Phonemic Error Analysis: Edit Distance Instead of Phonemic Error Rate

Public Deposited
Analytics

Downloadable Content

Download PDF
Request Version for Screen Reader
Last Modified
  • February 22, 2019
Creator
  • Smith, Michael
    • Affiliation: School of Medicine, Department of Allied Health Sciences, Division of Speech and Hearing Sciences
  • Haley, Katarina L.
    • Affiliation: School of Medicine, Department of Allied Health Sciences, Division of Speech and Hearing Sciences
  • Wambaugh, Julie
    • Other Affiliation: University of Utah
Abstract
  • Purpose: The purpose of this study was to determine convergent validity of edit distance and manual phoneme-level analysis in the quantification of phonemic errors. Background: The proportion of words produced without any errors, is often used as a convenient index of articulatory severity (Haley et al., 2012; Duffy et al., 2017). Phoneme-level analysis is far more precise in its report of the frequency of substitution, omission, and addition errors that a person makes when speaking (Cunningham et al., 2016; Haley et al., 2001; Odell et al., 1990, 1991; Strand et al., 2014). It is, however, time-consuming, in that it requires consistent judgments about error type and manual calculation, making applications limited. In this study, we apply an edit distance metric to compute the similarity automatically. The Levenshtein distance, the simplest of edit distances, quantifies the difference between two strings by summing the fewest number of omissions, additions, and substitutions required to transform one string into another. Previous applications in the area of linguistics have been used to quantify the difference between languages or dialects (Thije & Zeevaert, 2007). Methods: We analyzed two retrospective speech samples. The first sample consisted of 24 speakers with AOS who repeated 27 words with varied length and phonetic complexity on a motor speech evaluation (Duffy, 2013). The second sample consisted of 41 speakers with a WAB-based diagnosis of conduction aphasia who named 15 pictures while completing the short-form Boston Naming Test (Kaplan et al., 2001). For each word, we computed the edit distance and manual phoneme-level analysis. Edit distance was calculated using an online calculator (Holsinger, 2017), and the sum of the edit distance for all attempted words was divided by the total number of phonemes produced. Phonemic error rate calculations were completed by manually counting the frequency of omissions, additions, and substitutions and again dividing this sum by the total number of phonemes produced. Results: Convergent validity for the edit distance was excellent, as demonstrated by a negligible mean difference between the edit distance and phoneme-level analysis and a near perfect correlation between the two analyses. Discussion: Not only is convergent validity high for applications of the edit distance to phonemic analysis in people with aphasia and apraxia of speech; the measure is also more time-efficient and not vulnerable to user error. Because omissions, additions, and substitutions are counted independently, difficult clinical decisions must be made when counting phonemic errors manually. Due to the convenience of automated calculation and potential for inclusion in transcription software, clinical application is highly feasible and would extend to any area for which phonemic analysis is relevant, which includes a variety of speech production disorders in children and adults.
Date of publication
Keyword
DOI
Resource type
Rights statement
  • In Copyright
Conference name
  • University of North Carolina at Chapel Hill. Department of Allied Health Sciences. Division of Speech and Hearing Sciences. Student Research Day (9th: 2017: Chapel Hill, NC)
Language
Date uploaded
  • May 5, 2017

Relations

Parents:

This work has no parents.

In Collection:

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PhonemicErrorAnalysis.pdf 2019-04-30 Public Download