Any attempt to analyze poetry with numbers is going to make some readers uneasy. That reaction is understandable.
Poetry is written by people. It emerges from memory, imagination, history, desire, grief, anger, craft, instinct, revision, accident, and experience. It is read by people too, each bringing a different life to the page. No software can reproduce that exchange. No score can fully explain why a poem stays with someone for years, why one line suddenly feels unbearable, or why a familiar poem changes after a loss, a move, a marriage, or the passing of time.
VerseVAD was not created to compete with that human experience. It was created because that experience is worth examining more closely.
The purpose of VerseVAD is not to replace reading, interpretation, scholarship, or conversation. It is not designed to decide what a poem means, whether it is good, or how a reader ought to feel about it. Its purpose is more modest and, I hope, more useful: to help readers notice patterns, test impressions, challenge assumptions, and return to poems with better questions.
What VerseVAD Measures
VerseVAD examines selected features of poetic language. Depending on the analysis, it can look at emotional associations, concreteness, word familiarity, estimated age of acquisition, reading ease, sound, stress, rhyme, line structure, and other measurable characteristics.
These features are real parts of poems. A poet chooses words that carry emotional associations. Some poems rely heavily on physical objects, while others move through abstract ideas. Some use familiar everyday language; others depend on rare, archaic, technical, or invented vocabulary. Some create steady rhythmic patterns, while others resist regularity.
VerseVAD helps describe those features in a consistent way. For example, a reader may feel that one poem is more agitated than another. VerseVAD can examine whether the vocabulary also tends toward higher emotional activation. A poem may seem unusually image-driven, and concreteness data can help show whether its language is in fact more physical and sensory than that of a comparison poem. A reader may find a poem difficult and discover that its words are actually common and early-acquired, suggesting that the difficulty comes from syntax, ambiguity, form, or allusion rather than vocabulary.
That is the kind of work VerseVAD does best. It takes a broad impression and breaks it into more specific questions.
What the Scores Actually Mean
A VerseVAD score is not a verdict. It is a summary of one measurable feature under a particular set of analytical conditions. A valence score describes the positive or negative associations of matched words. It does not tell us whether the poem as a whole is happy or sad. An arousal score describes how calm or activated the vocabulary tends to be. It does not measure the full intensity of the reading experience. A concreteness score describes the balance between more physical and more abstract language. It does not determine whether the poem is vivid, clear, or artistically successful.
The same principle applies throughout the software. A Poetry Reading Ease score estimates linguistic accessibility. It does not measure intellectual depth. A frequency score describes how common the vocabulary is. It does not tell us whether the poem is easy to interpret. A metrical analysis identifies possible stress patterns. It does not decide how every reader will perform the poem aloud. The numbers describe parts of the language. They do not contain the whole poem.
This is why the most useful VerseVAD analysis usually begins after the score appears. Which words contributed most strongly? Where do they occur? Is repetition affecting the result? Does the pattern remain consistent across the poem, or does it change from one section to another? Does the result support the reader’s first impression, or complicate it? A score becomes meaningful through investigation.
Where Computational Analysis Falls Short
Poetry is difficult for computational tools for many of the same reasons it is rewarding for human readers. Words change in context. Irony can make a positive statement feel bitter. Negation can reverse the apparent meaning of a word. Metaphor moves meaning between unlike things. Historical usage may differ from contemporary usage. A speaker may say one thing while the poem quietly suggests another.
Most lexical datasets rate words individually. Poems rarely ask us to experience words individually. A lexicon may associate a word with fear, anger, trust, or disgust without understanding how that word functions in a particular line. A pronunciation dictionary may provide a standard stress pattern even when the meter invites a different performance. A readability formula may interpret a long poetic sentence as difficult while overlooking the familiarity of its words and the clarity of its images.
These limitations should not be hidden. They are part of the method. VerseVAD therefore reports match rates, word-level results, alternative weightings, and other information that helps users see how an outcome was produced. When an emotion category appears because of one questionable lexical match, the reader should be able to find that match. When a score depends heavily on a repeated word, that influence should remain visible. Transparency does not eliminate the limitations of computational analysis, but it prevents those limitations from masquerading as certainty.
A Surprising Result Is Not a Final Answer
Some of the most valuable results are the ones that initially look wrong. A dark poem may receive a higher valence score than expected. A seemingly simple poem may use less familiar vocabulary than another poem that feels more difficult. A poem associated with emotional intensity may show relatively low arousal at the word level. The temptation is either to dismiss the score or to treat it as a hidden truth revealed by the software. Neither response is especially useful.
Instead, the disagreement itself becomes the subject of inquiry. Perhaps the poem describes grief through the language of beauty, love, or memory. Perhaps the emotional force comes from rhythm, repetition, or narrative situation rather than explicitly emotional words. Perhaps the poem uses ordinary vocabulary in a syntactically difficult way. Perhaps irony creates distance between the literal language and the reader’s experience.
The result may also expose a limitation in the underlying data. That matters too. A computational tool should be open to correction. Its value does not depend on pretending that every output is meaningful. VerseVAD is not most useful when it confirms everything we already believe. It is useful when it helps us identify exactly where our assumptions and the measurable language diverge.
The Human Author Does Not Disappear
One concern about computational literary analysis is that it may flatten writers into datasets. A poet’s life, historical moment, formal choices, influences, and intentions can seem to vanish behind averages and charts. That is a real danger if the numbers are treated as self-sufficient.
VerseVAD is designed to work in the opposite direction. A pattern across a poet’s work should lead us toward historical and literary questions. If a poet’s emotional vocabulary changes across collections, we might ask what was happening in the poet’s life, politics, artistic development, or surrounding culture. If one poem appears unusually distant from the rest of a collection, we can investigate its form, subject, circumstances, and place within the book.
The data does not replace context. It helps us find places where context may matter most. The author remains a maker of choices, even when those choices are not fully conscious. Computational analysis can reveal recurring habits, contrasts, and changes, but it cannot explain them alone. Explanation still belongs to scholarship, biography, history, theory, and close attention to the poems themselves.
The Human Reader Does Not Disappear Either
Readers are not passive recipients of scores. They remain responsible for deciding what a result means, whether it is persuasive, and how it relates to the poem. Different readers may interpret the same pattern differently. One may see low dominance as vulnerability; another may see restraint. One may read high concreteness as sensory immediacy; another may notice how physical objects are being used symbolically. The data narrows certain questions, but it does not eliminate disagreement. Nor should it.
Literary interpretation is not weakened by multiple informed readings. It depends upon them. VerseVAD can give readers shared evidence to discuss, but it cannot settle every interpretive dispute. In that sense, the software is less like a judge and more like an additional pair of eyes. It notices some things very well, misses others entirely, and occasionally points toward something worth examining more closely.
Returning to Close Reading
The central goal of VerseVAD is to improve close reading, not move away from it. A good analysis should eventually return to individual words, lines, images, sounds, and formal choices. If a poem has unusually high arousal, where does that energy live? If its language is highly concrete, how are those physical objects arranged? If the vocabulary is unusually familiar, how does the poet create complexity from simple words? If a rhythmic pattern breaks, what happens at that moment in the poem?
The numerical result gives us a direction. The poem gives us the answer, or at least the material from which an answer can be argued. This is especially useful when working with larger collections. Human readers are excellent at close attention, but we are not equally good at remembering the distribution of thousands of words across hundreds of poems. Computational tools can help identify broad patterns that would be difficult to see unaided. Once those patterns are visible, the reader can return to selected poems and study them closely. The process moves back and forth: from reading to measurement, from measurement to comparison, and from comparison back to reading.
Another Way of Paying Attention
VerseVAD does not claim that poetry can be reduced to data. It begins from the opposite belief: poems are complex enough to reward many forms of attention. Close reading, historical research, biography, formal analysis, performance, archival study, theory, and computational methods can all illuminate different parts of a poem. None needs to become the only legitimate approach.
Numbers can sometimes create a false impression of authority, so they should be handled carefully. But refusing to measure anything does not necessarily protect poetry’s humanity. Used responsibly, measurement can help us notice the craft, patterns, and changes created by human writers. The aim is not to make poetry less mysterious. It is to make our questions more precise.
VerseVAD cannot tell us why a poem matters to a particular reader. It cannot decide whether a metaphor succeeds, whether a line is beautiful, or whether a poem deserves to last. It cannot replace the experience of reading slowly and being surprised. What it can do is show us something we may not have noticed. It can test a claim, complicate an impression, reveal a contrast, or point us toward a poem that deserves another look.
Then it steps aside, and the reading continues.