Whamit!

Many grammatical phenomena are replicative in the following sense: the featural information pertaining to some element A in a syntactic domain D is repeated on some other element B which stands in a c-command relation with A in D. For instance, in cases of clausal φ-agreement, the φ-features of a clausal argument (subject and/or object) are replicated on the clausemate verb. The syntactic operation of Agree in Minimalism (Chomsky, 2001) is specifically designed to capture grammatical replicativeness. This follows from the idea, hardwired into Agree, that syntactic relationships are fundamentally asymmetric, involving dependencies between an independent element and a dependent counterpart. The idea is that the defectiveness of a probe for some (potentially unary) set of features α triggers valuation/checking of α, under c-command, by a local goal which is specified for α. The only possible output of such an Agree operation is a representation involving replication of α across the probe & goal. Under a strongly Minimalist worldview, it is further assumed that all syntactic relationships are derived by Agree, understood in the sense above. This yields the following state-of-affairs: 1. All syntactic relationships are derived via Agree, and; 2. The only possible output of Agree is feature-replication across the Agreeing elements. Ergo. All syntactic relationships must be featurally replicative.

In this talk, I will argue that such a scenario strongly undergenerates. Partially and fully non-replicative processes in grammar do exist – a fairly uncontroversial point. Perhaps more controversially, I argue that a (proper) subset of non-replicative phenomena are (narrowly-)syntactic in nature (piggybacking on prior work in Bobaljik, 2008; Preminger, 2014; Levin, 2015; Yuan, To Appear, showing that case- marking (i) feeds φ-agreement; (ii) is syntactic, and (iii) involves case-competition, not case-licensing). Such cases are fatal to the strongly Minimalist world-view described above since they clearly cannot be derived under Agree, as it stands.

To accommodate these problematic cases, I develop a radically revised model of Agree (renamed RELATE to avoid ambiguity) which abandons the idea that syntactic relationships are (asymmetric) dependencies. RELATE is grounded on the notion that syntactic dependencies are restricted by a generalized OCP constraint that two syntactically local objects cannot be featurally indistinguishable at the interfaces (along the lines of Richards, 2010, with significant deviations). The corollary to this is that a syntactic link between two nodes A & B for some feature α must output a representation where A & B remain distinguishable at LF/PF wrt. some relevant feature β, where β ̸= α. I show that the new powerful algorithm also accurately predicts some long-observed replicative vs. non-replicative differences at LF and PF between local and long-distance anaphora crosslinguistically (Faltz, 1977; Jackendoff, 1992; Lidz, 2001; Reuland, 2011) as well as distinctness effects in predicate-nominal and small clause constructions (Longobardi, 1994; den Dikken, 2007). I believe the model may also be fruitfully extended to capture certain cases of switch-reference (e.g. in Washo, Arregi and Hanink, 2021) and cases of so-called “subset control” (Ackema and Neeleman, 2013) including of partial obligatory control.