Although advances in diffusion tensor imaging have enabled us to better study the anatomy of the inferior longitudinal fasciculus (ILF), its function remains poorly understood. Recently, it was suggested that the subcortical network subserving the language semantics could be constituted, in parallel with the inferior occipitofrontal fasciculus, by the left ILF, joining the posterior occipitotemporal regions to the temporal pole, then relayed by the uncinate fasciculus connecting the anterior temporal pole to the frontobasal areas. Nevertheless, this hypothesis was solely based on neurofunctional imaging, allowing a cortical mapping but with no anatomofunctional information regarding the white matter. Here, we report a series of 12 patients operated on under local anaesthesia for a cerebral low-grade glioma located within the left temporal lobe. Before and during resection, we used the method of intraoperative direct electrostimulation, enabling us to perform accurate and reliable anatomofunctional correlations both at cortical and subcortical levels. In order to map the ILF. Using postoperative MRI, we correlated these functional findings with the anatomical locations of the sites where language disturbances were elicited by stimulations, both at cortical and subcortical levels. Our goal was to study the potential existence of parallel and distributed language networks crossing the left dominant temporal lobe, subserved by distinct subcortical pathways--namely the inferior occipitofrontal fasciculus and the ILF. Intraoperative stimulation of the anterior and middle temporal cortex elicited anomia in four patients. At the subcortical level, semantic paraphasia were induced in seven patients during stimulation of the inferior occipitofrontal fasciculus, and phonological paraphasia was generated in seven patients by stimulating the arcuate fasciculus. Interestingly, subcortical stimulation never elicited any language disturbances when performed at the level of the ILF. In addition, following a transient postoperative language deficit, all patients recovered, despite the resection of at least one part of the ILF, as confirmed by control MRI. On the basis of these results, we suggest that the "semantic ventral stream" could be constituted by at least two parallel pathways within the left dominant temporal lobe: (i) a direct pathway, the inferior occipitofrontal fasciculus, that connects the posterior temporal areas and the orbitofrontal region, crucial for language semantic processing, since it elicits semantic paraphasia when stimulated; (ii) and also possibly an indirect pathway subserved by the ILF, not indispensable for language, since it can be compensated both during stimulation and after resection.