A blackboard model of accenting

Contrast in concept-to-speech generation

“…All approaches to contrast prediction discussed in the previous sections have in common that they regard the presence of a pair of 'alternative items' (called parallel, contrastive, or contrastible items) as a prerequisite for contrast. An important difference between the approaches is that both Prevost [1995] and Pulman [1997] claim that two alternatives should be at least 'of the same type', whereas in the theory of van Deemter [1994avan Deemter [ , 1998van Deemter [ , 1999, the only condition on contrastible items is inequality of denotations. A possible explanation for this difference is the following.…”

“…Such cases occur quite frequently in D2S-based applications, but the approaches discussed in Section 3 are not well suited to deal with them. Table IV shows two of the GoalGetter examples discussed in the previous sections, and the corresponding predictions of Prevost [1995], Pulman [1997] and van Deemter [1994avan Deemter [ , 1998van Deemter [ , 1999. Example (18) is not included in Table IV, because of its similarity to (17).…”

“…The theory of contrast put forward by van Deemter [1994avan Deemter [ , 1998van Deemter [ , 1999 is based on the determination of contrariety and equivalence. Van Deemter states that many cases of contrast can be attributed to parallelism (see e.g., Prüst [1992], Hobbs and Kehler [1997], Asher et al [2001]), but he points out that there are also many examples of contrast which lack parallelism.…”