Framing the End of the World. A Computational Analysis of Semantic Divergence in "Don't Look Up"

A linguistic analysis of language structures

This project analyzes the screenplay of Adam McKay's film Don't Look Up using Natural Language Processing (NLP) methods. The objective is to empirically demonstrate that the communicative failure depicted in the film is not merely a plot device, but a structural inevitability.

By operationalizing sociological theories (Schütz, Lakoff) into measurable data vectors, the analysis proves that the catastrophe is caused not by a lack of knowledge, but by disjointed "finite provinces of meaning" and a lack of semantic intersection.

---

Theoretical Framework

The data analysis is based on two theoretical approaches.:

1. Alfred Schütz (Phenomenological Sociology): The theory of Finite Provinces of Meaning. The characters inhabit separated realities (Science vs. Media/Politics) between which no translation is possible.
2. George Lakoff (Cognitive Linguistics): Theories of Framing and Hypocognition. The study investigates how threats ("Comet") are linguistically reframed into market opportunities ("Assets") and why society lacks the cognitive frames to grasp "The End" (Hypocognition).

---

Methodology & Tech Stack

• Language: Python 3.x
• Libraries: spaCy, scikit-learn, pandas, matplotlib, seaborn, nltk, wordcloud
• Model: en_core_web_md (utilizing 300-dimensional Word Embeddings)

Steps

1. Parsing: Extraction of dialogue from the raw screenplay & cleaning (removal of stage directions and metadata).
2. Clustering: Aggregation of characters into 4 sociological archetypes:

• 1. SCIENCE (Dr. Mindy, Dr. Oglethorpe)
• 2. ACTIVISM (Kate Dibiasky)
• 3. POPULISM (President Orlean)
• 4. CAPITALISM (Peter Isherwell)

3. Analysis:

• Vector Space Analysis: Calculation of Cosine Similarity between worldviews.
• Frequency Analysis: Deterministic evaluation of keywords (Dictionary Approach).
• Framing Analysis: Visualization of semantic orientation via Radar Charts.

---

Key Findings

1. The Semantic Gap (Heatmap)

Vector space analysis reveals that Science and Activism, though natural allies, share a semantic similarity of only 0.40. They utilize incompatible codes (Rationality vs. Morality). Simultaneously, Science exhibits a dangerous proximity to Capitalism (0.63), explaining its vulnerability to technocratic co-optation.

2. Competing Cognitive Frames (Radar Chart)

Following Lakoff's theory that different worldviews activate distinct conceptual metaphors, the radar chart shows minimal shared semantic space.

3. Divergent Relevance Structures (Bar Charts)

While scientists focus on the Object (Comet), politicians fixate on the Subject (Voters/Crowd). There is no intersection in their relevance structures. Furthermore, the analysis shows that terms like "Death" appear in the Capitalist cluster but are technocratically redefined (as "Evolution" or "Algorithm").

---

Installation & Usage

1. Clone Repository:

git clone https://github.com/peter2b/dont_look_up

pip install spacy pandas matplotlib seaborn nltk scikit-learn wordcloud

3. Download SpaCy Model: The medium-sized model is required for vector calculations:

python -m spacy download en_core_web_md

4. Run Analysis: Execute the Jupyter Notebook or the script:

jupyter notebook Frame_Analysis_DONT_LOOK_UP_fin.ipynb

---

License & Credits

This analysis is based on the original screenplay by Adam McKay&Sirota (Don't Look Up, 2021). Created for the seminar Inter-American Environmental Studies (25W 975.036) at the University of Graz.

Author: Peter Bauer Date: 15.2.2026