Linguistics in Göttingen - A platform for empirical and theoretical linguistics

Hedde Zeijlstra (Göttingen): Upward Agree

In Zeijlstra (2012), it was argued that the syntactic operation Agree can only take place between a probe that carries an uninterpretable feature and a goal that carries a matching interpretable feature when the goal is the closest potential goal that c-commands the probe. Similar ideas have been proposed by Wurmbrand (2012a,b, 2014). This version of Agree has been dubbed Upward Agree (UA), as opposed to Downward Agree (DA, where the probe c-commands the goal). Zeijlstra argues for UA on the basis, first, that some instances of syntactic agreement can only be analyzed as instances of UA (i.e. they cannot be analyzed as instances of DA); and second, that it is possible to analyze all other instances of Agree as instances of UA. These latter cases include both long-distance φ-agreement and Agree-driven movement effects ("EPP" effects), which have elsewhere been discussed as evidence in favor of DA.
UA has received some criticism, most notably by Preminger (2013), who presents and discusses two cases of long-distance φ-agreement, one from Tsez (Polinsky and Potsdam 2000), and one from substandard Basque (Etxepare 2006, Preminger 2009, 2011, 2013). Preminger argues that these instances of long-distance agreement cannot be accounted for by Zeijlstra's proposals, and so argue against a redefinition of Agree in terms of UA.
This paper addresses the question of whether UA can account for cases of apparently "downward" φ-agreement, including phenomena such as long-distance agreement. Contra Preminger, we argue that the conclusion that UA is untenable as a universal mechanism for agreement is unjustified - rather than suggesting that the UA proposal should be weakened, we argue that the apparent counterexamples to UA vanish once we adopt a more restrictive version of UA. We argue that this more restrictive version of UA, slightly modified from Zeijlstra (2012), is in fact uniquely able to account for notable asymmetries between upwards and downwards φ-agreement. These asymmetries include a typological link between downwards φ-agreement and Case, as already proposed by Baker (2008), and the observation that downwards φ-agreement is often in some sense defective as compared to upwards φ-agreement, as observed going back to work on Spec-Head feature checking, and noted in connection to the UA debate by Bjorkman (2011). We also show that this novel version of UA fares better in explaining EPP-effects.