Evidence for Phonological Constraints on Nuclear Accent Placement

1. Introduction

The distribution of pitch accents in speech is a topic of long-standing importance because it reflects the relationships among different levels of representation: phonology, syntax, semantics, and information structure. The broad tendency for accents to be located on new information is a classic observation, but formalizing the principles that govern accent placement in all cases has proven to be a difficult challenge. In this discussion note, we use experimental data to explore in depth the accent patterns on constructions with stranded prepositions. We extend the model of Schwarzschild 1999 to account for our findings.

Function words are prosodically weaker than content words and do not ordinarily carry accents. However, Ladd (1980) notes an interesting class of exceptions. These are cases in which a preposition has been stranded at the end of an infinitival clause in which all other words are given. Ladd’s dialogue (1) provides an example.

(1)

A: Why don’t you have some French toast?

B: There’s nothing to make French toast out of.

In 1B, the noun French toast is the default location for the nuclear pitch accent by virtue of being prosodically stronger than its neighbors. However, the prosodic prominence of the noun is weak because of its recent mention in the discourse, and the nuclear accent appears on the preposition.

Our study concerns a closely related but more tractable set of examples such as 2, in which contextual effects are more carefully controlled.

(2)

A: Are the children playing their game?

B: Paul took down the tent that they play their game in.

In 1, the accentability of the verb make is equivocal, as it is newly mentioned, but it is also a light verb which might be implied by the first clause. In our examples, the verb in the rejoinder is always given. We are concerned here only with the placement of the nuclear accent, in view of experimental results by Welby (2003) indicating that prenuclear accents are far less relevant to focus structure.

The two primary reference points for our theoretical analysis are Selkirk 1984, 1995 and Schwarzschild 1999. These proposals were developed against a backdrop of efforts to relate pitch accents to syntactic structure and information status that date back to Chomsky 1970. Liberman and Prince (1977), Ladd (1980), Pierrehumbert (1980), and Halle and Vergnaud (1987) all use autosegmental-metrical theory to handle the placement of accents in the default case where the sentence is entirely new. They also provide for shift of the accent to other locations, determined by the structure of the prosodic tree, in many of the more obvious cases of discourse givenness.

Selkirk (1984, 1995) advances beyond the works just cited by developing an explicit account of how focus structure propagates through syntactic trees. With certain simplifying assumptions, she predicts that the preposition is the only acceptable locus for a pitch accent in contexts like 2B. A closely related proposal is developed by Rochemont (1986, 1998).

The very significant contribution of Schwarzschild (1999) is a technical formulation of givenness. He proposes that accent placement is predictable for any sentence if the set of entailments of the discourse context is known. Based on those entailments, a rigorous semantic definition identifies syntactic constituents that bear a given relation to the discourse context. He uses optimality theory (Prince & Smolensky 1993, 2003) to formalize the relationship of givenness to accent placement. For any distribution of given and nongiven constituents in a syntactic tree, a set of hierarchically ranked, violable constraints determines how many accents are necessary and where they should occur. The account handles a number of cases that were problematic for Selkirk (1984, 1995), notably the case in which a nuclear accent is produced even if the entire intonational phrase is given. It makes explicit predictions about the placement of pitch accents in contexts like 2B, where a preposition is the only new information in its prosodic domain. Specifically, it predicts that an accent will occur on the preposition in such cases, since no other pattern satisfies the constraint hierarchy as well.

Schwarzschild (1999) and Selkirk (1984...