An ontology of systematic relations for a shared grammar of Slavic 
Tania AVGUSTINOVA 
Computational Linguistics, Saarland University 
P. O. Box 151150 
Saarbriicken, Germany, D-66041 
avgustinova @coli.uui-sb.de 
Hans USZKOREIT 
Language Technology Lab, DFKI 
Computational Linguistics, Saarland Univc,'sity 
Saarbr(icken, Germany, D-66041 
uszkoreit@dfki.dc 
Abstract 
Sharing portions of grammars across s greatly re- 
duces the costs of nutltilingual grammar engineering. Related 
s share a ntuch wider range of linguistic itff'ornuttio;t 
than typically assunwd in stamlard mttltilingttal gramtmtr 
atwhitectures. Taking grammatical relatedness seriously, we 
are particularly interested in designing lhtguistically motivated 
grammatical resottrces Jbr Slavic s to be used itz 
applied and theoretical computational linguistics, ht order to 
gain the pelwpective of a -family oriented gramntar 
desigtl, we consider {ttt arrtly of systematic relations that can 
hold between syntactical units. While the categorisation of 
primitive linguistic entities tends to be -specilic or 
even constrttction-.~pecific, the relations holding between them 
allow viii'lofts degrees of absltztction. On the basis of Slavic 
&tta, we show how a domain ontology conceptualising molpho- 
syntaclic "buildiltg blocks" can serve as a basis r~" a shared 
grotlt;nar of Slavic. 
Introduction 
In applied computational linguistics, the need for 
developing and utilising operational notions of 
shared grammars stems fi'om multilingual grammar 
engineering. If considerable portions of existing 
grammars can be reused for the specification ot7 new 
grammars, development eftbrts can be greatly re- 
duced. A shared grammar also facilitates the diffi- 
cult task of lnaintaining consistency within and 
across the individual parallel grammars. In machine 
translation, the specification of a shared grammar 
can furthermore be exploited for simplifying the 
transfer process. 
Without much ado, computational linguists engaged 
in multilingual grammar development have always 
tried to reduce their labour by importing existing 
grammar components in a simple "copy-paste- 
modify" fashion. But there were also a number of 
systematic attempts to create and describe shared 
grammars that are convincingly documented in 
publications. \[Kam88\] demonstrates the concept for 
a relatively restricted domain, the grammatical 
description of simple nominal expressions in five 
s. \[BOP88\] were able to exploit the gram- 
matical overlap of two Slavic s, for the 
design of a lean transfer process in Russian to 
Czech machine translation. In multilingual applica- 
tion development within Microsoft research, gram- 
mar sharing has extensively been exploited - 
\[Pin96\], \[GLPR97\]. 
However, all these approaches are rather opportun- 
istic in the sense that existing grammatical descrip- 
tions based on existing grammar models were ex- 
plored. We went a step further and started grammar 
design with a notion of a shared grammar for a 
family of related s. Pursuing the goal of 
designing linguistically motivated grammatical 
resources for Slavic s to be used in com- 
putational linguistics, one is inevitably confronted 
with primary problems stemming ti'om the t'act that 
different linguistic theories cut up grammars in 
quite different ways, and grammar formalisms differ 
in their degree of granularity. It cannot be expected, 
therefore, that the minimal differences between two 
s or their shared elements form easily 
identifiable units in the available -specific 
grammars. Therefore, an ontology conceptualising 
morphosyntactic "building blocks" would offer a 
solid basis for a shared grammar of Slavic in the 
sense of \[ASU99\]. Our use of the term ontology is 
fairly pragmatic, namely, as representing a formal 
shared conceptualisation of a particular domain of 
interest. It describes concepts relevant for the do- 
main, their relationships, as well as "axioms" about 
these concepts and relationships. Note that such a 
pragmatic approach does not presuppose any gen- 
eral all-encompassing ontology of  but 
rather "mini-ontologies" conceptualising the se- 
lected domain from various perspectives in a con- 
sistent way. The domain of interest in this project is 
the grammatical knowledge on Slavic morphosyn- 
tax contained in linguistic theories and linguistic 
descriptions. While the categorisation of primitive 
linguistic entities lends to be -specific or 
even construction-specific, the relations holding 
between them allow various degrees of abstraction. 
In order to gain the perspective of -family 
oriented grammar design, we will consider the array 
of .s3,stematic relations that can hold between syn- 
tactically significant items. 
Systematic relations 
Systematic relations motivate shared patterns of 
variation cross-linguistically as well as across con- 
structions. In a constraint-based theory like HPSG, 
where the grammatical properties of linguistic enti- 
ties are typically revealed in complex taxonomies, 
nothing in the formal apparatus would actually 
exclude the possibility to organise also the relations 
holding in syntactic constructions in a type hierar- 
chy. So, the type subsumption could be interpreted 
as modelling a continuum from general - and pre- 
sumably universal - systematic relations to more 
and still more specific instances of these relations 
resulting fi'om admissible cross-classifications. I In 
I The two types of edges connecting types in otu g,'aphical 
representation of hierarchies - 'square' and 'direct' - are signifi- 
cant. The former indicate possible conjunction of types, and thus 
introduce various dimensions of multiple inhe,'itance. The latter 
indicate disjunction of types within the respective dimension of 
classification. 
98 
our view, two orthogonal types of systematic rela- 
tions have to be distinguished: syntagmatics and 
alignment, since they appear to be universally rele- 
vant for the well-for|nedlmss of utterances in any 
 (Hierarchy I). 2 Syntagmatic relations play 
a constitutive role in syntax by establishing instant 
COlmections between linguistic entities in various 
constructions. There is a covert, meaningful dime,> 
sion of structural syntagmatics, a,+d a,l ()Vel"t., ,llOl"- 
phosyntactic, form-oriented dimension of combi- 
natorial syntagmatics. 
systematic relation I 
J 
I intonation syntagmatics alignnlent 
i \[ ~ ! i i"" 
combinatorial structural continuity directionality periphery 
overt covert , 
nlorp ~osyn actc ureaningM i 
1 ,icrnrchy I: Systcnmlic r¢lntions: dimensions of classification 
With respect to the alignment relation, which is 
responsible for the actual lincat distributiot+ of syn- 
tactically relevant itch|s, we assun|e that at least the 
continuity o1' syntactic units, the directionality of 
the head (or, more gene,'ally, of a certain syntacti- 
cally significant entity) its well as the l)eril)hery of at 
syntactically determined domain a,'c relevant di- 
n|cnsions of classification (ltierarchy 2). 
alignmout 
! 
continuRy directionality periphery 
contitluous disconOnuous tlonX-X X-nonX lel~ @ht 
(nonhead-HEAD) (HEAD-tionheadl 
Hierarchy 2: Alignment 
The continuity of syntactic units can be realised its 
immediate constituency (i.e. of type continuous) or 
as long-distance constituency (i.e. o1' type discon- 
tinue++s). The directionality accounts for situations 
where, e.g., Ihe head either li)llows the dependent or 
precedes it. In turn, the i~eriphcry of a syntactically 
determined domain can be left or right. 
Structural syntagmatics 
The structural syutagn|atic relations between two 
syntactic units is classified along two primary di- 
mensions which we call centHcity and taxis 
(Hierarchy 3). The chosen terms should be under- 
{~ tr stood in the context of distin~.uishin~, on the one 
hand, endocentric imd exocentric relations, and on 
the other hand, h37mtaxix and parataxis. The endo- 
centricity of a strt,ctural syntagmatic relation 
(between, e.g., ~ and \[~) presuptmses that one of the 
syntactic items involved in this relation (e.g., c0 
plays a pron|inent role. In contrast, the exocenlricity 
of a structural syntag,natic relation presupl)oses no 
assun+|ptions in this respect, hence, it can be viewed 
as the unmarked member of the centricity opposi- 
tion. The hypotaxis means that there is a depend- 
2 The intotmlional ol+ganisation of utterances by the way is 
anolher systematic relalion exhibiting ,his trait. A more thor- 
ough invcsligation of ,he in,onatiomd aspect would bc well 
beyond the morphosyntactic orientation of the present work. 
cncy of subol'di,mtion between the involved syntac- 
tic items, while the paramxis is net|tral in this re- 
spect and is regarded as the untnarked member of 
the taxis opposition. 
syntagmatics 
structural 
centlicity 
i 
taxis 
hypofnxis parataxis en(Jocenttlc exocenltlc 
I lierarchy 3: Cenhicity and taxis 
Consequently, if two linguistic entities belong to- 
gether fi'om the viewpoint of structural syntagtnat- 
its, they are involved in one of the following rela- 
tion types which arc obtained via+ admissible cross- 
classifications (Hierarchy 4). 
l,;ndocc||tric hypotaxis, or selection. The head- 
dependent co|fl'igu|'ation can be identified unambi- 
guously. Tim prominent element is also the domi- 
nating one in the subordination. 
Exoccntric hypotaxis, or modification. There is no 
l)t'olnincnt clone|on+ to unambigttously lake o\,oi" the 
role of a dolninating item in the subordination. Note 
that this is where (theory-sinusitis) linguistic con- 
ventions regarding the head-dependct+t configura- 
tion actually begin. 
Endocentric parataxis. There is a prominent elc- 
|nent in this relation, but no head-dependent con- 
figuration. 
Exocentric parataxis. In the relation holding be- 
tween lhc ilwolved linguistic entities there is neither 
a prominent element nor a head-dependent configu- 
ration. This is the tmlnarked case with respect to 
both ccnlricity and taxis. 
syntagmatics 
structural 
centricity taxis 
et\]docetltfic exocentfic hypctaxis parataxis 
selection modification endo-para exo-para 
" r, 
i licrarchy 4: Structural syntugmatics 
Due to the fact that there always is a principal or 
leading element in tile endocenlric relations, differ- 
ent linguistic theories typically agree on how to 
intel'pret these relations structurally. But there is no 
consensus -- often even within the same linguistic 
theory - on the structural interpretation of the 
exocentric relations. So, additional factors are usu- 
ally taken into consideration as st, pporting the in- 
+,'eduction of particular conventions. The latter, 
however, arc not always linguistically motiwtted, 
29 
tile choice is sometimes arbitrary and often due to 
theory-sl~ecific technical reasons. 
Combinatorial syntagmatics 
The combinatorial dimension in tile proposed tax- 
onomy (Hierarchy 5) largely corresponds, in our 
understanding, to what \[S&L95\] regard as mor- 
phological signalling of direct syntactic relations. 
syntagmatics 
I 
combinatorial 
. - - -- . . 
juxtaposition co-variation government 
Hicrm'chy 5: Combinatorial synt~matics 
Tile combinatorial syntagmatic relation of juxtapo- 
sition presupposes no overt morphological indica- 
tion. As to governmettt, it is traditionally under- 
stood as the determination by one element of the 
inflectional form of the other, i.e. form government. 
Its classical instance is, of course, case government. 
In \[Cann93\] (p. 51) these morpllosyntactic relation 
is formulated for some construction involving y and 
(5 ill tile following way: y governs (5 if (i) varying the 
inflectional l'orm of (5 while keeping T constant leads 
to ungrammaticality, and (ii) varying the form oft 
and keeping 8 constant makes no difference to 
grammaticality. The systematic co-variatioll o1' 
linguistic forms is typically rcalised as feature con- 
gruity, i.e. compatibility of values of identical 
grammatical categories of syntactically colnbined 
linguistic items. Ill our view, two general co- 
variation types must be distinguished (cf. Hierarchy 
6), namely, asymmetric and symmetric co-variation, 
with only the former actually corresponding to the 
traditional directional concept of agreement, e.g., 
\[Cor98\]. 
Cc-vatiatiorl 
asymmetric symmetric 
al 1~1 ('d ~I 
t__~ t__J 
llierarchy 6: Morphosyntaclic co-variation 
As the term suggests, the asymmetry of co-variation 
presupposes a controller-target conl'iguration. This 
is to be contrasted with the synunetry of co- 
variation which is not interpretable in these terms. 
Symmetric co-variation, in essence, would presume 
redundancy as if both co-varying syntactic items 
were controllers and targets at the same time. 
Endocentric hypotaxis (selection) 
The endocentric hypotaxis corresponds to the tradi- 
tional notion of selection. Even though not directly 
observable, it underlies specific morphosyntactic 
realities interpreted in Hierarchy 7 as resulting from 
a cross-classification with the combinatorial syn- 
tagmatic relation types. 
The traditional notion of subcategorisation call thus 
be viewed as a selection that is realised via govern- 
mont. Two general oplions are usually available 
across s lbr externalising the governed 
selection (i.e. subcategorisation) of nominal catego- 
ries in actual syntactic constructions. 
syntagmatics 
combinatorial structural 
co-variatien centricity taxis 
asymmetric symmemc juxtaposition govemmem ot,ao exo nypo L)ara i 
seleclion 
\[ agreement I \[matching marking SL~bcalogonsatlon 
cress-referencing relational case 
object eliticisation 
Hierarchy 7: Selection 
A typical definition of Ihe firsl one call be found in 
\[Blake94\]: "Case in its most een|ral manifestation is 
a system of marking dependent nouns for the type 
of relationship they bear to their beads". The rela- 
tional case explicating case government stands in 
opposition to tile so-called concordial case ob- 
served in ease-governed modification environments 
and presented in Hierarchy 8. The second option to 
externalise subcatcgorisation of nominal categories 
is the cross-referencing the syntactic function of the 
dependent at the head. It is actually confined to 
certain core grammatical relations and typically 
amounts to some kind of pronominal representation 
of these grammatical relations at tile head. As 
\[Blake94\] observes, the cross-referencing pronomi- 
nal elements serve as an alternative to case ill sig- 
nalling grammatical relations. In Slavic, there are 
two candidales for the second /ype of externalising 
a governed selection. On tile ()he hand, lh0 verb 
inl\]cction can possibly be interpreted as cross- 
referencing the sub.icct fnnetion, especially in Bul- 
garian where no relational case is realised on the 
dependent. On the other band, pronominal elitics 
can cross-reference the direct and the indirect object 
in Bulgarian verb complex, cf. \[Avg97a\], as well as 
the possessor relation in Bulgarian noun phrases, cf. 
\[Avg97b\]. Therefore, the systematic relation of 
object cliticisation can be viewed as a more specific 
instance of cross-referencing. In general, a nominal 
category representing a grammatical relation that is 
cross-referenced at the head selecting this nominal 
category need not be overtly realised. So, the cross- 
referenced noun phrase controlling the agreement 
can typically be omitted. 
In our view, the systematic relation of marking 
(Hierarchy 7) is an instance of juxtaposed selection, 
i.e. all endocentrie hypotaxis that is realised via 
juxtaposition. We promote a faMy extensive under- 
standing of this syntactic relation as involving vari- 
ous functional categories, including auxiliaries, 
particles, determiners, prepositions, conjunctions, 
etc. The notion of morphosyntactic marking is in- 
troduced in \[Avg97a\] for syntagmatic relations 
30 
holding in Bulgarian analyiic verb lbrms between 
the main verb as a syntactic (and semantic) head 
and the possibly nlultiplc auxiliary verbs ;is nlai+kers 
specifying it. The agreement between the verb and 
its subject or conlplcmcnt is interpreted in our tax- 
onotny as a selection real\]seal via asymmetric co- 
variation (agreement 1). It typically occurs in con\]- 
bination with the relation of subcatcgorisation (of. 
Table 1) which in s mnploying relational 
case is realiscd as case assignment, while it\] those 
cn\]l+~loying cross-reforencing ;is a syntactic function 
identification. 
entliy :i 
Ii(?ltll \] INDliX ~\] 
ii ,Nt,+xl,  
;';','&Ic,,+,+ I++l 
INI)I+X\[II 
Ilma ',wa,vor tzp I i II:AI)ICASE \[\] I \[NI)EX \[1~ j 
........ ~B\] INI)F:X 
('oFuht 
INH NtIMItER\[I\]\] • 
GI~NI)IiR ~2 \] 
entity Ii 
l ,.,,,.b I INI:I,\[I~\] 
lthraval'-h,xe,u , 
IIEAI)\]I+I!X I:t)Ie.MANT III:ADICASF{2\] IN) X\[\] 
"cliticized" i'ed~ 
('l l'llCS / III!AI)ZCASEI2\]\] 
"' " \IINI)I!X~ \] 1'"') 
'clilicized" holm 
,++,< <:, 
\L tNDExLU 1/ 
copttla 
INFI.\[1 \] 
predicative - ilolltl 
INIII~X\[ NUMItI'R~I\] \] 
predicatire adjecfire 
INI'I NIIMBI!I{~j\] 
" {;I.:NDiiR ~\] / 
systenmilc rdiltlttns 
relational case 
agreement 1 with subject a 
subcategori.valion 
agreement 1 with sttbiect a 
cro.vs-r@wem:ing 
agreement 1 with expcricnccr a 
cross-re./brencin,q, 
agreelnell\[ \] with contplol//Otll il 
cros.v-r@,rencing 
agreemcni 1 with specifier a 
SITvle langu'lges 
Russian, l'olish, Czech 
llulgarian 
relatimud case Polish, Czech 
agreement 1 with subject "l 
,~ttbcaleL, ori,valioH |/ulgal'ian 
agreement 1 with subject a 
Polish, Czech 
rehltio.al case 
agrcenlenl 1 with conipietncnt I} 
\] I ulgaritlll 
,s'ubcategori,~alioll 
ttgrccnlent \] ,,villi conlplcntcnt \]l 
Table 1: Subcategorisation and agroenlent I 
entity a entity b \[ 
,,,,i,,,,. I \[,,,,,.z, I INFI,\[li \] \[INFi,\[2}\] 
systematic rehttions 
marking 
malching 
sllbg;alegorisoIiOll 
agreement 1 with complcnlcnt b 
Slavic hulgtulges 
lhtlgarian 
? (l~,ttssian, Czech, Polish) 
l{ussian, Czech, Polish 
Table 2: Marking and matdfillg 
Finally, what we call matching corresponds to a 
selection rcaliscd via symmetric co-variation. Its 
most typical instance can be found where there is 
compatibility in person, number or gender between 
(possibly multiple) auxiliaries and a main verb. 
Matching usually co-occut's with a 1nat+king rchition 
(cf. Table 2) which, as shown in Hierarchy 7, is 
interpreted its a juxlaposcd selection. 
Exocentric hypotaxis (modification) 
The exoccntric hypotaxis corresponds to the mtdi- 
tional notion of modification. It nnderlies specific 
morphosyntactic realities resulling from a cross- 
classification with combinatorial syntagmalic rela- 
tion types (cf. Hierarchy 8). In all of then\], we are 
confronted with a +'elation of subordination in 
which, however, there is no indisputable prominent 
clcmenL 
It+ general we assume that there is no "case agree- 
ment". Rather, the regular compatibility of case 
specifications between the itwolved syntaclic items 
is due to a modification relation realised in a gov- 
erned environment. Ix+ other words, we cat\] regard 
concordial case as a typical instance of a governed 
modification. The asymmetric co-variation real\]sing 
a modification relation can be called concord, but 
let us refer to it - for the sake of sitnplicity - as 
ctgreement 2. In the majority of Slavic s, 
but obviously not in Bulgarian, this relation occurs 
in combination with concordial case (of. Table 3). 
Note that in our approach the treat+nun+ of nominal 
apposition would be parallel to that of the adjective- 
noun relations. 
syntaomatics 
combinatorial structural 
centricity co-vanarton 
asyRlll\]olec symmemc juxtaposition govemelenz endo 
taXiS 
exo nypo para 
l~odif~calion 
agreement 2 simile uxlaposed modificationtion governed modification \] tconcomt 
(case) adjunction secondary predicalion concord@ case 
predicative case adjunction 
I licrarchy 8: Modification 
The main difference between tl~e agt'eement 1 
(Hierarchy 7) and the agreement 2 (or concord) 
discussed here amounts to tile fact that these co- 
variation relations exhibit different ccntricity. 
Cross-classifying cxoccntric hypotaxis with sym- 
metric co-vat\]alien results in what cat+ be called 
simile and is typically observed in comparative 
31 
constructions, provided appropriate categories are 
available. This systematic relation differs fl'om that 
of parallelism (distinguished in Hierarehy 10) in 
being hypotactic in nature, and thus, an actual in- 
stance of modification. Similarly to the asymmetric 
agreement 2 (concord), tim simile relation co-occurs 
with concordial case, cf. ex. 6. The systematic rela- 
tion of (case) adjunction is an instance of juxta- 
I)osed mod(fication. 
entity a entity 1) 
adjective noun \[CASI'\[~ \] IIEAIIlCASli \[~ 
\] 
systematic relations 
cmlcordial case 
ngreement 2 (concord) 
Shtvic hmguages 
Russian, Polish, Czech 
governed mod~'c'ation Bulgarian 
agreement 2 (concord) 
com'ordial case Russian, Polish, Czech 
agreement 2 (concord) 
governed mod(fication Bulgarian 
ngreement 2 (concord) 
Table 3: Concordial case and agreement 2 (concord) 
Interestingly, the well-known "instrumental" prob- 
lem - i.e. whether we are confi'onted with a com- 
plement or a free adjt, nct- narrows down in our 
approach to a fluctuation between adjunction 
(juxtaposed modification - Hierarchy 8) and sub- 
categorisation (governed selection - Hierarchy 7), 
with the crucial point being merely a different ccn- 
tricity interpretation. Also the secondaly predica- 
tion (referring, typically, to the relation holding 
between a verb and a secondary controlled predica- 
tive) is a subtype of juxtaposed modification, with 
the predicative case adjunction as a more specific 
instance. As to the relation holding between the 
secondary predicative and the subject or the object, 
it is an instance o1' control and presupposes co- 
reference. The latter two concepts realise an endo- 
ccntric parataxis and are introduced in Hierarchy 9. 
Endocentric parataxis 
In the endoccntric parataxis there is a prominent 
item but no subordination relation. This allows us to 
model concepts like co-reference, correlation, co- 
marking (illustrated by ex. 6) and control as natu- 
rally resulting from a cross-classification with the 
combinatorial syntagmatic relation types. 
If an endocentric parataxis is revealed by an asym- 
metric co-variation, this results in co-reference. This 
systematic relation is tbund in relativisation (i.e. 
between a nominal category and the relative pro- 
noun introducing a relative clause that modifies this 
nominal category), in resumption (i.e. between a 
nominal category and the pronominal element re- 
suming it in a different syntactic domain), and in 
binding (i.e. between a pronoun and its antecedent). 
When, however, an endocentric parataxis is re- 
vealed in a symmetric co-variation, we can speak of 
correlation. But in both instances of co-variation, 
we arc confronted with paMng indices (or restricted 
parameters) el' referential objects. What we propose 
to distinguish as co-marking corresponds to endo- 
centrie parataxis that is realised via juxtaposition. 
So, it contrasts with the systematic relation of 
marking (presented in Hierarchy 7) only along the 
taxis dimension of structural syntagmatics inasmuch 
as there is no subordination relation between the 
involved syntactic items. As to the systematic rela- 
tion of syntactic control, it is registered in our tax- 
enemy as an cndocentrie parataxis resulting in a 
for,n government. In Bulgarian, it co-occurs with 
co-reference - cx. 3 and ex. 4. 
syntagmaties 
combinatorial srrucmra 
co-variation centricity taxis 
asymmernc symmetric luxraposliion government enao exo Ilypo \[;ara 
en~o-Dara 
co-referenoe correlallon co marking conlml 
rEativisalion resumption binding 
Ilierarchy 9: Endoccntric parataxis 
Exocentric parataxis 
The cxoeentric parataxis is the actual tnunarkcd 
case: there is neither a prominent dement nor a 
subordination relation belwcen the involved syntac- 
tic items. A cross-classification with combinatorial 
syntagmatic relation types allows us to encode fur- 
ther phenomena that are shown Hierarchy 10. 
syntagmatics 
t 
combinatorial structural 
\ t 
co-variation '\, centricity taxis 
\ 
asymmetric symmetric juxtaposibbn government endo exo hy#o para 
, \ 
,, , , 
exo-para ) 
\ 
agreement 3 \[ parallersm ' coordination co-dependence 
I (accord) I J \[ \] 
Hierarchy I0: Exocenlrie paralaxis 
The relation of co-dependence plays a crucial role 
in a number of constructions. It is an exoeentric 
parataxis that is realised via government, with a 
special requirement that all involved syntactic items 
have the same governor. In other words, these items 
are typically dependcnts of the same syntactic head. 
What we call agreement 3 (or accord) corresponds 
to an exocentric paralaxis that is rcaliscd via asym- 
metric co-variation. It regularly presupposes a co- 
32 
dependence relation (of. Table 4), and its most 
typical instance can be fotmd as a COlnpatibility in 
number or gender between the subject and the 
predicative in copular constructions. Another in- 
stance is lhe co-dependence relation holding be- 
lween a dependent of the primary predicate (i.e. the 
verb) and a secondary predicative in ex. 1 and cx. 2. 
When exocentric parataxis is externalised by a 
symmetric co-variation, we are confronted with 
parallelism. It co-occurs in ex. 5 with co- 
dependence. 
The coordinatioll relation is generally interprelable 
as an exocemric paratactic juxtaposition. 
mmn NUMBI{R~\[ INI)EX (;I!NI,ER\[2 l\] 
........ l\] 
INI)FX \[GENDI{I{ ~\] 1 \[ 
predicative noun 
predicative-adjective l 
INFI \[NI3MIIER0\]\] \[ 
........ 
INI)I:X \[NUMBH~ 
adjective 1 
~'stcma(ic relations 
co-d('/)glldetlce 
agreement 3 (accord) between subject a and 
complement b 
co-dependence 
agreement 3 (accord) between subject / 
object a and secondary predicalive I} 
~ s 
(Russian ?), V, ulgarian, 
Polish, Czech 
Russian, Bulgarian, t'olish, 
Czech 
Table 4: Co-dependence and a ;reement 3 (accold) 
Conclusion and prospects 
We have p,esented an approach of computational 
grammar design tlmt st, pports the notion of gram- 
mar sharing and, moreover, lends itself to the for- 
real linguistic description of individual s as 
well as  families. The basic building blocks 
of such a grammar were demonstrated with the 
example of Slavic hmguagcs. Grammars of this type 
can straightforwardly be extended and employed in 
,q ntlnlber o\[" developlncnt and rLiiiI.illle systems 
accommodating HPSG. Some of these systems have 
reached a parsing efficiency thai makes them suited 
for a wu'iety of applications, \]F()TU2000\[. Al- 
though the original motiwltion for the work calllO 
from applied research, the insights thai were gained 
on the differences between Slavic s, led lo 
new results in comparative linguistic dcscrip/ion. 
We expecl that psycholinguistic research on bilin- 
gt, alism and second  acquisition will 
greatly benefit from opportunities of modelling 
shared grammatical knowledge. The insights gained 
by such models will in turn be useful for CALL 
applications and for the computational treatment of 
cross- interference in gl'ananlar and style 
checking. 
charts) 
co-dependence 
agreement 3 (accord) \[SG.F\] 
relational case \[INST\] 
agreement 1 ISG.FI 
concordial case \[INST\] 
agrccnlcnt 2 (concord) \[SG.F\] 
devo~koj. 
gM-SG.F.INST 
33 
ex. 2 (Russian) 'They ordered him to come washed.' 
O/ti relational case \[NOMI 
3PL.NOM lagrecmcnt I \[PL\] 
veleli 
ordercd-PL 
relational case \[DAT\] 
cx. 3 (Bulgarian) 'John saw Mary ill (reportedly).' 
Ivan-3SG.MIVan 
ja 
ACC.SG.F 
ex. 4 (Bulgarian) 'You would come disguised (reportedly).' 
elllll 
3SG.M.I)AT 
ot~iect cliticisation 
vidjal 
saw-3SG.M 
subcategorisation \[INF\] ...... 
control co-dependence 
agreement 3 (accord) \[SG.M\] 
prijti predicative case adjunction \[INST\] 
come-lNF 
cross-rel~rencing 
agreement I \[3SG.F\] 
uno,O,m. 
washcd-SG.M.lNST 
subcategorisation 
Maria 
Mary-SG.F 
Ti ...... subcatcgorisation 
2SG i agreement I \[2SG\] 
si marking 
AUX.2SG matching \[2SG.F\] 
gtjala 
AUX-SG.F 
ex. 5 (Polish) '1 consider him to be nice / to be a fool.' 
Uwa~am relational case IACC\] 
consider-lSG 
go 
3SG.M.ACC 
control 
co-rclizrence \[SG.F\] 
secondary predication 
control 
co-rcl)rcnce \[SG.FI 
bolua. 
ilI-SG.F 
control 
co-rcl~rence \[SG\] 
marking control 
matching \[2SG.F\] co-rcli.zrcnce \[SG.F\] 
i i 
da marking 
particle 
dojdeg 
come-2SG 
secondary predication 
maskiraua. 
disguiscd-SG.F 
relational case \[prepositional ACCI 
co-dependence 
parallelism \[SG.M / 3SG.M\] 
Za marking 
lbr 
milego / durnia. 
nicc-SG.M.ACC / lbol-3SG.M.ACC 
cx. 6 (Russian) 'I suffered lbr him as for a son.' l j,, 
re at on  caseINOMl : 
1SO Ilagreement 1 \[sG\] ............. : ..... .... 
Z~ 
i2~r 
relational case \[prepositional ACC\] 
nego 
3SG.M.ACC 
stradala 
sufl?rcd-SG.F 
marking 
a@mction 
kak 
as 
CO=lllflrking 
concordial case \[ACC\] 
.... simile \[3SG.M\] 
mark n  
I 
Zll 
.~ylt tl 
son-3SG.M.ACC 
34 
Sample mmlyses (relational 
ex. 1 (Russian) 'She turned out a healthy girl.' 
Ona relational case \[NOM\] 
3SG.F.NOM agreement 1 SG.I'\[ 
okazalas' 
turncd-SG. F 
zdorovoj 
healthy-SG.F.INST 

References 

Avgustinova, T., W. Sktlt an(I 1|. Uszkoreit. Typologi- 
cal similarilies in ttPSG: a case sltldy orl Slavic verb dialhc- 
sis. In I'rzcpi6rkowski, A. and R. Borsley (cd.) Slavic in 
ItPSG. CSLI 1999:1-28 

Avguslinova, T. Word order and clilics in Bulgarian. 
Saarbri.icken l)isscrlalions in Computational IAnguislics and 
Imnguagc Technology, Volume 5, 1997 

Avgustinova, T. Clustering clitics in Bulgarian nomi- 
nal constituents. In: Proceedings of FDSL-2, Potsdam 1997 

Bemova, A., K. Oliva and J. Panevova. Some problems of machine translation between closely related s. In: Proceedings of COLING'88, Budapcsl 1988 

Blake, FL,I. Case. Cambridge Tcxlbooks in Linguis- 
tics. Cambridge University Press, 1994 

Cann, R. l'atlems of hcadedncss In: Corbcll, G., N. 
Fraser and S. McGlashan (ed.) tlcads in grammatical lhc- 
ory. Cambrktge Universily Iqvss, 1993.44-72 

Corl)cll, G. Agreement in Slavic. I'osition paper for Ihc 
workshol~ "Comparative Slavic Morphosyntax: The Slate of 
Ihe Art", Indiana University, 1998 

Flickingcr, I)., S. Ocpcn, J. Tsujii, and II. 
Uszkorci! (cd.).(in press) .Iournal of Nalural l~.lnguage t;,n- 
gincering. Special Issue on lffficicnt Processing with IfI'SG. 
Volume 6 (I). Cambridge, UK: Cambridge Univcrsily 
Press, 2000. 

Gamon, M., .C. Lozano, .I. Pinkham and T. Reutler. 
I'raclical experience with grammar sharing in inullilingual 
NI,I'. Technical report MSR-TR-97-16, Redmond 1997 

Kameyama, M. Atomization in grammar sharing. In: 
Proceedings of 26th Annual Meeting of ACL, New York 
1988 

l'iukham, J. Grammar Sharing in I"mnch and linglish. 
In: l'roccedings of IANLP 1996 

Schmidl, I'. and W. Lchfeldt. Kongruenz- Rektion - 
AdjunkliotL Systcmalischc trod historisehe Unlcrstlchtmgen 
zur allgcmcinen Morphosynlax und ztt den Wortfiigungcn 
(slovoso~%tat/ija) im Russischen. Specimina Philologiae 
S lavicae. Miinchen: Otto Sagner 1995 
