This thesis aims to understand the relationship between the absence of plural marking in the word itself (non-marking) and the eventual migration between plural patterns in the mental representation of the word groups in potential metaphorical context, ending in the following: -ão and -au, -éu, -al and -el, especially in (sub)groups whose expected pluralization would not be the most frequent for their morphological group or in words with low frequency of occurrence. Therefore, the representation of plural in the mental lexicon would be formed by two faces, which feed each other from probabilistic information stored in memory. In the first research front, the differences between the plural prescriptions for these groups in the standard pattern of Portugal and Brazil were verified from grammar and orthography manuals written by the Portuguese people since 1536 and by Brazilians from 1806 onwards. From that investigation, it was found that aspects of diatopic and diachronic orders directly influenced the prescription of each country, especially from the beginning of the nineteenth century, as well as it was observed that the current norms prescribed then even by grammarians of the same nationality presented criteria based on varied perspectives of the cultured language. On the second research front, based on theoretical-methodological assumptions of Dialectology (Rossi, 1967; Chambers; Trudgill, 1994; Cardoso, 2010), Sociolinguistics (Weinreich; Labov; Herzog, 2006[1968]) and Exemplar Theory (Bybee, 1985, 1988, 1991, 1996; Johnson, 1997; Pierrehumbert, 2001; Bybee, 2002, 2007), the mental representation of certain plural patterns and their correlation with the variables order of plural of the same words, frequency of occurrence, word, mesoregion of Bahia, sample/year, age group, level of education, level of contact with the public in the occupational market and degree of monitoring was verified. In the impossibility of applying the logistic regression test to all (sub)groups, considering the distributions of the data on the two sides of the plural representation separately and the significant number of controlled predictive variables, the chi-square test was applied using the R Platform (R Core Team, 2018). Thus, the words of these groups in the sample of the Linguistic Atlas of Brazil Project and in the Production and Perception Questionnaire – applied almost two decades after that one – were controlled, totaling a general sample of 124 native informants from 22 municipalities of Bahia State, distributed in its seven mesoregions. Based on this control, we have highlighted the findings of the correlation between the strengthening of the non-marking between the words with low frequency of occurrence and the consequent weakening of other plural patterns for those items in the mental representation of the informants, as well as, in general, the implementation of changes in some plural patterns of some words,  a reduction in non-marking and strengthening of double marking, -ons and -is among the residents of the coastal perimeter, with greater contact with the public in the work environment and level of education, as well as in contexts with a higher degree of monitoring, revealing how well these plural patterns are evaluated by Bahians, regardless of whether they are the strategies prescribed by the standard pattern.