Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Svetoslav Marinov
sedalti99@yahoo.com
Introduction
The basic notions in DG are that of a head (or governor) and its dependent two words in a sentence, and the binary relation that holds between
them. Zwicky (1985) gives six properties which can differentiate the head
(H) from its dependent (D) in a given construction (C). These criteria have
a mixed character some are semantic and some are syntactic properties.
The following list is based on Hudson (1990, pp. 106-7), who argues that
heads in many different constructions are congruent with one or several of
those properties:
1. C refers to a hyponym of what H refers to. E.g., big book refers to a
kind of book.
2. The semantic relation between H and D involves different part of their
respective meanings: Hs sense but Ds referent. E.g., in picture of a
girl, the referent of D (a girl) defines the sense of H (picture).
1
see
in the park?
In the sections that follow we will concentrate on the following points, which
are central and ubiquitous to most of present-day DG frameworks as well as
in the field of syntactic dependency parsing.
Syntactic analysis in terms of word-to-word relations
Head/dependet or governor/governee dichotomy among the words
Projectivity/non-projectivity of analysis
Rootedness and single-headedness
Labelled word-to-word relations
The Grammar of P
an.ini (ca. 400-500 B.C.)
segments; Dh
atup
a.tha, describing about 2000 verbal roots and their morphological and syntactic properties; Gan.ap
a.tha, a list of 261 lexical items.
Some of the 4000 rules in the As..ta
dhy
ay part of Pan.inis grammar deal
with the sentence structure. In this view of syntax there are no movements, deletions or transformations, deep and surface structure. His system
is a derivational one and goes through a number of levels but these deal
with the correct morphological and phonological realization of elements as
well as the roles and relations between the elements. The derivation of
a sentece starts from the level of meaning and ends with its phonological
form (Itkonen, 1991). The lexicon contains verbal and nominal stems. The
derivation is instantiated by choosing items from the lexicon and deciding
on the relations between the verbal root and the nominals. These relations
or roles are called k
arakas and according to Kiparsky (2002) there are six of
them Agent (kartr), Goal (karman), Recipient (sam
ana), Instrument
. prad
(karan.a), Locative (adhikaran.a) and Source (ap
ad
ana). The well-formedness
of the sentence stems from the proper applications of the rules, so that each
nominal is assigned a role, discernible in some morphological element. In this
respect the k
arakas mediate between meaning and morphosyntactic structure in the following way (Kiparsky, 2002, p.16):
Every k
araka must be expressed by a morphological element
No k
araka can be expressed by more than one morphological element
Every morphological element must express something
The following Sanskrit example demonstrates the syntactic structure of a
sentence, where the morphological elements expressing the k
arakas are boxed
and subscripted with the respective role. Since the instantiation/expression
of the roles happens in relation to the verb, hence the links between them.
(2)
kr.s.n.ah. paca ti
Krishna cooks
AGEN T
odan am
rice
GOAL
On the other hand, Itkonen (1991) illustrates the relations between the elements in a sentence at the level of meaning as (3) and at the level of
form as (4). In the latter case we have substituted the nominals with their
case-marking. The arrows represent the semantic and syntactic determination/governance respectively.
(3)
(4)
It would be too far-fetched to induce any dependency-based syntactic analysis, in its current incarnation, from the grammar of Pan.ini. What is relevant for the present task of looking at dependency ideas in retrospect can
be summed up as follows:
There are binary relations between a verb root and nominals, as well
as between nominals (Itkonen, 1991; Kiparsky, 2002).
The action, as expressed by the aspect of the verb, is central (Itkonen,
1991).
There are six (semantic/syntactic) roles which describe the binary relations (Misra, 1966; Itkonen, 1991; Kiparsky, 2002).
There can be only one role per nominal but a verb may enter into
relation with zero to six nominals (Itkonen, 1991; Kiparsky, 2002).
From the above summary we find support for notions like binary relations, rootedness (the action/verb being the root), labelled relations (i.e.
k
arakas) and single-headedness. It is interesing to note that at the level of
semantic dependency in the Meaning-Text Theory (Melcuk, 1988), there
are six actants or relations. These are labelled with digits 1 to 6.
a.
b.
idem
homo lapsus
heu hodie concidit
the same man having slipped alas today fell down
homo concidit
man fell down
The logician Boethius (ca. 480-524/6 A.D.) is the first person who coins
a special term for the supportive role of the minor word-classes (Percival,
1990). His use of the term determinatio is generic and refers to the relation
of all minor classes with respect to the major one.
To sum up, many of the ideas discussed in the works of Ancient grammarians and logicians can be subsumed under the modern understanding of
dependency. These include rootedness (i.e. the prior of the noun and the
verb), head-modifier relations (e.g. the adverb-verb relation), analysis in
terms of words only as well as a term for the head-dependent relation (i.e.
determinacio).
DG type of syntactic analyses are found in the Arab Linguistic tradition with
its chief luminary the grammarian Sbawaihi (d. ca. 793) and his seminal
work known as Al-Kit
ab (=The Book) (Itkonen, 1991). The need to give
explanation for the nominal inflections (i.e., the existence of Nominative,
Accusative and Genitive case) brings about a basic cause-effect scheme (7)
(Itkonen, 1991, p. 132).
(7)
b) A
a) construction
A
cause
B
effect
Following Owens (1988), Itkonen calles A a governor of B and B a dependent of A. This scheme gives explanation to clear-cut cases where a verb is
governing a noun in Nominative or Accusative, or a preposition governing a
noun in Genitive (8) and ultimately covers cases where a non-overt element
governs syntactically another element (9).
(8)
V N-nom,
(9)
(=A)
V N-acc,
Prep N-gen
B-{case/mode}
d.arab-tu-hu
hit-I-he
I hit him
Zaydun rajulun
Topic Comment
Zayd is a man
rajulan
kana zaydun
was ZaydN OM manACC
Zayd was a man
While authors like Owens (1988) finds support only for dependency analyses
in the Arab grammarians, others (e.g. Carter (1973)) defend the idea that
the syntactic analyses of both nominal and verbal sentences as proposed
by Sbawaihi and his followers are in essence Bloomfieldian type Immediate
Constituent analyses. Itkonen (1991), on the other hand, takes the moderate
view of assuming that Sbawaihi and followers operated with the two notions
we now call dependency and constituency, all depending on the type of
structure at hand.
From dependency point of view what is problematic in the Arabic syntactic theory are constructions involving non-accusative complements of the
verb. These will be introduced by prepositions (13).
(13)
marar-tu bi zayd-in
passed-I by Zayd-Gen
I passed Zayd
Following Owens (1988) and Itkonen (1991) it can be said that the Arab
grammarians were the proponents of the modern definition of dependency.
They differentiated between c a
mil=head and mac m
ul=dependent. Singleheadedness (i.e. every element can have only one governor) and projectivity
(i.e. no crossing-arcs) were principles explicitly present in the analyses by
the grammarians. As for the proposing by Sbawaihi of a covert element
governing the nouns in an N+N construction, Itkonen (1991) considers this
to support a transformational grammar approach. However, Melcuk (1988)
similarly assumes an empty category/element to be the head in certain N+N
case (e.g. Ya vrac=I (am) a doctor (Russian)).
In the European Linguistic tradition the earliest examples of DG-like syntactic analyses can be attributed to the Modistae (ca. 1260-1310) (Covington,
1984). The modistic grammar (or grammatica speculativa) described how
linking words together can build up whole sentences. The formation of the
sentence is divided into three successive steps constructio, establishing
links between the words; congruitas, application of well-formedness conditions on the links; perfectio, a final check on whether this is a complete
sentence (Covington, 1984). Any word-word relation is also classified as being transitive or intransitive, depending on whether the two words pertain
to the same real-world referent. Within each construction there are two
grammatical relations of primum to secundum and of dependens to terminans. In the former relation secundum presupposes the presence of primum.
As for the letter, a dependens is an unsaturated element, while a terminans is the element which saturates (Lepschy, 1994). Covington (1984)
claims that the dependensterminans relation is an extention of Petrus Helias concept of regimen, which according to Law (2003) is the concept of
government, where one word forces another to be in a particular form. In
transitive constructions the two sets of relations run in the same direction
while in intransitive constructions in the opposite. These can be illustrated
by the following analysis of a Latin sentence in (14), (Covington, 1984),
which means A man very swift on his feet runs through the forest. The
dashes lines exemplify the depenedens-terminans relations, and the rest the
primum-secundum relations.
(14)
Between the middle of the 14th and the middle of the 20th century, there
does not seem to be any single grammatical framework that relates to DG,
as it is with the frameworks reviewed in the last few sections. The only
references we find in the literature are based on Kruijff (2002) and Percival
(1990).
10
Conclusion
In this paper we traced the roots of DG or rather those central notions which
are omnipresent in the different DG frameworks, as well as important for
syntactic dependency parsing. These were notions like: syntactic analysis
in terms of word-to-word relations; head-dependent dichotomy, rootedness,
(non-)projectivity, single-headedness and labelled word-relataions.
We showed that many of these ideas were present in the works of a
number of ancient grammarians. The three, most worthy of mentioning are
the Sanskrit grammar by Pan.ini, the Arabic grammar by Sbawaihi and the
works of the Modistae. Still, the interpretations of the ideas by the ancient
grammarians are not uniform. Some modern authors see sentence analyses
in terms of constituency, while others in terms of dependencies. We believe
that many of theories we just reviewed support the moderate view that the
notion of word-to-word relations in one form or another can be correlated
with the modern notion of dependency.
References
Auroux, Sylvain, E. F. K. Koerner, Hans-Josef Niederehe, and Kees Versteegh, ed. 2000. History of the Language Sciences, volume 1. Walter de
Gruyter.
11
12