Siang-Tan-Ūi

Ēng lâi hun-sek bûn-pún (text) kán-tan koh ū-sū-sái. Tòng-chò gí-giân bô͘-sek (language model) lâi chò gí-im piān-sek (speech recognition) mā kài chán (Collins, 1996). Siang-tan-ūi sǹg sī N-tan-ūi (N-gram) ê 1 ê te̍k-lē.

Làng-phāng siang-tan-ūi (Gappy bigram, skipping bigram) kí 2 ê tan-ūi tiong-ng ū làng-phāng, chhiūⁿ kóng làng koè liân-chiap-jī (connecting word), iah sī kóng ti oá-loā bûn-hoat (dependency grammar) lāi-té beh bô͘-hóng oá-loā ê koan-hē.

Thâu-jī siang-tan-ūi (Head word bigram) tō sī 1 chióng ū bêng-khak oá-loā koan-hē ê làng-phāng siang-tan-ūi.

Tī bi̍t-bé-ha̍k (cryptography) ū 1 chióng siang-tan-ūi pîn-lu̍t kong-kek (bigram frequency attack), lī-iōng pîn-lu̍t hun-sek (frequency analysis) lâi kái-phoà àm-bé (cryptogram).

Nā chai-iáⁿ siang-tan-ūi ê ki-lu̍t kap thaû-chêng hit ê tan-ūi ê ki-lu̍t, lán tō ē-tit ēng Bayes tēng-lí (Bayes' theorem) lâi sǹg aū-piah hit ê tan-ūi ê tiâu-kiāⁿ ki-lu̍t:

${\displaystyle P(T_{n}|T_{n-1})={P(T_{n-1},T_{n}) \over P(T_{n-1})}}$ 

Iā tō sī kóng, nā chai-iáⁿ ${\displaystyle T_{n-1}}$  ê ki-lu̍t, án-ne ${\displaystyle T_{n}}$  ê ki-lu̍t ${\displaystyle P()}$  tō sī siang-tan-ūi ê ki-lu̍t ${\displaystyle P(T_{n-1},T_{n})}$  khì tû-í thâu-chêng tan-ūi ${\displaystyle T_{n-1}}$  ê ki-lu̍t.

Pún bûn-chiuⁿ sī chi̍t phiⁿ phí-á-kiáⁿ. Lí thang tàu khok-chhiong lâi pang-chō͘ Wiki Bân-lâm-gú. hián lūn pian

• Digraph
• N-tan-ūi (N-gram)