as amass, for which no VP-external
material has been posited. Secondly, if the bans on singular count-noun internal
arguments and definite specific internal arguments were due to na-s c-selection
features or to its being an event-quantifying functional element, they should not
be pragmatically cancelable. And on the other hand, if na- is resultative, the
cancelability of the restrictions is not surprising at all if only we assume that the
na-headed resultative small clause has the shape of a there-be-like predicate
(be in quantity (somewhere)) and the internal argument of na-verbs originates
as the subject of the there-be-like small clause; a cancelable ban on
definiteness effects is well-attested with what KEARNS (2000) calls there-be
constructions (typically an existential or representational there-be, sometimes a
task or list there-be) (cf. MCNALLY 2009).
2.4 Structure
Section 2.2 has provided the most standard kind of evidence for resultativity of a
prefixunselected objectsand section 2.3 has shown that even the quantity
and definiteness restrictions on the internal argument of na-verbs do not warrant
a VP-external quantifier-like treatment. The quantity restriction was shown to be
similar to what we find with resultative verbs like amass, where the restriction
arguably comes from the incorporated meaning of mass, and the cancelable
definiteness restriction is similar to what we find with there-be constructions.
With this in mind, I propose that the VP of a na-verb has the resultative
structure in (17), i.e. a standard resultative structure, which differs from the
structure of verbs like amass only in that its manner component is specified.
(17) VP
ru
V SC/ResultP
trg- ru
pluck DP PP
6
na- ...
If amass glosses with cause sth to be in/form a mass, na-verbs gloss with
something like cause sth to be in/form a mass by V-ing; for (17), cause sth to
be in/form a mass by plucking. Often, a na-verb can thus be interpreted as a
sort of creation verb, loosely understood, with a quantity of something coming
into existence or coming to exist at a particular location (cf. ROMANOVA 2007
and AUCER 2009 for data). AUCER (2009) discusses in detail how the meaning
of mass (i.e. the measure/quantity component) arises in the PP, suggesting that
the prefixal preposition has a null mass-like complement in the syntax.9
2.5 Stacked na- is resultative
It is well-known that at least in some cases and in some Slavic languages,
accumulative na- can be found stacked over a resultative prefix, as in (18).
In accounts that see accumulative na- as resultative (PIN 1994, BABKO-
MALAYA 1999, BISKUP 2007), such cases have gone undiscussed; in the rest of
the literature on na- (e.g. PERELSTVAIG 2006, ROMANOVA 2007, TATEVOSOV
2007, 2008), the capacity to stack has been seen as evidence that na-not just
when stacked but in its accumulative use in generalis not resultative.
Now, the previous sections have established that the accumulative na- must
be resultative. So unless one is ready to give up on the standard diagnostics of
resultativity, we are left, in principle, with two options: to treat the unstacked
and stacked na-s as radically different, one resultative and the other not, or to
devise an account in which both can be treated as resultative. The first option, of
course, will only make sense if the two na-s also behave radically differently.
However, when cases with a stacked na- are compared to the cases with an
__________
9 I assume that the genitival argument of na-verbs does not originate as an internal
argument of na- but as its external argument, i.e. the subject of the small clause. This is
unlike PERELTSVAIG (2006) and TATEVOSOV (2007), who seek to explain the genitive
case on this argument (among other things) as a direct consequence of its being selected
by na-. I refer the reader to AUCER (2009) for justification of this assumption, but also
note that as far as I can see, the choice is not crucial for the main claim here, which is
that na- and the genitive argument both originate in a resultative small clause.
400 Rok aucer
unstacked na-, they show all of the characteristics that the proponents of the VP-
external account typically consider as the defining characteristics of
accumulative na-. That is:
a) whether stacked or not, accumulative na- contributes the same vague-
measure meaning;
b) whether a verb has na- stacked over another prefix or not, its internal
argument cannot be a singular count noun (see 2.3 above for exceptions);
c) whether a verb has na- stacked over another prefix or not, its internal
argument must be nonspecific indefinite (see 2.3 above for exceptions);
d) whether a verb has na- stacked over another prefix or not, its internal
argument can, in some Slavic languages, and perhaps must, in other
Slavic languages, be in the genitive (see AUCER 2009 for details).
If na- exhibits all of the characteristics that the proponents of the VP-external
account typically consider as the defining characteristics of accumulative na-,
then it is clearly reasonable to assume that the prefix has the same origin in both
cases. And given that we determined that it must be resultative when stem-
adjacent, then it is reasonable to conjecture that it is resultative also when it is
stacked over another prefix.
Despite the parallel behavior on the four points above, it should be noted,
however, that the stacked and unstacked/stem-adjacent na- are marked by a
crucial difference: unlike the unstacked/stem-adjacent na-, the stacked na- does
not have selection-changing effects with respect to its input, i.e. the singly
prefixed verb. A nominal will only be an acceptable internal argument of a verb
with a stacked na- if it is already acceptable with the input of this na-, i.e. the
singly-prefixed verb without na-. For example, tickets is an acceptable
argument to Russian ot-ryvat tear off, and so it is also acceptable with na-ot-
ryvat; but there exist no cases with a stacked na- where the argument would
count as an unselected object with respect to the singly-prefixed verb, in the
sense of the data with an unstacked/stem-adjacent na- from section 2.2 above.
We thus need an account which will capture both the shared characteristics in a)
to d) above and the difference with respect to selection-changing effects.
At first sight, neither a simple VP-external account, as in (6b) above, nor a
simple resultative structure, as in (6a), can capture this (nor can, for that matter,
an account that would treat the stem-adjacent na- as resultative and the stacked
na- as VP-external, since the shared characteristics will be unaccounted for). In
what follows, I will propose a structure which will capture the shared
characteristics by treating both a stacked and an unstacked na- in the same
wayas a resultative prefixand the difference between them will follow from
argument sharing between the two resultative VPs of the doubly prefixed verb.
3. SVC-like structure
In 2.4 above, I analyzed verbs with an unstacked na- as amass-like resultatives.
Given the shared characteristics of unstacked and stacked na- (meaning
contribution, restrictions on the verbs internal argument), verbs with a stacked
na- should also be amass-like resultatives. Now, it is often observed that there
can be only one independent resultative secondary predicate per verb
(GOLDBERG 1995, RAPPAPORT & LEVIN 2001, DIKKEN 2003, RAMCHAND 2008,
etc.), (19).10
(19) a. Sam kicked Bill 9out of the room / 9black and blue / *out of the
room black and blue. (GOLDBERG 1995: 81)
b. send the letters 9up / 9away / *up away (DEN DIKKEN 2003)
In fact, this generalization appears strong enough to have been built into the
theory of resultative secondary predication (RAPPAPORT & LEVIN 2001,
SVENONIUS 2004, RAMCHAND 2008, etc.). Therefore, cases like (1) above, in
which the resultative na- stacks on top of another resultative prefix, present us
with two options: to give up the aforementioned generalization and with it the
mainstream analysis of resultatives, or to hypothesize that doubly-prefixed cases
like (1) containdespite appearancestwo resultative VPs, one of which is
headed by a null verb. Given that the generalization of one resultative per verb is
otherwise robust, giving it upand with it the widespread mainstream analysis
of resultativesseems undesirable. Moreover, for a related doubly-prefixed
case, the covert presence of two VPs has also been supported with various kinds
of independent syntactic evidence (AUCER 2009, 2010). Therefore, I will
simply assume here that positing the presence of a null verb is the right way to
go, and refer the reader for a longer defense of this position to AUCER (2009,
2010).
Now, looking at the interpretation of doubly-prefixed cases such as (1),
repeated below, it is evident that we have an amass-like event (i.e. producing
a lot of tickets), which is carried out in a certain manner, i.e. by tearing tickets
off the roll. Therefore, the two resultative VPs should be combined in such a
way that one will provide the main resultative event and the other its manner. I
follow several accounts of SVCs, which encode such a relation between the two
VPs of an SVC through adjunction of one VP (or AspP) to the other (BAKER &
STEWART 2002, MUYSKEN & VEENSTRA 2006, etc.); (1) will thus have the
structure in (20) (see ARSENIJEVI 2007 for a closely related proposal).
__________
10 Nested particles/PPs (run down into the woods), in which one PP represents a further
specification of the other and both belong to the same result, are a different case.
402 Rok aucer

(20) TP
ei
AspP
ru
VP
wo
AspP VP
ru ru
-yva- VP V SC
ru ty
V SC biletikovi na-
r- ty tickets on
tear biletikovi ot-
tickets off
In (20), then, we have two VPs/AspPs under one TP/AgrSP, very much like
what is known from much of the literature on SVCs. The adjoined VP/AspP
provides the VP from the main projection line with manner, and each VP comes
with its own prefix-headed resultative secondary predicate, so the generalization
of one result per verb is preserved and the mainstream analysis of resultatives
can easily be employed. As already noted above, interpretative, syntactic and
morphosyntactic evidence for two VPs/AspPs is clearest in a related doubly-
prefixed construction (AUCER 2009), and I will not review it here; at the same
time, there is neither morphosyntactic nor interpretative evidence for two
SubjectAgrPs or two TPs, so despite the presence of two VPs, (20) will only
contain one set of (the top part of) Vs extended projection.
Whereas the V of the manner VP is filled with the root tear, the V of the
main projection line is null. One may ask, of course, why this should be so and
why we do not find cases where both Vs would be realized overtly. As
mentioned in section 1, verbs in Slavic must occur inflected up to subject
agreementa feature that has been crosslinguistically linked to the absence of
standard SVCs (BAKER & STEWART 2002, MUYSKEN & VEENSTRA 2006, etc.)
and so two overtly realized verbs could not survive in a structure with a single
TP/AgrSP. On the other hand, assuming that the full-inflection condition on
Slavic verbs is essentially a PF requirement (whether originally stemming from
the verb or the affix), nothing will prevent SVC-like structures such as the one
in (20) if one of the Vs is null. Indeed, given that syntactic theory has
independently posited null verbs (cf. section 1 above), the existence of SVC-like
structures with one of the V's null thus actually becomes a predicted possibility
(when all other well-formedness requirements are also met).
As for the nature of the null verb, it can be either a phonologically null V
with some basic semantics (such as ACT/UNDERGO), or if one wishes to dispense
with lexical categories and replace V with a functional category, it can be a
functional node not filled by a root, with just its just functional content
interpreted. And since a result predicate seems to presuppose a dominating VP
(cf. RAMCHAND 2008), the null verb will be recoverable through the presence of
its resultative prefix na-. The same basic two-VP structure has also been argued
to capture two other doubly-prefixed constructions (AUCER 2009).11
4 SVC-like argument sharing between the VPs
Despite two VPs and two resultative prefixes, (1) above shows just one object. If
each result predicate requires a subject (RAMCHAND 2008), then it can only be
that the two result predicates in (1) obligatorily share their subject. This
requirement is captured if just like one TP/AgrSP, the structure in (20) also
contains just one sentential object-licensing projection (e.g. one AgrOP), thus
preventing two different objects. In (20), this argument sharing is marked with
tickets placed in both result predicates and coindexed.
Rather than being just some desperate magic to save the account, this
argument sharing actually correctly predicts the split in the behavior between
singly-prefixed and doubly-prefixed na-verbs that was briefly mentioned in
section 2.5. When na- is the only prefix, it licenses objects that are unselected
with respect to its input, i.e. the unprefixed verb root, (8). But since in strings
with a stacked na-, the object is a shared argument of both prefixes, a stacked
na- cannot license objects that are unselected with respect to its input, i.e. the
singly prefixed verb. And in turn, the presence of object sharing in strings with a
stacked na- further strengthens the proposed parallel between strings with a
stacked na- and SVCs, since object sharing is well-attested in several types of
SVCs (BAKER & STEWART 2002, MUYSKEN & VEENSTRA 2006, etc.).
5. Conclusion
I started out with the observation that Slavic languages predictably lack standard
SVCs, but that given the postulation of phonologically null verbs in syntactic
theory, one might expect that SVCs will be possible if one verb is null. I then
argued that doubly-prefixed accumulative na-verbs instantiate just this option:
__________
11 The claim that the two VPs are combined with adjunction faces some challenges, most
notably the availability of extraction. In the context of such a proposal for Edo SVCs,
BAKER & STEWART (2002, fn. 14) suggest that it may be that only full clauses count as
adjunct islands (cf. also TRUSWELL 2007). In fact, there exist several alternatives for
combining the VPs of SVCs (double-headedness, conjunction, complementation), but
each of these faces challenges in its own right, so I stick to the most common account.
404 Rok aucer
after showing that accumulative na- is resultative and na-verbs very much like
plain verbs such as amass, I then argued that when na- is stacked over another
resultative prefix, we have an amass-like verb whose manner is specified by its
singly-prefixed input. I proposed that despite appearances, doubly-prefixed na-
cases contain two VPsone adjoined to the otherunder a single TP/AgrSP, as
had been proposed for SVCs. This allowed me to uphold the generalization that
there can be only one resultative per verb, and with it the mainstream analysis of
resultatives. The parallel with SVCs turned out to be further supported with
argument sharing between the two VPs, which makes a correct prediction with
respect to a difference between singly- and doubly-prefixed na-verbs.
References
AIKHENVALD, A. V. (2006) Serial verb constructions in typological perspective. In:
AIKHENVALD, A. V. & R. M. V. DIXON (eds) Serial verb constructions. New York:
Oxford University Press.
ARSENIJEVI, B. (2007) A unified analysis of two classes of Slavic verb prefixes. In: BLAHO,
S. et al. (eds) Console XIV Proceedings. Leiden: Console.
BABKO-MALAYA, O. (1999) Zero Morphology: a Study of Aspect, Argument Structure and
Case. Ph.D. thesis, Rutgers University.
BAKER, M. & O. T. STEWART (2002) A serial verb construction without constructions. Ms.
Rutgers University.
BISKUP, P. (2007) P(refixe)s and P(reposition)s. To appear in: Proceedings of the 2nd
Congress of the SLS.
BORIK, O. (2008) Morphology-Semantics Interface: Dealing with Aspect. To appear in: York
Papers in Linguistics.
DIKKEN, M. DEN (2003) When particles wont part Ms. New York University.
FILIP, H. (2005) Measures and indefinites. In: CARLSON, G. N. & J. F. PELLETIER (eds.)
Reference and Quantification. The Partee Effect. Stanford: CSLI.
FILIP, H. (2000) The Quantization Puzzle. In: PUSTEJOVSKY, J. & C. TENNY (eds.) Events as
Grammatical Objects. Stanford: CSLI.
GOLDBERG, A. (1995) Constructions. Chicago: University of Chicago Press.
LARSON, R., M. DEN DIKKEN & P. LUDLOW (2006) Intensional Transitive Verbs and Abstract
Clausal Complementation. Stony Brook Occasional Papers in Linguistics 1: 57-105.
MARUI, F. & R. AUCER (2006a) On the intensional FEEL-LIKE construction in Slovenian: A
case of a phonologically null verb. Natural Language and Linguistic Theory 24: 1093-
1159.
MARUI, F. & R. AUCER (2006b) On the complement of the intensional transitive want.
Stony Brook Occasional Papers in Linguistics 1: 128-151.
MCCAWLEY, J. (1979) On identifying the remains of deceased clauses. In: McCawley, J. D.
Adverbs, vowels, and other objects of wonder. Chicago: University of Chicago Press.
MCNALLY, L. (2009) Existential sentences. To appear in: MAIENBORN, C. et al. (eds.)
Semantics: An International Handbook of Natural Language Meaning. Berlin: De
Gruyter.
PERELTSVAIG, A. (2006) Small Nominals. Natural Language & Linguistic Theory 24: 433-
500.
PIN, C. (1994) Accumulation and aspectuality in Polish. In: Gonzlez, M. (ed.)
Proceedings of the North East Linguistic Society 24 (vol. 2). Amherst, Mass.: GLSA.
RAPPAPORT HOVAV, M. & B. LEVIN (2001) An event-structure account of English resultatives.
Language 77/4: 766-797.
RAMCHAND, G. (2008) Verb meaning and the lexicon: A first phase syntax. Cambridge:
Cambridge University Press.
RIEMSDIJK, H C. VAN. (2002) The Unbearable Lightness of GOing. Journal of Comparative
Germanic Linguistics 5: 143-196.
ROMANOVA, E. (2007) Constructing perfectivity in Russian. PhD thesis, Universitetet i
Troms.
ROSS, J. R. (1979) To have have and not to have have. JAZAYERY, M. A. et al. (eds.) Linguistic
and literary studies in honor of Archibald A. Hill. Lisse: Peter de Ridder Press.
SPENCER, A. & M. ZARETSKAYA (1998) Verb prefixation in Russian as lexical subordination.
Linguistics 36: 1-39.
SVENONIUS, P. (2004) Slavic prefixes inside and outside the VP. Nordlyd 32/2: 205-253.
TATEVOSOV, S. (2007) Measuring individuals, partitioning events. In: KOSTA, P. & L.
SCHRCKS (eds.) Linguistic Investigations into Formal Description of Slavic
Languages. Frankfurt am Main: Peter Lang.
TATEVOSOV, S. (2008) Intermediate prefixes in Russian. In: ANTONENKO, A. et al. (eds.)
Formal Approaches to Slavic Linguistics 16: The Stony Brook Meeting 2007. Ann
Arbor: Michigan Slavic Publications.
TRUSWELL, R. (2007) Extraction from adjuncts and the structure of events. Lingua 117: 1355-
1377.
MUYSKEN, P. & T. VEENSTRA (2006) Serial verbs. In: EVERAERT, M. & H. VAN RIEMSDIJK
(eds.) The Blackwell Companion to Syntax. Volume IV. Malden, Mass.: Blackwell
Publishing.
AUCER, R. (2009) A VP-internal/resultative analysis of 4 VP-external uses of Slavic verbal
prefixes. PhD thesis, University of Ottawa. lingBuzz/000828
AUCER, R. (2010) The reflexive-introducing na- and the distinction between internal and
external Slavic prefixes. In: SMIRNOVA, A. et al. (eds.) Formal Studies in Slavic
Linguistics. Newcastle/Tyne: Cambridge Scholars Publ.
Rok aucer, of Nova Gorica, Institute for Cultural Studies, Vipavska 13, SI-
5000 Nova Gorica, Slovenia, rok.zaucer@p-ng.si
Potsdam Linguistic Investigations
Recherches Linguistique Potsdam
Edited by / Herausgegeben von / Edit par

Peter Kosta, Gerda Haler,
Lilia Schrcks and / und / et Nadine Thielemann
Band 1 Peter Kosta / Lilia Schrcks (eds.): Linguistics Investigations into Formal Description of
Slavic Languages. Contributions of the Sixth European Conference held at Potsdam Uni-
versity, November 30December 02, 2005. 2007.
Band 2 Lilia Schrcks: Binding and Discourse. Where Syntax and Pragmatics Meet. 2008.
Band 3 Christiane Hmmer: Synonymie bei phraseologischen Einheiten. Eine korpusbasierte Un-
tersuchung. 2009.
Band 4 Svetlana Friedrich: Definitheit im Russischen. 2009.
Band 5 Matthias Guttke: Strategien der Persuasion in der schriftkonstituierten politischen Kommu-
nikation. Dargestellt an Parteiprogrammen der Neuen Rechten in Polen. 2010.
Band 6 Peter Kosta / Lilia Schrcks (eds.): Formalization of Grammar in Slavic Languages. Con-
tributions of the Eighth International Conference on Formal Description of Slavic Lan-
guages FDSL VIII 2009. University of Potsdam, December 25, 2009. 2011.
www.peterlang.de
6 Potsdam Linguistic Investigations
This book assembles the contributions of the Eighth European Conference on
Formal Description of Slavic Languages (FDSL VIII) which took place from 2 nd
to 5 th December 2009 at the University of Potsdam. The concern was to bring
together excellent experienced but also young scholars who work in the field of Recherches Linguistiques Potsdam
formal description of Slavic languages. Besides that two workshops on typology
of Slavic languages and on the structure of DP/NP in Slavic were organized.
Edited by Peter Kosta
Herausgegeben von Gerda Haler
Edit par Lilia Schrcks
Nadine Thielemann
Peter Kosta / Lilia Schrcks (eds.) Formalization of Grammar in Slavic Languages

in Slavic Languages
Peter Kosta is Professor of Slavic Linguistics and Chair at the Slavic Department
at Potsdam University. Focus of research: Biolinguistics, Theory of Language,
Comparative Slavic Syntax and the Typology of Languages.
Lilia Schrcks, Ph.D., is since 2005 Associate Professor of Slavic linguistics
at the Institute of Slavic Studies, University of Potsdam. Her fields of research
include Slavic Syntax, Generative Syntax, Binding Theory, DP/NP, Identity
and Language.
PETERLANG
LANG
www.peterlang.de
PLI 06-Kosta-Schurks-261869-HCA5-TP.indd 1 16.05.11 11:15:00 Uhr

(Potsdam Linguistic Investigations _ Potsdamer Linguistische Untersuchungen _ Recherches Linguistiques à Potsdam 6) Peter Kosta, Lilia Schürcks-Formalization of Grammar in Slavic Languages_ Contributi

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Potsdam Linguistic Investigations _ Potsdamer Linguistische Untersuchungen _ Recherches Linguistiques à Potsdam 6) Peter Kosta, Lilia Schürcks-Formalization of Grammar in Slavic Languages_ Contributi

Uploaded by

Copyright:

Available Formats

6 Potsdam Linguistic Investigations

This book assembles the contributions of the Eighth European Conference on

Peter Kosta / Lilia Schrcks (eds.) Formalization of Grammar in Slavic Languages

www.peterlang.de ISBN 978-3-631-61869-1

PLI 06-Kosta-Schurks-261869-HCA5-TP.indd 1 16.05.11 11:15:00 Uhr

University of Potsdam, December 25, 2009

ISBN 978-3-653-01040-4 (eBook)

Die Reihe Potsdam Linguistic Investigations Potsdamer linguistische Untersuchungen

La serie Potsdam Linguistic Investigations Potsdamer linguistische Untersuchungen

I. Phonetics & Phonology 11

ALE BIAN: Structure of Syllables in Czech 13

II. Machine Translation 55

IV. Syntax 109

The Editors Potsdam, April 2011

Phonetics & Phonology

phonemes may be viewed as phonemes in a subsystem representing two or more

(1) occlusive fricative

syllable in Czech. Similarly, the four post-nuclear positions correspond to the

(b) In the pre-nuclear context, the mentioned phonemes must be imme-

4. Structure of consonantal combinations

phonemes may be added, either before or after it or both. A vowel may be

/bn/, /b / or /pm/ in which case phonemes from the archi-position E2 are

MULDER, J. (1968) Sets and Relations in Phonology. Oxford: Clarendon Press.

Ale Bian, Masaryk University, Department of Linguistics and Baltic Lan-

3500Hz, 5500Hz and 7500Hz, relatively equally dispersed in the perceptual

(3) Spectra of the two fricative series from kari 2009:75

alizations of /, , d/ and 35 of //. In this study we found an overwhelming in-

/plae/ *LEX plae *LEX hlae *LEX plae

Ma+gorzata E. avar, University of Zadar, Linguistics Department, Obala kralja

sl_v: slovo, slaviti, (sl

m_r: mrti, moriti, -mariti, -m rD, -mirati

b_d bljusti buditi b

m_r mrti moriti mariti m rD mirati

{non-reduced non-lengthened o-grade} (represented commonly by o,

V{x} = v1R +v2L + v3O

(8a) Example of minimal distance:

(8b) Example of zero distance:

(8c) Example of more than minimal distance:

Such distance between codes is known as Hamming distance10. Hence, any

(9b) 001 101

Ondej efk, Masaryk University,Department of Linguistics and Baltic Lan-

1st person 2nd person 3rd person sg&pl

Czech 143 462 8 18 264 167

Russian 5433 2361 24 334 5102 4131

Next we will demonstrate concrete examples of different types of dropness in

Following is the example of the copula construction for the languages:

(7) Mm koku (cz) -> (ru)

Reflexive suffixes -/- in Russian vs. reflexive particles in Czech

BLEU score is an automatic metric for evaluation of MT. The sentences

Because of the drawbacks of the automatic metrics of evaluation described

Natalia Klyueva, Charles University in Prague, Institute of Formal and Applied

slide passes the unergativity diagnostic (pseudo-passive) (PERMUTTER &

Russian unaccusativity data makes a clean cut of four types of intransitive

(7) Effects of Animacy by Verbs Class

Diagnostics Unaccus Unerg Trans Exist Variable

Gen of Neg yes yes No no no

Locative yes yes no no

feature clusters are derived which directly correspond to theta-roles (REINHART

(13) a. Mnogo travy naroslo v parke Unaccusative

(16) a. *Po mysli probealo v kadoi golove Unergatives

(19) a. po milicioneru stojalo na kadom vyhode

4.2 Gen of Neg, Locative Inversion

(28) a. Na dereve krasnelo jabloko

5. Conclusions and Implications

Ale Bian, Masaryk University, Department of Linguistics and Baltic Lan-

(3) Spectra of the two fricative series from kari 2009:75

alizations of /, , d/ and 35 of //. In this study we found an overwhelming in-

/plae/ LEX plae LEX hlae *LEX plae

m_r: mrti, moriti, -mariti, -mrD, -mirati

m_r mrti moriti mariti mrD mirati

Ondej efk, Masaryk University,Department of Linguistics and Baltic Lan-

(7) Mm koku (cz) -> (ru)

Reflexive suffixes -/- in Russian vs. reflexive particles in Czech

b. Ivani sebja*i/j skazal to Borisj narisoval t

(10) Dal li on komu-to /komu-nibud to-to/to-nibud?

(35) [DP[AP kakoj-to]i [D [NP ti [NP[AP znamenityj ][NP[AP detskij ] [NPvra]]]]]]

(51) OP2x [[[[[ DOCTOR x ] [ x FOR CHILDREN ]] [ FAMOUS x ]]

(6) a. *Kdo pozval na verek a koho pozval na verek?