Frequencies as an Aid to Segmentation in Language Acquisition.
Rebecca Sue Badger Horvath · Deep Blue (University of Michigan) · 1980
This study proposes a model for segmenting speech that might be used by children in the language acquisition process. Segmentation refers to the process of discerning word units in an unbroken stream of speech. The model of functioning described is a frequency model based on the assumption that the more frequently sounds are heard together, the more likely they are to cluster as a unit and the more likely these units will be words or clumps of words. An examination of basic theories of cognition and studies of infants' response to language indicate this model is a possible way to obtain the information necessary for segmenting. Speech of parents to and in the presence of their children was collected. The four children ranged in age from fourteen to twenty-one months in age. The parental speech was translated into a typewritten equivalent of phonemic symbols and typed into a computer with all external markers such as punctuation, stress, and spacing removed. A computer program was designed to count how many times phonemes and phoneme groups repeated and segment the parental speech on the basis of those repetitions. Sections of protocol, 33 to 350 words long, were segmented. The percentage of correct words and groups of words obtained independently for each section averaged 39% and ranged from 22% to 62%. A chi square comparison with sections r and omly segmented showed this to be highly significant. Three modifications of parental speech were observed. First, there was a tendency to choose words of similar sounds. Second, "target" words were repeatedly used in a variety of sentences. Finally, formula sentences with new words inserted in a salient position were used. It was theorized that the two structural modifications mentioned last would aid in the segmentation process.