ArticlePDF Available

Clitics in Grecia Salentina Greek: A dynamic account



In this paper, we attempt an analysis of Grecia Salentina Greek (GSG) clitics within the Dynamic Syntax (DS) framework (Kempson et al., 2001; Cann et al., 2005). We argue that the use of a parsing oriented framework can provide us with an account of a number of puzzling phenomena regarding GSG clitics and clitics in general. In specific, we will argue that the proclisis–enclisis alternation is the result of two different parsing triggers being available in the lexical entries for GSG clitics. The parsing triggers posited will be argued to be the outcome of a routinization process in the sense of Pickering and Garrod (2004), with the pragmatics atrophying over time within such a process (Bouzouita, 2008a). We furthermore discuss sequences of clitics, and argue that strict dat–acc ordering can be accounted by assuming an analysis where dative clitics as well as 1st/2nd person accusative clitics compete for the first fixed node in the tree, while 3rd person accusative clitics on the other hand do not. These assumptions will straightforwardly account for strict dat–acc ordering. Finally the Person Case Constraint (PCC) is taken to derive from the same ordering principles that give rise to dat–acc ordering by further positing that 1st/2nd person accusative like dative clitics compete for the first fixed position in the tree structure.
Clitics in Grecia Salentina Greek: A Dynamic Account
Stergios Chatzikyriakidis
King’s College, London
In this paper, we attempt an analysis of Grecia Salentina Greek (GSG) clitics within the Dynamic Syntax (DS) framework
(Kemspon et al., 2001; Cann et al., 2005). We argue that the use of a parsing oriented framework can provide us with an
account of a number of puzzling phenomena regarding GSG clitics and clitics in general. In specific, we will argue that
the proclisis - enclisis alternation is the result of two different parsing triggers being available in the lexical entries for GSG
clitics. The parsing triggers posited will be argued to be the outcome of a routinization process in the sense of Pickering and
Garrod (2004), with the pragmatics atrophying over time within such a process (Bouzouita, 2008a). We furthermore discuss
sequences of clitics, and argue that strict DAT-ACC ordering can be accounted by assuming an analysis where dative clitics as
well as 1st/2nd person accusative clitics compete for the first fixed node in the tree, while 3rd person accusative clitics on the
other hand do not. These assumptions will straightforwardly account for strict DAT -ACC ordering. Finally the Person Case
Constraint (PCC) is taken to derive from the same ordering principles that give rise to dat-acc ordering by further positing
that 1st/2nd person accusative like dative clitics compete for the first position in the tree structure.
Keywords: Dynamic Syntax; Clitics; Person Case Constraint; Greek Syntax
1. Introduction
Grecia Salentina Greek (GSG) is a term we are going to use in this paper to refer to one of the main varieties of the Greek
dialect Grico. This variety is spoken in a number of villages around the area of Lecce in Italy. This area is usually called
Grecia Salentina so we will refer to this variety as Grecia Salentina Greek. We do that in order to avoid data inconsistencies
with the other main variety, namely Calabrian Greek1. Even though these two varieties seem to agree to most (all?) features
regarding clitic positioning, there is a possibility that Calabrian Greek is different with respect to a number of features
regarding the PCC2. In that respect, and since we do not want to overgeneralize, we will base our paper on the analysis of
the clitic system of the variety spoken in the Grecia Salentina region. The data will be drawn from a small fieldwork visit
undertaken in July 2007 in 3 of the Greek speaking villages in the area of Grecia Salentina, namely Kalimera, Martignano
and Sternatia. Additional data sources will be cited when used.
The paper is structured as follows: In section 2, the distributional properties of GSG clitics3are presented. In section 3,
we briefly discuss the existing analysis on GSG. Before we proceed to the actual DS analysis, we firstly present the relevant
framework (section 4). Lastly, in sections 5 to 8 we present our account of GSG clitics using the Dynamic Syntax grammar
Email address: (Stergios Chatzikyriakidis)
1See Katsoyannou (1995) for a description of this variety.
2There is also the possibility that Calabrian Greek is different with respect to restructuring verb climbing, since in Calabrian Greek the use of the infinitive
is more widespread than in GSG. In that respect, it wouldn’t be implausible to expect restructuring verb climbing to be more generalized in this variety.
3It should be noted that an array of terms are used in this paper pre-theoretically in order to make communication with the readers easier and do not involve
any framework dependent assumptions. In that respect, we will use the terms ”clitic, clitic climbing, in situ, restructuring, pro - drop” in a pre-theoretical
sense, i.e. without necessarily accepting the standard assumptions that have given rise to these terms inthe first place.
Preprint submitted to Lingua March 13, 2009
2. The data
The clitic system of GSG resembles to a high degree the system of Standard Modern Greek (SMG). Clitics immediately
precede the verb except in cases where an imperative verb is present. In the latter case the clitic is encliticized to the verb:
(1) Ton
’He/She/It loves him.’
(2) *Gapa
loves ton.
’He/She/It loves him.’
(3) Tin
’I hit her.
(4) *Kopanitzo
hit tin.
’I hit her.
(5) Grafe
’Write it.’
(6) *To
’Write it.’
In a sequence of two clitics the order must obligatorily be DAT-ACC 4. The latter restriction holds for imperatives as well
(unlike SMG):
(7) Tu
’I gave it to him.’
(8) *To
’I gave it to him.’
(9) Tis
’I said that to her.’
(10) *To
’I said that to her.’
(11) Do
’Give it to me.
4In GSG thedative function is morphologically realized as genitive. In this paper,we will consistently use theterm ’dative’ instead of ’genitive’ in order
to make communication with the readers easier.
(12) *Do
mecldat it
’Give it to me.
(13) Pe
’Say it to me.’
(14) *Pe
’Say it to me.’
Nothing can intervene between the clitic and the verb. In a sequence of two clitics, the clitics must be adjacent:
(15) *Ton
now kopanitzo.
’Now I hit him.’
(16) *Tu
’I gave it to him.’
One rather idiosyncratic phenomenon of GSG is the existence of obligatory climbing with the modal verb sotzo ’can’.
It is crucial to note that the complement of the latter verb must be an infinitive, whereas the general verbal complementa-
tion strategy for all the other so-called restructuring5verbs in GSG involves the use of finite verbal forms preceded by the
subjunctive marker na6.
(17) To
can avorasi.
’We can buy it.’
(18) *Sotzume
can to
’We can buy it.’
(19) *Sotzume
can avorasi
’We can buy it.’
Climbing with other verbs that do involve climbing in a number of languages (Italian, Spanish) is not possible in GSG:
(20) Telume
want no(na-to)
’We want to buy it.’
(21) *To
want na
SUBJ avorasume.
’We want to buy it.’
The Person Case Constraint, widely spread across a number of different languages (Italian, Spanish, Georgian, Kiowa,
Basque to name a few), is also relevant for GSG. In GSG no 1st/2nd person accusative clitic can co-occur with a dative clitic:
5Restructuring verbs include modal, aspectual and motion verbs
6We won’t discuss Climbing in this paper. See Chatzikyriakidis (In preparation) for an analysis of obligatory climbing in GSG.
(22) *Su
’He gave me to you.’
(23) *Tu
’He gave me to him.’
(24) *Dizze
’Show me to her.
GSG, like SMG and unlike Italian, is a clitic doubling language. In that respect both Clitic Left Dislocation (CLLD) and
Clitic Doubling are grammatical7:
(25) To
’I know George.’
(26) Ta
know ta
’I know how things are.’
3. Previous analyses
The only analysis we know regarding GSG clitics is Condoravdi and Kiparksy (2002). This analysis however is a gen-
eral analysis for all Modern Greek (MG) dialects, and in that respect no dialect is analyzed in depth. But let us see what
Condoravdi and Kiparksy (2002) actually propose.
3.1. Condoravdi and Kiparsky
Condoravdi and Kiparksy (2002) propose a tripartite classification of clitic systems for all MG dialects based on the status
that clitics are assumed to have in each dialect:
(27) Type A dialects: Xmax clitics, syntactically adjoined to a maximal projection (Cappadocian Greek - CAG).
Type B dialects: X0clitics, syntactically adjoined to a lexical head (Pontic Greek - PG ).
Type C dialects: Lexical clitics, affixed words (SMG, GSG).
According to the above classification, GSG clitics like SMG clitics fall under category C. Clitics in that category as argued by
Condoravdi and Kiparsky (2002), are lexical affixes, word- to - word affixes to be more specific. The same phrase structure is
assumed to underlie all three categories. Type C clitics, being word to word affixes attach lexically to the left of a finite verb,
with subsequent head movement of the verb plus the clitic to Tns0as shown below:
7We won’t discuss any of these phenomena in this paper. For a DS analysis of CLLD and Clitic Doubling in SMG see Cann et al. (2005), Chatzikyriakidis
(2006) and Gregoromichelaki (Forthcoming).
(28) CP
Spec C’
Spec ΣP
Spec Σ
Spec’ TNS’
cl+V ...
Condoravdi and Kiparksy (2002) adduce evidence that clitics are in fact agreement affixes in type C dialects by comparing
them to subject agreement affixes. The latter are present only in finite verbal forms, a fact also true of clitics as Condoravdi
and Kiparksy (2002) argue. The exact reasoning is that clitics do not combine lexically with non - finite forms. However, this
is not true since clitics are possible with non - finite verbal forms, namely imperatives and gerunds or only gerunds assuming
that imperatives are finite. No matter what assumptions we make with respect to imperatives, such an analysis will give us
the wrong results. Assuming imperatives are non finite verbal forms, we should explain why clitics are possible with these
forms. Assuming they are finite verbal forms, the generalization that Condoravdi and Kiparsky (2002) propose, i.e. clitics
in C dialects attach lexically to the left of a finite verb, won’t capture the imperative case. Whatever the analysis of clitics
with non finite forms and whatever their stance on the imperative issue is, Condoravdi and Kiparsky will need two different
assumptions to cover the whole range of phenomena. This will basically mean that clitics are affixes with finite verbal forms
and something different with non finite verbal forms, something rather unmotivated. Furthermore, there is no discussion
about sequences of clitics and how more than one clitic lexically adjoins to the verb.
Summarizing, Condoravdi and Kiparksy (2002) propose an analysis that tries to capture the whole range of clitic
phenomena found in the dialects of MG by classifying these into three major categories with respect to their linguistic status.
However, their analysis does not go into the specifics of each dialect. In that respect, the peculiarities of each clitic system are
not discussed. In our case, a number of issues regarding GSG clitics remain open. Since there is no other analysis regarding
GSG clitics8we move on to propose our DS based analysis of GSG clitics. But before we do that, we will first briefly and
informally present the DS framework.
8There is a vast literature on clitics in SMG (Agouraki, 1993; Sportiche, 1993; Anagnostopoulou, 1994; Phillipaki, 1994; Philippaki and Spyropoulos,
1999; Terzi, 1999a,b; Mavrogiorgos, To appear, among others) which are to a large extent relevant to GSG. However it is not our intention to go into the
specifics of a number of analyses within the GB/minimalism tradition but rather to argue that once we shift to a parsing oriented framework a number of
puzzling phenomena receive a natural explanation. In that respect, we won’t discuss these analyses here.
4. An informal introduction to Dynamic Syntax
4.1. Basic intuitions behind DS
The basic assumption behind DS is that natural languages are interpreted via an incremental, word-to-word, left-to-right
cumulative construction of transparent semantic representations with the upper goal to finally construct a logical form of
type t (?Ty(t)). Such an interpretation is driven by means of monotonic tree growth, representing the attempt to model
the way information is processed in a time-linear, word-to-word manner. However, tree-structures in DS are considerably
different from those found in derivational or declarative frameworks like minimalism or HPSG respectively, in that they
are not inhabited by words as such, but rather from the representations of those words (Fodor, 1975). Furthermore, the
tree structure corresponding to the representation of the ending result of parsing a natural language string is a semantic
representation assigned to this natural language string with respect to some context. This semantic representation does not
correspond to word order but rather represents argument structure. However, the incremental left-to-right parsing via an array
of successive, monotonically growing tree structures, handles word order through the mere definition of incremental parsing.
The partial tree structures or the history of parsing stages are used to capture word order phenomena, since this whole process
is totally dependent on the way words are ordered9. In order for all these intuitions to be carried out, a number of formal tools
are employed.
4.2. The formal framework in a glance
4.2.1. LOFT, Tree decorations, Requirements
The parsing process is represented by means of binary trees underpinned by the Logic Of Finite Trees (LOFT, Blackburn
and Meyer - Viol, 2001). Left branches are addressed conventionally by adding 0 to the value of the mother node, while right
branches by adding 1. The position of a given node is expressed using the predicate Tn (standing for treenode) followed by
the actual treenode address. Furthermore two basic tree modalities, <>and <>, standing for the mother and daughter
relationship respectively, allow relations between the trees to be represented:
<1> T n(01),
<0> T n(0)
<1><0> T n(00)
<0><1> T n(011)
<1> T n(01)
Notice that a given treenode can be addressed from the perspective of a different treenode. For example <0><1>
T n(011) in the 010 node reads as follows: You will find treenode 011 if you first go up the 0 mother relation and then go
down the 1 daughter relation. The symbol, found in the 0 node in our example is called the pointer, and its basic function is
9An anonymous reviewer wonders if there is a different syntactic component apart from the parser in DS. The answer is that for DS the parser is the
grammar. Parametric variation in natural languages is accounted via the interactions of the lexical entries of a given language with general computational
and pragmatic actions. Computational actions are also subject to parametric variation. Some languages may exhibit different variants of computational
actions while others not. Furthermore, generation is assumed to involve the same mechanism assumed modelling parsing. We do not have the space to
provide examples of parametric variation in DS but the interested read will find vast examples of such variation in Kempson et al (2001), Cann et al (2005).
Furthermore, anyone interested to see the assumptions DS makes regarding the parser should consult Kemspon et al. (2001), Purver and Kempson (2004),
Cann et al (2005), Purver et al. (2006).
to track which node is the current node under construction any time during the parsing process10. Nodes in DS are inhabited
by a set of labels, conventionally called ”Tree Decorations”. The basic elements comprising this set are:
a. Formula value decorations. These are represented using the predicate Fo followed by the representation of the entity
in brackets, e.g. Fo(John’)11.
b. Type value decorations. These are represented using the predicate Ty followed by the type of the word/concept in
question in brackets, e.g. Ty(e)12.
A basic concept in the DS framework is that of requirements. Requirements can be seen as goals that need to be achieved.
Requirements have the general form ?Lai13 (e.g. ?Ty(e)). In order for a given parse to be successful, no outstanding require-
ments must exist in the ending tree. In that respect, requirements can be also seen as a device explaining ungrammaticality.
Example (29) shows a complete tree in DS. Notice that no outstanding requirements exist14:
Ty(e) Fo(Love’)(Maria’)(y),
Ty(e t)
Ty(e) Fo(Love’)(x)(y),
Ty(e (e t))
4.2.2. Computational - Lexical - Pragmatic rules/actions
The parsing process is driven by three kinds of rules/actions: a) Computational b) Lexical and c) Pragmatic rules/actions.
The former are general computational devices, comprising the basic tree construction mechanism. They always 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 or 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 etc. An example of a computational rule, the rule of COMPLETION, is
shown below:
(31) {...{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), ...}...}
The above rule moves the pointer to the mother node as soon as any type of requirement is satisfied on any of the daughters
of that mother node. The top part reads as follows: There is a node with treenode address Tn(n) and another one dominated
by Tn(n) ( <i>, T n(n)) that has a satisfied type requirement and also bears the pointer. The output description (second
line) presents a situation where the pointer has moved to the Tn(n) addressed node, with an additional statement that records
10The Pointer function is also one of the ways to account for ungrammaticality in DS. For instance, if the pointer is at a given node which has an
outstanding requirement for a type e expression to be found, and the next word that is parsed does not provide such a Type, by providing e.g. a Ty(et)
expression, the parse will abort rendering the whole string ungrammatical.
11The prime indicates that the concept and not the word John is represented.
12The difference between DS and most of the formal semantic theories with respect to typing is twofold. Firstly, DS has an additional type (cn) standing
for common noun, and furthermore there is no recursion on types. Types are a rather closed set. For a more detailed discussion on DS typing see Kemspon
et al. (2001), Cann et al. (2005).
13Where La stands for label and i>1. For a formal presentation of declarative units in DS consult Kemspon et al. (2001), chapter 7.
14The lambda terms in the Fo formulas have been excluded for ease of exposition.
15Where i=(0,1,*).
the daughter’s satisfied requirement (<i> T y(X)). There are a number of procedural computational rules like the one we
have just seen but we won’t go into the rest of them for reasons of space. The interested reader is referred to Kempson et al.
(2001), Cann et al (2005) for a detailed presentation of a number of computational rules. Additional computational rules will
be introduced if needed.
On the other hand, lexical rules are basically entries associated with a given word providing instructions on how the
parsing must or must not proceed. There are no general rules regarding the content of these instructions. They rather depend
on the syntactic nature of these words. However, there is a generalized schema involved in the way these words introduce
their content. This general procedural schema followed by a sample lexical entry is shown below:
(32) Lexical entry format
IF Trigger
T H EN Actions
ELSE Elsewhere statement
(33) Sample lexical entry for Bill
IF ?T y(e)
T H EN put(T y(e), F o(Bill0)),[]
ELSE Abort
Example (33) reads as follows: If you are in a node that has a type e requirement, then decorate this node with a type e
value and a formula value representing the concept given by the word ’Bill’. In any other case abort. 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. Other languages with overt noun case marking will have further restrictions
for their equivalent entry for Bill that will ensure that the proper noun will be parsed in the relevant node depending on case
marking. For example we can associate a given case marking with a fixed structural position by means of tree modalities as
shown below:
(34) Structural accusative lexical entry
IF ?T y(e), <1>?T y(et)
T H EN put(T y(e), F o(x0)),[]
ELSE Abort
The above entry will block a noun of the above type to be parsed in the subject node (00) simply because the condition
?T y(et)is not going to be satisfied16.
Lastly, pragmatic actions involve contextual information providing information with respect to the parsing process. One
very basic rule is the rule of SUBSTITUTION which updates a formula metavariable into a proper formula referring to some
entity in the context17. We won’t discuss any other pragmatic actions in this paper but the interested reader is directed to
Kemspon et al. (2001), Cann et al. (2005) for further information on pragmatic actions.
4.2.3. Underspecification, LINK
Central within the DS framework is the concept of underspecification, the idea that parts of natural language may not
be fully specified by the time they enter into the parsing procedure. Of course underspecification is not in itself a new
concept in linguistics. It has been extensively used the last 20 years by formal semanticists to deal with ambiguity and
16In DS binary trees, as already mentioned, do not encode word order but rather represent argument structure. In that respect the subject node is always in
the same position no matter what the actual word order is. This position is the 00 node. Given that, it is clear why the condition is not satisfied.
17See Kemspon et al. (2001), Cann et al. (2005) for a formal definition of the rule of SUBSTITUTION.
anaphora resolution. What is novel however, is that underspecification is moved into the area of syntax18, making the syntax
the dynamic part rather than the semantics. DS uses two types of underspecification: a) Content underspecification and b)
Structural underspecification. With respect to content underspecification, DS employs the use of metavariables, functioning
as mere content placeholders with a requirement that substitution of the metavariable will take place at later stages of the
parse. A classic example of content underspecification is pronouns. DS assumes that the lexical entry for a pronoun, say
he, involves the projection of a metavariableas the value the Formula takes. This metavariable must be updated as soon as a
proper formula value is provided by the context or bythe natural language string itself. The metavariable comes with person
and case restrictions depending on the pronoun. A requirement that a proper Fo value must be found ensures that the node
which bears the metavariable will eventually get a proper formula value. The lexical entry for the pronoun he is shown below:
(35) Lexical entry for he
IF ?T y(e), <1>?T y(t)
T H EN put(T y(e), F o(Umalesg ),?x.F o(x),[])
ELSE Abort
Structural underspecification on the other hand is represented in DS by employing a set of rules, the family of ADJUNC-
TION rules19. *ADJUNCTION effectively introduces a node, which position in the tree is not yet fixed at the time of its
introduction. To be more specific, the rule of *ADJUNCTION projects such an unfixed node from the initial ?Ty(t) node
which bears a requirement for an expression of type e to be found at that node:20
(36) {...{T n(a), ......?T y(t),3}}
{...{T n(a), ...?T y(t)},{<↑ ∗ > T n(a),?x.T n(x), ..., ?T y(e),3}}
The effect of the rule is shown schematically below:
<↑ ∗ >?T y(t),
An NP can be parsed on that unfixed node satisfying the type e requirement:
(38) Parsing ton Jani ’John’ on an unfixed node21
<↑ ∗ >?T y(t),
18There is however a similar notion, the notion of functional uncertainty in LFG (Bresnan, 2001).
19We will present two of the various variants of the ADJUNCTION rule here. Additional *ADJUNCTION rules will be introduced later on if needed.
20The kleene star operator is a way to encode underspecification in the modal language. <↑ ∗ > X reads as: X holds at a node above me or at the
current node. Using the opposite modality, i.e the daughter modality, the kleene star denotes the notion of dominance plus reflexiveness. The pure notion of
dominance is encoded by means of the kleene plus operator (+). In that respect <+> X reads as: There is a node above me where X holds. We will see
later on how we will exploit both of the operators in our analysis.
21We ignore determiners for the moment.
The ?x.Tn(x) restriction will ensure that the node must be fixed at a later stage in the parse22. The underspecified relation
<↑ ∗ >?T y(t)will enable the NP to be parsed in different structural positions. The *ADJUNCTION rule is a natural way to
encode this intuition. The *ADJUNCTION rule will account for long scrambling cases as well, since the tree modality used
does not restrict the full NP to apply in the local domain. A variant of *ADJUNCTION however, *LOCAL ADJUNCTION
will do just that, i.e. it will restricts the potential fixing sites of the node to local nodes. The rule is shown below:
(39) {...{T n(a), ..., ?T y(t),3}...}
{...{T n(a),?T y(t),}...{<0><↑ ∗ > T n(a),?x.T n(x), ..., ?T y(e),3}...}
The effect of the rule in tree notation is the following:
<0><↑ ∗ >?T y(t),
Notice that the modality has changed from <>to <0><
1>. This will ensure that the NP in question is parsed in
the local propositional domain 23. The two rules are used for long and short distance scrambling effects respectively. We will
see later on the relevance ofthese rules with respect to clitics.
While the ADJUNCTION rules involve the creation of an unfixed node that has a requirement for a specified treenode
address in the tree under construction to be found, LINK structures involve the construction of a second tree structure inde-
pendently of the initial one, which howeverposits a requirement for a shared term between the two trees. In order for LINK
structures to be modelled, we need to introduce two new modal operators, < L > and < L1>. The former refers to a
tree structure which is linked, as it is shown schematically in (43), by means of an arrow, with the current node, while the
latter refers to that node. LINK rules are also a family of rules, sharing the characteristics just mentioned. For demonstration
purposes we will present one of them. The latter comes in the form of two rules, the rules of TOPIC STRUCTURE INTRO-
DUCTION and TOPIC STRUCTURE REQUIREMENT 24 respectively. These two rules are used by (Cann et al., 2005) to
account for Hanging Topic Left dislocation structures(HTLD). The first rule effectively creates a LINK transition between the
initial node and a top node with a type e requirement. The rule is shown below:
(41) {{T n(0),?T y(t),}}
{{T n(0),?T y(t)}},{< L > T n(0),?T y(e),}
Notice that the above rule does not mention anything about a shared term. That is because there is no shared term at the
time the rule applies. The requirement for a shared term is introduced via the second rule the rule of TOPIC STRUCTURE
REQUIREMENT shown below. This rule applies as soon as the dislocated argument is parsed (as for Mary in our example)25:
(42) {{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)}
After the introduction of these two rules, we get the following:
22?x.Tn(x) reads as: There is a requirement for a proper treenode address (fixed) to be found. If the latter does not happen, then the parse cannot be
completed as at least one outstanding requirement will exist in the tree.
23Assuming that all argument nodes are the 0 nodes and an additional propositional domain will involve a type t in one of the argument nodes, this rule
will exclude cases where the NP is associated with an argument in the next propositional domain.
24The prototypical LINK rule is the rule of LINK *ADJUNCTION used by Cann et al. (2005) to account for relative clauses. We choose to present the
TOPIC STRUCTURE INTRODUCTION and TOPIC STRUCTURE REQUIREMENT rules instead, which are basically a variant of the prototypical LINK
rule. The interested reader is referred to Cann et al. (2005 - 85-94) for the prototypical LINK rule.
25The D modality stands for the downward modality that encodes the kleene star operator and can furthermore extend over LINK structures. In that
respect D is defined as D={↓0,1,,, L}.
(43) hLiT n(0),
F o(Mary0),
T y(e)
hL1iT n(n),?T y(t),?hDiF o(M ary 0)
In an HTLD construction, the dislocated NP will be parsed in the node where the LINK begins. After TOPIC STRUC-
TURE REQUIREMENT has applied, a requirement for the same Formula value provided in the first tree will be put in the
LINKed tree. This will ensure that a copy of the formula Mary will also be provided by the linguistic string, for example a
resumptive pronoun in English. There are a number of important things with respect to the general LINK rule, but we won’t
discuss them here since these are not relevant to the scope of the paper. The interested reader is however referred to Kempson
et al. (2001) and Cann et al. (2005) for more information on the various LINK rules.
5. A DS analysis - One clitic constructions
For DS, as already mentioned, natural language parsing is triggered by three types of actions, namely computational,
lexical and pragmatic actions. Lexical actions are basically the lexical entries that words are assumed to have. Clitics in our
analysis will have their own lexical entries. However, this does not mean that we a priori decide on the status of GSG clitics as
words. For DS every linguistic element, no matter how it is traditionally characterized, can be associated with a lexical entry.
It is often assumed in various DS analyses that affixes, that is traditionally non - words, can have their own lexical entries
(see for example the lexical entries for Japanese case suffixes given by Cann et al., 2005). Such a move is not contradictory
assuming that for DS lexical entries are just pieces of information on how parsing must or must not continue. In that respect,
any element that provides such information can have its own lexical entry. Within this line of reasoning, the affix - word
debate (Philippaki, 1998; Philippaki and Spyropoulos, 1999; Condoravdi and Kiparksy, 2002, among others) can be easily
side - stepped as irrelevant with respect to the DS framework, since we can give an account that will effectively be the same
no matter what our stance regarding the affix - word debate is. We will present an analysis of GSG clitics where the different
distributional properties of these are captured via the interactions of the lexical entries we will give for clitics and general
computational actions. These interactions are further underpinned by general pragmatic actions. But let us see in detail how
this can be done.
Clitics in GSG, resemble SMG clitics in that they appear preverbally with all verbal forms except imperatives. In the
latter case, they must appear postverbally. In ensuring proclisis, we need a mechanism that will not alow a clitic to be parsed
after a verb has already done so. In DS terms this means that no verb must exist when the clitic will enter the parsing process.
There is a rather straightforward way to do that. We will use the third person neuter accusative clitic to ’it’ to illustrate how.
In our brief introduction to the DS framework we have said that DS does not directly represent word order, but rather uses
binary trees to represent argument structure. In that respect, the position of the verbal argument in the tree is always the same
no matter the word order the sentence exhibits. As mentioned, what we need to exclude is the case where a verb has been
parsed first. On the other hand, we do want to include a number of cases where a preverbal subject or object has been parsed
first. We will propose two ways to do that, and we will see which one is more plausible as more phenomena are examined.
The first proposed restriction is shown below:
(44) First proposed restriction
IF ?T y(t)
T H EN I F [+]?x.T n(x)
T H EN ...
ELSE ...
ELSE ...
The above restriction reads as follows: If you are in a node which bears a type t requirement, then if all the nodes below
that node bear a requirement for a proper treenode address to be found, i.e. all the nodes dominated by the type t bearing node
are unfixed, then proceed to the actions (which we have not given yet). Such a restriction will give us the desired results. The
clitic cannot be parsed if a verb has already done so, since the verb will project a number of fixed nodes if parsed (See the
entry for indicative verbs in (63)) . It can however when a preverbal object has, since the restriction takes into consideration
only nodes that have already found their fixed position in the tree structure. A preverbal object will always involve either
an unfixed node or a LINKed structure. These will find their proper treenode addresses only after the verb is parsed. The
next thing we need to discuss is whether 3rd person accusative clitics in GSG are better analyzed as unfixed or as projecting
a fixed structure (Bouzouita, 2008b) for Spanish,(Chatzikyriakidis, 2006) for SMG). An unfixed node analysis for the 3rd
person accusative clitic will immediately imply that the clitic, being structurally unfixed, can receive more than one structural
position in the emerging tree. Based on the data we have for GSG, an accusative clitic is always linked to the direct object.
Our informants in Grecia Salentina noted that there are no double accusative verbs in GSG 26. Given these facts, the lexical
entry for the accusative neuter clitic will involve the projection of fixed structure:
(45) 3rd person accusative neuter clitic to ’it’
IF ?T y(t)
T H EN I F [+]?x.T n(x)
T H EN makeg o(<1>), makeg o(<0>);
put(F o(Vneut,sg ), T y(e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
Let us comment on the embedded THEN part of the rule a little bit. This is where the actual actions induced from the
lexical entry are encoded. The clitic builds both the 01 and the 010 node. It annotates the 010 node with a type e decoration and
projects a formula metavariable to be substituted from context or from the natural language string. The metavariable carries
presuppositions (i.e. the subscripts) which must be satisfied by the candidate value in order for the latter to be accepted as the
metavariable’s update. These presupositions are in effect restrictions on metavariable update. The second proposed trigger
on the other hand, involves a statement about the potential functor nodes only:
(46) Second proposed trigger
IF ?T y(t)
T H EN I F [+
1]?T y(x)
T H EN ...
ELSE ...
ELSE ...
The above restriction states that all functor nodes should bear a type requirement in order for the clitic to get parsed. In
case a verb is parsed, this will decorate a functor node with a type value. In that respect the above trigger won’tget satisfied,
and the clitic will be unable to get parsed. The full lexical entry is shown below:
(47) Alternative entry for to ’it’
IF ?T y(t)
T H EN I F [+]?T y(x)
26Katsoyannou (1995) mentions that in the other main Grico variety, Calabrian Greek, double accusative constuctions are restricted to just one verb, the
verb ”matenno/learn” and even in that case they are extremely rare (only one instance is found in her corpus). Even in SMG where a number of double
accusative verbs exist, a number of complications arise when we substitute full NP’s with clitics in these constructions. For example it is impossible to
substitute both of the accusative marked full NP’s with two accusative marked clitics. In case we do that, the clitic functioning as an indirect object must be
genitive marked. We remain agnostic as to what is going on in these constructions in SMG and we won’t discuss this issue further. The interested reader is
directed however to Anagnostopoulou (2001) for a detailed discussion on double accusative verbs in Modern Greek.
T H EN makeg o(<1>), makeg o(<0>);
put(F o(Vneut,sg ), T y(e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
Both of the entries we have given correctly predict sentences (1) and (2), repeated here as (48) and (49), to be grammatical
and ungrammatical respectively:
(48) Ton
’He loves him’.
(49) *Gapa
loves ton.
’He loves him.’
Furthermore, cases where a preverbal object or a subject has been parsed are also predicted to be grammatical. The
examples below are correctly captured by the entries we gave:
(50) Ton
’He loves George.’
(51) O
’He loves him.’
Assuming that preverbal objects are parsed using *ADJUNCTION or *LOCAL ADJUNCTION, no node with a proper
treenode address will exist below the initial type t requiring node. No type value will exist in any of the functor nodes either.
In case a preverbal subject is parsed, a LINKed or an *ADJUNCTed structure will be used. Thus, in that case, a separate tree
structure or an unfixed argument node is involved:
(52) After parsing a preverbal object
<↑ ∗ >?T y(t),
(53) After parsing a topic subject
hLiT n(0),
F o(Giorgos0),
T y(e)
hL1iT n(n),?T y(t),?h↓iF o(Giorgos0),
The two lexical entries we’ve given correctly capture preverbal positioning of the 3rd person accusative clitic to ’it’.
Before proceeding to the entries for genitive and 1st/2nd person accusative clitics, first we need to account for postverbal
clitic positioning with imperatives.
5.1. Enclitic positioning with imperatives
5.1.1. Minimalist analyses
It is a rather puzzling phenomenon that clitics in a wide range of languages, while proclitic in general, appear postverbally
with imperatives. A number of different proposals have been given for enclisis with imperatives within chomskyan frame-
works (Phillipaki, 1994; Rivero and Terzi, 1995; Terzi, 1999a,b, among others). The standard assumption in these analyses
is that imperatives in Greek exhibit movement to a functional projection past TP. This functional projection is CP in Rivero
(1994), Rivero and Terzi (1995), Terzi (1999a,b) and MP in Philippaki (1994). Clitics in these approaches are situated below
these projections or they move along with the verb to this projection as in Terzi (1999b). In that respect, what we get is
In their classic paper, Rivero and Terzi (1995) examine the behaviour of the imperative in Ancient Greek (AG), SMG and
Cypriot Greek (CG). The crucial difference between AG on the one hand and SMG - CG on the other according to Rivero
and Terzi (1995), is that in the latter two cases imperatives obligatorily move to C to satisfy their own requirements, while
in the former case they optionally move to the Wackernagel phrase (WP) to satisfy requirements of the W phrase. Let us see
what the proposed analysis actually predicts for CG and SMG. Rivero and Terzi (1995) assume the same clausal structure for
both SMG and CG. In this analysis negation is situated between C and the IP. Assuming clitics are situated in the IP domain,
V - C movement will result in a situation where the verb precedes the clitic, thus the enclisis. In the presence of negation,
movement is blocked assuming relativized minimality (Rizzi 1991). In order to account for the differences in positioning in
CG and SMG while retaining the same clausal architecture, Rivero and Terzi (1995) have to further assume that the clitic
must be licensed in the domain of a head with operator like properties in CG but not in SMG. In the absence of such a head,
the verb has to move to a position below C that Rivero and Terzi do not specify. However,in Rivero (1994) and Terzi (1999a)
such a head is specified to be a Mood Phrase (MP). The plausible question to ask given the latter fact, is why imperatives do
not move to that MP projection that licenses Mood features as its name suggests, but rather move to C instead. It seems worth
emphasizing that the same argument was raised by Roussou (2000) with respect to the same analysis. The answer seems to be
that if we abandon the V-C movement of imperatives and adopt a unified V-M movement approach for all verbs in CG, then
assuming as Rivero (1994) and Terzi(1999a,b) that NegP is higher than MP, we would expect negated imperatives to be licit.
A further problem with such an analysis concerns the status of the so - called operator like heads. One major discrepancy
involves the assumption that the complementizer pu carries operator like properties while oti does not. Without getting into
the debate of whether such a claim is true, Rivero and Terzi (1995) do not mention that complementizer oti exhibits variation
in positioning in CG as reported in Revithiadou (2006) and Chatzikyriakidis (In preparation)27. Under these variation data,
Terzi and Rivero have to assume that in some cases oti does and in some cases does not behave as an operator, something
rather strange. A further problem for such an analysis comes from languages like Later Medieval Greek (LMEG), a language
which under Rivero and Terzi (1995) should fall within the same category as CG, namely under the type of languages showing
imperatival distinct syntax as the lack of negated imperatives in LMEG would suggest (Pappas 2002). Assuming however the
same analysis as in CG, i.e. V-C movement of imperatives, then an apparent problem comes from LMEG examples where
proclisis with imperative verbs is possible in the presence of a focused element in the left periphery:
(54) Ala
(Pappas - 2001: 95).
’Tell me something else.
(55) Aγia
said (Pappas - 2004: 70).
’Call her holy.
Terzi (1999a) arguesfor the existence of a MoodP in Modern Greek as well. In fact, this is the projection where gerunds
move to. Then, as in the case of CG, the same question can be asked for SMG, i.e. why the presence of MP does not trigger
movement of imperatives to the M0and imperatives have to rather raise all the way up to C to check their mood features.
We won’t discuss these analyses any further since it is not our intention to give a detailed account of the analyses proposed
in minimalism or to further elaborate any of them. Our intention was to show that the existing analyses concerning Greek
27An anonymous reviewer asks whether this variation has something to do with the positioning of oti with respect to foci/topics. The data presented in
Revythiadou (2006) as well as those we have collected do not indicate any such correlation.
imperatives are not devoid of problems and to further propose an alternative DS analysis. The interested reader is however
referred to Phillipaki (1994), Rivero (1994), Rivero and Terzi (1995), Terzi (1999a,b) and Roussou (2000) for more details
and argumentation on the discussed analyses.
5.2. Imperatives in DS
Clitic positioning in DS, as we have already seen, is captured by imposing restrictions on the entry for the clitic that refer
to the current parse state of the tree. In the case of imperatives, an imperatival feature (IMP) is going to be used as a second
disjunctive trigger in the entry for clitics.
Assuming that an imperatival verb will project such a feature in the type t requiring node, enclitic positioning is effectively
captured assuming this second disjunctive trigger. The new entry for the 3rd person neuter accusative clitic ’to’ in GSG is
shown below:
(56) Entry for the 3rd person accusative clitic to ’it’ (imperatives included)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
IF Mood(Imp)
T H EN makeg o(<1>), makeg o(<0>);
put(F o(Vneut,sg ), T y(e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
(57) Alternative entry for the 3rd person accusative clitic to ’it’ (imperatives included)
IF ?T y(t)
T H EN I F [+]?T y(x)
IF Mood(Imp)
T H EN makeg o(<1>), makeg o(<0>);
put(F o(Vneut,sg ), T y(e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
Both the above lexical entries capture enclisis with imperatives. However, we should explain why an imperative verb
cannot be parsed after a clitic has already done so. Assuming a lexical entry in the same style as the lexical entry for an
indicative verb, i.e. an entry in which the verb starting from the initial node builds the whole propositional structure and
furthermore decorates the subject node with a type value and a formula metavariable, ungrammatical examples like (58) are
predicted to be grammatical. Nothing will stop an imperative verb to be parsed after a clitic in the same sense as an indicative
verb will:
(58) *To
’Show it!’
What we are going to argue is that imperative verbs (at least for SMG and GSG28) have a similar restriction to the one
we have given in the first of the two entries for the 3rd person accusative clitic. This means that imperatives compete for the
first fixed node in the domain. Assuming the ACTIONS for the 3rd person accusative clitic29, the imperative won’t be able
to get parsed due to that restriction. The rather peculiar characteristic of enclisis with imperatives can be attributed to the
impossibility of imperative verbs to be preceded by fixed nodes. In that respect the clitic has to follow rather than precede the
imperative. What is the nature of such restriction is something that we should discuss however. Looking back at the history
of clitic distribution from LMEG to the MG dialects, we can derive a natural explanation for this restriction. Looking for
example at LMEG, we do find categorical restrictions regarding clitic positioning but in most cases what we find is tendencies
or almost categorical restrictions (Pappas, 2004). For example, in the case of imperatives there is a clear tendency for enclitic
positioning, but this is by no means a categorical restriction as we have already seen in examples (59) and (60), repeated
(59) Ala
(Pappas - 2001: 95).
’Tell me something else.
(60) Aγia
said (Pappas - 2004: 70)
Call her holy
The fronted constituent is probably the reason for proclisis in the above examples. It is quite interesting to observe that no
Modern Greek dialect presents any variation with imperatives anymore. CG, a dialect which pretty much resembles LMEG
in terms of clitic distribution does not allow proclisis with imperatives no matter what the nature of the preceding element is:
(61) TO
’Give me the book.’
(62) *TO
’Give me the book.’
Enclisis with imperatives is categorical in the dialect we are examining as well. . In that respect, we can assume that the
entries for imperative verbs obtained a triggering restriction that would prevent proclisis. In the dialects we are interested
in, all proclitic triggers have collapsed into one and all clitic distribution is defined solely with respect to the verbal form.
However, the imperative restriction was retained in the entry for the clitics. The latter, given an entry for imperatives that
bans proclisis, gives us the desired distribution, i.e. a situation where enclisis is categorically restricted in case an imperative
is present. In order for proclitic triggers to collapse, what was needed was an overgeneralization of proclitic environments.
We argue that the pragmatic basis behind clitic positioning with fronted focused constituents was lost, giving rise to proclisis
in a number of other environments involving non focused objects and subjects, all adverbs and PP’s. The whole process can
be described as a stepwise routinization process in the sense of Pickering and Garrod (2004) with the pragmatics atrophying
over time within that process in the sense of Bouzouita (2008a,b). Such a process eventually led to a generalization of
proclisis with finite forms, effectively leading to a collapse of the preverbal triggers to just one. The fact that imperatives lack
subordinate conjunctions, or any of the particles triggering proclisis (na ’subjunctive marker’, min ’NEG’) and are less likely
to be preceded by focused constituents, is probably what prevented proclisis to further extend to imperatives. In that respect,
what we get is as system with two triggers, one for imperative and one for non imperative verbs. Within such an approach,
dialects like Pontic Greek (PG) which are strictly enclitic, followed the opposite route, i.e. a route where proclitic triggers
did not collapse to a general proclitic trigger but rather disappeared, while enclisis on the other hand was generalized to all
environments. Thus, in the case of PG the form of the verb does not play any role with respect to positioning anymore. This
could be made possible in the transition from Medieval Pontic Greek (MPG) to PG, since MPG as shown by Pappas (2001)
28We discuss imperatives in dialects where in general enclisis is the rule, i.e. CG and PG, in Chatzikyriakidis (In preparation).
29We will see how the clitics that will be treated as unfixed will be accommodated under that assumption.
displayed a system where the proclitic triggers where considerably fewer compared to LMEG or even Medieval Cypriot
Greek (MCG). Assuming that triggers are something like parsing heuristics, i.e. facilitators of the parsing process, both of
these evolutions led to greater simplicity by collapsing an array of disjunctive triggers into just two or one in the case of PG.
However, the fact that some languages collapsed a set of disjunctive triggers in favor of simplicity does not mean that all
languages or dialects are expected to behave like that. It is true that there is a strong tendency towards that direction, but there
are a number of Greek dialects which would immediately falsify such a claim, notably CG and Cappadocian Greek (CAG -
See Janse 1994). The latter dialects, with some degree of deviation, exhibit pretty much the same patterns in terms of clitic
positioning that LMEG did. In fact, these clitic systems seem to be rather stable, since almost all patterns of distribution of
LMEG are still obeyed in both dialects. However, this is not exactly the case. For example, comparing MCG to CG, someone
can easily realize that a number of proclitic triggers present in CG are absent in MCG, namely temporal expressions and
fronted constituents (Pappas, 2001, 2004). In that respect, the CG clitic system resembles more the LMEG system that its
Medieval ancestor did. Thus, despite the first appearances, the CG system has been in a state of change like all the other
MG dialects. The direction of change in each case seems to depend highly on the medieval ancestors of each of the modern
Returning to the actual way of encoding the triggering restrictions in DS and assuming the heuristic like role of parsing
triggers, it is highly expected that these triggers will be encoded in the same heuristic like way. For example, moving to a
dynamic syntactic model where each word is parsed in relation to the current state of the partial tree at the time these words
come into parse, every word will have to ensure that the given tree satisfies some restrictions imposed by its entry that will
correctly account for its distributional behavior. In the case of GSG, two different triggers are imposed, one that aborts in
case a verbal type is present in the tree and one that proceeds to the actions in case an imperative feature is present on the
initial type t requiring node. The former will give us proclitic distribution, while the latter enclitic distribution. Furthermore,
imperatives in GSG have a trigger, similar to the one we have given for third person accusative clitics in (44), that aborts in
case any fixed nodes are present in the tree. This will correctly predict that enclisis is categorical with imperatives in GSG.
The entries for an indicative monotransitive and an imperative monotransitive verb are shown below:
(63) Entry for an indicative monotransitive verb
IF ?T y(t)
T H EN put(T ns(T)), M ood(Ind);
makego(<1>); put(?T y(et));
makego(<1>); put(F o(verb0)),
T y(ee(t),[]);
go(<1>), makeg o(<0>); put(?T y(e));
go(<0>)(<1>), makeg o(<0>);
put(T y(e), F o(Upersnum),?x.F o(x))
ELSE Abort
(64) Lexical entry for an imperative monotransitive verb 30
IF ?T y(t)
T H EN I F [+]?x.T n(x)
T H EN put(Mood(Imp));
makego(<1>); put(?T y(et));
makego(<1>); put(F o(verb0)),
T y(ee(t),[]);
go(<1>), makeg o(<0>); put(?T y(e));
30Both verbs impose a restriction with respect to the update of the subject nodemetavariable. An indicative verb will project person and number restrictions
F o(Upersnum)while an imperative verb presumably only anumber restriction F o(Unum).
go(<0>)(<1>), makeg o(<0>);
put(T y(e), F o(Unum ),?x.F o(x))
ELSE Abort
ELSE Abort
The restriction [+]?x.T n(x)in the lexical entry for the imperative verb will prevent it from being parsed if any node
with a proper treenode address exists below the initial node when this comes into parse. This means that such a verb won’t
be able to get parsed after a clitic or in general any element that will project a fixed node(s) is parsed31. Example (58) is ruled
out with our new triggering restriction. It is easy to see why. The accusative clitic will have built the 01 and 010 node when
the imperative verb will enter the parsing procedure. Two nodes with proper treenode addresses will exist, and thus parsing
of the imperative verb will be blocked.
Concluding this section we should note that the explanation we provide with respect to the emergence of the clitic system
of GSG cannot be fully justified, unless a thorough investigation of the transition process from regional dialects of Medieval
Greek that gave rise to the dialect under question has been done. Specifically, within such an approach, one should expect
to find a stage in GSG or its ancestral Medieval dialect where proclisis gets generalized to more environmentsand enclisis is
gradually reduced. Indeed, this is what we find in 18th century Spanish (Bouzouita 2008a, 2008b). Such an investigation is
still in its first stages and we cannot say anything yet. We won’t pursue this issue further, since firstly it is not our intention
to give a full diachronic account of the facts in GSG and secondly as we’ve said earlier we have not yet the results of an
investigation regarding the Medieval clitic system of GSG.
6. Ordering of a sequence of two clitics
As we’ve already mentioned, GSG exhibits strict DAT -ACC32 order in both imperative and non imperative environments.
Examples (7), (8) and (11), (12) repeated here as (65), (66) and (67), (68) respectively, clearly illustrate the latter fact:
(65) Tu
’I gave it to him.’
(66) *To
’I gave it to him.’
(67) Do
’Give it to me.
(68) *Do
’Give it to me.
The first thing we need to do in order to capture DAT-ACC order, is to think how a lexical entry for genitive clitics should
be. Accusative clitics are always associated with the direct object node, so their position is fixed in the tree. For genitive clitics
on the other hand, one might argue (at least for GSG and SMG) that their interpretation is ambiguous between a direct and
an indirect object interpretation. Constructions involving the verb areso ’like’ are classic examples of constructions requiring
their sole object to be marked for dative. The following example is grammatical in both SMG and GSG:
31A welcoming result is that auxiliary verbs will be predicted to be impossible with imperatives, assuming that auxiliaries will project fixed nodes after
they are parsed. See Chatzikyriakidis (In preparation) for an analysis of auxiliaries in DS.
32We note once again that the two terms dative and genitive will be used interchangeably to denote the indirect object function. The reason for that is that
dative is morphologically realized as genitive in GSG.
(69) Mu
I like that
A small number of other monotransitive verbs also subcategorize for dative in both GSG and SMG:
(70) Tu
I talk to him
(71) Tu
I called him
This positioning duality of dative clitics can be accounted assuming that genitive clitics as opposed to accusative clitics
are structurally underspecified with respect to the position they will eventually occupy in the tree structure. However, there
are a number of restrictions as regards these positions. The first one is that the clitic must be interpreted locally. The other
restriction is that dative clitics can never function as subjects. These two facts have rather straightforward solutions within
DS. The first restriction can be implemented in DS by encoding the rule of *LOCAL ADJUNCTION in the THEN part
of the clitic’s entry. This will ensure that the clitic will be underspecified but will however fix its position locally33. On
the other hand, if we want to exclude clitics from being interpreted as subjects, we need to posit a variant of the *LOCAL
ADJUNCTION rule that will effectively do that. The underspecified address <0><
further restricted to <0><1><
1>, in order to avoid clitics being updated in the subject node. The next thing we have
to think is how we will ensure that dative clitics are always first in a sequence of two clitics. Assuming that all preverbal
arguments in GSG will use one of the *ADJUNCTION or LINK rule variants to get parsed and that accusative clitics will
always project fixed structure, the restriction ( [+]?x.T n(x)) we have used in the first entry we have given for the 3rd person
accusative clitic to ’it’ will give us the ordering facts. Putting all these assumptions together we get the following lexical
(72) Lexical entry for the first person dative clitic mu (preverbal cases only)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
T H EN makeg o(<1><
1>), makego(<0>);
put(F o(VSpeaker0), T y (e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
(73) Lexical entry for the first person dative clitic mu (imperatives included)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
IF Mood(Imp)
T H EN makeg o(<1><
1>), makego(<0>);
33The rule says that the node where the clitic is, is a node where one step across the 0 node must be taken in order to be reached. This means that the
clitic will be always interpreted as an argument ofthe local domain. The clitic cannot extend to another domain since a different domain will either involve
a LINK structure or more than one steps across the 0 daughter relation to be taken. See Cann et al. (2005) for a definition of locality in DS.
put(F o(VSpeaker0), T y (e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
Given the above entry, it is now a good chance to test the two alternative entries for the third person accusative clitic ((56)
and (57)) against the data. At a first glance, it seems that both the entries will give us the correct results with respect to DAT
-ACC ordering. Assuming a dative clitic has been parsed first, the triggering restrictions of both accusative entries seem to
be satisfied. The dative clitic, being an unfixed node, will satisfy the [+]?x.T n(x)restriction and thus an accusative clitic
will be able to get parsed after it. On the other hand, since the dative clitic won’t project any type value in any of the functor
nodes, the second restriction is also satisfied ([+
1]?x.T y(x)). Thus, it seems that both entries we’ve given for the third
person accusative clitic would work. Taking a closer look at the triggers however, someone will realize that this is not the
case, and in fact only the second entry is able to capture DAT -ACC order. Let us explain. The restricted version of *LOCAL
ADJUNCTION we’ve given in the THEN part of the entry for the dative clitic specifies that the parser should construct the
01 node followed by the construction of an unfixed node which is an arbitrary number of steps below the 1 relation, followed
by a step down the 0 relation. This means that the parser will always construct the 01 node. What effectively happens in that
case is, that even though we have an unfixednode projected by the clitic, a fixed node is projected too. If however the parser
always builds the 01 node in parsing a dative clitic, then the triggering point of the first entry we have given for the 3rd person
accusative clitic ([+]?x.T n(x)) won’t work, since a fixed node will have already been constructed, i.e. the 01 node. No
problem arises for the second entry we have given, since no type value will exist in any of the functor nodes after a dative
clitic has been parsed. We will thus keep the second alternative for the third person accusative clitic, even though it may
seem nicer from a framework internal point of view to have the same restriction for both clitics. We will see the relevance of
keeping this second alternative when we will discuss 1st/2nd person accusative clitics.
What is left now, is to see whether the above lexical entry can capture DAT-ACC ordering with imperatives as well. It is
easy to observe that this is not the case. Nothing stops a dative clitic to be parsed after an accusative clitic has already done so
in an imperative construction. Indeed, this is the case for SMG where clitic ordering in imperatives is free. The above lexical
entry will thus work for SMG but not for GSG. A minimal modification in the entry can however treat this overgeneration:
(74) Lexical entry for the first person dative clitic mu (imperatives included)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
IF Mood(Imp),[+
1][0]?T y(x)
T H EN makeg o(<1><
1>), makego(<0>);
put(F o(VSpeaker0), T y (e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
The restriction [+
1][0]?x.T y(x)requires that all object nodes should bear a type requirement. In case an accusative
clitic is parsed, this requirement won’t be satisfied since the accusative clitic will project a type value in the direct object
node. Note that preverbal objects or preverbal strong pronouns are not excluded. Assuming that preverbal objects are parsed
using one of the ADJUNCTION rules with MERGE taking place after the verb has been parsed or via LINK in which case
we are dealing with a different tree structure not accessible by the modalities given, no such problem arises. Notice that this
triggering restriction actually says that the clitic should be the first fixed argument in the tree structure. In that respect it looks
like a more specified version of the [+]?x.T n(x)restriction. The entries for the rest of the singular dative clitics are identical
to the one we’ve given above, provided we do the necessary changes regarding the presuppositions the Fo metavariables have
7. The person case constraint
The PCC is a phenomenon spread in a wide range of both related and unrelated to each other languages (Spanish, Italian,
Bantu Languages,Kiowa, Georgian, Basque to name a few). Various accounts have been given in the literature (Bonet, 1991,
1994; Haspemath, 2004; Anagnostopoulou, 2003, 2005; Adger and Harbour, 2007; Ormazabal and Romero, 2007, among
others). We briefly discuss two of the most recent syntactic accounts given before moving on to the analysis we propose.
7.1. Anagnostopoulou(2003, 2005), Adger and Harbour (2007)
In Anagnostopoulou (2003,2005), the PCC is the result of feature checking failure against one functional head. Both
clitics have to check their features against one functional head, the latter bearing a number and a person feature. Dative clitics
bear person features but not number features, first - second person accusative clitics bear both number and person features
and lastly third person accusative clitics bear only number features. Assuming checking against one head which can check
number and person features once, the only licit combinations are the ones where a 1st/2nd person accusative clitic does not
co-occur with a dative clitic. In a similar vein, Adger and Harbour (2007) argue that the PCC is the result of a first second
person accusative clitic carrying participant features and the presence of an Appl head. The generalization they propose bans
features present in the specifier of a functional head to be used as probes in the complement domain of that same head:
(75) Adger and Harbour’s Generalization
The requirements which a functional head requires its specifier to bear cannot be used as probes in the head’s
complement domain.
A first or second person accusative clitic is then excluded when an APPL head is present (i.e. in ditransitive constructions),
since the participant features of such a clitic will remain unchecked assuming (75). The following tree diagram represents a
ditransitive construction in Kiowa 35:
34Notice that ordering phenomena are dealt inside the lexicon, i.e. by triggering restrictions in the lexical entries and not via general computational rules.
This is a conscious decision, since we do not believe that some sort of universal clitic ordering exists. We would expect a number of other languages with
similar clitic phenomena to be dealt within the same analysis but we do not want to posit any strong universal ordering constraints for clitic sequences, at
least for the purposes of this paper. The interested reader is however directed to Manzini and Savoia (2004) for an interesting discussion plus references on
the clitic ordering issue.
35Kiowa is verb final. With the obvious modifications the tree diagram also applies to MG and Romance.
Each argument as shown above is checked against the closest c-commanding head. Thus, the direct object is checked
against Appl, the indirect object against v and the subject against Asp. In case an applicative head is present, its specifier will
contain participant features. However, this will exclude the possibility of the Appl head to probe for participant features in its
complement domain, thus the PCC.
Both the analyses presented share the intuition that the PCC is caused by feature checking failure. What is however
problematic is whether these features actually exist and if they do, how does someone decide which features to attribute to
which clitic. For example, Anagnostopoulou( 2003), (2005) assumes that third person dative clitics contrary to third person
accusative clitics do carry person features but this person feature they carry is specified as minus (-). In the case of third
person accusative clitics no such feature exists. The plausible question to ask then, is what is the difference between absence
of a feature and its negative specification and why two different specifications with respect to person have to be made for
the two clitic forms. If these specifications are given as such to maintain the proposed analysis, then the generality of the
account is altogether collapsing. At the moment, we do not find any other principled reason besides such a move. The same
reasoning applies to the assumption that dative clitics are defective heads and as such they do not check number features
in virtue of these not being being accessible for checking. This latter assumption is based on participial agreement data
where dative clitics do not trigger participial agreement. However, this last assumption has already been criticized by Bonet
(2007: footnote 14) for Catalan where only third person accusative clitics and not first second person accusative clitics trigger
participial agreement, and thus such evidence cannot be decisive.
Accordingly, in Adger and Harbour (2007) dative clitics are all assigned a participant feature. This is based on the
assumption that all indirect objects are animates. However, we believe that this is a very strong claim to make at least for MG
even though there is definitely a preference for animate indirect objects. However, constructions with an inanimate indirect
object are perfectly grammatical in MG. There is a tendency for animate indirect objects but this fact remains a tendency,not
a categorical restriction36:
(77) Tis eδose mia klotsia (tis kareklas).
herclgen gavepast3rdsg aacc kickacc thedat chairdat
’He kicked the chair.
36The same problem is reported for Catalan by Bonet (2007) in discussing the same analysis.
(78) Tis eδose mia klotsia (tis Marias).
herclgen gavepast3rdsg aacc kickacc thedat Marydat
’He kicked Mary.
Examples where the doubled inanimate NP is not present but is however present in the immediately preceding context are
also grammatical. This should not be the case according to Adger and Harbour (2007) since inanimate objects do not bear
participant features:
(79) A: Pos egine etsi to vivlio? B: Tu eδosa mia klotsia kata lathos.
how happened that the book itcldat gave a kick by mistake
A: ’Why is the book like that? B: I kicked it by mistake.’
(80) A: Pos ton halases ton ipologisti sto grafio? B: Apla, tu esteila enan io.
how itclacc damaged the computer in-the office simply itclgen sent a virus
A: ’How did you manage to destroy the computer in the office? B: I just sent it a virus.
We believe that even though a number of ditransitive constructions involving inanimates are somewhat degraded compared
to the ones involving animate NP’s, a generalization banning inanimates from ditransitive constructions is on the wrong track.
What exactly is going on in these constructions is something that we do not know since the data are far from being clear cut.
For example, there are a number of constructions involving inanimate indirect objects that are if not sharply ungrammatical,
question mark grammatical. Substituting the dative with the preposition se ’to’ plus an accusative NP, the sentence becomes
grammatical. The peculiar thing is that a dative clitic can be used to refer back to the PP construction37:
(81) A: ???ose mia efkeria tis /deltaimokratias. B: Θa tis δoso.
give a chance thedat democracydat FUT hercldat give-I
A: ’Give democracy a chance. B: I will.’
(82) A: ???ose mia efkeria sti δimokratia. B: Θa tis δoso.
give a chance to-theacc democracyacc FUT hercldat give-I
A: Give democracy a chance B: I will.
Further research is needed in order to understand what is the exact correlation between animacy and double object con-
structions. It is clear to us however that a strong generalization like the one given by Adger and Harbour (2007) cannot
be maintained. In that respect, at least for MG, the assumption that all indirect objects are interpreted as animates is rather
7.2. Ormazabal and Romero - 2007
Ormazabal and Romero (2007) dissociate the PCC into two different, according to them, phenomena. On the one hand,
there is a universal tendency of object agreement sensitivity to animacy while on the other hand there is a restriction on
agreement with multiple objects. Ormazabal and Romero (2007) go through data from a number of languages arguing that
the PCC, should be split into the following two generalizations :
(83) Object animacy realization Object relations, in contrast to subject and applied object relations, are sensitive to
(84) Object agreement constraint
If the verbal complex encodes verbal agreement, no other argument can be licensed through verbal agreement.
37An anonymous reviewer mentions that (81) improves if the order of the two objects is switched. We do not see any difference in grammaticality but
certainly this is something that needs to be further checked. The reviewer also notes that the example in question isnot that bad as it is. If this is so, then our
claim that the animacy restriction with indirect objects is a preference and not a constraint is further vindified.
In particular, the above two generalizations are argued to be adequate enough to capture the complexity that the PCC
exhibits. Ormazabal and Romero (2007) predict that in clitic languages where the PCC is active, it should be active only for
argumental clitics. In that respect, in a construction where the dative clitic is a non - argumental clitic, say an ethical dative,
the PCC should be inactive. Indeed that is what we find in Spanish. However, things are not the same in SMG and GSG since
the PCC remains active with ethical datives as well. The examples below from SMG clearly exemplify the latter claim:
(85) Mu
’They killed him (and I’m affected by it).’
(86) *Mu
’They killed you (and I’m affected by it).’
Speakers of GSG indicated the same for GSG. Ormazabal and Romero’s (2007) generalizations are inadequate with respect to
SMG and GSG, since their prediction is that the PCC should not be active with ethical datives in those two dialects. One other
further issue we need to discuss with respect to Ormazabal and Romero (2007) is the account they give regarding the third
person Spanish clitic lo. Since their second generalization does not allow two objects to agree with the verb, they argue for a
non - agreement, determiner - like analysis for lo to explain the grammaticality of sentences involving two argument clitics,
where the accusative clitic is lo.They use a number of examples to prove that lo does not agree with the verb. They present
data from doubling, to prove that lo is indeed a different kind of clitic compared to dative or 1st/2nd person accusative clitics.
The latter can only double in particular environments, and when they do they must be interpreted as [+]specific. However,
extending Ormazabal and Romero’s (2007) explanation to MG,once more things don’t seem to work. The reason is that the
equivalent MG clitic for lo, at least for SMG38, can double at least the same phrases that first and second person accusative
clitics do. The first two sentences are ungrammatical in Spanish, but however grammatical in SMG
(87) To
saw to
the spiti.
’We saw the house.’
(88) Tus
saw merikus
some sto
in maγazi.
’We saw some of them in the shop.’
(89) Mas
imagine merikus
some sti
in filaki.
’I imagine some of us in prison.’
(90) Tha
FUT sas
see merikus
some avrio.
’I will see some of you tomorrow.
As for specificity, even though it has been argued that indeed this is the case for MG doubling constructions, (Iatridou,
1995; Anagnostopoulou, 1997) there are plenty of sentences involving a bare quantifier plus doubling which do not exhibit
any specificity effect or optionally exhibit a specificity effect39:
(91) Polus
many anθropus
people den
NEG tus
’Many people do not care’ (specific or non - specific).
38I do not have any data from GSG regarding this construction. The clitic doubling data I have so far suggest that GSG shows pretty much the same be-
haviour with respect to clitic doubling. However, this needs to be further checked. CG on the other hand follows the same pattern as SMG (Chatzikyriakidis,
In preparation).
39The same observations are noted in Kallulli (2000).
(92) Mia
one kokini
red bluza
blouse ti
want afton
this ton
the kero.
’I need a red blouze at this time of the year’ (specific or non - specific).
(93) Merika
drink apopse.
’I would have some drinks tonight’ (specific or non - specific).
(94) Mia
one bluza
blouse θa
FUT tin
’I would buy a red blouse’ (non specific only).
(95) Opjoδipote
whichever traγuδi
song tu
whistle, tha
FUT to
’He will play any song you will whistle to him.’
The above examples from SMG suggest that n determiner like analysis of third person accusative clitics loses its empirical
support and cannot maintained at least for a number of MG dialects including the dominant SMG variety. What we will
attempt to do next, is to present an account of PCC within the DS framework based on the same assumptions we have already
made in our analysis of clitics so far.
7.3. A DS analysis
We’ve already given the entries for dative and 3rd person accusative clitics. Since we want to address the PCC, what is
left is to give the lexical entries for 1st/2nd person accusative clitics. Remember that when we were discussing the entries
for 3rd accusative clitics, two alternative entries were given (examples (56) and (57)) until one of them was rejected when
the entries for dative clitics were eventually given. The acute reader will remember that the reason for excluding the first of
the two entries, i.e. the entry shown in (56), was that it fallaciously predicted sentences like (7) repeated below as (96) to be
(96) Tu
giveP ast1stsg
’I gave it to him.’
This is exactly what we need in order to capture the PCC. We need an entry that will ban any combination of a 1st/2nd
accusative clitic with a dative. In that respect, the first entry we’ve given for third person accusative clitics turns out to be
relevant for the PCC. We will in that respect posit a lexical entry for the 1st/2nd person accusative clitics in the same lines we
did for the first of the two entries we proposed for 3rd person accusative clitics. The entry is shown below:
(97) Entry for the 1st person accusative clitic me (imperatives included)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
IF Mood(Imp),[+
1][0]?T y(x)
T H EN makeg o(<1>), makeg o(<0>);
put(F o(VSpeaker0), T y (e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
Notice that both 1st/2nd accusative as well as dative clitics involve the same restriction in their entry. The two clitics
compete for the first fixed node in the tree. If there is such a node, neither clitic can be parsed. With this assumption both the
PCC and ACC-DAT ungrammaticality is captured. On the other hand, the lexical entry we eventually kept for the third person
accusative clitic does not have the ”first fixed node” restriction but instead posits a different restriction. According to this
restriction every functor node must bear a type requirement, i.e. no verb must have been parsed by the time the clitic comes
into parse. In that respect, DAT -ACC order with 3rd person accusative clitics is predicted to be grammatical. ACC-DAT
order is blocked due to the ”first fixed node” requirement that dative clitics bear. Note that 1st/2nd person accusative clitics,
like dative clitics, carry an additional restriction to be the first fixedarguments in the tree in an imperative construction. This
will capture the PCC effects with imperatives as well. Generalizing we can say that the PCC is the result of competition
between 1st/2nd person accusative clitics and dative clitics for the first fixed position. In case of proclisis, this restriction is
stated as the first fixed node in the tree. Since imperatives also compete for the first fixed node, the restriction is modified
from ”first fixed node” in the tree to ”first fixed argument” in the tree in enclitic environments. It is obvious however, that the
first restriction entails the second i.e. the second restriction can be seen as a more specified version of the first.
One might wonder how can such an account explain the nature of the constraint under consideration. An anonymous
reviewer asks for example what do dative and first and second person accusative clitics have in common and thus exhibit the
same triggering restrictions. The answer is that identical parsing triggers can be used for a number of elements in DS that do
not necessarily have anything in common to each other. For example both a verb and a clitic in GSG will involve the same
initial triggering restriction, namely that the pointer should be at a type t requiring node in order for these to be parsed. On
the other hand, restrictions like the ”first fixed node” restriction can be seen as parsing shortcuts that cover a wide range of
phenomena without actually referring to any of them. Such a restriction will capture the ordering facts with respect to clitics
themselves, the proclitic nature of clitics in non imperative environments while it will predict preverbal objects or subjects to
be possible assuming that the latter will be parsed as either unfixed nodes or LINKed structures. Under the assumption that
the parser is the grammar, we actually expect these parsing shortcuts to exist. We furthermore expect some of these parsing
shortcuts to be the result of routinization processes minimizing parsing costs. In effect, a number of different phenomena
related to clitic positioning boil down to one single parsing restriction, a parsing facilitator, facilitating the parsing process
without actually referring to the phenomena it is meant to capture.
The above discussion naturally poses another very basic question. How general the PCC is, and if a great deal of generality
is indeed involved, how does the proposed analysis account for it? In the literature two versions of the PCC are used. One
is the strong version, the version that GSG clitics exhibit. There is however another version the weak version of the PCC
according to which, 1st/2nd person accusative clitics cannot combine with a 3rd person dative clitic but they can do so with a
1st/2nd person dative clitic. This is true for languages like Catalan even though disputed by some speakers (Bonet, 2007)40:
(98) Te m’ ha recomanat la Mireia.
tesg mesg has recommended the Mireia
a. ’Mireia has recommended me to you.’
b. ’Mireia has recommended you to me.’
There are furthermore languages like Romanian that obey none of the two versions of the PCC (see (Savescu, 2007) for
the relevant data). These types of languages pose a serious threat to analyses that predict PCC to be a universal constraint.
What we rather find, is person restriction tendencies, found in many clitic languages. We should also not forget that there
are clitic languages that do not exhibit any person restrictions like Polish (Franks and King, 2000; Haspemath, 2004)41. Our
claim is that person case restrictions are also the result of routinization. This means that some combinations of clitics became
calcified, routinized clusters while others did not. Since in order for routinization to occur, a relative high frequencyof such
expressions must occur, we concur with Haspemath (2004) that one of the reasons for such restrictions could be frequency
40Example from Bonet (2007).
41An anonymous reviewer asks whether Polish is a clitic language at all. It is true that polish clitics behave differently with respect to positioning to other
clitic languages like Romance or other West Slavic and Slavic in general languages (see Franks and King, 2000). However, such a fact does not mean that
Polish is not a clitic language. We should not forget that the term ”clitic” is an umbrella term for elements that cannot be classified as either words or affixes.
In that respect, it is normal to expect the elements that are characterized as clitics to have different properties in different languages, leaning more or less
towards words or affixes in each case. The exact differences of Polish clitics with other clitic systems is well beyond the scope of this paper. The interested
reader is however directed to Franks and King (2000) for a relevant discussion.
rates42. What other reasons may lie behind such restrictions is something that we do not know. We strongly believe however
that person is not the cause of such a phenomenon. The fact that some clitic sequences are sharply ungrammatical while the
individual elements comprising this sequence are perfectly grammatical on their own, suggests that a number of restrictions
should be employed in the grammar to capture the phenomenon. Different languages in that respect will require different
restrictions in order to capture the relevant facts or the PCC phenomena they might exhibit. There is a great deal of generality
associated with the actions the clitics induce in our analysis (check the THEN actions of accusative clitics for example) but
the triggering point in each case might be totally different. We believe that triggering points can become, as we’ve said,
parsing heuristics, pure facilitators of the parsing process. In that respect, conditions on tree unfolding or on partial trees are
perfectly legitimate. It is a good challenge to check whether our triggering restrictions extend to other languages as well, and
if not, what amount of modification will they need to do so. We cannot do that in this paper for obvious reasons of space. We
will leave this issue open as a subject of future research. See however Chatzikyriakidis (2006, In preparation), Kempson and
Cann (2007), Bouzouita (2008a,b) for DS analyses in other languages or different dialects of MG.
7.4. The case of first and second accusative plural clitics
In GSG, while 1st/2nd person singular clitics are distinctively case marked, i.e. two different forms corresponding to the
dative and the accusative clitic exist, the plural forms of these clitics exhibit case syncretism. In that respect two different
case markings correspond to one form. The table below shows the relevant facts:
1st 2nd 3rd
Sg accusative me se ton/tin/to
Pl accusative ma(s) sa(s) tus/tis - tes/ ta
Sg genitive mu su tu/tis
Pl genitive ma(s) sa(s) tos
Syncretized forms of the GSG clitic system can be straightforwardly accounted in DS, by giving a lexical entry within
the lines of the analysis of singular dative clitics. This entry will effectively be identical to the one we have already given for
singular dative clitics 43 :
(99) Lexical entry for the first person plural dative clitic mas (including imperatives)
IF ?T y(t)
T H EN I F [+]?x.T n(x)
IF Mood(Imp),[+
1][0]?T y(x)
T H EN makeg o(<1><
1>), makego(<0>);
put(F o(VSpeaker0), T y (e),?x.F o(x));
gof irst(?T y(t))
ELSE Abort
ELSE Abort
The syncretized forms in that respect can be very easily accounted in DS using the notion of structural underspecification,
namely encoding a variant of *LOCAL ADJUNCTION into the clitic’s lexical entry as we did in the case of singular dative
42This is however an assumption that must be further elaborated, showing that indeed in older stages of the language the clitic combinations that are now
banned, had full pronoun counterparts with lower frequency rates compared to the full pronoun counterparts of legitimate clitic clusters.
43There is something we should note however. The entry we have given for singular genitive clitics, i.e. an entry encoding locally restricted structural
underspecification, will predict that singular dative clitics will be also able to get parsed as arguments of monotransitive verbs requiring an accusative. There
is nothing to stop such a thing within the analysis given. On the other hand we do believe that the analysis given is in the right track. The overgeneration
caused by our entry can be easily treated assuming that the verb will impose a some kind of a requirement for a case filter to be found on the relevant node.
This case filter will presumably be projected by the accusative clitic but not from the dative clitic. The details of such an analysis are irrelevant to this paper.
The interested reader is referred to Chatzikyriakidis (In preparation) for further details on how this can be done.
8. Conclusion
In this paper we’ve argued for a parsing based analysis for GSG clitics. It was shown that a grammar formalism which
assumes incrementality and underspecification to be a part of the grammar formalism itself can help us account for various
otherwise problematic phenomena regarding clitics. The proclisis - enclisis alteration was argued to be the result of two
distinct parsing triggers being present in the entries for clitics. These two parsing triggers were argued to have derived from
a more complex, highly disjunctive entry via a stepwise routinization process in the sense of Pickering and Garrod (2004)
and Bouzouita (2008a). In general, we have given an analysis of clitics based on the notion of fixed and unfixed nodes. In
that respect, we’ve argued that dative and 1st/2nd person accusative clitics always compete for the first fixed node in the tree.
This immediately gave us the desired results regarding the PCC. On the other hand, 3rd person clitics were argued not to
carry such a restriction and thus do not compete with genitive clitics. Note however that this is asymmetrical, because dative
clitics do compete with them. Thus, DAT -ACC and ACC-DAT orders are predicted to be grammatical and ungrammatical
respectively. Lastly, the syncretised forms for first and second person plural clitics are straightforwardly accommodated in
DS by assuming that these clitics actually encode a weaker version of the *LOCAL ADJUNCTION rule, i.e. they project
unfixed nodes.
I’m indebted to my supervisor, Ruth Kempson, for invaluable comments and constructive critique on various parts of this
paper. I also thank the audiences at King’s college and Queen Mary, for useful comments and discussion. Particularly I want
to thank Eleni Gregoromichelaki, Miriam Bouzouita and Wilfried Meyer - Viol for helpful comments. Furthermore, two
anonymous reviewers are thanked for providing useful comments and critique on earlier drafts of this paper. The Arts and
Humanities Research Council (AHRC), the Leventis Foundation and the Dimitris Koufogiorgos Foundation are gratefully
thanked for providing partial funding related to parts of the work presented in this paper. Of course, any inconsistencies
remain my own.
Agouraki, G., 1993. Spec - Head Licensing: The Scope of the Theory. Phd Thesis, University College London.
Adger, D., Harbour, D., 2007. Syntax and Syncretisms of the Person Case Constraint. Syntax 10, 2-37.
Anagnostopoulou, E., 1994. Clitic Dependencies in Greek. Doctoral dissertation, University of Salzburg.
Anagnostopoulou, E., 1997. Clitic Left Dislocation and Contrastive Left Dislocation. In: Anagnostopoulou, E., Van Riemsd-
jik, H., and Zwarts, F. (eds), Materials on Left Dislocation (Linguistik Aktuell 14). Amsterdam:Benjamins, 151-192.
Anagnostopoulou, E., 2001. Two Classes of Double Object Verbs: The Role of Zero Morphology. In: van Oostendorp, M.,
Anagnostopoulou, E. (eds.), Progress in Grammar: 20 Years of Grammaticamodellen Meertens Institute, electronic book.
Anagnostopoulou, E., 2003. The Syntax of Ditransitives: Evidence from Clitics. Mouton de Gruyter.
Anagnostopoulou, E., 2005. Strong and Weak Person Restrictions: A Feature Checking Analysis. In: L., Heggie and F.,
Ordonez (eds), Clitics and Affixation. Amsterdam : Benjamins, 199 - 235.
Blackburn, P., Meyer - Viol, W., 1994. Linguistics, Logic and Finite Trees. Bulletin of Interest Group of Pure and Applied
Logics 2, 2-39.
Bonet, E., 1991. Morphology after Syntax: Pronominal Clitics in Romance. Doctoral dissertation, MIT.
Bonet, E., 1994. The Person Case Constraint: A Morphological Approach. In: H., Harley and C., Philips (eds), The morpho-
logy Syntax Connection. MIT, Cambridge, MA, 33-52.
Bonet, E., 2007. The Person Case Constraint and Repair Strategies. To appear in: Alessasndro, R., Fischer, S., Hrafnbjanson,
G. H. (eds), Person Restrictions. Mouton de Gruyter.
Bouzouita, M., 2008a. Modelling Syntactic Variation. In: Dialogo de la Lengua, I, 15-39.
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, J., 2001. Lexical - Functional Grammar. Oxford: Blackwell.
Cann, R., Kempson, R., Marten, L., 2005. The Dynamics of Language. Oxford: Elsevier.
Cardinaletti, A., Shlonsky, U., 2004. Clitic Positions and Restructuring in Italian. Linguistic Inquiry 35, 519-557.
Chatzikyriakidis, S., 2006.Clitics in Modern Greek : A Dynamic Account. MSc thesis, King’s college London.
Chatzikyriakidis, S., In preparation. Clitics in Four Dialects of Modern Greek : A dynamic account. Phd thesis, King’s
college, London.
Cinque, G., 1999. Adverbs and Functional heads: A Cross- Linguistic Perspective. New York, Oxford University press.
Cinque, G., 2006. Restructuring and Functional Heads. Oxford Studies in Comparative Syntax, Oxford university press.
Condoravdi, C., Kiparksy, P., 2002. Clitics and Clause Structure. Journal of Greek Linguistics, 2, 1-39.
Fodor, J., 1975. Language of Thought. Harvard university press, Cambridge.
Franks, S., King., T., H., 2000. A Handbook of Slavic Clitics. Oxford University Press.
Gregoromichelaki, E. (forthcoming). 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
Workshop. Forthcoming from John Benjamins.
Haspelmath, M., 2004. Explaining the Ditransitive Person - Role Constraint. Electronic form article. Available at: haspelmt/2004dpr.pdf.
Iatridou, S., 1995. Clitics and Island Effects. In: Bhatt, R., Garrett, S., Han, C. - H., Izvorsky. R. (eds), Penn working papers
in Linguistics 2,11-31.
Iatridou, S., 1996. To Have and Have Not. Proceedings of WCCFL XIV.
Janse., M., 1993. La Position des Pronoms Personnels Enclitiques en Grec Neo-Testamentaire la Lumire des Dialectes Neo-
Hellenique. In: Claude, Brixhe (ed.), La Koine Grecque Antique. I. Une Langue Introuvable. Nancy: Presses Universitaires
de Nancy, 1993, 83-121 (Etudes anciennes, 10).
Kallulli,D., 2000. In Beukema, F., Den Dikken. M., (eds.) Clitic Phenomena in European Languages, 209-248. Amsterdam:
John Benjamins.
Katsoyannou, M., 1995. Le Parler Greco de Galliciano (Italie): Description d’une Langue en Voie de Disparition. Phd thesis,
University of Paris VII.
Kayne, R., 1989. Null Subjects and Clitic Climbing. In: Jeggli, O., Safir, K.J (Eds), The Null Subject Parameter. Dordrecht,
Kempson, R., Meyer-viol, W., Gabbay, D., 2001. Dynamic Syntax: The Flow of Language Understanding. Blackwell Pub-
Kempson, R., Cann, R., 2007. Production Pressures, Syntactic Change and the Emergence of Clitic Pronouns. Ms., King’s
college London.
Manzini, R., Savoia, L., 2004. Clitics: Cooccurence and mutual exclusion patterns. In Rizzi, L. (ed.): The Structure of CP
and IP. The Cartography of Syntactic Structures Vol. 2. Oxford University Press.
Miller, P., Sag, I., 1997. French Clitic Movement without Clitics or Movement. Natural Language and Linguistic Theory 15,
Mavrogiorgos, M., To appear. C - Agreement and the EPP - Deriving Enclisis in Modern Greek. In: Proceedings of the 7th
International Conference on Greek Linguistics.
Monachesi, P., 1993. Object Clitics and Clitic Climbing in Italian HPSG Grammar. In: Proceedings of the Sixth Conference
of the European Chapter of the Association for Computational Linguistics. Utrecht, 437 - 442.
Monachesi, P., 1998a. The Morphosyntax of Romanian Cliticization. In: Coppen P., Van Halteren, H., Teunissen L. (eds.),
Proceedings of Computational Linguistics in The Netherlands 1997. Rodopi, Amsterdam-Atlanta, 99-118.
Monachesi, P., 1998b. Italian Restructuring Verbs: a Lexical Analysis. In: Hinrichs, E., Kathol, A., Nakazawa, T. (eds.),
Complex Predicates in Nonderivational Syntax. Syntax and Semantics 30, Academic Press, San Diego, 313-368.
Monachesi, P., 1999. The Syntactic Structure of Romanian Auxiliary (and Modal) Verbs. In: G., Bouma et al. (eds), Con-
straints and resources in natural language syntax and semantics. CSLI publications, Stanford.
Ormazabal, J., Romero, R., 2007. The Object Agreement Constraint. Natural Language and Linguistic Theory, 2007, 25.
Springer, 315-347.
Pappas, P., 2001. Weak Object Pronoun Placement in Later Medieval Greek: Intralinguistic Parameters Affecting Variation.
In: Ohio State University Working Papers in Linguistics (OSUWPL), 56, 79-106.
Pappas, P., 2004. Variation and Morphosyntactic Change in Greek: From Clitics to Affixes. Palgrave Studies in Language
Change and Language History, Palgrave.
Philippaki - Warburton, I., 1994. V Movement and Clitics in Modern Greek. In:Philippaki - Warburton, I., Nikolaidis, K.,
Sifianou, M. (eds), Themes in Greek Linguistics, Amsterdam: John Benjamins 53 - 60.
Philippaki - Warburton, I., 1998. Functional Categories and Modern Greek Syntax. The Linguistic Review 15,159-186.
Philippaki - Warburton, I., Spyropoulos, V., 1999. On the Boundaries of Inflection and Syntax. Yearbook of morphology
1998, 45-72.
Pickering, M., Garrod, S., 2004. Towards a Mechanistic Psychology of Dialogue. Brain and behavioral science 27.
Purver, M., Kempson, R., 2004. Incremental Parsing, or Incremental Grammar? In Proceedings of the ACL Workshop on
Incremental Parsing. Barcelona, Spain, July 2004, 74-81.
Purver, M.,Cann, R., Kempson, R., 2006. Grammars as Parsers: Meeting the Dialogue Challenge. Research on Language and
Computation 4(2-3), October 2006, 289-326.
Revythiadou, A., 2006. Prosodic Filters on Syntax: An Interface Account of Second Position Clitics. Lingua 116, 79-111.
Rivero, L., 1994. Clause Structure and V- Movement in the Languages of the Balkans. Natural Language and Linguistic
Theory 12, 63 - 120.
Rivero, L., Terzi, A., 1995. Imperatives, V-movement and Logical Mood. Journal of Linguistics, Volume 31, 2, 301-32.
Rizzi, L., 1982. Issues in Italian Syntax. Studies in Generative grammar 11. Foris Publications, Dordrecht.
Roussou, A., 2000. On the Left Periphery: Modal Particles and Complementizers. Journal of Greek Linguistics 2000, vol 1:
Savescu, O., 2007. Challenging the Person Case Constraint: Evidence from Romanian. In: Camacho, J., Flores-Ferran, N.,
Sanchez, L., Deprez, V., Jose Cabrera, M. (eds), Romance Linguistics 2006. Selected Papers from the 36th Linguistic
Symposium on Romance Languages (LSRL), New Brunswick, March-April 2006. John Benjamins.
Sperber, D., Wilson, D., 1994. Relevance: Communication and Cognition. Blackwell publishing.
Sportiche, D., 1993. Clitic Constructions. Ms, UCLA.
Terzi, A., 1999a. Cypriot Greek and their Positioning Restrictions. In: Alexiadou A., Horrocks, G. and Stavrou, M. (eds.),
Studies in Greek Syntax. Springer.
Terzi, A., 1999b. Clitic combinations, their Hosts and their Ordering. Natural Language and Linguistic Theory 17, 85-121.
... On this basis, they induce triggering of goals to build/linearise conceptual mechanisms ('ad-hoc concepts') classified as belonging to ontological types (e for entities in general, e s for events, e → (e s → t) for predicates, etc). In (9) above, the goal is realised as a prediction to process next a proposition of type t. This is shown as a one-node tree with the prediction ?T y(t) and the ICS's current focus of attention, the pointer ♦. ...
... Returning to the processing in (9), now NL-specific constraints kick in since the pointer ♦ is left at the argument node implementing the word-order restriction in English that the object needs to follow the verb: (9) ...hugged... ?T y(e), ♦ Hug ′ : e → (e → (es → t)) At this point, the word Mary can be processed to initiate the tracking of a contextually-identifiable individual (M ary ′ ) at the argument node internal to the predicate [for the view that such concepts are skill adaptations allowing the accumulation of knowledge about individuals, see 67]. After this step, everything is in place for the structural underspecification to be resolved, namely, the node annotated by who can now unify with the subject node of the predicate, which results in an ICS that includes the minimal content of an utterance of Who hugged Mary? imposed as a goal (?Q WH ) for the next action steps (either by the speaker or the hearer): The DS model assumes tight interlinking of NL perception and action: the predictions generating the sequence of trees above are equally deployed in comprehension and production. ...
... Over the long term, by iterated interaction coordinations among groups of individuals, successful processing paths become progressively routinised and grammaticalised/lexicalised, i.e. easily activated as whole sequences of basic actions (macros). Cross-linguistic and diachronic analyses in DS show how the appearance of distinct NLs arises through the establishment of different such routinisations [59,7,9] invested with social value. This also provides the possibility of explanatory modelling of recent cognitive evidence that processing and interaction constraints affect directly the design of the grammar itself [28,77]. ...
Full-text available
In this paper we present a view of natural language (NL) grammars compatible with enactive approaches to cognition. This perspective aims to directly model the group-forming properties of NL interactions. Firstly, NL communication is not taken as underpinned by convergence/common ground but modelled as the employment of flexible procedures enabling creative joint activities without overarching common goals. On this basis, we argue that a common non-individualistic pattern can be discerned across NL learning, individual and institutional NL change, and evolution. At all levels and stages, modelling of change relies on situated iteration leading to joint establishment and modification of practices. NL learning, change, and even NL emergence can all then be seen in gradualistic terms, with the higher-order organisation that incorporates NL grammars constituting an adaptive interactive system in continuity with the definition of living organisms as modelled in enactive approaches.
... SeeKatsoyannou (1995) for the Calabrian variety andChatzikyriakidis (2009aChatzikyriakidis ( , b, 2010a) for the clitic system of the Grecia Salentina variety. The standard reference grammar for both varieties isRohlfs (1977).2 ...
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.
... Il grico offre materiale importante anche per gli studi sulla morfologia: si ricordano tra gli altri Ralli-Melissaropoulou (2008) Lo studio di Frassanito (2010) mette in rapporto le forme di infinito "fossilizzato" del grico con le equivalenti forme participiali di SMG, individuando casi di "sincretismo" con forme finite del verbo in grico e collegandoli con un caso corrispondente in ghego (albanese). In Chatzikyriakidis (2009Chatzikyriakidis ( , 2010 si ha un'analisi dei clitici e del fenomeno del movimento lungo del clitico (clitic climbing) nel framework della Sintassi Dinamica. ...
Full-text available
The present study presents the results of the documentation and analysis of some aspects of the grammar of Griko. Griko is a Greek dialect spoken in Apulia, in the province of Lecce (Italy). Chapter 1 describes the present-day linguistic situation of Griko and offers an overview of the previous studies and a summery of the question of the origin. Empirical basis of the research and methodology are also described. Chapter 2 discusses the relationship between Griko and the languages belonging to the Balkan Sprachbund. Moreover, it presents in a descriptive way some morphosyntactic features of Griko possibly due to language contact with Italo-romance. Chapters 3 and 4 offer a closer examination and (tentative) analysis of two ‘Balkanisms’ in Griko, namely the replacement of infinitive and the double complementizer system. Empirical data are analysed in the framework of recent generative studies and in comparison with neighbouring Italo-romance dialects (Salentino, Southern Calabrian) and Balkan languages, primarily Greek.
... SeeKatsoyannou (1995) for the Calabrian variety andChatzikyriakidis (2009aChatzikyriakidis ( , b, 2010a) for the clitic system of the Grecia Salentina variety. The standard reference grammar for both varieties isRohlfs (1977).2 ...
Full-text available
In this article, 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 furtermore derive from a common linguistic ancestor, i.e. Koine Greek. The goal of this paper is to show that the transition from Koine Greek to the Medieval varieties and from the Medieval varieties to the respective modern ones can be explained assuming that routinization (in the sense of Pickering \& Garrod 2004) is one of the reasons behind syntactic change. The assumption will be 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 assuming that production/parsing mismatches might trigger syntactic change. This last assumption combined with a) the tendency to obtain parsing triggers in the entries for clitics that are as general as possible and b) the fact that the Medieval varieties under consideration already differed in minimal but crucial ways, provides us an explanation for the transition to the modern varieties.
... A number of approaches have been proposed for clitics in DS (Bouzouita 2008a(Bouzouita , 2008bChatzikyriakidis 2009aChatzikyriakidis , 2009bKempson & Chatzikyriakidis 2009). In all these analyses, positioning restrictions are defined as restrictions on the current parse state, while the actions projected by the clitic vary depending on the level of underspecification involved in each case. ...
... 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), Cann & Kempson (2008 and Gregoromichelaki (2010) among others. ...
Full-text available
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.
Full-text available
In this paper, we investigate a syntactic gap in the structure of the Balkan languages, the absence of Clitic Climbing, which we argue to be a consequence of a well-known Balkanism, namely the loss of the infinitive. For this purpose, we propose a division of the finite constructions that have replaced the infinitive in three categories: Restructuring (Raising), Control, and Subjunctivelike constructions. We also briefly discuss evidence for the existence of selective Clitic Climbing in older stages and in modern dialects of the Balkan languages.
Full-text available
This thesis investigates the syntax of so-called ‘dative’ arguments in Greek and the role of their abstract Case feature in their licensing, from a generative/minimalist perspective. The main claim of the thesis is that dative arguments of all types originate low, i.e. within the maximal projection of the root, in accordance with universal linking principles, and that all apparent variation regarding their realisation and their A-/A’- behaviour can be parameterised in terms of their Case feature and the way it is valued. The secondary claims/premises on which the main claim depends are: (a) a distinction between syntactically inactive and active inherent Case features, which are both possible for dative argument DPs, with purely structural Case being a third possibility cross-linguistically attested; (b) the assumption that minimality effects in phi-Agree must be relativised to Case features; (c) a movement analysis of dative shift; (d) a novel view of applicatives as elements that simply attract dative arguments to their specifier for Case-related reasons, rather than introducing/selecting them. On this view, applicatives are last resort elements and their possible heights of attachment are derivable from the event structure of the predicate. This theory of Appl attachment, coupled with a thematic hierarchy that distinguishes goals from non-goals (and experiencers) with respect to their base position derives the full typology of dative arguments. In support of these assumptions, this thesis draws on evidence from person restrictions in transitive contexts with datives and beyond (Chapter 2), which seem to be best accounted for if the argument affected by the restriction is treated as a (defective) intervener between the dative and an applicative head; the interference of (different types of) datives themselves with agreement relations in various configurations, in Greek as well as cross-linguistically (Chapter 3); the A-/A’-properties of dative arguments of all types in Greek and Romance and novel diagnostics for unpronounced copies with syntactic or interpretive effects (Chapter 4); the diachronic and cross-dialectal behaviour of dative arguments in Greek (Chapter 5), which confirms some empirical correlations that necessitate the assumptions listed above, most notably the generalisation that both (i) the strong Person Case Constraint, and (ii) minimality effects in Agree across datives imply the availability of active Case on indirect object DPs, which is minimally manifested by the existence of the dative-shifted/double-object construction.
Conference Paper
Full-text available
In this paper we argue that the Person Case Constraint, generally presumed to be an irreducible morpho-syntactic constraint on clitic pronoun combina-tions is a direct consequence of processing considerations. Adopting the Dynamic Syntax (DS) perspective of Cann et al. (2005) and Cann & Kempson (2008) in which syntax is defined as the monotonic incremental growth of semantic structure, with structural underspecification and update as the core syntactic notion, we argue that the PCC is wholly due to general restrictions on tree-growth: that these should underpin observed gaps in possible clitic combinations is due to clitics being calcified reflexes of previously available tree-growth update-sequences whose variability is the source of word order variation. We argue that the PCC is the consequence of a tree-logic restriction that only one unfixed node can be present in a tree at any stage in the tree growth process. Strong evidence for this account comes from Pontic Greek, whose preclusion of 3rd person clitic clusters emerges as a consequence of this constraint, in sharp contrast to feature-based accounts which, in being de-fined to match the license for these combinations in other languages, would directly preclude these data, thus pointing towards a feature-free account of the PCC.
This dissertation is a study of Spec-head Licensing within the Principles and Parameters framework. I examine Spec-head Licensing in three configurations, namely Polarity Item Licensing, Focussing and the Clitic-Construction. The language providing most of the data is Modern Greek (MG). I also discuss English and Romance data. In chapter II, I examine Existential Polarity Items (EPI's). I argue that there is a cluster of properties that can be attributed to propositional operators, the operators modifying the proposition in the philosopher's sense of the term. Namely, propositional operators license EPI's and give rise to inner island effects. The claim is advanced for MG and English. It is further suggested that there are interpretational differences between EPI anyone and someone. In chapter III, I consider properties of foci. I claim that the Focus-Criterion is different from the Wh-Criterion. It is also argued that foci always have scope over Neg. Finally, I present some evidence against Quantifier Raising. It is suggested that scope ambiguity between quantifiers is an epiphenomenon bearing crucially on the Focus-Criterion. In chapter IV, I look at the Clitic-Construction (CLC), the construction involving a DP and a matching clitic. I argue that the position occupied by the object DP in MG CLC is both an A- and A'-position. Clitic Doubling is claimed generally to involve syntactic verb-focussing. Complex Inversion in French is analysed as subject CLC. The parametrization of CLC is addressed next. MG has only object CLC while French has only subject CLC. I suggest that the presence versus absence of object CLC correlates with the Class 1 / Class 2 distinction of languages (Koopman and Sportiche (1991)). In chapter V, I attempt some generalizations on the previous discussion. The following seem to be the basic properties of Spec-head mechanisms: licensing of extraction, licensing of A-properties and licensing of operators.