Poetry has always invited us to notice patterns. We hear the recurrence of a sound, feel the movement of a rhythm, recognize an image returning in a different form, or sense that one poem is somehow calmer, darker, more forceful, or more immediate than another. Sometimes those impressions are easy to explain. At other times, they remain just beyond the reach of precise description. We know that two poems feel different, even when they concern similar subjects, but we may struggle to identify exactly where that difference lives. Is it in the vocabulary? The rhythm? The emotional associations of the words? The balance between concrete images and abstract ideas? The familiarity or rarity of the language?
VerseVAD began with a curiosity about those questions. I wanted a way to examine poetic language from another angle—not as a replacement for reading, interpretation, or literary scholarship, but as a companion to them. What might we discover if we could measure certain features of a poem, compare them across texts, and then return to the poem with a new set of questions? What patterns might become visible across a collection, a poet’s career, or a larger literary corpus? Could computational analysis help us articulate qualities that readers often perceive intuitively?
Those questions eventually became VerseVAD, an open-source application for the computational analysis of poetry. It brings together several established linguistic datasets and analytical methods in one workspace, allowing users to explore the emotional, lexical, stylistic, and prosodic features of individual poems and larger collections. The goal is not to reduce poetry to numbers. It is to use numbers to find patterns, surprises, and points of departure for deeper reading.
That distinction matters. A poem is not the sum of its scores. No numerical result can tell us whether a poem is beautiful, successful, moving, politically important, historically significant, or worthy of attention. No chart can settle the meaning of an image, account completely for irony, or reproduce the private associations that a reader brings to a word. VerseVAD is not designed to issue verdicts about poems. It is designed to help us ask better and more specific questions about how they are made.
Beginning with Emotional Language
The name VerseVAD comes from one of the project’s central analytical frameworks: valence, arousal, and dominance. These three dimensions are commonly used in affective research to describe the emotional associations attached to words.
Valence concerns positivity and negativity. A word such as joy generally carries a more positive valence than a word such as grief. Arousal concerns activation or intensity. Words associated with excitement, violence, panic, or exhilaration tend to have higher arousal, while words associated with stillness, rest, or quiet tend to have lower arousal. Dominance concerns power, control, and agency. Some words convey strength or command, while others suggest vulnerability, submission, or powerlessness.
VerseVAD can examine these dimensions across the language of a poem. That does not mean it determines the poem’s single, definitive emotional state. Instead, it shows the tendencies present in the poem’s vocabulary. A poem about death may contain unexpectedly positive language. A love poem may be emotionally agitated rather than soothing. A poem that appears serene at first glance may rely on words associated with vulnerability or diminished control. These results do not replace interpretation; they create interpretive problems worth investigating.
That is one of the pleasures of computational literary analysis: the most interesting result is not always the one that confirms what we already believe. Sometimes the value lies in the disagreement between our expectation and the data. When a poem produces an unexpected score, the next question is not simply, “Is the software wrong?” It may be, “What is happening in the poem’s language that produces this result?”
Perhaps the poem repeats a seemingly neutral word with a strong lexical rating. Perhaps its subject is bleak, but its imagery is filled with beauty, light, love, music, or memory. Perhaps the emotional effect of the poem depends on irony, negation, juxtaposition, or narrative development—features that a word-level lexicon cannot fully capture. Each possibility sends us back to the text.
VerseVAD also includes analyses based on named emotions, making it possible to examine the presence and intensity of language associated with categories such as joy, sadness, anger, fear, trust, anticipation, surprise, and disgust. Once again, these are not declarations about what the poem “really feels.” They are ways of mapping the emotional associations carried by its words.
This can be useful at the level of a single poem, but it becomes especially interesting across a larger body of work. Does a poet’s emotional vocabulary change from one collection to another? Are certain emotions unusually prominent in one period? Does a poem that seems atypical within a collection also appear distant from that collection’s broader emotional profile? These are questions that readers and scholars can explore through a combination of computational evidence and close reading.
Concrete Images and Abstract Ideas
Poetry often moves between the tangible and the conceptual. A poem may give us a wheelbarrow, a plum, a bird, a stone, a doorway, or a field. It may also give us love, memory, eternity, justice, grief, or despair. One of VerseVAD’s analytical areas examines this spectrum through lexical concreteness.
Concreteness ratings estimate how readily a word refers to something that can be experienced through the senses. Words such as apple, rain, or hand are relatively concrete. Words such as truth, destiny, or virtue are more abstract. Most poems contain some combination of the two, and the balance between them can tell us something about how the poem builds its world.
A high concreteness score does not make a poem better, clearer, or more vivid by definition. Nor does abstraction represent a flaw. Some poems depend upon precise physical images; others work through philosophical language, emotional states, symbolic ideas, or spiritual concepts. The usefulness of the measurement lies in comparison. We can ask how one poem differs from another, whether a poet becomes more or less concrete over time, or how imagery functions within a collection.
The distribution can be more revealing than the average alone. Two poems might have similar mean concreteness scores while arriving there in very different ways. One may use mostly moderately concrete language, while another alternates sharply between highly physical images and deeply abstract concepts. VerseVAD therefore examines not only central tendencies but also variation and distribution, helping users see the internal shape of the vocabulary rather than relying on a single number.
That pattern can lead directly back into interpretation. If a poem repeatedly moves from concrete objects into abstract reflection, where do those transitions occur? Does the physical world anchor the poem’s ideas, or does the poem gradually leave that world behind? Are abstract claims made persuasive through sensory detail? Does a sudden concrete image interrupt an otherwise philosophical passage? The computational result identifies the pattern; the poem explains its significance.
Familiar Words, Rare Words, and the Question of Difficulty
Another area of analysis concerns word frequency. Some poems are built largely from highly familiar, frequently encountered words. Others use rare, archaic, technical, regional, or highly specialized vocabulary. VerseVAD uses frequency data to help users examine how common or uncommon a poem’s language is.
This can contribute to discussions of accessibility and difficulty, but the relationship is not simple. A poem written with common words can still be extraordinarily difficult because of its syntax, imagery, allusions, fragmentation, ambiguity, or conceptual demands. Likewise, a poem containing unusual vocabulary may remain emotionally or narratively direct. Lexical rarity is only one form of difficulty.
Still, it is a meaningful feature. Comparing the vocabulary of William Carlos Williams with that of Gerard Manley Hopkins, for example, reveals something important even before we begin discussing their syntax, rhythm, or imagery. One may rely more heavily on familiar everyday language, while the other may present compounds, archaic forms, unusual constructions, or less frequent words. Frequency analysis gives us a way to describe that difference with greater precision.
VerseVAD can also examine the estimated age at which words are typically acquired. Age-of-acquisition data offers another perspective on linguistic familiarity. Some words are generally learned very early in life, while others tend to enter a person’s vocabulary later. A poem dominated by early-acquired words may create a surface of linguistic simplicity, even if its meaning is complex. Another poem may depend upon vocabulary that readers are less likely to encounter until adolescence or adulthood.
Again, these results should not be mistaken for judgments about who can or cannot understand a poem. Readers have different educational, cultural, linguistic, and personal backgrounds. A word that is unfamiliar to one reader may be ordinary to another. Age-of-acquisition data describes broad tendencies, not individual lives. Its value lies in helping us characterize the vocabulary a poem uses and compare that vocabulary across texts.
Together, frequency and age of acquisition allow us to ask more nuanced questions about poetic accessibility. Is the poem difficult because its words are rare, or despite the fact that its words are common? Does it create complexity from simple language? Does it use childlike vocabulary to address adult grief, desire, violence, or mortality? Does a poet’s diction become more specialized across successive collections? These questions are much richer than simply assigning a poem a label such as “easy” or “hard.”
Readability—and Why Poetry Resists It
VerseVAD also includes readability analysis, although poetry presents particular challenges for traditional readability formulas. Most readability measures were designed with prose in mind. They often rely on sentence length, word length, syllable count, and related features. Poems complicate nearly all of those assumptions.
What counts as a sentence when punctuation stretches across several lines? How should a formula interpret fragments, refrains, unconventional capitalization, or deliberate grammatical disruption? Is a long sentence necessarily difficult when its syntax is clear and musical? Is a short line necessarily simple? What happens when a poem uses familiar one-syllable words in a highly ambiguous structure?
Different readability formulas can produce very different results for the same poem because they define and weight difficulty differently. That disagreement is not merely a technical inconvenience. It can itself become informative. Rather than treating a readability score as an objective grade level stamped onto a poem, VerseVAD presents readability as one piece of evidence among many.
A low reading-ease score may point toward longer words or more complex sentence structures. It may also reflect formatting choices or poetic syntax that the formula was never designed to interpret. A comparison between poems can still be useful, particularly when the same method is applied consistently, but the results require context.
Poetry reminds us that readability and interpretive difficulty are not identical. A poem can be easy to pronounce and hard to understand. It can be syntactically simple and emotionally devastating. It can use familiar words while withholding narrative clarity. It can also use elaborate diction to express a relatively direct idea. VerseVAD does not resolve these tensions, but it can help make them visible.
Listening to the Poem
Poetry is not only a form of language on the page. It is also patterned sound. For that reason, VerseVAD includes tools for examining prosody: syllables, stress, meter, rhyme, and rhythmic regularity.
Prosodic analysis is one of the most technically difficult areas of computational poetry study because pronunciation is never entirely fixed. A dictionary can provide likely pronunciations, but poets and readers may stress words differently depending on context, dialect, historical period, performance, or metrical necessity. A word may have multiple accepted pronunciations. A poem may deliberately resist its apparent meter. A line may begin in a regular pattern and then disrupt it for emphasis.
VerseVAD approaches this uncertainty by treating pronunciation data as evidence rather than absolute law. It can identify possible syllable counts and stress patterns, surface alternate pronunciations, estimate likely metrical tendencies, and examine rhyme relationships. Where pronunciation dictionaries are incomplete or inappropriate, users can review coverage and apply scholarly overrides.
The point is not to pretend that a machine can hear a poem exactly as a human reader does. It cannot. The point is to give readers a structured way to inspect the sound patterns that contribute to poetic form. A stress sequence may reveal a strong underlying rhythm. A deviation may direct attention toward a particular word. A rhyme analysis may show that a poem uses full rhyme sparingly but relies heavily on slant rhyme or internal echo. These patterns can then be tested by reading the poem aloud.
Prosodic data is especially valuable when paired with other forms of analysis. Does a rise in emotionally intense language coincide with greater rhythmic irregularity? Do calmer sections use longer lines or more predictable stresses? Does a poet’s rhyme practice change across collections? Are certain emotional or semantic categories associated with particular sonic patterns? VerseVAD is built to encourage these kinds of connections.
From Individual Poems to Larger Patterns
A single poem can offer a rich field of analysis, but VerseVAD is also designed for corpus-level exploration. A corpus is simply a collection of texts gathered for study. It might include every poem in one book, several books by the same poet, a group of poems from a particular period, or works selected for comparison.
Corpus analysis allows us to move between the individual and the collective. We can examine the average profile of a collection and then ask which poems sit closest to or farthest from that center. We can identify works that appear highly characteristic of a poet’s broader vocabulary and those that seem unusual. We can compare collections, trace changes over time, or investigate differences between literary groups.
VerseVAD’s visual and statistical tools are intended to make these relationships easier to explore. A poem can be plotted in relation to a reference corpus, allowing users to examine its distance from the corpus centroid—the central point representing the corpus’s average profile across selected features. This does not tell us whether the poem is important or representative in a traditional literary-historical sense. It tells us how similar or dissimilar its measurable features are to those of the reference collection.
That distinction opens up productive possibilities. A famous poem may turn out to be linguistically atypical of the poet who wrote it. A relatively neglected poem may sit very near the center of a collection’s emotional and lexical profile. A poet’s early and late work may occupy noticeably different regions. A poem’s unusual position may reflect its subject, form, vocabulary, historical circumstances, or place within a sequence.
These findings are beginnings, not conclusions. Once a poem appears as an outlier, the meaningful question becomes why. What does it do differently? Is the difference visible in its diction, form, imagery, emotional tone, or subject matter? Does its distance support an existing scholarly argument, complicate one, or suggest a new avenue of research?
A Tool for More Than One Kind of Reader
VerseVAD is being developed with several overlapping audiences in mind. A literary scholar might use it to identify patterns across a large body of work or generate questions for a research project. A student might use it to compare two poems and support a close-reading assignment. A teacher might use it to demonstrate that terms such as “tone,” “difficulty,” or “imagery” can be broken into multiple observable features. A poet might use it to inspect patterns in a manuscript or compare different drafts. A curious reader might simply enjoy discovering something unexpected about a favorite poem.
No advanced background in statistics or computational linguistics should be required to begin exploring. At the same time, the methods and source data should remain visible enough for serious research. VerseVAD is intended to balance accessibility with transparency. Users should be able to see not only a result but also how it was produced, which words were matched, where dictionary coverage was incomplete, and how different analytical settings affect the outcome.
That transparency is especially important because computational tools can create an illusion of certainty. A number displayed to several decimal places may look authoritative even when it depends on incomplete data, methodological choices, or assumptions that deserve scrutiny. VerseVAD therefore emphasizes match rates, configurable options, token- and type-weighted results, stopword controls, pronunciation alternatives, and other forms of methodological context.
The goal is not to remove human judgment from analysis. It is to make that judgment more informed.
Returning to the Poem
The larger philosophy behind VerseVAD is simple: computation should return us to the poem.
A chart may show that one poem has unusually high arousal. The next step is to ask which words contribute to that pattern, where they occur, and what they are doing. A frequency analysis may reveal that a poem uses overwhelmingly common language. We can then examine how the poet creates complexity without lexical rarity. A concreteness result may show a movement from physical imagery toward abstraction. We can return to the poem and locate that movement line by line. A metrical analysis may identify a disruption in an otherwise regular pattern. Reading aloud may reveal whether the disruption creates hesitation, emphasis, violence, surprise, or emotional strain.
In each case, the numerical result matters because the poem matters. The data does not close interpretation. It reopens it.
This approach also acknowledges something important about literary reading: interpretation often begins in observation. Before making an argument, we notice recurrence, contrast, emphasis, absence, structure, and change. Computational methods can expand our powers of observation, especially across bodies of work too large to hold fully in memory. They can count what we cannot easily count, compare what we cannot easily compare, and reveal patterns that may otherwise remain hidden.
But patterns do not interpret themselves. They become meaningful through historical knowledge, formal analysis, theoretical perspective, and attentive reading. The most exciting work happens at the meeting point between those forms of understanding.
VerseVAD is an experiment in building that meeting place. It is a tool for exploring how poems feel, sound, move, and construct their worlds through language. It is also an evolving project, one that will continue to develop through testing, research, user feedback, and collaboration.
Most of all, it grows from a conviction that poetry can withstand examination from many directions. Measuring one feature of a poem does not diminish its mystery. Often, it makes the mystery more precise. It allows us to see that our initial impression was only one part of the poem’s design—and gives us a reason to read it again.