ArticlePDF Available

A dynamic account of clitic positioning in Cypriot Greek



In this paper, a full account of clitic positioning in Cypriot Greek is attempted within the framework of Dynamic Syntax. Firstly, it is shown that the existing approaches dealing with CG clitic positioning are inadequate to deal with the full range of clitic positioning phenomena as these are described by Pappas (2010) and Chatzikyriakidis (2010). Then it is argued that this complex system can be effectively captured assuming a lexical entry where three generalized parsing strategies, i.e. ways of structure building, function as lexical triggers for parsing CG clitics. Variation in positioning with the non-factive complementizer oti as well as the causal subordinator epidi are accounted for assuming that these elements can be parsed as either subordinators or coordinators. Furthermore, the challenge of providing an account for complex markers/subordinators formed with the coordinator tze is provided, arguing that the unexpected enclisis caused in these cases is due to the fact that these elements provide two separate linked domains where the first acts as the context in which the second is parsed (e.g. a negative context). Lastly, the account proposed will be shown to be grounded in historical considerations as well, arguing that the transition from a descriptively simpler system (that of Medieval Cypriot Greek, MCG) to a more complex one (CG) is only epiphenomenal, showing that the transition from MCG to CG involves simplification of the lexical entry for clitics.
A Dynamic Account of Clitic Positioning in Cypriot Greek
Stergios Chatzikyriakidis
Royal Holloway, University of London; Open University of Cyprus
In this paper, a full account of clitic positioning in Cypriot Greek is attempted within the framework of Dynamic Syntax.
Firstly, it is shown that the existing approaches dealing with CG clitic positioning are inadequate to deal with the full range
of clitic positioning phenomena as these are described by Pappas (2010) and Chatzikyriakidis (2010). Then it is argued
that this complex system can be effectively captured assuming a lexical entry where three generalized parsing strategies, i.e.
ways of structure building, function as lexical triggers for parsing CG clitics. Variation in positioning with the non-factive
complementizer oti as well as the causal subordinator epidi are accounted for assuming that these elements can be parsed as
either subordinators or coordinators. Furthermore, the challenge of providing an account for complex markers/subordinators
formed with the coordinator tze is provided, arguing that the unexpected enclisis caused in these cases is due to the fact that
these elements provide two separate linked domains where the first acts as the context in which the second is parsed (e.g. a
negative context). Lastly, the account proposed will be shown to be grounded in historical considerations as well, arguing
that the transition from a descriptively simpler system (that of Medieval Cypriot Greek, MCG) to a more complex one (CG)
is only epiphenomenal, showing that the transition from MCG to CG involves simplification of the lexical entry for clitics.
1. Introducing the data
Taking a look at the positioning system of CG, the major difference one notes between systems of the Standard Modern Greek
(SMG) type and of the CG type is that in the latter clitic positioning is not defined solely with respect to the form of the verb
(imperative-gerund vs indicative) but a number of non-verbal elements can affect clitic positioning as well. Specifically, CG
exhibits three different positioning environments: a) enclitic b) proclitic and c) variation environments. Clitics in CG are in
general enclitic in indicative and non-indicative contexts:
a. Poli
many anthropi
people kamnun
do.3PL to
it.CL-ACC sosta.
b. *Poli
many anthropi
people to
it.CL-ACC kamnun
do.3PL sosta.
‘Many people people do it right.’
Email address: Chatzikyriakidis)
Preprint submitted to Lingua January 24, 2012
a. Kamne
do.IMP to.
b. *To
it.CL-ACC kamne.
‘Do it.’
This is in contrast to what we find in SMG, where positioning is regulated by the form of the verb: indicative verbs are
associated with proclisis (3) while imperatives and gerunds with enclisis (4 and 5 respectively):
a. *Poli
many anthropi
people kanun
do.3PL to
it.CL-ACC sosta.
b. Poli
many anthropi
people to
it.CL-ACC kanun
do.3PL sosta.
‘Many people people do it right.’ [SMG]
a. Kane
do.IMP to.
b. *To
it.CL-ACC kane.
‘Do it.’
a. Kanontas
do.GER to...
b. *To
it.CL-ACC kanontas...
‘By doing it.’
However proclisis obtains in indicative environments in CG in case a number of elements appear at the left-periphery. These
elements include Wh elements and fronted constituents, NEG, modal/tense particles and subordinating conjunctions. These
proclisis-inducing elements are further subcategorized to those that obligatorily induce proclisis and to those that optionally
do so. In the former category, one finds: a)Wh-elements:
(6) Pios
who.NOM ton
him.CL-ACC ide
saw.3SG (*ton)?
‘Who saw him? ’
b) Modality/Tense/Mood/Negation markers:
(7) En
NEG ton
him.CL-ACC iksero
know.1SG (*ton).
‘I do not know him.
(8) Min
NEG ton
him.CL-ACC dis
see.2SG.PNP (*ton).
‘Do not see him.’
(9) Thelo
want.1SG na
SUBJ ton
him.CL-ACC do
see.1SG (*ton).
‘I want to see him.’
(10) Enna
FUT ton
him.CL-ACC do
see.1SG (*ton)
him.CL-ACC apopse.
‘I’m going to see him tonight.’
b) A number of subordinating conjunctions , notably conditional and temporal ones (11 and 12-13 respectively):
(11) An
if ton
him.CL-ACC dite
saw.2PL (*ton),...
‘If you see him...’
(12) Pu
when ton
him.CL-ACC ida
saw.1SG (*ton),
him.CL-ACC milisa
spoke.1SG tu.
‘When I saw him, I spoke to him.’
(13) Aman
if/when ton
him.CL-ACC ipopsiazete
is-suspicious (*ton),
him.CL-ACC gorazi
buy.3SG tis
her.CL-GEN lulufkja/fjora.
‘If/When she is suspicious of him, he buys her flowers.’
c) The factive complementizer pu:1
(14) Lipume
be-sorry.1SG pu
COMP ton
him.CL-ACC ides
saw.2SG (*ton).
‘I’m sorry that you saw him.’
(15) Thimume
remember.1SG pu
COMP ton
him.CL-ACC ides
saw.2SG (*ton).
‘I remember that you saw him.’
The second category involves elements that induce proclisis optionally. In this category, we firstly find fronted objects, fronted
subjects and fronted adverbs. It is worth noting that in the existing literature, these elements are categorized as belonging
to the first category (see e.g. Agouraki, 2001), i.e. elements that obligatorily trigger proclisis. The reasoning behind this
is based on the claim that fronted elements are always associated with proclisis when these are focused. In that respect,
the generalization in these analyses is that focused elements and not fronted elements trigger proclisis. However, Pappas
(2010) has disputed the claim that focused elements are always associated with proclitic positioning, arguing that focus is
not a prerequisite of proclisis with fronted elements. Following this observation, I believe that the correct description is that
fronted elements sometimes trigger proclisis and sometimes do not. The examples below are illustrative of this variation in
positioning with fronted elements. Examples 16 to 19 seem to show that focus is what triggers proclisis. However, examples
(20) and (21) show that focus does not always induce proclisis and, vice versa, that non-focused fronted elements can also
trigger proclisis (21):
1Note that given the subjunctive is expressed via the proclisis inducing subjunctive marker na,subjunctives are always associated with proclisis in CG.
(16) TUTO
this tu
him.CL-GEN edokes.
gave.1SG him.CL-GEN not the other
‘You gave him this one.’
(17) Tuto
this edokes
him.CL-GEN tu.
gave.1SG him.CL-GEN not the other
‘You gave him this one.’
(18) I
Mary.NOM ton
him.CL-ACC ikseri
know.3SG oi
not i
the.NOM Ioana.
‘Mary knows him, not Ioanna.’
(19) I
the.NOM Maria
Mary.NOM ikseri
him.CL-ACC ton.
‘Mary knows him.’
(20) Etsi/ETSI
that-way ton
him.CL-ACC ida.
‘I saw him like this/It was like this that I saw him.’
(21) Sinexia/SINEXIA
continuously eksetimaze
was-swearing.3SG me.
‘He was continuously swearing at me.’
In Pappas’ (2010) corpus, 15% of fronted coreferential OV constructions are associated with preverbal placement. The
standard assumption in the literature is that these constructions should be illicit given that doubling constructions are topics,
and thus non-focused. This is disputed by Pappas (2010) who gives the following Clitic Left Dislocated (CLLD) example
from his own corpus where proclisis is attested:
(22) Ki
and emena
me.DAT m
me.CL-DAT aresi.
‘I like this.’
It is certainly true that focus plays a role in CG clitic positioning and potentially proclisis is preferred when a fronted element
is focused. However, the data in Pappas (2010) show that this should not be taken as a strict constraint but rather as a
preference. In the approach taken in this paper, Pappas’ (2010)claim as regards the association of focus with proclisis will be
neatly explained by the dynamics of the framework involved, namely the fact that more than one strategy might be possible
for parsing one and the same word. Given that these are parsing strategies that do not involve features like e.g. focus or topic
features, the account predicts that variation with fronted elements should be possible no matter the pragmatic import in each
Besides variation found with fronted elements, which is also found in languages like Portuguese (Crysmann, 1997; Luis &
Otoguro, 2004 inter alia) or other dialects of Modern Greek, like e.g. Cretan Greek, there are a number of environments which
are unambiguous cases of variation no matter the descriptive or theoretical standpoint one takes towards these elements. These
concern subordinating conjunctions and complementizers that freely allow proclisis and enclisis without any interpretational
difference whatsoever. The first such element is the non-factive complementizer oti, which was firstly reported to trigger both
proclisis and enclisis in Revithiadou (2006):2
2One of the reviewers asks whether optionality means preference or chance. This is something that is not yet elucidated. Some speakers indicate that
there is preference towards one of the two options in a specific construction (e.g. a preference for enclisis in the case of factive complementizer oti) while
the rest indicate that there is nopreference for one structure against the other. While this is an important issue, I believe what matters the most in these cases
is that the speakers note that both proclitic and enclitic versions of the samesentence are perfectly licit CG sentences.
a. Ipen
said.3SG oti
COMP ethkiavasen
read.3SG to.
b. Ipen
said.3SG oti
it.CL-ACC ethkiavasen.
‘S/He/It said that s/he read it.’
Pappas (2010) as well as Chatzikyriakidis (2010) independently reported that variation is also possible with subordinates of
cause epiδi and γiati ‘why’, with Pappas further identifying variation with afu ‘given that’ as well:
a. Epidi
because enevriases
made-angry.2SG me,
me.CL-ACC en
NEG kamno
do.1SG tipota.
b. Epidi
because me
me.CL-ACC enevriases,
made-angry.2SG en
NEG kamno
do.1SG tipota.
‘I’m not doing anything because you made me angry.’
a. Giati
because kamni
makes.3SG me
me.CL-ACC tze
and niotho
feel.1SG orea.
b. Giati
because me
me.CL-ACC kamni
makes.3SG tze
and niotho
feel.1SG orea.
‘Because s/he/it makes me feel nice.’
a. Afu
since edimiurgisan
created.3PL tis
her.CL-DAT provlima.
‘Since rthey caused trouble to her.’ [Pappas, 2010: ex.9]
b. Afu
because me
me.CL-ACC kamni
makes.3SG tze
and niotho
feel.1SG orea.
‘Since s/he/it makes me feel nice.’
Lastly, it is worth looking at a number of environments that involve a negation-conjunction or a conjunction-conjunction
complex, the second part of the complex always being the coordinator tze ‘and’. The first part of the complex is always a
proclisis inducing element (the negation marker or a subordinate conjunction) whereas the second part of the complex is the
coordinating conjunction tze. In all these cases, proclisis is ungrammatical with enclisis being the only grammatical option:
a. En
NEG tze
and (*ton)
him.CL-ACC ida
saw.1SG ton.
b. An
if tze
and (*ton)
him.CL-ACC ida
saw.1SG ton.
‘Even though I saw him...’
1.1. Existing analyses
A number of analyses regarding clitic positioning in CG have been put forth by the years, notably Agouraki (1997, 2001),
Terzi (1999a, 1999b), Condoravdi & Kiparsky (2001) and Revithiadou (2006). With the exception of Revithiadou (2006),
all the other accounts are inadequate on independent empirical grounds, since they deal with a rather incomplete set of data
(e.g. no variation environments are discussed). Besides that, a number of technicalities of these accounts are problematic,
like e.g. the motivation for V movementin Terzi (1999a,b) and Agouraki (2001), or the existence of two NegPs in Agouraki
(2001). These approaches are discussed in detail in Revithiadou (2006), Chatzikyriakidis (2010) and Pappas (2010), showing
the problems associated with these accounts in full detail and as such will not be reviewed here again. The interested reader
is directed to Revithiadou (2006), Chatzikyriakidis (2010) and Pappas (2010) for more details on these approaches.
An interesting account of CG is that of Revithiadou (2006). Revithiadou (2006) departs from both ‘all syntax’ and ‘all
phonology’ accounts of clitic positioning and provides an account where phonology has a filtering role on syntax, i.e. there is
no direct intervention of phonology in the syntax like e.g. PF movement, but however phonology provides a number of well-
formedness rules that apply to pairs of syntactic constructions. Then, given such a pair, these rules choose the construction
that scores the less violations and discard the other. Revithiadou (2006) follows Boˇskovi´c (1995, 2001) who argues that
syntax assigns more than one copy of clitics and pre-specified phonological filters act as filters that eventually decide which
of the two copies will be pronounced. Then, Revithiadou (2006) proposes the following prosodic ranking hierarchy for CG,
i.e. a set of ranked well-formedness rules:
(28) Ranking of prosodic structure in CG
FAITH(acc) is a constraint which posits that inherent accent of the input must be preserved in the output (Revithiadou 2006:
87, apud Revithiadou 1999). EXH and NONREC are prosodic domination constraints. The former specifies that no Ci
immediately dominates Cjfor j<i-1 while the latter that no Cidominates Cjfor j=1. The rest of the constraints are alignment
constraints (see Revithiadou, 2006: 85 for the formal details). Revithiadou then proceeds and shows how these constraints
work and give rise to the various patterns in CG. For example, in case only a verb is present, enclisis is the only option
given that in case of proclisis, NONREC will be violated which is ranked higher than the PCON and WCON(R) in the enclitic
case. In cases of proclisis, there are two options for the clitic to be phrased. In case of a stressed function word, the clitic
appears as enclitic to that stressed element since this case incurs less violations than the respective construction where the
clitic encliticizes to the verb. In case of an unstressed function word, the cluster function word+clitic appears proclitic to the
verb, though the actual details are not given in the paper (it seems that both cases will incur a WCON(L) violation, but the
enclitic case will incur two violations of PCON and one of WCON(R) while the proclitic case only one violation of PCON).
The account works fine up to that point. However, the problem arises with variation constructions. Specifically, Revithiadou
argues that variation attested with oti, which is a stressed complementizer, is explained giventhe existence of two alternative
phrasing possibilities. It can be either phrased with the verb of the main clause or alternatively with the verb of the embedded
clause. The two different phrasing possibilities cause the two alternative patterns of clitic positioning. In case oti is phrased
along with the main verb, then enclisis obtains. However, if oti is phrased along with the embedded verb then the clitic is
enclitic to the oti and surfaces preverbally. Revithiadou does not show how these variation data will be captured under her
constraint ranking. But even if we assume that Revithiadou’s account indeed captures variation with oti, there are a number
of facts that would immediately falsify such an account. The first counter-evidence for such an account is constructions that
are the same as the oti constructions, the only difference being a different complementizer, e.g. sentences introduced with
the conditional conjunction ean (which is also stressed). In these constructions variation is not possible, the only possible
construction being proclisis (or preverbal position with the clitic enclitisizing to the subordinate conjunction):
(29) Ipen
said.3SG ean
read.3SG ethkiavasen
him.CL-ACC (*to).
’S/He said/asked if s/he read it.
It is not clear how the above will be handled assuming a phrasing account is used to explain variation with oti. Even worse,
there are constructions in which oti has only one phrasing option, i.e. being phrased with the embedded verb only, and still
both proclisis and enclisis are allowed. Such a situation obtains if the oti clause is left-dislocated, i.e. when oti appears as the
first element of the clause:
a. Oti
COMP ton
him.CL-ACC ikseri,
know.3SG xtes
yesterday mu
me to
it ipe.
b. Oti
COMP kseri
know.3SG ton,
him.CL-ACC xtes
yesterday mu
me to
it ipe.
’Yesterday s/he told me that s/he knows him.
It is not clear how cases like the above will be treated under an account which attributes variation to the different phrasing
possibilities of oti. Furthermore, the proclitic account of elements that bear inherent is also not devoid of problems. Pappas
(2010), in discussing Revithiadou (2006), argues that elements like enke (en tze) that carry inherent stress will be associated
with preverbal positioning in Revithiadou’s system contrary to fact. Furthermore, the phonological data from s-voicing and
e-deletion Revithiadou presents in order to support the claim that preverbal clitics can be either proclitic to the verb or enclitic
to a stressed function word are disputed by my speakers. Revithiadou claims that e-deletion and s-voicing between the verb
and the clitic is not possible when the clitic is enclitic to the function word and not proclitic to the verb. Thus Revithiadou says
that in e.g. enna tus maierepsi ‘S/he will cook for them’ s-voicing does not apply and we thus get [enna tus mairepsi] and
not [*enna tuz mairepsi]. However, my speakers indicated s-voicing to be the case here. Furthermore, as regards e-deletion,
Revithiadou claims that this is obligatory when the function word is unstressed and does not occur when the function word
is stressed, e.g pos to ftkiavase vs pj´
os to efkiavase ‘that s/he read it’ vs ‘who read it .’ My speakers indicated optionality of
e-deletion in both the stressed and the unstressed case, and thus noted that both pos to ftkiavase/pos to efkiavase and pj´
os to
os to fkiavase are possible. If my data are correct, then the basic spine of the account put forth by Revithiadou
(2006) needs to be completely rethought of. In case Revithiadou is correct and my data are just plainly wrong,3then still the
account proposed cannot as already said account for variation environments and predicts proclisis instead of enclisis for en
The task I’m going to take up in this paper, is to argue that a full account of CG can be given once a shift into a dynamic,
parsing-oriented framework has been made. In section 2 an introduction to the framework of Dynamic Syntax is given and in
section 3 a full account of the CG positioning system is given.
1.2. Interim note on data collection methodology
Before I proceed to the main part of this paper, I should make a note on data collection methodology, given that a number of
the data presented in this paper is the result of 3 year data collection that took place from 2007 to 2010 in discontinuous time
intervals. The participants were all native speakers of CG aged between 20-35, that have been raised and lived in Cyprus at
least until the age of 18 (or still live there). All speakers were also speakers of SMG. Data elicitation was done by using pre-
designed questionnaires which contained two different sections. The first section asked the participants to translate a number
of SMG sentences into the equivalent CG ones, and indicate whether more than one structural possibility is possible. The
second section asked the participants to posit grammaticality judgments as regards a number of pre-constructed CG sentences.
It must be noted that CG, despite being the native tongue of possibly most (if not all) Cypriot Greeks in both Cyprus and
diaspora, the official language of the Cypriot Greek state is SMG (Newton, 1972; Tsiplakou 2009, among others), and as
such modern day CG stands in a diglossic relationship to SMG (Tsiplakou, 2009, in press). Tsiplakou (2009) showed that
this diglossic relationship in modern day Cyprus is between SMG and an emergent pancypriot Koin´e. At the same time, all
3Given the uniformity of judgments from all the speakers I consulted in this case, this is something that I doubt.
regional varieties contributing to the formation of this new form of Cypriot, i.e. the Koin´e, are subject to levelling (Tsiplakou,
2009). I do not know at the present whether differences in clitic positioning are found among individual regional varieties
or whether the data gathered as presented in this paper present a clitic positioning system which is influenced by SMG.
Unfortunately, there is no way at present to test this. I believe that the data gathered, besides presenting some novel aspects
on clitic positioning (e.g. variation with subordinates of cause reported in Chatzikyriakidis (2010) and independently by
Pappas (2010), are a fairly accurate description of the CG positioning system, at least in the way it is spoken today by young
CG speakers given that the data selected were extremely uniform across speakers (8) and variation between these was almost
non-existent. It is however wise to keep in mind that this paper will not try to address regional differences with respect to
clitic positioning (if any). Therefore, it should not be taken for granted that the clitic positioning system described in this
thesis is uniform across all regional dialects of idioms of CG (which might be true but at the present it is just impossible to
2. DS preliminaries
The Dynamic Syntax (DS) framework (Kempson et al. 2001; Cann et al. 2005) is a processing oriented framework. The basic
assumption behind DS is that natural language syntax can be seen as the progressive accumulation of transparent semantic
representations with the upper goal being the construction of a logical propositional formula (a formula of type t). Such a
process is driven by means of monotonic tree growth, representing the attempt to model the way information is processed
in a time-linear, incremental, word-to-word manner. DS is a goal driven framework driven by means of requirements. For
example the starting point for every parse, as already mentioned, is a requirement to obtain a propositional formula of type t:
(31) The AXIOM
?T y(t),3
The parsing process is represented via means of binary trees, underpinned by the Logic of Finite Trees (LOFT, Blackburn
& Meyer-Viol, 1994), an expressive modal language that allows one to talk about any treenode from the perspective of any
treenode. LOFT uses two basic modalities, and , corresponding to the daughter and mother relation respectively. Left
nodes are addressed as 0 nodes, whereas right nodes as 1 nodes. Conventionally, nodes on the left correspond to the argument
nodes, i.e. the nodes in which the arguments will be represented, whereas the 1 nodes correspond to the functor nodes, i.e.
the nodes in which all the various types of predicates will be represented. The rootnode, defined as the sole node that does
not have a mother node, is given the treenode address 0. The two basic LOFT modalities (and ) are used in either their
existential or universal version (represented as h↓i and []respectively). The example below illustrates the flexibility of LOFT
by showing a binary tree where different nodes are addressed from the perspective of other nodes using treenode modalities:
(32) The LOFT modalities in action
T n(0),
h↓1iT n(01)
T n(00),
h↑0iT n(0)
T n(01),
h↑1ih↓0iT n(00)
T n(010),
h↑0ih↓1iT n(011)
T n(011),
h↑1iT n(01)
In the above tree, all nodes have a treenode address and a further statement identifying another node in the tree. For example,
the statement h↑0ih↓1iT n(011) found in the 010 node reads as: you will find treenode 011 if you take a step across the 0
mother relation followed by a step across the 1 daughter relation. Furthermore, the two kleene operators * and + are used
in combination with the basic tree modalities, denoting the reflexive transitive and the transitive closure of the modality in
each case. Thus, h↓iT n(x)reads as T n(x)holds at the current node or at a node below the current one (of arbitrary depth),
whereas h↓+iT n(x)reads as T n(x)holds on a node below the current one.
Treenodes are inhabited by labels expressing Type/Formula value information, tree-logic modalities and unfulfilled require-
ments of all the latter sort. The end result of every successful parse of a given natural language string involves a binary tree,
in which all the nodes have type and formula values and no outstanding requirements exist. Such a complete tree depicting
the parse of the intransitive struture O Gianis traguda ‘John sings’ in CG is shown below:4
(33) A complete parse of O Gianis traguda ‘John sings’
T y(t), F o(traguda0(gianis0)(, si, siRR< snow)),3
T y(es)
F o(, si, siRR< snow)
T y(est)
F o(λe.traguda0(gianis0)(e))
T y(e)
F o(gianis0)
T y(e(est))
F o(λx.λe.traguda0(x)(e))
A complete DS tree is a binary tree where every node carries a Formula and a Type (Fo and Ty respectively) value (among
potentially other information). In the above tree all nodes carry formula and type value information. Formula and Type values
combine via functional application and modus ponens respectively. The lower functor node (011 node) represents the node
where the Fo and Ty value of the word/concept of traγuδa is projected. In the same vein, the 010 node represents the node
where the subject is projected whereas the 00 node and the nodes below it represent the situation argument node, i.e. the
locus of tense/aspect information.5The formal details of the information on the situation argument node shall not bother us
here. However, the interested reader is directed to Cann (2011) and Chatzikyriakidis (2010, 2011) for a detailed explanation.
2.1. Lexical entries and Computational rules
Treenode decorations and requirements comprise the declarative part of the DS system. However, DS further employs a
number of procedural mechanisms that project the features of the declarative system. The basic devices behind the DS
procedural system, are lexical and transition/computational rules (henceforth computational rules). The first are language-
specific rules projected from the words of a given language. These are the lexical entries each word is assumed to involve. On
the other hand, computational rules are language-general rules and, as such, are assumed to hold for every language. Lexical
actions in DS are encoded in a simple algorithmic IF THEN ELSE format. The IF part contains a statement(s) that should
be true or false with respect to the partial tree at the time the word comes into parse. If this statement is satisfied, the actions
encoded in the THEN part of the entry are projected. There is an elsewhere statement (ELSE) that might contain alternative
information in case the first triggering point is not satisfied. The examples below present a sample entry for the proper noun
Bill in English:
4The situation argument node (the node with type eshas complex structure which we do not show for ease of exposition. This structure will be made
explicit later on if needed.
5This is based on the assumption that every sentence has an explicit situation argument. See Gregoromichelaki (2006), Cann (2011) and Tenny &
Pustejovsky (2000) for the relevant argumentation.
(34) Lexical entry for the proper noun Bill in English
IF ?T y(e)
THEN put((T y(e), F o(Bill0),[])
ELSE abort
The above example reads as follows: if you are at a node that has a type erequirement, then decorate this node with a type
evalue and a formula value representing the concept ‘Bill’. In any other case abort. In that sense, a proper noun like Bill
in English will be able to get parsed as soon as a node has a requirement for a type e. This will allow a word like Bill to be
parsed either as a subject or as an object in English. Languages with overt NP case marking are assumed to involve further
restrictions, encoded as output filters, that will ensure that an NP marked for structural accusative for example will end up on
the direct object node. Such an entry is shown below:
(35) Lexical entry for an NP marked for structural accusative
IF ?T y(e)
THEN put((T y(e), F o(x0),?h↑0iT y (et),[])
ELSE abort
In the above lexical entry, the accusative marked NP posits a requirement that the predicate node (T y(et)) is found in
case a step across the zero daughter relation is taken from the current node, in effect identifying the NP’s position with the
direct object node. Such an encoding of case has the advantage of not having to posit case features. Case under such an
account is seen as a restriction on the place in the overall configuration at which the term will end up.6Besides projecting
information on nodes already present in the partial tree, lexical entries can also construct nodes. For example, a lexical entry
for a monotransitive verb in a typical pro-drop language like CG will involve unfolding of the whole propositional template.
The verb starting from the type tnode will build the whole propositional template and further decorate the 011 node with a
type and formula value, representing the adicity and the LOT concept of the verb in question. The pointer is then assumed to
be returned to the type trequiring node. This will allow parsing of both VSO and VOS structures, since the pointer can move
to both the subject and the object node by ANTICIPATION as we shall see shortly. The lexical entry for the monotransitive
verb vlepi ‘sees’ in CG is shown below:
(36) Lexical entry for a monotransitive verb in CG7
IF ?T y(t)
THEN make(h↓1i); go(h↓1i); put(?T y(et);
make(h↓1i); go(h↓1i);
put(F o(λx.λy.verb0(x)(y)), T y(ee(t)),[]);
go(h↑1i); make(h↓0i); go(h↓0i); put(?T y(e));
go(h↑0ih↑1i); make(h↓0i); go(h↓0i);
put(T y(e), F o(Ux)), g ofirst(?T y (t))
ELSE abort
The effect of parsing vlepi in tree notation is shown below:
(37) Parsing a monotransitive verb in CG
6If one wants to be accurate, such an account of case functioning as an output filter will not disallow an NP from being parsed in other nodes that do not
satisfy the case restriction. However, if the NP is parsed on these nodes the case filter requirement will never be satisfied and thus the parse will never be
successful. Note that no account of case of determiners or adjectives is going to be given or discussed in this paper.
7The subscript x on the U metavariable represents the restrictions that the subject metavariable will bear. I will not get into the details as regards these
restrictions or subject agreement in general.
?T y(t),3
F o(Ux),?x.F o(x),
T y(e)?T y(et)
?T y(e)λx.λy .F o(verb0(x)(y)),
T y(e(et))
Verbs in CG, just like verbs in SMG, are assumed to build the whole propositional structure and decorate the subject node
with a type evalue and a formula metavariable. This is done in order to capture the subject pro-drop properties of the
language. Update of metavariables into proper Fo values can be done either by the context or by the natural language string
itself. Update by the context is done using the rule of SUBSTITUTION, which substitutes an Fo metavariable by a proper
Fo value from the context in case this Fo value does not appear anywhere in the local domain of the tree under construction.8
On the other hand, update by the natural language string itself is done when the lexical entry of a word provides an Fo value
for the Fo metavariable. Given these two possibilities of metavariable update, the subject can be overt or covert in CG. Note
that such an update from the context is not possible for the object node, since no metavariable and type value are projected
in the object node but only a type requirement. In that sense, overt material is needed in order to fill the object slot, thus no
object pro-drop being possible in CG. There is a great deal of parametric variation with respect to the entries of lexical verbs
in different languages in DS. For example, non pro-drop languages like English will project a requirement for a type ein both
the subject node and object nodes (see Cann et al., 2005: 48), while full pro-drop languages like Japanese or presumably
Latin (on the assumption that Latin is a full pro-drop language) will project a type value and a formula metavariable in both
the subject and the object nodes (Kempson & Kurosawa, forthcoming: 7). Lexical actions are thus the domain of parametric
variation in DS. As such, syntactic differences across languages are assumed to derive from different lexical specifications,
i.e. different lexicons.
The algorithmic format of lexical entries can be of arbitrary complexity. Embedded IF THEN schemas can appear within the
initial IF THEN schema and so on. A lexical entry with the structure shown below is perfectly legitimate:
(38) An embedded IF THEN lexical entry format
IF ...
THEN ...
ELSE abort
ELSE abort
A number of partial recursions on specific parts of the algorithm may be used. For example, an entry containing a disjunctive
IF part will be denoted by separating the two disjunctive parts with the symbol |:
(39) An embedded IF THEN with a disjunctive IF part
8See Cann et al., 2005: chapter 2 for the exact definition of SUBSTITUTION.
IF ...
THEN IF ...|
THEN ...
ELSE abort
ELSE abort
Computational rules are general computational devices, comprising the basic tree construction mechanism. Formally, com-
putational rules involve an input and an output description. The former designates where the pointer must be along with
information about the node that the pointer is on as well as information about other nodes with respect to the pointer node,
while the latter shows the transformation of the input in terms of requirements, adding nodes, pointer movement and so on.
Computational rules are assumed to be a closed set of rules universally available to every language. This closed set of rules
are basically rules that help the parsing process unfold. These involve pointer movement rules, rules that perform functional
application and modus ponens (for formulas and types respectively) or rules that get rid of requirements as soon as these are
satisfied. A characteristic case of a computational rule is the rule of ELIMINATION. This rule basically performs functional
application on formulas and modus ponens on types, in case both the argument and the function nodes bear complete formula
and type value:9
?T y(t),h↓0iT y(e)
F o(Ux),
?x.F o(x),
T y(e)
?T y(et),3
T y(e),
F o(N P 0)
F o(λx.λy.verb0(x)(y)),
T y(e(et))
7→ ?T y(t),h↓0iT y(e)
F o(U),
?x.F o(x),
T y(e)
T y(et),h↓0iT y(e),
F o(λy.verb0(N P 0)(y)),3
T y(e)
F o(N P 0)
F o(λx.λy.verb0
T y(e(et))
Additional rules involve ANTICIPATION, which moves the pointer to a daughter node in case an unfulfilled requirement
exists, COMPLETION, which moves the pointer to the mother node in case a type requirement is satisfied in the daughter
node and THINNING, which eliminates a requirement as soon this is satisfied. The rules in their formal format are shown in
the appendix.
2.2. Structural underspecification - Unfixed nodes
2.2.1. Structural Underspecification - The ADJUNCTION Rules
A central concept in DS is the assumption that natural languages are to a large extent underspecified regarding both content
and structure. And while content underspecification has been largely employed within the formal semantics literature of the
9We show the effect of these rules in tree notation. See appendix A for the actual rules.
past 30 years, no attempts to move underspecification into the area of syntax have been made.10 We have already seen how DS
deals with semantic underspecification by employing metavariables as content placeholders. Structural underspecification on
the other hand, is encoded in DS using a family of computational rules, the ADJUNCTION rules. The function of these rules
is to introduce unfixed nodes, i.e. nodes that have not found their fixed position in the tree at the time of their introduction.
The first of the ADJUNCTION rules we are going to see is the rule of *ADJUNCTION. According to this rule, an unfixed
node marked with a type erequirement is projected from a type trequiring node. The effect of the above rule in tree notation
is shown below:
(41) The effect of the *ADJUNCTION rule
T n(n),?T y(t)
h↑iT n(n),
?T y(e),
?x.T n(x),3
The pointer is left at the type erequiring node. The node is structurally underspecified since it does not carry a fixed treenode
address. The only thing the node “knows” as regards its treenode position is that somewhere up above or at the current node,
Tn(n) must be found.11 The pointer is left at the type erequiring unfixed node. At that point, given that a lexical entry for an
NP will have a type erequiring trigger, it can be parsed on that unfixed node. This will be on a node whose structural position
in the tree is not yet known. The *ADJUNCTION rule works neatly for left dislocated structures like OV focus structures.12
In these cases, the preposed object is assumed to be parsed on an unfixed node:
(42) Parsing tuto to vivlio ‘this book’ in tuto to vivlio agorasa ‘This is the book I bought’ on an unfixed node13
?T y(t), T n(n)
h↑iT n(n),
T y(e), F o(viv lio0),
?x.T n(x),3
The pointer moves up to the type trequiring node via COMPLETION and the verb can now be parsed. The structure we get
after parsing the verb, given the entry for monotransitive verbs shown in (36), is the following:
(43) Parsing agorasa ‘hit’ in tuto to vivlio agorasa ‘This is the book I bought’
10With the exception possibly being the notion offunctional uncertainty, formalized within Lexical-Functional Grammar (Bresnan, 2001).
11Notice that the reflexive satisfaction of the * in which the node unifies with its host trivially is not possible. Tn(n) is the treenode address of the type t
requiring node. In that sense, the only way such a reflexive satisfaction will hold is in case the unfixed node unifies with the type t requiring node. However,
such unification is impossible, given the incompatible specifications of the respective type requirements (?e and ?t).
12Even though object FOCUS is not an option in CG in general butonly with specific NPs like for example, deictic elements (see Tsiplakou et al., 2007).
13I ignore determiners for the moment.
?T y(t), T n(n),3
h↑ ∗iT n(n),
T y(e), F o(viv lio0),
?h↑0iT y(t),?x.T n(x)
F o(USpeaker0),?x.F o(x),
T y(e)?T y(et)
?T y(e)F o(λx.λy.agorasa0(x)(y)),
T y(e(et))
In the above structure there is an unfixed node with a type and a formula value and an open slot decorated with a type e
requirement (the 010 node). It is at that point that a process of unification between the unfixed node and the fixed object node
(010) can take place using MERGE. MERGE is a computational rule which unifies two nodes just in case one of the two
updates the treenode address of the other.14 The notion of update is defined by treenode address entailment. If a treenode
address entails another treenode address, then the former can be seen as an update of the latter. In that sense an underspecified
address like h↓iTn(a) can be updated to a more specified address like h↓iTn(a) but not vice versa. Furthermore, the two nodes
must not bear any conflicting specifications. In example (43), the treenode address of the direct object node (010) can be a
proper update of the underspecified address the unfixed node carries. Furthermore, the fact that the 010 node has a proper
treenode address will eliminate the requirement of the unfixed node that a fixed treenode address should be found (?xT n(x)).
Note that no conflicting specifications exist, since the fixed node (DU’) has a type erequirement and the unfixed node (DU)
a type evalue. Given that a type evalue is an update of the type erequirement, no conflict arises. The trees below display the
tree structure before and after MERGE has applied:15
(44) Before and after MERGE of the unfixed node
F o(vivlio0),
T y(e),
?x.T n(x)
h↑i?T y(t)
F o(USpeaker0),?x.F o(x),
T y(e)?T y(et)
?T y(e),3F o(agorasa0),
T y(e(et))
?T y(t)
F o(USpeaker0),?x.F o(x),
T y(e)?T y(et)
F o(vivlio0), T y (e),3F o(λx.λy.agorasa0(x)(y)),
T y(e(et))
14See Appendix A for the actual rule.
15The steps of THINNING, eliminating the type erequirement as well as the requirement ?xT n(x)are not shown.
Notice that the rule of *ADJUNCTION will also work for cases where the left dislocated element is embedded in a comple-
ment clause. The modality associated with the *ADJUNCTION rule makes MERGE of the unfixed node with a type enode
that is deeply nested possible. The only requirement is that this node must be found within the same tree structure. There are
a number of other ADJUNCTION rules but I will not present them here for reasons of both relevance and space. However,
the interested reader is directed to Cann et al. (2005) and Chatzikyriakidis (2010) for more details on the ADJUNCTION
family of rules.
2.3. Parsing in Context - LINK Structures
Besides the tree structures in which each sentence involvesa single tree (regardless of tree embedding), DS also makes use of
pairs of trees which are linked to each other via a relation called LINK. LINK structures involve two separate tree structures
linked by means of an arrow relation (LINK), that share in most of the cases a term. The node from which the LINK starts can
be seen as setting the context in which the LINKed tree is going to be parsed. Examples of LINK relations include relative
clauses, in which case the relative clause is parsed within the context of the head or HTLD constructions in which case the
HTLD sentence is parsed within the context of having parsed the left-dislocated element first. LINK structures have a variety
of uses in DS. Let us illustrate LINK structures in more detail by looking at an HTLD example. In order to analyze HTLD
constructions, Cann et al. (2005) define two rules which link a type enode where the dislocated element is parsed to a type
t requiring node, where the rest of the HTLD structure is parsed. The first rule introduces a LINK transition from a type
erequiring node to a type trequiring node, while leaving the pointer in the first of the two. The second rule introduces a
requirement for a shared term as soon as the dislocated element is parsed. In parsing an HTLD sentence like the one shown in
(46), we apply the first rule (TOPIC STRUCTURE INTRODUCTION, see Appendix A) that introduces the type erequiring
node. The NP is parsed on that node and then the second rule (TOPIC STRUCTURE REQUIREMENT, see Appendix A)
takes effect moving the pointer to the type trequiring node and positing (on the same node) a requirement that a copy of the
formula found in the node where the LINK begins must be found somewhere in the LINKed tree or to a tree LINked to the
LINKed tree (?hDiF o(Giorgos0)):
(46) O
the.NOM Giorgos,
George.NOM gnorizo
know.1SG ton.
‘I know George’
(47) hLiT n(0),
F o(giorg os0),[]>,
T y(e)
hL1iT n(n),?T y(t),?hDiF o(giorgos0)
3. A DS analysis
3.1. Preliminaries on clitics in DS
Being a lexicalized framework,DS assumes parametric variation in naturallanguages to reside in the lexicon. Therefore, one
could plausibly point out that DS, like any other syntactic framework, is vulnerable to the affix-word debate, since a decision
with respect to the status of elements like clitics will have further consequences on whether such elements will have their own
lexical entry or rather be part of the lexical entry of a larger unit (the verbal complex in this case). However, this assumption
is only superficially correct since in DS every element, no matter its traditional characterization (as a word or affix), can have
its own lexical entry. The only requirement is that this element provides distinct procedural information on how the parse
should/must or should not/must not proceed. This might sound controversial but becomes fairly straightforward assuming
that for DS lexical entries are just pieces of information on how the parsing process proceeds. In that respect, every element
providing such information can have its own lexical entry in DS. A classic illustration of that is the analysis of Japanese case
suffixes put forth by Cann et al. (2005). In this analysis, Japanese case suffixes involve a separate lexical entry and are not part
of the entry for the NP preceding these suffixes. Their role is to fix the locally unfixed node the preceding NP is parsed on, in
order for multiple scrambling to be allowed (see Cann et al., 2005: 236-240 for details).16 Given this view on lexical entries,
the affix-word debate is largely irrelevant to DS, since any DS account can be compatible with any pre-theoretical decision
on the affix-word debate. It is rather uncontroversial under this view that clitics will provide distinct procedural infromation
with respect to the parsing process, thus involving their own lexical entries. This is the stance I am going to take in this paper,
following earlier approaches within the same framework notably Bouzouita (2008a,b), Cann et al. (2005), Chatzikyriakidis
(2009a,b, 2010, 2011), Cann & Kempson (2008) and Gregoromichelaki (2010) among others.
3.2. The enclitic environments
I will begin our discussion on CG clitic positioning by using the 3rd person accusative clitic to ‘it’ as the role model. One
of the issues that need to be resolved as regards this type of clitic is the actions projected via its lexical entry. Following
Chatzikyriakidis (2010), I take 3rd person accusative clitics to project fixed structure. This would basicaly mean that 3rd
person accusative clitics are always interpreted as direct objects. This seems to be correct for CG, since no double accusative
constructions exist in CG. In that sense in parsing a 3rd person clitic in CG, the actions found in the lexical entry for the clitic
build the 01,011 and 0110 (the direct object node) nodes and decorate the latter node with a type value (type e) and a formula
metavariable along with a proper formula value to be found later on and return the pointer to the initial type t node. The effect
of parsing a 3rd person accusative clitic is shown below:
(48) The effect of parsing a 3rd person accusative clitic
?T y(t),3
?T y(x)
?T y(x)
F o(Ux),
T y(e),
?x.F o(x)
With the actions projected by the clitic in line, the next step is to see how enclitic positioning is going to be captured.
Hopefully, there is a straightforward way to formalize the entry in order for enclitic positioning to be captured. What we need
to do is provide a triggering point which will basically abort in case a functor node bears a type value, i.e. in case any kind of
verb has already been parsed. The lexical entry for CG clitics capturing enclitic positioning only is shown below:
(49) Lexical entry capturing enclitic positioning in CG:17
16This is due to the “no more than one unfixed nodeat a time constraint”, put forth in Cann & Kempson, 2008 and Chatzikyriakidis & Kempson, 2010.
17The bold parentheses indicate optionality of application depending on the assumption one makes as regards the presence or not of the higher situation
node. In the absence of these higher nodes, application of the make/go functions inside the bold parentheses is not performed. On the contrary, in the
presence of the higher situation nodes, application of the make/go functions proceeds normally.
IF ?T y(t)
1iT y(x)
THEN (make((h↓1i); go(h↓1i)); )
make(h↓1i); go(h↓1i);
make(h↓0i); go(h↓0i)
put(T y(e), F o(Ux),?x.F o(x);
gof irst(?T y(t)))
ELSE abort
ELSE abort
The above entry will correctly predict sentences like (1a.) and (2a.) to be grammatical according to fact. Assuming that a
verb has been parsed, a verbal type will be projected in a functor node. Which functor node that will be depends on the verb’s
adicity. For example parsing of a transitive verb will result in the projection of a verbal type in the 011 node:
(50) Parsing a transitive verb (the higher situation nodes are omitted for ease of exposition)
?T y(t),3
T y(e),
F o(Ux),?x.F o(x)?T y(et)
?T y(e)T y(e(et)),
F o(x0)
At this point the clitic comes into parse. Both its triggers are satisfied, since the pointer is at the type trequiring node and
a type value exists in one of the functor nodes, and as such the parsing process can continue.18 The same mechanism is at
play in parsing imperative verbs. Imperative verbs will also project a type value in a functor node. In that sense, enclitic
positioning with imperatives is predicted in exactly the same way as with indicatives. Of course, such an entry will not suffice
to capture the whole range of positioning since proclitic positioning is still left unaccounted for.19 The next step is to look at
the way proclitic positioning is going to be captured.
3.3. The Proclitic Environments
The set of proclitic triggers found in CG is anything but homogeneous, as it consists of a number of differently functioning el-
ements: modality/tense markers, wh-elements, fronted constituents and subordinating conjunctions among others. However,
a number of generalizations can be captured by looking into the nature of these triggers more carefully.
18The meticulous reader should have noted that the entry for the clitic has a mechanism for building nodes (make). However, in the case we are discussing,
these nodes are already there by means of parsing the verb. The nodes built from the entry of the clitic will basically collapse with the already existing nodes,
given that they do not carry any incompatible descriptions. This is a general feature ofthe underlying tree description language according to which any two
nodes with compatible descriptions can harmlessly coalesce. An anonymous reviewer is thanked for pointing this out to me.
19Note that such an entry will suffice for systems like Pontic Greek (PG),where clitics appear postverbally in all environments (Drettas, 1997; Chatzikyr-
iakidis, 2010 among others).
3.3.1. Unfixed Nodes as Proclitic Triggers
In the introduction to the DS framework, we have discussed the use of structural underspcification via means of the AD-
JUNCTION rules. Even though, there are a number of variants of the ADJUNCTION rule we only presented one variant,
*ADJUNCTION. Surprisingly *ADJUNCTION can be used in order to capture a generalization across elements triggering
proclisis in CG. Bouzouita (2008a,b) in discussing clitic positioning in MedSp, argued that one of proclitic triggers in MedSp
is the existence of an unfixed node. This was based on standard assumptions in DS, according to which Wh-elements and
fronted constituents are or can be parsed on an unfixed node (Kempson et al., 2001; Cann et al., 2005). This assumption
extends straightforwardly to CG, since the similarity of clitic positioning in the two systems is striking. In that sense, the
first generalization that can be drawn is that the existence of an unfixed node constitutes a trigger for proclisis. Note that this
proclitic trigger does not refer to the actual content of the elements triggering proclisis but rather to the parsing strategy used
(unfixed node). But let us see in detail the basis of this intuition. Let us start with Wh-elements:
(51) Pios
who.NOM ton
him.CL-ACC ikseri
knows.3SG ton
the.ACC Giorko?
‘Who knows George?’
Following Kempson et al. (2001), I take wh-elements to decorate an unfixed node, marking the initial node with a Q feature, its
role being the identification of the wh-element as an interrogative wh-element. The effect is that the unfixed node is decorated
with a type value and a specialized formula metavariable WH. The entry below is the lexical entry for pios ‘who.NOM’ in
(52) Lexical entry for pios in CG
IF ?T y(e)
THEN IF h↑i?T y (t),?x.T n(x)
THEN go(h↑i); put(Cat(Q));
go(h↓i); put(WH, T y(e),
?h↑0iT y(t),[])
ELSE abort
ELSE abort
Notice the existence of the case filter ?h↑0iT y(t), identifying the unfixed node as immediately dominated by a type tnode,
i.e. as being the subject node (in effect a nominative case filter).
In the same vein, fronted constituents can be modelled as decorating unfixed nodes. A fronted NP at the left periphery can
be parsed on an unfixed node, finding its position in the tree later on in the parse. The same reasoning is used for fronted
adverbs or any other fronted element. Within this line of reasoning, we can formulate a trigger which will allow the clitic to
be parsed in case an unfixed node is present in the tree structure. The entry below captures enclitic positioning plus proclitic
positioning with elements parsed on an unfixed node:
(53) Lexical entry containing the enclitic and the unfixed node trigger:
IF ?T y(t)
1iT y(x)|
h↓iT y(x),?x.T n(x)
THEN (make((h↓1i); go(h↓1i)); )
make(h↓1i); go(h↓1i);
make(h↓0i); go(h↓0i)
put(T y(e), F o(Ux),?x.F o(x);
gof irst(?T y(t)))
ELSE abort
ELSE abort
Such a trigger will ensure that the clitic will appear before the verb in case an unfixed node is present in the tree structure. Note
that the specification is not restricted to unfixed nodes that are type specific. Actually, the specification proposed (following
the formulation found in Bouzouita, 2008a) does not refer to the type value of the unfixed node or to its formula value. Such
a specification will allow us to account for cases where a strong pronoun is parsed on an unfixed node, in which case a
metavariable instead of a proper formula value will have been projected and proclisis obtains:
(54) Ego
him.GEN edoka
gave.1SG lefta.
‘It was me that I gave him money.
However, given the nature of inclusive disjunction, a clitic can be parsed in case both of its triggers are satisfied, i.e. in
case both a verbal type and an unfixed node exist. Such a fact will fallaciously predict that variation is possible with e.g.
Wh-elements. In that sense, we will have to encode an exclusive version of disjunction according to which the parse cannot
proceed in case both triggering points are satisfied. The revised lexical entry is shown below (exclusive disjunction is noted
as ||):
(55) Updated lexical entry containing the enclitic and the unfixed node trigger:
IF ?T y(t)
1iT y(x)||
h↓iT y(x),?x.T n(x)
THEN (make((h↓1i); go(h↓1i)); )
make(h↓1i); go(h↓1i);
make(h↓0i); go(h↓0i)
put(T y(e), F o(Ux),?x.F o(x);
gof irst(?T y(t)))
ELSE abort
ELSE abort
Note that the use of unfixed nodes for the parse of fronted constituents is one of the options of parsing these elements in DS
and as such will not predict categorical proclisis with fronted constituents but on the contrary will predict variation in this
case. This is indeed the case with fronted subjects, adverbs and PPs, and has been shown to be the case for CLLD as well
(Pappas, 2010). Such a situation is easily predicted given the availability of different processing strategies for the one and
the same element, given that fronted elements can be parsed by making use of more than one alternative parsing strategies
(unfixed node and LINK), and given that only the unfixed node strategy is associated with proclisis, we expect variation to
be possible. Furthermore, the assumption that non-focused fronted elements can also trigger proclisis is also predicted given
such an account, since the argument in the account given is that proclisis is associated with a processing strategy, i.e. the
unfixed node strategy, and not with a some kind of pragmatic focus feature or a specific strategy for parsing focus structures.
The fact that focused elements and proclisis-inducing may to a largeextent overlap is a wholly differentissue.20
3.4. Subordinating Conjunctions, Modality/Tense marker, Negation
As already mentioned, subordinating conjunctions, with the exception of causal subordinators, trigger proclisis in CG. For
these specific both proclisis and enclisis are possible. The crucial question to ask here is what is the general characteristic that
links all these conjunctions. In order to answer this, we need to see how subordination is analyzed in DS. Gregoromichelaki
20Note that the availability of different parsing strategies is one of the main assumptions made in DS and is argued to be one of the reasons of diachronic
change in both Bouzouita (2008a) and Chatzikyriakidis (2010). The argument there is that this availability of alternative parsing options may lead to
production/parsing mismatches that in effect can lead to syntactic change (see Bouzouita, 2008a and Chatzikyriakidis, 2010: chapter 8). See also section 4
in this paper.
(2006), in analyzing conditionals in DS, assumes that subordination includes pairs of LINKed trees sharing a situation ar-
gument. For example, in conditional structures, the IF clause is assumed to be linked with the THEN clause via means of
a LINK relation. The LINK relation is built via the ACTIONS of the conditional conjunction if. The lexical entry for if,
links two situation argproclisis/enclisisument nodes with the requirement that these two are shared. The assumption is that
the situation argument node of the subordinate will act as the context in which the main sentence (or the THEN clause in
conditionals) is going to be parsed. For example, in the case of conditionals the context in which the THEN clause is going
to be true is the set of all situations of the type depicted in the IF clause.21 In a similar way, Chatzikyriakidis (2010), argues
that subordinating conjunctions LINK two independent trees and project a requirement for a situation argument node in the
LINKed tree, i.e. the tree where the subordinate clause is going to be parsed. A way to look at this requirement, is to think
of it as an action for the establishment of a new propositional domain. Parsing one of the subordinate conjunctions signals a
new propositional domain and a way to signal this domain is by projecting a requirement for a situation argument node, i.e.
the need for a new situation/event. There is a lot of formal detail behind this intuition, but this will not bother us here for
reasons of relevance, since a DS internal discussion is highly irrelevant for the needs of this paper. However, the full formal
details of such a proposal can be found in Chatzikyriakidis (2010). What is needed for the needs of the present analysis is the
assumption that conditional/temporal/cause22 and in general subordinating conjunctions build the internal type esrequiring
node of the LINKed tree and leave the pointer at the most local type tnode. Given this property of subordinating conjunctions
to build the type esrequiring node, we can assume a lexical trigger which proceeds in case a type esrequirement exists in
the immediately dominated argument daughter. This lexical trigger is again an item neutral trigger that refers to a parsing
strategy rather than to a specific lexical item. Given the existence of a situation argument node requirement, the clitic can be
parsed. The lexical entry with the added trigger is shown below:
(56) Lexical entry containing the enclitic, the unfixed and the type esrequiring node trigger:
IF ?T y(t)
1iT y(x)||
h↓iT y(x),?x.T n(x)|
h↓0i?T y(es)
THEN (make((h↓1i); go(h↓1i)); )
make(h↓1i); go(h↓1i);
make(h↓0i); go(h↓0i)
put(T y(e), F o(Ux),?x.F o(x);
gof irst(?T y(t)))
ELSE abort
ELSE abort
The type esrequiring trigger correctly captures proclisis found with subordinating conjunctions. Surprisingly, the same line
of reasoning will turn out to be extremely relevant for modality/tense markers as well. Let us explain by reverting to the
simplest type of English example: John sang. In DS, tense and aspect properties are projected inside a complex situation
argument node. This node, following standard DS assumptions with respect to NP structure, is assumed to involve complex
structure. In the example below, illustrating the complete parse of the sentence John sang, tense properties are introduced in
the lower functor node (noted as [4] in the tree), specifying that the event described (the singing event) took place in the past:
(57) Parsing John sang
21See Gregoromichelaki, 2006 on how the quantification over situations is achieved in this case.
22Variation presented with this subordinator will be dealt with later on.
T y(t), F o(sing0(john0)(, si, siRR< snow)),3
T y(es)[1]
F o(, si, siRR< snow)
T y(cns)
F o(si, siRR< snow)
T y(es)[2]
F o(si)
T y(escns)
F o(λ.e0(e0, e0RR< snow))
T y(es), F o(R)[3] T y (es(escns))[4]
F o(λ.eλ.e0(e0, e0ee < snow))
T y(cnses)
F o(λP.(, P ))
T y(est)
F o(λy.sing0)(e)John0)])
T y(e)
F o(john0)
T y(e(est))
F o(λx.λe.sing0(x)(e))
Given this analysis of tense and aspect, let us think how the lexical entries for modality/tense markers are going to look like.
Without going into the exact details of what is the exact contribution of these elements in terms of tense or aspect, we can
safely assume that all these elements will project this kind of information in the internal functor node (the node marked [4]).
This means that all these elements will have to build this node as well, given that they appear before the verb and giventhat
the internal functor node will not have been constructed by the time they will come into parse. This last fact has the further
consequence that the higher argument node (noted with [1] in the tree structure) has to be built as well, decorated with an es
requirement.23 The final consequence of all these assumptions is that when any of these elements is parsed, and assuming
that the pointer is returned to the initial type trequiring node after parsing any of these elements, the clitic can be parsed
given the trigger shown in (55), i.e. in the presence of a type esrequirement. As already mentioned, the exact details of the
lexical entries of these elements are not going to be fleshed out here, since whatever these details will be, the clitic positioning
analysis can effectively stay the same. The relevant part common to all these entries is the construction of the es[1] node and
its decoration with a type esrequirement.24 Given these assumptions, proclisis is predicted to obtain in case any modal/tense
marker is parsed.
3.5. The Complementizer pu
The factive complementizer pu is another environment where proclisis arises in CG.25 The example below exemplifies the
relevant facts:
23Note that this is requirement for the type and not the actual type itself. When the verb is going to be parsed the rest of the information needed to compile
the [1] node into a node with a complete type will be provided. The stadnard assumption is that when a verb, let us say an indicative verb is parsed, it
provides all the necessary information to compile the situation argument node. Given this, after the verb has been parsed and having acquired a type in the
situation argument node, the requirement for this node iseliminated and thus cannot function as a trigger anymore.
24The interested reader is however directed to Giannakidou (2007) for a discussion involving the formal semantics of the subjunctive marker na plus the
SMG future/mood particle θa.
25Terzi (1999a: fn 19) mentions that some of the Cypriot speakers consulted indicated that enclisis with pu is also possible. I got the same judgments from
one of the speakers consulted. All the other speakers indicated that proclisis is the only option. In this paper, I will assume that clitic placement is strictly
proclitic with the factive complementizer pu. It will become evident when the analysis for the non-factive complementizer oti will be given, how such an
account can be expanded to cover variation with pu, in case such variation is proven to exist in CG. See the relevant discussion as regards variation with oti.
(58) Lipume
be-sorry.1SG pu
COMP ton
him.CL-ACC ides
saw.2SG (*ton).
‘I’m sorry that you saw him.’
Proclisis with pu will be captured in the same sense proclisis with subordinating conjunctions is.26 Under this analysis, parsing
of pu will build the situation argument node decorating it with a requirementfor a estype. The difference with subordinating
conjunctions will lie in the fact that this requirement will be built in the main tree and not in a LINKed tree as it is the case with
subordinating conjunctions. Such an analysis will unify the analysis of pu with the one given for subordinating conjunctions.
Evidence that this account is on the correct track, is the fact that pu as well as subordinating conjunctions cannot be followed
by any of the subjunctive markers na or as. With the trigger of these markers being the non-existence of a situation node,
parsing of pu will immediately block further parsing of any of these two markers according to fact:
(59) *Lipame
be-sorry.1SG pu
COMP na/as
SUBJ ton
him.CL-ACC dis.
‘I’m sorry that you saw him.’
Under this analysis, the actions of the complementizer pu will involve the construction of the situation argument node along
with its decoration with a requirement for a estype:27
(60) Sample lexical entry for pu
IF ?T y(t),h↑i>
THEN make(h↓0i); go(h↓0i); put(?T y(es));
...gof irst(?T y(t))
ELSE abort
The above entry correctly captures proclisis with pu.
3.6. Negation
The last proclitic environment we are going to look at concerns cases involving the negation particles en/min, ‘not’. As
already said, proclisis obtains in the presence of these elements as exemplified below:
(61) En
NEG ton
him.CL-ACC iksero
know.1SG (*ton).
‘I do not know him.
(62) Min
NEG ton
him.CL-ACC dis
see.2SG.PNP (*ton).
‘Do not see him.’
In Bouzouita (2008a,b), proclisis found in negative environments is accounted for by using a negative feature [+NEG]. This
feature is assumed to be projected by the negation marker in the type trequiring node. However, there is a way to avoid the
[+NEG] trigger and still be able to capture negation. There are two types of negation in CG, one used in indicative and one
in subjunctive or negative imperative environments:
26See Chatzikyriakidis, 2010 for an alternative DS proposal based on Roussou’s (2007) proposal that factive pu is the same as relative pu.
27The statement h↑i> says that in order for pu to be parsed, structure must exist above this node. This captures the fact that pu is parsed in an embedded
type t node (in other words it introduces a verbal complement) and not onthe initial type t requiring node.
(63) En/*min
NEG ton
him.CL-ACC ida.
‘I did not see him.’
(64) Thelo
want.1SG na
SUBJ min/*en
NEG pais.
‘I want you not to go/ I do not want you to go.’
(65) Na
SUBJ min
NEG pais.
‘Do not go.’
Furthermore, the negation marker min when combined with the Perfective Non Past (PNP) verbal form can give rise to a
negative imperative reading. In that sense, sentence (65) is also possible without the subjunctive marker na as witness the
example below:
(66) Min
NEG pais.
‘Do not go.’
Thus, it seems that the different types of negation interact with modality/tense/aspect in different ways. Giving an analysis
of negation in CG might then involve the contribution of a number of restrictions as regards modality or aspect by the two
different negation markers that will capture the different semantics in each case as well as the combinatorial properties of the
two. Of course, a study of what these restrictions actually are and how to formally encode them is well beyond the scope of
this paper. However, given that modality/tense/aspect information is encoded inside the complex situation argument node,
it is quite straightforward to assume that the negation markers will project their restrictions inside this node. Whatever the
exact locus of the projection of this information (presumably the lower functor node), negation markers will have to built the
higher type esnode. In that respect, the lexical entry for negation markers will have to project at least a [+NEG] feature as
well as the type esrequiring node. The lexical entry is shown below:
(67) Lexical entry for negation
IF ?T y(t)
THEN make(h↓0i); go(h↓0i);
put(?T y(es)), ...;
gof irst(?T y(t)),put([+N EG]).
ELSE abort
Given the above entry, we do not have to refer to the [+NEG] feature to capture negation, since proclisis with negation will
be captured by the same mechanisms used for modality/tense markers as well as subordinating conjunctions, namely the type
esrequiring trigger. In that sense, the entry for the clitic stays the same as in (56).
With a third trigger added, we will have to see its interaction with the already presented triggers (the unfixed node and
the enclitic trigger). The first thing we need to look at is the interaction of the type esrequiring trigger with the other
proclitic trigger, namely the unfixed node trigger, i.e. whether constructions where both triggers are present are allowed
in CG or not. Such structures are commonly found in CG and can involve a number of different structures, e.g. a fronted
object+negation+clitic, a subordinating conjunction+fronted object+clitic, a wh+future particle+clitic and a number of similar
constructions. In these cases, both triggers are satisfied, so the interpretation of the disjunction must be inclusive. An exclusive
interpretation of disjunction will undergenerate, since it will rule out examples like the ones below:
(68) An
if en
NEG ton
him.CL-ACC iksero...
‘If I do not know him...’
(69) Pios
who enna
FUT ton
him.CL-ACC piasi
catch tilefono?
‘Who is going to call him?’
In the above example both an unfixed node and a type esrequiring daughter node will exist when the clitic comes into parse.
Assuming an exclusive interpretation of the disjunction between the unfixed node trigger and the type esrequiring node
trigger, examples like the above would be ruled out contrary to fact. Thus, an inclusive interpretation of disjunction is needed
here. On the contrary, the interaction of the type esrequiring trigger or/and the unfixed node trigger with the enclitic trigger is
of the same type as the one between the unfixed node trigger and the enclitic trigger, i.e. the case where both a type estrigger
and the enclitic trigger are satisfied will vastly overgenerate, since it will predict constructions like the ones shown below to
be grammatical contrary to fact:
(70) *An
if dite
saw.2PL ton,...
‘If you see him...’
(71) *Pu
when ida
saw.1SG ton,
him.CL-ACC milisa
spoke.1SG tu.
‘When I saw him, I spoke to him.’
Thus, interpretation of disjunction in this case should be inclusive. The lexical entry thus stays as in (56).
3.7. Variation environments
3.7.1. The Case of oti
One of the facts that make clitic positioning in CG rather complex, is the existence of environments where variation in clitic
positioning is possible. The most striking of these environments is the case of the non-factive complementizer oti ‘that’. This
complementizer, unlike its factive counterpart discussed above, can trigger both proclisis and enclisis. The relevant facts are
repeated below:
a. Ipen
said.3SG oti
COMP ethkiavasen
read.3SG to.
b. Ipen
said.3SG oti
it.CL-ACC ethkiavasen.
‘S/He said that s/he read it.’
Proclisis with oti is straightforwardly captured assuming an analysis in line with the approach I proposed for complementizer
pu, i.e. an analysis where oti builds the situation argument node and decorates it with a type esrequirement:
(73) Sample lexical entry for oti
IF ?T y(t),h↑i>
THEN make(h↓0i); go(h↓0i); put(?T y(es));
...gof irst(?T y(t))
ELSE abort
However, only half the problem is solved, since oti can also involve enclitic positioning. Nothing in the entry we have given
above will predict this fact. So, what can explain enclitic positioning with oti? Looking at the diachrony of oti, specifically its
behaviour in Medieval Modern Greek (MMG) and Medieval Cypriot Greek (MCG), might give us a way out of the problem.
Clitic positioning with oti in MMG is prevalently postverbal, with 30 of the 38 tokens showing postverbal positioning and
only 8 of them preverbal (Pappas, 2004). The situation in the Cypriot chronicles of Makhairas and Boustronios with respect
to clitic positioning of oti is quite harder to tell. Pappas (2004) does not give a count for oti for MCG. A first look at the
chronicle of Makhairas, reveals a number of cases involving oti. However, in their vast majority, these cases involve the use
of oti as a causal subordinator and not as a non-factive complementizer. However, in both its guises, oti can be found with
both proclisis and enclisis, with enclisis being prevalent in both cases (5/8 for non-factive oti and 36/40 for causal oti). I
do not know if and how positioning of causal oti has influenced positioning with complementizer oti or vice versa. What is
rather interesting is Pappas’ (2004: 36) discussion concerning the classification of oti in the category clause initial. Pappas
provides two interesting views on the function of oti, one given by Mackridge (1985) and one given by Jannaris (1968). Both
these writers, argue that oti is actually more of a coordinating conjunction rather than a subordinating one. Mackridge states
that oti shows a pattern of “incomplete subordination” while Jannaris (1968) that in many instances oti “corresponds to our
modern colon(;)...”. The use of oti as a coordinating rather than a subordinating conjunction might give us a way out of the
variation problem. I will then assume that oti, can be also parsed as a coordinating conjunction. In this scenario, oti will
build a LINK transition in the same sense coordinating conjunctions do in DS. There are a number of issues with respect to
such a transition. Coordinating conjunctions induce this LINK transition from a type complete to a type requiring node of the
same type. However, in the case of oti things are different, since no complete type tnode will exist by the time oti will come
into parse. This is because a main verb like ksero ‘know’ will subcategorize for an NP subject and a VP object. Therefore,
when the complementizer oti will come into parse in a sentence like (74), no complete type tnode will exist, since the value
for the VP object (the internal type trequiring node) will have not yet been provided:28
(74) O
the Giorkos
George kseri
know.3SG oti
COMP kseris
know.2SG ton.
‘George knows you know him.’
(75) Before parsing oti in o Giorkos kseri oti kseris ton
?T y(t)
T y(e),
F o(giorkos0)?T y(et)
?T y(t),3T y(e(et)),
F o(λx.λy.kser i0(x)(y))
28Again, the situation nodes are omitted of ease of exposition.
The pointer is at the internal type trequiring node. The subject has been parsed but no complete type tcan be given until the
embedded clause is parsed. I assume in that respect that oti projects a type tvalue and a formula metavariable (F o(U)) in the
embedded type tnode with the requirement that this formula metavariable gets substituted by a proper formula value that is
the same as the one found on the top of the LINKed tree (?x.F o(x)∧ hLiF o(x). Then, a LINK relation is created and the
LINKed tree is created and decorated with a type trequirement (make(hLi); go(hLi); put(?T y(t))). The full lexical entry is
shown below:
(76) Final lexical entry for oti
IF ?T y(t),h↑i>
THEN make(h↓0i); go(h↓0i); put(?T y(es))
...gof irst(?T y(t))|
put(T y(t), F o(U),?x.F o(x)∧ hLiF o(x))
make(hLi); go(hLi); put(?T y(t))
ELSE abort
Given this lexical entry, the result of parsing (74) using the LINK strategy is shown below:
(77) Parsing o Giorkos kseri oti kseris ton ‘George knows that you know him.’
?T y(t)
T y(e),
F o(giorkos0)?T y(et)
T y(t), F o(U),
?x.F o(x)∧ hLiF o(x)
T y(e(et)),
F o(λx.λy.kser i0(x)(y))
T y(t), F o(kseris(gianis0)(stergios0)),3
The only possible substituent for the Fo(U) metavariable in the embedded type t node given the situation above is the Fo
formula value of the top node of the LINKed tree (F o(kseris(gianis0)(stergios0))). Given substitution, we get the correct
semantic results. Thus, treating oti both as a regular complementizer and as a coordinating conjunction will predict variation
to be possible.29
3.8. The Case of epidi
Subordinate clauses introduced with the causal subordinator epidi (giati and afu as well) also exhibit variation in CG. The
relevant data are repeated below:
29For those speakers that accept variation with pu, it can be argued that parsing of pu as a coordinating conjunction is also possible, possibly in analogy
to non-factive oti.
a. Epidi
because enevriases
made-angry.2SG me,
me.CL-ACC en
NEG kamno
do.1SG tipota.
b. Epidi
because me
me.CL-ACC enevriases,
made-angry.2SG en
NEG kamno
do.1SG tipota.
‘I’m not doing anything because you pissed me off.’
The variation attested with epidi is problematic, since one would expect epidi to pattern with the rest of the subordinate
conjunctions in terms of clitic positioning and exhibit only proclisis rather than variant positioning. Thus, assuming an
analysis similar to the one I have argued for the rest of the subordinating conjunctions, capturing proclisis is the easy case.
Under such an analysis, the causal subordinator will project a type esrequirement that will function as the clitic’s proclitic
trigger. However, this is again half the analysis since we also need to capture enclisis. How is enclisis going to be captured?
In order to answer this question, it would be good to look at the properties of the causal subordinator in question or causal
subordinators in general. The first thing worth mentioning is that this variant positioning exhibited by epidi has been also
attested for the equivalent subordinator in Medieval Spanish (Bouzouita, 2008a). The cause subordinator ca ‘because’ in
MedSp shows the same characteristics, exhibiting both proclisis and enclisis (see Bouzouita, 2008a for the actual examples).
Bouzouita (2008a: chapter 3), following a generally accepted line of view (Lapeza 1978; Menendez Pidal 1980), notes that
sentences introduced with ca can function as either subordinate or coordinate sentences. The reason for this structural duality
can be traced back to Latin where two different kinds of causal relations existed, one pattering with coordinate and one with
subordinate structures. The same structural duality is found in Ancient Greek (AG, Tzartzanos, 1940). It is worth mentioning
that the variation attested with epidi in CG is also attested in MCG, so variation is not an innovation of the modern CG
system. A quick check in the chronicle of Makhairas returned four instances of clitics appearing in epidi clauses. Half of
them exhibited proclisis and the other half enclisis, so variation with epidi was already at stake in MCG. Of course, this kind
of evidence is not conclusive for the assumption that epidi can be parsed as a coordinating conjunction as well. 30 However,
what I am going to propose is that epidi, just like oti can be parsed as both a subordinating and a coordinating conjunction.
Whether this behaviour has its roots in the structural duality of AG subordinates of cause or it is an innovation of the MCG
or even the modern CG system is something that for the moment cannot be proven. Going back to the analysis of epidi let us
start with the proclitic case involving epidi. In this case, parsing epidi will give us a type esrequirement, and thus proclisis
will obtain. In the second case, epidi will be parsed as a coordinating conjunction where no such a requirement will exist.
Actually, the analysis of epidi as a coordinating conjunction will be much closer to the DS analyses of classic coordinating
conjunctions like and or but than the analysis provided for oti. This is because oti was assumed to build a LINK relation from
a type trequiring node to another type trequiring node or, according to the alternative analysis provided, the LINK relation
was created from a type complete node but with no complete formula value (metavariable). This is unlike the analyses of
coordinate conjunctions like and or but where the LINK relation is built from a type and formula complete node into a type
trequiring node. In the case of epidi however, one can maintain an analysis where the causal conjunction builds a LINK
relation from a type and formula complete node to a type trequiring one. Let us see how this will be done. Say we want to
parse a subordinate of cause structure involving a clitic in enclitic position followed by the main clause:
(79) En
NEG su
you.CL-GEN milao,
talk epidi
because enevriases
made-angry.2SG me.
‘Because you made me angry, I’m not talking to you.’
First we parse the main clause obtaining a type tand a formula value in the initial node. At that point epidi comes into parse
and creates a LINK relation from the type tcomplete node to a type trequiring node and leaves the pointer at the latter node.
30Another way to explain the variant positioning exhibited with causal subordinates is to attribute it to variant positioning found with the complementizer
oti. Given that oti could be either a non-factive complementizer or a subordinate of cause conjunction, variation exhibited by non-factive oti could spread to
causal oti and from causal oti to subordinates of cause in general. However, no data are available to vindicate such a speculation and as such at the moment
it cannot be anything more than a speculation.
The pointer is left at the type trequiring node of the LINKed tree. No proclitic trigger exists and thus parsing of the clitic is
impossible. On the other hand, the verb can be parsed since its triggering point (?T y(t)) is satisfied, and now parsing of the
clitic is possible giving rise to enclitic positioning:
(80) After parsing epidi
T y(t), F o(miliso(maria0(sterg ios0))
T y(e),
F o(stergios0)?T y(et)
T y(t), F o(maria0)T y(e(et)),
F o(λx.λy.miliso0)(x)(y)
T y(t), F o(enevriases(stergios0)(maria0)),3
At that point the LINK evaluation rule for causal subordinates applies. A LINK evaluation rule is basically a rule that defines
the exact semantic correlation of the two LINKed trees (e.g. conjunction in coordinated structures). The LINK evaluation
rule for causal subordinates will be minimally different to the one assumed for coordinate conjunctions like and or but in that
it links the two sentences via a causal relation:31
(81) LINK evaluation rule for epidi
{{T n(X), T y(t), Fo(α), ...},{hL1iT n(X), T y (t), F o(β),3...}...}
{{T n(X), T y(t), Fo(α)F o(β),3, ...},{hL1iT n(X), T y (t), F o(β), ...}...}
The symbol is taken to encode the causal relation.32 Formally F o(α)F o(β)reads as: F o(α)has a cause in case
F o(β)is true. In simpler terms, F o(α)is the cause of F o(β).33 The analysis of causal subordinates in line with coordinating
conjunctions will give us enclisis than proclisis. Given that subordinates of cause can be also parsed as regular subordinates,
variation with epidi is predicted to be possible. The same line of reasoning can be applied to the other causal subordinators
like giati and afu, ‘because’ and ‘since (causal)’. 34
3.9. Complex Conjunctions/marker with tze
In this section, I will take a look at the behaviour of complex subordinators and the complex negation marker ‘en tze’
with respect to positioning. All these complex subordinators/markers are consisted of two parts. The first part is a marker or
subordinate conjunction, while the second part is the coordinate conjunction tze. The first part of the conjunction is comprised
31Similar rules are assumed for relative, conditional and temporal clauses. See Gregoromichelaki (2006) and Chatzikyriakidis, 2010 for more details.
32Taken from Litschitz (1998).
33The formal details of what a causal relation is are not going to be discussed here. The interested reader is directed to Lewis (1986) for a discussion of
causal relations and Liftschitz (1998), Shafer (1998) for causal logic models using situation calculus and event trees respectively.
34The difference in pragmatic import associated with the use of the different causal subordinators need not bother us here.
of proclitic triggers (e.g. negation marker en or subordinate conjunctions), however the complex conjunction/marker as a
whole fails to trigger proclisis and enclisis obtains instead. For example, negation en and the conditional conjunction ean,
as discussed earlier, categorically trigger proclisis. However, the complex particles/markers en tze ‘NEG’ and ean tze ‘even
though’ fail to do so and enclisis is found in these cases instead. The culprit behind this behaviour seems to be the coordinating
conjunction tze, which in a way cancels triggering of preverbal positioning. Indeed, this is the analysis already pursued in
Agouraki (2001), where en tze is argued to involve conjunction of two CPs. The negation marker occupies the head of the
CP1 phrase, while conjunction tze occupies the SpecCP2, the verb occupying CP20. However, in order for tze to cancel out
a proclitic trigger, it needs to follow this trigger and not just be within the same domain with this element. For example,
coordination between two clauses where the second of these involves the negation marker following tze regularly trigger
proclisis as the example below shows:
(82) Tuto
this en
is alithja
truth tze
and en
NEG ton
him.CL-ACC adiko
blame pu
COMP alakse
changed.3SG gnomi.
‘This is true and I do not blame him for changing his mind.’
A good way to test whether tze really cancels out proclisis in general or this is just an idiosyncratic behaviour of the complex
elements involving tze, is to try and find cases where tze appears inside the clause rather than in its usual position at the
beginning of the clause. Fortunately in Modern Greek and CG as well, there are a number of constructions where tze is not
used as a coordinating conjunction, but rather starts off the sentence. Tze is followed by an affirmative sentence, followed by
another instance of tze, followed by the negative version of the affirmative sentence. Such examples are shown below::
(83) Tze
and ksero
know.1SG ton
him.CL-ACC tze
and en
NEG ton
him.CL-ACC ksero.
‘I know him but not too well (free translation)’
(84) Tze
and sevome
respect.1SG ton
him.CL-ACC tze
and ektimo
appreciate.1SG ton.
‘I both respect and have an appreciation for him’
So, how are structures like the above are going to help us decide whether tze cancels out proclisis across the board when
following proclitic triggers? Well, fortunately structures like these can be embedded within subordinate clauses, in which
case they follow the subordinate conjunction. The crucial sentence in these cases is the affirmative one given that the negative
one will be proclitic in any case. Indeed, as we see from the example below, tze cancels out proclisis that is induced by the
conditionals ean and otan respectively:
(85) Ean
if tze
and theli
want.3SG to
it.CL-ACC tze
and en
NEG to
it.CL-ACC theli,
want.3SG kalitera
better na
SUBJ min
NEG ton
‘If he does not want it that much, let us not push him.’
(86) Otan
when kapion
someone tze
and sevese
respect.2SG ton
him.CL-ACC tze
and ektimas
appreciate.CL-ACC ton,
him.CL-ACC tote
then tzinos
he enna
‘When you both respect and appreciate someone, s/he is going to pay you back.’
In effect when tze follows any of the proclitic triggers, then it cancels out proclisis. Now, in giving an analysis of en tze, we
must first look at the elements comprising the complex negation particle en tze and see what the existing DS analyses are,
if any, with respect to each of these elements separately. The negation marker has already received an analysis. According
to this, the negation marker builds the situation argument node and then returns to the type trequiring node where it further
projects a [+NEG] feature on that node. On the other hand, coordinating conjunctions have traditionally been characterized
in DS as LINKed structures, the mainstream assumption being that they induce a LINK relation from a type complete node
to a node with a requirement of the same type (Cann et al., 2005). The reasoning behind such treatment is that coordinating
conjunctions start a new domain, be it sentential, verbal or nominal, an assumption rather independent from the theoretical
model used here. However, no actions are induced in that same domain by coordinating conjunctions in contrast to subordinate
conjunctions which further introduce structure in that new LINKed domain. In that sense, a clitic following a coordinating
conjunction is a sentence initial clitic, since it is the first element inducing structure in the domain. The reason that proclisis
cannot obtain with en tze, and enclisis is only possible might very well be attributed to this latter fact, i.e. that tze actually
starts a new domain different than the one en is parsed. In that sense, the situation argument node projected by the negation
marker is parsed in a different domain, given that tze will induce a LINK transition from the tree where negation is parsed to
another tree structure. The first thing we need to look at in formalizing this assumption, is the node from which the LINK
transition associated with tze is going to be created. When tze comes into parse in a sentence like (85), the only word already
parsed will be the negation operator en:
(87) After parsing en
?T y(t),[+N E G],
?T y(es)
The pointer returns to the type trequiring node. The problem is that now, no node is type complete. In that sense, the
assumption that coordinating conjunctions project a LINK relation from a type complete node to a node requiring the same
type cannot be maintained in this case. The other way to look at the problem is to assume that these complex elements are
parsed as one element and not separately. Such an assumption can be further backed by the fact that nothing can intervene
between the two elements in all these constructions and furthermore, while the two elements can appear on their own,
the interpretation that they give rise to in combination is not a compositional accumulation of the content of the entries
of each of the separate elements. For example, an tze ‘even though’, cannot be captured assuming an analysis where the
two elements are treated as separate entries, since in this case what we get is the actions induced by the conditional an
plus the actions of the coordinating conjunction tze, which by no means capture the semantics of concession. Of course,
none of these two diagnostics is enough to guarantee that the two elements involve one single lexical entry, since examples
involving two inseparable elements but still are treated as separate entries or cases where a combination of two words does
not give to compositional meaning are all over the place in natural language. Examples of the first kind include cases
of mood/tense particles and the verb or the clitic and the verb in Greek, whereas examples of the second involve various
idiomatic expressions.35 For the moment, it is impossible to dwell into a discussion of whether or not these elements are
to be parsed as a single entry or not. For the moment I will make the assumption that one lexical entry is indeed involved
in these cases. Given an analysis for en tze as involving one lexical entry, we can assume that en tze first marks the type t
requiring node with the [+NEG] feature while it projects a type t value and a formula metavariable in that same node with the
requirement that the value for the metavariable’s substituent is to be shared with the formula value of the LINKed tree, as in
the case of oti. Then, a LINK transition to a type trequiring node is created. This new LINKed domain will be the domain in
which the rest of the sentence is going to be parsed. What I am then assuming is that the sentence is parsed within the context
of a negative specification, within the context of negation. In that sense, such an analysis is quite close to the DS analysis of
HTLD or relative clause structures. Left dislocated arguments in HTLD constructions or the relativized elements in relative
clauses can be seen as setting the context in which the rest of the clause in HTLD structures or the relative clause in relative
35An anonymous reviewer is gratefully thanked for reminding me that the arguments used are enough to prove that the cases in question are indeed cases
where one complex lexical entry is involved.
clause structures are going to be parsed. In the same sense, en tze can be seen as providing the context (a negative context) in
which the sentence is to be parsed. The lexical entry for en tze is shown below:
(88) Lexical entry for en tze
IF ?T y(t)
THEN put([+N EG], T y (t), F o(U),?x.F o(x)∧ hLiFo(x));
make(hLi); go(hLi); put(?T y(t))
ELSE abort
The result in tree notation is shown below:
(89) After parsing en tze
?T y(t),[+N E G], T y (t), F o(U),?x.F o(x)∧ hLiF o(x) ?T y(t)
Notice that no type esrequiring node is assumed to be projected by en tze. This is done for reasons of simplicity only, since it
does not make any difference in the analysis presented, given that the type esrequiring node will be projected in the domain
where the LINK starts, i.e. in a tree different to the one where the clitic is going to be parsed as we will see. This type es
requiring node cannot constitute a proclitic trigger, as the clitic will be parsed in a different domain, namely the domain of
the LINKed tree. After parsing en tze the pointer is at the type trequiring node of the LINKed tree. Let us say we want to
parse en tze ida ton ‘I did not see him’. En tze is parsed first giving us the structure just discussed. Then, the rest of the
sentence is parsed in the LINKed type trequiring domain. Assuming that all metavariables (subject and object) have been
substituted by values from the context, we end up with the following structure (the metavariablein the type t node of the first
tree is substituted by the value found in the LINKed type t node):
(90) After metavariable substitution
T y(t), F o(ida(giani)(stergios0)),[+N EG]T y(t), F o(ida(giani)(stergios0))
The core of the proposed analysis is that the complex negation particle en tze links a type t node with a type t requiring one.
The node where the LINK starts (the type t node), carries the [+NEG] feature. The pointer ends up at type trequiring node
after parsing en tze. It is at that domain where the rest of the sentence is parsed. The sentence is parsed given a negative
context, i.e. the complex negation marker sets a negative context against which the sentence is going to be parsed. This
treatment of complex negation is an effective way to get out of the enclisis problem. Proclisis is not possible with en tze
since the clitic is parsed in a domain where no proclitic triggers exist (the clitic will be parsed on the LINKed tree). In
that respect, the only possibility with en tze is enclisis. The same reasoning can be argued to apply for the other complex
markers/conjunctions like ean tze ‘even though’. Ean tze, in the same sense as en tze, will be parsed as one complex element
linking a type trequiring node with another node. A LINK evaluation rule will then do the appropriate copying, so that the
semantics of concession are captured. There are a number of details that need to be taken care of for the case of ean tze,
since concessive clauses just like subordinate clauses will already be linked with the consequent clause via a LINK relation.
I will not flesh out the exact details of how the ean tze conjunction (or other complex conjunctions with tze) works. What is
crucial, is that there is a way to account for these cases fairly easy within DS.
3.10. Clitic Positioning with Imperative verbs
We have provided a mechanism for capturing the full range of clitic positioning restrictions in CG. However, all these posi-
tioning restrictions are relevant for non-imperative verbs only, since in imperative environments these restrictions disappear
and enclisis is the case across the board. The examples below shown cases of imperative sentences where a fronted constituent
is present but proclisis cannot obtain:
(91) TUTO
this pe
tell.2SG.IMP mu/
me.CL-DAT *TUTO mu pe.
’Tell me this.
(92) TORA
now fer
bring.2SG.IMP mu
me.DAT to/
it.CL-ACC TORA mu to fer(e).
‘Bring it to me now.
Proclisis with mood/tense markers or the negation marker are in principle not possible since these elements are incompatible
with imperatives in general. Given this situation someone might argue that what the account provided does is to account
for only a subpart of the clitic positioning behaviour of CG, given that clitic positioning with imperatives is not captured.
However, as we will see the lexical entry we have proposed for CG will work for imperatives as well. As the reader might
already inferred the difference in the behaviour between indicative and imperative verbs as regards clitic positioning will
reside in the lexical entries for the two different types of verbs. As we have just mentioned, imperative verbs are in principle
incompatible with tense/mood markers or the negation marker in CG. In this sense, we will need a mechanism irrelevant
of the behaviour of clitics that will rule out the cases where a mood/tense or negation marker is followed by an imperative
verb. The obvious way to rule these cases lexically is via a restriction in the entry for imperative verbs that will abort in case
a requirement for a estype exists in the 0 node below the type tnode, given that all these elements will project this node
along with the type esrequirement (see the discussion on ?). Thus, the lexical entry for the imperative verb in CG, or even in
languages with similar behaviour like for example SMG will have to have a trigger that aborts in case such a situation occurs:
(93) Lexical entry for imperative verbs in CG (ACTIONS are omitted)
IF ?T y(t)
THEN IF h↓0i?T y(es)
ELSE abort
ELSE abort
Of course the above entry does not suffice to exclude cases where an unfixed node has been parsed first followed by the clitic,
and then the imperative verb. Furthermore, we cannot exclude these cases on the basis of the unfixed node strategy given that
cases where a fronted element parsed on an unfixed node followed by an imperative exist in CG:
(94) TON
George.ACC (i)ksero.
’Tell me this.
(95) TORA
now ela.
‘Come now.
Chatzikyriakidis (2009) argues that imperatives in Grecia Salentina Greek (SMG as well) involve a restriction in their entries
for imperative verbs that basically aborts in case any fixed node is present in the tree structure:36
36See Chatzikyriakidis (2009) for the relevant argumentation.
(96) The restriction as given by Chatzikyriakidis (2009)
IF ?T y(t)
THEN IF [+]?x.T n(x)
ELSE abort
ELSE abort
The restriction basically says that all nodes below carry a requirement for a proper treenode address, thus are all unfixed.37
This restriction includes the restriction we have provisionally provided for excluding tense/mood and negation markers, since
it aborts in case any fixed node is present. However, this restriction does more than this, since it provides us with an account of
the asymmetry in clitic positioning between imperative and non-imperative verbs. Let us see why. Positing such a restriction,
one will not only exclude cases where a tense/mood or negation marker has already been parsed (since a fixed node will exist
in this case) but also in general all cases in which a fixed node is present when the verb comes into parse. Parsing a clitic, will
always involve the projection of a fixed node, thus cases where an unfixed node plus a clitic have been parsed will be ruled
out for imperatives.38 Thus, assuming that imperatives in CG involve the restriction as this is posited for GSG and SMG in
Chatzikyriakidis (2009, 2010), the asymmetry in the behaviour of clitic positioning in imperative and non-imperative verbs
is explained.
3.11. Some notes on the Diachrony of the Cypriot Greek positioning System
The modern CG system emerged from its medieval counterpart, Medieval Cypriot Greek (MCG). Clitic positioning in MCG
was quite similar to the positioning found today. There is however a major difference between the two systems. MCG in
contrast to CG does not exhibit proclisis with fronted elements. The data below taken from Pappas (2004) are illustrative:
(97) Clitic positioning in the Cypriot Chronicles (vols. 1-56, adapted from Pappas, 2004)
Environment Preverbal Postverbal
Clause initial 0 208
Reduplicated object 0 15
Function word 101 3
Fronted constituent 1 14
Subject 0 29
Gerund 0 19
Imperative 0 4
As we can see, only one instance out of 15 of a fronted constituent triggers proclisis. The same goes for subjects even though
we do not have a separate count for fronted subjects. Thus, the two systems differ in that in MCG the factor fronted element
does not play a role in clitic positioning, whereas in CG a fronted element can triggerproclisis.
The account proposed for CG involves the use of three generalized parsing strategies that are used as triggers for parsing the
clitic. These strategies do not refer to the actual elements parsed but are general actions of structure building. For example
the unfixed node trigger refers to the existence of a structurally underspecified element whereas the enclitic trigger refers
to the presence of any functor (verbal) type. This way of dealing with the complex positioning of CG, besides providing
37Note that in case where no nodes exist below the current type t requiring one, the statement is trivially true, given that universal quantification does not
carry any sort of existential statement.
38I assume that dative clitics, even though underspecified, always project a fixed node. This is done in order to capture the fact that these clitics even
though underspecified never function as subjects. In this sense, I follow Chatzikyriakidis (2009, 2010) rather than Kempson & Cann (2008.
an economical account of the facts, is further backed by the evidence of MCG and the transition from MCG to CG. As we
have seen, in MCG fronted constituents do not trigger proclisis. Proclisis with these elements is captured via the unfixed
node trigger, which furthermore captures proclisis with Wh elements. MCG like CG exhibits proclisis with Wh elements.
Thus, the lexical entry for MCG would still involve the unfixed node trigger but this would have to be specified to apply
to Wh elements only, i.e. the lexical entry would exclusively refer to the parsing strategy of the unfixed node as applied to
Wh elements. Given this, the transition from MCG to CG involves the generalization of the unfixed trigger to apply for all
elements that can be parsed on unfixed nodes. In effect we have a transition from an item-specific trigger (the unfixed node
strategy as applied to Wh elements only) to an item-neutral one (general unfixed node strategy). The transition from MCG to
CG is shown below:
(98) The transition from MCG to CG
IF ?T y(t)
1iT y(x)|
h↓i?x.T n(x)F o(W H)|
h↓0i?T y(es)
ELSE abort
ELSE abort
IF ?T y(t)
1iT y(x)||
h↓i?x.T n(x)|
h↓0i?T y(es)
ELSE abort
ELSE abort
As you see in the above entries the unfixed node trigger specialized to apply for Wh elements only in the MCG entry
(h↓i?x.T n(x)F o(W H)) is generalized to all elements parsed on an unfixed node in the CG entry by dropping the
second part of the conjunction (h↓i?x.T n(x)). In that respect, the transition from MCG to CG involves the emergence of
a new proclitic environment, i.e. the fronting constituent environment. Such a transition is easily explainable given the DS
perspective as a process where a parsing strategy acting as a parsing trigger for the parse of a specific class of elements only
(Wh elements) gets generalized and serves as a generalized parsing trigger for all other elements using this strategy (Wh
elements plus fronted constituents). What is rather welcomed in such an analysis is that even though the CG system looks
more complicated on a descriptive basis than the MCG one, since it involves more proclitic environments than CG, the entry
for CG is less complicated than the respective MCG one. In that respect, the passage from MCG to CG involved reducing
the complexity of the entry as well. It is worth noting that the transition from MCG to CG, as shown in Chatzikyriakidis
(2010) parallels the transition to other MG dialects via their respective medieval ancestors, with routinization (in the sense of
Pickering & Garrod, 2004) and parsing trigger generalization being the leading factors of change in all cases. Reasons of space
do not allow me to further expand on the issue in this paper. However, the interested reader is directed to Chatzikyriakidis
(2010) for a quite detailed sketch of the transition from Koine Greek to the Medieval Greek dialectal systems and from the
latter systems to the modern positioning ones. Furthermore, the interested reader is directed to Bouzouita (2008) for a similar
account as regards the transition from Medieval to Modern Spanish.
4. Some general notes on clitic positioning and clitics in general
Abstracting away from the technicalities of the proposed account, one might plausibly ask what this account has to offer with
respect to the nature and properties of clitics. I consider that DS accounts of clitics (this account included) have a number of
advantages over other theoretical frameworks and can explain some interesting aspects of the syntax of clitics from a wholly
new perspective. Firstly, what I consider very important is that the word-affix debate, as already discussed in 3.1, is irrelevant
for DS, since assuming that lexical entries are basically just packages of information, what we need in order to assume that
a lingusitic element has a separate lexical entry, is that this element provides distinct procedural information as regards the
parsing process. Such information is definitely conveyed by clitics whether these are considered words or not. In this sense,
no pre-theoretical decision on the issue needs to be taken in order to provide an account of clitics within the DS framework.
What is more, the account will be compatible with both options (clitics being words and clitics being affixes) and as such we
avoid the problem of an analysis being dependent on whatever pre-theoretical decision we take on the issue. Furthermore,
clitics under the account proposed, and pretty much all DS accounts on clitics, are treated as ordinary pronouns in that they
are assumed to project a type value and a formula metavariable that needs to be updated from the context. However, what
is exceptional as regards clitics is that these create their own structure, and, in so doing, they have become more complex
than just being a pronoun occurring in a certain position. Thus, the lexical entry for a pronoun has as its triggering point the
node in which type and formula information is going to be provided (?Ty(e)). However, in the case of clitics, this triggering
point is a type trequiring node. Starting from this node, clitics create their own structure in which they project type and
formula information (by building a fixed node or an unfixed one). This behaviour of clitics is independently needed given
the locality restrictions associated with clitics. Given these restrictions, a type erequiring trigger for clitics will predict that
locality violations in the case of clitics should be fine contrary to fact (given parsing of the clitic on a node introduced by
*ADJUNCTION). This behavior, the fact that clitics induce their own structure, is one of the differences between clitics and
pronouns and follows the natural observation as regards positioning of pronouns on the one hand and clitics on the other, the
former being freer compared to the latter. Lastly, the whole adjunct vs argument debate as regards the status of clitics and the
NPs (or DPs depending on the framework used) that are doubled by clitics in doubling constructions is again not something
that constitutes a problem in giving a DS analysis. Given that clitics project a Fo metavariable in one of the object nodes, a
proper formula value needs to act as a substituent for this metavariable. This substituent can very well come from the context
in which case no doubling occurs, but can however also come from the natural language string itself, in which case doubling
occurs by unification of the two nodes, the one that hosts the clitic and the unfixed node that hosts the doubled NP.39
Clitic positioning in a lexicalized framework like DS is defined within the lexical entries for clitics in each case, i.e. it is
basically treated as a lexical phenomenon. From a synchronic point of view positioning restrictions can be seen as a number
of generalized triggering points that if satisfied regulate clitic positioning. From a diachronic point of view, these restrictions
can be seen as the result of a number of processes, including routinization in the sense of Picekring & Garrod (2004) as well
as parsing/production (speaker/hearer) mismatches due to the availability of different parsing strategies for parsing one and
the same element. Indeed, similar claims have been made for languages like Spanish (Bouzouita, 2008) and an example of the
transition from MCG to CG has been given in this paper as well. Typologically, at least in the case of Greek, we are dealing
with three types of systems as Condoravdi & Kiparsky (2001) have already pointed out. The SMG one, the CG one and the
Pontic Greek (PG) one. The three systems can be described as involving the same set of actions but different triggering points
(see Chatzikyriakidis, 2010: chapters 3,4,5 for more details). These triggering points may seem arbitrary from a synchronic
point of view but make sense once one looks at the forms the respective medieval dialects had. Thus, typologically different
systems in Greek are described via different triggering points, the latter being dependent on the form of the medieval dialect
in each case. In the same vein, similar clitic systems in different language families e.g. the systems of Modern Spanish and
SMG should have developed from pretty much similar ancestoral systems, which is to a large extent true for the cases in
question (check Chatzikyriakidis, 2010 for Greek and Bouzouita, 2008 or Spanish). Of course, this last claim needs to be
further tested.
5. Conclusions
In this paper, I provided an account of CG positioning by arguing that the apparent complexity of proclitic positioning can
be derived assuming two generalized proclitic triggers. Following Bouzouita (2008), the first proclitic trigger was argued to
be the existence of an unfixed node, i.e. that the presence of an unfixed node constitutes a trigger for parsing the clitic. This
assumption correctly captures variation with fronted elements, given that these can be parsed using more than one parsing
strategy. In that sense, the account correctly predicts the claim made in Pappas (2010), namely that focused fronted elements
do not necessarily trigger proclisis. Building on assumptions by Gregoromichelaki (2006) and Cann (2011), according to
which every sentence contains an obligatory situation argument, I argued that the second proclitic trigger consists in the
existence of a type esrequiring node. I then assumed that subordinating conjunctions, modal/tense-aspect markers and
negation project a requirement for such a type. Enclisis on the other hand is captured assuming a generalized trigger which
allows the clitic to be parsed in case any verbal type has been parsed first. These three triggers are adequate enough to
39This is based on the assumption that clitics allow more structure to be built below the node they are hosted. If this is not allowed, then no doubling is
possible. This is presumably the difference between doubling and non-doubling languages, as well as the reason why strong pronouns cannot be doubled
(see Cann et al., 2005: chapter 4; Gregoromichelaki, 2010 among others).
capture the complexity of the CG positioning system as regards strict enclitic and proclitic environments. Then, variation in
positioning with oti and subordinates of cause was captured assuming that these elements can be also parsed as coordinating
conjunctions (i.e. LINK structures). Lastly, the account was shown to be grounded in diachronic considerations as well,
showing that the CG positioning system derived from the respective MCG one via parsing trigger generalization, according
to which an item specific trigger changes to an item neutral one.
This paper has benefited from a number of discussions with Ruth Kempson, Ronnie Cann, Stavroula Tsiplakou, Theoni
Neokleous. Spyros Armosti and Theoni Neokleous are also thanked for commenting on some of the CG data as well as
providing judgments. The AHRC is gratefully thanked for funding part of this work. Lastly, Alex Deligianidis is thanked for
providing crucial technical advice on parts of this paper.
Agouraki, Y., 1997. On the Enclisis-Proclisis Alternation. In: Drachman, G., Malikouti-Drachman, A., Fykias, J., & Klidi,
C., (eds), Proceedings of the 2nd International Conference on Greek Linguistics. Graz, W. Neugebauer Verlag, 393-404.
Agouraki, Y., 2001. The position of clitics in Cypriot Greek. In: Ralli, A., Joseph, B., & Janse, M., (eds), Proceedings of the
First International Conference on Greek Linguistics, 1-18.
Blackburn, P. & Meyer-Viol, W., 1994. Linguistics, Logic and Finite Trees.
Boˇskovi´c, ˇ
Z., 1995. Participle Movement and Second Position Cliticization in Serbo-Croatian. Lingua 96, 245-266.
Boˇskovi´c, ˇ
Z., 2001. On the Nature of the Syntax-Phonology Interface: Cliticization and Related Phenomena. Elsevier.
Bouzouita, M., 2008a. The Diachronic Development of Spanish Clitic Placement. Phd Thesis, King’s College, London.
Bouzouita, M., 2008b. At the Syntax-Pragmatics Interface: Clitics in the History of Spanish. In: Cooper, R., Kempson, R.
(eds), Language Evolution and Change. Cambridge, Cambridge University Press.
Bresnan, B., 2001. Lexical-Functional Syntax. Oxford, Blackwell.
Cann, R., 2011. Towards an Account of the English Auxiliary System: Building Interpretations Incrementally. In: Kempson
R., Gregoromichelaki, E., & Howes C., (eds), The Dynamics of Lexical Interfaces. Stanford, CSLI publications.
Cann, R., & Kempson, R., 2008. Production Pressures, Syntactic Change and the Emergence of Clitic Pronouns. In: Cooper,
R. & Kempson, R., (eds), Language in Flux: Dialogue Coordination, Language Variation, Change and Evolution, London,
College Publications, 221-263.
Cann, R., Kempson, R., & Marten, L., 2005. The Dynamics of Language. Oxford, Elsevier.
Chatzikyriakidis, S., 2006. Clitics in Modern Greek: A Dynamic Account. MSc thesis, King’s college London.
Chatzikyriakidis, S., 2009a. Clitics in Grecia Salentina Greek: A Dynamic Account. Lingua 119. Elsevier, 1939-1968.
Chatzikyriakidis, S., 2009b. The Person Case Constraint in Modern Greek: A Unified Dynamic Syntax Account. Newcastle
Working Papers in Linguistics 15, 40-62.
Chatzikyriakidis, S., 2010. Clitics in 4 Dialects of Modern Greek: A Dynamic Account. PhD thesis, King’s College, London.
Chatzikyriakidis, S., 2011. A Dynamic Account of Clitic Climbing: A First Sketch. In: Kempson R., Gregoromichelaki, E.
(eds), The Dynamics of Lexical Interfaces. Stanford, CSLI publications.
Condoravdi, C., & Kiparksy, P., 2002. Clitics and Clause Structure. Journal of Greek Linguistics 2, 1-39.
Crysmann, B., 1997. Parallel Morpho-Syntactic Constraints in European Portuguese Cliticization. In: Butt, M., & Holloway
King, T., (eds.), proceedings of the LFG97 Conference. CSLI publications, Stanford.
Gregoromichelaki, E., 2006. Conditionals in Dynamic Syntax. Phd thesis, King’s College, London.
Gregoromichelaki, E., 2010. A Dynamic Perspective on Left-Right Asymmetries: CLLD and Clitic Doubling in Greek.
In: Rightward Movement in a Comparative Perspective: Selected Papers from the DGfS-AG Rightward Movement Work-
shop. Amsterdam, John Benjamins. Jannaris, A.N., 1968. Historical Greek Grammar. Georg Olms Verlagsbuchhandlung:
Kempson, R., & Kurosawa, A., forthcoming. At the Syntax Pragmatics Interface: Japanese Relative Clause Construal. In:
Hoshi., H. (ed.), Language Mind and Brain. Kuroshio publishers.
Kempson, R., Meyer-viol, & W., Gabbay, D., 2001. Dynamic Syntax: The Flow of Language Understanding. Oxford, Black-
Kordoni, V., 2004. Between shifts and alternations: Ditransitive constructions. In: M¨uller, S., Proceedings of the HPSG04
Conference. Stanford, CSLI publications.
Lewis, D., 1986. Philosophical Papers, volume II. New York, Oxford University Press.
Lifschitz, V, 1997. On the Logic of Causal Explanation. Artificial Intelligence96, 451-465.
Luis, A. & R. Otoguro. 2004. Proclitic Contexts in European Portuguese and their Effect on Clitic Placement. In: Proceedings
of LFG04, CSLI on-line Publications.
Mackridge, P., 1985. The Modern Greek Language: a Descriptive Analysis of Standard Modern Greek. Oxford University
Press, Oxford.
Newton, B., 1972. Cypriot Greek: Its Phonology and Inflections. The Hague: Mouton.
Pappas, P., 2004. Variation and Morphosyntactic Change in Greek: From Clitics to Affixes. Palgrave Studies in Language
Change and Language History, Palgrave.
Pappas, P. A., 2010. Object clitic placement in the history of Cypriot Greek. In: Heselwood, B., Upton, C. (Eds.), Proceedings
of Methods XIII: Papers from the Thirteenth International Conference on Methods in Dialectology 2008. Frankfurt: Peter
Paraskeua, M., 2007. A Dynamic Syntax Account of Object Clitics in Cypriot Greek. Msc Thesis, King’s College, London.
Pickering, M., & Garrod, S., 2004. Towards a Mechanistic Psychology of Dialogue. Brain and behavioral science 27.
Revithiadou, A., 2006. Prosodic Filters on Syntax: An Interface Account of Second Position Clitics. Lingua 116, 79-111.
Roussou, A., 2000. On the Left Periphery: Modal Particles and Complementizers. Journal of Greek Linguistics 1, 65-94.
Roussou, A., 2008. Selecting complementizers. Ms., University of Patras.
Shafer, G., 1998. Causal Logic. Proceedings of the 13th European Conference on Artificial Intelligence (ECAI 98), John
Wiley & Sons, 1998, 711-720.
Tenny, C., & Pustejovsky., J., 2000. A History of Events in Linguistic Theory. In: Tenny, C., & Pustejovsky., J. (eds.), Events
as Grammatical Objects. Stanford, CSLI publications.
Terzi, A., 1999a. Cypriot Greek clitics and their Positioning Restrictions. In: Alexiadou A., Horrocks, G. & Stavrou, M.
(eds.), Studies in Greek Syntax. Dordrecht, Kluwer, 227-240.
Terzi, A., 1999b. Clitic combinations, their Hosts and their Ordering. Natural Language and Linguistic Theory 17, 85-121
Tzartzanos, A., 1940. Sintaktiko tis Arxeas Elinikis Glosis. [Syntax of the Ancient Greek Language]. Kiriakidis, Thessaloniki,
Tsiplakou, S., 2009. Code-switching and Code-mixing between Related Varieties: Establishing the Blueprint. The Interna-
tional Journal of the Humanities 6, Issue 12, 49-66.
Tsiplakou, S., in press. Competing grammars in Cypriot Greek? Evidence from language alternation. Proceedings of the 18th
International Conference on Historical Linguistics, Montreal, 6-11 August 2007. Amsterdam, John Benjamins.
Tsiplakou, S., Panagiotidis, P. & Grohmann, K.K., 2007. Properties of Cypriot Greek Wh-question formation. In: Proceedings
of the 7th International Conference on Greek Linguistics, 83-98.
Appendix A. Computational Rules
{...{h↓0i(F o(a), T y(X)),h↓1i(F o(b), T y (XY), ..., 3)}...}
{...{F o(b(a)), T y(Y),h↓0i(F o(a), T y (X)),h↓1i(Fo(b), T y(XY), ..., 3)}...}
Condition:h↓ii?φdoes not hold and i ∈ {0,1}
{...{T n(n), ...},{<i>, T n(n), ..., T y (X).., 3}...}
{...{T n(n), ..., <i> T y(X), ..., 3},{<i> T n(n), ..., T y(X), ...}...}
Where i (0,1,*)
{...{T n(n), ..., 3},{h↑iT n(n),?X...}...}
{...{T n(n), ...},{h↑iT n(n),?X..., 3}...}
{...{..., X, ..., ?X, ..., 3}...}
{...{..., X, ..., 3}...}
{...{T n(n),?T y(t),3}}
{...{T n(n),?T y(t)},{h↑iT n(n),?T y(e),?x.T n(x),3}}
{...{..., DU, DU 0, ...}...}
{...{..., DU tDU 0, ...}...}
Where 3DU’ and DU DU’ is consistent
{{T n(0),?T y(t),}}
{{T n(0),?T y(t)}},{< L > T n(0),?T y(e),}
{{T n(0),?T y(t)},{< L > T n(0), F o(a), T y(e),}
{{T n(0),?T y(t),?< D > F o(a),}},{< L > T n(0), F o(a), T y(e)}
... Enclisis in cg and sg 3.1 Overview I start the discussion by giving an overview of (linear) enclisis in cg and sg (for phonological proclisis/enclisis, see Klavans 1985; for cg, see Revithiadou 2006Revithiadou , 2007. For descriptions of cg enclisis, see Agouraki 1997Agouraki , 2001Agouraki , 2010Chatzikyriakidis 2010Chatzikyriakidis , 2012Condoravdi & Kiparsky 2001;Mavrogiorgos 2010Mavrogiorgos , 2013Pappas 2004;Philippaki-Warburton 1995Revithiadou 2006Revithiadou , 2007Rivero 1994;Rivero & Terzi 1995;Terzi 1999a,b). The main generalization to keep in mind is the following: in sg, enclisis strictly correlates with lack of person and/or restricted person specification on the verbal host (see Mavrogiorgos 2010). ...
... Whereas an typically correlates with proclisis, dʒe always triggers enclisis. On the other hand, an dʒe (on its own) triggers enclisis (and the same applies to other complex complementizers containing dʒe; see also Agouraki 2001;Chatzikyriakidis 2012). Why should this be the case? ...
... That the structural position of a complementizer is crucial for clitic positioning becomes most clearly evident in the case of optional complementizers. An example of an optional complementizer was provided in (29b), which is repeated here for convenience: In cg, optional complementizers include among others óti and pos 'that' , or epiðí and γiatí 'because' (see also Agouraki 2001;Chatzikyriakidis 2012;Revithiadou 2006). Their common property is that they seem to be compatible with both proclisis and enclisis. ...
This paper pursues the idea, originally proposed by Landau (2007), that the Extended Projection Principle is pf related on the basis of Greek enclisis. It is argued that the complementary distribution pattern attested with Cypriot Greek finite enclisis derives from the fact that the first head h c-selecting tp has a morpho-syntactic requirement, and a related pf/prosodic requirement subject to an Economy Condition. The former derives merger of an x or xp copy at h, while the latter ensures that only one of the two copies gets spelled-out. Non-finite h triggers obligatory enclisis in both Cypriot and Standard Greek, as it contains only affixal morphemes, which is further supported by Medieval Greek non-finite enclisis. The parameterization of h along with potential implications are also discussed.
... 'S/He/It loves him.' [SMG] ( The second type of system is the one exhibited by varieties like Cypriot Greek in the case of Greek (CG) (Revithiadou and Spyropoulos 2008;Chatzikyriakidis 2012), while its Romance counterparts are languages like European Portuguese (Crysmann 2001;Luıs et al. 2004), Asturian Spanish (Crysmann 2001;Luıs et al. 2004) Medieval Spanish (Bouzouita 2008) and Medieval Greek (Pappas 2004b;Soltic 2013). In these varieties, clitic positioning is in general enclitic (7) unless a number of proclitic triggers appear at the left periphery. ...
... Note that examples (14)-(16) can appear with enclisis as well. SeePappas (2008Pappas ( , 2012 andChatzikyriakidis (2012) for the relevant data. For a detailed assessment of these variation environments using a ...
... In this respect, the claim that there is a tendency towards proclitic positioning does not seem to be contradicted byPappas' claim. 27 For more details as regards the DS analysis of CG seeChatzikyriakidis (2012). ...
Full-text available
In this paper, the historical development of the clitic systems of Standard Modern, Cypriot and Pontic Greek are discussed. These three varieties not only present the whole range of variation one can find across clitic systems in Greek but, furthermore, derive from a common linguistic ancestor, i.e. Koine Greek. This paper argues that the transition from Koine Greek to the Medieval varieties and from the Medieval varieties to the respective modern ones can be explained by making the assumption that routinization (in the sense of Pickering & Garrod 2004) and parsing/hearer assymetries are two important factors behind syntactic change. The claim is that the transition from Koine to the Medieval Greek varieties involves the emergence of a clitic system with encoded syntactic constraints out of a freer one, where clitic positioning was regulated by pragmatic preferences rather than syntactic constraints. Then, the transition to the modern varieties from the respective medieval ones is explained, at least partly, on the assumption that production/parsing mismatches are capable of triggering syntactic change. This last assumption combined with: a) the tendency to obtain more generalized parsing triggers for parsing the individual clitics and b) the fact that the Medieval varieties in question differ in minimal but crucial ways, provides us an explanation for the transition to the modern varieties.
... Pronominal object clitic placement differs in the two varieties in question: in Cypriot Greek pronominal object clitics generally 10 appear in the second position in the clause (Agouraki, 2001;Chatzikyriakidis, 2012;Mavrogiorgos, 2013;Neokleous, 2015;Tsiplakou, 2017), while in Standard Greek they appear preverbally if the verb is finite and postverbally with imperatives or gerunds, which do not have tense features (imperatives are only marked for aspect and agreement): ...
... 10. In Cypriot Greek clitics appear in second position when the first item in the clause is (a) the verb, (b) some item in CP or below, e.g., the Mood marker [na], the future/mood marker [ˈenːa], the negation marker [en], wh-expressions, focus clefts (see below) and most complementisers; there is variation between proclisis and enclisis with preverbal subjects, preverbal topics and the complementisers [ˈoti] 'that' and [ʝaˈti] 'because'(Agouraki, 2001;Chatzikyriakidis, 2012;Leivada et al., 2017;Tsiplakou, 2017). ...
... AG is the most typical example of Standard Modern Greek, the official variety of Greek taught at schools and employed in the administration in Greece and Cyprus. The two varieties differ primarily in acoustic representation, although the other linguistic domains may also vary (Chatzikyriakidis, 2012;Grohmann and Leivada, 2012;Terkourafi, 2001). Sonorant sounds differ in the two varieties, as shown by early dialectal research on Greek dialects (Menardos, 1894;Newton, 1972aNewton, , 1972bVagiakakos, 1973b) and by a recent study by Themistocleous (2019) that used artificial neural networks and identified these two varieties based on a single sonorant sound with 81% classification accuracy. ...
This study aims to determine the effect of language varieties on the spectral distribution of stressed and unstressed sonorants (nasals /m, n/, lateral approximants /l/, and rhotics /r/) and their coarticulatory effects on adjacent sounds. To quantify the shape of the spectral distribution, we calculated the spectral moments from the sonorant spectra of nasals /m, n/, lateral approximants /l/, and rhotics /r/ produced by Athenian Greek and Cypriot Greek speakers. To estimate the co-articulatory effects of sonorants on the adjacent vowels' F1 - F4 formant frequencies, we developed polynomial models of the adjacent vowel's formant contours. All sonorants (including /m/ and /n/) had distinct effects on adjacent vowel formant contours, especially for F3 and F4. The study highlights that the combination of spectral moments and coarticulatory effects of sonorants determines sonorants' linguistic (stress and phonemic category) and sociolinguistic (language variety) characteristics. It also provides the first comparative acoustic analysis of Athenian Greek and Cypriot Greek sonorants.
... AG is the most typical example of Standard Modern Greek, the official variety of Greek taught at schools and employed in the administration in Greece and Cyprus. The two varieties differ mainly in their phonetics but there are also morphosyntactic, lexicosemantic, and pragmatic differences (Chatzikyriakidis, 2012;Grohmann & Leivada, 2012;Terkourafi, 2001). Sonorant sounds differ in the two varieties, as shown by early dialectal research on Greek dialects (Menardos, 1894;Newton, 1972aNewton, , 1972bVagiakakos, 1973b) and by a recent study by Themistocleous (2019) that used artificial neural networks and identified these two varieties based on a single sonorant sound with 81% classification accuracy. ...
Full-text available
The aim of this study is to determine the effect of language varieties on the spectral distribution of stressed and unstressed sonorants (nasals /m, n/, lateral approximants /l/, and rhotics /r/) and on their coarticulatory effects on adjacent sounds. To quantify the shape of the spectral distribution, we calculated the spectral moments from the sonorant spectra of nasals /m, n/, lateral approximants /l/, and rhotics /r/ produced by Athenian Greek and Cypriot Greek speakers. To estimate the co-articulatory effects of sonorants on the adjacent vowels' F1 - F4 formant frequencies, we developed polynomial models of the adjacent vowel's formant contours. We found significant effects of language variety (sociolinguistic information) on the spectral moments of each sonorant /m/, /n/, /l/, /r/ (except between /m/ and /n/) and on the formant contours of the adjacent vowel. All sonorants (including /m/ and /n/) had distinct effects on adjacent vowel's formant contours, especially for F3 and F4. The study highlights that the combination of spectral moments and coarticulatory effects of sonorants determines linguistic (stress and phonemic category) and sociolinguistic (language variety) characteristics of sonorants. It also provides the first comparative acoustic analysis of Athenian Greek and Cypriot Greek sonorants.
... Clitic placement is a case in point. In Standard Greek pronominal object clitics appear preverbally if the verb form has tense features, but postverbally if the verb form is an imperative or a gerund (which are only marked for aspect and agreement); in contrast, Cypriot Greek displays clitic-second/Wackernagel or, alternatively, Tobler-Mussafia effects (Chatzikyriakidis 2012;Mavrogiorgos 2013;Neokleous 2015): However, in the sentences in (21)-(23) we have some typical examples of exceptional clitic placement or unexpected standard-like proclisis: tin eθeˈorun in lieu of the expected Cypriot clitic-second structure eθeˈoɾun tin 'I considered it' in (21); to ˈeʃi in lieu of the clitic-second structure ˈeʃi to in (22); and mas efoˈitʃazen 'he would scare us' in lieu of the expected cliticsecond structure efoˈitʃazen mas in (23) : 11, Tsiplakou 2017 Such structures are by now quite frequent in oral corpora (Leivada et al. 2017 and their frequency is confirmed by experimental studies (Pappas 2014; see Figure 3 below), while similar trends towards proclisis occur in child language (see Grohmann et al. 2017 for a summary of relevant research). The fact that preverbal clitics occur in otherwise Cypriot phonological and morohological environments suggests that exceptional clitic placement is becoming a part of the grammatical system of the koine, contributing to its hybrid character. ...
... Of special interest here is the syntactic variation in the data from the other core syntactic area where Cypriot differs radically from Standard Greek, namely pronominal clitic placement. As is well-known, the generalization is that in Standard Greek proclisis (clitic placement in the immediately preverbal position) depends on the finiteness of the verb form, hence gerunds and imperatives trigger enclisis; Cypriot Greek displays clitic-second/ Wackernagel or, alternatively, Tobler-Mussafia effects (Horrocks, 1990;Terzi, 1999;Agouraki, 2001;Condoravdi and Kiparsky, 2002;Pappas, 2004Pappas, , 2014Revithiadou, 2006Revithiadou, , 2008Tsiplakou, 2006;Chatzikyriakidis, 2010Chatzikyriakidis, , 2012Mavrogiorgos, 2010Mavrogiorgos, , 2013Grohmann, 2011;Neokleous, 2015;Grohmann et al., 2017 among others). As with the other variants, while the Cypriot structure, enclisis, was the preferred option, the standard-like strategy of proclisis without a triggering element in C or below, i.e., exceptional clitic placement, was certainly present in that extensive sample of Cypriot Greek oral production. ...
Full-text available
This article explores the concept of gradient bilectalism by capitalizing on insights from recent developments in second language acquisition, particularly the suggestion that aspects of the syntax–discourse interface that are not easily accessible to the learner may lead to fossilization, even at end state. I explore the implications of this suggestion for bilectal grammars by examining the ways in which speakers of Cypriot Greek do syntactic focusing in Standard Greek. The phenomenon is structurally different in the two varieties of Greek examined: clefting is the Cypriot syntactic focusing strategy par excellence while in Standard Greek the relevant strategy is movement of the focused item to an immediately preverbal position. Interestingly, this focusing strategy is largely unattested in the acrolectal or standard-like production of bilectal Greek Cypriot speakers; on the contrary, the preferred strategy for syntactic focusing appears to be clefting, as is indicated by data from spontaneous speech. Quantitative data from a questionnaire survey presented in this article confirm that such “residual clefting” persists even at end state, which in turn suggests imperfect acquisition of the relevant structural aspect of Standard Greek, the second variety of these otherwise bilectal speakers. The data invite an approach couched within the Interface Hypothesis, and the argument is put forward that, being a structure at the interface between syntax and other modules or cognitive domains (semantics, pragmatics, and discourse), focusing in the target variety is vulnerable as regards acquisition.
... As one final step, the parsing of de, which is obligatorily associated with building an inverse LINK transition across from such a completed propositional node to a node in a distinct tree on which its value is dependent, duly builds such a transition from the node decorated with the propositional formula projected from the immediately preceding clausal sequence now to be constructed as LINKed to a node of type e. This might seem at this juncture to be precluded because such a relation between nodes is already in place, however, in the DS framework, nothing prevents a tree from being induced more than once: it is merely that such a construction step cannot be kept distinct from that first instance of the relation, as exactly the same tree configuration will result (see Chatzikyriakidis and Kempson 2011;Chatzikyriakidis 2012;Chatzikyriakidis and Gibson 2017 for detailed justification). In all such cases, there is thus harmless collapse of the two putatively distinct transitions as long as the decorations induced by this duplicating process are compatible with its earlier construction. ...
This paper addresses the challenge of Chinese cleft structures, involving a pairing of the particles shi and de, which in different combinations display a variety of focus-related effects and different potentials for ambiguity: clefts and pseudo-clefts in particular differ only in order of the elements. We argue that retaining conventional assumptions necessarily involves positing unrelated structures and multiple ambiguities, leaving the systematicity of variation unexplained; and we go on to argue that it is only by turning to a dynamic framework in which syntax is defined as mechanisms for incremental build-up of interpretation that an integrated characterisation of these effects is made possible. Adopting the Dynamic Syntax framework (Cann et al 2005), we argue that shi and de induce procedures for incremental build-up of construal which feed and can be fed by other such procedures; and we show how the array of effects both in clefts and pseudo-clefts can be shown to follow from the dynamics of building up interpretation reflecting online processing.
... Like him, we remain open on the issue of whether such representationalism may turn out to be ultimately eliminable. 8 For example,Kempson and Kiaer (2010) for Korean multiple scrambling,Gregoromichelaki (2013b) andChatzikyriakidis (2010Chatzikyriakidis ( , 2012Chatzikyriakidis ( , 2017 for various clitic related phenomena in Greek and dialects of Greek,Bouzouita (2008) for the diachrony of the Spanish clitic system, Cann (2011) for idiosyncracies of the English auxiliary system,Cann and Wu (2011) for idiosyncracies in Chinese passive forms,Sehraku (2013) on Japanese clefts and the restricted optionality of case marking,Gibson (2015) for idiosyncratic left-peripheral word orderings in Rangi, a typologically exceptional Bantu language. ...
... (Grohmann, 2014a, p. 4) CG and SMG differ most obviously in their phonetics, (morpho)phonology, and lexicon (e.g., Newton, 1972;Theodorou, 2007;Arvaniti, 2010). As for morphosyntax, there are also a large number of differences, but it is clitic placement that has arguably drawn the greatest attention (e.g., Agouraki, 1997Agouraki, , 2001Terzi, 1999a,b;Revithiadou, 2006;Revithiadou and Spyropoulos, 2008;Chatzikyriakidis, 2010Chatzikyriakidis, , 2012Pappas, 2012Pappas, , 2014. The following provides a brief overview of clitic placement in the two varieties. ...
Full-text available
This paper examines the development of object clitic placement by children acquiring Cypriot Greek. Greek-speaking Cyprus is sociolinguistically characterized by diglossia between two varieties of Greek, the local Cypriot Greek and the official Standard Modern Greek. Arguably as a result of this situation, clitics may be placed post- (enclisis) or preverbally (proclisis) in the same syntactic environment; while the former is a property of Cypriot Greek, the latter is typically considered an effect of the standard language. The following issues are investigated here: (a) how such bilectal speakers distinguish between the two Greek varieties with respect to clitic placement; (b) how the acquisition of clitics develops over time; (c) how, and which, sociolinguistic factors determine clitic placement; and (d) how schooling may affect clitic placement. To address (a)–(d), a sentence completion task was used to elicit clitic productions, administered to 431 children around Cyprus ranging from 2;8 to 8;11. The C5.0 machine- learning algorithm was employed to model the interaction of (socio-)linguistic factors on the development of clitic placement. The model shows that speakers acquire the relevant features very early, yet compartmentalization of form and function according to style emerges only as they engage in the larger speech community. In addition, the effects of sociolinguistic factors on clitic placement appear gradually.
The theoretical domain of investigation of this volume is the nature of the syntax-phonology interface. The empirical domain of investigation is cliticization in South Slavic. The volume also examines several phenomena that raise theoretical issues related to those involved in South Slavic cliticization, namely, multiple wh-fronting in Slavic and Romanian, Germanic V-2, object shift and stylistic fronting in Scandinavian, and negation in Romance.The central theoretical questions considered in the volume are how syntax and phonology interact with each other and whether PF can affect word order. It is argued that PF does affect word order, but not through actual PF movement.The volume makes new proposals concerning the structural representation of clitics and the nature of clitic clustering. It also provides an account of the second position effect and teases apart the role of syntax and phonology in cliticization and the second position phenomenon.
"Lexical-Functional Syntax" is the definitive text for Lexical-Functional Grammar in the field of syntax.