Google's Whisk experiment decomposes images into "Subject," "Scene," and "Style," allowing surgical transfer and manipulation. What if writing, too, could be distilled to its fundamental axes? I propose five "eigenvalues" of expression. Not categories, but orthogonal dimensions. From these, any text might be reborn through a generative process.
Decomposing Text
1. Content
The meaning of the text—its topics, facts, events, ideas, and any background knowledge it assumes.
Examples:
• A travelogue describing Paris's landmarks (assumes readers know basic geography)
• A mystery story about a missing heirloom (introduces characters, motives, clues)
• An op-ed arguing for climate action (presents data, policies, values)
2. Structure
The logical or narrative organization—how ideas unfold, how paragraphs/sections connect, any footnotes or sidebars.
Examples:
• Inverted pyramid (headline → key facts → details)—typical of news articles
• Three-act story arc (setup → conflict → resolution)—typical of short fiction
• Point-counterpoint essay (thesis → opposing view → rebuttal → conclusion)
3. Voice
The author's lexical and syntactic signature—choice of words, sentence patterns, dialects, code-switching, jargon level.
Examples:
• Conversational blog ("you'll love how easy this is!") vs. academic paper ("the data suggest…")
• A teen's text-message style mixing slang and emoji vs. a formal business memo
• A bilingual narrator pausing mid-sentence to quote in Spanish then back to English
4. Mood
he text's emotional and interpersonal vibe—its tone, attitude, formality, and how it invites the reader to feel or respond.
Examples:
• Warm and encouraging ("Keep going—you've got this!")
• Stern and admonishing ("This behavior is unacceptable.")
• Neutral and factual ("The survey recorded 45% participation.")
5. Format
The type and size of the document—its genre conventions and overall length, which dictate things like presence of headings, paragraphs, or dialogue.
Examples:
• 300-word blog post with subheadings vs. 2,000-word magazine feature with pull-quotes
• A letter (date, greeting, sign-off) vs. a screenplay (character cues, scene breaks)
• A five-paragraph essay vs. a free-form personal reflection
Discussion
This framework naturally extends beyond traditional text. Consider how different media introduce new eigenvalues.
- Hyperlinked documents add a navigational dimension: the reader's agency in choosing paths through content.
- Technical documentation with version control introduces a temporal dimension: how ideas evolve through tracked changes and commit histories.
- Visual documents bring a spatial dimension: the semantic weight of placement, whitespace, and typography.
- Interactive notebooks like Jupyter add an executable dimension: where code, output, and narrative interweave in real time.
- Collaborative platforms introduce a social dimension: multiple voices negotiating meaning through comments, edits, and attributions.
The emergence of AI writing systems presents a particularly rich case. Here we find a generative dimension that encompasses the prompt, the context, the model's response, the human's steering, and the iterative refinement process. This isn't simply another voice or format; it's a new mode of authorship where intention and interpretation dance between human and machine.
These observations suggest fertile ground for systematic investigation. How might we decompose music into fundamental dimensions beyond melody, harmony, and rhythm? What eigenvalues define film beyond cinematography, narrative, and sound design? Could we identify the orthogonal components of performance art, where body, space, time, and audience participation intersect? Most urgently, as natural language becomes a programming medium and AI systems become co-authors, we need rigorous frameworks to understand these hybrid forms of expression. The work begins with careful observation: collecting examples, identifying patterns, and testing whether our proposed dimensions truly capture what makes each medium distinct. Only through such systematic analysis can we build tools that genuinely augment human creativity rather than merely mimic its surface features.
Appendix
Eigenvalues of this article
Dimension | Eigenvalue of This Article |
---|---|
Content | A conceptual proposal of five orthogonal “eigenvalues” (Content, Structure, Voice, Mood, Format) for generative text. |
Structure | Blog-style layout: brief intro → five numbered sections (each defining and exemplifying a dimension) → extended reflection and appendix. |
Voice | Informative-academic yet approachable: uses precise terminology alongside conversational touches and examples. |
Mood | Analytical and speculative: invites readers to theorize, experiment and extend the framework to other media. |
Format | Web essay with headings, bullets, prose paragraphs and an appendix; moderate length suitable for online reading. |