William Badecker
Johns Hopkins University
The elements of sentence structure (phrasal dependencies, grammatical agreement, etc.) are inherently local in their formal character, and yet processing such dependencies requires tracking across unbounded amounts of lexical and structural planning. Evidence of how this information is tracked can be found in a familiar performance error: Agreement Attraction (Bock & Miller, 1991; Bock, et al., 2001). On some recent accounts, the syntactic contribution to agreement attraction is explained by positing a feature percolation mechanism: Attraction arises whenever the features of a local noun (e.g., plural number) migrate erroneously from a lexical source to the root node of the Subject phrase. Once an errant feature finds its way to this root node, the mistake is accommodated by subsequent feature copying (or checking) procedures (Vigliocco et al., 1996; Hartsuiker et al., 2001). I will present cross-linguistic experimental evidence (Slovak and English) indicating that the features that participate in agreement attraction remain part of their lexical source. E.g., Gender attraction depends on Nominative/Accusative indeterminacy in both the head and local noun, even though this morpho-syntactic ambiguity should be eliminated if the case feature percolates to an intermediate node that requires one or the other feature value.
Putative effects for syntactic distance (Hartsuiker et al, 2001; Frank et al., 1997) have also been offered as evidence for the percolation based production model. Results from our own studies and from Nicol (1995) show that the apparent effect derives from competition within planning increments.
I present an alternative model in which the agreement calculation respects the lexical integrity of sources. Lexical (and phrasal) agreement sources are flagged by activation-based tracking mechanisms, but this give rise to a recurrent weakness: Activation can only distinguish an element against a background of comparative dormancy. Other sources of activation for local nouns may cause them to intrude on the agreement calculation. In addition to effects for morpho-syntactic indeterminacy, this model predicts another finding: Attraction is more likely when a lexical element in a conjoined local NP is itself plural (as in "the report of the explosion(s) and the fire") since it adds an additional source of distracting plural activation. This lexical source contributes over and above that of the conjoined NP. In contrast, in percolation-based theories, conjoined NPs will affect the agreement calculation in a uniform manner, showing plural features externally, regardless of their internal content.
Percolation fails to capture these and other properties of agreement attraction, while the activation based model of tracking provides a natural explanation.