Condit-Schultz, N., (2026) “The Question of Melismas: Commentary on Lee et al. (2025)”, Empirical Musicology Review 20(2), 113-114. doi: https://doi.org/10.18061/emr.7407

INTRODUCTION

LEE’S “Tone-Melody Correspondence in Cantopop: Patterns of Directionality and Interval Size from a Corpus Study” is an excellent piece of empirical musicology. Lee’s analysis replicates previous findings that “similar motion” between language and music (accordance between lexical tone and melodic tone) is clearly favored by Cantonese song writers, but also reveals new nuances and details that have not been previously reported: notably that melodic unisons (“level”) are rarely paired with changing tones, while level tones are quite often paired with melodic motion. No research is perfect—it would always be better if we could have 1,000 pieces transcribed instead of “only” 105!—but Lee’s approach is well thought out as well as carefully and clearly explained and justified.

However, since I have this lovely opportunity to write a commentary on this paper, there is one area where I think Lee’s methodology might be improved upon in future work: the treatment of melisma. Since melismas are quite rare in the music being studied here (less than 3% of notes), I don’t think this issue makes any difference to Lee’s conclusions. However, for future work, I would advocate for a different approach to melisma. I can attest from experience that melismas, especially when combined with “grace notes,” are a difficult problem for the computational analysis of song. There is no obviously correct way to treat them, which is no doubt why (as Lee observes) nobody can agree on how to treat them. I don’t think I have all the answers—the best methodology always depends on the specific research question—but I do suspect that the approach Lee takes in this case is not optimal.

Lee points to Malawey’s (2020) description of melismas as “virtuosic showcases.” This sort of melisma is certainly something that figures in various forms of pop music, including Cantopop. However, most melismas in pop music are short and thus don’t fit this description at all—and I have found that this is true in Lee’s Cantopop data as well. In the following, I will refer to a “melisma” as any place a Chinese-language syllable[2] is set to two or more notes. From my own quick analysis, 748 of the melismas in Lee’s Cantopop corpus involve only two notes, with 129 three-note melismas and 30 four-note melismas[3]. The longest melisma I can find is only nine notes, with only 20 longer than 4 notes in the whole corpus. What’s more, in 501 of the two-notes melismas, the two notes have the same duration. None of this is consistent with a “showy” virtuosic lines that can easily be discounted.

The notion that we can easily dismiss melismatic notes as being showy (and thus probably “non-structural”), even with an appeal to Lerdahl and Jackendoff (1983), seems like a stretch to me. However, even if we can make a musicological argument for excluding melismas, extending this notion to linguistic prosody is highly suspect in my opinion. The idea of skipping over (ignoring) “non-structural” pitches in a language utterance is, to my knowledge, not consistent with linguistic theories of prosody. What’s more, the two “contour tones” in Cantonese actually naturally involve a change of pitch through a single syllable.

Despite these concerns, I agreed to write this commentary not because of a pretty trivial methodological critique for which I can’t really offer a good solution. Rather, it seems to me that melismas present a very interesting untapped resource for studying tone-melody correspondence which I’d like to explore (of course, melismas are too rare to base a proper analysis on, but it is still interesting to see what we can find in Lee’s Cantopop melisma data). Specifically, the Cantonese “contour tones” (T2 and T6) involve a change of pitch within a single syllable. It seems intuitive to me (albeit, from the etic perspective of a non-tone-language speaker) that melismas would be the ideal place for these tones to be realized. To explore this idea, I parsed Lee’s Cantopop corpus[3] and isolated all 501 two-note, equal-duration melismas and then tallied the melodic intervals (in semitones) between the two notes. The results can be seen in Table 1:

Contrary to the hypothesis that the ascending tones T2 and T5 would be executed as ascending intervals, a tendency for two-note melismas to descend appears to be overwhelming. I did poke around a bit more with other melismas in the dataset, but there doesn’t seem to be any evidence here that T2 or T5 are more likely to ascend, despite their idealized contours. For example, there are many examples of T2 words—which are supposed to rise through three registers when spoken—being sung with a descending melodic contour. Does this reflect a form of contrary tonal counterpoint? Or perhaps this is direct evidence that melismas are indeed ignored in the perception of tones in Cantonese song? I will have to defer to the judgment of Cantonese speakers (and preferably, Cantonese linguists) and for larger datasets of melismas to be analyzed.

From Table 1, it is notable that T1 is much more common than the other tones. This seems to be consistent with the distribution of tones in natural Cantonese speech (see https://github.com/jane-lisy/cantfreq), though T6 seems unexpectedly common compared to speech data. It would be worthwhile, in future research, to compare how frequent tone “unisons” occur in natural Cantonese speech compared to song.

ACKNOWLEDGMENTS

This article was copyedited by Jonathan Tang and layout-edited by Niels Chr. Hansen.

REFERENCES

Lee, J. (2025). “Tone-Melody Correspondence in Cantopop: Patterns of Directionality and Interval Size from a Corpus Study”, Empirical Musicology Review, 20(2), 95-112. https://doi.org/10.18061/emr.6598

Lerdahl, F., & Jackendoff, R. S. (1983). A Generative Theory of Tonal Music. MIT Press. https://doi.org/10.7551/mitpress/12513.001.0001 

Malawey, V. (2020). A Blaze of Light in Every Word: Analyzing the Popular Singing Voice. Oxford University Press. https://doi.org/10.1093/oso/9780190052201.001.0001