Despite the growing interest in endangered languages, relatively little attention has been paid to the ways in which the structure of these languages is conditioned by the language shift setting, even among conservative older speakers. This thesis investigates how the social circumstances of language endangerment---which include disrupted intergenerational transmission, loss of a cohesive speech community, pressure to master a new dominant language, and stigmatization of the traditional language---can have significant grammatical effects. I investigate morphosyntactic variation among different groups of speakers of the highly endangered polysynthetic indigenous language Chukchi, which is spoken in northeastern Russia. Following a series of disruptive social and educational policies implemented in the mid-20th century, speakers of Chukchi rapidly shifted to Russian; today, virtually all speakers are bilingual in Russian and transmission of Chukchi to children has ceased entirely. In order to systematically compare linguistic patterns among speakers of different backgrounds (proficient older speakers, attriting speakers, and young L2 or heritage learners), I utilize a combination of traditional fieldwork techniques and controlled experimental production tasks. I focus on several distinct reflexes of the encoding of argument structure, which cuts across multiple morphosyntactic domains and thus affords us the opportunity to examine not only individual grammatical changes due to language shift, but also system-wide grammatical restructuring that can only be seen as a direct result of the modern sociolinguistic setting. Modern Chukchi speakers evidence variation across the following domains: agreement marking, morphological and syntactic ergativity, valency-changing derivational morphology, verbal and nominal incorporation, and argument drop. While older, highly proficient speakers display patterns that are largely consistent with existing grammatical descriptions, attriting speakers and L2 speakers show deviations from the expected patterns, though not always in identical ways. Attriting and L2 speakers reanalyze agreement marking across different dimensions, and while both groups make little productive use of verbal derivation and incorporation, this tendency is more pronounced among L2 learners. However, these varieties are alike in that the changes present in the grammars of these speakers are entirely consistent with cross-linguistic tendencies and a shift away from a polysynthetic configuration. Furthermore, while similar changes in other moribund languages have often been characterized as "linguistic loss," the Chukchi data show that as certain features are lost, speakers innovate new patterns to replace them, often making use of existing resources in the language (rather than borrowing from or replicating patterns in the contact language).