In lieu of an abstract, here is a brief excerpt of the content:

  • The Prosodic Phrasing of Clause-Final Prepositional Phrases
  • Eileen Fitzpatrick

Spoken language is not produced in a continuous flow; it is broken up into phrases. An understanding of phrase-boundary placement is critical for comprehension and of great importance in text-to-speech technology. The knowledge that speakers use to determine phrasal boundaries has been attributed in the literature to many seemingly competing factors, syntactic, semantic, phonological, discourse, and pragmatic. This article reports on a study of the boundaries of a single type of data, clause-final prepositional phrases (PPs). The study was done to improve the phrasing of a text-to-speech synthesizer. The syntactic constituency of the PP and its length as measured in accented syllables account for an overwhelming majority of the data. The few exceptions to this account fall into natural categories of semantics, discourse, and pragmatics, which suggests they have the status of marked forms.*

1. The problem

The domain over which an intonation contour extends is crucial to the communication of the intended meaning of an utterance, as example 1 illustrates.

(1) After he ate my cat Freddy took a nap.

If the first domain in 1 extends over after he ate, then the cat is the diner, but if it extends over after he ate my cat, then the cat is the dinner. Typically, a speaker indicates the boundary of an intonation domain with a pause, a pitch perturbation and/or syllable lengthening. A writer would indicate the boundary in 1 with a comma, but punctuation does not always indicate the end of a contour. No punctuation, for example, indicates that 2 is normally spoken with a boundary after well.

(2) Waiters who remember well serve orders correctly.

The knowledge that speakers and readers use to determine intonation contour boundaries has been characterized in the linguistic literature as ‘not well understood’ (Selkirk 1995) and ‘notoriously elusive’ (Ladd 1996), chiefly because so many seemingly competing factors have been adduced as correlates of intonation phrase boundaries.

Since phrase boundaries aid comprehension, text-to-speech (tts) systems attempt to mark these boundaries by inserting pauses as well as changes in segment duration and pitch. But because of a poor understanding of boundary placement, tts systems lack principle-based prosodic boundary rules,1 which gives these systems their run-on delivery and contributes to their artificial quality. However, since most tts synthesizers allow boundary markers to be inserted either by hand or by rule, these systems provide a good means of explicitly testing the aspects of the knowledge of the speaker/reader that contribute to prosodic phrasing.

At the risk of being one more blind man putting a hand to the elephant, our group at Bell Labs tested several hypotheses about the determinants of the prosodic phrase by building rules based on Gee and Grosjean’s (1983) prosodic phrase model into a [End Page 544] tts system and seeing where the system performed incorrectly.2 If we could identify where the rules rendered inappropriate phrasing, we could improve the system and perhaps shed more light on the nature of prosodic phrasing.

One of the interesting places where our system performed poorly was in rendering pauses around clause-final constituents. For example, it generated an unnatural period of silence before the italicized phrase in 3.

(3) The speaker pronounced the names of the characters on the left.

This inappropriate phrasing indicated a circumscribed problem to solve and an opportunity to examine the proposed linguistic correlates of prosodic phrasing. In an effort to render clause-final constituents correctly, we examined the prosodic phrasing of clause-final prepositional phrases (PPs) in a self-contained corpus. This article discusses the results of that investigation and, in particular, the contribution of phrasal length based on accented syllable count that characterizes most of the data.3 This work is intended as an empirical study, one that can form the foundation for further theoretical development.

2. The study

The study of prepositional phrases grew out of tests on an experimental system (Bachenko et al. 1986) to determine phrase-level prosody for textual input to the Olive-Liberman text-to-speech synthesizer (Olive & Liberman 1985). The original system, following Gee and Grosjean’s...

pdf

Share