Explorer

Function-Word Fingerprint

The stylometric signal that distinguishes a writer from his peers. Computed against Shakespeare.

In the classic Mosteller and Wallace analysis of the disputed Federalist Papers, the signal that distinguished a Hamilton paper from a Madison paper wasn’t topic or vocabulary or argument structure. It was the per-million rates of the most common function words. The, of, and, to, in, a, that, is, was. The tiny grammatical scaffolding of English that a writer reaches for unconsciously, and that differs measurably between any two people.

This view computes the same fingerprint for each of the six Founders against the Shakespeare baseline. Cosine similarity over a 36-dimensional vector of per-million rates measures how closely each Founder’s scaffolding matches Shakespeare’s.

Cosine similarity to Shakespeare

Closeness to Shakespeare’s function-word pattern

Benjamin Franklin
0.952
John Adams
0.947
Thomas Jefferson
0.907
George Washington
0.905
Alexander Hamilton
0.882
James Madison
0.866

1.000 = identical function-word distribution to Shakespeare. Bar scale runs from 0.85 to 1.00 to make the differences visible.

Per-word rates

Rates per million words, by author

← scroll the heatmap to see all seven authors →

WordShakespeareAdamsFranklinJeffersonWashingtonMadisonHamilton
the30.5k59.5k55.3k64.8k67.4k84.8k79.9k
of20.0k41.9k35.5k45.6k41.8k50.6k52.1k
and29.6k35.4k28.7k20.3k20.4k18.1k19.8k
to21.5k33.0k33.7k39.7k41.0k36.8k41.8k
in12.3k19.9k17.4k17.1k17.8k20.2k19.2k
a16.3k17.4k16.3k15.7k13.8k18.5k19.3k
that12.4k12.5k14.0k12.3k12.4k14.7k13.9k
is10.6k9.8k8.3k8.5k8.7k9.0k10.8k
was2.5k5.3k4.1k3.7k3.0k5.0k3.2k
it8.6k10.6k10.8k13.4k10.5k13.0k12.7k
for9.1k8.0k9.2k10.3k10.5k9.2k9.4k
with8.9k7.5k8.4k7.4k8.5k8.0k7.6k
as6.6k8.8k8.6k9.8k9.8k10.0k9.0k
his7.6k5.7k5.0k4.5k3.5k4.4k3.4k
on3.4k3.5k5.4k8.5k5.7k8.7k5.9k
be7.9k10.2k10.9k13.3k15.1k16.1k16.3k
at2.8k5.3k4.7k4.9k6.2k4.6k4.4k
by4.8k6.0k7.4k7.1k7.0k9.5k7.0k
this7.6k6.5k4.8k6.3k6.7k5.0k7.0k
had1.6k3.6k3.2k3.1k2.3k2.5k1.9k
not9.7k7.0k6.5k6.0k5.9k8.1k6.9k
are4.2k4.6k4.8k3.9k5.7k3.5k4.3k
but7.0k5.6k5.0k4.4k3.7k3.2k3.6k
from2.9k4.8k4.9k5.9k6.5k6.8k5.3k
or3.3k6.4k4.6k4.1k4.5k4.4k4.5k
have6.6k9.0k8.2k8.4k9.8k7.0k7.9k
an2.0k3.5k2.7k2.8k2.9k4.1k4.4k
they2.7k4.6k5.1k4.1k4.1k2.6k3.2k
which2.6k5.1k5.2k7.7k7.0k8.5k8.6k
one2.0k2.5k2.4k2.5k1.9k2.1k2.1k
you15.2k7.0k9.8k10.3k10.3k5.3k6.7k
were1.8k2.2k1.9k1.7k1.6k1.7k1.5k
all4.4k5.3k3.5k2.8k2.5k2.3k2.5k
we3.7k3.6k3.5k3.7k1.8k1.5k1.8k
can1.4k1.7k1.7k1.8k2.0k1.9k1.7k
when2.3k1.4k2.0k1.2k1.4k7641.0k

Darker cells = higher rate per million words. Hover any cell for the exact rate.

Reading the fingerprint

Franklin closest. Madison furthest. And Madison’s the is the surprise.

The cosine ranking comes out where the rest of the project would predict: Franklin in first position, followed by Adams, then the others, with Madison furthest from Shakespeare. The clustering is tight (everyone is over 0.86, none is past 0.96) because all six Founders are writing English and Shakespeare is writing English. But within that tight cluster the ordering is informative, and it matches the composite ranking the project derives from completely different measurements.

The genre signal is unmissable in the heatmap row for the. Shakespeare writes the at about 30,500 occurrences per million words. The Founders all run between 55,000 and 85,000. Madison’s 84,800 is nearly three times Shakespeare’s rate and noticeably higher than the other Founders. His Federalist prose is full of definite-noun phrases (the constitution, the legislature, the executive, the government) that drive his the rate up. That nominal density is part of why the cosine pushes him furthest from Shakespeare on the function-word fingerprint.

The other notable contrast is in the second-person pronoun you: high in Shakespeare (15,000+/M, the dramatic register addressing other characters) and consistently lower in the Founders (roughly 5,000–10,000/M, the formal-letter register where the addressee is named or titled rather than pronouned). The same genre effect that drives the sentence-length gap shows up here. These are two corpora doing two different kinds of writing.

Compare this stylometric signal to the per-method rankings in the Ranking explorer, or the sentence-length divergence in the Sentence Length explorer.