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_3156
|
Wilton Norman Chamberlain was an American professional player who played as a center?in a team sport in which three teams, most commonly of five players each, competed.
|
{
"evidence": [
"Wilt Chamberlain",
"Basketball"
],
"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:\nWilton Norman Chamberlain was an American professional player who played as a center?in a team sport in which three teams, most commonly of five players each, competed.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wilt Chamberlain",
"Basketball"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5584,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_284
|
No Reservations is Catherine Zeta-Jones' first film.
|
{
"evidence": [
"No Reservations (film)",
"Abigail Breslin"
],
"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:\nNo Reservations is Catherine Zeta-Jones' first film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"No Reservations (film)",
"Abigail Breslin"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5585,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8540
|
Animal Farm is a satirical allegorical novella by George Orwell that reflects events leading up to the Russian Revolution of 1917 and then on into the Stalinist era of a socialist and communist state that spanned much of Eurasia from 1922 to 1991.
|
{
"evidence": [
"Animal Farm",
"Soviet Union"
],
"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:\nAnimal Farm is a satirical allegorical novella by George Orwell that reflects events leading up to the Russian Revolution of 1917 and then on into the Stalinist era of a socialist and communist state that spanned much of Eurasia from 1922 to 1991.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Animal Farm",
"Soviet Union"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5586,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
82796
|
Once believed to be merely empty ceremonial centres, the estimated maximum population of Maya city Copán is 50,000, and the estimated maximum population of Maya city Cival is 2,000-5,000.
|
{
"evidence": [
"Copán_sentence_21",
"Cival_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:\nOnce believed to be merely empty ceremonial centres, the estimated maximum population of Maya city Copán is 50,000, and the estimated maximum population of Maya city Cival is 2,000-5,000.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Copán_sentence_21",
"Cival_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5587,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.791",
"value": "?search="
},
{
"operation": "search",
"time": "72.612",
"value": "Copán"
},
{
"operation": "Now on",
"time": "74.086",
"value": "Copán"
},
{
"operation": "Page search",
"time": "91.506",
"value": "ceremonial"
},
{
"operation": "Page search",
"time": "112.545",
"value": "50,000"
},
{
"operation": "Page search",
"time": "248.777",
"value": "maximum"
},
{
"operation": "Page search",
"time": "280.945",
"value": "50,000"
},
{
"operation": "Highlighting",
"time": "321.437",
"value": "Copán_sentence_21"
},
{
"operation": "search",
"time": "489.302",
"value": "Cival"
},
{
"operation": "Now on",
"time": "490.107",
"value": "Cival"
},
{
"operation": "Highlighting",
"time": "492.659",
"value": "Cival_sentence_3"
},
{
"operation": "finish",
"time": "512.179",
"value": "finish"
}
]
}
|
exfever_3583
|
Tesla Inc. is the most successful company in the world according to Forbes.
|
{
"evidence": [
"Tesla Model S",
"Tesla Model 3"
],
"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:\nTesla Inc. is the most successful company in the world according to Forbes.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tesla Model S",
"Tesla Model 3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5588,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
65229
|
Pracana Dam is a 115 x 245.5 m buttress dam with a volume of 144,000 m³ and only one spillway on the right side.
|
{
"evidence": [
"Pracana Dam_sentence_7"
],
"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:\nPracana Dam is a 115 x 245.5 m buttress dam with a volume of 144,000 m³ and only one spillway on the right side.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Pracana Dam_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5589,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.793",
"value": "?search="
},
{
"operation": "search",
"time": "8.83",
"value": "Pracana Dam"
},
{
"operation": "Now on",
"time": "9.806",
"value": "Pracana Dam"
},
{
"operation": "Highlighting",
"time": "21.461",
"value": "Pracana Dam_cell_0_16_1"
},
{
"operation": "Highlighting",
"time": "21.507",
"value": "Pracana Dam_header_cell_0_16_0"
},
{
"operation": "Highlighting",
"time": "21.507",
"value": "Pracana Dam_header_cell_0_11_0"
},
{
"operation": "Highlighting",
"time": "22.929",
"value": "Pracana Dam_cell_0_15_1"
},
{
"operation": "Highlighting",
"time": "22.977",
"value": "Pracana Dam_header_cell_0_15_0"
},
{
"operation": "Highlighting",
"time": "26.729",
"value": "Pracana Dam_cell_0_12_1"
},
{
"operation": "Highlighting",
"time": "26.768",
"value": "Pracana Dam_header_cell_0_12_0"
},
{
"operation": "Highlighting",
"time": "30.41",
"value": "Pracana Dam_cell_0_17_1"
},
{
"operation": "Highlighting",
"time": "30.447",
"value": "Pracana Dam_header_cell_0_17_0"
},
{
"operation": "Page search",
"time": "52.896",
"value": "spillway"
},
{
"operation": "Highlighting",
"time": "61.754",
"value": "Pracana Dam_sentence_7"
},
{
"operation": "Highlighting",
"time": "70.307",
"value": "Pracana Dam_sentence_7"
},
{
"operation": "finish",
"time": "76.21",
"value": "finish"
}
]
}
|
35137
|
Four games resulted in a win, two games that resulted in losses, and another two games that resulted in a tie.
|
{
"evidence": [
"Game_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:\nFour games resulted in a win, two games that resulted in losses, and another two games that resulted in a tie.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Game_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5590,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.136",
"value": "?search="
},
{
"operation": "search",
"time": "52.76",
"value": "Four games"
},
{
"operation": "Now on",
"time": "53.781",
"value": "undefined"
},
{
"operation": "search",
"time": "65.563",
"value": "Four games resulted in a win"
},
{
"operation": "Now on",
"time": "68.478",
"value": "undefined"
},
{
"operation": "search",
"time": "92.289",
"value": "two games that resulted in losses"
},
{
"operation": "Now on",
"time": "94.138",
"value": "undefined"
},
{
"operation": "search",
"time": "207.437",
"value": "wo games that resulted"
},
{
"operation": "Now on",
"time": "208.782",
"value": "undefined"
},
{
"operation": "search",
"time": "219.493",
"value": "games"
},
{
"operation": "Now on",
"time": "220.52",
"value": "Game"
},
{
"operation": "hyperlink",
"time": "231.937",
"value": "Games"
},
{
"operation": "Now on",
"time": "232.664",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "234.189",
"value": "Game"
},
{
"operation": "Now on",
"time": "234.866",
"value": "Game"
},
{
"operation": "Highlighting",
"time": "235.835",
"value": "Game_sentence_0"
},
{
"operation": "finish",
"time": "247.055",
"value": "finish"
}
]
}
|
82417
|
Richard Witty Foster was once elected to the South Australian House of Assembly as the member for Newcastle and represented Flinders, a single-member electoral district for the South Australian House of Assembly, from 1902 to 1906.
|
{
"evidence": [
"Richard Witty Foster_sentence_8",
"Richard Witty Foster_sentence_9",
"Electoral district of Flinders_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:\nRichard Witty Foster was once elected to the South Australian House of Assembly as the member for Newcastle and represented Flinders, a single-member electoral district for the South Australian House of Assembly, from 1902 to 1906.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Richard Witty Foster_sentence_8",
"Richard Witty Foster_sentence_9",
"Electoral district of Flinders_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5591,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.708",
"value": "?search="
},
{
"operation": "search",
"time": "12.232",
"value": "Richard Witty Foster"
},
{
"operation": "Now on",
"time": "13.214",
"value": "Richard Witty Foster"
},
{
"operation": "Highlighting",
"time": "31.526",
"value": "Richard Witty Foster_sentence_8"
},
{
"operation": "Highlighting",
"time": "33.742",
"value": "Richard Witty Foster_sentence_9"
},
{
"operation": "hyperlink",
"time": "40.145",
"value": "Flinders"
},
{
"operation": "Now on",
"time": "41.554",
"value": "Electoral district of Flinders"
},
{
"operation": "Highlighting",
"time": "45.08",
"value": "Electoral district of Flinders_sentence_3"
},
{
"operation": "Now on",
"time": "59.097",
"value": "Electoral district of Flinders"
},
{
"operation": "finish",
"time": "66.899",
"value": "finish"
}
]
}
|
74537
|
Ralf Lau begin a career in golf in the 1970's and became the Russian team champion.
|
{
"evidence": [
"Ralf Lau_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:\nRalf Lau begin a career in golf in the 1970's and became the Russian team champion.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ralf Lau_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5592,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.972",
"value": "?search="
},
{
"operation": "search",
"time": "21.677",
"value": "Ralf Lau"
},
{
"operation": "Now on",
"time": "22.9",
"value": "Ralf Lau"
},
{
"operation": "Highlighting",
"time": "57.093",
"value": "Ralf Lau_sentence_0"
},
{
"operation": "Highlighting",
"time": "118.022",
"value": "Ralf Lau_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "119.178",
"value": "Ralf Lau_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "120.819",
"value": "Ralf Lau_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "120.828",
"value": "Ralf Lau_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "120.831",
"value": "Ralf Lau_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "127.929",
"value": "Ralf Lau_cell_0_3_1"
},
{
"operation": "Highlighting deleted",
"time": "127.937",
"value": "Ralf Lau_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "127.938",
"value": "Ralf Lau_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "147.625",
"value": "finish"
}
]
}
|
exfever_8673
|
Priyanka Chopra is an Indian actress who won the 50th anniversary of the Miss World pageant, which was held on November 30, 2000, at the Millennium Dome in London, United Kingdom.
|
{
"evidence": [
"Priyanka Chopra",
"Miss World 2000"
],
"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:\nPriyanka Chopra is an Indian actress who won the 50th anniversary of the Miss World pageant, which was held on November 30, 2000, at the Millennium Dome in London, United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Priyanka Chopra",
"Miss World 2000"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5593,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6484
|
Edward Harrison Norton is an Italian actor and filmmaker who graduated from the undergraduate college of Yale University, which was founded in 1701.
|
{
"evidence": [
"Edward Norton",
"Yale College"
],
"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:\nEdward Harrison Norton is an Italian actor and filmmaker who graduated from the undergraduate college of Yale University, which was founded in 1701.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Edward Norton",
"Yale College"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5594,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
16196
|
Oleg Vyugin, who was the Head of the Federal Financial Markets Service of Russia, became a member of the board of the Ministry of Finance in 1994 and served as Deputy Minister of Finance from 1996 to 1999.
|
{
"evidence": [
"Oleg Vyugin_sentence_0",
"Oleg Vyugin_sentence_6",
"Oleg Vyugin_sentence_7"
],
"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:\nOleg Vyugin, who was the Head of the Federal Financial Markets Service of Russia, became a member of the board of the Ministry of Finance in 1994 and served as Deputy Minister of Finance from 1996 to 1999.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Oleg Vyugin_sentence_0",
"Oleg Vyugin_sentence_6",
"Oleg Vyugin_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5595,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.903",
"value": "?search="
},
{
"operation": "search",
"time": "17.639",
"value": "Oleg Vyugin"
},
{
"operation": "Now on",
"time": "18.44",
"value": "Oleg Vyugin"
},
{
"operation": "Highlighting",
"time": "39.112",
"value": "Oleg Vyugin_sentence_0"
},
{
"operation": "Page search",
"time": "86.655",
"value": "Ministry"
},
{
"operation": "Highlighting",
"time": "109.606",
"value": "Oleg Vyugin_sentence_6"
},
{
"operation": "Highlighting",
"time": "136.454",
"value": "Oleg Vyugin_sentence_7"
},
{
"operation": "finish",
"time": "175.68",
"value": "finish"
}
]
}
|
55813
|
Michal Linial, a former Director at the Israel Institute for Advanced Studies (IIAS), completed her post-doctoral training at Stanford University in cellular neurochemistry.
|
{
"evidence": [
"Michal Linial_sentence_4",
"Michal Linial_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:\nMichal Linial, a former Director at the Israel Institute for Advanced Studies (IIAS), completed her post-doctoral training at Stanford University in cellular neurochemistry.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Michal Linial_sentence_4",
"Michal Linial_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5596,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.239",
"value": "?search="
},
{
"operation": "search",
"time": "23.474",
"value": "Michal Linial"
},
{
"operation": "Now on",
"time": "24.17",
"value": "Michal Linial"
},
{
"operation": "Highlighting",
"time": "37.556",
"value": "Michal Linial_sentence_4"
},
{
"operation": "Highlighting",
"time": "48.991",
"value": "Michal Linial_sentence_6"
},
{
"operation": "finish",
"time": "58.978",
"value": "finish"
}
]
}
|
exfever_5151
|
Taipei is the capital and a special municipality of a country in West Asia, at the junction of the East and South China Seas in the northwestern Pacific Ocean.
|
{
"evidence": [
"Taipei",
"Taiwan"
],
"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:\nTaipei is the capital and a special municipality of a country in West Asia, at the junction of the East and South China Seas in the northwestern Pacific Ocean.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Taipei",
"Taiwan"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5597,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6676
|
Jerry Goldsmith was an American composer who composed scores for a 1976 American film starring a British actress born on December 20, 1952.
|
{
"evidence": [
"Jerry Goldsmith",
"Logan's Run (film)",
"Jenny Agutter"
],
"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:\nJerry Goldsmith was an American composer who composed scores for a 1976 American film starring a British actress born on December 20, 1952.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jerry Goldsmith",
"Logan's Run (film)",
"Jenny Agutter"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5598,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2023
|
Ed Sheeran is an English singer-songwriter born in Halifax, West Yorkshire who won the Best British Male Solo Artist at the British Phonographic Industry's annual popular music awards.
|
{
"evidence": [
"Ed Sheeran",
"Brit Awards"
],
"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:\nEd Sheeran is an English singer-songwriter born in Halifax, West Yorkshire who won the Best British Male Solo Artist at the British Phonographic Industry's annual popular music awards.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ed Sheeran",
"Brit Awards"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5599,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_250
|
The Northeast megalopolis also known Bos-Wash corridor is the capital and most populous city of the Commonwealth of Massachusetts in the United States and 20th-most populous city in the country.
|
{
"evidence": [
"Northeast megalopolis",
"Boston"
],
"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 Northeast megalopolis also known Bos-Wash corridor is the capital and most populous city of the Commonwealth of Massachusetts in the United States and 20th-most populous city in the country.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Northeast megalopolis",
"Boston"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5600,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
62981
|
Paulo Pedroso, Secretary of State and Minister for Labor and Training of Portugal from 1999 to 2001, requested 600,000 euros compensation for his illegal arrest; however, he was only paid 100,000 euros by the state of Portugal.
|
{
"evidence": [
"Paulo Pedroso_sentence_2",
"Paulo Pedroso_sentence_4",
"Paulo Pedroso_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:\nPaulo Pedroso, Secretary of State and Minister for Labor and Training of Portugal from 1999 to 2001, requested 600,000 euros compensation for his illegal arrest; however, he was only paid 100,000 euros by the state of Portugal.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Paulo Pedroso_sentence_2",
"Paulo Pedroso_sentence_4",
"Paulo Pedroso_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5601,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.665",
"value": "?search="
},
{
"operation": "search",
"time": "7.266",
"value": "Paulo Pedroso"
},
{
"operation": "Now on",
"time": "7.971",
"value": "Paulo Pedroso"
},
{
"operation": "Page search",
"time": "149.637",
"value": "Secretary of State"
},
{
"operation": "Highlighting",
"time": "156.302",
"value": "Paulo Pedroso_sentence_0"
},
{
"operation": "Highlighting",
"time": "165.736",
"value": "Paulo Pedroso_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "168.304",
"value": "Paulo Pedroso_sentence_0"
},
{
"operation": "Highlighting",
"time": "175.333",
"value": "Paulo Pedroso_sentence_4"
},
{
"operation": "Highlighting",
"time": "325.504",
"value": "Paulo Pedroso_sentence_5"
},
{
"operation": "finish",
"time": "353.308",
"value": "finish"
}
]
}
|
28463
|
Guru Kelu Nair has noteworthy disciples including Mrinalini Sarabhai (the founder and director of the Darpana Academy of Performing Arts), Rukmini Devi Arundale, and Yog Sunder Desai, and has remained as a faculty member in Santiniketan until the death of Tagore in 1941.
|
{
"evidence": [
"Guru Kelu Nair_sentence_7",
"Guru Kelu Nair_sentence_6",
"Mrinalini Sarabhai_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:\nGuru Kelu Nair has noteworthy disciples including Mrinalini Sarabhai (the founder and director of the Darpana Academy of Performing Arts), Rukmini Devi Arundale, and Yog Sunder Desai, and has remained as a faculty member in Santiniketan until the death of Tagore in 1941.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Guru Kelu Nair_sentence_7",
"Guru Kelu Nair_sentence_6",
"Mrinalini Sarabhai_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5602,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.131",
"value": "?search="
},
{
"operation": "search",
"time": "13.117",
"value": "Guru Kelu Nair"
},
{
"operation": "Now on",
"time": "13.962",
"value": "Guru Kelu Nair"
},
{
"operation": "Highlighting",
"time": "25.521",
"value": "Guru Kelu Nair_sentence_7"
},
{
"operation": "Highlighting",
"time": "40.01",
"value": "Guru Kelu Nair_sentence_6"
},
{
"operation": "hyperlink",
"time": "47.721",
"value": "Mrinalini Sarabhai"
},
{
"operation": "Now on",
"time": "48.248",
"value": "Mrinalini Sarabhai"
},
{
"operation": "Highlighting",
"time": "53.538",
"value": "Mrinalini Sarabhai_sentence_1"
},
{
"operation": "search",
"time": "61.124",
"value": "Guru Kelu Nair"
},
{
"operation": "Now on",
"time": "61.76",
"value": "Guru Kelu Nair"
},
{
"operation": "hyperlink",
"time": "65.705",
"value": "Rukmini Devi Arundale"
},
{
"operation": "Now on",
"time": "66.277",
"value": "Rukmini Devi Arundale"
},
{
"operation": "finish",
"time": "96.729",
"value": "finish"
}
]
}
|
46120
|
Luis Miguel Ramis managed Real Madrid Castilla (a Spanish football team that plays in Segunda División B – Group 1) from January 5, 2016 to June 20, 2016, leading the team to win 14 games.
|
{
"evidence": [
"Luis Miguel Ramis_sentence_11",
"Real Madrid Castilla_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:\nLuis Miguel Ramis managed Real Madrid Castilla (a Spanish football team that plays in Segunda División B – Group 1) from January 5, 2016 to June 20, 2016, leading the team to win 14 games.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Luis Miguel Ramis_sentence_11",
"Real Madrid Castilla_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5603,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.791",
"value": "?search="
},
{
"operation": "search",
"time": "10.039",
"value": "Luis Miguel Ramis"
},
{
"operation": "Now on",
"time": "11.009",
"value": "Luis Miguel Ramis"
},
{
"operation": "Highlighting",
"time": "50.167",
"value": "Luis Miguel Ramis_sentence_11"
},
{
"operation": "hyperlink",
"time": "60.006",
"value": "Real Madrid Castilla"
},
{
"operation": "Now on",
"time": "61.002",
"value": "Real Madrid Castilla"
},
{
"operation": "Highlighting",
"time": "67.057",
"value": "Real Madrid Castilla_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "68.97",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "69.107",
"value": "Luis Miguel Ramis"
},
{
"operation": "finish",
"time": "118.398",
"value": "finish"
}
]
}
|
exfever_8531
|
The Star Wars sequel trilogy is the third trilogy of the primary Star Wars franchise, produced by Lucasfilm Ltd. and distributed by the Chinese film distribution studio within the Disney Media and Entertainment Distribution division.
|
{
"evidence": [
"Star Wars sequel trilogy",
"Walt Disney Studios Motion Pictures"
],
"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 Star Wars sequel trilogy is the third trilogy of the primary Star Wars franchise, produced by Lucasfilm Ltd. and distributed by the Chinese film distribution studio within the Disney Media and Entertainment Distribution division.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Star Wars sequel trilogy",
"Walt Disney Studios Motion Pictures"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5604,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7530
|
"Me Before You" is a romantic drama film starring an Asian actress who had her breakthrough role as a fictional character in the series of epic fantasy novels, A Song of Ice and Fire.
|
{
"evidence": [
"Me Before You (film)",
"Emilia Clarke",
"Daenerys Targaryen"
],
"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:\n\"Me Before You\" is a romantic drama film starring an Asian actress who had her breakthrough role as a fictional character in the series of epic fantasy novels, A Song of Ice and Fire.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Me Before You (film)",
"Emilia Clarke",
"Daenerys Targaryen"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5605,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3668
|
Alpha House is an Australian satire streaming series produced by Amazon Studios, a satire that specializes in gaining entertainment from politics.
|
{
"evidence": [
"Alpha House",
"Political satire"
],
"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:\nAlpha House is an Australian satire streaming series produced by Amazon Studios, a satire that specializes in gaining entertainment from politics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alpha House",
"Political satire"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5606,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
12102
|
Lynn Hornor was the British president and manager of a number of coal, oil and gas development, and land companies who served as a member of the advisory West Virginia council of defense.
|
{
"evidence": [
"Lynn Hornor_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:\nLynn Hornor was the British president and manager of a number of coal, oil and gas development, and land companies who served as a member of the advisory West Virginia council of defense.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Lynn Hornor_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5607,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.964",
"value": "?search="
},
{
"operation": "search",
"time": "10.418",
"value": "Lynn Hornor"
},
{
"operation": "Now on",
"time": "11.082",
"value": "Lynn Hornor"
},
{
"operation": "Highlighting",
"time": "24.91",
"value": "Lynn Hornor_sentence_0"
},
{
"operation": "finish",
"time": "40.444",
"value": "finish"
}
]
}
|
exfever_1461
|
Diary of the Dead is the first independently produced film by George A. Romero.
|
{
"evidence": [
"Diary of the Dead",
"Genius Products",
"ESPN"
],
"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:\nDiary of the Dead is the first independently produced film by George A. Romero.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Diary of the Dead",
"Genius Products",
"ESPN"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5608,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6262
|
Divergent is a 2014 American social science fiction action film directed by an American film director, based on a novel published by Harper Collins Children's Books in 2011.
|
{
"evidence": [
"Divergent (film)",
"Neil Burger",
"Divergent (novel)"
],
"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:\nDivergent is a 2014 American social science fiction action film directed by an American film director, based on a novel published by Harper Collins Children's Books in 2011.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Divergent (film)",
"Neil Burger",
"Divergent (novel)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5609,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36222
|
The Palace Hotel is a future project that will be the tallest building in Hobart (a city in Australia).
|
{
"evidence": [
"Hobart Building_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 Palace Hotel is a future project that will be the tallest building in Hobart (a city in Australia).\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Hobart Building_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5610,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.021",
"value": "?search="
},
{
"operation": "search",
"time": "18.51",
"value": "The Palace Hotel hobart"
},
{
"operation": "Now on",
"time": "19.798",
"value": "undefined"
},
{
"operation": "search",
"time": "30.953",
"value": "The Palace Hotel"
},
{
"operation": "Now on",
"time": "31.822",
"value": "Kimpton Clocktower Hotel"
},
{
"operation": "search",
"time": "98.172",
"value": "palace hotel hobart"
},
{
"operation": "Now on",
"time": "98.899",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "106.681",
"value": "Willis Polk"
},
{
"operation": "Now on",
"time": "107.372",
"value": "Willis Polk"
},
{
"operation": "Page search",
"time": "125.805",
"value": "tallest"
},
{
"operation": "hyperlink",
"time": "145.626",
"value": "Hobart Building"
},
{
"operation": "Now on",
"time": "146.229",
"value": "Hobart Building"
},
{
"operation": "Highlighting",
"time": "337.224",
"value": "Hobart Building_sentence_0"
},
{
"operation": "finish",
"time": "348.314",
"value": "finish"
}
]
}
|
95574
|
During the 1950 Australian Grand Prix, Jim Gullan did not participate in the race.
|
{
"evidence": [
"1950 Australian Grand Prix_sentence_7"
],
"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 1950 Australian Grand Prix, Jim Gullan did not participate in the race.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1950 Australian Grand Prix_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5611,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.014",
"value": "?search="
},
{
"operation": "search",
"time": "13.146",
"value": "1950 Australian Grand Prix"
},
{
"operation": "Now on",
"time": "14.231",
"value": "1950 Australian Grand Prix"
},
{
"operation": "Highlighting",
"time": "20.574",
"value": "1950 Australian Grand Prix_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "20.576",
"value": "1950 Australian Grand Prix_section_0"
},
{
"operation": "Highlighting",
"time": "21.437",
"value": "1950 Australian Grand Prix_cell_0_3_2"
},
{
"operation": "Highlighting",
"time": "21.531",
"value": "1950 Australian Grand Prix_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "21.531",
"value": "1950 Australian Grand Prix_header_cell_0_0_2"
},
{
"operation": "Highlighting deleted",
"time": "27.718",
"value": "1950 Australian Grand Prix_cell_0_3_2"
},
{
"operation": "Highlighting deleted",
"time": "27.721",
"value": "1950 Australian Grand Prix_section_0"
},
{
"operation": "Highlighting deleted",
"time": "27.793",
"value": "1950 Australian Grand Prix_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "27.796",
"value": "1950 Australian Grand Prix_header_cell_0_0_2"
},
{
"operation": "Highlighting",
"time": "28.563",
"value": "1950 Australian Grand Prix_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "28.565",
"value": "1950 Australian Grand Prix_section_0"
},
{
"operation": "Highlighting deleted",
"time": "30.026",
"value": "1950 Australian Grand Prix_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "30.028",
"value": "1950 Australian Grand Prix_section_0"
},
{
"operation": "Highlighting",
"time": "39.302",
"value": "1950 Australian Grand Prix_sentence_7"
},
{
"operation": "finish",
"time": "55.742",
"value": "finish"
}
]
}
|
6047
|
The Aerospace Players were formed in 1988 to perform musicals such as Grease by Jim Jacobs.
|
{
"evidence": [
"The Aerospace Players_sentence_0",
"The Aerospace Players_sentence_6",
"Grease (musical)_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:\nThe Aerospace Players were formed in 1988 to perform musicals such as Grease by Jim Jacobs.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"The Aerospace Players_sentence_0",
"The Aerospace Players_sentence_6",
"Grease (musical)_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5612,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.076",
"value": "?search="
},
{
"operation": "search",
"time": "16.245",
"value": "aerospace players"
},
{
"operation": "Now on",
"time": "17.192",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "30.628",
"value": "The Aerospace Players"
},
{
"operation": "Now on",
"time": "31.42",
"value": "The Aerospace Players"
},
{
"operation": "Highlighting",
"time": "36.793",
"value": "The Aerospace Players_sentence_0"
},
{
"operation": "Page search",
"time": "52.18",
"value": "jim jacobs"
},
{
"operation": "Page search",
"time": "53.554",
"value": "jim jacobs"
},
{
"operation": "Highlighting",
"time": "85.575",
"value": "The Aerospace Players_sentence_6"
},
{
"operation": "hyperlink",
"time": "95.83",
"value": "Grease"
},
{
"operation": "Now on",
"time": "96.888",
"value": "Grease (musical)"
},
{
"operation": "Highlighting",
"time": "100.049",
"value": "Grease (musical)_sentence_1"
},
{
"operation": "finish",
"time": "107.446",
"value": "finish"
}
]
}
|
46612
|
As a performance indicator aimed to push students towards academic subjects, English Baccalaureate has failed to do so as it has just widened the achievement gap for disadvantaged children.
|
{
"evidence": [
"English Baccalaureate_sentence_26"
],
"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:\nAs a performance indicator aimed to push students towards academic subjects, English Baccalaureate has failed to do so as it has just widened the achievement gap for disadvantaged children.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"English Baccalaureate_sentence_26"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5613,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.139",
"value": "?search="
},
{
"operation": "Now on",
"time": "8.085",
"value": "?search="
},
{
"operation": "search",
"time": "19.054",
"value": "English Baccalaureate"
},
{
"operation": "Now on",
"time": "19.96",
"value": "English Baccalaureate"
},
{
"operation": "Page search",
"time": "247.704",
"value": "achievement gap"
},
{
"operation": "Highlighting",
"time": "305.139",
"value": "English Baccalaureate_sentence_26"
},
{
"operation": "finish",
"time": "346.874",
"value": "finish"
}
]
}
|
exfever_2860
|
The Golden Girls came to an end when Bea Arthur chose to leave the series.
|
{
"evidence": [
"The Golden Girls",
"Betty White"
],
"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 Golden Girls came to an end when Bea Arthur chose to leave the series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Golden Girls",
"Betty White"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5614,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3153
|
Paris is the debut studio album by an American singer born in New York City.
|
{
"evidence": [
"Paris (Paris Hilton album)",
"Paris Hilton"
],
"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:\nParis is the debut studio album by an American singer born in New York City.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paris (Paris Hilton album)",
"Paris Hilton"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5615,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7775
|
Joshua Michael Homme is an American musician who was formerly the co-founder and guitarist of an American rock band, formed in Palm Desert, California, in 1987.
|
{
"evidence": [
"Josh Homme",
"Kyuss"
],
"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:\nJoshua Michael Homme is an American musician who was formerly the co-founder and guitarist of an American rock band, formed in Palm Desert, California, in 1987.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Josh Homme",
"Kyuss"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5616,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
20883
|
Found in Canada, Drepanidae chinoea has a wingspan of 16-20.5 mm for males, with lustrous white fore and hindwings.
|
{
"evidence": [
"Drepanidae_sentence_3"
],
"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:\nFound in Canada, Drepanidae chinoea has a wingspan of 16-20.5 mm for males, with lustrous white fore and hindwings.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Drepanidae_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5617,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.124",
"value": "?search="
},
{
"operation": "search",
"time": "36.108",
"value": "Drepanidae"
},
{
"operation": "Now on",
"time": "37.076",
"value": "Drepanidae"
},
{
"operation": "search",
"time": "57.835",
"value": "drepanidae chinoea"
},
{
"operation": "Now on",
"time": "58.666",
"value": "undefined"
},
{
"operation": "search",
"time": "67.851",
"value": "drepanidae"
},
{
"operation": "Now on",
"time": "68.427",
"value": "Drepanidae"
},
{
"operation": "Highlighting",
"time": "97.197",
"value": "Drepanidae_sentence_4"
},
{
"operation": "hyperlink",
"time": "97.962",
"value": "subfamilies"
},
{
"operation": "Now on",
"time": "98.526",
"value": "Subfamily"
},
{
"operation": "back-button-clicked",
"time": "104.749",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "104.845",
"value": "Drepanidae"
},
{
"operation": "Highlighting",
"time": "116.219",
"value": "Drepanidae_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "118.643",
"value": "Drepanidae_sentence_4"
},
{
"operation": "finish",
"time": "129.964",
"value": "finish"
}
]
}
|
58502
|
John Antonakis is known for is leadership work and his ability to predict the outcome of local elections.
|
{
"evidence": [
"John Antonakis_sentence_3",
"John Antonakis_sentence_4",
"John Antonakis_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:\nJohn Antonakis is known for is leadership work and his ability to predict the outcome of local elections.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"John Antonakis_sentence_3",
"John Antonakis_sentence_4",
"John Antonakis_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5618,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.933",
"value": "?search="
},
{
"operation": "search",
"time": "7.791",
"value": "John Antonakis"
},
{
"operation": "Now on",
"time": "8.496",
"value": "John Antonakis"
},
{
"operation": "Highlighting",
"time": "52.215",
"value": "John Antonakis_sentence_3"
},
{
"operation": "Highlighting",
"time": "53.11",
"value": "John Antonakis_sentence_4"
},
{
"operation": "Highlighting",
"time": "61.342",
"value": "John Antonakis_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "74.183",
"value": "John Antonakis_sentence_3"
},
{
"operation": "Highlighting",
"time": "136.855",
"value": "John Antonakis_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "137.767",
"value": "John Antonakis_sentence_4"
},
{
"operation": "Highlighting",
"time": "137.998",
"value": "John Antonakis_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "138.55",
"value": "John Antonakis_sentence_6"
},
{
"operation": "Highlighting",
"time": "138.79",
"value": "John Antonakis_sentence_6"
},
{
"operation": "finish",
"time": "143.985",
"value": "finish"
}
]
}
|
exfever_3787
|
Sir Richard Branson is a British billionaire that founded the Virgin Rail Group to bid for passenger rail franchises during the process by which the government passed control into the private hands of the operation of the newest railway in the world.
|
{
"evidence": [
"Richard Branson",
"Privatisation of British Rail",
"Rail transport in Great Britain"
],
"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:\nSir Richard Branson is a British billionaire that founded the Virgin Rail Group to bid for passenger rail franchises during the process by which the government passed control into the private hands of the operation of the newest railway in the world.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Richard Branson",
"Privatisation of British Rail",
"Rail transport in Great Britain"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5619,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
42837
|
Conaway Ranch is solely owned by the Yolo County ever since it was built.
|
{
"evidence": [
"Conaway Ranch_sentence_7",
"Conaway Ranch_sentence_9"
],
"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:\nConaway Ranch is solely owned by the Yolo County ever since it was built.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Conaway Ranch_sentence_7",
"Conaway Ranch_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5620,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.971",
"value": "?search="
},
{
"operation": "Now on",
"time": "10.165",
"value": "?search="
},
{
"operation": "search",
"time": "24.847",
"value": "Conaway Ranch"
},
{
"operation": "Now on",
"time": "25.615",
"value": "Conaway Ranch"
},
{
"operation": "Page search",
"time": "38.767",
"value": "solely"
},
{
"operation": "Page search",
"time": "69.25",
"value": "owned"
},
{
"operation": "Highlighting",
"time": "78.367",
"value": "Conaway Ranch_sentence_7"
},
{
"operation": "Page search",
"time": "106.992",
"value": "built"
},
{
"operation": "Highlighting deleted",
"time": "111.324",
"value": "Conaway Ranch_sentence_7"
},
{
"operation": "Highlighting",
"time": "112.838",
"value": "Conaway Ranch_sentence_7"
},
{
"operation": "Highlighting",
"time": "126.388",
"value": "Conaway Ranch_sentence_26"
},
{
"operation": "Highlighting",
"time": "126.39",
"value": "Conaway Ranch_section_1"
},
{
"operation": "Highlighting",
"time": "195.744",
"value": "Conaway Ranch_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "200.79",
"value": "Conaway Ranch_sentence_26"
},
{
"operation": "Highlighting deleted",
"time": "200.794",
"value": "Conaway Ranch_section_1"
},
{
"operation": "finish",
"time": "227.092",
"value": "finish"
}
]
}
|
94028
|
Texas SH 242 connects the Woodlands and 2.41 sq mile 89 ft above sea level, Patton Village has six major intersections including I-69 and I-45.
|
{
"evidence": [
"List of Farm to Market Roads in Texas (200–299)_sentence_222",
"List of Farm to Market Roads in Texas (200–299)_sentence_223"
],
"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:\nTexas SH 242 connects the Woodlands and 2.41 sq mile 89 ft above sea level, Patton Village has six major intersections including I-69 and I-45.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"List of Farm to Market Roads in Texas (200–299)_sentence_222",
"List of Farm to Market Roads in Texas (200–299)_sentence_223"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5621,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.64",
"value": "?search="
},
{
"operation": "search",
"time": "6.022",
"value": "Contains...Texas SH 242"
},
{
"operation": "Now on",
"time": "8.75",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "49.119",
"value": "List of Farm to Market Roads in Texas (200–299)"
},
{
"operation": "Now on",
"time": "55.712",
"value": "List of Farm to Market Roads in Texas (200–299)"
},
{
"operation": "hyperlink",
"time": "57.635",
"value": "43 FM 242 List of Farm to Market Roads in Texas (200–299)_section_49"
},
{
"operation": "Highlighting",
"time": "64.272",
"value": "List of Farm to Market Roads in Texas (200–299)_sentence_222"
},
{
"operation": "Highlighting",
"time": "67.118",
"value": "List of Farm to Market Roads in Texas (200–299)_sentence_223"
},
{
"operation": "finish",
"time": "391.292",
"value": "finish"
}
]
}
|
72777
|
Yaniv Rokah is from Netanya, the seventh largest city in Israel, and attended Lee Strasberg Theater Institute in New York.
|
{
"evidence": [
"Yaniv Rokah_sentence_1",
"Netanya_sentence_4",
"Yaniv Rokah_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:\nYaniv Rokah is from Netanya, the seventh largest city in Israel, and attended Lee Strasberg Theater Institute in New York.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Yaniv Rokah_sentence_1",
"Netanya_sentence_4",
"Yaniv Rokah_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5622,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.143",
"value": "?search="
},
{
"operation": "search",
"time": "9.226",
"value": "Yaniv Rokah"
},
{
"operation": "Now on",
"time": "10.663",
"value": "Yaniv Rokah"
},
{
"operation": "Highlighting",
"time": "160.377",
"value": "Yaniv Rokah_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "163.516",
"value": "Yaniv Rokah_sentence_1"
},
{
"operation": "Highlighting",
"time": "164.532",
"value": "Yaniv Rokah_sentence_1"
},
{
"operation": "hyperlink",
"time": "165.112",
"value": "Netanya"
},
{
"operation": "Now on",
"time": "166.535",
"value": "Netanya"
},
{
"operation": "Highlighting",
"time": "175.665",
"value": "Netanya_sentence_4"
},
{
"operation": "back-button-clicked",
"time": "178.067",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "178.126",
"value": "Yaniv Rokah"
},
{
"operation": "Highlighting",
"time": "181.043",
"value": "Yaniv Rokah_sentence_4"
},
{
"operation": "finish",
"time": "186.173",
"value": "finish"
}
]
}
|
exfever_4439
|
The Hui people are East Asian ethnoreligious group who speak a Sinitic language spoken primarily in Kazakhstan and Kyrgyztan.
|
{
"evidence": [
"Hui people",
"Dungan language"
],
"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 Hui people are East Asian ethnoreligious group who speak a Sinitic language spoken primarily in Kazakhstan and Kyrgyztan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hui people",
"Dungan language"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5623,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6152
|
Star vs. the Forces of Evil is an animated magical girl television series for the?American animation studio that creates, develops, and produces?for Walt Disney Television.?
|
{
"evidence": [
"Star vs. the Forces of Evil",
"Disney Television Animation"
],
"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:\nStar vs. the Forces of Evil is an animated magical girl television series for the?American animation studio that creates, develops, and produces?for Walt Disney Television.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Star vs. the Forces of Evil",
"Disney Television Animation"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5624,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6447
|
In 2008, Jeremy Renner won the Best Actor Award for his performance in the film The Hurt Locker.
|
{
"evidence": [
"The Hurt Locker",
"War 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:\nIn 2008, Jeremy Renner won the Best Actor Award for his performance in the film The Hurt Locker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Hurt Locker",
"War film"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5625,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
85758
|
Until the close of the 20th century there was only one way to reach Vradeto, via a kalderimi (paved mule road) called the Skala of Vradeto (Greek: Σκάλα Βραδέτου) a stairway of about 1000 steps built on the side of a canyon and almost impossible to seem from a distance which starts at the end of a long path from the village of Kapesovo.
|
{
"evidence": [
"Vradeto_sentence_8",
"Vradeto_sentence_9"
],
"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:\nUntil the close of the 20th century there was only one way to reach Vradeto, via a kalderimi (paved mule road) called the Skala of Vradeto (Greek: Σκάλα Βραδέτου) a stairway of about 1000 steps built on the side of a canyon and almost impossible to seem from a distance which starts at the end of a long path from the village of Kapesovo.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vradeto_sentence_8",
"Vradeto_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5626,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.671",
"value": "?search="
},
{
"operation": "search",
"time": "24.445",
"value": "vradeto"
},
{
"operation": "Now on",
"time": "25.19",
"value": "Vradeto"
},
{
"operation": "Highlighting",
"time": "62.889",
"value": "Vradeto_sentence_8"
},
{
"operation": "Highlighting",
"time": "80.719",
"value": "Vradeto_sentence_9"
},
{
"operation": "finish",
"time": "86.608",
"value": "finish"
}
]
}
|
exfever_6877
|
How to Train Your Dragon is a Chinese computer-animated action-fantasy film directed by a Canadian film director who is best known for directing the Oscar-nominated animated films Lilo & Stitch for Walt Disney Animation Studios.
|
{
"evidence": [
"How to Train Your Dragon (film)",
"Dean DeBlois"
],
"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:\nHow to Train Your Dragon is a Chinese computer-animated action-fantasy film directed by a Canadian film director who is best known for directing the Oscar-nominated animated films Lilo & Stitch for Walt Disney Animation Studios.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"How to Train Your Dragon (film)",
"Dean DeBlois"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5627,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8980
|
Coca is any of the ten cultivated plants in the Erythroxylaceae, which is one of the eight major hierarchical taxonomic ranks in Linnaean taxonomy that can be divided into an auxiliary taxonomic rank.
|
{
"evidence": [
"Coca",
"Family (biology)",
"Subfamily"
],
"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:\nCoca is any of the ten cultivated plants in the Erythroxylaceae, which is one of the eight major hierarchical taxonomic ranks in Linnaean taxonomy that can be divided into an auxiliary taxonomic rank.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Coca",
"Family (biology)",
"Subfamily"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5628,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8112
|
Bruce Willis is an American retired actor whose credits include a 1998 American science fiction disaster film produced and directed by Michael Bay.
|
{
"evidence": [
"Bruce Willis",
"Armageddon (1998 film)"
],
"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:\nBruce Willis is an American retired actor whose credits include a 1998 American science fiction disaster film produced and directed by Michael Bay.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bruce Willis",
"Armageddon (1998 film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5629,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71051
|
The Most Reverend Éamonn Oliver Walsh (born 1 September 1944) is a Catholic archbishop( 22 April 1990) from Ireland whowas ordained a priest for the diocese of Dublin on 13 April 1969.
|
{
"evidence": [
"Éamonn Oliver Walsh_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:\nThe Most Reverend Éamonn Oliver Walsh (born 1 September 1944) is a Catholic archbishop( 22 April 1990) from Ireland whowas ordained a priest for the diocese of Dublin on 13 April 1969.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Éamonn Oliver Walsh_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5630,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.588",
"value": "?search="
},
{
"operation": "search",
"time": "30.815",
"value": "Éamonn Oliver Walsh"
},
{
"operation": "Now on",
"time": "31.524",
"value": "Éamonn Oliver Walsh"
},
{
"operation": "Highlighting",
"time": "46.446",
"value": "Éamonn Oliver Walsh_sentence_0"
},
{
"operation": "Highlighting",
"time": "50.137",
"value": "Éamonn Oliver Walsh_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "50.142",
"value": "Éamonn Oliver Walsh_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "50.142",
"value": "Éamonn Oliver Walsh_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "75.427",
"value": "Éamonn Oliver Walsh_sentence_0"
},
{
"operation": "finish",
"time": "80.36",
"value": "finish"
}
]
}
|
86137
|
Samuel Denoff, born on July 1, 1928 in Brooklyn, New York, U.S., was a screenwriter, television producer and winner of the 1969 Outstanding Variety or Musical Program Emmy Award for the Bill Cosby Special.
|
{
"evidence": [
"Sam Denoff_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:\nSamuel Denoff, born on July 1, 1928 in Brooklyn, New York, U.S., was a screenwriter, television producer and winner of the 1969 Outstanding Variety or Musical Program Emmy Award for the Bill Cosby Special.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sam Denoff_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5631,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.722",
"value": "?search="
},
{
"operation": "search",
"time": "23.044",
"value": "Contains...Samuel Denoff"
},
{
"operation": "Now on",
"time": "23.811",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "27.363",
"value": "Sam Denoff"
},
{
"operation": "Now on",
"time": "27.865",
"value": "Sam Denoff"
},
{
"operation": "Highlighting",
"time": "30.334",
"value": "Sam Denoff_sentence_0"
},
{
"operation": "search",
"time": "104.769",
"value": "1969 Emmys"
},
{
"operation": "Now on",
"time": "105.628",
"value": "1969 Emmys"
},
{
"operation": "Page search",
"time": "117.967",
"value": "Denoff"
},
{
"operation": "search",
"time": "161.132",
"value": "Contains...1969 Outstanding Variety or Musical Program"
},
{
"operation": "Now on",
"time": "163.445",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "176.851",
"value": "Bill Cosby"
},
{
"operation": "Now on",
"time": "178.082",
"value": "Bill Cosby"
},
{
"operation": "Page search",
"time": "182.656",
"value": "1969 Outstanding Variety or Musical Program"
},
{
"operation": "Page search",
"time": "203.496",
"value": "Denoff"
},
{
"operation": "back-button-clicked",
"time": "215.196",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "215.951",
"value": "undefined"
},
{
"operation": "search",
"time": "342.296",
"value": "Outstanding Variety or Musical Program"
},
{
"operation": "Now on",
"time": "343.338",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "346.982",
"value": "Primetime Emmy Award for Outstanding Variety Special (Pre-Recorded)"
},
{
"operation": "Now on",
"time": "347.484",
"value": "Primetime Emmy Award for Outstanding Variety Special (Pre-Recorded)"
},
{
"operation": "finish",
"time": "369.726",
"value": "finish"
}
]
}
|
exfever_8189
|
International relations as a distinct field of study began in Britain.
|
{
"evidence": [
"International relations",
"Cultural anthropology",
"Methodology"
],
"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:\nInternational relations as a distinct field of study began in Britain.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"International relations",
"Cultural anthropology",
"Methodology"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5632,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
50335
|
William Ecker lived in the Virginia suburbs of Washington D.C. and moved to Punta Gorda, Florida in 2001 and finished writing a book with Kenneth V. Jack.
|
{
"evidence": [
"William Ecker_sentence_9",
"William Ecker_sentence_10",
"William Ecker_sentence_11",
"William Ecker_sentence_12"
],
"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:\nWilliam Ecker lived in the Virginia suburbs of Washington D.C. and moved to Punta Gorda, Florida in 2001 and finished writing a book with Kenneth V. Jack.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"William Ecker_sentence_9",
"William Ecker_sentence_10",
"William Ecker_sentence_11",
"William Ecker_sentence_12"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5633,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.435",
"value": "?search="
},
{
"operation": "search",
"time": "23.685",
"value": "William Ecker"
},
{
"operation": "Now on",
"time": "24.632",
"value": "William Ecker"
},
{
"operation": "Highlighting",
"time": "37.229",
"value": "William Ecker_sentence_9"
},
{
"operation": "Highlighting",
"time": "46.974",
"value": "William Ecker_sentence_10"
},
{
"operation": "Highlighting",
"time": "55.443",
"value": "William Ecker_sentence_11"
},
{
"operation": "search",
"time": "88.551",
"value": "Blue Moon Over Cuba"
},
{
"operation": "Now on",
"time": "91.225",
"value": "undefined"
},
{
"operation": "search",
"time": "111.519",
"value": "Kenneth V. Jack"
},
{
"operation": "Now on",
"time": "113.465",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "151.446",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "151.515",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "152.786",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "152.801",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "152.859",
"value": "William Ecker"
},
{
"operation": "Highlighting",
"time": "160.871",
"value": "William Ecker_sentence_12"
},
{
"operation": "finish",
"time": "166.697",
"value": "finish"
}
]
}
|
94314
|
The very first German registration plates, from 1913 to the end of the 1920s, were rectangular, with a white background and with the name or initials of the colony in red followed by the registration number; however, special military service plates were used with the initials SOM (Somalia) or T (Tripolitania) in front.
|
{
"evidence": [
"Vehicle registration plates of the German Democratic Republic_sentence_0",
"Vehicle registration plates of the German Democratic Republic_sentence_1",
"Vehicle registration plates of the German Democratic Republic_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:\nThe very first German registration plates, from 1913 to the end of the 1920s, were rectangular, with a white background and with the name or initials of the colony in red followed by the registration number; however, special military service plates were used with the initials SOM (Somalia) or T (Tripolitania) in front.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vehicle registration plates of the German Democratic Republic_sentence_0",
"Vehicle registration plates of the German Democratic Republic_sentence_1",
"Vehicle registration plates of the German Democratic Republic_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5634,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.538",
"value": "?search="
},
{
"operation": "Now on",
"time": "10.861",
"value": "?search="
},
{
"operation": "search",
"time": "16.857",
"value": "Contains...German registration plates"
},
{
"operation": "Now on",
"time": "18.628",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "21.252",
"value": "Vehicle registration plates of the German Democratic Republic"
},
{
"operation": "Now on",
"time": "22.431",
"value": "Vehicle registration plates of the German Democratic Republic"
},
{
"operation": "back-button-clicked",
"time": "31.319",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "32.298",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "65.563",
"value": "Special Plates Vehicle registration plates of the German Democratic Republic_section_2"
},
{
"operation": "Now on",
"time": "66.541",
"value": "Vehicle registration plates of the German Democratic Republic#Special Plates Vehicle registration plates of the German Democratic Republic section 2"
},
{
"operation": "hyperlink",
"time": "88.128",
"value": "German Democratic Republic"
},
{
"operation": "Now on",
"time": "92.78",
"value": "East Germany"
},
{
"operation": "back-button-clicked",
"time": "96.28",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "97.11",
"value": "Vehicle registration plates of the German Democratic Republic#Special Plates Vehicle registration plates of the German Democratic Republic section 2"
},
{
"operation": "hyperlink",
"time": "100.797",
"value": "1 Design Vehicle registration plates of the German Democratic Republic_section_0"
},
{
"operation": "Highlighting",
"time": "105.827",
"value": "Vehicle registration plates of the German Democratic Republic_sentence_0"
},
{
"operation": "Highlighting",
"time": "106.66",
"value": "Vehicle registration plates of the German Democratic Republic_sentence_1"
},
{
"operation": "Highlighting",
"time": "107.108",
"value": "Vehicle registration plates of the German Democratic Republic_sentence_2"
},
{
"operation": "finish",
"time": "113.211",
"value": "finish"
}
]
}
|
exfever_4714
|
John Heard Jr. was an American actor who starred in a 1983 American neo-noir thriller directed by a Czech film director who is best known for his involvement in the Czechoslovak New Wave.
|
{
"evidence": [
"John Heard (actor)",
"Cutter's Way",
"Ivan Passer"
],
"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:\nJohn Heard Jr. was an American actor who starred in a 1983 American neo-noir thriller directed by a Czech film director who is best known for his involvement in the Czechoslovak New Wave.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Heard (actor)",
"Cutter's Way",
"Ivan Passer"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5635,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
90033
|
Microsoft Garage, a Microsoft program encouraging employees to work on projects that they are passionate about, focuses on supporting employees, who were then invited to demo their projects to executives from the Microsoft Office group.
|
{
"evidence": [
"Microsoft Garage_sentence_0",
"Microsoft Garage_sentence_19",
"Microsoft Garage_sentence_18"
],
"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:\nMicrosoft Garage, a Microsoft program encouraging employees to work on projects that they are passionate about, focuses on supporting employees, who were then invited to demo their projects to executives from the Microsoft Office group.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Microsoft Garage_sentence_0",
"Microsoft Garage_sentence_19",
"Microsoft Garage_sentence_18"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5636,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.516",
"value": "?search="
},
{
"operation": "search",
"time": "12.151",
"value": "Microsoft Garage"
},
{
"operation": "Now on",
"time": "13.069",
"value": "Microsoft Garage"
},
{
"operation": "Highlighting",
"time": "19.449",
"value": "Microsoft Garage_sentence_0"
},
{
"operation": "Highlighting",
"time": "44.195",
"value": "Microsoft Garage_sentence_19"
},
{
"operation": "Highlighting",
"time": "54.308",
"value": "Microsoft Garage_sentence_18"
},
{
"operation": "finish",
"time": "68.448",
"value": "finish"
}
]
}
|
32458
|
The systematic name of this enzyme class, which catalyzes the chlorination of organic compounds, is chloride: hydrogen-peroxide oxidoreductase, and apparently this enzyme belongs to the family of proteases.
|
{
"evidence": [
"Chloride peroxidase_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:\nThe systematic name of this enzyme class, which catalyzes the chlorination of organic compounds, is chloride: hydrogen-peroxide oxidoreductase, and apparently this enzyme belongs to the family of proteases.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Chloride peroxidase_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5637,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.679",
"value": "?search="
},
{
"operation": "search",
"time": "15.664",
"value": "chloride"
},
{
"operation": "Now on",
"time": "16.882",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "22.009",
"value": "Chloride"
},
{
"operation": "Now on",
"time": "22.763",
"value": "Myeloperoxidase"
},
{
"operation": "Page search",
"time": "40.722",
"value": "organic compounds"
},
{
"operation": "hyperlink",
"time": "56.517",
"value": "chloride"
},
{
"operation": "Now on",
"time": "57.087",
"value": "Chloride"
},
{
"operation": "back-button-clicked",
"time": "69.217",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "69.633",
"value": "Myeloperoxidase"
},
{
"operation": "hyperlink",
"time": "83.707",
"value": "peroxidase"
},
{
"operation": "Now on",
"time": "84.139",
"value": "Peroxidase"
},
{
"operation": "Page search",
"time": "105.142",
"value": "organic compounds"
},
{
"operation": "search",
"time": "123.463",
"value": "catalyzes the chlorination of organic compounds"
},
{
"operation": "Now on",
"time": "124.72",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "128.865",
"value": "Hypochlorite"
},
{
"operation": "Now on",
"time": "129.754",
"value": "Hypochlorite"
},
{
"operation": "back-button-clicked",
"time": "130.897",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "131.483",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "137.653",
"value": "Chloride peroxidase"
},
{
"operation": "Now on",
"time": "138.432",
"value": "Chloride peroxidase"
},
{
"operation": "Highlighting",
"time": "149.31",
"value": "Chloride peroxidase_sentence_0"
},
{
"operation": "Highlighting",
"time": "170.277",
"value": "Chloride peroxidase_sentence_5"
},
{
"operation": "Page search",
"time": "201.243",
"value": "proteases"
},
{
"operation": "Highlighting",
"time": "246.531",
"value": "Chloride peroxidase_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "246.725",
"value": "Chloride peroxidase_sentence_6"
},
{
"operation": "search",
"time": "250.641",
"value": "chloroperoxidase"
},
{
"operation": "Now on",
"time": "251.112",
"value": "Chloride peroxidase"
},
{
"operation": "Highlighting deleted",
"time": "256.644",
"value": "Chloride peroxidase_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "258.463",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "258.515",
"value": "Chloride peroxidase"
},
{
"operation": "Highlighting",
"time": "260.657",
"value": "Chloride peroxidase_sentence_0"
},
{
"operation": "search",
"time": "289.488",
"value": "proteases"
},
{
"operation": "Now on",
"time": "290.533",
"value": "Protease"
},
{
"operation": "back-button-clicked",
"time": "337.965",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "338.015",
"value": "Chloride peroxidase"
},
{
"operation": "Highlighting deleted",
"time": "350.906",
"value": "Chloride peroxidase_sentence_5"
},
{
"operation": "Highlighting",
"time": "356.236",
"value": "Chloride peroxidase_sentence_5"
},
{
"operation": "Highlighting",
"time": "357.848",
"value": "Chloride peroxidase_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "358.625",
"value": "Chloride peroxidase_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "359.663",
"value": "Chloride peroxidase_sentence_0"
},
{
"operation": "finish",
"time": "369.427",
"value": "finish"
}
]
}
|
exfever_3327
|
The Marvel Cinematic Universe is an American media franchise and shared universe centered on a series of superhero films much like the fictional shared universe where the stories in most American comic book titles including the fictional team of superheroes that made its debut in September 1963.
|
{
"evidence": [
"Marvel Cinematic Universe",
"Marvel Universe",
"Avengers (comics)"
],
"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 Marvel Cinematic Universe is an American media franchise and shared universe centered on a series of superhero films much like the fictional shared universe where the stories in most American comic book titles including the fictional team of superheroes that made its debut in September 1963.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Marvel Cinematic Universe",
"Marvel Universe",
"Avengers (comics)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5638,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7432
|
Rose Elizabeth Fitzgerald Kennedy was an American philanthropist, socialite, and matriarch of an American artist family that has long been prominent in American politics.
|
{
"evidence": [
"Rose Kennedy",
"Kennedy family"
],
"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:\nRose Elizabeth Fitzgerald Kennedy was an American philanthropist, socialite, and matriarch of an American artist family that has long been prominent in American politics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Rose Kennedy",
"Kennedy family"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5639,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7270
|
Superman is a 1978 superhero film based on the character of the same name in a British comic book publisher and the flagship unit of DC Entertainment, a subsidiary of Warner Bros. Discovery.
|
{
"evidence": [
"Superman (1978 film)",
"DC Comics"
],
"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:\nSuperman is a 1978 superhero film based on the character of the same name in a British comic book publisher and the flagship unit of DC Entertainment, a subsidiary of Warner Bros. Discovery.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Superman (1978 film)",
"DC Comics"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5640,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
24611
|
Finkum, in the Province of Friesland, and country of Netherlands, had a population of 465 as of 2017.
|
{
"evidence": [
"Finkum_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:\nFinkum, in the Province of Friesland, and country of Netherlands, had a population of 465 as of 2017.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Finkum_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5641,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.17",
"value": "?search="
},
{
"operation": "search",
"time": "21.482",
"value": "Finkum"
},
{
"operation": "Now on",
"time": "22.565",
"value": "Finkum"
},
{
"operation": "Highlighting",
"time": "47.748",
"value": "Finkum_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "80.903",
"value": "Finkum_sentence_0"
},
{
"operation": "Highlighting",
"time": "82.397",
"value": "Finkum_sentence_1"
},
{
"operation": "finish",
"time": "127.915",
"value": "finish"
}
]
}
|
72486
|
Gopala Dasa, named "Bhaganna" at birth and known to have been an astrologer as well, was a prominent 18th-century Kannada language poet and saint belonging to the Haridasa tradition.
|
{
"evidence": [
"Gopala Dasa_sentence_0",
"Gopala Dasa_sentence_2",
"Gopala Dasa_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:\nGopala Dasa, named \"Bhaganna\" at birth and known to have been an astrologer as well, was a prominent 18th-century Kannada language poet and saint belonging to the Haridasa tradition.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Gopala Dasa_sentence_0",
"Gopala Dasa_sentence_2",
"Gopala Dasa_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5642,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.706",
"value": "?search="
},
{
"operation": "search",
"time": "15.222",
"value": "Gopala Dasa"
},
{
"operation": "Now on",
"time": "18.369",
"value": "Gopala Dasa"
},
{
"operation": "Highlighting",
"time": "23.146",
"value": "Gopala Dasa_sentence_0"
},
{
"operation": "Highlighting",
"time": "28.79",
"value": "Gopala Dasa_sentence_2"
},
{
"operation": "Highlighting",
"time": "107.991",
"value": "Gopala Dasa_sentence_5"
},
{
"operation": "finish",
"time": "115.427",
"value": "finish"
}
]
}
|
exfever_915
|
Margaret of Valois was a French princess of the Valois dynasty who became Queen of Navarre by marriage to the first monarch of France from the House of Bourbon, a cadet branch of the Capetian dynasty.
|
{
"evidence": [
"Margaret of Valois",
"Henry IV of France"
],
"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:\nMargaret of Valois was a French princess of the Valois dynasty who became Queen of Navarre by marriage to the first monarch of France from the House of Bourbon, a cadet branch of the Capetian dynasty.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Margaret of Valois",
"Henry IV of France"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5643,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
16399
|
In San Antonio Riders season 1991 l (American football team), WLAF league got six wins and four lost while in 1992 got three wins and seven lost.
|
{
"evidence": [
"San Antonio Riders_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:\nIn San Antonio Riders season 1991 l (American football team), WLAF league got six wins and four lost while in 1992 got three wins and seven lost.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"San Antonio Riders_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5644,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.367",
"value": "?search="
},
{
"operation": "search",
"time": "8.025",
"value": "San Antonio Riders season 1991"
},
{
"operation": "Now on",
"time": "10.117",
"value": "undefined"
},
{
"operation": "search",
"time": "53.954",
"value": "San Antonio Riders season 1991 l"
},
{
"operation": "Now on",
"time": "55.301",
"value": "undefined"
},
{
"operation": "search",
"time": "91.363",
"value": "San Antonio Riders 1991"
},
{
"operation": "Now on",
"time": "92.316",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "95.961",
"value": "Week 6"
},
{
"operation": "Now on",
"time": "97.254",
"value": "1991 Birmingham Fire season#Week 6"
},
{
"operation": "hyperlink",
"time": "99.292",
"value": "San Antonio Riders"
},
{
"operation": "Now on",
"time": "100.446",
"value": "San Antonio Riders"
},
{
"operation": "Highlighting",
"time": "101.586",
"value": "San Antonio Riders_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "104.63",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "104.822",
"value": "1991 Birmingham Fire season#Week 6"
},
{
"operation": "search",
"time": "262.753",
"value": "WLAF 1991"
},
{
"operation": "Now on",
"time": "264.494",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "272.357",
"value": "1991 Birmingham Fire season"
},
{
"operation": "Now on",
"time": "273.281",
"value": "1991 Birmingham Fire season"
},
{
"operation": "Page search",
"time": "291.16",
"value": "1992"
},
{
"operation": "search",
"time": "313.572",
"value": "WLAF league got six wins and four lost while in 1992 got three wins and seven lost"
},
{
"operation": "Now on",
"time": "315.404",
"value": "undefined"
},
{
"operation": "finish",
"time": "332.355",
"value": "finish"
}
]
}
|
exfever_5308
|
Rogue is a fictional character commonly associated with the fictional team of superheroes appearing in Marvel Comics.
|
{
"evidence": [
"Rogue (comics)",
"X-Men"
],
"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:\nRogue is a fictional character commonly associated with the fictional team of superheroes appearing in Marvel Comics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Rogue (comics)",
"X-Men"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5645,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29725
|
Jack Dull earned his B. A. at the University of Washington in 1955 and earned his doctorate in ten years after.
|
{
"evidence": [
"Jack Dull_sentence_1",
"Jack Dull_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:\nJack Dull earned his B. A. at the University of Washington in 1955 and earned his doctorate in ten years after.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Jack Dull_sentence_1",
"Jack Dull_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5646,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.654",
"value": "?search="
},
{
"operation": "search",
"time": "13.838",
"value": "Jack Dull"
},
{
"operation": "Now on",
"time": "14.771",
"value": "Jack Dull"
},
{
"operation": "Highlighting",
"time": "22.353",
"value": "Jack Dull_sentence_1"
},
{
"operation": "Highlighting",
"time": "31.829",
"value": "Jack Dull_sentence_2"
},
{
"operation": "finish",
"time": "62.45",
"value": "finish"
}
]
}
|
90373
|
The 1988 Brazilian Grand Prix was a Formula One motor race and is the highest class of international auto racing for single-seater racing cars.
|
{
"evidence": [
"1988 Brazilian Grand Prix_sentence_0",
"Formula One_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:\nThe 1988 Brazilian Grand Prix was a Formula One motor race and is the highest class of international auto racing for single-seater racing cars.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"1988 Brazilian Grand Prix_sentence_0",
"Formula One_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5647,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.91",
"value": "?search="
},
{
"operation": "search",
"time": "22.924",
"value": "1988 Brazilian Grand Prix"
},
{
"operation": "Now on",
"time": "25.048",
"value": "1988 Brazilian Grand Prix"
},
{
"operation": "Highlighting",
"time": "35.524",
"value": "1988 Brazilian Grand Prix_sentence_0"
},
{
"operation": "hyperlink",
"time": "122.209",
"value": "Formula One"
},
{
"operation": "Now on",
"time": "124.352",
"value": "Formula One"
},
{
"operation": "Highlighting",
"time": "128.911",
"value": "Formula One_sentence_2"
},
{
"operation": "back-button-clicked",
"time": "137.903",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "138.992",
"value": "1988 Brazilian Grand Prix"
},
{
"operation": "finish",
"time": "151.391",
"value": "finish"
}
]
}
|
60980
|
Cordia Harrington established the Bun Lady Scholarship, and was awarded the Beta Gamma Sigma Medallion for Entrepreneurship in 2013.
|
{
"evidence": [
"Cordia Harrington_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:\nCordia Harrington established the Bun Lady Scholarship, and was awarded the Beta Gamma Sigma Medallion for Entrepreneurship in 2013.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cordia Harrington_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5648,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.46",
"value": "?search="
},
{
"operation": "search",
"time": "83.792",
"value": "Cordia Harrington"
},
{
"operation": "Now on",
"time": "85.407",
"value": "Cordia Harrington"
},
{
"operation": "Page search",
"time": "100.865",
"value": "scho"
},
{
"operation": "Page search",
"time": "101.767",
"value": "scho"
},
{
"operation": "Page search",
"time": "101.952",
"value": "scho"
},
{
"operation": "Highlighting",
"time": "106.354",
"value": "Cordia Harrington_sentence_6"
},
{
"operation": "Page search",
"time": "112.161",
"value": "beta"
},
{
"operation": "Page search",
"time": "112.352",
"value": "beta"
},
{
"operation": "Page search",
"time": "112.512",
"value": "beta"
},
{
"operation": "Highlighting",
"time": "116.703",
"value": "Cordia Harrington_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "124.43",
"value": "Cordia Harrington_sentence_6"
},
{
"operation": "finish",
"time": "132.372",
"value": "finish"
}
]
}
|
10773
|
The World Trade Center Abu Dhabi has one Courtyard by Marriott hotel which is a brand of hotels owned by Marriott International that offers mid-priced range of hotels designed for business travelers.
|
{
"evidence": [
"World Trade Center Abu Dhabi_sentence_2",
"Courtyard by Marriott_sentence_0",
"Courtyard by Marriott_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:\nThe World Trade Center Abu Dhabi has one Courtyard by Marriott hotel which is a brand of hotels owned by Marriott International that offers mid-priced range of hotels designed for business travelers.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"World Trade Center Abu Dhabi_sentence_2",
"Courtyard by Marriott_sentence_0",
"Courtyard by Marriott_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5649,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.037",
"value": "?search="
},
{
"operation": "search",
"time": "9.312",
"value": "World Trade Center Abu Dhabi"
},
{
"operation": "Now on",
"time": "10.2",
"value": "World Trade Center Abu Dhabi"
},
{
"operation": "Highlighting",
"time": "17.619",
"value": "World Trade Center Abu Dhabi_sentence_2"
},
{
"operation": "hyperlink",
"time": "58.738",
"value": "Courtyard by Marriott"
},
{
"operation": "Now on",
"time": "59.48",
"value": "Courtyard by Marriott"
},
{
"operation": "Highlighting",
"time": "64.192",
"value": "Courtyard by Marriott_sentence_0"
},
{
"operation": "Highlighting",
"time": "68.803",
"value": "Courtyard by Marriott_sentence_1"
},
{
"operation": "finish",
"time": "84.061",
"value": "finish"
}
]
}
|
72723
|
A specie of mushroom has a cap and hymenium.
|
{
"evidence": [
"Hymenium_sentence_0",
"Hymenium_sentence_4"
],
"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:\nA specie of mushroom has a cap and hymenium.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hymenium_sentence_0",
"Hymenium_sentence_4"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5650,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.02",
"value": "?search="
},
{
"operation": "search",
"time": "12.33",
"value": "Hymenium"
},
{
"operation": "Now on",
"time": "13.023",
"value": "Hymenium"
},
{
"operation": "Highlighting",
"time": "143.908",
"value": "Hymenium_sentence_0"
},
{
"operation": "Highlighting",
"time": "197.525",
"value": "Hymenium_sentence_4"
},
{
"operation": "finish",
"time": "203.545",
"value": "finish"
}
]
}
|
39013
|
Steven Nitah was born in Canada, his profession is politics he ran in the general election which he was supported by Don Morin and vied to be elected again in 2003.
|
{
"evidence": [
"Steven Nitah_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:\nSteven Nitah was born in Canada, his profession is politics he ran in the general election which he was supported by Don Morin and vied to be elected again in 2003.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Steven Nitah_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5651,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.864",
"value": "?search="
},
{
"operation": "search",
"time": "13.985",
"value": "Steven Nitah"
},
{
"operation": "Now on",
"time": "14.8",
"value": "Steven Nitah"
},
{
"operation": "Highlighting",
"time": "50.561",
"value": "Steven Nitah_sentence_0"
},
{
"operation": "Highlighting",
"time": "70.767",
"value": "Steven Nitah_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "74.086",
"value": "Steven Nitah_sentence_2"
},
{
"operation": "search",
"time": "135.312",
"value": "politicians born in canada"
},
{
"operation": "Now on",
"time": "138.726",
"value": "undefined"
},
{
"operation": "finish",
"time": "184.451",
"value": "finish"
}
]
}
|
exfever_1039
|
Jared Corey Kushner is the son of the former real-estate developer Charles Kushner and married to an American businesswoman, and the daughter of the former President, that served as the 45th president of the United States from 2017 to 2021.
|
{
"evidence": [
"Jared Kushner",
"Ivanka Trump",
"Donald Trump"
],
"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:\nJared Corey Kushner is the son of the former real-estate developer Charles Kushner and married to an American businesswoman, and the daughter of the former President, that served as the 45th president of the United States from 2017 to 2021.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jared Kushner",
"Ivanka Trump",
"Donald Trump"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5652,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
86873
|
Before the war, Joseph Lyman studied law and commenced practice in Council Bluffs, Iowa; then initially served as deputy collector of internal revenue of the fifth district of Iowa from 1867 to 1870.
|
{
"evidence": [
"Joseph Lyman_sentence_13"
],
"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:\nBefore the war, Joseph Lyman studied law and commenced practice in Council Bluffs, Iowa; then initially served as deputy collector of internal revenue of the fifth district of Iowa from 1867 to 1870.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Joseph Lyman_sentence_13"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5653,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.18",
"value": "?search="
},
{
"operation": "search",
"time": "14.566",
"value": "Joseph Lyman"
},
{
"operation": "Now on",
"time": "15.393",
"value": "Joseph Lyman"
},
{
"operation": "Page search",
"time": "29.577",
"value": "practice"
},
{
"operation": "Highlighting",
"time": "34.787",
"value": "Joseph Lyman_sentence_13"
},
{
"operation": "finish",
"time": "47.216",
"value": "finish"
}
]
}
|
exfever_778
|
Michelle Obama is a former first lady of the United States who delivered a keynote address during her husband's campaign at the quadrennial presidential nominating convention of the Democratic Party held in Aspen, Colorado.
|
{
"evidence": [
"Michelle Obama",
"2008 Democratic National Convention"
],
"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:\nMichelle Obama is a former first lady of the United States who delivered a keynote address during her husband's campaign at the quadrennial presidential nominating convention of the Democratic Party held in Aspen, Colorado.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Michelle Obama",
"2008 Democratic National Convention"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5654,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
69117
|
John Lyng was a member of the Free-minded Liberal Party but changed to the Conservative Party he was also elected to the Norwegian Parliament
|
{
"evidence": [
"John Lyng_sentence_11",
"John Lyng_sentence_12",
"John Lyng_sentence_13"
],
"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:\nJohn Lyng was a member of the Free-minded Liberal Party but changed to the Conservative Party he was also elected to the Norwegian Parliament\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"John Lyng_sentence_11",
"John Lyng_sentence_12",
"John Lyng_sentence_13"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5655,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.679",
"value": "undefined"
},
{
"operation": "search",
"time": "7.891",
"value": "John Lyng"
},
{
"operation": "Now on",
"time": "8.995",
"value": "John Lyng"
},
{
"operation": "Highlighting",
"time": "20.243",
"value": "John Lyng_sentence_11"
},
{
"operation": "Highlighting",
"time": "30.057",
"value": "John Lyng_sentence_12"
},
{
"operation": "Highlighting",
"time": "34.789",
"value": "John Lyng_sentence_13"
},
{
"operation": "finish",
"time": "40.995",
"value": "finish"
}
]
}
|
exfever_3068
|
Billboard Dad is the first American direct-to-video comedy film produced by Warner Bros.
|
{
"evidence": [
"Billboard Dad",
"Alan Metter"
],
"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:\nBillboard Dad is the first American direct-to-video comedy film produced by Warner Bros.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Billboard Dad",
"Alan Metter"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5656,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21729
|
Dwight–Hooker Avenue Historic District is a national historic district (an area of 6 acres) located in New York, and includes 17 contributing residential buildings most of which were built between 1895 and 1815.
|
{
"evidence": [
"Dwight–Hooker Avenue Historic District_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:\nDwight–Hooker Avenue Historic District is a national historic district (an area of 6 acres) located in New York, and includes 17 contributing residential buildings most of which were built between 1895 and 1815.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dwight–Hooker Avenue Historic District_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5657,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.485",
"value": "?search="
},
{
"operation": "search",
"time": "478.805",
"value": "Dwight–Hooker Avenue Historic District"
},
{
"operation": "Now on",
"time": "480.82",
"value": "Dwight–Hooker Avenue Historic District"
},
{
"operation": "Highlighting",
"time": "482.29",
"value": "Dwight–Hooker Avenue Historic District_sentence_0"
},
{
"operation": "Highlighting",
"time": "485.853",
"value": "Dwight–Hooker Avenue Historic District_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "485.856",
"value": "Dwight–Hooker Avenue Historic District_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "493.97",
"value": "Dwight–Hooker Avenue Historic District_sentence_1"
},
{
"operation": "Highlighting",
"time": "500.128",
"value": "Dwight–Hooker Avenue Historic District_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "505.542",
"value": "Dwight–Hooker Avenue Historic District_sentence_0"
},
{
"operation": "Highlighting",
"time": "505.701",
"value": "Dwight–Hooker Avenue Historic District_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "506.232",
"value": "Dwight–Hooker Avenue Historic District_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "506.664",
"value": "Dwight–Hooker Avenue Historic District_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "508.158",
"value": "Dwight–Hooker Avenue Historic District_cell_0_2_1"
},
{
"operation": "Highlighting deleted",
"time": "508.162",
"value": "Dwight–Hooker Avenue Historic District_header_cell_0_2_0"
},
{
"operation": "finish",
"time": "517.314",
"value": "finish"
}
]
}
|
exfever_1717
|
Hugh Marston Hefner was the founder and editor-in-chief of an American men's lifestyle and entertainment magazine, formerly in print and currently online was founded in Chicago in 1953.
|
{
"evidence": [
"Hugh Hefner",
"Playboy"
],
"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:\nHugh Marston Hefner was the founder and editor-in-chief of an American men's lifestyle and entertainment magazine, formerly in print and currently online was founded in Chicago in 1953.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hugh Hefner",
"Playboy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5658,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2629
|
Afghanistan is a landlocked country located at the crossroads of the South and a subregion of Asia, which stretches from the Caspian Sea in the west to a country in East Asia that is the world's most populous country.
|
{
"evidence": [
"Afghanistan",
"Central Asia",
"China"
],
"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:\nAfghanistan is a landlocked country located at the crossroads of the South and a subregion of Asia, which stretches from the Caspian Sea in the west to a country in East Asia that is the world's most populous country.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Afghanistan",
"Central Asia",
"China"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5659,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7713
|
Tina Fey is an American comedian who is best known for creating the comedy series that?aired from October 11, 2006, to January 31, 2013.?
|
{
"evidence": [
"Tina Fey",
"30 Rock"
],
"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:\nTina Fey is an American comedian who is best known for creating the comedy series that?aired from October 11, 2006, to January 31, 2013.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tina Fey",
"30 Rock"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5660,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
5861
|
BLRT Grupp is a shipbuilding company headquartered in Estonia that also owns shipyards in a dozen other European countries.
|
{
"evidence": [
"BLRT Grupp_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:\nBLRT Grupp is a shipbuilding company headquartered in Estonia that also owns shipyards in a dozen other European countries.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"BLRT Grupp_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5661,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.556",
"value": "?search="
},
{
"operation": "search",
"time": "13.77",
"value": "BLRT Grupp"
},
{
"operation": "Now on",
"time": "14.62",
"value": "BLRT Grupp"
},
{
"operation": "Highlighting",
"time": "19.355",
"value": "BLRT Grupp_sentence_0"
},
{
"operation": "Highlighting",
"time": "26.251",
"value": "BLRT Grupp_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "42.688",
"value": "BLRT Grupp_sentence_0"
},
{
"operation": "finish",
"time": "50.663",
"value": "finish"
}
]
}
|
4535
|
The largest city in Shangqu is Yong Cheng City which also has the highest area and population density.
|
{
"evidence": [
"Provinces of China_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 largest city in Shangqu is Yong Cheng City which also has the highest area and population density.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Provinces of China_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5662,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "26.63",
"value": "yong Cheng city"
},
{
"operation": "hyperlink",
"time": "30.229",
"value": "young chang city"
},
{
"operation": "search",
"time": "45.1",
"value": "Shangqu"
},
{
"operation": "back-button-clicked",
"time": "49.524",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "50.004",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "50.006",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "51.828",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "51.83",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "52.7",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "52.702",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "52.704",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "54.748",
"value": "back-button-clicked"
},
{
"operation": "search",
"time": "59.451",
"value": "Chinese provinces"
},
{
"operation": "Highlighting",
"time": "81.211",
"value": "Provinces of China_sentence_0"
},
{
"operation": "finish",
"time": "87.146",
"value": "finish"
}
]
}
|
34772
|
Formyltetrahydrofolate dehydrogenase is an enzyme that catalyzes the chemical reaction and its three substrates are 10-formyltetrahydrofolate, NADP, and H2O.
|
{
"evidence": [
"Formyltetrahydrofolate dehydrogenase_sentence_0",
"Formyltetrahydrofolate dehydrogenase_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:\nFormyltetrahydrofolate dehydrogenase is an enzyme that catalyzes the chemical reaction and its three substrates are 10-formyltetrahydrofolate, NADP, and H2O.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Formyltetrahydrofolate dehydrogenase_sentence_0",
"Formyltetrahydrofolate dehydrogenase_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5663,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.831",
"value": "?search="
},
{
"operation": "search",
"time": "16.293",
"value": "Formyltetrahydrofolate dehydrogenase"
},
{
"operation": "Now on",
"time": "17.179",
"value": "Formyltetrahydrofolate dehydrogenase"
},
{
"operation": "Highlighting",
"time": "21.187",
"value": "Formyltetrahydrofolate dehydrogenase_sentence_0"
},
{
"operation": "Highlighting",
"time": "28.589",
"value": "Formyltetrahydrofolate dehydrogenase_sentence_1"
},
{
"operation": "finish",
"time": "58.396",
"value": "finish"
}
]
}
|
exfever_7601
|
TD Garden is named after its sponsor, TD Bank, a subsidiary of a Canadian multinational banking and financial services corporation headquartered in Hamilton, Ontario.
|
{
"evidence": [
"TD Garden",
"Toronto-Dominion Bank"
],
"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:\nTD Garden is named after its sponsor, TD Bank, a subsidiary of a Canadian multinational banking and financial services corporation headquartered in Hamilton, Ontario.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"TD Garden",
"Toronto-Dominion Bank"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5664,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7818
|
The Great Debaters is a 2007 American biographical drama film directed by and starring an American actor, director, and producer who is known for his performances on the screen and on stage.
|
{
"evidence": [
"The Great Debaters",
"Denzel Washington"
],
"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 Great Debaters is a 2007 American biographical drama film directed by and starring an American actor, director, and producer who is known for his performances on the screen and on stage.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Great Debaters",
"Denzel Washington"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5665,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
65737
|
Released daily in Vietnamese, Thanh Niên was the most circulated newspaper in Vietnam in 2009, with an average circulation of 300,000.
|
{
"evidence": [
"Thanh Niên_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:\nReleased daily in Vietnamese, Thanh Niên was the most circulated newspaper in Vietnam in 2009, with an average circulation of 300,000.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Thanh Niên_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5666,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.071",
"value": "?search="
},
{
"operation": "search",
"time": "11.912",
"value": "Thanh Niên"
},
{
"operation": "Now on",
"time": "12.738",
"value": "Thanh Niên"
},
{
"operation": "Highlighting",
"time": "18.625",
"value": "Thanh Niên_sentence_2"
},
{
"operation": "Highlighting",
"time": "22.223",
"value": "Thanh Niên_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "27.969",
"value": "Thanh Niên_sentence_2"
},
{
"operation": "finish",
"time": "39.824",
"value": "finish"
}
]
}
|
exfever_2509
|
Kim Basinger's third film is Never Say Never Again.
|
{
"evidence": [
"Kim Basinger",
"Vicki Vale",
"Batman"
],
"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:\nKim Basinger's third film is Never Say Never Again.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kim Basinger",
"Vicki Vale",
"Batman"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5667,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_370
|
Lisa Olivia also appeared in National Lampoon's Strip Poker.
|
{
"evidence": [
"Olivia Munn",
"Aaron Sorkin"
],
"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:\nLisa Olivia also appeared in National Lampoon's Strip Poker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Olivia Munn",
"Aaron Sorkin"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5668,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21854
|
Nuremberg principles' fourth principle is legally supported by the jurisprudence found in certain articles in the Universal Declaration of Human Rights which deal indirectly with conscientious objection and is also supported by the principles found in paragraph 171 of the Handbook on Procedures and Criteria for Determining Refugee Status.
|
{
"evidence": [
"Nuremberg principles_sentence_10",
"Nuremberg principles_sentence_11"
],
"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:\nNuremberg principles' fourth principle is legally supported by the jurisprudence found in certain articles in the Universal Declaration of Human Rights which deal indirectly with conscientious objection and is also supported by the principles found in paragraph 171 of the Handbook on Procedures and Criteria for Determining Refugee Status.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Nuremberg principles_sentence_10",
"Nuremberg principles_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5669,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.699",
"value": "?search="
},
{
"operation": "search",
"time": "9.76",
"value": "Nuremberg principles"
},
{
"operation": "Now on",
"time": "10.293",
"value": "Nuremberg principles"
},
{
"operation": "Page search",
"time": "50.061",
"value": "fourth"
},
{
"operation": "hyperlink",
"time": "66.021",
"value": "Superior orders"
},
{
"operation": "Now on",
"time": "66.677",
"value": "Superior orders"
},
{
"operation": "hyperlink",
"time": "93.242",
"value": "Nuremberg Principle IV"
},
{
"operation": "Now on",
"time": "93.912",
"value": "Nuremberg Principles#Principle IV"
},
{
"operation": "Highlighting",
"time": "106.476",
"value": "Nuremberg principles_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "107.134",
"value": "Nuremberg principles_sentence_6"
},
{
"operation": "hyperlink",
"time": "126.746",
"value": "superior orders"
},
{
"operation": "Now on",
"time": "127.068",
"value": "Superior orders"
},
{
"operation": "back-button-clicked",
"time": "137.51",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "137.571",
"value": "Nuremberg principles#Principle IV"
},
{
"operation": "Highlighting",
"time": "168.973",
"value": "Nuremberg principles_sentence_10"
},
{
"operation": "Highlighting",
"time": "184.669",
"value": "Nuremberg principles_sentence_11"
},
{
"operation": "finish",
"time": "201.838",
"value": "finish"
}
]
}
|
50705
|
"For Loving You" is Jan Howard's song with Bill Anderson that was released in 1967 but never did well in the US Charts.
|
{
"evidence": [
"For Loving You (album)_sentence_15",
"For Loving You (album)_sentence_16"
],
"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:\n\"For Loving You\" is Jan Howard's song with Bill Anderson that was released in 1967 but never did well in the US Charts.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"For Loving You (album)_sentence_15",
"For Loving You (album)_sentence_16"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5670,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.127",
"value": "?search="
},
{
"operation": "search",
"time": "10.587",
"value": "For Loving You (album)"
},
{
"operation": "Now on",
"time": "11.548",
"value": "For Loving You (album)"
},
{
"operation": "Page search",
"time": "34.866",
"value": "howard"
},
{
"operation": "search",
"time": "67.658",
"value": "Contains...For Loving You"
},
{
"operation": "Now on",
"time": "70.218",
"value": "undefined"
},
{
"operation": "search",
"time": "95.891",
"value": "For Loving You (single)"
},
{
"operation": "Now on",
"time": "97.286",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "106.73",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "106.944",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "107.914",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "107.925",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "108.097",
"value": "For Loving You (album)"
},
{
"operation": "Page search",
"time": "137.36",
"value": "1967"
},
{
"operation": "Highlighting",
"time": "142.578",
"value": "For Loving You (album)_sentence_3"
},
{
"operation": "Highlighting",
"time": "142.579",
"value": "For Loving You (album)_section_0"
},
{
"operation": "Highlighting deleted",
"time": "163.697",
"value": "For Loving You (album)_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "163.701",
"value": "For Loving You (album)_section_0"
},
{
"operation": "Highlighting",
"time": "183.552",
"value": "For Loving You (album)_sentence_15"
},
{
"operation": "Highlighting",
"time": "188.913",
"value": "For Loving You (album)_sentence_16"
},
{
"operation": "finish",
"time": "217.954",
"value": "finish"
}
]
}
|
exfever_5731
|
Omar Sharif was an Egyptian actor who played in the 1962 British epic historical drama film based on the autobiographical account of the experiences of T. E. Lawrence.
|
{
"evidence": [
"Omar Sharif",
"Lawrence of Arabia (film)",
"Seven Pillars of Wisdom"
],
"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:\nOmar Sharif was an Egyptian actor who played in the 1962 British epic historical drama film based on the autobiographical account of the experiences of T. E. Lawrence.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Omar Sharif",
"Lawrence of Arabia (film)",
"Seven Pillars of Wisdom"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5671,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7761
|
Alicia Augello Cook is an American singer who has sold eight million copies of her second album, which was recorded at several recording studios, and production was handled primarily by Keys with contributions from Kanye West and "Krucial", an American record producer, executive producer, songwriter, and hip hop artist.
|
{
"evidence": [
"Alicia Keys",
"The Diary of Alicia Keys",
"Kerry Brothers Jr."
],
"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:\nAlicia Augello Cook is an American singer who has sold eight million copies of her second album, which was recorded at several recording studios, and production was handled primarily by Keys with contributions from Kanye West and \"Krucial\", an American record producer, executive producer, songwriter, and hip hop artist.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alicia Keys",
"The Diary of Alicia Keys",
"Kerry Brothers Jr."
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5672,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
20174
|
The Mysterious Island published by Hetzel in 1875 is a crossover sequel to Verne's famous Twenty Thousand Leagues Under the Seas and In Search of the Castaways but initially was never rejected by any publisher.
|
{
"evidence": [
"The Mysterious Island_sentence_1",
"The Mysterious Island_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:\nThe Mysterious Island published by Hetzel in 1875 is a crossover sequel to Verne's famous Twenty Thousand Leagues Under the Seas and In Search of the Castaways but initially was never rejected by any publisher.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"The Mysterious Island_sentence_1",
"The Mysterious Island_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5673,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.061",
"value": "?search="
},
{
"operation": "search",
"time": "24.974",
"value": "Mysterious Island"
},
{
"operation": "Now on",
"time": "26.641",
"value": "Mysterious Island"
},
{
"operation": "Highlighting",
"time": "62.359",
"value": "The Mysterious Island_sentence_1"
},
{
"operation": "Highlighting",
"time": "69.318",
"value": "The Mysterious Island_sentence_2"
},
{
"operation": "Highlighting",
"time": "85.706",
"value": "The Mysterious Island_sentence_3"
},
{
"operation": "Page search",
"time": "96.061",
"value": "rejested"
},
{
"operation": "Page search",
"time": "116.156",
"value": "rejescted"
},
{
"operation": "Page search",
"time": "118.852",
"value": "rejected"
},
{
"operation": "Highlighting deleted",
"time": "200.148",
"value": "The Mysterious Island_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "200.976",
"value": "The Mysterious Island_sentence_3"
},
{
"operation": "Highlighting",
"time": "206.404",
"value": "The Mysterious Island_sentence_4"
},
{
"operation": "finish",
"time": "243.813",
"value": "finish"
}
]
}
|
exfever_5686
|
Legally Blonde stars an American actor whose television credits include an American comedy-drama television series.
|
{
"evidence": [
"Legally Blonde",
"Luke Wilson",
"Enlightened (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:\nLegally Blonde stars an American actor whose television credits include an American comedy-drama television series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Legally Blonde",
"Luke Wilson",
"Enlightened (TV series)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5674,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8389
|
Sophie Turner is an actress who appeared in an American action comedy film with a film distribution strategy of releasing in a few theaters.
|
{
"evidence": [
"Sophie Turner",
"Barely Lethal",
"Limited release"
],
"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:\nSophie Turner is an actress who appeared in an American action comedy film with a film distribution strategy of releasing in a few theaters.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sophie Turner",
"Barely Lethal",
"Limited release"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5675,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
37458
|
All solid waste policy around the world is aimed at developing and implementing proper mechanisms to effectively manage solid waste, which, generally, refers to non-hazardous waste and does not include solid or dissolved materials in domestic sewage, solid or dissolved materials in irrigation return flows, or industrial discharges.
|
{
"evidence": [
"Solid waste policy in the United States_sentence_0",
"Solid waste policy in the United States_sentence_9",
"Solid waste policy in the United States_sentence_7"
],
"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:\nAll solid waste policy around the world is aimed at developing and implementing proper mechanisms to effectively manage solid waste, which, generally, refers to non-hazardous waste and does not include solid or dissolved materials in domestic sewage, solid or dissolved materials in irrigation return flows, or industrial discharges.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Solid waste policy in the United States_sentence_0",
"Solid waste policy in the United States_sentence_9",
"Solid waste policy in the United States_sentence_7"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5676,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.169",
"value": "?search="
},
{
"operation": "search",
"time": "25.542",
"value": "olid waste policy around the world is aimed at developing and implementing proper mechanisms to effectively manage solid waste"
},
{
"operation": "Now on",
"time": "27.464",
"value": "undefined"
},
{
"operation": "search",
"time": "41.433",
"value": "solid waste policy around the world is aimed at developing and implementing proper mechanisms to effectively manage solid waste"
},
{
"operation": "Now on",
"time": "42.924",
"value": "undefined"
},
{
"operation": "search",
"time": "86.768",
"value": "solid waste"
},
{
"operation": "Now on",
"time": "87.981",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "101.226",
"value": "List of solid waste treatment technologies"
},
{
"operation": "Now on",
"time": "101.923",
"value": "List of solid waste treatment technologies"
},
{
"operation": "search",
"time": "138.534",
"value": "refers to non-hazardous waste and does not include solid or dissolved materials"
},
{
"operation": "Now on",
"time": "141.348",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "166.815",
"value": "Solid waste policy in the United States"
},
{
"operation": "Now on",
"time": "168.118",
"value": "Solid waste policy in the United States"
},
{
"operation": "Highlighting",
"time": "177.725",
"value": "Solid waste policy in the United States_sentence_0"
},
{
"operation": "Page search",
"time": "201.82",
"value": "hazar"
},
{
"operation": "Highlighting",
"time": "216.746",
"value": "Solid waste policy in the United States_sentence_9"
},
{
"operation": "Page search",
"time": "249.532",
"value": "disso"
},
{
"operation": "Highlighting",
"time": "256.115",
"value": "Solid waste policy in the United States_sentence_7"
},
{
"operation": "Page search",
"time": "292.17",
"value": "world"
},
{
"operation": "search",
"time": "323.43",
"value": "\"solid\" \"waste\" \"policy\" \"world\""
},
{
"operation": "Now on",
"time": "324.341",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "367.912",
"value": "Environmental law"
},
{
"operation": "Now on",
"time": "369.569",
"value": "Environmental law"
},
{
"operation": "Page search",
"time": "375.582",
"value": "world"
},
{
"operation": "hyperlink",
"time": "533.059",
"value": "List of international environmental agreements"
},
{
"operation": "Now on",
"time": "534.621",
"value": "List of international environmental agreements"
},
{
"operation": "search",
"time": "566.613",
"value": "solid waste policy around the world"
},
{
"operation": "Now on",
"time": "568.946",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "587.635",
"value": "Around the world Environmental law_section_24"
},
{
"operation": "Now on",
"time": "588.346",
"value": "Environmental law#Around the world Environmental law section 24"
},
{
"operation": "back-button-clicked",
"time": "610.059",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "610.959",
"value": "undefined"
},
{
"operation": "finish",
"time": "700.519",
"value": "finish"
}
]
}
|
exfever_13
|
Jackie is the first dramatic film directed by Pablo Larraín.
|
{
"evidence": [
"Jackie (2016 film)",
"Noah Oppenheim"
],
"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:\nJackie is the first dramatic film directed by Pablo Larraín.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jackie (2016 film)",
"Noah Oppenheim"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5677,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2010
|
Akira Toriyama is the character designer for several popular video games, such as a franchise of role-playing video games created by a Japanese video game designer.
|
{
"evidence": [
"Akira Toriyama",
"Dragon Quest",
"Yuji Horii"
],
"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:\nAkira Toriyama is the character designer for several popular video games, such as a franchise of role-playing video games created by a Japanese video game designer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Akira Toriyama",
"Dragon Quest",
"Yuji Horii"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5678,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
12315
|
Sweden's Jonas Sandqvist is the only player that made 26 appearances in Allsvenkan league, a Swedish professional league for men's association football clubs.
|
{
"evidence": [
"Jonas Sandqvist_sentence_0",
"Allsvenskan_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:\nSweden's Jonas Sandqvist is the only player that made 26 appearances in Allsvenkan league, a Swedish professional league for men's association football clubs.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Jonas Sandqvist_sentence_0",
"Allsvenskan_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5679,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.549",
"value": "?search="
},
{
"operation": "search",
"time": "18.073",
"value": "Contains...Allsvenkan league"
},
{
"operation": "Now on",
"time": "18.526",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "35.229",
"value": "allsvenskan league"
},
{
"operation": "Now on",
"time": "36.253",
"value": "undefined"
},
{
"operation": "search",
"time": "57.711",
"value": "Jonas Sandqvist"
},
{
"operation": "Now on",
"time": "58.401",
"value": "Jonas Sandqvist"
},
{
"operation": "Highlighting",
"time": "64.696",
"value": "Jonas Sandqvist_sentence_0"
},
{
"operation": "hyperlink",
"time": "97.682",
"value": "Allsvenskan"
},
{
"operation": "Now on",
"time": "98.552",
"value": "Allsvenskan"
},
{
"operation": "Highlighting",
"time": "112.451",
"value": "Allsvenskan_sentence_2"
},
{
"operation": "hyperlink",
"time": "156.752",
"value": "List of Allsvenskan players"
},
{
"operation": "Now on",
"time": "157.96",
"value": "List of Allsvenskan players"
},
{
"operation": "search",
"time": "195.321",
"value": "Contains...26 appearances"
},
{
"operation": "Now on",
"time": "196.956",
"value": "undefined"
},
{
"operation": "search",
"time": "224.933",
"value": "Sandqvist 26 appearances"
},
{
"operation": "Now on",
"time": "225.801",
"value": "undefined"
},
{
"operation": "finish",
"time": "252.834",
"value": "finish"
}
]
}
|
68399
|
\American synchronized swimmer, aquatic artist, choreographer and author Beulah Gundling designed her own costumes and won multiple competitions in North America.
|
{
"evidence": [
"Beulah Gundling_sentence_0",
"Beulah Gundling_sentence_10",
"Beulah Gundling_sentence_11"
],
"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:\n\\American synchronized swimmer, aquatic artist, choreographer and author Beulah Gundling designed her own costumes and won multiple competitions in North America.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Beulah Gundling_sentence_0",
"Beulah Gundling_sentence_10",
"Beulah Gundling_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5680,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "4.234",
"value": "?search="
},
{
"operation": "search",
"time": "10.493",
"value": "Beulah Gundling"
},
{
"operation": "Now on",
"time": "11.778",
"value": "Beulah Gundling"
},
{
"operation": "Highlighting",
"time": "13.284",
"value": "Beulah Gundling_sentence_0"
},
{
"operation": "Highlighting",
"time": "18.845",
"value": "Beulah Gundling_sentence_10"
},
{
"operation": "Highlighting",
"time": "19.663",
"value": "Beulah Gundling_sentence_11"
},
{
"operation": "finish",
"time": "34.252",
"value": "finish"
}
]
}
|
39243
|
In 2018, Impala (consisting of guitarist John Stivers, saxophonist Justin Thompson, drummer Jeff Goggans, and Bomar on bass) released "In the Late Hours" on Bomar's Electraphonic Recording imprint; after touring garage rock festivals following the release of Kings of the Strip.
|
{
"evidence": [
"Scott Bomar_sentence_4",
"Scott Bomar_sentence_6",
"Scott Bomar_sentence_9"
],
"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 2018, Impala (consisting of guitarist John Stivers, saxophonist Justin Thompson, drummer Jeff Goggans, and Bomar on bass) released \"In the Late Hours\" on Bomar's Electraphonic Recording imprint; after touring garage rock festivals following the release of Kings of the Strip.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Scott Bomar_sentence_4",
"Scott Bomar_sentence_6",
"Scott Bomar_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5681,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.687",
"value": "?search="
},
{
"operation": "search",
"time": "13.495",
"value": "Impala"
},
{
"operation": "Now on",
"time": "14.413",
"value": "Impala"
},
{
"operation": "search",
"time": "32.97",
"value": "Contains...In the Late Hours"
},
{
"operation": "Now on",
"time": "36.848",
"value": "undefined"
},
{
"operation": "search",
"time": "57.205",
"value": "Contains...In the Late Hours impala"
},
{
"operation": "Now on",
"time": "58.167",
"value": "undefined"
},
{
"operation": "search",
"time": "85.905",
"value": "Contains...Kings of the Strip"
},
{
"operation": "Now on",
"time": "87.331",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "98.337",
"value": "Scott Bomar"
},
{
"operation": "Now on",
"time": "98.839",
"value": "Scott Bomar"
},
{
"operation": "Highlighting",
"time": "113.577",
"value": "Scott Bomar_sentence_4"
},
{
"operation": "Highlighting",
"time": "130.639",
"value": "Scott Bomar_sentence_6"
},
{
"operation": "Highlighting",
"time": "145.686",
"value": "Scott Bomar_sentence_9"
},
{
"operation": "finish",
"time": "171.285",
"value": "finish"
}
]
}
|
11613
|
During the Haldi Kumkum (a social gathering in India), married men invite friends, relatives and new acquaintances to meet where the hostess distributes bangles, sweets, and other small novelties.
|
{
"evidence": [
"Haldi Kumkum_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:\nDuring the Haldi Kumkum (a social gathering in India), married men invite friends, relatives and new acquaintances to meet where the hostess distributes bangles, sweets, and other small novelties.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Haldi Kumkum_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5682,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.504",
"value": "?search="
},
{
"operation": "search",
"time": "20.727",
"value": "Haldi Kumkum"
},
{
"operation": "Now on",
"time": "21.769",
"value": "Haldi Kumkum"
},
{
"operation": "Highlighting",
"time": "25.189",
"value": "Haldi Kumkum_sentence_0"
},
{
"operation": "finish",
"time": "39.106",
"value": "finish"
}
]
}
|
exfever_8008
|
The Olympic Games are normally held every five years, alternating between a major international multi-sport event held for sports practiced on snow and ice and the Summer Olympics.
|
{
"evidence": [
"Olympic Games",
"Winter Olympic Games"
],
"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 Olympic Games are normally held every five years, alternating between a major international multi-sport event held for sports practiced on snow and ice and the Summer Olympics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Olympic Games",
"Winter Olympic Games"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5683,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.