Speech planning involves different steps in order to transform a conceptual message into speech. These include establishing structural relations among constituents (i.e., relational information), and selecting the appropriate lexical items to convey the intended message (non-relational elements). These two types of information are often interrelated and confounded. Therefore, the precise way relational and non-relational information are computed when undertaking linguistic encoding is still not clear. In this talk I will explore how the pre-linguistic message undergoes linguistic encoding, and what kind of information (relational or non-relational) is prioritized in doing so. To do so, I will present data from two projects aimed to teasing apart the effects of word order in sentence planning. First, I will present data in Kaqchikel language, a Mayan language spoken in Guatemala, which canonical word order is VOS, with SVO being widely used. After it, I will present data in the production planning of Relative Clauses in Japanese (a head-final language), in comparison with Spanish (a head-initial language). Sentence planning was explored by means of the eye-tracking method while participants described pictures. Results in both cases point to a planning process that prioritizes structural relations over access to lexical elements in order in the planning of complex structures, with room for flexibility when the grammar of the language, or the order selected, allows so. Implications of these results for models of speech production will be discussed.