id
stringlengths 1
12
| question
stringlengths 22
566
| golden_answers
dict | data_source
stringclasses 2
values | prompt
listlengths 1
1
| ability
stringclasses 6
values | reward_model
dict | extra_info
dict | metadata
dict |
|---|---|---|---|---|---|---|---|---|
exfever_5960
|
Drop Dead Gorgeous is a 2000 American satirical black comedy mockumentary film starring talented artists, including an American actress and singer born in Atlanta.
|
{
"evidence": [
"Drop Dead Gorgeous (film)",
"Brittany Murphy"
],
"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:\nDrop Dead Gorgeous is a 2000 American satirical black comedy mockumentary film starring talented artists, including an American actress and singer born in Atlanta.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Drop Dead Gorgeous (film)",
"Brittany Murphy"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5284,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1045
|
Divergent is a 2014 film directed by Neil Burger, based on the debut novel of Italian novelist Veronica Roth, published by Harper Collins Children's Books in 2011.
|
{
"evidence": [
"Divergent (film)",
"Divergent (novel)"
],
"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:\nDivergent is a 2014 film directed by Neil Burger, based on the debut novel of Italian novelist Veronica Roth, published by Harper Collins Children's Books in 2011.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Divergent (film)",
"Divergent (novel)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5285,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
33373
|
Acta Neuropathologica is a monthly published journal which discusses topics related to Neuropathology.
|
{
"evidence": [
"Acta Neuropathologica_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:\nActa Neuropathologica is a monthly published journal which discusses topics related to Neuropathology.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Acta Neuropathologica_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5286,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.953",
"value": "?search="
},
{
"operation": "search",
"time": "803.223",
"value": "Acta Neuropathologica"
},
{
"operation": "Now on",
"time": "805.37",
"value": "Acta Neuropathologica"
},
{
"operation": "Highlighting",
"time": "822.554",
"value": "Acta Neuropathologica_sentence_0"
},
{
"operation": "Highlighting",
"time": "842.124",
"value": "Acta Neuropathologica_table_caption_0"
},
{
"operation": "Highlighting",
"time": "843.267",
"value": "Acta Neuropathologica_cell_0_0_1"
},
{
"operation": "Highlighting",
"time": "843.292",
"value": "Acta Neuropathologica_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "851.612",
"value": "Acta Neuropathologica_cell_0_0_1"
},
{
"operation": "Highlighting deleted",
"time": "851.638",
"value": "Acta Neuropathologica_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "857.015",
"value": "Acta Neuropathologica_cell_0_0_1"
},
{
"operation": "Highlighting",
"time": "857.05",
"value": "Acta Neuropathologica_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "866.358",
"value": "Acta Neuropathologica_table_caption_0"
},
{
"operation": "Highlighting deleted",
"time": "867.667",
"value": "Acta Neuropathologica_cell_0_0_1"
},
{
"operation": "Highlighting deleted",
"time": "867.691",
"value": "Acta Neuropathologica_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "884.927",
"value": "finish"
}
]
}
|
exfever_8048
|
Virginia is a state in the Mid-Atlantic region, between the system of mountains in the eastern to a northeastern continent in the Northern Hemisphere and almost entirely within the Western Hemisphere and Atlantic Coast.
|
{
"evidence": [
"Virginia",
"Appalachian Mountains",
"North America"
],
"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:\nVirginia is a state in the Mid-Atlantic region, between the system of mountains in the eastern to a northeastern continent in the Northern Hemisphere and almost entirely within the Western Hemisphere and Atlantic Coast.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Virginia",
"Appalachian Mountains",
"North America"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5287,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71045
|
Carroll County Broadcasting Company retained ownership of WTTR AM and FM, until the station was sold to Shamrock Broadcasting in the early 1980s; then Dwight Dingle, known for decades as the radio "Voice of Baltimore County", managed WTTR from 1974 until his death.
|
{
"evidence": [
"WTTR_sentence_14",
"WTTR_sentence_15"
],
"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:\nCarroll County Broadcasting Company retained ownership of WTTR AM and FM, until the station was sold to Shamrock Broadcasting in the early 1980s; then Dwight Dingle, known for decades as the radio \"Voice of Baltimore County\", managed WTTR from 1974 until his death.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"WTTR_sentence_14",
"WTTR_sentence_15"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5288,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.962",
"value": "?search="
},
{
"operation": "search",
"time": "17.874",
"value": "WTTR AM and FM,"
},
{
"operation": "Now on",
"time": "19.031",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "21.802",
"value": "WTTR"
},
{
"operation": "Now on",
"time": "22.364",
"value": "WTTR"
},
{
"operation": "Highlighting",
"time": "59.471",
"value": "WTTR_sentence_14"
},
{
"operation": "Highlighting",
"time": "72.607",
"value": "WTTR_sentence_15"
},
{
"operation": "search",
"time": "94.472",
"value": "Dwight Dingle"
},
{
"operation": "Now on",
"time": "97.513",
"value": "undefined"
},
{
"operation": "search",
"time": "124.766",
"value": "Voice of Baltimore"
},
{
"operation": "Now on",
"time": "129.105",
"value": "undefined"
},
{
"operation": "search",
"time": "156.028",
"value": "Baltimore COunty"
},
{
"operation": "Now on",
"time": "157.764",
"value": "Baltimore County, Maryland"
},
{
"operation": "Page search",
"time": "165.769",
"value": "Dwight"
},
{
"operation": "search",
"time": "222.053",
"value": "Dwight Dingle"
},
{
"operation": "Now on",
"time": "223.079",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "225.646",
"value": "WTTR"
},
{
"operation": "Now on",
"time": "226.081",
"value": "WTTR"
},
{
"operation": "finish",
"time": "265.078",
"value": "finish"
}
]
}
|
31632
|
David Bowie Is had a 75,000 private collection that Geoffrey Marsh was allowed to access for free.
|
{
"evidence": [
"David Bowie Is_sentence_7",
"David Bowie Is_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:\nDavid Bowie Is had a 75,000 private collection that Geoffrey Marsh was allowed to access for free.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"David Bowie Is_sentence_7",
"David Bowie Is_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5289,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.638",
"value": "?search="
},
{
"operation": "search",
"time": "18.604",
"value": "David Bowie"
},
{
"operation": "Now on",
"time": "19.723",
"value": "David Bowie"
},
{
"operation": "search",
"time": "57.066",
"value": "Geoffrey Marsh"
},
{
"operation": "Now on",
"time": "57.93",
"value": "Geoffrey Marsh"
},
{
"operation": "hyperlink",
"time": "74.51",
"value": "Geoffrey Marsh (museum director)"
},
{
"operation": "Now on",
"time": "75.014",
"value": "Geoffrey Marsh (museum director)"
},
{
"operation": "hyperlink",
"time": "100.53",
"value": "David Bowie Is"
},
{
"operation": "Now on",
"time": "101.083",
"value": "David Bowie Is"
},
{
"operation": "Highlighting",
"time": "128.314",
"value": "David Bowie Is_sentence_7"
},
{
"operation": "Highlighting",
"time": "133.174",
"value": "David Bowie Is_sentence_8"
},
{
"operation": "finish",
"time": "148.62",
"value": "finish"
}
]
}
|
15238
|
In Avalanche Sharks, which was starred by American fashion model, actress, and singer Kate Nauta, the group tries to convince Dale, Lars and the town mayor about the sharks while Adam was fired and forced to leave the station.
|
{
"evidence": [
"Avalanche Sharks_sentence_37",
"Kate Nauta_sentence_0",
"Avalanche Sharks_sentence_17",
"Avalanche Sharks_sentence_18"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn Avalanche Sharks, which was starred by American fashion model, actress, and singer Kate Nauta, the group tries to convince Dale, Lars and the town mayor about the sharks while Adam was fired and forced to leave the station.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Avalanche Sharks_sentence_37",
"Kate Nauta_sentence_0",
"Avalanche Sharks_sentence_17",
"Avalanche Sharks_sentence_18"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5290,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.584",
"value": "?search="
},
{
"operation": "search",
"time": "9.369",
"value": "Avalanche Sharks"
},
{
"operation": "Now on",
"time": "10.315",
"value": "Avalanche Sharks"
},
{
"operation": "Highlighting",
"time": "67.102",
"value": "Avalanche Sharks_sentence_37"
},
{
"operation": "hyperlink",
"time": "69.794",
"value": "Kate Nauta"
},
{
"operation": "Now on",
"time": "70.364",
"value": "Kate Nauta"
},
{
"operation": "Highlighting",
"time": "75.175",
"value": "Kate Nauta_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "76.871",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "76.969",
"value": "Avalanche Sharks"
},
{
"operation": "Highlighting",
"time": "109.004",
"value": "Avalanche Sharks_sentence_17"
},
{
"operation": "Highlighting",
"time": "112.113",
"value": "Avalanche Sharks_sentence_18"
},
{
"operation": "finish",
"time": "122.504",
"value": "finish"
}
]
}
|
52544
|
Born in Antwerp, artist Tobias Verhaecht (whose father Cornelis van Haecht was also an artist) traveled to Florence, Italy where he first worked.
|
{
"evidence": [
"Tobias Verhaecht_sentence_3",
"Tobias Verhaecht_sentence_4",
"Tobias Verhaecht_sentence_6"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBorn in Antwerp, artist Tobias Verhaecht (whose father Cornelis van Haecht was also an artist) traveled to Florence, Italy where he first worked.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Tobias Verhaecht_sentence_3",
"Tobias Verhaecht_sentence_4",
"Tobias Verhaecht_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5291,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.569",
"value": "undefined"
},
{
"operation": "search",
"time": "12.627",
"value": "Tobias Verhaecht"
},
{
"operation": "Now on",
"time": "13.55",
"value": "Tobias Verhaecht"
},
{
"operation": "Highlighting",
"time": "20.003",
"value": "Tobias Verhaecht_sentence_3"
},
{
"operation": "Highlighting",
"time": "23.345",
"value": "Tobias Verhaecht_sentence_4"
},
{
"operation": "Highlighting",
"time": "37.316",
"value": "Tobias Verhaecht_sentence_6"
},
{
"operation": "finish",
"time": "83.447",
"value": "finish"
}
]
}
|
exfever_5531
|
Mr. Nobody is a 2019 science fiction starring a German and American actress known for her roles in the film as Helen in the epic war film Troy.
|
{
"evidence": [
"Mr. Nobody (film)",
"Diane Kruger"
],
"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:\nMr. Nobody is a 2019 science fiction starring a German and American actress known for her roles in the film as Helen in the epic war film Troy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mr. Nobody (film)",
"Diane Kruger"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5292,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
8617
|
The I See You Tour, the English indie band which began within Sweden, sold out in The Forum and in the Bill Graham Civic Auditorium.
|
{
"evidence": [
"I See You Tour_sentence_0",
"The xx_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:\nThe I See You Tour, the English indie band which began within Sweden, sold out in The Forum and in the Bill Graham Civic Auditorium.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"I See You Tour_sentence_0",
"The xx_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5293,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.559",
"value": "?search="
},
{
"operation": "search",
"time": "9.697",
"value": "I See You Tour"
},
{
"operation": "Now on",
"time": "10.4",
"value": "I See You Tour"
},
{
"operation": "Highlighting",
"time": "54.024",
"value": "I See You Tour_sentence_0"
},
{
"operation": "hyperlink",
"time": "99.406",
"value": "The xx"
},
{
"operation": "Now on",
"time": "100.283",
"value": "The xx"
},
{
"operation": "Highlighting",
"time": "103.414",
"value": "The xx_sentence_2"
},
{
"operation": "finish",
"time": "113.069",
"value": "finish"
}
]
}
|
exfever_8561
|
Lisa Valerie Kudrow portrayed Phoebe Buffay in an American television sitcom created by David Crane and an American television writer and producer.
|
{
"evidence": [
"Lisa Kudrow",
"Friends",
"Marta Kauffman"
],
"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:\nLisa Valerie Kudrow portrayed Phoebe Buffay in an American television sitcom created by David Crane and an American television writer and producer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lisa Kudrow",
"Friends",
"Marta Kauffman"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5294,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_46
|
James VI and I was King of England and Ireland as James I from the accession that followed the death of James's cousin, the Queen of England and Ireland, from 17 November 1558 until 1604.
|
{
"evidence": [
"James VI and I",
"Union of the Crowns",
"Elizabeth I of England"
],
"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:\nJames VI and I was King of England and Ireland as James I from the accession that followed the death of James's cousin, the Queen of England and Ireland, from 17 November 1558 until 1604.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"James VI and I",
"Union of the Crowns",
"Elizabeth I of England"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5295,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
49488
|
Unité Spéciale de la Police is the police tactical unit of the Grand Ducal Police that was created in Luxembourg in 1999.
|
{
"evidence": [
"Unité Spéciale de la Police_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:\nUnité Spéciale de la Police is the police tactical unit of the Grand Ducal Police that was created in Luxembourg in 1999.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Unité Spéciale de la Police_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5296,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.041",
"value": "?search="
},
{
"operation": "search",
"time": "9.943",
"value": "Unité Spéciale de la Police"
},
{
"operation": "Now on",
"time": "10.906",
"value": "Unité Spéciale de la Police"
},
{
"operation": "Highlighting",
"time": "14.357",
"value": "Unité Spéciale de la Police_sentence_0"
},
{
"operation": "finish",
"time": "33.584",
"value": "finish"
}
]
}
|
81497
|
Stuart Fullerton became a museum curator after leaving the Air Force and was the curator of education at Disney World.
|
{
"evidence": [
"Stuart Fullerton_sentence_6",
"Stuart Fullerton_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:\nStuart Fullerton became a museum curator after leaving the Air Force and was the curator of education at Disney World.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Stuart Fullerton_sentence_6",
"Stuart Fullerton_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5297,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.999",
"value": "?search="
},
{
"operation": "search",
"time": "18.841",
"value": "Stuart Fullerton"
},
{
"operation": "Now on",
"time": "19.528",
"value": "Stuart Fullerton"
},
{
"operation": "Highlighting",
"time": "31.208",
"value": "Stuart Fullerton_sentence_6"
},
{
"operation": "Highlighting",
"time": "52.223",
"value": "Stuart Fullerton_sentence_7"
},
{
"operation": "Highlighting",
"time": "67.471",
"value": "Stuart Fullerton_sentence_11"
},
{
"operation": "Highlighting deleted",
"time": "96.56",
"value": "Stuart Fullerton_sentence_11"
},
{
"operation": "finish",
"time": "193.95",
"value": "finish"
}
]
}
|
3969
|
33rd Alabama Infantry Regiment originated in Milk, Alabama with 15 officers and 200 soldiers.
|
{
"evidence": [
"33rd Alabama Infantry Regiment_sentence_11"
],
"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:\n33rd Alabama Infantry Regiment originated in Milk, Alabama with 15 officers and 200 soldiers.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"33rd Alabama Infantry Regiment_sentence_11"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5298,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "16.039",
"value": "33rd Alabama Infantry Regiment"
},
{
"operation": "Page search",
"time": "44.6",
"value": "Milk, Alabama"
},
{
"operation": "Highlighting",
"time": "82.13",
"value": "33rd Alabama Infantry Regiment_sentence_11"
},
{
"operation": "Page search",
"time": "125.88",
"value": "Milk"
},
{
"operation": "finish",
"time": "175.561",
"value": "finish"
}
]
}
|
31506
|
Jenga World Tour was universally panned by critics for its physics system and expensive price, with IGN, an American video game and entertainment media website operated by IGN Entertainment Inc., giving the game a 1.7 out of 10.
|
{
"evidence": [
"Jenga World Tour_sentence_7",
"IGN_sentence_1",
"Jenga World Tour_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:\nJenga World Tour was universally panned by critics for its physics system and expensive price, with IGN, an American video game and entertainment media website operated by IGN Entertainment Inc., giving the game a 1.7 out of 10.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Jenga World Tour_sentence_7",
"IGN_sentence_1",
"Jenga World Tour_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5299,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.892",
"value": "?search="
},
{
"operation": "search",
"time": "11.302",
"value": "Jenga World Tour"
},
{
"operation": "Now on",
"time": "12.38",
"value": "Jenga World Tour"
},
{
"operation": "Highlighting",
"time": "36.486",
"value": "Jenga World Tour_sentence_7"
},
{
"operation": "hyperlink",
"time": "44.052",
"value": "IGN"
},
{
"operation": "Now on",
"time": "44.851",
"value": "IGN"
},
{
"operation": "Highlighting",
"time": "56.036",
"value": "IGN_sentence_1"
},
{
"operation": "back-button-clicked",
"time": "58.82",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "59.074",
"value": "Jenga World Tour"
},
{
"operation": "Highlighting",
"time": "61.005",
"value": "Jenga World Tour_sentence_8"
},
{
"operation": "finish",
"time": "92.348",
"value": "finish"
}
]
}
|
22355
|
Opryland USA was an amusement park owned by Gaylord Entertainment Company and located in Memphis, Tennessee,
|
{
"evidence": [
"Opryland USA_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:\nOpryland USA was an amusement park owned by Gaylord Entertainment Company and located in Memphis, Tennessee,\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Opryland USA_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5300,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.562",
"value": "?search="
},
{
"operation": "search",
"time": "14.687",
"value": "Opryland USA"
},
{
"operation": "Now on",
"time": "15.949",
"value": "Opryland USA"
},
{
"operation": "Highlighting",
"time": "34.139",
"value": "Opryland USA_sentence_0"
},
{
"operation": "finish",
"time": "47.372",
"value": "finish"
}
]
}
|
1467
|
UK Singles Chart records and statistics were first compiled in 1969, and Frankie Laine had the first single to spend the most weeks at number one.
|
{
"evidence": [
"UK Singles Chart records and statistics_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:\nUK Singles Chart records and statistics were first compiled in 1969, and Frankie Laine had the first single to spend the most weeks at number one.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"UK Singles Chart records and statistics_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5301,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "990.423",
"value": "Frankie Laine"
},
{
"operation": "search",
"time": "1229.612",
"value": "UK Singles Chart records"
},
{
"operation": "Page search",
"time": "1246.98",
"value": "frank laine"
},
{
"operation": "Highlighting",
"time": "1293.885",
"value": "UK Singles Chart records and statistics_sentence_0"
},
{
"operation": "Highlighting",
"time": "1393.648",
"value": "UK Singles Chart records and statistics_sentence_6"
},
{
"operation": "Highlighting",
"time": "1393.668",
"value": "UK Singles Chart records and statistics_section_0"
},
{
"operation": "Highlighting",
"time": "1393.68",
"value": "UK Singles Chart records and statistics_section_3"
},
{
"operation": "Highlighting",
"time": "1414.351",
"value": "UK Singles Chart records and statistics_sentence_1"
},
{
"operation": "hyperlink",
"time": "1416.636",
"value": "Official Charts Company"
},
{
"operation": "back-button-clicked",
"time": "1418.236",
"value": "back-button-clicked"
},
{
"operation": "Highlighting",
"time": "1425.14",
"value": "UK Singles Chart records and statistics_sentence_1"
},
{
"operation": "finish",
"time": "1456.829",
"value": "finish"
}
]
}
|
exfever_6517
|
Jarhead is an American biographical war drama film directed by a Spanish film and stage director who also directed an original West End stage musical for the first time with Charlie and the Chocolate Factory.
|
{
"evidence": [
"Jarhead (film)",
"Sam Mendes"
],
"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:\nJarhead is an American biographical war drama film directed by a Spanish film and stage director who also directed an original West End stage musical for the first time with Charlie and the Chocolate Factory.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jarhead (film)",
"Sam Mendes"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5302,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
85868
|
Vasily Dokuchaev known as the father of soil science also helped develop soil science in America.
|
{
"evidence": [
"Vasily Dokuchaev_sentence_1",
"Vasily Dokuchaev_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVasily Dokuchaev known as the father of soil science also helped develop soil science in America.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vasily Dokuchaev_sentence_1",
"Vasily Dokuchaev_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5303,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.715",
"value": "?search="
},
{
"operation": "search",
"time": "116.389",
"value": "Vasily Dokuchaev"
},
{
"operation": "Now on",
"time": "116.901",
"value": "Vasily Dokuchaev"
},
{
"operation": "Highlighting",
"time": "125.841",
"value": "Vasily Dokuchaev_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "131.401",
"value": "Vasily Dokuchaev_sentence_0"
},
{
"operation": "Highlighting",
"time": "132.823",
"value": "Vasily Dokuchaev_sentence_1"
},
{
"operation": "Highlighting",
"time": "135.152",
"value": "Vasily Dokuchaev_sentence_2"
},
{
"operation": "finish",
"time": "252.341",
"value": "finish"
}
]
}
|
exfever_1219
|
Chris Evans is an American actor who starred in a 2007 superhero film that was made by digitally generating animated images.
|
{
"evidence": [
"Chris Evans (actor)",
"TMNT (film)",
"Computer animation"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nChris Evans is an American actor who starred in a 2007 superhero film that was made by digitally generating animated images.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Chris Evans (actor)",
"TMNT (film)",
"Computer animation"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5304,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1478
|
Dracula is a novel by a Belgian author.
|
{
"evidence": [
"Dracula",
"Bram Stoker"
],
"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:\nDracula is a novel by a Belgian author.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dracula",
"Bram Stoker"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5305,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7055
|
Petyr Baelish is a fictional character in the A Song of Ice and Fire series of fantasy novels by an American novelist, screenwriter, television producer and short story writer.
|
{
"evidence": [
"Petyr Baelish",
"George R. R. Martin"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPetyr Baelish is a fictional character in the A Song of Ice and Fire series of fantasy novels by an American novelist, screenwriter, television producer and short story writer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Petyr Baelish",
"George R. R. Martin"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5306,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
44004
|
Fielding Graduate University is a private university in Santa Barbara, California; which offers postgraduate and doctoral studies primarily through in-person classes.
|
{
"evidence": [
"Fielding Graduate University_sentence_0",
"Fielding Graduate University_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:\nFielding Graduate University is a private university in Santa Barbara, California; which offers postgraduate and doctoral studies primarily through in-person classes.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Fielding Graduate University_sentence_0",
"Fielding Graduate University_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5307,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.328",
"value": "?search="
},
{
"operation": "search",
"time": "37.843",
"value": "Fielding Graduate University"
},
{
"operation": "Now on",
"time": "39.054",
"value": "Fielding Graduate University"
},
{
"operation": "Highlighting",
"time": "40.478",
"value": "Fielding Graduate University_sentence_0"
},
{
"operation": "Highlighting",
"time": "44.031",
"value": "Fielding Graduate University_sentence_1"
},
{
"operation": "finish",
"time": "78.332",
"value": "finish"
}
]
}
|
exfever_8307
|
Curly Top is the first film directed by Irving Cummings.
|
{
"evidence": [
"Curly Top (film)",
"Heidi (1937 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:\nCurly Top is the first film directed by Irving Cummings.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Curly Top (film)",
"Heidi (1937 film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5308,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2970
|
Jordan Nathaniel Marcel Knight is an American singer-songwriter who rose to fame in the 1980s and 1990s, using a falsetto style of singing influenced by an American Philadelphia soul group that achieved its greatest chart success in the 1970s.
|
{
"evidence": [
"Jordan Knight",
"The Stylistics"
],
"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:\nJordan Nathaniel Marcel Knight is an American singer-songwriter who rose to fame in the 1980s and 1990s, using a falsetto style of singing influenced by an American Philadelphia soul group that achieved its greatest chart success in the 1970s.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jordan Knight",
"The Stylistics"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5309,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6721
|
Monster Hunter is Capcom's highest selling game.
|
{
"evidence": [
"Super Street Fighter II Turbo",
"Capcom",
"Mega Man"
],
"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:\nMonster Hunter is Capcom's highest selling game.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Super Street Fighter II Turbo",
"Capcom",
"Mega Man"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5310,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3919
|
Split is James McAvoy's second film of the year 2016.
|
{
"evidence": [
"Split (2016 American film)",
"Psychological thriller"
],
"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:\nSplit is James McAvoy's second film of the year 2016.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Split (2016 American film)",
"Psychological thriller"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5311,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
64607
|
The Gold nanoparticles in chemotherapy range in size depending on which therapy they are being used for; in photothermal cancer therapy, many silver nanoparticle molecules are used in each test and they must all be uniform in size.
|
{
"evidence": [
"Gold nanoparticles in chemotherapy_sentence_6"
],
"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 Gold nanoparticles in chemotherapy range in size depending on which therapy they are being used for; in photothermal cancer therapy, many silver nanoparticle molecules are used in each test and they must all be uniform in size.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Gold nanoparticles in chemotherapy_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5312,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.647",
"value": "?search="
},
{
"operation": "search",
"time": "19.927",
"value": "Gold nanoparticles in chemotherapy"
},
{
"operation": "Now on",
"time": "30.035",
"value": "Gold nanoparticles in chemotherapy"
},
{
"operation": "Highlighting",
"time": "55.165",
"value": "Gold nanoparticles in chemotherapy_sentence_6"
},
{
"operation": "finish",
"time": "70.044",
"value": "finish"
}
]
}
|
8203
|
In 1961, Orson Welles was not able to attend a retrospective of his films organized by Peter Bogdanovich for the Museum of Modern Art.
|
{
"evidence": [
"This is Orson Welles_sentence_7",
"This is Orson Welles_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:\nIn 1961, Orson Welles was not able to attend a retrospective of his films organized by Peter Bogdanovich for the Museum of Modern Art.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"This is Orson Welles_sentence_7",
"This is Orson Welles_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5313,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.092",
"value": "?search="
},
{
"operation": "search",
"time": "57.823",
"value": "Orson Welles"
},
{
"operation": "Now on",
"time": "60.203",
"value": "Orson Welles"
},
{
"operation": "Page search",
"time": "67.779",
"value": "peter bogdanovich"
},
{
"operation": "Page search",
"time": "177.608",
"value": "museum of mod"
},
{
"operation": "search",
"time": "264.944",
"value": "this is orson welles"
},
{
"operation": "Now on",
"time": "266.4",
"value": "This Is Orson Welles"
},
{
"operation": "Page search",
"time": "278.247",
"value": "attend"
},
{
"operation": "Highlighting",
"time": "282.545",
"value": "This is Orson Welles_sentence_7"
},
{
"operation": "Highlighting",
"time": "283.136",
"value": "This is Orson Welles_sentence_8"
},
{
"operation": "finish",
"time": "291.258",
"value": "finish"
}
]
}
|
exfever_1663
|
Jean Grey is a fictional character created by an American comic book artist widely regarded as one of the medium's major innovators.
|
{
"evidence": [
"Jean Grey",
"Jack Kirby"
],
"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:\nJean Grey is a fictional character created by an American comic book artist widely regarded as one of the medium's major innovators.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jean Grey",
"Jack Kirby"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5314,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_882
|
At Middleton is a 2013 film that had a world premiere at one of the top film festivals in North America.
|
{
"evidence": [
"At Middleton",
"Seattle International Film Festival"
],
"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:\nAt Middleton is a 2013 film that had a world premiere at one of the top film festivals in North America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"At Middleton",
"Seattle International Film Festival"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5315,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_374
|
The Foo Fighters were the most popular American rock band in 2000.
|
{
"evidence": [
"Foo Fighters",
"Grammy Award",
"Phonograph"
],
"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 Foo Fighters were the most popular American rock band in 2000.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Foo Fighters",
"Grammy Award",
"Phonograph"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5316,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4264
|
Laura Linney is an American actress who received nominations for a 1953 play by American playwright Mel Gibson.
|
{
"evidence": [
"Laura Linney",
"The Crucible"
],
"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:\nLaura Linney is an American actress who received nominations for a 1953 play by American playwright Mel Gibson.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Laura Linney",
"The Crucible"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5317,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
5449
|
Zhanhe District is a district in Pingdingshan, Taiwan Province, People's Republic of China.
|
{
"evidence": [
"Zhanhe District_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:\nZhanhe District is a district in Pingdingshan, Taiwan Province, People's Republic of China.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Zhanhe District_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5318,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.08",
"value": "?search="
},
{
"operation": "search",
"time": "15.574",
"value": "Zhanhe District"
},
{
"operation": "Now on",
"time": "16.465",
"value": "Zhanhe District"
},
{
"operation": "Highlighting",
"time": "19.649",
"value": "Zhanhe District_sentence_0"
},
{
"operation": "Highlighting",
"time": "23.929",
"value": "Zhanhe District_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "23.955",
"value": "Zhanhe District_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "23.963",
"value": "Zhanhe District_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "42.981",
"value": "Pingdingshan"
},
{
"operation": "Now on",
"time": "43.861",
"value": "Pingdingshan"
},
{
"operation": "back-button-clicked",
"time": "52.916",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "53.13",
"value": "Zhanhe District"
},
{
"operation": "Highlighting deleted",
"time": "74.889",
"value": "Zhanhe District_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "74.92",
"value": "Zhanhe District_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "74.93",
"value": "Zhanhe District_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "77.227",
"value": "finish"
}
]
}
|
exfever_4939
|
Spider-Man 3 is the final installment in Raimi's Spider-Man trilogy and the sequel to Spider-Man and the 2005 American superhero film directed by Sam Raimi and written by Alvin Sargent.
|
{
"evidence": [
"Spider-Man 3",
"Spider-Man 2"
],
"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:\nSpider-Man 3 is the final installment in Raimi's Spider-Man trilogy and the sequel to Spider-Man and the 2005 American superhero film directed by Sam Raimi and written by Alvin Sargent.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Spider-Man 3",
"Spider-Man 2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5319,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
24889
|
Dan Severson was a fighter pilot who succeeded Tim O'Driscoll (previous mayor of St. Joseph, Minnesota) as a member of the Minnesota House of Representatives from the 14A district.
|
{
"evidence": [
"Tim O'Driscoll_sentence_8"
],
"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:\nDan Severson was a fighter pilot who succeeded Tim O'Driscoll (previous mayor of St. Joseph, Minnesota) as a member of the Minnesota House of Representatives from the 14A district.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Tim O'Driscoll_sentence_8"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5320,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.81",
"value": "?search="
},
{
"operation": "search",
"time": "10.752",
"value": "Dan Severson"
},
{
"operation": "Now on",
"time": "11.749",
"value": "Dan Severson"
},
{
"operation": "Highlighting",
"time": "73.559",
"value": "Dan Severson_sentence_0"
},
{
"operation": "Highlighting",
"time": "73.561",
"value": "Dan Severson_title"
},
{
"operation": "Highlighting",
"time": "80.979",
"value": "Dan Severson_sentence_2Dan Severson_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "95.234",
"value": "Dan Severson_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "96.131",
"value": "Dan Severson_sentence_2Dan Severson_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "96.134",
"value": "Dan Severson_title"
},
{
"operation": "Page search",
"time": "105.715",
"value": "fighter pilot"
},
{
"operation": "Highlighting",
"time": "108.982",
"value": "Dan Severson_sentence_8"
},
{
"operation": "Highlighting",
"time": "108.984",
"value": "Dan Severson_title"
},
{
"operation": "Highlighting",
"time": "108.985",
"value": "Dan Severson_section_0"
},
{
"operation": "Page search",
"time": "120.972",
"value": "Tim O'Driscoll"
},
{
"operation": "Highlighting",
"time": "124.452",
"value": "Dan Severson_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "124.479",
"value": "Dan Severson_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "124.48",
"value": "Dan Severson_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "124.481",
"value": "Dan Severson_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "128.235",
"value": "Tim O'Driscoll"
},
{
"operation": "Now on",
"time": "129.393",
"value": "Tim O'Driscoll"
},
{
"operation": "Highlighting",
"time": "146.388",
"value": "Tim O'Driscoll_sentence_8"
},
{
"operation": "Page search",
"time": "178.658",
"value": "mayor"
},
{
"operation": "finish",
"time": "193.688",
"value": "finish"
}
]
}
|
exfever_822
|
Winona Ryder is an American actress who rose to prominence with major roles in a 1990 American fantasy film directed by Bruce Lee.
|
{
"evidence": [
"Winona Ryder",
"Edward Scissorhands"
],
"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:\nWinona Ryder is an American actress who rose to prominence with major roles in a 1990 American fantasy film directed by Bruce Lee.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Winona Ryder",
"Edward Scissorhands"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5321,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
26119
|
A demo mode is usually programmed into the floor model's firmware (a specific class of computer software that provides the low-level control for a device's specific hardware), and is usually activated by retail stores on their consumer electronics floor models.
|
{
"evidence": [
"Demo mode_sentence_1",
"Firmware_sentence_0",
"Demo mode_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:\nA demo mode is usually programmed into the floor model's firmware (a specific class of computer software that provides the low-level control for a device's specific hardware), and is usually activated by retail stores on their consumer electronics floor models.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Demo mode_sentence_1",
"Firmware_sentence_0",
"Demo mode_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5322,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.531",
"value": "?search="
},
{
"operation": "search",
"time": "10.049",
"value": "demo mode"
},
{
"operation": "Now on",
"time": "10.966",
"value": "Demo mode"
},
{
"operation": "Highlighting",
"time": "29.74",
"value": "Demo mode_sentence_1"
},
{
"operation": "hyperlink",
"time": "33.766",
"value": "firmware"
},
{
"operation": "Now on",
"time": "34.452",
"value": "Firmware"
},
{
"operation": "Highlighting",
"time": "36.827",
"value": "Firmware_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "43.083",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "43.129",
"value": "Demo mode"
},
{
"operation": "Highlighting",
"time": "45.782",
"value": "Demo mode_sentence_2"
},
{
"operation": "finish",
"time": "54.413",
"value": "finish"
}
]
}
|
59625
|
Twomey, first elected to the Newfoundland assembly in 1975, served stints as Minister of Health from 1984 to 1988, and as Minister of Public Works and Services from 1988 to 1989.
|
{
"evidence": [
"Hugh Twomey_sentence_8",
"Hugh Twomey_sentence_9"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTwomey, first elected to the Newfoundland assembly in 1975, served stints as Minister of Health from 1984 to 1988, and as Minister of Public Works and Services from 1988 to 1989.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Hugh Twomey_sentence_8",
"Hugh Twomey_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5323,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.251",
"value": "?search="
},
{
"operation": "search",
"time": "66.258",
"value": "Twomey"
},
{
"operation": "Now on",
"time": "67.183",
"value": "Twomey"
},
{
"operation": "back-button-clicked",
"time": "85.404",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "85.61",
"value": "?search="
},
{
"operation": "search",
"time": "91.897",
"value": "Twomey"
},
{
"operation": "Now on",
"time": "92.89",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "110.5",
"value": "Twomey"
},
{
"operation": "Now on",
"time": "111.05",
"value": "Twomey"
},
{
"operation": "search",
"time": "188.715",
"value": "Newfoundland assembly 1975"
},
{
"operation": "Now on",
"time": "190.657",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "205.03",
"value": "Hugh Twomey"
},
{
"operation": "Now on",
"time": "206.002",
"value": "Hugh Twomey"
},
{
"operation": "Page search",
"time": "253.344",
"value": "assembl"
},
{
"operation": "Page search",
"time": "253.576",
"value": "assembl"
},
{
"operation": "Highlighting",
"time": "256.899",
"value": "Hugh Twomey_sentence_1"
},
{
"operation": "Highlighting",
"time": "267.854",
"value": "Hugh Twomey_sentence_8"
},
{
"operation": "Highlighting deleted",
"time": "269.161",
"value": "Hugh Twomey_sentence_1"
},
{
"operation": "Highlighting",
"time": "274.575",
"value": "Hugh Twomey_sentence_9"
},
{
"operation": "finish",
"time": "298.964",
"value": "finish"
}
]
}
|
exfever_7108
|
Emma Watson is an American actress who starred in a 2010 American comedy-drama film directed by Anna Boden.
|
{
"evidence": [
"Emma Roberts",
"It's Kind of a Funny Story (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:\nEmma Watson is an American actress who starred in a 2010 American comedy-drama film directed by Anna Boden.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Emma Roberts",
"It's Kind of a Funny Story (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5324,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1589
|
Color of Night is an Indian film classified as a film subgenre defined as a thriller with a thematic basis in illicit romance or erotic fantasy.
|
{
"evidence": [
"Color of Night",
"Erotic thriller"
],
"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:\nColor of Night is an Indian film classified as a film subgenre defined as a thriller with a thematic basis in illicit romance or erotic fantasy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Color of Night",
"Erotic thriller"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5325,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4992
|
Liverpool is a professional soccer?club that plays with no one allowed to touch the ball except the major positions in association baseball.
|
{
"evidence": [
"Liverpool F.C.",
"Association football",
"Goalkeeper (association football)"
],
"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:\nLiverpool is a professional soccer?club that plays with no one allowed to touch the ball except the major positions in association baseball.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Liverpool F.C.",
"Association football",
"Goalkeeper (association football)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5326,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
939
|
Athara Hazari is a city in Tehsil of Jhang District, Punjab, Pakistan near the Trimmu Barrage with Muhammad Mohsin Alam as its incumbent Assistant Commissioner.
|
{
"evidence": [
"Athara Hazari_sentence_0",
"Athara Hazari_sentence_1",
"Athara Hazari_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:\nAthara Hazari is a city in Tehsil of Jhang District, Punjab, Pakistan near the Trimmu Barrage with Muhammad Mohsin Alam as its incumbent Assistant Commissioner.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Athara Hazari_sentence_0",
"Athara Hazari_sentence_1",
"Athara Hazari_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5327,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "28.623",
"value": "Athara Hazari Tehsil"
},
{
"operation": "search",
"time": "44.865",
"value": "Athara Hazari"
},
{
"operation": "Highlighting",
"time": "61.28",
"value": "Athara Hazari_sentence_0"
},
{
"operation": "Highlighting",
"time": "69.852",
"value": "Athara Hazari_sentence_1"
},
{
"operation": "Highlighting",
"time": "85.911",
"value": "Athara Hazari_sentence_4"
},
{
"operation": "finish",
"time": "111.952",
"value": "finish"
}
]
}
|
exfever_466
|
Amy Winehouse's debut studio album, Frank, was released by a multinational music label owned by Universal Music Group.
|
{
"evidence": [
"Frank (Amy Winehouse album)",
"Island Records"
],
"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:\nAmy Winehouse's debut studio album, Frank, was released by a multinational music label owned by Universal Music Group.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Frank (Amy Winehouse album)",
"Island Records"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5328,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
67433
|
Gibson Bay in Canada is located in the Foxe Basin, a shallow basin north of the Ungava Bay that spends most of the year blocked by ice.
|
{
"evidence": [
"Gibson Bay_sentence_0",
"Foxe Basin_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:\nGibson Bay in Canada is located in the Foxe Basin, a shallow basin north of the Ungava Bay that spends most of the year blocked by ice.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Gibson Bay_sentence_0",
"Foxe Basin_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5329,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.093",
"value": "?search="
},
{
"operation": "search",
"time": "35.767",
"value": "Gibson Bay"
},
{
"operation": "Now on",
"time": "36.957",
"value": "Gibson Bay"
},
{
"operation": "Highlighting",
"time": "58.209",
"value": "Gibson Bay_sentence_0"
},
{
"operation": "Highlighting",
"time": "163.601",
"value": "Gibson Bay_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "163.604",
"value": "Gibson Bay_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "163.604",
"value": "Gibson Bay_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "167.194",
"value": "Foxe Basin"
},
{
"operation": "Now on",
"time": "168.596",
"value": "Foxe Basin"
},
{
"operation": "Highlighting",
"time": "174.505",
"value": "Foxe Basin_sentence_1"
},
{
"operation": "Page search",
"time": "192.543",
"value": "ungava"
},
{
"operation": "search",
"time": "203.169",
"value": "Ungava Bay"
},
{
"operation": "Now on",
"time": "204.09",
"value": "Ungava Bay"
},
{
"operation": "Page search",
"time": "214.273",
"value": "foxe basin"
},
{
"operation": "finish",
"time": "227.899",
"value": "finish"
}
]
}
|
88614
|
During the second half of the century, history painters combined a local influence from Rubens with knowledge of classicism and Italian Baroque qualities; then later in the century, many painters turned to Anthony van Dyck as a major influence like Pieter Thijs and Lucas Franchoys the Younger.
|
{
"evidence": [
"Flemish Baroque painting_sentence_27",
"Flemish Baroque painting_sentence_29",
"Flemish Baroque painting_sentence_30"
],
"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:\nDuring the second half of the century, history painters combined a local influence from Rubens with knowledge of classicism and Italian Baroque qualities; then later in the century, many painters turned to Anthony van Dyck as a major influence like Pieter Thijs and Lucas Franchoys the Younger.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Flemish Baroque painting_sentence_27",
"Flemish Baroque painting_sentence_29",
"Flemish Baroque painting_sentence_30"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5330,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.195",
"value": "?search="
},
{
"operation": "search",
"time": "73.762",
"value": "Lucas Franchoys the Younger"
},
{
"operation": "Now on",
"time": "75.913",
"value": "Lucas Franchoys the Younger"
},
{
"operation": "search",
"time": "147.619",
"value": "Peter Paul Rubens"
},
{
"operation": "Now on",
"time": "151.085",
"value": "Peter Paul Rubens"
},
{
"operation": "search",
"time": "255.651",
"value": "Baroque period"
},
{
"operation": "Now on",
"time": "262.047",
"value": "Baroque"
},
{
"operation": "search",
"time": "307.41",
"value": "Italian Baroque"
},
{
"operation": "Now on",
"time": "310.095",
"value": "Italian Baroque"
},
{
"operation": "search",
"time": "618.53",
"value": "Flemish Baroque painting"
},
{
"operation": "Now on",
"time": "624.785",
"value": "Flemish Baroque painting"
},
{
"operation": "Highlighting",
"time": "635.087",
"value": "Flemish Baroque painting_sentence_27"
},
{
"operation": "Highlighting",
"time": "639.549",
"value": "Flemish Baroque painting_sentence_29"
},
{
"operation": "Highlighting",
"time": "646.332",
"value": "Flemish Baroque painting_sentence_30"
},
{
"operation": "finish",
"time": "657.707",
"value": "finish"
}
]
}
|
exfever_6182
|
Henry John Deutschendorf Jr. was an American singer-songwriter who began his music career with music groups whose genre included traditional music and the contemporary genre that evolved from the former during the 20th-century folk revival.
|
{
"evidence": [
"John Denver",
"Folk music"
],
"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:\nHenry John Deutschendorf Jr. was an American singer-songwriter who began his music career with music groups whose genre included traditional music and the contemporary genre that evolved from the former during the 20th-century folk revival.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Denver",
"Folk music"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5331,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
45709
|
Zhu Gaoxu, is wrongly thought to be the son of Yongle Emperor, and having reigned from 1395 - 1404 before being promoted to Prince of Han and reigning from 1404 -1426
|
{
"evidence": [
"Zhu Gaoxu_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:\nZhu Gaoxu, is wrongly thought to be the son of Yongle Emperor, and having reigned from 1395 - 1404 before being promoted to Prince of Han and reigning from 1404 -1426\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Zhu Gaoxu_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5332,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.997",
"value": "?search="
},
{
"operation": "search",
"time": "22.181",
"value": "Zhu Gaoxu"
},
{
"operation": "Now on",
"time": "22.875",
"value": "Zhu Gaoxu"
},
{
"operation": "Page search",
"time": "39.584",
"value": "wrongly"
},
{
"operation": "Highlighting",
"time": "48.264",
"value": "Zhu Gaoxu_sentence_0"
},
{
"operation": "finish",
"time": "72.503",
"value": "finish"
}
]
}
|
34099
|
XMPP is closed protocol and is only implemented in commercial software.
|
{
"evidence": [
"XMPP_sentence_6"
],
"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:\nXMPP is closed protocol and is only implemented in commercial software.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"XMPP_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5333,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.76",
"value": "?search="
},
{
"operation": "search",
"time": "6.88",
"value": "XMPP"
},
{
"operation": "Now on",
"time": "7.559",
"value": "XMPP"
},
{
"operation": "Page search",
"time": "15.465",
"value": "open"
},
{
"operation": "Highlighting",
"time": "27.267",
"value": "XMPP_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "28.319",
"value": "XMPP_sentence_5"
},
{
"operation": "Highlighting",
"time": "29.052",
"value": "XMPP_sentence_6"
},
{
"operation": "finish",
"time": "36.265",
"value": "finish"
}
]
}
|
35009
|
KJPG ended its simulcast with KKGO-FM, their call sign was changed to KTRJ and it began airing Catholic Christmas music.
|
{
"evidence": [
"KJPG_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:\nKJPG ended its simulcast with KKGO-FM, their call sign was changed to KTRJ and it began airing Catholic Christmas music.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"KJPG_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5334,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.805",
"value": "?search="
},
{
"operation": "search",
"time": "4.107",
"value": "KJPG"
},
{
"operation": "Now on",
"time": "5.117",
"value": "KJPG"
},
{
"operation": "Highlighting",
"time": "9.536",
"value": "KJPG_sentence_5"
},
{
"operation": "Highlighting",
"time": "15.032",
"value": "KJPG_sentence_6"
},
{
"operation": "Highlighting",
"time": "36.238",
"value": "KJPG_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "40.124",
"value": "KJPG_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "44.309",
"value": "KJPG_sentence_6"
},
{
"operation": "finish",
"time": "48.358",
"value": "finish"
}
]
}
|
37462
|
Richard Reinhardt "Richie Ramone"(born August 11, 1957 in Passaic, New Jersey) is a musician, songwriter and the fastest, most powerful bassist of the punk rock band the Ramones from February 1983 until August 1987..
|
{
"evidence": [
"Richie Ramone_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:\nRichard Reinhardt \"Richie Ramone\"(born August 11, 1957 in Passaic, New Jersey) is a musician, songwriter and the fastest, most powerful bassist of the punk rock band the Ramones from February 1983 until August 1987..\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Richie Ramone_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5335,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.042",
"value": "?search="
},
{
"operation": "search",
"time": "16.45",
"value": "Richard Reinhardt"
},
{
"operation": "Now on",
"time": "17.197",
"value": "Richie Ramone"
},
{
"operation": "Highlighting",
"time": "28.72",
"value": "Richie Ramone_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "28.732",
"value": "Richie Ramone_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "28.732",
"value": "Richie Ramone_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "28.732",
"value": "Richie Ramone_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "92.389",
"value": "Richie Ramone_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "97.443",
"value": "Richie Ramone_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "97.453",
"value": "Richie Ramone_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "97.454",
"value": "Richie Ramone_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "97.456",
"value": "Richie Ramone_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "137.262",
"value": "finish"
}
]
}
|
exfever_3636
|
The 19th G7 summit was held in Tokyo where the heads of the richest industrialized countries, like a country primarily located in North America, were included.
|
{
"evidence": [
"19th G7 summit",
"United States"
],
"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 19th G7 summit was held in Tokyo where the heads of the richest industrialized countries, like a country primarily located in North America, were included.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"19th G7 summit",
"United States"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5336,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
83067
|
In 1979, recommended by physicist Qian Linzhao, Zhongchao went to Cambridge University in the United Kingdom to study theoretical physics; then received a doctorate from Cambridge University in 1984.
|
{
"evidence": [
"Wu Zhongchao_sentence_5",
"Wu Zhongchao_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:\nIn 1979, recommended by physicist Qian Linzhao, Zhongchao went to Cambridge University in the United Kingdom to study theoretical physics; then received a doctorate from Cambridge University in 1984.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Wu Zhongchao_sentence_5",
"Wu Zhongchao_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5337,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.251",
"value": "?search="
},
{
"operation": "search",
"time": "16.648",
"value": "Contains... Zhongchao went to Cambridge University in the United Kingdom to study theoretical physics"
},
{
"operation": "Now on",
"time": "18.236",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "277.345",
"value": "Wu Zhongchao"
},
{
"operation": "Now on",
"time": "278.236",
"value": "Wu Zhongchao"
},
{
"operation": "Highlighting",
"time": "413.068",
"value": "Wu Zhongchao_sentence_5"
},
{
"operation": "Highlighting",
"time": "417.363",
"value": "Wu Zhongchao_sentence_7"
},
{
"operation": "finish",
"time": "424.05",
"value": "finish"
}
]
}
|
exfever_2226
|
Sleeping Beauty is an American erotic drama film at The 64th Cannes Film Festival held from 11 to 22 May 2011.
|
{
"evidence": [
"Sleeping Beauty (2011 film)",
"2011 Cannes Film Festival"
],
"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:\nSleeping Beauty is an American erotic drama film at The 64th Cannes Film Festival held from 11 to 22 May 2011.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sleeping Beauty (2011 film)",
"2011 Cannes Film Festival"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5338,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71541
|
The SWAC commissioner announced that the game had been officially sanctioned, however the bowl was not revived and was the conference's of Division II playoffs.
|
{
"evidence": [
"SWAC Championship Game_sentence_16",
"SWAC Championship Game_sentence_17"
],
"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 SWAC commissioner announced that the game had been officially sanctioned, however the bowl was not revived and was the conference's of Division II playoffs.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"SWAC Championship Game_sentence_16",
"SWAC Championship Game_sentence_17"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5339,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.168",
"value": "?search="
},
{
"operation": "search",
"time": "21.155",
"value": "Contains...SWAC commissioner"
},
{
"operation": "Now on",
"time": "22.517",
"value": "undefined"
},
{
"operation": "search",
"time": "30.502",
"value": "SWAC Championship Game"
},
{
"operation": "Now on",
"time": "32.632",
"value": "SWAC Championship Game"
},
{
"operation": "Highlighting",
"time": "192.591",
"value": "SWAC Championship Game_sentence_16"
},
{
"operation": "Highlighting",
"time": "198.127",
"value": "SWAC Championship Game_sentence_17"
},
{
"operation": "finish",
"time": "216.432",
"value": "finish"
}
]
}
|
exfever_7787
|
Tokyo Ghoul is a 2017 Japanese dark fantasy action film based on the manga series Tokyo Ghoul by a Japanese manga artist best known for his subgenre of fantasy literary, artistic, and cinematic works that incorporate disturbing and frightening themes of fantasy.
|
{
"evidence": [
"Tokyo Ghoul (film)",
"Sui Ishida",
"Dark fantasy"
],
"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:\nTokyo Ghoul is a 2017 Japanese dark fantasy action film based on the manga series Tokyo Ghoul by a Japanese manga artist best known for his subgenre of fantasy literary, artistic, and cinematic works that incorporate disturbing and frightening themes of fantasy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tokyo Ghoul (film)",
"Sui Ishida",
"Dark fantasy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5340,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
15796
|
Lucas Victor Baldin (born on June 5, 1991) never played football.
|
{
"evidence": [
"Lucas Baldin_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:\nLucas Victor Baldin (born on June 5, 1991) never played football.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Lucas Baldin_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5341,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.752",
"value": "?search="
},
{
"operation": "search",
"time": "11.59",
"value": "Contains...Lucas Victor Baldin"
},
{
"operation": "Now on",
"time": "12.445",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "14.475",
"value": "Lucas Baldin"
},
{
"operation": "Now on",
"time": "15.534",
"value": "Lucas Baldin"
},
{
"operation": "Highlighting",
"time": "17.902",
"value": "Lucas Baldin_sentence_0"
},
{
"operation": "hyperlink",
"time": "33.197",
"value": "footballer"
},
{
"operation": "Now on",
"time": "34.328",
"value": "Association football"
},
{
"operation": "back-button-clicked",
"time": "41.346",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "41.457",
"value": "Lucas Baldin"
},
{
"operation": "finish",
"time": "61.59",
"value": "finish"
}
]
}
|
36323
|
Velous, a stage name based on a comment about his marvelous production skills, released his debut mixtape called Velocity The Mixtape in 2011 which only took success after a few years since its release.
|
{
"evidence": [
"Velous_sentence_0",
"Velous_sentence_13",
"Velous_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:\nVelous, a stage name based on a comment about his marvelous production skills, released his debut mixtape called Velocity The Mixtape in 2011 which only took success after a few years since its release.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Velous_sentence_0",
"Velous_sentence_13",
"Velous_sentence_14"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5342,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.026",
"value": "undefined"
},
{
"operation": "search",
"time": "8.977",
"value": "Velous"
},
{
"operation": "Now on",
"time": "10.067",
"value": "Velous"
},
{
"operation": "Highlighting",
"time": "13.31",
"value": "Velous_sentence_0"
},
{
"operation": "Highlighting",
"time": "48.883",
"value": "Velous_sentence_14"
},
{
"operation": "Highlighting deleted",
"time": "128.879",
"value": "Velous_sentence_14"
},
{
"operation": "Highlighting",
"time": "140.71",
"value": "Velous_sentence_13"
},
{
"operation": "Highlighting",
"time": "143.344",
"value": "Velous_sentence_14"
},
{
"operation": "finish",
"time": "147.269",
"value": "finish"
}
]
}
|
89940
|
Kaczerginski, who is responsible for preserving over 250 Holocaust songs, dedicated much of his time after the start of the First World War to collecting pre-war Yiddish songs and songs of the Holocaust in order to save Yiddishkeit from destruction.
|
{
"evidence": [
"Shmerke Kaczerginski_sentence_6"
],
"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:\nKaczerginski, who is responsible for preserving over 250 Holocaust songs, dedicated much of his time after the start of the First World War to collecting pre-war Yiddish songs and songs of the Holocaust in order to save Yiddishkeit from destruction.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Shmerke Kaczerginski_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5343,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.648",
"value": "?search="
},
{
"operation": "search",
"time": "10.468",
"value": "Contains...Kaczerginski"
},
{
"operation": "Now on",
"time": "11.733",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "22.105",
"value": "Shmerke Kaczerginski"
},
{
"operation": "Now on",
"time": "23.267",
"value": "Shmerke Kaczerginski"
},
{
"operation": "Page search",
"time": "31.339",
"value": "holocaust"
},
{
"operation": "Highlighting",
"time": "38.345",
"value": "Shmerke Kaczerginski_sentence_7"
},
{
"operation": "Highlighting",
"time": "43.851",
"value": "Shmerke Kaczerginski_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "47.439",
"value": "Shmerke Kaczerginski_sentence_7"
},
{
"operation": "finish",
"time": "61.668",
"value": "finish"
}
]
}
|
68211
|
Frederick Borden, the Minister of Militia and Defence, announced in May that a 200-strong force would be deployed to Fort Selkirk (a former trading post on the Yukon River); and one of the arguments put forward in favour of this option was that it was much cheaper than sending additional police, who enjoyed higher pay than soldiers.
|
{
"evidence": [
"Yukon Field Force_sentence_20",
"Yukon Field Force_sentence_21",
"Fort Selkirk_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:\nFrederick Borden, the Minister of Militia and Defence, announced in May that a 200-strong force would be deployed to Fort Selkirk (a former trading post on the Yukon River); and one of the arguments put forward in favour of this option was that it was much cheaper than sending additional police, who enjoyed higher pay than soldiers.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Yukon Field Force_sentence_20",
"Yukon Field Force_sentence_21",
"Fort Selkirk_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5344,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.232",
"value": "undefined"
},
{
"operation": "search",
"time": "20.163",
"value": "Contains...200-strong force would be deployed to Fort Selkirk"
},
{
"operation": "Now on",
"time": "20.614",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "24.212",
"value": "Deployment Yukon Field Force_section_2"
},
{
"operation": "Now on",
"time": "24.921",
"value": "Yukon Field Force#Deployment Yukon Field Force section 2"
},
{
"operation": "Highlighting",
"time": "36.093",
"value": "Yukon Field Force_sentence_20"
},
{
"operation": "Highlighting",
"time": "42.967",
"value": "Yukon Field Force_sentence_21"
},
{
"operation": "hyperlink",
"time": "45.65",
"value": "Fort Selkirk"
},
{
"operation": "Now on",
"time": "47.627",
"value": "Fort Selkirk"
},
{
"operation": "Highlighting",
"time": "49.984",
"value": "Fort Selkirk_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "52.581",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "52.721",
"value": "Yukon Field Force#Deployment Yukon Field Force section 2"
},
{
"operation": "finish",
"time": "82.432",
"value": "finish"
}
]
}
|
exfever_6282
|
Halle Berry is one of the highest-paid actress of all time.
|
{
"evidence": [
"Halle Berry",
"Miss World 1986",
"Denmark"
],
"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:\nHalle Berry is one of the highest-paid actress of all time.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Halle Berry",
"Miss World 1986",
"Denmark"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5345,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_127
|
Paul Bettany is best known for portraying Vision and voicing J.A.R.V.I.S. in the Marvel Cinematic Universe film, a 2018 American superhero film produced by Marvel Studios and distributed by Walt Disney Studios Motion Pictures.
|
{
"evidence": [
"Paul Bettany",
"Avengers: Infinity War"
],
"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:\nPaul Bettany is best known for portraying Vision and voicing J.A.R.V.I.S. in the Marvel Cinematic Universe film, a 2018 American superhero film produced by Marvel Studios and distributed by Walt Disney Studios Motion Pictures.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Bettany",
"Avengers: Infinity War"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5346,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
73687
|
In the 1989 British Formula Three Championship, Allan McNish was initially declared as the champion after Brabham was disqualified and was never reinstated even after a court hearing.
|
{
"evidence": [
"1989 British Formula Three Championship_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 1989 British Formula Three Championship, Allan McNish was initially declared as the champion after Brabham was disqualified and was never reinstated even after a court hearing.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1989 British Formula Three Championship_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5347,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "3.651",
"value": "?search="
},
{
"operation": "search",
"time": "13.847",
"value": "1989 British Formula Three Championship"
},
{
"operation": "Now on",
"time": "16.392",
"value": "1989 British Formula Three Championship"
},
{
"operation": "Highlighting",
"time": "34.879",
"value": "1989 British Formula Three Championship_sentence_1"
},
{
"operation": "Highlighting",
"time": "38.198",
"value": "1989 British Formula Three Championship_sentence_2"
},
{
"operation": "Highlighting",
"time": "45.431",
"value": "1989 British Formula Three Championship_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "47.238",
"value": "1989 British Formula Three Championship_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "48.821",
"value": "1989 British Formula Three Championship_sentence_1"
},
{
"operation": "finish",
"time": "55.738",
"value": "finish"
}
]
}
|
52497
|
Vijay Singh, the 2000 champion of the Masters Tournament, headlined the second round with a dominant, seven-under 65 to bolt up to the top of the leaderboard at −9; then Tiger Woods never headlined the five players.
|
{
"evidence": [
"2000 Masters Tournament_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:\nVijay Singh, the 2000 champion of the Masters Tournament, headlined the second round with a dominant, seven-under 65 to bolt up to the top of the leaderboard at −9; then Tiger Woods never headlined the five players.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"2000 Masters Tournament_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5348,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.89",
"value": "?search="
},
{
"operation": "search",
"time": "21.364",
"value": "Vijay Singh (disambiguation)"
},
{
"operation": "Now on",
"time": "22.005",
"value": "Vijay Singh (disambiguation)"
},
{
"operation": "hyperlink",
"time": "32.835",
"value": "Vijay Pal Singh"
},
{
"operation": "Now on",
"time": "33.233",
"value": "Vijay Pal Singh"
},
{
"operation": "search",
"time": "52.14",
"value": "Singh Masters Tournament"
},
{
"operation": "Now on",
"time": "54.486",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "68.34",
"value": "Masters Tournament"
},
{
"operation": "Now on",
"time": "69.979",
"value": "Masters Tournament"
},
{
"operation": "Page search",
"time": "75.214",
"value": "2000"
},
{
"operation": "Page search",
"time": "77.161",
"value": "2000"
},
{
"operation": "search",
"time": "107.091",
"value": "2000 Masters Tournament"
},
{
"operation": "Now on",
"time": "108.63",
"value": "2000 Masters Tournament"
},
{
"operation": "Page search",
"time": "115.167",
"value": "Singh"
},
{
"operation": "Highlighting",
"time": "122.315",
"value": "2000 Masters Tournament_sentence_1"
},
{
"operation": "search",
"time": "193.402",
"value": "2000 Masters Tournament second round"
},
{
"operation": "Now on",
"time": "195.285",
"value": "undefined"
},
{
"operation": "search",
"time": "214.354",
"value": "2000 Masters Tournament"
},
{
"operation": "Now on",
"time": "214.901",
"value": "2000 Masters Tournament"
},
{
"operation": "Page search",
"time": "221.913",
"value": "dominant"
},
{
"operation": "Page search",
"time": "245.312",
"value": "-9"
},
{
"operation": "search",
"time": "329.238",
"value": "Masters Tournament Singh"
},
{
"operation": "Now on",
"time": "329.952",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "337.41",
"value": "PGA Championship"
},
{
"operation": "Now on",
"time": "342.175",
"value": "PGA Championship"
},
{
"operation": "Page search",
"time": "345.572",
"value": "singh"
},
{
"operation": "Highlighting",
"time": "368.262",
"value": "PGA Championship_cell_1_21_1"
},
{
"operation": "Highlighting",
"time": "368.263",
"value": "PGA Championship_title"
},
{
"operation": "Highlighting",
"time": "368.263",
"value": "PGA Championship_section_3"
},
{
"operation": "Highlighting",
"time": "368.265",
"value": "PGA Championship_section_4"
},
{
"operation": "Highlighting",
"time": "368.268",
"value": "PGA Championship_header_cell_1_0_1"
},
{
"operation": "Highlighting",
"time": "370.016",
"value": "PGA Championship_cell_1_21_0"
},
{
"operation": "Highlighting",
"time": "370.022",
"value": "PGA Championship_header_cell_1_0_0"
},
{
"operation": "finish",
"time": "419.361",
"value": "finish"
}
]
}
|
exfever_3509
|
Tropic Thunder is a 2008 film starring an American actor and stand-up comedian best known for his role in the film Roll Bounce in 2005.
|
{
"evidence": [
"Tropic Thunder",
"Brandon T. Jackson"
],
"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:\nTropic Thunder is a 2008 film starring an American actor and stand-up comedian best known for his role in the film Roll Bounce in 2005.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tropic Thunder",
"Brandon T. Jackson"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5349,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1276
|
Futurama is a series that follows the adventures of the professional slacker Philip J. Fry, who is working alongside the one-eyed Leela and robot one of the main characters in the animated television series.
|
{
"evidence": [
"Futurama",
"Bender (Futurama)"
],
"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 a series that follows the adventures of the professional slacker Philip J. Fry, who is working alongside the one-eyed Leela and robot one of the main characters in the animated television series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Futurama",
"Bender (Futurama)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5350,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5112
|
Ari Levine's first production with Bruno Mars was the song Treasure.
|
{
"evidence": [
"Treasure (Bruno Mars song)",
"Unorthodox Jukebox"
],
"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:\nAri Levine's first production with Bruno Mars was the song Treasure.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Treasure (Bruno Mars song)",
"Unorthodox Jukebox"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5351,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
32601
|
Wendell Holland was first assigned to the Naviti tribe when he joined Survivor: Ghost Island.
|
{
"evidence": [
"Wendell Holland_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:\nWendell Holland was first assigned to the Naviti tribe when he joined Survivor: Ghost Island.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Wendell Holland_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5352,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.169",
"value": "undefined"
},
{
"operation": "search",
"time": "13.262",
"value": "Wendell Holland"
},
{
"operation": "Now on",
"time": "15.521",
"value": "Wendell Holland"
},
{
"operation": "hyperlink",
"time": "18.393",
"value": "2.1 Ghost Island Wendell Holland_section_2"
},
{
"operation": "Highlighting",
"time": "20.762",
"value": "Wendell Holland_sentence_7"
},
{
"operation": "search",
"time": "45.523",
"value": "Contains...Survivor"
},
{
"operation": "Now on",
"time": "48.595",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "53.9",
"value": "Wendell Holland"
},
{
"operation": "Now on",
"time": "54.23",
"value": "Wendell Holland"
},
{
"operation": "Page search",
"time": "63.776",
"value": "Naviti"
},
{
"operation": "Page search",
"time": "65.007",
"value": "Naviti"
},
{
"operation": "Highlighting",
"time": "85.4",
"value": "Wendell Holland_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "88.945",
"value": "Wendell Holland_sentence_1"
},
{
"operation": "finish",
"time": "97.196",
"value": "finish"
}
]
}
|
45918
|
Terje Rypdal, a musician who plays drums, was born in 1947 in Oslo (the capital and most populous city of Norway) but has yet to land a recording contract.
|
{
"evidence": [
"Terje Rypdal_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:\nTerje Rypdal, a musician who plays drums, was born in 1947 in Oslo (the capital and most populous city of Norway) but has yet to land a recording contract.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Terje Rypdal_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5353,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.88",
"value": "?search="
},
{
"operation": "search",
"time": "17.27",
"value": "Terje Rypdal"
},
{
"operation": "Now on",
"time": "18.602",
"value": "Terje Rypdal"
},
{
"operation": "Highlighting",
"time": "181.693",
"value": "Terje Rypdal_sentence_0"
},
{
"operation": "finish",
"time": "197.776",
"value": "finish"
}
]
}
|
46387
|
Fred Puffer was the son of Customs deputy collector Alfred Puffer and began his athletic career in 1887 with YMCA and the Olympic Athletic Club.
|
{
"evidence": [
"Fred Puffer_sentence_2",
"Fred Puffer_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:\nFred Puffer was the son of Customs deputy collector Alfred Puffer and began his athletic career in 1887 with YMCA and the Olympic Athletic Club.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Fred Puffer_sentence_2",
"Fred Puffer_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5354,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.283",
"value": "?search="
},
{
"operation": "search",
"time": "8.792",
"value": "Fred Puffer"
},
{
"operation": "Now on",
"time": "9.657",
"value": "Fred Puffer"
},
{
"operation": "Highlighting",
"time": "70.307",
"value": "Fred Puffer_sentence_2"
},
{
"operation": "Highlighting",
"time": "77.638",
"value": "Fred Puffer_sentence_3"
},
{
"operation": "finish",
"time": "82.695",
"value": "finish"
}
]
}
|
89350
|
Or83b's function is only one, that is, to interact with specific odor stimuli independently of other odorant receptors (ORs).
|
{
"evidence": [
"Or83b odorant receptor_sentence_2",
"Or83b odorant receptor_sentence_3",
"Or83b odorant receptor_sentence_4",
"Or83b odorant receptor_sentence_5"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOr83b's function is only one, that is, to interact with specific odor stimuli independently of other odorant receptors (ORs).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Or83b odorant receptor_sentence_2",
"Or83b odorant receptor_sentence_3",
"Or83b odorant receptor_sentence_4",
"Or83b odorant receptor_sentence_5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5355,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.342",
"value": "?search="
},
{
"operation": "search",
"time": "9.038",
"value": "Or83b odorant receptor"
},
{
"operation": "Now on",
"time": "10.184",
"value": "Or83b odorant receptor"
},
{
"operation": "Highlighting",
"time": "22.746",
"value": "Or83b odorant receptor_sentence_4"
},
{
"operation": "Highlighting",
"time": "40.174",
"value": "Or83b odorant receptor_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "44.846",
"value": "Or83b odorant receptor_sentence_4"
},
{
"operation": "Highlighting",
"time": "45.534",
"value": "Or83b odorant receptor_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "48.524",
"value": "Or83b odorant receptor_sentence_5"
},
{
"operation": "Highlighting",
"time": "49.182",
"value": "Or83b odorant receptor_sentence_3"
},
{
"operation": "Highlighting",
"time": "49.886",
"value": "Or83b odorant receptor_sentence_4"
},
{
"operation": "Highlighting",
"time": "50.43",
"value": "Or83b odorant receptor_sentence_5"
},
{
"operation": "finish",
"time": "56.486",
"value": "finish"
}
]
}
|
16085
|
Hamed Abderrahaman Ahmad is a Spanish terrorist who was arrested in Pakistan for fighting for al-Qaeda in Afghanistan.
|
{
"evidence": [
"Hamed Abderrahaman Ahmad_sentence_0",
"Hamed Abderrahaman Ahmad_sentence_3",
"Hamed Abderrahaman Ahmad_sentence_4",
"Hamed Abderrahaman Ahmad_sentence_31",
"Hamed Abderrahaman Ahmad_sentence_32",
"Hamed Abderrahaman Ahmad_sentence_33",
"Hamed Abderrahaman Ahmad_sentence_34",
"Hamed Abderrahaman Ahmad_sentence_5",
"Hamed Abderrahaman Ahmad_sentence_6"
],
"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:\nHamed Abderrahaman Ahmad is a Spanish terrorist who was arrested in Pakistan for fighting for al-Qaeda in Afghanistan.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hamed Abderrahaman Ahmad_sentence_0",
"Hamed Abderrahaman Ahmad_sentence_3",
"Hamed Abderrahaman Ahmad_sentence_4",
"Hamed Abderrahaman Ahmad_sentence_31",
"Hamed Abderrahaman Ahmad_sentence_32",
"Hamed Abderrahaman Ahmad_sentence_33",
"Hamed Abderrahaman Ahmad_sentence_34",
"Hamed Abderrahaman Ahmad_sentence_5",
"Hamed Abderrahaman Ahmad_sentence_6"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5356,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.166",
"value": "?search="
},
{
"operation": "search",
"time": "11.334",
"value": "Hamed Abderrahaman Ahmad"
},
{
"operation": "Now on",
"time": "12.215",
"value": "Hamed Abderrahaman Ahmad"
},
{
"operation": "Highlighting",
"time": "36.966",
"value": "Hamed Abderrahaman Ahmad_sentence_0"
},
{
"operation": "Highlighting",
"time": "39.002",
"value": "Hamed Abderrahaman Ahmad_sentence_3"
},
{
"operation": "Highlighting",
"time": "41.314",
"value": "Hamed Abderrahaman Ahmad_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "59.882",
"value": "Hamed Abderrahaman Ahmad_sentence_5"
},
{
"operation": "Highlighting",
"time": "60.714",
"value": "Hamed Abderrahaman Ahmad_sentence_4"
},
{
"operation": "Highlighting",
"time": "62.011",
"value": "Hamed Abderrahaman Ahmad_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "132.998",
"value": "Hamed Abderrahaman Ahmad_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "135.862",
"value": "Hamed Abderrahaman Ahmad_sentence_4"
},
{
"operation": "Highlighting",
"time": "138.813",
"value": "Hamed Abderrahaman Ahmad_sentence_4"
},
{
"operation": "Highlighting",
"time": "202.495",
"value": "Hamed Abderrahaman Ahmad_sentence_31"
},
{
"operation": "Highlighting",
"time": "203.815",
"value": "Hamed Abderrahaman Ahmad_sentence_32"
},
{
"operation": "Highlighting",
"time": "215.248",
"value": "Hamed Abderrahaman Ahmad_sentence_33"
},
{
"operation": "Highlighting",
"time": "216.144",
"value": "Hamed Abderrahaman Ahmad_sentence_34"
},
{
"operation": "Highlighting",
"time": "245.122",
"value": "Hamed Abderrahaman Ahmad_sentence_5"
},
{
"operation": "Highlighting",
"time": "247.009",
"value": "Hamed Abderrahaman Ahmad_sentence_6"
},
{
"operation": "finish",
"time": "309.043",
"value": "finish"
}
]
}
|
exfever_5672
|
My Own Private Idaho is a 1991 American independent adventure drama film loosely based on Shakespeare's Henry IV, Part 1, Henry IV, Part 2, and a historical play that focuses on events before and after the Battle of Agincourt.
|
{
"evidence": [
"My Own Private Idaho",
"Henry V (play)"
],
"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:\nMy Own Private Idaho is a 1991 American independent adventure drama film loosely based on Shakespeare's Henry IV, Part 1, Henry IV, Part 2, and a historical play that focuses on events before and after the Battle of Agincourt.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"My Own Private Idaho",
"Henry V (play)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5357,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3380
|
Los Angeles is the largest city in California, which is known for its wet summers and mild, wet winters.
|
{
"evidence": [
"Los Angeles",
"Mediterranean climate"
],
"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:\nLos Angeles is the largest city in California, which is known for its wet summers and mild, wet winters.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Los Angeles",
"Mediterranean climate"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5358,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
38497
|
Zoe Dunning, a Navy vessel commander, announced she was lesbian after another service member announced they were homosexual as well, leading to Dunning being recommended for honorable discharge.
|
{
"evidence": [
"Commander (United States)_sentence_7",
"Commander (United States)_sentence_13",
"Commander (United States)_sentence_14",
"Zoe Dunning_sentence_13",
"Zoe Dunning_sentence_12",
"Zoe Dunning_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:\nZoe Dunning, a Navy vessel commander, announced she was lesbian after another service member announced they were homosexual as well, leading to Dunning being recommended for honorable discharge.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Commander (United States)_sentence_7",
"Commander (United States)_sentence_13",
"Commander (United States)_sentence_14",
"Zoe Dunning_sentence_13",
"Zoe Dunning_sentence_12",
"Zoe Dunning_sentence_14"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5359,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.964",
"value": "?search="
},
{
"operation": "search",
"time": "9.085",
"value": "Zoe Dunning"
},
{
"operation": "Now on",
"time": "9.947",
"value": "Zoe Dunning"
},
{
"operation": "Page search",
"time": "46.402",
"value": "vessel"
},
{
"operation": "Highlighting",
"time": "82.702",
"value": "Zoe Dunning_sentence_10"
},
{
"operation": "Highlighting",
"time": "83.914",
"value": "Zoe Dunning_sentence_11"
},
{
"operation": "Highlighting",
"time": "193.343",
"value": "Zoe Dunning_sentence_16"
},
{
"operation": "Highlighting deleted",
"time": "274.482",
"value": "Zoe Dunning_sentence_16"
},
{
"operation": "Highlighting",
"time": "276.001",
"value": "Zoe Dunning_sentence_12"
},
{
"operation": "Highlighting",
"time": "323.004",
"value": "Zoe Dunning_sentence_13"
},
{
"operation": "Highlighting deleted",
"time": "324.772",
"value": "Zoe Dunning_sentence_13"
},
{
"operation": "Page search",
"time": "334.022",
"value": "Meinhold"
},
{
"operation": "Highlighting deleted",
"time": "355.708",
"value": "Zoe Dunning_sentence_12"
},
{
"operation": "Highlighting",
"time": "357.869",
"value": "Zoe Dunning_sentence_13"
},
{
"operation": "Highlighting",
"time": "360.068",
"value": "Zoe Dunning_sentence_12"
},
{
"operation": "Highlighting",
"time": "371.021",
"value": "Zoe Dunning_sentence_14"
},
{
"operation": "Highlighting",
"time": "429.927",
"value": "Zoe Dunning_sentence_8"
},
{
"operation": "Highlighting",
"time": "431.239",
"value": "Zoe Dunning_sentence_9"
},
{
"operation": "Highlighting",
"time": "435.048",
"value": "Zoe Dunning_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "435.936",
"value": "Zoe Dunning_cell_0_6_1"
},
{
"operation": "Highlighting",
"time": "435.949",
"value": "Zoe Dunning_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "435.949",
"value": "Zoe Dunning_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "439.4",
"value": "Zoe Dunning_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "440.823",
"value": "Zoe Dunning_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "440.832",
"value": "Zoe Dunning_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "441.791",
"value": "Zoe Dunning_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "442.711",
"value": "Zoe Dunning_cell_0_5_1"
},
{
"operation": "Highlighting",
"time": "442.72",
"value": "Zoe Dunning_header_cell_0_5_0"
},
{
"operation": "Highlighting deleted",
"time": "448.839",
"value": "Zoe Dunning_cell_0_6_1"
},
{
"operation": "Highlighting deleted",
"time": "448.846",
"value": "Zoe Dunning_header_cell_0_6_0"
},
{
"operation": "Highlighting deleted",
"time": "449.4",
"value": "Zoe Dunning_cell_0_5_1"
},
{
"operation": "Highlighting deleted",
"time": "449.406",
"value": "Zoe Dunning_header_cell_0_5_0"
},
{
"operation": "Highlighting deleted",
"time": "450.272",
"value": "Zoe Dunning_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "450.279",
"value": "Zoe Dunning_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "450.282",
"value": "Zoe Dunning_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "462.131",
"value": "Commander"
},
{
"operation": "Now on",
"time": "463.174",
"value": "Commander (United States)"
},
{
"operation": "hyperlink",
"time": "470.7",
"value": "1.2 Naval Commander (United States)_section_2"
},
{
"operation": "Highlighting",
"time": "476.745",
"value": "Commander (United States)_sentence_7"
},
{
"operation": "Highlighting",
"time": "497.121",
"value": "Commander (United States)_sentence_13"
},
{
"operation": "Highlighting",
"time": "512.538",
"value": "Commander (United States)_sentence_14"
},
{
"operation": "back-button-clicked",
"time": "517.884",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "518.192",
"value": "Zoe Dunning"
},
{
"operation": "Highlighting deleted",
"time": "526.492",
"value": "Zoe Dunning_sentence_8"
},
{
"operation": "Highlighting deleted",
"time": "528.611",
"value": "Zoe Dunning_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "533.978",
"value": "Zoe Dunning_sentence_10"
},
{
"operation": "Highlighting deleted",
"time": "535.186",
"value": "Zoe Dunning_sentence_11"
},
{
"operation": "Highlighting deleted",
"time": "581.589",
"value": "Zoe Dunning_sentence_12"
},
{
"operation": "Highlighting deleted",
"time": "582.444",
"value": "Zoe Dunning_sentence_13"
},
{
"operation": "Highlighting deleted",
"time": "583.068",
"value": "Zoe Dunning_sentence_14"
},
{
"operation": "Highlighting",
"time": "593.101",
"value": "Zoe Dunning_sentence_13"
},
{
"operation": "Highlighting",
"time": "594.38",
"value": "Zoe Dunning_sentence_12"
},
{
"operation": "Highlighting",
"time": "595.628",
"value": "Zoe Dunning_sentence_14"
},
{
"operation": "finish",
"time": "615.134",
"value": "finish"
}
]
}
|
41689
|
Carlos Villanueva can play basketball and soccer, but he never excelled in the latter so he pursued the NBA.
|
{
"evidence": [
"Charlie Villanueva_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:\nCarlos Villanueva can play basketball and soccer, but he never excelled in the latter so he pursued the NBA.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Charlie Villanueva_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5360,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.892",
"value": "?search="
},
{
"operation": "search",
"time": "18.285",
"value": "Carlos Villanueva"
},
{
"operation": "Now on",
"time": "19.226",
"value": "Carlos Villanueva"
},
{
"operation": "hyperlink",
"time": "38.079",
"value": "Carlos Villanueva López"
},
{
"operation": "Now on",
"time": "38.507",
"value": "Carlos Villanueva López"
},
{
"operation": "back-button-clicked",
"time": "95.578",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "95.621",
"value": "Carlos Villanueva"
},
{
"operation": "Highlighting",
"time": "101.934",
"value": "Carlos Villanueva_item_0_2"
},
{
"operation": "Highlighting",
"time": "101.935",
"value": "Carlos Villanueva_title"
},
{
"operation": "Highlighting",
"time": "103.376",
"value": "Carlos Villanueva_item_0_3"
},
{
"operation": "Highlighting deleted",
"time": "104.086",
"value": "Carlos Villanueva_item_0_3"
},
{
"operation": "Highlighting deleted",
"time": "104.688",
"value": "Carlos Villanueva_item_0_2"
},
{
"operation": "Highlighting deleted",
"time": "104.689",
"value": "Carlos Villanueva_title"
},
{
"operation": "hyperlink",
"time": "105.898",
"value": "Carlos Villanueva (footballer)"
},
{
"operation": "Now on",
"time": "106.367",
"value": "Carlos Villanueva (footballer)"
},
{
"operation": "Page search",
"time": "135.028",
"value": "basketball"
},
{
"operation": "back-button-clicked",
"time": "139.992",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "140.045",
"value": "Carlos Villanueva"
},
{
"operation": "hyperlink",
"time": "141.899",
"value": "Carlos Villanueva (baseball)"
},
{
"operation": "Now on",
"time": "142.335",
"value": "Carlos Villanueva (baseball)"
},
{
"operation": "Page search",
"time": "150.387",
"value": "basketball"
},
{
"operation": "Page search",
"time": "171.219",
"value": "soccer"
},
{
"operation": "back-button-clicked",
"time": "211.075",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "211.122",
"value": "Carlos Villanueva"
},
{
"operation": "search",
"time": "492.139",
"value": "charlie Villanueva"
},
{
"operation": "Now on",
"time": "494.152",
"value": "Charlie Villanueva"
},
{
"operation": "back-button-clicked",
"time": "520.882",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "520.926",
"value": "Carlos Villanueva"
},
{
"operation": "hyperlink",
"time": "532.439",
"value": "Carlos Villanueva (footballer)"
},
{
"operation": "Now on",
"time": "532.754",
"value": "Carlos Villanueva (footballer)"
},
{
"operation": "search",
"time": "561.285",
"value": "Carlos Villanueva nba"
},
{
"operation": "Now on",
"time": "565.013",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "593.723",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "593.799",
"value": "Carlos Villanueva (footballer)"
},
{
"operation": "back-button-clicked",
"time": "596.179",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "596.182",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "596.234",
"value": "Carlos Villanueva"
},
{
"operation": "hyperlink",
"time": "597.83",
"value": "Carlos Villanueva (baseball)"
},
{
"operation": "Now on",
"time": "598.127",
"value": "Carlos Villanueva (baseball)"
},
{
"operation": "Page search",
"time": "629.735",
"value": "soccer"
},
{
"operation": "Page search",
"time": "629.945",
"value": "soccer"
},
{
"operation": "Page search",
"time": "637.431",
"value": "football"
},
{
"operation": "Page search",
"time": "637.729",
"value": "football"
},
{
"operation": "back-button-clicked",
"time": "641.964",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "642.03",
"value": "Carlos Villanueva"
},
{
"operation": "search",
"time": "650.632",
"value": "Carlos Villanueva Fuentes"
},
{
"operation": "Now on",
"time": "651.642",
"value": "Carlo Villanueva Fuentes"
},
{
"operation": "back-button-clicked",
"time": "659.179",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "659.217",
"value": "Carlos Villanueva"
},
{
"operation": "search",
"time": "674.272",
"value": "charlie Villanueva"
},
{
"operation": "Now on",
"time": "674.714",
"value": "Charlie Villanueva"
},
{
"operation": "Highlighting",
"time": "678.223",
"value": "Charlie Villanueva_sentence_0"
},
{
"operation": "finish",
"time": "687.722",
"value": "finish"
}
]
}
|
exfever_275
|
The Great Outdoors is a film produced by an American filmmaker and author of humorous essays and stories for American humor magazine that ran from 1970 to 1999.
|
{
"evidence": [
"The Great Outdoors (film)",
"John Hughes (filmmaker)",
"National Lampoon (magazine)"
],
"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 Great Outdoors is a film produced by an American filmmaker and author of humorous essays and stories for American humor magazine that ran from 1970 to 1999.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Great Outdoors (film)",
"John Hughes (filmmaker)",
"National Lampoon (magazine)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5361,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4690
|
James Mangold's debut film is Copland.
|
{
"evidence": [
"James Mangold",
"Logan (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:\nJames Mangold's debut film is Copland.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"James Mangold",
"Logan (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5362,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
11809
|
Fullstem is an animal.
|
{
"evidence": [
"Chamaechaenactis_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:\nFullstem is an animal.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Chamaechaenactis_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5363,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.639",
"value": "?search="
},
{
"operation": "search",
"time": "8.113",
"value": "Fullstem"
},
{
"operation": "Now on",
"time": "9.678",
"value": "Fullstem"
},
{
"operation": "Highlighting",
"time": "15.363",
"value": "Chamaechaenactis_sentence_0"
},
{
"operation": "finish",
"time": "22.114",
"value": "finish"
}
]
}
|
exfever_2106
|
Ralph Nathaniel Twisleton-Wykeham-Fiennes first achieved success as a Shakespeare interpreter onstage at one of the United Kingdom's most prominent publicly funded performing arts venues.
|
{
"evidence": [
"Ralph Fiennes",
"Royal National Theatre"
],
"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:\nRalph Nathaniel Twisleton-Wykeham-Fiennes first achieved success as a Shakespeare interpreter onstage at one of the United Kingdom's most prominent publicly funded performing arts venues.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ralph Fiennes",
"Royal National Theatre"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5364,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2581
|
"Just the Way You Are" is the second solo single that was certified thirteen times platinum.
|
{
"evidence": [
"Just the Way You Are (Bruno Mars song)",
"The Smeezingtons",
"California"
],
"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:\n\"Just the Way You Are\" is the second solo single that was certified thirteen times platinum.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Just the Way You Are (Bruno Mars song)",
"The Smeezingtons",
"California"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5365,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
46241
|
The Contribution of the Aviation Industry to the UK Economy was published in 1999, and The Economic Contribution of the Aviation Industry in the UK, co-sponsored by the national tourist agency VisitBritain, was published in October 2006.
|
{
"evidence": [
"Environmental impact of aviation in the United Kingdom_sentence_40",
"Environmental impact of aviation in the United Kingdom_sentence_41"
],
"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 Contribution of the Aviation Industry to the UK Economy was published in 1999, and The Economic Contribution of the Aviation Industry in the UK, co-sponsored by the national tourist agency VisitBritain, was published in October 2006.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Environmental impact of aviation in the United Kingdom_sentence_40",
"Environmental impact of aviation in the United Kingdom_sentence_41"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5366,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.653",
"value": "undefined"
},
{
"operation": "search",
"time": "13.092",
"value": "Contains...Contribution of the Aviation Industry to the UK Economy"
},
{
"operation": "Now on",
"time": "13.939",
"value": "undefined"
},
{
"operation": "search",
"time": "37.835",
"value": "Contains...The Economic Contribution of the Aviation Industry in the UK"
},
{
"operation": "Now on",
"time": "39.156",
"value": "undefined"
},
{
"operation": "search",
"time": "85.546",
"value": "Contains...VisitBritain"
},
{
"operation": "Now on",
"time": "86.637",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "96.658",
"value": "VisitEngland"
},
{
"operation": "Now on",
"time": "97.377",
"value": "VisitEngland"
},
{
"operation": "Highlighting",
"time": "108.843",
"value": "VisitEngland_item_0_1"
},
{
"operation": "Highlighting",
"time": "108.846",
"value": "VisitEngland_title"
},
{
"operation": "Highlighting",
"time": "108.848",
"value": "VisitEngland_section_1"
},
{
"operation": "Highlighting deleted",
"time": "109.559",
"value": "VisitEngland_item_0_1"
},
{
"operation": "Highlighting deleted",
"time": "109.561",
"value": "VisitEngland_title"
},
{
"operation": "Highlighting deleted",
"time": "109.564",
"value": "VisitEngland_section_1"
},
{
"operation": "search",
"time": "125.443",
"value": "Contains...Contribution of the Aviation Industry"
},
{
"operation": "Now on",
"time": "126.983",
"value": "undefined"
},
{
"operation": "search",
"time": "172.804",
"value": "UK economy"
},
{
"operation": "Now on",
"time": "175.151",
"value": "Economy of the United Kingdom"
},
{
"operation": "search",
"time": "240.419",
"value": "Contains...The Contribution of the Aviation Industry to the UK Economy was published in 1999"
},
{
"operation": "Now on",
"time": "242.184",
"value": "undefined"
},
{
"operation": "search",
"time": "292.554",
"value": "Contains...The Economic Contribution of the Aviation Industry"
},
{
"operation": "Now on",
"time": "293.618",
"value": "undefined"
},
{
"operation": "search",
"time": "317.365",
"value": "Contains...the national tourist agency VisitBritain"
},
{
"operation": "Now on",
"time": "319.451",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "331.189",
"value": "Environmental impact of aviation in the United Kingdom"
},
{
"operation": "Now on",
"time": "332.167",
"value": "Environmental impact of aviation in the United Kingdom"
},
{
"operation": "Highlighting",
"time": "348.827",
"value": "Environmental impact of aviation in the United Kingdom_sentence_40"
},
{
"operation": "Highlighting",
"time": "357.786",
"value": "Environmental impact of aviation in the United Kingdom_sentence_41"
},
{
"operation": "finish",
"time": "377.681",
"value": "finish"
}
]
}
|
exfever_713
|
Nick Jonas is an American singer-songwriter and a member of the band Jonas Brothers, which became prominent figures on the American pay television channel and the flagship property of Disney Branded Television.
|
{
"evidence": [
"Nick Jonas",
"Disney Channel"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNick Jonas is an American singer-songwriter and a member of the band Jonas Brothers, which became prominent figures on the American pay television channel and the flagship property of Disney Branded Television.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nick Jonas",
"Disney Channel"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5367,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21789
|
Marie-Marguerite-Denis or Marguerite Canal was the first woman in France to conduct an orchestra and the first woman to receive the First Grand Prix of Rome.
|
{
"evidence": [
"Marguerite Canal_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:\nMarie-Marguerite-Denis or Marguerite Canal was the first woman in France to conduct an orchestra and the first woman to receive the First Grand Prix of Rome.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Marguerite Canal_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5368,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.128",
"value": "?search="
},
{
"operation": "search",
"time": "10.505",
"value": "Marie-Marguerite-Denis"
},
{
"operation": "Now on",
"time": "11.579",
"value": "undefined"
},
{
"operation": "search",
"time": "16.032",
"value": "Marguerite Canal"
},
{
"operation": "Now on",
"time": "16.945",
"value": "Marguerite Canal"
},
{
"operation": "Highlighting",
"time": "21.395",
"value": "Marguerite Canal_sentence_3"
},
{
"operation": "Highlighting",
"time": "23.89",
"value": "Marguerite Canal_sentence_0"
},
{
"operation": "Highlighting",
"time": "32.305",
"value": "Marguerite Canal_sentence_4"
},
{
"operation": "finish",
"time": "41.342",
"value": "finish"
}
]
}
|
exfever_248
|
John Nash was one of the foremost British architects of the era of period in British history from 1714 to c. 1830–37.
|
{
"evidence": [
"John Nash (architect)",
"Georgian era"
],
"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:\nJohn Nash was one of the foremost British architects of the era of period in British history from 1714 to c. 1830–37.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Nash (architect)",
"Georgian era"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5369,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3088
|
Lisa Marie Presley is the only child of an Australian businesswoman and actress with the King of Rock and Roll.
|
{
"evidence": [
"Lisa Marie Presley",
"Priscilla Presley",
"Elvis Presley"
],
"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:\nLisa Marie Presley is the only child of an Australian businesswoman and actress with the King of Rock and Roll.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lisa Marie Presley",
"Priscilla Presley",
"Elvis Presley"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5370,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
61323
|
Originally owned by Starke Christian Education Radio & TV, the station started broadcasting in 1982 before changing ownership multiple times and call signs until 2018 when its call sign was changed to WTLG.
|
{
"evidence": [
"WLSF_sentence_9"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOriginally owned by Starke Christian Education Radio & TV, the station started broadcasting in 1982 before changing ownership multiple times and call signs until 2018 when its call sign was changed to WTLG.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"WLSF_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5371,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.683",
"value": "?search="
},
{
"operation": "search",
"time": "19.047",
"value": "WTLG"
},
{
"operation": "Now on",
"time": "19.984",
"value": "K-Love"
},
{
"operation": "Page search",
"time": "37.407",
"value": "starke"
},
{
"operation": "Page search",
"time": "66.862",
"value": "2081"
},
{
"operation": "Page search",
"time": "69.943",
"value": "2018"
},
{
"operation": "search",
"time": "115.077",
"value": "WLSF"
},
{
"operation": "Now on",
"time": "116.05",
"value": "WLSF"
},
{
"operation": "Highlighting",
"time": "151.613",
"value": "WLSF_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "153.149",
"value": "WLSF_sentence_3"
},
{
"operation": "Highlighting",
"time": "153.855",
"value": "WLSF_sentence_2"
},
{
"operation": "Highlighting",
"time": "154.368",
"value": "WLSF_sentence_3"
},
{
"operation": "Highlighting",
"time": "170.452",
"value": "WLSF_sentence_4"
},
{
"operation": "Highlighting",
"time": "172.307",
"value": "WLSF_sentence_5"
},
{
"operation": "Highlighting",
"time": "173.601",
"value": "WLSF_sentence_6"
},
{
"operation": "Highlighting",
"time": "174.788",
"value": "WLSF_sentence_7"
},
{
"operation": "Highlighting",
"time": "175.372",
"value": "WLSF_sentence_8"
},
{
"operation": "Highlighting",
"time": "176.352",
"value": "WLSF_sentence_9"
},
{
"operation": "Highlighting",
"time": "204.278",
"value": "WLSF_sentence_9"
},
{
"operation": "finish",
"time": "211.441",
"value": "finish"
}
]
}
|
exfever_3347
|
Eunice Kathleen Waymon's music spanned styles including classical, jazz, blues, folk, R&B, gospel, and a genre of popular music that originated in its modern form during the mid-1950s in the United States and the United Kingdom.
|
{
"evidence": [
"Nina Simone",
"Pop music"
],
"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:\nEunice Kathleen Waymon's music spanned styles including classical, jazz, blues, folk, R&B, gospel, and a genre of popular music that originated in its modern form during the mid-1950s in the United States and the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nina Simone",
"Pop music"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5372,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3530
|
Thomas Black is one of the highest-paid?American actors.
|
{
"evidence": [
"Jack Black",
"Orange County (film)",
"Comedy 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:\nThomas Black is one of the highest-paid?American actors.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jack Black",
"Orange County (film)",
"Comedy film"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5373,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7481
|
Major League Soccer is the first professional men's soccer league in the U.S.
|
{
"evidence": [
"Major League Soccer",
"United States Soccer Federation"
],
"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:\nMajor League Soccer is the first professional men's soccer league in the U.S.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Major League Soccer",
"United States Soccer Federation"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5374,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
86694
|
The Alberdingk family consisted of a great-grandson named Jonas Alberding, and a son named Johannes Heinrich Alberding.
|
{
"evidence": [
"Alberdingk_sentence_3",
"Alberdingk_sentence_4"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Alberdingk family consisted of a great-grandson named Jonas Alberding, and a son named Johannes Heinrich Alberding.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Alberdingk_sentence_3",
"Alberdingk_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5375,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.462",
"value": "?search="
},
{
"operation": "search",
"time": "14.784",
"value": "Alberdingk"
},
{
"operation": "Now on",
"time": "15.73",
"value": "Alberdingk"
},
{
"operation": "Highlighting",
"time": "40.238",
"value": "Alberdingk_sentence_3"
},
{
"operation": "Highlighting",
"time": "41.481",
"value": "Alberdingk_sentence_4"
},
{
"operation": "finish",
"time": "59.615",
"value": "finish"
}
]
}
|
61485
|
Silambarasan released the film, Kuththu, which received favorable reviews, and his final release was Manmadhan (which he wrote,) which was a major success and propelled him to stardom.
|
{
"evidence": [
"Silambarasan_sentence_24",
"Silambarasan_sentence_26",
"Silambarasan_sentence_24"
],
"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:\nSilambarasan released the film, Kuththu, which received favorable reviews, and his final release was Manmadhan (which he wrote,) which was a major success and propelled him to stardom.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Silambarasan_sentence_24",
"Silambarasan_sentence_26",
"Silambarasan_sentence_24"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5376,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.824",
"value": "?search="
},
{
"operation": "search",
"time": "17.494",
"value": "Silambarasan"
},
{
"operation": "Now on",
"time": "19.819",
"value": "Silambarasan"
},
{
"operation": "Highlighting",
"time": "102.965",
"value": "Silambarasan_sentence_24"
},
{
"operation": "Highlighting",
"time": "131.596",
"value": "Silambarasan_sentence_26"
},
{
"operation": "Highlighting",
"time": "132.66",
"value": "Silambarasan_sentence_24"
},
{
"operation": "finish",
"time": "171.923",
"value": "finish"
}
]
}
|
exfever_4925
|
Beautiful is a 2000 comedy-drama film starring an American former child actress, best known for being "The Pepsi Girl".
|
{
"evidence": [
"Beautiful (2000 film)",
"Hallie Eisenberg"
],
"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:\nBeautiful is a 2000 comedy-drama film starring an American former child actress, best known for being \"The Pepsi Girl\".\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Beautiful (2000 film)",
"Hallie Eisenberg"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5377,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
40362
|
Nezamysl was one of seven mythical princes whose names were recorded in Cosmas chronicle and then transmitted into historical works, but they are not connected to the Ducal Rotunda of the Virgin Mary dating back to the 11th or 12th century.
|
{
"evidence": [
"Nezamysl_sentence_0",
"Nezamysl_sentence_1",
"Nezamysl_sentence_2",
"Nezamysl_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:\nNezamysl was one of seven mythical princes whose names were recorded in Cosmas chronicle and then transmitted into historical works, but they are not connected to the Ducal Rotunda of the Virgin Mary dating back to the 11th or 12th century.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Nezamysl_sentence_0",
"Nezamysl_sentence_1",
"Nezamysl_sentence_2",
"Nezamysl_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5378,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.676",
"value": "?search="
},
{
"operation": "search",
"time": "25.585",
"value": "Nezamysi"
},
{
"operation": "Now on",
"time": "26.4",
"value": "undefined"
},
{
"operation": "search",
"time": "45.632",
"value": "Cosmas chronicle"
},
{
"operation": "Now on",
"time": "46.836",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "50.173",
"value": "Přemysl the Ploughman"
},
{
"operation": "Now on",
"time": "50.687",
"value": "Přemysl the Ploughman"
},
{
"operation": "back-button-clicked",
"time": "54.54",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "55.127",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "57.069",
"value": "Nezamysl"
},
{
"operation": "Now on",
"time": "57.581",
"value": "Nezamysl"
},
{
"operation": "Highlighting",
"time": "60.861",
"value": "Nezamysl_sentence_0"
},
{
"operation": "Highlighting",
"time": "66.857",
"value": "Nezamysl_sentence_1"
},
{
"operation": "Highlighting",
"time": "81.426",
"value": "Nezamysl_sentence_2"
},
{
"operation": "Highlighting",
"time": "98.954",
"value": "Nezamysl_sentence_3"
},
{
"operation": "finish",
"time": "132.839",
"value": "finish"
}
]
}
|
92666
|
Tamil Solidarity is an international group fighting for the rights of unsheltered canine and feline founded in 2009.
|
{
"evidence": [
"Tamil Solidarity_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:\nTamil Solidarity is an international group fighting for the rights of unsheltered canine and feline founded in 2009.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Tamil Solidarity_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 5379,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.235",
"value": "?search="
},
{
"operation": "search",
"time": "9.761",
"value": "Tamil Solidarity"
},
{
"operation": "Now on",
"time": "10.588",
"value": "Tamil Solidarity"
},
{
"operation": "Highlighting",
"time": "29.608",
"value": "Tamil Solidarity_sentence_0"
},
{
"operation": "finish",
"time": "56.1",
"value": "finish"
}
]
}
|
exfever_6770
|
Holly Hunter is an American actress, who appeared in the 1993 drama film The Piano, which earned her the award presented annually by the Academy of Motion Picture Arts and Sciences.
|
{
"evidence": [
"Holly Hunter",
"Academy Award for Best Actress"
],
"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:\nHolly Hunter is an American actress, who appeared in the 1993 drama film The Piano, which earned her the award presented annually by the Academy of Motion Picture Arts and Sciences.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Holly Hunter",
"Academy Award for Best Actress"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5380,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_105
|
Saving Private Ryan is a 1998 American epic war film that is set during Operation Overlord, the Allied operation that launched the successful invasion of German-occupied Western Europe during World War II.
|
{
"evidence": [
"Saving Private Ryan",
"Invasion of Normandy"
],
"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:\nSaving Private Ryan is a 1998 American epic war film that is set during Operation Overlord, the Allied operation that launched the successful invasion of German-occupied Western Europe during World War II.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Saving Private Ryan",
"Invasion of Normandy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5381,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
77393
|
Harry Lubse (born 23 September 1951) is a retired football striker who represented the Netherlands at the 1978 FIFA World Cup.
|
{
"evidence": [
"Harry Lubse_sentence_0",
"Harry Lubse_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:\nHarry Lubse (born 23 September 1951) is a retired football striker who represented the Netherlands at the 1978 FIFA World Cup.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Harry Lubse_sentence_0",
"Harry Lubse_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 5382,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.705",
"value": "?search="
},
{
"operation": "search",
"time": "16.898",
"value": "Harry Lubse"
},
{
"operation": "Now on",
"time": "17.613",
"value": "Harry Lubse"
},
{
"operation": "Highlighting",
"time": "23.018",
"value": "Harry Lubse_sentence_0"
},
{
"operation": "Highlighting",
"time": "28.049",
"value": "Harry Lubse_sentence_2"
},
{
"operation": "Highlighting",
"time": "39.57",
"value": "Harry Lubse_sentence_2"
},
{
"operation": "Highlighting",
"time": "42.817",
"value": "Harry Lubse_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "42.821",
"value": "Harry Lubse_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "42.821",
"value": "Harry Lubse_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "45.153",
"value": "Harry Lubse_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "45.156",
"value": "Harry Lubse_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "56.641",
"value": "Harry Lubse_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "56.645",
"value": "Harry Lubse_header_cell_0_4_0"
},
{
"operation": "finish",
"time": "78.889",
"value": "finish"
}
]
}
|
exfever_4712
|
Ava Gardner had a stroke in 1986, which left her paralyzed down much of her left side and severely weakened her.
|
{
"evidence": [
"Seven Days in May",
"Ava Gardner",
"Mogambo"
],
"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:\nAva Gardner had a stroke in 1986, which left her paralyzed down much of her left side and severely weakened her.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Seven Days in May",
"Ava Gardner",
"Mogambo"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 5383,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.