The capacity to flexibly combine words into a virtually infinite number of well-formed sentences is uniquely human and rests upon syntax (Berwick et al., 2013; Friederici, 2017). At the formal level, linguistic units are recursively combined into constituents according to syntactic rules and grammatical categorical information (Chomsky, 1995). At the neural level, neuroimaging studies support the role of Broca’s area in syntactic composition purely based on grammatical categorical information (Goucha & Friederici, 2015; Zaccarella & Friederici, 2015b). Furthermore, both the latency and attention-independent nature of the Early Left Anterior Negativity (ELAN) event-related potential (ERP) component elicited by syntactic categorical violations indicate that the analysis of syntactic categorical information is a fast and automatic process (Friederici, 2011). However, at present it remains unclear how syntactic composition takes place at the cognitive level. The present work examines how Broca’s area might incrementally build syntactic structures. Two hypotheses, namely categorical prediction or bottom-up integration, are investigated focusing on basic two-word structures, in order to highlight compositional processes while limiting extra-linguistic (e.g., working memory) demands (Pylkkänen, 2020; Zaccarella & Friederici, 2015b). Study 1 adapted a two-word ERP paradigm with syntactic categorical violations (Hasting & Kotz, 2008), shown to elicit a two-word equivalent of the ELAN component termed Early Syntactic Negativity (ESN). Study 1 successfully elicited an ESN effect with the stimulus list and timing of events employed. Study 2 tested the causal role of Broca’s area in syntactic categorical prediction, by simultaneously employing online transcranial magnetic stimulation (TMS) and ERP measures. Crucially, the ESN effect was not affected by the functional disruption of Broca’s area by TMS at the predictive stage. Study 3 tested whether evidence for syntactic categorical prediction can be found at the behavioral level, employing a two-word masked syntactic priming paradigm (Berkovitch & Dehaene, 2019; Pyatigorskaya et al., 2023) to focus on automatic linguistic processes. A significant masked syntactic priming effect was observed, which however did not stem from facilitation in processing well-formed structures compared to a neutral context but rather from inhibition of the ungrammatical structures compared to a neural condition. This result converges on earlier observations made at the agreement level (Friederici & Jacobsen, 1999) and is not compatible with a central role of predictive processes in automatic categorical analysis. Finally, a comprehensive Review of two-word studies on syntactic processing characterized syntactic composition as a two-step and rule-based combinatorial process, achieved with a central role of Broca’s area and the posterior temporal lobe in the representation and combination of syntactic features, relying on efficient bottom-up integration rather than top-down prediction in the formation of constituents. Building on the results of the three studies and the review work, the present thesis puts forward a neurocognitive model of two-word incremental syntactic composition upon which future research can be based.