This project includes data and code for the following paper: **Schmidt, H., Tran, S., Medaglia, J.D., Ulichney, V., Mitchell, W.J., & Helion, C. (under review). Conversational Linguistic Features Inform Social-Relational Inference. *PsyArXiv https://doi.org/10.31234/osf.io/fn4my*** Additional information can be found on [GitHub][1]. ---------- ## Abstract ## Whether it is the first day of school or a new job, individuals often find themselves in situations where they must learn and integrate into novel social relationships. However, the mechanisms through which individuals evaluate the strength and nature of these existing relationships – social-relational inference – remain unclear. We posit that linguistic features of conversations may help individuals evaluate social relationships and may be associated with social-relational inference. Leveraging a naturalistic behavioral experiment (57 adults; 34,735 observations), participants watched a mid-season episode of Survivor and evaluated the observed dyadic relationships between contestants. We employed novel person- and stimulus-focused approaches to 1) investigate social-relational inference similarity between participants, 2) examine the association between distinct linguistic features and social-relational inference, and 3) explore the relationship between early season conversation similarity and later perceived relationship formation. We found high pairwise participant response similarity across three experimental conditions, distinct associations between relational judgments and linguistic features, including semantic similarity, sentiment, and clout, and no evidence of an association between early conversation similarity and later friendship inference. These findings suggest that naturalistic conversational content is both a potential mechanism of social-relational inference and a promising avenue for future research. ---------- ## Data ## The `data` folder contains all de-identified data. ---------- ## Scripts ## All scripts associated with this project are included. Analyses were performed in R and Python: R scripts are formatted as .qmd files and Python scripts are formatted as .ipynb files. R analytic scripts are numbered for reproducibility. Clout scores are calculated using the Linguistic Inquiry and Word Count (LIWC) software, semantic similarity is calculated using the Universal Sentence Encoder (4) in Python, and sentiment is calculated using the `sentimentr` package in R. - `1-data-cleaning-and-setup` - Loads raw task data, calculates sentiment scores, loads similarity (see USE subfolder) and clout scores, combines all data - `2-rsa-behavioral` - Examines task response similarity across participants using representational similarity analysis (RSA) - `3-bayesian-clout-modeling` - Fits Bayesian statistical models to examine association between social-relational inference and conversational clout - `3-bayesian-sentiment-modeling` - Fits Bayesian statisical models to examine association between social-relational inference and conversational sentiment - `3-bayesian-similarity-modeling` - Fits Bayesian statistical models to examine association between social-relational inference and conversational semantic similarity - `4-all-features-modeling` - Fits LASSO regression model to examine overlapping contributions of all three conversational linguistic features - `5-whole-season-transcription-wrangling` - Formats transcriptions for first five episodes in the Survivor season - `6-whole-season-bayesian-similarity-modeling` - Fits Bayesian statistical models to examine association between early semantic similarity and later social-relational inference - `7-supplemental` - Creates some supplemental plots (others are created in earlier scripts) - `8-extract-posterior-draws` - Extract posteriors from Bayesian models and create plots ---------- ## Output ## Please note that the output folder does not currently contain any files. However, if you download the project and wish to recreate any analyses, all figures, tables, and models will be saved into this folder. ---------- Please direct any comments or questions to **Helen Schmidt** at helen_schmidt@temple.edu. Please feel free to use any of these scripts, but unfortunately I am unable to provide support if you adapt them to your own data. If you notice a bug or error, please tell me! [1]: https://github.com/hschmidt12/Conversational-Linguistic-Features-Inform-Social-Relational-Inference