id
stringlengths 1
12
| question
stringlengths 22
566
| golden_answers
dict | data_source
stringclasses 2
values | prompt
listlengths 1
1
| ability
stringclasses 6
values | reward_model
dict | extra_info
dict | metadata
dict |
|---|---|---|---|---|---|---|---|---|
exfever_1151
|
Clint Eastwood is an American actor and composer who rose to international fame with his role as the "Man with No Name" in the "Dollars Trilogy" of the broad subgenre of Western films produced in Asia.
|
{
"evidence": [
"Clint Eastwood",
"Spaghetti Western"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nClint Eastwood is an American actor and composer who rose to international fame with his role as the \"Man with No Name\" in the \"Dollars Trilogy\" of the broad subgenre of Western films produced in Asia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Clint Eastwood",
"Spaghetti Western"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 288,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
80507
|
Paul Gerrie was searching for suitable to open a winery, he visited France whose capital is Paris and Oregon before hiring a consultant to look for an appropriate location.
|
{
"evidence": [
"Cristom Vineyards_sentence_2",
"Cristom Vineyards_sentence_3",
"Cristom Vineyards_sentence_4",
"Cristom Vineyards_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPaul Gerrie was searching for suitable to open a winery, he visited France whose capital is Paris and Oregon before hiring a consultant to look for an appropriate location.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cristom Vineyards_sentence_2",
"Cristom Vineyards_sentence_3",
"Cristom Vineyards_sentence_4",
"Cristom Vineyards_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 289,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.234",
"value": "undefined"
},
{
"operation": "search",
"time": "11.926",
"value": "Paul Gerrie"
},
{
"operation": "Now on",
"time": "13.427",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "29.932",
"value": "Cristom Vineyards"
},
{
"operation": "Now on",
"time": "31.051",
"value": "Cristom Vineyards"
},
{
"operation": "Highlighting",
"time": "36.453",
"value": "Cristom Vineyards_sentence_2"
},
{
"operation": "Highlighting",
"time": "44.47",
"value": "Cristom Vineyards_sentence_3"
},
{
"operation": "Highlighting",
"time": "71.68",
"value": "Cristom Vineyards_sentence_4"
},
{
"operation": "Highlighting",
"time": "79.316",
"value": "Cristom Vineyards_sentence_5"
},
{
"operation": "finish",
"time": "100.896",
"value": "finish"
}
]
}
|
exfever_724
|
Zootopia opened to record-breaking box offices in several countries, and earned a worldwide gross of over $1 billion, making it the fourth-highest-grossing film of 2016.
|
{
"evidence": [
"Zootopia",
"Computer animation"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nZootopia opened to record-breaking box offices in several countries, and earned a worldwide gross of over $1 billion, making it the fourth-highest-grossing film of 2016.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Zootopia",
"Computer animation"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 290,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_8665
|
Legion is a 2010 British film starring an actor known for his role in a 1998 romantic comedy film directed by Nancy Meyers.
|
{
"evidence": [
"Legion (2010 film)",
"Dennis Quaid",
"The Parent Trap (1998 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLegion is a 2010 British film starring an actor known for his role in a 1998 romantic comedy film directed by Nancy Meyers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Legion (2010 film)",
"Dennis Quaid",
"The Parent Trap (1998 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 291,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
1346
|
The population of the European Union, with some five hundred million residents, accounts for a highly populated, culturally diverse union of 27 member states and about 20–25 million residents (3%) are members of diasporas of non-European origin.
|
{
"evidence": [
"European Union_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe population of the European Union, with some five hundred million residents, accounts for a highly populated, culturally diverse union of 27 member states and about 20–25 million residents (3%) are members of diasporas of non-European origin.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"European Union_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 292,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "48.524",
"value": "European Union"
},
{
"operation": "Highlighting",
"time": "113.729",
"value": "European Union_sentence_2"
},
{
"operation": "Highlighting",
"time": "122.196",
"value": "European Union_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "131.624",
"value": "European Union_sentence_2"
},
{
"operation": "finish",
"time": "143.978",
"value": "finish"
}
]
}
|
53194
|
Learning about the exploits of the men in the outfit, General Pavel Melik-Shahnazaryan advised Bagramyan to return to Tiflis (capital of the country of Georgia) to enroll in the Praporshchik Military Academy; but in order to attend the school, he needed to satisfy the academy's requirement of having completed school at a gymnasium.
|
{
"evidence": [
"Ivan Bagramyan_sentence_27",
"Ivan Bagramyan_sentence_28",
"Tbilisi_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLearning about the exploits of the men in the outfit, General Pavel Melik-Shahnazaryan advised Bagramyan to return to Tiflis (capital of the country of Georgia) to enroll in the Praporshchik Military Academy; but in order to attend the school, he needed to satisfy the academy's requirement of having completed school at a gymnasium.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Ivan Bagramyan_sentence_27",
"Ivan Bagramyan_sentence_28",
"Tbilisi_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 293,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.019",
"value": "?search="
},
{
"operation": "search",
"time": "6.561",
"value": "Pavel Melik-Shahnazaryan"
},
{
"operation": "Now on",
"time": "7.255",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "12.651",
"value": "Ivan Bagramyan"
},
{
"operation": "Now on",
"time": "13.752",
"value": "Ivan Bagramyan"
},
{
"operation": "Highlighting",
"time": "19.295",
"value": "Ivan Bagramyan_sentence_27"
},
{
"operation": "Highlighting",
"time": "40.171",
"value": "Ivan Bagramyan_sentence_28"
},
{
"operation": "search",
"time": "54.304",
"value": "georgia (country)"
},
{
"operation": "Now on",
"time": "55.766",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "57.637",
"value": "Parliament of Georgia (country)"
},
{
"operation": "Now on",
"time": "58.495",
"value": "Parliament of Georgia (country)"
},
{
"operation": "search",
"time": "66.175",
"value": "georgia"
},
{
"operation": "Now on",
"time": "67.052",
"value": "Georgia"
},
{
"operation": "search",
"time": "91.51",
"value": "Tbilisi"
},
{
"operation": "Now on",
"time": "102.81",
"value": "Tbilisi"
},
{
"operation": "Highlighting",
"time": "118.531",
"value": "Tbilisi_sentence_0"
},
{
"operation": "finish",
"time": "124.679",
"value": "finish"
}
]
}
|
4149
|
IrAero and UVT Aero, both headquartered in St. Petersberg, offer only domestic flights through the Strigino International Airport.
|
{
"evidence": [
"IrAero_sentence_0",
"UVT Aero_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIrAero and UVT Aero, both headquartered in St. Petersberg, offer only domestic flights through the Strigino International Airport.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"IrAero_sentence_0",
"UVT Aero_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 294,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "22.552",
"value": "IrAero and UVT Aero,"
},
{
"operation": "search",
"time": "39.013",
"value": "Strigino International Airport"
},
{
"operation": "Page search",
"time": "47.845",
"value": "domestic"
},
{
"operation": "Page search",
"time": "71.804",
"value": "IrAero"
},
{
"operation": "hyperlink",
"time": "93.247",
"value": "IrAero"
},
{
"operation": "Highlighting",
"time": "107.671",
"value": "IrAero_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "111.368",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "114.704",
"value": "UVT Aero"
},
{
"operation": "Highlighting",
"time": "118.891",
"value": "UVT Aero_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "125.96",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "135.879",
"value": "Saint Petersburg"
},
{
"operation": "back-button-clicked",
"time": "146.575",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "151.067",
"value": "IrAero"
},
{
"operation": "Page search",
"time": "160.036",
"value": "headqu"
},
{
"operation": "Page search",
"time": "174.659",
"value": "Petersberg"
},
{
"operation": "finish",
"time": "197.509",
"value": "finish"
}
]
}
|
32309
|
In January 2004, CryptoLogic Ltd., listed on the Toronto Stock Exchange (CRY) in 1998 and on the NASDAQ (CRYP) in 2000, announced the licensing agreement with an online gambling company which operates the world's largest online betting exchange, for the launch of an exclusive online poker game.
|
{
"evidence": [
"CryptoLogic_sentence_11",
"CryptoLogic_sentence_8",
"Betfair_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn January 2004, CryptoLogic Ltd., listed on the Toronto Stock Exchange (CRY) in 1998 and on the NASDAQ (CRYP) in 2000, announced the licensing agreement with an online gambling company which operates the world's largest online betting exchange, for the launch of an exclusive online poker game.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"CryptoLogic_sentence_11",
"CryptoLogic_sentence_8",
"Betfair_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 295,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.1",
"value": "?search="
},
{
"operation": "search",
"time": "16.908",
"value": "CryptoLogic"
},
{
"operation": "Now on",
"time": "17.829",
"value": "CryptoLogic"
},
{
"operation": "Page search",
"time": "34.708",
"value": "stock exchange"
},
{
"operation": "hyperlink",
"time": "42.121",
"value": "Toronto Stock Exchange"
},
{
"operation": "Now on",
"time": "42.849",
"value": "Toronto Stock Exchange"
},
{
"operation": "back-button-clicked",
"time": "44.833",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "45.006",
"value": "CryptoLogic"
},
{
"operation": "Highlighting",
"time": "48.053",
"value": "CryptoLogic_sentence_11"
},
{
"operation": "Page search",
"time": "62.781",
"value": "nasdaq"
},
{
"operation": "Page search",
"time": "88.323",
"value": "gambling"
},
{
"operation": "Highlighting",
"time": "120.416",
"value": "CryptoLogic_sentence_8"
},
{
"operation": "hyperlink",
"time": "123.752",
"value": "Betfair"
},
{
"operation": "Now on",
"time": "124.401",
"value": "Betfair"
},
{
"operation": "Highlighting",
"time": "127.466",
"value": "Betfair_sentence_0"
},
{
"operation": "finish",
"time": "193.415",
"value": "finish"
}
]
}
|
32782
|
Vasileios Sapountzaki attended the University of Athens, where he studied languages and music, and he retired from service on 8 January 1745.
|
{
"evidence": [
"Vasileios Sapountzakis_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVasileios Sapountzaki attended the University of Athens, where he studied languages and music, and he retired from service on 8 January 1745.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vasileios Sapountzakis_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 296,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.63",
"value": "?search="
},
{
"operation": "search",
"time": "19.1",
"value": "Vasileios Sapountzakis"
},
{
"operation": "Now on",
"time": "19.948",
"value": "Vasileios Sapountzakis"
},
{
"operation": "Highlighting",
"time": "54.415",
"value": "Vasileios Sapountzakis_sentence_3"
},
{
"operation": "finish",
"time": "66.232",
"value": "finish"
}
]
}
|
29100
|
Rapid economic development in Hong Kong during the 1970s is attributed to a shrinking labor force.
|
{
"evidence": [
"Hong Kong_sentence_226"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRapid economic development in Hong Kong during the 1970s is attributed to a shrinking labor force.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hong Kong_sentence_226"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 297,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.674",
"value": "?search="
},
{
"operation": "search",
"time": "29.159",
"value": "hong kong 1970"
},
{
"operation": "Now on",
"time": "30.769",
"value": "undefined"
},
{
"operation": "search",
"time": "68.966",
"value": "hong kong shrinking labor force"
},
{
"operation": "Now on",
"time": "70.521",
"value": "undefined"
},
{
"operation": "search",
"time": "105.19",
"value": "Rapid economic development in Hong Kong during the 1970s"
},
{
"operation": "Now on",
"time": "108.645",
"value": "undefined"
},
{
"operation": "search",
"time": "156.498",
"value": "1970 hong kong economy"
},
{
"operation": "Now on",
"time": "157.571",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "190.904",
"value": "Economy of China"
},
{
"operation": "Now on",
"time": "192.226",
"value": "Economy of China"
},
{
"operation": "search",
"time": "204.213",
"value": "Economy of Hong Kong"
},
{
"operation": "Now on",
"time": "205.245",
"value": "Economy of Hong Kong"
},
{
"operation": "Page search",
"time": "209.736",
"value": "1970"
},
{
"operation": "search",
"time": "258.102",
"value": "hong kong"
},
{
"operation": "Now on",
"time": "259.696",
"value": "Hong Kong"
},
{
"operation": "Page search",
"time": "291.018",
"value": "1970"
},
{
"operation": "Page search",
"time": "294.769",
"value": "1970"
},
{
"operation": "Page search",
"time": "385.248",
"value": "labor force"
},
{
"operation": "Highlighting",
"time": "476.072",
"value": "Hong Kong_sentence_226"
},
{
"operation": "finish",
"time": "491.367",
"value": "finish"
}
]
}
|
exfever_651
|
IZombie is an American supernatural procedural crime drama television series developed by an American television producer who's also has included credits in an American sitcom television series that is based on the British comedy series of the same name that was created by Chris Niel.
|
{
"evidence": [
"IZombie (TV series)",
"Diane Ruggiero",
"Free Agents (U.S. TV series)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIZombie is an American supernatural procedural crime drama television series developed by an American television producer who's also has included credits in an American sitcom television series that is based on the British comedy series of the same name that was created by Chris Niel.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"IZombie (TV series)",
"Diane Ruggiero",
"Free Agents (U.S. TV series)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 298,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
53530
|
In the 1990s, WMIO, founded by Maria Isabel Ortiz Avilés, which is the sister of radio entrepreneur David Ortiz, was a repeater for WFID (a radio station licensed to serve Río Piedras, Puerto Rico, established in 1958), and later with WIAC-FM, until was sold to Bestov Broadcasting on April 14, 1999.
|
{
"evidence": [
"WMIO_sentence_4",
"WMIO_sentence_6",
"WFID_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn the 1990s, WMIO, founded by Maria Isabel Ortiz Avilés, which is the sister of radio entrepreneur David Ortiz, was a repeater for WFID (a radio station licensed to serve Río Piedras, Puerto Rico, established in 1958), and later with WIAC-FM, until was sold to Bestov Broadcasting on April 14, 1999.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"WMIO_sentence_4",
"WMIO_sentence_6",
"WFID_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 299,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.759",
"value": "?search="
},
{
"operation": "search",
"time": "15.621",
"value": "WMIO"
},
{
"operation": "Now on",
"time": "20.043",
"value": "WMIO"
},
{
"operation": "Highlighting",
"time": "386.222",
"value": "WMIO_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "386.709",
"value": "WMIO_sentence_0"
},
{
"operation": "Page search",
"time": "394.399",
"value": "maria"
},
{
"operation": "Highlighting",
"time": "399.229",
"value": "WMIO_sentence_4"
},
{
"operation": "Highlighting",
"time": "406.65",
"value": "WMIO_sentence_6"
},
{
"operation": "hyperlink",
"time": "418.501",
"value": "WFID"
},
{
"operation": "Now on",
"time": "419.623",
"value": "WFID"
},
{
"operation": "Highlighting",
"time": "422.82",
"value": "WFID_sentence_1"
},
{
"operation": "finish",
"time": "430.891",
"value": "finish"
}
]
}
|
exfever_990
|
Sewardesses provide routine services and respond to emergencies to ensure the safety and comfort of airline passengers.
|
{
"evidence": [
"Pan Am (TV series)",
"Jet Age"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSewardesses provide routine services and respond to emergencies to ensure the safety and comfort of airline passengers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pan Am (TV series)",
"Jet Age"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 300,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
34187
|
Red Sea Works is the world's fourth-largest producer and supplier of potash products and it has customers in over 60 countries.
|
{
"evidence": [
"Dead Sea products_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRed Sea Works is the world's fourth-largest producer and supplier of potash products and it has customers in over 60 countries.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Dead Sea products_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 301,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.652",
"value": "?search="
},
{
"operation": "search",
"time": "9.358",
"value": "Red Sea Works"
},
{
"operation": "Now on",
"time": "11.392",
"value": "undefined"
},
{
"operation": "search",
"time": "35.476",
"value": "potash products"
},
{
"operation": "Now on",
"time": "36.825",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "45.416",
"value": "Dead Sea products"
},
{
"operation": "Now on",
"time": "45.996",
"value": "Dead Sea products"
},
{
"operation": "Highlighting",
"time": "61.976",
"value": "Dead Sea products_sentence_3"
},
{
"operation": "finish",
"time": "84.392",
"value": "finish"
}
]
}
|
exfever_797
|
In Dubious Battle is a 2016 drama film starring Franco and an American singer who rose to prominence as Alex Russo in the television series Wizards of Waverly Place on an American pay television channel that serves as the flagship property of Disney Branded Television.
|
{
"evidence": [
"In Dubious Battle (film)",
"Selena Gomez",
"Disney Channel"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn Dubious Battle is a 2016 drama film starring Franco and an American singer who rose to prominence as Alex Russo in the television series Wizards of Waverly Place on an American pay television channel that serves as the flagship property of Disney Branded Television.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"In Dubious Battle (film)",
"Selena Gomez",
"Disney Channel"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 302,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_355
|
Terminator Salvation is the 16th film released by Columbia Pictures internationally.
|
{
"evidence": [
"Terminator Salvation",
"John Brancato and Michael Ferris"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTerminator Salvation is the 16th film released by Columbia Pictures internationally.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Terminator Salvation",
"John Brancato and Michael Ferris"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 303,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
41970
|
Winter Saloon also known as Harm's Bar is a historic structure that has yet to be placed on the National Register of Historic Places.
|
{
"evidence": [
"Winter Saloon_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWinter Saloon also known as Harm's Bar is a historic structure that has yet to be placed on the National Register of Historic Places.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Winter Saloon_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 304,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.168",
"value": "?search="
},
{
"operation": "search",
"time": "9.167",
"value": "Winter Saloon"
},
{
"operation": "Now on",
"time": "9.855",
"value": "Winter Saloon"
},
{
"operation": "Highlighting",
"time": "12.218",
"value": "Winter Saloon_sentence_0"
},
{
"operation": "Highlighting",
"time": "13.058",
"value": "Winter Saloon_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "13.805",
"value": "Winter Saloon_sentence_0"
},
{
"operation": "finish",
"time": "29.313",
"value": "finish"
}
]
}
|
972
|
The Bucharest Metro Line M5 (opened on 15 September 2020) construction was started on the first section of the M5 and the expected construction cost was €708.6 million.
|
{
"evidence": [
"Bucharest Metro Line M5_sentence_0",
"Bucharest Metro Line M5_sentence_2",
"Bucharest Metro Line M5_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Bucharest Metro Line M5 (opened on 15 September 2020) construction was started on the first section of the M5 and the expected construction cost was €708.6 million.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Bucharest Metro Line M5_sentence_0",
"Bucharest Metro Line M5_sentence_2",
"Bucharest Metro Line M5_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 305,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "29.843",
"value": "Bucharest Metro Line M5"
},
{
"operation": "Highlighting",
"time": "36.942",
"value": "Bucharest Metro Line M5_sentence_0"
},
{
"operation": "Highlighting",
"time": "55.116",
"value": "Bucharest Metro Line M5_sentence_2"
},
{
"operation": "Highlighting",
"time": "55.881",
"value": "Bucharest Metro Line M5_sentence_3"
},
{
"operation": "finish",
"time": "92.028",
"value": "finish"
}
]
}
|
exfever_8520
|
Curtis James Jackson III is an American rapper who signed to an American record label founded by Paul Rosenberg and a rapper who is credited with popularizing hip hop in middle America and is critically acclaimed as one of the greatest rappers of all time.
|
{
"evidence": [
"50 Cent",
"Shady Records",
"Eminem"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCurtis James Jackson III is an American rapper who signed to an American record label founded by Paul Rosenberg and a rapper who is credited with popularizing hip hop in middle America and is critically acclaimed as one of the greatest rappers of all time.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"50 Cent",
"Shady Records",
"Eminem"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 306,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_3954
|
SoCal is a geographic and cultural region that generally comprises the southern portion of a state in the Southern United States.
|
{
"evidence": [
"Southern California",
"California"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSoCal is a geographic and cultural region that generally comprises the southern portion of a state in the Southern United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Southern California",
"California"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 307,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
62288
|
During the second half of the century, history painters combined a foreign influence from Rubens with knowledge of classicism and Italian Baroque qualities; then later in the century, many painters turned to Anthony van Dyck as a major influence.
|
{
"evidence": [
"Flemish Baroque painting_sentence_27"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDuring the second half of the century, history painters combined a foreign influence from Rubens with knowledge of classicism and Italian Baroque qualities; then later in the century, many painters turned to Anthony van Dyck as a major influence.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Flemish Baroque painting_sentence_27"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 308,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.143",
"value": "?search="
},
{
"operation": "search",
"time": "134.041",
"value": "\"During the second half of the century, history painters combined a foreign influence from Rubens with knowledge of classicism and Italian Baroque qualities\""
},
{
"operation": "Now on",
"time": "135.469",
"value": "undefined"
},
{
"operation": "search",
"time": "162.543",
"value": "\"influence from Rubens with knowledge of classicism and Italian Baroque qualities\""
},
{
"operation": "Now on",
"time": "163.172",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "164.676",
"value": "Flemish Baroque painting"
},
{
"operation": "Now on",
"time": "165.355",
"value": "Flemish Baroque painting"
},
{
"operation": "Page search",
"time": "169.681",
"value": "influence from Rubens with knowledge of classicism and Italian Baroque qualities"
},
{
"operation": "Highlighting",
"time": "173.536",
"value": "Flemish Baroque painting_sentence_27"
},
{
"operation": "finish",
"time": "196.378",
"value": "finish"
}
]
}
|
exfever_2401
|
Coldplay is a British rock band whose members include an English singer-songwriter who was a co-founder of the rock band Coldplay.
|
{
"evidence": [
"Coldplay",
"Chris Martin"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nColdplay is a British rock band whose members include an English singer-songwriter who was a co-founder of the rock band Coldplay.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Coldplay",
"Chris Martin"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 309,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
23396
|
Odisha FC Reserves and Academy was first headed by Francisco Perez Lazaro as coach and the 2017–18 I-League 2nd Division was their first season in the league.
|
{
"evidence": [
"Odisha FC Reserves and Academy_sentence_4",
"Odisha FC Reserves and Academy_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOdisha FC Reserves and Academy was first headed by Francisco Perez Lazaro as coach and the 2017–18 I-League 2nd Division was their first season in the league.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Odisha FC Reserves and Academy_sentence_4",
"Odisha FC Reserves and Academy_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 310,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.041",
"value": "?search="
},
{
"operation": "search",
"time": "11.823",
"value": "Odisha FC Reserves and Academy"
},
{
"operation": "Now on",
"time": "12.778",
"value": "Odisha FC Reserves and Academy"
},
{
"operation": "Page search",
"time": "23.329",
"value": "perez"
},
{
"operation": "Highlighting",
"time": "39.385",
"value": "Odisha FC Reserves and Academy_sentence_4"
},
{
"operation": "Highlighting",
"time": "54.877",
"value": "Odisha FC Reserves and Academy_sentence_5"
},
{
"operation": "finish",
"time": "70.587",
"value": "finish"
}
]
}
|
exfever_4569
|
The Mughal Empire stretched from the outer fringes of a country in South Asia.
|
{
"evidence": [
"Mughal Empire",
"Bangladesh"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Mughal Empire stretched from the outer fringes of a country in South Asia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mughal Empire",
"Bangladesh"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 311,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4770
|
Jet Li is a Chinese martial artist and current Wushu champion who has been a naturalized citizen of a sovereign island city-state in maritime Southeast Asia.
|
{
"evidence": [
"Jet Li",
"Singapore"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJet Li is a Chinese martial artist and current Wushu champion who has been a naturalized citizen of a sovereign island city-state in maritime Southeast Asia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jet Li",
"Singapore"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 312,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4503
|
One World Trade Center is the new skyscraper in the United States that has the same name as the North Tower of the large complex in the Financial District of Lower Manhattan designed by Minoru Yamasaki.
|
{
"evidence": [
"One World Trade Center",
"World Trade Center (1973–2001)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOne World Trade Center is the new skyscraper in the United States that has the same name as the North Tower of the large complex in the Financial District of Lower Manhattan designed by Minoru Yamasaki.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"One World Trade Center",
"World Trade Center (1973–2001)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 313,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
11704
|
Equinox Group owns Pure Yoga and politically controversial SoulCycle, both introduced to Japan in 2008.
|
{
"evidence": [
"SoulCycle_sentence_0",
"SoulCycle_sentence_1",
"SoulCycle_sentence_48",
"SoulCycle_sentence_49",
"Equinox Group_sentence_0",
"Equinox Group_sentence_9"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEquinox Group owns Pure Yoga and politically controversial SoulCycle, both introduced to Japan in 2008.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"SoulCycle_sentence_0",
"SoulCycle_sentence_1",
"SoulCycle_sentence_48",
"SoulCycle_sentence_49",
"Equinox Group_sentence_0",
"Equinox Group_sentence_9"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 314,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.09",
"value": "?search="
},
{
"operation": "search",
"time": "12.889",
"value": "equinox group"
},
{
"operation": "Now on",
"time": "13.949",
"value": "Equinox Group"
},
{
"operation": "Highlighting",
"time": "16.97",
"value": "Equinox Group_sentence_0"
},
{
"operation": "Highlighting",
"time": "33.266",
"value": "Equinox Group_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "39.612",
"value": "Equinox Group_sentence_9"
},
{
"operation": "Highlighting",
"time": "44.959",
"value": "Equinox Group_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "46.684",
"value": "Equinox Group_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "46.696",
"value": "Equinox Group_header_cell_0_7_0"
},
{
"operation": "hyperlink",
"time": "57.259",
"value": "SoulCycle"
},
{
"operation": "Now on",
"time": "58.357",
"value": "SoulCycle"
},
{
"operation": "Highlighting",
"time": "63.43",
"value": "SoulCycle_sentence_0"
},
{
"operation": "Highlighting",
"time": "67.758",
"value": "SoulCycle_sentence_1"
},
{
"operation": "Highlighting",
"time": "170.238",
"value": "SoulCycle_sentence_48"
},
{
"operation": "Highlighting",
"time": "171.106",
"value": "SoulCycle_sentence_49"
},
{
"operation": "hyperlink",
"time": "202.96",
"value": "Equinox Group"
},
{
"operation": "Now on",
"time": "203.802",
"value": "Equinox Group"
},
{
"operation": "Highlighting",
"time": "205.873",
"value": "Equinox Group_sentence_0"
},
{
"operation": "Highlighting",
"time": "215.776",
"value": "Equinox Group_sentence_9"
},
{
"operation": "finish",
"time": "242.183",
"value": "finish"
}
]
}
|
69632
|
The United States Congress ignored the articles in the Saturday Review of Literature.
|
{
"evidence": [
"Saturday Review (U.S. magazine)_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe United States Congress ignored the articles in the Saturday Review of Literature.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Saturday Review (U.S. magazine)_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 315,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "20.818",
"value": "?search="
},
{
"operation": "search",
"time": "37.087",
"value": "Saturday Review of Literature"
},
{
"operation": "Now on",
"time": "50.55",
"value": "Saturday Review (U.S. magazine)"
},
{
"operation": "Highlighting",
"time": "82.699",
"value": "Saturday Review (U.S. magazine)_sentence_1"
},
{
"operation": "finish",
"time": "89.85",
"value": "finish"
}
]
}
|
14765
|
Joe Robbie (principal founder of the Miami Dolphins) was a lawyer born July 7, 1916 and died in January 7, 2000 with one spouse.
|
{
"evidence": [
"Joe Robbie_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJoe Robbie (principal founder of the Miami Dolphins) was a lawyer born July 7, 1916 and died in January 7, 2000 with one spouse.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Joe Robbie_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 316,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.147",
"value": "?search="
},
{
"operation": "search",
"time": "12.07",
"value": "Joe Robbie"
},
{
"operation": "Now on",
"time": "12.991",
"value": "Joe Robbie"
},
{
"operation": "Highlighting",
"time": "17.52",
"value": "Joe Robbie_sentence_0"
},
{
"operation": "finish",
"time": "43.275",
"value": "finish"
}
]
}
|
exfever_1444
|
Where the Wild Things Are is a 2009 fantasy adventure drama film that combines live-action, performers in costumes, and a modern variant of the automaton and is often used for the portrayal of characters in films and theme park attractions.
|
{
"evidence": [
"Where the Wild Things Are (film)",
"Animatronics"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWhere the Wild Things Are is a 2009 fantasy adventure drama film that combines live-action, performers in costumes, and a modern variant of the automaton and is often used for the portrayal of characters in films and theme park attractions.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Where the Wild Things Are (film)",
"Animatronics"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 317,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
89077
|
Born 16 September 1969 in Bangalore, Karnataka, Pramila Bhatt won five out of seven games during her time as an ODI cricketer captain.
|
{
"evidence": [
"Pramila Bhatt_sentence_0",
"Pramila Bhatt_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBorn 16 September 1969 in Bangalore, Karnataka, Pramila Bhatt won five out of seven games during her time as an ODI cricketer captain.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Pramila Bhatt_sentence_0",
"Pramila Bhatt_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 318,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.145",
"value": "?search="
},
{
"operation": "search",
"time": "19.179",
"value": "Pramila Bhatt"
},
{
"operation": "Now on",
"time": "20.076",
"value": "Pramila Bhatt"
},
{
"operation": "Highlighting",
"time": "173.475",
"value": "Pramila Bhatt_sentence_0"
},
{
"operation": "Highlighting",
"time": "191.176",
"value": "Pramila Bhatt_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "191.927",
"value": "Pramila Bhatt_sentence_4"
},
{
"operation": "Highlighting",
"time": "192.792",
"value": "Pramila Bhatt_sentence_2"
},
{
"operation": "finish",
"time": "211.564",
"value": "finish"
}
]
}
|
88219
|
The 2004 Lumberjacks, led by fifth-year head coach Doug Adkins, played home games at the Redwood Bowl in Arcata, California and finished the season with a record of five wins and five losses.
|
{
"evidence": [
"2004 Humboldt State Lumberjacks football team_sentence_0",
"2004 Humboldt State Lumberjacks football team_sentence_2",
"2004 Humboldt State Lumberjacks football team_sentence_3",
"2004 Humboldt State Lumberjacks football team_sentence_4"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 2004 Lumberjacks, led by fifth-year head coach Doug Adkins, played home games at the Redwood Bowl in Arcata, California and finished the season with a record of five wins and five losses.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"2004 Humboldt State Lumberjacks football team_sentence_0",
"2004 Humboldt State Lumberjacks football team_sentence_2",
"2004 Humboldt State Lumberjacks football team_sentence_3",
"2004 Humboldt State Lumberjacks football team_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 319,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.322",
"value": "?search="
},
{
"operation": "search",
"time": "13.276",
"value": "Contains...2004 Lumberjacks"
},
{
"operation": "Now on",
"time": "14.087",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "18.693",
"value": "2004 Humboldt State Lumberjacks football team"
},
{
"operation": "Now on",
"time": "20.015",
"value": "2004 Humboldt State Lumberjacks football team"
},
{
"operation": "Highlighting",
"time": "22.759",
"value": "2004 Humboldt State Lumberjacks football team_sentence_0"
},
{
"operation": "Highlighting",
"time": "31.343",
"value": "2004 Humboldt State Lumberjacks football team_sentence_2"
},
{
"operation": "Highlighting",
"time": "34.64",
"value": "2004 Humboldt State Lumberjacks football team_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "34.645",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "34.645",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "41.42",
"value": "2004 Humboldt State Lumberjacks football team_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "41.423",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "41.424",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "42.796",
"value": "2004 Humboldt State Lumberjacks football team_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "42.799",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "42.799",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "46.131",
"value": "2004 Humboldt State Lumberjacks football team_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "48.574",
"value": "2004 Humboldt State Lumberjacks football team_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "48.577",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "48.578",
"value": "2004 Humboldt State Lumberjacks football team_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "53.666",
"value": "2004 Humboldt State Lumberjacks football team_sentence_4"
},
{
"operation": "finish",
"time": "62.209",
"value": "finish"
}
]
}
|
exfever_2009
|
Cornell was forced to undergo the emergency surgery after a routine examination due to years of abuse, his larynx had virtually disappeared, as had the surrounding vocal cords.
|
{
"evidence": [
"Temple of the Dog",
"Chris Cornell",
"Soundgarden"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCornell was forced to undergo the emergency surgery after a routine examination due to years of abuse, his larynx had virtually disappeared, as had the surrounding vocal cords.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Temple of the Dog",
"Chris Cornell",
"Soundgarden"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 320,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_8734
|
Samsung Life Insurance is a South Korean multinational insurance company and a subsidiary of a multinational manufacturer based in South America.
|
{
"evidence": [
"Samsung Life Insurance",
"Samsung"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSamsung Life Insurance is a South Korean multinational insurance company and a subsidiary of a multinational manufacturer based in South America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Samsung Life Insurance",
"Samsung"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 321,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4478
|
Paul William Walker IV was an American singer known for his role as a fictional character and one of the main protagonists of the Fast & Furious franchise.
|
{
"evidence": [
"Paul Walker",
"Brian O'Conner"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPaul William Walker IV was an American singer known for his role as a fictional character and one of the main protagonists of the Fast & Furious franchise.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Walker",
"Brian O'Conner"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 322,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
49166
|
Following a renovation, the passenger building of Sondrio railway station, currently managed by Rete Ferroviaria Italiana (the owner of Italy's railway network), consists of the following public spaces: ticketing, bar, kiosk and waiting room, while the station's building containing the public conveniences has a now a different style.
|
{
"evidence": [
"Sondrio railway station_sentence_2",
"Sondrio railway station_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFollowing a renovation, the passenger building of Sondrio railway station, currently managed by Rete Ferroviaria Italiana (the owner of Italy's railway network), consists of the following public spaces: ticketing, bar, kiosk and waiting room, while the station's building containing the public conveniences has a now a different style.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sondrio railway station_sentence_2",
"Sondrio railway station_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 323,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.873",
"value": "?search="
},
{
"operation": "search",
"time": "12.356",
"value": "Sondrio railway station"
},
{
"operation": "Now on",
"time": "13.312",
"value": "Sondrio railway station"
},
{
"operation": "Highlighting",
"time": "23.505",
"value": "Sondrio railway station_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "23.508",
"value": "Sondrio railway station_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "23.508",
"value": "Sondrio railway station_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "37.038",
"value": "Sondrio railway station_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "40.245",
"value": "Sondrio railway station_cell_0_3_1"
},
{
"operation": "Highlighting deleted",
"time": "40.248",
"value": "Sondrio railway station_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "40.249",
"value": "Sondrio railway station_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "51.101",
"value": "Sondrio railway station_sentence_3"
},
{
"operation": "Highlighting",
"time": "52.654",
"value": "Sondrio railway station_sentence_2"
},
{
"operation": "Highlighting",
"time": "53.475",
"value": "Sondrio railway station_sentence_3"
},
{
"operation": "finish",
"time": "72.319",
"value": "finish"
}
]
}
|
exfever_2086
|
The electric chair is a method of execution originating in the United States that has also been used in several countries, including an archipelagic country in Southeast Asia.?
|
{
"evidence": [
"Electric chair",
"Philippines"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe electric chair is a method of execution originating in the United States that has also been used in several countries, including an archipelagic country in Southeast Asia.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Electric chair",
"Philippines"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 324,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
10338
|
Andrzej Seweryn is a Polish director and actor who only appears in local Polish movies.
|
{
"evidence": [
"Andrzej Seweryn_sentence_2"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAndrzej Seweryn is a Polish director and actor who only appears in local Polish movies.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Andrzej Seweryn_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 325,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.604",
"value": "?search="
},
{
"operation": "search",
"time": "10.492",
"value": "Andrzej Seweryn"
},
{
"operation": "Now on",
"time": "11.261",
"value": "Andrzej Seweryn"
},
{
"operation": "Highlighting",
"time": "14.707",
"value": "Andrzej Seweryn_sentence_1"
},
{
"operation": "Highlighting",
"time": "77.428",
"value": "Andrzej Seweryn_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "83.17",
"value": "Andrzej Seweryn_sentence_1"
},
{
"operation": "finish",
"time": "97.7",
"value": "finish"
}
]
}
|
34850
|
The soundtrack album by Konami Kukeiha Club was released on November 01, 1999 worldwide, and it comes with a single disc with a total length of 30:15.
|
{
"evidence": [
"Kid Dracula (1990 video game)_sentence_11"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe soundtrack album by Konami Kukeiha Club was released on November 01, 1999 worldwide, and it comes with a single disc with a total length of 30:15.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Kid Dracula (1990 video game)_sentence_11"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 326,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.648",
"value": "?search="
},
{
"operation": "search",
"time": "15.968",
"value": "Konami Kukeiha Club"
},
{
"operation": "Now on",
"time": "17.032",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "62.571",
"value": "Michiru Yamane"
},
{
"operation": "Now on",
"time": "63.556",
"value": "Michiru Yamane"
},
{
"operation": "Page search",
"time": "68.984",
"value": "The soundtrack album by Konami Kukeiha Club was released on November 01, 1999 worldwide, and it comes with a single disc with a total length of 30"
},
{
"operation": "Page search",
"time": "79.817",
"value": "Konami Kukeiha Club"
},
{
"operation": "back-button-clicked",
"time": "133.945",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "134.703",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "137.601",
"value": "Kenichiro Fukui"
},
{
"operation": "Now on",
"time": "138.126",
"value": "Kenichiro Fukui"
},
{
"operation": "Highlighting",
"time": "151.392",
"value": "Kenichiro Fukui_item_7_21"
},
{
"operation": "Highlighting",
"time": "151.394",
"value": "Kenichiro Fukui_title"
},
{
"operation": "Highlighting",
"time": "151.394",
"value": "Kenichiro Fukui_section_1"
},
{
"operation": "Highlighting",
"time": "151.395",
"value": "Kenichiro Fukui_section_3"
},
{
"operation": "Highlighting deleted",
"time": "151.546",
"value": "Kenichiro Fukui_item_7_21"
},
{
"operation": "Highlighting deleted",
"time": "151.549",
"value": "Kenichiro Fukui_title"
},
{
"operation": "Highlighting deleted",
"time": "151.55",
"value": "Kenichiro Fukui_section_1"
},
{
"operation": "Highlighting deleted",
"time": "151.551",
"value": "Kenichiro Fukui_section_3"
},
{
"operation": "Highlighting",
"time": "152.099",
"value": "Kenichiro Fukui_item_7_21"
},
{
"operation": "Highlighting",
"time": "152.1",
"value": "Kenichiro Fukui_title"
},
{
"operation": "Highlighting",
"time": "152.101",
"value": "Kenichiro Fukui_section_1"
},
{
"operation": "Highlighting",
"time": "152.101",
"value": "Kenichiro Fukui_section_3"
},
{
"operation": "Highlighting deleted",
"time": "152.95",
"value": "Kenichiro Fukui_item_7_21"
},
{
"operation": "Highlighting deleted",
"time": "152.953",
"value": "Kenichiro Fukui_title"
},
{
"operation": "Highlighting deleted",
"time": "152.955",
"value": "Kenichiro Fukui_section_1"
},
{
"operation": "Highlighting deleted",
"time": "152.956",
"value": "Kenichiro Fukui_section_3"
},
{
"operation": "search",
"time": "161.456",
"value": "Final Fantasy VII Advent Children Original Soundtrack"
},
{
"operation": "Now on",
"time": "162.621",
"value": "Final Fantasy VII Advent Children Original Soundtrack"
},
{
"operation": "Highlighting",
"time": "163.296",
"value": "Music of the Final Fantasy VII series_sentence_2"
},
{
"operation": "Highlighting",
"time": "163.297",
"value": "Music of the Final Fantasy VII series_title"
},
{
"operation": "Highlighting deleted",
"time": "165.134",
"value": "Music of the Final Fantasy VII series_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "165.136",
"value": "Music of the Final Fantasy VII series_title"
},
{
"operation": "Page search",
"time": "175.657",
"value": "Konami Kukeiha Club"
},
{
"operation": "search",
"time": "219.104",
"value": "Suikoden"
},
{
"operation": "Now on",
"time": "220.176",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "229.346",
"value": "Suikoden (video game)"
},
{
"operation": "Now on",
"time": "230.996",
"value": "Suikoden (video game)"
},
{
"operation": "Page search",
"time": "235.094",
"value": "sound"
},
{
"operation": "search",
"time": "399.483",
"value": "Contains...Konami Kukeiha Club"
},
{
"operation": "Now on",
"time": "399.978",
"value": "undefined"
},
{
"operation": "search",
"time": "416.787",
"value": "Konami Kukeiha Club"
},
{
"operation": "Now on",
"time": "417.226",
"value": "undefined"
},
{
"operation": "search",
"time": "431.955",
"value": "Konami"
},
{
"operation": "Now on",
"time": "433.469",
"value": "Konami"
},
{
"operation": "search",
"time": "442.821",
"value": "konami club"
},
{
"operation": "Now on",
"time": "443.396",
"value": "undefined"
},
{
"operation": "search",
"time": "480.133",
"value": "Konami Square Wave Club"
},
{
"operation": "Now on",
"time": "481.836",
"value": "undefined"
},
{
"operation": "search",
"time": "490.526",
"value": "Contains...\"Konami Square Wave Club\""
},
{
"operation": "Now on",
"time": "490.919",
"value": "undefined"
},
{
"operation": "search",
"time": "551.209",
"value": "Contains...Kukeiha Club"
},
{
"operation": "Now on",
"time": "552.091",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "556.604",
"value": "Kenichiro Fukui"
},
{
"operation": "Now on",
"time": "556.818",
"value": "Kenichiro Fukui"
},
{
"operation": "hyperlink",
"time": "563.178",
"value": "Konami"
},
{
"operation": "Now on",
"time": "563.537",
"value": "Konami"
},
{
"operation": "Page search",
"time": "568.974",
"value": "club"
},
{
"operation": "back-button-clicked",
"time": "585.54",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "585.607",
"value": "Kenichiro Fukui"
},
{
"operation": "back-button-clicked",
"time": "586.506",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "586.509",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "587.11",
"value": "undefined"
},
{
"operation": "search",
"time": "596.932",
"value": "Contains...Konami Kukeiha Club"
},
{
"operation": "Now on",
"time": "597.512",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "600.134",
"value": "Kid Dracula (1990 video game)"
},
{
"operation": "Now on",
"time": "600.687",
"value": "Kid Dracula (1990 video game)"
},
{
"operation": "Highlighting",
"time": "614.551",
"value": "Kid Dracula (1990 video game)_sentence_11"
},
{
"operation": "back-button-clicked",
"time": "632.953",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "633.619",
"value": "undefined"
},
{
"operation": "finish",
"time": "648.171",
"value": "finish"
}
]
}
|
58542
|
Dimitrije Mitrinović's father ran an experimental farm and was not a part of the Austro-Hungarian government.
|
{
"evidence": [
"Dimitrije Mitrinović_sentence_2"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDimitrije Mitrinović's father ran an experimental farm and was not a part of the Austro-Hungarian government.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dimitrije Mitrinović_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 327,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.209",
"value": "?search="
},
{
"operation": "search",
"time": "4.365",
"value": "Dimitrije Mitrinović"
},
{
"operation": "Now on",
"time": "5.272",
"value": "Dimitrije Mitrinović"
},
{
"operation": "Highlighting",
"time": "8.793",
"value": "Dimitrije Mitrinović_sentence_2"
},
{
"operation": "finish",
"time": "18.009",
"value": "finish"
}
]
}
|
exfever_41
|
Stranger than Fiction is a 2001 American fantasy comedy-drama film written by an American screenwriter who also wrote the internationally-performed stage play Good Canary.
|
{
"evidence": [
"Stranger than Fiction (2006 film)",
"Zach Helm"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nStranger than Fiction is a 2001 American fantasy comedy-drama film written by an American screenwriter who also wrote the internationally-performed stage play Good Canary.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stranger than Fiction (2006 film)",
"Zach Helm"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 328,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_278
|
Demi Lovato's second guest appearance on a studio album is Nick Jonas.
|
{
"evidence": [
"Nick Jonas (album)",
"Island Records",
"Independent record label"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDemi Lovato's second guest appearance on a studio album is Nick Jonas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nick Jonas (album)",
"Island Records",
"Independent record label"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 329,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_910
|
St. Francois Mountains of Missouri is the third highest point in America.
|
{
"evidence": [
"Ozarks",
"United States"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSt. Francois Mountains of Missouri is the third highest point in America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ozarks",
"United States"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 330,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_5282
|
The lead actors of the film Mrigayaa were Mithun Chakraborty and an African actress known for work in Tollywood.
|
{
"evidence": [
"Mrigayaa",
"Mamata Shankar",
"Cinema of West Bengal"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe lead actors of the film Mrigayaa were Mithun Chakraborty and an African actress known for work in Tollywood.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mrigayaa",
"Mamata Shankar",
"Cinema of West Bengal"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 331,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
39321
|
Arthur Dooley was enlisted in the Irish Guards and became a piper in the regiment's band, then later joined the Palestine Liberation Organization, an organization founded in 1964 to the "liberation of Palestine" through armed struggle, with much of its violence aimed at Israeli civilians.
|
{
"evidence": [
"Arthur Dooley_sentence_2",
"Arthur Dooley_sentence_3",
"Palestine Liberation Organization_sentence_2"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nArthur Dooley was enlisted in the Irish Guards and became a piper in the regiment's band, then later joined the Palestine Liberation Organization, an organization founded in 1964 to the \"liberation of Palestine\" through armed struggle, with much of its violence aimed at Israeli civilians.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Arthur Dooley_sentence_2",
"Arthur Dooley_sentence_3",
"Palestine Liberation Organization_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 332,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.627",
"value": "undefined"
},
{
"operation": "search",
"time": "37.208",
"value": "Lost Killers"
},
{
"operation": "Now on",
"time": "38.847",
"value": "Lost Killers"
},
{
"operation": "search",
"time": "47.932",
"value": "Arthur Dooley"
},
{
"operation": "Now on",
"time": "49.221",
"value": "Arthur Dooley"
},
{
"operation": "Highlighting",
"time": "56.712",
"value": "Arthur Dooley_sentence_2"
},
{
"operation": "Highlighting",
"time": "62.395",
"value": "Arthur Dooley_sentence_3"
},
{
"operation": "hyperlink",
"time": "64.572",
"value": "Palestine Liberation Organization"
},
{
"operation": "Now on",
"time": "67.85",
"value": "Palestine Liberation Organization"
},
{
"operation": "Highlighting",
"time": "67.939",
"value": "Palestine Liberation Organization_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "69.628",
"value": "Palestine Liberation Organization_sentence_1"
},
{
"operation": "Highlighting",
"time": "70.443",
"value": "Palestine Liberation Organization_sentence_2"
},
{
"operation": "back-button-clicked",
"time": "75.965",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "76.035",
"value": "Arthur Dooley"
},
{
"operation": "finish",
"time": "92.971",
"value": "finish"
}
]
}
|
29068
|
Michael Vlamis, an American actor and screenwriter, played "Hipster Jerkwad" on a 2014 episode of New Girl.
|
{
"evidence": [
"Michael Vlamis_sentence_0",
"Michael Vlamis_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMichael Vlamis, an American actor and screenwriter, played \"Hipster Jerkwad\" on a 2014 episode of New Girl.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Michael Vlamis_sentence_0",
"Michael Vlamis_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 333,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.352",
"value": "?search="
},
{
"operation": "search",
"time": "15.717",
"value": "Michael Vlamis"
},
{
"operation": "Now on",
"time": "17.04",
"value": "Michael Vlamis"
},
{
"operation": "Highlighting",
"time": "21.504",
"value": "Michael Vlamis_sentence_0"
},
{
"operation": "Highlighting",
"time": "26.662",
"value": "Michael Vlamis_sentence_1"
},
{
"operation": "hyperlink",
"time": "28.001",
"value": "New Girl"
},
{
"operation": "Now on",
"time": "29.168",
"value": "New Girl"
},
{
"operation": "back-button-clicked",
"time": "86.658",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "86.748",
"value": "Michael Vlamis"
},
{
"operation": "hyperlink",
"time": "99.93",
"value": "New Girl"
},
{
"operation": "Now on",
"time": "100.486",
"value": "New Girl"
},
{
"operation": "Page search",
"time": "128.05",
"value": "Michael Vlamis"
},
{
"operation": "Page search",
"time": "133.103",
"value": "Michael Vlamis"
},
{
"operation": "search",
"time": "136.733",
"value": ""
},
{
"operation": "Now on",
"time": "137.23",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "143.507",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "143.74",
"value": "New Girl"
},
{
"operation": "Highlighting",
"time": "170.797",
"value": "New Girl_cell_1_2_3"
},
{
"operation": "Highlighting",
"time": "170.798",
"value": "New Girl_title"
},
{
"operation": "Highlighting",
"time": "170.799",
"value": "New Girl_section_9"
},
{
"operation": "Highlighting",
"time": "170.799",
"value": "New Girl_section_10"
},
{
"operation": "Highlighting",
"time": "170.802",
"value": "New Girl_header_cell_1_2_0"
},
{
"operation": "Highlighting",
"time": "170.802",
"value": "New Girl_header_cell_1_1_0"
},
{
"operation": "Highlighting",
"time": "170.803",
"value": "New Girl_header_cell_1_0_3"
},
{
"operation": "Highlighting deleted",
"time": "177.986",
"value": "New Girl_cell_1_2_3"
},
{
"operation": "Highlighting deleted",
"time": "177.987",
"value": "New Girl_title"
},
{
"operation": "Highlighting deleted",
"time": "177.988",
"value": "New Girl_section_9"
},
{
"operation": "Highlighting deleted",
"time": "177.99",
"value": "New Girl_section_10"
},
{
"operation": "Highlighting deleted",
"time": "177.994",
"value": "New Girl_header_cell_1_2_0"
},
{
"operation": "Highlighting deleted",
"time": "177.995",
"value": "New Girl_header_cell_1_1_0"
},
{
"operation": "Highlighting deleted",
"time": "177.995",
"value": "New Girl_header_cell_1_0_3"
},
{
"operation": "finish",
"time": "223.406",
"value": "finish"
}
]
}
|
exfever_6885
|
Eva Longoria is least known as Gabrielle Solis on the ABC television series created by Marc Cherry.
|
{
"evidence": [
"Eva Longoria",
"Desperate Housewives"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEva Longoria is least known as Gabrielle Solis on the ABC television series created by Marc Cherry.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Eva Longoria",
"Desperate Housewives"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 334,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_1308
|
Saif Ali Khan achieved success with his roles in the Indian action comedy film directed by Sameer Malkan.
|
{
"evidence": [
"Saif Ali Khan",
"Main Khiladi Tu Anari"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSaif Ali Khan achieved success with his roles in the Indian action comedy film directed by Sameer Malkan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Saif Ali Khan",
"Main Khiladi Tu Anari"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 335,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_1113
|
Ice hockey is a distance winter team sport played on a frozen body of water and/or hardened chemicals where people can ice skate or play winter sports.
|
{
"evidence": [
"Ice hockey",
"Ice rink"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIce hockey is a distance winter team sport played on a frozen body of water and/or hardened chemicals where people can ice skate or play winter sports.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ice hockey",
"Ice rink"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 336,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
37548
|
Dobson was born neither in the small fishing village of Keiss nor in Caithness, then in the 1890s, Dobson had moved to the Cardiff Docks area where he had become a coal trimmer.
|
{
"evidence": [
"Tom Dobson_sentence_5"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDobson was born neither in the small fishing village of Keiss nor in Caithness, then in the 1890s, Dobson had moved to the Cardiff Docks area where he had become a coal trimmer.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Tom Dobson_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 337,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.21",
"value": "?search="
},
{
"operation": "search",
"time": "13.961",
"value": "Dobson"
},
{
"operation": "Now on",
"time": "15.26",
"value": "Dobson"
},
{
"operation": "search",
"time": "36.5",
"value": "\"Dobson\" \"keiss\""
},
{
"operation": "Now on",
"time": "37.23",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "45.464",
"value": "Tom Dobson"
},
{
"operation": "Now on",
"time": "46.152",
"value": "Tom Dobson"
},
{
"operation": "Highlighting",
"time": "58.145",
"value": "Tom Dobson_sentence_5"
},
{
"operation": "finish",
"time": "92.332",
"value": "finish"
}
]
}
|
32189
|
Having a building that can be attributed to County Commissioner J.M. May, Thorpe Ford Covered Bridge was where he attempted to travel—from Rosedale to Rockville—after his election.
|
{
"evidence": [
"Thorpe Ford Covered Bridge_sentence_5"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHaving a building that can be attributed to County Commissioner J.M. May, Thorpe Ford Covered Bridge was where he attempted to travel—from Rosedale to Rockville—after his election.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Thorpe Ford Covered Bridge_sentence_5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 338,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.038",
"value": "?search="
},
{
"operation": "search",
"time": "50.621",
"value": "J.M. May"
},
{
"operation": "Now on",
"time": "52.814",
"value": "undefined"
},
{
"operation": "search",
"time": "67.524",
"value": "County Commissioner J.M. May"
},
{
"operation": "Now on",
"time": "69.166",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "72.019",
"value": "Thorpe Ford Covered Bridge"
},
{
"operation": "Now on",
"time": "72.595",
"value": "Thorpe Ford Covered Bridge"
},
{
"operation": "Highlighting",
"time": "79.069",
"value": "Thorpe Ford Covered Bridge_sentence_4"
},
{
"operation": "Highlighting",
"time": "161.838",
"value": "Thorpe Ford Covered Bridge_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "176.709",
"value": "Thorpe Ford Covered Bridge_sentence_4"
},
{
"operation": "finish",
"time": "189.853",
"value": "finish"
}
]
}
|
89579
|
The Burlingame Treaty included articles that gave the Chinese the right to free travel in the US and gave the two nations reciprocal access to education.
|
{
"evidence": [
"Burlingame Treaty_sentence_14",
"Burlingame Treaty_sentence_15"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Burlingame Treaty included articles that gave the Chinese the right to free travel in the US and gave the two nations reciprocal access to education.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Burlingame Treaty_sentence_14",
"Burlingame Treaty_sentence_15"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 339,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.444",
"value": "?search="
},
{
"operation": "search",
"time": "15.142",
"value": "Burlingame Treaty"
},
{
"operation": "Now on",
"time": "15.921",
"value": "Burlingame Treaty"
},
{
"operation": "Page search",
"time": "87.779",
"value": "free travel"
},
{
"operation": "Page search",
"time": "125.596",
"value": "education"
},
{
"operation": "Highlighting",
"time": "133.307",
"value": "Burlingame Treaty_sentence_14"
},
{
"operation": "Highlighting",
"time": "135.674",
"value": "Burlingame Treaty_sentence_15"
},
{
"operation": "finish",
"time": "141.406",
"value": "finish"
}
]
}
|
32086
|
William Shield (the son of William Shield) only played the piano.
|
{
"evidence": [
"William Shield_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWilliam Shield (the son of William Shield) only played the piano.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"William Shield_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 340,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.876",
"value": "?search="
},
{
"operation": "search",
"time": "18.889",
"value": "William Shield"
},
{
"operation": "Now on",
"time": "19.848",
"value": "William Shield"
},
{
"operation": "Highlighting",
"time": "24.096",
"value": "William Shield_sentence_1"
},
{
"operation": "finish",
"time": "44.163",
"value": "finish"
}
]
}
|
44695
|
Battle of Red Ford happened in 1294, and resulted in the Victory for Alasdair MacDubhgaill.
|
{
"evidence": [
"Battle of Red Ford_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBattle of Red Ford happened in 1294, and resulted in the Victory for Alasdair MacDubhgaill.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Battle of Red Ford_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 341,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.217",
"value": "?search="
},
{
"operation": "search",
"time": "10.535",
"value": "Batle of the Red Ford"
},
{
"operation": "Now on",
"time": "11.555",
"value": "undefined"
},
{
"operation": "search",
"time": "49.222",
"value": "Contains...battle of the red f*"
},
{
"operation": "Now on",
"time": "51.757",
"value": "undefined"
},
{
"operation": "Page search",
"time": "85.744",
"value": "\"Battle of the red ford\"\""
},
{
"operation": "search",
"time": "89.274",
"value": ""
},
{
"operation": "Now on",
"time": "89.995",
"value": "undefined"
},
{
"operation": "search",
"time": "119.478",
"value": "Alasdair MacDubhgaill"
},
{
"operation": "Now on",
"time": "120.719",
"value": "Alexander of Argyll"
},
{
"operation": "search",
"time": "163.096",
"value": "insource"
},
{
"operation": "Now on",
"time": "167.412",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "169.895",
"value": "Battle of Red Ford"
},
{
"operation": "Now on",
"time": "170.589",
"value": "Battle of Red Ford"
},
{
"operation": "Highlighting",
"time": "180.626",
"value": "Battle of Red Ford_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "180.63",
"value": "Battle of Red Ford_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "180.63",
"value": "Battle of Red Ford_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "195.522",
"value": "Battle of Red Ford_cell_0_2_1"
},
{
"operation": "Highlighting deleted",
"time": "195.527",
"value": "Battle of Red Ford_header_cell_0_2_0"
},
{
"operation": "Highlighting deleted",
"time": "195.528",
"value": "Battle of Red Ford_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "199.401",
"value": "Battle of Red Ford_sentence_0"
},
{
"operation": "finish",
"time": "244.561",
"value": "finish"
}
]
}
|
94511
|
During First Day Hikes, participants must hike with a guide at all time, including while horseback riding and mountain biking.
|
{
"evidence": [
"First Day Hikes_sentence_12",
"First Day Hikes_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDuring First Day Hikes, participants must hike with a guide at all time, including while horseback riding and mountain biking.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"First Day Hikes_sentence_12",
"First Day Hikes_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 342,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.205",
"value": "?search="
},
{
"operation": "search",
"time": "17.974",
"value": "First Day Hikes"
},
{
"operation": "Now on",
"time": "18.785",
"value": "First Day Hikes"
},
{
"operation": "Page search",
"time": "31.659",
"value": "guide"
},
{
"operation": "page-search-reset",
"time": "57.265",
"value": "page-search-reset"
},
{
"operation": "Highlighting",
"time": "68.844",
"value": "First Day Hikes_sentence_12"
},
{
"operation": "Page search",
"time": "86.706",
"value": "with a guide"
},
{
"operation": "Page search",
"time": "89.613",
"value": "with a guide"
},
{
"operation": "Highlighting",
"time": "106.193",
"value": "First Day Hikes_sentence_0"
},
{
"operation": "search",
"time": "199.666",
"value": "Contains...First Day Hikes, participants must hike with a guide at all time"
},
{
"operation": "Now on",
"time": "203.1",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "218.784",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "219.099",
"value": "First Day Hikes"
},
{
"operation": "finish",
"time": "256.55",
"value": "finish"
}
]
}
|
79941
|
José Antonio Romero Morilla was appointed director of the youth setup and returned to the Verdiblancos' reserve team, replacing fired Pepe Puche.
|
{
"evidence": [
"José Antonio Romero Morilla_sentence_6",
"José Antonio Romero Morilla_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJosé Antonio Romero Morilla was appointed director of the youth setup and returned to the Verdiblancos' reserve team, replacing fired Pepe Puche.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"José Antonio Romero Morilla_sentence_6",
"José Antonio Romero Morilla_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 343,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.051",
"value": "?search="
},
{
"operation": "search",
"time": "61.623",
"value": "José Antonio Romero Morilla"
},
{
"operation": "Now on",
"time": "62.662",
"value": "José Antonio Romero Morilla"
},
{
"operation": "Highlighting",
"time": "128.516",
"value": "José Antonio Romero Morilla_sentence_6"
},
{
"operation": "Page search",
"time": "219.042",
"value": "youth"
},
{
"operation": "Highlighting",
"time": "226.324",
"value": "José Antonio Romero Morilla_sentence_4"
},
{
"operation": "Highlighting",
"time": "234.156",
"value": "José Antonio Romero Morilla_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "262.765",
"value": "José Antonio Romero Morilla_sentence_4"
},
{
"operation": "finish",
"time": "362.382",
"value": "finish"
}
]
}
|
76828
|
Rebecca Judd, who is a fashion ambassador for Myer, was criticized for looking anorexic but was defended by Chris Judd for he believed that anorexia should not be ashamed of.
|
{
"evidence": [
"Rebecca Judd_sentence_12",
"Rebecca Judd_sentence_9",
"Rebecca Judd_sentence_10"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRebecca Judd, who is a fashion ambassador for Myer, was criticized for looking anorexic but was defended by Chris Judd for he believed that anorexia should not be ashamed of.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Rebecca Judd_sentence_12",
"Rebecca Judd_sentence_9",
"Rebecca Judd_sentence_10"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 344,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.176",
"value": "?search="
},
{
"operation": "search",
"time": "15.023",
"value": "Rebecca Judd"
},
{
"operation": "Now on",
"time": "15.766",
"value": "Rebecca Judd"
},
{
"operation": "hyperlink",
"time": "35.541",
"value": "a revealing red Tarvydas dress"
},
{
"operation": "Now on",
"time": "36.099",
"value": "Red Tarvydas dress of Rebecca Twigley"
},
{
"operation": "back-button-clicked",
"time": "78.663",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "78.69",
"value": "Rebecca Judd"
},
{
"operation": "Page search",
"time": "85.992",
"value": "anorex"
},
{
"operation": "Page search",
"time": "100.631",
"value": "myer"
},
{
"operation": "Highlighting",
"time": "103.459",
"value": "Rebecca Judd_sentence_12"
},
{
"operation": "Highlighting",
"time": "109.514",
"value": "Rebecca Judd_sentence_9"
},
{
"operation": "Page search",
"time": "133.919",
"value": "same"
},
{
"operation": "Page search",
"time": "136.903",
"value": "sham"
},
{
"operation": "Highlighting",
"time": "142.088",
"value": "Rebecca Judd_sentence_10"
},
{
"operation": "finish",
"time": "150.957",
"value": "finish"
}
]
}
|
exfever_6340
|
Paul Thomas Anderson is an English film director who made his feature-film debut with a 1996 American crime film starring Philip Baker Hall, John C. Reilly, Gwyneth Paltrow and Samuel L. Jackson.
|
{
"evidence": [
"Paul Thomas Anderson",
"Hard Eight (film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPaul Thomas Anderson is an English film director who made his feature-film debut with a 1996 American crime film starring Philip Baker Hall, John C. Reilly, Gwyneth Paltrow and Samuel L. Jackson.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Thomas Anderson",
"Hard Eight (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 345,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
75028
|
The Ländler folk dance was never popular in Vienna despite efforts by Josef Lanner.
|
{
"evidence": [
"Ländler_sentence_9",
"Ländler_sentence_10",
"Ländler_sentence_11",
"Ländler_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Ländler folk dance was never popular in Vienna despite efforts by Josef Lanner.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ländler_sentence_9",
"Ländler_sentence_10",
"Ländler_sentence_11",
"Ländler_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 346,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.751",
"value": "?search="
},
{
"operation": "search",
"time": "15.112",
"value": "Contains...Ländler folk dance"
},
{
"operation": "Now on",
"time": "16.404",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "20.351",
"value": "Landler"
},
{
"operation": "Now on",
"time": "21.381",
"value": "Landler"
},
{
"operation": "hyperlink",
"time": "23.53",
"value": "Ländler"
},
{
"operation": "Now on",
"time": "24.365",
"value": "Ländler"
},
{
"operation": "Page search",
"time": "35.254",
"value": "vienna"
},
{
"operation": "Highlighting",
"time": "85.021",
"value": "Ländler_sentence_9"
},
{
"operation": "Highlighting",
"time": "85.855",
"value": "Ländler_sentence_10"
},
{
"operation": "Highlighting",
"time": "87.37",
"value": "Ländler_sentence_11"
},
{
"operation": "Highlighting deleted",
"time": "88.191",
"value": "Ländler_sentence_11"
},
{
"operation": "Highlighting",
"time": "89.802",
"value": "Ländler_sentence_11"
},
{
"operation": "Highlighting",
"time": "104.468",
"value": "Ländler_sentence_0"
},
{
"operation": "finish",
"time": "107.153",
"value": "finish"
}
]
}
|
exfever_5827
|
30 Seconds to Mars is the 3rd studio album by American rock band Thirty Seconds to Mars that was first released by an American independent record label and imprint label based in Los Angeles, California.
|
{
"evidence": [
"30 Seconds to Mars (album)",
"Immortal Records"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n30 Seconds to Mars is the 3rd studio album by American rock band Thirty Seconds to Mars that was first released by an American independent record label and imprint label based in Los Angeles, California.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"30 Seconds to Mars (album)",
"Immortal Records"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 347,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_8585
|
To Kill a Mockingbird is a 1962 American drama film based on the novel by an American novelist who won the 1961 award for achievements in newspaper, online journalism, literature, and musical composition within the United Kingdom.
|
{
"evidence": [
"To Kill a Mockingbird (film)",
"Harper Lee",
"Pulitzer Prize"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTo Kill a Mockingbird is a 1962 American drama film based on the novel by an American novelist who won the 1961 award for achievements in newspaper, online journalism, literature, and musical composition within the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"To Kill a Mockingbird (film)",
"Harper Lee",
"Pulitzer Prize"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 348,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_6862
|
Asia is Earth's largest and most populous continent, located primarily in the half of Earth that is north of the Equator.
|
{
"evidence": [
"Asia",
"Northern Hemisphere"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAsia is Earth's largest and most populous continent, located primarily in the half of Earth that is north of the Equator.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Asia",
"Northern Hemisphere"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 349,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
91025
|
During his presidency, Rodrigo Duterte signed many laws, including the General Appropriations Act of 2017, 2018, and 2019, whose principal sponsor was Loren Legard.
|
{
"evidence": [
"New People's Army_sentence_8",
"New People's Army_sentence_105"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDuring his presidency, Rodrigo Duterte signed many laws, including the General Appropriations Act of 2017, 2018, and 2019, whose principal sponsor was Loren Legard.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"New People's Army_sentence_8",
"New People's Army_sentence_105"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 350,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.266",
"value": "?search="
},
{
"operation": "search",
"time": "32.497",
"value": "Contains... Rodrigo Duterte"
},
{
"operation": "Now on",
"time": "34.164",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "48.916",
"value": "New People's Army"
},
{
"operation": "Now on",
"time": "50.912",
"value": "New People's Army"
},
{
"operation": "search",
"time": "166.904",
"value": "General Appropriations Act"
},
{
"operation": "Now on",
"time": "170.377",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "232.063",
"value": "Pantaleon Alvarez"
},
{
"operation": "Now on",
"time": "233.218",
"value": "Pantaleon Alvarez"
},
{
"operation": "search",
"time": "301.562",
"value": "President Duterte"
},
{
"operation": "Now on",
"time": "303.116",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "336.756",
"value": "New People's Army"
},
{
"operation": "Now on",
"time": "337.371",
"value": "New People's Army"
},
{
"operation": "Highlighting",
"time": "405.154",
"value": "New People's Army_sentence_8"
},
{
"operation": "Highlighting",
"time": "522.457",
"value": "New People's Army_sentence_105"
},
{
"operation": "search",
"time": "560.574",
"value": "Loren Legard"
},
{
"operation": "Now on",
"time": "562.169",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "582.748",
"value": "loren legarda"
},
{
"operation": "Now on",
"time": "584.319",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "664.859",
"value": "Senate of the Philippines"
},
{
"operation": "Now on",
"time": "667.76",
"value": "Senate of the Philippines"
},
{
"operation": "finish",
"time": "818.845",
"value": "finish"
}
]
}
|
exfever_4396
|
Same-sex marriage is the marriage of two people of the opposite sex where they allow those couples to jointly adopt through judicial decisions that often determine whether they can serve as parents either individually or as a couple.
|
{
"evidence": [
"Same-sex marriage",
"LGBT adoption"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSame-sex marriage is the marriage of two people of the opposite sex where they allow those couples to jointly adopt through judicial decisions that often determine whether they can serve as parents either individually or as a couple.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Same-sex marriage",
"LGBT adoption"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 351,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_232
|
The 1983–84 NBA season ended with the Boston Celtics winning, beating an American professional team in a sport in which two teams compete with the objective of shooting through the defender's hoop.
|
{
"evidence": [
"1983–84 NBA season",
"Los Angeles Lakers",
"Basketball"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 1983–84 NBA season ended with the Boston Celtics winning, beating an American professional team in a sport in which two teams compete with the objective of shooting through the defender's hoop.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"1983–84 NBA season",
"Los Angeles Lakers",
"Basketball"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 352,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
93331
|
Joseph Osgood Barrett, a prominent medium, spiritualist and author, was expelled from the Universalist ministry for his unorthodox beliefs and moved to Madison, Wisconsin in 1872.
|
{
"evidence": [
"Joseph Osgood Barrett_sentence_5"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJoseph Osgood Barrett, a prominent medium, spiritualist and author, was expelled from the Universalist ministry for his unorthodox beliefs and moved to Madison, Wisconsin in 1872.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Joseph Osgood Barrett_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 353,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.169",
"value": "?search="
},
{
"operation": "search",
"time": "15.842",
"value": "Joseph Osgood Barrett"
},
{
"operation": "Now on",
"time": "16.38",
"value": "Joseph Osgood Barrett"
},
{
"operation": "Highlighting",
"time": "20.409",
"value": "Joseph Osgood Barrett_sentence_0"
},
{
"operation": "Highlighting",
"time": "29.769",
"value": "Joseph Osgood Barrett_sentence_4"
},
{
"operation": "Highlighting",
"time": "41.129",
"value": "Joseph Osgood Barrett_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "43.913",
"value": "Joseph Osgood Barrett_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "47.673",
"value": "Joseph Osgood Barrett_sentence_0"
},
{
"operation": "Highlighting",
"time": "131.519",
"value": "Joseph Osgood Barrett_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "131.942",
"value": "Joseph Osgood Barrett_sentence_1"
},
{
"operation": "Highlighting",
"time": "133.19",
"value": "Joseph Osgood Barrett_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "133.598",
"value": "Joseph Osgood Barrett_sentence_1"
},
{
"operation": "finish",
"time": "191.51",
"value": "finish"
}
]
}
|
exfever_6044
|
Quinoa is the largest plant in the amaranth family.
|
{
"evidence": [
"Quinoa",
"Flowering plant"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nQuinoa is the largest plant in the amaranth family.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Quinoa",
"Flowering plant"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 354,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_5075
|
Water is vital for all known forms of life, even though it provides neither food, energy, nor organic micronutrients, a form of matter having a constant chemical composition and characteristic properties used by an organism to survive.
|
{
"evidence": [
"Water",
"Nutrient",
"Chemical substance"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWater is vital for all known forms of life, even though it provides neither food, energy, nor organic micronutrients, a form of matter having a constant chemical composition and characteristic properties used by an organism to survive.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Water",
"Nutrient",
"Chemical substance"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 355,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4653
|
Margaret Constance "Maisie" Williams is an Irish actress who appeared in an American fantasy drama television series created by David Benioff and D. B. Weiss for HBO.
|
{
"evidence": [
"Maisie Williams",
"Game of Thrones"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMargaret Constance \"Maisie\" Williams is an Irish actress who appeared in an American fantasy drama television series created by David Benioff and D. B. Weiss for HBO.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Maisie Williams",
"Game of Thrones"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 356,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_5691
|
Thomas Grant Gustin is best known for his role as Barry Allen, an American superhero television drama series airing on The CW.
|
{
"evidence": [
"Grant Gustin",
"The Flash (2014 TV series)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThomas Grant Gustin is best known for his role as Barry Allen, an American superhero television drama series airing on The CW.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Grant Gustin",
"The Flash (2014 TV series)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 357,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
46977
|
Hurricane Ivan tornado outbreak happened the most in Franklin, Georgia.
|
{
"evidence": [
"Hurricane Ivan_sentence_53",
"Hurricane Ivan_sentence_167"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHurricane Ivan tornado outbreak happened the most in Franklin, Georgia.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hurricane Ivan_sentence_53",
"Hurricane Ivan_sentence_167"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 358,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.735",
"value": "?search="
},
{
"operation": "search",
"time": "10.804",
"value": "Hurricane Ivan"
},
{
"operation": "Now on",
"time": "11.924",
"value": "Hurricane Ivan"
},
{
"operation": "hyperlink",
"time": "51.78",
"value": "3.5.3 Elsewhere in the United States Hurricane Ivan_section_13"
},
{
"operation": "Page search",
"time": "85.71",
"value": "tornado"
},
{
"operation": "Highlighting",
"time": "100.908",
"value": "Hurricane Ivan_sentence_53"
},
{
"operation": "Page search",
"time": "112.821",
"value": "georgia"
},
{
"operation": "hyperlink",
"time": "134.055",
"value": "Georgia"
},
{
"operation": "Now on",
"time": "135.259",
"value": "Georgia (U.S. state)"
},
{
"operation": "back-button-clicked",
"time": "136.66",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "136.768",
"value": "Hurricane Ivan#Elsewhere in the United States Hurricane Ivan section 13"
},
{
"operation": "Page search",
"time": "144.677",
"value": "georgia"
},
{
"operation": "Highlighting",
"time": "178.307",
"value": "Hurricane Ivan_sentence_167"
},
{
"operation": "Page search",
"time": "205.853",
"value": "Franklin, Georgia"
},
{
"operation": "Page search",
"time": "236.516",
"value": "Franklin, Georgia"
},
{
"operation": "search",
"time": "247.162",
"value": "Franklin, Georgia"
},
{
"operation": "Now on",
"time": "248.186",
"value": "Franklin, Georgia"
},
{
"operation": "Page search",
"time": "267.947",
"value": "ivan"
},
{
"operation": "back-button-clicked",
"time": "272.195",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "272.31",
"value": "Hurricane Ivan#Elsewhere in the United States Hurricane Ivan section 13"
},
{
"operation": "finish",
"time": "294.171",
"value": "finish"
}
]
}
|
79748
|
During the episode Odor in the Court, Betty is sassy to her family and becomes addicted to a perfume.
|
{
"evidence": [
"Odor in the Court_sentence_26",
"Odor in the Court_sentence_38"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDuring the episode Odor in the Court, Betty is sassy to her family and becomes addicted to a perfume.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Odor in the Court_sentence_26",
"Odor in the Court_sentence_38"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 359,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.053",
"value": "?search="
},
{
"operation": "search",
"time": "14.582",
"value": "Odor in the Court"
},
{
"operation": "Now on",
"time": "15.353",
"value": "Odor in the Court"
},
{
"operation": "Page search",
"time": "26.263",
"value": "betty"
},
{
"operation": "Page search",
"time": "54.015",
"value": "sassy"
},
{
"operation": "Page search",
"time": "65.058",
"value": "family"
},
{
"operation": "Page search",
"time": "114.153",
"value": "perfume"
},
{
"operation": "Highlighting",
"time": "149.671",
"value": "Odor in the Court_sentence_26"
},
{
"operation": "Page search",
"time": "193.458",
"value": "family"
},
{
"operation": "Highlighting",
"time": "222.454",
"value": "Odor in the Court_sentence_38"
},
{
"operation": "finish",
"time": "292.92",
"value": "finish"
}
]
}
|
52172
|
Oxenbridge, neither an English landowner nor an administrator from Sussex referred to as the Brede Giant and the Sussex Ogre, was said to be invulnerable to metal weapons and only wooden weapons would kill him.
|
{
"evidence": [
"Goddard Oxenbridge_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOxenbridge, neither an English landowner nor an administrator from Sussex referred to as the Brede Giant and the Sussex Ogre, was said to be invulnerable to metal weapons and only wooden weapons would kill him.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Goddard Oxenbridge_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 360,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.941",
"value": "?search="
},
{
"operation": "search",
"time": "25.116",
"value": "Oxenbridge"
},
{
"operation": "Now on",
"time": "25.83",
"value": "Oxenbridge"
},
{
"operation": "hyperlink",
"time": "30.923",
"value": "Goddard Oxenbridge"
},
{
"operation": "Now on",
"time": "31.328",
"value": "Goddard Oxenbridge"
},
{
"operation": "Highlighting",
"time": "37.083",
"value": "Goddard Oxenbridge_sentence_0"
},
{
"operation": "finish",
"time": "114.362",
"value": "finish"
}
]
}
|
exfever_4592
|
Rome was formerly known as the capital city that became famous for its architectural masterpieces, such as a Late Antique place of worship in Istanbul, designed by Isidore of Miletus and Anthemius of Tralles.
|
{
"evidence": [
"Istanbul",
"Constantinople",
"Hagia Sophia"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRome was formerly known as the capital city that became famous for its architectural masterpieces, such as a Late Antique place of worship in Istanbul, designed by Isidore of Miletus and Anthemius of Tralles.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Istanbul",
"Constantinople",
"Hagia Sophia"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 361,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_8593
|
Half of the population of the world has pets.
|
{
"evidence": [
"Pet",
"Animal"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHalf of the population of the world has pets.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pet",
"Animal"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 362,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4978
|
Michael Fassbender is the youngest German-born Irish actor and racing driver.
|
{
"evidence": [
"Michael Fassbender",
"Academy Awards",
"1st Academy Awards"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMichael Fassbender is the youngest German-born Irish actor and racing driver.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Michael Fassbender",
"Academy Awards",
"1st Academy Awards"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 363,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4200
|
Kajal Aggarwal is an actress, model, and also producer.
|
{
"evidence": [
"Kajal Aggarwal",
"Darling (2010 film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nKajal Aggarwal is an actress, model, and also producer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kajal Aggarwal",
"Darling (2010 film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 364,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_3435
|
Sir Roger Moore was the first English actor to be knighted by the Queen.
|
{
"evidence": [
"Roger Moore",
"Live and Let Die (film)",
"Tom Mankiewicz"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSir Roger Moore was the first English actor to be knighted by the Queen.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Roger Moore",
"Live and Let Die (film)",
"Tom Mankiewicz"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 365,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_6112
|
Ginger Baker pioneered a style of music from non-Western countries that includes subgenres like ethnic fusion, which has been popularized by artists including an American musician.?
|
{
"evidence": [
"Ginger Baker",
"World music",
"Ry Cooder"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nGinger Baker pioneered a style of music from non-Western countries that includes subgenres like ethnic fusion, which has been popularized by artists including an American musician.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ginger Baker",
"World music",
"Ry Cooder"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 366,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4130
|
Community is a television sitcom set at a community college in a state in the Mountain West subregion of the United States.
|
{
"evidence": [
"Community (TV series)",
"Colorado"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCommunity is a television sitcom set at a community college in a state in the Mountain West subregion of the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Community (TV series)",
"Colorado"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 367,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
57391
|
The Lairawn Baptist Association originated from the independent Baptist Churches that emerged as early as 1952 because of the increase in population of the Chin people.
|
{
"evidence": [
"Lairawn Baptist Association_sentence_15",
"Lairawn Baptist Association_sentence_13",
"Lairawn Baptist Association_sentence_31"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Lairawn Baptist Association originated from the independent Baptist Churches that emerged as early as 1952 because of the increase in population of the Chin people.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Lairawn Baptist Association_sentence_15",
"Lairawn Baptist Association_sentence_13",
"Lairawn Baptist Association_sentence_31"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 368,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.462",
"value": "?search="
},
{
"operation": "search",
"time": "47.02",
"value": "The Lairawn Baptist Association"
},
{
"operation": "Now on",
"time": "48.071",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "69.353",
"value": "Lairawn Baptist Association"
},
{
"operation": "Now on",
"time": "70.417",
"value": "Lairawn Baptist Association"
},
{
"operation": "Highlighting",
"time": "109.568",
"value": "Lairawn Baptist Association_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "144.284",
"value": "Lairawn Baptist Association_sentence_3"
},
{
"operation": "Highlighting",
"time": "241.356",
"value": "Lairawn Baptist Association_sentence_15"
},
{
"operation": "Highlighting",
"time": "437.593",
"value": "Lairawn Baptist Association_sentence_3"
},
{
"operation": "Highlighting",
"time": "470.507",
"value": "Lairawn Baptist Association_sentence_8"
},
{
"operation": "Highlighting",
"time": "479.506",
"value": "Lairawn Baptist Association_sentence_11"
},
{
"operation": "Highlighting",
"time": "482.245",
"value": "Lairawn Baptist Association_sentence_13"
},
{
"operation": "Highlighting",
"time": "500.573",
"value": "Lairawn Baptist Association_sentence_20"
},
{
"operation": "Highlighting",
"time": "527.825",
"value": "Lairawn Baptist Association_sentence_24"
},
{
"operation": "Highlighting",
"time": "530.086",
"value": "Lairawn Baptist Association_sentence_23"
},
{
"operation": "Highlighting deleted",
"time": "606.628",
"value": "Lairawn Baptist Association_sentence_11"
},
{
"operation": "Highlighting",
"time": "714.483",
"value": "Lairawn Baptist Association_sentence_31"
},
{
"operation": "finish",
"time": "1108.21",
"value": "finish"
}
]
}
|
exfever_343
|
Imagine Dragons is an Asian pop-rock band that released their debut studio album through KIDinaKORNER and Interscope Records.
|
{
"evidence": [
"Imagine Dragons",
"Night Visions"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nImagine Dragons is an Asian pop-rock band that released their debut studio album through KIDinaKORNER and Interscope Records.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Imagine Dragons",
"Night Visions"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 369,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_4643
|
Mediterranean cuisine is the food and methods of preparation used by the people of the Mediterranean Basin which the idea of a Mediterranean cuisine originates with the cookery writer, an American cookery writer who has the influence on British cooking extended to professional as well as domestic cooks, and chefs and restaurateurs of later generations including a British-South African restaurateur, chef, and novelist who is known as the Chancellor of Queen Margaret University, Edinburgh.
|
{
"evidence": [
"Mediterranean cuisine",
"Elizabeth David",
"Prue Leith"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMediterranean cuisine is the food and methods of preparation used by the people of the Mediterranean Basin which the idea of a Mediterranean cuisine originates with the cookery writer, an American cookery writer who has the influence on British cooking extended to professional as well as domestic cooks, and chefs and restaurateurs of later generations including a British-South African restaurateur, chef, and novelist who is known as the Chancellor of Queen Margaret University, Edinburgh.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mediterranean cuisine",
"Elizabeth David",
"Prue Leith"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 370,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_760
|
The Ten Commandments is the first and only American religious drama film.
|
{
"evidence": [
"The Ten Commandments (1956 film)",
"Epic film"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Ten Commandments is the first and only American religious drama film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Ten Commandments (1956 film)",
"Epic film"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 371,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
22037
|
Portuguese Brazilians' history dates back to the 18th century when hundreds of thousand Portuguese left for the Portuguese colony of Brazil during the gold rush.
|
{
"evidence": [
"Portuguese Brazilians_sentence_43"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPortuguese Brazilians' history dates back to the 18th century when hundreds of thousand Portuguese left for the Portuguese colony of Brazil during the gold rush.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Portuguese Brazilians_sentence_43"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 372,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.874",
"value": "?search="
},
{
"operation": "search",
"time": "12.926",
"value": "Portuguese Brazilians' history"
},
{
"operation": "Now on",
"time": "14.925",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "17.747",
"value": "History and overview Brazilians_section_1"
},
{
"operation": "Now on",
"time": "20.23",
"value": "Brazilians#History and overview Brazilians section 1"
},
{
"operation": "hyperlink",
"time": "32.09",
"value": "Portuguese settlers"
},
{
"operation": "Now on",
"time": "33.86",
"value": "Portuguese Brazilians"
},
{
"operation": "Page search",
"time": "66.913",
"value": "gold rush"
},
{
"operation": "Highlighting",
"time": "81.87",
"value": "Portuguese Brazilians_sentence_43"
},
{
"operation": "finish",
"time": "111.136",
"value": "finish"
}
]
}
|
exfever_93
|
Colin Trevorrow was raised in Oakland, California, where he began making short films at the age of 12.
|
{
"evidence": [
"Michael Giacchino",
"Jon Watts",
"Marvel Cinematic Universe"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nColin Trevorrow was raised in Oakland, California, where he began making short films at the age of 12.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Michael Giacchino",
"Jon Watts",
"Marvel Cinematic Universe"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 373,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
63199
|
Newsjack was first broadcast on BBC Radio 7 (a British digital radio station) and had an "Open door" policy on writing and was designed to give new writers an opportunity
|
{
"evidence": [
"Newsjack_sentence_4",
"Newsjack_sentence_5",
"Newsjack_sentence_6",
"BBC Radio 4 Extra_sentence_2",
"BBC Radio 4 Extra_sentence_6"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNewsjack was first broadcast on BBC Radio 7 (a British digital radio station) and had an \"Open door\" policy on writing and was designed to give new writers an opportunity\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Newsjack_sentence_4",
"Newsjack_sentence_5",
"Newsjack_sentence_6",
"BBC Radio 4 Extra_sentence_2",
"BBC Radio 4 Extra_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 374,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.449",
"value": "?search="
},
{
"operation": "search",
"time": "12.632",
"value": "Newsjack"
},
{
"operation": "Now on",
"time": "13.608",
"value": "Newsjack"
},
{
"operation": "Highlighting",
"time": "21.999",
"value": "Newsjack_sentence_4"
},
{
"operation": "Highlighting",
"time": "24.562",
"value": "Newsjack_sentence_5"
},
{
"operation": "Highlighting",
"time": "42.599",
"value": "Newsjack_sentence_6"
},
{
"operation": "hyperlink",
"time": "73.595",
"value": "BBC Radio 7"
},
{
"operation": "Now on",
"time": "74.5",
"value": "BBC Radio 4 Extra"
},
{
"operation": "Highlighting",
"time": "92.035",
"value": "BBC Radio 4 Extra_sentence_2"
},
{
"operation": "Highlighting",
"time": "98.56",
"value": "BBC Radio 4 Extra_sentence_6"
},
{
"operation": "finish",
"time": "117.097",
"value": "finish"
}
]
}
|
exfever_8822
|
Matthew William Sorum is a singer who's also a former member of an American synthpop band fronted by singer-songwriter Tori Amos.
|
{
"evidence": [
"Matt Sorum",
"Y Kant Tori Read"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMatthew William Sorum is a singer who's also a former member of an American synthpop band fronted by singer-songwriter Tori Amos.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Matt Sorum",
"Y Kant Tori Read"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 375,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_2787
|
Venice is the capital of the biggest city on the Adige River in Veneto, India, with 258,031 inhabitants.
|
{
"evidence": [
"Venice",
"Veneto",
"Verona"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVenice is the capital of the biggest city on the Adige River in Veneto, India, with 258,031 inhabitants.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Venice",
"Veneto",
"Verona"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 376,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_3056
|
Black Sabbath was an English rock band formed in a city that is located in the larger West Midlands region in England approximately 100 miles from the capital and largest city of England and the United Kingdom.
|
{
"evidence": [
"Black Sabbath",
"Birmingham",
"London"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBlack Sabbath was an English rock band formed in a city that is located in the larger West Midlands region in England approximately 100 miles from the capital and largest city of England and the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Black Sabbath",
"Birmingham",
"London"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 377,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_1565
|
The Streets of San Francisco is Michael Douglas' fifth television series.
|
{
"evidence": [
"Michael Douglas",
"University of California, Santa Barbara",
"University of California, Los Angeles"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Streets of San Francisco is Michael Douglas' fifth television series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Michael Douglas",
"University of California, Santa Barbara",
"University of California, Los Angeles"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 378,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_467
|
Machu Picchu also has strict entrance rules to reduce the effects of tourism.
|
{
"evidence": [
"Machu Picchu",
"Cordillera Oriental (Peru)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMachu Picchu also has strict entrance rules to reduce the effects of tourism.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Machu Picchu",
"Cordillera Oriental (Peru)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 379,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_566
|
Margaret Constance "Maisie" Williams was an American actress who appeared in television programs, including the British docudrama television film written by Ben Chanan and David Lobatto.
|
{
"evidence": [
"Maisie Williams",
"Cyberbully (2015 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMargaret Constance \"Maisie\" Williams was an American actress who appeared in television programs, including the British docudrama television film written by Ben Chanan and David Lobatto.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Maisie Williams",
"Cyberbully (2015 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 380,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_279
|
India is a federal union comprising 28 states and 8 union territories.
|
{
"evidence": [
"Limbu language",
"India"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIndia is a federal union comprising 28 states and 8 union territories.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Limbu language",
"India"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 381,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_1581
|
Selena Marie Gomez is an English singer, actress, and producer who rose to prominence for her lead role as a fictional character and the protagonist of the Disney Channel sitcom Wizards of Waverly Place.
|
{
"evidence": [
"Selena Gomez",
"Alex Russo"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSelena Marie Gomez is an English singer, actress, and producer who rose to prominence for her lead role as a fictional character and the protagonist of the Disney Channel sitcom Wizards of Waverly Place.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Selena Gomez",
"Alex Russo"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 382,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
exfever_5446
|
Jackie Chan, the founder of the People's Republic of China, asserted that the Chinese people are a nation and promoted cultural and national unity.
|
{
"evidence": [
"Mao Zedong",
"Chinese nationalism"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJackie Chan, the founder of the People's Republic of China, asserted that the Chinese people are a nation and promoted cultural and national unity.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mao Zedong",
"Chinese nationalism"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 383,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
95645
|
Fettiplace (allegedly of Norman descent) included John Fettiplace who was a London draper who possessed the manors of East Shefford and had two sons.
|
{
"evidence": [
"Fettiplace_sentence_0",
"Fettiplace_sentence_10"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFettiplace (allegedly of Norman descent) included John Fettiplace who was a London draper who possessed the manors of East Shefford and had two sons.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Fettiplace_sentence_0",
"Fettiplace_sentence_10"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 384,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.036",
"value": "?search="
},
{
"operation": "search",
"time": "19.195",
"value": "Fettiplace"
},
{
"operation": "Now on",
"time": "20.704",
"value": "Fettiplace"
},
{
"operation": "Highlighting",
"time": "26.328",
"value": "Fettiplace_sentence_0"
},
{
"operation": "Highlighting",
"time": "37.277",
"value": "Fettiplace_sentence_8"
},
{
"operation": "Highlighting",
"time": "44.253",
"value": "Fettiplace_sentence_9"
},
{
"operation": "Highlighting",
"time": "48.11",
"value": "Fettiplace_sentence_10"
},
{
"operation": "Highlighting deleted",
"time": "50.028",
"value": "Fettiplace_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "50.953",
"value": "Fettiplace_sentence_8"
},
{
"operation": "finish",
"time": "60.71",
"value": "finish"
}
]
}
|
exfever_4315
|
South Park is an American animated sitcom that revolves around four boys such as a fictional character and one of the four main characters who debuted on television when South Park first aired on July 13, 1997.
|
{
"evidence": [
"South Park",
"Kenny McCormick"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSouth Park is an American animated sitcom that revolves around four boys such as a fictional character and one of the four main characters who debuted on television when South Park first aired on July 13, 1997.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"South Park",
"Kenny McCormick"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 385,
"split": "test_on"
}
|
{
"annotator_operations": []
}
|
12971
|
Ashton Kutcher, an actor, model, and producer famous for his character in That '70s Show, has taken on the role of producer for productions like Spread and Killers.
|
{
"evidence": [
"Ashton Kutcher_sentence_0",
"Ashton Kutcher_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAshton Kutcher, an actor, model, and producer famous for his character in That '70s Show, has taken on the role of producer for productions like Spread and Killers.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Ashton Kutcher_sentence_0",
"Ashton Kutcher_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 386,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.573",
"value": "?search="
},
{
"operation": "search",
"time": "58.308",
"value": "CURRENT CLAIM"
},
{
"operation": "Now on",
"time": "59.978",
"value": "undefined"
},
{
"operation": "search",
"time": "69.843",
"value": "ashton"
},
{
"operation": "Now on",
"time": "70.708",
"value": "Ashton"
},
{
"operation": "search",
"time": "86.434",
"value": "Ashton Kutcher"
},
{
"operation": "Now on",
"time": "87.869",
"value": "Ashton Kutcher"
},
{
"operation": "Highlighting",
"time": "163.455",
"value": "Ashton Kutcher_sentence_0"
},
{
"operation": "Highlighting",
"time": "181.456",
"value": "Ashton Kutcher_sentence_1"
},
{
"operation": "Page search",
"time": "191.614",
"value": "'70s"
},
{
"operation": "Page search",
"time": "590.345",
"value": "'famous"
},
{
"operation": "Page search",
"time": "594.391",
"value": "famous"
},
{
"operation": "finish",
"time": "601.542",
"value": "finish"
}
]
}
|
16597
|
Richard Osbaldeston and his grandfather were not politicians, graduated MA from St John's College, Cambridge and was a chaplain to King George II.
|
{
"evidence": [
"Richard Osbaldeston_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRichard Osbaldeston and his grandfather were not politicians, graduated MA from St John's College, Cambridge and was a chaplain to King George II.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Richard Osbaldeston_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 387,
"split": "test_on"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.552",
"value": "?search="
},
{
"operation": "search",
"time": "15.197",
"value": "Richard Osbaldeston (Attorney General)"
},
{
"operation": "Now on",
"time": "16.138",
"value": "Richard Osbaldeston (Attorney General)"
},
{
"operation": "hyperlink",
"time": "41.868",
"value": "Richard Osbaldeston"
},
{
"operation": "Now on",
"time": "42.818",
"value": "Richard Osbaldeston"
},
{
"operation": "Highlighting",
"time": "78.881",
"value": "Richard Osbaldeston_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "113.699",
"value": "Richard Osbaldeston_sentence_2"
},
{
"operation": "Highlighting",
"time": "115.466",
"value": "Richard Osbaldeston_sentence_4"
},
{
"operation": "finish",
"time": "187.702",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.