This paper focuses on the manual creation of context-‐dependent natural-‐language definitions in EcoLexicon, a terminological knowledge base on the Environment. Given the interdisciplinary nature of the environmental domain, many concepts in EcoLexicon show a high degree of multidimensionality. In other words, this means that concepts can be described from many different perspectives. For such concepts, a single definition that encompasses the whole environmental domain is not informative enough because not all environmental domains describe concepts in the same fashion. For that reason, we propose the creation of flexible definitions. A flexible definition is a system of definitions for the same concept composed of a general environmental definition with a set of recontextualized definitions (definitions that describe a concept from the viewpoint of a certain subject domain). This approach is based on category definitional templates and conceptual frames that provide a consistent way of managing and representing the dimensions of contextually-‐variable concepts in terminological definitions.
LexiCon Research Group. Department of Translation and Interpreting, University of Granada
Calle Buensuceso n.o 11, 18071 Granada, Spain
A conceptual system is considered to be multidimensional
when its concepts are categorized according to different
characteristics, and thus showing their different dimensions
In a terminological knowledge base for translators, which is the case of EcoLexicon, the framework for this research, multidimensional knowledge representation allows the user to acquire a better insight into a given concept. This is very useful for translators because they may need to translate terms that represent concepts viewed from very different points of view.
For the representation of multidimensionality in terminological definitions, our proposal consists of the creation of several natural-language definitions for a given concept, each one describing the concept from a different subdomain of the discipline of the Environment. As a consequence, the concept to be defined is situated in different conceptual frames, which also affects which knowledge is represented in the definitions. 2
EcoLexicon (http://ecolexicon.ugr.es) is a terminological knowledge base on the environment. It is concept-oriented and multilingual. So far, it has 3,533 concepts and 18,798 terms in English, Spanish, German, French, Russian, Modern Greek, and Dutch as well as linguistic and phraseological information for each term. The main target users of EcoLexicon are translators, who must undoubtedly understand what they read and write in subject fields where they are not experts but need to sound like they were. This entails that they need to acquire new specialized knowledge in a very short time. To enhance knowledge acquisition, conceptual information in EcoLexicon is stored and represented in different ways.
On the one hand, specialized knowledge is represented by
means of conceptual networks codified in terms of
conceptual propositions in the form of a triple (CONCEPT
relation CONCEPT), for instance, <SAND type-of SEDIMENT>
or <MORTAR made-of SAND>. The conceptual relations used
in EcoLexicon include both hierarchical (hypernymic and
meronymic) and non-hierarchical relations, some of which
are domain-specific. Concept nature (OBJECT, PROCESS, or
PROPERTY) determines the combinatorial potential of
concepts by means of a closed inventory of conceptual
relations
On the other hand, conceptual information is also shown in the form of natural language definitions in English and Spanish, which are based on the most prototypical conceptual propositions established by the concept to be defined. Additionally, domain-specific knowledge is also presented in the form of images and videos. 2.1
The theoretical and methodological framework of
EcoLexicon is called Frame-Based Terminology
As well as being linked by conceptual relations, concepts
are also organized in the underlying conceptual framework
of the Environmental Event (Fig. 1), where the most generic
categories of the domain are related to each other in an
action-environment interface
The extraction of specialized knowledge for its
subsequent representation is carried out semi-automatically
combining a top-down and bottom-up approach
On the other hand, the top-down approach consists of extracting information from the definitions of concepts in other specialized terminological resources and reference material. 2.2
EcoLexicon has always focused on representing the
multidimensionality of specialized knowledge, given the
interdisciplinary nature of the environmental domain.
However, the simultaneous representation of all conceptual
dimensions often leads to an information overload in the
conceptual networks
On Fig. 2, all the dimensions of SAND are represented at the same time. SAND is a contextually-variable concept that can have different characteristics and be linked to other concepts depending on the context. However, all the properties of the concept are never activated at the same time in real scenarios. For instance, SAND can be categorized as a subtype of SEDIMENT, a SOIL_COMPONENT, a FILTRATION MEDIUM, etc. in different contexts, but not all of them simultaneously. Contextual variation also affects other types of relation, such as functional ones. Sand, in the CIVIL ENGINEERING domain, is related to CONSTRUCTION, whereas in WATER TREATMENT, it is linked to WATER_FILTRATION.
Thus, the problem of overinformation can be resolved by recontextualizing the relational behaviour of concepts (León Araúz, 2009). Recontextualization is thus the process by which triples are stored and represented in a knowledge base, according to the restrictions imposed by different situational contexts. This approach makes it possible to account for the multidimensionality of concepts and, at the same time, to avoid information overload.
In EcoLexicon, the notion of situational context is related to the constraints imposed by environmental subdomains, such as HYDROLOGY, GEOLOGY, OCEANOGRAPHY, CIVIL ENGINEERING, etc.1
Recontextualization is performed by specifying which
conceptual propositions (triples) are prototypically activated
in each domain. This allows the representation of conceptual
1 For the complete list of contextual domains in EcoLexicon, see
In EcoLexicon, definitions are regarded as mini-knowledge
representations
The representativeness of each feature is determined by
the category assigned to the concept being defined. Each
category has a set of representative conceptual relations that
describe it. They are specified in the category definitional
template
Category definitional templates are schematic
representations of the most prototypical relations established
by the concepts that are members of the same category.
They guide the formulation of definitions. They are encoded
in the form of a slot-filler table like Martin’s frame-based
definitions
type-of located-at made-of
CONSTRUCTION SHORELINE MATERIAL made-of has-function has-function
Hard coastal defence structure made of concrete, wood, steel and/or rock perpendicular to the shoreline built to protect a shore area, retard littoral drift, reduce longshore transport and prevent beach erosion. type-of located-at
Category definitional templates are created by combining a bottom-up and top-down approach On the one hand, the top-down approach signifies that the membership in toplevel categories partly determines the configuration of the definition. On the other hand, we also take into account the extension of a category (bottom-up approach), because a category is not only determined by its superordinates but also by its members. Consequently, before defining a concept, it is necessary to categorize it and then analyze the other members of the category so as to modulate the template inherited from superordinate categories.
For concepts with a high level of contextual variation, a
single definition that encompasses the whole environmental
domain is not sufficiently informative, as is the case of these
definitions of SAND in different environmental
terminological resources:
• Mineral rock fragments (sediment) which have a
particle size between 0.06 millimetres and 2.0
millimetres, which is between −1.0 and 4.0 on the phi
(φ) scale. [A Dictionary of Environment and
Conservation
These definitions of SAND are not very useful for a translator dealing with the concept of SAND in different environmental subdomains. For instance, in CIVIL ENGINEERING, it is important to know the different functions of sand, and in SOIL SCIENCES, how sand affects the properties of the soils in which it can be found. Furthermore, no consensus seems to exist regarding SAND hypernyms (i.e. FRAGMENT, SEDIMENT, MATERIAL), because they are also source of contextual variation, which shows that knowledge is not naturally structured in clear-cut taxonomies.
For that reason, we propose the creation of ‘flexible definitions’. A flexible definition is a system of definitions for the same concept composed of a general environmental definition along with a set of recontextualized definitions derived from it, which situate the concept in different domains.
Since flexible definitions follow the same premises used
in the recontextualization of conceptual networks (section
2.2.), they account for the systemic factor in definition
building. According to
Recontextualized definitions are standalone, and thus convey all the necessary information to define a concept in a certain domain, independently of the other definitions in the set. Table 3 presents an extract of the flexible definition of SAND2:
al Definition
Mineral material consisting mainly of particles of quartz ranging in size of 0.05-2 mm.
Sediment consisting mainly of particles of quartz ranging in size of 0.05-2 mm that is part of the soil and can be found in great quantities in beaches, river beds, the seabed, and deserts.
Unconsolidated inorganic soil component consisting mainly of particles of quartz ranging in size of 0.05-2 mm that are the result of weathering and erosion. It renders soils light, acidic, and permeable. 2 Not all domains are included in this example.
consisting mainly of particles of quartz ranging in size of 0.05-2 mm that is mixed with cement, lime and other materials to produce concrete and mortar.
In a flexible definition, the general environmental definition encodes the basic meaning present in all contextual domains and the recontextualized definitions can be considered a variation of it. For this reason, the general environmental definition includes those propositions shared by all the recontextualized definitions (e.g., in the definition of SAND: <SAND made_of QUARTZ>)3. 4.1
One of the main difficulties posed by flexible definitions is that, contrary to what one might think, even hypernyms are subject to contextual variation. Quite understandably, this can impair feature inheritance in a hierarchy. As shown in Table 3, SAND is categorized in different ways depending on how the concept is prototypically conceived in each domain.
Since a coherent hierarchy needs to be specified before the defining process in order to assure correct feature inheritance4, in the case of flexible definitions, each contextual domain requires its own hierarchy. The main information sources that determine how to categorize a concept are the definition of the concept in other terminological resources and KP-based corpus analysis.
On the one hand, extracting the hypernyms of a concept from other resources has its limitations. The first is the fact that it is not usual to find various definitions for the same concept in resources that focus on different domains. For instance, definitions of SAND can be found in Geology and Soil Sciences dictionaries and glossaries. But it is unusual to find an entry for SAND in Water Treatment or Meteorology resources, since the concept is less prototypical in the latter.
On the other hand, however, KP-based corpus analysis
However, all the hypernym candidates extracted with
these two methods can only be used as a guide. Concepts
can be categorized in several ways even in the same
knowledge domain. In fact, many of the categories that can
be extracted with these two methods could be considered ad
hoc categories
The main guidelines for the structuring of recontextualized hierarchies in EcoLexicon are coherence (for correct feature inheritance once all the data is implemented in an ontology) and the activation of the most prototypical underlying conceptual frame. 4.2
According to the principle of cognitive economy
As for recontextualized definitions, the choice of genus is
even more important because by categorizing a concept as a
member of a contextual domain, it is inserted into a specific
conceptual frame. Such a frame takes the form of a
description that relates different conceptual categories.
Whereas in FrameNet
Unlike in
When SAND is categorized as SEDIMENT in GEOLOGY, SOIL_COMPONENT in SOIL SCIENCES, and CONSTRUCTION_AGGREGATE in CIVIL ENGINEERING this situates it in the frames of SEDIMENTATION (Table 4), SOIL_PROPERTIES (Table 5), and COMPOSITE_MATERIAL_PRODUCTION (Table 6), respectively.
Contextual domain: GEOLOGY 1. A MATERIAL suffers WEATHERING and EROSION and, as a consequence, becomes a SEDIMENT. 2. NATURAL_AGENTS transport the SEDIMENT. 3. The SEDIMENT is deposited in a
GEOGRAPHIC_FEATURE.
Contextual domain: SOIL SCIENCES SOIL is composed of SOIL_COMPONENTS that determine the SOIL’S PHYSICAL, CHEMICAL, and BIOLOGICAL_PROPERTIES.
In the CIVIL ENGINEERING hierarchy, SAND is categorized as a CONSTRUCTION_AGGREGATE, which itself is a subordinate of MATERIAL. Therefore, the category definitional template for MATERIAL affects the category definitional template of CONSTRUCTION_AGGREGATE, and the latter can be used, in turn, for the definition of SAND.
Table 7 and 8 show the category definitional templates for MATERIAL and CONSTRUCTION_AGGREGATE.
type-of made-of component-of
MATERIAL PHYSICAL OBJECT SUBSTANCE
PHYSICAL OBJECT
component-of Material consisting of particles that is mixed with a matrix to produce composite material to be used for construction. type-of made-of
MATERIAL SUBSTANCE_PARTICLES COMPOSITE_MATERIAL (when mixed with MATRIX) has-attribute has-function
NATURAL/ARTIFICIAL
CONSTRUCTION
The category definitional template for CONSTRUCTION_AGGREGATE is partly determined by its superordinate concept MATERIAL. Therefore, it activates the relations made-of and component-of. The underlying COMPOSITE_MATERIAL_PRODUCTION frame (Table 6) is the reason why the concepts COMPOSITE_MATERIAL and MATRIX are part of one of the values in the template. The relations has-attribute and has-function, as well as the specification that a <CONSTRUCTION_AGGREGATE is madeof PARTICLES>, are included in the template because of the subordinate concepts of CONSTRUCTION_AGGREGATE. An analysis of its category members such as SAND, GRAVEL, or SLAG reveals that such information is relevant for the description of the category.
Table 9 shows the definition of SAND after the application of the category definitional template of CONSTRUCTION_AGGREGATE.
SAND Natural construction aggregate consisting mainly of particles of quartz ranging in size of 0.05-2 mm that is mixed with cement, lime and other materials to produce concrete and mortar. type-of CONSTRUCTION_AGGREGATE made-of (0.05-2 MM) QUARTZ_PARTICLES component- MORTAR/CONCRETE (when mixed with of CEMENT/LIME) has-attribute NATURAL
As can be observed in Table 9, the proposition <SAND has-function CONSTRUCTION> does not appear either in the definition or in the template for SAND. This is because this proposition is inherited from CONSTRUCTION_AGGREGATE, and it is thus implicit. The other relations are represented in the definition because they are a further specification of the category definitional template of CONSTRUCTION_AGGREGATE.
A single definition is not sufficient to describe multidimensional concepts that participate in many different conceptual frames. Context is an essential factor in the choice of definitional information. Depending on the context, a concept may be categorized differently and therefore establish a link to a different conceptual frame. This underlying conceptual frame guides the configuration of the category definitional template to be used in the defining process.
Because of the inherent limitations of using a closed inventory of conceptual relations, category definitional templates are not as expressive as natural language. Thus, there is the need to nuance the information in the templates. Although the configuration of category definitional templates and frames can be time-consuming, we plan to streamline these tasks in the future by formalizing all this information in an ontology. Our approach based on category definitional templates and frames provides a consistent way of managing and representing the different dimensions of contextually-variable concepts in terminological definitions. This enhances knowledge acquisition in terminological knowledge bases because it affords users a clearer and more coherent vision of each concept and its contextualized meaning in different knowledge domains.
This research was funded by the Spanish Ministry of Economy and Competitiveness (Project FFI 2011-22397) and the Spanish Ministry of Education, Culture, and Sports (FPU Program AP2009-4519).