Code to replicate analyses in Bergen & Trott (submitted)

About the paper

This paper is an extension to an earlier publication (Trott & Bergen, 2018), which found:

1. Comprehenders are more likely to interpret a potential indirect request (e.g. "It's cold in here") as a request when the speaker can be inferred to be unaware of an obstacle to fulfilling the request (e.g. a broken heater).
2. Participants with higher mentalizing ability, as measured by the Short Story Task (Dodell-Feder et al, 2013), showed a stronger effect of Speaker Awareness. That is, better mentalizers were more likely to modulate their pragmatic interpretations as a function of what a speaker could be inferred to know.

Previous work left a critical question open: is the effect of mentalizing due primarily to better mentalizers being more likely to sample information about a speaker's knowledge states in the first place, or are better mentalizers also more likely to deploy this information?

In the current work, we manipulate both the task participants are asked to perform (Experiments 1a-1b), as well as how explicitly a speaker's knowledge states are given in a passage (Experiment 2), and find that mentalizing predicts both sampling and deployment rate, provided the task explicitly involves knowledge in some way. This suggests that comprehenders might recruit their mentalizing capacity flexibly, in a task-dependent manner.

Data

The aggregated critical data is found under Experiment1/data and Experiment2/data respectively.

Note that this data does not include identifying information about a participant's gender or age. Please contact Sean Trott (sttrott at ucsd dot edu) separately for demographic information.

Primary variables

The following factors are consistent across both experiments:

Condition: was the speaker aware or unaware of an obstacle to fulfilling a request?
Factor levels: Speaker Unaware, Speaker Aware

answer: participant response (yes or no).
(Note that for Experiment 1, the meaning of this variable changes depending on experimental group; see notes below.)

Order: trial order (continuous, 1-8 or 1-16)

Experiment 1

exp_group: Participants in Experiment were assigned to one of two experimental groups: Inference or Knowledge
Individual difference variables:

• exp.inf.mean: Mean explicit mental state reasoning score (averaged across both coders).
• rc.mean: Mean reading comprehension score (averaged across both coders).
• rc.bins: Binned reading comprehension (created in experiment1_analysis.Rmd)
• spon.true: Spontaneous mental state reasoning scores, including tiebreaker codes.

numeric_categorization2: recoded version of answer, to align yes responses across experimental groups.
In the Knowledge group, a "Yes" response would by default be correct in Speaker Aware condition, e.g. "Yes, the speaker is aware of the obstacle"; whereas in the Inference group, a "Yes" response would by default be correct in the Speaker Unaware condition, e.g. "Yes, the speaker is making a request". This recoding inverts Knowledge answers, such that a "Yes" now corresponds to: "Yes, the speaker is unaware of the obstacle". This allows for direct comparison across experimental groups in which the same expected outcome corresponds to each condition.

Experiment 2

Individual difference variables:

• exp_inf_mean: Mean explicit mental state reasoning score (averaged across both coders).
• rc_mean: Mean reading comprehension score (averaged across both coders).
• rc_bins_1: Binned reading comprehension (created in experiment1_analysis.Rmd)
• spon_true: Spontaneous mental state reasoning scores, including tiebreaker codes.

Random factors

• subject: subject ID
• stimNum: item number

Analyses

The analyses can be found in the .Rmd files:

• Experiment1/experiment1_analysis.Rmd
• Experiment2/experiment2_analysis.Rmd