Abstract The proposal presented in this study seeks to properly represent natural language to ontologies and vice-versa. Therefore, the semi-automatic creation of a lexical database in Brazilian Portuguese containing morphological, syntactic, and semantic information that can be read by machines was proposed, allowing the link between structured and unstructured data and its integration into an information retrieval model to improve precision. The results obtained demonstrated that the methodology can be used in the risco financeiro (financial risk) domain in Portuguese for the construction of an ontology and the lexical-semantic database and the proposal of a semantic information retrieval model. In order to evaluate the performance of the proposed model, documents containing the main definitions of the financial risk domain were selected and indexed with and without semantic annotation. To enable the comparison between the approaches, two databases were created based on the texts with the semantic annotations to represent the semantic search. The first one represents the traditional search and the second contained the index built based on the texts with the semantic annotations to represent the semantic search. The evaluation of the proposal was based on recall and precision. The queries submitted to the model showed that the semantic search outperforms the traditional search and validates the methodology used. Although more complex, the procedure proposed can be used in all kinds of domains.