Supporting material for Arnold (2024) Why does recency guide pronoun comprehension? |

Supporting material for:

Arnold, J. E. (under review). Why does recency guide pronoun comprehension? It’s not just topicality, attention, or predictability. Ms., UNC Chapel Hill.

This project was supported by NSF grant #1651000 to J. Arnold

Abstract

It is well known that comprehenders assume that pronouns are more likely to refer to recently-mentioned referents than less-recent referents, for example in Ana ordered a croissant with her latte. Liz ordered a cappuccino. She grabbed a table by the window, people are more likely to assign she to Liz than Ana. We seek to understand why this bias occurs by testing which features can modulate this recency bias, and how these features relate to the predictability of the referents. We consider three proposals about what underlies the recency bias: 1) recent things tend to be topical; 2) recent things tend to be attended by the speaker and listener; and 3) recent thing tend to be predictable. Results confirm that recency is a strong constraint for pronoun comprehension, and this bias can be modulated by pointing (which is an explicit signal of the speaker’s meaning) but gazing (a signal of the speaker’s attention) has a smaller effect. Discourse topicality, signaled through repeated mention, had very little effect. Exp. 2 demonstrated that predictability has a complicated relationship with these constraints. In sum, the strongest effects on pronoun comprehension were the most proximal signals: recent mention in subject position and social signals about the speaker’s meaning and attention. While recency is associated with topicality, attention, and predictability, it cannot be reduced to any one dimension.

Data and models can be found at: https://osf.io/y45qp/?view_only=acdd5afc7d5f40fc9698261e08566964

An example video stimulus for Experiment 2 is shown below in each condition. The audio stimuli for Exp. 1 were the same as for a subset of the stories in Exp. 2 (but without the videos). The stimuli for Experiment 3 were truncated versions of the Experiment 2 videos.

Discourse topic recent / Gaze to neutral