It has often been suggested that the syllable is the basic unit underlying the organization of speech production planning in adults (see among others, Levelt’s Mental Syllabary hypothesis or Guenther et al’s GODIVA model). The goal of this study is to quantitively evaluate differences in coarticulation patterns within and across syllables using GEPPETO-OC (GEstures shaped by the Physics and by PErceptually-oriented Targets combined with Optimal Control, i.e an Optimal Control version of the GEPPETO model). Unlike other speech production models, GEPPETO-OC controls a biomechanical model of the vocal tract, and, as such, integrates the crucial influence of orofacial biomechanics on the articulatory movements and, then, on the spectro-temporal characteristics of speech. Furthermore, it integrates optimality because motor commands are computed by minimizing a sensorimotor cost combining both neurosmuscular effort (global change in muscle commands) and sensorial penalty in reaching phonemerelated goals defined in a multisensory space associating auditory, proprioceptive and tactile inputs. Thus, coarticulatory effects are expected to emerge in this model since trajectories reflect both a representation of biomechanics and a higher-order effort minimization. In order to evaluate variations in coarticulation patterns, three hypotheses are tested and compared in bi-syllabic words: (1) cost optimization based on the planning of four phonemes (C1V1C2V2) (phoneme-based planning); (2) separated intrasyllabic cost optimization for two syllables C1V1 and C2V2 (Intrasyllabic planning) (3) optimization based on the optimal combination of optimal intrasyllabic planning (optimal mental syllabary based planning). In order to illustrate the work in progress, preliminary results are presented, which correspond to 20 simulations of sequence /taki/. Figure 1 shows the distribution of the cost values after optimisation. As expected cost values in simulations based on Hypothesis 1 (Hyp.1) are significantly smaller than those based on Hypothesis 2 (Hyp.2). In case of Hypothesis 3 (Hyp.3) the average is similar to the one of Hyp.1, but the variance is clearly smaller. Further investigation is required to understand this last observation. Fig 2 shows the tongue position reached for consonant /k/ in the three experimental conditions. In Hyp.2, /k/ is articulated more front than in the context of Hyp.1 and Hyp.3. This is consistent with the planning hypotheses: in Hyp.1 and Hyp.3 the posterior phonetic context associated with vowel /a/ influences the articulation of consonant /k/ in the following syllable (retentive coarticulation); in Hyp.2 there is no interaction across the syllables, in such a way that the optimal planning within syllable /ki/ alone is not at all affected by the preceding context. Taking into account the variance of the predictions with the set of 20 simulations no significant difference exists between Hyp.1 and Hyp3. for the articulation of consonant /k/. Further analysis on a larger amount of bi-syllabic words are in progress to analyse more in depth coarticulation patterns throughout the whole sequence, in particular at the level of muscle activations. In general, we obviously expect stronger anticipatory effects of /i/ during the production of /a/ in the phoneme-based planning (Hyp.1) than in the purely intrasyllabic planning (Hyp.2). More interestingly, we expect from the analysis of muscle recruitments and time course of movements in different parts of the tongue to clarify how optimal planning based on mental syllabary (Hyp.3) can be differentiated from optimal phoneme-based planning. Comparisons of the obtained simulations with articulatory and physiological data could help disentangling these two hypotheses of speech motor planning.