Abstract

Native speakers occasionally report incorrect interpretations of language inputs. For instance, when presented with implausible passive sentences, such as "The dog was bitten by the man", participants sometimes report incorrectly that the dog is the agent of the sentence (Ferreira, 2003). According to the good-enough approach, this error occurs because the language comprehension system sometimes relies on rapid heuristics in favour of structure to determine the meaning of a sentence. This thesis tested the hypothesis that the passive misinterpretation effect is not an error of comprehension, but arises only after interpretation, when participants formulate a thematic judgment of the sentence. It was hypothesized that such thematic judgments involve metalinguistic thinking that can be decoupled from language comprehension processes. This thesis reports the results of two research manuscripts designed to test this hypothesis. In both manuscripts, the methodology involved comparisons of L1 and L2 speakers in single and dual task conditions, which required participants to either report the agents of aurally presented sentences (single task), or to do so while also maintaining six digits in WM (dual task). Because L2 comprehension is more dependent on metalinguistic knowledge than L1 comprehension (Paradis, 2004), it was hypothesized that language status would modulate thematic judgment errors. In Manuscript 1 (Chapter 2), I report the results of two between-subject experiments—one single task and one dual task—which showed that indeed language status modulated the effect. In the single task experiment, there was no difference between L1 and L2 groups. However, in a dual task experiment, the L2 group performed better on the language component of the task (thematic assignment) and worse on the WM component of the task (digit recall). This illustrated that attention allocation to language inputs (at the expense of other tasks) may buttress thematic judgment. In Manuscript 2 (Chapter 4), I report the results of a within-subjects experiment with bilinguals designed to stabilize attention allocation to the different task components, and better isolate the language processing mechanisms that differentiate L1-versus-L2 performance. The data provided evidence for a dissociation between semantic composition and metalinguistic processing in the form of a language-by-load crossover interaction. Bilinguals were better at retrieving the agents of passive sentences in their L2 in the single task condition, but were worse in their L2 in the dual task condition. The pattern suggested that thematic judgments are supported by metalinguistic processes. L2 performance is better in the single task condition because L2 comprehension entails higher initial engagement of metalinguistic representations. In contrast, L1 comprehension depends on implicit semantic compositional processing, which engenders switch costs associated with initiating metalinguistic judgments after interpretation. Critically, because metalinguistic processing is dependent on the control system, the WM load in the dual task condition interfered selectively with L2 comprehension. Thus, L2 performance declined precipitously, while L1 processing remained stable. I conclude that in a native language, “good-enough” heuristics bias metalinguistic conceptualizations of thematic roles post interpretation, and that the underlying semantic composition of passive sentences is fully and faithfully achieved upon an initial analysis by the linguistic system.