Context Specificity of Semantic Interference Effects in Production When we talk about related ideas, the coactivation of several semantic concepts leads to the coactivation of their lexical representations. This concurrent activation manifests as an increased difficulty retrieving a name after having recently used a related one (e.g, Abdel Rahman et al., 2011; Kleinman et al., 2015), suggesting an incremental learning mechanism is at play. However, after decades of research, there is a lack of consensus about the underlying nature of this phenomenon. Our research suggests that incremental learning may not apply to a central lexicon. Rather, interference may be more task and context-specific than current theories assume. An influential model of semantic interference implements incremental learning as modulation of conceptual-to-lexical connections based on experience (Oppenheim et al., 2010). Connections to selected names are strengthened, and those to activated but unspoken items are weakened. Figure 1 shows an elaboration of this type of model for both taxonomic (i.e., categorical) and thematic contexts, with thematic contexts requiring context specification. We conducted multi-phase picture naming experiments to investigate if the proposed incremental learning mechanism cumulates across cyclic naming and continuous naming tasks for both taxonomically and thematically related materials. The key experiment consisted of three phases. In the first phase, participants either generated or read 64 picture names with related items spread over quartiles. In the second phase, they cyclically named groups of 4 taxonomically or thematically related items as well as unrelated controls. In a final continuous phase, they generated all names once more. If incremental learning effects apply to a central lexicon, semantic interference should accumulate both within and across phases. We also predicted more interference for strongly activated taxonomically related items compared to thematically related items. Contrary to expectations, we found strong cumulative interference for thematic as well as for taxonomic relations in the initial continuous name generation (Figure 2). This is surprising given the absence of thematic labels, but it predicts carryover for both types of material to future retrievals. In fact, we found that the accrued interference costs carried over to all conditions in a following cyclic naming phase, but interference due to relatedness was only present for taxonomic contexts (Figure 3). Interference in this phase was also noncumulative. This lack of transfer between naming tasks supports the view that interference in cyclic naming is distinct from that found in continuous naming. Further strengthening this conclusion, a second continuous generation phase following cyclic naming showed renewed cumulative interference but no carry-over, such that the experience of relatedness in the cyclic phase had no effect on retrieval in the subsequent continuous phase, not even for the large taxonomic relatedness effect (Figure 4). These findings highlight the limitations of current models of semantic interference as well as the tasks used to study it. The thematic data raise new questions about the continuous naming task, and the taxonomic findings support the view that cyclic naming does not engage incremental learning (Belke et al., 2005). Regardless, the near absence of transfer effects poses a puzzle for current models of incremental learning, given that repeatedly selecting items for production is thought to create central changes in conceptual-to-lexical mappings. The context- specificity of adaptation is however consistent with other recent research on consolidation of new experiences in lexical memory (e.g., Gaskell et al., 2019). Conceptualizing semantic interference as an episodic phenomenon is also supported by recent findings in our own lab that failed to find item-specific transfer of interference across taxonomic and thematic contexts in cyclic naming, suggesting that semantic interference is more context-bound than previously thought. If this conclusion is correct, it raises new questions about the integration of new generative experiences into central lexical representations. Thank you!