Proceedings of the Workshop on Frontiers in Linguistically Annotated Corpora 2006, pages 78–85,
Sydney, July 2006. c©2006 Association for Computational Linguistics
On Distance between Deep Syntax and Semantic Representation
V´aclav Nov´ak
Institute of Formal and Applied Linguistics
Charles University
Praha, Czech Republic
novak@ufal.mff.cuni.cz
Abstract
We present a comparison of two for-
malisms for representing natural language
utterances, namely deep syntactical Tec-
togrammatical Layer of Functional Gen-
erative Description (FGD) and a seman-
tic formalism, MultiNet. We discuss the
possible position of MultiNet in the FGD
framework and present a preliminary map-
ping of representational means of these
two formalisms.
1 Introduction
The Prague Dependency Treebank 2.0 (PDT 2.0)
described in Sgall et al. (2004) contains a large
amount of Czech texts with complex and inter-
linked morphological (2 million words), syntactic
(1.5M words), and complex semantic (tectogram-
matical) annotation (0.8M words); in addition,
certain properties of sentence information struc-
ture and coreference relations are annotated at the
semantic level.
The theoretical basis of the treebank lies in the
Functional Generative Description (FGD) of lan-
guage system by Sgall et al. (1986).
PDT 2.0 is based on the long-standing Praguian
linguistic tradition, adapted for the current
computational-linguistics research needs. The
corpus itself is embedded into the latest annotation
technology. Software tools for corpus search, an-
notation, and language analysis are included. Ex-
tensive documentation (in English) is provided as
well.
An example of a tectogrammatical tree from
PDT 2.0 is given in  gure 1. Function words are
removed, their function is preserved in node at-
tributes (grammatemes), information structure is
annotated in terms of topic-focus articulation, and
every node receives detailed semantic label corre-
sponding to its function in the utterance (e.g., ad-
dressee, from where, how often, . . . ). The square
node indicates an obligatory but missing valent.
The tree represents the following sentence:
Letos
d15d15 d33d33d67d67
d67 se
d29d29d60d60
d60sna z· 
d15d15d1d1d2d2d2d2
o
d15d15
n·avrat
d15d15
do
d15d15
politiky.
d15d15
This year he tries to return to politics.
(1)
_ .
 
          
_
 
. . .
.
_
 . .
. . .
_
 
.
_ .
 
.
t-ln94200-123-p12s3
root
letos
t TWHEN basic
adv.denot.ngrad.nneg
#PersPron
t ACT
n.pron.def.pers
anim sg 3 basic
sna it_se enunc
f PRED
v decl disp0 ind
proc it0 res0 sim
nÆvrat
f PAT
n.denot
inan sg
politika
f DIR3 basic
n.denot
fem sg
Figure 1: Tectogrammatical tree of sentence (1)
1.1 MultiNet
The representational means of Multilayered Ex-
tended Semantic Networks (MultiNet), which are
78
described in Helbig (2006), provide a universally
applicable formalism for treatment of semantic
phenomena of natural language. To this end, they
offer distinct advantages over the use of the clas-
sical predicate calculus and its derivatives. The
knowledge representation paradigm and semantic
formalism MultiNet is used as a common back-
bone for all aspects of natural language process-
ing (be they theoretical or practical ones). It is
continually used for the development of intelligent
information and communication systems and for
natural language interfaces to the Internet. Within
this framework, it is subject to permanent practical
evaluation and further development.
The semantic representation of natural language
expressions by means of MultiNet is mainly in-
dependent of the considered language. In con-
trast, the syntactic constructs used in different
languages to describe the same content are ob-
viously not identical. To bridge the gap be-
tween different languages we can employ the deep
syntactico-semantic representation available in the
FGD framework.
An example of a MultiNet structure is given in
 gure 2. The  gure represents the following dis-
course:
Max gave his brother several apples.
This was a generous gift.
Four of them were rotten. (2)
MultiNet is not explicitly model-theoretical and
the extensional level is created only in those situ-
ations where the natural language expressions re-
quire it. It can be seen that the overall structure
of the representation is not a tree unlike in Tec-
togrammatical representation (TR). The layer in-
formation is hidden except for the most important
QUANT and CARD values. These attributes con-
vey information that is important with respect to
the content of the sentence. TR lacks attributes
distinguishing intensional and extensional infor-
mation and there are no relations like SUBM de-
noting relation between a set and its subset.
Note that the MultiNet representation crosses
the sentence boundaries. First, the structure repre-
senting a sentence is created and then this structure
is assimilated into the existing representation.
In contrast to CLASSIC (Brachman et al., 1991)
and other KL-ONE networks, MultiNet contains a
prede ned  nal set of relation types, encapsula-
tion of concepts, and attribute layers concerning
cardinality of objects mentioned in discourse.
In Section 2, we describe our motivation for ex-
tending the annotation in FGD to an even deeper
level. Section 3 lists the MultiNet structural coun-
terparts of tectogrammatical means. We discuss
the related work in Section 4. Section 5 deals with
various evaluation techniques and we conclude in
Section 6.
2 FGD layers
PDT 2.0 contains three layers of information about
the text (as described in Haji c (1998)):
Morphosyntactic Tagging. This layer represents
the text in the original linear word order with
a tag assigned unambiguously to each word
form occurence, much like the Brown corpus
does.
Syntactic Dependency Annotation. It contains
the (unambiguous) dependency representa-
tion of every sentence, with features describ-
ing the morphosyntactic properties, the syn-
tactic function, and the lexical unit itself. All
words from the sentence appear in its repre-
sentation.
Tectogrammatical Representation (TR). At
this level of description, we annotate every
(autosemantic non-auxiliary) lexical unit
with its tectogrammatical function, position
in the scale of the communicative dynamism
and its grammatemes (similar to the mor-
phosyntactic tag, but only for categories
which cannot be derived from the word’s
function, like number for nouns, but not its
case).
There are several reasons why TR may not be
suf cient in a question answering system or MT:
1. The syntactic functors Actor and Patient dis-
allow creating inference rules for cognitive
roles like Affected object or State carrier. For
example, the axiom stating that an affected
object is changed by the event ((v AFF o) →
(v SUBS change.2.1)) can not be used
in the TR framework.
2. There is no information about sorts of con-
cepts represented by TR nodes. Sorts (the
upper conceptual ontology) are an important
source of constraints for MultiNet relations.
Every relation has its signature which in turn
79
Figure 2: MultiNet representation of example discourse (2)
reduces ambiguity in the process of text anal-
ysis and inferencing.
3. Lexemes of TR have no hierarchy which lim-
its especially the search for an answer in a
question answering system. In TR there is
no counterpart of SUB, SUBR, and SUBS
MultiNet relations which connect subordi-
nate concepts to superordinate ones and indi-
vidual object representatves to corresponding
generic concepts.
4. In TR, each sentence is isolated from the
rest of the text, except for coreference arrows
heading to preceding sentences. This, in ef-
fect, disallows inferences combining knowl-
edge from multiple sentences in one infer-
ence rule.
5. Nodes in TR always correspond to a word
or a group of words in the surface form of
sentence or to a deleted obligatory valency
of another node. There are no means for
representing knowledge generated during the
inference process, if the knowledge doesn’t
have a form of TR. For example, consider ax-
iom of temporal precedence transitivity (3):
(a ANTE b) ∧ (b ANTE c) → (a ANTE c)
(3)
In TR, we can not add an edge denoting
(a ANTE c). We would have to include a
proposition like  a precedes c as a whole
new clause.
For all these reasons we need to extend our text
annotation to a form suitable to more advanced
tasks. It is shown in Helbig (2006) that MultiNet
is capable to solve all the above mentioned issues.
Helbig (1986) describes a procedure for auto-
matic translation of natural language utterances
into MultiNet structures used in WOCADI tool for
German. WOCADI uses no theoretical intermedi-
ate structures and relies heavily on semantically
annotated dictionary (HagenLex, see Hartrumpf et
al. (2003)).
In our approach, we want to take advantage of
existing tools for conversions between layers in
FGD. By combining several simpler procedures
for translation between adjacent layers, we can im-
prove the robustness of the whole procedure and
the modularity of the software tools. Moreover,
the process is divided to logical steps correspond-
ing to theoretically sound and well de ned struc-
tures. On the other hand, such a multistage pro-
cessing is susceptible to accumulation of errors
made by individual components.
3 Structural Similarities
3.1 Nodes and Concepts
If we look at examples of TR and MultiNet struc-
tures, at  rst sight we can see that the nodes of
TR mostly correspond to concepts in MultiNet.
However, there is a major difference: TR does not
include the concept encapsulation. The encapsu-
lation in MultiNet serves for distinguishing def-
initional knowledge from assertional knowledge
about given node, e.g., in the sentence  The old
man is sleeping , the connection to old will be in
the de nitional part of man, while the connection
to the state is sleeping belongs to the assertional
80
part of the concept representing the man. In TR,
these differences in content are represented by dif-
ferences in Topic-Focus Articulation (TFA) of cor-
responding words.
There are also TR nodes that correspond to no
MultiNet concept (typically, the node representing
the verb  be ) and TR nodes corresponding to a
whole subnetwork, e.g., Fred in the sentence  Fred
is going home. , where the TR node representing
Fred corresponds to the subnetwork1 in  gure 3.
SUB
human
ATTR
SUB
 rst name
VAL
fred
G01
Figure 3: The MultiNet subnetwork correspond-
ing to TR node representing Fred
3.2 Edges, relations and functions
An edge of TR between nodes that have their
conceptual counterparts in MultiNet always corre-
sponds to one or more relations and possibly also
some functions. In general, it can be said that
MultiNet representation of a text contains signif-
icantly more connections (either as relations, or as
functions) than TR, and some of them correspond
to TR edges.
3.3 Functors and types of relations and
functions
There are 67 functor types in TR (see Haji cov·a
et al. (2000) for description), which correspond to
94 relation types and 19 function types in Multi-
Net (Helbig, 2006). The mapping of TR functions
to MultiNet is given in table 1:
TR functor MultiNet counterpart
ACMP ASSOC
ACT AFF, AGT, BENF, CSTR, EXP,
MEXP, SCAR
ADDR ORNT
ADVS SUBST, OPPOS
AIM PURP
APP ASSOC, ATTCH
continued . . .
1In fact the concept representing the man is the concept
G01, i.e. only one vertex. However, the whole network cor-
responds to the TR node representing Fred.
TR functor MultiNet counterpart
APPS EQU, NAME
ATT MODL
AUTH AGT, ORIG
BEN BENF
CAUS CAUS, JUST
CNCS CONC
CM *ITMS, MODL
COMPL PROP except for sentential com-
plements
COND COND
CONFR OPPOS
CONJ *IMTS-I, *TUPL
CONTRA OPPOS
CONTRD CONC
CPR *COMP
CRIT METH, JUST, CIRC, CONF
CSQ CAUS, JUST, GOAL
DIFF *MODP, *OP
DIR1 ORIGL, ORIG
DIR2 VIA
DIR3 DIRCL, ELMT
DISJ *ALTN2, *VEL2
EFF MCONT, PROP, RSLT
EXT QMOD
HER AVRT
ID NAME
INTT PURP
LOC LOC, LEXT
MANN MANNR, METH
MAT ORIGM
MEANS MODE, INSTR
MOD MODL
OPER *OP, TEMP
ORIG AVRT, INIT, ORIGM, ORIGL,
ORIG
PARTL MODL
PAT AFF, ATTR, BENF, ELMT,
GOAL, OBJ, PARS, PROP,
SSPE, VAL
PREC REAS, OPPOS
REAS CAUS, GOAL
REG CONF
RESL CAUS, GOAL
RESTR *DIFF
RHEM MODL
RSTR PROP, ATTR
SUBS SUBST
continued . . .
81
TR functor MultiNet counterpart
TFHL DUR
TFRWH TEMP
THL DUR
THO QUANT layer
TOWH SUBST, TEMP
TPAR TEMP, DUR
TSIN STRT
TTILL FIN
TWHEN TEMP
Table 1: Mapping of TR functors to MultiNet
There are also TR functors with no appropriate
MultiNet counterpart: CPHR, DENOM, DPHR,
FPHR, GRAD, INTF, PAR, PRED and VOCAT
Table 2 shows the mapping from MultiNet rela-
tions to TR functors:
MultiNet TR counterpart
Relations:
AFF PAT, DIR1
AGT ACT
ANTE TWHEN
ARG1/2/3 ACT, PAT, . . .
ASSOC ACMP, APP
ATTCH APP
ATTR RSTR
AVRT ORIG, ADDR, DIR1
BENF BEN
CAUS CAUS, RESL, REAS, GOAL
CIRC CRIT
CONC CNCS
COND COND
CONF REG, CRIT
CSTR ACT
CTXT REG
DIRCL DIR3
DUR TFHL, PAR, THL
ELMT DIR3, DIR1
EXP ACT
FIN TTILL
GOAL see RSLT, DIRCL and PURP
IMPL CAUS
INIT ORIG
INSTR MEANS
JUST CAUS
LEXT LOC
LOC LOC
MANNR MANN
continued . . .
MultiNet TR counterpart
MCONT PAT, EFF
MERO see PARS, ORIGM, *ELMT,
*SUBM and TEMP
METH MANN, CRIT
MEXP ACT
MODE see INSTR, METH and
MANNR
MODL MOD, ATT, PARTL, RHEM
NAME ID, APPS
OBJ PAT
OPPOS CONTRA
ORIG ORIG, DIR1, AUTH
ORIGL DIR1
ORIGM ORIG
ORNT ADDR
PROP COMPL, RSTR
PROPR COMPL, RSTR
PURP AIM
QMOD RSTR
REAS see CAUS, JUST and IMPL
RPRS LOC, MANN
RSLT PAT, EFF
SCAR ACT
SITU see CIRC and CTXT
SOURC see INIT, ORIG, ORIGL,
ORIGM and AVRT
SSPE PAT
STRT TSIN
SUBST SUBS
SUPPL PAT
TEMP TWHEN
VAL RSTR, PAT
VIA DIR2
Functions:
∗ALTN1 CONJ
∗ALTN1 DISJ
∗COMP CPR, grammateme DEGCMP
∗DIFF RESTR
∗INTSC CONJ
∗ITMS CONJ
∗MODP MANN
∗MODQ RHEM
∗MODS MANNR
∗NON grammateme NEGATION
∗ORD grammateme NUMERTYPE
∗PMOD RSTR
∗QUANT MAT, RSTR
continued . . .
82
MultiNet TR counterpart
∗SUPL grammateme DEGCMP
∗TUPL CONJ
∗UNION CONJ
∗VEL1 CONJ
∗VEL2 DISJ
Table 2: Mapping of MultiNet relations to TR
There are also MultiNet relations and functions
with no counterpart in TR (stars at the begin-
ning denote a function): ANLG, ANTO, CHEA,
CHPA, CHPE, CHPS, CHSA CHSP, CNVRS,
COMPL, CONTR, CORR, DISTG, DPND, EQU,
EXT, HSIT, MAJ, MIN, PARS, POSS, PRED0,
PRED, PREDR, PREDS, SETOF, SUB, SYNO,
VALR, *FLPJ and *OP.
From the tables 1 and 2, we can conclude that
although the mapping is not one to one, the prepro-
cessing of the input text to TR highly reduces the
problem of the appropriate text to MultiNet trans-
formation. However, it is not clear how to solve
the remaining ambiguity.
3.4 Grammatemes and layer information
TR has at its disposal 15 grammatemes, which
can be conceived as node attributes. Note that
not all grammatemes are applicable to all nodes.
The grammatemes in TR roughly correspond to
layer information in MultiNet, but also to speci c
MultiNet relations.
1. NUMBER. This TR grammateme is trans-
formed to QUANT, CARD, and ETYPE at-
tributes in MultiNet.
2. GENDER. This syntactical information is not
transformed to the semantic representation
with the exception of occurences where the
grammateme distinguishes the gender of an
animal or a person and where MultiNet uses
SUB relation with appropriate concepts.
3. PERSON. This verbal grammateme is re-
 ected in cognitive roles connected to the
event or state and is semantically super uous.
4. POLITENESS has no structural counterpart
in MultiNet. It can be represented in the con-
ceptual hierarchy of SUB relation.
5. NUMERTYPE distinguishing e.g.  three 
from  third and  one third is transformed to
corresponding number and also to the manner
this number is connected to the network.
6. INDEFTYPE corresponds to QUANT and
VARIA layer attributes.
7. NEGATION is transformed to both FACT
layer attribute and *NON function combined
with modality relation.
8. DEGCMP corresponds to *COMP and
*SUPL functions.
9. VERBMOD: imp value is represented by
MODL relation to imperative, cdn value is
ambiguous not only with respect to facticity
of the condition but also with regard to other
criteria distinguishing CAUS, IMPL, JUST
and COND relatinos which can all result in
a sentence with cdn verb. Also the FACT
layer attribute of several concepts is affected
by this value.
10. DEONTMOD corresponds to MODL rela-
tion.
11. DISPMOD is semantically super uous.
12. ASPECT has no direct counterpart in Multi-
Net. It can be represented by the interplay
of temporal speci cation and RSLT relation
connecting an action to its result.
13. TENSE is represented by relations ANTE,
TEMP, DUR, STRT, and FIN.
14. RESULTATIVE has no direct counterpart
and must be expressed using the RSLT rela-
tion.
15. ITERATIVENESS should be represented by
a combination of DUR and TEMP rela-
tions where some of temporal concepts have
QUANT layer information set to several.
3.5 TFA, quantifiers, and encapsulation
In TR, the information structure of every utterance
is annotated in terms of Topic-Focus Articulation
(TFA):
1. Every autosemantic word is marked c, t, or
f for contrastive topic, topic, or focus, re-
spectively. The values can distinguish which
part of the sentence belongs to topic and
which part to focus.
2. There is an ordering of all nodes according to
communicative dynamism (CD). Nodes with
lower values of CD belong to topic and nodes
83
with greater values to focus. In this way, the
degree of  aboutness is distinguished even
inside topic and focus of sentences.
MultiNet, on the other hand, doesn’t contain
any representational means devoted directly to
representation of information structure. Neverthe-
less, the differences in the content of sentences dif-
fering only in TFA can be represented in MultiNet
by other means. The TFA differences can be re-
 ected in these categories:
• Relations connecting the topic of sentence
with the remaining concepts in the sentence
are usually a part of de nitional knowledge
about the concepts in the topic, while the re-
lations going to the focus belong to the asser-
tional part of knowledge about the concepts
in focus. In other words, TFA can be re ected
in different values of K TYPE attribute.
• TFA has an effect on the identi cation of
presuppositions (Peregrin, 1995a) and allega-
tions (Haji cov·a, 1984). In case of presuppo-
sition, we need to know about them in the
process of assimilation of new information
into the existing network in order to detect
presupposition failures. In case of allegation,
there is a difference in FACT attribute of the
allegation.
• The TFA has an in uence on the scope of
quanti ers (Peregrin, 1995b; Haji cov·a et al.,
1998). This information is fully transformed
into the quanti er scopes in MultiNet.
4 Related Work
There are various approaches trying to analyze
text to a semantic representation. Some of them
use layered approach and others use only a sin-
gle tool to directly produce the target struc-
ture. For German, there is the above mentioned
WOCADI parser to MultiNet, for English, there
is a Discourse Representation Theory (DRT) ana-
lyzer (Bos, 2005), and for Czech there is a Trans-
parent Intensional Logic analyzer (Hor·ak, 2001).
The layered approaches: DeepThought
project (Callmeier et al., 2004) can combine
output of various tools into one representation.
It would be even possible to incorporate TR and
MultiNet into this framework. Meaning-Text
Theory (Bolshakov and Gelbukh, 2000) uses
an approach similar to Functional Generative
Description (  Zabokrtsk·y, 2005) but it also has no
layer corresponding to MultiNet.
There were attempts to analyze the seman-
tics of TR, namely in question answering system
TIBAQ (Jirk u and Haji c, 1982), which used TR di-
rectly as the semantic representation, and Kruijff-
Korbayov·a (1998), who tried to transform the TFA
information in TR into the DRT framework.
5 Evaluation
It is a still open question how to evaluate systems
for semantic representation. Basically, three ap-
proaches are used in similar projects:
First, the coverage of the system may serve as a
basis for evaluation. This criterion is used in sev-
eral systems (Bos, 2005; Hor·ak, 2001; Callmeier
et al., 2004). However, this criterion is far from
ideal, because it’s not applicable to robust systems
and can not tell anything about the quality of re-
sulting representation.
Second, the consistency of the semantic repre-
sentation serves as an evaluation criterion in Bos
(2005). It is a desired state to have a consistent
representation of texts, but there is no guarantee
that a consistent semantic representation is in any
sense also a good one.
Third, the performance in an application
(e.g., question answering system) is another cri-
terion used for evaluating a semantic representa-
tion (Hartrumpf, 2005). A problem in this kind
of evaluation is that we can not separate the eval-
uation of the formalism itself from the evaluation
of the automatic processing tools. This problem
becomes even bigger in a multilayered approach
like FGD or MTT, where the overall performance
depends on all participating transducers as well as
on the quality of the theoretical description. How-
ever, from the user point of view, this is so far
the most reliable form of semantic representation
evaluation.
6 Conclusion
We have presented an outline of a procedure that
enables us to transform syntactical (tectogrammat-
ical) structures into a fully equipped knowledge
representation framework. We have compared
the structural properties of TR and MultiNet and
found both similarities and differences suggest-
ing which parts of such a task are more dif cult
and which are rather technical. The comparison
shows that for applications requiring understand-
84
ing of texts (e.g., question answering system) it is
desirable to further analyze TR into another layer
of knowledge representation.
Acknowledgement
This work was supported by Czech Academy
of Science grant 1ET201120505 and by Czech
Ministry of Education, Youth and Sports project
LC536. The views expressed are not necessarily
endorsed by the sponsors. We also thank anony-
mous reviewers for improvements in the  nal ver-
sion.

References
Igor Bolshakov and Alexander Gelbukh. 2000. The
Meaning-Text Model: Thirty Years After. Interna-
tional Forum on Information and Documentation,
1:10 16.
Johan Bos. 2005. Towards Wide-Coverage Se-
mantic Interpretation. In Proceedings of Sixth In-
ternational Workshop on Computational Semantics
IWCS-6, pages 42 53.
Ronald J. Brachman, Deborah L. McGuinness, Pe-
ter F. Patel-Schneider, Lori Alperin Resnick, and
Alex Borgida. 1991. Living with CLASSIC: When
and How to Use a KL-ONE-like Language. In John
Sowa, editor, Principles of Semantic Networks: Ex-
plorations in the representation of knowledge, pages
401 456. Morgan-Kaufmann, San Mateo, Califor-
nia.
Ulrich Callmeier, Andreas Eisele, Ulrich Schcurrency1afer, and
Melanie Siegel. 2004. The DeepThought Core Ar-
chitecture Framework. In Proceedings of LREC,
May.
Jan Haji c. 1998. Building a Syntactically Anno-
tated Corpus: The Prague Dependency Treebank. In
E. Haji cov·a, editor, Issues of Valency and Meaning.
Studies in Honour of Jarmila Panevov´a, pages 106 
132. Karolinum, Charles University Press, Prague,
Czech Republic.
Eva Haji cov·a, Jarmila Panevov·a, and Petr Sgall.
2000. A Manual for Tectogrammatic Tagging of
the Prague Dependency Treebank. Technical Re-
port TR-2000-09, ·UFAL MFF UK, Prague, Czech
Republic. in Czech.
Eva Haji cov·a, Petr Sgall, and Barbara Partee. 1998.
Topic-Focus Articulation, Tripartite Structures, and
Semantic Content. Kluwer, Dordrecht.
Eva Haji cov·a. 1984. Presupposition and Allegation
Revisited. Journal of Pragmatics, 8:155 167.
Sven Hartrumpf, Hermann Helbig, and Rainer Oss-
wald. 2003. The Semantically Based Computer
Lexicon HaGenLex  Structure and Technological
Environment. Traitement automatique des langues,
44(2):81 105.
Sven Hartrumpf. 2005. University of hagen at qa@clef
2005: Extending knowledge and deepening linguis-
tic processing for question answering. In Carol
Peters, editor, Results of the CLEF 2005 Cross-
Language System Evaluation Campaign, Work-
ing Notes for the CLEF 2005 Workshop, Wien,currency1
Osterreich. Centromedia.
Hermann Helbig. 1986. Syntactic-Semantic Analy-
sis of Natural Language by a New Word-Class Con-
trolled Functional Analysis. Computers and Artifi-
cial Inteligence, 5(1):53 59.
Hermann Helbig. 2006. Knowledge Representation
and the Semantics of Natural Language. Springer-
Verlag, Berlin Heidelberg.
Ale s Hor·ak. 2001. The Normal Translation Algorithm
in Transparent Intensional Logic for Czech. Ph.D.
thesis, Faculty of Informatics, Masaryk University,
Brno, Czech Republic.
Petr Jirk u and Jan Haji c. 1982. Inferencing and search
for an answer in TIBAQ. In Proceedings of the 9th
conference on Computational linguistics – Volume
2, pages 139 141, Prague, Czechoslovakia.
Ivana Kruijff-Korbayov·a. 1998. The Dynamic Po-
tential of Topic and Focus: A Praguian Approach
to Discourse Representation Theory. Ph.D. thesis,
·UFAL, MFF UK, Prague, Czech Republic.
Jaroslav Peregrin. 1995a. Topic, Focus and the Logic
of Language. In Sprachtheoretische Grundlagen f¨ur
die Computerlinguistik (Proceedings of the Goettin-
gen Focus Workshop, 17. DGfS), Heidelberg. IBM
Deutschland.
Jaroslav Peregrin. 1995b. Topic-Focus Articulation
as Generalized Quanti cation. In P. Bosch and
R. van der Sandt, editors, Proceedings of “Focus and
natural language processing”, pages 49 57, Heidel-
berg. IBM Deutschland.
Petr Sgall, Eva Haji cov·a, and Jarmila Panevov·a. 1986.
The Meaning of the Sentence in Its Semantic and
Pragmatic Aspects. D. Reidel Publishing company,
Dodrecht, Boston, London.
Petr Sgall, Jarmila Panevov·a, and Eva Haji cov·a. 2004.
Deep Syntactic Annotation: Tectogrammatical Rep-
resentation and Beyond. In A. Meyers, editor, Pro-
ceedings of the HLT-NAACL 2004 Workshop: Fron-
tiers in Corpus Annotation, pages 32 38, Boston,
Massachusetts, USA. Association for Computa-
tional Linguistics.
Zden ek  Zabokrtsk·y. 2005. Resemblances between
Meaning-Text Theory and Functional Generative
Description. In Proceedings of the 2nd Interna-
tional Conference of Meaning-Text Theory, pages
549 557.
