The role of prosody in the English dative alternation

Abstract

Does prosody influence the ordering of syntactic constituents? One syntactic phenomenon where prosody has occasionally been suggested to play a role is the English dative alternation. Building on proposals in earlier literature, we present a model of the dative alternation based on prosodic constraints interacting in terms of Optimality Theory. The model predicts the possible constituent orderings as well as the relative well-formedness of each ordering for verb phrases of different prosodic types. Data from two corpora that represent informal written and spoken English show that these predictions are largely confirmed. We conclude that the dative alternation exhibits prosodic effects that are mostly gradient and variable, yet entirely systematic. The prosodic hypothesis is thus a serious contender that must be taken into account in any attempt to explain the dative alternation in English.

Keywords:

• Dative alternation
• Prosody
• Optimality Theory

View correction statement:

Corrigendum

Acknowledgements

This work was partially funded by a grant from the Office of the Vice Provost for Undergraduate Education, Stanford University, and by the Department of Linguistics, New York University.

Some of this material is covered in an earlier version that appeared as Anttila (2008). This paper has benefited from presentations at the Berkeley Phonetics and Phonology Forum (May 2005), the Indiana University Phonology Fest (June 2006), the 26th West Coast Conference on Formal Linguistics at the University of California, Berkeley (April 2007), the LSA Linguistic Institute at Stanford University (July 2007), the GLOW Colloquium at Newcastle, U.K. (March 2008), the Stanford Phonology Workshop (April 2008), the Cornell Workshop on Theoretical and Empirical Approaches to Prosody (April 2008), the 39th Meeting of the North East Linguistic Society at Cornell University (November 2008), and at linguistics colloquia at the University of California, Santa Cruz (November 2007), University of Southern California (March 2008), the University of California, San Diego (May 2008), and the University of California, Los Angeles (November 2009). We thank the audiences for their feedback.

We are deeply indebted to Philipp Angermeyer, Rahul Balusu, and Peter Liem for compiling the Blogspot corpus in the summer of 2004, Anna Cueni and Gabe Recchia for compiling the Switchboard dative alternation corpus, Curtis Andrus for programming the T-Order Generator, and Joan Bresnan for sharing her insights and corpora as well as helping us with statistics and R. We also thank Max Bane, Uriel Cohen Priva, Vivienne Fong, Florian Jaeger, Dan Jurafsky, Paul Kiparsky, Anthony Kroch, Beth Levin, Peter Sells, Tom Wasow, Arnold Zwicky, and members of the Stanford Spoken Syntax Lab for useful input. We are solely responsible for any errors.

Notes

¹A different formulation is given in Truckenbrodt (2007) whose constraint StressXP requires that each syntactic phrase contain a beat of stress on the level of the prosodic phrase.

²An anonymous reviewer notes that in spoken language pronouns can be accented, citing Wolters and Beaver (2001), and concludes that the unstressed pronoun assumption is only a very rough approximation. Following the standard view in autosegmental-metrical phonology (see e.g., Ladd, 1996), we make a distinction between stress and pitch accent: stress denotes metrical prominence; pitch accents are tones. It is true that pitch accents provide a powerful diagnostic for lexical stress (see e.g., Hayes, 1995, pp. 10–11 for examples from English), but they cannot be identified with lexical stress. We thus cannot conclude from the existence of accented pronouns that pronouns have lexical stress. Indeed, Wolters and Beaver (2001) point out that in most cases accent on pronouns appears to be optional and interpretable as a sign of rhetorical contrast.

³Under our descriptive taxonomy, one long verb turns out to have three feet: (ad)(mini)(ster). If this description is correct, and if (ad)(mini)(ster) behaves prosodically like a two-foot verb, e.g. (de)(liver), we would need to say that *Ternary is violated by prosodic phrases that are ternary or longer.

⁴The Multiple Grammars Theory is a generic theory of variation that can be implemented in different ways. For example, Stochastic Optimality Theory (Boersma & Hayes, 2001) posits grammatical constraints with numerical weights called ranking values. The ranking value of each constraint is fixed, but at evaluation time a small amount of normally distributed noise is added to each constraint. This leads to limited variation in performance: constraints that are close together in weight are highly likely to reverse their mutual ranking from evaluation to evaluation, whereas constraints that are far apart in weight are highly unlikely to do so. The speaker's competence thus consists of a single grammar with numerically weighted constraints, but in actual performance we have multiple grammars evaluated in the standard optimality-theoretic fashion.

⁵Since the different length variables are correlated, including them all in the model results in collinearity. Including only orthographic words and primary stresses results in a better model. To determine this, we used the vif() function from R's Design library which shows collinearity to be within acceptable limits (< 4). We thank Joan Bresnan and an anonymous reviewer for drawing our attention to this problem and for suggesting ways to deal with it.