RELATE-ing assimilation & dissimilation in syntax

Sandhya Sundaresan

Stony Brook University

Lattimore 201

Grammatical phenomena are fundamentally distinguished wrt. whether they are assimilatory or dissimilatory. The former involve operations which result in some (potentially unary) set of features α being replicated on two elements in a local domain: e.g. vowel harmony in phonology (see e.g. Hayes, Zuraw, Siptár, and Londe, 2009, a.m.o.) or variable-binding in semantics. The latter involve operations which result in two local elements bearing features α and β where α 6= β, e.g. OCP effects in phonology (McCarthy, 1986) or Condition B & C effects in semantics (Reinhart, 1983; Safir, 2004).

Curiously, operations in syntax have been (more or less tacitly) assumed to be only assimilatory, not dissimilatory. In fact, the formal operation of Agree in Minimalism (Chomsky, 2001) is hardwired to always and only yield outputs that are assimilatory in nature. Agree is based on the idea that syntactic relationships are asymmetric dependencies between a probe (deficient for some feature α) & a goal (bearing α): feature-copying of α onto the probe by the goal redeems this asymmetry. Agree is thus crash-proof: the syntax will output all and only those outcomes involving replication of α across the probe & goal. Under a strongly Minimalist worldview, whereby Agree and Merge are the only syntactic operations, this predicts that all syntactic relationships should be assimilatory. However, this prediction is falsified in the case of phenomena which are both syntactic and dissimilatory (e.g. Preminger, 2014; Levin, 2015; Yuan, To Appear, for case). It is arguably also falsified in other instances, e.g. in cases of local anaphora & predicate-nominal constructions (Longobardi, 1994; den Dikken, 2007), & Person Case Constraint effects (Bonet, 1991; Anagnostopoulou, 2003; Nevins, 2007) crosslinguistically.

In this talk, I develop a radically revised model of Agree (renamed RELATE), which can derive both assimilatory & dissimilatory phenomena in syntax. RELATE abandons the idea that syntactic relationships are (asymmetric) dependencies between a probe & matching goal. Rather, two syntactic objects A and B may be non-trivially linked under RELATE if one is valued & the other is unvalued (as under standard models of Agree), but also if both are unvalued. The only restriction underlying RELATE is the following generalized OCP constraint (along the lines of Richards, 2010, with significant deviations): a syntactic relation between two objects A and B may not output a representation where A and B are featurally indistinguishable for the purposes of interpretation at LF & PF. Corollary: 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 6= α. I show that the new powerful algorithm not only allows us to present a unified derivation for assimilatory processes like φ-agreement and dissimilatory ones like (dependent) case-marking, but also accurately predicts some long-observed assimilatory vs. dissimilatory differences at LF and PF between local and long-distance anaphora crosslinguistically (Faltz, 1977; Jackendoff, 1992; Lidz, 2001; Reuland, 2011).