Upload folder using huggingface_hub

night5_json/w42_typology_binding.json

@@ -0,0 +1,350 @@
+{
+  "experiment": "W42 wider-typology no-binding (C3 cross-construction agent/patient binding in typologically diverse case-marking langs jpn/hin/tur/rus/arb)",
+  "shared_operators": "eng-fit R_pos (prepool rotation) + R_role (tae_comp_W_order polar); ridge-E REFIT per language on SONAR-translated native sentences",
+  "config": {
+    "langs": "jpn_Jpan,hin_Deva,tur_Latn,rus_Cyrl,arb_Arab",
+    "device": "cuda",
+    "n_ridge_src": 1400,
+    "ridge_trans_max_seq_len": 48,
+    "n_role_pairs": 48,
+    "min_tok": 4,
+    "max_pos": 64,
+    "seed": 0,
+    "out": "/root/w42_typology_binding.json"
+  },
+  "prereg": {
+    "prediction": "no-binding holds typologically: z reads SURFACE POSITION not thematic role even in case-marked langs (cross-construction role AUC ~chance, surface first-mention transfers). EXCEPT possibly where the case PARTICLE is a readable token (jpn が/を, hin ने/को): any role signal is a LEXICAL case-marker readout (z_bag reads it; collapses on particle neutralization), NOT abstract role.",
+    "falsified_if": "z robustly binds thematic role cross-order in a case-marked language (cross-construction role AUC > .7 invariant to scrambling) WITHOUT it being the case-particle token (z_bag at chance AND survives neutralization) => real abstract thematic binding exists in some languages."
+  },
+  "per_language": {
+    "jpn_Jpan": {
+      "lang": "jpn_Jpan",
+      "ridge_E_ids": 5735,
+      "n_native_ridge_sents": 1400,
+      "native_sent_examples": [
+        "分類を決定する重要な要因を理解することで 雇用者は 誤った分類のリスクを最小限に抑え 労働法の遵守を保証することができます",
+        "従業員または独立した契約者として分類されるかどうかは,労働者,雇用者,そしてより広範な経済にとって重要な影響があります.",
+        "分類に注意深く注意深く取り組むことで 雇用者は誤った分類の罠を回避し 従業員とのより強い より持続可能な関係を築くことができます"
+      ],
+      "binding_marked": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.8927,
+        "role_auc_within_passive": 0.906,
+        "role_auc_pooled_z": 0.7144,
+        "role_auc_zbag_negctrl": 0.4992,
+        "role_auc_ci_field": 0.688,
+        "role_shuffled_null": 0.554,
+        "cross_construction_role_auc_mean": 0.0171,
+        "cross_construction_role_train_act_test_pas": 0.0126,
+        "cross_construction_role_train_pas_test_act": 0.0217,
+        "cross_construction_SURFACE_auc_mean": 0.9829,
+        "cross_construction_zbag_role_auc_mean": 0.3872,
+        "zbag_within_quad_max_dev_mean": 0.0417,
+        "example_quad": [
+          "裁判官が警備員を警告した。",
+          "警備員が裁判官を警告した。",
+          "警備員は裁判官に警告された。",
+          "裁判官は警備員に警告された。"
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "binding_neutralized": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.8903,
+        "role_auc_within_passive": 0.8738,
+        "role_auc_pooled_z": 0.5698,
+        "role_auc_zbag_negctrl": 0.5009,
+        "role_auc_ci_field": 0.5656,
+        "role_shuffled_null": 0.5472,
+        "cross_construction_role_auc_mean": 0.0009,
+        "cross_construction_role_train_act_test_pas": 0.0009,
+        "cross_construction_role_train_pas_test_act": 0.0009,
+        "cross_construction_SURFACE_auc_mean": 0.9991,
+        "cross_construction_zbag_role_auc_mean": 0.4255,
+        "zbag_within_quad_max_dev_mean": 0.0203,
+        "example_quad": [
+          "裁判官警備員警告した。",
+          "警備員裁判官警告した。",
+          "警備員裁判官警告された。",
+          "裁判官警備員警告された。"
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "case_marker_neutralization": {
+        "marked_within_min": 0.8927,
+        "neutralized_within_min": 0.8738,
+        "within_drop_on_neutralize": 0.0189,
+        "marked_xc_role": 0.0171,
+        "neutralized_xc_role": 0.0009,
+        "xc_drop_on_neutralize": 0.0162,
+        "marked_zbag_particle_leak": false,
+        "interpretation": "role signal does NOT depend on the particle token (order-only); binding is surface-positional regardless of overt case"
+      },
+      "verdict": {
+        "cross_construction_role_auc": 0.0171,
+        "surface_first_mention_transfers": 0.9829,
+        "no_binding_holds": true,
+        "binds_thematic_abstractly": false,
+        "lexical_case_marker_readout": false,
+        "summary": "NO-BINDING (surface-positional; cross-construction role at/below chance)"
+      }
+    },
+    "hin_Deva": {
+      "lang": "hin_Deva",
+      "ridge_E_ids": 4052,
+      "n_native_ridge_sents": 1400,
+      "native_sent_examples": [
+        "वर्गीकरण को निर्धारित करने वाले प्रमुख कारकों को समझकर, नियोक्ता गलत वर्गीकरण के जोखिमों को कम कर सकते हैं और श्रम कानूनों का अनुपालन सुनिश्चित कर सकते हैं।",
+        "क्या एक व्यक्ति को एक कर्मचारी या एक स्वतंत्र ठेकेदार के रूप में वर्गीकृत किया जाता है, इसका श्रमिक, नियोक्ता और व्यापक अर्थव्यवस्था पर महत्वपूर्ण प्रभाव पड़ता है।",
+        "वर्गीकरण के साथ सावधानी और विवरण पर ध्यान देने से, नियोक्ता वर्गीकरण के गलतफहमी से बच सकते हैं और अपने कर्मचारियों के साथ मजबूत, अधिक स्थायी संबंध बना सकते हैं।"
+      ],
+      "binding_marked": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.8699,
+        "role_auc_within_passive": 0.858,
+        "role_auc_pooled_z": 0.6662,
+        "role_auc_zbag_negctrl": 0.5015,
+        "role_auc_ci_field": 0.6313,
+        "role_shuffled_null": 0.5569,
+        "cross_construction_role_auc_mean": 0.165,
+        "cross_construction_role_train_act_test_pas": 0.0508,
+        "cross_construction_role_train_pas_test_act": 0.2791,
+        "cross_construction_SURFACE_auc_mean": 0.9477,
+        "cross_construction_zbag_role_auc_mean": 0.5028,
+        "zbag_within_quad_max_dev_mean": 0.0291,
+        "example_quad": [
+          "न्यायाधीश ने रक्षक को चेताया।",
+          "रक्षक ने न्यायाधीश को चेताया।",
+          "रक्षक को न्यायाधीश द्वारा चेताया गया।",
+          "न्यायाधीश को रक्षक द्वारा चेताया गया।"
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "binding_neutralized": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.8441,
+        "role_auc_within_passive": 0.8096,
+        "role_auc_pooled_z": 0.5744,
+        "role_auc_zbag_negctrl": 0.4997,
+        "role_auc_ci_field": 0.5536,
+        "role_shuffled_null": 0.5532,
+        "cross_construction_role_auc_mean": 0.0189,
+        "cross_construction_role_train_act_test_pas": 0.0252,
+        "cross_construction_role_train_pas_test_act": 0.0126,
+        "cross_construction_SURFACE_auc_mean": 0.9811,
+        "cross_construction_zbag_role_auc_mean": 0.5013,
+        "zbag_within_quad_max_dev_mean": 0.0284,
+        "example_quad": [
+          "न्यायाधीश रक्षक चेताया।",
+          "रक्षक न्यायाधीश चेताया।",
+          "रक्षक न्यायाधीश चेताया गया।",
+          "न्यायाधीश रक्षक चेताया गया।"
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "case_marker_neutralization": {
+        "marked_within_min": 0.858,
+        "neutralized_within_min": 0.8096,
+        "within_drop_on_neutralize": 0.0484,
+        "marked_xc_role": 0.165,
+        "neutralized_xc_role": 0.0189,
+        "xc_drop_on_neutralize": 0.1461,
+        "marked_zbag_particle_leak": false,
+        "interpretation": "role signal does NOT depend on the particle token (order-only); binding is surface-positional regardless of overt case"
+      },
+      "verdict": {
+        "cross_construction_role_auc": 0.165,
+        "surface_first_mention_transfers": 0.9477,
+        "no_binding_holds": true,
+        "binds_thematic_abstractly": false,
+        "lexical_case_marker_readout": false,
+        "summary": "NO-BINDING (surface-positional; cross-construction role at/below chance)"
+      }
+    },
+    "tur_Latn": {
+      "lang": "tur_Latn",
+      "ridge_E_ids": 5288,
+      "n_native_ridge_sents": 1400,
+      "native_sent_examples": [
+        "Sınıflandırmayı belirleyen kilit faktörleri anlarsanız, işverenler yanlış sınıflandırma riskini en aza indirebilir ve iş kanunlarına uymayı sağlayabilir.",
+        "Bir kişinin işçi veya bağımsız sözleşmeli olarak sınıflandırılması, işçi, işveren ve daha geniş bir ekonomi için önemli etkilere sahiptir.",
+        "Klasifikasyonu dikkatle ve ayrıntılara dikkatle yaklaşarak, işverenler yanlış sınıflandırmanın tehlikelerinden kaçınabilir ve işçilerle daha güçlü, daha sürdürülebilir ilişkiler kurabilirler."
+      ],
+      "binding_marked": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.8739,
+        "role_auc_within_passive": 0.8818,
+        "role_auc_pooled_z": 0.7077,
+        "role_auc_zbag_negctrl": 0.5514,
+        "role_auc_ci_field": 0.7083,
+        "role_shuffled_null": 0.5331,
+        "cross_construction_role_auc_mean": 0.0202,
+        "cross_construction_role_train_act_test_pas": 0.0334,
+        "cross_construction_role_train_pas_test_act": 0.0069,
+        "cross_construction_SURFACE_auc_mean": 0.9798,
+        "cross_construction_zbag_role_auc_mean": 0.0399,
+        "zbag_within_quad_max_dev_mean": 0.0859,
+        "example_quad": [
+          "Bekçi hâkimi uyardı.",
+          "Hâkim bekçiyi uyardı.",
+          "Hâkim bekçi tarafından uyarıldı.",
+          "Bekçi hâkim tarafından uyarıldı."
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "verdict": {
+        "cross_construction_role_auc": 0.0202,
+        "surface_first_mention_transfers": 0.9798,
+        "no_binding_holds": true,
+        "binds_thematic_abstractly": false,
+        "lexical_case_marker_readout": false,
+        "summary": "NO-BINDING (surface-positional; cross-construction role at/below chance)"
+      }
+    },
+    "rus_Cyrl": {
+      "lang": "rus_Cyrl",
+      "ridge_E_ids": 5229,
+      "n_native_ridge_sents": 1400,
+      "native_sent_examples": [
+        "Понимая ключевые факторы, определяющие классификацию, работодатели могут минимизировать риски неправильной классификации и обеспечить соблюдение трудовых законов.",
+        "Независимо от того, классифицируется ли человек в качестве работника или независимого подрядчика, это оказывает значительное влияние на работника, работодателя и общую экономику.",
+        "При подходе к классификации с осторожностью и вниманием к деталям работодатели могут избежать ловушек неправильной классификации и построить более прочные, более устой"
+      ],
+      "binding_marked": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.886,
+        "role_auc_within_passive": 0.9078,
+        "role_auc_pooled_z": 0.7651,
+        "role_auc_zbag_negctrl": 0.6616,
+        "role_auc_ci_field": 0.7929,
+        "role_shuffled_null": 0.4717,
+        "cross_construction_role_auc_mean": 0.0314,
+        "cross_construction_role_train_act_test_pas": 0.0369,
+        "cross_construction_role_train_pas_test_act": 0.026,
+        "cross_construction_SURFACE_auc_mean": 0.9686,
+        "cross_construction_zbag_role_auc_mean": 0.0243,
+        "zbag_within_quad_max_dev_mean": 0.0901,
+        "example_quad": [
+          "Рабочий обвинил сторожа.",
+          "Сторож обвинил рабочего.",
+          "Сторож был обвинён рабочим.",
+          "Рабочий был обвинён сторожем."
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": true
+      },
+      "verdict": {
+        "cross_construction_role_auc": 0.0314,
+        "surface_first_mention_transfers": 0.9686,
+        "no_binding_holds": true,
+        "binds_thematic_abstractly": false,
+        "lexical_case_marker_readout": true,
+        "summary": "LEXICAL CASE-MARKER READOUT (within-role rides particle in z_bag; cross-construction surface)"
+      }
+    },
+    "arb_Arab": {
+      "lang": "arb_Arab",
+      "ridge_E_ids": 2999,
+      "n_native_ridge_sents": 1400,
+      "native_sent_examples": [
+        "من خلال فهم العوامل الرئيسية التي تحدد التصنيف، يمكن لأصحاب العمل تقليل مخاطر التصنيف الخاطئ وضمان الامتثال لقوانين العمل.",
+        "إما أن يتم تصنيف الفرد كموظف أو متعاقد مستقل له آثار مهمة على الموظف، والرب، والاقتصاد الأوسع.",
+        "من خلال التعامل مع التصنيف بعناية وا��اهتمام بالتفاصيل، يمكن لأصحاب العمل تجنب أخطاء التصنيف الخاطئ وبناء علاقات أقوى وأكثر"
+      ],
+      "binding_marked": {
+        "n_rows": 192,
+        "n_pairs": 48,
+        "role_auc_within_active": 0.6989,
+        "role_auc_within_passive": 0.8319,
+        "role_auc_pooled_z": 0.6389,
+        "role_auc_zbag_negctrl": 0.494,
+        "role_auc_ci_field": 0.6371,
+        "role_shuffled_null": 0.5052,
+        "cross_construction_role_auc_mean": 0.1663,
+        "cross_construction_role_train_act_test_pas": 0.112,
+        "cross_construction_role_train_pas_test_act": 0.2205,
+        "cross_construction_SURFACE_auc_mean": 0.8337,
+        "cross_construction_zbag_role_auc_mean": 0.503,
+        "zbag_within_quad_max_dev_mean": 0.0674,
+        "example_quad": [
+          "حذر الحارس القاضي.",
+          "حذر القاضي الحارس.",
+          "حُذر القاضي من قبل الحارس.",
+          "حُذر الحارس من قبل القاضي."
+        ],
+        "thematic_generalisation": false,
+        "surface_positional": true,
+        "within_construction_binding": true,
+        "zbag_particle_leak": false
+      },
+      "verdict": {
+        "cross_construction_role_auc": 0.1663,
+        "surface_first_mention_transfers": 0.8337,
+        "no_binding_holds": true,
+        "binds_thematic_abstractly": false,
+        "lexical_case_marker_readout": false,
+        "summary": "NO-BINDING (surface-positional; cross-construction role at/below chance)"
+      }
+    }
+  },
+  "typology_verdict": {
+    "per_language_xc_role_auc": {
+      "jpn_Jpan": 0.0171,
+      "hin_Deva": 0.165,
+      "tur_Latn": 0.0202,
+      "rus_Cyrl": 0.0314,
+      "arb_Arab": 0.1663
+    },
+    "per_language_surface_transfer": {
+      "jpn_Jpan": 0.9829,
+      "hin_Deva": 0.9477,
+      "tur_Latn": 0.9798,
+      "rus_Cyrl": 0.9686,
+      "arb_Arab": 0.8337
+    },
+    "per_language_summary": {
+      "jpn_Jpan": "NO-BINDING (surface-positional; cross-construction role at/below chance)",
+      "hin_Deva": "NO-BINDING (surface-positional; cross-construction role at/below chance)",
+      "tur_Latn": "NO-BINDING (surface-positional; cross-construction role at/below chance)",
+      "rus_Cyrl": "LEXICAL CASE-MARKER READOUT (within-role rides particle in z_bag; cross-construction surface)",
+      "arb_Arab": "NO-BINDING (surface-positional; cross-construction role at/below chance)"
+    },
+    "languages_with_abstract_thematic_binding": [],
+    "languages_lexical_case_marker_readout": [
+      "rus_Cyrl"
+    ],
+    "languages_no_binding": [
+      "jpn_Jpan",
+      "hin_Deva",
+      "tur_Latn",
+      "rus_Cyrl",
+      "arb_Arab"
+    ],
+    "no_binding_holds_typologically": true,
+    "falsifier_status": "NOT FALSIFIED: no-binding holds typologically (z reads surface position, not thematic role, even in case-marked langs; any marked-case role signal is lexical/within-construction and does not generalise cross-construction)",
+    "headline": "Cross-construction agent/patient role AUC across typologically diverse case-marking languages. KEY: Japanese が/を (lexical case particles) — does z read the morphologically-marked role abstractly, or only as a within-construction particle readout? See per_language_summary + case_marker_neutralization."
+  }
+}