Production preferences cannot be understood without reference to communication

MacDonald (2013) proposes that comprehenders are sensitive to statistical patterns in their language input (Claim 1). These patterns are hypothesized to result from speakers' preferences in production, aggregated over the population (Claim 2). Production preferences are taken to be primarily determined by biases that serve production ease, thereby improving fluency (Claim 3). These three claims, together constituting the core of the PDC, are an ambitious endeavor to tie together several lines of research in psycholinguistics and linguistics. Here, I focus on the second and third claim, that it is predominantly “production ease,” rather than communicative pressures, that drives production preferences and hence language form (M, p. 13; cf. Bard et al., 2000; Ferreira and Dell, 2000; Arnold, 2008; Ferreira, 2008; Lam and Watson, 2010). In contrast, I argue that production preferences and language form are unlikely to be understood without reference to communication. Specifically, production preferences are the result of at least two competing type of biases: biases toward production ease and biases toward ease, or at least success, of comprehension (Zipf, 1949). I refer to a weak version of the second type of bias as robust information transfer.1 Two hypotheses about how robust information transfer might affect production preferences are often conflated in the literature. First, speakers might continuously “estimate” their interlocutors' beliefs and structure their utterances based on these estimates. This claim, often referred to as audience design, is what production researchers (incl. M) tend to have in mind when they reject the idea that production preferences are affected by communicative biases. Many consider this claim implausible because production seems too demanding to allow additional computations (Ferreira, 2008). I share Tanenhaus's position that such intuitions are often misleading (Tanenhaus, 2013). Here, however, I pursue an alternative hypothesis, that communicative biases affect production preferences through learning and generalization across previous experiences (building on Jaeger and Ferreira, in press).