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 |
|---|---|---|---|---|---|---|---|---|
74579
|
Deon Kipping released his Billboard ranking album, I Just Want to Hear You, in September 4, 2012 under Verity Records, an American gospel music-focused record label.
|
{
"evidence": [
"Deon Kipping_sentence_9",
"Deon Kipping_sentence_10",
"Deon Kipping_sentence_11",
"Verity Records_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDeon Kipping released his Billboard ranking album, I Just Want to Hear You, in September 4, 2012 under Verity Records, an American gospel music-focused record label.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Deon Kipping_sentence_9",
"Deon Kipping_sentence_10",
"Deon Kipping_sentence_11",
"Verity Records_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 284,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.347",
"value": "?search="
},
{
"operation": "search",
"time": "9.704",
"value": "Deon Kipping"
},
{
"operation": "Now on",
"time": "10.725",
"value": "Deon Kipping"
},
{
"operation": "Highlighting",
"time": "28.667",
"value": "Deon Kipping_sentence_9"
},
{
"operation": "Highlighting",
"time": "30.249",
"value": "Deon Kipping_sentence_10"
},
{
"operation": "Highlighting",
"time": "31.558",
"value": "Deon Kipping_sentence_11"
},
{
"operation": "hyperlink",
"time": "46.148",
"value": "Verity Records"
},
{
"operation": "Now on",
"time": "46.66",
"value": "Verity Records"
},
{
"operation": "Highlighting",
"time": "47.901",
"value": "Verity Records_sentence_0"
},
{
"operation": "finish",
"time": "55.338",
"value": "finish"
}
]
}
|
33564
|
Nor Sanavongsay (born on February 28, 1975) initially resettled with his parents, one brother, and one sister in the town of Kingsport, Tennessee in 1979, when the Sanavongsay family came to the United States in the aftermath of the Laotian Civil War.
|
{
"evidence": [
"Nor Sanavongsay_sentence_0",
"Nor Sanavongsay_sentence_1",
"Nor Sanavongsay_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:\nNor Sanavongsay (born on February 28, 1975) initially resettled with his parents, one brother, and one sister in the town of Kingsport, Tennessee in 1979, when the Sanavongsay family came to the United States in the aftermath of the Laotian Civil War.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Nor Sanavongsay_sentence_0",
"Nor Sanavongsay_sentence_1",
"Nor Sanavongsay_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 285,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.711",
"value": "?search="
},
{
"operation": "search",
"time": "15.632",
"value": "Nor Sanavongsay"
},
{
"operation": "Now on",
"time": "17.099",
"value": "Nor Sanavongsay"
},
{
"operation": "Highlighting",
"time": "21.63",
"value": "Nor Sanavongsay_sentence_0"
},
{
"operation": "Highlighting",
"time": "32.33",
"value": "Nor Sanavongsay_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "35.685",
"value": "Nor Sanavongsay_sentence_0"
},
{
"operation": "Highlighting",
"time": "43.915",
"value": "Nor Sanavongsay_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "66.987",
"value": "Nor Sanavongsay_sentence_2"
},
{
"operation": "Highlighting",
"time": "67.527",
"value": "Nor Sanavongsay_sentence_1"
},
{
"operation": "Highlighting",
"time": "68.087",
"value": "Nor Sanavongsay_sentence_2"
},
{
"operation": "finish",
"time": "93.159",
"value": "finish"
}
]
}
|
exfever_7724
|
SpongeBob SquarePants is an Canadian comedy television series made using animation techniques with a common series title, usually related to one another.
|
{
"evidence": [
"SpongeBob SquarePants",
"Animated series"
],
"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:\nSpongeBob SquarePants is an Canadian comedy television series made using animation techniques with a common series title, usually related to one another.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"SpongeBob SquarePants",
"Animated series"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 286,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
9384
|
Ifor Bowen Lloyd was educated at Winchester and Exeter College, Oxford where he was a member of the Library Committee of the Oxford Union Society, as well as, received a BA in Modern History in 1924.
|
{
"evidence": [
"Ifor Bowen Lloyd_sentence_3",
"Ifor Bowen Lloyd_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:\nIfor Bowen Lloyd was educated at Winchester and Exeter College, Oxford where he was a member of the Library Committee of the Oxford Union Society, as well as, received a BA in Modern History in 1924.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ifor Bowen Lloyd_sentence_3",
"Ifor Bowen Lloyd_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 287,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.557",
"value": "?search="
},
{
"operation": "search",
"time": "6.181",
"value": "Ifor Bowen Lloyd"
},
{
"operation": "Now on",
"time": "6.855",
"value": "Ifor Bowen Lloyd"
},
{
"operation": "Highlighting",
"time": "11.883",
"value": "Ifor Bowen Lloyd_sentence_3"
},
{
"operation": "Highlighting",
"time": "27.784",
"value": "Ifor Bowen Lloyd_sentence_4"
},
{
"operation": "finish",
"time": "61.063",
"value": "finish"
}
]
}
|
exfever_7957
|
Steven Victor Tallarico is a record producer.
|
{
"evidence": [
"Steven Tyler",
"Dream On (Aerosmith song)",
"Aerosmith (album)"
],
"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:\nSteven Victor Tallarico is a record producer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Steven Tyler",
"Dream On (Aerosmith song)",
"Aerosmith (album)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 288,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4358
|
The human condition is all of the characteristics and key events that compose the essentials of animal existence, which has been pondered and analyzed from many perspectives, including the study of general and fundamental questions.
|
{
"evidence": [
"Human condition",
"Philosophy"
],
"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 human condition is all of the characteristics and key events that compose the essentials of animal existence, which has been pondered and analyzed from many perspectives, including the study of general and fundamental questions.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Human condition",
"Philosophy"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 289,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
80725
|
Released on 1 November 2006, Noël, the debut album from Japanese singer, Ayaka, peaked at number one on the Weekly Charts of the Oricon Sales Chart, with debut sales of 350,580 units.
|
{
"evidence": [
"First Message_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:\nReleased on 1 November 2006, Noël, the debut album from Japanese singer, Ayaka, peaked at number one on the Weekly Charts of the Oricon Sales Chart, with debut sales of 350,580 units.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"First Message_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 290,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.323",
"value": "?search="
},
{
"operation": "search",
"time": "20.3",
"value": "Noël"
},
{
"operation": "Now on",
"time": "21.244",
"value": "Noel"
},
{
"operation": "search",
"time": "55.169",
"value": "Ayaka"
},
{
"operation": "Now on",
"time": "56.02",
"value": "undefined"
},
{
"operation": "search",
"time": "81.255",
"value": "Noel Ayaka"
},
{
"operation": "Now on",
"time": "82.181",
"value": "undefined"
},
{
"operation": "search",
"time": "109.649",
"value": "Contains...Noël, the debut album from Japanese singer, Ayaka"
},
{
"operation": "Now on",
"time": "110.653",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "139.187",
"value": "noel the debut album from japanese singer ayala"
},
{
"operation": "Now on",
"time": "139.929",
"value": "undefined"
},
{
"operation": "search",
"time": "153.846",
"value": "Noel Album"
},
{
"operation": "Now on",
"time": "155.153",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "170.319",
"value": "Noel"
},
{
"operation": "Now on",
"time": "170.866",
"value": "Noel"
},
{
"operation": "hyperlink",
"time": "178.905",
"value": "4.1 Music Noel_section_4"
},
{
"operation": "search",
"time": "213.824",
"value": "japanese singer ayaka"
},
{
"operation": "Now on",
"time": "214.902",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "221.296",
"value": "Ayaka Sasaki"
},
{
"operation": "Now on",
"time": "222.248",
"value": "Ayaka Sasaki"
},
{
"operation": "hyperlink",
"time": "229.035",
"value": "2 Discography Ayaka Sasaki_section_1"
},
{
"operation": "back-button-clicked",
"time": "270.053",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "270.724",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "276.85",
"value": "First Message"
},
{
"operation": "Now on",
"time": "277.637",
"value": "First Message"
},
{
"operation": "Highlighting",
"time": "288.806",
"value": "First Message_sentence_0"
},
{
"operation": "finish",
"time": "332.67",
"value": "finish"
}
]
}
|
exfever_5096
|
Futurama is an American animated science fiction sitcom created by an American cartoonist, writer, producer, and animator who created the comic strip Life in Hell.
|
{
"evidence": [
"Futurama",
"Matt Groening"
],
"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:\nFuturama is an American animated science fiction sitcom created by an American cartoonist, writer, producer, and animator who created the comic strip Life in Hell.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Futurama",
"Matt Groening"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 291,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
20722
|
In 2013, Vandana Vishwas released a music album called 'Monologues', and released another album called 'Parallels' in 2017.
|
{
"evidence": [
"Vandana Vishwas_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:\nIn 2013, Vandana Vishwas released a music album called 'Monologues', and released another album called 'Parallels' in 2017.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Vandana Vishwas_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 292,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.179",
"value": "?search="
},
{
"operation": "search",
"time": "7.976",
"value": "Vandana Vishwas"
},
{
"operation": "Now on",
"time": "8.962",
"value": "Vandana Vishwas"
},
{
"operation": "Highlighting",
"time": "20.749",
"value": "Vandana Vishwas_sentence_5"
},
{
"operation": "Highlighting",
"time": "27.671",
"value": "Vandana Vishwas_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "30.65",
"value": "Vandana Vishwas_sentence_5"
},
{
"operation": "finish",
"time": "47.27",
"value": "finish"
}
]
}
|
88661
|
NADPH—hemoprotein reductase belongs to the family of oxidoreductases and has more than three cofactors.
|
{
"evidence": [
"NADPH—hemoprotein reductase_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:\nNADPH—hemoprotein reductase belongs to the family of oxidoreductases and has more than three cofactors.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"NADPH—hemoprotein reductase_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 293,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.067",
"value": "?search="
},
{
"operation": "search",
"time": "10.442",
"value": "NADPH—hemoprotein reductase"
},
{
"operation": "Now on",
"time": "11.298",
"value": "NADPH—hemoprotein reductase"
},
{
"operation": "Highlighting",
"time": "14.495",
"value": "NADPH—hemoprotein reductase_sentence_2"
},
{
"operation": "Highlighting",
"time": "23.247",
"value": "NADPH—hemoprotein reductase_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "24.238",
"value": "NADPH—hemoprotein reductase_sentence_2"
},
{
"operation": "finish",
"time": "27.717",
"value": "finish"
}
]
}
|
15862
|
The 1928 Summa, an S-type asteroid, was discovered on September 21, 1938 and is named after the Battle of Summa.
|
{
"evidence": [
"1928 Summa_sentence_3",
"1928 Summa_sentence_1",
"1928 Summa_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 1928 Summa, an S-type asteroid, was discovered on September 21, 1938 and is named after the Battle of Summa.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"1928 Summa_sentence_3",
"1928 Summa_sentence_1",
"1928 Summa_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 294,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.996",
"value": "?search="
},
{
"operation": "search",
"time": "12.861",
"value": "1928 Summa"
},
{
"operation": "Now on",
"time": "13.662",
"value": "1928 Summa"
},
{
"operation": "Highlighting",
"time": "27.215",
"value": "1928 Summa_sentence_3"
},
{
"operation": "Highlighting",
"time": "32.213",
"value": "1928 Summa_sentence_1"
},
{
"operation": "Highlighting",
"time": "37.273",
"value": "1928 Summa_sentence_2"
},
{
"operation": "finish",
"time": "75.126",
"value": "finish"
}
]
}
|
47004
|
San Diego Padres in the California Sports Hall of Fame include Steve Garvey, an American former professional baseball player born in 1948, Tony Gwynn, Dave WInfield and Fernando Valenzuela.
|
{
"evidence": [
"Steve Garvey_sentence_1",
"Tony Gwynn_sentence_2",
"Tony Gwynn_sentence_458"
],
"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:\nSan Diego Padres in the California Sports Hall of Fame include Steve Garvey, an American former professional baseball player born in 1948, Tony Gwynn, Dave WInfield and Fernando Valenzuela.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Steve Garvey_sentence_1",
"Tony Gwynn_sentence_2",
"Tony Gwynn_sentence_458"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 295,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.228",
"value": "?search="
},
{
"operation": "search",
"time": "17.538",
"value": "San Diego Padres in the California Sports Hall of Fame"
},
{
"operation": "Now on",
"time": "19.856",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "75.274",
"value": "San Diego Hall of Champions San Diego Padres_section_16"
},
{
"operation": "Now on",
"time": "76.205",
"value": "San Diego Padres#San Diego Hall of Champions San Diego Padres section 16"
},
{
"operation": "Highlighting",
"time": "89.226",
"value": "San Diego Padres_cell_4_0_0"
},
{
"operation": "Highlighting",
"time": "89.227",
"value": "San Diego Padres_title"
},
{
"operation": "Highlighting",
"time": "89.228",
"value": "San Diego Padres_section_9"
},
{
"operation": "Highlighting",
"time": "89.228",
"value": "San Diego Padres_section_15"
},
{
"operation": "Highlighting deleted",
"time": "93.321",
"value": "San Diego Padres_cell_4_0_0"
},
{
"operation": "Highlighting deleted",
"time": "93.322",
"value": "San Diego Padres_title"
},
{
"operation": "Highlighting deleted",
"time": "93.324",
"value": "San Diego Padres_section_9"
},
{
"operation": "Highlighting deleted",
"time": "93.325",
"value": "San Diego Padres_section_15"
},
{
"operation": "search",
"time": "108.528",
"value": "California Sports Hall of Fame"
},
{
"operation": "Now on",
"time": "110.381",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "120.897",
"value": "Ben Lyons"
},
{
"operation": "Now on",
"time": "121.65",
"value": "Ben Lyons"
},
{
"operation": "back-button-clicked",
"time": "132.986",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "133.685",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "139.96",
"value": "Christian Okoye"
},
{
"operation": "Now on",
"time": "140.699",
"value": "Christian Okoye"
},
{
"operation": "back-button-clicked",
"time": "162.217",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "162.441",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "162.443",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "162.578",
"value": "San Diego Padres#San Diego Hall of Champions San Diego Padres section 16"
},
{
"operation": "hyperlink",
"time": "217.176",
"value": "Baseball Hall of Fame"
},
{
"operation": "Now on",
"time": "218.017",
"value": "National Baseball Hall of Fame and Museum"
},
{
"operation": "back-button-clicked",
"time": "219.913",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "220.044",
"value": "San Diego Padres#San Diego Hall of Champions San Diego Padres section 16"
},
{
"operation": "Page search",
"time": "243.668",
"value": "California Sports Hall of Fame"
},
{
"operation": "Page search",
"time": "260.17",
"value": "steve garvey"
},
{
"operation": "search",
"time": "295.703",
"value": "Steve Garvey"
},
{
"operation": "Now on",
"time": "296.665",
"value": "Steve Garvey"
},
{
"operation": "Highlighting",
"time": "303.961",
"value": "Steve Garvey_sentence_1"
},
{
"operation": "Page search",
"time": "332.552",
"value": "california"
},
{
"operation": "search",
"time": "402.175",
"value": "Tony Gwynn"
},
{
"operation": "Now on",
"time": "406.161",
"value": "Tony Gwynn"
},
{
"operation": "Highlighting",
"time": "420.952",
"value": "Tony Gwynn_sentence_2"
},
{
"operation": "Page search",
"time": "431.733",
"value": "california"
},
{
"operation": "Highlighting",
"time": "472.713",
"value": "Tony Gwynn_sentence_458"
},
{
"operation": "search",
"time": "497.239",
"value": "Dave WInfield"
},
{
"operation": "Now on",
"time": "498.414",
"value": "Dave Winfield"
},
{
"operation": "Page search",
"time": "506.544",
"value": "californai"
},
{
"operation": "Page search",
"time": "510.152",
"value": "california"
},
{
"operation": "hyperlink",
"time": "521.24",
"value": "2.6 Honors and awards Dave Winfield_section_7"
},
{
"operation": "finish",
"time": "541.487",
"value": "finish"
}
]
}
|
91133
|
Rensselaer Polytechnic Institute's most recen tresearch center is the Center for Biotechnology and Interdisciplinary Studies, located on 15th street between RPI’s Playhouse and Academy Hall which targets biologics, and CBIS has produced over 2,000 peer-reviewed publications.
|
{
"evidence": [
"Rensselaer Polytechnic Institute_sentence_0",
"Center for Biotechnology and Interdisciplinary Studies_sentence_0",
"Center for Biotechnology and Interdisciplinary Studies_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRensselaer Polytechnic Institute's most recen tresearch center is the Center for Biotechnology and Interdisciplinary Studies, located on 15th street between RPI’s Playhouse and Academy Hall which targets biologics, and CBIS has produced over 2,000 peer-reviewed publications.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Rensselaer Polytechnic Institute_sentence_0",
"Center for Biotechnology and Interdisciplinary Studies_sentence_0",
"Center for Biotechnology and Interdisciplinary Studies_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 296,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.392",
"value": "undefined"
},
{
"operation": "search",
"time": "34.77",
"value": "Rensselaer Polytechnic Institute"
},
{
"operation": "Now on",
"time": "37.209",
"value": "Rensselaer Polytechnic Institute"
},
{
"operation": "Highlighting",
"time": "46.646",
"value": "Rensselaer Polytechnic Institute_sentence_0"
},
{
"operation": "Page search",
"time": "86.554",
"value": "Center for Biotechnology and Interdisciplinary Studies"
},
{
"operation": "Page search",
"time": "86.732",
"value": "Center for Biotechnology and Interdisciplinary Studies"
},
{
"operation": "hyperlink",
"time": "154.108",
"value": "Center for Biotechnology and Interdisciplinary Studies"
},
{
"operation": "Now on",
"time": "155.467",
"value": "Center for Biotechnology and Interdisciplinary Studies"
},
{
"operation": "Highlighting",
"time": "160.199",
"value": "Center for Biotechnology and Interdisciplinary Studies_sentence_0"
},
{
"operation": "Highlighting",
"time": "160.752",
"value": "Center for Biotechnology and Interdisciplinary Studies_sentence_1"
},
{
"operation": "Page search",
"time": "220.358",
"value": "biologics"
},
{
"operation": "Page search",
"time": "220.512",
"value": "biologics"
},
{
"operation": "Page search",
"time": "223.624",
"value": "biologics"
},
{
"operation": "Page search",
"time": "223.823",
"value": "biologics"
},
{
"operation": "Page search",
"time": "224.015",
"value": "biologics"
},
{
"operation": "Page search",
"time": "224.191",
"value": "biologics"
},
{
"operation": "Page search",
"time": "224.368",
"value": "biologics"
},
{
"operation": "search",
"time": "306.771",
"value": "biologics"
},
{
"operation": "Now on",
"time": "308.586",
"value": "Biologics"
},
{
"operation": "back-button-clicked",
"time": "325.911",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "326.55",
"value": "Center for Biotechnology and Interdisciplinary Studies"
},
{
"operation": "Page search",
"time": "333.029",
"value": "biolog"
},
{
"operation": "Page search",
"time": "333.379",
"value": "biolog"
},
{
"operation": "Page search",
"time": "333.579",
"value": "biolog"
},
{
"operation": "Page search",
"time": "333.78",
"value": "biolog"
},
{
"operation": "Page search",
"time": "333.972",
"value": "biolog"
},
{
"operation": "Page search",
"time": "334.163",
"value": "biolog"
},
{
"operation": "Page search",
"time": "334.324",
"value": "biolog"
},
{
"operation": "Page search",
"time": "445.919",
"value": "cbis"
},
{
"operation": "Page search",
"time": "446.384",
"value": "cbis"
},
{
"operation": "Page search",
"time": "446.567",
"value": "cbis"
},
{
"operation": "Page search",
"time": "446.742",
"value": "cbis"
},
{
"operation": "Page search",
"time": "446.918",
"value": "cbis"
},
{
"operation": "Page search",
"time": "447.07",
"value": "cbis"
},
{
"operation": "Page search",
"time": "447.214",
"value": "cbis"
},
{
"operation": "Page search",
"time": "475.76",
"value": "publi"
},
{
"operation": "Page search",
"time": "476.11",
"value": "publi"
},
{
"operation": "Page search",
"time": "476.322",
"value": "publi"
},
{
"operation": "Page search",
"time": "476.535",
"value": "publi"
},
{
"operation": "Page search",
"time": "543.852",
"value": "revi"
},
{
"operation": "Page search",
"time": "544.106",
"value": "revi"
},
{
"operation": "Page search",
"time": "544.323",
"value": "revi"
},
{
"operation": "Page search",
"time": "544.498",
"value": "revi"
},
{
"operation": "Page search",
"time": "544.708",
"value": "revi"
},
{
"operation": "Page search",
"time": "544.898",
"value": "revi"
},
{
"operation": "Page search",
"time": "545.09",
"value": "revi"
},
{
"operation": "Page search",
"time": "545.281",
"value": "revi"
},
{
"operation": "Page search",
"time": "545.49",
"value": "revi"
},
{
"operation": "Page search",
"time": "545.658",
"value": "revi"
},
{
"operation": "finish",
"time": "679.53",
"value": "finish"
}
]
}
|
exfever_8572
|
Gil Sharone is an American drummer and member of an American experimental alternative rock/post-rock group.
|
{
"evidence": [
"Gil Sharone",
"Team Sleep"
],
"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:\nGil Sharone is an American drummer and member of an American experimental alternative rock/post-rock group.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gil Sharone",
"Team Sleep"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 297,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5098
|
Unforgettable is a 2017 American drama thriller film starring an American actress who made her film debut in the film The Ugly Truth.
|
{
"evidence": [
"Unforgettable (2017 film)",
"Katherine Heigl"
],
"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:\nUnforgettable is a 2017 American drama thriller film starring an American actress who made her film debut in the film The Ugly Truth.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Unforgettable (2017 film)",
"Katherine Heigl"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 298,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
12091
|
Surcuolm is a village in Germany with a historical population of 167 in 1850 that continuously dropped to 130 in 2000.
|
{
"evidence": [
"Surcuolm_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:\nSurcuolm is a village in Germany with a historical population of 167 in 1850 that continuously dropped to 130 in 2000.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Surcuolm_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 299,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.116",
"value": "?search="
},
{
"operation": "search",
"time": "13.1",
"value": "Suruolm"
},
{
"operation": "Now on",
"time": "13.712",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "16.89",
"value": "Surcuolm"
},
{
"operation": "Now on",
"time": "17.582",
"value": "Surcuolm"
},
{
"operation": "Highlighting",
"time": "24.571",
"value": "Surcuolm_sentence_0"
},
{
"operation": "finish",
"time": "87.244",
"value": "finish"
}
]
}
|
16257
|
Aristocrat Frau von Zeppelin (descendant of Count Ferdinand von Zeppelin, creator of the Zeppelin aircraft) felt the band's use of Zeppelin was insulting and dishonoured her family name, as a result, hostility toward the rock group continued on their next tour of the country in early 1970.
|
{
"evidence": [
"Led Zeppelin European Tour 1970_sentence_3",
"Led Zeppelin European Tour 1970_sentence_10",
"Led Zeppelin European Tour 1970_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:\nAristocrat Frau von Zeppelin (descendant of Count Ferdinand von Zeppelin, creator of the Zeppelin aircraft) felt the band's use of Zeppelin was insulting and dishonoured her family name, as a result, hostility toward the rock group continued on their next tour of the country in early 1970.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Led Zeppelin European Tour 1970_sentence_3",
"Led Zeppelin European Tour 1970_sentence_10",
"Led Zeppelin European Tour 1970_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 300,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.505",
"value": "?search="
},
{
"operation": "search",
"time": "21.832",
"value": "Frau von Zeppelin"
},
{
"operation": "Now on",
"time": "22.818",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "37.135",
"value": "Led Zeppelin European Tour 1970"
},
{
"operation": "Now on",
"time": "37.795",
"value": "Led Zeppelin European Tour 1970"
},
{
"operation": "Highlighting",
"time": "67.863",
"value": "Led Zeppelin European Tour 1970_sentence_3"
},
{
"operation": "Highlighting",
"time": "76.678",
"value": "Led Zeppelin European Tour 1970_sentence_10"
},
{
"operation": "Highlighting",
"time": "85.134",
"value": "Led Zeppelin European Tour 1970_sentence_11"
},
{
"operation": "finish",
"time": "137.638",
"value": "finish"
}
]
}
|
84897
|
Gressbanen is a stadium located at Holmen in Oslo, Norway and during the summer it has real turf and is used for association football and in the winter it has artificial ice and fields bandy.
|
{
"evidence": [
"Gressbanen_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:\nGressbanen is a stadium located at Holmen in Oslo, Norway and during the summer it has real turf and is used for association football and in the winter it has artificial ice and fields bandy.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Gressbanen_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 301,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.801",
"value": "?search="
},
{
"operation": "search",
"time": "6.89",
"value": "Gressbanen"
},
{
"operation": "Now on",
"time": "7.849",
"value": "Gressbanen"
},
{
"operation": "Highlighting",
"time": "11.665",
"value": "Gressbanen_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "15.134",
"value": "Gressbanen_sentence_0"
},
{
"operation": "Highlighting",
"time": "19.429",
"value": "Gressbanen_sentence_1"
},
{
"operation": "finish",
"time": "28.177",
"value": "finish"
}
]
}
|
60037
|
Alessia Zecchini was born 1992 in Rome, she started her freediving career when she was 13 years old but never won any medal.
|
{
"evidence": [
"Alessia Zecchini_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:\nAlessia Zecchini was born 1992 in Rome, she started her freediving career when she was 13 years old but never won any medal.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Alessia Zecchini_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 302,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.66",
"value": "?search="
},
{
"operation": "search",
"time": "9.342",
"value": "Alessia Zecchini"
},
{
"operation": "Now on",
"time": "10.107",
"value": "Alessia Zecchini"
},
{
"operation": "Highlighting",
"time": "19.028",
"value": "Alessia Zecchini_sentence_0"
},
{
"operation": "Highlighting",
"time": "37.068",
"value": "Alessia Zecchini_sentence_2"
},
{
"operation": "Highlighting",
"time": "55.811",
"value": "Alessia Zecchini_sentence_5"
},
{
"operation": "Highlighting",
"time": "59.539",
"value": "Alessia Zecchini_sentence_5"
},
{
"operation": "finish",
"time": "81.155",
"value": "finish"
}
]
}
|
exfever_2953
|
The Apple Store is a chain of retail stores that sells a digital media player and microconsole that is developed and marketed by Apple Inc.
|
{
"evidence": [
"Apple Store",
"Apple TV"
],
"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 Apple Store is a chain of retail stores that sells a digital media player and microconsole that is developed and marketed by Apple Inc.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Apple Store",
"Apple TV"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 303,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
22320
|
Russia produces the most raspberry's out of the five countries that produce it worldwide.
|
{
"evidence": [
"Raspberry_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:\nRussia produces the most raspberry's out of the five countries that produce it worldwide.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Raspberry_sentence_7"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 304,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.628",
"value": "?search="
},
{
"operation": "search",
"time": "35.633",
"value": "List of largest producing countries of agricultural commodities"
},
{
"operation": "Now on",
"time": "38.074",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "44.278",
"value": "Sanitary paper"
},
{
"operation": "Now on",
"time": "44.692",
"value": "Sanitary paper"
},
{
"operation": "search",
"time": "81.496",
"value": "Raspberry"
},
{
"operation": "Now on",
"time": "82.383",
"value": "Raspberry"
},
{
"operation": "Highlighting",
"time": "87.28",
"value": "Raspberry_sentence_7"
},
{
"operation": "search",
"time": "148.319",
"value": "List of largest producing countries of agricultural commodities"
},
{
"operation": "Now on",
"time": "149.554",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "154.453",
"value": "Make in India"
},
{
"operation": "Now on",
"time": "155.327",
"value": "Make in India"
},
{
"operation": "finish",
"time": "175.487",
"value": "finish"
}
]
}
|
7119
|
A total quota of 120 athletes was allowed to compete in the Short track speed skating at the 2014 Winter Olympics, Russia was allowed the minimum even if they were the host.
|
{
"evidence": [
"Short track speed skating at the 2014 Winter Olympics – Qualification_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:\nA total quota of 120 athletes was allowed to compete in the Short track speed skating at the 2014 Winter Olympics, Russia was allowed the minimum even if they were the host.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Short track speed skating at the 2014 Winter Olympics – Qualification_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 305,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.525",
"value": "?search="
},
{
"operation": "search",
"time": "11.141",
"value": "2014 Winter Olympics"
},
{
"operation": "Now on",
"time": "12.106",
"value": "2014 Winter Olympics"
},
{
"operation": "Page search",
"time": "15.607",
"value": "speed skating"
},
{
"operation": "hyperlink",
"time": "31.313",
"value": "Speed skating"
},
{
"operation": "Now on",
"time": "31.943",
"value": "Speed skating at the 2014 Winter Olympics"
},
{
"operation": "back-button-clicked",
"time": "80.473",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "80.652",
"value": "2014 Winter Olympics"
},
{
"operation": "search",
"time": "95.258",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "Now on",
"time": "95.997",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "Page search",
"time": "114.727",
"value": "120"
},
{
"operation": "Highlighting",
"time": "147.541",
"value": "Short track speed skating at the 2014 Winter Olympics_sentence_0"
},
{
"operation": "Highlighting",
"time": "147.544",
"value": "Short track speed skating at the 2014 Winter Olympics_title"
},
{
"operation": "Highlighting deleted",
"time": "148.907",
"value": "Short track speed skating at the 2014 Winter Olympics_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "148.91",
"value": "Short track speed skating at the 2014 Winter Olympics_title"
},
{
"operation": "Highlighting",
"time": "149.437",
"value": "Short track speed skating at the 2014 Winter Olympics_sentence_0"
},
{
"operation": "Highlighting",
"time": "149.44",
"value": "Short track speed skating at the 2014 Winter Olympics_title"
},
{
"operation": "Highlighting deleted",
"time": "149.884",
"value": "Short track speed skating at the 2014 Winter Olympics_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "149.887",
"value": "Short track speed skating at the 2014 Winter Olympics_title"
},
{
"operation": "hyperlink",
"time": "151.021",
"value": "Short track speed skating"
},
{
"operation": "Now on",
"time": "151.744",
"value": "Short track speed skating at the Winter Olympics"
},
{
"operation": "Page search",
"time": "179.374",
"value": "120"
},
{
"operation": "search",
"time": "193.661",
"value": "2014 winter olympics"
},
{
"operation": "Now on",
"time": "194.781",
"value": "2014 Winter Olympics"
},
{
"operation": "Page search",
"time": "202.602",
"value": "120"
},
{
"operation": "Page search",
"time": "219.406",
"value": "short tracj"
},
{
"operation": "Page search",
"time": "221.174",
"value": "short track"
},
{
"operation": "search",
"time": "256.403",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "Now on",
"time": "256.961",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "search",
"time": "329.807",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "Now on",
"time": "330.687",
"value": "Short track speed skating at the 2014 Winter Olympics"
},
{
"operation": "search",
"time": "356.31",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification"
},
{
"operation": "Now on",
"time": "357.285",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification"
},
{
"operation": "Highlighting",
"time": "359.107",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification_sentence_2"
},
{
"operation": "Highlighting",
"time": "396.093",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "411.173",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification_sentence_2"
},
{
"operation": "Now on",
"time": "432.368",
"value": "Short track speed skating at the 2014 Winter Olympics – Qualification"
},
{
"operation": "finish",
"time": "441.557",
"value": "finish"
}
]
}
|
88101
|
According to David R. Adler of Allmusic, Just So Happens was innovate, rewarding, and not repetitive.
|
{
"evidence": [
"Just So Happens_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:\nAccording to David R. Adler of Allmusic, Just So Happens was innovate, rewarding, and not repetitive.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Just So Happens_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 306,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.923",
"value": "?search="
},
{
"operation": "search",
"time": "26.353",
"value": "Just So Happens"
},
{
"operation": "Now on",
"time": "28.019",
"value": "Just So Happens"
},
{
"operation": "Highlighting",
"time": "32.093",
"value": "Just So Happens_sentence_1"
},
{
"operation": "finish",
"time": "55.03",
"value": "finish"
}
]
}
|
77405
|
In 2011, with an estimated cost of less than US$100 million, the Guggenheim in Abu Dhabi had begun its construction with a Roman architectural style but was later suspended.
|
{
"evidence": [
"Guggenheim Abu Dhabi_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:\nIn 2011, with an estimated cost of less than US$100 million, the Guggenheim in Abu Dhabi had begun its construction with a Roman architectural style but was later suspended.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Guggenheim Abu Dhabi_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 307,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.815",
"value": "?search="
},
{
"operation": "search",
"time": "13.988",
"value": "Contains...Guggenheim in Abu Dhabi"
},
{
"operation": "Now on",
"time": "15.157",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "30.359",
"value": "Early history Guggenheim Abu Dhabi_section_0"
},
{
"operation": "Now on",
"time": "31.223",
"value": "Guggenheim Abu Dhabi#Early history Guggenheim Abu Dhabi section 0"
},
{
"operation": "Page search",
"time": "56.01",
"value": "roman"
},
{
"operation": "Page search",
"time": "107.294",
"value": "100"
},
{
"operation": "search",
"time": "120.99",
"value": "with an estimated cost of less than US$100 million, the Guggenheim in Abu Dhabi had begun its construction with a Roman architectural style but was later suspended"
},
{
"operation": "Now on",
"time": "124.373",
"value": "undefined"
},
{
"operation": "search",
"time": "134.597",
"value": "begun its construction with a Roman architectural style but was later suspended"
},
{
"operation": "Now on",
"time": "137.361",
"value": "undefined"
},
{
"operation": "search",
"time": "152.919",
"value": "Contains...Guggenheim in Abu Dhabi"
},
{
"operation": "Now on",
"time": "153.686",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "163.66",
"value": "Gulf Labor"
},
{
"operation": "Now on",
"time": "164.493",
"value": "Gulf Labor"
},
{
"operation": "back-button-clicked",
"time": "170.98",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "171.874",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "174.739",
"value": "Guggenheim Abu Dhabi"
},
{
"operation": "Now on",
"time": "175.546",
"value": "Guggenheim Abu Dhabi"
},
{
"operation": "Highlighting",
"time": "179.519",
"value": "Guggenheim Abu Dhabi_sentence_3"
},
{
"operation": "finish",
"time": "319.582",
"value": "finish"
}
]
}
|
exfever_5290
|
Pluto also has a quasi-satellite.
|
{
"evidence": [
"Pluto",
"International Astronomical Union"
],
"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:\nPluto also has a quasi-satellite.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pluto",
"International Astronomical Union"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 308,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
70157
|
Melora Walters married and later divorced Alex Vendller.
|
{
"evidence": [
"Melora Walters_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:\nMelora Walters married and later divorced Alex Vendller.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Melora Walters_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 309,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.911",
"value": "?search="
},
{
"operation": "search",
"time": "21.303",
"value": "Melora Walters"
},
{
"operation": "Now on",
"time": "22.92",
"value": "Melora Walters"
},
{
"operation": "Highlighting",
"time": "106.811",
"value": "Melora Walters_sentence_2"
},
{
"operation": "Highlighting",
"time": "108.891",
"value": "Melora Walters_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "178.538",
"value": "Melora Walters_sentence_2"
},
{
"operation": "finish",
"time": "296.586",
"value": "finish"
}
]
}
|
21478
|
János Szerednyei, a local priest, was among the Tarcal residents rounded up for deportation in 1945 and sent to Vorosilovka labor camp.
|
{
"evidence": [
"János Szerednyei_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:\nJános Szerednyei, a local priest, was among the Tarcal residents rounded up for deportation in 1945 and sent to Vorosilovka labor camp.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"János Szerednyei_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 310,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.891",
"value": "?search="
},
{
"operation": "search",
"time": "14.405",
"value": "store near mt everest"
},
{
"operation": "Now on",
"time": "16.203",
"value": "undefined"
},
{
"operation": "search",
"time": "27.133",
"value": "János Szerednyei"
},
{
"operation": "Now on",
"time": "28.606",
"value": "János Szerednyei"
},
{
"operation": "Highlighting",
"time": "34.379",
"value": "János Szerednyei_sentence_0"
},
{
"operation": "finish",
"time": "132.329",
"value": "finish"
}
]
}
|
13557
|
Macvey Napier, editor of the review from 1829, acted as Librarian to the Signet Library.
|
{
"evidence": [
"Macvey Napier_sentence_7",
"Macvey Napier_sentence_8"
],
"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:\nMacvey Napier, editor of the review from 1829, acted as Librarian to the Signet Library.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Macvey Napier_sentence_7",
"Macvey Napier_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 311,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.568",
"value": "?search="
},
{
"operation": "search",
"time": "4.487",
"value": "Macvey Napier"
},
{
"operation": "Now on",
"time": "5.18",
"value": "Macvey Napier"
},
{
"operation": "Page search",
"time": "18.32",
"value": "editor"
},
{
"operation": "Highlighting",
"time": "25.554",
"value": "Macvey Napier_sentence_7"
},
{
"operation": "Highlighting",
"time": "34.397",
"value": "Macvey Napier_sentence_8"
},
{
"operation": "finish",
"time": "255.391",
"value": "finish"
}
]
}
|
exfever_8573
|
The Mummy is a 2017 American fantasy action-adventure film directed by Alex Kurtzman and written by an American screenwriter, film director, and producer who made his directorial debut with "Jack Reacher."
|
{
"evidence": [
"The Mummy (2017 film)",
"Christopher McQuarrie"
],
"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 Mummy is a 2017 American fantasy action-adventure film directed by Alex Kurtzman and written by an American screenwriter, film director, and producer who made his directorial debut with \"Jack Reacher.\"\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Mummy (2017 film)",
"Christopher McQuarrie"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 312,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
86230
|
Werder Bremen made 10 appearances in the 2006 season with Palmeiras, a Brazilian professional football club based in the city of São Paulo, in the district of Perdizes.
|
{
"evidence": [
"SV Werder Bremen_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWerder Bremen made 10 appearances in the 2006 season with Palmeiras, a Brazilian professional football club based in the city of São Paulo, in the district of Perdizes.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"SV Werder Bremen_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 313,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.1",
"value": "?search="
},
{
"operation": "search",
"time": "19.363",
"value": "Werder Bremen"
},
{
"operation": "Now on",
"time": "22.309",
"value": "SV Werder Bremen"
},
{
"operation": "search",
"time": "44.932",
"value": "Werder Bremen II"
},
{
"operation": "Now on",
"time": "46.364",
"value": "SV Werder Bremen II"
},
{
"operation": "search",
"time": "55.3",
"value": "Werder Bremen"
},
{
"operation": "Now on",
"time": "56.357",
"value": "Werder Bremen"
},
{
"operation": "Highlighting",
"time": "65.024",
"value": "SV Werder Bremen_sentence_1"
},
{
"operation": "search",
"time": "87.729",
"value": "FC palmeiras"
},
{
"operation": "Now on",
"time": "89.622",
"value": "undefined"
},
{
"operation": "search",
"time": "100.355",
"value": "palmeiras FC"
},
{
"operation": "Now on",
"time": "101.3",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "106.926",
"value": "C.R.D. Libolo"
},
{
"operation": "Now on",
"time": "108.444",
"value": "C.R.D. Libolo"
},
{
"operation": "search",
"time": "119.135",
"value": "bremen"
},
{
"operation": "Now on",
"time": "123.369",
"value": "Bremen"
},
{
"operation": "search",
"time": "140.859",
"value": "palmeiras"
},
{
"operation": "Now on",
"time": "146.012",
"value": "Sociedade Esportiva Palmeiras"
},
{
"operation": "Highlighting",
"time": "152.834",
"value": "Sociedade Esportiva Palmeiras_sentence_2"
},
{
"operation": "Highlighting",
"time": "152.836",
"value": "Sociedade Esportiva Palmeiras_title"
},
{
"operation": "hyperlink",
"time": "181.516",
"value": "5 Players Sociedade Esportiva Palmeiras_section_35"
},
{
"operation": "Page search",
"time": "197.362",
"value": "2006"
},
{
"operation": "Highlighting",
"time": "206.046",
"value": "Sociedade Esportiva Palmeiras_sentence_54"
},
{
"operation": "Highlighting",
"time": "206.053",
"value": "Sociedade Esportiva Palmeiras_section_0"
},
{
"operation": "Highlighting",
"time": "206.055",
"value": "Sociedade Esportiva Palmeiras_section_4"
},
{
"operation": "Highlighting",
"time": "222.189",
"value": "Sociedade Esportiva Palmeiras_sentence_135"
},
{
"operation": "Highlighting",
"time": "222.198",
"value": "Sociedade Esportiva Palmeiras_section_10"
},
{
"operation": "search",
"time": "238.833",
"value": "werder"
},
{
"operation": "Now on",
"time": "241.26",
"value": "Werder"
},
{
"operation": "hyperlink",
"time": "243.659",
"value": "1 People Werder_section_0"
},
{
"operation": "Highlighting",
"time": "248.707",
"value": "Werder_item_0_0"
},
{
"operation": "Highlighting",
"time": "248.709",
"value": "Werder_title"
},
{
"operation": "Highlighting",
"time": "248.71",
"value": "Werder_section_0"
},
{
"operation": "Highlighting",
"time": "249.586",
"value": "Werder_item_0_1"
},
{
"operation": "Highlighting",
"time": "250.479",
"value": "Werder_item_0_2"
},
{
"operation": "Highlighting",
"time": "251.445",
"value": "Werder_item_0_3"
},
{
"operation": "Highlighting deleted",
"time": "259.842",
"value": "Werder_item_0_0"
},
{
"operation": "Highlighting deleted",
"time": "260.464",
"value": "Werder_item_0_1"
},
{
"operation": "Highlighting deleted",
"time": "261.065",
"value": "Werder_item_0_2"
},
{
"operation": "Highlighting deleted",
"time": "261.672",
"value": "Werder_item_0_3"
},
{
"operation": "Highlighting deleted",
"time": "261.676",
"value": "Werder_title"
},
{
"operation": "Highlighting deleted",
"time": "261.68",
"value": "Werder_section_0"
},
{
"operation": "search",
"time": "265.754",
"value": "bremen"
},
{
"operation": "Now on",
"time": "268.348",
"value": "Bremen"
},
{
"operation": "search",
"time": "309.554",
"value": "palmeiras FC"
},
{
"operation": "Now on",
"time": "311.207",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "315.34",
"value": "C.R.D. Libolo"
},
{
"operation": "Now on",
"time": "316.328",
"value": "C.R.D. Libolo"
},
{
"operation": "search",
"time": "325.426",
"value": "FC palmeiras"
},
{
"operation": "Now on",
"time": "326.374",
"value": "undefined"
},
{
"operation": "finish",
"time": "341.075",
"value": "finish"
}
]
}
|
70408
|
Mildam (located in Heerenveen, Friesland in the Netherlands) has a population of around 685 in 2018 and a landmark is a windmill named De Tjongermolen.
|
{
"evidence": [
"Mildam_sentence_0",
"Mildam_sentence_1",
"Mildam_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:\nMildam (located in Heerenveen, Friesland in the Netherlands) has a population of around 685 in 2018 and a landmark is a windmill named De Tjongermolen.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Mildam_sentence_0",
"Mildam_sentence_1",
"Mildam_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 314,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.731",
"value": "?search="
},
{
"operation": "search",
"time": "11.023",
"value": "Mildam"
},
{
"operation": "Now on",
"time": "12.361",
"value": "Mildam"
},
{
"operation": "Highlighting",
"time": "23.528",
"value": "Mildam_sentence_0"
},
{
"operation": "Highlighting",
"time": "33.043",
"value": "Mildam_sentence_1"
},
{
"operation": "Highlighting",
"time": "43.887",
"value": "Mildam_sentence_3"
},
{
"operation": "hyperlink",
"time": "48.999",
"value": "De Tjongermolen"
},
{
"operation": "Now on",
"time": "49.811",
"value": "De Tjongermolen, Mildam"
},
{
"operation": "back-button-clicked",
"time": "53.811",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "53.898",
"value": "Mildam"
},
{
"operation": "finish",
"time": "74.998",
"value": "finish"
}
]
}
|
exfever_3827
|
Chester Charles Bennington is a member of a Canadian rock band that rose to international fame with their debut studio album, Hybrid Theory.
|
{
"evidence": [
"Chester Bennington",
"Linkin Park"
],
"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:\nChester Charles Bennington is a member of a Canadian rock band that rose to international fame with their debut studio album, Hybrid Theory.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Chester Bennington",
"Linkin Park"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 315,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76037
|
Cloncurry Solar Farm was scrapped because of light pollution concerns but on 2 December 2011 Ingenero Pty Ltd have been named to install a 2 MW photovoltaic power system.
|
{
"evidence": [
"Cloncurry Solar Farm_sentence_2",
"Cloncurry Solar Farm_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:\nCloncurry Solar Farm was scrapped because of light pollution concerns but on 2 December 2011 Ingenero Pty Ltd have been named to install a 2 MW photovoltaic power system.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cloncurry Solar Farm_sentence_2",
"Cloncurry Solar Farm_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 316,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.346",
"value": "undefined"
},
{
"operation": "search",
"time": "10.167",
"value": "Cloncurry Solar Farm"
},
{
"operation": "Now on",
"time": "16.269",
"value": "Cloncurry Solar Farm"
},
{
"operation": "Highlighting",
"time": "27.641",
"value": "Cloncurry Solar Farm_sentence_2"
},
{
"operation": "Highlighting",
"time": "31.145",
"value": "Cloncurry Solar Farm_sentence_3"
},
{
"operation": "finish",
"time": "52.871",
"value": "finish"
}
]
}
|
35083
|
Due to th Jan Mayen Fracture Zone the deep water masses of the Greenland Current are recirculated within the Greenland sea where the deep water encounters the Jan Mayen Ridge and are deflected eastward towards the interior of the Greenland Sea Gyre.
|
{
"evidence": [
"East Greenland Current_sentence_21",
"East Greenland Current_sentence_22"
],
"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:\nDue to th Jan Mayen Fracture Zone the deep water masses of the Greenland Current are recirculated within the Greenland sea where the deep water encounters the Jan Mayen Ridge and are deflected eastward towards the interior of the Greenland Sea Gyre.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"East Greenland Current_sentence_21",
"East Greenland Current_sentence_22"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 317,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.698",
"value": "?search="
},
{
"operation": "search",
"time": "5.843",
"value": "Jan Mayen Fracture Zone"
},
{
"operation": "Now on",
"time": "7.316",
"value": "undefined"
},
{
"operation": "search",
"time": "21.931",
"value": "East Greenland Current"
},
{
"operation": "Now on",
"time": "22.829",
"value": "East Greenland Current"
},
{
"operation": "Highlighting",
"time": "28.134",
"value": "East Greenland Current_sentence_22"
},
{
"operation": "Highlighting deleted",
"time": "33.241",
"value": "East Greenland Current_sentence_22"
},
{
"operation": "Highlighting",
"time": "33.882",
"value": "East Greenland Current_sentence_21"
},
{
"operation": "Highlighting",
"time": "47.46",
"value": "East Greenland Current_sentence_22"
},
{
"operation": "finish",
"time": "78.269",
"value": "finish"
}
]
}
|
60231
|
DVCAM is an XDCAM format which uses an MXF container, a format in professional digital video defined by a set of SMPTE standards.
|
{
"evidence": [
"DV_sentence_1",
"DV_sentence_42"
],
"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:\nDVCAM is an XDCAM format which uses an MXF container, a format in professional digital video defined by a set of SMPTE standards.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"DV_sentence_1",
"DV_sentence_42"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 318,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.908",
"value": "?search="
},
{
"operation": "search",
"time": "12.505",
"value": "DVCAM"
},
{
"operation": "Now on",
"time": "14.372",
"value": "DV"
},
{
"operation": "Highlighting",
"time": "130.054",
"value": "DV_sentence_42"
},
{
"operation": "hyperlink",
"time": "225.585",
"value": "DV (disambiguation)"
},
{
"operation": "Now on",
"time": "226.856",
"value": "DV (disambiguation)"
},
{
"operation": "hyperlink",
"time": "245.465",
"value": "DV"
},
{
"operation": "Now on",
"time": "246.41",
"value": "DV"
},
{
"operation": "Highlighting",
"time": "250.032",
"value": "DV_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "335.486",
"value": "DV_sentence_42"
},
{
"operation": "Highlighting",
"time": "336.917",
"value": "DV_sentence_42"
},
{
"operation": "finish",
"time": "363.012",
"value": "finish"
}
]
}
|
90869
|
The Blue Shield of California was created in 1938 by California Medical Association, which is a professional organization representing nearly 50,000 physicians founded in 1856.
|
{
"evidence": [
"Blue Shield of California_sentence_4",
"California Medical Association_sentence_0",
"California Medical Association_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 Blue Shield of California was created in 1938 by California Medical Association, which is a professional organization representing nearly 50,000 physicians founded in 1856.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Blue Shield of California_sentence_4",
"California Medical Association_sentence_0",
"California Medical Association_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 319,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.904",
"value": "?search="
},
{
"operation": "search",
"time": "13.196",
"value": "Blue Shield of California"
},
{
"operation": "Now on",
"time": "13.982",
"value": "Blue Shield of California"
},
{
"operation": "Highlighting",
"time": "22.432",
"value": "Blue Shield of California_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "22.434",
"value": "Blue Shield of California_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "29.577",
"value": "Blue Shield of California_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "30.591",
"value": "Blue Shield of California_cell_0_2_1"
},
{
"operation": "Highlighting deleted",
"time": "30.593",
"value": "Blue Shield of California_header_cell_0_2_0"
},
{
"operation": "hyperlink",
"time": "58.424",
"value": "California Medical Association"
},
{
"operation": "Now on",
"time": "59.104",
"value": "California Medical Association"
},
{
"operation": "Highlighting",
"time": "62.526",
"value": "California Medical Association_sentence_0"
},
{
"operation": "Highlighting",
"time": "63.526",
"value": "California Medical Association_sentence_1"
},
{
"operation": "finish",
"time": "70.357",
"value": "finish"
}
]
}
|
23939
|
John Player Classic is wrongly believed to have taken place four years between 1970 and 1973, one of the winners is Bob Charles.
|
{
"evidence": [
"John Player Classic_sentence_0",
"John Player Classic_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:\nJohn Player Classic is wrongly believed to have taken place four years between 1970 and 1973, one of the winners is Bob Charles.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"John Player Classic_sentence_0",
"John Player Classic_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 320,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.846",
"value": "?search="
},
{
"operation": "Now on",
"time": "24.96",
"value": "?search="
},
{
"operation": "search",
"time": "46.297",
"value": "john player classic"
},
{
"operation": "Now on",
"time": "47.29",
"value": "John Player Classic"
},
{
"operation": "search",
"time": "156.959",
"value": "john player classic wrongly"
},
{
"operation": "Now on",
"time": "158.814",
"value": "undefined"
},
{
"operation": "search",
"time": "185.546",
"value": "john player classic 1971"
},
{
"operation": "Now on",
"time": "186.578",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "193.127",
"value": "John Deere Classic"
},
{
"operation": "Now on",
"time": "194.538",
"value": "John Deere Classic"
},
{
"operation": "search",
"time": "226.678",
"value": "john player classic"
},
{
"operation": "Now on",
"time": "227.502",
"value": "John Player Classic"
},
{
"operation": "Highlighting",
"time": "238.965",
"value": "John Player Classic_sentence_0"
},
{
"operation": "Highlighting",
"time": "245.856",
"value": "John Player Classic_sentence_3"
},
{
"operation": "finish",
"time": "291.841",
"value": "finish"
}
]
}
|
38080
|
In the 2011 census in Africa, there were 839 people.
|
{
"evidence": [
"South African National Census of 2011_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 the 2011 census in Africa, there were 839 people.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"South African National Census of 2011_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 321,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.571",
"value": "?search="
},
{
"operation": "search",
"time": "42.467",
"value": "2011 census africa"
},
{
"operation": "Now on",
"time": "44.214",
"value": "undefined"
},
{
"operation": "search",
"time": "50.281",
"value": "africa 2011 census"
},
{
"operation": "Now on",
"time": "51.275",
"value": "undefined"
},
{
"operation": "search",
"time": "61.319",
"value": "839 people 2011 census"
},
{
"operation": "Now on",
"time": "62.955",
"value": "undefined"
},
{
"operation": "search",
"time": "72.657",
"value": "africa census"
},
{
"operation": "Now on",
"time": "73.533",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "78.756",
"value": "Census in Botswana"
},
{
"operation": "Now on",
"time": "79.636",
"value": "Census in Botswana"
},
{
"operation": "hyperlink",
"time": "83.675",
"value": "Census"
},
{
"operation": "Now on",
"time": "84.793",
"value": "Census"
},
{
"operation": "back-button-clicked",
"time": "91.455",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "91.542",
"value": "Census in Botswana"
},
{
"operation": "search",
"time": "171",
"value": "census of africa"
},
{
"operation": "Now on",
"time": "172.933",
"value": "undefined"
},
{
"operation": "search",
"time": "194.386",
"value": "2011 census"
},
{
"operation": "Now on",
"time": "195.184",
"value": "2011 census"
},
{
"operation": "hyperlink",
"time": "250.069",
"value": "South African National Census of 2011"
},
{
"operation": "Now on",
"time": "251.349",
"value": "South African National Census of 2011"
},
{
"operation": "Page search",
"time": "263.099",
"value": "839"
},
{
"operation": "Highlighting",
"time": "277.985",
"value": "South African National Census of 2011_sentence_0"
},
{
"operation": "finish",
"time": "280.163",
"value": "finish"
}
]
}
|
exfever_2316
|
George Tuska was an American comic book and newspaper comic strip artist best known for his 1940s work on an American comic book series published between 1942 and 1955 by Lev Gleason Publications.
|
{
"evidence": [
"George Tuska",
"Crime Does Not Pay (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:\nGeorge Tuska was an American comic book and newspaper comic strip artist best known for his 1940s work on an American comic book series published between 1942 and 1955 by Lev Gleason Publications.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"George Tuska",
"Crime Does Not Pay (comics)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 322,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21968
|
The Puerto Cabezas Airport's improvement plans include a new air terminal that has a total area of 450 m with terminal facilities that are modern and air-conditioned.
|
{
"evidence": [
"Puerto Cabezas Airport_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:\nThe Puerto Cabezas Airport's improvement plans include a new air terminal that has a total area of 450 m with terminal facilities that are modern and air-conditioned.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Puerto Cabezas Airport_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 323,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.014",
"value": "?search="
},
{
"operation": "search",
"time": "12.203",
"value": "Puerto Cabezas Airport improvement"
},
{
"operation": "Now on",
"time": "13.34",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "14.906",
"value": "Improvement plans Puerto Cabezas Airport_section_0"
},
{
"operation": "Now on",
"time": "15.967",
"value": "Puerto Cabezas Airport#Improvement plans Puerto Cabezas Airport section 0"
},
{
"operation": "Highlighting",
"time": "23.356",
"value": "Puerto Cabezas Airport_sentence_7"
},
{
"operation": "finish",
"time": "37.975",
"value": "finish"
}
]
}
|
exfever_5823
|
Jim Parsons is an American actor who appears in a sitcom for an American television network with the broadcast station that originates a television network.
|
{
"evidence": [
"Jim Parsons",
"CBS",
"Flagship (broadcasting)"
],
"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:\nJim Parsons is an American actor who appears in a sitcom for an American television network with the broadcast station that originates a television network.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jim Parsons",
"CBS",
"Flagship (broadcasting)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 324,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
47798
|
Jumber Jikia (born in 1950) studied at Tbilisi School of Art and worked on a statue of Oliver.
|
{
"evidence": [
"Jumber Jikia_sentence_2",
"Jumber Jikia_sentence_3",
"Jumber Jikia_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:\nJumber Jikia (born in 1950) studied at Tbilisi School of Art and worked on a statue of Oliver.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Jumber Jikia_sentence_2",
"Jumber Jikia_sentence_3",
"Jumber Jikia_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 325,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.916",
"value": "?search="
},
{
"operation": "search",
"time": "12.997",
"value": "Jumber Jikia"
},
{
"operation": "Now on",
"time": "14.702",
"value": "Jumber Jikia"
},
{
"operation": "Highlighting",
"time": "18.169",
"value": "Jumber Jikia_sentence_2"
},
{
"operation": "Highlighting",
"time": "21.017",
"value": "Jumber Jikia_sentence_3"
},
{
"operation": "hyperlink",
"time": "26.595",
"value": "Giorgi Margvelashvili"
},
{
"operation": "Now on",
"time": "27.435",
"value": "Giorgi Margvelashvili"
},
{
"operation": "back-button-clicked",
"time": "29.043",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "29.099",
"value": "Jumber Jikia"
},
{
"operation": "Highlighting",
"time": "30.019",
"value": "Jumber Jikia_sentence_4"
},
{
"operation": "finish",
"time": "36.814",
"value": "finish"
}
]
}
|
exfever_2845
|
Ghostbusters is a 2019 American supernatural comedy film directed by an American actor, comedian and filmmaker who is known for creating the American science fiction comedy streaming television series that premiered on April 14, 2015.
|
{
"evidence": [
"Ghostbusters (2016 film)",
"Paul Feig",
"Other Space"
],
"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:\nGhostbusters is a 2019 American supernatural comedy film directed by an American actor, comedian and filmmaker who is known for creating the American science fiction comedy streaming television series that premiered on April 14, 2015.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ghostbusters (2016 film)",
"Paul Feig",
"Other Space"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 326,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
26854
|
After losing the first five games of the 1931–32 regular season, the Toronto Maple Leafs, whose colours are navy blue and white, fired head coach Art Duncan and replaced him with Dick Irvin, who had an 8–3–2 record in his first month behind the bench.
|
{
"evidence": [
"1931–32 Toronto Maple Leafs season_sentence_5",
"1931–32 Toronto Maple Leafs season_sentence_6",
"1931–32 Toronto Maple Leafs season_sentence_8"
],
"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:\nAfter losing the first five games of the 1931–32 regular season, the Toronto Maple Leafs, whose colours are navy blue and white, fired head coach Art Duncan and replaced him with Dick Irvin, who had an 8–3–2 record in his first month behind the bench.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1931–32 Toronto Maple Leafs season_sentence_5",
"1931–32 Toronto Maple Leafs season_sentence_6",
"1931–32 Toronto Maple Leafs season_sentence_8"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 327,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.912",
"value": "?search="
},
{
"operation": "search",
"time": "15.203",
"value": "Toronto Maple Leafs"
},
{
"operation": "Now on",
"time": "21.341",
"value": "Toronto Maple Leafs"
},
{
"operation": "hyperlink",
"time": "46.541",
"value": "1.2.1 Opening of Maple Leaf Gardens (1930s) Toronto Maple Leafs_section_3"
},
{
"operation": "Page search",
"time": "98.944",
"value": "art"
},
{
"operation": "Page search",
"time": "117.407",
"value": "art duncan"
},
{
"operation": "Page search",
"time": "143.978",
"value": "fir"
},
{
"operation": "Page search",
"time": "144.601",
"value": "fire"
},
{
"operation": "hyperlink",
"time": "238.212",
"value": "Dick Irvin"
},
{
"operation": "Now on",
"time": "240.889",
"value": "Dick Irvin"
},
{
"operation": "Page search",
"time": "289.738",
"value": "coach"
},
{
"operation": "hyperlink",
"time": "337.046",
"value": "Conn Smythe"
},
{
"operation": "Now on",
"time": "339.981",
"value": "Conn Smythe"
},
{
"operation": "search",
"time": "353.561",
"value": "art duncan"
},
{
"operation": "Now on",
"time": "354.833",
"value": "Art Duncan"
},
{
"operation": "Highlighting",
"time": "419.021",
"value": "Art Duncan_sentence_29"
},
{
"operation": "Highlighting",
"time": "419.026",
"value": "Art Duncan_title"
},
{
"operation": "Highlighting",
"time": "419.028",
"value": "Art Duncan_section_2"
},
{
"operation": "Highlighting deleted",
"time": "422.608",
"value": "Art Duncan_sentence_29"
},
{
"operation": "Highlighting deleted",
"time": "422.615",
"value": "Art Duncan_title"
},
{
"operation": "Highlighting deleted",
"time": "422.619",
"value": "Art Duncan_section_2"
},
{
"operation": "Highlighting",
"time": "424.217",
"value": "Art Duncan_sentence_29"
},
{
"operation": "Highlighting",
"time": "424.223",
"value": "Art Duncan_title"
},
{
"operation": "Highlighting",
"time": "424.224",
"value": "Art Duncan_section_2"
},
{
"operation": "hyperlink",
"time": "426.664",
"value": "Toronto Maple Leafs"
},
{
"operation": "Now on",
"time": "427.957",
"value": "Toronto Maple Leafs"
},
{
"operation": "Page search",
"time": "443.387",
"value": "1931"
},
{
"operation": "search",
"time": "561.627",
"value": "1931-32 Toronto Maple Leafs"
},
{
"operation": "Now on",
"time": "565.364",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "567.566",
"value": "1931–32 Toronto Maple Leafs season"
},
{
"operation": "Now on",
"time": "572.255",
"value": "1931–32 Toronto Maple Leafs season"
},
{
"operation": "Highlighting",
"time": "582.787",
"value": "1931–32 Toronto Maple Leafs season_sentence_5"
},
{
"operation": "Highlighting",
"time": "585.334",
"value": "1931–32 Toronto Maple Leafs season_sentence_6"
},
{
"operation": "Highlighting",
"time": "606.47",
"value": "1931–32 Toronto Maple Leafs season_sentence_8"
},
{
"operation": "finish",
"time": "634.752",
"value": "finish"
}
]
}
|
exfever_4152
|
Herbert Marshall McLuhan began his teaching career as a professor of English at several universities in the United States in 1945.
|
{
"evidence": [
"Marshall McLuhan",
"Winnipeg"
],
"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:\nHerbert Marshall McLuhan began his teaching career as a professor of English at several universities in the United States in 1945.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Marshall McLuhan",
"Winnipeg"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 328,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
41611
|
Nabiha Bensouda is an American artist that only sang folk songs.
|
{
"evidence": [
"Nabiha_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:\nNabiha Bensouda is an American artist that only sang folk songs.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Nabiha_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 329,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.397",
"value": "?search="
},
{
"operation": "search",
"time": "11.998",
"value": "Nabiha Bensouda"
},
{
"operation": "Now on",
"time": "13.066",
"value": "Nabiha"
},
{
"operation": "Highlighting",
"time": "26.77",
"value": "Nabiha_sentence_0"
},
{
"operation": "finish",
"time": "35.11",
"value": "finish"
}
]
}
|
exfever_5884
|
Gwen Renée Stefani is an American singer-songwriter and co-founder of the band No Doubt, whose singles include a song on their third studio album, produced by Blake Shelton.
|
{
"evidence": [
"Gwen Stefani",
"Spiderwebs (song)"
],
"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:\nGwen Renée Stefani is an American singer-songwriter and co-founder of the band No Doubt, whose singles include a song on their third studio album, produced by Blake Shelton.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gwen Stefani",
"Spiderwebs (song)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 330,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4574
|
Sansa Stark is a fictional character introduced in the first novel of A Song of Ice and Fire that won an annual set of literary awards voted on by readers of the science fiction and fantasy magazine Locus.
|
{
"evidence": [
"Sansa Stark",
"A Game of Thrones",
"Locus Award"
],
"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:\nSansa Stark is a fictional character introduced in the first novel of A Song of Ice and Fire that won an annual set of literary awards voted on by readers of the science fiction and fantasy magazine Locus.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sansa Stark",
"A Game of Thrones",
"Locus Award"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 331,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
1667
|
The Alabama State Janitor has been a political office since 1819, as the head banker for the State of Alabama, handling deposits, withdrawals, redemptions of state warrants and investments of state Beginning in 1861, the treasurer was elected biennially from until 1868 and Arthur Bingham was then elected by popular vote for a term of two years.
|
{
"evidence": [
"Alabama State Treasurer_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 Alabama State Janitor has been a political office since 1819, as the head banker for the State of Alabama, handling deposits, withdrawals, redemptions of state warrants and investments of state Beginning in 1861, the treasurer was elected biennially from until 1868 and Arthur Bingham was then elected by popular vote for a term of two years.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Alabama State Treasurer_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 332,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "13.528",
"value": "alabama state janitor"
},
{
"operation": "search",
"time": "56.854",
"value": "alabama state treasurer"
},
{
"operation": "Highlighting",
"time": "73.334",
"value": "Alabama State Treasurer_sentence_0"
},
{
"operation": "finish",
"time": "136.516",
"value": "finish"
}
]
}
|
exfever_4140
|
The Vampire Diaries is a Japanese supernatural teen drama television series that premiered on an American English-language commercial broadcast television, a limited liability joint venture between the CBS Entertainment Group unit of Paramount Global.
|
{
"evidence": [
"The Vampire Diaries",
"The CW"
],
"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 Vampire Diaries is a Japanese supernatural teen drama television series that premiered on an American English-language commercial broadcast television, a limited liability joint venture between the CBS Entertainment Group unit of Paramount Global.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Vampire Diaries",
"The CW"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 333,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1323
|
The Voice is the first singing competition television show in America.
|
{
"evidence": [
"The Voice (U.S. TV series)",
"Universal Music Group"
],
"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 Voice is the first singing competition television show in America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Voice (U.S. TV series)",
"Universal Music Group"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 334,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
86500
|
Wuhu ranked first in terms of GDP, with the Primary Industry being its biggest contributor, among the 16 prefectures in Anhui Province.
|
{
"evidence": [
"Anhui_sentence_69",
"Anhui_sentence_90",
"Anhui_sentence_91"
],
"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:\nWuhu ranked first in terms of GDP, with the Primary Industry being its biggest contributor, among the 16 prefectures in Anhui Province.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Anhui_sentence_69",
"Anhui_sentence_90",
"Anhui_sentence_91"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 335,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.735",
"value": "?search="
},
{
"operation": "search",
"time": "17.697",
"value": "Anhui Province"
},
{
"operation": "Now on",
"time": "19.673",
"value": "Anhui"
},
{
"operation": "hyperlink",
"time": "42.961",
"value": "6.1.3 Wuhu Economic and Technological Development Zone Anhui_section_10"
},
{
"operation": "search",
"time": "118.887",
"value": "Wuhu"
},
{
"operation": "Now on",
"time": "119.818",
"value": "WUHU"
},
{
"operation": "hyperlink",
"time": "123.867",
"value": "Wuhu"
},
{
"operation": "Now on",
"time": "124.246",
"value": "Wuhu (disambiguation)"
},
{
"operation": "Highlighting",
"time": "129.421",
"value": "Wuhu (disambiguation)_item_0_0"
},
{
"operation": "Highlighting",
"time": "129.424",
"value": "Wuhu (disambiguation)_title"
},
{
"operation": "Highlighting deleted",
"time": "139.961",
"value": "Wuhu (disambiguation)_item_0_0"
},
{
"operation": "Highlighting deleted",
"time": "139.962",
"value": "Wuhu (disambiguation)_title"
},
{
"operation": "search",
"time": "149.483",
"value": "Contains...Wuhu ranked first in terms of GDP"
},
{
"operation": "Now on",
"time": "151.383",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "163.015",
"value": "Wuhu Economic and Technological Development Zone Anhui_section_10"
},
{
"operation": "Now on",
"time": "164.698",
"value": "Anhui#Wuhu Economic and Technological Development Zone Anhui section 10"
},
{
"operation": "hyperlink",
"time": "176.866",
"value": "6.1.3 Wuhu Economic and Technological Development Zone Anhui_section_10"
},
{
"operation": "search",
"time": "259.786",
"value": "Contains...he Primary Industry being its biggest contributor"
},
{
"operation": "Now on",
"time": "262.417",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "273.842",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "274.02",
"value": "Anhui#Wuhu Economic and Technological Development Zone Anhui section 10"
},
{
"operation": "hyperlink",
"time": "303.63",
"value": "6.1 Major economic and technological development zones Anhui_section_7"
},
{
"operation": "Highlighting",
"time": "351.038",
"value": "Anhui_sentence_69"
},
{
"operation": "Highlighting",
"time": "443.641",
"value": "Anhui_sentence_90"
},
{
"operation": "Highlighting",
"time": "454.04",
"value": "Anhui_sentence_91"
},
{
"operation": "finish",
"time": "490.939",
"value": "finish"
}
]
}
|
exfever_7168
|
Wrath of the Titans was the highest-grossing film in 2012.
|
{
"evidence": [
"Wrath of the Titans",
"Rosamund Pike"
],
"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:\nWrath of the Titans was the highest-grossing film in 2012.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wrath of the Titans",
"Rosamund Pike"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 336,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6453
|
Ohtani was the first Japanese pitcher to win the MVP award.
|
{
"evidence": [
"Major League Baseball Most Valuable Player Award",
"Major League Baseball"
],
"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:\nOhtani was the first Japanese pitcher to win the MVP award.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Major League Baseball Most Valuable Player Award",
"Major League Baseball"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 337,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3021
|
A newspaper is a periodical publication containing written information about current events, including an article about a recently deceased person.
|
{
"evidence": [
"Newspaper",
"Obituary"
],
"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:\nA newspaper is a periodical publication containing written information about current events, including an article about a recently deceased person.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Newspaper",
"Obituary"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 338,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
62179
|
Shasta Lake is the largest city in Shasta Cascade region in California.
|
{
"evidence": [
"Redding, California_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:\nShasta Lake is the largest city in Shasta Cascade region in California.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Redding, California_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 339,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.514",
"value": "?search="
},
{
"operation": "search",
"time": "16.194",
"value": "Shasta Lake, California"
},
{
"operation": "Now on",
"time": "17.361",
"value": "Shasta Lake, California"
},
{
"operation": "Highlighting",
"time": "60.91",
"value": "Shasta Lake, California_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "60.911",
"value": "Shasta Lake, California_title"
},
{
"operation": "Highlighting",
"time": "60.921",
"value": "Shasta Lake, California_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "60.922",
"value": "Shasta Lake, California_header_cell_0_5_0"
},
{
"operation": "search",
"time": "179.431",
"value": "Shasta Cascade Region"
},
{
"operation": "Now on",
"time": "180.972",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "203.601",
"value": "Redding, California"
},
{
"operation": "Now on",
"time": "205.66",
"value": "Redding, California"
},
{
"operation": "Highlighting",
"time": "223.336",
"value": "Redding, California_sentence_4"
},
{
"operation": "finish",
"time": "282.251",
"value": "finish"
}
]
}
|
63130
|
Camp Century was an Arctic United States military scientific research base east of Thule Air Base under Project Iceworm which aimed to build a network of mobile nuclear missile launch sites under the Norway ice sheet
|
{
"evidence": [
"Project Iceworm_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:\nCamp Century was an Arctic United States military scientific research base east of Thule Air Base under Project Iceworm which aimed to build a network of mobile nuclear missile launch sites under the Norway ice sheet\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Project Iceworm_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 340,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.797",
"value": "?search="
},
{
"operation": "search",
"time": "12.506",
"value": "Camp Century"
},
{
"operation": "Now on",
"time": "13.452",
"value": "Camp Century"
},
{
"operation": "Highlighting",
"time": "16.741",
"value": "Camp Century_sentence_0"
},
{
"operation": "Highlighting",
"time": "16.742",
"value": "Camp Century_title"
},
{
"operation": "Highlighting",
"time": "26.304",
"value": "Camp Century_sentence_1"
},
{
"operation": "Highlighting",
"time": "30.455",
"value": "Camp Century_sentence_3"
},
{
"operation": "hyperlink",
"time": "40.108",
"value": "Project Iceworm"
},
{
"operation": "Now on",
"time": "40.929",
"value": "Project Iceworm"
},
{
"operation": "Highlighting",
"time": "44.352",
"value": "Project Iceworm_sentence_0"
},
{
"operation": "finish",
"time": "61.804",
"value": "finish"
}
]
}
|
34346
|
At the beginning of operation, the FedEx Asia-Pacific Hub in China employed more than 800 people and operated 136 flights a week, despite it being one of the smaller FedEx hubs outside the United States.
|
{
"evidence": [
"Guangzhou Baiyun International Airport_sentence_27"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAt the beginning of operation, the FedEx Asia-Pacific Hub in China employed more than 800 people and operated 136 flights a week, despite it being one of the smaller FedEx hubs outside the United States.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Guangzhou Baiyun International Airport_sentence_27"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 341,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.232",
"value": "?search="
},
{
"operation": "search",
"time": "47.903",
"value": "Contains...Fedex asia-pacific hub"
},
{
"operation": "Now on",
"time": "49.674",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "67.25",
"value": "FedEx Asia-Pacific hub Guangzhou Baiyun International Airport_section_7"
},
{
"operation": "Now on",
"time": "68.747",
"value": "Guangzhou Baiyun International Airport#FedEx Asia-Pacific hub Guangzhou Baiyun International Airport section 7"
},
{
"operation": "Highlighting",
"time": "125.694",
"value": "Guangzhou Baiyun International Airport_sentence_26"
},
{
"operation": "Highlighting deleted",
"time": "195.075",
"value": "Guangzhou Baiyun International Airport_sentence_26"
},
{
"operation": "Highlighting",
"time": "195.974",
"value": "Guangzhou Baiyun International Airport_sentence_27"
},
{
"operation": "finish",
"time": "325.167",
"value": "finish"
}
]
}
|
exfever_7088
|
Hilary Ann Swank had her breakthrough by starring as Julie Pierce in the fourth installment in The Karate Kid franchise, following the 1989 American martial arts drama film.
|
{
"evidence": [
"Hilary Swank",
"The Next Karate Kid",
"The Karate Kid Part III"
],
"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:\nHilary Ann Swank had her breakthrough by starring as Julie Pierce in the fourth installment in The Karate Kid franchise, following the 1989 American martial arts drama film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hilary Swank",
"The Next Karate Kid",
"The Karate Kid Part III"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 342,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3134
|
Radioactive is an alternative rock song with elements of electronic dance music that originated in East London.
|
{
"evidence": [
"Radioactive (Imagine Dragons song)",
"Dubstep"
],
"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:\nRadioactive is an alternative rock song with elements of electronic dance music that originated in East London.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Radioactive (Imagine Dragons song)",
"Dubstep"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 343,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
57398
|
Malcolm Claiborne along with 25 of 29 African Americans elected in Georgia Legislature in 1868(when the State Capitol was moved to Atlanta) were denied seats by their white colleagues until federal intervention allowed them in 1870.
|
{
"evidence": [
"Malcolm Claiborne_sentence_1",
"Malcolm Claiborne_sentence_2",
"Atlanta_sentence_43"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMalcolm Claiborne along with 25 of 29 African Americans elected in Georgia Legislature in 1868(when the State Capitol was moved to Atlanta) were denied seats by their white colleagues until federal intervention allowed them in 1870.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Malcolm Claiborne_sentence_1",
"Malcolm Claiborne_sentence_2",
"Atlanta_sentence_43"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 344,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.527",
"value": "undefined"
},
{
"operation": "search",
"time": "44.886",
"value": "Malcolm Claiborne"
},
{
"operation": "Now on",
"time": "45.625",
"value": "Malcolm Claiborne"
},
{
"operation": "Highlighting",
"time": "52.2",
"value": "Malcolm Claiborne_sentence_1"
},
{
"operation": "Highlighting",
"time": "284.833",
"value": "Malcolm Claiborne_sentence_2"
},
{
"operation": "search",
"time": "297.427",
"value": "Atlanta"
},
{
"operation": "Now on",
"time": "299.269",
"value": "Atlanta"
},
{
"operation": "Page search",
"time": "303.888",
"value": "1868"
},
{
"operation": "Highlighting",
"time": "310.737",
"value": "Atlanta_sentence_43"
},
{
"operation": "finish",
"time": "369.658",
"value": "finish"
}
]
}
|
exfever_6591
|
The second season of "Psych" stars an American actor, director, and screenwriter as Shawn Spencer, a fake psychic detective.
|
{
"evidence": [
"Psych (season 2)",
"James Roday"
],
"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 second season of \"Psych\" stars an American actor, director, and screenwriter as Shawn Spencer, a fake psychic detective.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Psych (season 2)",
"James Roday"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 345,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4885
|
Audrey Horne is the half-sister of a fictional character in the ABC television series Twin Peaks that was portrayed by Peggy Lipton.
|
{
"evidence": [
"Audrey Horne",
"Donna Hayward"
],
"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:\nAudrey Horne is the half-sister of a fictional character in the ABC television series Twin Peaks that was portrayed by Peggy Lipton.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Audrey Horne",
"Donna Hayward"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 346,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6973
|
Gilmore Girls is the most successful show on The WB.
|
{
"evidence": [
"Gilmore Girls",
"Lauren Graham"
],
"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:\nGilmore Girls is the most successful show on The WB.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gilmore Girls",
"Lauren Graham"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 347,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
51294
|
UEFA Euro 1988 squads had 20 players, the person in first position was Sjaak Troost (born 28 August 1959).
|
{
"evidence": [
"UEFA Euro 1988_sentence_78"
],
"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:\nUEFA Euro 1988 squads had 20 players, the person in first position was Sjaak Troost (born 28 August 1959).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"UEFA Euro 1988_sentence_78"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 348,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.1",
"value": "?search="
},
{
"operation": "search",
"time": "17.025",
"value": "UEFA Euro 1988"
},
{
"operation": "Now on",
"time": "23.624",
"value": "UEFA Euro 1988"
},
{
"operation": "Highlighting",
"time": "50.162",
"value": "UEFA Euro 1988_sentence_78"
},
{
"operation": "hyperlink",
"time": "139.88",
"value": "UEFA Euro 1988 Final"
},
{
"operation": "Now on",
"time": "140.945",
"value": "UEFA Euro 1988 Final"
},
{
"operation": "search",
"time": "168.576",
"value": "Sjaak Troost"
},
{
"operation": "Now on",
"time": "171.481",
"value": "Sjaak Troost"
},
{
"operation": "hyperlink",
"time": "183.816",
"value": "1988"
},
{
"operation": "Now on",
"time": "185.052",
"value": "UEFA Euro 1988"
},
{
"operation": "back-button-clicked",
"time": "197.033",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "197.147",
"value": "Sjaak Troost"
},
{
"operation": "hyperlink",
"time": "198.943",
"value": "UEFA European Championship"
},
{
"operation": "Now on",
"time": "203.676",
"value": "UEFA European Championship"
},
{
"operation": "back-button-clicked",
"time": "207.624",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "207.706",
"value": "Sjaak Troost"
},
{
"operation": "hyperlink",
"time": "220.447",
"value": "UEFA Euro 1988"
},
{
"operation": "Now on",
"time": "221.11",
"value": "UEFA Euro 1988"
},
{
"operation": "finish",
"time": "311.316",
"value": "finish"
}
]
}
|
48239
|
Hanjin Venezia collided with the Delta Tower of the San Franciso-Oakland Bay Bridge but was not the vessel responsible for the Cosco Busan oil spill.
|
{
"evidence": [
"Hanjin Venezia_sentence_1",
"Hanjin Venezia_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:\nHanjin Venezia collided with the Delta Tower of the San Franciso-Oakland Bay Bridge but was not the vessel responsible for the Cosco Busan oil spill.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hanjin Venezia_sentence_1",
"Hanjin Venezia_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 349,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.367",
"value": "?search="
},
{
"operation": "search",
"time": "25.307",
"value": "Hanjin Venezia"
},
{
"operation": "Now on",
"time": "26.319",
"value": "Hanjin Venezia"
},
{
"operation": "Page search",
"time": "38.036",
"value": "Delta Tower"
},
{
"operation": "Highlighting",
"time": "40.407",
"value": "Hanjin Venezia_sentence_1"
},
{
"operation": "Page search",
"time": "87.649",
"value": "spill"
},
{
"operation": "Highlighting",
"time": "90.23",
"value": "Hanjin Venezia_sentence_2"
},
{
"operation": "finish",
"time": "113.542",
"value": "finish"
}
]
}
|
exfever_8487
|
Heartbreak Ridge is an American war film directed by Clint Eastwood whose title comes from a month-long battle in the Korean War which took place between September 13th and October 15th, 1951.
|
{
"evidence": [
"Heartbreak Ridge",
"Battle of Heartbreak Ridge"
],
"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:\nHeartbreak Ridge is an American war film directed by Clint Eastwood whose title comes from a month-long battle in the Korean War which took place between September 13th and October 15th, 1951.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Heartbreak Ridge",
"Battle of Heartbreak Ridge"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 350,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5597
|
Doctor Stephen Strange is a fictional character appearing in a thick book, averaging 32 pages and containing comics published by Marvel Comics.
|
{
"evidence": [
"Doctor Strange",
"American comic book"
],
"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:\nDoctor Stephen Strange is a fictional character appearing in a thick book, averaging 32 pages and containing comics published by Marvel Comics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Doctor Strange",
"American comic book"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 351,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5745
|
A shed is typically a simple, double-story roofed structure that is used for a working area that typically contains a workbench, hand tools, power tools, and other hardware.
|
{
"evidence": [
"Shed",
"Workshop"
],
"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:\nA shed is typically a simple, double-story roofed structure that is used for a working area that typically contains a workbench, hand tools, power tools, and other hardware.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Shed",
"Workshop"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 352,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
85622
|
Antipathy, observed by Sophie Bryant (an Anglo-Irish mathematician, educator, feminist and activist), is based on the positive character of others.
|
{
"evidence": [
"Sophie Bryant_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:\nAntipathy, observed by Sophie Bryant (an Anglo-Irish mathematician, educator, feminist and activist), is based on the positive character of others.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Sophie Bryant_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 353,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.231",
"value": "?search="
},
{
"operation": "search",
"time": "58.866",
"value": "sophie bryant"
},
{
"operation": "Now on",
"time": "59.956",
"value": "Sophie Bryant"
},
{
"operation": "Page search",
"time": "65.994",
"value": "obser"
},
{
"operation": "Highlighting",
"time": "70.404",
"value": "Sophie Bryant_sentence_0"
},
{
"operation": "Page search",
"time": "76.338",
"value": "charac"
},
{
"operation": "search",
"time": "150.16",
"value": "Antipathy"
},
{
"operation": "Now on",
"time": "150.841",
"value": "Antipathy"
},
{
"operation": "finish",
"time": "166.463",
"value": "finish"
}
]
}
|
exfever_1911
|
Hugh John Mungo Grant is a British actor who received attention for his performances including a 1993 British-American drama film directed by James Ivory.
|
{
"evidence": [
"Hugh Grant",
"The Remains of the Day (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:\nHugh John Mungo Grant is a British actor who received attention for his performances including a 1993 British-American drama film directed by James Ivory.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hugh Grant",
"The Remains of the Day (film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 354,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
18368
|
Nevada was the first state to sanction the use of the gas chamber with its first execution in May 1924.
|
{
"evidence": [
"Gas chamber_sentence_34"
],
"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:\nNevada was the first state to sanction the use of the gas chamber with its first execution in May 1924.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Gas chamber_sentence_34"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 355,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.34",
"value": "?search="
},
{
"operation": "search",
"time": "17.326",
"value": "Nevada"
},
{
"operation": "Now on",
"time": "18.881",
"value": "Nevada"
},
{
"operation": "Page search",
"time": "25.723",
"value": "gas chamber"
},
{
"operation": "search",
"time": "55.167",
"value": "evada was the first state to sanction"
},
{
"operation": "Now on",
"time": "57.084",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "61.403",
"value": "evada was the first stage to sanctions"
},
{
"operation": "Now on",
"time": "61.818",
"value": "undefined"
},
{
"operation": "search",
"time": "78.429",
"value": "Death penalty US"
},
{
"operation": "Now on",
"time": "80.089",
"value": "Death penalty US"
},
{
"operation": "Page search",
"time": "90.151",
"value": "nevada"
},
{
"operation": "Page search",
"time": "137.183",
"value": "may 1924"
},
{
"operation": "Page search",
"time": "187.21",
"value": "gas chamber"
},
{
"operation": "Page search",
"time": "190.027",
"value": "gas chamber"
},
{
"operation": "search",
"time": "195.877",
"value": "Gas chamber"
},
{
"operation": "Now on",
"time": "196.968",
"value": "Gas chamber"
},
{
"operation": "search",
"time": "206.397",
"value": "may 1924"
},
{
"operation": "Now on",
"time": "207.341",
"value": "May 1924"
},
{
"operation": "search",
"time": "215.378",
"value": "nevada"
},
{
"operation": "Now on",
"time": "216.025",
"value": "Nevada"
},
{
"operation": "search",
"time": "230.617",
"value": "Gas chamber"
},
{
"operation": "Now on",
"time": "231.197",
"value": "Gas chamber"
},
{
"operation": "Page search",
"time": "280.261",
"value": "nevada"
},
{
"operation": "Highlighting",
"time": "303.824",
"value": "Gas chamber_sentence_34"
},
{
"operation": "hyperlink",
"time": "309.525",
"value": "Nevada"
},
{
"operation": "Now on",
"time": "310.096",
"value": "Nevada"
},
{
"operation": "Page search",
"time": "320.247",
"value": "gas chamber"
},
{
"operation": "finish",
"time": "358.889",
"value": "finish"
}
]
}
|
exfever_1063
|
Karan Kumar Johar is the only Indian filmmaker who has received more than 5 Filmfare Awards.
|
{
"evidence": [
"Karan Johar",
"Kuch Kuch Hota Hai"
],
"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:\nKaran Kumar Johar is the only Indian filmmaker who has received more than 5 Filmfare Awards.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Karan Johar",
"Kuch Kuch Hota Hai"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 356,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
82504
|
The 1979 season of the Cleveland Indians implemented a Farm system for their Minor League Baseball with levels AAA, AA, and A represented by Tacoma Tugs, Chattanooga Lookouts, and Waterloo Indians, respectively.
|
{
"evidence": [
"Kenesaw Mountain Landis_sentence_338",
"Kenesaw Mountain Landis_sentence_348",
"Kenesaw Mountain Landis_sentence_347"
],
"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 1979 season of the Cleveland Indians implemented a Farm system for their Minor League Baseball with levels AAA, AA, and A represented by Tacoma Tugs, Chattanooga Lookouts, and Waterloo Indians, respectively.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Kenesaw Mountain Landis_sentence_338",
"Kenesaw Mountain Landis_sentence_348",
"Kenesaw Mountain Landis_sentence_347"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 357,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.453",
"value": "undefined"
},
{
"operation": "search",
"time": "24.912",
"value": "Contains...Cleveland Indians implemented a Farm system for their Minor League"
},
{
"operation": "Now on",
"time": "26.585",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "107.51",
"value": "Kenesaw Mountain Landis"
},
{
"operation": "Now on",
"time": "110.208",
"value": "Kenesaw Mountain Landis"
},
{
"operation": "Highlighting",
"time": "476.816",
"value": "Kenesaw Mountain Landis_sentence_338"
},
{
"operation": "Highlighting",
"time": "564.298",
"value": "Kenesaw Mountain Landis_sentence_348"
},
{
"operation": "Highlighting",
"time": "581.314",
"value": "Kenesaw Mountain Landis_sentence_347"
},
{
"operation": "finish",
"time": "691.999",
"value": "finish"
}
]
}
|
exfever_1392
|
Burnt Offerings was distributed by Metro-Goldwyn-Mayer along with United Artists.
|
{
"evidence": [
"Robert Singer (producer)",
"Burnt Offerings (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:\nBurnt Offerings was distributed by Metro-Goldwyn-Mayer along with United Artists.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Robert Singer (producer)",
"Burnt Offerings (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 358,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6624
|
Damages is an American legal thriller that premiered on FX and aired for three seasons before moving to an American multichannel video programming distributor based in El Segundo, California.
|
{
"evidence": [
"Damages (TV series)",
"DirecTV"
],
"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:\nDamages is an American legal thriller that premiered on FX and aired for three seasons before moving to an American multichannel video programming distributor based in El Segundo, California.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Damages (TV series)",
"DirecTV"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 359,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3193
|
Imagine Dragons is an Australian pop rock band from the largest city within a xeric desert in the rain shadow of the Sierra Nevada mountains.
|
{
"evidence": [
"Imagine Dragons",
"Las Vegas",
"Mojave Desert"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nImagine Dragons is an Australian pop rock band from the largest city within a xeric desert in the rain shadow of the Sierra Nevada mountains.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Imagine Dragons",
"Las Vegas",
"Mojave Desert"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 360,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1542
|
Pulp Fiction was a high-grossing film written and directed by Quentin Tarantino.
|
{
"evidence": [
"Pulp Fiction",
"Roger Avary",
"The Rules of Attraction (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:\nPulp Fiction was a high-grossing film written and directed by Quentin Tarantino.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pulp Fiction",
"Roger Avary",
"The Rules of Attraction (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 361,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7069
|
Major film have dominated both American cinema and the global film industry. U.S. studios have benefited from a strong competitive advantage gained by the initial significant occupant of a market segment.
|
{
"evidence": [
"Major film studio",
"First-mover advantage"
],
"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:\nMajor film have dominated both American cinema and the global film industry. U.S. studios have benefited from a strong competitive advantage gained by the initial significant occupant of a market segment.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Major film studio",
"First-mover advantage"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 362,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
19416
|
The Martham railway station was on a line that went through Great Yarmouth and closed in 1959.
|
{
"evidence": [
"Martham railway station_sentence_1",
"Martham railway station_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 Martham railway station was on a line that went through Great Yarmouth and closed in 1959.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Martham railway station_sentence_1",
"Martham railway station_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 363,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.494",
"value": "?search="
},
{
"operation": "search",
"time": "14.875",
"value": "Martham railway station"
},
{
"operation": "Now on",
"time": "15.485",
"value": "Martham railway station"
},
{
"operation": "Highlighting",
"time": "24.181",
"value": "Martham railway station_sentence_1"
},
{
"operation": "Highlighting",
"time": "29.499",
"value": "Martham railway station_sentence_2"
},
{
"operation": "finish",
"time": "167.888",
"value": "finish"
}
]
}
|
18248
|
Socialist Propaganda League was a Democratic group that supported capitalism.
|
{
"evidence": [
"Socialist Propaganda League_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:\nSocialist Propaganda League was a Democratic group that supported capitalism.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Socialist Propaganda League_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 364,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.886",
"value": "?search="
},
{
"operation": "search",
"time": "23.538",
"value": "Socialist Propaganda League"
},
{
"operation": "Now on",
"time": "24.017",
"value": "Socialist Propaganda League"
},
{
"operation": "Highlighting",
"time": "84.971",
"value": "Socialist Propaganda League_sentence_1"
},
{
"operation": "hyperlink",
"time": "97.743",
"value": "Socialist Propaganda League of America"
},
{
"operation": "Now on",
"time": "98.367",
"value": "Socialist Propaganda League of America"
},
{
"operation": "search",
"time": "132.627",
"value": "Socialist Propaganda League"
},
{
"operation": "Now on",
"time": "133.451",
"value": "Socialist Propaganda League"
},
{
"operation": "Page search",
"time": "139.503",
"value": "democratic"
},
{
"operation": "finish",
"time": "157.485",
"value": "finish"
}
]
}
|
exfever_5930
|
Her is an American science-fiction film directed by Spike Jonze whose genre involves romantic love stories that focus on the affectionate romantic involvement of the main characters.
|
{
"evidence": [
"Her (film)",
"Romance 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:\nHer is an American science-fiction film directed by Spike Jonze whose genre involves romantic love stories that focus on the affectionate romantic involvement of the main characters.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Her (film)",
"Romance film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 365,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8906
|
Stockard Channing, a Canadian actress, is known for playing Betty Rizzo in a 1978 American musical romantic comedy film based on the 1971 musical of the same name.
|
{
"evidence": [
"Stockard Channing",
"Grease (film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nStockard Channing, a Canadian actress, is known for playing Betty Rizzo in a 1978 American musical romantic comedy film based on the 1971 musical of the same name.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stockard Channing",
"Grease (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 366,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1583
|
Matthew Groening is an American cartoonist who created the American animated series that is epitomized by a nuclear family, consisting of married couple Homer and Marge and their three children, Bart, Lisa, and Maggie.
|
{
"evidence": [
"Matt Groening",
"The Simpsons",
"Simpson family"
],
"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:\nMatthew Groening is an American cartoonist who created the American animated series that is epitomized by a nuclear family, consisting of married couple Homer and Marge and their three children, Bart, Lisa, and Maggie.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Matt Groening",
"The Simpsons",
"Simpson family"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 367,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3972
|
Asia has most Christian people than Africa.
|
{
"evidence": [
"Christianity",
"Life of Jesus in the New Testament"
],
"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:\nAsia has most Christian people than Africa.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Christianity",
"Life of Jesus in the New Testament"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 368,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5411
|
A Game of Thrones had a novella, Blood of the Dragon, comprising one of its characters cited by an American daily newspaper as one of the author's finest creations.
|
{
"evidence": [
"A Game of Thrones",
"Daenerys Targaryen",
"The New York Times"
],
"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:\nA Game of Thrones had a novella, Blood of the Dragon, comprising one of its characters cited by an American daily newspaper as one of the author's finest creations.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"A Game of Thrones",
"Daenerys Targaryen",
"The New York Times"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 369,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
48032
|
Souha Hatono, known for treating both sides at the Satsuma Rebellion equally, in accord with the spirit of International Red Cross, was a Japanese physician specializing in surgery.
|
{
"evidence": [
"Souha Hatono_sentence_0",
"Souha Hatono_sentence_1",
"Souha Hatono_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:\nSouha Hatono, known for treating both sides at the Satsuma Rebellion equally, in accord with the spirit of International Red Cross, was a Japanese physician specializing in surgery.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Souha Hatono_sentence_0",
"Souha Hatono_sentence_1",
"Souha Hatono_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 370,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.13",
"value": "undefined"
},
{
"operation": "search",
"time": "197.833",
"value": "Souha Hatono"
},
{
"operation": "Now on",
"time": "199.238",
"value": "Souha Hatono"
},
{
"operation": "Highlighting",
"time": "203.255",
"value": "Souha Hatono_sentence_0"
},
{
"operation": "Highlighting",
"time": "205.228",
"value": "Souha Hatono_sentence_1"
},
{
"operation": "Highlighting",
"time": "216.028",
"value": "Souha Hatono_sentence_3"
},
{
"operation": "finish",
"time": "230.322",
"value": "finish"
}
]
}
|
15860
|
Nachum Ish Gamzu was the teacher of Rabbi Akiva, and is known by interpreting the whole Torah according to the rule of "ribbui u-mi'uṭ".
|
{
"evidence": [
"Nachum Ish Gamzu_sentence_6",
"Nachum Ish Gamzu_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:\nNachum Ish Gamzu was the teacher of Rabbi Akiva, and is known by interpreting the whole Torah according to the rule of \"ribbui u-mi'uṭ\".\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Nachum Ish Gamzu_sentence_6",
"Nachum Ish Gamzu_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 371,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.728",
"value": "?search="
},
{
"operation": "search",
"time": "5.756",
"value": "nachum ish gamzu"
},
{
"operation": "Now on",
"time": "6.671",
"value": "Nachum Ish Gamzu"
},
{
"operation": "Highlighting",
"time": "218.142",
"value": "Nachum Ish Gamzu_sentence_6"
},
{
"operation": "Highlighting",
"time": "224.444",
"value": "Nachum Ish Gamzu_sentence_7"
},
{
"operation": "finish",
"time": "233.331",
"value": "finish"
}
]
}
|
28631
|
In the 2016 season of the Minnesota Vikings, Michael Griffin was signed in a one year.three million dollar contract, previously playing on the Detroit Lions.
|
{
"evidence": [
"Michael Griffin (American football)_sentence_70"
],
"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 the 2016 season of the Minnesota Vikings, Michael Griffin was signed in a one year.three million dollar contract, previously playing on the Detroit Lions.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Michael Griffin (American football)_sentence_70"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 372,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.408",
"value": "?search="
},
{
"operation": "search",
"time": "210.353",
"value": "2016 Minnesota Vikings season"
},
{
"operation": "Now on",
"time": "211.557",
"value": "2016 Minnesota Vikings season"
},
{
"operation": "Page search",
"time": "215.647",
"value": "griffin"
},
{
"operation": "hyperlink",
"time": "231.663",
"value": "Michael Griffin"
},
{
"operation": "Now on",
"time": "232.616",
"value": "Michael Griffin (American football)"
},
{
"operation": "Page search",
"time": "238.942",
"value": "contract"
},
{
"operation": "Highlighting",
"time": "257.775",
"value": "Michael Griffin (American football)_sentence_70"
},
{
"operation": "Page search",
"time": "289.281",
"value": "1.3"
},
{
"operation": "Page search",
"time": "308.534",
"value": "lions"
},
{
"operation": "Now on",
"time": "328.749",
"value": "Michael Griffin (American football)"
},
{
"operation": "Now on",
"time": "379.825",
"value": "Michael Griffin (American football)"
},
{
"operation": "finish",
"time": "387.975",
"value": "finish"
}
]
}
|
36878
|
In the 2010 Tesoro Anacortes Refinery disaster, the pressure-containing shell of the heat exchanger burst at its weld seams; the ensuing explosion was so violent that many in Anacortes felt the shock wave across Fidalgo Bay (an island neither in Skagit County nor in Washington).
|
{
"evidence": [
"Fidalgo Island_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:\nIn the 2010 Tesoro Anacortes Refinery disaster, the pressure-containing shell of the heat exchanger burst at its weld seams; the ensuing explosion was so violent that many in Anacortes felt the shock wave across Fidalgo Bay (an island neither in Skagit County nor in Washington).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Fidalgo Island_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 373,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.686",
"value": "?search="
},
{
"operation": "search",
"time": "16.332",
"value": "2010 Tesoro Anacortes Refinery disaster"
},
{
"operation": "Now on",
"time": "17.325",
"value": "2010 Tesoro Anacortes Refinery disaster"
},
{
"operation": "Highlighting",
"time": "30.765",
"value": "2010 Tesoro Anacortes Refinery disaster_sentence_3"
},
{
"operation": "Highlighting",
"time": "30.765",
"value": "2010 Tesoro Anacortes Refinery disaster_title"
},
{
"operation": "Highlighting",
"time": "30.766",
"value": "2010 Tesoro Anacortes Refinery disaster_section_0"
},
{
"operation": "Highlighting",
"time": "38.521",
"value": "2010 Tesoro Anacortes Refinery disaster_sentence_4"
},
{
"operation": "hyperlink",
"time": "42.458",
"value": "Fidalgo Bay"
},
{
"operation": "Now on",
"time": "43.009",
"value": "Fidalgo Island"
},
{
"operation": "Highlighting",
"time": "46.755",
"value": "Fidalgo Island_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "48.784",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "48.884",
"value": "2010 Tesoro Anacortes Refinery disaster"
},
{
"operation": "Highlighting deleted",
"time": "50.975",
"value": "2010 Tesoro Anacortes Refinery disaster_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "52.683",
"value": "2010 Tesoro Anacortes Refinery disaster_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "52.685",
"value": "2010 Tesoro Anacortes Refinery disaster_title"
},
{
"operation": "Highlighting deleted",
"time": "52.686",
"value": "2010 Tesoro Anacortes Refinery disaster_section_0"
},
{
"operation": "finish",
"time": "63.929",
"value": "finish"
}
]
}
|
exfever_4092
|
Beauty and the Beast is a 2017 American musical fantasy film co-produced by an American film production company headquartered in Burbank, California.
|
{
"evidence": [
"Beauty and the Beast (2017 film)",
"Mandeville Films"
],
"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:\nBeauty and the Beast is a 2017 American musical fantasy film co-produced by an American film production company headquartered in Burbank, California.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Beauty and the Beast (2017 film)",
"Mandeville Films"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 374,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
74710
|
Flossie Bailey, who was the president of the local branch of the NAACP, tried to obtain police protection when two African American men, Thomas Shipp and Abram Smith were hanged from a tree after a mob broke out in Grant County in August 1930.
|
{
"evidence": [
"Flossie Bailey_sentence_16",
"Flossie Bailey_sentence_17"
],
"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:\nFlossie Bailey, who was the president of the local branch of the NAACP, tried to obtain police protection when two African American men, Thomas Shipp and Abram Smith were hanged from a tree after a mob broke out in Grant County in August 1930.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Flossie Bailey_sentence_16",
"Flossie Bailey_sentence_17"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 375,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.29",
"value": "?search="
},
{
"operation": "search",
"time": "11.053",
"value": "Flossie Bailey"
},
{
"operation": "Now on",
"time": "12.229",
"value": "Flossie Bailey"
},
{
"operation": "Highlighting",
"time": "75.497",
"value": "Flossie Bailey_sentence_2"
},
{
"operation": "Highlighting",
"time": "147.024",
"value": "Flossie Bailey_sentence_17"
},
{
"operation": "Highlighting deleted",
"time": "149.465",
"value": "Flossie Bailey_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "166.846",
"value": "Flossie Bailey_sentence_17"
},
{
"operation": "Highlighting",
"time": "167.353",
"value": "Flossie Bailey_sentence_16"
},
{
"operation": "Highlighting",
"time": "169.601",
"value": "Flossie Bailey_sentence_17"
},
{
"operation": "finish",
"time": "189.593",
"value": "finish"
}
]
}
|
69837
|
Recorded in Clovis, New Mexico in 1959 with producer Norman Petty, Terry Teene's first album was released by Warwick Records in the UK.
|
{
"evidence": [
"Terry Teene_sentence_6",
"Terry Teene_sentence_7",
"Single (music)_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:\nRecorded in Clovis, New Mexico in 1959 with producer Norman Petty, Terry Teene's first album was released by Warwick Records in the UK.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Terry Teene_sentence_6",
"Terry Teene_sentence_7",
"Single (music)_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 376,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.988",
"value": "?search="
},
{
"operation": "search",
"time": "27.92",
"value": "Terry Teene"
},
{
"operation": "Now on",
"time": "29.098",
"value": "Terry Teene"
},
{
"operation": "Highlighting",
"time": "74.773",
"value": "Terry Teene_sentence_6"
},
{
"operation": "Highlighting",
"time": "82.849",
"value": "Terry Teene_sentence_7"
},
{
"operation": "hyperlink",
"time": "126.106",
"value": "single"
},
{
"operation": "Now on",
"time": "130.027",
"value": "Single (music)"
},
{
"operation": "Highlighting",
"time": "137.217",
"value": "Single (music)_sentence_0"
},
{
"operation": "finish",
"time": "152.016",
"value": "finish"
}
]
}
|
93519
|
In the Colgate Red Raiders 1993 season under Andrew Kerr, they compiled a six-one-one record and shut out five of the eight opponents; Andrew Kerr took them to a perfect season the previous year with a nine-zero record, they were not scored upon, and they were named national champions.
|
{
"evidence": [
"1993 Colgate Red Raiders football team_sentence_2",
"Andy Kerr Stadium_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:\nIn the Colgate Red Raiders 1993 season under Andrew Kerr, they compiled a six-one-one record and shut out five of the eight opponents; Andrew Kerr took them to a perfect season the previous year with a nine-zero record, they were not scored upon, and they were named national champions.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1993 Colgate Red Raiders football team_sentence_2",
"Andy Kerr Stadium_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 377,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.575",
"value": "?search="
},
{
"operation": "search",
"time": "13.571",
"value": "Contains...Colgate Red Raiders 1993 season"
},
{
"operation": "Now on",
"time": "15.286",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "19.015",
"value": "1993 Colgate Red Raiders football team"
},
{
"operation": "Now on",
"time": "19.669",
"value": "1993 Colgate Red Raiders football team"
},
{
"operation": "Highlighting",
"time": "35.321",
"value": "1993 Colgate Red Raiders football team_sentence_2"
},
{
"operation": "Highlighting",
"time": "43.175",
"value": "1993 Colgate Red Raiders football team_header_cell_0_5_0"
},
{
"operation": "Highlighting deleted",
"time": "46.946",
"value": "1993 Colgate Red Raiders football team_header_cell_0_5_0"
},
{
"operation": "hyperlink",
"time": "49.024",
"value": "Andy Kerr Stadium"
},
{
"operation": "Now on",
"time": "49.586",
"value": "Andy Kerr Stadium"
},
{
"operation": "Highlighting",
"time": "67.409",
"value": "Andy Kerr Stadium_sentence_3"
},
{
"operation": "back-button-clicked",
"time": "71.102",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "71.157",
"value": "1993 Colgate Red Raiders football team"
},
{
"operation": "finish",
"time": "163.531",
"value": "finish"
}
]
}
|
80591
|
Hackers posted fake remarks on the official Qatar News Agency attributed to the Emir of Qatar, the emir is quoted praising Iran but the remarks were widely publicized in the various Arab news media.
|
{
"evidence": [
"Qatar News Agency_sentence_12",
"Qatar News Agency_sentence_16"
],
"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:\nHackers posted fake remarks on the official Qatar News Agency attributed to the Emir of Qatar, the emir is quoted praising Iran but the remarks were widely publicized in the various Arab news media.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Qatar News Agency_sentence_12",
"Qatar News Agency_sentence_16"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 378,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.828",
"value": "?search="
},
{
"operation": "search",
"time": "121.457",
"value": "Emir of Qatar"
},
{
"operation": "Now on",
"time": "122.435",
"value": "Emir of Qatar"
},
{
"operation": "search",
"time": "159.975",
"value": "Qatar News Agency"
},
{
"operation": "Now on",
"time": "160.699",
"value": "Qatar News Agency"
},
{
"operation": "Page search",
"time": "167.821",
"value": "hackers"
},
{
"operation": "Highlighting",
"time": "305.291",
"value": "Qatar News Agency_sentence_19"
},
{
"operation": "Highlighting",
"time": "306.3",
"value": "Qatar News Agency_sentence_18"
},
{
"operation": "Highlighting",
"time": "321.672",
"value": "Qatar News Agency_sentence_12"
},
{
"operation": "Highlighting deleted",
"time": "329.048",
"value": "Qatar News Agency_sentence_18"
},
{
"operation": "Highlighting deleted",
"time": "329.708",
"value": "Qatar News Agency_sentence_19"
},
{
"operation": "Highlighting",
"time": "339.83",
"value": "Qatar News Agency_sentence_16"
},
{
"operation": "finish",
"time": "345.772",
"value": "finish"
}
]
}
|
32633
|
Judiciary of Somalia proceedings are ordinarily private, situations can cause the court to go from public to private.
|
{
"evidence": [
"Judiciary of Somalia_sentence_18"
],
"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:\nJudiciary of Somalia proceedings are ordinarily private, situations can cause the court to go from public to private.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Judiciary of Somalia_sentence_18"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 379,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.643",
"value": "?search="
},
{
"operation": "search",
"time": "7.337",
"value": "Judiciary of Somalia"
},
{
"operation": "Now on",
"time": "8.254",
"value": "Judiciary of Somalia"
},
{
"operation": "Highlighting",
"time": "79.19",
"value": "Judiciary of Somalia_sentence_18"
},
{
"operation": "finish",
"time": "92.623",
"value": "finish"
}
]
}
|
exfever_1421
|
The Book of Mormon is a musical comedy by Robert Lopez, Matt Stone, and an American actor, animator, writer, producer, director, and composer.
|
{
"evidence": [
"The Book of Mormon (musical)",
"Trey Parker"
],
"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 Book of Mormon is a musical comedy by Robert Lopez, Matt Stone, and an American actor, animator, writer, producer, director, and composer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Book of Mormon (musical)",
"Trey Parker"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 380,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4839
|
Vera Ann Farmiga is an American actress best known for portraying paranormal investigator Lorraine Warren in a 2016 American supernatural horror film directed by James Wan.
|
{
"evidence": [
"Vera Farmiga",
"The Conjuring 2"
],
"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:\nVera Ann Farmiga is an American actress best known for portraying paranormal investigator Lorraine Warren in a 2016 American supernatural horror film directed by James Wan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Vera Farmiga",
"The Conjuring 2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 381,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7315
|
Tobey Maguire is an American actor whose other films include an American Revisionist Western film written by Ang Lee.
|
{
"evidence": [
"Tobey Maguire",
"Ride with the Devil (film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTobey Maguire is an American actor whose other films include an American Revisionist Western film written by Ang Lee.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tobey Maguire",
"Ride with the Devil (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 382,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
12706
|
Mosquito (Spanish and Portuguese for "little fly") control has three toxicants one of which is DDT.
|
{
"evidence": [
"Mosquito_sentence_4",
"DDT_sentence_14"
],
"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:\nMosquito (Spanish and Portuguese for \"little fly\") control has three toxicants one of which is DDT.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Mosquito_sentence_4",
"DDT_sentence_14"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 383,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.558",
"value": "?search="
},
{
"operation": "search",
"time": "4.258",
"value": "Mosquito"
},
{
"operation": "Now on",
"time": "5.372",
"value": "Mosquito"
},
{
"operation": "Highlighting",
"time": "10.377",
"value": "Mosquito_sentence_4"
},
{
"operation": "Page search",
"time": "23.738",
"value": "toxicants"
},
{
"operation": "Page search",
"time": "32.832",
"value": "toxicant"
},
{
"operation": "Page search",
"time": "46.985",
"value": "DDT"
},
{
"operation": "search",
"time": "83.85",
"value": "DDT"
},
{
"operation": "Now on",
"time": "84.912",
"value": "DDT"
},
{
"operation": "Page search",
"time": "92.077",
"value": "Mosquito"
},
{
"operation": "Highlighting",
"time": "104.573",
"value": "DDT_sentence_14"
},
{
"operation": "hyperlink",
"time": "163.826",
"value": "National Toxicology Program"
},
{
"operation": "Now on",
"time": "164.507",
"value": "National Toxicology Program"
},
{
"operation": "Page search",
"time": "169.237",
"value": "mosq"
},
{
"operation": "finish",
"time": "177.285",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.