Dear list members,
I thought I would contribute to the ongoing discussion of observations in the linguistic literature regarding claims of the prevalence of “minor third” intervals in speech intonation patterns. Because my response is rather lengthy, I have divided it up into two parts. The first part of my response deals specifically with references to musical intervals in the linguistic literature. The second part gives additional comments on linguistic theory relevant to making connections between empirical facts about speech intonation and music.
First, I am very familiar with linguistic theories of intonation and with the literature in this area and I can confirm that the widespread assumption that minor thirds are prevalent in speech is just that, an assumption, and is not backed up by any studies of which I am aware. Moreover, I have discussed this issue with Bob Ladd who has himself pointed out that the minor third associated with the English “calling contour” described by Liberman (1975) and others (which is appropriate for use in a context such as “An- na! Dinner’s ready!”) does not always involve a fixed interval of a minor third. Rather, it is possible for English speakers to produce the entire contour in an “elevated” pitch range, so that the pitch interval is much smaller, something like one or two semitones (a minor or major second in musical terminology). This tends to lend support for the original proposal that pitch intervals in speech may be divided into “smaller” and “larger” intervals.
Nevertheless, other examples from linguistics suggest further analogies with musical intervals, but the size of intervals seem to differ across linguistic systems. For example, Chao (1947) describes Cantonese Tone 1 as being four semitones higher than Tone 3 and Tone 3 as being two semitones higher than Tone 6. (Anyone interested in this observation should also have a look at Wong and Diehl’s excellent study of Cantonese tone perception in the Journal of Speech, Language, and Hearing Research 2003, Vol. 46, 413-421). Moreover, in a recent conversation with David Odden, a prominent authority on the lexical tone systems of African languages, he shared that many “downstepping” contours in these systems involve a minor third. In contrast, Goldsmith (1976) describes Mende as a language in which these downsteps involved an interval of about a semitone. Like other list members, I have been somewhat frustrated with the lack of hard data to back up some of these observations.
Second, to anyone entering into serious inquiry of the linguistic theories behind speech intonation patterns, be aware that there are a number of assumptions in much of the linguistic literature which present theory-internal obstacles to making connections with music. Particularly problematic is a core assumption of autosegmental theory, which has been the dominant theory of linguistic tone in descriptive linguistics since the mid-1970’s. The basic assumption of this theory is that linguistic tones are represented in a manner that is essentially identical to speech phonemes. The implication of this assumption is that no direct analogy is possible between the relative heights of tones in sequence and the notion of musical contour (whether a note is higher than, lower than, or at the same level as the previous note - cf. Dowling and Fujitani 1971). Instead, relative height in linguistic tone systems is assumed to be represented in terms of a supposedly universal linguistic principle known as the Obligatory Contour Principle (OCP). A review article by Odden (1995) suggests that the bulk of evidence across languages is inconsistent with this proposed principle, yet alternative proposals have been slow in coming.
The lack of any theoretical codification of the notion of relative heights of tones in autosegmental theory was shown in my recent Ph.D. dissertation to lead to logical problems in the description of F0 patterns, particularly for subsequent theories of English intonation which have adopted the core assumptions of autosegmental theory, e.g., Pierrehumbert (1980). In absence of any explicit codification of relative height in the phonological theory, it was necessary to posit that multiple phonetic parameters mediate the representation between abstract linguistic categories and F0 values. I show that these mediating parameters in Pierrehumbert (1980) are largely or wholly unconstrained, which essentially renders the theory empirically untestable. Note that the popular ToBI system of English intonation transcription is essentially is a modified version of the theory of Pierrehumbert (1980) which assumes implicit restrictions on theoretical parameters alluded to above; as a result, ToBI does make some testable predictions.
I’ll also mention that my dissertation proposes an alternative linguistic theory which makes explicit connections to music theory. In this theory, which I have termed “tone interval theory”, tones are assumed to be represented in terms of abstractions of frequency ratios called “tone intervals”. A tone interval, I, specifies a relationship of relative height and/or distance between a tone, T, and a referent pitch level, r, where I = T/r. Then a particular tone interval specifies whether a tone is higher than (I > 1), lower than (I < 1) or equal to (I = 1) the referent pitch level. In this way the theory makes an explicit connection between musical contour and speech intonation. The dissertation can be found on my web site under “Publications” for anyone interested. It also includes results from several perception and production studies using native English speakers which test the proposed categories of English intonation proposed in the ToBI system of intonation transcription against the predictions of tone interval theory, showing support for the latter. I recently presented this work at the Linguistics Society of America (LSA) meeting in early January.
Thanks to the original poster, Jeremy Day-O’Connell, for raising these issues in the present forum.
Laura Dilley, Ph.D.
Post-doctoral Research Associate
The Ohio State University