Agreement, quadratic-weighted Cohen's kappa, n=24
(the 4 items with no model answer are excluded: both coders were told to score them 0)
  intent_inference           kappa 0.88   exact 92%
  social_appropriateness     kappa 0.30   exact 62%
  unsupported_assumptions    kappa 0.70   exact 83%
  semantic_preservation      kappa 0.89   exact 92%
  pooled                     kappa 0.72
  totals: correlation 0.88, mean absolute difference 0.71

Coder 1
  language, context implicit       -1.14  [-3.86, +2.29]  n=7
  language, context explicit       -1.71  [-4.00, +1.00]  n=7
  context, English                 +1.14  [-2.29, +4.57]  n=7
  context, Roman Urdu              +0.57  [-0.86, +2.00]  n=7
  language, implicit, clean only   -3.25  [-5.75, -1.50]  n=4
  per scenario, English implicit to Roman Urdu implicit:
    ambiguity            8.0 -> 5.0
    code_switch          8.0 -> 5.0  (truncated response in this scenario)
    deference            0.0 -> 8.0  (truncated response in this scenario)
    face_saving          0.0 -> 0.0  (truncated response in this scenario)
    family_obligation    8.0 -> 7.0
    favor_indirect       8.0 -> 6.0
    kinship_context      8.0 -> 1.0

Coder 2
  language, context implicit       -0.43  [-3.57, +3.00]  n=7
  language, context explicit       -0.29  [-2.29, +2.00]  n=7
  context, English                 +0.86  [-2.57, +4.29]  n=7
  context, Roman Urdu              +1.00  [-0.86, +3.14]  n=7
  language, implicit, clean only   -2.25  [-6.00, +0.00]  n=4
  per scenario, English implicit to Roman Urdu implicit:
    ambiguity            8.0 -> 7.0
    code_switch          8.0 -> 6.0  (truncated response in this scenario)
    deference            0.0 -> 8.0  (truncated response in this scenario)
    face_saving          0.0 -> 0.0  (truncated response in this scenario)
    family_obligation    8.0 -> 8.0
    favor_indirect       8.0 -> 8.0
    kinship_context      8.0 -> 0.0

Mean of both coders
  language, context implicit       -0.79  [-3.71, +2.64]  n=7
  language, context explicit       -1.00  [-3.00, +1.50]  n=7
  context, English                 +1.00  [-2.43, +4.43]  n=7
  context, Roman Urdu              +0.79  [-0.86, +2.57]  n=7
  language, implicit, clean only   -2.75  [-5.88, -0.75]  n=4
  per scenario, English implicit to Roman Urdu implicit:
    ambiguity            8.0 -> 6.0
    code_switch          8.0 -> 5.5  (truncated response in this scenario)
    deference            0.0 -> 8.0  (truncated response in this scenario)
    face_saving          0.0 -> 0.0  (truncated response in this scenario)
    family_obligation    8.0 -> 7.5
    favor_indirect       8.0 -> 7.0
    kinship_context      8.0 -> 0.5