Read the preprint here:

van der Walle, H. A., Margulis, E. H., Jakubowski, K. (2026). Framed Listening: Can accompanying contextual cues shape Music-Influenced Mental Content?. PsyArXiv. https://doi.org/10.31234/osf.io/zup69_v1.

Framed Listening

Does contextual framing shape what listeners think about while hearing music?

This repository contains all data, analysis code, and outputs for the Framed Listening project (Hazel A. van der Walle, PhD, Music, Durham University). The study investigates whether brief contextual framing cues — descriptions of the contextual situation in which a piece of music may be typically heard (e.g., bar, concert, movie, video game) — shape the semantic content of listeners' music-influenced mental content (MIMC), and whether this varies across musical genres.

https://hazelvdw.github.io/context-framed-listening/

---

Study Overview

Participants listened to sixteen musical excerpts from four genres (80s-pop, Electronic, Jazz, Metal) while being given one of four contextual framing cues. After each clip they described any thoughts, images, or mental experiences they had. A supplementary study collected independent believability ratings for all 64 clip–context pairings.

Semantic similarity of MIMC descriptions was quantified using two complementary NLP pipelines:

• MiniLM (all-MiniLM-L6-v2) — sentence-embedding semantic similarity at the individual response level
• TF-IDF — lexical similarity at the aggregated document level (combMIMC)

Three research questions were addressed:

1. RQ1 — Do contextual framing cues shape MIMC similarity?
2. RQ2 — Does music or context drive greater MIMC convergence? Does combined alignment produce additional convergence?
3. RQ3 — Does musical genre moderate the context effect?

---

Repository Structure

context-framed-listening/
│
├── dataMAINstudy.csv              # Main study raw data (participant MIMC descriptions)
├── dataSUPPstudy.csv              # Supplementary study believability ratings
├── LICENSE
│
├── NLP_analyses/                  # Analysis pipeline (run in order)
│   ├── text-prep.qmd              # 1. Spell-checking, preprocessing, combMIMC aggregation
│   ├── TFIDF-analysis.qmd         # 2. TF-IDF cosine similarity (combMIMC level)
│   ├── SEmb-analysis.qmd          # 3. Sentence-embedding similarity (individual level)
│   ├── model-comparison.qmd       # 4. Cross-model comparison (9 models × 2 levels)
│   ├── requirements.txt           # Python minimum-version dependencies
│   ├── constraints.txt            # Python exact-version pins
│   ├── DESCRIPTION                # R package metadata (used by renv)
│   ├── renv.lock                  # R exact-version lockfile
│   └── REPRODUCIBILITY.md         # Full setup and reproduction instructions
│
├── NLP_outputs/                   # Generated outputs (not tracked by git if large)
│   ├── combMIMC_lvl1a.csv         # Aggregated MIMC, light preprocessing + text_original
│   ├── combMIMC_lvl1b.csv         # Aggregated MIMC, light preprocessing (thought words kept)
│   ├── combMIMC_lvl2a.csv         # Aggregated MIMC, aggressive preprocessing + text_original
│   ├── combMIMC_lvl2b.csv         # Aggregated MIMC, aggressive preprocessing (thought words kept)
│   ├── dataMIMC_preprocessed.csv  # Individual MIMC responses (all preprocessing levels)
│   ├── dataMIMC.csv               # Cleaned individual MIMC data
│   ├── clipContextDescrStats.csv  # Descriptive statistics by clip × context
│   ├── misspellings_df.csv        # Flagged misspellings for manual review
│   ├── misspelling_c...csv        # Applied correction mapping
│   ├── model_comparison/          # Cross-model comparison outputs and embedding caches
│   ├── SentenceEmbedding/         # MiniLM outputs and embedding cache
│   ├── TFIDF/                     # TF-IDF outputs
│   └── supp_study/                # Supplementary study analysis outputs
│
├── PsychoPy-files/                # Experiment files
│   ├── PsychoPy-exp-master.js
│   └── PsychoPy-exp-master.psyexp
│
└── supp-study/                    # Supplementary study analysis
    ├── suppstudy-analysis.qmd
    ├── suppstudy-analysis.html
    ├── ptp-descriptive-data.html
    └── output/suppstudy/          # Figures and exported tables

---

Analysis Pipeline

The four QMD files in NLP_analyses/ should be run in order:

Step	File	Purpose
1	text-prep.qmd	Spell-check, stop word removal, lemmatisation, combMIMC aggregation
2	model-comparison.qmd	Cross-model comparison: 9 models × 2 data levels, rank correlation, top/bottom pair inspection
3	TFIDF-analysis.qmd	TF-IDF vectorisation and cosine similarity at the combMIMC level
4	SEmb-analysis.qmd	Sentence-embedding similarity at the individual level; LMMs; t-SNE

Each file includes a version-check chunk that confirms all dependencies are correctly installed before any analysis runs.

---

Preprocessing Levels

Two preprocessing pipelines are applied, tailored to model requirements:

Level	Used for	What is removed
lvl1	MiniLM (light)	Custom stop words; punctuation and word forms preserved
lvl2	TF-IDF (aggressive)	Custom stop words + NLTK English stop words + lemmatisation; punctuation and case lowered

---

Reproducibility

See REPRODUCIBILITY.md for full setup instructions. In brief:

Python:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

R:

install.packages("renv")
renv::restore()
# If knitr/rmarkdown are missing after restore:
renv::install(c("knitr", "rmarkdown", "reticulate"))
renv::snapshot()

Rendering:

quarto render text-prep.qmd
quarto render TFIDF-analysis.qmd
quarto render SEmb-analysis.qmd
quarto render model-comparison.qmd

The Word2Vec model (~1.6 GB) and all sentence embeddings are cached locally on first run; subsequent runs load from cache. See REPRODUCIBILITY.md for cache management.

---

Dependencies

Type	Key packages
Python	pandas, numpy, scikit-learn, sentence-transformers, torch, gensim, nltk, pyspellchecker, matplotlib, seaborn, wordcloud
R	lme4, lmerTest, MuMIn, emmeans, ARTool, randomForest, MASS, ggplot2, dplyr, dunn.test
Quarto	≥ 1.4
Python	3.10–3.12
R	≥ 4.2

---

Citation

If you use this code or data, please cite:

van der Walle, H. A., Margulis, E. H., Jakubowski, K. (2026). Framed Listening: Can accompanying contextual cues shape Music-Influenced Mental Content?. PsyArXiv. https://doi.org/10.31234/osf.io/zup69_v1.

---

Licence

Code: MIT Data and written materials: CC BY 4.0