Perm Stat :: Normalizations (1/2)

• We permutated sentences instead of n-grams

  • Because sentences are intra-dependent

• Normalizing is crucial

  • to make corrections for differences in sentence-length

• Two normalizations were used

  • The basic in-permutation normalization
    • much weight for frequent, significant extreme n-grams
  • Between-permutations normalization
    • extreme significant n-grams regardless of frequency

Notes: