Objective of present paper is to classify the comma uses focused in grammatical aspects and from a computational linguistic perspective is proposed. From this objective, some theoretical aspects based on grammatical criteria are showed, and the following classification of the comma functions are established: (i) indicator comma: it points enumerations and ellipsis, (ii) bounding comma: it delimits incidental clauses (appositions, vocatives, etcetera), and (iii) comma for disambiguation: it avoids confusion in expressions that could present more than a interpretation. Afterwards, a formalization and a computational implementation are made with the objective of getting a method of automatic detection for comma functions. In relation to the computational work, the software Smorph and Post Smorph Module (MPS) were used. Smorph analyzes the characters chain morphologically, giving an output with the morphological and categorical assignation for each occurrence according to the features declared. The Smorph output is the MPS input, and MPS analyzes the lemmas chain given by the morphological analysis, with composition, decomposition, and correspondence rules declared by the user. It was worked with a corpora composed of journalist articles that had 10,000 words. Results showed measures of precision, recall and f-measure that are adequate enough to confirm the validity of the proposed classification.