6. CONCLUSIONS

This doctoral thesis analyzed the effects of contextual variation in environmental concepts with a focus on its representation in terminological definitions. Specifically, we sought to characterize the conceptual phenomena resulting from contextual variation and to develop guidelines on how to reflect them in terminological definitions from the extraction of knowledge to the actual writing of the definition.

By applying a cognitive linguistics approach, the following general conclusions on terminological definitions were derived from our research:

• Defining a term is not the same as describing its meaning because terms do not have meaning outside of real events. They have semantic potential.
• A term’s semantic potential is the raw material for its definition, not its object. The semantic potential is not the object because this would mean that defining a term would involve describing all the conceptual content that the term can activate. This is not viable because a term’s semantic potential corresponds to a vast quantity of information that cannot be measured and which is also never activated entirely in real events.
• The object of a definition is a subset of the term’s semantic potential, which corresponds to a premeaning. A premeaning is an intermediate stage between semantic potential and meaning. It is an abstraction of the meaning that the term could have in a given context. Thus, the chosen premeaning depends on the contextual constraints applied to the definition.
• Defining a term entails offering a definition for each concept contained in the semantic potential of the term. In the definition, a relevant part of the concept (a premeaning) and the frames that it activates are described.
• Defining a concept entails the description of a relevant subset of its traits (a premeaning) along with the background knowledge that always accompanies this subset of conceptual content.
• In view of prototype effects in conceptual representation, the notion of differentiae in definitions needs to be redefined. Differentiae are not necessary and sufficient features, but rather relevant features.
• In line with Seppälä (2012), the factors that determine trait relevance in terminological definitions can be grouped in three dimensions: (i) ontological (based on the ontological type of the definiendum), functional (based on the users and resource), and contextual (based on contextual constraints).

One the specific objectives of this work was to determine the components of context that affect specialized meaning construction and also the ones that affect conceptual representation in terminological definitions. This led us to the following conclusions:

• Since the object of the terminological definition is an abstraction of the situated meanings of a term under certain contextual constraints (i.e., a premeaning), the context associated with a premeaning is also an abstraction of real contexts (i.e., a precontext).
• There are two main types of context: (i) the mental phenomenon (conceptual context); (ii) the situation (situational context). A precontext is an abstraction of a conceptual context, which is composed of the interpretation of the situational context and background knowledge.
• Context comprises linguistic context, discursive context, sociocultural context, and spatial-temporal context. The precontext for terminological definitions includes linguistic constraints, thematic constraints, cultural constraints, ideological constraints, and diachronic constraints.
• Domains, in their sense of knowledge field, allow the systematic characterization of thematic constraints in terminological definitions. They can be understood as macroframes that guide knowledge organization and categorization of in a given conceptual area.

We also sought to characterize the phenomena resulting from contextual variation in the environmental domain. Our conclusions were the following:

• Contextual variation can be characterized as including three different phenomena: modulation, perspectivization, and subconceptualization. These phenomena are additive in that all concepts experience modulation; some concepts also undergo perspectivization; and finally, a few concepts additionally are subjected to subconceptualization.
• Modulation is the type of contextual variation that only alters minor characteristics of a concept that are neither necessary nor prototypical. These alterations are not represented in a terminological definition.
• Perspectivization is the type of contextual variation that results in the change of the level of prototypicality of certain traits of a concept consistently in relation to the general environmental premeaning.
• Subconceptualization is the type of contextual variation in which the extension of the concept in relation to the general environmental premeaning is modified.

Regarding the management of the previously mentioned phenomena and, more generally, the development of guidelines for the creation of flexible terminological definition, we obtained the following conclusions:

• By feeding a term extractor with domain-specific corpora and comparing the output, it is possible to obtain a list of contextually variable candidate terms. This list also contains polysemous terms.
• The extraction of domain-specific contextonyms from corpus has proven to be the most effective way of identifying the most relevant traits that characterize contextualized premeanings.
• Domain-specific definitions in most terminological resources do not always reflect the role of a concept in the domain. Therefore, they are less useful than contextonyms to identify how concepts are construed differently depending on the domain.
• The distinction between the three types of contextual variation described in this work is fuzzy. Consequently, determining the kind of contextual variation must be carried out according to the needs of the user and the characteristics of the resource in which the definitions are to be inserted.
• The main difference between perspectivation and subconceptualization regarding the application in flexible terminological definitions is that a contextualized definition for a subconceptualization represents as necessary a trait that in the general environmental definition does not have necessary status. This is not possible in a contextualized definition of a premeaning that is only a perspective.
• A common problem with perspectivization is that many concepts participate in a broad range of frames, even if only one contextual domain is taken into account. This phenomenon is what we have called hyperversatility.
• A special type of hyperversatility is superordinate hyperversatility, which occurs when a concept has many subordinate concepts and each subordinate participate in different frames. As a consequence, the concept behaves as a superordinate-level concept in the domain in question.
• Both cases of hyperversatility seem to be common phenomena. They might be the reason why frame-based resources tend to avoid or inconsistently represent certain types of concept, such as artifacts.
• The creation of a contextualized definition for a hyperversatile or superordinate hyperversatile concept implies that a summary of the main roles of the concept in the corresponding contextual domain needs to be provided in the definition. This entails a longer process of documentation for the terminographer.
• The identification of subconceptualizations is hindered by the fact that conceptual boundaries are not clear-cut, and by the lack of entrenchment of the general environmental premeaning or of the subconceptualization itself. Furthermore, the hierarchical organization of subconceptualizations does not always match conventionalized domain hierarchies.
• A subconceptualization’s extension can correspond to another concept’s extension. In this case, the contextualized definition should refer to that other concept if it is a relevant concept in that domain.
• If a subconceptualization’s extension corresponds to a concept that is not relevant in that domain or does not correspond to another concept, the contextualized definition will represent as necessary the characteristic that gives rise to the context-specific change of extension.
• The choice of the genus of a flexible definition should be guided by the coherence with the rest of the resource and with the frames that the concept activates in accordance with the applied contextual restrictions.
• Genus candidates for terminological definitions can be obtained by means of hypernymic knowledge-patterns and by extracting them from the definitions of other resources. Each contextual domain needs to have its own conceptual hierarchy that reflects how concepts are categorized in that domain.
• The genus of a definition should convey the role of the concept in the most relevant frame it activates. Superordinate concepts can be classified into qualia roles in order to facilitate the selection of genus.
• Frames can be integrated into definitions by means of framing propositions. Unlike conventional definitional propositions that only link the definiendum to other concept, framing propositions relate the concepts activated in the definition with other concepts.
• Since all generic-specific propositions entail property inheritance, concepts may inherit from several superordinate concepts at the same time in a definition.

Finally, our proposal of flexible terminological definition contributes to the improvement of the quality of terminological definitions for these three reasons:

• With our approach, the user is presented with a definition tailored for the domain that he/she has chosen, multiplying the probabilities that the definition will offer him/her the information he/she needs.
• As the analysis of the definitions from other resources showed, terminological definitions are sometimes not properly adapted to the domain from which they are supposed to be written. In our approach, the knowledge represented in each contextualized definition has been chosen based on the result of the analysis of contextonyms. This ensures that the definition actually reflects how the definiendum is construed in that domain, which might differ from the viewpoint adopted in the environment as a whole or in other environmental subdomains.
• The genus and differentiae selection in flexible definitions is guided by contextonym-based analysis and the aim of inserting the concept in the frames in which it participates. By representing concepts in context-specific hierarchies and frames, the flexible terminological definition provides a knowledge representation that better resembles the human conceptual system than traditional terminological definitions. As a consequence, a flexible definition not only provides more relevant information, but it also accomplishes this in a way that potentially facilitates and enhances knowledge acquisition.

As for future work, we plan to address certain limitations that were encountered during this research. First of all, the way in which conceptual propositions are encoded in EcoLexicon has proven not to be entirely suitable for the representation of the information in flexible definitions. To this respect, an ontology-based system for the management of definitional templates could provide more expressiveness and flexibility with the additional advantage that property inheritance would be automatic, and inferences could be implemented.

This ontology-based system for the management of definitional templates could also encode frames. Nonetheless, further study is necessary to determine the best way in which frames can be formalized in order to support the creation of flexible definitions. The problem of hyperversatility will certainly present a challenge in this respect.

Another limitation that will be addressed in future work is the fact that the creation of flexible terminological definitions is a laborious process that would benefit from being streamlined. More specifically, the contextonym-based analysis of contextual variation is time-consuming. The development of a black list filtering terms that are not informative (such as example, call or require) would speed up the analysis. However, this list would have to be carefully compiled because certain common English words that may apparently be too generic, such as use or part, are also indicative of important relations (for example, functional or meronymic in the case of use and part).

Additionally, contextonym analysis could be further improved if morphological derivations on the lists were grouped together, if complex terms were also included as contextonyms, or by developing a system that would automatically compare contextonyms in different domains. Furthermore, we are aware that the effectiveness of contextonym analysis using SketchEngine as presented in this work is currently limited to non- polysemous terms. A possible solution would involve the use of word- disambiguation techniques to create sense-specific contextonym lists.

The methodology for the extraction of domain-specific definitional knowledge could also be enhanced by the development of other knowledge-pattern-based sketch grammars to extract conceptual relations other than generic-specific ones, such as functional, causal, or meronymic.

Finally, we envision extending the scope of the study of contextual variation and its application to terminological definitions in two ways. On the one hand, since our analysis was limited to ENTITIES designated by simple terms, we propose to include ATTRIBUTES and PROCESSES, as well as concepts associated with complex terms. On the other hand, it would be relevant to analyze contextual variation in specialized domains other than the environment and to incorporate other languages, especially from a contrastive point of view.

We believe that this research has contributed to the understanding of contextual variation in terminology and has shown effective ways of representing it in terminological definitions. Nonetheless, having welcomed context into the realm of terminological definitions, countless possibilities open up, and new research avenues await to be explored.