Writing guide · 8 min read
Sound and Line: What the Engine Hears in a Poem
Line rhythm, rhyme density, sound device density, and end word strength measure the music of verse. Here is exactly what each one counts, and the places it hears wrong.
By Alyssa Glasco, Founder · Published
Three of the poetry readings measure sound. Line Rhythm counts syllables and words per line and reports how much they vary. Rhyme Density reports the share of lines whose ending rhymes with another line. Sound Device Density reports how often neighboring words share an opening consonant or a vowel.
All three are counts. None of them is a judgment about whether the poem works. Free verse that scores zero on rhyme is not failing at anything.
Line Rhythm
For every line the engine counts syllables and words, then reports the coefficient of variation across the poem: the spread of your line lengths relative to their average. A poem in strict blank verse produces a very low number. A poem in open free verse produces a high one. Both are forms.
Syllables come from the CMU Pronouncing Dictionary, which carries about 134,000 words with their pronunciations. Words it does not carry, which for most poets means invented names and coinages, fall back to a vowel-group estimate. If your poem leans hard on invented vocabulary, treat the syllable counts as approximate.
The reading needs at least two lines. A poem submitted as one unlineated block reports nothing rather than reporting perfect regularity, which is what a single line would otherwise imply.
Rhyme Density, and where it hears wrong
The engine takes the final word of each line, reduces it to a rime key, and counts how many lines share a key. The share of lines that match another line is the density.
The rime key is built from spelling, with a handful of normalizations for common English patterns. That is worth knowing because it produces two specific errors, and you will see both if you look for them.
It counts eye rhymes as rhymes. End two lines on love and moveand the engine records a rhyme, because the letters agree even though the sounds do not. And it misses true rhymes that are spelled differently. End two lines on eight and late and the engine records nothing, though any reader hears the rhyme.
Slant rhyme, internal rhyme, and rhyme across stanza breaks are all outside what this reading measures. If rhyme is central to what you are doing, use the number as a rough count of full end rhyme and trust your ear over it.
Sound Device Density
Within each line, the engine strips function words, then checks whether a content word shares its opening consonant with another content word in the same line (alliteration) or its vowel with another (assonance). The reported figure is the share of content words involved in either.
The comparison is on letters, not phonemes, and it works line by line rather than across the poem. Sound that carries across a line break will not appear here.
What to do with all three
Read them together with the poem in front of you. A high line-rhythm variation and a low rhyme density describe most contemporary free verse, and describing is all they are doing. What matters is whether the numbers match what you thought you were writing. A poem you intended as tightly measured that reports wide variation is worth reading aloud again.
Open a poem in the editor and run Grade this passage. Poetry is graded against its own benchmark set: sentence-based readings like average sentence length and paragraph rhythm are hidden entirely for verse, because line breaks make them meaningless.
End Word Strength
The fourth poetry reading looks at where your lines land. For each line the engine takes the last word and looks it up in the Brysbaert concreteness norms, a table of just under 40,000 words rated from 1 (abstract) to 5 (concrete). Your score is the average across the lines it could rate. The benchmark is a minimum of 2.8.
The reasoning is that a line ending is a position of emphasis, and a concrete word holds that position better than an abstraction. Stone lands. Meaningtends not to.
Two limits are worth knowing. The metric needs real line breaks: a poem pasted as a single paragraph is one line, so the engine measures one word and reports it as your whole score. And the averaging only covers words the table contains, so a poem ending lines on invented names, archaisms, or proper nouns is measured on whatever remains.
Concreteness is not strength. An abstract closing word can be exactly right, and a poem that ends every line on a hard noun can read as a list. The number tells you which register you are working in, not whether the choice was correct.