The Role of Phonological Structure on Generalization of Trained to Untrained Words in People with Aphasia

University Of Washington, Seattle WA

Investigators

Linked publications & trials

Abstract

Improving the communication abilities of people with aphasia is of vital importance. One aspect of treatment for PWA who have anomia is single-word treatments to increase their vocabulary size. Current treatment protocols rely on the process of generalization to treat not just a set of trained words, but also related, untrained words. Specifically, spreading activation from the trained words propagates through memory to untrained words, making the untrained words easier to retrieve and produce later in time. However, generalization of trained words to untrained words has been limited at best. This proposal tests a new method for treatment word selection that would enhance generalization of trained words to untrained words, namely by taking into consideration the network structure of words in memory. Specifically, words in memory can be represented as a network where words are connected to each other if they share similarity. Many different types of language networks have been examined, but this proposal will focus on a phonological network structure that connects words if they share all but one phoneme (through addition, substitution, or deletion). The structure that emerges through this definition of phonological similarity has been shown to influence several spoken word perception and production processes, and word learning in healthy children and adults. Three measures of phonological network structure will be examined in this proposal to increase generalization of trained words to untrained words in PWA. Path length is the number of connections that must be crossed to get from one word to another in the network. Spreading activation diminishes in strength with each successive step, and therefore, low path lengths are essential to the success of generalization. Degree is the number of immediate connections, or neighbors, of a word. Words with many neighbors can receive more spreading activation than words with few neighbors, increasing its chances for generalization. Clustering coefficient is the amount of interconnectedness among the neighbors of a word. Words with a high amount of interconnectedness among its neighbors will have more reverberation of spreading activation in its local neighborhood, increasing the likelihood of generalization, whereas words with little or no interconnectedness among its neighbors will be unable to retain spreading activation in its local neighborhood. In sum, this proposal tests the hypothesis that untrained words will receive a greater strength of spreading activation when they are closely connected to a trained word, and have high degree and high clustering coefficient. The influence of path length will be assessed via and archival data analysis of existing treatment data from 58 PWA. The influence of degree and clustering coefficient will be assessed via repetition priming experiments with 26 PWA. The findings of this proposal will inform treatment stimulus selection that will increase generalization of trained words to untrained words. Clinicians and researchers will be able to select a smaller, smarter corpus of trained stimuli that will benefit patients? communication abilities and well-being.