Mapping prosody onto meaning – the case of information structure in American English^*

* TM and JC conceived the idea for this research and developed the experimental design. TM generated stimuli, carried out the experiment, and assembled the data. TR analyszed the data statistically and archived materials, data, and scripts in online repositories. TR and JC wrote the manuscript.

ABSTRACT

Prosody is a central part of human speech, with prosodic modulations of the signal expressing important communicative functions. Yet, the exact mechanisms of how listeners map prosodic aspects of the speech signal onto speaker-intended discourse functions are only poorly understood. Here we present three perception experiments that test the mapping between the prosodic form of a heard utterance and possible information structural categories (here: focus and givenness) determined by a discourse context. Results suggest varying degrees of accuracy dependent on the specific information structure categories that are presented to the listener in the experiment (the target and the competitor). Moreover, listeners are sometimes biased towards or against certain discourse contexts. These biases are compatible with the idea that listeners infer speaker intentions based not only on bottom-up processing of acoustic cues but also on probabilistic knowledge about how likely prosodic forms co-occur with specific discourse contexts.

KEYWORDS:

• Prosody
• intonation
• focus
• information structure
• American English

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Timo B. Roettger http://orcid.org/0000-0003-1400-2739

Notes

* TM and JC conceived the idea for this research and developed the experimental design. TM generated stimuli, carried out the experiment, and assembled the data. TR analyszed the data statistically and archived materials, data, and scripts in online repositories. TR and JC wrote the manuscript.

1 The findings from these experiments are also discussed in Mahrt (2018), with qualitative comparisons across experimental conditions.

2 The ToBI labels represent the one or two most frequent pitch accents produced on the subject and object nouns, over the nine sentence stimuli in each focus condition.

3 Our decision to administer both tasks (1C-2P, 2C-1P) to the same participants in Experiment 3, rather than run two separate experiments with the expanded, scalar response set, were driven by practical constraints of time and money. We were also interested in testing the feasibility of combining both tasks in one experiment, to approximate the design of earlier experiments, testing categorical perception of phoneme contrasts, in which identification and discrimination experiments are administered to the same participants. Mahrt (2018, p. 30) reports on a pilot experiment using both tasks, involving 45 subjects recruited from Mechanical Turk. In the pilot experiment, a third of the participants first did the identification task followed by the discrimination task, another third did the tasks in the opposite order, and the final third did them with the tasks interleaved in random sequence over trials. The same set of stimuli were presented to all participants. The results of the pilot showed that the task done second resulted in higher accuracy than when it was performed first, with intermediate accuracy for participants in the interleaved tasks condition. On the basis of these findings, Experiment 3 adopted the interleaved task design.

4 This adaptive behaviour is in line with language users adapting readily to their immediate local context in syntax (e.g. Fine, Jaeger, Farmer, & Qian, 2013; Jaeger & Snider, 2013), pragmatics (e.g. Grodner & Sedivy, 2011; Yildirim, Degen, Tanenhaus, & Jaeger, 2016), and, most importantly, in speech (e.g. Kleinschmidt & Jaeger, 2015; Norris et al., 2003).