Accent Uncertainty (*?)

As mentioned above, a given string of words may form separate accentual phrases or may group together to form one large accentual phrase. It is most common for unaccented words to group together with adjacent words, while accented words tend to form separate phrases. However, there are cases in which even accented words can join with adjacent words in the same phrase. In such cases, the left-most accented word retains its accentual fall, and the accents to the right in the phrase are totally deleted. Since the difference between a very subordinate accent and deletion of an accent can be subtle, it may be difficult for the labeller to decide whether there is indeed an accent present on a given word (and thus forms a separate phrase), or if the accent has been deleted (i.e. ``totally dephrased'' with the preceding words). This is especially the case when the pitch range in which the word is realized is very reduced.

In these cases where a word is lexically specified as accented, but upon consideration of both the sound and F0 records the labeller is uncertain whether the speaker indeed produced an accent, a ``*?'' label should be marked in the tone tier. This label simply means ``I don't know if the speaker actually produced an accent'', and thus should not be used in cases where the labeller feels that the accent has been totally deleted. Like the H*+L label, *? should be placed within the mora marked for lexical accent.

Utterance narabu shows an example of labeller uncertainty about the accentuation of the final verb /sima'su/ `do'. It is common in Tokyo Japanese for final predicates to be produced in a reduced pitch range, thus making it difficult to see or hear the accentual fall. It is ambiguous whether the verb has been dephrased together with the preceding words (and thus the accent deleted), or has been produced in a separate phrase with a very narrow pitch range. Here, the *? is marked to indicate this ambiguity. (Section 4.6 will discuss in detail the break index label ``2--'' which accompanies this tonal uncertainty.)

In example utterance curtain, on the other hand, the labeller is not uncertain about the accentuation of the final verb /tukema'su/ `attach', but rather feels that the accent has been totally deleted. Therefore, no marking (H*+L nor *?) appears on the verb, and the break index (BI 1) indicates that the verb belongs to the same phrase as the preceding words (see section 4.2 below).

<<curtain>> ``I will attach a curtain to the window.''

The use of the *? uncertainty label is highly subjective, and labeller opinions about whether a word has an accent or not may vary. Since there is no right or wrong answer, labellers should not hesitate to mark *? if they are uncertain. The ``degeneration'' and total dephrasing of accented words in Japanese is an interesting area of research (see [Maekawa 1994]), and with these relevant locations flagged by *?, it will then be possible to search through large labelled databases to pull out examples for further research.