A comparison of four vowel overlap metrics

  • Author(s) / Creator(s)
  • The overlap of vowel categories is a fairly common linguistic phenomenon. But, it can be difficult to judge whether two supposedly distinct vowel categories have merged or not. One tool that a researcher may use is a quantification of the overlap based on acoustic properties of recorded vowel tokens. However, there a serveral to choose from, and not much research as been carried out comparing the options. This study presents such a comparison between four proposed quantitative methods for assessing vowel merger: the sepctral overlap assessment metric (Wassink, 2006), the a posteroiri probability-based overlap metric (Morrison, 2008), the vowel overlap with convex hull assesment method (Haynes & Taylor, 2014), and the Pillai score from running a MANOVA (Hay, Warren, & Drager, 2006). A Monte Carlo technique was used to run the measures 1000 times on simulated data sets in each of a variety of potential overlap scenarios. In each tested overlap scenario, the measures were compared to each other in terms of their accuracy (mean absolute error from the desired overlap value) and precision (mean absolute deviation from the median). The performance and shortcomings of each measure are discussed, and the results of the simulations suggest that the a posteriori probability-based overlap metric performs the best overall. A recommendation is then made to use the a posteriori probability- based overlap metric in general cases.

    Presented at the Alberta Conference on Linguistics 2017 in Edmonton, AB.

  • Date created
  • Subjects / Keywords
  • Type of Item
    Conference/Workshop Presentation
  • DOI
  • License
    Attribution-ShareAlike 4.0 International