Children's learning biases for person systems

Person pronominal systems categorize entities based on their role in the context of speech (e.g., speaker, addressee, and non-participant other). The frequency of different person systems across languages is uneven: some person systems are quite common while others are rare or even unattested (e.g., Cysouw 2003). The cross-linguistic regularities of person systems have been taken to (at least partly) reflect constraints on how we represent the pronominal referents. A common view is that attested systems are grounded in semantic similarity; that is, they exploit pre-existing semantic or conceptual distinctions, such as the contrast between speaker and non-speakers, and the contrast between speech participants and non-participants (e.g., Harley & Ritter 2002, Harbour 2016, Ackema & Neelman 2018). In many of these theories, semantic distinctions are cashed out in terms of features. For example, several theories propose the existence of a feature speaker and a feature participant, which combined can account for many of the attested systems in the typology (e.g., Harbour 2016). Along these lines, Maldonado and Culbertson (2022) argue that both a bias for feature/semantic similarity and a specific bias for speaker distinctiveness shape the learning of novel pronominal systems. Specifically, they report a preference for systems that distinguish participants from non-participants, and a particularly strong preference for systems that distinguish the speaker from non-speakers; however, systems organized based on distinction between addressee and non-addressees (which are also typological unattested) are found to be more difficult to infer. Similarly, in a non-linguistic task, Maldonado et al. (2023) found that learners from diverse linguistic backgrounds tend to categorize person space based on the speaker/non-speaker distinction and participant/non-participant distinction, with the speaker/non-speakers distinction having a higher weight. Participants showed a much weaker bias to categorize the space based on the addressee/non-addressee distinction. Taken together, these results suggest that not all distinctions are equally natural; in particular, the speaker and participant features (or conceptual primitives) are more relevant to the representation of person systems, in line with many feature theories (e.g., Harbour, 2016). However, previous studies have focused exclusively on adult participants; consequently, it remains unclear how early these biases emerge. Crucially, young children have less experience with both the language and world, and also a more limited cognitive capacity, and hence might be less sensitive to the distinctions than adults. If they also show these two biases, it would provide stronger evidence for the claim that these two biases are conceptual constraints at work. In the present experiment, we investigate child learners’ preferences for pronominal systems, asking whether the two biases described above already manifest in the early stages of language development. Specifically, we test whether English-speaking 5–6-year-old children show preferences among pronominal systems based on (i) a speaker/non-speaker distinction (homophony between second- and third-person singular pronouns), (ii) a participant/non-participant distinction (homophony between first- and second-person singular pronouns), and (iii) an addressee/non-addressee distinction (homophony between first- and third-person singular pronouns).