This project analyzes language attitudes toward English accents using the Stella Survey dataset. Non-native English speakers rated multiple accent recordings on two perceptual dimensions:

• Status (prestige / competence)
• Solidarity (friendliness / approachability)

The project includes a full reproducible pipeline:

1. Data cleaning + reshaping from raw survey CSV
2. Exploratory statistical analysis of hypotheses
3. Publication-style visualizations and summary tables in Quarto

H1. Standard vs. Non-standard accents

• Standard accents will score higher on status
• Non-standard accents will score higher on solidarity

H2. Speaker gender effects

• Male vs female speakers may be rated differently on status/solidarity.

H3. RP vs GA

• UK Received Pronunciation (RP) will score higher on status than US General American (GA).

/data
 stella_survey.csv
 stella_survey_clean.rds
/outputs
 01_analysis.docx
/scripts
 00_data_cleaning.qmd
 01_analysis.qmd
README.md

The cleaning script:

• reads the raw CSV (treating -99 as missing)

• filters for interview-mode participants

• selects participant metadata + rating variables

• reshapes wide → long with pivot_longer()

• splits recording IDs into:

  • recording_id
  • country (UK vs US)
  • standard (standard vs non-standard)
  • speaker_gender

• recodes participant variables (gender, student status)

• parses numeric years of English instruction

• converts key variables to factors

• saves a tidy dataset as .rds

Output:

data/stella_survey_clean.rds

The analysis script:

• loads the cleaned .rds
• creates participant-level metadata (deduplicated)
• computes descriptive summaries per hypothesis
• produces plots with ggplot2 + viridis palettes
• formats tables with flextable and kableExtra
• renders a report to .docx

• Status: Standard accents received higher status ratings than non-standard accents.
• Solidarity: Standard and non-standard accents showed no meaningful difference.

Interpretation: Standardness strongly affects perceived prestige, but not perceived warmth.

• Male speakers were rated higher on status.
• Female speakers were rated higher on solidarity.

Interpretation: Accent perception is influenced by gendered expectations (authority vs approachability).

• UK (RP) accents scored higher on status and solidarity than US (GA) accents.

Interpretation: Participants show a consistent preference for UK English across both prestige and affective dimensions.

• Majority aged 20–26
• Gender imbalance: ~67% male, ~33% female
• Most participants were not students
• English instruction clustered around 8 years
• English contact was polarized (many very low, many high)

These characteristics likely shape the attitude patterns above.

install.packages(c(
 "tidyverse", "readr", "knitr", "kableExtra",
 "flextable", "viridis"
))

Render:

scripts/00_data_cleaning.qmd

This regenerates the cleaned dataset.

Render:

scripts/01_analysis.qmd

This regenerates all figures and tables.

• Data cleaning & wrangling (filtering, selecting, pivoting, recoding)
• Regex-based metadata extraction from recording IDs
• Tidy long-format survey design
• Exploratory statistical analysis
• Data visualization (ggplot2, accessible palettes)
• Publication-ready reporting (Quarto → Word, flextable)
• Reproducible workflow structuring
• Linguistic interpretation of perceptual data

The dataset contains anonymized survey responses only (age, gender, learning history, ratings). No personally identifying information is included.

Ersin Gültekin M.A. Linguistics — University of Freiburg Computational Linguistics / NLP / Quantitative Linguistics