2.12.1 - The alternatives (alt) tier
(Note for 2006 draft: all content on the alt tier is still under development, and the conventions described are subject to change)
The alt tier, a recent addition to the ToBI system, allows the labeller to indicate places where more than one label was seriously considered. There may be times when a labeller spends a comparatively long time determining which label to use for a particular tone or break. The alternatives tier allows an outlet for the labeller to include both labels, so that she can move forward with more speed, and less frustration.
Use of the alternatives tier, along with other markers of uncertainty which will be discussed later in this section, also provides a record about which regions and points in time in the speech signal are harder to label, so that researchers may investigate the sources of ambiguity. Without use of any uncertainty markers, the labels used would not differentiate between those stretches of speech which are straightforward to label, and those which contain ambiguities.
back to top
2.12.2 - Uncertainty regarding pitch accent type
A common situation of uncertainty arises when the labeller has determined that a syllable is pitch-accented, but has trouble deciding which pitch accent type best accounts for the pitch pattern observed. Most often, the labeller is considering between two specific pitch accent types, such as H* vs. L+H*. The labeller might feel that the best label to use would be H*, but strongly considered using L+H*. In such a case, the labeller should use the label H*? in the tones tier, and put L+H* in the alt tier. The ? symbol indicates, simultaneously, that the labeller was uncertain about the choice of pitch accent label, and that the alternatives tier has been used to indicate the other pitch accent that the labeller considered. The example <indeed>, shown in Figure 2.12.1, contains two instances where the labeller was not certain about which pitch accent type to label: first on the syllable -deed of indeed, and then later in the file, on the monosyllabic word clouds. In the first full intonational phrase of the file (shown in Figure 2.12.1) with the words "and indeed," there is a strong prominence that is marked by a peak in the pitch track on the syllable -deed of indeed; the syllable is clearly pitch-accented. However, the labeller was not certain whether the pitch accent on -deed should be labelled with the single-tone H*, or with the bitonal L+H*. The pitch at the beginning of the phrase starts out moderately high in the speaker's pitch range, and rises slightly into the first syllable of indeed, -in, and then rises to a peak in the pitch-accented syllable. There is an apparent fall in F0 shown in the pitch track at the end of the syllable in-, but this can be attributed to segmental effects. In fact, these segmental effects between the two syllables of the word indeed make the F0 tricky to interpret. The rise into the pitch accented syllable appears to be quite sharp, as would be expected from the bitonal pitch accent L+H*. However, the labeller here felt that the pitch accent sounded more like an exaggerated or emphatic production of a single-tone H* pitch accent.
Play indeed.wav:
In the second Intonational Phrase of the file <indeed>, shown in full in Figure 2.12.2 , below, the labeller has used the label L+H*? on the word clouds. In this case, the option considered by the labeller was L+!H* (where the bitonal has a downstepped High). In this case, the labeller perceived the pitch height of the peak of the pitch-accented word clouds as being about the same as the pitch height reached in the preceding pitch accent on the -lec- of electron. However, the pitch track shows that the F0 of the peak on clouds as lower than the peak of the first pitch accent. The labeller gave precedence to the choice of L+H* (with no downstep), and put that label in the tones tier, along with the "?" marker. Then, she put her second choice, L+!H* in the alt tier.
Play indeed.wav:
back to top
2.12.3 - The X*? label
While most often a labeller will have a first choice of label when considering two alternatives, there may be times when the labeller will be unable to decide which pitch accent type to put in the tones tier. The label X*? can be used for such cases. Whenever possible, the labeller should still use the alternatives tier, and list both alternatives considered there. For example, if the labeller can't decide between L+H* and H*, and isn't happy committing to either one for the tones tier, the labeller can put X*? in the tones tier, and both L+H* and H* in the alt tier. The labeller should align a single point in the alt tier with the X*? in the tones tier, and list both alternatives at that point, separated by a "pipe" ("|") symbol, eg. "H*|L+H*".
Table 2.12.2: Commonly confusable pairs of pitch accents
-
H* vs L+H*:
The pitch accented syllable clearly has a High prominence, but the rise may be more gradual than expected for the bitonal L+H*, but steeper than would be expected for a single-tone H*
-
L* vs H* (in a compressed pitch range):
The syllable is clearly prominent, but the pitch track is more or less flat, with no clear peaks (as often occurs with H*) or dips (as one would expect for L*), and the labeller is uncertain whether the region in question is low enough in the speaker's pitch range to justify Low tones (L*) or in a mid-range, such as would happen with H*
-
L+H* vs L*+H:
There is a distinct low and a distinct high associated with a pitch accent, but the labeller is uncertain which tone, L or H is more strongly associated with the prominence
-
H* vs !H*:
A pitch accent following another High tone pitch accent (such as H* or L*) is somewhat lower in f0 than that preceding accent, but the labeller is uncertain that the tone is perceptually lower
-
H+!H* vs !H*:
The labeller is uncertain whether a peak leading to the !H* prominence is associated with that pitch accent, or if it is associated with some preceding High tone pitch accent (such as a late peak after a H* or L+H*)
back to top
2.12.4 - Uncertainty about whether or not a syllable is pitch-accented
There will be cases of uncertainty relating to pitch accent that are not only about which type of pitch accent to use, but about whether to use a pitch accent label at all. The label *? is used for just such cases, and its use will be discussed below.
Levels of prominence:
As discussed in Chapter 2.3, syllables in spoken language can be produced with a range of degrees of prominence: it is not the case, for example, that all syllables which do not bear a pitch accent are equally weak. Some syllables can be prominent compared to neighboring weaker syllables, but still less prominent than a syllable bearing a pitch accent. To increase the complexity, not all pitch accented syllables are equally prominent. For example, it may be the case that in an intonational phrase with two pitch accents, one pitch-accented syllable may be much perceptually stronger, and more salient than the other. The less-strong pitch-accented syllable is still pitch-accented, nevertheless.
There may be times, though, where it is not clear whether a syllable, which has some degree of prominence, is in fact pitch-accented. The example <august> shows an instance where the labeler was not certain whether a particular syllable was pitch-accented or not. In this example, shown in Figure 2.12.3, the labeler has used the *? label on the word how in the second intermediate phrase, for how long in that file. In this example, the labeler's use of the L* pitch accent symbol on the word long shows that she was certain that the word was pitch accented, and that the pitch patterns were best captured by the Low pitch accent. However, the labeler felt that the word how could also be pitch-accented. While not as prominent as long, the word seems to be fairly prominent. The pitch track also shows some ambiguity: there is a slight rise in f0 at the beginning of the word how. While this is potentially the result of a pitch-tracking error, the labeler may have been uncertain whether the rise in pitch, combined with the sense that the word how is fairly prominent, was evidence of a pitch accent on that syllable. The labeller addressed this uncertainty by using the *? label in the tones tier. Then, in the alt tier, the labeller put the type of pitch accent label that she would use had she felt confident that indeed the syllable was pitch-accented.
Play august.wav:
Sources of uncertainty/ambiguity:
There are a number of fairly common phenomena that make it difficult to determine whether or not a certain syllable is pitch-accented. It is sometimes the case that a labeller will be uncertain whether a strong syllable between two (clearly) pitch-accented syllables also contains a pitch accent. This can happen when the syllable in question occurs during a stretch of high pitch between two H* pitch accents, or comparably during a stretch of low pitch between two L* pitch accents.
The example <marmalade-starq> shows an example of a *? label in a region between two L* pitch accents. In this example, there is a Low-tone pitch accent on the -an- of Marianna, and then another Low-tone pitch accent on the mar- of marmalade. The pitch lowers into the first of these two pitch accents, and stays low through the second. Notice how one of the syllables between these two pitch accents, the monosyllabic word made, is also fairly strong, and could potentially also have a Low (L*) pitch accent: the signal is ambiguous here. The labeller has captured this ambiguity, and resulting uncertainty, by using the *? label in the tones tier, and L* in the alt tier.
Play marmalade-starq.wav:
The example <married-starq>, shown in Figure 2.12.5, shows a parallel case to the one discussed above: in this case, a strong syllable produced in a stretch of high pitch between two syllables with High (H*) pitch accents could potentially also be pitch-accented with H*. Here, while the An- of Anna and the Len- of Lenny both are clearly pitch-accented (with H* pitch accents) the syllable mar- of married could also potentially be pitch accented; the label *? is used for this syllable in the tones tier, and H* in the alt tier.
Play married-starq.wav:
Other common contexts for *?:
- after one or more !H* pitch accents, such that the local pitch range has been greatly reduced
-
when the speaker is generally using reduced pitch range
- when there is a phrase final full-vowel syllable, especially a monosyllabic word
- a phrase-initial full-vowel syllable, especially starting with a vowel (glottal stop)
- 2 consecutive full-vowel syllables: both seem strong, and it may be hard to tell which of the two is pitch accented—there may be times when there is pitch movement in the area of these two syllables that could be associated with one or the other or both
back to top