Digital humanities · Software engineering

Searching for Poetry in Latin Prose

A research tool developed for the Leiden University Centre for the Arts in Society (LUCAS) to help uncover possible hidden poetry fragments in classical Latin prose.

Purpose: Detect poetic fragments hidden in Latin prose
Analysis engine: LSTM-based AI model
Supported meters: Hexameter, pentameter, iambic, trochaic, and more

Searching for Poetry in Latin Prose tool showing a prose input, meter filters, and detected poetry candidates with scansion results — The analysis workspace combines text submission, candidate review, metrical scansion, filtering, and export in one interface.

Hidden poetry detection

Finding poetic rhythm inside continuous prose

Classical prose may contain embedded or unattributed quotations from poetry. Finding them manually requires researchers to inspect large amounts of text syllable by syllable. This tool scans Latin prose automatically and extracts fragments that resemble poetic meter, directing attention to the most promising candidates.

Users can paste prose directly or upload a text file, making the application suitable for both quick experiments and longer textual datasets. Detected candidates can then be reviewed, filtered, and exported for further academic analysis.

Scansion engine

Transforming text into metrical patterns

At the centre of the system is an LSTM neural network that assigns scansion labels to Latin text fragments. It predicts whether syllables are long, short, or an elision, transforming prose into a sequence that can be compared with known poetic patterns.

Each candidate is presented with its predicted meter, metrical distance, confidence information, and syllable-level markings. Long syllables, short syllables, elisions, and uncertain areas are highlighted directly in the text so Latinists can inspect why a passage was selected rather than treating the model as a black box.

Capabilities

Key features

LSTM-powered scansion

Predict long, short, and uncertain syllable patterns in Latin text fragments.

Hidden poetry detection

Extract prose fragments that may match classical poetic rhythm.

Multiple meter support

Search for hexameter, pentameter, hendecasyllable, iambic, and trochaic forms.

Interactive filters

Refine candidates by meter, confidence, syllables, elisions, metrical distance, and more.

Visual result analysis

Inspect scansion markings and confidence information directly in each candidate fragment.

Research-ready workflow

Paste text or upload a file, analyse candidates, and export the findings for later review.

Meter support

Nine metrical forms in one search

The tool can detect hexameter, pentameter, hendecasyllable, and a range of iambic and trochaic forms. Its modular structure allows more meters to be added in the future.

Core forms

Hexameter
Pentameter
Hendecasyllable

Iambic

Trimeter
Senarius
Septenarius
Octonarius

Trochaic

Septenarius
Octonarius

Candidate filtering

Move from broad detection to focused review

A detailed filtering system helps researchers reduce noise and test different interpretations of the results. Criteria can be combined to isolate stronger candidates or investigate specific metrical characteristics.

Pattern and structure filters

Meter type and maximum metrical distance
Minimum and maximum syllable count
Short and long syllable counts
Number of feet and half-meter options

Candidate quality filters

Minimum and maximum confidence
Minimum and maximum elisions
Repeated vowel-type limits
Monosyllabic ending filter

Technology

An analysis pipeline made accessible on the web

The final product connects an Angular frontend to a Flask backend and PyTorch model, turning a computational scansion pipeline into a research interface for Latinists without programming experience.

PyTorch · Angular · TypeScript · Flask · Python · GitHub Projects · Scrum

Personal contribution: my work focused on the Angular application structure, analysis page, filtering and export controls, and overall frontend management.