Academic journal article Psychonomic Bulletin & Review

Paws + Cause = Pause? Memory Load and Memory Blends in Homophone Recognition

Academic journal article Psychonomic Bulletin & Review

Paws + Cause = Pause? Memory Load and Memory Blends in Homophone Recognition

Article excerpt

Previous research has shown that separate information sources, activated closely together in time, can induce errors suggestive of memory blends. In Experiment 1, homophones were used to induce such memory errors. In a study task, participants made relatedness judgments to word pairs that included homophones (e.g., PAWS-BEAR). During this task, one group (study-similar) maintained memory loads with words that were orthographically similar to the presented homophones (e.g., JAWS). Another group (test-similar) maintained memory loads similar to those obtained for the study homophone's alternate spelling (e.g., CAUSE). A third group maintained no memory load during the task. In a surprise recognition test, participants were presented both previously viewed homophones (PAWS) and non-presented alternate spellings (PAUSE). We hypothesized that partially activated alternate spellings, in conjunction with the orthographically similar memory words, would result in the creation of blended memories. The results followed suit: The test-similar condition produced significantly elevated false alarms, relative to both the study-similar and the no-load conditions. Experiment 2 replicated these results while including memory loads to control for potential orthographic confusions. The findings are discussed in terms of multiple, episodic memories later activated as single, blended memories.

A basic assumption of almost all memory models, including multiple-trace (Hintzman, 1986; Shiffrin & Steyvers, 1997) and spreading-activation models (Collins & Loftus, 1975; McClelland & Rumelhart, 1985), is that related items activate each another. For example, in spreading-activation models, perception of a target word results in concurrent activation of surrounding related words. More recent models propose that, with sufficient similarity, information from separate events may be retrieved together, causing memory blends, "memory" for nonexperienced stimuli or events (e.g., Arndt & Hirshman, 1998; Stark & McClelland, 2000).

False memory errors have been frequently observed in the context of single words. Using the now popular Deese (1959), or DRM, paradigm, Roediger and McDermott (1995,2000) had people study semantically related words (e.g., bed, rest, awake, etc.), all related to a critical nonpresented word (e.g., sleep). Afterward, participants falsely recalled such critical words with the same frequency as they did legitimate targets. A natural account assumes that study words repeatedly activate such critical lures. During recall, people erroneously attribute the previous activation of critical words to their inclusion in study lists. Such findings exist for words that are similar along other, nonsemantic dimensions, such as orthography (Pesta, Murphy, & Sanders, 2001) or phonology (Sommers & Lewis, 1999). Recently, Watson, Balota, and Roediger (2003) presented lists containing both semantic and phonological associates. For those hybrid lists, false recall of critical lures exhibited over-additivity, exceeding the level expected from the simple sum of errors produced by separate semantic and phonological lists. Watson et al. interpreted these results in terms of separate lexical dimensions combining into potent false memories.

Depending on the original encoding task, words may be represented in memory with emphasis on their semantic, orthographic (visual), or phonological dimensions. In a recognition test, all three dimensions may be evaluated. In distributed models, words may be represented as convolved patterns of activation across these dimensions. If words are similar or are related along one dimension, it is assumed that their representations overlap. Thus, multiple words may be activated via shared units. For example, research on homophones (words with different spellings but shared pronunciation, such as PAIN and PANE) has produced findings indicating the activation of multiple orthographic representations through shared phonological units. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.