As we have seen, Intonational Phrases frequently contain more than a single pitch accent. In previous sections we have seen four pitch accent labels: the two single-tone accents (H* and L*) and two bitonal accents (L+H* and L*+H). Theoretically, there are no constraints within the ToBI system on how these pitch accents may combine in an Intonational Phrase: any pitch accent label of the set seen so far may precede or follow any of the pitch accent labels in that same set. For example, H* may be followed or preceded by another H*, or by L* or L+H* or L*+H.

However, the !H* pitch accent label ("downstepped H star") is dependent on the immediately preceding labels. It indicates that there is a specific tonal relationship between the prominence labelled !H* and the preceding pitch accent. Specifically, a downstepped H pitch accent indicates that the tone of the prominent syllable is realized by a perceptually lower tone than the High tone target of an immediately preceding pitch accent: the tone has "stepped down" from the preceding High. While the !H* is realized by a lower pitch than the preceding High, it is distinct from the L* pitch accent, which is characterized by a pitch excursion down towards the bottom of the speaker's pitch range (for that utterance). The !H* does not necessarily approach the bottom of the speaker's range. For example, in <another-banana1>, the H* on the -noth- syllable of another is followed by !H* on the -nan- of banana. The phrase ends with a Low phrase tone and a Low boundary tone (L-L%), which are realized by additional lowering of the F0 in the final syllable (-na of banana). (Note that the pitch track shows what looks to be a rise in the F0 at the end of banana, but this is a pitch-tracking error.)

Play another-banana1.wav: - download another-banana1.TextGrid


Figure 2.6.1 : H* !H* L-L%

The example <clean_slate>, below shows another instance of the same tones (H* !H* L-L%), this time realized on two single-syllable words, "clean slate." Here, the pitch track shows a nice "step down" from the word clean, labelled with H*, to the word slate, labelled with !H*. (The audible fall in pitch into creaky voice at the end of the word slate, which is the realization of the L-L% phrase tone-boundary tone combination, is not captured by the pitch tracker.)

Play clean_slate.wav: - download clean_slate.TextGrid


Figure 2.6.2 : another example of H* !H* L-L%

The !H* label may follow any of the pitch accents which contain an H target, such as H*, L+H* or L*+H. The !H* label will therefore never immediately follow a (single-tone) L* pitch accent, as that pitch accent has no High tone target. The !H* label will likewise never be the first pitch accent in a phrase.

In general, pitch accents labelled !H* are realized in the middle of the speaker's pitch range for that Intonational Phrase; however, they are not considered "mid" tones. Like L+H* and H*, the !H* is a pitch accent for which the prominence is considered to be signalled by a High tonal target, which is why the H symbol is used with the star symbol. As with the other pitch accents with H targets, the actual realization of the High tone can vary quite a lot (i.e. speakers can speak with a relatively low pitch or a high pitch, but still produce High tones). This is particularly apparent in cases where one !H* pitch accent is followed by another !H*. The second !H* means that the target of the this prominence is lower relative to the target of the first !H*.

For example, in Figure 2.6.3 <privateryan> ("Saving Private Ryan") the last prominent (pitch-accented) syllable of the phrase (the Ry- of Ryan) is produced with a lower F0 than that of the previous prominence (on the Pri- of Private) which is in turn lower than the F0 of the first prominent syllable of the phrase (the Sav- of Saving). This relationship in the pitch height is reflected in the use of the use of the !H* labels on the last two pitch accents of the phrase, giving the tone labels H* !H* !H* L-L%. By the way, notice that the pitch range (75-150 Hz, indicted on the right vertical axis in the F0 pane) for this speaker is much more compressed than the range seen in other examples. It is useful to adjust the range for each speaker in order to see the F0 excursions. Otherwise, in another display, such as one from 75-400 Hz, the F0 might appear nearly flat.

Play privateryan.wav: - download privateryan.TextGrid


Figure 2.6.3 : Two sequential !H* pitch accents

The same tone labels (where the pitch accents labels are an H* followed by !H* followed by another !H*) can be seen in the example Figure 2.6.4 <fencedmeadow_150>. There are no theoretical limits within the ToBI system as to how many times such downstepping can occur within a phrase.

Play fencedmeadow_150.wav: - download fencedmeadow_150.TextGrid


Figure 2.6.4 : Another example of two sequential !H* pitch accents

On the other hand, there is no special term or diacritic for a second High pitch accent that is realized with a higher f0 than its immediate predecessor. In this case, the second, higher pitch accent would be simply labeled as another H*, as in the following examples:

Play really1.wav: - download really1.TextGrid


Figure 2.6.5 : H* !H* L-L% compared to H* H* L-L%

In this example, the first rendition of That's really illuminating has two pitch accents, an H* on real- and a !H* on -lum- similar to the examples of downstep shown above. However, the second rendition has an H* on real- followed by a higher H* on -lum-. The differences in heights of H*'s in an intonational phrase is not distinctive, that is, medium High H*'s don't convey different information than higher H*'s. However, the term "upstep" is sometimes used casually or in other contexts (see box: What About "Upstep"?).

Figure 2.6.6 : Two more examples of H* followed by another H* realized with a higher f0

Play thought.wav: - download thought.TextGrid


Figure 2.6.6A : "That's what I thought."

Play tree1house.wav: - download tree1house.TextGrid


Figure 2.6.6B : "My classmate, who lives in a treehouse, was written up in Atlantic."

In this last example, note the middle Intonational Phrase (who lives in a treehouse) where the H* on treehouse is higher than the H* on lives.

What about "Upstep"? The term "upstep" has sometimes been used in ToBI literature (eg. the Labelling Guide), but not in a way corresponding to the term "downstep." Specifically, it has been used not for pitch accents, but in relation to phrase tone-boundary tone combinations. The High phrase tone (H-) is said to "upstep" the following boundary tone. For example, the Low boundary tone (L%) in a H-L% combination is not realized as a lowering in pitch, as the H- phrase tone has "stepped it up." The High boundary tone (H%) is also considered "upstepped" after a H phrase tone (H-): the rise in pitch at the boundary for H-H% is higher than the pitch of H% when following a Low phrase tone (L-H%). In a sequence of two pitch accents with H targets, we have seen that it is often the case that the second pitch accent will be realized with a higher F0 than the first. While downstep is explicitly marked with the ! symbol, there is no analogous marking of "upstep" in a sequence of two pitch accents. back to top

In addition, once a speaker has produced a !H*, there is no requirement to continue to downstep for the rest of the Intonational Phrase. A speaker can produce a !H* and follow it with an H*. In the following example:

Play saving_down_up_S1.wav: - download saving_down_up_S1.TextGrid


Figure 2.6.7A : H* following !H*

Play saving_down_up_S1.wav: - download saving_down_up_S1.TextGrid


Figure 2.6.7B : a sequence of !H* produced by the same speaker as the previous example

One other optional ToBI label was used in saving-down-up-S1: the < in saving. As you may have noticed, the peak of the High Tone doesn't always align with, e.g. the center of the accented syllable's vowel. Many labelers wish to note when a peak is particularly late with respect to the vowel and will mark the < to indicate a late peak. This label is optional and therefore not consistently marked throughout the ToBI community.

In earlier discussions of pitch accent (whether in the single-tone H* or L*, or in the bitonals L+H* and L*+H), we have tried as much as possible to give examples where the pitch movement associated with the pitch accent is clearly visible in the pitch contour. However, it is not the case that each pitch accent will have a clearly distinguishable pitch movement of its own. Sometimes the pitch event associated with a pitch accent is not a movement, such as a High or Low culminating in a "peak" or "valley," but an absence of movement toward another target. For example, in an Intonational Phrase with an H* ending in a Low phrase tone-Low boundary tone combination (L-L%), the fall into the L phrase tone is expected immediately after the high pitch associated with a High pitch accent (H*). This is the case in <armani6>, shown below, where the F0 begins to fall rather sharply immediately following the end of the pitch-accented syllable (the -man- of Armani). Notice that by the time the speaker reaches the word millionaire, the F0 has nearly reached the bottom of the speaker's pitch range for that utterance. In contrast, <armani5> the F0 between the H* on the -man- of Armani and the !H* pitch accent on the mil- of millionaire, stays comparatively high. This second pitch accent, while it doesn't have a "peak" of its own, effectively prevents the F0 from falling into the L- phrase tone until later: after the second pitch-accent. In this example, even the first H* pitch accent does not have a clear "peak" in the F0. Many single-tone H* pitch accents are realized with fairly flat (but still high) F0 regions.

Figure 2.6.8 : H* L-L% vs. H* !H* L-L%

Play armani6.wav: - download armani6.TextGrid


Figure 2.6.8A : armani6, H* L-L%

Play armani5.wav: - download armani5.TextGrid


Figure 2.6.8B : armani5, H* !H* L-L%

back to top

What triggers "downstep"? This section is taken from notes by Mary Beckman, used at the ToBI clinic at Ohio State University. The trigger for downstep in English is very controversial. Pierrehumbert (1980) first proposed that downstep is triggered by any sequence of alternating L and H tones (as in the African tone languages where it was first described), whereas Beckman & Pierrehumbert (1986) proposed that it is only the alternating sequences of bitonal pitch accents that triggers it. Note that in both of these models, there is a bitonal accent H*+L which is missing in ToBI, for which the trailing "L" target is not very low in the pitch range. So an Intonational Phrase contour that in ToBI would be transcribed as H* !H* L-L% would be represented in the Pierrehumbert (1980) or Beckman & Pierrehumbert (1986) models as H*+L H* L-L%, with the downstep that is explicitly marked in ToBI being understood as an automatic consequence of the choice of a bitonal H*+L accent. Ladd (1983, 1986) objected to this characterization of downstep, because it meant positing this H*+L accent that typically doesn't have an obvious L target. Gussenhoven and his colleagues have also objected to this, and have always maintained that downstep is a facultative choice that a speaker can make to signal greater cohesion within a prosodic phrase (see, e.g., the van den Berg, Gussenhoven, & Rietveld, 1992, paper in LabPhon2). They have a way of representing it that differs from ToBI, in terms of a combination of a H*L accent type for which the L tone typically represents the L- phrase tone, but in cases where the accent is not the nuclear tone, it detaches and reattaches to the following H*L accent, or simply detaches and is a "floating" tone that is realized as downstep. In this system, downstep is represented not as a mark on the accent tones, but as the choice of "float" among the three-way opposition among ways to deal with the L tone (i.e., "spread" for a nuclear accent vs. "reattach" for a what we would transcribe as the leading L tone of a L+H* accent vs. "float" for what we would transcribe as a H* or L+H* accent before a downstepped !H* accent). Ladd (1986) proposed a different representation again -- namely that downstep be modeled as one term in a binary opposition between two different "register tone" relationships -- adopting Clements's tree model of downstep in African tone languages. So the short story is that no one is sure what triggers downstep. There are competing theories, but no definitive data to decide among them. Since there is this controversy about the theory, what we decided to do back in 1990, at the first workshop where we started to devise the English ToBI system, was to sidestep the controversy completely, by explicitly marking downstep on the first H tone that is realized in the downstepped pitch range. So downstep can follow any pitch accent that has a H target in it. If you want to read the description of how we decided this, by the way, you can look at the Beckman, Hirschberg, & Shattuck-Hufnagel chapter that will appear in the Jun (2004) book. back to top

Section 2.6 Exercises: !H*

A. & B.: Listening and Labelling exercises:

For this section's exercises, there are 5 batches containing labelled soundfiles (.wav and .TextGrid). If you don't get assigned a batch in class, please email Alejna at abrugos@bu.edu to ask for one. When possible, work with a partner or small group. Each batch will be assigned to several different people.

In each batch, you will find:

1. 15 pairs of soundfiles and labelled TextGrids, all containing examples where one or more labeller used the !H* pitch accent label.

2. An excel spreadsheet which has: a) each filename in the batch b) the tone labels (in sequence) and c) the word labels. It may be useful to print this file.

Notes about the files: The files contain mostly read speech, but each batch should have one or two excerpts taken from longer spontaneous speech files. Each batch should have a range in terms of the number of pitch accents in a phrase, and more than one phrase accent-boundary tone combination. The files should only contain labels that have been introduced so far, but it's possible that some unfamiliar labels may have passed through the "filter." Additionally, these files may contain actual errors.

Your task:

1. Listen carefully to each soundfile before you look at the pitch track or TextGrid. Try to pick out which words/syllables have a !H* pitch accent (if any), and which syllable/word has the initial pitch accent with a High element (which must be in the phrase if there is a !H*). You might want to mark on the spreadsheet the syllables that you hear as pitch-accented. (Evaluate the files in any order you choose—it may be fun/useful to evaluate pairs of files with the same text/words. Each batch should have at least one such pair.)

2. Look at the soundfile and TextGrid together. Compare and evaluate your choices to the labels in the TextGrid. Do they agree? Where there is disagreement, consider which labels you prefer, and why. (Please don't consider the existing TextGrids as "answers.")

3. Adjust the TextGrid labels (as needed) according to your decisions, and save it with your initials appended to the filename. If you make no changes to the TextGrid, you need not resave.

4. Feel free (and please do!) to make adjustments to the TextGrids beyond those involving pitch accents: many of these files may have errors, omissions, sloppy alignments, or other strangeness. Adjust them according to the conventions we have used to date. Save these TextGrids with your initials appended as well.

5. Pick out 2 or 3 files from your batch (more if you like) that you feel have clear instances of !H*--and that you think a new labeller would be able to label. Mark these in your Excel sheet, and upload the spreadsheet to your anita directory with your initials appended to the filename. Also make a note of any files that made you want to pull your hair our, and that you think would be very difficult for a new labeller. Any other comments you'd like to make about the files in your batch are also welcome!

C. Further exercises:

Produce and record your own !H* examples.

1. Try the following text and tone sequences. How many sequential !H* pitch accents can you produce? (Add more words if you can keep going...)
A banana. H* L-L%
Another banana. H* !H* L-L%
Another yellow banana. H* !H* !H* L-L%
Another little yellow banana. H* !H* !H* !H* L-L%

2. Pick any file(s) from section 2.6, or a favorite from the listening/labelling exercises above, and reproduce in your own voice. Save and upload your versions to anita using the filename ex6c2, plus a keyword and your initials.