Turning points, tonal targets, and the English L- phrase accent

Abstract

This study of the alignment of L- in the H* L- H% contour of American English finds the strongest predictor of the location of the retracted phrase-accent “elbow” to be the location of the accent-related F0 peak, rather than one of a set of metrical “attractors” investigated. A strong correlation between peak height and elbow sharpness suggests that the elbow may arise from interaction between peak height and perception-based constraints on the implementation of postnuclear deaccenting: F0 must become low immediately after the Nuclear-Pitch-Accented syllable, to prevent the percept of a pitch accent on a following accentable syllable. Relativised Area Under the Curve (R-AUC), a global measure for characterising F0 contours, distinguishes this H* L- from contrasting contours (e.g., H* L*) more reliably than a turning point-based approach focused on the location/scaling of an F0 elbow. The R-AUC measure is sensitive to perceptually significant, but difficult-to-quantify, changes in contour shape, while avoiding the pitfalls of purely shape-based theories.

Keywords:

• Intonation
• Phonology

Acknowledgements

We wish to thank Michael Wagner, as well as two anonymous reviewers, for valuable feedback on an earlier draft of this paper. We also wish to thank Mariapaola D'Imperio, Sónia Frota, Bob Ladd, Lisa Selkirk, Susan Hertz, Draga Zec, and other audience members at the Cornell Prosody conference in April 2008 for insightful comments and suggestions regarding this work. We gratefully acknowledge the support of the National Science Foundation, Grants #BCS-0643134, #BCS 0643054 and #BCS 0643019.

Notes

¹To borrow the famous words of US Supreme Court Justice Potter Stewart (Jacobellis v. Ohio, 378 US 184, 197 [1964]).

²Throughout, we will use the term turning point to refer to a concrete event in the F0 curve, and the terms tone and tonal target to refer to phonological specifications of pitch level. Likewise, we will use the terms align and alignment to refer to the phonetic synchronisation of a turning point and its segmental anchor, and the terms associate and association to refer to the relationship between a phonological tonal autosegment and the Tone-Bearing Unit (TBU) on which it docks.

³As discussed below in footnote 12, prominence relations in English are known to be subject to alterations under certain circumstances generally relating to concerns of eurhythmy. While this will become important later in this text, for purposes of the assumptions made here, we know of no reason why the prominence relations we describe would be altered in the contexts we have constructed, nor do we have any evidence that subjects implemented such reversals at any point.

⁴One speaker with a particularly narrow pitch range typically produced very little final rise in his productions. Many tokens concluded with a rise of 1 semitone or less, which was perceived as largely flat by the authors. Other tokens had no distinguishable final rise in the F0 contour, nor did they have a final fall. As tokens from this speaker formed a continuum from virtually flat to slightly rising to more audibly rising, all such tokens were included as rise-fall-rise contours, and deemed variations in that speaker's unique, rather flaccid production of the desired contour.

⁵This dependence of the duration of pitch movements on independently varying segmental durations lies at the heart of the Autosegmental-metrical argument against traditional “configuration-based” approaches to the characterization of F0 contours.

⁶An anonymous reviewer questions our dismissal of the .533 correlation coefficient for the relationship between vowel duration and elbow location as indicative of an alignment relationship between the elbow and the pitch-accented syllable. The reviewer notes that we do accept a correlation coefficient of .592 for vowel duration and peak location as indicative of an alignment relationship between the peak and the pitch-accented syllable, and that since elbows are difficult to measure, we might expect the elbow data to be noisier than the peak data, resulting in the slightly lower correlation coefficient, even while the elbow was in fact aligning phonetically with the pitch-accented syllable. The important point here, however, is not the absolute value of any given correlation coefficient, but rather the relative strengths of correlations that offer competing explanations for the phenomenon at hand. The fact that the correlation between peak location and elbow location is stronger than that between vowel duration and elbow location (together with the data from the multiple regression including both vowel duration and peak location as variables) indicates that there is a significant amount of variation in elbow location that we can account for with reference to peak location, but not with reference to vowel duration. (Note too that noise in the elbow data in this case seems not to impede the stronger correlation result.) As to why there should be a relationship as strong as r =.533 between vowel duration and elbow location in the first place, this is likely just because vowel duration and peak location, the two competing predictors in this instance, are highly collinear: If peaks are aligned with some point in the pitch-accented syllable, such that peak location is correlated with vowel duration, and if elbows are aligned with respect to peaks, such that elbow location and peak location are correlated, then elbow location and vowel duration will also necessarily be correlated. However, because vowel duration and peak location are not perfectly collinear, the relationship between vowel duration and elbow location will necessarily be weaker than that between peak location and elbow location. Put another way, much of the variation we observe in peak location can be explained with reference to vowel duration. But since some of this variation is either random, or attributable to other sources, if elbow location is in fact determined by peak location, then we expect there to be variation in the placement of the elbow that cannot be accounted for with reference to vowel duration. As we have shown, this is indeed the case.

⁷Indeed, as Lisa Selkirk reminds us (personal communication), this was the essence of Bruce's original proposal for the phrase accent in Swedish.

⁸It is also not an effect evidenced by all of our subjects: 2 of 15 subjects showed no trace of it, and for six others, the difference in question was on the order of 5 ms or less.

⁹In fact, the number used in this correlation is just the slope of the first line, which, given the relative lack of variation in the slope of the line representing the low, flat trough, is a strong enough proxy for elbow “sharpness” in general. (The correlation between slope of the first line and change in slope between the two lines calculated directly was over 0.9 for all speakers.)

¹⁰The four regression lines that seem to follow roughly the same path, beginning at the upper left corner of the chart in Figure 13 represent the four male speakers that took part in the study. The remaining 11 lines represent females. Note that the male speaker with the highest peak F0 values has far steeper slope values than do females with comparable peak F0, resulting in horizontal overlap, but vertical separation, in their plotted results. This is because the male speaker is throughout the elicitation at the top of his pitch range, while the female speaker with comparable F0 is nearer the bottom of hers. An enthusiastic male, in other words, has similar peak F0 to a bored female, with the difference in slope of the fall back to baseline distinguishing the two.

¹¹Note that this predicts that for lower-peak utterances, and also for shorter utterances, the contrast between H* L- and H* L* may become perceptually ambiguous. We believe that this is so, and are currently designing studies to test this prediction.

¹²It is important to point out here that we are considering the first syllables of both nóminàtor and nòminátion type words to be equally accentable. It is true that under most circumstances, if an –ATION word is to be pitch-accented, the pitch accent will fall on the primary-stressed penult, and not the secondary-stressed initial. That said, it is a well-known fact about English that prominence relations in this sense are malleable, with such factors as rhythm and position in the phrase contributing to frequent prominence reversals in words of this type (Shattuck-Hufnagel, Ostendorf, & Ross, 1994). It is thus common enough for syllables of this type to be pitch-accented. (Should Newt Gingrich ever receive the GOP nomination for president, for example, we might quite grammatically refer to him as Nóminàtion Néwt.) It may also be of relevance that at the moment when a listener encounters the first syllable of the word nomination, both nomination and nominator are still equally part of the relevant cohort for lexical access, all the more so since prosodic cues to their disambiguation appear to be otherwise severely attenuated in postnuclear position.

¹³Note that the calculation of AUC for our contours is somewhat idealised, assuming as it does linear interpolations between peak, E1, and the attractor midpoint. This assumption allows us to calculate AUC for the region as the sum of the areas of the three geometric figures represented in the shaded region of Figure 14. (Namely, two right triangles and a rectangle.) It would of course be better to calculate AUC directly using integration, or to approximate it using Riemann sums based on successive, short-interval F0 samples over the window in question in our data. We are in fact doing this in several projects in progress, with strong results. For the time being, though, the straight-line stylisation approach seems a reasonable approximation of what we actually recorded. (It also sounds quite natural when resynthesised, as ‘t Hart 1991, for example, argues that it should.)