analyze_results.py

import json, random, statistics, sys
from collections import defaultdict

PATH = sys.argv[1] if len(sys.argv) > 1 else "results_scored.jsonl"
DIMS = ["intent_inference", "social_appropriateness",
        "unsupported_assumptions", "semantic_preservation"]

rows = [json.loads(x) for x in open(PATH, encoding="utf-8") if x.strip()]

missing = [r["id"] for r in rows if not all(d in r for d in DIMS)]
if missing:
    sys.exit(f"missing score fields on: {missing}")
bad = [r["id"] for r in rows if not all(int(r[d]) in (0, 1, 2) for d in DIMS)]
if bad:
    sys.exit(f"scores outside 0-2 on: {bad}")

for r in rows:
    r["total"] = sum(int(r[d]) for d in DIMS)

pairs = defaultdict(dict)
by_cond = defaultdict(list)
for r in rows:
    pairs[r["scenario_id"]][r["condition"]] = r["total"]
    by_cond[r["condition"]].append(r["total"])

ORDER = ["implicit_english", "roman_urdu_mixed",
         "explicit_english", "explicit_context_added"]

def ci(values, n=10000, seed=42):
    rng = random.Random(seed)
    means = sorted(statistics.mean([rng.choice(values) for _ in values])
                   for _ in range(n))
    return means[int(.025 * n)], means[int(.975 * n)]

def diffs(a, b):
    return [p[a] - p[b] for p in pairs.values() if a in p and b in p]

def report(label, a, b):
    d = diffs(a, b)
    if not d:
        return
    lo, hi = ci(d)
    print(f"{label:34s} {statistics.mean(d):+.3f}  "
          f"95% CI [{lo:+.3f}, {hi:+.3f}]  n={len(d)}")

print("Mean total score out of 8, by condition")
for c in ORDER:
    v = by_cond.get(c, [])
    if v:
        print(f"  {c:26s} {statistics.mean(v):.3f}   n={len(v)}")

print("\nPaired differences")
report("language, context implicit", "roman_urdu_mixed", "implicit_english")
report("language, context explicit", "explicit_context_added", "explicit_english")
report("context, English", "explicit_english", "implicit_english")
report("context, Roman Urdu", "explicit_context_added", "roman_urdu_mixed")

inter = [(p["explicit_context_added"] - p["roman_urdu_mixed"])
         - (p["explicit_english"] - p["implicit_english"])
         for p in pairs.values() if len(p) == 4]
if inter:
    lo, hi = ci(inter)
    print(f"\n{'interaction, recovery in RU vs EN':34s} {statistics.mean(inter):+.3f}  "
          f"95% CI [{lo:+.3f}, {hi:+.3f}]  n={len(inter)}")

print("\nPer-dimension means by condition")
hdr = "  " + " ".join(f"{d[:12]:>13s}" for d in DIMS)
print(f"  {'':26s}{hdr}")
for c in ORDER:
    sel = [r for r in rows if r["condition"] == c]
    if not sel:
        continue
    cells = " ".join(f"{statistics.mean([int(r[d]) for r in sel]):>13.2f}" for d in DIMS)
    print(f"  {c:26s}  {cells}")