Dependency in Linguistic Description

Igor Mel'uk

It depends!
[the main principle of human scienceand of human life]

Table of contents
Introduction.................................................................................................. 2
Chapter I: Preliminaries .................................................................................... 3
1. Auxiliary Notions.................................................................................... 3
2. Basic Assumptions .................................................................................. 5
3. Illustrations of Sentence Structures: Semantic, Syntactic, and Morphological ............... 6
Chapter II: Three Major Types of Linguistic Dependency ............................................. 9
1. General Remarks..................................................................................... 9
2. Semantic Dependency ............................................................................... 10
2.1. The Concept of Semantic Dependency..................................................... 10
2.2. The Logical and Linguistic Properties of Semantic Dependency........................ 11
3. Morphological Dependency......................................................................... 13
3.1. The Concept of Morphological Dependency .............................................. 13
3.2. The Logical and Linguistic Properties of Morphological Dependency ................. 13
3.3. The Three Major Subtypes of Morphological Dependency.............................. 16
4. Syntactic Dependency ............................................................................... 22
4.1. General Remarks ............................................................................. 22
4.2. The Rationale for Syntactic Dependency................................................... 23
4.3. The Concept of Syntactic Dependency..................................................... 25
4.3.1. Criteria A: SSynt-Connectedness................................................... 26
4.3.2. Criteria B: SSynt-Dominance ....................................................... 28
4.3.3. Criteria C: Labeled SSynt-Dependencies .......................................... 34
4.4. The Logical and Linguistic Properties of Syntactic Dependency........................ 41
4.5. Some Non-Definitorial Properties of Synt-Governors and Synt-Dependents......... 42
4.6. The Absolute Head of the Synt-Structure of a Sentence ................................. 45
4.7. The Three Major Classes of Syntactic Dependencies..................................... 49
4.8. Syntactic Dependencies of a Language: Surface-Syntactic Relations of English...... 53
5. Possible Combinations of the Three Types of Linguistic Dependency........................ 59
6. Correlations between the Three Types of Linguistic Dependency ............................. 66
Chapter III: Syntactic Dependency........................................................................ 67
1. Current Fallacies Concerning Syntactic Dependency............................................ 67
1.1. 'Double Dependency' ........................................................................ 67
1.1.1. Relative Pronouns .................................................................... 67
1.1.2. Raisings................................................................................ 71
1.1.3. Subordination of Coordinate Expressions......................................... 72
1.2. 'Mutual Dependency' ........................................................................ 72
1.3. 'No Dependency'............................................................................. 72
1.4. 'Insufficient Dependency' ................................................................... 73
2. Syntactic Dependency in Action: Eight Illustrative Case Studies............................... 74
2.1. Russian Numeral Phrases ................................................................... 74
2.2. A Russian 'Approximation'-Marking Preposition........................................ 76
2.3. Determiners as Heads?....................................................................... 77
2.4. Romance Clitics............................................................................... 78
2.5. AUX + V Phrases, English-Style .......................................................... 78
2.6. Conjoined Nominal Phrases N + CONJ + N ............................................. 80
2.7. Russian 'Exotic' Coordination of Interrogative/Negative Pronouns ................... 81
 2

2.8. Elliptical Constructions ...................................................................... 82

3. Advantages of Syntactic Dependency ............................................................. 82
4. Syntactic Dependency and Syntactic Constituency .............................................. 89
4.1. Comparing Syntactic Dependency with Syntactic Constituency ........................ 89
4.2. Crossing Syntactic Dependency with Syntactic Constituency........................... 91
5. Insufficiency of Syntactic Dependency: Coordination........................................... 92
6. Syntactic Dependency in Computational Linguistics ............................................ 95
Acknowledgments .......................................................................................... 95
References
............................................................................................................. 10
6

Abbreviations and notations

-A : actant L(w) : lexeme to which the wordform w belongs
ACC : accusative L : a particular language
ADJ : adjective Morph- : morphological
ADV : adverb MV : Main Verb
AUX : auxiliary verb N : noun
C : (inflectional) category NUM : numeral
C(w) : (inflectional) category of the word- r : a particular Surface-Syntactic Relation
form w PERF : perfect
C : constituency PREP : preposition
CONJ : conjunction Rel : relation
D : dependency S- : surface-
D- : deep- -S : structure
DET : determiner Sem- : semantic
DirO : Direct Object Synt- : syntactic
g : grammeme : syntactics (of a linguistic sign)
IndirO : Indirect Object w : wordform
L : lexeme w(L) : wordform which belongs to the lexeme L

Introduction
One of the most vital and, at the same time, the most visible characteristics of human speech
is a VERY HIGH DEGREE OF ORGANIZATION of utterances. (Nothing astonishing, if we remember
that (information) means, strictly speaking, (degree of organization).) More specifically, all the units
which constitute the utterancelet us limit ourselves here, for simplicity's sake, to wordforms
are arranged by the speaker in well-specified configurations, according to numerous complicated
rules, which make up the central part of any language: namely, its syntax. Putting this in a different
way, all wordforms within an utterance are always related or linked among themselves. This fact is
obvious to any speaker, independently of his educational level or general knowledge.
Thus, in English, we have to say I love you, rather than *I you love, as one does in Russian
or French (Ja tebja ljublju /Je t'aime), or *Love I you, or *Me you love (still in the sense of (I love
you)!), or *I loves you , etc. It is clear that the position and the form of the pronouns I JA, JE and
YOU TY, TU depend on the verb, while the form of the verb depends on I JA, JE. To make a
long story short, the wordforms in an utterance are linked by DEPENDENCIES: one wordform must
 3

depend on another for its linear position and its grammatical form. That is how the concept of
dependency appears in linguistics.
Just from these few words it becomes evident to what extent dependency is important for
linguistic description. It is one of the most basic concepts of linguistics, situated on the same level
of basicness as, say, the signified, the signifier, the syntactics, and the linguistic sign: to speak in a
way that will guarantee the transmission of information, the speaker has first to select necessary
signs (the paradigmatic axis) and then to arrange the signs selected into linear sequence (the syntag-
matic axis). The arrangement of signs on the syntagmatic axisi.e. the signs' temporal sequence
is controlled by dependencies between them. Thus, linguistic dependency merits an in-depth study,
which I hope to offer in what follows.
Two important warnings: First, not all the relations between wordforms in utterances are de-
pendencies. For instance, the coreference relation between wordforms father and Hull in the
sentence When John saw his father, Hull Senior was busy repairing the fence (father and Hull refer
to the same person) is not a dependency. I will limit myself here to dependency relations.
Second, dependency in language is of different types. This is, however, not easily seen on
the surfacehence the widespread confusion of these different types; the failure to distinguish
them clearly results in many an incongruous or outright false statement. I will keep the different
types of linguistic dependency apart as strictly as possible.
The paper is divided in three chapters:
Chapter I supplies the introductory information: auxiliary notions, basic assumptions our dis-
cussion is based on, and detailed illustrations of linguistic representations proposed.
Chapter II discusses the three major types of linguistic dependency: semantic, syntactic, and
morphological. After formulating the definitions, the properties of each type of dependency are
described in parallel, their subtypes are specified, and a review of their 14 possible combinations in
a sentence is presented.
Chapter III concentrates on syntactic dependency. Four current fallacies concerning syntactic
dependency are analyzed, and eight case studies are givento illustrate the effect of our criteria for
establishing syntactic dependencies. A cursory comparison with constituency representation
follows; the cases in which 'pure' syntactic dependency proves to be insufficient are discussed.
The chapter ends with remarks on the use of syntactic dependency in computational linguistics.

Chapter I: Preliminaries
1. Auxiliary Notions
The logical analysis of the concept 'dependency in language' requires the following fourteen
underlying notions:1
 4

1. Utterance: a speech segment which is sufficiently autonomous; it can appear between two
major pauses, constitutes a prosodic unit and its internal structure is governed by linguistic rules; it
is also perceived by speakers as 'something that exists in the language.' An utterance is a word-
form, a phrase, a clause, or a sentence.
2. Wordform: a minimal utterance [= not containing other utterances]; in a prototypical case, it
is a disambiguated word [= a lexeme] taken in a specific inflectional form; for instance,
[to] SPEAK is a lexeme, while speak, speaks, spoke, spoken, etc. are its wordforms.2 The word-
form is the ultimate unit in this article: only linguistic dependencies between wordforms are consi-
dered, but not those between wordform parts [= morphs and other signs of the morphological
level] or between wordform configurations [= phrases or clauses].
3. Phrase: an utterance consisting of several wordforms (as a limiting case, it can be one
wordform).3
4. Clause: a phrase that is grammatically organized in essential respects as a sentence; it can
constitute a (simple) sentence by itself or be a constituent part of a sentence. A clause always
contains a finite ( tensed) verb.
5. Sentence: a maximal utterance, which is a complete communication unit. (Two or more
sentences are a sequence of utterances.) The sentence constitutes the upper limit of analysis in this
article: only linguistic dependencies between wordforms within a sentence are considered, to the
exclusion of those between wordforms from different sentencessuch as, for instance, semantic
and anaphoric dependencies.
6 - 8. Semantic predicate, semantic name, argument of a predicate:
the notions themselves and the way they are used in linguistics are borrowed from the language of
predicate calculus. A (semantic) predicate is a 'binding' meaning, which is somehow
incomplete without other meaningsit has open 'slots' where other meanings should be inserted.
A meaning that is not a predicate is a (semantic) name. Predicates refer to actions, activities,
events, processes, states, properties, relations, localizations, quantities, etc.; their linguistic expres-
sions can belong to any part of speech. Semantic names refer to objects (including beings),
substances, and points in time and space; their expressions are nouns.
A meaning that is inserted into an open slot of a predicate is called its argument; the
traditional notation for a predicate P and its argument a is P(a). Thus, Leo is sleeping is
represented as SLEEP(LEO). A predicate can have several arguments: P(a1 ; a2 ; a3 ; ...); e.g.,
SEND takes three arguments, cf. Leo sent a letter to Alan = SEND(LEO ; LETTER ; ALAN). The
number and the nature of possible arguments of a predicate must be fully specified in its description
in one way or another, e.g., by ordering or numbering the arguments, so that, e.g., HIT(LEO ;
ALAN) HIT(ALAN ; LEO). A predicate with its arguments can itself be an argument of another
predicate, this phenomenon being recursive:
 5

Leo knows that Alan is in love with Helen = KNOW(LEO, BE-IN-LOVE(ALAN, HELEN));
I think that Leo knows that Alan is in love with Helen =
THINK(I, KNOW(LEO, BE-IN-LOVE(ALAN, HELEN))); etc.
9-10. Inflectional category: a set of mutually opposed inflectional meanings, called
grammemes, such that the selection of one of them is obligatory for lexemes of a given class (e.g.,
in English, number for a noun, with grammemes (SG ) and (PL), or tense for a finite verb, with
grammemes (PRES), (PAST), (FUT)).
11-13. Syntactics: one of the three components of any linguistic sign, in particular of a
wordform; it specifies the cooccurrence of the sign that is not determined by its signified nor by its
signifier (i.e. more or less arbitrary cooccurrence). The syntactics of a sign is represented as a set
of features, each of which admits mutually exclusive values.
14. Passive syntactic valency of a lexeme/of a phrase: a set of syntactic roles which
the lexeme/the phrase can take in larger constructions (maybe with some inflectional modifications).
In other words, the passive syntactic valency of a lexeme/a phrase is its syntactic distribution.
Passive syntactic valency is normally defined for major classes of lexemes, known as parts of
speech. Thus, the passive syntactic valency of the English noun is as follows: 1) the syntactic
subject of a finite verb, 2) the Dir(ect) O(bject) [= DirO] of a transitive verb, 3) the Indir(ect)
O(bject) [= IndirO] of a special verb (send Father a letter), 4) the complement of a copula, 5) the
object of a preposition, 6) the first member of a nominal compound (computer program), 7) an
address, 8) a fronted topic, etc.4

2. Basic Assumptions
Assumption 1: LEVELS OF SENTENCE REPRESENTATION. A sentence has representations on
four levels: semantic, syntactic, morphological, and phonological. (In what follows, the phonolo-
gical representation will be left out. However, phonological dependency is considered in a number
of works: thus, see Durand (ed.) 1986, Anderson/Ewen 1987, and rnason 1989.) Each repre-
sentation reflects a set of such properties of the sentence that are of the same nature and belong to
the level in question. The same is true of any non-minimal part of the sentencee.g. the clause or
the phrase.
A sentence representation is a set of formal objects called structures, each of which is res-
ponsible for a particular aspect of sentence organization at the given level. Thus:
Sem(antic) Representation = Sem-S(tructure); Sem-Comm(unicative) S; Sem-Rhetorical S>
D(eep)-Synt(actic) Representation = <DSyntS; DSynt-CommS; DSynt-Anaph(orical) S; DSynt-Pros(odic) S>
S(urface)-Synt(actic) Representation = <SSyntS; SSynt-CommS; SSynt-AnaphS; SSynt-ProsS>
DMorph(ological) Representation = <DMorphS; DMorph-ProsS>
SMorph Representation = <SMorphS; SMorph-ProsS>
 6

Assumption 2: SENTENCE STRUCTURE. The central part of a sentence representation, called its
(central) structure, appears formally as a labeled graph, whose vertices, or nodes, represent linguis-
tic units of the corresponding level, and whose arcs represent relations between these units.
It is here that the notion of linguistic dependency comes into play: the major type of relation
between linguistic units in a sentence structure is dependency.
Assumption 3: DEEP VS. SURFACE DISTINCTION. On the syntactic and the morphological level
the Deep and the Surface sublevels of the sentence structure are distinguished: the former is aimed
at meaning and expresses explicitly all relevant semantic distinctions; the latter is aimed at form and
expresses explicitly all relevant formal distinctions. (For more on the Deep vs. Surface distinction,
see Mel'uk 1988: 59-72.)

3. Illustrations of Sentence Structures: Semantic, Syntactic, and Morphological

In order to show the reader how linguistic dependencies 'work,' sentence structures that use
various types of dependency will be presentedbefore introducing the corresponding concepts
formally. These illustrations will be referred to when discussing dependencies later on.
Consider the English sentence (1) and its structures at different levels (2)-(5):
(1) For decades, cocoa farming has escaped such problems by moving to new areas in the
tropics.
(2) The Semantic Structure [= SemS] of (1) [drastically simplified; thus, semantic grammemes
number for nouns, tense for verbsare not indicated]
(move)
1 2
1 (farm) (areas) (tropics)
(means) 1 2
2 1 1 2
(cocoa) (located)
1 (new)
(escape) (decades)
2 2
1 2
(such) 1
(problems) (duration)

[The blank '' instead of a semanteme attached to a node means that the corresponding
Sem(antic) A(ctant) is not specified.]

The Semantic Structure of a sentence is a network whose nodes represent meanings

and are labeled with semantemes (roughly, lexical meanings of the language); its arcs represent
predicate-to-argument relations and are labeled with numbers identifying an argument with respect
to its predicate. These arcs correspond to semantic dependencies, see Ch. II, 2, p. 00.
 7

(3) The Deep-Syntactic Structure [= DSyntS] of (1)

FOR ESCAPE active, indic, pres, perf

ATTR
II I II ATTR
FARMINGsg, indef
PROB- BY
LEM pl, indef
II ATTR II
DECADEpl, indef
COCOA sg, indef MOVE ger
SUCH I II
AREA pl, indef
FARMING sg, indef ATTR ATTR
IN
NEW
II

TROPICS pl, def

The Deep-Syntactic Structure of a sentence is a tree whose nodes are labeled with
the full lexemes of the sentencesuch that there is a one-to-one correspondence between DSynt-
nodes and full lexemes; the arcs of this tree, called branches, are labeled with names of abstract
universal Deep-Syntactic Relations. Their numberacross all languagesis about 10: six actantial
DSyntRels (I, II, ..., VI ), an attributive (ATTR ), a coordinative (COORD ), and an appenditive
DSyntRel (APPEND).5 DSyntRels are of course particular subtypes of syntactic depen-
dency; see Ch. II, 3, p. 00.
The diagram of (3) indicates the coreference link between the two occurrences of FARMING
(by a dashed bi-directional arrow). This indication belongs to the Deep-Syntactic Anaphoric Struc-
ture, mentioned above (and not considered in this paper); it does not interfere with genuine
syntactic dependencies represented in the DSyntS.
The Surface-Syntactic Structure of a sentence (see next page) is also a tree
whose nodes are labeled with all the lexemes of the sentence (including all auxiliary and 'structural'
words)again there being a one-to-one correspondence between the SSynt-nodes and the lexemes;
the arcs of this tree, also called branches, are labeled with names of language-specific Surface-
Syntactic Relations, each of which represents a particular construction of the language (their
number, in an average language, is somewhere around 50; see a list of SSyntRels of English in Ch.
II, 4.8, p. 00). SSyntRels also are particular subtypes of syntactic dependency.
 8

(4) The Surface-Syntactic Structure [= SSyntS] of (1)

FOR HAVEindic, pres

circumstantial
prepositional subjectival auxiliary
FARMINGsg ESCAPEpast participle
compositive circumstantial
direct-
DECADEpl objectival BY
COCOAsg PROBLEMpl prepositional

modificative MOVEger
prepositional-
SUCH objectival

TO
prepositional

AREApl
modificative attributive

IN
NEW
prepositional

THE determinative TROPICSpl

(5) The Deep-Morphological Structure [= DMorphS] of (1)

FOR DECADEpl COCOAsg FARMINGsg

HAVEind, pres, sg, 3 ESCAPEppart SUCH PROBLEMpl

BY MOVEger TO NEW AREApl IN THE TROPICpl.

The Deep-Morphological Structure of a sentence is a string of lexico-morpholo-

gical representations of all its wordforms; its arcs are, so to speak, degenerated: they specify only
the strict linear ordering of wordforms ('w1 immediately precedes w2'), so that they need not be
indicated explicitly. In sharp contrast to the SemS and the D-/S-SyntSs, the DMorphS of a sentence
does not represent morphological dependencies between its wordforms: morphological
dependencies are not explicitly presented in a special structure because they are not universal (see
Ch. II, 3.2, item f, p. 00). They are computedby means of syntactic rules of the languageon
 9

the basis of syntactic dependencies, that is, from the SSyntS, and transcoded into grammemes that
appear in the DMorphS and are expressed in the corresponding wordforms.6
Thus, the SemS (2) shows semantic dependencies between (the meanings of) the wordforms
of sentence (1), while the DSyntS (3) and the SSyntS (4) show the Deep and Surface syntactic
dependencies between the wordforms of (1); morphological dependencies are not shown explicitly.
But given the morphological poverty of English, there is only one case of morphological dependen-
cy in (1): the wordform has depends morphologicallyfor the singular and 3rd personon
farming. The Russian sentence (6), which is a close translation equivalent of (1), contains many
examples of morphological dependency (its major typesagreement, government, and congruence
are considered in Ch. II, 3.3, p. 00ff.):
(6) V teenie desjatiletij, kultura kakao ne znala tix problem blagodarja rasprostraneniju na
novye territorii v tropikax.
Here,
desjatiletij [GEN] (decades) depends for its case on v teenie (during)
[government];
kultura [NOM] (farming) depends for its case on [ne] znala (has-escaped)
[government];
znala [SG, FEM] (has-escaped) depends for its number and gender on
kultura (farming) [agreement];
tix [PL, GEN] (such) depends for its number and case on
problem (problems) [agreement]; etc.
In Russian, unlike English, almost all the wordforms of a sentence may be linked by
morphological dependencies.

Chapter II: Three Major Types of Linguistic Dependency

1. General Remarks
I will consider three types of syntagmatic dependency relations between wordforms in a sen-
tence: semantic dependency [= Sem-D ]7, syntactic dependency [= Synt-D ], and morphological
dependency [= Morph-D ], as distinguished in Mel'uk (1964, 1979: 13, 1981, 1988: 105-149)
and developed in Nichols 1986. I will leave aside paradigmatic relations between wordforms, such
as synonymy, antonymy or derivation, and syntagmatic relations of a different nature, such as:
all kinds of lexical correspondences, e.g., between a word and a preposition it requires (insist
- on, borrow - from, central - to), or between a noun and its classifier (e.g., in Vietnamese an
animate noun takes the classifier CON and an inanimate noun, the classifier CI, with some ex-
ceptionssuch as con sng (river); in Malay, nouns take different classifiers according, roughly, to
their semantic class: tiga helas kemeja (three shirts) vs. tiga ekor ajam (three chickens) vs. tiga
 10

batang rokok (three cigarettes), etc.; let it be emphasized that no morphology is involved in such
lexical correspondences);
the anaphoric relation (coreferentiality: between a pronoun and its antecedent or
between two nouns sharing the same referent; lexical identity: between a pronoun of the
type of THAT and its antecedent, as in my hat and that of my friend);
the inclusion relation (between a phrase and its constituents),
the ordering relation (between wordforms, phrases, and clauses);
the communicative dominance relation (between semantic units in a semantic representation).
I will deal only with DIRECT dependencies, without indicating this explicitly every time.
Dependency is by definition a non-symmetrical relation, of the same type as implication: one
element 'presupposes' in some sense the other, but generally speaking not vice versa. Therefore,
dependency is denoted by an arrow: w w means that w depends on w ; w is said to be
1 2 2 1 1
a/the governor of w2, and w2 a dependent of w1. Other terms used to designate the governor
in a dependency relation include: head, regent, ruler; here, however, only the term governor will be
used. The term head, extremely popular in the literature, has the following defect: it is natural to
speak of the head of a phrase/clause/sentence, but the expression ?the head of this wordform
meaning (the governor of this wordform) seems much less convenient. The concept of head is inhe-
rited from phrase-structure syntax and carries with it unnecessary connotations (implying constitu-
ency). Moreover, governor of phrase P head of phrase P: P's governor is outside of P, P's head
is inside of P, so that in (7) the head of the phrase P = abc is the unit b, while P's governor is the
unit d:

(7) d a b c
Therefore, in this article the term head is used only in the precise sense (the Synt-head of a phrase/a
clause/a sentence), never in the sense of the Synt-governor. (Cf. Hudson 1993a: 274-275, on the
head of a phrase vs. head of a wordform terminological problem.)
An alternative term for dependent is satellite.
Because of its intermediate natureit is 'squeezed' between semantics and morphology,
Synt-D is the most difficult type of linguistic dependency to grasp; therefore, it will be treated after
Sem-D and Morph-D.

2. Semantic Dependency

2.1. The Concept of Semantic Dependency

As stated in Chapter I and illustrated in (2), the meaning of a sentence can be represented
using the formalism of the predicate calculus. We say that an argument of a predicate semantically
 11

depends on its predicate, and for P(a) we write Psema. As I have said, an argument of a predi-
cate P1 can be another predicate P2 with its own arguments a2-1, a2-2, a2-3, ...:
P1(P2(a2-1 ; a2-2 ; a2-3 ; ...))
In this case, we write P1semP2, P2sema2-1, P2sema2-2, P2sema2-3, etc.
The arc between the predicate and its argument carries the number of the argument: P1a1,
P2a2, etc. The meaning of the sentence Leo sent a letter to Alan can then be represented (leav-
ing grammemes aside) as
(send)
1 2 3

(Leo) (Alan)
(letter)
From this, we immediately obtain the definition of Sem-D between wordforms w1 and w2
in a sentence.

Definition 1: Semantic dependency

The wordform w 2 is said to semantically depend on the wordform w 1 in the given
sentence if the meaning of w1 is a predicate and the meaning of w2 is an argument of it in this
sentence: (w 1)((w 2)).
I write, as convened above, w 1semw 2.
A Sem-dependent of a wordform is called its Sem-Actant.

2.2. The Logical and Linguistic Properties of Semantic Dependency

a) Sem-D is anti-symmetrical: w sem w entails (w sem w ), i.e. *w sem w .
1 2 1 2 1 2
The meaning of a wordform (or any other type of meaning) cannot be an argument of the meaning
of another wordform and, at the same time, have the latter as its own argument.
sem
b) Sem-D is anti-reflexive: * ; a meaning cannot be its own argument. The anti-reflexivity of
w
the Sem-D follows from its anti-symmetry.
c) Sem-D is, generally speaking, neither transitive, nor anti-transitive: in most cases,
w sem w and w sem w entails neither w sem w , nor (w sem w ).
1 2 2 3 1 3 1 3
Thus, from I saw [= w1] Alan's [= w3] wife [= w2] it does not follow that I saw Alan as well, but
it does not follow, either, that I did not (I could).
However, in some cases, i.e. for some predicates, Sem-D is transitive:
w sem w and w sem w entails w sem w .
1 2 2 3 1 3
 12

A typical example is the predicate ([to] order) in the sentence I order [= w ] him [= w ] to go [=
1 3
w ],8 which has the following SemS:
2
(order)

1 3
2

(he)
(I)
1

(go)
Thus, the SemS may contain an undirected circuit (such a circuit is shown in boldface in the above
diagram), but not a cycle, i.e. a directed circuit in which all the arrows point in the same direction.
Finally, in some other cases, Sem-D is anti-transitive:
w sem w and w sem w (in a sentence) entails (w sem w ).
1 2 2 3 1 3
Thus, in I wrote down [= w 1] Alan's [= w 3] address [= w 2], it is clear that (w 1semw 3).
Another example of the same type is I heard [= w1] that Alan [= w3] came [= w2] home.
d) Sem-Ds must be typed, or labeled: a Sem-D arc has to be supplied with the symbol identify-
ing the corresponding argument. In the present approach, this is a purely distinctive number: it does
not carry meaning by itself; thus, an arc i expresses different semantic roles with different pre-
dicates. (The actual semantic role of an argument of the predicate (w) is specified by the semantic
decomposition of (w). For instance, (X kills Y) (X, by acting upon Y, causes that Y dies), which
shows that X is the Agent and the Causer, while Y is the Undergoer.) In other approaches, the
symbols on Sem-arcs can be meaningful: e.g., 'Agent,' 'Perceiver,' 'Beneficiary,' etc. Since this
does not affect my reasoning in any essential way, I will not deal with this issue here.
e) Sem-D does not presuppose the uniqueness of the governor: a wordform can semantically
depend simultaneously on many other wordforms, i.e. many different meanings can be predicated
about one meaning at the same time:
(renowned)
(little) (nice)
1
[a] nice little hotel renowned [for its comfort] = 1 1

(hotel)
f) Sem-D is universal in the following three respects: it is present in all languages; it appears in
all sentences of a language; and it embraces all full wordforms of a sentence (this means that in a
sentence, Sem-D s always form a connected structure, such that there is a Sem-'path' between
any wordform and any other wordform). Cf. the Sem-Ds in the SemS of (2).
 13

3. Morphological Dependency

3.1. The Concept of Morphological Dependency

In many languages (but by no means in all!), a wordform w2 in a sentence can take a particu-
lar morphological form, or inflect, under the impact of another wordform, w 1, of the sentence.
Thus, in I am well vs. You are well the verb BE has different forms because of its subject [agree-
ment, cf. 3.3, Definition 2.1]; in German, after the preposition NACH (after/to) a noun is in the
dative (nach dem Fest (after the feast)), but after WEGEN (because of) it is in the genitive (wegen
des Festes (because-of the feast)) [government, cf. 3.3, Definition 2.2]. Technically, in such
cases a grammeme g of an inflectional category C of w2 is determined by some properties of w1.
This leads to the following definition.

Definition 2.1: Morphological dependency

The wordform w2 is said to morphologically depend on the wordform w1 in the given
sentence if at least one grammeme of w2 is determined by w1.
I write w 1 morph w 2.

3.2. The Logical and Linguistic Properties of Morphological Dependency

a) Morph-D is, strictly speaking, anti-symmetrical. In most cases (agreement of an ADJ with an
N, government of the case of an N by the a V or a PREP, etc.), w morph w entails
1 2
(w 1 morph w 2 ).
However, in some cases, Morph-D is, so to speak, symmetricalor rather reciprocal: a
wordform w2 can be inflected, for the inflectional category C1, as a function of the wordform w1,
and, at the same time, w1 must be inflected, for a DIFFERENT category C2, as a function of w2, so
that w 1 morph w 2 entails w 1 morph w 2 ; as a result, w 1 morph w 2 obtains, with C1
C 2. This property does not contradict the anti-symmetry of dependency in general, since this
reciprocity of Morph-D is possible only with respect to different Cs.
Examples
(8) a. Russian
dv+a stol +a (two tables)
two MASC.NOM table[MASC] SG.GEN
The noun stola morphologically depends for its singular and genitive case on the numeral dva,
while dva depends on stola for its masculine gender; cf. pjat stol+ov [PL.GEN] (five tables) or
dvadcat odin stol+ [SG.NOM], lit. (twenty-one table); dv+e [FEM.NOM] krovati (two beds).
[Here and below, I boldface the grammemes of the wordform w which are determined by the
wordform w', as well as their markers.]
 14

b. Georgian
i. Gogi+ m +zrdi +s me
Gogi NOM 1SG.OBJ bring.up PRES .3SG.S U B me.DAT
(Gogi brings me up).
vs.
Gogi+ gv +zrdi +s v e n
Gogi NOM 1PL.OBJ bring.up PRES . 3SG . S U B we.DAT
(Gogi brings us up).

ii. Gogi+m ga +m +zard +a me

Gogi ERG COMPL 1SG.OBJ bring.up AOR .3SG.SUB me.NOM
(Gogi brought him/her up).
vs.
Gogi+m ga +g v +zard +a v e n
Gogi ERG COMPL 1PL.OBJ bring.up AOR .3SG.SUB we.NOM
(Gogi brought us up).
Both Synt-actants depend morphologically on the verb for their case (which is imposed
by the syntactic type of the verb and its tense: the present vs. the aorist), while the verb
depends morphologically on the actants for its person/number (of the subject and of the
DirO).
morph

b) Morph-D is anti-reflexive: * , i.e. the inflection of a wordform cannot be determined

w
by the wordform itself. However, the inflection of a wordform w can be conditioned by some pro-
perties of w itself, which is not a case of Morph-D .9 Thus, in Alutor, in the ergative construction
of a transitive verb, the Synt-subject w is in the instrumental if w is not a human proper name, and
in the locative if it is:
(9) a. Alutor (Chukchee-Kamchatkan family, Russia)
i. npNav+a +l/us7qiv +nin + qlavul+
old.woman SG.INSTR 3SG.SUB went.to.see 3.OBJ SG man SG.NOM
(An old woman went to see [her] man).

ii. Miti+nak +l/us7qiv +nin + qlavul+

Miti SG.LOC 3SG.SUB went.to.see 3.OBJ SG man SG.NOM
(Miti went to see [her] man).
The case of the Synt-subject is determined here by the verbbut conditionally, i.e. according to the
indicated property of the subject noun.
Another example of a slightly different kind comes from Arabic:
 15

b. Arabic
J rafq+u vs. J rafq+aab +
oh friend NOM oh friend A C C father 1SG
(Oh, friend!) (Oh, friend of my father!)
The case of the address noun N is determined by the vocative particle J, but according to whether
or not N has its own nominal Synt-dependent: if such a dependent is absent, N is in the nominative;
if the dependent is present, N is in the accusative.
c) Morph-D is neither transitive nor anti-transitive. In most cases, Morph-D is anti-transitive:
w morphw and w morphw (in one sentence) entails (w morphw ).
1 2 2 3 1 3
Thus, in Rus. Ja viu [= w1] krasivuju [= w3] knigu [= w2] (I see [a] beautiful book) there is no
Morph-D between the verb and the adjective.
There are, however, cases where Morph-D is transitive:
w morph w and w morph w entails w morph w .
1 2 2 3 1 3
An example of a transitive Morph-D (again, for different inflectional categories and different gram-
memes) is found in Russian:
(10) Russian
Ja zna +l +a ego molod+ym
I know PAST FEM he.SG.ACC young SG.MASC.INSTR
(I [a woman] knew him young).
vs.
Ja zna +l +a e molod+oj
I know PAST FEM she.SG.ACC young SG.FEM.INSTR
(I [a woman] knew her young).
vs.
Ja zna +l +a ix molod+ymi
I know PAST FEM they.PL.ACC young PL.INSTR
(I [a woman] knew them young).
Here, ego/e/ix [= w 2 ] depends on znala (knew) [= w 1 ] for its accusative case, while
molodym/molodoj/molodymi [= w3] depends on ego/e/ix for its number and gender, and on znala
for its instrumental case.10
d) Similarly to Sem-D , Morph-D must be also typed (= labeled) : if w1morphw2, then in
order to fully specify this Morph-D , we have to indicate the inflectional category C(w2) whose
grammeme is imposed by w . Thus, the labeling of Morph-D s is meaningful rather than purely
1
distinctive, as is the case with Sem-D.
e) Morph-D does not presuppose the uniqueness of the governor: a wordform can morphologic-
ally depend simultaneously on several other wordformsfor different inflectional categories of
course. Cf. (10), where w3 depends morphologically on w1 and w2 at the same time (with transi-
 16

tivity of Morph-D); another example of Morph-D with multiple governors (without transitivity of
Morph-D ) is (11a), p. 00.
f) Morph-D is not universal: in many languages it does not exist at all; in a language where it
does exist it is not present in all sentences; and in a sentence where it is present it does not neces-
sarily embrace every wordform (that is, in a sentence Morph-Ds do not form, generally speaking, a
connected structure: there are wordforms that are not morphologically linked to the rest of the
sentence).

3.3. The Three Major Subtypes of Morphological Dependency

There are three major subtypes of Morph-D : agreement, government, and congruence
(Mel'uk 1993).11 Here are the corresponding definitions; in all of them the wordform w depends
2
morphologically on the wordform w1 according to the inflectional category C2. The wordform w1
is called the controller, and the wordform w2, the target of the Morph-D in question. In
the examples below, the controller is boxed, and the controlled grammeme and its marker in the
target are boldfaced.

Definition 2.2: Agreement

The wordform w is said to agree with the wordform w in the inflectional category C if the
2 1 2
following two conditions are simultaneously satisfied:
1) the wordform w2 is not a substitute pronoun12 coreferential with w1;
2) a grammeme g2(w2), such that g2C2, is selected depending upon:
(a) either a grammeme g1(w1), such that g1C1 and C2 is bound to C1;
(b) or the value 1 of a feature 1 of the syntactics of w1, this 1 being one of the follow-
ing three features of syntactics:
agreement class, pronominal person, or pronominal number;
(c) or some semantic components of w1 or some properties of its referent.

Comments
1. An inflectional category C2 is said to be bound to the inflectional category C1 if
(roughly) C2 exists in L exclusively to 'reflect' C1. Thus, adjectival number and adjectival case are
bound to nominal number and nominal case. (The relation to be bound to is by no means
symmetrical: C1 is not bound to C2.)
2. The agreement class A is (roughly) a subset of lexemes of the same part of speech
(essentially, of nouns) such that in any context the following three conditions are simultaneously
satisfied:
1) if any two wordforms wi and wj of A impose on a third wordform w a grammeme of a cate-
gory C(w), they impose on w the same grammeme geC(w);
 17

2) if in a sentence a wordform w of A simultaneously imposes on different wordforms w i

morphologically depending on it a grammeme of a category C(wi), it imposes on all wi the same
grammeme geC(wi);
3) the grammeme g is not imposed by anything except the wordforms of A.
Agreement class is a generalization of grammatical gender (as in Indo-European languages) and of
nominal class (as in Bantu and Daghestanian); agreement classes are defined and established in par-
ticular languages prior to and independently of the notion of agreement (Mel'uk 1993: 323-324,
1996b: 206-211).
3. Condition 1 separates agreement from congruence, see below. Condition 2a foresees
agreement with GRAMMEMES of the controller (e.g., agreement of an ADJ with an N in number
and case). Condition 2b foresees agreement with FEATURES of the SYNTACTICS of the controller:
e.g., agreement of an ADJ with an N in gender; or agreement of a V with a pronominal Synt-actant
in person and number, the latter being syntactic features of a pronoun). Condition 2c foresees what
is known as SEMANTIC AGREEMENT (Rus. Na+a vra skazal+a (our[FEM ] doctor [a woman]
said[FEM]), where, in spite of the fact that the noun VRA is masculine, the agreeing adjective and
the verb are both in the feminine, because in this sentence VRA refers to a woman).

Examples
(11) a. In Akhvakh (North-Caucasian, Daghestan, Russia; Boguslavskaja 1991: 11), an
adjective or a participle which is used as a restrictive modifier of a noun and, at the same
time, has a complement or an actant of its own agrees both with this complement/actant (in
nominal class) and the modified noun (again in class); the first agreement is shown by a
prefix, and the second by a suffix:
mina + b +ai +da +we a +ssua
hekw
head[III] SG.NOM III white ADJECT(ivizer) I man[I] SG.DAT
lit. (head white to-man) = (to a white-haired man)
[mina (head) is a complement of the adjective ai (white): mina bai (head-wise white)]

roa + b +eeq+ida +je jae +

book[III] SG.NOM III bring ADJECT II girl[II] SG.NOM
lit. (book bringing girl) = (a girl who is bringing a book)

b. In Old Georgian, a noun N2 in the genitive that syntactically depends on another noun
N1 agrees with N1 in case; as a result, N2 has two case suffixes: the marker of its own
genitive and the marker of the second, 'agreeing' case, cf.:
neb++ita mrt+ +isa +jta, lit. (by-[the-]will of-God)
will SG INSTR God SG GEN INSTR
 18

c. In Kayardild (Australia), all the objects and complements of the verb agree with it in
tense/mood (Evans 1988: 221-222):
daNga+a bargi+da t1uNgal+ +i nara+Nuni +y
man NOM chop NON-FUT tree ACC NON-FUT knife INSTR NON-FUT
(The man just chopped/is chopping the tree with a knife).
vs.
daNga+a bargi+du t1uNgal+ +u nara+Nuni +wu
man NOM chop FUT tree ACC FUT knife INSTR FUT
(The man will chop the tree with a knife).
vs.
daNga+a bargi+dara t1uNgal+ +ina nara+Nuni +na
man NOM chop PAST tree ACC P A S T knife INSTR P A S T
(The man (had) chopped the tree with a knife).
For more examples of 'exotic' agreement see Kibrik 1977 amd Anderson 1992: 103-118.

Definition 2.3: Government

The wordform w2 is said to be governed by the wordform w1 in the inflectional category
C1 if a grammeme g2(w2), such that g2C2, is selected depending upon:
(a) either a grammeme g1(w1), such that g1C1 and C2 is not bound to C1;
(b) or a value 1 of a feature 1 of the syntactics of w1, this 1 being neither agreement class,
nor pronominal person, nor pronominal number.

Comment
Condition (a) foresees government by a grammeme of the controller. These are 'exotic' cases
of government: e.g., the comparative that governs the case of the comparand (Rus. siln+ee smerti
[GEN] (stronger than death)) or the tense of the verb governing the case of its actants, see examples
(12a-b). Condition (b) foresees government by a feature of the syntactics of the controller; it se-
parates such government from syntactics-induced agreement. These are 'normal' cases of govern-
ment: e.g., a verb or a preposition governing the grammatical case of a complement.

Examples
(12) a. In Georgian, a transitive verb in the present/imperfect governs the nominative of the sub-
ject and the dative of the DirO; if the verb is in the aorist, the subject takes the ergative and
the DirO, the nominative; the verb in the perfect governs the dative of the subject and the
nominative of the DirO. However, the agreement of the verb does not change: it always
agrees with its subject and with its DirO (in person and number), if the latter is not of the
3rd person:
Gogi+ ceril +s cer +s
Gogi NOM letter.SG DAT write PRES.3SG.SUB
 19

(Gogi is-writing [a] letter).

vs.
Gogi+m ceril +i da +cer +a
Gogi ERG letter.SG NOM COMPLETIVE write AOR.3SG.SUB
(Gogi wrote [a] letter).
vs.
Gogi+s ceril +i da +u +cer +i +a
Gogi DAT letter.SG NOM COMPLETIVE 3person write PERF 3SG.SUB
(Gogi (apparently) has-written [a] letter).
["3 person" in the last example denotes the grammeme (for the other) of a special inflectional
category of Georgian: the version, which signals the person for whose 'benefit' the event in
question takes place; this grammeme must be obligatorily present in perfect forms.]

b. In Hindi, a transitive verb in the present governs the nominative of the subject and the
nominative/dative of the DirO (the dative seems syntactically optional); if the verb is in
the perfect, the subject takes the ergative and the DirO remains in the nominative/dative.
But, unlike Georgian, the agreement of the verb changes depending on the tense: in the
present, the verb agrees with the subject, but in the perfect either it agrees with the DirO
(if the DirO is in the nominative) or it takes the unmarked form of the 3rd person
singular masculine (if the DirO is in the dative).
Lar>k + kitb + par>h+t hai
boy[MASC] NOM book[FEM] NOM read IMPF.MASC.SG AUX.PRES.3SG

([The] boy reads [a] book). analytical present form

vs.
Lar>ke +ne kitb + par>h+
boy[MASC] ERG book[FEM] NOM read PERF.FEM.3SG
([The] boy read [a] book).
vs.
Lar>k +ne Sit + dekh+
boys[MASC] ERG Sita[FEM] NOM see PERF.FEM.3SG
([The] boys saw Sita).
vs.
Lar>kiy +ne Sit +k o dekh+
girls[FEM] ERG Sita[FEM] DAT see PERF.MASC.3SG
([The] girls saw Sita).
c. In Russian, the infinitive in a special 'impossibility' construction governs the dative of its
semantic 'subject:'
Mn+e tu knigu ne proest
I DAT this book not read.PERF.INF
(I will not be able to read this book).
 20

Alen+u tuda ne dojti

Alan SG.DAT till.there not reach.walking.PERF.INF
(Alan will not be able to walk till there).

d. In Hungarian, the subordinate conjunction HOGY (that), when it syntactically depends

on a verb of volition, requires the imperative of the Main Verb of the subordinate com-
pletive clause:
Azt akarta, hogy lassan jr+j
this-ACC want-PAST.3SG that slowly walk IMPER.2SG
lit. ([S/he]this wanted that [you-SG] slowly walk).
Let it be emphasized that it is impossible to define agreement and government in a simpler
wayfor instance, following the traditional view that agreement is a correspondence between the
inflectional form of a lexeme and the inflectional form of another lexeme, while government is a
correspondence between the inflectional form of a lexeme and lexicographic properties of another
lexeme. This viewpoint is simply wrong: many types of agreement involve lexicographic properties
of the controller (gender, nominal class, animacy), and many types of government are determined
by the inflectional form of the controller (cf., among others, examples (12a-b)).

Definition 2.4: Congruence

The wordform w2 is said to be congruent with the wordform w1 in the inflectional cate-
gory C2 if w 2 is a substitute pronoun coreferential with w 1 and a grammeme g2(w 2), such
that g2C2, is selected depending on w1.

Comments
1. Congruence is, so to speak, a particular case of agreement, namely, 'agreement in absen-
tia:' while genuine agreement obtains between an ADJ/a V and the N it combines with syntagma-
tically, congruence obtains between a substitute pronoun and the N it replaces. Agreement marks
semantic and/or syntactic Ds within the borders of a clause, and congruence marks anaphoric links,
basically outside the borders of a clause. For congruence, correspondence according to the meaning
(rather than according to grammatical properties of the controller) is especially typical. Recall that
congruence is not a syntactic dependency, but a morphological one; therefore, the typical absence
of its controller in the clause is not a problem.
2. Congruence presupposes the choice of a particular inflectional form of a given lexeme.
Thus, in Spanish, the noun caballo (horse) [MASC, SG] is replaced with the pronoun l (he), mosca
(fly) [FEM, SG] with ella (she), caballos [MASC, PL] with ellos, and moscas [FEM, PL] with ellas,
and this is congruence: l, ella, ellos, and ellas are forms of one lexeme (= L), which is inflected
for gender and number. (The same state of affairs obtains in any language in which substitute pro-
 21

nouns grammatically distinguish gender and/or number: Romance, Slavic, Semitic, Bantu lan-
guages.) However, the choice between different pronominal lexemes as a function of w 1 to be
replaced is not congruence. Thus, in English, general or Alan is replaced by HE, sister or battle-
ship, by SHE, and warning or fly, by IT; but HE, SHE and IT are different lexemes rather than
inflectional forms of the same lexemebecause English has no inflectional category of gender. The
selection of the appropriate lexeme has to do with lexical correspondences, mentioned in 1, p. 00,
not with congruencebecause no Morph-D is involved (no grammeme is imposed).

Examples
(13) a. French
Nous tudions un suffixe [MASC .SG ] et deux alternances [FEM.PL] ; nous traiterons
celui-l [MASC . SG ] immdiatement, et nous analyserons celles-ci [ FEM . PL ] a u
chapitre suivant
(We will study a suffix and two alternations; we will deal with the former right away, and
we will analyze the latter in the next chapter).
The wordforms celui and celles are inflectional forms of the lexeme CELUI, so that their choice
illustrates congruence. (In contrast, the English wordforms former and latter belong to two diffe-
rent lexemes, and therefore their use is not related to congruence.)

b. In Bushong (Bantu), a noun is replaced by different inflectional forms of the same sub-
stitue pronoun lexeme -N ((s)he, it, they), namelyby the form of the corresponding no-
minal class:
I aa +n replaces a noun of the class I;
II baa +n replaces a noun of the class II;
III muu +n replaces a noun of the class III;
IV mii +n replaces a noun of the class IV; etc.

To conclude this subsection, let me state three reasons that underlie the intuitive desire of a
linguist to distinguish these three types of Morph-D (cf. also 6, p. 00):
1) A morphological reason: under agreement and congruence (which is a particular case of
agreement) the target 'reflects' some properties of the controller; under government, this cannot
happen.
2) A semantic reason: under agreement the target is prototypically the Sem-governor of the con-
troller, which is its Sem-actant; under government the target is prototypically the Sem-dependent of
the controller, i.e. its Sem-actant; under congruence the target and the controller cannot be linked by
a Sem-D: they are coreferential.
 22

3) A syntactic reason: under agreement the target can be or not be linked by a direct Synt-D to
the controller; under government the target is necessarily linked by a direct Synt-D to the controller;
under congruence the target and the controller cannot be linked by a Synt-D.

4. Syntactic Dependency

4.1. General Remarks

Paraphrasing R. Jakobson, we can say that Sem-D is directly related to meaning and there-
fore it is conceivable or understandable, while Morph-D is directly related to (phonological) form
and therefore it is perceivable. The Synt-D is, however, not DIRECTLY related either to meaning or
to formit is more abstract, more indirect than Sem-D and Morph-D, and, as a consequence, more
questionable; even its mere existence needs justification.
Syntactic dependency has been used to talk about the structure of sentences from Antiquity
and throughout the Middle Ages to modern times. All respectable pre-20th century grammatical
traditions in syntax have been based on it, as has much language teaching. By the 8th century, Arab
grammarians (e.g. Sbawaih, who died in 798) already distinguished the governor vs. the depen-
dent in syntax and used this distinction when formulating rules for word order and inflection
(Owens 1988: 79-81). One finds dependency trees as a means of describing sentence structure in
German syntax books from the 19th century (Weber 1992: 13). In point of fact, constituency repre-
sentation in syntax, i.e. what became known as phrase-structure, was first introducedand that,
almost exclusively in the domain of the English language!in the early 20th century. The depen-
dency approach [= D-approach] was properly presented for the first time in Tesnire 1959 (the first
sketch of Tesnire's theory appearing in Tesnire 1934; see Ch. II); this profound treatise made
syntactic dependency available for serious theorizing. Yet, due to the dominance of Chomskian
Transformational-Generative Grammarwhich used, as its main syntactic tool, the phrase-struc-
ture representation (i.e. constituency)the D-approach did not become popular in modern linguis-
tics until the beginning of the 1980s.
Nevertheless, starting in the early 60s and over a period of about ten years, a number of pub-
lications which laid foundations for the D -approach had appeared (Hays 1960 [1961], 1964a, b,
Lecerf 1960, Fitialov 1962, 1968, Mel'uk 1962, 1963, 1964a, b, Iordanskaja 1963, 1967, Padu-
eva 1964, Gaifman 1965, Baumgrtner 1965, 1970, Marcus 1965a, b, Robinson 1970a, b,
Heringer 1970). All these studies were more or less inspired by computational applications of lin-
guisticsprimarily machine translation and other types of computer text processing. Gradually, the
field grew into real theoretical research, continuing to rely heavily on computer applications (e.g.,
Kunze/Priess 1967-1971, Sgall et al. 1969, Goralkov 1973, Machov 1975, Kunze 1975,
Hudson 1976, 1980a, b, 1984, Garde 1977, Korhonen 1977, Schubert 1987). And more recently,
several general linguistic theories have emerged, based partially or completely on the D-approach,
 23

including Case Grammar (Fillmore 1968, Anderson 1977), Meaning-Text Theory (Mel'uk 1974
[1999], 1979, 1988, 1997b), Lexical-Functional Grammar (Bresnan 1982), Relational Grammar
(Perlmutter 1983), Word Grammar (Hudson 1984, 1990), Functional Generative Description
(Sgall et al. 1986, Petkevi 1995), Lexicase Theory (Starosta 1988)cf. Hudson 1993b: 330-332.
Cognitive Grammar (Langacker 1987, 1991, 1997) is also dependency-oriented. One finds a few
university manuals which use the D -approach (e.g., Matthews 1981, Tarvainen 1981, Weber
1992). The description of German syntax in Engel 1977 [1982, 1994] and the syntactic part of
Engel 1988one of the most authoritative German reference grammarsare developed explicitly
within the D -approach (see especially Engel 1988: 21-26). A formal D -grammar for German
syntax (a list of rules and theoretical discussion) is found in Heringer 1996.
Let it be clear that, when speaking of the D-approach in what follows, I mean exclusively a
DEPENDENCY REPRESENTATION of the structure of sentences rather than a DEPENDENCY GRAM-
MAR, or a logical device consisting of rules that ensure the generation/parsing of sentences. The
two notions are of course logically related, but should be kept distinct. (Cf. Hudson 1993a: 266-
269 on the difference between syntactic heads in sentence structure and syntactic heads in grammar
rules.)

4.2. The Rationale for Syntactic Dependency

The notion of Synt-D is proper to Syntactic Structure [= SyntS]: a formal object used to
depict the organization of a sentence as opposed to its meaning, which is the target of the Semantic
Structure [= SemS]. Synt-Ds are building blocks of a SyntS, and so it will be useful to start with a
brief characterization of the latter.

Formal considerations
The SyntS of a sentence is called upon to 'mediate' between its SemS and its Morph(ologic-
al)S. The SemS is formalized as an arbitrary (n-dimensional) graph, i.e. a network, as we see in
(2). The MorphS is a 1-dimensional (linear) graph, i.e. a string, cf. (5). The SyntS constitutes a
convenient bridge between the SemS and the MorphS: under text synthesisthat is, in the transi-
tion from meaning to textthe SyntS must be easily produced from the Sem-network and easily
converted into the Morph-string; under analysisthat is, in the transition from text to meaningit
must allow for ease of the inverse operations. The simplest formal object that satisfies these
requirements is a 2-dimensional (planar) graph, i.e. a tree. Networks are relatively easy to
arborize, and trees are easy to linearize (text synthesis); vice versa, strings are relatively
easy to arborize, and trees are easy to convert to networks (text analysis). In other words, the Synt-
tree is the most convenient intermediary between the Sem-network and the Morph-string. That is
how the idea of SyntS as a dependency tree composed of lexemes is formally arrived at. If the
SyntS is a tree, then any of its arcs, or branches, represents an anti-reflexive, anti-symmetrical
 24

and anti-transitive binary relation between lexemesi.e. a Synt-D relation. This reasoning leads us
to the notion of Synt-dependency as an order relation (see Definition 3.5, 4.3, p. 00) and to the
notion of dependency tree as an appropriate formalism for the representation of SyntSs (see the end
of 4.4, p. 00).
What has just been said should not be construed as a proposal to consider the dependency
tree as an artifact of the linguist, a figment of his imaginationwithout any claim to psychological
reality. On the contrary, I think that the dependency tree is a reasonably good model of how
sentences are organized in the brain of the speakers. The dependency tree is proposed here as an
exclusive means for representing the SyntSs of sentences exactly because I believe that my brain is
using it when I am writing these lines.

Substantive considerations
Now I will consider the problem of SyntS from another angle. Suppose we want to represent
the SyntS of the sentence Leo knows that Alan is in love with Helen. There are exactly four types
of linguistic means that this sentence uses to express its meaning: lexemes, order of lexemes (i.e.
word order), prosody, and inflection. Note that:
1) there do not exist other types of linguistic means that could be used to express meanings;
2) these four types of linguistic means are used by all languages in all sentenceswith the
notable exception of inflection, which does not exist in quite a few languages and which, even in
the languages where it does exist, does not appear in all sentences and on all the wordforms;
3) each of these means can be used either as a direct expression of meaning, i.e. in a SEMANTIC
CAPACITY , or without a direct relation to meaningthat is, purely in order to indicate links
between wordforms in the sentence, i.e. in a SYNTACTIC CAPACITY, see Table 1.

Linguistic means used in semantic capacity used in syntactic capacity

full words (for, decades, cocoa, governed prepositions and conjunctions (as in
lexical units farming, escape, the, when, etc.) depend on, to order that, etc.)
arrangements that mark commu- arrangements that mark syntactic construc-
word order nicative structure (theme ~ rheme, tions : N + N, PREP + N, ADJ + N, V + N
given ~ new, etc.) [= DirO], etc.
prosodies that mark question vs. prosodies that mark borders of constituents
prosody assertion, focus, emphasis, ...,
irony, threat, tenderness, etc.
number in nouns; aspect and tense case in nouns; person and number in verbs;
inflection in verbs gender, number and case in adjectives (agree-
ment and government categories)

Table 1: Linguistic Expressive Means and Their Possible Uses

Non-lexical means used in syntactic capacity (shown in a double box in Table 1) should not
appear in a SyntS:13 they are MEANS USED TO EXPRESS the SyntS, therefore they cannot be part of
it. All of them appear closer to surface, in the DMorphS of the sentence, while the SyntS must re-
place them with a formal simple homogeneous device. This device must be able to encode the
 25

future linear arrangement of wordforms, i.e. word order, in an explicit, clear and elegant way.
Note that syntactic prosody applies to a previously ordered sequence of wordforms, and inflection
is absent in many cases, so that these two linguistic means are secondary from the viewpoint of the
SyntS. The SyntS has to tell us, first of all, where to position a wordform w2: before or after ano-
ther wordform w1and then give us more details about mutual positions of different wordforms
which have to be positioned on the same side of w1. The most economical way to do this is using a
binary anti-reflexive, anti-symmetrical, and anti-transitive relation between the wordforms of the
sentencean order relation (in the logical sense). This is nothing but a Synt-D; thus, we have once
again, this time via substantive reasoning, come to the same conception of Synt-D relation.
As a bridge between the SemS and the DMorphS of a sentence, the D-/S-SyntS must encode
all the relevant semantic contrasts that are expressed on the surface and all the relevant formal con-
trasts that carry meaning. Therefore, the specific Synt-D relations that are introduced for a given
language must be such as to satisfy this requirement.

4.3. The Concept of Syntactic Dependency

What is of special importance for a good understanding of syntactic dependency is the fact
that at the very beginning it was not, and even now it is still not always, rigorously distinguished
from Sem-D and Morph-D. Linguists would often talk about dependency tout court, aiming at the
Synt-D, but in actual fact taking in a mixture of the three.
Since Synt-D is an abstract formal concept, it is not as easy to define as Sem-D and Morph-
D . Three groups of criteria for Synt-D have to be introduced; but first, let me emphasize that for
simplicity's sake I will be dealing exclusively with Surface-Syntactic [= SSynt-]dependency. The
results can be easily generalized to cover Deep-Syntactic dependency as well.
To establish a SSynt-D relation between two wordforms in a sentence we need (at least):
A. Criteria for SSynt-CONNECTEDNESS of the two wordforms (= for the presence of a SSynt-
D between them).
B. Criteria for the SSynt-DOMINANCE between the two wordforms (= for the orientation of the
SSynt-D between them).
C. Criteria for the specific TYPE of the given SSynt-D between the two wordforms (= for the
type of the SSynt-relation between them; as will be shown, to ensure a proper representation of
syntactic structure of utterances, we have to distinguish, in a particular language, many different
specific types of Synt-D).
These criteria are necessary, but unfortunately not sufficient. Thus, when establishing the
types of SSynt-Ds in a given language, the researcher has to invoke relevant linguistic properties of
different dependents subsumed under the same SSyntRel (Iordanskaja & Mel'uk 2000). While
deciding on the presence and orientation of SSynt-dependencies some heuristic principles have to
 26

be recurred to (see Ch. III, 1.1.1, p. 00). And more than anything else, reasoning by analogy
remains the most necessary tool: the description, in terms of Synt-D , of a 'dubious' phrase P1
should correspond to the SSynt-description adopted for the similar phrase P2 where the situation is
clearer or outright obvious.

4.3.1. Criteria A: SSynt-Connectedness

First of all, one has to know whether two wordforms in a sentence, w1 and w2, are syntac-
tically directly linkedor not. To answer this question, we need Criteria A1 and A2.

Criterion A1: Linear arrangement of wordforms

Wordforms w and w considered in a communicatively neutral sentence of language L have a
1 2
direct Synt-D link between them only if the linear position in this sentence of one of them must
be specified with respect to the other.

Comments
1. In languages where word order is used semanticallyamong other things, to express
communicative organization (the Rheme/Theme division, the Old vs. New, Focalization, Empha-
sis, Contrast, etc.)Criterion A1 applies in a limited way: it has to be applied only to communicati-
vely neutral expressions, i.e. to expressions without extractions, permutations or other communica-
tive transformations of all kinds.
2. When we say that the linear position of the wordform w1 is specified with respect to the
wordform w2, this means that w2 either precedes w1, or follows it, or else can precede or follow
it (optionally or under some conditions). Thus, if in a language, the manner adverbial can indiscri-
minately precede or follow the verb it modifies, we still formulate the possible positions of the
adverbial with respect to the verb, rather than the other way around.
3. The wordform determining the linear position of the other is not necessarily its Synt-go-
vernor (cf. item B below): thus, in the phrase PREPN, it is the Synt-governor PREP that is
positioned with respect to its Synt-dependent N. Of course to say that a PREP precedes the N it
introduces is logically equivalent to saying that an N introduced by a PREP follows this PREP.
However, linguistically these two statements are not equally acceptable: since a noun can appear
without a preposition, while a preposition cannot appear without a noun, it is more natural to
specify the place of the preposition with respect to the noun than the other way around. (By the
way, the etymology of the word preposition is a witness to exactly this intuition: it is an element
that is PRE-posed to the noun.)
4. In some 'exotic' cases Criterion A1 has to be applied, so to speak, with special caution.
Thus, in Serbo-Croatian, in the construction VAuxVnon-fin (Vnon-fin is a Past Participle in the
compound past tense, and an Infinitive or a clause with the conjunction DA (that) in the compound
 27

future) it is VAux that is the SSynt-governor: Ja sampisao (I [MASC.SG] have written), Ti sipi-
sala (You [SG.FEM] have written), Ja upisati or Ja uda piem (I will write), etc. (I cannot cite
here the arguments in favor of this description: see Milievi 2000: 00-00.) However, the linear
position of G [= VAux] is determined without any reference to its D [= PPart/Inf/DA-clause]: VAux
is a clitic, and Serbo-Croatian clitics are placed as a cluster, roughly speaking, after the first consti-
tuent of the clause, whatever the syntactic class and syntactic role of this constituent (some more
specific conditions apply). Nevertheless, BEFORE the clitics are placed where they belong, the VAux
cliticthe SSynt-head (= top node) of the clauseserves as the reference point for the linear place-
ment of all the other clause elements, exactly in the same way as any other top node of a clause
does. Therefore, even if one of the wordforms w 1 and w 2 is a clitic, Criterion A1 is still fully
applicable (but cum grano salis).

Criterion A2: Potential prosodic unity

Wordforms w and w considered in a given sentence of the language L can have a direct Synt-
1 2
D link between them only if:
(a) Either w1 and w2 can form an utterance in L , i.e. a special prosodic unita phrase
[general case]:
e.g., N + V, N + ADJ, V + N, PREP + N, ADV + ADJ, NUM + N, etc.
(b) Or the wordforms w 1 and w 2 cannot form a phrase, but the wordforms w 1, w 2 and a
set of wordforms W can, such that 1) in this phrase, w1 is the Synt-head and 2) w2 and
W also form a phrase in which w2 is the Synt-head [special case]:
e.g., escape [= w1] from [= w2] many problems [= the set W], where escape from many
problems is a phrase in which escape is the Synt-head; from many problems is also a
phrase in which from is the Synt-head; therefore, we say that escape and from have a
direct syntactic link: escapesyntfrom.14
Of course not every prosodic unit in an actual sentence is a phrase; the concept of phrase
needs an elaborate definition, which is outside of my frame here, because it is a concept of the
Deep-Morphological, rather than Syntactic, level. I take it to be one of my indefinibilia, see Ch. I,
1, p. 00. However, recall that a phrase must be perceived by speakers as existing in the language,
whatever this means.

Criteria A1 and A2 must of course not contradict each other. For instance, in (1), for has to
be positioned before decades, and escaped after has, etc.: therefore Criterion A1 indicates the pre-
sence of a Synt-D in these pairs. Criterion A2 does not contradict this: in (1), for decades is a
phrase of English, and so is has escaped (but, e.g., *to new is not); therefore, in for decades and
has escaped the wordforms can be linked by a Synt-D. Again in (1), by moving is positioned after
 28

escaped, and bybefore moving (Criterion A1); escaped by moving is a phrase, with escape as the
Synt-head, and so is by moving, where the preposition by is the Synt-head; therefore, by Criterion
A2-b, escaped and by can be linked by a Synt-D. Both criteria are again fulfilled.

For Criteria A to be satisfied, that is, for there to be w1 syntw2, both Criterion A1 and
Criterion A2 must be satisfied.

4.3.2. Criteria B: SSynt-Dominance

Next, in each pair of wordforms w 1 and w 2 which are syntactically directly linked in the
sentence under consideration, one of them syntactically dominates the other, i.e. is its Synt-gover-
nor. In the phrase w syntw the Synt-governor is the wordform that determinesat least,
1 2
to a greater extent than the other wordform (its Synt-dependent)different properties of the
phrase according to Criteria B1, B2 and B3.

Criterion B1: The passive SSynt-valency of the phrase

In the phrase w1syntw2, the wordform w1 is the Synt-governor of w2 if the passive SSynt-
valency of the whole phrase is determined by the passive Synt-valency of w1 to a greater extent
than by that of w2.
To put it differently, the passive Synt-valency of w1syntw2 is rather that of w1 than that
of w2; the Synt-head of a phrase determines more than any other of its elements all the external
syntactic links of the phrase. (For passive SSynt-valency, see Ch. I, 1, item 14, p. 00.)
Note that Criterion B1 does not require EXACT distributional equivalence between the Synt-
head of a phrase and the whole phrase, as is the case in some similar approaches. For us, it is
sufficient if, in the phrase w syntw , the wordform w contributes to the passive SSynt-
1 2 1
valency of w 1syntw 2 more than w 2 does.

Examples
(14) a. The passive SSynt-valency (= the distribution) of the phrase for decades is fully determin-
ed by the preposition; therefore, forsyntdecades.
b. Similarly, a phrase like has escaped or does not escape shows the distribution of, or
plays the same Synt-role as, has/does (i.e. that of a finite, or tensed, verb) rather than
that of the past participle escaped or the infinitive escape; therefore,
hassyntescaped, doessyntescape.
c. The phrase Sir Wanner has the passive SSynt-valency of Wanner, not that of Sir: I see
Sir Wanner ~ I see Wanner ~ *I see Sir; therefore, SirsyntWanner.
 29

By analogy with such nouns as Sir or Mister all other nouns possible in this construc-
tion are treated alike:
GeneralsyntWanner, ProfessorsyntWanner, PresidentsyntWanner, etc.15
Let us consider now a more difficult case where it is not immediately obvious what element is
the SSynt-governor.
(15) a. French
un drle de garon (a strange boy)
a.SG.MASC strange.SG.MASC of boy[MASC].SG
une drle de voiture (a strange car)
a.SG.FEM strange.SG.FEM of car[FEM].SG
ces drles de garons (these strange boys)
this.PL.MASC strange.PL.MASC of boy[MASC].PL
ces drles de voitures (these strange cars)
this.PL.FEM strange.PL.FEM of car[FEM].PL
The passive SSynt-valency of the phrase drle de garon (ADJ+de+N) is that of a noun and
not that of an adjective; what should be taken as the head of the phrases in (15a)?
Solution I: the noun (GARON, VOITURE) is the head. The internal SSynt-structure of the
synt
phrase is as follows: drle s y n
t de garon . We have then to treat DE not as a preposition, but

as a special adjectival marker (homophonous with the preposition DE and depending on the adjec-
tive). The adjective agrees in gender and number with its SSynt-governor, which is the rule in
French.
Solution II: the adjective (DRLE) is the head. The internal SSynt-structure of the phrase is as
follows: drlesyntdesyntgaron. We have then to admit that an adjective of such a type
(French has a handful of those: DRLE, CHOUETTE, VACHE16) has bizarre SSynt-properties: it
can be the head of a noun phrase, while governing a DE-phrase and agreeing with the dependent
noun of this DE-phrase, instead of with its own SSynt-governor.
Solution I requires the postulation of a special grammatical elementan adjectival marker
DEwhich does not exist in French elsewhere, i.e. outside of the construction under analysis.
Solution II, on the contrary, requires only the admission of a special character of three French
adjectives, which has to be admitted anyway (since even under Solution I, such a construction will
be possible only with these adjectives). Moreover, the construction with an adjective that heads an
NP while governing a DE-phrase and agreeing with the dependent noun does exist in French inde-
pendently: le plus intelligentsyntdes garons (the most intelligent of the boys) ~ la plus intel-
ligentesyntdes filles (the most intelligent of the girls); it is an absolutely regular and completely
productive construction. Therefore, Solution II has to be preferred. (As we see, the decision is
again arrived at by analogy.)
 30

While the construction of the type un drle de garon is very restricted in French, it is quite
productive in Sardinian:
b. Sardinian
unu bette de pittsinnu (a big boy)
a.SG.MASC big.SG.MASC of boy[MASC].SG
una ruja de mkkina (a red car)
a.SG.FEM red.SG.FEM of car[FEM].SG
sa manna de ampulla (the big bottle)
the.SG.MASC big.SG.FEM of bottle[FEM].SG
udda de mkkina (that car)
that.SG.FEM of car[FEM].SG
Although I do not have enough data on Sardinian, I think that all the phrases of the type illus-
trated in (15b) have to be described on the SSynt-level as it is proposed for (15a)with the
adjective as the Synt-head of the phrase:
unu syntbettesyntdesyntpittsinnu.
Thus, the examples in (15) show that in more complicated, 'exotic' cases one has to proceed
with utmost caution. The main tool here is ANALOGY with more normal (= more current, less res-
tricted) constructions. We have to make decisions that will agree with accepted descriptions and try
to relegate the eccentricities to restricted sets of phenomena, without allowing these eccentricities to
spread on more normal areas of the language.

Criterion B2: The morphological links between the elements of the phrase and
its external context
If in the phrase w 1syntw 2, in which the passive SSynt-valency does not allow us to esta-
blish the Synt-governor, it is w 1 is the Synt-governor of w 2 if w 1 controls the inflection of
wordforms external to the phrase or w1's inflection is controlled by such wordforms.

The wordform w 1 is called the morphological contact point of the phrase

w 1 synt w 2 .

Examples
(16) a. The Russian phrase jubka-tany, lit. (skirt-pants), does not allow for the application of
Criterion B1 (both its members are nouns and have the same passive Synt-valencies); but
Criterion B2 singles out jubka as the Synt-governor: ta [SG.FEM ] jubka-tany byla
[ SG.FEM ]... (this skirt-pants was...), where the external agreement is with jubka
[ F E M . S G ], and not with t a n y [P L ] * t i j u b k a - tany byli...; therefore,
jubkasynttany.
 31

b. In the phrase v tate Nebraska (in [the] state [of] Nebraska), tat is declined regularly
(tat, tata, tate, ...) in conformity with external context, while Nebraska remains in
the nominative (v Nebraske, but *v tate Nebraske); thus tat is here the morphological
contact point, and it is again picked by Criterion B2 as the Synt-governor:
tatsyntNebraska.

c. Similarly, in the phrase of the type udo-jabloko, lit. (miracle-apple), jabloko (apple) is
the Synt-governor, since it is declined according to the requirements of the external con-
text while udo remains invariable: udo-jabloka, udo-jabloku, ..., udo-jabloki, u-
do-jablokami, ... Thus, we have udo syntjabloko.

d. In the phrase [pjat] kilogrammov kolbasy ([five] kilos of-sausage), the noun kilogram-
mov is the Synt-governor, since it is the morphological contact point:
[s pjatju] kilogrammamisyntkolbasy ([with 5] kilos [of] sausage),
[v pjati] kilogrammaxsyntkolbasy ([in 5] kilos [of] sausage), etc.

e. Likewise in Germ. [zwei] Glser Wein, lit. ([two] glasses [of] wine), the Synt-gover-
nor is Glser, which is the morphological contact point:
i. [zu diesen zwei] Glser+n Wein, lit. ([to these two] glasses [of] wine), where Glsern
is in the dative, imposed by the preposition ZU;
ii. Dies+e [PL] zwei Glser Wein sind [PL] notwendig (These two glasses [of] wine are
necessary), where Glser [PL] imposes the plural grammeme on the adjective and on the
verb. Therefore, GlsersyntWein.
By analogy, the same structure is accepted in the cases where the measure noun remains
invariable:
drei KilosyntBrot (three kilos [of] bread),
vierzig GrammsyntFleisch (forty grams [of] meat).

f. In Dutch, the situation is slightly different from that in German: here, the Nmeasure does
not inflect under the impact of the external context (it has no case forms), but when in
plural, it imposes plural agreement on the verb:
Twee glazen wijn zijn [PL] *is [SG] noodzakelijk (two glasses [of] wine are necessary);
therefore, in Dutch we also have glazensyntwijn.
But in semantically equivalent phrases of Chinese, which has no inflection at all, the Synt-Ds
are different, see (17b).
 32

Criterion B3: The semantic content of the phrase

In the phrase w 1syntw 2, in which neither the passive SSynt-valency nor the morphology
allow us to establish the Synt-governor, w1 is the Synt-governor of w2 if w1syntw2 means
(a kind/an instance of w1) rather than (a kind/an instance of w2).

Examples
(17) a. In jam sandwich, the Synt-governor is sandwich, because "jam sandwich refers to a kind
of sandwich, rather than to a kind of jam" (Hudson 1990: 98).

b. In Chinese, where no inflection exists, the phrase sh bng ru (ten pounds [of] meat)
consists of morphologically invariable wordforms. Here again, Criterion B3 applies: sh
bng ru refers to an instance of meat, not to an instance of pounds, so ru (meat) is the
Synt-governor: sh syntbng syntru.

One can say (with Zwicky 1991: 4) that in a two-word phrase the Synt-governor is the syn-
tactic category determinant, orif there is no such syntactic determinantthe morphological beha-
vior determinant, orin case both syntactic and morphological determinants are absentthe se-
mantic content determinant. In one word (Bazell 1949: 11), the Synt-governor is more PROMINENT
than its Synt-dependent, namelymore prominent syntactically, or else morphologically, or at least
semantically.
Most approaches dealing with Synt-D s require concord between these properties, i.e. be-
tween Criteria B1-B3. In sharp contrast, in the Meaning-Text Theory such a concord is not requir-
ed. Only Criterion B1 is genuinely syntactic; B2 is morphological, and B3 semantic. And we know
already that the orientations of Sem-D, Synt-D and Morph-D can differ (cf. 5); therefore, we must
expect that these criteria will be in conflict more often than not. For me, Criteria B1-B3 form a
hierarchy: B1 > B2 > B3. Thus, if Criterion B1 is applicable, its indication is sufficient. Only if it
is not applicable (because w1 and w2 are both of the same part of speech and thus have the same
passive SSynt-valency), Criterion B2 appliesbut only in a language having inflection and only
for w1 and w2 with different morphological properties. Otherwise, Criterion B3 applies. There-
fore, these criteria are never applied together (= simultaneously) and, as a result, they cannot con-
tradict each other. To put it in a slightly different form: The ability of Synt-governors to control the
inflectional form of their Synt-dependents/to have their own inflectional form controlled by a Synt-
dependent or their ability to be or not to be semantically dominant should not be taken into account
when deciding on the Synt-governor status of a wordform: morphological and semantic properties
of heads are, as already stated, freely distributed among Synt-governors and Synt-dependents, so
that a consistent combination of these properties cannot be expected.
 33

For Criteria B to be satisfied, at least one of the Criteria B1-B3 must be satisfied, such that
other Criteria B higher in the hierarchy are not applicable.

The criteria for the orientation of Synt-D ('Head-vs.-Dependent' problem) are thoroughly
discussed in Pittman 1948, Zwicky 1985, 1991, Hudson 1987, 1990: 106-107, and in Corbett et
al. (eds) 1993. For a more rigorous formulation of Criterion B1, see Mel'uk 1988: 132-135.
Criteria B1-B3 call for the following two important remarks.
First, Criteria B1-B3 are LANGUAGE-SPECIFIC: if, in the phrase X + Y of language L (X and
Y being of different parts of speech), these criteria pick X as the Synt-governori.e. we have
XYthis will not necessarily be the case for a synonymous construction with similar parts of
speech in some other language. Thus in Russian and German NmeasureN, because Nmeasure is the
morphological contact point (cf. [v pjati] kilogramm+ax kolbasy in (16d) and [zu diesen zwei]
Glser+n Wein in (16e)); yet it does not follow that N syntactically depends on a quantifying
Nmeasure in any language: in a language where the Nmeasure does not inflect under the impact of an
'external' wordform and does not itself control the inflection of the quantified N, Criterion B3
picks this N as the Synt-governor: cf. (17b), where we have N N in Chinese.
measure
Second, Criteria B1-B3 are INHERENTLY INSUFFICIENT: there are cases where all the three
fail. This must happen where, in a phrase X + Y, both X and Y are of the same part of speech,
neither does inflect nor can impose different inflections, and both are semantically 'equal.' Take,
for example, the Russian phrase of the type vera utrom, lit. (yesterday morning), or segodnja
popoze, lit. (today later). Both wordforms in this phrase are adverbs, both have no morphology,
and both denote time; which one is the Synt-governor? Note that both are equally omissible: Alen
priexal vera (Alan came yesterday) and Alen priexal utrom (Alan came in the morning). In such
cases, a more or less arbitrary solution imposes itself: the preceding element will be taken as the
Synt-governor, so that we have verasyntutrom, segodnjasyntpopoze. However, there
could be semantic motivation for this solution, after all: (yesterday) and (today) are in a sense more
important than (in the morning) and (later), since (yesterday)/(today) denotes a whole day, of which
(in the morning)/(later) is but a part. Then Criterion B3 applies: (yesterday morning) is a particular
moment of yesterday, and (today later) a particular moment of today. We can also invoke a syntactic
consideration: yesterday morning represents a kind of coordination, and in coordinate strings, the
subsequent element depends on the preceding one.
An even more problematic case is that of compound numerals in languages where numerals
are morphologically invariable themselves and do not govern special inflections of the quantified
nouns.17 Take, for instance, Fr. soixante-neuf (69). Since both its components are numerals, Crite-
rion B1 is not applicable (soixante and neuf have the same passive SSynt-valency); since almost all
French numerals have no morphology and do not affect the morphology of the noun quantified,
 34

Criterion B2 is not applicable, either; finally, their meanings are strictly of the same type (=
numbers), so that neither Criterion B3 can be used. The only way open is then to reason by
analogy. The compound numeral soixante et un, lit. (60 and 1) (and a few others with 1 as the last
digit), would suggest the Synt-dominance soixanteetun; by analogy with regular conjoined
strings of the type AlanandLeo or beautifulandintelligent. But then consider two facts that
contradict this solution:
The numeral UN (one) agrees in gender with the noun quantified: vingt et un garons [MASC]
(twenty-one boys) vs. vingt et une [FEM] filles (twenty-one girls); according to Criterion B2, it is
UN that must be the Synt-head.
Take the ordinals, such as soixante et unime (sixty first) or soixante-cinquime (sixty-fifth)
(similarly, soixante et onzime, lit. (60 and 11th) = (71st) et quatre-vingt-onzime, lit. (80-11th) =
(91st)); here the Synt-governor is, according to Criterion B1, clearly the ordinal numeral unime
(1st), cinquime (5th) and onzime (11th), i.e. the last numeral in a compound ordinal:
troiscentsoixantecinquime (365th),
troiscentsoixanteetonzime (371st), etc.
Then, continuing our analogy and taking these two facts into account, we arrive at the same SSyntS
in compound cardinals: troiscentsoixantecinq (365). And, of course,
troiscentsoixanteetun (361).
In a language like German, where some numerals are regularly linked by a conjunction (und
(and)), this gives the following Synt-structures:
dreihundertfnfundsechzigster (365th), lit. (three hundred five and sixtieth),
where sechzigster (sixtieth) is clearly the Synt-head of the compound ordinal numeral; in a similar
way, dreihundertfnfundsechzig (365), lit. (three hundred five and sixty).
It is possible that elements like (and) (Fr. et, Germ. und) that appear within compound nume-
rals should not be considered coordinate conjunctions; then the SyntSs shown above would look
less exotic; cf. the Chukchee marker of compound numerals in (24c), p. 00.

4.3.3. Criteria C: Labeled SSynt-Dependencies

Last, for each pair of wordforms w syntw which are syntactically directly linked in a
1 2
particular direction, one has to know exactly which specific type of the Synt-D links them. In order
to represent successfully the SyntSs of utterances, we have to use DIFFERENT types of Synt-D .
Thus, IsyntLOVEsyntYOU does not distinguish between (I love you) and (You love me);
[JOHN] I syntSEND syntYOU can be implemented as (John sends me to you) or (John
sends you to me); etc. In all these and a host of similar cases, different types of Synt-D s, or
different Surface-Syntactic Relations [= SSyntRels] have to be distinguished:
Ir1LOVEr2YOU, [JOHN] Ir2SENDr3YOU, ...
 35

As we see, SSyntRels must be labeled, the label being meaningful (as is the case with
Morph-D): the label r of a SSyntRel refers to a family of specific syntactic constructions which im-
plement, in the DMorphS of the sentence, the SSyntRel r. Thus, consider the label subj(ectival)
of a SSyntRel in Russian, i.e. the SSyntRel that appears in phrases of the type
V fin[= w ] subjN[= w ]
1 2
(Malik prixodit/Devoka prixodit (The boy comes)/(The girl comes); Malik pril/Devoka prila
(The boy came)/(The girl came)). The label subj identifies a set of SSynt-rules that make the finite
verb w1 agree with the noun w2 in person and number (if the verb is in the present or the future)
or in number and gender (if the verb is in the past or the subjunctive); these rules also position w2
[= N], with respect to w 1 [= V]. In other words, the SSyntRel subjectival is the signified
(= Saussure's signifi) of every construction in this family; generally speaking, a SSyntRel is a
component of a linguistic sign, whose signifier is the construction in question (an ordered pair of
lexemic classes with particular morphological characteristics).18
In phrases of the form w1rw2, the Synt-D that links the two wordforms can be labeled r
(i.e., it can be the SSyntRel r) only if it satisfies the following three criteria: C1-C3. If at least one
of Criteria C1-C3 is not satisfied, the presumed SSyntRel r should be split in two (or more)
SSyntRels.

Criterion C1: Absence of semantic contrast

Notations: w(L) is a wordform of lexeme L (wi and wj can be different or identical); O is
the operation of linguistic union, which links signs, in particularwordforms, according to their
syntactics and general rules of L .

A SSyntRel must not describe two different phrases

w i(L1)Ow j(L2) and w m (L1)Ow n(L2), where L1 syntL 2,
which 1) contrast semantically [(wi(L1)Owj(L2)) (wm(L1)Own(L2))]
and 2) differ formally only by some syntactic means of expression (i.e. by word order, by
syntactic prosody, or by syntactic grammemes).

Criterion C1 corresponds to what is known in linguistics as 'the minimal pair test,' which is
used in phonology (= two phones cannot be relegated to one phoneme if they are the only distingui-
shers of the signifiers of two semantically contrasting wordforms), morphology, and semantics.

Examples
(18) a. In Russian, the construction DESJATrDOLLAR has two different implementations
with different meanings:
desjat dollarov (10 dollars) vs. dollarov desjat (approximately 10 dollars);
 36

the formal difference between the two phrases is purely syntactic: word order; therefore,
the presumed SSyntRel r must be split in two different SSyntRels:
DESJATquantitativeDOLLAR desjat dollarov
DESJATapprox-quantitDOLLAR dollarov desjat

b. In English, the construction STARSrVISIBLE also has two different implemen-

tations with different meanings (Quirk et al. 1985: 419; cf. Mel'uk/Pertsov 1988: 136-
137):
the visible stars (stars that can be seen in principle)
vs.
the stars visible (stars that can be seen currently);
the formal difference between the two phrases is again purely syntactic: word order; there-
fore, there are two different SSyntRels as well:
VISIBLE modificativeSTARS the visible stars
VISIBLE post-modificativeSTARS the stars visible
Cf. other pairs with the same semantic/formal difference: navigable rivers ~ rivers navig-
able, a written word ~ a word written, a sitting figure ~ the people sitting, etc.

Criterion C2: Syntactic substitutability

The first formalization of the SSynt-substitutability of syntactic subtrees as a means for
establishing SSyntRels was proposed by the German researcher J. Kunze (Kunze 1972: 23; see
also Kunze 1975: 5.3, p. 235ff): the so-called Kunze property. I start with presenting it here, in
order to show that a weaker version of it must be preferred.
Let there be, in L , lexemes L(X), L(Y), ... of syntactic classes X, Y, ..., complete SSynt-
configurations (Z) and (W) (i.e. subtrees having as their top nodes lexemes L(Z) and L(W)), and a
SSyntRel r.

Definition 3.1: The Kunze Property

A SSyntRel r has the Kunze Property if, and only if, for any pair of SSynt-configurations L(X)
r (Z) and L(Y) r (W) , replacing (Z) by (W) and vice versa does not affect their
19
SYNTACTIC well-formedness.

To put it differently, for a SSyntRel that has the Kunze property, any of its potential Ds can
be attached to any of its potential Gs (= all Ds of a SSyntRel are mutually substitutable in all
SSyntSs salva correctione). In Mel'uk 1988: 142 it was required that any SSyntRel in any L has
the Kunze property.20 Now, however, I think that the Kunze property is too rigid, since it does not
allow for some desirable generalizations. For instance, it does not admit the same SSyntRel for
nominal and infinitival SSynt-Subjects, as in the following French sentences:
 37

(19) La courserfatigue, lit. (The running tires).

Courirrfatigue, lit. (To-run tires).
Since far from any verb in French takes an infinitive as its SSynt-Subject (*Pleuvoir m'a surpris,
has caught me (out))), the SSyntRel r in (19) does not possess the Kunze property:
lit. (To-rain
with L = SURPRENDRE, (Z) = NP (e.g., La pluiersurprend) and L(Y) = FATIGUER,
(X)
(W) = Infinitival Phrase (Courirrfatigue), the replacement produces the syntactically ill-formed
configuration *VinfrSURPRENDRE ([to] catch N (out)). As a result, using the Kunze property
leads to having two different SSyntRels for nominal and infinitival SSynt-Subjects (as stated in
Kunze 1975: 279). But I think that in (19) the SSyntRel r should not be split: all the SSynt-
Subjects, whether nominal or infinitival, share a set of important unique properties, and it is
preferable to describe all of them by the same SSyntRel.
Therefore, it is proposed to use the quasi-Kunze Property, which is weaker: substitutabi-
lity is required only in one direction and only by at least one particular subtree (which is not a sub-
stitute pronoun, since substitute pronounssee Footnote 12, p. 00constitute a 'secondary'
syntactic class, being introduced by a 'transformational' rule), rather than in both directions and by
any subtree. (The concept of the quasi-Kunze Property has been elaborated jointly with L. Iordan-
skaja; it is introduced in Iordanskaja & Mel'uk 2000. Another weaker version of the Kunze
Property was considered in Mel'uk 1977: 261.)

Definition 3.2: The Quasi-Kunze Property

A SSyntRel r has the quasi-Kunze Property if, and only if, there exists in L a syntactic class (
part of speech) X, which is different from substitute pronouns and such that for any SSynt-
configuration Lr (Y) , replacing (Y) by (X) (but not necessarily vice versa!) in any
SSyntS does not affect its syntactic well-formedness.

The element (X) that 'passes' with any governor of the SSyntRel r is nothing else but the
PROTOTYPICAL D of the SSyntRel r.
The SSyntRel r in (19) possesses the quasi-Kunze property, since this r has a prototypical D:
a prepositionless nounbecause in French any finite verb admits a nominal SSynt-Subject.21 As a
result, the SSyntRel r is allowed: this is the subjectival SSyntRel.
Let it be emphasized that, while in definitions 3.1 and 3.2 the G is a particular lexeme,
(Y)
is considered UP TO THE SYNTACTIC CLASS. Thus, for instance, different prepositions are not dis-
tinguished: the SSyntRel r in the phrases insistron, supplyrwith and comparerto has
the quasi-Kunze propertybecause a PREP+N phrase can be substituted for its D with any of
these verbs, provided the appropriate preposition is chosen according to the verb's Government
Pattern.
 38

Criterion C2 can now be formulated as follows:

Any SSyntRel r must possess the quasi-Kunze Property.

Criterion C2 is a particular case of what is known as the 'substitution criterion,'

or 'substitution test,' except that here we deal with the substitution of subtrees which
have to hang on the same SSyntRel r.

Examples
(20) a. In Russian, in the phrases vrPari (to Paris) and tobyritat (in-order-to read) the
presumed SSyntRel r does not possess the quasi-Kunze Property: *tobyrPari,
*vritat (Russian has no prototypical D for this SSyntRel: no element can pass with
both a preposition and a conjunction);
therefore, there are two different SSyntRels:
V prepositionalPARI
and
TOBY conjunctional-infinitivalITAT.

b. In English, in the phrases haverbeen and bergoing the presumed SSyntRel r does
not possess the quasi-Kunze Property: *havergoing and *berbeen;
therefore, there are two different SSyntRels:
HAVE perfect-analyticalBEEN
and
BE progressive-analytical GOING.

Criterion C3: Repeatability with the same Synt-governor

Let me consider the possible number of occurrences of a given SSyntRel r with the same
governor. In this respect, any r of language L can be either non-repeatable or unlimitedly
repeatable.

Definition 3.3: Non-repeatable SSyntRel

A SSyntRel r is non-repeatable if, and only if, no more than one branch labeled r can start from
any G(overnor).
In other words, in a given sentence of L , a G of a non-repeatable r can have only one D (=
one clause element) of the corresponding type. For instance, actantial SSyntRels whose Ds are
marked by purely syntactic means (word order, prosody, inflection)such as the subj and the
dir(ect)-obj(ectival) SSyntRelsare obligatorily non-repeatable: otherwise, they would violate
Criterion C1, because their Ds would contrast semantically, while differing only in syntactic
 39

means. (Only actantial SSyntRels whose Ds are marked by lexical means, that is, by different pre-
positionssuch as the oblique-objectival SSyntRel can be repeatable.) 22
An important warning: In some languages, a clause element can be DUPLICATED by what is
called a resumptive clitic. Such is, for instance, the D of the dir-obj SSyntRel in Spanish,
where we have the construction of the type (21a):
dir-obj

(21) a. Sp. A Alain ledir-objveo todos los das, lit. (Alan him [I] see every day)
[a human DirO in Spanish is introduced by the preposition A (to)].
We do not consider pronominal duplication of a clause element as repeatability, since such duplica-
tion has a grammaticized character and is 'orthogonal' to the genuine cooccurrence of SSyntRels,
since the noun and the clitic that duplicates it are necessarily coreferential; in spite of expressions of
the type (21a), the dir-obj SSyntRel is considered non-repeatable in Spanish. Similarly, in spite of
(21b), the indir-objectival SSyntRel is also considered non-repeatable in French:
indir-obj

b. Fr. mes enfants, je leur indir-objpermets tout,

lit.(To my children, I permit them everything).
Here is another example of grammaticized duplication by clitic:
c. Albanian
(i) Msuesi u foli fmijve, lit. (The-teacher to-them talked to-the-kids),
where fmijve and u also are both IndirOs.
(ii) Njerzit m pan mua, lit. (The-people me saw me),
where mua and m are both DirOs.
In all such cases, the corresponding SSyntRel is considered non-repeatable.

Definition 3.4: (Unlimitedly) Repeatable SSyntRel

A SSyntRel r is (unlimitedly) repeatable if, and only if, several branches labeled r can start from
a G (of course with the exclusion of resumptive clitics).
For instance, the modificative and the circumstantial SSyntRels in English are unlimitedly
repeatable; so is the obl(ique)-obj(ectival) SSyntRel. For a repeatable SSyntRel r the number of
branches labeled r that can start from a G in any particular case is theoretically unlimited, although
in practice, this number can be limited either by pragmatic considerations or by the lexicographic
properties of concrete Gs, for instance, by their Government Patternas is the case with the obl-
obj SSyntRel; this number cannot be limited by any general syntactic factors.
In other words, a SSyntRel cannot be LIMITEDLY repeatable (without being constrained by
Government Pattern of the G).
 40

Now I can formulate Criterion C3:

Any SSyntRel r must be unlimitedly repeatable or non-repeatable.

As is always the case, exceptions are possible. Thus, in English, the relative SSyntRel is
non-repeatable: generally speaking, a noun cannot have more than one relative clause. There is,
however, a contradicting phenomenon: two restrictive relative clauses with the same noun are
possible under specific conditions in highly colloquial speech, cf. (22):
(22) a. A student [who comes to my class ]1 [who broke the news to me]2 left the building.
b. We are in the room [I will never forget ]1 [where she kissed me for the first time]2.
If we decidein spite of their marginalityto consider such facts, they can be fully and exactly
circumscribed. Therefore, they constitute a legitimate exception, which does not prevent us from
declaring the relative SSyntRel non-repeatable in English. (This case has been brought to my atten-
tion by L. Iomdin.)

Criterion C3 corresponds roughly to the 'cooccurrence test,' used in linguistics on

all levels of analysis. In morphology, an element of a morphological category is either non-
repeatable (tense or number in English or French) or unlimitedly repeatable (the causative in
Turkish). When we see, for instance, just two possible repetitionslike nominal case suffixes in
Basque or Georgian, we speak of two different case categories (semantic case vs. syntactic case;
governed case vs. agreeing case).

Example
(23) In Persian, we find extremely widespread expressions of the following type:
Rmin+rrkardrbedr,
Ramin DirO made awakening [Noun]
lit. ([He/she/it] made [the] awakening Ramin) = (He/she/it awoke Ramin).
These expressions are built on verbal collocations of the type bedr kard (awakening make) =
(wake) or dars dad, lit. (lesson give) = (teach), which, although they seem to include a DirO, such
as bedr or dars, behave as transitive verbs and takeas a wholea 'genuine' DirO (since the
suffix -r is an unmistakable marker of DirO with verbs meaning (kill), (see), (build), etc.).
The presumed SSyntRel r [direct-objectival?] in such expressions would be limitedly repeat-
ablejust twice, while no obvious naturally-looking conditions can be formulated; at the same
time, this phenomenon can by no means be treated as an exception. Therefore, there are two diffe-
rent SSyntRels:
RMIN dir-obj(ectival)KARDquasi-dir-objBEDR.
The nominal element in verbal collocations of the above type is considered to be a Quasi-
Direct Object. Here is another similar example (Lazard 1994: 93):
 41

Samvar ate kardand

samovar fire do.PAST.3PL,
lit.
([They] samovar fire did) = (They lit a samovar),
with the SSyntS SAMVARdir-objKARDquasi-dir-objATE.
A very similar situation exists in Korean (O'Grady 1991: 236):
John+i enehak +ul kongpwu+lul hay+ss +ta
NOM linguistics ACC study ACC do PAST DECL(arative)
lit.
(John linguistics study made) = (John studied linguistics)
[(make [the] study) = ([to] study)].
The SSyntS is here as follows: ENEHAKdir-objHAYquasi-dir-objKONGPWU.

For Criteria C to be satisfied, all three criteria C1 - C3 must be satisfied.

The SSyntRels of a language form a systematic inventory, just like phonemes or inflectional
grammemes; Criteria C1-C3 are part of a methodology for establishing SSyntRels' inventories.
Note Criteria C1 and C2 are paradigmatic, while Criterion C3 is syntagmatic.23

Now we are ready for a definition of Synt-D.

Definition 3.5: Syntactic dependency

The wordform w 2 is said to syntactically depend on the wordform w 1 via SSyntRel r in a
sentence if the three groups of Criteria A, B and C are satisfied for this pair of wordforms and r.
I write w 1rw 2.

4.4. The Logical and Linguistic Properties of Syntactic Dependency

a) Synt-D is anti-symmetrical: w syntw entails (w syntw ), i.e. *w syntw .
1 2 1 2 1 2
This means that a wordform w1 cannot be the Synt-governor of another wordform w2 and simul-
taneously have w2 as its own Synt-governor. This follows from our decision to use the dependen-
cy tree as the formalism for the representation of Synt-structures. Moreover, since w1syntw2
signals that one of the two wordforms, e.g. w2, is linearly positioned with respect to the other, i.e.
w1, it is paradoxical to claim that at the same time w1syntw2, so that w1 is linearly positioned
with respect to w2.
synt
b) Synt-D is anti-reflexive: * . This means that a wordform cannot be linearly positioned
w
with respect to itself. As with Sem-D, anti-reflexivity of Synt-D follows from its anti-symmetry.
c) Synt-D is anti-transitive:
w syntw and w syntw (in one sentence) entails (w syntw ).
1 2 2 3 1 3
 42

Otherwise, the principle of the unique governorsee below, item e)would be violated. This
does not preclude, however, the presence of an INDIRECT Synt-D between w1 and w3: w3 is part
of the Synt-subtree hanging from w1.
d) Synt-Ds must be distinctively labeled: to properly represent Mary loves John, in the phrases
Maryr love and Johnr love the SSyntRels r and r must be different; otherwise the
1 2 1 2
semantic contrast will not be preserved in the SSyntS. (The SSyntS MaryrlovesrJohn does
not show who loves whom.)
e) Synt-D presupposes the uniqueness of the governor: a wordform can syntactically depend
only on ONE other wordform (or be independent, as is the top node of a Synt-tree).
f) Synt-D is universal in the following three respects: it is present in all languages; it appears in
all sentences of a language; and it embraces all wordforms of a sentence (that is, for a sentence,
Synt-Ds always form a connected structurelike Sem-Ds, but unlike Morph-Ds).
The logical properties of Synt-D as defined above correspond to the fact that Synt-D s be-
tween the wordforms of a sentence form a dependency tree: a connected graph in which 1)
each node can depend only on one other node (= the uniqueness of the Synt-governor) and 2) one
and only one node does not depend on anythingthe top node, or the root of the SSyntS (=
the presence of the absolute head). The linear order of the nodes in the SSyntS is of course not
defined; in this way, the D -description of the SSyntS consistently separates the SSynt-links
between wordforms and the linear order of the latter. (Word order is computed by syntactic rules of
the lan guage on the basis of Synt-Ds.)
Examples of Deep-Synt-D s and Surface-Synt-D s, i.e. DSyntRels and SSyntRels, are given
in the structures (3) and (4). For a detailed description of the SSyntRels of English, see Mel'uk/
Pertsov 1987: 85-156 (and 4.8 below), as well as Apresjan et al. 1992: 71-121; the inventories of
SSyntRels for Russian are found in Mel'uk 1974: 221-235, and Apresjan et al. 1989, 1992: 204-
208; for the inventories of SSyntRels ( 'dependent types') for German, Danish, Polish, Bangla,
Finnish, Hungarian, Japanese, and Esperanto, see Maxwell/Schubert 1989; a list of Synt-Ds, illus-
trated in English, is attached to Petkevi 1995. A sketch of syntactic word order rules based on
Synt-D s for Russian, see Mel'uk 1967 and 1974: 260-302; see also Sgall et al. 1995 (for Czech
and German).

4.5. Some Non-Definitorial Properties of Synt-Governors and Synt-Dependents

Synt-governors and Synt-dependents possess three important properties, which, however,
cannot be taken as definitorial: some Synt-governors and Synt-dependents in particular languages
do not have them. Nevertheless, these properties are sufficiently characteristic of Synt-governors
and Synt-dependents so that they can be resorted to as convenient HEURISTIC means. These proper-
ties are omissibility, cooccurrence control, and incorporability.
 43

Omissibility
This is the most important non-definitorial property that distinguishes Synt-governors and
Synt-dependents. Typically, in the configuration w syntw , the Synt-dependent w can be
1 2 2
omitted without affecting the Synt-correctness of the SSyntS (and without producing an ellipsis),
while the Synt-governor w1 cannot. Such is the case in the constructions
ADJN, NNgen, VPREP + N, X Conjcoord+ Y
and a few others. (Let it be emphasized that we speak here of omissibility in the Synt-structure, not
in the actual sentence.) But this is not always the case:
The Synt-dependent may be obligatory (= non-omissible): either in some contexts (e.g., the
DET in a DETN construction), or alwaysas in exocentric constructions (e.g., the N in a
PREPN construction). Cf., for instance, non-omissible adjectives in phrases like a man of
various talents.
The Synt-governor can be omissible: for example, 1) the Russian preposition OKOLO (about)
with a numeral phrase (okolo trx tonn (about three tons) is syntactically equivalent to tri tonny) or
the English prepositional configuration from - to, again with a numeral phrase (from three to six
girls is syntactically equivalent to six girls); 2) the English subordinate conjunction THAT (John
knows that Mary is in town is syntactically equivalent to John knows Mary is in town).

Cooccurrence (= Subcategorization) Control

Typically, in the configuration w syntw , it is the Synt-governor w that is subcatego-
1 2 1
rized for by the Synt-governor w of the whole phrase. To put it differently, the lexicographic des-
cription of w must take into account some properties of w1, but not of w2. Thus, if a verb admits
a noun as its actant, the lexicographic properties of the noun may be relevant (this verb admits only
human nouns, or only mass nouns, etc.); but it is not the case that a verb admits as its actant a noun
24
with a particular determinersay, only with EVERY, or only with A/AN, etc. This fact points to
N as the Synt-governor in the constructions DETN or ADJN. Similarly, in the construction
CONJsubordVsubord (...whether [he] comes, ...that [I] am [here]), it is CONJsubord that deter-
mines the subcategorization of the MV in the matrix clause: some verbs take WHETHER, some
others take THAT, etc.; but the verb of the subordinate clause is immaterial in this respect. Conse-
quently, we have V CONJ [V ].
matrix subord subord
More generally, the Synt-governor w1 tends to subcategorize for its Synt-dependent w2 (i.e.
w 1 tends to determine the choice of w 2): we say manybooks, but muchnoise, etc.; or else
dependon, but borrowfrom, etc.
 44

Incorporability
Typically, a language with incorporation manifests two phenomena concerning the orientation
of Synt-D in a configuration w syntw :
1 2
Internal incorporability. If w2 can be incorporated into w1, and not the other way around,
then w 1 is the Synt-governor of w 2 ; if w 2 has its own dependents, they can be incorporated
together with it into w1 or remain stranded in the sentence (as a function of the language and the
context). Well-known examples include the incorporation of actants into the verb and of modifying
adjectives into the nouns. Cf., for instance (the incorporated stem is boldfaced):
(24) a. Chukchee (Chukchee-Kamchatkan family, Russia)
n +tur +qine+tekupre+te (with [a] new net)
ADJ new 3SG INSTR net INSTR
vs.
tur+kupre+te (with [a] new-net)
new net INSTR
[Non-incorporated adjectives in Chukchee have a special prefix nbb -, marking them as ad-
jectives, and a person/number suffix.]
External incorporability. If w 1 (or both w 1 and w 2, but not w 2 alone) can be incorporated
into the Synt-governor w of the whole phrase, then w1 is the Synt-governor of w2; we can thus
have [w+w 1]-syntw 2, but not *[w+w 2] syntw 1. Again, if w 2 has its own dependents, they
can be incorporated with it into w or remain stranded; but it seems impossible to have a Synt-
dependent of wi incorporated, while wi itself is not (Payne 1993):
b. Southern Tiwa (Kiowa-Tanoan family, New Mexico, USA)
Wisi seuan+inbi +mu+ +ban (I saw two men).
two man PL 1SG see PAST
vs.
Wisi bi +seuan+mu=+ban, lit. (I two man-saw).
two 1SG man see PAST
vs.
Seuan+in *bi +wisi+m u= +ban (I men two-saw).
man PL 1SG two see PAST
Thus, in the phrase wisi seuanin (two men) we have wisiseuanin, because seuanin can be
incorporated alone into the verb, while wisi alone cannot.
c. Chukchee
Nireqqlikkin amNrootken parollili +t (forty-eight gloves)
forty eight extra glove PL.NOM
vs.
Nireqqlikkin amNrootken parol+lele +t (to forty-eight gloves)
forty eight extra glove PL.DAT
 45

[Incorporation of numerals into the quantified N is obligatory in Chukchee if this N is in an

oblique case.]
For a compound numeral, only the marker of compound numerals parol, meaning (extra,
added), is incorporated, other components of the numeral remaining 'outside.' (We see that the
noun stem lili is modified to lele in the incorporated form: this is the effect of vowel harmony under
the influence of the incorporated element.) Therefore, the marker parol is the Synt-head of the
whole numeral, so that the SSynt-dependencies in a Chukchee compound numeral are as follows:
NireqqlikkinamNrootkenparol
forty eight extra = (48)

As we see, from a logical standpoint, incorporability could be a definitorial property of Synt-

governors, if it weren't for the restricted character of incorporation itself: it is far from being
universal, since it is not found in a majority of languages; therefore, it cannot be used as a general
criterion for the orientation of Synt-D.

Other non-definitorial properties of Synt-governors (listed as early as Pittman 1948) include

CLASS SIZE (a Synt-governor belongs, as a rule, to a larger word-class than its dependent), VER-
SATILITY (a Synt-governor appears in a greater variety of syntactic environments), FREQUENCY (a
particular Synt-governor is less frequent than a particular dependent), as well as some others.
However, all of them are violated by many types of Synt-governors, so that they can be used as
heuristic coonsiderations only.

4.6. The Absolute Head of the Synt-Structure of a Sentence

Since Synt-D presupposes uniqueness of the governor (= no wordform in the sentence can
depend syntactically on more than one other wordform), the SyntS of a sentence must have one
absolute head, or a top nodea wordform which does not syntactically depend on anything and
on which all the other wordforms of the sentence depend (directly or indirectly). Practically speak-
ing, in most versions of the D -approach known to me, in a complete clause/a complete sentence
this role is filled by the finite, or tensed, verbthe Main Verb (at least in languages that
obligatorily have one in each complete clause/sentence, cf. below).25 Thus, in the DSyntS (3),
where any form of the MV, even an analytical one, is represented by a single node, the top node of
the sentence is the verb ESCAPE (in the finite form of the Present Perfect); in the SSyntS (4),
where each wordform, including the auxiliaries, is represented by a separate node, the top node is
the auxiliary verb HAVE (in the finite form of the Present Indefinite). The choice of the MV as the
Synt-head of the sentence is by no means arbitrary: the finite verb is, on Criteria B1-B3, the Synt-
governor vis--vis all its partners in the sentence, and in this way it ends up as the absolute head.
Let us consider the application of Criteria B1-B3 to the MV of a sentence.
 46

By Criterion B1, the finite verb is the governor of the subject, since the passive Synt-valency
of the phrase SubjectsyntMV is determined by the verb: for a phrase to be insertable in the con-
struction I know that... (or any similar context), it has to contain a finite verb; with respect to the
phrases ObjectsyntMV or CircumstantialsyntMV the syntactically dominant status of the verb
is obvious (and has never been doubted). To this, two arguments can be added:
In many languages, subjectless sentences exist (Chinese, Japanese, Lezgian): for instance, in
the Lezgian sentence Meqizva, lit. (Cold-is) = (It is cold) no Synt-Subject is possible, even a zero
onethe Lezgian verb knows no agreement, so that nothing would justify positing a zero dummy
subject. Even in languages where the subject is not omissible, such as English or French, the
imperative sentence uses a finite verb, but has no surface subject; in PRO-drop languages (Spanish,
Polish, ...), sentences without an overt subject are quite typical (Sp. Est muy ocupado (He is very
busy) is a current example). Sentences without objects and circumstantials are even more wide-
spread. However, languages that admit full sentences without the MV, or more precisely, without a
Synt-predicate, are not known (at least, to me). Thus, the presence of the MV (more generally, of a
Synt-predicate) is the necessary and sufficient condition for the existence of a 'genuine' sentence.
The Sem-valency and the active Synt-valency of the MV determine the syntactic organization
of the sentence/the clause. Thus, if the MV is SLEEP, only one Sem-Actant is possible and, conse-
quently, the clause allows only the Synt-Subject; with SEE, two Sem-Actants and, consequently, a
Synt-Subject and a DirO are necessary; KISS involves three Sem-Actants ((who kisses what part
of whom)), but there can be two Synt-Actants (the Synt-Subject and the DirO: either with the
Possessor depending syntactically on the DirO or the bodypart being not mentioned) or three Synt-
actants (the Synt-Subject, the DirO and an Oblique Object): Alan kissed Helen's hand/Helen vs.
Alan kissed Helen on the forehead.
Strictly speaking, we do not need to try Criteria B2 and B3, since Criterion B1 establishes the
MV as the top node of a sentence/a clause beyond any doubt; however, I will do this here in order
to show that in this case they all agree. By Criterion B2, it is the finite verb that is the morpho-
logical contact point in a subordinate clause (minus the complementizer); for instance:
In French, after the conjunction QUOIQUE (although), the MV of the subordinate clause has
to be in the subjunctive: quoiqu'il soit *est malade, lit. (although he should-be ill)).
In French and English, after the conjunction SI/IF the MV of the subordinate clause has to be
in the present, even if it refers to the future: S'il vient *viendra demain .../If he comes *will
come tomorrow...
If a clause is nominalized in order to be used in the Synt-Structure as a noun, it is its MV that
actually undergoes the nominalization: After John arrived, ... After John's arrival, ...
 47

And, finally, by Criterion B3 the whole sentence is semantically reducible to its MV rather
than to its Synt-Subject; thus, John works at IBM is more an instance of work that an instance of
John or of IBM.
However, two complications arise in connection with the Main-Verb-as-the-Synt-Head-of-
the-Sentence principle: zero verb forms and verbless sentences.

Zero verb forms

What is the top node of the SyntS of the Russian sentence (25a), which does not contain any
overt verb at all?
(25) a. Leo moj drug, lit. (Leo my friend) = (Leo is my friend).
Our first clue is that as soon as this sentence is transposed into the past, the future, the subjunctive
or the imperative, a wordform of the verb BYT ([to] be) obligatorily appears:
b. Leo byl moim drugom (Leo was my friend).
Leo budet moim drugom (Leo will-be my friend).
Leo byl by moim drugom (Leo would be my friend).
Leo, bud moim drugom! (Leo, be my friend!)
Since (25a) stands in an obvious paradigmatic relation to (25b), we conclude that the meaning
(present indicative) is expressed in (25a) by a zero wordform or, to put it differently, that the verb
BYT has a zero wordform in the present indicative. The SSyntS of (25a) looks then as follows:
c. BYTpres

subjectival copular

DRUGsg
LEO modificative

MOJ
BYTpres is expressed by a zero signifier on the SMorph-level only; thus it does not create a pro-
blem for the D-Synt-Structure of a sentence.
See Mel'uk 1988: 303ff or 1995a: 169ff on zero verb forms in syntax.

Verbless sentences
In quite a few languages, a full sentence does not have to include a finite verb. Thus, in
Turkic languages, an equative or locative sentence in the present of the indicative ((John is a doctor/
John is Canadian/John is in the room)) does not admit a finite verb ([to] be); instead, the predicative
noun or adjective is supplied with a predicative suffix, which thus marks its Synt-role. In Salishan
languages (West Coast, Canada), in particular, in Lushootseed, all types of full sentences are pos-
sible without a finite verb andunlike Turkicwithout any morphological marker of predicativity.
 48

(26) a. Turkish26
Ingiliz+im ocuk+lar+ Ankara+da +dir+lar
English 1 S G kid PL NOM Ankara LOC 3 PL
(I [am] English). (Kids [are] in Ankara).

Ev + +de +sin Asker+siniz

house SG LOC 2 S G soldier 2PL
(YouSG [are] in the house). (YouPL [are] soldiers).

b. Lushootseed (Beck 1997: 98ff; the syntactic predicate is underlined)

i. s/uladxw ti/i lit. (Salmon that) = (That [is/was] a salmon).
salmon that
ii. sali/ ti// sqwigwac lit. (Two this deer) = (These deer [are/were] two).
two this deer
iii. dxw/al t hud t s +xwitil / t biac
into the burning the NOMINALIZER fall of the meat
lit. (Into the fire [is/was] the fall(ing) of the meat) =
(The meat falls/fell into the fire).
This situation is typical for other Salishan languages as well: any lexeme, whatever its part of
speech, can be turned into the syntactic predicate, provided it is rhematic (in Salishan languages,
the SyntS of sentences very closely parallels their communicative structure). In such sentences, the
top node of the SSyntS can really be anythingfor instance, here is the SSyntS of (26b-iii); it has
as its top node the preposition DXw/AL meaning (into):
DXw/AL (into)

subjectival prepositional
w
SX ITIL (falling) HUD (burning)
determinative
agentive
determinative
/ (of)
T (the)
prepositional T (the)

BIAC (meat)
determinative

T (the)
To take into account languages with verbless sentences, we have to generalize our Main-
Verb-as-the-Head-of-the-Sentence principle. This is readily done:
The top node of the SyntS of a sentence is its main, or primary, Synt-Predicate, whatever its
surface realization.
 49

In the languages of what Whorf called the 'Standard Average European' type the Synt-predi-
cate of a full-fledged clause is (almost) invariably a finite verb. However, even these languages
have 'incomplete' clauses of different types, in which the top node of the SyntS is not a finite verb,
but a noun, an infinitive, an adverb:
Best wishes to you and your family; Down with Saddam Hussein!; Yours sincerely;
Rus. Mne e domoj idti, lit. (To-me still home to-go) = (I still have to go home);
Fr. Et elle de rire, lit. (And she to laugh) = (She broke out laughing), Heureusement qu'elle
est venue ! , lit. (Luckily that she has come!) = (Thank goodness she has come!);
Germ. "Ich kann wieder Farben unterscheiden", so Charlotte Falk, lit. (I can distinguish colors
again," so Ch. F.) = (... says/said Ch. F.); etc.
Language-specific rules define the admissible top node for each of these 'minor' sentence
types.

4.7. The Three Major Classes of Syntactic Dependencies

There are three MAJOR classes of Synt-Ds, recognized universally: complementation, modifi-
cation, and coordination. (Complementation and modification are particular cases of subordina-
tion.) SPECIFIC types of Synt-Ds, i.e. Deep and Surface SyntRels, introduced above, are distribut-
ed between these major classes.
Complementation, modification and coordination have been discussed for a long time (cf.,
e.g., Matthews 1981: 147-167, Lehmann 1985, Zwicky 1993), so here I simply formulate the cor-
responding definitions 3.6-3.8. Note that on the SSynt-level there are several Synt-Ds to which the
distinction of these three classes of Synt-D does not apply in a clear-cut way; we have to allow for
some SSynt-Ds that belong to neither of these classes (isreading or fromto [as in from two to
five pounds...]; in such a case, I will speak of ancillary SSyntRels). On the DSynt-level,
however, the distinction between complementation, modification and coordination creates no pro-
blems.
In all of the following definitions, the wordform w2 depends syntactically on the wordform
w 1 in the given sentence: w 1syntw 2.
Each one of Definitions 3.6-3.8 is approximate: it covers only the PROTOTYPICAL cases. In
order to take into account all possible cases, I would have to add more conditions and thus make
the definitions more complex; but in the present context, it seems not worthwhile.
 50

Definition 3.6: Complementation

The wordform w is a complement, or a Synt-Actant, of the wordform w if w is also
2 1 2
a Sem-dependent of w1:
synt
w1 w2
sem
Complementation is always controlled by the active valency of the Synt-governor;
formally, this means that the complements of the wordform w must be specified in the lexical entry
of L(w), more preciselyin its Government Pattern. In other words, a complement of w corres-
ponds necessarily to a DSyntA(w), the inverse being untrue: a DSyntA(w) can be expressed, as we
will see, by a modifier of w.
Examples (Synt-actants, or complements, are in boldface): Alan loves Helen; during [the]
meeting; worth [a] trip; This must be [seen]; but [not] Helen.

Comment
Definition 3.6 does not cover, for instance, the construction where a Synt-actant w2 of the
wordform w1 depends semantically on a different wordform w3 which also stands in a comple-
mentation relation to w1: e.g., He believed [= w1] John [= w2] to be sick [= w3]. Here, John is a
DSynt-actant of believe, without being its Sem-actant, that is, without depending on believe seman-
tically. (On the discrepancy between Sem- and DSynt-actants of the same lexeme, as well as on
cases where a Sem-actant of w is implemented as a modifier of w, see, e.g., Boguslavskij 1985:
10-19 and 1996: 23-43.)

Definition 3.7: Modification

The wordform w is a modifier, or a Synt-attribute, of the wordform w if w is a
2 1 2
Sem-governor of w1:
synt
w1 w2
sem
Modification is typically not controlled by the active valency of the Synt-governor; this means
that the modifiers of the wordform w are, as a rule, not specified in the lexical entry of L(w).
Examples (modifiers are in boldface): good friend, love passionately; only him; not serious;
wrote in [Stuttgart]; wrote when [he was in Stuttgart].

Comments
1. Definition 3.7 does not cover, for instance, the construction where a SSynt-modifier w2
of the wordform w1 depends semantically on it, since w2 expresses one of w1's Sem- and DSynt-
actants: e.g., French [= w2] participation [= w1]. Here, French depends both syntactically and
 51

semantically on participation, but in spite of this it is a SSynt-modifier of participation rather than

its Synt-actant (it is an adjective, and adjectives are modifiers by definitionbecause of their
specific SSynt-behavior). Another similar example is a book [= w 1] hard [= w 2] to find: the
adjective hard depends syntactically on book, but semantically bears on find; however, it is
considered as a modifier of book.
The opposition 'complementation ~ modification' underlies, in an obvious way, the problem
of distinguishing between actants ( complements, Germ. Ergnzungen, Rus. dopolnenija27)
and circumstantials ( modifiers, Germ. Angaben, Rus. obstojatelstva). This distinction,
first introduced explicitly probably in Tesnire 1959, is discussed in Engel 1977: 98-103, 158-179,
and in Helbig 1992: 75-98 (with rich bibliography).
2. Interestingly, as indicated in Bazell 1949: 7-9, some languages strongly prefer comple-
mentation, while some others stress modification. Thus, Turkic languages or Japanese formally
mark the complements, using morphological government: in particular, they possess developed
case systems. On the other hand, Bantu languages formally mark only modification: they actively
use agreement and completely lack cases; even in complementation constructions, they make the
Synt-governor (say, the verb) agree with its Objects, leaving the latter unmarked. Of course many
languages mix the two techniques in different proportions; thus, Classic and most Slavic languages
richly mark both complementation ( government) and modification ( agreement).

Definition 3.8: Coordination

The wordform w2 is a conjunct of the wordform w1 if, and only if, semantically neither of
them depends on the other (w 1 and w 2 are not directly related semantically), but w 1 and w 2
both are or at least can be Sem-dependents of a semanteme (and) (or of (or), or else of any of
their semantic 'parents,' like (but), etc.) while syntactically w2 depends on w1:
w 1 synt w 2
( sem sem )
(and)

Comments
1. The coordination of w1 and w2 can be of two types:
Either DIRECT coordination, where w1 and w2 have a direct Synt-D between them:
w 1 coord w 2 ;
this coordination is called asyndetic ((conjunctionless)).
Examples: Alan, Leo, Helen; eat, drink, sing, dance; [something] red, [not] white.
Or INDIRECT coordination, where w 1 and w 2 are syntactically linked via a conjunction
CONJcoord:
 52

w 1 coordCONJ coord conjuctw 2.

Examples: Alan and Helen; either Alan or Leo; eat and drink, but not sing and dance; red, but
[not] white.
2. As two preceding definitions, Definition 3.8 does not cover all cases of coordination.
Thus, on the SSynt-level, formally coordinate structures can be used to express DSynt-subordina-
tion. For instance, in Russian we have izlovilsja i ukusil, lit. ([he] managed and bit) = ([he] ma-
naged to bite), where the conjunct ukusil implements the DSynt-A II of izlovilsja (example from
Boguslavskij 1996: 28-32); a similar English example: Try and catch the train.28

The above distinctions between the three major classes of Synt-Ds are reflected in the DSynt-
component of the Meaning-Text model by the three-pronged division of the DSyntRels: actan-
tial (I, II, ..., VI) = complementation vs. attributive (ATTR) = modification vs. coor-
dinative (COORD) = coordination; see Mel'uk 1988: 63-65.
The inclusion relations between the three major classes of Synt-D can be shown in the
following diagram:
SYNTACTIC DEPENDENCY

SUBORDINATION COORDINATION

COMPLEMENTATION MODIFICATION
These classes of Synt-D were clearly distinguished by L. Bloomfield (1933: 194-198; I
change here Bloomfiled's formulations, without modifying has main idea):
In a complementation phrase w1 + w2 the passive Synt-valency of the whole phrase is
different from those of both of its elements, although it is determined by the passive Synt-valency
of one of them, namely that of the Synt-head of the phrase.
In a modification phrase w1 + w2 the passive Synt-valency of the whole phrase is the
same as that of one of its elements, namely that of its Synt-head.
In a coordination phrase w 1 + w 2 +...+ w n the passive Synt-valency of the whole
phrase is the same as that of each of its elements (= conjuncts; we abstract of course from conjunc-
tions, if there are any).
The constructions manifesting the SyntRels of the first class, i.e. complementation, are called
exocentric; the constructions manifesting the SyntRels of the second and third classes, i.e.
modification and coordination, are called endocentric.
 53

For the SURFACE SyntS, a fourth major class of SSyntRel is neededto link 'syntactically-
induced' wordforms (so-called structural words, chunks of idioms, parts of compound numerals,
etc.), which do not appear in the Deep-SyntS and cannot be covered by the dependencies of the
three above-mentioned classes. As proposed at the beginning of Subsection 4.7, I will call these
SSyntRels ancillary, to emphasize their 'subservient' character.

4.8. Syntactic Dependencies of a Language: Surface-Syntactic Relations of English

In order to give the reader a better idea about SSyntRels, as they can be used in a description
of a language, I cite here a tentative list of SSyntRels of English, takenwith some corrections and
additionsfrom Mel'uk/Pertsov 1987: 85-160. No claims are laid as to completeness of this list;
its purpose is strictly illustrative.
In the examples, the SSynt-dependent is boldfaced and words not participating in the con-
struction illustrated are included in brackets.
For a better surveyability of the list, the SSyntRels are grouped as follows:
First, they are divided into subordinate and coordinate ones.
Second, the subordinate SSyntRels are subdivided into two subsets:
clausal SSyntRels, or those that can hold between (the heads of) noun, verb, adjective, and
adverb phrases (they can also appear within these phrases); and
phrasal SSyntRels, or those that can hold only between the elements within phrases (never
between phrases).
Third, inside each subdivision, the line is drawn between valency-controlled SSyntRels (that
necessarily embody complementation) and non-valency-controlled SSyntRels (that can be either
modificative or ancillary).

I. Subordinate SSyntRels: 1 - 51

CLAUSE-LEVEL (= CLAUSAL) SSYNTRELS: 1 - 22

Valency-controlled SSyntRels: Complementation (1 - 15)
Actantial SSyntRels
1. Subjectival:
I subjam...; Smoking subjis [dangerous].
That [Alan comes]subjis [clear].
Itsubjis [clear that Alan comes].
2. Quasi-Subjectival:
[It] is [clear]quasi-subjthat [Alan comes].
3. Direct-Objectival:
seesdir-objme; [to have] writtendir-objnovels;
 54

[Helen] wantsdir-objAlan [to read];

worthdir-obj[a] trip; preferdir-obj[her] staying [home];
explain [to him]dir-objthat [Alan was absent];
makedir-objit [possible to neutralize the consequences]
4. Quasi-Direct-Objectival:
make [it possible]quasi-dir-objto [neutralize the consequences]
5. Indirect-Objectival:
givesindir-objAlan /him [some money];
convince [Alan]indir-objthat [he should work less]
6. Oblique-Objectival-1:
dependsobl-obj1on [Alan]; respectobl-obj1for [Alan];
translationobl-obj1from [Lushootseed into Polish];
7. Oblique-Objectival-2:
translation [from Lushootseed]obl-obj2into [Polish]
8. Infinitival-Objectival:
can inf-objread; want inf-objto [read];
[Helen] wants [Alan]inf-objto [read]; [Helen] makes [Alan]inf-objread;
[her] desireinf-objto [come]
9. Completive:
find [this]compleasy; consider [Alan]complhappy;
make [it] complpossible; make [Helen] compl[a good] wife
10. Copular:
be copuleasy; be copul[a] teacher;
becopulwithout [a hat]; seemcopulin [a difficult position]
11. Agentive:
writtenagentby [Alan]; arrivalagentof [Alan];
shootingagentof [the hunters: (the hunters shoot)];
[a] translationagentby [Alan];
[I like] for [Alan to]agentplay [cards].
12. Patientive:
translationpatientof [this text];
shootingpatientof [the hunters: (the hunters are shot)]
Copredicative SSyntRels
13. Subject-copredicative: [Alan] returnedsubj-copredrich.
14. Object-copredicative:
[Alan] likes [Helen]obj-copredslim.
 55

[Alan] hammered [the coin]obj-copredflat.

Comparative SSyntRel
15. Comparative:
oldercomparthan [Leo]; [Alan loves Helen] morecomparthan [Leo];
more [important]comparthan [Leo]; as [important]comparas [Leo];
Non-Valency-controlled SSyntRels: Modification (16 - 22)
Absolutive SSyntRel
16. Absolute-predicative:
[His first] attemptabs-pred[a] failure, [Alan ...].
[He went out, his] angerabs-predgone.
[He went out, his] gunabs-predin [his left hand].
Adverbial SSyntRels
17. Adverbial:
walkadverbfast; [will] writeadverb[next] week;
delveadverbdeeply; [He] worksadverbthere in [this office].
[He ]went [out,]adverb[his] gun [in his left hand].
With [her paper finished, Helen]adverbcan afford this trip.
18. Modificative-adverbial:
[As always] elegant, mod-adverb[Alan] walked [away].
19. Appositive-adverbial:
[An old] man, appos-adverb[Alan] works [less].
20. Attributive-adverbial:
Abroad, attr-adverb[Alan] works [less].
Sentential SSyntRels
21. Parenthetical:
Oddly, parenth[Alan] works [less].
As [we have known for some time,] parenth[Alan] works [less].
To [give an example,]parenth[I] consider [now nominal suffixes].
22. Adjunctive:
OK, adjunct[I] agree.

PHRASE-LEVEL (= PHRASAL) SSYNTRELS: 23 - 51

General Phrase SSyntRels
Non-valency-controlled SSyntRels: Modification
23. Restrictive:
still restrtaller, most restrfrequent;
 56

notrestrhere; [Alan has] just restrarrived.

Noun Phrase SSyntRels

Valency-controlled SSyntRels: Complementation
24. Elective:
[the] poorestelectamong [peasants]; [the] best [ones]electfrom [these boys]
Mixed Type SSyntRels = Valency-controlled / Non-Valency-controlled: Modification
25. Possessive:
Alan's possarrival; Alan's possbed
26. Compositive:
man[-machine]composinteraction; carcomposrepair;
noun composphrase; color composblind
Non-Valency-controlled SSyntRels: Modification
27. Determinative:
my determbed, a determbed, those determbeds
28. Quantitative:
three quantbeds; [three num-junct] thousand quantpeople
29. Modificative:
comfortable modif beds, visible modif stars, French modif production
30. Post-modificative:
stars post-modifvisible (vs. visible stars)
31. Descriptive-Modificative:
[these] beds, descr-modifcomfortable [and not expensive], ...
32. Relative:
[the] paperrelat[that I] read [yesterday]; [the] paperrelat[I] read [yesterday];
the girlrelat[who] came [first]
33. Descriptive-Relative:
[this] paper,descr-relat[which I] read [yesterday];
Alan,descr-relat[who] loves [her so much]
34. Appositive:
General apposWanner; [the] term appos'suffix'
35. Descriptive-Appositive:
[This] termdescr-appos ('suffix') [will be considered later].
[You forget about] me,descr-appos[your] mother.
Alan appos[the] Powerful;
youdescr-apposchildren, wedescr-apposlinguists
 57

36. Sequential:
mansequentmachine [interaction]; fiftysequentto [seventy dollars]
37. Attributive:
learnersattrwith [different backgrounds]; dressattrof [a beautiful color];
[a] manattr[the same] age; yearsattrof [war], [the] bedattrof [Alain]
38. Descriptive-Attributive:
[Professor] Wanner,descr-attrfrom [Stuttgart, was also present].

Prepositional Phrase SSyntRels

A valency-controlled SSyntRel: Complementation
39. Prepositional:
inpreposbed, withoutprepos[three hundred] dollars
A non-valency-controlled SSyntRel: Complementation (by analogy)
40. Prepositional-infinitival:
to prepos-infinitgo [to bed]

Verb Phrase (= Analytical Form) SSyntRels

Non-valency-controlled SSyntRels: Ancilliary
41. Perfect-analytical:
has perf-analytwritten, has perf-analytbeen [beaten]
42. Progressive-analytical:
was progr-analyt writing
43. Passive-analytical:
was pass - analyt written

Conjunction Phrase SSyntRels29

Valency-controlled SSyntRels: Complementation
44. Subordinate-Conjunctional:
[Suppose] thatsubord-conj[Alan] comes.
[so] as [not]subord-conjto [irritate Leo]
45. Coordinate-Conjunctional:
[Alan] andcoord-conjHelen
46. Comparative-Conjunctional:
than compar-conj Helen; as compar-conj always
47. Absolute-Conjunctional:
Ifabs-conj[a] pronoun, [the grammatical subject may...]
while abs-conjin [bed]
 58

Word-like Phrase SSyntRels

Non-valency-controlled SSyntRels: Ancillary
48. Verb-junctive:
give verb-junctup, bring verb-junctdown
49. Numeral-junctive:
fifty num-junct three; fifty num-junct third
50. Binary-junctive:
if... bin-junct then...; the [more...] bin-junct the [more...];
till bin-junct after; from [...] bin-junct to [...];
either [...] bin-junct or [...]
51. Colligative:
[is] dealtcolligwith [stranded prepositions]

II. Coordinate SSyntRels: 52-53

Non-valency-controlled SSyntRels: Coordination
52. Coordinative:
Alancoordand [Leo]; Alancoordbut [not Leo];
rich,coordintelligentcoordand [beautiful]
53. Quasi-coordinative:
[Alan was] abroad quasi -coord without a penny quasi -coord in a desperate situation.
[These moneys we keep hidden under a loose] board quasi-coordunder the floor quasi-co-
ord under a chamber pot quasi - coord under my friend's bed (T. Capote, "A Christmas
Memory").

Comment
As suggested above (4.7, Comment after Definition 3.6), some of the modification class
SSyntRels can be valency-controlled, so that their dependents correspond to DSynt-Actants of their
governors:
mydetermarrival IIARRIVE;
Americanmodifparticipation AMERICAIPARTICIPATE;
treat [someone]adverbfriendly TREATIIIFRIENDLY;
income compos tax INCOMEIITAX;
etc.
Similarly, the coordinative SSyntRel can be valency-controlled:
trycoordand [come] TRYIICOME.
 59

In point of fact, the correlation between complementation and modification, as well as be-
tween complementation/modification and coordination on the DSynt- and SSynt-levels is complex
and cannot be discussed here in depth.

5. Possible Combinations of the Three Types of Linguistic Dependency

The three types of linguistic syntagmatic dependency that we are studyingsemantic, syn-
tactic, and morphologicalare logically independent of each other, which means that they can co-
occur in all possible combinations. Thus, two wordforms in a sentence can be related by a Sem-D
with no Synt-D or Morph-D between them (a); or they can have a Sem-D and, at the same time, an
inverse Synt-D , with still no Morph-D (b); or there can be a Synt-D with a Morph-D having the
same orientation, but no Sem-D (c); etc.
(a) w 1 sem w 2 (b) w 1 sem w 2 (c) w 1 synt w 2
synt morph
All in all, there are fourteen logically possible combinations of direct Sem-D , Synt-D and
Morph-D between two wordforms, w1 and w2, of a sentence (cf. Mel'uk 1964, 1988: 118-128);
all of them are actually found in languages and will be illustrated below.

CASE 1: w1 w2, i.e. no syntagmatic dependency whatsoever between two wordforms.

Example: The wordforms cocoa and new in (1).

CASE 2: w semw , i.e. two wordforms are linked by a Sem-D , unsupported by any Synt-D
1 2
or Morph-D .
Example: The wordforms farming and problems in (1) are semantically directly relatedfarming
depends on problems ((problems are-for farming)), yet there is no Synt-D or Morph-D between
them. Another example could be an expression of the type He drives me mad, where we have
madsemme, while syntactically and morphologically me and mad are not linked.

CASE 3: w 1 syntw 2, i.e. two wordforms are linked by a Synt-D , but there is no Sem-D or
Morph-D between them.
Examples
(27) a. In Japanese, a numeral or a quantitative adverb, while bearing semantically on the SSynt-
Subject or the DirO as in (Five people were injured) or (He reads many books) (andfor
numeralsmorphologically depending on it), depends syntactically on the verb, with
which it has neither semantic nor morphological links, cf.:
i. Sono ziko +de keganin +ga go+nin deta,
this accident LOC injured.people SUBJ(ective) five CLASS(ifier) emerged
lit. (In this accident, injured-people five-lysyntemerged) =
(In this accident, five people were injured).
 60

ii. Nihongo+no hon +o takusan yomimasita,

Japanese GEN book ACC many read-PAST
lit. (Japanese-language books many-lysyntread) =
([I] have read many Japanese books).
For a detailed characterization of this Quantifier Float construction and its relations to other
numeral constructions in Japanese, see Kim 1995; cf. also Case 8, (31b).

b. In Russian, a numeral adverb of the type VDVOM (being-two), VTROM (being-three),

etc. is used in a similar construction,where this adverb semantically bears on the SSynt-
Subject of the clause, having with it no syntactic or morphological links:
My sideli [na beregu] vetverom,
lit. (We satsyntbeing-four [on the-shore]).
The same holds about most of Floating Quantifiers of different types in various languages.

c. In English, French and many other languages, a measure noun used as a DirO depends
syntactically on the verb, but does not have a semantic or morphological link with it
(semantically the verb dominates the noun quantified):
John bought synt[five] kilos [of potatoes].
Fr. Jean a achetsynt[cinq] kilos [de pommes de terre].
Cf. Case 9, example (32b).

d. Any conjoined elements that are morphologically invariable, as, e.g., Alan, Helen, Leo,
are linked syntactically without any direct semantic or morphological link between them.

CASE 4: w morphw , i.e. two wordforms are linked by a Morph-D only, without any Sem-
1 2
D or Synt-D .
Examples
(28) a. In Tabassaran (Eastern Caucasian), the M(ain) V(erb) can agree with the 1st/2nd person
Possessor of the SSynt-Subject, the Possessor being not directly related to the verb
semantically or syntactically, cf.:
I mudur uvuhna he+b+gnu+ji (Our kid ran away to your place)
Our goat-kid[II] to-you left II left 1 P L
where the verb hegnu (ran away, fled) agrees in class with mudur (class II, the class-
marking infix in the verb is -b-), but in person and number with i (our).
The same type of agreement of the MV is also characteristic, among others, of Chickasaw
(Muscogean, USA), Wichita (Caddoan, USA), Tangut, and Maithili (Indian).
 61

b. In Maasai (Nilotic, Kenya), an infinitive which semantically and syntactically depends on

the MV agrees in number with the Synt-Subject of the MV (rather than with its own
understood [= semantic] subject):
trto OltUNn /IltU;Nn a +mU;k nish
help-PERF-1SG.SUBJ.3.OBJ the-man.SG.NOM/the-man.PL.NOM INF.SG brew beer.SG.NOM
([I] helped the-man/the-men to-brew [SG] the-beer).
vs.
KI;trto OltUNn /IltU;Nn a +mUk nish
help-PERF-1PL.SUBJ.3.OBJ the-man.SG.NOM/the-man.PL.NOM INF.PL brew beer.SG.NOM
([We] helped the-man/the-men to-brew [PL] the-beer).

c. In Alutor, a transitive verb of perception on which syntactically depends a DirO clause

can show object-agreement either with the SSynt-Subject or with the DirO of this clause
(depending on the communicative role of the former and the latter):
i. Qbmavb+nak na +la?u+tkbni +bbt
Qamav SG.LOC 3SG.SUB see PRES 2SG.OBJ

bb n +annb +kblatbtkb+na +wwi qura+wwi

you.SG INSTR 2SG.SUB harness 3.OBJ PL reindeer PL
lit.
(Qamav sees-you you are-harnessing reindeer)
= (Qamav sees YU harnessing the reindeer).

ii. Qbmavb+nak +la?u+tkbni +nina + w w i

Qamav SG.LOC 3SG.SUBJ see PRES 3.OBJ P L

bn +annb +kblatbtkb+na +wwi qura+w w i

you.SG INSTR 2SG.SUB harness 3.OBJ PL reindeer P L
lit. (Qamav sees-them you are-harnessing reindeer)
= (Qamav sees you harnessing the RINDEER).
[The Alutor transitive verb enters in an ergative construction, with the SSynt-Subject in the loca-
tive, if it is a human proper name, and in the instrumental otherwise (cf. (9a), p. 00); na- is a 3sg
subject marker if the DirO is neither in the 3rd person nor 1sg, and - is a 3sg subject marker if the
DirO is in the 3rd person or 1sg. A verb of perception can also agree with its DirO clause as a
whole, showing 3sg object agreement; this case is, however, irrelevant in the present context.]

CASE 5: w sem w i.e. two wordforms are linked by a Sem-D and a Synt-D ,
1 2
synt oriented the same way, but no Morph-D is present.
This is a typical situation with nominal objects in caseless languages: e.g., the wordforms escape
and problems in (1).
 62

CASE 6: w sem i.e. two wordforms are linked by a Sem-D and a Synt-D , this
1 w
synt 2 time oriented the opposite ways, with no Morph-D present.
Examples
(29) a. An adjective and the modified noun in ADJ + N phrase in a language where adjectives are
invariable, cf. new and areas in (1).

b. In Lushootseed, a PREP + NUM phrase syntactically depends on the clause predicate

(underlined in the example), which is its Sem-dependent, and there is no Morph-D be-
tween them:
/ bs +s+q w b+q w bay / lgw/ / ti // b +sali /
POSS dog dog PL by this two
lit. ([They are] dog-dog-possessor + s by these two) =
([They] have two dogs) (example of D. Beck).

CASE 7: w sem i.e. two wordforms are linked by a Sem-D and a Morph-D ,
1 morph
w2
oriented the same way, but no Synt-D is present.
Example: In a language where Clitic Raising exists, a cliticin the SSyntScan semantically and
morphologically depend on an infinitive, while there will be no direct Synt-dependency between
them, the clitic being a Synt-dependent of a higher verb, which 'hosts' it, cf.:
(30) Sp. Le syntquisiera poder enviar este libro,
lit. (To-him [I] would-like to-be-able to-send this book).
Semantically, le (to him) depends as an actant on enviar ([to] send); its dative form is also
imposed by this verb, so that morphologically le also depends on enviar.

CASE 8: w sem i.e. two wordforms are linked by a Sem-D and a Morph-D ,
w2
1 morph oriented the opposite ways, without any direct Synt-D.
Examples
(31) a. An attributive or copredicative adjective and the Subject/the DirO in a language with
adjectival agreement illustrate Case 8. The adjective shows agreement with the Subject/the
DirO, which is its Sem-dependent, and there is no direct Synt-link between them:
Fr. Elle semblait fatigue (She seemed tired),
where semantically elle depends on fatigue [= (fatigu)((elle))], but morphologically
fatigue depends on elle for its singular and feminine; syntactically, the two are not
directly related.
Similar examples: Fr. Elle est rentre heureuse (She returned happy), Il buvait son th
froid/sa tisane froide (He drank his tea cold/his herbal tea cold), etc.
For a detailed analysis of the copredicative construction, see Nichols 1978.
 63

b. A Japanese numeral w can bear semantically on a noun w and morphologically de-

1 2
pend on it (the form of the numeralmore precisely, its classifying suffixis determined
by the noun), while syntactically it depends on the verb (cf. Case 3, example (27a)):
Uma +o ip+piki synttot +ta ([He] took one horse).
horse ACC one CLASS take PAST
Hagaki +o ip+pon synttot +ta ([He] took one postcard).
postcard ACC one CLASS take PAST
K i p p u +o iti+maisynttot +ta ([He] took one ticket).
ticket ACC one CLASS take PAST

CASE 9: w synt i.e. two wordforms are linked by a Synt-D and a Morph-D ,
w2
1 morph oriented the same way, with no Sem-D between them.
Examples
(32) a. In Latin, the construction ab urbe condita, lit. (since [the] city founded) = (since the
founding of the city), the preposition ab (since) syntactically and morphologically domin-
ates the noun urbe, while semantically it bears on (conditio) = (founding).

b. A measure noun used as a DirO in a language with cases depends syntactically and mor-
phologically on the verb, but does not have a semantic link with it:
Rus. Ivan kupil tonnu kirpiej (Ivan bought a ton of bricks).
Cf. Case 3, example (27b).

CASE 10: w synt i.e. two wordforms are linked by a Synt-D and a Morph-D ,
w2
1 morph oriented the same way, again with no Sem-D between them.

Example: A phasic or copular verb and its Synt-Subject. The verb syntactically dominates the sub-
ject, but morphologically depends on it (= agrees with it in person and number), while there is
no Sem-D between this verb and its subject, because the subject semantically depends on the
lexical verb, cf.:
(33) Alan begins to read or Alan seems to read,
where Alan syntactically depends on begin/seem, morphologically dominates it, and seman-
tically depends on read: (read)((Alan)).

CASE 11: sem i.e. two wordforms are linked by all three types of depen-
w1 synt w 2 dency, oriented all the same way.
morph
Example: A verb and its nominal object in a language with cases, cf. Rus. problem with respect to
(ne) znat in (6).
 64

CASE 12: sem i.e. two wordforms are linked by all the three types of depen-
w1 synt w 2 dency, with Sem-D and Morph-D oriented the same way,
morph while Synt-D goes in the opposite direction.

Examples
(34) a. A postnominal modifier and the modified noun in a language having what is known as
izafa construction. Cf. Persian daftar+e nav, lit. (workbook new), where nav [= w 1]
bears semantically on daftar [= w2] and imposes on it a special form (= the izafa suffix
-e), while being syntactically its dependent.

b. A negative particle and the negated verb in a language where the particle requires a spe-
cial form of the verb. Thus, in Arabic the particle lam (NEG.COMPL(etive).PAST) requires
the jussive, while the particle lan (NEG.COMPL.FUT) requires the subjunctive (la (NEG.IN-
COMPL(etive).PRES) is neutral in this respect):
ja+ktub+u ([he] writes) ~ la ja+ktub+u ([he] does not write)
IND(icative)
vs.
kataba ([he] wrote) ~ lam ja+ktub+ ([he] did not write)
JUSS(ive)
vs.
sa+ja+ktubu ([he] will write) ~ lan ja+ktub+a ([he] will not write)
SUBJ(unctive)
Semantically, the negative particle LAM/LAN bears on the verb and morphologically con-
trols its form; but syntactically, it depends on the verb.

CASE 13: sem i.e. two wordforms are linked by all the three types of depen-
w1 synt w 2 dency, with Sem-D and Morph-D oriented the same way,
morph while Synt-D goes in the opposite direction.

Examples
(35) a. A verb and its nominal actant in a language with polypersonal agreement of the verb, but
no nominal cases, such as, e.g., Abkhaz (West Caucasian), where the MV agrees in
nominal class and number with the SSynt-Subject, the DirO and IndirO:
Sara Nada i+l + s+teit awqw
I Nadsha it her I gave book
(I gave Nadsha [a] book).
Here, nouns and pronouns have no case inflection themselves, but impose agreement on
the verb, whose prefixes cross-reference these three SSynt-actants.

b. Agreement of the participle in an analytical verb form with the preposed DirO in French:
 65

les fleur +s que je t' ai offert+es

the flower[FEM] PL that I to-you have given PL.FEM
(the flowers that I have given to you),
where que (that) semantically and syntactically depends on offertes (given), but mor-
phologically controls its gender and number (QUE gets its gender and number from its
antecedent, FLEURS, so that, in final analysis, OFFERTES is feminine and plural be-
cause of FLEURS; yet, technically speaking, it agrees with QUE).
In point of fact, the situation here is more complex, since que is an accusative form, imposed
by the transitive verb offrir ([to] give); so that que morphologically depends on offertes at the same
time that offertes depends on que. This is then a case of reciprocal morphological dependency.

CASE 14: sem i.e. two wordforms are linked by all three types of depen-
w1 synt w 2 dency, of which Synt-D and Morph-D are oriented the same
morph way, but in the opposite direction with respect to Sem-D.

Example: An agreeing adjective and the modified noun in a language with adjectival agreement
(Slavic, Romance, Semitic, German, etc.), where the adjective bears semantically on the noun,
but syntactically and morphologically depends on it.

Consistent distinction of the three types of dependency allows for some elegant formulations,
for instance:
The adjective as a part of speech can be characterized in terms of Sem-D vs. Synt-D (see Beck
1998):
In a prototypical case, an adjective semantically dominates the noun on which it depends syntac-
tically. (Morph-D can go both ways or be absent altogether: cf. Cases 6, 12, and 14.)
Similarly, for the adverb (replacing 'noun' with 'verb or adjective').
Taking into account the three types of linguistic dependency, Zwicky (1993) presents the dif-
ferences between complements and modifiers in a compact form:
Properties Complement Modifier
Semantic argument predicate
Syntactic obligatory optional
unique repeatable
Morphological agreement controller agreement target
government target government controller
The properties stated in this table hold only in the most current, prototypical cases; as has been
shown above, the syntactic and morphological properties of complements vs. modifiers can in prin-
ciple be inverted. However, the semanticdefinitorialproperty is stable.
 66

In the literature, one finds heated debates concerning the split of head-related properties be-
tween different sentence elements, which presumably makes the identification of heads difficult
and/or dubious: a given element seems to be the head in one respect, but the dependent in another
one. However, if one distinguishes the three types of dependency and uses Criteria B1-B3 in the
hierarchical way (p. 00), such a split is logically impossible. Thus, a Synt-head must be determin-
ed exclusively according to the properties of SYNTACTIC heads; it is irrelevant whether or not it has
properties of Sem-heads or Morph-heads (as the 14 combinations above show, in many cases a
Synt-head does not have such properties.)

6. Correlations between the Three Types of Linguistic Dependency

The three types of dependency are linked by the following correlations (these correlations
hold only for PROTOTYPICAL cases of morphological agreement and government and are no more
than tendencies).

Sem- D vs. Morph-D

Sem-governors agree morphologically with their Sem-dependents;
Sem-governors govern morphologically their Sem-dependents.
This is the Keenan's principle (Keenan 1974: 298-303 and 1978: 94-98); cf. Zwicky's
slogan: 'Functors are agreement targets and government triggers' (1991: 2).

Synt- D vs. Morph-D

If w2 morphologically agrees with w1, then w1 and w2 sometimes are, and sometimes are
not, linked by a direct Synt-D (there also may be no Sem-D between w1 and w2).
If w 2 is morphologically governed by w 1, then w 1 and w 2 are always linked by a direct
Synt-D (however, a Sem-D can be absent).
As can be seen in our review of theoretically possible cases, in the configuration
w syntw , the Morph-D s can go both ways: the Synt-governor can be either the controller or
1 2
the target of a Morph-D . The same holds with respect to the linear position: rules for positioning
can also go both ways, such that in some cases the linear position of the Synt-dependent w2 must
be stated with respect to its Synt-governor w 1 (ADJN, NV, ADVV, etc.), and in some
others the linear position of the Synt-governor w1 must be stated with respect to its Synt-dependent
w2 (PREPN, AUXV, CONJV, etc.). N : The reference point for linear positioning of the
one of two syntactically-linked elements XsyntY is the element which can appear without the
other, the inverse being untrue. Thus, in ADJsyntN, N can appear without ADJ, but not vice
versa, which means that ADJ is positioned with respect to N, its SSynt-governor. Similarly, for
PREPsyntN, N can be used without a PREP, but a PREP never appears without its N; there-
 67

fore, the position of PREP is stated with respect to N, its SSynt-dependent (cf. 4.3.1, Criterion
A1, Comment 1, p. 00).
Sem-D and Synt-D are global in that they 'embrace' ALL the wordforms in a sentence; there-
fore, they are represented explicitly in the SemS and the D-/S-SyntS of the sentence. On the other
hand, Morph-D is not global in this sense: it does not 'embrace' all the words in a sentence (in
addition, it is by no means present in all sentences and can be altogether absent from a language);
therefore, no special structure is foreseen in which it would be explicitly represented: Morph-Ds are
computed by syntactic rules of the language during the SSyntS DMorphS transition and encoded
in the DMorphS via corresponding grammemes.

Chapter III: Syntactic Dependency

Among the three types of linguistic dependency that we are studying, it is the Synt-D that
attracts the lion's share of attention; it is, beyond any doubt, the most important type of dependency
and, at the same time, the most controversial one. I will discuss the Synt-D additionally, touching
on the following points:
Some false dogmas on the subject of Synt-D (1).
Analysis of some constructions 'difficult' for the attribution of Synt-D (2).
Advantages of Synt-D (3).

1. Current Fallacies Concerning Syntactic Dependency

One finds in the literature a number of criticisms leveled at the D-approach in syntax; these
can be grouped under four headings: 'double dependency,' 'mutual dependency,' 'no dependen-
cy,' and 'insufficient dependency.' I will consider below examples of each in order to show that
these criticisms are unjustified, since they stem from the confusion of different types of dependency
or from using unlabeled dependencies.

1.1. 'Double Dependency'

There are three typical cases where many see double syntactic dependency: relative pronouns,
raisings and subordination of coordinate expressions.

1.1.1. Relative Pronouns

In the man whom we saw/the car which we saw, many linguistsfor instance, Tesnire
1959: 560 and Hudson 1990: 117say that the relative pronoun syntactically depends both on the
MV of the relative clause (here, saw) and on its own antecedent (here, man/car; this second Synt-D
is shown by a boldfaced branch); at the same time, the relative pronoun being the Synt-head of the
relative clause syntactically governs the MV on which it, at the same time, syntactically depends:
 68

MAN CAR

SAW SAW

WHO WE WHICH WE
Were it so, this would be a problem for the D -approach, since it would mean the violation of the
uniqueness-of-Synt-governor principle, as well as the principle of the absence of cycles in the
Synt-structure. This would, in turn, destroy a clear understanding of the substantive nature of
Synt-D , which is supposed to specify the linear positioning of one of its members with respect to
the otherand nothing else.
However, the representation above is simply a case of confusion between different types and/
or levels of dependency. I think that in the SSyntS the relative pronoun depends SYNTACTICALLY
only on the MV of the relative clause, while standing in an anaphoric relation to its antecedent; and
in many languages the relative pronoun also has a Morph-D with its antecedent (namely, congru-
ence). This masks the fact that the Synt-head of a relative clause is its finite MV, and by no means
the relative pronoun: it is only the presence of a finite verb in a clause that licenses the speaker to
use this clause as a relative, and it is this use that imposes the pronominalization of the relativized
clause element, which thus becomes a marker of relativization. Here is the SSyntS of a relative
clause as proposed in the Meaning-Text theory (the dashed arrow is part of SSynt-Anaphoric Struc-
ture):
MAN CAR

SAW SAW

WHO WE WHICH WE
It is, however, obvious that the relative pronoun has indeed a double syntactic nature: it is
both a Synt-dependent of the MV of the relative and, at the same time, the marker of the relative.
This leads some researchers to split the relative pronoun into two abstract lexical elements, one of
which represents the Synt-head of the relative clause (its MV depends on this element), while the
other occupies its legitimate dependent Synt-position with respect of the MV of the relative. Thus,
Engel (1977: 234-235 [1988: 292-293]), following the proposal of Tesnire (1959: 561), repre-
sents the SSyntS of the German relative clause der Mann, der Birnen verkauft (the man who sells
pears) by splitting the relative pronoun DER (that) [= (which/who)] into the relative marker part D-
and the pronominal anaphoric part -ER (he), obtaining something like the man that he [= der] sells
pears and thus avoiding double dependency:
 69

der Mann

d-

verkaufen

er Birnen
Relative clauses with a separate expression of the relative marker and the pronominal anapho-
ric element (= resumptive pronoun) exist in many languages, for instance, in Arabic, Turkish, Al-
banian, Persian, Middle High German, Provenal, etc. (see, e.g., Suer 1998). But this is exactly
what proves that there is no need for such a tour de force in English, French or German: here, the
syntax of the relative clause is different. The double role of the relative pronoun in these languages
is reflected on different levels of representation in terms of the three types of dependency plus the
separate anaphoric relation. As far as the Synt-D is concerned, the relative pronoun does not de-
pend on its antecedent and does depend on the MV of the relative clause:
On the one hand, it does not Synt-depend on its antecedent because the antecedent of a rela-
tive pronoun and the pronoun itself cannot form a phrase; thus, *[a] man whom and *[a] car which
are by no means phrases of English. See Criterion A2, Ch. II, 4.3, p. 00.
On the other hand, some properties of the relative pronoun clearly point to its dependent
Synt-role within the relative. The most important in this respect is the fact that relativization may be
restricted by the dependent Synt-role of the relative pronoun: for instance, in some languages relati-
vization is possible only if the would-be relative pronoun is the SSynt-Subject, or if it is the SSynt-
Subject or the DirO, or if it is the SSynt-Subject, the DirO or the IndirO, etc. Thus, the type of the
Synt-D of the relative pronoun on the MV of the relative clause is crucial. To this, one could add,
for instance, that the relative pronoun can be omitted in some languages (as in the man I saw or the
man I talk with) without any effect on the relative; omissibility is a typical feature of Synt-depen-
dentsalthough it happens to the Synt-heads as well (Ch. II, 4.5). Also, in some languages, the
relative clause is marked by a special form of the MV of the relative, without any relative pronoun
(Bantu).
But my strongest arguments against the double dependency of a relative pronoun are as
follows:

Deep-Synt-Structure of the Relative Clause. In the DSyntS, the (future) relative clause has no
relative pronoun at allonly its nominal source N is allowed to appear there. And this N syntactic-
ally depends of course only on the MV of its clause, being anaphorically related to its antecedent (as
 70

is the case with all substitute, i.e. anaphoric, pronouns). When in the transition DSynt SSyntS
this N is replaced with the corresponding relative pronoun, what could be the reason to add another
Synt-D between it and its antecedent? I can see no one. This consideration can be formalized by the
following heuristic priciple:

As-Little-DSynt SSynt-Restructuring-as-Possible Principle

When deciding on the SSyntS of a phrase/a clause, the reasercher should maintain for it the
same orientation of syntactic D s as in the DSyntSexcept in cases of obvious necessity to
reverse the dependencies, which have to be explicitly justified. (In other words, the default case
must be that a DSynt-D remains a SSynt-D.)

The SSynt-Structure of the Related Interrogative Clause. The full-fledged sentence Who wants
a lift? has the SSyntS with the top node WANT, and this is for me an important argument in favor
of establishing the same top node in the corresponding relative [the boy] who wants a liftbecause
I adopt the following principle:

Always-the-same-SSyntS Principle
When deciding on the SSyntS of a phrase/a clause, the reasercher should try to maintain for it
always the same SSyntS no matter where this phrase/this clause appears in a larger formation.

Therefore, the same phrase who wants a lift in a sentence of the type Who wants a lift has to
sign upthis time, a headless relativehas the SSyntS with the finite verb as its top
node. Thus, if I have accepted the MV as the head of an independent interrogative clause, I want
this clause to have the same SSyntS even when it is used as a relative. If I have accepted the MV as
the head of a 'normal' relative clause, I prefer to treat the corresponding headless relative in the
same way; and so forth. This means, among other things, that in English, the finite, or 'tensed,'
verb has in its passive valency the role of the head of such phrases (= actually, full-fledged clauses)
that can be used as equivalents of noun phrasesunder specific conditions, of course (such as the
presence of relative pronouns).
The phrase who wants a lift is a partial syntactic equivalent of a noun phraseit can be, e.g.,
a SSynt-Subject. The phrase what Alan bought in a sentence of the type What Alan bought is im-
portantagain, a headless relativealso has a finite verb as its top node: Alanboughtwhat.
It is also a partial syntactic equivalent of a noun phrase, since it can be the SSynt-Subject, or the
DirO of the MV, or else depend on a preposition:

What Alanbought is important.

 71

I like what Alan bought.

for Alanboughtwhat = For what Alan bought [I could pay him $ 15].
The phrase whatever apples Alan bought (Van Langendonck 1994: 256), which is as well syntac-
tically equivalent to a noun phrase, has a similar SSyntSin the sense that its top node is the finite
verb bought and the WH-pronoun depends on it (in this case, indirectly):
whateverapplesboughtAlan.
(The SSyntS of whatever apples that Alan bought is different, its SSynt-head being APPLES:

whateverapples thatboughtAlan.)
The situation is the same with indirect-interrogative pronouns, as in I wonder whom you love
or He asked what book Alan had brought. Such a pronoun depends syntacticallydirectly or indi-
rectlyonly on the MV of the completive clause, although there is a direct Sem-D between the
pronoun and the MV of the matrix clause: I wondersemwhom... and He askedsemwhat...
(for more on the representation of indirect questions of this type, see Kahane/Mel'uk 1999).
To conclude, let it be mentioned that in the D-descriptions of various languages (English, Da-
nish, Esperanto, etc.) for a Machine Translation system (Schubert 1987: 100-102, Maxwell/ Schu-
bert 1989), the relative pronoun is treated as a Synt-dependent of the MV of the relative clause.

1.1.2. Raisings
It is also said that in sentences of the type He keeps talking, the pronoun he depends both on
keeps and talking, because it is the subject of both (cf. Hudson 1988b: 194ff; the construction is
even commonly known as 'Subject Raising'). However, if Sem-D and Synt-D are distinguished,
this reasoning does not apply: he is the SYNTACTIC Subject of keeps (HE controls the agreement of
KEEP, is positioned with respect to KEEP, undergoes inversion with DO) but the 1ST SEMANTIC
actant of talking (this allows one to account easily for cooccurrence restrictions: *Something talks,
etc.). At the SSynt-level, there is no direct Synt-link between he and talking: *he talking is not a
phrase of English. The situation is slightly different with 'meteorological' verbs: in It keeps rain-
ing, the impersonal pronoun it is the SSynt-subject of keeps, but it does not appear in the SemS nor
in the DSyntS, since it is semantically empty; it is inserted in the SSyntS by a special rule, based on
the lexical entry for [to] RAIN, so that the question of its semantic compatibility with the verb does
not even rise (the verb [to] RAIN has no semantic actant).
 72

1.1.3. Subordination of Coordinate Expressions

In sentences of the type Alan reads books, newspapers and magazines, the elements newspa-
pers and magazines are often said to have two Synt-governors each: newspapers depends on books
and on the verb reads, while magazines depends on and and again on the verb reads. Formally,
then, the DSyntS of such a sentence must be as follows:
READ

I II II II

COORD COORD II
ALAN BOOKS NEWSPAPERS AND MAGAZINES
'Duplicating' Synt-D s are shown by bolder branches. This type of representation is discussed in
detail on basis of Russian data in Sannikov 1989: 32-41.
Whichever the advantages of this representation,30 it reflects again a confusion of different
types of D s: in point of fact, 'duplicating' branches show Morph-D s (if any: Alan sees Helen and
me *I) and Sem-D s; they do not have the same nature as the coordinate Synt-D s, whichas all
Synt-D sessentially specify linear positioning of wordforms. The intuition that I would like to
capture in the case of coordinate, or conjoined, strings on the Surface-Synt-level is not that every
element of a conjoined phrase depends 'in parallel' on the same Synt-governor, but rather that a
conjoined phrase as a whole depends on its Synt-governor via its Synt-head (= its first element, see
1.3 below).

1.2. 'Mutual Dependency'

Fairly often, grammarians insist on mutual dependency between the MV of a clause and its
SSynt-Subject, saying that even if it is the MV that 'represents' the whole clause, the Subject con-
trols the form of the verb (The cat is sleeping vs. The cats are sleeping); moreover, the Subject
and the MV constitute a communicative unit consisting of a theme/topic and a rheme/comment.
Again, such statements are due to confusion between different levels of dependency: the fact that
the Subject depends on the MV syntactically does not prevent the MV from depending on its
Subject morphologically. In many languages the MV agrees not only with the Subject but with the
DirO (and sometimes with the IndirO) as well: cf. (35a) above, p. 00; this, however, does not belie
the universally accepted syntactic status of objects as dependents of the MV.

1.3. 'No Dependency'

While some linguists treat coordination by means of double dependencies (1.1 above), it is
also frequently said (Matthews 1981: 196, Hudson 1988a: 314) that there is no Synt-D at all within
conjoined, or coordinate, expressions: in Leo and Alan [came], as well as in Leo or Alan [will do
it] nothing is the Synt-head. This viewpoint goes back to Tesnire 1959: 339ff.31 Once again,
 73

Synt-D is being confused with subordination (which is a particular case of Synt-D). Leo and Alan
is a phrase of English, and so is and Alan, while *Leo and is not (the fact that a pause is possible
after ANDas, for instance, in Leo and, || believe me or not, || Marga...does not impart to the
expression *Leo and the status of a phrase; it still is not an utterance of English). The phrase Leo
and Alan has thus the passive Synt-valency of Leo, and not that of and Alan, the passive Synt-
valency of the phrase and Alan being determined by and rather than by Alan (the phrases such as ...
and Alan, ... or Alan, ... but not Alan etc. can be only conjoined constituents, and this property
comes from the coordinate conjunction); therefore the Synt-Ds in Leo and Alan are as follows:
LEO coordinativeAND conjuctionalALAN.
In a conjunctionless coordinate phrase such as Leo, Alan, Helen the Synt-Ds are
LEOcoordinativeALANcoordinativeHELEN.
The Synt-head of a conjoined phrase is, at least in English and similar languages, its first ele-
ment (independently of the presence/absence of a coordinate conjunction). Note that in a number of
languages, the first element in a coordinate string has some special properties. Thus, in some Bantu
languages, only the first verb in a coordinate string of verbs (stood up, drank his coffee, took the
book and left) has a complete morphological marking, including tense; all the following verbs are in
a specialconjunctiveform, which precludes the expression of tense. In Nias (Indonesia),
in a string of coordinated nouns, only the first noun is inflected according to the external context,
while all the others remain in the unmarked nominative; etc. The proposed D-description of coordi-
nate phrases is shared, for instance, by Engel (1982: 263ff). For more on Synt-D in connection
with coordination, see 5 below; a concise review of possible solutions to the problem of D -
description of coordination is presented in Schubert 1987: 104-119.32

1.4. 'Insufficient Dependency'

Many linguists have criticized the D-approach for its incapacity to express what they call the
multi-layer character of syntactic structure. Thus, in Alan gives an apple to Helen, APPLE is some-
how closersyntactically, not linearly!to GIVE than is HELEN; and ALAN has the loosest link
to the verb ('external argument,' as it is known in some approaches). Without going into a discus-
sion of what this syntactic closeness really means, I can point out simply that all such distinctions
are easily and naturally expressed via the names of D-/S-SyntRels:
GIVE GIVE
I III subj prepos-obj
II dir-obj
TO
ALAN ALAN
APPLE HELEN APPLE prepos

HELEN
 74

Labeled SSyntRels guarantee the capacity of any D -description to state the 'closeness' (or
'remoteness')of course, in the structural senseof any elements of the sentence.

2. Syntactic Dependency in Action: Eight Illustrative Case Studies

Let us consider now several particular syntactic constructions and show how they are des-
cribed using Synt-D; I will mainly concentrate on the ORIENTATION of Synt-D, i.e. on the problem
of Synt-head. Since in some cases all theoretical arguments advanced in support of this or that
viewpoint concerning the treatment of a given phrase in terms of Synt-D seem inconclusive, I will
try to argue, when need be, based on the RULES necessary to produce the construction in question
within the framework of a stratificational multilevel semantic-oriented linguistic model (more
specifically, the Meaning-Text Model). I will try to show that the opposite decision concerning the
choice of the Synt-governor entails a complexification of the processing rules.

2.1. Russian Numeral Phrases

In Russian, a numeral phrase NUM + N shows rather complex behavior:
if the numeral does not end in ODIN (1) and is in the nominative or the accusative, the noun is
in the genitive and its number depends on the numeral (with DVA (2), TRI (3), and ETYRE (4) or
any numeral that ends in these three23, 32, 44, ..., 1452, etc.the noun is in the singular,
while with all other numerals it is in the plural);
if the numeral ends in ODIN (1) (e.g., 1231), no matter in what case it is, the number of the
noun is singular;
if the case of the numeral (other than ODIN (1)) is the nominative or the accusative, and if it is
(or ends in) DVA, it agrees with the noun in gender; etc.
This complexity engendered much discussion concerning the orientation of Synt-D in the
NUM + N phrase; all logically possible solutions have actually been proposed (NUMN;
NUMN; NUM N; in the nominative and the accusative it is NUMN, in other cases
NUMN; etc.). In actual fact, the orientation of Synt-D in Russian numeral phrases is always
NUMN, since the passive Synt-valency of the phrase is obviously that of N, and not that of
NUM. What obscures the picture is again confounding the Synt-D with variegated Morph-D s
(Mel'uk 1985: 59-102; for the opposite viewNUMN, i.e. the numeral is the Synt-head, see
Corbett 1993).
To make my point clearer, I will describe in parallel the production of two phrases, one with
a genuine numeral DVA (two) and another one with a measure noun KUA (heap, a lot):
[On prol] dva romana ([He read] two novels)
and
[On prol] kuu romanov ([He read] a-lot of-novels).
 75

In the SemS, both phrases have a similar representation, where quantification appears as any
semantic modification would:
(read) (quantity) (read) (quantity)
2 1 2 2 1 2

(novels) (two) (novels) (a lot)

In the SemS DSyntS transition, the direct DSynt-link between ITAT (read) and ROMA-
NY (novels) is preserved, such that the quantifying expression remains a modifier of ROMANY
(via the DSyntRel ATTR):
ITAT ITAT

II II

ROMANY ROMANY
ATTR ATTR

DVA KUA
This allows one to take care of all cooccurrence constraints holding between the verb and the DirO
noun, including collocational constraints expressed in terns of Lexical Functions.
Under the DSyntS SSyntS transition, the situation changes: the NUM DVA remains sub-
ordinated to the quantified N, while the Nmeasure KUA becomes the SSynt-governor of it:
ITAT ITAT

direct-objectiva direct-objectival

ROMANY KUA
quantitative completive

DVA ROMANY
This is done since, from a purely SSynt-viewpoint, the phrase itat kuu ([to] read a-lot) behaves
exactly as any other pair V(transitive)dir-objN: the N is inflected and positioned as any regular
DirO should. On the other hand, kua romanov behaves as any other pair NcomplN does. Be-
cause of this, for itat kuu romanov, the inversion of the Synt-D between KUA and ROMANY
is justified by a considerable economy of rules, which otherwise would have to be doubled: a
special set of rules would be needed to describe the treatment of a quantifying modifier (= KUA)
that behaves as a DirO and another set of rules for the treatment of a DirO (= ROMANY) that
behaves as an adnominal complement. But for itat dva romana nothing justifies such an inversion:
the extremely complex rules that compute the inflection of the NUM, of the quantified N and even-
tually of some depending adjectives remain the same, whichever element is taken to be the Synt-
 76

head in the phrase NUM + N (see these rules in Mel'uk 1985: 162-210). Therefore, the overall
simplicity of DSynt-rules requires not to invert the Synt-D between NUM and N. Ergo, on both the
DSynt- and SSynt-levels, we have NUMN.
The treatment proposed for itat kuu romanov ([to] read a lot of novels) can be easily
extended to cover all the constructions including quantifying expressions, in particular the measure
phrase, as in Rus. vypit tri litra piva ([to] have drunk three liters of beer), Eng. have eaten ten
dollars of bagels, Fr. faire deux heures de sieste, lit. ([to] have two hours of nap), etc. In the
DSyntS, the measure phrase is represented as a modifier of the noun quantifiedin the same way
as the synonymous expressions vypit pivo v koliestve trx litrov, lit. ([to] have drunk beer in
quantity of three liters), Eng. have eaten bagels for ten dollars, Fr. faire une sieste de deux
heures, lit. ([to] have a nap of two hours). It is on the level of Surface-Syntax Structure that the
inversion of dependency takes place.

2.2. A Russian 'Approximation'-Marking Preposition

In the Russian phrase okolo pjati kilogramm (about five kilos) the preposition OKOLO (lit.
(close to), here (approximately)), is the Synt-head of the phrase: without it, the numeral phrase has
the exact distribution, i.e. the passive Synt-valency, of a noun, but with OKOLO the numeral
phrase can only be used as the Synt-Subject or DirO. Thus, the phrase with okolo cannot be the
complement of a preposition (*dlja okolo pjati kilogramm (for about five kilos)) or an IndirO
(*raven okolo pjati kilogrammam ([is] equal to about five kilos)). Therefore, we obtain, on the
SSynt-level, okolokilogrammpjat. (In English about (approximately) does not have the same
properties: for about five kilos is OK; and therefore its SSynt-status is different:
aboutfivekilos.) This representation is buttressed by the complete identity in syntactic behavior
of this okolo and all other 'genuine' Russian prepositions; thus, all of them, together with the
numeral, follow the noun quantified in the approximate-quantitative construction: dlja pjati
kilogramm (for five kilos) ~ kilogramm dlja pjati (for approximately five kilos) and okolo pjati
kilogramm (about five kilos) ~ kilogramm okolo pjati (approximately about five kilos).
A similar construction exists in Latin:
(36) Lat. Circa quingentos Romanorum cecid +erunt
around 500.ACC Roman.PL.GEN fall.PERF 3PL
(About 500 Romans fell).
The preposition CIRCA governs the case (namely, the accusative) of quingenti (500), as all Latin
prepositions do: it is the Synt-head of the phrase circa quingentos Romanorum; however, it is
omissible without any syntactic effect on the phrase.
In Russian, as in Latin, taking the preposition OKOLO/CIRCA to be the Synt-governor of
the numeral phrase as any other preposition is allows us to avoid writing special syntactic rules to
treat these syntactically quite ordinary prepositions, which are only semantically 'deviant' (they
 77

manifest a kind of adverbial meaning: semantically, they are monoactantialin contrast to genuine
prepositions, which are biactantial).

2.3. Determiners as Heads?

Several linguists argue that in the DET + N phrase the determiner is the Synt-head: thus, in
English we should have THEsyntN, ANYsyntN, etc. (Hudson 1984, 1990: 271ff, Hew-
son 1991; cf. also Vennemann 1977: 270, 296). I cannot analyze their argumentation in depth, but
within the framework expounded above, such a description is unacceptable, and this, for the
following three reasons.
First, the passive Synt-valency of the phrase the dog is that of the noun dog, not of the article
the. If in some syntactic positions DOG cannot appear without an article (or any other determiner),
this happens because articles and the determiners in general are analytical exponents of grammemes
of an inflectional categorynamely, of the definiteness of the nounand in these positions an
English noun cannot be used without a marker of its definiteness, just as a Latin noun cannot be
used without a case-number suffix. The expression *Dog is barking is ungrammatical, indepen-
dently of the fact that its SSyntS is well-formed: the problem here is the incorrect grammatical form
of the lexeme DOG, very much like the bad expression *The dogs is barking, where the SSyntS is
also 100% correct, but the grammatical form is is not. Ergo, the SSyntS is here the syntdog,
anysyntdog, etc.; cf. the phrase that (stupid) John, which has the distribution of John and not
that of the determiner that, or Dogs are faithful, where the noun dogs appears without any overt
determiner.
Second, it is necessary to reflect the parallelism in the syntactic behavior of such phrases as
the dog, this dog and Alan's dog; are we prepared to say that this and Alan's are the SSynt-
governors of dog?
Third, analytical exponents of grammemes of a lexeme in most cases syntactically depend on
this lexemein the SSyntS, since they do not appear at all in the DSyntS (auxiliary verbs that ex-
press the grammemes of tense, mode, aspect or voice constitute an important exception, see below,
2.5). For instance, in Tagalog an analytical case marker of an Nang [NOM], ng [OBL] and sa
[DAT]syntactically depends on the noun, while the preposition (that syntactically dominates N)
imposes the choice of the case: in the DSyntS, we have, for instance, PARA (for)syntBABAE
(woman), and in the SSyntS, PARAsyntBABAEsyntSA, which gives para sa babae (for
[the] woman) (the preposition PARA requires the dative). In languages where the plural of a noun
is expressed by a separate word (Dryer 1989), this plural exponent equally depends syntactically on
the noun: Yapese (Austronesian) ea pisyntkaarroo neey (the PL car this) = (these cars) or Mixe
(Mexico) he pi' mi&synt /aHk& (the little boy PL) = (the little boys). Cf. also the Russian par-
ticle BY that expresses the subjunctive of a verb on which it depends: Ja by syntpoexal (I
 78

would go). In Hawaiian (and other Polynesian languages) all markers of the verb's inflectional
grammemes are analytical and syntactically depend on the verb: uasynthele au (COMPL(etive)
go I) = (I went), ke syntkali syntnei au (PROGR (essive) wait I) = (I am waiting), e synt
kali (IMPER(ative) wait) = (Wait!), etc. Considering an analytical grammeme marker as a Synt-
governor in the SSyntS would entail a restructuring of the DSyntS, where this marker does not
appear at all; but why engage in something complex when one can easily do with something
simple? Since the article is a particular case of an analytical grammeme marker, it should be consi-
dered a Synt-dependent of the noun. (The solution DETsyntN is successfully defended in Van
Langendonck 1994; for a different treatment of the DET + N phrase in Salishan, see Beck 1997:
109-118.)

2.4. Romance Clitics

Clitics in French (and other Romance languages, where Clitic Raising exists) pose a difficulty
for a D-description: the clitic does not always syntactically depend on the same wordform on which
its source [= the noun the clitic replaces] depends. Thus, compare (37a) with (37b), where the clitic
changes Synt-governor vis--vis that of its source:
(37) a. Elle a t fidle Pierre (She has been faithful to Peter).
b. Elle luia t fidle, lit. (She to-him has been faithful).
On the Surface-Synt-level, where clitics first appear (the Deep-Synt-level admits only nominal
sources of clitics-to-be), a clitic depends syntactically on its host word, with which it forms a
possible utterance (= a prosodic unit, i.e. a phrase, as in Sp. lo ve, lit. ([(s)he] it sees) or le da, lit.
([(s)he] to-him/to-her gives)) and with respect to which it is linearly positioned; in Romance lan-
guages, the host of a clitic in the SSyntS is not necessarily the same element on which the source of
the clitic depends in the DSyntS. The 'new,' i.e. Surface-Syntactic, governor of the clitic is com-
puted by special rules of the DSyntS SSyntS transition; roughly speaking, it is the Synt-head of
a dependency chain on the last element of which the source of the clitic depends.

2.5. AUX + V Phrases, English-Style

What is the orientation of Synt-D in the phrase AUX + V in English (and similar languages)
AUXsyntV or AUX syntV? As before, I will try to argue based on the rules necessary to
produce the phrase in question from a SemS. Suppose we want to have the sentence Alan has slept.
Starting with a SemS
(Alan) o1o (sleep)
(plus the indication of time), the semantic rules of Lexicalization and Arborization construct the
DSyntS of the form
ALAN oIo SLEEP .
sg ind, pres, perf
 79

The compound, or analytical, form of the verb is represented, at this level, as one node directly
linked to the subject node by the Deep-SyntRel I; thus, all the lexical selection constraints (that may
exist between the verb and its Synt-Subject) can be easily accounted for. In the SSyntS, the DSynt-
node
o SLEEPind, pres, perf
is expanded into
HAVE operfect-analyticalo SLEEPpast.participle
ind, pres
by the following Deep-Syntax rule:
L o HAVE(V) ind, pres operf-analyto L(V)past.participle
(V)ind, pres, perf
From the purely logical viewpoint, we can take HAVE as the SSynt-head of the phrase
AUX + V (as shown in our illustration) or as the SSynt-dependent: for this rule itself it makes no
difference. But for all the Surface-Syntax rules which have to compute the inflections on HAVE,
linearly position it in the sentence and check the well-formedness of the SSyntS the difference is
quite significant. If HAVE is considered to be the SSynt-head, all the SSynt-rules that apply to any
pair NsubjV will automatically apply to the auxiliary HAVEas they do to any verb in the
role of SSynt-head, including the non-auxiliary HAVE. More specifically, aspects under which an
auxiliary (BE, DO and HAVE) must be treated as any 'normal' finite verb of English include:
Agreement with the Subject (I have written vs. He has written as I read vs. He reads;
including all complex cases of the type The committee has/have, etc.).
Word order, in particular, inversion (Have I? as Can I?).
Being the only verb in the clause ([I know that] he has as [I know that] he works; or in tags:
He has not gone, has he?; He works, doesn't he?).
Carrying the grammemes of mood and tense.
Parallelisms with the non-auxiliary BE, DO and HAVE (He is astonished vs. He was asto-
nished by John; He does his work vs. He does work; He has arrested John vs. He has John
arrested; He has gone vs. He has to go).
Parallelisms with semi-auxiliaries such as GET (He got robbed), KEEP (He keeps going) or
GOING TO (He is going to read), which have to be treated in a similar way to the genuine auxi-
liaries.
On the other hand, there are no idiosyncratic SSynt-properties of English auxiliaries that would
require any specific rule to deal with them.
However, if the auxiliary HAVE (or BE, or DO) is not taken to be the SSynt-head of the
AUX + V phrase, a bunch of additional rules have to be written to deal with a finite auxiliary verb
which is not the SSynt-head of the clause, as well as with a non-finite verb form which is. There is
no justification for such useless multiplication of entities; ergo, on the SSynt-level, we have
 80

AUXV (which corresponds to Criterion B1). For more argumentation in favor of AUXV, see
Hudson 1987: 117-118 (English) and Milievi 2000: 00 (Serbo-Croatian).

2.6. Conjoined Nominal Phrases N + CONJ + N

According to Criterion B1, in such a phrase as Alan and Helen the Synt-head is ALAN: the
passive Synt-valency of the phrase is the same as that of the noun ALAN (rather than that of the
phrase AND HELEN). Nevertheless, in many syntactic frameworks AND is considered the Synt-
head of a coordinate string, ALAN and HELEN being its actants: ALANANDHELEN (the
same description is applied to all coordinate conjunctions: e.g., Schubert 1987: 104ff; cf. Footnote
32, p. 00).
Once again, purely theoretical argumentation proved to be of little use here, so I will reason
from the viewpoint of the rules that have to synthesize such strings.
If we take the conjunction to be the SSynt-head of the coordinate string we run into the
following difficulties.
First, to be able to specify the linear order of conjuncts, which in many cases is significant,
both Synt-Ds starting from CONJ have to carry different labels.
Second, rules for the conjoined strings without a coordinate conjunctionsuch as Alan,
Helen, Leomust be completely different, or else a 'dummy' artificial conjunction has to be added
to the SyntS.
Third, the linguistic rules that deal with the SSynt-Subject, the DirO, the IndirO, the comple-
ment of a preposition, the Proplepsis (= Fronted Topic), and the likein one word, with an N
have to be repeated for the conjunction! And this will be sufficient only for the conjunction that
links two nouns; for conjunctions linking lexemes of other parts of speech still other additional
rules are needed. More than that: these rules must be extremely complex, since they have, e.g., to
assign grammatical case to the conjunction and then percolate it to the nouns linked by the conjunc-
tion, etc.
Fourth, must the conjunction be the Synt-head on the DSynt-level? Presumably so, since this
is closer to its semantic role. Then all the selectional constraints acting between the verb and the
conjoined nouns will not be easily checkableagain an unnecessary complication.
In case we take the initial (= first) element of the conjoined phrase to be its Synt-head, no
rules dealing with nouns, verbs, adjectives or adverbs have to be doubled and no additional rules
are required (just specific rules for the conjunction and the second conjoined element), nor any
dummy added. At the same time, absolutely nothing is lost. Why should then anyone want to add
complexity without any visible gain? Ergo, on both the DSynt- and SSynt-levels, we have, respec-
tively:
X COORDCONJIIX
1 2
 81

and
X coordinative CONJ conjuctional X .
1 2

2.7. Russian 'Exotic' Coordination of Interrogative/Negative Pronouns

In Russian, the interrogative and negative pronouns which bear different Deep-Syntactic rela-
tions to the governing verb are allowed to form a coordinate string in the Surface-Syntactic struc-
ture (in the Deep-SyntS there are no direct syntactic links between these elements: they are subor-
dinated to the verb 'in parallel'):
(38) a. Kto, komu i em pomog?,
lit. (Who, to whom and with what helped?).
b. Nikto, nikomu i niem ne pomog,
lit. (Nobody, to nobody and with nothing not helped).
To represent the phrase kto, komu i em on the SSynt-level simply as all other coordinate phrases
are represented, that is, as
ktocoordkomu coordiconjunctem,
is insufficient. In a 'regular' coordinate construction any SSynt-dependent element plays with res-
pect to the SSynt-governor of the whole coordinate string the same SSynt-role as its SSynt-Gover-
nor itself; but in this case, kto [NOM] is the SSynt-Subject, but komu [DAT] is an IndirO and em
[INSTR] is another IndirO of the verb pomog (helped); accordingly, all three pronouns are inflected
differently. To account for this, it has been proposed (Sannikov 1989) to use double dependency,
namely to add to the SyntS above the indication of the direct Synt-D of each pronoun on the verb
pomog. But these added Synt-Ds do not have the same substantive nature as the Synt-Ds covering
the coordination in this case: the added links are needed only to compute the Morph-Ds (under syn-
thesis) and the Sem-D s (under analysis). However, as we have seen, Morph-D s and Sem-D s can
link two wordforms that do not have a direct Synt-D between them. Therefore, it is preferable to
introduce some special SSyntRels just for this very special construction: coord-subj, coord-dir-obj,
coord-indir-obj, etc. Such SSyntRels indicate, in a natural way, the exotic character of this coor-
dinate phrase and encode the 'actual' SSynt-roles of its 'displaced' elements.
A similar method can be used in comparative constructions (Savvina 1976). For instance, in
Russian, the two following comparative constructions have to be distinguished in their SSyntSs in
the following way:
(39) [Ja ljublju Mau bole,] emconjunct-subjVan+ja
(I love Masha more than Vanya [does]).
vs.
[Ja ljublju Mau bole,] emconjunct-dir-objVan+ju
(I love Masha more than [I love] Vanya).33
 82

Another possibility to represent the SSyntS of these constructions would be to consider the gram-
matical case of the SSynt-dependent in such coordinate or comparative strings as semantically
meaningful and admit it into the SSyntS of these constructions; this is, however, too technical a
point to be discussed here.

2.8. Elliptical Constructions

How should one describe in the SSyntS common gappings of the type Alan went to Singa-
pore and Leo to Paris? Since the expression and Leo to Paris is not a phrase of English, it cannot
such as it isbe assigned a well-formed SSyntS. It is a 'mutilated' expression, which lost its top
node, in this case a finite verb,a 'non-canonical conjunct,' in Hudson's (1988: 305) terminolo-
gy. But before the verb is deleted during the transition SSyntS DMorphS, it imposes on its
dependentswhere appropriategovernment-induced inflection grammemes. Therefore, to repre-
sent the SSyntS of the expression in question in terms of Synt-D , one has to use an empty node
that stands for the elided verb; this node is marked with a blank , linked by an anaphoric
relation (shown by a dashed line) to its antecedent, in this case, the verb GO:
(40)

GO AND
coord conjuct
subjectival prepos-objectival subjectival prepos-objectival
TO TO
ALAN LEO
prepositional prepositional

SINGAPORE PARIS
In the DSyntS, the elided node is present and labeled with the appropriate lexeme, in our case, with
the verb GO.Note that the anaphoric relation between the empty node and the GO node is not one
of coreference (the two nodes are not coreferential); it is a relation of lexical identity.
This is how the SSyntS of elliptical expressionsor, more precisely, of non-canonical con-
juncts is represented in the Meaning-Text approach (see also below, the last paragraphs of 3).
Such a 'dynamic' way of reflecting ellipseswhich are, after all, operationscorresponds to
Lobin's (1993: 111ff) proposal to use a procedural description for all coordinate structures, not just
for ellipses.34

3. Advantages of Syntactic Dependency

The remarks that follow are very sketchy and superficial: a systematic discussion of the
advantages of the D-approach and its comparison with the constituency, or C -, approach would re-
 83

quire another paper, perhaps longer than the present one. Still I think that these remarks can be use-
fulthey point at least in the right direction.
Let me begin with two GENERAL considerations. First, in a linguistic description that takes
semantics into account seriously, the D-approach in syntax imposes itself, since it ensures a much
better fit of syntactic structure with semantic structure, where dependencies are universally recog-
nized (most versions of predicate calculus language used in semantics are, in point of fact, D-bas-
ed). A lack of interest in semantics and the postulate that syntax is autonomous are main factors that
have lead to the dominance of C -representations in syntax. In a theory where the Synt-structure of a
sentence is produced (roughly) from the Sem-structure of this sentence, this Sem-structure being
written in terms of Sem-D s, it is much more natural to see the Synt-structure as being based on
Synt-D s.
Second, a D-representation with labeled SyntRels is formally more powerful than a 'pure' C -
representationin the sense that the former allows one to present all relevant syntactic details much
better than the latter. As a result, linguists have been forced, practically from the beginning, to spe-
cify heads of the constituents as opposed to satellites (e.g., Pittman 1948) and the relations between
them. But in a C -representation, as soon as one starts marking heads and indicating types of
SyntRels between heads and satellites, the heavy machinery of constituencyparticularly, non-
terminal nodes, numerous empty nodes, and artificial ordering of elements in the SyntSbecomes
useless, because redundant: all these pieces of information can be easily computed, if and when
needed, from the D s specified. Because of this, most modern syntactic theoriessuch as Perl-
mutter's Relational Grammar, Bresnan's Lexical-Functional Grammar or Pollard
and Sag's Head-Driven Phrase-Structure Grammarare moving fast in the direction
of the D-approach.
To these general considerations, one can add a number of SPECIFIC ones. Namely, there are
at least five important linguistic phenomena for the description of which Synt-D is really crucial:
valency, voice, restricted lexical cooccurrence, word order, and ellipses of all types. (I am not
implying that the C -approach cannot handle them; but the D-approach does it, I think, in a more na-
tural and therefore more economical way.)
1) Valencyor, more precisely, active valencyis a property of lexemes: a lexeme
opens 'slots' for other lexemes that it 'attracts' as its dependents. Linguistic valency is obviously a
metaphor based on valency in chemistry: atoms have valencies to link with other atoms and thus
form molecules. In much the same way, a lexeme has semantic, syntactic and morphological
valencies to link with other lexemes. Lexemes Li that 'fill' the valencies of the lexeme L depend on
it, exactly in the sense in which dependency has been defined above. Actually, valency and
dependency are related in a very direct way; cf. Baumgrtner 1970: 62ff and also Eichinger/Eroms
(eds) 1995. Active valency is of course not the only 'source' of dependencythere is passive
 84

valency of lexemes as well; however, active valency shows the convenience of using Synt-D in an
especially graphic way.
2) The inflectional category of voice is crucial to the understanding of semantics, syntax and
morphology. Voice grammemes mark the change of the basic diathesis of the verb, i.e. the
correspondence between its semantic and syntactic actants (Mel'uk 1997a), or, to put it different-
ly, between its Sem- and Synt-dependents. No wonder, then, that voice and voice-related categori-
es are much better described in the D-approach; in particular, they have been the focus of research
within the framework of Perlmutter's Relation Grammar or Foley/Van Valin's Function and Refe-
rence Grammar far more than in any C -based theory.
3) For a systematic description of restricted lexical cooccurrence, or collo-
cations, the apparatus of Lexical Functions is proposed (olkovskij/Mel'uk 1967,
Mel'uk 1996a). Each collocation is described as having the structure f(x) = y, where f is a
particular lexical function, x is a lexical unit which is the base of the collocation, and y, a set of
(more or less synonymous) lexical units each of which is the collocateit expresses, contin-
gent on x, the meaning of f. Here are a few examples:
Intensifier Support Verb Realization Verb
Magn(smoker) = inveterate Oper1(favor) = do Real1(goal) = achieve
Magn(sleep) = like a log Oper1(order) = give Real3(order) = execute
Magn(hot) = burning Oper2(exam) = take Real2(exam) = pass
The number of lexical functions is about 60, and they are universal; their values, on the contrary,
are of course language-dependentthey are specified, for each language and each lexical unit, in a
special lexicon. Using them greatly facilitates lexicalization in the transition SemS DSyntS
SSyntS, when the appropriate collocates have to be selected.
Now, as is easy to see, the lexical-functional dependency between the base lexeme of a collo-
cation and the collocate lexeme is supported by a Synt-D between them. Thus, Magn(armed) = to
the teeth, and armedsyntto the teeth; similarly, Oper (visit) = [to] pay, and paysyntvisit,
1
or Real2(exam) = pass, and passsyntexam. For each lexical function, a particular Synt-D be-
tween its base and its collocate is specified. Outside of Synt-D, there is no economical way to des-
cribe the collocations properly.
4) Synt-D is especially convenient for the description of word order. Using Synt-D s forces
the linguist to separate strictly and consistently the hierarchical (= genuinely syntactic) 'order' from
the linear order, which is a surface means for the expression of the former. Thus, the main task of
natural language syntaxlinearizing a two-dimensional Synt-structure (explicitly formulated in
Tesnire 1959: 19-20)can be solved with much more ease with Synt-Ds than in any other way.
The advantages of Synt-Ds for the description of word order can be resumed in three points:
 85

First, word order rules can be easily formulated in terms of positioning a Synt-dependent
with respect to its Synt-governor (before or after it). Again, Tesnire (1959: 22-25, 32-33) stated
this fact explicitly, dividing languages in centripetal (where a Synt-dependent precedes its
Synt-governor) and centrifugal (where a Synt-dependent follows its Synt-governor); both can
be so consistently or inconsistently. For some languages, this allows for very compact
formulations; e.g., in Japanese all Synt-dependents precede their governors, in Welsh almost all
Synt-dependents (the only exception being the article y) follow their governors (Hudson 1990:
105):
(41) a. Japanese [a consistently centripetal language]
Itiban takai siraga+de+no sensei+wa kono omosirokunai hon+o kai+ta
very tall gray-haired professor this boring book wrote
lit.
(Very tall gray-haired professor this boring book wrote).
b. Welsh [a consistently centrifugal language]
Ysgrifennodd athro tal iawn a gwallt llwyd ganddo y llyfr undonnog hwm
wrote professor tall very and hair gray to-him the book boring this
lit. (Wrote professor tall very and hair gray to-him the book boring this).
But even in languages where the linear distribution of Synt-governors vs. Synt-dependents is
not as clear-cut as in Japanese or Welsh, that is, in 'inconsistent' languages, resorting to these no-
tions helps to state the word-order rules. Thus, in Arabic the majority of Synt-dependents follow
their governors, with the notable exception of the demonstratives and numerals; in Hungarian, the
majority of Synt-dependents precede their governors, with the notable exception of the relative
clause; etc. Such facts were theoretically discussed already in Trubetzkoy 1939 and practically used
in numerous language manuals and descriptive grammars; cf. an outline of word order typology in
Xolodovi 1966. In this vein, a relatively complete description of word order in Synt-D -terms
(within simple clauses) was proposed for Russian (Mel'uk 1967, 1974 [1999]: 260-302).
Second, Synt-D has allowed for the discovery (Lecerf 1960, Hays 1960) of an important
property of word order in all languages, called projectivity. If we supply an average sentence
with its SyntS written in terms of Synt-D and draw a perpendicular from each wordform to the
node that represents it in the SyntS, then:
1) no branches of the SyntS intersect;
2) no branch intersects with a projection perpendicular.
Let me illustrate this with sentence (1), associating its surface form with its SSyntS (next page).
As one can easily see, the sentence appears as a 'projection' of the SSyntS such that
SSyntS's branches cross neither each other nor the projection perpendiculars; hence the name
"projectivity".
 86

For decades, cocoa farming has escaped such problems by moving to new areas in the tropics
Figure 1: Sentence (1) and its SSynt-Structure
An absolute majority of sentences in most languages are projective. Taking this fact into ac-
count ensures a more elegant formulation of word-order rules and greatly facilitates the analysis and
synthesis of texts: with the exception of particular cases (see immediately below), only projective
sentences must be produced from a given SSyntS, and only SSyntSs that guarantee projectivity
must be associated with a given sentence.
However, projectivity can be systematically violated in many special cases, for instance:
1) English

the most interesting paper in the collection

The culprit here is the superlative marker of the adjective; cf. the representation in a tree form:

the most interesting paper in the collection

2) French

la fille dont je connais le pre,

lit. (the girl whose I know the father) = (the girl whose father I know)
The culprit is the extracted relative pronoun dont:
 87

la fille dont je connais le pre

3) Serbo-Croatian

Verski mu je uitelj odvratio,

lit. (Of-faith to-him has [actually, (is)] the-teacher answered)
= (The teacher of faith has answered to him).
The culprits are two 'displaced' cliticsa dative pronoun mu (to him) and the auxiliary verb je (is):

Verski mu je uitelj odvratio

I could indicate much more constructions in different languages where non-projectivity is
possible or obligatory; all such cases must be isolated and specified in concrete linguistic descrip-
tions. Now, while within the D -approach the representation of non-projective structures poses no
problem whatsoever, the C -approach is unable to represent non-projective structures without some
additional (and relatively clumsy) machineryfor instance, transformations. In this respect, the
D-approach is again superior to its rival.
Third, the D-approach is much less rigid (than the C -approach) and has the inherent ability to
accommodate easily what is known as 'non-configurationality' and long-range dependencies. The
perturbations introduced into the word order of a sentence by its Communicative Structure
Frontings, Extractions, Postponings, etc. plus all sorts of 'displacements' in such languages as
German or Russiancan wreak havoc on a C -structure, since even the closest-knit phrases can be
torn apart and permuted. On the other hand, D-structure, without linearity and contiguity, is totally
insensitive to such permutations: they happen in the linearized DMorphS of the sentence and do not
at all affect the SSyntS. The reason is obviousa strict and complete separation of hierarchical (=
syntactic) and linear links in the D-approach. As a result, the D-approach does not know problems
in representing discontinuities, which, in the literal sense of the word, simply do not exist in a D -
structure. Thus, the sentence Which violins are these sonatas easy to play on? is assigned a
natural SSynt-structure written in D -terms (Fig. 2, next page). The linear break of the phrase on
which violins is produced by a word order rule that puts the phrase which violins in the first linear
position in the sentence, i.e. extracts it (during the transition SSyntR DMorphR). Note that
such an extraction is not possible for a similar SSyntS of Fig. 3 (next page): the result *Which
sonatas are these violins easy to play on? is ungrammatical and can be precluded by imposing all
 88

the necessary conditions on the extraction rule. (The SSyntS of Fig. 3 can be only realized with the
WH-word in situ: These violins are easy to play on WHICH sonatas?) See Hudson 1988b: 199ff on
the problem of extraction with a D-framework.
5) As Nichols 1993 has shown, ellipses, i.e. constituent-reducing operations, can be
conveniently characterized in terms of Synt-D. Thus, four languages studied by NicholsRussian,
Nunggubuyu (Australian), English, and Chechen-Ingush (North-Caucasian)differ with respect
to their preferences in the domain of constituent-reducing: Russian prefers to remove Synt-heads,
Nunggubuyu does it more frequently with Synt-dependents, English removes both with equal ease,
while Chechen-Ingush does neither (which means that it has few ellipses). Cf. (42):
(42) a. Rus. A Maka emu po morde, lit. (And M. to-him on the-mug) = (And M. gave him a
blow in the face), where the top nodea Synt-head, which is a verb meaning ([to] hit) =
([to] give a blow),is elided.
b. Nung. Angugu n/galima; n/galima, lit. ([He] water fetched-for-him; fetched-for-
him), where the top nodea verbal Synt-head meaning ([to] fetch)is repeated by the
narrator for more expressivity, but with its dependent (water) elided.
c. Eng. Leo is from Chernigovsky, and Alan from Paris,
where the top nodethe Synt-head of the second conjunct clause (the verb BE)is
elidedor, more precisely, factored out;
or
Susan is fond of, while Marga looks askance at, profanity,
where the Synt-dependent of fond of is elided/factored out (Russian does not admit this
type of dependent removal).
d. In Chechen-Ingush, the answer to the question (What did he give his son?) must be
Sowat dennad, lit. ([He a] gift gave), rather than simply *Sowat ([a] gift), which is
the norm in the other three languages: Chechen-Ingush does not tolerate the removal of
Synt-heads. Even the sentence meaning (Good!/OK!) must contain the verbal Synt-
head: Dika du!, lit. (Good is).

All the five above phenomena are related to the trend in modern linguistics that Hudson
(1990) aptly dubbed lexicalism: putting at the center of the linguistic description facts about
lexical units rather than facts about syntactic constructions, so that the lexicon is at last given a place
of honor in linguistic studies; cf. as well Hudson 1983, 1984 and Mel'uk 1995b. Stressing the
importance of the lexicon goes quite well with the D-approach in syntax, because in this approach
all the links are established between wordforms and based, in the final analysis, on their
lexicographic properties.
 89

BE BE

SONATAS EASY VIOLINS EASY

TO TO
THIS THIS
PLAY PLAY

ON SONATAS
ON
VIOLINS
WHICH

WHICH
Figure 2: Figure 3:
The SSyntS of the sentence The SSyntS of the sentence
Which violins are these sonatas easy to play on? These violins are easy to play on which sonatas?

4. Syntactic Dependency and Syntactic Constituency

I am not offering here a comparison in depth of D - and C-approaches in syntax (cf. Venne-
mann 1977, Nichols 1978, Hudson 1980a, b, Dahl 1980, Matthews 1981: 71-95, Mel'uk 1988:
13-17, Sgall/Panevov 1988-89); I will, however, briefly touch upon two topics relevant to such a
comparison: D -approach vs. 'pure' C -approach in syntax and hybridization of D - and C -ap-
proaches.

4.1. Comparing Syntactic Dependency with Syntactic Constituency

To compare the D -approach in syntax to the C -approach (also known as the 'Phrase-Struc-
ture' approach), one needs to make precise the concept of syntactic constituent. Let me
first take the simplest, or 'naive,' interpretation of constituent as a linearly ordered string of actual
wordforms that shows a prosodic and semantic unity (i.e., a constituent an actual phrase) and
consider constituency exclusively as based on contiguity. Such constituents are not syntactic units
in the sense that the Synt-structure of a sentence cannot be described in terms of these formations:
they are linear, prosodic and morphological IMPLEMENTATIONS of (fragments of) the SSyntS,
rather than part of it. The legitimate place of such constituents is in the Deep-Morphological
structure of the sentence. (Cf. Langacker 1997 for a convincing discussion of the role constituency
plays in language on the semantic and phonological levels, while it has no place on the syntactic
level of sentence representation. According to Langacker, syntactic structure must ensure the
 90

correspondence between semantic and phonological constituents, especially in cases of unavoidable

numerous and variegated mismatches, due to the strictly linear character of human speech, which
has to convey utterly 'non-linear' meaning. As a result, syntactic structure itself cannot be repre-
sented in terms of constituents.)
If, on the other hand, we consider the 'sophisticated' concept of a constituent as a set of
lexemes that 'go together,' this set taken before linearization, prosodization and morphologization,
then, in order for constituents to be able to carry relevant information about word order and inflec-
tion, each constituent has to have both its head and its constituent type specified; this means, more
or less, indicating the type of the Synt-relation between the constituent's elements. But no sooner is
this done than we have a D -representation! Or, to be more precise: a 'sophisticated' C -representa-
tion carries all D -information PLUS some other characteristics of the sentence represented. The
question is then whether we need these extra data to be explicitly present in the Synt-structure of the
sentence. The answer depends of course on our main theoretical postulates. I, for one, proceed
from the postulate that every part of a linguistic representation must be as homogeneous and as
compact as possible; in other words, phenomena of different nature should be represented in diffe-
rent components of the representationso that a multi-layered representation must be preferred
over a 'unified' one. If this postulate is accepted, then specifying the Synt-heads and the type of
SyntRels between the sentence elements makes all other attributes of the C-approach redundant and
therefore superfluousI mean, particularly, 1) non-terminal nodes and 2) the categorization of
constituents in the SyntS.
Non-terminal nodes indicate the Synt-constituents, but, as I have said, the constituents can be
computed from the D -representation and are needed only on a closer-to-surface levelon the
DMorph-level; therefore, they should not be present in the Synt-structure.
The categorization of the elements of a Synt-structure, i.e. the syntactic class and other syn-
tactic features of lexemes, should not be part of the SyntS, either: this is not syntactic, but lexico-
graphic information; as such it should remain 'behind the scenes,' in the lexicon's entries for the
lexemes involved.
So, if one follows the above postulatethat is, stops specifying non-terminal nodes and
keeps lexicographic information in the lexicon, rather than in the SyntSthen nothing remains of
the 'classical' C -approach in syntax.
To avoid misunderstandings, it would probably be worthwhile to formulate the following
two provisos concerning the problem of constituency in the D-approach.
1. The D-approach does not negate the existence of constituentsthey do of course exist and
have the primary importance for any complete linguistic description. (I mean here constituents as
real linguistic itemsstrings of wordforms with an appropriate prosody, not as formal abstract
entities that are automatically specified by any dependency tree as projections of complete subtrees.)
 91

However, their place is not in the SyntS, but rather, as pointed out above, in the DMorphS of the
sentence: in the SSyntS DMorphS transition, linearization is carried out in terms of constituents
that have to be computed from the SSyntS; and prosodization affects the constituents after they
have been linearized.
2. The D-approach extensively uses standard subtrees (non-linearized and non-mor-
phologized), which specify different constructions that behave identically from the viewpoint of
surface syntax. Such are, for instance, NUMP or APPROXsources of the numeral and numeral-
approximate phrases: three kilos ~ about/over three kilos ~ more than three kilos ~ from three to ten
kilos ~ ... (Mel'uk/Pertsov 1987: 487-489). Another example is V, or the verbal nucleus:
a chain consisting of verbs and some special non-verbal elements allowing for particular operations
in which it participates as a whole (Kahane/Mel'uk 1999). However, again, first, standard
subtrees are not constituents; and second, their place is not in the SyntS of a sentence, but in (the
syntactic rules of) the linguistic model, which identifies them in the SyntS and processes them as
specified.

4.2. Crossing Syntactic Dependency with Syntactic Constituency

For many years, linguists have been talking about the integration of both approachesthat
is, they have been looking for a hybrid between D - and C -representations to be used in syntax
(e.g., Baumgrtner 1970, Vennemann 1977). The incentive for such an integration comes primari-
ly from the problems related to representing COORDINATION in the D-approach (see 5), as well as
to some other linguistic phenomena such as EXTRACTION (I know which girl you told my wife
Alan was going out with, the extracted component being boldfaced; extraction happens under
focusing, relativization, or interrogation), ANALYTICAL FORMS (verbal and nominal, i.e. AUXV
and DETN: has been detected; the book), IDIOMS, COLLOCATIONS (among others, with what is
known as light verbs: make headway, pay a visit, launch an attack, Germ. zur Auffhrung brin-
gen, lit. (to-the carrying-out bring) = ([to] carry out)), and the like. The main idea is to introduce for
any of these syntactic constructions a special type of subtree that is allowedas a wholeto occu-
py one node of a dependency tree. In this way, the linguist tries to capture the intuition that such a
set of wordforms depends on or governs other wordforms as a unit. The first full-fledged specific
proposal for a 'mixed' D -/C -representation of this typeby means of so-called syntactic
groupswas advanced in Gladkij 1966 and 1968. A similar device is put forward in Lobin
1993: 42ff and 1995 (under the name of complex elements). The most recent move in this direction
is, as far as I know, Kahane 1997, where the concept of bubble is introduced: a subset of nodes
of a D-tree which is allowed to be treated as a node, while having inside a completely specified D-
structure of its own, including other bubbles. It is natural that linguists feel the need for some
 92

formation of this type; but to what extent such a 'hybridization,' or rather, 'extension' or
'enrichment,' of D-trees is welcome remains to be seen.
In particular, using multistructural and multilevel representations (cf. 3) allows for elegant
solutions of many problems that otherwise have to be treated via bubble-like entities. Thus, the dif-
ficulties of representation related to various extractions can be overcome in a natural way by re-
course to the Communicative Structure (Kahane/Mel'uk 1999). Similarly, the special character of
AUX + V or DET + N phrases, as well as of idiomatic phrases (= full phrasemes) such as with
respect to or the same, is reflected by the fact that in the DSyntS all these phrases are represented
each by one single node. Collocations are described, as pointed out in III-3, p. 00, in terms of
lexical functions, which makes explicit the specific character of the former. For instance, in the
DSyntS, a phrase such as pay a visit or do a favor is represented as
Oper oIIo VISIT or Oper oIIo FAVOR
1 1
Here, Oper1 is the symbol of a lexical function which specifies for a deverbal noun the support
verb that joins this noun as its DirO to its subject; Oper1's values, as those of the other LFs, are
given in the lexical entries for nouns:
Oper1(VISIT) = pay [ART ~] Oper1(FAVOR) = do [ART ~]
To sum up: For the time being, I believe that more progress is needed in the domain of the D-
approach to syntax before we can determine where and how to use this or that element of the C -
approach within the D -framework. However, what is already clear is that a SINGLE D -tree is not
sufficient to represent all the information that might be necessary at the syntactic level. The linguis-
tic model I propose uses TWO D-treesnamely the D- and S-SyntS; in addition, it has recourse to
a separate Communicative Structure. And that is not all: in some specific cases, more special
machinery is used (groupings, see immediately below).

5. Insufficiency of Syntactic Dependency: Coordination

If we agree to use two levels of representation for syntactic structures, that is, the DSyntS
and SSyntS, plus Communicative Structures on both levels, then a pure D-representation in syntax
seems to be sufficient for all syntactic phenomena, except for one type of construction, and that is
in the domain of COORDINATION (cf. Hudson 1990: 97ff and Lobin 1993 on a special place coordi-
nation occupies with respect to Synt-D ). The problem arises because the following situation is
possible: A wordform w 'relates' either to a whole conjoined phrase or just to its Synt-head alone,
such that the two constructions are morphologically, linearly or prosodically distinct and have
different meanings; however, within the strict D -approach, both types of structure can be shown
only by the direct Synt-D of w on the Synt-head of the conjoined phrase (the 'pure' D -formalism
does not allow for the dependency on a phrase as a whole). Consequently, one SSyntS written in
 93

terms of Synt-D s corresponds in such cases to two different meanings, which is not admissible.
Here are two examples:
(43) a. The SyntS oldmenand women represents two meanings, that is
either a meaning that corresponds to a joint reading [the adjective bears on both nouns]:
(old {men + women}) = ({old men} and {old women});
or a meaning that corresponds to a disjoint reading [the adjective bears on one noun only]:
({old men} and women [the women are not necessarily old]).

b. The SyntS [He is ] nottall andfat also represents two different meanings:
either ([he is] not {tall and fat}) [joint reading: he is neither tall nor fat];
or ([he is] {not tall} and {fat}) [disjoint reading: he is not tall, but fat].35
In such cases, different surface implementations that formally distinguish intended meanings
are in principle available (depending on the language and particular lexical means used; '||' stands
for a pause):
in (43a), old men and women [without a pause] vs. old men || and women;
in (43 b), He is not || tall and fat vs. He is not tall || and fat.
The semantic contrast accompanied by a formal contrast requires that the semantic distinction be
maintained in the SSyntS (cf. Criterion C1, p. 00). The only way to do sosticking to 'pure'
Synt-D, that is, using exclusively D-formalism and without admitting multiple Synt-dependencies
seems to be to label differently the SSyntRels involved, i.e. to have in (43a) something like
oldmodifmenandwomen for the disjoint reading
and
oldcoord-modifmenandwomen for the joint reading.
However, this solution is no good: First, it is not natural enough linguistically; among other things,
it entails doubling all SSyntRels that can link Synt-dependents to conjoined phrases. Second, it is
not sifficient formally: it cannot help in the case of more than two conjuncts, such as in hungry
men, and women, and children (hungry {men, women and children}) vs. ({hungry {men and
women}} and children). Therefore, a real alternative is to complement the 'pure' D-approach with
groupingsspecification, within the SyntS, of the D-subtrees relevant in such cases (Mel'uk
1974[1999]: 214-216, 1988: 28-33). For instance:
oldmodifmenandwomen (without grouping)
stands for the disjoint reading (({old men} and women)), but
oldmodif[menandwomen] (with a grouping indicated by square brackets)
for the joint reading ((old men and old women)).
For (43b), we will also write two different SSyntSs:
He is not restrtallandfat (for the disjoint reading: (He is {not tall} and fat))
 94

and
He is not restr[tallandfat] (for the joint reading: (He is not {tall and fat})).
Note that a grouping is not a constituent in the strict sense: there is no higher node to repre-
sent it as a whole (because a grouping is not a projection of a complete subtree), and it does not
participate in D-links as such (because in a consistent D-approach, only single nodes do; this is in
contrast to the approach advanced in Kahane 1997, where a configuration of nodes in a dependency
treea bubblecan be treated as a bona fide node). As we see in the example
old modif[menandwomen],
the branch modif leaves the node men within the grouping, but not the grouping as such. 36
An overall theory of coordination within the D-approach is put forth in Lobin 1993. The main
idea is to consider syntactic coordination as a dynamic phenomenon and to describe itremaining
within D-syntaxessentially based on OPERATIONS of structure reduction and linearization (rather
than on static SyntS representations); the book also offers a thorough review of coordinate con-
structions of German.
I would like to mention two other syntactic phenomena where groupings in the SSyntS may
be necessary:
The first one is 'layered,' or recursive, modification:
expensive {Japanese cars} vs. Japanese {expensive cars}
The linear order of adjectives is here not arbitrary: it reflects the successive, or stepwise, inclusion
of sets of the objects on which bear the modifiers, and is thus semantically relevant. (The problem
is again created by 'quasi-coordination,' i.e. by co-subordination.) Under the D-approach, both ex-
pressions have the same SSyntS:
CARS
modificative modificative

EXPENSIVE JAPANESE
so that a semantic difference is lost. It is not, however, clear to me whether this difference should
be accounted for in the SSyntS as such (then groupings are needed) or rather in the Syntactic-Com-
municative or Referential Structure (and then groupings in the SSyntS are avoided). For the time
being, I prefer the second solution; if it is adopted, either we have to introduce a special Sem-Com-
municative opposition, which will represent the order of (sub)set inclusions, orand this seems
more naturalwe have to use different referential indications (S. Kahane): in the first, but not the
second, reading, the meaning (Japanese cars) has a direct referent: the set of (all) Japanese cars is
characterized by the property of being expensive; the situation is inverse in the second reading: the
set of (all) expensive cars is characterized as being Japanese.
 95

The second phenomenon is modification of the type [a] typical American woman *an Ame-
rican typical woman or [a] former German minister *a German former minister. These cases re-
semble the syntactic behavior of quantifiers, which also deserves a special study from the view-
point of D -representation. However, it is possible to deal with the adjectives of the TYPICAL or
FORMER type on the basis of their special lexicographic properties: they can be linked to their
Synt-governor by the same SSyntRel as any other adjective, but their positioning is controlled by
their lexicographic features. Therefore, in this case, groupings in the SSyntS are not necessary.

6. Syntactic Dependency in Computational Linguistics

As pointed out above, the D-approach was born more or less out of the necessity to describe
languages for computer processing, especially so in the domain of Machine Translation. (In my
personal case, I had to reinvent dependencies in 1956-57since Tesnire was then not yet known
in the Soviet Unionwhile working on French-to-Russian and Hungarian-to-Russian Machine
Translation systems. Using dependencies was the only feasible approach to formal syntactic des-
cription of languages with 'free' word order, such as Russian and Hungarian.) Therefore it would
be important to discuss here the use of Synt-Ds in computer applications of modern linguistics, in
the first place for synthesis and analysis of sentences, as well as studies of formal properties of
syntactic D-representations. Unfortunately, this is quite a special topic, which would require much
additional competence, research and spaceand I am not in a position to implicate myself to the
necessary degree. The reader will have to be satisfied with a few references to the work where the
respective problems are covered: Kunze (ed.) 1982, Schubert 1987, Maxwell/Schubert (eds) 1988,
Apresjan et al 1989, 1992, Covington 1990, Fraser/Hudson 1992 (with further bibliography),
Badia 1993, Rambow/Joshi 1997, and Lombardo/Lesmo 1998.

Acknowledgments
This article was written in May-July 1998 during my stay in Germany under a Humboldt
Foundation scholarship and reworked several times since. The final version of it has profited a
great deal from my joint work with L. Iordanskaja (Iordanskaja & Mel'uk 2000): I have included
in the present article several of our common results and findings. The text has been read and com-
mented upon by D. Beck, I. Bolshakov, N. Brker, P. Hellwig, R. Hudson, L. Iomdin, L. Ior-
danskaja, S. Kahane, H. Lobin, J. Milievi, E. Savvina, L. Wanner, and D. Weiss; N. Pertsov
has caught several mistakes and incongruities at the very last moment. The remarks and objections
of D. Beck and S. Kahane have been especially deep and constructive and have lead to many
substantial changes in the text. I did my best to take into account all the suggestions of my friends
and colleagues, and I am happy to extend my heartfelt gratitude to all of them, as well as to the
Humboldt Foundation; the usual disclaimers apply of course.
 96

Notes

1
(I, 1, p. 00) The formulations that follow are not really rigorous definitions, but rather
approximate characterizations, which are hopefully sufficient for the purposes of the present paper.

2
(I, 1, p. 00) In case of compounding or incorporation, a wordform may represent two or more
lexemes. This complication is, however, irrelevant for our purposes here.

3
(I, 1, p. 00) I allow myself, stretching the terminology a bit, to use the term phrase for the
structural representation of an actual phrase; thus I will speak of 'the ADJ + N phrase,' meaning the
set of phrases like intelligent child, expensive houses, former minister, blue sky, etc. This is
simply a convenient abbreviation.

4
(I, 1, p. 00) Note that the notion of passive Synt(actic)-valency cannot be reduced to that of part
of speech. First, passive Synt-valency characterizes not only lexemes, but phrases as well, to
which I think the notion of part of speech is not applicable. Second, and more importantly, passive
Synt-valency of a lexeme L is determined, generally speaking, by the part of speech of L only
partially: syntactic features of L play here a crucial role. That is, two lexemes of the same part of
speech may have different passive Synt-valencies because of their syntactic features (
subcategorization). Thus, nouns like MONTH, WEEK or DAY may appear in the duration
construction with a verb (work the whole month, travel day after day, etc.), in which other nouns
are impossible; this fact is expressed by the syntactic feature temp assigned to such nouns. For
more on syntactic features, see Mel'uk/Pertsov 1987: 471ff.

5
(I, 2, p. 00) Some further DSyntRels might be needed: e.g., the qualificative attributive
DSyntRel as opposed to the restrictive attributive DSyntRel; or a special DSyntRel for Direct
Speech. However, this problem cannot be dealt with here.

6
(I, 2, p. 00) The other structure of the DMorphR of a sentence is the Deep-Morphological
Prosodic Structure, which specifies the pauses, i.e. phonological phrases, as well as intonation
contours, phrase and sentence stresses, etc. It is here that what are known as constituents in the
strict sense of the term first appear. Cf. 4.1, p. 00ff.

7
(II, 1, p. 00) In point of fact, Sem-D holds between lexical MEANINGS (of wordforms), i.e.
between semantemes in the Semantic Structure rather than between actual wordforms in an actual
 97

sentence. However, I have allowed myself this abus du langage in order to be able to compare
different Ds between wordforms, doing this in a parallel fashion.

8
(II, 2.2, p. 00) The English verb [to] ORDER is such that if its meaning takes as the 2nd Sem-
argument the meaning (go), which in turn has the 1st Sem-argument (he), then the meaning ([to]
order) has to take (he) as the 3rd Sem-argument (such verbs are known as 'subject-to-object raising
verbs'). This shows the transitivity of the Sem-D under consideration.

9
(II, 3.2, p. 00) According to Definition 2, Morph-D means the imposition of a grammeme. A
wordform cannot impose a grammeme upon itself, but it can, by some of its properties, condition
the choice between several grammemes imposed on it by a different wordform.

10
(II, 3.2, p. 00) The Russian verb ZNAT ([to] know) is such that if it has a Morph-dependent
w2 which has a Morph-dependent w3 of its own, thenunder specific syntactic conditions (w2 is
a DirO, etc.)w 3 is a Morph-dependent of ZNAT as well. This shows the transitivity of the
Morph-D under consideration.
11
(II. 3.3, p. 00) For a different analysis of the corresponding notions (and a rich bibliography),
see Schmidt/ Lehfeldt 1995.Recall that agreement and government have been treated for a long
time as types of SYNTACTIC dependency, which created confusion.

12
(II. 3.3, p. 00) Substitute, or anaphoric, pronouns are pronouns of the type HE, SHE, IT,
THEY, and all the relative pronouns, which replace nouns: a substitute pronoun is always used
instead of a noun, so it is really a PRO -noun. Substitute pronouns must be distinguished from
personal pronouns of the type I, YOU, WE, which never replace a noun.

13
(II. 4.2, p. 00) Lexical means used in syntactic capacity, i.e. what is known as 'structural,' or
'empty,' words, complicate the picture without affecting the essence of my reasoning: they do not
appear in the DSyntS, but they are present in the SSyntSsince they are separate wordforms, and
the SSyntS is supposed to represent all the wordforms actually found in the sentence. To keep my
formulations as simple as possible I leave the lexical means used in a syntactic capacity out of the
discussion.

14
(II, 4.3.1, p. 00) Here is a more complex case (brought to my attention by N. Pertsov): Rus. k
domu [= w 1], cvet kryi kotorogo menja razdraaet [= w 2] (to [the] house [the] color of [the]
roof of which irritates me), where cvet kryi kotorogo razdraaet is W. The wordforms w1, w2
and W constitute a phrase: domu, cvet kryi kotorogo razdraaet, whose Synt-head is domu; w2
 98

and W also constitute a phrase: cvet kryi kotorogo razdraaet, whose Synt-head is razdraaet;
therefore, domu and razdraaet are directly linked by a Synt-dependency:
domusyntrazdraaet.

15
(II, 4.3.2, p. 00) But in I saw the Pope John-Paul the Synt-D s are different: since I saw the
Pope is perfectly grammatical, we have the PopesyntJohn-Paul.

16
(II, 4.3.2, p. 00) Here are two more examples (for a detailed analysis of the construction in
question, further examples and a bibliography, see Gaatone 1988):
un vache de garon (an impressive boy)
a.SG.MASC impressive.SG.MASC of boy[MASC].SG
une vache de voiture (an impressive car)
a.SG.FEM impressive.SG.FEM of car[FEM].SG
ce chouette de garon (this nice boy)
this.SG.MASC nice.SG.MASC of boy[MASC].SG
cette chouette de voiture (this nice car)
a.SG.FEM nice.SG.FEM of car[FEM].SG
Some French expletive interjections can also appear, along with the above-mentioned adjec-
tives, as the SSynt-head of this construction: cette nom d'un chien de machine (this darned ma-
chine), cette bon sang de Julie (this bloody J.), ces sacr nom de Danois (these bloody Danes),
ma nom de Dieu de parole d'honneur (my damned word of honor), ce putain de garon (this
bloody boy), etc. The construction has the SSyntS of the following form:
CE synt[NOM D'UN CHIEN]syntDEsyntMACHINE;
the determiner agrees in gender and number with the noun rather than with its own SSynt-governor
the head of the phrase, which, unlike an adjective, cannot 'borrow' the gender and number from
the noun. A similar English construction (a bitch of a problem , 'Ulysses' is murder to read, etc.) is
analyzed in McCawley 1987.
Let it be emphasized that the construction illustrated in (15) is different from such construc-
tions as ce cochon de Polytte (this swine of P.), l'imbcile de ton mari (the fool of your husband)
or ce fou de prof, lit. (this crazy of professor), where the head is a noun (it can be a nominalized
adjective, but it is anyway a noun). In (15), the head adjective cannot be nominalized: *un drle,
*un chouette, etc.

17
(II, 4.3.2, p. 00) Otherwise, numerals do not create problems. Thus, in Russian, in estdesjat
tri (63) the Synt-head is tri, because in compound numerals the last (= rightmost) numeral is the
morphological contact point: estdesjattri stol+a, but estdesjatpjat (65) stol+ov and
 99

estdesjatodin (61) stol+. This means that Criterion B2 applies here and indicates the Synt-
governor.

18
(II, 4.3.3, p. 00) A SSyntRel is by no means a meaning; but a signified is not necessarily a
meaningit can even be a command to perform some modifications in the syntactics of a sign (as
is the case with voices). I cannot, however, enter here in the discussion of the types of linguistic
signifieds.

19
(II, 4.3.3, p. 00) A violation of semantic/lexical constraints is not considered as syntactic ill-
formedness. Thus, cf. inside the car, but *inside Stuttgart or according to Leo, but *according to
the car; however, the starred phrases are considered as syntactically well-formed (PREP + N being
a legitimate phrase of English).

20 (II, 4.3.3, p. 00) A property similar to the Kunze property was used for the identification of
SSyntRels in the METATAXIS system (see Schubert 1989: 10: "Interchangeable dependents are
grouped in classes and the relations that are definitional for these classes are given names").

21
(II, 4.3.3, p. 00) The SSynt-Subject of impersonal verbs (PLEUVOIR ([to] rain), NEIGER
([to] snow), etc.)the 'impersonal' ILis considered as a particular case of noun (= a pronominal
noun, which is not a substitute pronoun). Note that with the Kunze property, SSynt-Subjects in Il
[= Alain, i.e. a substitute pronoun] dort (He is sleeping) and Il pleut (It is raining) must be
described by two different SSyntRels.

22
(II, 4.3.3, p. 00) It is sometimes claimed that even actantial SSyntRels can be repeatable. The
best-known example is the repeatability of the dir-obj SSyntRel in Kinyarwanda: it is said that in
this language, a clause can have up to three DirOs (Kimenyi 1980: 229); cf.:
(i) Umo +gre +r +ubak+iish +iriz +a b+ana umu+gabo inzu
Class I woman I PRES build CAUS BENEF CONT II children I man house
(The woman, on behalf of the children, is making the man build the house).
A detailed analysis of 'repeated DirOs' in Kinyarwanda in Gary & Keenan 1977: 87-94 shows that
indeed all of them possess the same relevant linguistic properties, which set them off with respect
to oblique objects: they passivize, reflexivize and relativize, they can be cross-referenced in the
verb, etc. And yet, in our framework, all three of them cannot be considered DirOs, because they
contrast semantically, that is, they violate our Criterion C1. The presumed dir-obj SSyntRel in
Kinyarwanda has to be split into three different SSyntRels, which are, so to speak, the subtypes of
an abstract SSyntRel: the dir-obj SSyntRel, the caus-dir-obj SSyntRel and the benef-dir-obj
SSyntRel. In this way, the commonality of their important properties is explicitly shown.
 100

Similarly, in Sanskrit, two objects in the accusative (= 'double accusatives') cannot be both
DirOs, either:
(ii) T yajamnam > vc+aya +ti
them-ACC sacrificer-ACC name CAUS IND.PRES.3SG
([He] makes [the] sacrificer name them).
T is a DirO, but yajamnam> must be described by a different SSyntRel: the caus-dir-obj one.
In other languages (e.g., Latin, German and Serbo-Croatian) the situation with 'double
accusatives' is even clearer:
(iii) a. Lat. Quis music+am [ACC]?docuit?Epaminond+am [ACC]?,
lit.(Who taught Epaminondas music?) = (Who taught music to Epaminondas?)
or
Me [ACC]?rogavit?sententi+am [ACC],
lit. ([He] asked me opinion) = (He asked me for my opinion).

b. Serb.-Cr. Ta slika m+e [ACC]?kota?hiljad+u [ACC] maraka

(The painting costs me one thousand mark).

c. Germ. Was [ACC]?fragst [du]?mich [ACC]? (What are you asking me?)
All these sentences do not have two DirOs: the two accusatives do not display the same
syntactic behavior. Thus, in (iii-b) me is omissible, while hiljadu is not: Ta slika kota hiljadu
maraka vs. *Ta slika me kota; this shows that me is here an IndirO, despite its accusative form. In
(iii-c), only mich is the DirO, was being an OblO; etc.
For more on multiple objects in Latin, Ancient Greek and Modern Hebrew, see Lazard 1994:
89-96.
An interesting case of double accusatives is found in Korean (O'Grady 1991):
(iv) Kay +ka John+ul son +ul mwul+ess +ta
Dog NOM ACC hand ACC bite PAST DECLAR
(The dog bit Johns hand).
There can even be multiple double accusatives:
(v) John+i mwune+lul tali+lul kkuth pwupwun+ul cokum+ul cal+lass+ta
NOM octopus ACC leg ACC end part ACC bit ACC cut PAST DECL
(John cut the octopus on the end part of the leg a bit).
But only the first one in such a chain of accusatives is a DirO; all the others behave like adver-
bials (OGrady 1991: 74-75, 77-78): they 1) cannot accept modifiers and 2) cannot be permuted
with the DirO, cf. (iv) vs. *Kay ka son +ul John+ul mwulessta.
 101

23
(II, 4.3.3, p. 00) Along with Criteria C1-C3, the researcher can use the following heuristic test
in order to establish the type of a SSyntRel:
Coordinability with one SSynt-governor
Within a coordinated phrase D1coordD2 which is subordinated as a whole to a SSynt-gover-
nor G, each element must in principle bear the same SSyntRel r to G:
if GrD1coordD2, then GrD1 and GrD2.

Examples
(i) French
a. Il craint d'tre dcouvert et que l'administration le punisse, lit . (He fears to be
discovered and that the administration punish him).
b. Il veut partir et aussi que je parte avec lui, lit. (He wants to leave and that I leave with
him).
c. le rendement augmente successivement et par degr, lit. (The yield rises successively
and by degrees).
In (i), the boldfaced phrases stand in the same SSyntRel to the Main Verb.
Unfortunately, this test cannot be raised to the rank of a genuine formal criterion: coordi-
nationat any rate, in many languagesis strongly semantically motivated; therefore, in some
cases, syntactically different clause elements can be coordinated, while in some other cases identical
clause elements cannot. Here are a few examples.

Coordination of different clause elements (cf. Grevisse 1993: 371):

French
(ii) a. Elle vieillissait dans l'aisance et entoure de considration
(She was aging in well-being and surrounded with consideration).
b. augmentation successive et par degr, lit. (a rise successive and by degree).
c. Je me demande si et sous quelles conditions on pourra regler le problme
(I ask myself whether and under what conditions it will be possible to solve the problem).
(iii) (Defrancq 1998: 118-119)
a. Je me demande qui travaille et o (I ask myself who is-working and where).
b. Je me demande qui dcide et quoi (I ask myself who decides and what).
(iv) couper les cheveux trs court et de faon ce qu'ils ne lui tombent pas sur le
front
Other examples can be drawn from Russian:
(v) a. to otkrytie bylo sdelano v Anglii i anglianinom,
 102

lit. (This discovery was made in England and by an Englishman).

b. Ja govorju s potom i o pote, lit. (I talk with a poet and about a poet).
c. Nikto, nikomu i nikogda ne pomogaet, lit. (Nobody, to nobody and never helps).
(Russian coordinate constructions of this 'exotic' type are described in detail in Sannikov 1989: 14-
20.)
Impossible coordination of identical clause elements:
(v) French
a. *Ils taient cinq et trs blonds, lit. (They were five and very blond).
b. *des plats franais et exquis (French and exquisit dishes).
c. *Tout le monde prfre le repos maintenant et partir plus tard (Everybody prefers
the rest now and to leave later).
Consequently, the result of coordination test can serve as an argument in favor of or
against a particular solution (especially in less obvious cases); but the test as such cannot be
accepted as a rigorous criterion. Cf. the discussion of the role coordination plays in establishing
grammatical relations in Sag et al. 1985 (I am neither an authority on this subject nor trying to
portray myself as one, Pat was awarded the prize and very upset about it, and the like) and
Hudson 1988.

24
(II, 4.5, p. 00) A particular syntactic or communicative role may require a noun in a particular
inflectional form, for instance, (DEF(inite)) or (INDEF(inite)); thus, in French, the boldfaced quasi-
subject in the construction Il est venu 10 tudiants, lit. (It has come 10 students), may be only
indefinite. D. Beck pointed out to me another interesting example: in Lushootseed, the negative
predicate xwi! ([to] be not) requires its actant to be in the subjunctive and have the hypothetical
determiner kwi:
xwi! kwi gw +ad +s +!ed
be.not DET SUBJ 2SG NOM(inalizer) eat
lit. (Is-not your eating) = (You did not eat).
Yet, I think, in all such cases the Synt-governor DIRECTLY requires a particular form of its depen-
dentrather than the presence of a particular dependent of its dependent.

25
(II, 4.6, p. 00) This view was held, at least in Europe, as early as in the 13th-14th centuries.
Weber 1992: 13 speaks of Siger von Kortrijk, who preached the absolute dominance of the finite
verb in a sentence around 1300; cf. the following remark by Nicol Macchiavelli in 1516:
'...dicono che chi considera bene le 8 parti de l'orazione...troverr che quella che si chiama verbo
la catena e il nervo de la lingua;' quoted in Koch/Krefeld 1991, V. For objections to the status of
the Main Verb as the Synt-head of the sentence, see Hewson 1992: 49-51; these objections are
 103

(again and again) due to the confusion of different types of D (syntactic, morphological, and
communicative).

26
(II, 4.6, p. 00) In Turkish we cannot postulate a zero copula form in the present based on
paradigmatic considerations, as we have done for Russian. The main reason is that the forms in
(26a) contain the marker of predicativity that precludes the use of the copula: in the past tense of the
indicative, both the expression with the marker of predicativity but without copula and the
expression with the copula I(-mek) ([to] be) but without a marker of predicativity are possible (the
latter being typical of colloquial speech, while the former is current in the written language):
ocuk+tu +m and ocuk i+di +m
kid PAST 1SG kid be PAST 1SG
both meaning (I was a kid). Note that the verb I(-mek) has no present tense.

27
(II, 4.7, p. 00) In modern linguistic literature, the terms arguments or terms (vs. non-
arguments/non-terms) are also current for the corresponding concept. I prefer avoiding them in
linguistics, to reserve their use for logic: arguments/terms of a predicate.

28
(II, 4.7, p. 00) Languages also differ with respect to the meanings they allow to be coordinated.
Cf. numerous examples of coordinate phrases in Latin which should be translated with subordinate
phrases in French (Tesnire 1959: 315-316): Lat. orare atque obsecrare ([to] pray and-also implore)
~ Fr. prier instamment ([to] pray insistently), Lat. interdicit atque imperat (He forbids and-also
orders) ~ Fr. Il dfend expressment (He forbids expressly), Lat. diuellere ac distrahere ([to]
separate and-also tear-apart) ~ Fr. sparer violamment ([to] separate violently), Lat. doctrina et ratio
(teaching and method) ~ Fr. un enseignement mthodique (a methodical teaching), Lat. studium et
aures (favor and ears) ~ Fr. une oreille favorable (a favorable ear), etc. However, the study of the
relationship between coordination and subordination falls outside the scope of this paper.
29
(II, 4.8, p. 00) For an argumentation in favor of the SSyntS ConjMV see Hudson 1987: 119-
121.

30
(III, 1.1.3, p. 00) One of its disadvantages is immediately clear: it presupposes the repeatability
of actantial dependencies, which contradicts the postulate of unicity of each actant, widely shared
by linguists of all schools of thought.

31
(III, 1.3, p. 00) Tesnire distinguishes jonction, which is coordination, from connexion,
which is subordination. In Tesnire's trees (= stemmas) conjoined elements are linked horizontally,
showing in this way their equal nature. Each of them is then subordinated to the same Synt-gover-
 104

nor. Among other examples of how the coordination is represented, one finds in Tesnire 1959:
345 the following complex structure with parallel Synt-D s between the elements of coordinate
phrases which themselves are not linked by Synt-Ds:
Les matres, les pdagogues et les ducateurs donnent, rptent et ressassent des avis, des conseils
et des avertissements aux coliers, aux collgiens et aux lycens
(Teachers, pedagogues and educators give, repeat and trot out opinions, pieces of advice and
warnings to the school kids, college students and high-school students).
The structure proposed by Tesnire is as follows (only a part of it is represented here):
DONNER RPTER RESSASSER

MATRES PDAGOGUES DUCATEURS

32
(III, 1.3, p. 00) The solution Schubert himself prefers is to take as the head of a conjoined
phrase the coordinate conjunction: LEOANDALAN. But this solution is unacceptable for me;
see 2.6.

33
(III, 2.7, p. 00) Note that, for instance, in English the situation is different, because of the
impossibility of *I like her more than he:
I like her more thanconjunct[he] does vs. I like her more thanconjuncthim
For this case, we do not need special conjunctional SSyntRels.

34
(III, 2.8, p. 00) Alternatively, the elision process could be relegated to a 'later' stage, that is, to
the SSyntS DMorphS transition; then no artificial nodes in the SSyntS would be required for the
representation of such ellipses. For the time being, I do not see any serious objections to such a
strategy. The only reason for which I keep ellipsis in the SSyntS of a sentence is the intuitive
feeling that a complete sentence and a sentence with ellipsis are different SYNTACTICALLY and,
therefore, this difference should be reflected on a SSynt-level. In any event, I consider this an open
question.

35
(III, 5, p. 00) We take the negative particle NOT in this example to be a Synt-dependent of
TALL rather than of BE; cf. He is, as everybody knows since the period when ..., not tall and
fat vs. He isnot isn't, as everybody knows since the period when ..., tall and fat .

36
(III, 5, p. 00) Three remarks concerning relevant aspects of groupings seem in order.
 105

Dependence on the head of a coordinate string vs. dependence on the whole string
These two cases of dependency are distinguished in the proposed SSyntS with groupings by
including into a grouping all 'private' (= disjoint) dependents of its head: thus, for
(old {{fat men} and women}),
where (old) bears on the whole conjoined string (joint reading), but (fat) on (men) only (disjoint
reading), we write
modif

old [fatmenandwomen].
Dependence of the head of a coordinate string vs. dependence of the whole string
Here again, groupings allow for efficient disambiguation. Let us consider the following
French example (Abeill 1997a: 19): Paul rvait d'acheter et collectionner des pistolets an-
glais (P. was dreaming of buying and collecting English pistols). The boldfaced conjoined string
of infinitives depends on the verb rver ([to] dream) as a wholeit has a shared DirO pistolets
anglais (English pistols); therefore, the preposition DE that introduces the infinitive need not to be
repeated (joint reading). However, if the two conjoined infinitives do not depend on rver as a
whole, the preposition has to be repeated: Paul rvait de voyager et de collectionner des
pistolets anglais (P. was dreaming of traveling and collecting English pistols) *Paul rvait de
voyager et collectionner des pistolets anglais (disjoint reading). This difference is readily
expressed using groupings: for the joint reading, we write rvait [acheter etcollectionner
]des pistolets anglais, and for the disjoint one, rvaitvoyageretcollectionnerdes
pistolets anglais.

Multiple coordinate conjunctions depending on the same Governor

This is another problematic case for 'pure' dependency. Thus, consider the expressions (i) -
(ii), where both conjunctionsAND and ORsyntactically depend on the same noun (MEN, in
this case):
(i) {men and women} or children
vs.
(ii) {men or children} and women,
These expressions clearly have different meanings; however, in terms of pure dependency, both
have the same SSyntS:
 106

MEN

AND OR

WOMEN CHILDREN
We thus see that a pure-dependency SSyntS is unable to preserve the intended meaning in cases of
such a type. In order to distinguish (i) and (ii) in the SSyntS, we need groupingsand there is no
other way to achieve this goal:

(i') [MENANDWOMEN] ORCHILDREN

(ii') [MENORCHILDREN] ANDWOMEN

References
Anderson, John (1977): On Case Grammar. Prolegomena to a Theory of Grammatical Relations.
London.
Anderson, John/Ewen, Colin (1987): Principles of Dependency Phonology. Cambridge: Cam-
bridge University Press.
Anderson, Stephen (1992): A-Morphous Morphology. Cambridge: Cambridge University Press.
Apresjan, Jurij/Boguslavskij, Igor/Iomdin, Leonid et al. (1989): Lingvistieskoe obespeenie
sistemy TAP-2. Moscow: Nauka.
Apresjan, Jurij/Boguslavskij, Igor/Iomdin, Leonid et al. (1992): Lingvisitieskij processor dlja
slonyx informacionnyx sistem. Moscow: Nauka.
rnason, Kristjn (1989): Dependency Phonology. Linguistics, 27, 319-339.
Badia, Toni (1993): Dependency and Machine Translation. In: Eynde, Frank van (ed.), Linguistic
Issues in Machine Translation, London/New York: Pinter, 105-137.
Baumgrtner, Klaus (1965): Spracherklrung mit den Mitteln der Abhngigkeitsstruktur. Beitrge
zur Sprachkunde und Informationsverarbeitung, 5, 31-53.
Baumgrtner, Klaus (1970): Konstituenz und Dependenz. Zur Integration der beiden grammati-
schen Prinzipien. In: H. Steger (Hg.), Vorschlge zu einer strukturellen Grammatik des
Deutschen, Darmstadt, 57-77.
Bazell, Charles (1949): Syntactic Relations and Linguistic Typology. Cahiers F. de Sasussure, 8:
5-
 107

Beck, David (1997): Theme, Rheme, and Communicative Structure in Lushootseed and Bella Coo-
la. In: L. Wanner (ed.), Recent Trends in Meaning-Text Theory, Amsterdam/Philadelphia,
93-135.
Beck, David (1998): Adjectives and the Organization of Lexical Inventories. Toronto Working Pa-
pers in Linguistics, 16: 2.
Bloomfield, Leonard (1933): Language. New York: Holt, Rinehart and Winston.
Boguslavskaja, Olga (1991): Razdvoenie grammatieskix kategorij isla i soglasovatelnogo klassa
i aspektnye konstrukcii v dagestanskix jazykax. In: Tipologija grammatieskix kategorij,
Leningrad: Nauka, 10-12.
Boguslavskij, Igor (1985): Issledovanija po sintaksieskoj semantike. Moskva: Nauka.
Boguslavskij, Igor (1996): Sfera dejstvija leksieskix edinic. Moskva: kola Jazyki russkoj
kultury.
Bresnan, Joan (ed.) (1982): The Mental Representation of Grammatical Relations.Cambridge,
MALondon: The MIT Press.
Corbett, Greville (1993): The Head of Russian Numeral Expressions. In: Corbett et al. (eds) 1993,
11-35.
Corbett, Greville/Fraser, Norman/McGlashan, Scott (eds) (1993): Heads in Grammatical Theory.
Cambridge.
Covington, Michael (1990): A Dependency Parser for Variable-word-order Languages. Compu-
tational Linguistics, 16: 4, 234-236.
Dahl, sten (1980): Some Arguments for Higher Nodes in Syntax: A Reply to Hudson's 'Consti-
tuency and Dependency.' Linguistics, 18: 5/6, 485-488.
Dryer, Matthew (1989): Plural Words. Linguistics, 27, 865-895.
Durand, Jacques (ed.) (1986): Dependency and Non-linear Phonology. LondonDover, N.H.:
Croom Helm.
Eichinger, Ludwig/Eroms, Hans-Werner (eds) (1995): Dependenz und Valenz. Hamburg.
Engel, Ulrich (1977): Syntax der deutschen Gegenwartssprache. Berlin: Erich Schmidt. 307 pp.
[3rd rev. ed.: 1994.]
Engel, Ulrich (1988): Deutsche Grammatik. Heidelberg: Julius Groos/Tokyo: Sansyusya. 888 pp.
Evans, Nicholas (1988): Odd Topic Marking in Kayardild. In: P. Austin (ed.), Complex Sentence
Construction in Australian Languages, Amsterdam/Philadelphia: Benjamins, 219-266.
Fillmore, Charles (1968): The Case for Case. In: E. Bach/R. Harms (eds), Universals in Linguistic
Theory, New York: Holt, Rinehart, Winston, 1-88.
Fitialov, Sergej (1962): O modelirovanii sintaksisa v strukturnoj lingvistike. In: S. aumjan (ed.),
Problemy strukturnoj lingvistiki, Moskva, 100-114.
 108

Fitialov, Sergej (1968): Ob kvivalentnosti grammatik NS i grammatik zavisimostej. In: S. au-

mjan (ed.), Problemy strukturnoj lingvistiki-1967, Moskva, 71-102.
Fraser, Norman/Hudson, Richard (1992): Inheritance in Word Grammar. Computational Linguis-
tics, 18: 2, 133-158.
Gaatone, David (1988): Cette coquine de construction: remarques sur trois structures affectives du
franais. Travaux de linguistique, 17: 159-176.
Gaifman, Haim (1965): Dependency Systems and Phrase Structure Systems. Information and
Control, 8, 304-337.
Garde, Paul (1977): Ordre linaire et dpendance syntaxique : contribution une typologie. Bulletin
de la Socit de linguistique de Paris, 72: 1, 1-19.
Gary, Judit/Keenan, Edward (1977): On Collapsing Grammatical Relations in Universal Grammar.
In: P. Cole/J. Sadock (eds), Grammatical Relations [= Syntax and Semantics, 8], New York
etc.: Academic Press, 83-120.
Gladkij, Aleksej (1966): Lekcii po matematieskoj lingvistike dlja studentov NGU. Novosibirsk.
[French translation: Leons de linguistique mathmatique, fasc. 1, 1970, Paris.]
Gladkij, Aleksej (1968): Ob opisanii sintaksieskoj struktury predloenija. Computational Linguis-
tics 7, Budapest, 21-44.
Goralkov, Alla (1973): On One Type of Dependency Grammar. In: W. Klein/A. von Stechow
(eds), Functional Generative Grammar in Prague, Kronberg: Skriptor, 64-81.
Hays, David (1960): Basic Principles and Technical Variations in Sentence Structure Determina-
tion. Santa Monica, CA. [Reprinted in: C. Cherry (ed.), Information Theory, 1961,
Washington, 367-374.]
Hays, David (1964a): Connectability Calculations, Syntactic Functions and Russian Syntax.
Mechanical Translation, 8, 32-51. [Reprinted in: D. Hays (ed.), Readings in Automatic
Language Processing, 1966, New York, 107-125.]
Hays, David (1964b): Dependency Theory: A Formalism and some Observations. Language, 40:
4, 511-525.
Helbig, Gerhard (1992): Probleme der Valenz- und Kasustheorie. Tbingen: Niemeyer. 194 pp.
Heringer, Hans Jrgen (1970): Einige Ergebnisse und Probleme der Dependenzgrammatik. Der
Deutschunterricht, 4, 42-98.
Heringer, Hans Jrgen (1993a): [Dependency Theory] Basic Ideas and the Classic Model. In:
Jacobs et al. 1993: 298-315.
Heringer, Hans Jrgen (1993a): [Dependency Theory] Formalized Models. In: Jacobs et al. 1993:
315-328.
Heringer, Hans Jrgen (1996): Deutsche Syntax Dependentiell. Tbingen: Stafenburg. 292 SS.
Hewson, John (1991): Determiners as Heads. Cognitive Linguistics, 2: 4, 317-337.
 109

Hewson, John (1992): Review Article: R. Hudson, English Word Grammar and R. Langacker,
Concept, Image, and Symbol. The Cognitive Basis of Grammar. Canadian Journal of
Linguistics, 37: 1, 41-53.
Hudson, Richard (1976): Arguments for a Non-transformational Grammar. Chicago.
Hudson, Richard (1980a): Constituency and Dependency. Linguistics, 18: 3/4, 179-198.
Hudson, Richard (1980b): A Second Attack on Constituency: A Reply to Dahl. Linguistics, 18:
5/6, 489-504.
Hudson, Richard (1983): Word Grammar. In: Sh. Hattori/K. Inoue (eds), Proceedings of the
XIIIth International Congress of Linguistis, Aug. 29-Sept. 4, 1982, Tokyo, Tokyo, 89-101.
Hudson, Richard (1984): Word Grammar. Oxford.
Hudson, Richard (1987): Zwicky on Heads. Journal of Linguistics, 23, 109-132.
Hudson, Richard (1988a): Coordination and Grammatical Relations. Journal of Linguistics, 24,
303-342.
Hudson, Richard (1988b): Extraction and Grammatical Relations. Lingua, 76, 177-208.
Hudson, Richard (1990): English Word Grammar. Oxford: Basil Blackwell. 445 pp.
Hudson, Richard (1993a): Do We Have Heads in Our Minds? In: Corbett et al. (eds) (1993): 266-
291.
Hudson, Richard (1993b): Recent Developments in Dependency Theory. In: Jacobs et al. (eds)
(1993): 329-338.
Iordanskaja, Lidija (1963): O nekotoryx svojstvax pravilnoj sintaksieskoj struktury (na materiale
russkogo jazyka). Voprosy Jazykoznanija, n 4, 102-112.
Iordanskaja, Lidija (1967): Avtomatieskij sintaksieskij analiz. Tom II: Mesegmentnyj sintaksi-
eskij analiz. Novosibirsk: Nauka.
Iordanskaja, Lidija/Mel'uk, Igor (2000): Towards the Notion of Surface-Syntactic Relation.
Jacobs, Joachim/von Stechow, Arnim/Sternefeld, Wofgang/Vennemann, Theo (eds) (1993): Syn-
tax. An International Handbook of Contemporary Research. Vol. 1. BerlinNew York: W.
de Gruyter.
Kahane, Sylvain (1997): Bubble Trees and Syntactic Representations. In: T. Becker/H.-U. Krieger
(eds), Proceedings of the 5th Meeting of Mathematics of Language (MOL 5), Saarbrcken,
70-76.
Kahane, Sylvain/Mel'uk, Igor (1999): Synthse des phrases extraction en franais contem-
porain. t.a.l., 40: 2, 25-85.
Keenan, Edward (1974): The Functional Principle: Generalizing the Notion of 'Subject-of'.
CLS-10, 298-310.
Keenan, Edward (1978): Logical Semantics and Universal Grammar. Theoretical Linguistics, 5,
83-107.
 110

Kimenyi, Alexander (1980): Relational Grammar of Kinyarwanda. Berkeley: UCLA Press.

Kibrik, Aleksandr (1977): O sootnoenii ponjatija sintaksieskogo podinenija s ponjatijami sogla-
sovanija, upravlenija i primykanija. In: Problemy teoretieskoj i ksperimentalnoj lingvistiki,
Moskva: Izd-vo MGU, 161-179. [Reprinted in: A. Kibrik, Oerki po obim i prikladnym
voprosam jazykoznanija (universalnoe, tipovoe i specifieskoe v jazyke), Moskva: Izd-vo
MGU, 102-123.]
Kim, Alan (1995): Word Order at the Noun Phrase Level in Japanese: Quantifier Constructions and
Discourse Functions. In: P. Downing/M. Noonan (eds), Word Order in Discourse, Amster-
dam/Philadelphia: Benjamins, 199-246.
Koch, Peter/Krefeld, Thomas (eds) (1991): Connexiones Romanicae. Dependenz und Valenz in
romanischen Sprachen. Tbingen.
Korhonen, Jarmo (1977): Studien zu Dependenz, Valenz und Satzmodell. Teil I. Theorie und Pra-
xis der Beschreibung der deutschen Gegenwartssprache. Frankfurt am Main.
Kunze, Jrgen (1972): Die Komponenten der Darstellung syntaktischer Strukturen in einer Abhn-
gigkeitsgrammatik. Prague Bulletin of Mathematical Linguistics, 18, 15-27.
Kunze, Jrgen (1975): Abhngigkeitsgrammatik. Berlin: Akademie-Verlag.
Kunze, Jrgen/Priess, W. (1967-1971): Versuch eines objektivierten Grammatikmodells. Zeit-
schrift fr Phonetik, Sprachwissenschaft und Kommunikationsforschung, 20: 415-448, 21:
421-466, 23: 347-378, 24: 373-402.
Kunze, Jrgen (ed.) (1982): Automatische Analyse des Deutschen. Berlin: Akademie-Verlag.
Langacker, Ronald (1987): Foundations of Cognitive Grammar, vol. 1. Theoretical Prerequisites.
Stanford: Stanford University Press.
Langacker, Ronald (1991): Foundations of Cognitive Grammar, vol. 2. Descriptive Application.
Stanford: Stanford University Press.
Langacker, Ronald (1997): Constituency, Dependency, and Conceptual Grouping. Cognitive Lin-
guistics, 8: 1, 1-32.
Lazard, Gilbert (1994): L'actance. Paris: Les Presses Universitaires de France.
Lecerf, Yves (1960): Programme des conflits, modle des conflits. Traduction automatique, 1: 4,
11-18; 1: 5, 17-36.
Lehmann, Christian (1985): On Grammatical Relationality. Folia Linguistica, 19, 67-109.
Lobin, Henning (1993): Koordinationssyntax als prozedurales Phnomen. Tbingen.
Lobin, Henning (1995): Komplexe Elemente Indizien aus Nominalphrase und Verbalkomplex.
In: Eichinger/Eroms (eds) 1995, 117-133.
Lombardo, Vincenzo/Lesmo, Leonardo (1998): Formal Aspects and Parsing Issues of Dependency
Theory. In: COLING-ACL '98 (Proceedings of the Conference), vol. II, 787-793.
 111

McCawley, James (1987): A Case of Syntactic Mimicry. In: R. Dirven/V. Fried (eds.), Functional-
ism in Linguistics, Amsterdam/Philadelphia: Benjamins, 459-470.
Machov, Svatava (1975): Die Abhngigkeitsgrammatiken. Einfhrung in die generative Gram-
matik (Prager Autorengruppe), Kronberg: Skriptor, 146-154.
Marcus, Solomon (1965a): Sur la notion de projectivit. Zeitschrift fr mathematische Logik und
Grundlagen der Mathematik, 11, 181-192.
Marcus, Solomon (1965b): Sur une description axiomatique des liens syntaxiques. Zeitschrift fr
mathematische Logik und Grundlagen der Mathematik, 11, 291-296.
Matthews, Peter (1981): Syntax. Cambridge etc.: Cambridge University Press.
Maxwell, Dan/Schubert, Klaus (eds.) (1989): Metataxis in Practice. Dependency Syntax for Multi-
lingual Machine Translation. DordrechtProvidence: Foris.
Mel'uk, Igor (1962): Ob algoritme sintaksieskogo analiza jazykovyx tekstov (obie principy i
nekotorye itogi). Mainnyj perevod i prikladnaja lingvistika, 7, 45-87.
Mel'uk, Igor (1963): Avtomatieskij analiz tekstov (na materiale russkogo jazyka). In: Slavjan-
skoe jazykoznanie, Moskva: Nauka, 477-509.
Mel'uk, Igor (1964a): Tipy svjazej medu lementami teksta i tipologija jazykov. In: L.I. Rojzen-
zon (red.), Materialy konferencii Aktualnye voprosy sovremennogo jazykoznanija i ling-
vistieskoe nasledie E.D. Polivanova, tom I, Samarkand: Samarkandskij Un-tet, 57-59.
Mel'uk, Igor (1964b): Avtomatieskij sintaksieskij analiz. Tom I: Obie principy. Vnutriseg-
mentnyj sintaksieskij analiz. Novosibirsk: Nauka.
Mel'uk, Igor (1967): L'ordre des mots dans la synthse automatique du texte russe. T.A. In-
formations, n 2, 65-84.
Mel'uk, Igor (1974): Opyt teorii lingvistieskix modelej Smysl Tekst. Moskva: Nauka.
[Reprinted by kola Jazyki russkoj kultury in 1999.]
Mel'uk, Igor (1977): O tipax poverxnostno-sintaksieskix otnoenij (tri kriterija razlienija). Rus-
sian Linguistics, 3: 3-4, 245-270.
Mel'uk, Igor (1979): Studies in Dependency Syntax. Ann Arbor: Karoma.
Mel'uk, Igor (1981): Types de dpendance syntagmatique entre les mots-formes d'une phrase.
Bulletin de la Socit de linguistique de Paris, 76: 1, 1-59.
Mel'uk, Igor (1985): Poverxnostnyj sintaksis russkix islovyx vyraenij. Wien: WSA.
Mel'uk, Igor (1988): Dependency Syntax. Albany, NY: SUNY Press.
Mel'uk, Igor (1993): Agreement, Government, Congruence. Lingvistic Investigationes, 17, 307
-373.
Mel'uk, Igor (1995a): The Russian Language in the Meaning-Text Perspective. MoscowVien-
na: kola Jazyki russkoj kulturyWiener Slawistischer Almanach.
 112

Mel'uk, Igor (1995b): The Future of the Lexicon in Linguistic Description and the Explanatory
Combinatorial Dictionary. In: Ik-Hwan Lee (ed.), Linguistics in the Morning Calm 3 (Se-
lected Papers from SiICOL-1992), Seoul: Hanshin, 181-270.
Mel'uk, Igor (1996a): Lexical Functions: A Tool for the Description of Lexical Relations in the
Lexicon. In: L. Wanner (ed.), Lexical Functions in Lexicography and Natural Language Pro-
cessing, Amsterdam/Philadelphia: Benjamins, 37-102.
Mel'uk, Igor (1996b): Cours de morphologie gnrale. Vol. 3. Troisime partie : Moyens mor-
phologiques. Quatrime partie : Syntactiques morphologiques. Montral Paris: Les Presses
de l'Universit de Montral CNRS.
Mel'uk, Igor (1997a): Cas grammaticaux, construction verbale de base et voix en massa : vers
une meilleure analyse de concepts. Bulletin de la Socit de linguistique de Paris, 92: 1, 49-
113.
Mel'uk, Igor (1997b): Vers une linguistique Sens-Texte. Leon inaugurale. Paris: Collge de
France.
Mel'uk, Igor/Pertsov, Nikolaj (1987): Surface Syntax of English. A Formal Model within the
Meaning-Text Framework. Amsterdam: Benjamins.
Milievi, Jasmina (2000): Linear Placement of Serbian Clitics: A Description within a Dependen-
cy-Based Approach. In: Wanner (ed.), ...
Nichols, Johanna (1978): Double Dependency? CLS-14, 326-339.
Nichols, Johanna (1986): Head-marking and Dependent-marking Grammar. Language, 62: 1, 56-
119.
Nichols, Johanna (1993): Heads in Discourse: Structural Versus Functional Centricity. In: Corbett
et al. (eds) 1993, 164-185.
OGrady, William (1991): Categories and Case. The Sentence Structure of Korean. Amsterdam/
Philadelphia: Benjamins.
Owens, Jonathan (1988): The Foundations of Grammar: An Introduction to Mediaeval Arabic
Grammatical Theory. Amsterdam: Benjamins.
Padueva, Elena (1964): O sposobax predstavlenija sintaksieskoj struktury predloenija. Voprosy
Jazykoznanija, n 2, 99-113.
Perlmutter, David (ed.) (1983): Studies in Relational Grammar I. Chicago London: The Univer-
sity of Chicago Press.
Petkevi, Vladimr (1995): A New Formal Specification of Underlying Structures. Theoretical
Linguistics, 21: 1, 7-61.
Pittman, Richard (1948): Nuclear Structures in Linguistics. Language, 24: 287-202.
Quirk, Randolph/Greenbaum, Sidney/Leech, Geoffrey/Svartvik, Jan (1985): A Comprehensive
Grammar of English Language. London and New York.
 113

Rambow, Owen/Joshi, Aravind (1997): A Formal Look at Dependency Grammars and Phrase-
Structure Grammars, with Special Consideration of Word-Order Phenomena. In: L. Wanner
(ed.). Recent Trends in Meaning-Text Theory, Amsterdam/Philadelphia: Benjamins, 167-
190.
Robinson, Jane (1970a): A Dependency-based Transformational Grammar. In: Actes du X-me
Congrs international des linguistes (Bucarest, 1967), 2, Bucarest, 807-813.
Robinson, Jane (1970b): Dependency Structures and Transformational Rules. Language, 46: 2,
259-285.
Sannikov, Vladimir (1989): Russkie soinitelnye konstrukcii: semantika, pragmatika, sintaksis.
Moskva: Nauka.
Savvina, Elena (1976): Fragment modeli rusaskogo poverxnostnogo sintaksisa. III. Sravnitelnye
konstrukcii (sravnitelnye i sojuznye sintagmy). Nauno-texnieskaja informacija, serija 2, n
1, 38-43.
Schmidt, Peter/Lehfeldt, Werner (1995): Kongruez, Rektion, Adjunktion. Semantische und histor-
ische Untersuchungen zur allgemeinen Morphosyntax und zu den Wortfgungen (slovoso-
etanija) im Russischen. Mnchen: Sagner.
Schubert, Klaus (1987): Metataxis. Contrastive Dependency Syntax for Machine Translation.
DordrechtProvidence, RI: Foris. [Rev.: B. Sigurd, Studia Linguistica, 1988, 42: 2, 181-
184.]
Sgall, Petr/Nebesk, Ladislav/Goralkov, Alla/Hajiov, Eva (1969): A Functional Approach to
Syntax in Generative Description of Language. New York.
Sgall, Petr/Hajiov, Eva/Panevov, Jarmila (1986):The Meaning of the Sentence in its Semantic
and Pragmatic Aspects. Prague.
Sgall, Petr/Panevov, Jarmila (1988-89): Dependency SyntaxA Challenge. Theoretical Linguis-
tics, 15:1, 73-86.
Sgall, Petr/Pfeiffer, Oskar/Dressler, Wofgang/Puek, M. (1995): Experimental Research on Syste-
mic Ordering. Theoretical Linguistics, 21:2/3, 197-239.
Starosta, Stanley (1988): The Case for Lexicase. London.
Suer, Margarita (1998): Resumptive Restrictive Relatives: A Crosslinguistic Perspective. Lan-
guage, 74: 2, 335-364.
Tarvainen, Kalevi (1981): Einfhrung in die Dependenzgrammatik. Tbingen.
Tesnire, Lucien (1934): Comment construire une syntaxe. Bulletin de la Facult des lettres,
Universit de Strassbourg, 7, 219-229.
Tesnire, Lucien (1959): lments de syntaxe structurale. Paris: Klincksieck.
 114

Trubetzkoy, Nikolaj (1939): Le rapport entre le dtermin, le dterminant et le dfini. In: Mlanges
de linguistique offerts Charles Bally, 75-82. [Reprinted in: E. Hamp et al. (eds), Readings
in linguistics, II, 1966, 133-138.]
Van Langendonck, Willy (1994): Determiners as Heads? Cognitive Linguistics, 5: 3, 243-259.
Vennemann, Theo (1977): Konstituenz und Dependenz in einigen neueren Grammatiktheorien.
Sprachwissenschaft, 2: 3, 259-301.
Weber, Heinz (1992): Dependenzgrammatik. Ein Arbeitsbuch. Tbingen.
Xolodovi, Aleksandr (1966): K tipologii porjadka slov. Naunye doklady vysej kolyFilologi-
eskie nauki, n3. [Reprinted in: A.A. Xolodovi, Problemy grammatieskoj teorii, 1979,
Leningrad: Nauka, 255-268.]
Zwicky, Arnold (1985): Heads. Journal of Linguistics, 21: 1, 1-30.
Zwicky, Arnold (1993): Heads, Bases and Functors. In: Corbett et al. (eds) 1993: 292-315.
olkovskij, Aleksandr & Igor Meluk (1967): O semantieskom sinteze. Problemy kibernetiki,
19: 177-238. [There is a translation into French: T.A.Informations, 1970, n 2, 1-85.]