# Rules Rules link words in code, without a model. They only add links between words that have no link yet, only one-to-one, and only when the match is unique in the sentence pair. Rule links are sure, with confidence 0.95 (0.98 for identical symbols) and agent `rules`. | Rule | Links when | |---|---| | `identical` | the same numeral or symbol (or other form without letters) occurs once on each side; punctuation only with `link_punctuation: true`, because alignment guidelines such as Ugarit's never align it | | `names` | a proper noun on one side and a word on the other have matching consonant skeletons (transliterations from MISC `LTranslit`/`Translit`, lemma, form; Greek romanised; similarity ≥ 0.9, short skeletons equal), with no equally good rival | | `lexicon` | the lemma's top translation has at least `min_count` (5) links and covers at least `min_share` (95%) of **all** the lemma's occurrences, the lemma occurs once in its sentence and the translation once (unlinked) in the other | The lexicon rule counts only links with confidence ≥ 0.96, above what rules write, so rules never feed themselves; links that agents give high confidence do count, so the rule grows during a run. ```bash mcaligner rules $SRC -a pred=... --target pred # add links mcaligner rules $SRC -a silver=...:ro --measure silver --show-errors ``` `--measure` proposes links for every pair of a reference set, starting from no links and with the pair left out of the evidence, and reports each rule's precision. ## Measured precision On the 155 verses of Ephesians against the silver links: | Rule | Links | In silver | After reading the disagreements | |---|---|---|---| | identical | 0 | – | the Greek text has no punctuation | | names | 20 | 18 | 20/20 correct: both disagreements are silver errors | | lexicon | 129 | 127 | 128/129 correct | Looser lexicon settings add links at lower precision: (3, 0.9) gives 229 links with 3 real errors; (5, 0.8) 265 links at 94.7% against silver.