This thesis investigated the relationship between sentence production and comprehension. A combination of behavioural and neuroimaging techniques were used to examine the extent to which sentence production and comprehension engage common or distinct mechanisms, with specific focus on the processes engaged by semantic/syntactic competition. Behavioural studies in Chapter 2 indicated that high-competition cases were more difficult to understand and produce than low-competition cases, and that difficulty varied as a function of the number of alternative associations entertained during performance in both tasks. In Chapter 3, an fMRI study indicated that production and comprehension shared a common competition mechanism within left inferior frontal gyrus (LIFG). However, they engage distinctive networks that interact with LIFG, with production eliciting a larger network including areas involved in sentence planning and memory retrieval. Further asymmetries across tasks were revealed in Chapter 4, in which behavioural results and neural networks were compared across adults and adolescents. This study also demonstrated the occurrence of shifts in the neural networks involved in competition resolution throughout development, thereby providing a strong link between poor behavioural performance and the underdevelopment of pre-frontal inhibitory mechanisms in adolescents. Chapter 5 used an improved experimental paradigm from that in Chapters 3 and 4. The results showed that production elicits more activity than comprehension in the dorsal language route thus confirming the engagement of task-specific control processes. Interestingly, this study also revealed a common area of LIFG involved in both tasks, but also differences within LIFG, suggesting the possibility of task-specific circuitry. Together, the findings suggest that production and comprehension share fronto-temporal areas that store and manage abstract linguistic associations between words and structures. However, they differ in the manner in which linguistic information is used, as is evident by the recruitment of distinct networks. Implications for models of language processing are discussed.