Which Subject Islands Will the Acceptability of Improve with Repeated Exposure?

Since Chomsky & Miller (1963) and Chomsky (1965), linguists have recognized that a grammatical sentence can be deemed as unacceptable for many extra-grammatical reasons, including memory limitations, context, and stylistic factors. More recently, evidence is mounting that sentences of moderate or even low acceptability can become more acceptable with repeated exposure (Hiramatsu, 1999; Snyder, 2000; Luka & Barsalou, 2005; Francom, 2009; Crawford, 2011a,b), a phenomenon that is sometimes referred to as satiation. In other words, sentence acceptability can increase with repeated exposure. An analogous effect seems to arise in on-line sentence processing. As Braze (2002) and Hofmeister et al. (2013) show, reading times for sentences with low acceptability can decrease with repeated exposure. In this work we focus on the controversy surrounding the existence of satiation in Subject Island violations. Whereas Snyder (2000), Hiramatsu (1999, 2000), and Francom (2009) found evidence for satiation in Subject Islands, others have failed to replicate this result (Sprouse, 2009; Crawford, 2011a,b). In particular, Sprouse (2007, 2009) argues that the Subject Island satiation effects found in Snyder (2000) might be due to a confound, created by the fact that Snyder’s design was not properly balanced. Since the ungrammatical sentences outnumbered the grammatical ones, this could have led participants to employ an equalization strategy, balancing out their yes/no responses. Sprouse (2007:123) goes on to argue that non-satiation is the expected result: since ungrammatical sentences have no licit representation (they cannot be constructed by the available mental computations), it follows that extra-grammatical factors that affect the acceptability – and are predicated on the existence of a representation – such as syntactic priming, should not affect the acceptability of ungrammatical sentences. In other words, if there is no licit representation for a subject island violation, then there is nothing to satiate. We show that certain Subject Islands reliably satiate, including subject islands with non-derived subjects, which indicates that Sprouse (2009) and Crawford (2011) have overstated the generality of their findings. Our results are consistent with Francom (2009) and Hiramatsu (2000), who observe that the satiation effect is not an across-the-board phenomenon; satiation does not occur for all sentence types equally. For example, Subject Island satiation seems to be harder with transitive verbs (e.g. what does Clinton worry that an ally of will boycott the airline?). This finding is corroborated by the reading time studies in Kravtchenko et al. (2009); Polinsky et al. (2013), where unaccusative subjects are shown to make for weaker islands than other types of predicates. The number of exposures may also play a role in satiation. For example, Hiramatsu (2000) did not find satiation effects with 5 exposures, but did find them with 7 exposures. And finally, the complexity of the items may also be important. Hofmeister et al. (2013) report that satiation vanishes when items are made too complex. The structure of this paper is as follows. In Section 2 we briefly discuss Subject Island effects and their gradient acceptability. In Section 3 we describe several experiments where satiation of Subject Islands was reliably obtained. These experiments rule out the possibility for confounds like the ‘equalization strategy’ discussed by Sprouse (2009). Section 4 discusses the findings.