Listener-oriented phonetic reduction and theory of mind

ABSTRACT

Predictable words tend to be phonetically reduced relative to unpredictable words. Under “listener-oriented” accounts of this phenomenon, the talker has tacit knowledge of their interlocutor's mental state. These theories consequently predict that individual variation in theory of mind is related to magnitude of probabilistic phonetic reduction. The current study tests this prediction for three acoustic variables (word duration, vowel duration, and vowel dispersion) in two definitions of predictability (contextual predictability and discourse mention). A relationship between individual variation in theory of mind and phonetic reduction was observed only for semantic predictability, and in the direction opposite to that predicted by listener-oriented theories. Taken together, these results are not consistent with the predictions of a strong interpretation of listener-orientation in speech production.

KEYWORDS:

• Listener modelling
• discourse mention
• semantic predictability
• vowel dispersion
• duration

Acknowledgments

I am grateful to Mary Beckman, Cynthia Clopper, Susanne Gahl, Becca Morley, Shari Speer, and two anonymous reviewers for comments and feedback on earlier versions of this paper. This work has benefited from feedback from audiences at Indiana University; the Laboratoire de Sciences Cognitives et Psycholinguistique; the New Zealand Institute for Language, Brain, and Behaviour; the Ohio State Center for Cognitive and Brain Sciences; the University of Edinburgh; the 18th and 19th Mid-Continental Phonetics and Phonology Conferences; the 22nd Manchester Phonology Meeting; and the 88th Annual Meeting of the LSA. I would also like to thank McKenna Reeher and Amy Roberts for their assistance with data collection and coding. All errors are my own.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. See also Silverman (2012) on the development of anti-homophony constraints in phonology: “Anti-homophony is thus not an active pressure for which there is an abundance of overt evidence. Rather, it is a passive result of the pressures that inherently act upon the interlocutionary process.” (p. 147, emphasis in original.)

2. Note that weaker interpretations of the listener-oriented account exist which do not necessarily require interlocutor-specific knowledge but instead rely on “generic listener” models (Turk, 2010). It is difficult to reconcile this approach with the classic literature on audience design in collaborative tasks (e.g. Brennan & Clark, 1996; Clark & Wilkes-Gibbs, 1986; Isaacs & Clark, 1987), which illustrate that talkers consistently employ listener-modelling in their choice of referring expression. The weaker forms of the listener-oriented accounts do not make explicit the conditions under which we expect talkers to use listener-specific knowledge, making it difficult to test the validity of this account.

3. Although the concept of ToM is usually regarded as originating with Premack and Woodruff (1978), the related concepts of metarepresentation and mentalising have been part of biological thinking for many decades; see chapter 12 of Hobhouse (1901) for an early treatment of the subject.

4. Clopper et al.'s (2017) materials were, in turn, adapted from those of Baker and Bradlow (2009).

5. The original coding system outlined by Baron–Cohen, Wheelwright, Skinner et al. (2001) uses a binary scoring system, collapsing the distinction between “strongly (dis)agree” and “(dis)agree”. Following more recent work (e.g. Stevenson & Hart, 2017; Yu, 2010), this study used a Likert-style coding system which takes into account all four values. Possible scores on the AQ measured this way range from 50 to 200.

6. Equivalent rectangular bandwidth (ERB), is a psychoacoustically motivated scale similar to the Bark scale.

7. This correlation is significant even after applying a Bonferroni correction: $\alpha =.05/6\approx .008>.001$.

8. Pilot testing revealed that participants found presentations of 3500 ms in the clear speech condition to be “too fast”. This lengthened time also implicitly encouraged the participants to speak clearly.

9. Whether or not these adaptations are actually helpful for speech intelligibility is a separate question (see e.g. Ferguson, 2004; Picheny et al., 1985).

10. Indeed, it is likewise difficult to reconcile this weaker interpretation with the existing literature on audience design in collaborative tasks. In a classic task in the action-language tradition, two interlocutors worked together to arrange a set of tangram figures (Clark & Wilkes-Gibbs, 1986). During the course of the task, the interlocutors developed “conceptual pacts” whereby the name of a particular figure was conventionalised, such as “the ice-skater” (Brennan & Clark, 1996). A similar study by Isaacs and Clark (1987) used New York City landmarks instead of tangrams, and had dyads of two experts (native New Yorkers) or an expert and a novice (people who had never visited New York). They found that interlocutors are extremely adept at quickly determining the knowledge level of their dialogue partner, and tailoring their productions accordingly. These findings illustrate that talkers consistently employ listener-modelling in their choice of referring expression. The weak form of the listener-oriented account quoted above holds that talkers can do listener-modelling when needed but that it is not required. This version of the theory does not make explicit the conditions under which we expect talkers to use listener-specific knowledge, which means that testing the validity of this account is difficult.

11. Regardless of the fine structure of the subscales, however, the full AQ score has been demonstrated to be reliable (e.g. Austin, 2005), to compare favourably with other measures of autistic traits (Armstrong & Iarocci, 2013), and to not be easily explainable in terms of personality traits (Wakabayashi, Baron-Cohen, & Wheelwright, 2006), suggesting that the AQ does indeed measure cognitive style rather than simply idiosyncratic personal preferences.

Additional information

Funding

This research was partially supported by a Presidential Fellowship from the Ohio State University Graduate School.