Login
You're viewing the clockwork.monster public feed.
  • Lilaclilacperegrine
    Aug 9, 2026, 3:51 AM

    I had an idea for an automatic translator that would take two texts in any language and output a map between them of which words mean what

    This is the result of English -> English
    (Pride and Prejudice -> Sense and Sensibility)

    as you can see, after the first 5 words it goes downhill fast
    not sure how to fix that one

    (no machine learning involved btw)

    A map showing what words map to what as defined by my program. The first five have "the" mapping to "the" etc, until the 6th where it mixes up "i" and "a", and it just gets worse from there.
    💬 1🔄 0⭐ 1

Replies

  • Lilaclilacperegrine
    Aug 9, 2026, 4:37 AM

    this is based off a markov model
    if we say the transition matrix is M, and that the new text has different words but the same underlying structure (like words are replaced), then M' = P*M*P^T for a permutation matrix P.

    Then, we can take an eigenvector decomposition of both and use that to reverse-engineer P.

    💬 1🔄 0⭐ 0
  • Lilaclilacperegrine
    Aug 9, 2026, 5:16 AM

    ofc solving for eigenvalues means they can take on an arbitrary phase factor, so i need to find them and correct that before comparing them

    💬 0🔄 0⭐ 0