This study, which can be included in the post-structuralist field, has as its hypothesis the possibility of establishing categories capable of making the following kind of evaluation possible: minimally consensual evaluations of poetry translations. Therefore, when taking into account two or more translations of a poem, submitted to two or more evaluators who adopt uniform categories of analysis, their evaluations, though not identical, will have some relevant aspects in common. The search for consensus is carried out through annotation, one of the activities of Computational Linguistics, which identify and classify a certain linguistic phenomenon, using labels, tags, categories, in a given corpus, so as to achieve a certain goal. The aims of this dissertation are (a) to provide tools for evaluating translations of poems in a minimally consensual way;and (b) to define, systematize and validate the categories of the semantic-lexical level, and to describe and confirm the categories of the formal level (metric and rhymic levels) and of the poetic field of sound resources, so that they all can be used to support minimally consensual evaluations of poetry translation.The research consists of three steps. In the first, annotation is used as a methodology in the search of consensus. In this step, annotations of poems and their translations were made by different annotators. By comparing these annotations searching for consensus, categories were confirmed/validated or reformulated. The consensus allows confirmation and validation, but absence of consensus suggests reformulations and refinements. In the second step, Britto s methodology, with some additional observations, was used to analyse the results from the first step. This methodology aims at verifying whether the most relevant aspects of the formal and semantic-lexical levels and of the poetic field of sound resources (alliterations, assonances, and the like) of the poems have been re-created in the translations. And in the third step, the results of the second one are used in order to produce resources for a possible evaluation of poetry translations: among two translations of the sonnet 130 by Shakespeare, which one would be the most faithful. The hypothesis was proved concerning the formal level and the poetic field of sound resources, but wasn t proved regarding the semantic-lexical level. The validation of categories of the semantic-lexical level can go on in future researches, which may or may not result in the possibility of agreement among evaluations of this level. Aim (a) was achieved. And aim (b) was partially achieved: the definition, systematization and validation of the categories of the semantic-lexical level started in this research, and can go on in future studies. This research can be seen as a contribution not only to the field of poetry translation, but also to Computational Linguistics. Regarding the first, annotation requires interpretations and decision-making processes, thus highlighting the possible interpretations and decision-making processes used during translation. And, concerning the second, the development of a tool for poem annotation, and of a metric for evaluation of poetry translation, using the categories present in this dissertation, may be achieved through a partnership with Computer Engineering.