SOEL: Supporting Ontology Engineering with Large Language Models

The infl uence of Large Language Models (LLMs) is currently boosting and reshaping current practices in numerous fi elds, for example computer science. Due to their high capabilities in natural language understanding, their impact has been particularly evident in the fi eld of Natural Language Processing. However, other areas of Artifi cial Intelligence, like knowledge representation and ontology engineering, may also benefi t signifi cantly from these technologies. However, developing ontologies is still a challenging and time-consuming task even though many tools have been developed by the scientifi c community to assist developers, for example, for formalizing tests to assess requirements, creating human-readable documentation for ontologies, ontology assessment, etc. In general, a signifi cant manual eff ort is still required from ontology developers for conceptualizing, reusing, and validating existing ontologies.To alleviate such ontology engineering tasks, several works have arisen proposing the use of LLMs to support the generation of competency questions, the conceptualization of models from text, the identifi cation of concept alignments, etc.). However, the tasks addressed in previous works are usually defi ned in an heterogeneous manner, with diff erent scope, inputs, expected outputs and metrics.In summary, there are no reference evaluation tasks (e.g., assessment is purely qualitative) reference datasets or benchmarks that help compare existing work against future eff orts. To address this situation, SOEL aims to explore the application and adaptation of (Large) Language Models in Ontology Engineering tasks by creating open reference datasets, defi ning evaluation tasks and comparison benchmarks through adapting state-of-the-art LLM models. Although SOEL’s approach is independent of a specifi c domain, it is planned to assess the results in three specifi c domains in IoT (smart buildings), Linguistics (defi nition of terminologies), and Open Science (description of scientifi c artifacts and experiments).