Contour clustering: A field-data-driven approach for documenting and analysing prototypical f0 contours

Abstract

This paper reports an automatic data-driven analysis for describing prototypical intonation patterns, particularly suitable for initial stages of prosodic research and language description. The approach has several advantages over traditional ways to investigate intonation, such as the applicability to spontaneous speech, language- and domain-independency, and the potential of revealing meaningful functions of intonation. These features make the approach particularly useful for language documentation, where the description of prosody is often lacking. The core of this approach is a cluster analysis on a time-series of f0 measurements and consists of two scripts (Praat and R, available from https://constantijnkaland.github.io/contourclustering/). Graphical user interfaces can be used to perform the analyses on collected data ranging from spontaneous to highly controlled speech. There is limited need for manual annotation prior to analysis and speaker variability can be accounted for. After cluster analysis, Praat textgrids can be generated with the cluster number annotated for each individual contour. Although further confirmatory analysis is still required, the outcomes provide useful and unbiased directions for any investigation of prototypical f0 contours based on their acoustic form.

Automatic intonational contour clustering in Patwin

“…This study used the Contour Clustering (CC) toolkit (Kaland 2021) to automatically cluster Patwin intonation contours. In the first stage, contours were demarcated using Praat TextGrids (Boersma and Weenik 2024); F0 information was then extracted for each contour at 20 equidistant intervals.…”

“…Automated prosodic clustering, such as that of Kaland (2021) therefore appeals for two reasons: 1) to streamline the researcher's process of becoming familiar with a new (and often large) body of data, and 2) to provide the researcher with a set of intonational contours present in the language, which can then be further investigated. Kaland (2021) additionally adds that automatic clustering is warranted to curtail researcher bias.…”

“…Automated prosodic clustering, such as that of Kaland (2021) therefore appeals for two reasons: 1) to streamline the researcher's process of becoming familiar with a new (and often large) body of data, and 2) to provide the researcher with a set of intonational contours present in the language, which can then be further investigated. Kaland (2021) additionally adds that automatic clustering is warranted to curtail researcher bias. Because most researchers are not native speakers of the language they work with, there is a high potential for the researcher to either miss important patterns that are not analogous in their native language, or to ascribe undue meaning to contours are analogous to native patterns.…”

“…Using automated clustering is therefore not simply a matter of time efficiency, but also ensures a level of quantitative proof for the contours posited by the researcher. Previous uses of Kaland (2021) include Babinski and Bowern (2022), who use CC to identify additional phrase types in Bardi, a Nyulnyulan language from Northern Australia.…”

Phonetic and Phono-Lexical Accuracy of Non-Native Tone Production by English-L1 and Mandarin-L1 Speakers

“…Although it was relatively unequivocal to determine whether a production in the picture-naming task should be labeled “rise” or “fall” based on an auditory impression and on F0 trajectory observations, it was sometimes less clear whether a production should be labeled as “mid-level” or “low-level.” To disambiguate these tone productions, the Contour Clustering Graphical User Interface (Kaland, 2021) was used to determine in a more objective way whether participants produced distinct mid-level and low-level tone categories. The Contour Clustering GUI carries out an automated data-driven cluster analysis that groups together F0 contours based on their acoustic similarities into a defined number of clusters (see Kaland, 2021 for details).…”