On the Reliability of Rhythm Metrics

In the past decade or so, various metrics of vocalic and consonantal variability have been used to quantify linguistic rhythm, often yielding disparate results. The reliability of several such metrics (percentage of vocalic intervals and consonantal standard deviation, pairwise variability indices, and variation coefficient) was tested using materials from stress-timed English and German, syllable-timed Spanish and Italian, Korean, an unclassified language, and Greek, which has resisted classification. The materials for each language were divided into three subsets: an uncontrolled subset of sentences excerpted from a representative author of each language, a subset exhibiting as much as possible “stress-timing” properties (complex syllable structure and extensive vocalic variability), and a subset exhibiting as much as possible “syllable-timing” properties (simple syllable structure and limited vocalic variability). The results suggest that rhythmic scores can be severely affected by the choice of materials, especially in languages such as Italian, in which it is easy to avoid or accentuate variability (e.g., by excluding or including geminates). Variation coefficient scores were the most resistant to the manipulation of materials but failed to show statistical differences across most of the languages examined. The overall results cast doubt on the reliability of metric scores as indicators of timing and linguistic rhythm.