Function words

Function words (the, of, and, to, was) carry grammatical structure rather than content meaning. They form a small closed class, used at high frequency and largely unconsciously, which makes their usage rates powerful evidence in authorship attribution.

The numbers

Definition

Function words are the words that carry grammatical structure rather than dictionary meaning: articles (the, a), pronouns (he, she), prepositions (in, of, under), conjunctions (and, but, if), auxiliary verbs (would, could, had) and particles (up, out). They contrast with content words, the nouns, main verbs, adjectives and adverbs that carry a sentence's substance.

Function words form a closed class: it is very uncommon for a language to gain a new preposition or article, whereas the open class of content words admits new nouns and verbs every day. The set of distinct function words is small, but its members occur far more frequently in running text than content words do.

Why authorship analysis leans on them

Kestemont's (2014) survey of stylometry gives three linked reasons. Function words are very frequent, so even modest samples yield reliable statistics. They are largely independent of topic: every author must use 'the' and 'of' whatever the subject. And they are processed with little conscious attention, which makes an author's function-word profile hard to disguise and hard for an imitator to copy. Mosteller and Wallace's 1964 Federalist study, built on exactly this shift from rare, striking features to a language's most frequent words, is credited as the start of modern non-traditional authorship attribution.

The topic-independence is not absolute. Pronoun frequencies in particular correlate with genre and narrative perspective, which is why many studies remove pronouns ('pronoun culling') before comparing texts.

The Federalist Papers case

The Federalist Papers are 85 essays by Alexander Hamilton, James Madison and John Jay, published under the shared pseudonym Publius. The authorship of 12 essays was disputed between Hamilton and Madison. In 1964 the statisticians Frederick Mosteller and David Wallace attacked the problem with Bayesian inference over word rates measured in the essays of known authorship.

They settled on about 30 discriminating words, dominated by function words, precisely because both men wrote on the same topics in a deliberately shared persona: content words could not separate them, but grammatical habits could. Hamilton used 'upon' at an average rate of 3.24 per 1,000 words while Madison barely used it at all; Madison was almost twice as likely as Hamilton to use 'by'; Hamilton wrote 'while' where Madison consistently wrote 'whilst'.

After validating the method on essays of known authorship, their model attributed all twelve disputed essays to Madison, with odds reaching about 160 billion to 1 in the strongest case. The result corroborated the historian Douglass Adair's 1944 assignments and remains the accepted attribution.

Worked example

Classify each word of this 12-word sentence: 'The report was sent to the manager before the meeting had started.'

Function words: the (three times), was, to, before, had; that is 7 tokens. Content words: report, sent, manager, meeting, started; that is 5 tokens. Function-word share: 7 / 12 = 58%.

More than half of an ordinary English sentence is grammatical scaffolding. Because that scaffolding is produced constantly and unconsciously, small per-author differences in its rates, like Hamilton's fondness for 'upon', accumulate into a measurable fingerprint.

Sources

More terms

ScriptGrain · Why ScriptGrain · Sample voice profiles · The Journal · Support