Social biases of mention order

Abstract

What can first mention, the entity mentioned first in an utterance, tell us about group identification, self-other equivalences and assumptions of agency? In the field of psycholinguistics, language is not routinely studied in relation to social identity or social contexts. Yet, language is a fundamental social behaviour that is often influenced by culture and society. Through a series of production experiments, this thesis shows that a speaker who is describing an interaction such as a hug taking place between someone from their ingroup and someone from their outgroup will demonstrate a tendency to mention the ingroup member first. Through investigating the effects of social biases in production, we also demonstrate that mention order has considerable effects on comprehension through the assumptions of agency that it creates. Thus, this thesis presents evidence establishing the importance of considering a speaker’s identity and the identities of the people they are talking about in studies of production and comprehension.

In Chapter 2, we show how relative similarity, according to gender and race, affect mention order decisions, testing two hypotheses. First, that mention order would reflect conventional social hierarchies resulting in a ‘man-first’ and a ‘white-first’ bias, and second, that speakers would show a ‘Like Me’ bias to mention the person of their own gender or of their own race first, over the person from their outgroup. In Experiments 1, 2 and 4, Black and white women and men viewed and described symmetrical interactions (like meeting or hugging) between two people who either differed in gender (one woman and one man) or in race (one Black person and one white person) and we measured who they mentioned first. Results gave some support for a man-first bias, however, they generally supported the second hypothesis. Importantly, we found a Like Me effect even when both names had the same grammatical and semantic role, as in conjoined sentences (“Beth and Fred were hugging”), and when the second-mentioned name signified agency as in passive sentences (“Beth was being hugged by Fred”). We did not find strong evidence to suggest that these mention order biases were driven by perceptions of agency based on the visual stimuli, for instance that men were perceived as looking more like agents than women, or that people perceived the person of their own gender or race as the more likely agents (Experiment 3).

To test the effect of mention order biases on comprehension, Chapter 3 (Experiment 5) investigated how first mention influences perceptions of agency. Participants viewed transitive and conjoined sentences describing symmetrical interactions and rated who they thought was the most likely agent of the event. We found that participants tended to rate the first-mentioned name as the most likely agent. When a woman and a man’s name were paired, the order effect was modulated by gender so that the first-mentioned agent bias was stronger when it was a man’s name rather than a woman’s name.

In Chapter 4, we tested the Like Me bias further using animacy as a factor of relative similarity. We show that speakers demonstrate a robust ‘human-first’ bias when describing symmetrical interactions between a human and a robot (Experiment 6). However, when we used narrative to alter beliefs about these two groups, making the robots more similar in qualities to the speaker and the humans as less similar, non-human even, we significantly reduced this human-first effect (Experiment 7). Our findings reveal important effects of dehumanisation on language production.

The studies in this thesis demonstrate the importance of a multidisciplinary approach to understanding bias and language.