SÜNİ İNTELLEKTLƏ HÜQUQİ TƏRCÜMƏDƏ AXICILIQ VƏ MƏNA DƏQİQLİYİ

Abstract

Generative artificial intelligence can produce fluent Persian drafts of Azerbaijani legal sentences, but fluency alone does not establish that legal meaning has been preserved. This theoretical article develops a clause-level contrastive reading procedure for examining that distinction. It draws on legal translation scholarship and translation quality assessment frameworks, without testing any particular system. The analysis focuses on five relations that may be altered without making the Persian sentence ungrammatical: obligation and permission, the roles of legal actors, the order of events, negation, and the scope of a restrictive expression. Five short Azerbaijani–Persian pairs constructed for this article compare a meaning-preserving formulation with a fluent alternative that expresses a different rule. These pairs are analytical illustrations rather than recorded AI outputs, so they support no claim about the frequency of model errors. The proposed procedure begins by identifying the actor, action, modal force, condition, negation, and restriction in the Azerbaijani source. The reviewer then checks whether the same relations hold in the Persian draft before considering stylistic naturalness and terminology. The article's contribution is a limited and teachable way to distinguish target-language fluency from source–target meaning correspondence. It can structure the review of AI-assisted legal translation drafts and classroom discussion, while leaving full-document interpretation and empirical model evaluation to further work.