The inverse frequency effect in syntactic priming

Abstract

Repetition is a central phenomenon of behavior. In the language sciences, a common form of such repetition is syntactic priming, a tendency to reuse or better comprehend a structure because of its structural similarity to a previously experienced structure ("prime") (Bock, 1986). This dissertation investigated two alternative proposals about the mechanism behind syntactic priming: (a) residual activation models that emphasize short-term residual activation from a memory representation of a priming structure (Pickering & Branigan, 1998) and (b) implicit learning models that assume adults continue to learn mappings between message-level representations and abstract syntactic representations (e.g., Bock & Griffin, 2000; Chang, Dell, & Bock, 2006). Implicit leaning models learn more from input sentences that are unexpected given the model's prior experience. Importantly, predictions of the implicit learning models and residual activation models differ with respect to the magnitude of priming effects from an infrequent prime structure versus from a frequent prime structure. According to residual activation models, each exposure to a syntactic structure leads to a constant increase in the activation of that structure, which in turn increases the probability of producing or the ease of comprehending that structure. The amount of activation each structure receives is independent of its previous frequency, and thus there should be an equal amount of priming for the frequent structure and the infrequent structure. Implicit learning models, on the other hand, predict the infrequent structure generates a larger error signal when expectations are violated, and thus there should be greater priming for the infrequent structure than for the frequent structure. This is referred to as an inverse frequency effect. This dissertation used the inverse frequency effect as a vehicle to better understand the mechanisms underlying syntactic priming in both spoken production and reading comprehension. There is limited evidence about the inverse frequency effect in production (e.g., Bernolet & Hartsuiker, 2010; Ferreira, 2003; Hartsuiker, Bernolet, Schoonbaert, Speybroeck, & Vanderelst, 2008) and even less in comprehension (e.g., Kim, Kathleen, & Tanenhaus, 2014). Moreover, much remains to be understood on how the inverse frequency effect works on different type of constructions. Three studies examined this topic in two types of constructions (each involve two alternative structures). In production (Experiment 1), the prime sentence immediately preceded the target preamble within each prime-target pair. There were three prime conditions: an infrequent structure, an intransitive sentence that served as a neutral baseline, and a frequent structure. Participants were instructed to read each prime sentence aloud and, for target preambles, to create a sentence completion. Our data showed that the infrequent prime structure created greater syntactic priming relative to a neutral baseline, whereas the frequent prime structure created less. This evidence of an inverse frequency effect in production suggested an implicit learning mechanism and cannot be explained by residual activation models alone.A further analysis on the production data separated the inhibitory effect (decreasing the competing structure) from the priming effect (increasing the primed structure) for a given prime. For one type of the constructions (a sentence-complement structure SC and a direct object structure DO), our data revealed an inhibitory effect in syntactic priming - the infrequent primes SC not only increased the production of its own structure, but also decreased the production of its competing structure DO. Experiment 2 and 3 examined if similar syntactic priming patterns and the inverse frequency effect in production could be observed in self-paced reading studies. The stimuli in Experiment 2 were the same as Experiment 1's, except that two different endings were added onto each target fragment, one for each alternative for the construction (frequent target and infrequent target). Experiment 3, an extension of Experiment 2, attempted to further uncover how prime verb bias could moderate the magnitude of syntactic priming by adding half SC-biased priming verbs whose verb-specific structural preference did not match with the general structural preference. Even though readers were sensitive to basic factors affecting processing difficulty (e.g., word length, item order, etc.), no priming effect was observed at all in these two comprehension studies. Although we ended up being unable to examine syntactic priming patterns in comprehension, the inverse frequency effect in production (Experiment 1) helps adjudicate among the main competing accounts of the mechanisms underlying syntactic priming. This also provides the first evidence of the inhibitory effect in syntactic priming and thus calls for a competition-based structural selection component in current syntactic priming models. --Author's abstract