Formalizing Specialized Knowledge Events in Satellite Ontologies


Pamela Faber Pilar León Araúz Arianne Reimerink
University of Granada University of Granada University of Granada
Department of Translation
and Interpreting
Department of Translation
and Interpreting
Department of Translation
and Interpreting
Buensuceso, 11 (Granada) Buensuceso, 11 (Granada) Buensuceso, 11 (Granada)
pfaber@ugr.es pleon@ugr.es arianne@ugr.es






Abstract
There is a great need for explicit models of
semantic information (terminologies and
background knowledge) in order to support
information exchange. One approach to de-
scribing information semantics is through on-
tologies. Domain-specific ontologies and
general ontologies should depend on each
other since specialized knowledge is based on
and derived from general knowledge. This
paper describes a proposal on how EcoLexi-
con, a visual thesaurus of environmental sci-
ence, can be linked to FunGramKB, a
multipurpose knowledge base that has been
specifically designed for natural language
understanding with modules for lexical,
grammatical, and conceptual knowledge. We
show how the dynamism of environmental
concepts can benefit from a formal descrip-
tion in meaning postulates and their inclusion
in FunGramKB Cognicon scripts. This would
lead to the automatic generation of flexible
conceptual networks and definitional tem-
plates across different contexts.
1 Introduction
There is a clear need for explicit models of se-
mantic information (terminologies) to facilitate
information exchange. One approach to this is
through ontologies, which can be regarded as
shared models of some domain that encode a
view which is common to a set of users.
A domain-specific ontology, which is com-
posed of both concepts and instances within a
certain field, along with their relations and prop-
erties, is a new medium for the storage and
propagation of specialized knowledge (Hsieh et
al. 2010). Nevertheless, one of the problems with
specialized knowledge bases is that they are cre-
ated as stand-alone products, and appear to have
little or no connection with the general knowl-
edge represented in upper-level ontologies.
Upper-level ontologies are composed of gen-
eral concepts and properties, and are a valuable
tool for the contextualization of domain-specific
ontologies since they can and should be extended
so as to make explicit the link between general
and specialized knowledge (Tripathi and Babaie,
2008). This facilitates the acquisition and reuse
of the data.
EcoLexicon is a visual thesaurus of environ-
mental science, whose knowledge is gradually
evolving towards the status of a formal ontology
(León et al., 2009, 2010) and is ready to be con-
nected to other ontologies and resources. In this
paper, we focus on a first approach to the inte-
gration of EcoLexicon into FunGramKB, a mul-
tipurpose lexico-conceptual knowledge base for
NLP systems (Periñan and Arcas, 2010). In sec-
tion 2, the conceptual structure of both EcoLexi-
con and FunGramKB is addressed. In section 3,
we explain how the COREL formalism of Fun-
GramKB can be applied to SEDIMENT and to the
complex environmental event WATER
TREATMENT.
2 Conceptual Representation
2.1 FunGramKB
FunGramKB is an online environment for the
semiautomatic construction of a multipurpose
lexico-conceptual knowledge base for NLP sys-
101
Long papers of the 9th International Conference on Terminology and Artificial Intelligence, TIA 2011, pages 101–107
Paris, 8–10 November 2011

tems (Periñan and Arcas, 2010). It is both multi-
functional (designed for various NLP tasks) as
well as multilingual. FunGramKB has a lexical
level and a grammatical level, which are lan-
guage-specific, and a conceptual level, which is
not. This paper focuses on the integration of
EcoLexicon and FunGramKB at the conceptual
level. The subdivision of the conceptual level is
based on the combination of prototypicality and
temporality, which results in a typology of four
different conceptual schemata:
1. proto-microstructures (meaning postulates):
prototypical knowledge, which does not take
into account the time factor, such as the con-
ceptual representation of song;
2. proto-macrostructures (scripts): prototypical
knowledge that implies the passage of time,
such as the description of eating at a restau-
rant;
3. bio-microstructures (snapshots): instances of
entities, such as the Eiffel Tower;
4. bio-macrostructures (stories): for example,
the construction of the Eiffel Tower.
The conceptual level in FunGramKB is com-
posed of the following: (i) an Ontology of con-
cepts defined with meaning postulates; (ii) a
Cognicon with procedural knowledge stored as
scripts; (iii) an Onomasticon with information
about instances of entities and events in the form
of stories and snapshots. All of these are encoded
in COREL (COnceptual REpresentation Lan-
guage), a representation language based on
meaning postulates.
The FunGramKB ontology is a concept taxon-
omy, derived from linguistic concepts, in which
interlingual differences in syntactic constructions
do not involve conceptual differences. It is gen-
eral-purpose, and not domain-specific. However,
since expert knowledge stems from general
knowledge, it can be extended to include special-
ized knowledge by establishing links to satellite
domain-specific ontologies.
The concepts of FunGramKB belong to three
levels. The upper level is composed of 42 meta-
concepts, marked with the symbol #. These
metaconcepts are distributed in three subontolo-
gies: #ENTITY, #EVENT, and #QUALITY. Despite
the difference in names, this upper level is the
same in EcoLexicon (OBJECT, PROCESS, and
ATTRIBUTE).
Concepts at the middle level are marked by +
(e.g. +BOOK_00). These concepts are used in the
meaning postulates that define basic concepts
and terminal concepts, and also encode the selec-
tion restrictions in thematic frames. The third
level is composed of terminal concepts, marked
by $ (e.g. $METEORITE_00). The difference be-
tween basic and terminal concepts is that basic
concepts are used to define other concepts in
meaning postulates, whereas terminal concepts
are not. Evidently, in the satellite ontologies for
specialized knowledge, terminal concepts in
FunGramKB will have to be extended.
The Ontology is grounded on a spiral model,
where conceptual promotion and demotion can
occur between the basic and terminal levels
(Periñan and Arcas, 2010). Terminal concepts
can thus become basic concepts when the inclu-
sion of a new language or in this case, a more
specialized conceptual content, demands a more
specific world model. Inversely, basic concepts
can be demoted to terminal concepts in the case
that they are not used to describe other concepts.
However, the metaconceptual level always re-
mains stable.
An example of a meaning postulate of a basic
concept is shown in (1): +LEAVE_00 in the di-
mension of #MOTION.
(1) + (e1: +MOVE_00 (x1)Agent (x2)Theme (x3)Location (x4)Origin
(x5)Goal (f1: (e2: +BE_02 (x2)Theme (x4)Location (f2:
+IN_00)Position))Condition (f3: (e3: +be_02 (x5)Theme (x4)Location
(f4: +OUT_00)Position))Condition)
This representation is composed of three
events (e1, e2, and e3) (Mairal and Periñan, 2009).
In the first event, an Agent (x1) causes another
entity (x3) with the role of Theme) to move from
an Origin (x4) to a Goal (x5), provided two other
events occur. The second event states that the
Theme should be located at the Origin, whereas
the third event states that the Goal (x5) should be
located outside the Origin.
Although the granularity of this type of se-
mantic description is coarse-grained in compari-
son with standard lexicography, it is fine-grained
in comparison with the axioms in other formal
ontologies.
FunGramKB integrates semantic knowledge
from the ontology with procedural knowledge
from the Cognicon. In the Cognicon, proto-
macrostructures or scripts are structured into one
or more predications within a linear temporal
framework. In (2) the first nine predications are
shown of a proto-macrostructure (Periñan and
Arcas, 2010: 2669).
(2)
@EATING_AT_RESTAURANTS
*(e1: +ENTER_00 (x1: +CUSTOMER_00)Agent (x1)Theme
(x2)Location (x3)Origin (x4: +RESTAURANT_00)Goal (f1: (e2:
+BE_01 (x1)Theme (x5: +HUNGRY_00)Attribute))Reason)
102