Communication is fundamental to the human condition but is impaired in life-altering ways for more than 2.4 million people with aphasia (PWA) in the United States.1 PWA identify discourse-level communication (i.e. language in use) as a high priority for treatment.2 The central premise of most aphasia treatments is that restoring language at the phoneme, word, and/or sentence level will generalize to discourse.3 However, treatment-related discourse outcomes are modest, poorly understood, and vary greatly between PWA.4,5 In response, this proposal will investigate mechanisms of action and cognitive predictors of multilevel discourse outcomes in Semantic Feature Analysis (SFA). While SFA is one of the most common treatments for aphasia,6–8 it is unclear how SFA improves discourse-level communication, and for whom.9 SFA improves word-level communication by (a) spreading activation across semantic networks,10 thus improving word-finding ability (restorative mechanism) and/or (b) training self-cueing and strategic responses to instances of anomia, thus improving language use (compensatory mechanism).11–13 Preliminary analysis found that intensive SFA improved the informativeness14 of discourse overall, and for 19/44 PWA (5I01RX000832-05, MPI: Dickey). However, informativeness alone cannot differentiate which mechanism(s) are responsible for discourse generalization in SFA. This study proposes a multilevel discourse analysis15,16 of archival discourse outcomes across two high- intensity SFA studies (5I01RX000832-05 and 1R01DC017475-01A1; combined n = 60). Multilevel discourse measures include: semantic and verbal paraphasias, predicate-argument structure, lexical diversity, informativeness, and global coherence.16,17 Aim 1 will evaluate the role of restorative and compensatory mechanisms on multilevel discourse outcomes. Patterns of correspondence and non-correspondence across discourse outcome measures will delineate the role of restorative and compensatory mechanisms in treatment-related discourse improvements. Aim 2 will examine cognitive predictors of multilevel discourse outcomes: semantic memory, episodic memory, attention, and executive function. Examining the relationships between cognitive factors and discourse outcome measures will establish how each cognitive process supports 1) restorative and/or compensatory mechanisms and 2) improvements at multiple levels of discourse. The proposed study has the potential to improve the clinical effectiveness and implementation of SFA by determining how and for whom SFA improves discourse-level communication. Ultimately, improving discourse- level communication may help to reduce the profound communication and psychosocial consequences of aphasia. This proposal will integrate training in discourse analysis, multivariate statistics, comparative effectiveness research, and professional development to accelerate the applicant’s independence as a clinician- scientist in pursuit of his long-term research goa...