Cognitive Psychology, 2nd Ed. Chapter 12 Language Comprehension Word Recognition Recognition of 30,000 to 80,000 words in 200 ms involves bottom-up processing. Low level visual analysis in the occipital cortex plus specialized analysis of word forms (words and pseudowords but not letter strings) in the left posterior cortex. N400 ERPs to semantically anomalous words reveals parallel top-down processing. Poor readers fail to use global text structure in comprehension and are slower in word recognition. Word Frequency Effect (WFE) Refers to high-frequency words being recognized faster and more accurately than low-frequency words. The task requires naming the word aloud. 200 ms recognition time and vocabulary size suggests parallel search models. The magnitude of the frequency effect is greater for exception words compared to regular words. Connectionist Models of WFE A hidden layer model using back propagation can learn to associate orthography to phonology. The errors made by the network were high for low frequency exception words, matching the slow RT shown by human subjects to these items. Dual Route Model of WFE Recognition can proceed by a direct path from graphemes to an orthographic lexicon to the lexical-semantic system or by an indirect path of grapheme to phoneme conversion. The direct lexical path is faster than the indirect conversion path. For low frequency words (e.g.,pint), the indirect path interferes with the correct pronunciation. For high frequency words, recognition is complete via the direct path before the indirect conversion path can interfere (e.g., have). Dyslexia Surface dyslexics fail to read exception words correctly but can read nonwords. This could mean a selective breakdown in the direct lexical path. Phonological dyslexics read entirely by sight vocabulary. Nonwords and low frequency rare words cannot be read. This could mean a breakdown of the indirect conversion path. Sentence Comprehension Comprehension involves building mental representations similar to constructing a building. For example, a topic sentence is fixated longer than other sentences, allowing the reader to lay a foundation. Suppositions and inferences go beyond the literal words in constructing an accurate mental representation of the author’s or speaker’s intentions. Suppositions A simple assertion (e.g., the star is above the plus) takes 1,450 ms to comprehend. To comprehend a negative, one must first presuppose the simple assertion and then deny it, taking an additional 300 ms (e.g., the star is not above the plus). Inferences Anaphoric reference is illustrated by the problem of linking a pronoun to its antecedent reference noun (e.g., William went for a walk. He meandered through Dante’s square.). Activation of referents and suppression of nonreferents through inhibition result in the proper linkage. Inferences Given-new strategy in reading assumes that writers mark information already understood as opposed to new information. New information (someone meandered); old information (the person is William) Explicit marking is quickest to grasp (William < bard < he). Inferences Polysemy is the property of language that a single word can have more than one meaning. Homonyms (bank—river or finance?) and metaphors (time flies—how can it without wings?) illustrates the problem. Potential meanings are activated and then selectively inhibited in word recognition. Priming can directly activate nonliteral readings of metaphors. Discourse Comprehension True discourse is referentially coherent, meaning the words of each sentence refer unambiguously to the others sentences. Local cohesion helps provide this (e.g., anaphora). Global frameworks representing the theme or topic also help. Global Frameworks Micropropositions are the individual propositions of a sentence. Macropropositions are the schema-based generalizations that summarize main points or the gist of the text. Story grammars, causal relations, and mental models are alternative theoretical accounts of global structures. Process Models of Reading Propositions from each sentence are used to construct a textbase representation. A situation model is constructed in parallel using input from the text and from semantic long-term memory. Macropropositions summarize the textbase and situation model. Process Models of Reading Fixations (200-350 ms) and rapid eye movements or saccades provide snapshots of 4 characters to the left and up to 15 characters to the right of each fixation point. 80% of content and 20% of function words are fixated. With time to build mental structures, 250-300 wpm is a normal reading rate. Process Models of Reading Immediacy Assumption: an interpretation is immediately attached to each word fixated. N400 ERPs to anomolous words reflect an attempt to assign meaning immediately. Eye-Mind Assumption: the duration of fixation varies proportionally with the amount of information that must be processed (e.g., a grammatically complex sentence takes more time and activates more neural tissue than a simple sentence). Speed Reading is Trained Skimming Start fixations at the right of first word and take the last fixation prior to the end of the sentence. Focus on content words and infer function words. Avoid regressive eye movements. Avoid subvocalization.