Published online by Cambridge University Press: 01 January 2026
Since Du Bois's (1987b) seminal paper, ergative alignment in morphosyntax has been claimed to correlate with a characteristic constellation of argument realization in discourse: both intransitive subjects (S) and transitive objects (P) serve to introduce new referents via full noun phrases (NPs), while transitive subjects (A) are dispreferred for this function and are thus mostly realized as pronouns or zero (e.g. Dixon 1995, Du Bois et al. 2003, Goldberg 2004). This ergative patterning in discourse is generally accounted for in terms of information-management strategies employed by speakers in dealing with the cognitive demands of introducing and monitoring referents in discourse. These claims have recently been questioned by Everett (2009), whose data (English and Portuguese) show no support for the claimed ergative bias in discourse and raise doubts about explanations in terms of information management. The present article subjects the claims of an ergative bias in discourse to more rigorous testing, drawing on the largest database compiled to date (nineteen spoken-language corpora from fifteen typologically diverse languages), and assesses the explanatory frameworks. We find that, with the exception of Du Bois's original Sakapultek data, there is very little evidence for the postulated ergative pattern in natural spoken-language discourse crosslinguistically. Although our findings do confirm low levels of full NPs in the A role (Du Bois's ‘Non-lexical A’ constraint), we concur with Everett (2009) that the semantic feature [±human] provides an empirically more sound and conceptually more economical account than earlier explanations framed in terms of information management. Finally, we address the plausibility of emergentist claims for a diachronic link between ergative alignment in morphosyntax and information flow in discourse. The raw data used in this article and extensive exemplification of the methodology employed are available as online supplementary materials.
The research reported in this article was supported by a Discovery Early Career Researcher Award from the Australian Research Council to Stefan Schnell (grant DE120102017, 2012-2015) and through internal research funding by the University of Bamberg. Schnell also thanks the Vera'a community from North Vanuatu for their support and engagement in the ongoing collaborative documentation of the Vera'a language, which was supported by two grants in the VolkswagenStiftung's DoBeS program (grants II/81 898 and II/84 316, 2006-2012). Most of this research was undertaken while Schnell was a member of the Centre for Research on Language Diversity at La Trobe University; thanks for helpful comments and discussion go to Anthony Jukes, Stephen Morey, and Pavel Ozerov. Earlier versions of this article were presented on various occasions, and we would like to thank the audiences at the following venues: LACITO, Paris Sorbonne (May 2012); Linguistic seminar, School of Languages and Linguistics, University of Melbourne (July 2012); University of Konstanz (May 2013); and the Ludwig-Maximilian University, Munich (2013). We thank the following colleagues for collaboration on the GRAID annotation scheme: Johanna Andrees, Ulrike Mosel, Florian Siegl, Claudia Wegener, Dagmar Jung, Hanna Thiele, and Nils Schiborr. For assistance with data handling and statistical analyses we thank Jenny Herzky and Nils Schiborr. The raw data is deposited at the Multi-CAST website hosted by the Language Archive at the University of Cologne (LAC; https://lac.uni-koeln.de/en/); our thanks to Felix Rau and Jonathan Blumtritt from LAC for technical assistance. The online supplementary materials can be accessed at http://muse.jhu.edu/article/628202/pdf. All remaining errors and shortcomings are our responsibility.