Compare texts
Repeated sequences
Compare two selected works by looking for identical runs of words.
What the terms mean
- Token
- One word of the text as it stands, counted separately each time it occurs. μῆνιν ἄειδε θεά is three tokens.
- Form
- The written shape of a word, accents and endings included, as opposed to the dictionary word it belongs to. λέγει and λέγω are two forms of one word.
- n-gram
- A run of n consecutive tokens. A 3-gram is any three words that follow one another. Searching for shared n-grams is how the same turn of phrase is found in two works.
- Stop-word
- A word so common — καί, δέ, τό — that its presence says nothing about a borrowing. It still belongs to the sequence, but adds nothing to the score.
Texts and method
Select at least two works in the catalogue.
Diacritics are ignored by default here, unlike in word lists and statistics: the same words change accent with their position (grave or acute, enclitics) and between editions, and one differing accent would break a shared run.
Exposed stop-list
A sequence must contain at least two words outside this editable list. Stop-words remain part of the exact sequence but do not increase its improbability score.
Lemma sequences rely on Sôtêr, the automatic lemmatiser, having annotated the whole corpus. When it hesitates between several lemmas for a word, all of them are kept together; none is picked at random.
Results
Shared sequences
Scoring method
The score is the sum of the negative base-10 logarithms of add-one-smoothed token probabilities in the two compared texts, excluding the visible stop-list. A larger score means that the shared sequence is less probable under this explicit independence model. It is a ranking aid, not proof of dependence or direction of borrowing.