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_427
|
Wentworth is an Australian television drama series that initially focuses on the early days in prison of Bea Smith, a character portrayed by a New Zealand stage and screen actress known for her role as Rachel Greene.
|
{
"evidence": [
"Wentworth (TV series)",
"Danielle Cormack"
],
"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:\nWentworth is an Australian television drama series that initially focuses on the early days in prison of Bea Smith, a character portrayed by a New Zealand stage and screen actress known for her role as Rachel Greene.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wentworth (TV series)",
"Danielle Cormack"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6084,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21377
|
Found in China, Drepanidae chinoea has a wingspan of 16-20.5 mm for males, with lustrous white fore- and hindwings.
|
{
"evidence": [
"Drepanidae_sentence_3"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFound in China, Drepanidae chinoea has a wingspan of 16-20.5 mm for males, with lustrous white fore- and hindwings.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Drepanidae_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6085,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.619",
"value": "?search="
},
{
"operation": "search",
"time": "27.18",
"value": "Drepanidae chinoea"
},
{
"operation": "Now on",
"time": "27.845",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "91.159",
"value": "drepanidae chingola"
},
{
"operation": "Now on",
"time": "91.716",
"value": "undefined"
},
{
"operation": "search",
"time": "112.482",
"value": "Contains...Drepanidae chinoe"
},
{
"operation": "Now on",
"time": "112.938",
"value": "undefined"
},
{
"operation": "search",
"time": "131.282",
"value": "Drepanidae"
},
{
"operation": "Now on",
"time": "132.134",
"value": "Drepanidae"
},
{
"operation": "Highlighting",
"time": "148.513",
"value": "Drepanidae_sentence_3"
},
{
"operation": "finish",
"time": "163.631",
"value": "finish"
}
]
}
|
exfever_3005
|
Dhoom 3 is a 2013 Indian romantic thriller film in an Indo-Aryan language spoken chiefly in the Hindi Belt region.
|
{
"evidence": [
"Dhoom 3",
"Hindi"
],
"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:\nDhoom 3 is a 2013 Indian romantic thriller film in an Indo-Aryan language spoken chiefly in the Hindi Belt region.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dhoom 3",
"Hindi"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6086,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
40947
|
Ye Olde Cinder House was semi-detached, it was sold in 1990s and is located in Derbyshire which is a county in England.
|
{
"evidence": [
"Ye Olde Cinder House_sentence_0",
"Ye Olde Cinder House_sentence_7",
"Ye Olde Cinder House_sentence_6",
"Derbyshire_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:\nYe Olde Cinder House was semi-detached, it was sold in 1990s and is located in Derbyshire which is a county in England.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Ye Olde Cinder House_sentence_0",
"Ye Olde Cinder House_sentence_7",
"Ye Olde Cinder House_sentence_6",
"Derbyshire_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6087,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.065",
"value": "?search="
},
{
"operation": "search",
"time": "16.641",
"value": "Ye Olde Cinder House"
},
{
"operation": "Now on",
"time": "17.432",
"value": "Ye Olde Cinder House"
},
{
"operation": "Highlighting",
"time": "29.037",
"value": "Ye Olde Cinder House_sentence_0"
},
{
"operation": "Page search",
"time": "41.448",
"value": "sold"
},
{
"operation": "Highlighting",
"time": "45.711",
"value": "Ye Olde Cinder House_sentence_7"
},
{
"operation": "Highlighting",
"time": "53.688",
"value": "Ye Olde Cinder House_sentence_6"
},
{
"operation": "hyperlink",
"time": "61.125",
"value": "Derbyshire"
},
{
"operation": "Now on",
"time": "61.881",
"value": "Derbyshire"
},
{
"operation": "Highlighting",
"time": "69.843",
"value": "Derbyshire_sentence_2"
},
{
"operation": "finish",
"time": "103.345",
"value": "finish"
}
]
}
|
exfever_1129
|
The War of the Worlds is Miranda Otto's first science fiction action film.
|
{
"evidence": [
"War of the Worlds (2005 film)",
"Amblin Entertainment"
],
"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 War of the Worlds is Miranda Otto's first science fiction action film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"War of the Worlds (2005 film)",
"Amblin Entertainment"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6088,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
64471
|
According to Aleister Crowley, the last three Aeons within the system of Thelema have been (1) the Aeon of Isis, (2) the Aeon of Osiris, and (3) the current Aeon of Horus which began in 1904 with the writing of The Book of the Law
|
{
"evidence": [
"Aeon (Thelema)_sentence_31",
"Aeon (Thelema)_sentence_32",
"Aeon (Thelema)_sentence_7",
"Aeon (Thelema)_sentence_14",
"Aeon (Thelema)_sentence_18"
],
"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:\nAccording to Aleister Crowley, the last three Aeons within the system of Thelema have been (1) the Aeon of Isis, (2) the Aeon of Osiris, and (3) the current Aeon of Horus which began in 1904 with the writing of The Book of the Law\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Aeon (Thelema)_sentence_31",
"Aeon (Thelema)_sentence_32",
"Aeon (Thelema)_sentence_7",
"Aeon (Thelema)_sentence_14",
"Aeon (Thelema)_sentence_18"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6089,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.698",
"value": "?search="
},
{
"operation": "search",
"time": "23.733",
"value": "Contains...system of Thelema"
},
{
"operation": "Now on",
"time": "25.579",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "37.177",
"value": "Choronzon"
},
{
"operation": "Now on",
"time": "38.768",
"value": "Choronzon"
},
{
"operation": "search",
"time": "61.009",
"value": "Aleister Crowley"
},
{
"operation": "Now on",
"time": "78.509",
"value": "Aleister Crowley"
},
{
"operation": "Page search",
"time": "103.499",
"value": "aeon"
},
{
"operation": "hyperlink",
"time": "122.906",
"value": "Thelema"
},
{
"operation": "Now on",
"time": "132.65",
"value": "Thelema"
},
{
"operation": "search",
"time": "179.513",
"value": "Aeon of Osiris"
},
{
"operation": "Now on",
"time": "184.459",
"value": "Aeon (Thelema)"
},
{
"operation": "hyperlink",
"time": "255.935",
"value": "2 Differing views Aeon (Thelema)_section_5"
},
{
"operation": "Highlighting",
"time": "259.346",
"value": "Aeon (Thelema)_sentence_31"
},
{
"operation": "Highlighting",
"time": "260.551",
"value": "Aeon (Thelema)_sentence_32"
},
{
"operation": "search",
"time": "286.511",
"value": "crowley aeons"
},
{
"operation": "Now on",
"time": "288.614",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "303.806",
"value": "Aleister Crowley Thelema_section_9"
},
{
"operation": "Now on",
"time": "304.38",
"value": "Thelema#Aleister Crowley Thelema section 9"
},
{
"operation": "back-button-clicked",
"time": "344.838",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "345.504",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "355.826",
"value": "Aeonic Cosmology and Nazism Order of Nine Angles_section_5"
},
{
"operation": "Now on",
"time": "365.91",
"value": "Order of Nine Angles#Aeonic Cosmology and Nazism Order of Nine Angles section 5"
},
{
"operation": "Page search",
"time": "373.918",
"value": "crowley"
},
{
"operation": "hyperlink",
"time": "401.051",
"value": "Thelemic"
},
{
"operation": "Now on",
"time": "401.637",
"value": "Thelema"
},
{
"operation": "Page search",
"time": "410.381",
"value": "aeon*"
},
{
"operation": "search",
"time": "468.439",
"value": "Aeon of Horus"
},
{
"operation": "Now on",
"time": "469.105",
"value": "Aeon (Thelema)"
},
{
"operation": "Highlighting",
"time": "490.643",
"value": "Aeon (Thelema)_sentence_7"
},
{
"operation": "Highlighting",
"time": "495.57",
"value": "Aeon (Thelema)_sentence_14"
},
{
"operation": "Highlighting",
"time": "499.299",
"value": "Aeon (Thelema)_sentence_18"
},
{
"operation": "finish",
"time": "524.941",
"value": "finish"
}
]
}
|
5774
|
Íte grew up as an only child leading her to live in the Killeedy community alone where she was offered a big grant of land.
|
{
"evidence": [
"Íte of Killeedy_sentence_16"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nÍte grew up as an only child leading her to live in the Killeedy community alone where she was offered a big grant of land.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Íte of Killeedy_sentence_16"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6090,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.486",
"value": "?search="
},
{
"operation": "search",
"time": "16.69",
"value": "Íte"
},
{
"operation": "Now on",
"time": "17.274",
"value": "Íte"
},
{
"operation": "search",
"time": "52.184",
"value": "Killeedy"
},
{
"operation": "Now on",
"time": "52.878",
"value": "Killeedy"
},
{
"operation": "search",
"time": "74.166",
"value": "Ita"
},
{
"operation": "Now on",
"time": "74.854",
"value": "Ita"
},
{
"operation": "hyperlink",
"time": "80.703",
"value": "2 People Ita_section_1"
},
{
"operation": "hyperlink",
"time": "86.783",
"value": "Íte of Killeedy"
},
{
"operation": "Now on",
"time": "87.561",
"value": "Íte of Killeedy"
},
{
"operation": "Highlighting",
"time": "107.568",
"value": "Íte of Killeedy_sentence_16"
},
{
"operation": "finish",
"time": "138.112",
"value": "finish"
}
]
}
|
exfever_1968
|
Fight Club is a 1999 American film directed by David Fincher and starring Edward Norton plays the character with a split personality.
|
{
"evidence": [
"Fight Club",
"The Narrator (Fight Club)"
],
"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:\nFight Club is a 1999 American film directed by David Fincher and starring Edward Norton plays the character with a split personality.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Fight Club",
"The Narrator (Fight Club)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6091,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1466
|
The Ten Commandments was the first religious drama film based on the 1949 novel Prince of Egypt by Dorothy Clarke Wilson.
|
{
"evidence": [
"The Ten Commandments (1956 film)",
"Cecil B. DeMille",
"Paramount Pictures"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Ten Commandments was the first religious drama film based on the 1949 novel Prince of Egypt by Dorothy Clarke Wilson.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Ten Commandments (1956 film)",
"Cecil B. DeMille",
"Paramount Pictures"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6092,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
66842
|
Alejo Sauras, a Spanish actor best known as Paul Martinez in Los Serrano, spent his childhood in Madrid and studied aviation electronics for a while before he started to pursue acting.
|
{
"evidence": [
"Alejo Sauras_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:\nAlejo Sauras, a Spanish actor best known as Paul Martinez in Los Serrano, spent his childhood in Madrid and studied aviation electronics for a while before he started to pursue acting.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Alejo Sauras_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6093,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.489",
"value": "?search="
},
{
"operation": "search",
"time": "16.49",
"value": "Alejo Sauras"
},
{
"operation": "Now on",
"time": "17.154",
"value": "Alejo Sauras"
},
{
"operation": "Highlighting",
"time": "19.659",
"value": "Alejo Sauras_sentence_0"
},
{
"operation": "finish",
"time": "135.117",
"value": "finish"
}
]
}
|
exfever_7412
|
Green Lantern was the first comic book superhero created by DC Comics.
|
{
"evidence": [
"John Stewart (comics)",
"Green Lantern"
],
"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:\nGreen Lantern was the first comic book superhero created by DC Comics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Stewart (comics)",
"Green Lantern"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6094,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
16581
|
Švenčionys Ghetto had 566 Jews - some of which established contracts with active Soviet partisans, who frustrated economical exploitation plans of the Germans towards Soviet territories, while the others joined Jewish partisans.
|
{
"evidence": [
"Švenčionys Ghetto_sentence_0",
"Švenčionys Ghetto_sentence_2",
"Švenčionys Ghetto_sentence_14",
"Švenčionys Ghetto_sentence_15",
"Švenčionys Ghetto_sentence_16"
],
"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:\nŠvenčionys Ghetto had 566 Jews - some of which established contracts with active Soviet partisans, who frustrated economical exploitation plans of the Germans towards Soviet territories, while the others joined Jewish partisans.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Švenčionys Ghetto_sentence_0",
"Švenčionys Ghetto_sentence_2",
"Švenčionys Ghetto_sentence_14",
"Švenčionys Ghetto_sentence_15",
"Švenčionys Ghetto_sentence_16"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6095,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.15",
"value": "?search="
},
{
"operation": "search",
"time": "42.804",
"value": "Švenčionys Ghetto"
},
{
"operation": "Now on",
"time": "43.745",
"value": "Švenčionys Ghetto"
},
{
"operation": "Highlighting",
"time": "45.773",
"value": "Švenčionys Ghetto_sentence_0"
},
{
"operation": "Highlighting",
"time": "54.809",
"value": "Švenčionys Ghetto_sentence_2"
},
{
"operation": "Page search",
"time": "65.839",
"value": "566"
},
{
"operation": "Highlighting",
"time": "68.477",
"value": "Švenčionys Ghetto_sentence_14"
},
{
"operation": "Highlighting",
"time": "81.125",
"value": "Švenčionys Ghetto_sentence_15"
},
{
"operation": "Page search",
"time": "116.059",
"value": "economical exploitation"
},
{
"operation": "Page search",
"time": "165.649",
"value": "economical exploitation"
},
{
"operation": "Page search",
"time": "177.556",
"value": "Jewish partisans"
},
{
"operation": "Highlighting",
"time": "179.933",
"value": "Švenčionys Ghetto_sentence_16"
},
{
"operation": "finish",
"time": "218.774",
"value": "finish"
}
]
}
|
15550
|
The Kingdom of Tooro is a Constitutional Monarchy form of government system currently lead by King Patrick David Matthew Kaboyo Rwamuhokya Olimi III, who took the throne in 1995.
|
{
"evidence": [
"Tooro Kingdom_sentence_1",
"Tooro Kingdom_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 Kingdom of Tooro is a Constitutional Monarchy form of government system currently lead by King Patrick David Matthew Kaboyo Rwamuhokya Olimi III, who took the throne in 1995.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Tooro Kingdom_sentence_1",
"Tooro Kingdom_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6096,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.362",
"value": "?search="
},
{
"operation": "search",
"time": "11.802",
"value": "Contains...kingdom of tooro"
},
{
"operation": "Now on",
"time": "13.352",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "17.817",
"value": "Rwenzururu movement"
},
{
"operation": "Now on",
"time": "20.043",
"value": "Rwenzururu movement"
},
{
"operation": "Page search",
"time": "29.95",
"value": "tooro"
},
{
"operation": "Highlighting",
"time": "33.144",
"value": "Rwenzururu movement_sentence_2"
},
{
"operation": "Highlighting",
"time": "33.145",
"value": "Rwenzururu movement_title"
},
{
"operation": "hyperlink",
"time": "43.653",
"value": "Tooro"
},
{
"operation": "Now on",
"time": "45.095",
"value": "Tooro Kingdom"
},
{
"operation": "Highlighting",
"time": "54.279",
"value": "Tooro Kingdom_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "55.407",
"value": "Tooro Kingdom_cell_0_5_1"
},
{
"operation": "Highlighting",
"time": "55.428",
"value": "Tooro Kingdom_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "55.429",
"value": "Tooro Kingdom_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "67.403",
"value": "Tooro Kingdom_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "73.484",
"value": "Tooro Kingdom_cell_0_5_1"
},
{
"operation": "Highlighting deleted",
"time": "73.505",
"value": "Tooro Kingdom_header_cell_0_5_0"
},
{
"operation": "Highlighting deleted",
"time": "73.507",
"value": "Tooro Kingdom_header_cell_0_0_0"
},
{
"operation": "Page search",
"time": "109.782",
"value": "kaboyo"
},
{
"operation": "Highlighting",
"time": "119.497",
"value": "Tooro Kingdom_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "147.757",
"value": "Tooro Kingdom_sentence_2"
},
{
"operation": "Highlighting",
"time": "160.616",
"value": "Tooro Kingdom_sentence_2"
},
{
"operation": "finish",
"time": "172.42",
"value": "finish"
}
]
}
|
92356
|
Richard I of England, son of King Henry II of England and brother-in-law of William II of Sicily through the marriage of the later with his sister Joan of England, would prove to be just as adventurous as his kinsman.
|
{
"evidence": [
"Richard I of England_sentence_1",
"Richard I of England_sentence_3",
"Joan of England, Queen of Sicily_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:\nRichard I of England, son of King Henry II of England and brother-in-law of William II of Sicily through the marriage of the later with his sister Joan of England, would prove to be just as adventurous as his kinsman.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Richard I of England_sentence_1",
"Richard I of England_sentence_3",
"Joan of England, Queen of Sicily_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6097,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.693",
"value": "undefined"
},
{
"operation": "search",
"time": "6.955",
"value": "Richard I of England"
},
{
"operation": "Now on",
"time": "9.099",
"value": "Richard I of England"
},
{
"operation": "Highlighting",
"time": "11.444",
"value": "Richard I of England_sentence_1"
},
{
"operation": "Highlighting",
"time": "23.204",
"value": "Richard I of England_sentence_3"
},
{
"operation": "Page search",
"time": "54.061",
"value": "William II"
},
{
"operation": "Page search",
"time": "109.306",
"value": "brother-in-law"
},
{
"operation": "Page search",
"time": "129.41",
"value": "joan"
},
{
"operation": "Page search",
"time": "130.068",
"value": "joan"
},
{
"operation": "hyperlink",
"time": "141.776",
"value": "Queen Joan of Sicily"
},
{
"operation": "Now on",
"time": "142.526",
"value": "Joan of England, Queen of Sicily"
},
{
"operation": "Highlighting",
"time": "150.562",
"value": "Joan of England, Queen of Sicily_sentence_3"
},
{
"operation": "back-button-clicked",
"time": "160.723",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "160.854",
"value": "Richard I of England"
},
{
"operation": "Page search",
"time": "180.127",
"value": "adventu"
},
{
"operation": "Page search",
"time": "180.769",
"value": "adventu"
},
{
"operation": "hyperlink",
"time": "209.459",
"value": "Queen Joan of Sicily"
},
{
"operation": "Now on",
"time": "209.764",
"value": "Joan of England, Queen of Sicily"
},
{
"operation": "back-button-clicked",
"time": "346.001",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "346.132",
"value": "Richard I of England"
},
{
"operation": "Page search",
"time": "354.456",
"value": "kinsman"
},
{
"operation": "Page search",
"time": "372.601",
"value": "kinsman"
},
{
"operation": "Page search",
"time": "401.7",
"value": "adventurous"
},
{
"operation": "finish",
"time": "584.79",
"value": "finish"
}
]
}
|
17161
|
Dans Paris was produced by Paulo Branco and written by Christophe Honoré, and it premiered in France.
|
{
"evidence": [
"Dans Paris_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:\nDans Paris was produced by Paulo Branco and written by Christophe Honoré, and it premiered in France.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dans Paris_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6098,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.839",
"value": "?search="
},
{
"operation": "search",
"time": "10.662",
"value": "Dans Paris"
},
{
"operation": "Now on",
"time": "11.68",
"value": "Dans Paris"
},
{
"operation": "Highlighting",
"time": "49.179",
"value": "Dans Paris_sentence_3"
},
{
"operation": "hyperlink",
"time": "118.733",
"value": "2006 Cannes Film Festival"
},
{
"operation": "Now on",
"time": "119.644",
"value": "2006 Cannes Film Festival"
},
{
"operation": "Page search",
"time": "127.581",
"value": "premier"
},
{
"operation": "finish",
"time": "234.233",
"value": "finish"
}
]
}
|
34473
|
Ramsden, a British painter who lives and works in London, studied art at Leith School of Art (LSA) and Edinburgh College of Art, then continued her education and graduated in 2013 from the three-year postgraduate art school at the Royal Academy of Arts, RA, in London.
|
{
"evidence": [
"Mary Ramsden_sentence_0",
"Mary Ramsden_sentence_2",
"Mary Ramsden_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:\nRamsden, a British painter who lives and works in London, studied art at Leith School of Art (LSA) and Edinburgh College of Art, then continued her education and graduated in 2013 from the three-year postgraduate art school at the Royal Academy of Arts, RA, in London.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Mary Ramsden_sentence_0",
"Mary Ramsden_sentence_2",
"Mary Ramsden_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6099,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.488",
"value": "?search="
},
{
"operation": "search",
"time": "13.442",
"value": "Ramsden"
},
{
"operation": "Now on",
"time": "14.332",
"value": "Ramsden"
},
{
"operation": "hyperlink",
"time": "60.771",
"value": "Ramsden, Orpington"
},
{
"operation": "Now on",
"time": "61.55",
"value": "Ramsden, Orpington"
},
{
"operation": "search",
"time": "78.85",
"value": "Ramsden"
},
{
"operation": "Now on",
"time": "79.519",
"value": "Ramsden"
},
{
"operation": "search",
"time": "113.424",
"value": "Ramsden british painter"
},
{
"operation": "Now on",
"time": "115.068",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "148.186",
"value": "Mary Ramsden"
},
{
"operation": "Now on",
"time": "148.877",
"value": "Mary Ramsden"
},
{
"operation": "Highlighting",
"time": "154.055",
"value": "Mary Ramsden_sentence_0"
},
{
"operation": "Highlighting",
"time": "162.196",
"value": "Mary Ramsden_sentence_2"
},
{
"operation": "Highlighting",
"time": "168.685",
"value": "Mary Ramsden_sentence_3"
},
{
"operation": "finish",
"time": "313.546",
"value": "finish"
}
]
}
|
585
|
The aircraft, developed from the Avid Flyer and the Denney Kitfox, has a 4130 steel tube frame fuselage and a wing constructed from copper tubing, all covered in doped fabric.
|
{
"evidence": [
"Carlson Sparrow_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe aircraft, developed from the Avid Flyer and the Denney Kitfox, has a 4130 steel tube frame fuselage and a wing constructed from copper tubing, all covered in doped fabric.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Carlson Sparrow_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6100,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "16.97",
"value": "The aircraft, developed from the Avid Flyer and the Denney Kitfox, has a 4130 steel tube frame fuselage and a wing constructed from copper tubing, all covered in doped fabric."
},
{
"operation": "search",
"time": "33.4",
"value": "4130 steel tube frame fuselage"
},
{
"operation": "hyperlink",
"time": "38.689",
"value": "Carlson Sparrow"
},
{
"operation": "Highlighting",
"time": "61.273",
"value": "Carlson Sparrow_sentence_3"
},
{
"operation": "finish",
"time": "116.595",
"value": "finish"
}
]
}
|
56368
|
Spermatocytes have only one type.
|
{
"evidence": [
"Spermatocyte_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:\nSpermatocytes have only one type.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Spermatocyte_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6101,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.158",
"value": "?search="
},
{
"operation": "search",
"time": "32.306",
"value": "Spermatocytes"
},
{
"operation": "Now on",
"time": "37.604",
"value": "Spermatocytes"
},
{
"operation": "Highlighting",
"time": "42.256",
"value": "Spermatocyte_sentence_3"
},
{
"operation": "finish",
"time": "49.298",
"value": "finish"
}
]
}
|
40029
|
Dany Rigoulet is a former French figure skater who competed in ladies singles and retired in 1960.
|
{
"evidence": [
"Dany Rigoulot_sentence_0",
"Dany Rigoulot_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:\nDany Rigoulet is a former French figure skater who competed in ladies singles and retired in 1960.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Dany Rigoulot_sentence_0",
"Dany Rigoulot_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6102,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.396",
"value": "?search="
},
{
"operation": "search",
"time": "21.865",
"value": "Contains...Dany Rigoulet"
},
{
"operation": "Now on",
"time": "23.323",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "25.673",
"value": "Dany Rigoulot"
},
{
"operation": "Now on",
"time": "26.224",
"value": "Dany Rigoulot"
},
{
"operation": "Highlighting",
"time": "30.442",
"value": "Dany Rigoulot_sentence_0"
},
{
"operation": "Highlighting",
"time": "44.044",
"value": "Dany Rigoulot_sentence_1"
},
{
"operation": "finish",
"time": "55.429",
"value": "finish"
}
]
}
|
79426
|
The adenosine A2A receptor does not interact with proteins.
|
{
"evidence": [
"Adenosine A2A receptor_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:\nThe adenosine A2A receptor does not interact with proteins.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Adenosine A2A receptor_sentence_11"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6103,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.123",
"value": "?search="
},
{
"operation": "search",
"time": "11.293",
"value": "Adenosine A2A receptor"
},
{
"operation": "Now on",
"time": "13.042",
"value": "Adenosine A2A receptor"
},
{
"operation": "Highlighting",
"time": "32.206",
"value": "Adenosine A2A receptor_sentence_11"
},
{
"operation": "finish",
"time": "47.969",
"value": "finish"
}
]
}
|
exfever_6741
|
Resident Evil is a Korean horror video game series and media franchise created by a Japanese video game developer and publisher that created a number of multi-million-selling game franchises.
|
{
"evidence": [
"Resident Evil",
"Capcom"
],
"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:\nResident Evil is a Korean horror video game series and media franchise created by a Japanese video game developer and publisher that created a number of multi-million-selling game franchises.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Resident Evil",
"Capcom"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6104,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
67479
|
Andrew Proud was assistant priest at Bishop's Hatfield ministry for nine years, then became Rector at St. Mary's, East Barnet for two years.
|
{
"evidence": [
"Andrew Proud_sentence_7",
"Andrew Proud_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:\nAndrew Proud was assistant priest at Bishop's Hatfield ministry for nine years, then became Rector at St. Mary's, East Barnet for two years.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Andrew Proud_sentence_7",
"Andrew Proud_sentence_8"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6105,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.801",
"value": "?search="
},
{
"operation": "search",
"time": "15.733",
"value": "Andrew Proud"
},
{
"operation": "Now on",
"time": "17.8",
"value": "Andrew Proud"
},
{
"operation": "Page search",
"time": "63.749",
"value": "assistant"
},
{
"operation": "Highlighting",
"time": "69.889",
"value": "Andrew Proud_sentence_7"
},
{
"operation": "Highlighting",
"time": "85.47",
"value": "Andrew Proud_sentence_8"
},
{
"operation": "finish",
"time": "98.968",
"value": "finish"
}
]
}
|
exfever_4139
|
Geraldine Leigh Chaplin is the daughter of an English comic actor, filmmaker, and composer who rose to fame in the era of silent film.
|
{
"evidence": [
"Geraldine Chaplin",
"Charlie Chaplin"
],
"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:\nGeraldine Leigh Chaplin is the daughter of an English comic actor, filmmaker, and composer who rose to fame in the era of silent film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Geraldine Chaplin",
"Charlie Chaplin"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6106,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
37203
|
John Lithgow called Souls in Metal a "jalopy of a book," and considered the editor borderline illiterate.
|
{
"evidence": [
"Souls in Metal_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:\nJohn Lithgow called Souls in Metal a \"jalopy of a book,\" and considered the editor borderline illiterate.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Souls in Metal_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6107,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.686",
"value": "?search="
},
{
"operation": "search",
"time": "18.016",
"value": "Souls in Metal"
},
{
"operation": "Now on",
"time": "18.721",
"value": "Souls in Metal"
},
{
"operation": "Highlighting",
"time": "37.315",
"value": "Souls in Metal_sentence_6"
},
{
"operation": "finish",
"time": "45.503",
"value": "finish"
}
]
}
|
exfever_2585
|
Love Affair 1994 was the last movie by Katharine Hepburn.
|
{
"evidence": [
"Katharine Hepburn",
"Connecticut",
"New York (state)"
],
"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:\nLove Affair 1994 was the last movie by Katharine Hepburn.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Katharine Hepburn",
"Connecticut",
"New York (state)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6108,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1796
|
Stranger Things is an American television series in a genre of science fiction intended to terrify, surprise, or disgust the audience.
|
{
"evidence": [
"Stranger Things",
"Horror fiction"
],
"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:\nStranger Things is an American television series in a genre of science fiction intended to terrify, surprise, or disgust the audience.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stranger Things",
"Horror fiction"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6109,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5830
|
Black Mirror is a British television series created by Charlie Brooker, a film series that spans different genres and presents a different story and a different set of characters in each episode, season, segment, or short.
|
{
"evidence": [
"Black Mirror",
"Anthology series"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBlack Mirror is a British television series created by Charlie Brooker, a film series that spans different genres and presents a different story and a different set of characters in each episode, season, segment, or short.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Black Mirror",
"Anthology series"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6110,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
17457
|
The RLVM is primary regulator of the sympathetic nervous system and is notably involved in the baroreflex and It receives inhibitory GABAergic input from the caudal ventrolateral medulla (CVLM).
|
{
"evidence": [
"Rostral ventrolateral medulla_sentence_4",
"Rostral ventrolateral medulla_sentence_2",
"Rostral ventrolateral medulla_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe RLVM is primary regulator of the sympathetic nervous system and is notably involved in the baroreflex and It receives inhibitory GABAergic input from the caudal ventrolateral medulla (CVLM).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Rostral ventrolateral medulla_sentence_4",
"Rostral ventrolateral medulla_sentence_2",
"Rostral ventrolateral medulla_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6111,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.973",
"value": "?search="
},
{
"operation": "search",
"time": "11.926",
"value": "RVLM"
},
{
"operation": "Now on",
"time": "13.049",
"value": "Rostral ventrolateral medulla"
},
{
"operation": "Highlighting",
"time": "20.129",
"value": "Rostral ventrolateral medulla_sentence_0"
},
{
"operation": "Highlighting",
"time": "29.68",
"value": "Rostral ventrolateral medulla_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "43.304",
"value": "Rostral ventrolateral medulla_sentence_0"
},
{
"operation": "Highlighting",
"time": "45.88",
"value": "Rostral ventrolateral medulla_sentence_2"
},
{
"operation": "Highlighting",
"time": "50.841",
"value": "Rostral ventrolateral medulla_sentence_3"
},
{
"operation": "finish",
"time": "90.655",
"value": "finish"
}
]
}
|
33189
|
Boops boops, a species of seabream native neither to the eastern Atlantic nor the eastern Antarctic and commercially fished, with 37,830 tonnes taken in 2008, has variously been described as a rudimentary intersex organism.
|
{
"evidence": [
"Boops boops_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:\nBoops boops, a species of seabream native neither to the eastern Atlantic nor the eastern Antarctic and commercially fished, with 37,830 tonnes taken in 2008, has variously been described as a rudimentary intersex organism.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Boops boops_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6112,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.19",
"value": "?search="
},
{
"operation": "search",
"time": "15.344",
"value": "Boops boops"
},
{
"operation": "Now on",
"time": "16.682",
"value": "Boops boops"
},
{
"operation": "Highlighting",
"time": "45.949",
"value": "Boops boops_sentence_0"
},
{
"operation": "Page search",
"time": "70.338",
"value": "Antarctic"
},
{
"operation": "finish",
"time": "136.116",
"value": "finish"
}
]
}
|
exfever_3885
|
War of the Worlds is Steven Spielberg's most successful directed film.
|
{
"evidence": [
"War of the Worlds (2005 film)",
"H. G. Wells"
],
"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:\nWar of the Worlds is Steven Spielberg's most successful directed film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"War of the Worlds (2005 film)",
"H. G. Wells"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6113,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6680
|
In Dubious Battle is a 2016 drama film starring an American actor and singer who gained recognition for composing the music for Back Street Boys.
|
{
"evidence": [
"In Dubious Battle (film)",
"Nat Wolff"
],
"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:\nIn Dubious Battle is a 2016 drama film starring an American actor and singer who gained recognition for composing the music for Back Street Boys.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"In Dubious Battle (film)",
"Nat Wolff"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6114,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
55619
|
A419 road is a primary route between Chiseldon and is managed and maintained by a private company.
|
{
"evidence": [
"A419 road_sentence_0",
"A419 road_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nA419 road is a primary route between Chiseldon and is managed and maintained by a private company.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"A419 road_sentence_0",
"A419 road_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6115,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.764",
"value": "?search="
},
{
"operation": "search",
"time": "3.378",
"value": "A419"
},
{
"operation": "Now on",
"time": "5.217",
"value": "A419 road"
},
{
"operation": "Highlighting",
"time": "10.515",
"value": "A419 road_sentence_0"
},
{
"operation": "Highlighting",
"time": "14.64",
"value": "A419 road_sentence_1"
},
{
"operation": "finish",
"time": "20.209",
"value": "finish"
}
]
}
|
exfever_5766
|
The Shape of Water is a 2017 American romantic fantasy film directed by a Canadian film director best known for his Academy Award-winning fantasy films.
|
{
"evidence": [
"The Shape of Water (film)",
"Guillermo del Toro"
],
"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 Shape of Water is a 2017 American romantic fantasy film directed by a Canadian film director best known for his Academy Award-winning fantasy films.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Shape of Water (film)",
"Guillermo del Toro"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6116,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
60626
|
Vicente Fernández is Caucasian.
|
{
"evidence": [
"Vicente Fernández_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:\nVicente Fernández is Caucasian.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vicente Fernández_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6117,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.299",
"value": "?search="
},
{
"operation": "search",
"time": "8.742",
"value": "Vicente Fernández"
},
{
"operation": "Now on",
"time": "10.116",
"value": "Vicente Fernández"
},
{
"operation": "Highlighting",
"time": "33.11",
"value": "Vicente Fernández_sentence_3"
},
{
"operation": "Page search",
"time": "97.704",
"value": "caucasian"
},
{
"operation": "finish",
"time": "120.574",
"value": "finish"
}
]
}
|
71890
|
Aphanisma is a beach-dwelling plant native to the coastline of Baja California and southern California, and is a succulent saline-adapted plant.
|
{
"evidence": [
"Aphanisma_sentence_1",
"Aphanisma_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:\nAphanisma is a beach-dwelling plant native to the coastline of Baja California and southern California, and is a succulent saline-adapted plant.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Aphanisma_sentence_1",
"Aphanisma_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6118,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "18.299",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "266.222",
"value": "Aphanisma"
},
{
"operation": "Now on",
"time": "267.055",
"value": "Aphanisma"
},
{
"operation": "Highlighting",
"time": "269.396",
"value": "Aphanisma_sentence_1"
},
{
"operation": "Highlighting",
"time": "278.493",
"value": "Aphanisma_sentence_2"
},
{
"operation": "finish",
"time": "487.768",
"value": "finish"
}
]
}
|
8526
|
Cycle of poverty has a lot of disadvantages such as when poor people do not have the resources they need to get out of poverty; therefore, increasing the chance for their poverty and making it impossible to break the cycle.
|
{
"evidence": [
"Cycle of poverty_sentence_4",
"Cycle of poverty_sentence_5",
"Cycle of poverty_sentence_7",
"Cycle of poverty_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:\nCycle of poverty has a lot of disadvantages such as when poor people do not have the resources they need to get out of poverty; therefore, increasing the chance for their poverty and making it impossible to break the cycle.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cycle of poverty_sentence_4",
"Cycle of poverty_sentence_5",
"Cycle of poverty_sentence_7",
"Cycle of poverty_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6119,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.68",
"value": "?search="
},
{
"operation": "search",
"time": "12.661",
"value": "Cycle of poverty"
},
{
"operation": "Now on",
"time": "13.979",
"value": "Cycle of poverty"
},
{
"operation": "Page search",
"time": "23.365",
"value": "disadvantage"
},
{
"operation": "Highlighting",
"time": "25.225",
"value": "Cycle of poverty_sentence_4"
},
{
"operation": "Highlighting",
"time": "28.891",
"value": "Cycle of poverty_sentence_5"
},
{
"operation": "Highlighting",
"time": "34.981",
"value": "Cycle of poverty_sentence_7"
},
{
"operation": "Highlighting",
"time": "67.395",
"value": "Cycle of poverty_sentence_8"
},
{
"operation": "finish",
"time": "90.143",
"value": "finish"
}
]
}
|
exfever_2022
|
Richard Branson is a British billionaire, an entrepreneur, and a person who has achieved immense wealth through the ownership of multiple lines of enterprise.
|
{
"evidence": [
"Richard Branson",
"Business magnate"
],
"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:\nRichard Branson is a British billionaire, an entrepreneur, and a person who has achieved immense wealth through the ownership of multiple lines of enterprise.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Richard Branson",
"Business magnate"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6120,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3325
|
Paul McCartney gained worldwide fame as co-lead vocalist, co-songwriter and bassist for an English rock band.
|
{
"evidence": [
"Paul McCartney",
"The Beatles"
],
"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 McCartney gained worldwide fame as co-lead vocalist, co-songwriter and bassist for an English rock band.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul McCartney",
"The Beatles"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6121,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6680
|
In Dubious Battle is a 2016 drama film starring an American actor and singer who gained recognition for composing the music for Back Street Boys.
|
{
"evidence": [
"In Dubious Battle (film)",
"Nat Wolff"
],
"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:\nIn Dubious Battle is a 2016 drama film starring an American actor and singer who gained recognition for composing the music for Back Street Boys.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"In Dubious Battle (film)",
"Nat Wolff"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6122,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5056
|
Pirates of the Caribbean is a series of fantasy swashbuckler films produced by Jerry Bruckheimer and based on Walt Disney's dark ride tells the story of a band of pirates at a theme park found at Disneyland Canada in Marne-la-Vallée, France.
|
{
"evidence": [
"Pirates of the Caribbean (film series)",
"Pirates of the Caribbean (attraction)",
"Disneyland Park (Paris)"
],
"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:\nPirates of the Caribbean is a series of fantasy swashbuckler films produced by Jerry Bruckheimer and based on Walt Disney's dark ride tells the story of a band of pirates at a theme park found at Disneyland Canada in Marne-la-Vallée, France.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pirates of the Caribbean (film series)",
"Pirates of the Caribbean (attraction)",
"Disneyland Park (Paris)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6123,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1168
|
Doug Liman shot a commercial for Nike.
|
{
"evidence": [
"Edge of Tomorrow",
"Doug Liman"
],
"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:\nDoug Liman shot a commercial for Nike.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Edge of Tomorrow",
"Doug Liman"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6124,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
16541
|
Five tucker-class of destroyers, the fourth of five classes of destroyers that were known as the "thousand tonners", were commissioned for the United States Navy in 1916.
|
{
"evidence": [
"Tucker-class destroyer_sentence_0",
"Tucker-class destroyer_sentence_5",
"Tucker-class destroyer_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:\nFive tucker-class of destroyers, the fourth of five classes of destroyers that were known as the \"thousand tonners\", were commissioned for the United States Navy in 1916.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Tucker-class destroyer_sentence_0",
"Tucker-class destroyer_sentence_5",
"Tucker-class destroyer_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6125,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.109",
"value": "?search="
},
{
"operation": "search",
"time": "91.317",
"value": "Tucker-class destroyer"
},
{
"operation": "Now on",
"time": "92.256",
"value": "Tucker-class destroyer"
},
{
"operation": "Highlighting",
"time": "94.68",
"value": "Tucker-class destroyer_sentence_0"
},
{
"operation": "Highlighting",
"time": "116.862",
"value": "Tucker-class destroyer_sentence_5"
},
{
"operation": "Highlighting",
"time": "120.44",
"value": "Tucker-class destroyer_sentence_1"
},
{
"operation": "finish",
"time": "129.505",
"value": "finish"
}
]
}
|
exfever_1160
|
Ireland is a country in north-western Asia whose capital is its largest city, which is in the province that comprises the ancient kingdoms of Meath, Leinster, and Osraige.
|
{
"evidence": [
"Republic of Ireland",
"Dublin",
"Leinster"
],
"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:\nIreland is a country in north-western Asia whose capital is its largest city, which is in the province that comprises the ancient kingdoms of Meath, Leinster, and Osraige.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Republic of Ireland",
"Dublin",
"Leinster"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6126,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
33020
|
Gotenc is in the Kostel municipality, the municipality seat being the village of Vas, of Russia.
|
{
"evidence": [
"Municipality of Kostel_sentence_0",
"Municipality of Kostel_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:\nGotenc is in the Kostel municipality, the municipality seat being the village of Vas, of Russia.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Municipality of Kostel_sentence_0",
"Municipality of Kostel_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6127,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.661",
"value": "undefined"
},
{
"operation": "search",
"time": "17.811",
"value": "Gotenc"
},
{
"operation": "Now on",
"time": "18.687",
"value": "Gotenc"
},
{
"operation": "Highlighting",
"time": "23.09",
"value": "Gotenc_sentence_0"
},
{
"operation": "Highlighting",
"time": "23.092",
"value": "Gotenc_title"
},
{
"operation": "hyperlink",
"time": "48.345",
"value": "Municipality of Kostel"
},
{
"operation": "Now on",
"time": "48.817",
"value": "Municipality of Kostel"
},
{
"operation": "Page search",
"time": "57.877",
"value": "seat"
},
{
"operation": "Highlighting",
"time": "62.942",
"value": "Municipality of Kostel_sentence_3"
},
{
"operation": "Highlighting",
"time": "62.945",
"value": "Municipality of Kostel_section_0"
},
{
"operation": "Page search",
"time": "93.048",
"value": "vas"
},
{
"operation": "Highlighting deleted",
"time": "98.216",
"value": "Municipality of Kostel_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "98.22",
"value": "Municipality of Kostel_section_0"
},
{
"operation": "Highlighting",
"time": "101.698",
"value": "Municipality of Kostel_sentence_1"
},
{
"operation": "hyperlink",
"time": "105.739",
"value": "Vas"
},
{
"operation": "Now on",
"time": "106.244",
"value": "Vas, Kostel"
},
{
"operation": "Page search",
"time": "116.273",
"value": "russia"
},
{
"operation": "Highlighting",
"time": "123.314",
"value": "Vas, Kostel_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "123.315",
"value": "Vas, Kostel_title"
},
{
"operation": "Highlighting",
"time": "125.811",
"value": "Vas, Kostel_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "125.821",
"value": "Vas, Kostel_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "125.821",
"value": "Vas, Kostel_header_cell_0_0_0"
},
{
"operation": "back-button-clicked",
"time": "186.848",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "187.055",
"value": "Municipality of Kostel"
},
{
"operation": "Highlighting",
"time": "209.596",
"value": "Municipality of Kostel_sentence_0"
},
{
"operation": "Highlighting",
"time": "210.899",
"value": "Municipality of Kostel_sentence_1"
},
{
"operation": "finish",
"time": "263.844",
"value": "finish"
}
]
}
|
exfever_6739
|
Sir Timothy Miles Bindon Rice is an English lyricist and author who is best known for his collaborations with an American composer and impresario of musical theatre.
|
{
"evidence": [
"Tim Rice",
"Andrew Lloyd Webber"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSir Timothy Miles Bindon Rice is an English lyricist and author who is best known for his collaborations with an American composer and impresario of musical theatre.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tim Rice",
"Andrew Lloyd Webber"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6128,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2349
|
Abby Beethoven Sciuto is a fictional character from the NCIS television series, where in season 10, a young Abby was shown in a flashback portrayed by an American actress who is best known as Lizzie in AMC's The Walking Dead.
|
{
"evidence": [
"Abby Sciuto",
"Brighton Sharbino"
],
"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:\nAbby Beethoven Sciuto is a fictional character from the NCIS television series, where in season 10, a young Abby was shown in a flashback portrayed by an American actress who is best known as Lizzie in AMC's The Walking Dead.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Abby Sciuto",
"Brighton Sharbino"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6129,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3640
|
Antonio Lucio Vivaldi was an Italian who organized and often financed concerts, plays, or operas, similar to a film or television producer in the stage arts.
|
{
"evidence": [
"Antonio Vivaldi",
"Impresario"
],
"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:\nAntonio Lucio Vivaldi was an Italian who organized and often financed concerts, plays, or operas, similar to a film or television producer in the stage arts.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Antonio Vivaldi",
"Impresario"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6130,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
57880
|
De Veaux Woods State Park has baseball diamonds and a playground, a small area of desert is present inside and the park is open year-round.
|
{
"evidence": [
"De Veaux Woods State Park_sentence_8",
"De Veaux Woods State Park_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:\nDe Veaux Woods State Park has baseball diamonds and a playground, a small area of desert is present inside and the park is open year-round.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"De Veaux Woods State Park_sentence_8",
"De Veaux Woods State Park_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6131,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.052",
"value": "?search="
},
{
"operation": "search",
"time": "11.023",
"value": "De Veaux Woods State Park"
},
{
"operation": "Now on",
"time": "11.777",
"value": "De Veaux Woods State Park"
},
{
"operation": "Highlighting",
"time": "30.791",
"value": "De Veaux Woods State Park_sentence_8"
},
{
"operation": "Highlighting",
"time": "41.182",
"value": "De Veaux Woods State Park_sentence_9"
},
{
"operation": "finish",
"time": "50.235",
"value": "finish"
}
]
}
|
exfever_1436
|
Christopher Marlowe is the oldest translator during the reign of Queen Elizabeth I.
|
{
"evidence": [
"Christopher Marlowe",
"Elizabethan era"
],
"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:\nChristopher Marlowe is the oldest translator during the reign of Queen Elizabeth I.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Christopher Marlowe",
"Elizabethan era"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6132,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
69988
|
Genetic linkage can be explained by the experiment by Bateson, Saunders, and Punnett, a British geneticist.
|
{
"evidence": [
"Genetic linkage_sentence_14",
"Genetic linkage_sentence_15",
"Genetic linkage_sentence_16",
"Genetic linkage_sentence_17"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nGenetic linkage can be explained by the experiment by Bateson, Saunders, and Punnett, a British geneticist.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Genetic linkage_sentence_14",
"Genetic linkage_sentence_15",
"Genetic linkage_sentence_16",
"Genetic linkage_sentence_17"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6133,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.881",
"value": "?search="
},
{
"operation": "search",
"time": "62.899",
"value": "Genetic linkage"
},
{
"operation": "Now on",
"time": "63.933",
"value": "Genetic linkage"
},
{
"operation": "Page search",
"time": "153.086",
"value": "Bateson"
},
{
"operation": "Highlighting",
"time": "318.706",
"value": "Genetic linkage_sentence_14"
},
{
"operation": "Highlighting",
"time": "325.838",
"value": "Genetic linkage_sentence_15"
},
{
"operation": "Highlighting",
"time": "328.953",
"value": "Genetic linkage_sentence_16"
},
{
"operation": "Highlighting",
"time": "332.913",
"value": "Genetic linkage_sentence_17"
},
{
"operation": "Highlighting deleted",
"time": "335.893",
"value": "Genetic linkage_sentence_17"
},
{
"operation": "Highlighting",
"time": "336.39",
"value": "Genetic linkage_sentence_17"
},
{
"operation": "finish",
"time": "374.604",
"value": "finish"
}
]
}
|
9793
|
The businesses, communities, forest and natural resources (primary resources) and our personal lives do not need a sustainable management.
|
{
"evidence": [
"Sustainable management_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:\nThe businesses, communities, forest and natural resources (primary resources) and our personal lives do not need a sustainable management.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sustainable management_sentence_8"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6134,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.488",
"value": "?search="
},
{
"operation": "search",
"time": "32.017",
"value": "Sustainable management"
},
{
"operation": "Now on",
"time": "32.872",
"value": "Sustainable management"
},
{
"operation": "Page search",
"time": "39.411",
"value": "forest"
},
{
"operation": "Highlighting",
"time": "66.884",
"value": "Sustainable management_sentence_8"
},
{
"operation": "finish",
"time": "226.038",
"value": "finish"
}
]
}
|
71814
|
Introduced in 1970, Triumph 1500, a small front-wheel drive car that was produced by Standard-Triumph from 1970 to 1973, is the first car engine of any brand to have a capacity of 1493 cc.
|
{
"evidence": [
"Triumph 1500_sentence_0",
"Triumph 1500_sentence_11",
"Triumph 1500_sentence_12"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIntroduced in 1970, Triumph 1500, a small front-wheel drive car that was produced by Standard-Triumph from 1970 to 1973, is the first car engine of any brand to have a capacity of 1493 cc.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Triumph 1500_sentence_0",
"Triumph 1500_sentence_11",
"Triumph 1500_sentence_12"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6135,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.359",
"value": "?search="
},
{
"operation": "search",
"time": "9.989",
"value": "Triumph 1500"
},
{
"operation": "Now on",
"time": "11.199",
"value": "Triumph 1500"
},
{
"operation": "Highlighting",
"time": "14.373",
"value": "Triumph 1500_sentence_0"
},
{
"operation": "Highlighting",
"time": "60.587",
"value": "Triumph 1500_sentence_11"
},
{
"operation": "Highlighting",
"time": "61.782",
"value": "Triumph 1500_sentence_12"
},
{
"operation": "finish",
"time": "94.557",
"value": "finish"
}
]
}
|
exfever_8294
|
Lil Wayne and B.G. joined the hip hop group "Hot Boys" with Turk and an American rapper best known for his work with Cash Money Records.
|
{
"evidence": [
"Lil Wayne",
"Juvenile (rapper)"
],
"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:\nLil Wayne and B.G. joined the hip hop group \"Hot Boys\" with Turk and an American rapper best known for his work with Cash Money Records.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lil Wayne",
"Juvenile (rapper)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6136,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
60156
|
Fred Swanton (1862–1940) was an entrepreneur and real estate developer who served as mayor of Santa Cruz, California from 1927 until 1933, Swanton promoted the expansion of Santa Cruz as a beach resort city and In 1900 he invested in the Klondike Gold Rush, but by then the profits had been made.
|
{
"evidence": [
"Fred Swanton_sentence_0",
"Fred Swanton_sentence_1",
"Fred Swanton_sentence_15"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFred Swanton (1862–1940) was an entrepreneur and real estate developer who served as mayor of Santa Cruz, California from 1927 until 1933, Swanton promoted the expansion of Santa Cruz as a beach resort city and In 1900 he invested in the Klondike Gold Rush, but by then the profits had been made.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Fred Swanton_sentence_0",
"Fred Swanton_sentence_1",
"Fred Swanton_sentence_15"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6137,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.739",
"value": "?search="
},
{
"operation": "search",
"time": "6.571",
"value": "Fred Swanton"
},
{
"operation": "Now on",
"time": "7.827",
"value": "Fred Swanton"
},
{
"operation": "Highlighting",
"time": "16.345",
"value": "Fred Swanton_sentence_0"
},
{
"operation": "Highlighting",
"time": "21.894",
"value": "Fred Swanton_sentence_1"
},
{
"operation": "Highlighting",
"time": "38.342",
"value": "Fred Swanton_sentence_15"
},
{
"operation": "finish",
"time": "44.229",
"value": "finish"
}
]
}
|
exfever_6334
|
Star Search is an Asian television show hosted also by a host on the ABC game show, which was aired from September 30, 1957, to November 15, 1957.
|
{
"evidence": [
"Star Search",
"Ed McMahon",
"Who Do You Trust?"
],
"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:\nStar Search is an Asian television show hosted also by a host on the ABC game show, which was aired from September 30, 1957, to November 15, 1957.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Star Search",
"Ed McMahon",
"Who Do You Trust?"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6138,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8972
|
Logic is an American rapper who released his debut album on October 21, 2014 via Def Jam Recordings.
|
{
"evidence": [
"Logic (musician)",
"Under Pressure (album)"
],
"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:\nLogic is an American rapper who released his debut album on October 21, 2014 via Def Jam Recordings.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Logic (musician)",
"Under Pressure (album)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6139,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6828
|
The 2002 Royal Rumble was the 14th annual Royal Rumble professional wrestling event produced by the World Wrestling Federation that was held in the capital and most populous city of the U.S. state of Georgia.
|
{
"evidence": [
"Royal Rumble (2002)",
"Atlanta"
],
"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 2002 Royal Rumble was the 14th annual Royal Rumble professional wrestling event produced by the World Wrestling Federation that was held in the capital and most populous city of the U.S. state of Georgia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Royal Rumble (2002)",
"Atlanta"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6140,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2741
|
Con Man is an Australian comedy web series that was acquired by an American basic cable channel owned by the NBCUniversal Television and Streaming division of Comcast through NBCUniversal Cable Entertainment.
|
{
"evidence": [
"Con Man (web series)",
"Syfy"
],
"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:\nCon Man is an Australian comedy web series that was acquired by an American basic cable channel owned by the NBCUniversal Television and Streaming division of Comcast through NBCUniversal Cable Entertainment.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Con Man (web series)",
"Syfy"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6141,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3294
|
Deng Xiaoping was a Chinese communist revolutionary who was the founder of the People's Republic of China and developer of Mao Zedong Thought, a branch of Marxism–Leninism developed to bring about a socialist revolution in the agricultural, pre-industrial society of the Republic of China.
|
{
"evidence": [
"Mao Zedong",
"Maoism"
],
"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:\nDeng Xiaoping was a Chinese communist revolutionary who was the founder of the People's Republic of China and developer of Mao Zedong Thought, a branch of Marxism–Leninism developed to bring about a socialist revolution in the agricultural, pre-industrial society of the Republic of China.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mao Zedong",
"Maoism"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6142,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
33324
|
COPS6 gene that has been shown to interact with EIF3S6, encodes COP9 signalosome protein that functions as a positive regulator of E3 ubiquitin ligases.
|
{
"evidence": [
"COPS6_sentence_6",
"COPS6_sentence_1",
"COPS6_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:\nCOPS6 gene that has been shown to interact with EIF3S6, encodes COP9 signalosome protein that functions as a positive regulator of E3 ubiquitin ligases.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"COPS6_sentence_6",
"COPS6_sentence_1",
"COPS6_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6143,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.866",
"value": "?search="
},
{
"operation": "search",
"time": "9.195",
"value": "COPS6 gene"
},
{
"operation": "Now on",
"time": "11.217",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "12.793",
"value": "COPS6"
},
{
"operation": "Now on",
"time": "13.23",
"value": "COPS6"
},
{
"operation": "Highlighting",
"time": "16.874",
"value": "COPS6_sentence_0"
},
{
"operation": "Highlighting",
"time": "23.195",
"value": "COPS6_sentence_6"
},
{
"operation": "Highlighting",
"time": "43.041",
"value": "COPS6_sentence_2"
},
{
"operation": "Highlighting",
"time": "115.167",
"value": "COPS6_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "120.158",
"value": "COPS6_sentence_2"
},
{
"operation": "Highlighting",
"time": "156.019",
"value": "COPS6_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "184.362",
"value": "COPS6_sentence_0"
},
{
"operation": "finish",
"time": "209.06",
"value": "finish"
}
]
}
|
exfever_2112
|
Chantelle Whitney Brown-Young is a Jamaican Indonesian fashion model and public spokesperson on a long-term skin condition called leukoderma.
|
{
"evidence": [
"Winnie Harlow",
"Vitiligo"
],
"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:\nChantelle Whitney Brown-Young is a Jamaican Indonesian fashion model and public spokesperson on a long-term skin condition called leukoderma.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Winnie Harlow",
"Vitiligo"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6144,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
64157
|
PA2G4 is a gene that encodes RNA-binding protein that has nothing to do with growth regulation and is present in complexes.
|
{
"evidence": [
"PA2G4_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:\nPA2G4 is a gene that encodes RNA-binding protein that has nothing to do with growth regulation and is present in complexes.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"PA2G4_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6145,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.607",
"value": "?search="
},
{
"operation": "search",
"time": "8.258",
"value": "PA2G4"
},
{
"operation": "Now on",
"time": "9.052",
"value": "PA2G4"
},
{
"operation": "Highlighting",
"time": "21.848",
"value": "PA2G4_sentence_1"
},
{
"operation": "finish",
"time": "33.553",
"value": "finish"
}
]
}
|
exfever_2439
|
Wars of the Roses is a fight between supporters of two rival cadet branches of the royal House of Plantagenet: Lancaster and a cadet branch of the English royal House of Plantagenet that has five of its members became kings of England in the late 15th century
|
{
"evidence": [
"Wars of the Roses",
"House of York"
],
"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:\nWars of the Roses is a fight between supporters of two rival cadet branches of the royal House of Plantagenet: Lancaster and a cadet branch of the English royal House of Plantagenet that has five of its members became kings of England in the late 15th century\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wars of the Roses",
"House of York"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6146,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1343
|
Tommy Lee Jones is an American actor most notable for his role in the Men in Black film series as one of the two main protagonists of the Men in Black franchise that was voiced in the animated series by an American actor best known for using his voice in different animated series.
|
{
"evidence": [
"Tommy Lee Jones",
"Agent K",
"Gregg Berger"
],
"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:\nTommy Lee Jones is an American actor most notable for his role in the Men in Black film series as one of the two main protagonists of the Men in Black franchise that was voiced in the animated series by an American actor best known for using his voice in different animated series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tommy Lee Jones",
"Agent K",
"Gregg Berger"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6147,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1245
|
The Midwestern United States is situated in between the regions comprising the westernmost states of the United Kingdom.
|
{
"evidence": [
"Midwestern United States",
"Western United States"
],
"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 Midwestern United States is situated in between the regions comprising the westernmost states of the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Midwestern United States",
"Western United States"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6148,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
44372
|
Adam Gaudette is an American professional ice hockey player currently playing for the Vancouver Canucks and previously played with the Northeastern University Huskies of the NCAA.
|
{
"evidence": [
"Adam Gaudette_sentence_0",
"Adam Gaudette_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAdam Gaudette is an American professional ice hockey player currently playing for the Vancouver Canucks and previously played with the Northeastern University Huskies of the NCAA.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Adam Gaudette_sentence_0",
"Adam Gaudette_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6149,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.61",
"value": "?search="
},
{
"operation": "search",
"time": "9.319",
"value": "Adam Gaudette"
},
{
"operation": "Now on",
"time": "10.373",
"value": "Adam Gaudette"
},
{
"operation": "Highlighting",
"time": "13.215",
"value": "Adam Gaudette_sentence_0"
},
{
"operation": "Highlighting",
"time": "26.755",
"value": "Adam Gaudette_sentence_1"
},
{
"operation": "finish",
"time": "39.183",
"value": "finish"
}
]
}
|
58310
|
Ly Erg is a fairy in Scottish folklore associated with the Glenmore Forest (a remnant of the Caledonian Forest) who is dressed as a soldier.
|
{
"evidence": [
"Ly Erg_sentence_0",
"Glenmore Forest Park_sentence_0",
"Ly Erg_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLy Erg is a fairy in Scottish folklore associated with the Glenmore Forest (a remnant of the Caledonian Forest) who is dressed as a soldier.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Ly Erg_sentence_0",
"Glenmore Forest Park_sentence_0",
"Ly Erg_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6150,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.065",
"value": "?search="
},
{
"operation": "search",
"time": "14.085",
"value": "Ly Erg"
},
{
"operation": "Now on",
"time": "14.728",
"value": "Ly Erg"
},
{
"operation": "Highlighting",
"time": "17.779",
"value": "Ly Erg_sentence_0"
},
{
"operation": "hyperlink",
"time": "23.746",
"value": "Glenmore Forest"
},
{
"operation": "Now on",
"time": "24.256",
"value": "Glenmore Forest Park"
},
{
"operation": "Highlighting",
"time": "26.736",
"value": "Glenmore Forest Park_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "30.583",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "30.708",
"value": "Ly Erg"
},
{
"operation": "Highlighting",
"time": "36.196",
"value": "Ly Erg_sentence_1"
},
{
"operation": "finish",
"time": "188.665",
"value": "finish"
}
]
}
|
exfever_7539
|
Selena Marie Gomez is a Filipino singer and actress who appeared in a 2009 American made-for-television comedy-drama film directed by Lev L. Spiro that also classified as a film genre with fantastic themes, usually magic, supernatural events, mythology, folklore, or exotic fantasy worlds.
|
{
"evidence": [
"Selena Gomez",
"Wizards of Waverly Place: The Movie",
"Fantasy 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:\nSelena Marie Gomez is a Filipino singer and actress who appeared in a 2009 American made-for-television comedy-drama film directed by Lev L. Spiro that also classified as a film genre with fantastic themes, usually magic, supernatural events, mythology, folklore, or exotic fantasy worlds.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Selena Gomez",
"Wizards of Waverly Place: The Movie",
"Fantasy film"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6151,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
34104
|
The Kulm Cross initiated by Frederick William III of Prussia on December 1813 and worn on silver and NCOs was awarded to 12,066 people, with only 7,131 surviving soldiers to 1816 able to obtain the reward.
|
{
"evidence": [
"Kulm Cross_sentence_2",
"Kulm Cross_sentence_4",
"Kulm Cross_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:\nThe Kulm Cross initiated by Frederick William III of Prussia on December 1813 and worn on silver and NCOs was awarded to 12,066 people, with only 7,131 surviving soldiers to 1816 able to obtain the reward.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Kulm Cross_sentence_2",
"Kulm Cross_sentence_4",
"Kulm Cross_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6152,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.811",
"value": "?search="
},
{
"operation": "search",
"time": "14.871",
"value": "Kulm Cross"
},
{
"operation": "Now on",
"time": "15.783",
"value": "Kulm Cross"
},
{
"operation": "Highlighting",
"time": "22.839",
"value": "Kulm Cross_sentence_2"
},
{
"operation": "Highlighting",
"time": "41.845",
"value": "Kulm Cross_sentence_4"
},
{
"operation": "Highlighting",
"time": "55.921",
"value": "Kulm Cross_sentence_7"
},
{
"operation": "finish",
"time": "110.756",
"value": "finish"
}
]
}
|
85285
|
Kerstin Kaiser (born 16 July 1960 in Stralsund) is a German politician for the left wing party, Die Linke who worked as a teacher at the party school of the Socialist Unity Party of Germany and became an informer for the Stasi under the code name "Kathrin".
|
{
"evidence": [
"Kerstin Kaiser_sentence_0",
"Kerstin Kaiser_sentence_1",
"Kerstin Kaiser_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:\nKerstin Kaiser (born 16 July 1960 in Stralsund) is a German politician for the left wing party, Die Linke who worked as a teacher at the party school of the Socialist Unity Party of Germany and became an informer for the Stasi under the code name \"Kathrin\".\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Kerstin Kaiser_sentence_0",
"Kerstin Kaiser_sentence_1",
"Kerstin Kaiser_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6153,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.719",
"value": "?search="
},
{
"operation": "search",
"time": "8.817",
"value": "Kerstin Kaiser"
},
{
"operation": "Now on",
"time": "9.706",
"value": "Kerstin Kaiser"
},
{
"operation": "Highlighting",
"time": "15.33",
"value": "Kerstin Kaiser_sentence_0"
},
{
"operation": "Highlighting",
"time": "62.361",
"value": "Kerstin Kaiser_sentence_1"
},
{
"operation": "Highlighting",
"time": "64.093",
"value": "Kerstin Kaiser_sentence_2"
},
{
"operation": "finish",
"time": "76.009",
"value": "finish"
}
]
}
|
exfever_3939
|
Breaking Bad is an American neo-Western crime drama television series created and produced by Vince Gilligan, starring Bryan Cranston as Walter White, an underpaid, overqualified, and dispirited high-school coach who teaches about the scientific study of the properties and behavior of matter.
|
{
"evidence": [
"Breaking Bad",
"Chemistry"
],
"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:\nBreaking Bad is an American neo-Western crime drama television series created and produced by Vince Gilligan, starring Bryan Cranston as Walter White, an underpaid, overqualified, and dispirited high-school coach who teaches about the scientific study of the properties and behavior of matter.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Breaking Bad",
"Chemistry"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6154,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
72927
|
WNVM is a radio station licensed to serve Venezuela, it is owned by New Life Broadcasting and the station is relayed through booster station in Cayey.
|
{
"evidence": [
"WNVM_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:\nWNVM is a radio station licensed to serve Venezuela, it is owned by New Life Broadcasting and the station is relayed through booster station in Cayey.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"WNVM_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6155,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.348",
"value": "?search="
},
{
"operation": "search",
"time": "132.769",
"value": "WNVM"
},
{
"operation": "Now on",
"time": "133.654",
"value": "WNVM"
},
{
"operation": "Highlighting",
"time": "146.683",
"value": "WNVM_sentence_0"
},
{
"operation": "finish",
"time": "185.885",
"value": "finish"
}
]
}
|
exfever_5333
|
"7 Things" is a song by an Asian singer and television personality who is known for her distinctive raspy voice.
|
{
"evidence": [
"7 Things",
"Miley Cyrus"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n\"7 Things\" is a song by an Asian singer and television personality who is known for her distinctive raspy voice.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"7 Things",
"Miley Cyrus"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6156,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
42752
|
Nick Papadimitriou, who designed the Middlesex County Council website in 2007, co-hosted two series of the radio show Ventures and Adventure in Topography on Resonance FM, a London-based non-profit community radio station.
|
{
"evidence": [
"Nick Papadimitriou_sentence_4",
"Nick Papadimitriou_sentence_5",
"Resonance FM_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:\nNick Papadimitriou, who designed the Middlesex County Council website in 2007, co-hosted two series of the radio show Ventures and Adventure in Topography on Resonance FM, a London-based non-profit community radio station.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Nick Papadimitriou_sentence_4",
"Nick Papadimitriou_sentence_5",
"Resonance FM_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6157,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.891",
"value": "?search="
},
{
"operation": "search",
"time": "12.378",
"value": "Nick Papadimitriou"
},
{
"operation": "Now on",
"time": "13.069",
"value": "Nick Papadimitriou"
},
{
"operation": "Highlighting",
"time": "26.909",
"value": "Nick Papadimitriou_sentence_4"
},
{
"operation": "Highlighting",
"time": "49.812",
"value": "Nick Papadimitriou_sentence_5"
},
{
"operation": "hyperlink",
"time": "75.759",
"value": "Resonance FM"
},
{
"operation": "Now on",
"time": "76.317",
"value": "Resonance FM"
},
{
"operation": "Highlighting",
"time": "79.414",
"value": "Resonance FM_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "82.739",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "82.79",
"value": "Nick Papadimitriou"
},
{
"operation": "finish",
"time": "124.357",
"value": "finish"
}
]
}
|
69462
|
George Arbuthnot Scott played cricket.
|
{
"evidence": [
"George Arbuthnot Scott_sentence_0",
"George Arbuthnot Scott_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:\nGeorge Arbuthnot Scott played cricket.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"George Arbuthnot Scott_sentence_0",
"George Arbuthnot Scott_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6158,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.763",
"value": "?search="
},
{
"operation": "search",
"time": "8.829",
"value": "George Arbuthnot Scott"
},
{
"operation": "Now on",
"time": "22.582",
"value": "George Arbuthnot Scott"
},
{
"operation": "Highlighting",
"time": "32.378",
"value": "George Arbuthnot Scott_sentence_0"
},
{
"operation": "Highlighting",
"time": "93.778",
"value": "George Arbuthnot Scott_sentence_3"
},
{
"operation": "finish",
"time": "133.161",
"value": "finish"
}
]
}
|
38428
|
Yoshiaki Numata is a former world Junior Lightweight boxing champion which is considered a combat sport in which two people, usually wearing protective gloves, throw punches at each other for a predetermined amount of time in a boxing ring.
|
{
"evidence": [
"Yoshiaki Numata_sentence_0",
"Boxing_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:\nYoshiaki Numata is a former world Junior Lightweight boxing champion which is considered a combat sport in which two people, usually wearing protective gloves, throw punches at each other for a predetermined amount of time in a boxing ring.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Yoshiaki Numata_sentence_0",
"Boxing_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6159,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.663",
"value": "?search="
},
{
"operation": "search",
"time": "3.682",
"value": "Yoshiaki Numata"
},
{
"operation": "Now on",
"time": "4.545",
"value": "Yoshiaki Numata"
},
{
"operation": "Highlighting",
"time": "14.563",
"value": "Yoshiaki Numata_sentence_0"
},
{
"operation": "hyperlink",
"time": "25.372",
"value": "boxing"
},
{
"operation": "Now on",
"time": "26.551",
"value": "Boxing"
},
{
"operation": "Highlighting",
"time": "72.801",
"value": "Boxing_sentence_2"
},
{
"operation": "finish",
"time": "86.164",
"value": "finish"
}
]
}
|
54569
|
Carl Abel was a German comparative philologist from Berlin who studied Linguistic Essays in 1880 and also acted as Ilchester lecturer on comparative lexicography at the University of Oxford.
|
{
"evidence": [
"Carl Abel_sentence_1",
"Carl Abel_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:\nCarl Abel was a German comparative philologist from Berlin who studied Linguistic Essays in 1880 and also acted as Ilchester lecturer on comparative lexicography at the University of Oxford.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Carl Abel_sentence_1",
"Carl Abel_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6160,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.901",
"value": "?search="
},
{
"operation": "search",
"time": "9.854",
"value": "Carl Abel"
},
{
"operation": "Now on",
"time": "11.088",
"value": "Carl Abel"
},
{
"operation": "Highlighting",
"time": "17.848",
"value": "Carl Abel_sentence_1"
},
{
"operation": "Highlighting",
"time": "30.675",
"value": "Carl Abel_sentence_2"
},
{
"operation": "finish",
"time": "46.106",
"value": "finish"
}
]
}
|
79035
|
Pyotr Kolodin(born 23 September 1930) was born in Novovasilyevka, Soviet Union and graduated from Military Academy of Engineering and Radioengineering in 1950
|
{
"evidence": [
"Pyotr Kolodin_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:\nPyotr Kolodin(born 23 September 1930) was born in Novovasilyevka, Soviet Union and graduated from Military Academy of Engineering and Radioengineering in 1950\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Pyotr Kolodin_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6161,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.088",
"value": "?search="
},
{
"operation": "search",
"time": "9.87",
"value": "Pyotr Kolodin"
},
{
"operation": "Now on",
"time": "10.517",
"value": "Pyotr Kolodin"
},
{
"operation": "Highlighting",
"time": "22.559",
"value": "Pyotr Kolodin_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "24.853",
"value": "Pyotr Kolodin_sentence_0"
},
{
"operation": "Highlighting",
"time": "26.298",
"value": "Pyotr Kolodin_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "27.116",
"value": "Pyotr Kolodin_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "27.118",
"value": "Pyotr Kolodin_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "27.118",
"value": "Pyotr Kolodin_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "62.215",
"value": "Pyotr Kolodin_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "63.714",
"value": "Pyotr Kolodin_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "63.716",
"value": "Pyotr Kolodin_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "63.717",
"value": "Pyotr Kolodin_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "64.791",
"value": "Pyotr Kolodin_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "64.793",
"value": "Pyotr Kolodin_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "64.793",
"value": "Pyotr Kolodin_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "85.542",
"value": "Pyotr Kolodin_cell_0_3_1"
},
{
"operation": "Highlighting deleted",
"time": "85.545",
"value": "Pyotr Kolodin_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "85.545",
"value": "Pyotr Kolodin_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "90.199",
"value": "finish"
}
]
}
|
exfever_2133
|
Cleopatra is the Queen of Egypt and companion of the king of the modern Greek kingdom of Macedon.
|
{
"evidence": [
"Cleopatra",
"Alexander the Great"
],
"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:\nCleopatra is the Queen of Egypt and companion of the king of the modern Greek kingdom of Macedon.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Cleopatra",
"Alexander the Great"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6162,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
95566
|
From January 2017 to June 2017, there were eleven incidents involving cow vigilante violence (the use of physical force in the name of a predominantly African religious and political movement aiming to protect cows), wherein 18 people were killed and 20 people were wounded.
|
{
"evidence": [
"Cow vigilante violence in India_sentence_0",
"Cow protection movement_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:\nFrom January 2017 to June 2017, there were eleven incidents involving cow vigilante violence (the use of physical force in the name of a predominantly African religious and political movement aiming to protect cows), wherein 18 people were killed and 20 people were wounded.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Cow vigilante violence in India_sentence_0",
"Cow protection movement_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6163,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.341",
"value": "?search="
},
{
"operation": "search",
"time": "28.307",
"value": "\"cow vigilante violence\""
},
{
"operation": "Now on",
"time": "29.485",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "40.534",
"value": "Cow vigilante violence in India"
},
{
"operation": "Now on",
"time": "41.858",
"value": "Cow vigilante violence in India"
},
{
"operation": "Highlighting",
"time": "46.523",
"value": "Cow vigilante violence in India_sentence_0"
},
{
"operation": "hyperlink",
"time": "55.698",
"value": "cow protection"
},
{
"operation": "Now on",
"time": "57.282",
"value": "Cow protection movement"
},
{
"operation": "Highlighting",
"time": "60.05",
"value": "Cow protection movement_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "62.019",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "62.207",
"value": "Cow vigilante violence in India"
},
{
"operation": "Highlighting deleted",
"time": "64.476",
"value": "Cow vigilante violence in India_sentence_0"
},
{
"operation": "hyperlink",
"time": "65.195",
"value": "cow protection"
},
{
"operation": "Now on",
"time": "65.559",
"value": "Cow protection movement"
},
{
"operation": "back-button-clicked",
"time": "68.418",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "68.595",
"value": "Cow vigilante violence in India"
},
{
"operation": "Highlighting",
"time": "69.746",
"value": "Cow vigilante violence in India_sentence_0"
},
{
"operation": "hyperlink",
"time": "70.393",
"value": "cow protection"
},
{
"operation": "Now on",
"time": "70.766",
"value": "Cow protection movement"
},
{
"operation": "Highlighting deleted",
"time": "71.725",
"value": "Cow protection movement_sentence_0"
},
{
"operation": "Highlighting",
"time": "71.886",
"value": "Cow protection movement_sentence_0"
},
{
"operation": "finish",
"time": "109.459",
"value": "finish"
}
]
}
|
exfever_6598
|
Dracula is the shortest novel by Bram Stoker.
|
{
"evidence": [
"Dracula",
"Jonathan Harker"
],
"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:\nDracula is the shortest novel by Bram Stoker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dracula",
"Jonathan Harker"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6164,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
77878
|
Islington-City Centre West is a commercial and residential neighbourhood and it exits in Brown's Line.
|
{
"evidence": [
"Islington-City Centre West_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:\nIslington-City Centre West is a commercial and residential neighbourhood and it exits in Brown's Line.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Islington-City Centre West_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6165,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.965",
"value": "?search="
},
{
"operation": "search",
"time": "7.706",
"value": "Islington-City Centre West"
},
{
"operation": "Now on",
"time": "9.209",
"value": "Islington-City Centre West"
},
{
"operation": "Highlighting",
"time": "12.174",
"value": "Islington-City Centre West_sentence_0"
},
{
"operation": "Page search",
"time": "24.202",
"value": "exit"
},
{
"operation": "Page search",
"time": "39.323",
"value": "line"
},
{
"operation": "Highlighting deleted",
"time": "54.079",
"value": "Islington-City Centre West_sentence_0"
},
{
"operation": "Highlighting",
"time": "54.655",
"value": "Islington-City Centre West_sentence_1"
},
{
"operation": "Highlighting",
"time": "58.023",
"value": "Islington-City Centre West_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "58.551",
"value": "Islington-City Centre West_sentence_1"
},
{
"operation": "Page search",
"time": "79.688",
"value": "brown's"
},
{
"operation": "Page search",
"time": "83.421",
"value": "brown"
},
{
"operation": "Highlighting",
"time": "88.465",
"value": "Islington-City Centre West_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "89.762",
"value": "Islington-City Centre West_sentence_0"
},
{
"operation": "finish",
"time": "93",
"value": "finish"
}
]
}
|
36058
|
The main building of Agios Andreas Hospital, which is the second largest hospital in Patras, was struck by an earthquake in 2008 and was restored with additional equipment.
|
{
"evidence": [
"Agios Andreas Hospital_sentence_4"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe main building of Agios Andreas Hospital, which is the second largest hospital in Patras, was struck by an earthquake in 2008 and was restored with additional equipment.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Agios Andreas Hospital_sentence_4"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6166,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.785",
"value": "?search="
},
{
"operation": "search",
"time": "12.671",
"value": "Agios Andreas Hospital"
},
{
"operation": "Now on",
"time": "13.519",
"value": "Agios Andreas Hospital"
},
{
"operation": "Highlighting",
"time": "63.462",
"value": "Agios Andreas Hospital_sentence_4"
},
{
"operation": "finish",
"time": "76.913",
"value": "finish"
}
]
}
|
exfever_3460
|
Kate Nash is an English singer and actress who launched her music career in 2005 and starred in the 2013 British comedy film directed by Rachel Hirons.
|
{
"evidence": [
"Kate Nash",
"Powder Room (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:\nKate Nash is an English singer and actress who launched her music career in 2005 and starred in the 2013 British comedy film directed by Rachel Hirons.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kate Nash",
"Powder Room (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6167,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4382
|
Milla Jovovich debuted in the 1988 TV movie “The Night Train to Kathmandu” with Disney Channel production.
|
{
"evidence": [
"Milla Jovovich",
"London",
"Roman Empire"
],
"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:\nMilla Jovovich debuted in the 1988 TV movie “The Night Train to Kathmandu” with Disney Channel production.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Milla Jovovich",
"London",
"Roman Empire"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6168,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4910
|
Novak Djokovic is a Serbian amateur player in a racket sport that can be played individually against a single opponent or between two teams of two players each.
|
{
"evidence": [
"Novak Djokovic",
"Tennis"
],
"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:\nNovak Djokovic is a Serbian amateur player in a racket sport that can be played individually against a single opponent or between two teams of two players each.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Novak Djokovic",
"Tennis"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6169,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
25040
|
The Turku Coastal Regiment (formed on 10.9.1939 as Turku Sector) was a Finnish coastal artillery unit operating in the Turku area and Archipelago Sea which was later disbanded on 30.6.1998.
|
{
"evidence": [
"Turku Coastal Regiment_sentence_1",
"Turku Coastal Regiment_sentence_0",
"Turku Coastal Regiment_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Turku Coastal Regiment (formed on 10.9.1939 as Turku Sector) was a Finnish coastal artillery unit operating in the Turku area and Archipelago Sea which was later disbanded on 30.6.1998.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Turku Coastal Regiment_sentence_1",
"Turku Coastal Regiment_sentence_0",
"Turku Coastal Regiment_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6170,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.794",
"value": "?search="
},
{
"operation": "search",
"time": "21.826",
"value": "Turku Coastal Regiment"
},
{
"operation": "Now on",
"time": "22.975",
"value": "Turku Coastal Regiment"
},
{
"operation": "Highlighting",
"time": "77.994",
"value": "Turku Coastal Regiment_sentence_1"
},
{
"operation": "Highlighting",
"time": "82.745",
"value": "Turku Coastal Regiment_sentence_0"
},
{
"operation": "Highlighting",
"time": "90.319",
"value": "Turku Coastal Regiment_sentence_3"
},
{
"operation": "finish",
"time": "99.163",
"value": "finish"
}
]
}
|
36395
|
The LKL IV is a two-seater aircraft with the fuel tanks in the fuselage from Poland that had it's last flight in 1932 and was a little longer that the LKL V.
|
{
"evidence": [
"LKL IV_sentence_20"
],
"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 LKL IV is a two-seater aircraft with the fuel tanks in the fuselage from Poland that had it's last flight in 1932 and was a little longer that the LKL V.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"LKL IV_sentence_20"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6171,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.707",
"value": "?search="
},
{
"operation": "search",
"time": "11.525",
"value": "LKL IV"
},
{
"operation": "Now on",
"time": "12.253",
"value": "LKL IV"
},
{
"operation": "Highlighting",
"time": "16.582",
"value": "LKL IV_sentence_0"
},
{
"operation": "Highlighting",
"time": "63.03",
"value": "LKL IV_sentence_20"
},
{
"operation": "Highlighting deleted",
"time": "67.334",
"value": "LKL IV_sentence_0"
},
{
"operation": "finish",
"time": "91.869",
"value": "finish"
}
]
}
|
exfever_8537
|
Breaking Bad is an American neo-Western crime drama television series created and produced by an American writer, producer, and director, also known as the writer and producer of an American science fiction drama television series.
|
{
"evidence": [
"Breaking Bad",
"Vince Gilligan",
"The X-Files"
],
"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:\nBreaking Bad is an American neo-Western crime drama television series created and produced by an American writer, producer, and director, also known as the writer and producer of an American science fiction drama television series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Breaking Bad",
"Vince Gilligan",
"The X-Files"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6172,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4535
|
Steve Buscemi is an Italian actor known for his role in Reservoir Dogs, directed by a popular director known for his films that are characterized by frequent references to popular culture.
|
{
"evidence": [
"Steve Buscemi",
"Quentin Tarantino"
],
"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:\nSteve Buscemi is an Italian actor known for his role in Reservoir Dogs, directed by a popular director known for his films that are characterized by frequent references to popular culture.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Steve Buscemi",
"Quentin Tarantino"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6173,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
90238
|
Marie Henning was a member of the Women's Policy department and belonged to the Soviet sponsored Red Aid workers and her mother became a member of the Hamburg Parliament
|
{
"evidence": [
"Marie Henning_sentence_8",
"Marie Henning_sentence_11",
"Marie Henning_sentence_23",
"Marie Henning_sentence_24",
"Marie Henning_sentence_25",
"Marie Henning_sentence_26",
"Marie Henning_sentence_27",
"Marie Henning_sentence_28"
],
"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:\nMarie Henning was a member of the Women's Policy department and belonged to the Soviet sponsored Red Aid workers and her mother became a member of the Hamburg Parliament\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Marie Henning_sentence_8",
"Marie Henning_sentence_11",
"Marie Henning_sentence_23",
"Marie Henning_sentence_24",
"Marie Henning_sentence_25",
"Marie Henning_sentence_26",
"Marie Henning_sentence_27",
"Marie Henning_sentence_28"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 6174,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.002",
"value": "?search="
},
{
"operation": "search",
"time": "0.016",
"value": "Marie Henning"
},
{
"operation": "Now on",
"time": "0.018",
"value": "Marie Henning"
},
{
"operation": "Highlighting",
"time": "0.038",
"value": "Marie Henning_sentence_6"
},
{
"operation": "Highlighting",
"time": "0.045",
"value": "Marie Henning_sentence_7"
},
{
"operation": "Highlighting",
"time": "0.054",
"value": "Marie Henning_sentence_8"
},
{
"operation": "Highlighting deleted",
"time": "0.055",
"value": "Marie Henning_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "0.056",
"value": "Marie Henning_sentence_6"
},
{
"operation": "Highlighting",
"time": "0.068",
"value": "Marie Henning_sentence_11"
},
{
"operation": "hyperlink",
"time": "0.124",
"value": "Hamburg Parliament (\"Hamburgische Bürgerschaft\")"
},
{
"operation": "Now on",
"time": "0.126",
"value": "Hamburg Parliament"
},
{
"operation": "back-button-clicked",
"time": "0.129",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "0.129",
"value": "Marie Henning"
},
{
"operation": "finish",
"time": "0.141",
"value": "finish"
},
{
"operation": "Now on",
"time": "1622734280.8",
"value": "Marie Henning"
},
{
"operation": "Highlighting",
"time": "1622734286.164",
"value": "Marie Henning_sentence_23"
},
{
"operation": "Highlighting",
"time": "1622734287.05",
"value": "Marie Henning_sentence_24"
},
{
"operation": "Highlighting",
"time": "1622734289.005",
"value": "Marie Henning_sentence_25"
},
{
"operation": "Highlighting",
"time": "1622734289.976",
"value": "Marie Henning_sentence_26"
},
{
"operation": "Highlighting",
"time": "1622734291.011",
"value": "Marie Henning_sentence_27"
},
{
"operation": "Highlighting",
"time": "1622734291.639",
"value": "Marie Henning_sentence_28"
},
{
"operation": "finish",
"time": "1622734294.056",
"value": "finish"
}
]
}
|
34342
|
Although documents of religious instruction have been written since the beginning of Christianity and a catechism is typically an assemblage of these smaller documents into one large compilation, catechism itself is not in question-and-answer format, but a source on which to base such catechisms.
|
{
"evidence": [
"Catechism of the Catholic Church_sentence_17",
"Catechism of the Catholic Church_sentence_18",
"Catechism of the Catholic Church_sentence_19"
],
"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:\nAlthough documents of religious instruction have been written since the beginning of Christianity and a catechism is typically an assemblage of these smaller documents into one large compilation, catechism itself is not in question-and-answer format, but a source on which to base such catechisms.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Catechism of the Catholic Church_sentence_17",
"Catechism of the Catholic Church_sentence_18",
"Catechism of the Catholic Church_sentence_19"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6175,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.995",
"value": "undefined"
},
{
"operation": "search",
"time": "37.415",
"value": "Religious instruction"
},
{
"operation": "Now on",
"time": "38.927",
"value": "Catechism"
},
{
"operation": "Highlighting",
"time": "45.075",
"value": "Catechism_sentence_0"
},
{
"operation": "Highlighting",
"time": "45.076",
"value": "Catechism_title"
},
{
"operation": "Highlighting",
"time": "52.734",
"value": "Catechism_sentence_1"
},
{
"operation": "Highlighting",
"time": "137.468",
"value": "Catechism_sentence_2"
},
{
"operation": "Highlighting",
"time": "164.99",
"value": "Catechism_sentence_9"
},
{
"operation": "Highlighting",
"time": "164.992",
"value": "Catechism_section_0"
},
{
"operation": "hyperlink",
"time": "223.016",
"value": "Heidelberg Catechism"
},
{
"operation": "Now on",
"time": "223.728",
"value": "Heidelberg Catechism"
},
{
"operation": "back-button-clicked",
"time": "279.523",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "280.38",
"value": "Catechism"
},
{
"operation": "search",
"time": "300.812",
"value": "Contains... catechism is typically an assemblage of these smaller documents"
},
{
"operation": "Now on",
"time": "302.6",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "310.471",
"value": "Catechism of the Catholic Church"
},
{
"operation": "Now on",
"time": "311.078",
"value": "Catechism of the Catholic Church"
},
{
"operation": "Highlighting",
"time": "343.001",
"value": "Catechism of the Catholic Church_sentence_17"
},
{
"operation": "Highlighting",
"time": "344.718",
"value": "Catechism of the Catholic Church_sentence_18"
},
{
"operation": "Highlighting",
"time": "351.074",
"value": "Catechism of the Catholic Church_sentence_19"
},
{
"operation": "finish",
"time": "438.258",
"value": "finish"
}
]
}
|
32476
|
Helena Wells was an American novelist, writer and daughter of Robert Wells, a Loyalist, bookbinder, bookseller, and then a printer
|
{
"evidence": [
"Helena Wells_sentence_1",
"Helena Wells_sentence_2",
"Helena Wells_sentence_4",
"Helena Wells_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHelena Wells was an American novelist, writer and daughter of Robert Wells, a Loyalist, bookbinder, bookseller, and then a printer\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Helena Wells_sentence_1",
"Helena Wells_sentence_2",
"Helena Wells_sentence_4",
"Helena Wells_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6176,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.574",
"value": "?search="
},
{
"operation": "search",
"time": "9.243",
"value": "Helena Wells"
},
{
"operation": "Now on",
"time": "10.879",
"value": "Helena Wells"
},
{
"operation": "Highlighting",
"time": "13.416",
"value": "Helena Wells_sentence_1"
},
{
"operation": "Highlighting",
"time": "20.59",
"value": "Helena Wells_sentence_2"
},
{
"operation": "Highlighting",
"time": "40.877",
"value": "Helena Wells_sentence_4"
},
{
"operation": "Highlighting",
"time": "55.767",
"value": "Helena Wells_sentence_5"
},
{
"operation": "finish",
"time": "86.99",
"value": "finish"
}
]
}
|
28332
|
Cui Ren was promoted to China League One but failed to make his debut against Shanghai East Asia (chairman Zhang Min).
|
{
"evidence": [
"Cui Ren_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:\nCui Ren was promoted to China League One but failed to make his debut against Shanghai East Asia (chairman Zhang Min).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cui Ren_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6177,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.005",
"value": "?search="
},
{
"operation": "search",
"time": "30.972",
"value": "Cui Ren"
},
{
"operation": "Now on",
"time": "32.093",
"value": "Cui Ren"
},
{
"operation": "Highlighting",
"time": "50.595",
"value": "Cui Ren_sentence_1"
},
{
"operation": "Page search",
"time": "63.596",
"value": "debut"
},
{
"operation": "Highlighting",
"time": "71.383",
"value": "Cui Ren_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "72.783",
"value": "Cui Ren_sentence_1"
},
{
"operation": "finish",
"time": "88.925",
"value": "finish"
}
]
}
|
exfever_2431
|
Evolution is the second album released by Sabrina Carpenter, who began recording in 2015, shortly after the launch of her first studio album, a song with the contemporary genre that evolved from the former during the 30th-century.
|
{
"evidence": [
"Evolution (Sabrina Carpenter album)",
"Eyes Wide Open (Sabrina Carpenter album)",
"Folk music"
],
"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:\nEvolution is the second album released by Sabrina Carpenter, who began recording in 2015, shortly after the launch of her first studio album, a song with the contemporary genre that evolved from the former during the 30th-century.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Evolution (Sabrina Carpenter album)",
"Eyes Wide Open (Sabrina Carpenter album)",
"Folk music"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6178,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2618
|
Dead Man Down is a 2013 American neo-noir crime thriller film starring an English actor that appeared in the Agent Carter series on an American multinational commercial broadcast television network that serves as the flagship property of Walt Disney Television.
|
{
"evidence": [
"Dead Man Down",
"Dominic Cooper",
"American Broadcasting Company"
],
"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:\nDead Man Down is a 2013 American neo-noir crime thriller film starring an English actor that appeared in the Agent Carter series on an American multinational commercial broadcast television network that serves as the flagship property of Walt Disney Television.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dead Man Down",
"Dominic Cooper",
"American Broadcasting Company"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6179,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
79618
|
In 1911-12, Bill Noble played in the Kangaroo tour of Great Britain, and joined the Balmain Tigers as captain for his final season the following year.
|
{
"evidence": [
"Bill Noble_sentence_4",
"Bill Noble_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn 1911-12, Bill Noble played in the Kangaroo tour of Great Britain, and joined the Balmain Tigers as captain for his final season the following year.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Bill Noble_sentence_4",
"Bill Noble_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6180,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.419",
"value": "?search="
},
{
"operation": "search",
"time": "19.624",
"value": "Bill Noble"
},
{
"operation": "Now on",
"time": "20.952",
"value": "Bill Noble"
},
{
"operation": "Highlighting",
"time": "27.984",
"value": "Bill Noble_sentence_4"
},
{
"operation": "Highlighting",
"time": "56.347",
"value": "Bill Noble_sentence_5"
},
{
"operation": "finish",
"time": "113.777",
"value": "finish"
}
]
}
|
exfever_8717
|
Spider-Man first appeared in the anthology comic book Amazing Fantasy #15 in the period of artistic advancement and widespread commercial success in the mainstream following the era of American comic books from 1938 to 1956.
|
{
"evidence": [
"Spider-Man",
"Silver Age of Comic Books",
"Golden Age of Comic Books"
],
"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:\nSpider-Man first appeared in the anthology comic book Amazing Fantasy #15 in the period of artistic advancement and widespread commercial success in the mainstream following the era of American comic books from 1938 to 1956.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Spider-Man",
"Silver Age of Comic Books",
"Golden Age of Comic Books"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6181,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8072
|
Keturah was a wife and a concubine of the common patriarch of the Abrahamic religions, including Judaism, Christianity, and an Abrahamic monotheistic religion whose text, the Quran, is considered by followers, known as Muslims, to be the word of God.
|
{
"evidence": [
"Keturah",
"Abraham",
"Islam"
],
"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:\nKeturah was a wife and a concubine of the common patriarch of the Abrahamic religions, including Judaism, Christianity, and an Abrahamic monotheistic religion whose text, the Quran, is considered by followers, known as Muslims, to be the word of God.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Keturah",
"Abraham",
"Islam"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 6182,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
88815
|
Profilin 1 (encoded by the PFN1 gene) encoded by this gene is a ubiquitous actin and is thought to deregulate actin polymerization and it is associated with Miller-Dieker syndrome.
|
{
"evidence": [
"Profilin 1_sentence_0",
"Profilin 1_sentence_1",
"Profilin 1_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:\nProfilin 1 (encoded by the PFN1 gene) encoded by this gene is a ubiquitous actin and is thought to deregulate actin polymerization and it is associated with Miller-Dieker syndrome.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Profilin 1_sentence_0",
"Profilin 1_sentence_1",
"Profilin 1_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 6183,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.022",
"value": "?search="
},
{
"operation": "search",
"time": "21.815",
"value": "Profilin 1"
},
{
"operation": "Now on",
"time": "22.347",
"value": "Profilin 1"
},
{
"operation": "Highlighting",
"time": "29.477",
"value": "Profilin 1_sentence_0"
},
{
"operation": "Highlighting",
"time": "36.185",
"value": "Profilin 1_sentence_1"
},
{
"operation": "Highlighting",
"time": "42.136",
"value": "Profilin 1_sentence_2"
},
{
"operation": "finish",
"time": "57.055",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.