Causal Inference and Language Comprehension : : Event- Related Potential Investigations
- Author(s): Davenport, Tristan S.
- et al.
The most important information conveyed by language is often contained not in the utterance itself, but in the interaction between the utterance and the comprehender's knowledge of the world and the current situation. This dissertation uses psycholinguistic methods to explore the effects of a common type of inference - causal inference - on language comprehension. In 8 experiments, I investigate the effects of causal inference on the neuro-cognitive processes that occur during word processing (Experiments 1 -5) and the hemispheric basis of these processing effects (Experiments 6-8). The goal of Experiments 1-3 was to compare competing theoretical frameworks of language processing with respect to the ordering of "high-level" (causal inferential) and "low-level" (lexical association) context effects on word processing. To that end, participants listened to two-sentence short stories encouraging a causal inference, each followed by a probe word related to some aspect of the context story. ERP results showed that causal information affected word processing earlier than lexical associative information, and that lexical association effects were suppressed in discourse contexts. These results supported dynamic processing theories of the kind inspired by simple recurrent networks. Experiments 4 and 5 tested the impact of causal relatedness on multiple, semi-redundant discourse cues embedded in sentences. This study investigated whether causal inferences build up over time across several words, or if a full-fledged inference can be activated in response to a single critical word. Results indicated that different participants activated inferences in qualitatively different ways. Some showed evidence of predictive inference, while others showed evidence of a drawn-out inference activation process covering several cues to discourse implausibility. These results reflect individual differences in inference activation that are unrelated to common metrics of processing ability. Experiments 6-8 tested the hypothesis of a right hemisphere (RH) advantage for activating causal inferences. Results indicated that neither hemisphere had a processing advantage for causal related information, although left hemisphere (LH) experienced facilitated processing for strong lexical associations. This finding suggests causal inference processing is balanced between the two hemispheres, and that causal inference deficits in RH lesion patients are related to a dominant LH tendency to focus on local semantic relationships