💬 NLP

From graphemic dependence to lexical structure: a Markovian perspective on Dante's Commedia

This study uses a four-state Markov chain model based on vowel-consonant encoding to demonstrate that Dante's *Divine Comedy* exhibits a progressive shift in graphemic dependency from *Inferno* to *Paradiso*, revealing a structured relationship between local symbolic patterns, lexical organization, and the poem's overall macro-structure.

Original authors: Angelo Maria Sabatini

Published 2026-04-27

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Angelo Maria Sabatini

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are looking at a massive, intricate mosaic made of millions of tiny tiles. If you stand too close, you only see individual colors. If you stand too far away, you only see the big picture. This paper is about a scientist who found a way to look at the "texture" of the tiles to understand how the artist (Dante) built the masterpiece.

Here is the breakdown of the study in plain English:

1. The "DNA" of Words (The V/C Encoding)

Instead of reading Dante’s Divine Comedy for its meaning, the researcher stripped the poem down to its barest bones. He ignored the actual words and looked only at the pattern of Vowels (V) and Consonants (C).

The Analogy: Imagine listening to a song, but instead of hearing the lyrics, you only hear the rhythm of the drumbeat (the consonants) and the melody of the flute (the vowels). Even without the words, you can still feel the "vibe" or the tempo of the music. This is what the researcher did with the text.

2. The "Memory" of the Text (The Markov Chain)

The researcher used a mathematical tool called a Markov Chain to see how much one letter "remembers" the one before it. Does a vowel usually follow a consonant, or do they like to hang out in long groups? He called this "Memory Depth."

The Analogy: Think of a person walking through a forest. If they always follow a straight path, they have "high memory"—their next step is very predictable based on their last one. If they are dancing wildly and changing directions every second, they have "low memory." The researcher found that as Dante moves from the dark, chaotic Inferno toward the bright, orderly Paradiso, the "walking pattern" of the letters becomes more structured and predictable.

3. The "Probes" (Finding the Patterns)

The researcher looked for specific "trigrams"—little three-letter patterns (like CVC or VVC). He called these "Graphemic Probes." Some patterns were like "glue" that held words together, while others were like "bridges" that connected one word to the next.

The Analogy: Imagine you are inspecting a brick wall. Some patterns of bricks are used to build the solid center of a brick (intra-lexical), while other patterns only appear where two bricks touch (word boundaries). By tracking these patterns, the researcher could see how the "construction style" of the poem changed.

4. The "Anchors" (Connecting Math to Meaning)

Finally, he wanted to make sure this math wasn't just nonsense. He used AI to find which actual words were most common in each section (Inferno, Purgatorio, Paradiso). He then "anchored" his math patterns to these words.

The Analogy: It’s like finding out that a specific "drumbeat" pattern you discovered always happens when the lyrics mention "fire" or "darkness." By linking the rhythm to the meaning, he proved that the mathematical changes he found weren't accidents—they were part of Dante's deliberate stylistic journey.

The Big Picture

The paper concludes that Dante’s poem isn't just a story; it is a mathematical progression.

As the soul travels from Hell to Heaven, the very "texture" of the language changes. The words get longer, the "clunky" shortcuts (like apostrophes) disappear, and the patterns of vowels and consonants become more harmonious and organized. The researcher proved that you can detect the spiritual ascent of a soul just by looking at the mathematical heartbeat of the letters.

Technical Summary: From Graphemic Dependence to Lexical Structure

1. Problem Statement

Traditional computational literary analysis (stylometry) often focuses on high-level features like word frequencies, syntax, or semantic embeddings to classify genres or authors. However, there is a lack of research exploring how the most fundamental, low-level symbolic structures—specifically the sequence of vowels and consonants—reflect the large-scale organizational and stylistic shifts in a monumental poetic work. The study seeks to determine if a minimalist graphemic representation can capture the structural evolution of Dante’s Divine Comedy across its three cantiche (Inferno, Purgatorio, and Paradiso).

2. Methodology

The author employs a multi-layered statistical approach, moving from symbolic encoding to probabilistic modeling and lexical anchoring:

Symbolic Encoding (V/C): The text is stripped of semantic content and encoded as a binary sequence of Vowels (V) and Consonants (C).
Custom Tokenization: To handle the complexities of Dante’s volgare (vernacular), a rule-based tokenizer was developed to classify apostrophe-bearing forms (e.g., apheresis, apocope, clitic elision, and crasis). This ensures that orthographic "compressions" are treated as distinct structural events rather than noise.
Markovian Modeling:
- Two-state model: Captures global serial dependence.
- Four-state model: Based on overlapping symbol pairs (VV, VC, CV, CC), this captures dependencies at the trigram level, providing a more granular view of local "memory."
- Memory Depth (MD): A statistical index derived from the dispersion coefficient ( $C_F$ ) to quantify the degree of local dependency (persistence vs. alternation).
Graphemic Probes: Trigram configurations (e.g., CCC, VCV) were analyzed to see if their frequency profiles correlate with the poem's progression.
Lexical Anchoring & Classification:
- Supervised Learning: Two classifiers—sPLS-DA (Sparse Partial Least Squares Discriminant Analysis) and EN-MNLR (Multinomial Logistic Regression with Elastic-Net)—were used to identify terms that characterize each cantica.
- Integration: The study links the statistical "probes" (trigrams) to "anchors" (high-ranking lexical terms) to see if the mathematical patterns correspond to identifiable words.

3. Key Contributions

A Minimalist Framework: Demonstrates that complex literary structures can be detected using extremely reduced symbolic representations (V/C) without needing deep semantic or syntactic parsing.
The "Probe-Anchor" Bridge: Introduces a method to link low-level stochastic dynamics (Markov chains) to high-level linguistic features (lexical distribution).
Orthographic-Structural Link: Establishes a formal connection between orthographic phenomena (apostrophe usage/token length) and the mathematical "memory" of the text.

4. Results

Directional Shift in Dependency: There is a consistent increase in Memory Depth (MD) from the Inferno to the Paradiso. This indicates a shift in the local dependency structure of the poem.
Orthographic "Distension": The Inferno is characterized by high apostrophe usage and shorter tokens (compressed forms), whereas the Paradiso shows a "smoothing" effect with longer tokens and fewer contractions.
Trigram Behavior:
- Intra-lexical probes: Patterns like str or nte are mostly confined within single words.
- Boundary-driven probes: Patterns with multiple transitions (e.g., VCV or CVC) frequently occur across word boundaries, reflecting the interaction between adjacent tokens.
Successful Cantica Separation: The classification models achieved high accuracy ( $\approx 89\%$ ), successfully distinguishing the three parts. Crucially, the Purgatorio was identified as a transitional region, showing a monotonic progression in both lexical and graphemic space toward the Paradiso.
Semantic Alignment: The lexical anchors identified by the models align with classical literary interpretations: the Inferno focuses on corporeality (teeth, shoulders), the Purgatorio on movement (steps, mountain), and the Paradiso on transcendence (light, grace, stars).

5. Significance

This research proves that local symbolic dynamics are not random; they are deeply intertwined with the large-scale narrative and stylistic architecture of a text. By showing that a simple Markovian model can "track" the stylistic evolution of Dante’s work, the paper provides a new, interpretable framework for computational philology. It suggests that the "texture" of a poem (its vowel-consonant rhythm) is a measurable proxy for its thematic and formal progression, bridging the gap between quantitative mathematics and qualitative literary criticism.