In organizations, the business process modeling is of big importance to report, understand and automate processes. These organizations usually come unstructured documents and difficult to understand by analysts. The extraction models or fragments process from textual descriptions may contribute to minimizing the effort required to process modeling. In this context, this paper proposes an approach to generate text-oriented process from the text in natural language. This paper proposes to investigate the structure of a text in natural language must show to from this one can extract fragments or process models. Based on the study of grammar classes accomplished in the context of this research, there is no established or standardized order in which the grammar classes should be shown in the text in natural language. Thus, preliminary results of this research show that by mapping rules and correlations between words represent the grammatical classes indicate a process element, through keywords and/or verb tenses.