Physician poetry research

Digital Humanities

Digital methods that help make the patterns of physician writing more available for study, beginning with the complete contemporary corpus of 421 poems by Jenelle Jindal, MD.

Methodology

Visualizing literary corpus through vector embedding and dimensionality reduction. A study in semantic proximity.

Loading visualization data...

Scroll to zoom, drag to pan. Click nodes to read poems.

Displaying 421 Poems

Why digital methods belong here

Some questions become clearer when we can look across an entire corpus rather than one poem at a time. Digital methods help reveal patterns of form, attention, relation, and classification that can then return us to closer reading. One of the field's central ideas, "distant reading," coined by Franco Moretti at Stanford, holds that patterns of theme, language, and form emerge across a body of work that no single reading can reveal.

These visualizations place that idea in practice. Each poem is encoded as a vector embedding that captures its semantic meaning. Poems with similar imagery, tone, or subject sit closer together — and the collection becomes a navigable landscape.

Computation does not replace interpretation here. It helps surface questions, relationships, and structures that deserve further thought. For physicians: this is the same class of technique behind clinical NLP and large language models, turned toward a different question. Not diagnosis, but interpretation. Not patient records, but poems informed by a life in medicine.

Click any point to read the poem. The map invites distant reading. The poem returns you to close reading. Both are valuable perspectives.

Methodology

Visualization Techniques

The goal is not to flatten poems into data. The goal is to make literary patterns more available for research while preserving the complexity that poetry keeps intact. Every chart below analyzes the same dataset: 421 published poems by a single physician, each applying a different computational technique.

Browse the corpus →

Loading visualization data...

Technique

The Parallel Coordinates Plot

Parallel coordinates map each item as a polyline crossing multiple vertical axes — one per feature. Invented by Alfred Inselberg in the 1980s, the technique displays eight or more dimensions simultaneously, revealing correlations no lower-dimensional projection can capture.

For poetry, each polyline becomes a structural signature: word count, line length, stanza count, and semantic coordinates together trace the shape of a poem’s form and meaning at a glance.

Key Citations

Inselberg, A. (1985). “The Plane with Parallel Coordinates.” The Visual Computer, 1(4), 69–91. Also: Inselberg, A. (2009). Parallel Coordinates: Visual Multidimensional Geometry and Its Applications. Springer. Springer
Drucker, J. (2014). Graphesis: Visual Forms of Knowledge Production. Harvard University Press. Harvard UP
Piper, A. (2018). Enumerations: Data and Literary Study. University of Chicago Press. Chicago UP
Jockers, M. L. (2013). Macroanalysis: Digital Methods and Literary History. University of Illinois Press. Illinois UP
Underwood, T. (2019). Distant Horizons: Digital Evidence and Literary Change. University of Chicago Press. Chicago UP
Moretti, F. (2005). Graphs, Maps, Trees: Abstract Models for a Literary History. Verso. Verso Books

Loading visualization data...

Click and drag on any axis to brush-select a range. Click an axis to clear. Hover lines to identify poems.

Each line traces one poem across eight dimensions of form and meaning

Technique

The Word Tree

The word tree is a computational concordance: a visualization that displays every context in which a chosen word appears across a corpus, arranged as a branching tree whose limbs represent the different phrases that follow. The technique bridges two traditions central to literary scholarship. Close reading attends to the texture of individual passages; distant reading, as theorized by Franco Moretti, reveals patterns visible only at the scale of an entire body of work. A word tree holds both scales in view at once, letting a reader see every instance of a word simultaneously while preserving the surrounding language that gives each instance its meaning.

In this sense it is the digital heir to the printed concordance, a tool that has accompanied serious readers of the Bible, Shakespeare, and other canonical texts for centuries. Its computational form, introduced by Wattenberg and Viégas in 2008, makes the concordance interactive and visual, turning a static index into an instrument of exploration.

Key Citations

Wattenberg, Martin and Fernanda Viégas (2008). “The Word Tree, an Interactive Visual Concordance.” IEEE Transactions on Visualization and Computer Graphics 14(6): 1221–1228. doi:10.1109/TVCG.2008.172
Sinclair, Stéfan and Geoffrey Rockwell. Voyant Tools. Web-based text analysis platform, 2003–present. voyant-tools.org
Rockwell, Geoffrey and Stéfan Sinclair (2016). Hermeneutica: Computer-Assisted Interpretation in the Humanities. Cambridge, MA: MIT Press. MIT Press
Moretti, Franco (2005). Graphs, Maps, Trees: Abstract Models for a Literary History. London: Verso. Verso Books
Drucker, Johanna (2014). Graphesis: Visual Forms of Knowledge Production. Cambridge, MA: Harvard University Press. Harvard UP
Haider, Thomas (2024). “Understanding Poetry Using Natural Language Processing Tools: A Survey.” Digital Scholarship in the Humanities 39(2): 500–532. Oxford Academic

Loading visualization data...

Type a word or click a suggestion. Click any branch to see the poem context.

Each branch traces how a single word ramifies into different contexts, making visible the patterns of thought and expression that run through an entire body of poetic work.

Technique

The Stream Graph

A stream graph is a stacked area chart whose colored layers flow outward from a shared baseline, evoking tributaries that swell and narrow over time. Each stream represents one thematic tag; its width at any point reflects how many poems carry that tag within the time window.

Because most poems were written in 2019, the timeline uses equal-count bins rather than calendar intervals, revealing how thematic emphasis shifted across the body of work regardless of publication tempo.

Key Citations

Havre, S., Hetzler, E., & Nowell, L. (2000). “ThemeRiver: Visualizing Thematic Changes in Large Document Collections.” IEEE InfoVis 2000. doi:10.1109/INFVIS.2000.885098
Byron, L. & Wattenberg, M. (2008). “Stacked Graphs — Geometry & Aesthetics.” IEEE TVCG, 14(6), 1245–1252. doi:10.1109/TVCG.2008.166
Jockers, M. L. & Mimno, D. (2013). “Significant Themes in 19th-Century Literature.” Poetics, 41(6), 750–769. doi:10.1016/j.poetic.2013.08.005
Moretti, F. (2005). Graphs, Maps, Trees: Abstract Models for a Literary History. Verso. Verso Books
Piper, A. (2018). Enumerations: Data and Literary Study. University of Chicago Press. Chicago UP

Loading visualization data...

Hover over any time bin to see the tag breakdown. Width encodes poem count.

Five thematic currents traced across the full chronology of a physician-poet’s work

Technique

The Chord Diagram

The chord diagram maps relationships between categories as ribbons connecting arcs around a ring. Originally developed for comparative genomics under the name Circos, it reveals co-occurrence, flow, and mutual dependency within small sets of categories.

Here, ribbons connect themes that appear together in the same poem. A wider ribbon means more poems share that pair of themes. The arcs show each theme’s total presence across the corpus.

Key Citations

Krzywinski, M. et al. (2009). “Circos: An Information Aesthetic for Comparative Genomics.” Genome Research, 19(9), 1639–1645. circos.ca
Moretti, F. (2011). “Network Theory, Plot Analysis.” New Left Review, 68, 80–102. New Left Review
D’Ignazio, C. & Klein, L. F. (2020). Data Feminism. MIT Press. Open access
Drucker, J. (2014). Graphesis: Visual Forms of Knowledge Production. Harvard University Press. Harvard UP

Loading visualization data...

Hover over any arc to see its connections. Click ribbons to isolate a pair.

Each ribbon traces the poems where two themes meet, its width the measure of their shared presence

Technique

Semantic Zoom

Semantic zoom changes representation qualitatively — not merely in magnification — as the viewer moves between scales. At the broadest level, labeled regions and density contours; at intermediate zoom, individual data points; at the closest, textual detail on demand.

This operationalizes Shneiderman’s visualization mantra: “overview first, zoom and filter, then details on demand.”

Key Citations

Shneiderman, B. (1996). “The Eyes Have It: A Task by Data Type Taxonomy for Information Visualizations.” IEEE Symposium on Visual Languages, 336–343. doi:10.1109/VL.1996.545307
McInnes, L. et al. (2018). “UMAP: Uniform Manifold Approximation and Projection.” JOSS 3(29), 861. arXiv:1802.03426
Perlin, K. & Fox, D. (1993). “Pad: An Alternative Approach to the Computer Interface.” ACM SIGGRAPH ’93, 57–64. doi:10.1145/166117.166125
Smilkov, D. et al. (2016). “Embedding Projector.” NIPS 2016 Workshop. arXiv:1611.05469

Loading visualization data...

Scroll to zoom, drag to pan. At close zoom, click any poem to read it.

From overview to detail: the same embedding space at three levels of resolution

Technique

Similarity Heatmap & Dendrogram

The similarity matrix encodes the complete set of distances between every pair of poems, preserving relationships that any 2D or 3D projection must distort. Reordering rows and columns by hierarchical clustering reveals block-diagonal structure: dense squares along the diagonal mark groups of thematically kindred poems.

The dendrogram — a tree built by agglomerative clustering — shows the same structure as a branching genealogy of stylistic and thematic kinship, exposing sub-groups the manual tag system may not capture.

Key Citations

Burrows, J. (2002). “‘Delta’: A Measure of Stylistic Difference.” Literary and Linguistic Computing, 17(3), 267–287. doi:10.1093/llc/17.3.267
Eder, M. (2017). “Visualization in Stylometry.” Digital Scholarship in the Humanities, 32(1), 50–64. doi:10.1093/llc/fqv061
Karsdorp, F., Kestemont, M. & Riddell, A. (2021). Humanities Data Analysis. Princeton University Press. Companion site
Jockers, M. L. (2013). Macroanalysis: Digital Methods and Literary History. University of Illinois Press. Illinois UP

Loading visualization data...

Toggle between heatmap and dendrogram views. Hover cells or click branches to explore.

The unmediated ground truth of inter-poem similarity, ordered by hierarchical clustering

Technique

Topic Evolution

Topic evolution traces how computationally discovered themes wax and wane across a body of work, making legible the thematic arc of a literary career. The cumulative view shows absolute growth; the proportional view reveals how the mix shifts.

The technique draws on topic modeling and the distant-reading paradigm, which argues that patterns of literary change become visible only at scales beyond any individual reading.

Key Citations

Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). “Latent Dirichlet Allocation.” JMLR, 3, 993–1022. jmlr.org
Underwood, T. (2019). Distant Horizons: Digital Evidence and Literary Change. University of Chicago Press. Chicago UP
Grootendorst, M. (2022). “BERTopic: Neural Topic Modeling.” arXiv:2203.05794. arxiv.org
Drucker, J. (2014). Graphesis: Visual Forms of Knowledge Production. Harvard University Press. Harvard UP

Loading visualization data...

Toggle between cumulative and proportional views. Hover to see tag breakdown at any point.

Five thematic threads traced through chronological time, from first poem to last

Technique

Unsupervised Clustering

k-means clustering partitions the corpus into groups based on embedding proximity alone, without reference to the human-assigned tag system. The algorithm tests multiple values of k and selects the one that maximizes silhouette score — a measure of how cohesive each cluster is relative to its nearest neighbor.

A language model then reads the most representative poems in each cluster and proposes a thematic label. The resulting groupings often cross-cut the five-tag taxonomy, revealing sub-themes and affinities the original categorization does not capture.

Key Citations

Lloyd, S. (1982). “Least Squares Quantization in PCM.” IEEE Transactions on Information Theory, 28(2), 129–137. doi:10.1109/TIT.1982.1056489
Arthur, D. & Vassilvitskii, S. (2007). “k-means++: The Advantages of Careful Seeding.” Proceedings of SODA 2007, 1027–1035. ACM DL
Rousseeuw, P. J. (1987). “Silhouettes: A Graphical Aid to the Interpretation and Validation of Cluster Analysis.” Journal of Computational and Applied Mathematics, 20, 53–65. doi:10.1016/0377-0427(87)90125-7
Grootendorst, M. (2022). “BERTopic: Neural Topic Modeling with a Class-based TF-IDF Procedure.” arXiv:2203.05794. arXiv

Loading visualization data...

Toggle between cluster and tag coloring. Click cluster labels to explore. Scroll to zoom, drag to pan.

Natural groupings discovered by k-means in 3,072-dimensional embedding space, often crossing the boundaries of the five-tag taxonomy

Interactive

Thematic Corridors

Type a phrase — a feeling, a clinical scenario, a line of verse — and watch the corpus respond. Poems semantically close to your query light up on the map; everything else fades. A labeled dot marks where your text would sit among the poems.

Technique

Semantic Query Projection

Your query is embedded in real time via the Gemini embedding model, then compared against every poem in the Cloudflare Vectorize index. The top 20 nearest neighbors are highlighted on the UMAP projection, with opacity proportional to similarity score.

The “your search” dot is positioned at the score-weighted centroid of the results — an approximation of where your text would appear if we ran full UMAP dimensionality reduction on the augmented corpus.

Key Citations

Mikolov, T. et al. (2013). “Efficient Estimation of Word Representations in Vector Space.” arXiv:1301.3781
McInnes, L. et al. (2018). “UMAP: Uniform Manifold Approximation and Projection.” JOSS 3(29), 861. arXiv:1802.03426
Lee, J. et al. (2024). “Gemini Embedding Models.” Google DeepMind. Google AI

Loading visualization data...

Try: “grief in the exam room”, “watching a child sleep”, “the weight of a diagnosis”

Real-time semantic search projects your query into the embedding space, illuminating the poems closest in meaning

Interactive

Poem Journeys

Pick a starting poem and a destination. The system traces a path through the corpus, hopping from neighbor to neighbor, each step bridging the thematic distance. A literary shortest-path search through embedding space.

Technique

Dijkstra Traversal

Each poem knows its 15 nearest semantic neighbors (precomputed via cosine similarity over 3,072-dimensional Gemini embeddings). The pathfinder uses Dijkstra’s algorithm: at each hop, it expands the current poem’s neighbors and scores candidates by accumulated cost — the sum of (1 − similarity) along the path — guaranteeing the optimal shortest path through the similarity graph.

A visited set prevents cycles. Maximum path length: 10 hops. The paths surface poems that exist in the thematic spaces between obvious clusters — connective tissue no keyword search would find.

Key Citations

Dijkstra, E. W. (1959). “A Note on Two Problems in Connexion with Graphs.” Numerische Mathematik, 1(1), 269–271. doi:10.1007/BF01386390
Piper, A. (2018). Enumerations: Data and Literary Study. University of Chicago Press. Chicago UP

Loading visualization data...

Try journeys between distant poems: a clinical piece and a nature lyric, or a meditation and a narrative.

Each hop follows the strongest similarity link between neighboring poems, tracing the connective tissue of the corpus

Interactive

3D Semantic Cosmos

The full corpus rendered as a three-dimensional point cloud. Orbit, rotate, and zoom to explore the semantic landscape from any angle — relationships flattened in 2D become visible when the third dimension is restored.

Loading visualization data...

Drag to orbit, scroll to zoom, hover for poem titles. Click any point to read the poem.

The full corpus in three dimensions — preserving semantic relationships that 2D projections must compress

Glossary

Technical Definitions & Methodology

Vector Embedding

A mathematical representation of a poem's meaning, encoded as a sequence of numbers. Poems with similar themes, imagery, or language produce similar vectors.

Dimensionality Reduction

A technique for projecting high-dimensional data (thousands of numerical values per poem) into two or three dimensions so it can be visualized on screen.

Distant Reading

A method of literary analysis, developed by Franco Moretti at Stanford, that studies patterns across large collections of texts rather than focusing on individual works.

Close Reading

The traditional practice of careful, detailed attention to the language, structure, and meaning of a single text.

Semantic Proximity

The degree to which two texts share meaning. In the visualizations, poems that are semantically proximate appear close together.

Latent Space

The abstract mathematical space in which vector embeddings exist. Visualizations are projections from this high-dimensional latent space into visible dimensions.

Computational Criticism

The use of computational tools — algorithms, statistics, machine learning — to analyze and interpret literary works.

NLP (Natural Language Processing)

The branch of artificial intelligence concerned with enabling computers to understand, interpret, and generate human language.

Corpus

A structured collection of texts assembled for analysis. Here, the corpus consists of the poems published on poetdoctor.com.

Cluster

A group of poems that the algorithm identifies as thematically or linguistically related, appearing as a neighborhood in the visualization.

The explanatory text on this website was written with AI assistance.
All poems on this site are written by hand, without AI.