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_709
|
The Bible is a compilation of text of a variety of forms, originally written in an archaic form of language in the Canaanite branch of semintic language.
|
{
"evidence": [
"Bible",
"Biblical Hebrew"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Bible is a compilation of text of a variety of forms, originally written in an archaic form of language in the Canaanite branch of semintic language.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bible",
"Biblical Hebrew"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7684,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2907
|
Seven Days in May stars different artists, including an American actor who was regarded as one of the most celebrated stars of the 1930s.
|
{
"evidence": [
"Seven Days in May",
"Fredric March"
],
"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:\nSeven Days in May stars different artists, including an American actor who was regarded as one of the most celebrated stars of the 1930s.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Seven Days in May",
"Fredric March"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7685,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8318
|
Superman is a supervillain created by Jerry Siegel and Joe Shuster that appears in comic books published by an American comic book publisher.
|
{
"evidence": [
"Superman",
"DC Comics"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSuperman is a supervillain created by Jerry Siegel and Joe Shuster that appears in comic books published by an American comic book publisher.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Superman",
"DC Comics"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7686,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
72422
|
Jeroen Kampschreur(born April 1999 in Leiderdorp) practiced wheelchair basketball and won three gold medals on the 2017 World Para Alpine Skiing Championships.
|
{
"evidence": [
"Jeroen Kampschreur_sentence_0",
"Jeroen Kampschreur_sentence_6",
"Jeroen Kampschreur_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:\nJeroen Kampschreur(born April 1999 in Leiderdorp) practiced wheelchair basketball and won three gold medals on the 2017 World Para Alpine Skiing Championships.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Jeroen Kampschreur_sentence_0",
"Jeroen Kampschreur_sentence_6",
"Jeroen Kampschreur_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7687,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.379",
"value": "?search="
},
{
"operation": "search",
"time": "48.9",
"value": "Contains...Jeroen Kampschreur("
},
{
"operation": "Now on",
"time": "51.073",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "52.719",
"value": "Jeroen Kampschreur"
},
{
"operation": "Now on",
"time": "53.739",
"value": "Jeroen Kampschreur"
},
{
"operation": "Highlighting",
"time": "60.374",
"value": "Jeroen Kampschreur_sentence_0"
},
{
"operation": "Highlighting",
"time": "77.167",
"value": "Jeroen Kampschreur_sentence_6"
},
{
"operation": "Highlighting",
"time": "80.849",
"value": "Jeroen Kampschreur_sentence_7"
},
{
"operation": "finish",
"time": "94.449",
"value": "finish"
}
]
}
|
exfever_5312
|
Home Alone is a 1990 American comedy film directed by Chris Columbus and stars include Joe Pesci and who is often regarded as one of the most successful is generally applied to youth acting on stage or in movies or on television.
|
{
"evidence": [
"Macaulay Culkin",
"Child actor"
],
"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:\nHome Alone is a 1990 American comedy film directed by Chris Columbus and stars include Joe Pesci and who is often regarded as one of the most successful is generally applied to youth acting on stage or in movies or on television.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Macaulay Culkin",
"Child actor"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7688,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3380
|
Los Angeles is the largest city in California, which is known for its wet summers and mild, wet winters.
|
{
"evidence": [
"Los Angeles",
"Mediterranean climate"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLos Angeles is the largest city in California, which is known for its wet summers and mild, wet winters.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Los Angeles",
"Mediterranean climate"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7689,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4853
|
The golden jackal is listed as the first Least Concern on the IUCN Red List.
|
{
"evidence": [
"Gray wolf",
"Golden jackal",
"Subspecies"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe golden jackal is listed as the first Least Concern on the IUCN Red List.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gray wolf",
"Golden jackal",
"Subspecies"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7690,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
80031
|
Joe Robbie was a lawyer born July 7, 1916 and died in January 7, 2000 with one spouse.
|
{
"evidence": [
"Joe Robbie_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJoe Robbie was a lawyer born July 7, 1916 and died in January 7, 2000 with one spouse.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Joe Robbie_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7691,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.79",
"value": "?search="
},
{
"operation": "search",
"time": "30.814",
"value": "Joe Robbie"
},
{
"operation": "Now on",
"time": "31.446",
"value": "Joe Robbie"
},
{
"operation": "Highlighting",
"time": "34.193",
"value": "Joe Robbie_sentence_0"
},
{
"operation": "finish",
"time": "66.369",
"value": "finish"
}
]
}
|
exfever_339
|
Lee Child is a American author who writes thriller novels and is best known for his novel, the protagonist of a series of crime thrillers.
|
{
"evidence": [
"Lee Child",
"Jack Reacher"
],
"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:\nLee Child is a American author who writes thriller novels and is best known for his novel, the protagonist of a series of crime thrillers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lee Child",
"Jack Reacher"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7692,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
26525
|
Jim Paxson is the cousin of former NBA player Jim Paxson Sr. and proceeded to earn All-America honors as a rookie after averaging more than 23 points per game..
|
{
"evidence": [
"Jim Paxson_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJim Paxson is the cousin of former NBA player Jim Paxson Sr. and proceeded to earn All-America honors as a rookie after averaging more than 23 points per game..\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Jim Paxson_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7693,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.663",
"value": "?search="
},
{
"operation": "search",
"time": "27.02",
"value": "Jim Paxson"
},
{
"operation": "Now on",
"time": "28.114",
"value": "Jim Paxson"
},
{
"operation": "Highlighting",
"time": "59.554",
"value": "Jim Paxson_sentence_4"
},
{
"operation": "finish",
"time": "216.662",
"value": "finish"
}
]
}
|
exfever_6752
|
Boston is the capital and most populous city of the Commonwealth of Massachusetts in the USA, a country primarily located in North America which consists of 60 states, a federal district, five major unincorporated territories, 326 Indian reservations, and nine minor outlying islands.
|
{
"evidence": [
"Boston",
"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:\nBoston is the capital and most populous city of the Commonwealth of Massachusetts in the USA, a country primarily located in North America which consists of 60 states, a federal district, five major unincorporated territories, 326 Indian reservations, and nine minor outlying islands.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Boston",
"United States"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7694,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
23061
|
Born in Borås, Johan Rasmussen has competed in three World Championships and the 1994 Olympics and made his World Cup debut in March 1990 with a 21st place in Örnsköldsvik, and made three starts in 1991.
|
{
"evidence": [
"Johan Rasmussen_sentence_4",
"Johan Rasmussen_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:\nBorn in Borås, Johan Rasmussen has competed in three World Championships and the 1994 Olympics and made his World Cup debut in March 1990 with a 21st place in Örnsköldsvik, and made three starts in 1991.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Johan Rasmussen_sentence_4",
"Johan Rasmussen_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7695,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.757",
"value": "?search="
},
{
"operation": "search",
"time": "21.559",
"value": "Johan Rasmussen"
},
{
"operation": "Now on",
"time": "22.307",
"value": "Johan Rasmussen"
},
{
"operation": "Highlighting",
"time": "49.512",
"value": "Johan Rasmussen_sentence_2"
},
{
"operation": "Highlighting",
"time": "66.907",
"value": "Johan Rasmussen_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "107.281",
"value": "Johan Rasmussen_sentence_2"
},
{
"operation": "Highlighting",
"time": "136.107",
"value": "Johan Rasmussen_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "218.208",
"value": "Johan Rasmussen_sentence_6"
},
{
"operation": "Highlighting",
"time": "229.548",
"value": "Johan Rasmussen_sentence_6"
},
{
"operation": "finish",
"time": "257.973",
"value": "finish"
}
]
}
|
90958
|
Jose Gutierez Xtravaganza appeared in the December 1988 issue of American Vodue magazine alongside Naomi Campbell; then in 1989, he participated in the Love Ball, an elaborate celebrity fundraiser organized by the Design Industry Foundation for AIDS.
|
{
"evidence": [
"Jose Gutierez Xtravaganza_sentence_21",
"Jose Gutierez Xtravaganza_sentence_22"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJose Gutierez Xtravaganza appeared in the December 1988 issue of American Vodue magazine alongside Naomi Campbell; then in 1989, he participated in the Love Ball, an elaborate celebrity fundraiser organized by the Design Industry Foundation for AIDS.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Jose Gutierez Xtravaganza_sentence_21",
"Jose Gutierez Xtravaganza_sentence_22"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7696,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.765",
"value": "?search="
},
{
"operation": "search",
"time": "6.703",
"value": "Jose Gutierez Xtravaganza"
},
{
"operation": "Now on",
"time": "7.528",
"value": "Jose Gutierez Xtravaganza"
},
{
"operation": "Highlighting",
"time": "13.24",
"value": "Jose Gutierez Xtravaganza_sentence_21"
},
{
"operation": "Highlighting",
"time": "202.301",
"value": "Jose Gutierez Xtravaganza_sentence_22"
},
{
"operation": "finish",
"time": "211.103",
"value": "finish"
}
]
}
|
71899
|
Hickie, born in 1964 in Winnipeg into a military family, is a member of the Saskatchewan Party and was elected to represent the electoral district of Prince Albert Carlton in the Legislative Assembly of Saskatchewan in the 2007 election.
|
{
"evidence": [
"Darryl Hickie_sentence_0",
"Darryl Hickie_sentence_3",
"Darryl Hickie_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHickie, born in 1964 in Winnipeg into a military family, is a member of the Saskatchewan Party and was elected to represent the electoral district of Prince Albert Carlton in the Legislative Assembly of Saskatchewan in the 2007 election.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Darryl Hickie_sentence_0",
"Darryl Hickie_sentence_3",
"Darryl Hickie_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7697,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.733",
"value": "?search="
},
{
"operation": "search",
"time": "14.946",
"value": "Hickie"
},
{
"operation": "Now on",
"time": "17.397",
"value": "Hickie"
},
{
"operation": "hyperlink",
"time": "22.569",
"value": "Darryl Hickie"
},
{
"operation": "Now on",
"time": "24.197",
"value": "Darryl Hickie"
},
{
"operation": "Highlighting",
"time": "31.357",
"value": "Darryl Hickie_sentence_0"
},
{
"operation": "Highlighting",
"time": "45.42",
"value": "Darryl Hickie_sentence_3"
},
{
"operation": "Highlighting",
"time": "64.505",
"value": "Darryl Hickie_sentence_2"
},
{
"operation": "search",
"time": "167.526",
"value": "Legislative Assembly of Saskatchewan"
},
{
"operation": "Now on",
"time": "168.962",
"value": "Legislative Assembly of Saskatchewan"
},
{
"operation": "Page search",
"time": "252.876",
"value": "prince albert"
},
{
"operation": "search",
"time": "319.977",
"value": "2007 Saskatchewan general election"
},
{
"operation": "Now on",
"time": "320.839",
"value": "2007 Saskatchewan general election"
},
{
"operation": "Page search",
"time": "332.457",
"value": "prince albert"
},
{
"operation": "finish",
"time": "375.725",
"value": "finish"
}
]
}
|
47280
|
Melora Walters married Alex Vendller in 2008 but would divorce him two years later.
|
{
"evidence": [
"Melora Walters_sentence_2",
"Melora Walters_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:\nMelora Walters married Alex Vendller in 2008 but would divorce him two years later.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Melora Walters_sentence_2",
"Melora Walters_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7698,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.92",
"value": "?search="
},
{
"operation": "search",
"time": "9.92",
"value": "Melora Walters"
},
{
"operation": "Now on",
"time": "10.967",
"value": "Melora Walters"
},
{
"operation": "Highlighting",
"time": "236.69",
"value": "Melora Walters_sentence_2"
},
{
"operation": "Highlighting",
"time": "238.648",
"value": "Melora Walters_sentence_3"
},
{
"operation": "finish",
"time": "243.957",
"value": "finish"
}
]
}
|
88459
|
Steve Rushton, a British solo artist, and formerly co-lead singer bassist for the pop punk band, Son of Dork, was also in a band called Mr. Cheerful who was heavily influenced by Green Day and Elvis, and participate in the UK final for the 2003 Junior Eurovision Song Contest.
|
{
"evidence": [
"Steve Rushton_sentence_0",
"Steve Rushton_sentence_4",
"Steve Rushton_sentence_5",
"Steve Rushton_sentence_6",
"Steve Rushton_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:\nSteve Rushton, a British solo artist, and formerly co-lead singer bassist for the pop punk band, Son of Dork, was also in a band called Mr. Cheerful who was heavily influenced by Green Day and Elvis, and participate in the UK final for the 2003 Junior Eurovision Song Contest.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Steve Rushton_sentence_0",
"Steve Rushton_sentence_4",
"Steve Rushton_sentence_5",
"Steve Rushton_sentence_6",
"Steve Rushton_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7699,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.768",
"value": "undefined"
},
{
"operation": "search",
"time": "5.3",
"value": "Steve Rushton"
},
{
"operation": "Now on",
"time": "6.156",
"value": "Steve Rushton"
},
{
"operation": "Highlighting",
"time": "7.386",
"value": "Steve Rushton_sentence_0"
},
{
"operation": "Highlighting",
"time": "16.167",
"value": "Steve Rushton_sentence_4"
},
{
"operation": "Highlighting",
"time": "16.964",
"value": "Steve Rushton_sentence_5"
},
{
"operation": "Highlighting",
"time": "20.148",
"value": "Steve Rushton_sentence_6"
},
{
"operation": "Highlighting",
"time": "25.02",
"value": "Steve Rushton_sentence_7"
},
{
"operation": "finish",
"time": "32.295",
"value": "finish"
}
]
}
|
37015
|
André Leducq worked his way to his Tour de France success.
|
{
"evidence": [
"André Leducq_sentence_3",
"André Leducq_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:\nAndré Leducq worked his way to his Tour de France success.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"André Leducq_sentence_3",
"André Leducq_sentence_4"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7700,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.874",
"value": "?search="
},
{
"operation": "search",
"time": "8.952",
"value": "André Leducq"
},
{
"operation": "Now on",
"time": "9.903",
"value": "André Leducq"
},
{
"operation": "Highlighting",
"time": "64.649",
"value": "André Leducq_sentence_3"
},
{
"operation": "Highlighting",
"time": "66.048",
"value": "André Leducq_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "70.622",
"value": "André Leducq_sentence_4"
},
{
"operation": "Highlighting",
"time": "71.575",
"value": "André Leducq_sentence_4"
},
{
"operation": "finish",
"time": "180.494",
"value": "finish"
}
]
}
|
exfever_544
|
Elise Demetria Neal is an American actress who starred as Yvonne Hughley in the Hughleys sitcom on an Italian broadcast television network that launched on January 16, 1995.
|
{
"evidence": [
"Elise Neal",
"UPN"
],
"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:\nElise Demetria Neal is an American actress who starred as Yvonne Hughley in the Hughleys sitcom on an Italian broadcast television network that launched on January 16, 1995.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Elise Neal",
"UPN"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7701,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8644
|
Ellen Burstyn earned a nomination for Best Actress at an Academy Award for her appearance in the 1963 American supernatural horror film also starring a Swedish-French actor who appeared in more than 150 films and several television series in many languages.
|
{
"evidence": [
"Ellen Burstyn",
"The Exorcist (film)",
"Max von Sydow"
],
"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:\nEllen Burstyn earned a nomination for Best Actress at an Academy Award for her appearance in the 1963 American supernatural horror film also starring a Swedish-French actor who appeared in more than 150 films and several television series in many languages.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ellen Burstyn",
"The Exorcist (film)",
"Max von Sydow"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7702,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3317
|
In 1998, Tom Hardy became a model at the age of 21.
|
{
"evidence": [
"Tom Hardy",
"Drama Centre London",
"Federation of Drama Schools"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn 1998, Tom Hardy became a model at the age of 21.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tom Hardy",
"Drama Centre London",
"Federation of Drama Schools"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7703,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
42542
|
Ashurst Gilbert graduated from Brasenose College, Cambridge and married Mary Anne Wintle.
|
{
"evidence": [
"Ashurst Gilbert_sentence_2",
"Ashurst Gilbert_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAshurst Gilbert graduated from Brasenose College, Cambridge and married Mary Anne Wintle.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ashurst Gilbert_sentence_2",
"Ashurst Gilbert_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7704,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "4.484",
"value": "undefined"
},
{
"operation": "search",
"time": "13.3",
"value": "Ashurst Gilbert"
},
{
"operation": "Now on",
"time": "14.735",
"value": "Ashurst Gilbert"
},
{
"operation": "Highlighting",
"time": "29.257",
"value": "Ashurst Gilbert_sentence_2"
},
{
"operation": "Highlighting",
"time": "33.907",
"value": "Ashurst Gilbert_sentence_4"
},
{
"operation": "finish",
"time": "48.223",
"value": "finish"
}
]
}
|
72074
|
The Rob Roy Glacier in New Zealand covers most of the headwalls encircling a valley and during spring it only stays frozen causing small avalanches and bigger blocks of ice can tumble.
|
{
"evidence": [
"Rob Roy Glacier_sentence_6"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Rob Roy Glacier in New Zealand covers most of the headwalls encircling a valley and during spring it only stays frozen causing small avalanches and bigger blocks of ice can tumble.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Rob Roy Glacier_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7705,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.813",
"value": "?search="
},
{
"operation": "search",
"time": "9.3",
"value": "Rob Roy Glacier"
},
{
"operation": "Now on",
"time": "10.232",
"value": "Rob Roy Glacier"
},
{
"operation": "Highlighting",
"time": "31.036",
"value": "Rob Roy Glacier_sentence_5"
},
{
"operation": "Highlighting",
"time": "46.108",
"value": "Rob Roy Glacier_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "46.948",
"value": "Rob Roy Glacier_sentence_5"
},
{
"operation": "finish",
"time": "59.395",
"value": "finish"
}
]
}
|
72541
|
Fall for Your Type by Jamie Foxx is a somber track with a downtempo ballad that contains an "icy cold", hip hop landscape.
|
{
"evidence": [
"Fall for Your Type_sentence_0",
"Fall for Your Type_sentence_9",
"Fall for Your Type_sentence_10"
],
"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:\nFall for Your Type by Jamie Foxx is a somber track with a downtempo ballad that contains an \"icy cold\", hip hop landscape.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Fall for Your Type_sentence_0",
"Fall for Your Type_sentence_9",
"Fall for Your Type_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7706,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.803",
"value": "?search="
},
{
"operation": "search",
"time": "9.35",
"value": "Fall for Your Type"
},
{
"operation": "Now on",
"time": "10.045",
"value": "Fall for Your Type"
},
{
"operation": "Highlighting",
"time": "20.782",
"value": "Fall for Your Type_sentence_0"
},
{
"operation": "Highlighting",
"time": "54.476",
"value": "Fall for Your Type_sentence_9"
},
{
"operation": "Highlighting",
"time": "60.794",
"value": "Fall for Your Type_sentence_10"
},
{
"operation": "finish",
"time": "94.223",
"value": "finish"
}
]
}
|
exfever_7579
|
Behemoth was an English rock band formed in Birmingham in 1968 by the British guitarist who was ranked 25th on Rolling Stone magazine's list of the "100 Greatest Guitarists of All Time."
|
{
"evidence": [
"Black Sabbath",
"Tony Iommi"
],
"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:\nBehemoth was an English rock band formed in Birmingham in 1968 by the British guitarist who was ranked 25th on Rolling Stone magazine's list of the \"100 Greatest Guitarists of All Time.\"\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Black Sabbath",
"Tony Iommi"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7707,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
41912
|
London Customs Convention, officially titled the Netherlands–Belgium–Luxembourg Customs Convention, is a treaty that was signed on September 5, 1944 at London which was ratified in The Hague in 1947, but it was eventually revoked after a year due to a stronger treaty.
|
{
"evidence": [
"London Customs Convention_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLondon Customs Convention, officially titled the Netherlands–Belgium–Luxembourg Customs Convention, is a treaty that was signed on September 5, 1944 at London which was ratified in The Hague in 1947, but it was eventually revoked after a year due to a stronger treaty.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"London Customs Convention_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7708,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.777",
"value": "?search="
},
{
"operation": "search",
"time": "4.287",
"value": "London Customs Convention"
},
{
"operation": "Now on",
"time": "5.382",
"value": "London Customs Convention"
},
{
"operation": "Highlighting",
"time": "87.715",
"value": "London Customs Convention_sentence_0"
},
{
"operation": "Highlighting",
"time": "98.059",
"value": "London Customs Convention_sentence_4"
},
{
"operation": "finish",
"time": "121.923",
"value": "finish"
}
]
}
|
exfever_7847
|
Jane Austen's Emma is an adaptation of the 1815 novel of the same name, starring an English actress and model who has appeared in British costume dramas, including a 2000 period romantic drama film directed by Mel Gibson.
|
{
"evidence": [
"Emma (1996 TV film)",
"Kate Beckinsale",
"The Golden Bowl (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:\nJane Austen's Emma is an adaptation of the 1815 novel of the same name, starring an English actress and model who has appeared in British costume dramas, including a 2000 period romantic drama film directed by Mel Gibson.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Emma (1996 TV film)",
"Kate Beckinsale",
"The Golden Bowl (film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7709,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8856
|
Etan Cohen is a film director.
|
{
"evidence": [
"Tropic Thunder",
"Ben Stiller",
"The Secret Life of Walter Mitty (2013 film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEtan Cohen is a film director.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tropic Thunder",
"Ben Stiller",
"The Secret Life of Walter Mitty (2013 film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7710,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3856
|
Demi Gene Moore is an Australian actress who appeared in an American daytime television soap opera listed in Guinness World Records as the longest-running American soap opera in production.
|
{
"evidence": [
"Demi Moore",
"General Hospital"
],
"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:\nDemi Gene Moore is an Australian actress who appeared in an American daytime television soap opera listed in Guinness World Records as the longest-running American soap opera in production.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Demi Moore",
"General Hospital"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7711,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
73952
|
University of Veterinary Medicine and Pharmacy in Košice began its teaching activities with its first lecture on 5 October 1949 and it became the first school of veterinary medicine in Slovakia and the second one in Czechoslovakia.
|
{
"evidence": [
"University of Veterinary Medicine and Pharmacy in Košice_sentence_6",
"University of Veterinary Medicine and Pharmacy in Košice_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:\nUniversity of Veterinary Medicine and Pharmacy in Košice began its teaching activities with its first lecture on 5 October 1949 and it became the first school of veterinary medicine in Slovakia and the second one in Czechoslovakia.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"University of Veterinary Medicine and Pharmacy in Košice_sentence_6",
"University of Veterinary Medicine and Pharmacy in Košice_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7712,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.528",
"value": "?search="
},
{
"operation": "search",
"time": "13.309",
"value": "University of Veterinary Medicine and Pharmacy in Košice"
},
{
"operation": "Now on",
"time": "15.737",
"value": "University of Veterinary Medicine and Pharmacy in Košice"
},
{
"operation": "Highlighting",
"time": "26.392",
"value": "University of Veterinary Medicine and Pharmacy in Košice_sentence_6"
},
{
"operation": "Highlighting",
"time": "31.262",
"value": "University of Veterinary Medicine and Pharmacy in Košice_sentence_7"
},
{
"operation": "finish",
"time": "37.013",
"value": "finish"
}
]
}
|
exfever_698
|
Shirley MacLaine was the first actress awarded the AFI Life Achievement Award in 2012.
|
{
"evidence": [
"Shirley MacLaine",
"Richmond, Virginia"
],
"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:\nShirley MacLaine was the first actress awarded the AFI Life Achievement Award in 2012.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Shirley MacLaine",
"Richmond, Virginia"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7713,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2227
|
Entourage is a 2015 American comedy film written, directed and produced by an American podcaster, screenwriter and film and TV director, who attended Tulane University.
|
{
"evidence": [
"Entourage (film)",
"Doug Ellin"
],
"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:\nEntourage is a 2015 American comedy film written, directed and produced by an American podcaster, screenwriter and film and TV director, who attended Tulane University.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Entourage (film)",
"Doug Ellin"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7714,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
90193
|
Stunfest 2016 was a fighting game tournament in France, won by Japan's Yusuke Momochi.
|
{
"evidence": [
"Stunfest 2016_sentence_3",
"Stunfest 2016_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:\nStunfest 2016 was a fighting game tournament in France, won by Japan's Yusuke Momochi.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Stunfest 2016_sentence_3",
"Stunfest 2016_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7715,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.28",
"value": "?search="
},
{
"operation": "search",
"time": "11.281",
"value": "Stunfest 2016"
},
{
"operation": "Now on",
"time": "15.867",
"value": "Stunfest 2016"
},
{
"operation": "Highlighting",
"time": "19.083",
"value": "Stunfest 2016_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "26.706",
"value": "Stunfest 2016_sentence_0"
},
{
"operation": "Highlighting",
"time": "27.606",
"value": "Stunfest 2016_sentence_3"
},
{
"operation": "Highlighting",
"time": "36.135",
"value": "Stunfest 2016_sentence_0"
},
{
"operation": "finish",
"time": "59.061",
"value": "finish"
}
]
}
|
exfever_7247
|
Warren Casey wrote and composed the stage musical Grease, together with an American actor, composer, lyricist, and writer for the theatre.
|
{
"evidence": [
"Warren Casey",
"Jim Jacobs"
],
"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:\nWarren Casey wrote and composed the stage musical Grease, together with an American actor, composer, lyricist, and writer for the theatre.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Warren Casey",
"Jim Jacobs"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7716,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
1924
|
Jerry Fuchs (1974-2009), was the main singer and manager of the Beatles.
|
{
"evidence": [
"Jerry Fuchs_sentence_0",
"Jerry Fuchs_sentence_1",
"The Beatles_sentence_7"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJerry Fuchs (1974-2009), was the main singer and manager of the Beatles.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Jerry Fuchs_sentence_0",
"Jerry Fuchs_sentence_1",
"The Beatles_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7717,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "23.424",
"value": "Jerry Fuchs"
},
{
"operation": "Highlighting",
"time": "214.449",
"value": "Jerry Fuchs_sentence_0"
},
{
"operation": "Highlighting",
"time": "220.955",
"value": "Jerry Fuchs_sentence_1"
},
{
"operation": "search",
"time": "257.633",
"value": "Beatles"
},
{
"operation": "hyperlink",
"time": "280.077",
"value": "5 Personnel The Beatles_section_28"
},
{
"operation": "Highlighting",
"time": "282.24",
"value": "The Beatles_sentence_588"
},
{
"operation": "Highlighting",
"time": "282.458",
"value": "The Beatles_section_28"
},
{
"operation": "hyperlink",
"time": "282.959",
"value": "List of members of bands featuring members of the Beatles"
},
{
"operation": "back-button-clicked",
"time": "335.203",
"value": "back-button-clicked"
},
{
"operation": "Highlighting",
"time": "354.954",
"value": "The Beatles_sentence_7"
},
{
"operation": "finish",
"time": "365.622",
"value": "finish"
}
]
}
|
95662
|
Zapatourism, a touristic phenomenon in Africa that has started to lose support, attracts tourists by the message, image, or policy of the Zapatista movement which claims to provide autonomy, dignity, and freedom of the local indigenous communities.
|
{
"evidence": [
"Zapaturismo_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:\nZapatourism, a touristic phenomenon in Africa that has started to lose support, attracts tourists by the message, image, or policy of the Zapatista movement which claims to provide autonomy, dignity, and freedom of the local indigenous communities.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Zapaturismo_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7718,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.876",
"value": "?search="
},
{
"operation": "search",
"time": "9.917",
"value": "Zapatourism"
},
{
"operation": "Now on",
"time": "10.849",
"value": "Zapaturismo"
},
{
"operation": "Highlighting",
"time": "27.845",
"value": "Zapaturismo_sentence_0"
},
{
"operation": "finish",
"time": "49.545",
"value": "finish"
}
]
}
|
exfever_6841
|
Beautiful is a 2000 American comedy-drama film directed by an American actress who has received various accolades.
|
{
"evidence": [
"Beautiful (2000 film)",
"Sally Field"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBeautiful is a 2000 American comedy-drama film directed by an American actress who has received various accolades.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Beautiful (2000 film)",
"Sally Field"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7719,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7513
|
Kaitlin Olson won a Primetime Emmy Award for Outstanding Actress in a Short-Form Comedy or Drama Series for her role in The Mick.
|
{
"evidence": [
"Kaitlin Olson",
"Fox Broadcasting Company"
],
"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:\nKaitlin Olson won a Primetime Emmy Award for Outstanding Actress in a Short-Form Comedy or Drama Series for her role in The Mick.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kaitlin Olson",
"Fox Broadcasting Company"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7720,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7268
|
Solange Knowles is the youngest who has released her first studio album.
|
{
"evidence": [
"Solange Knowles",
"Bring It On: All or Nothing",
"Comedy film"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSolange Knowles is the youngest who has released her first studio album.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Solange Knowles",
"Bring It On: All or Nothing",
"Comedy film"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7721,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
3551
|
The SpVgg Neckarelz achieved three gold medals in the club's recent season performances from 2003-2017.
|
{
"evidence": [
"SpVgg Neckarelz_sentence_11"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe SpVgg Neckarelz achieved three gold medals in the club's recent season performances from 2003-2017.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"SpVgg Neckarelz_sentence_11"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7722,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "71.487",
"value": "SpVgg Neckarelz"
},
{
"operation": "Page search",
"time": "118.705",
"value": "gold"
},
{
"operation": "Highlighting",
"time": "330.491",
"value": "SpVgg Neckarelz_sentence_11"
},
{
"operation": "finish",
"time": "333.066",
"value": "finish"
}
]
}
|
16052
|
In May 2002, Sogani launched his Signature Brand of Lifestyle Accessories under the label Studio Vibhor Sogan; then in 2006, he designed a 2100 sq.m pavilion for DRDO.
|
{
"evidence": [
"Vibhor Sogani_sentence_7"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn May 2002, Sogani launched his Signature Brand of Lifestyle Accessories under the label Studio Vibhor Sogan; then in 2006, he designed a 2100 sq.m pavilion for DRDO.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Vibhor Sogani_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7723,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.555",
"value": "?search="
},
{
"operation": "search",
"time": "17.642",
"value": "Contains...Sogani"
},
{
"operation": "Now on",
"time": "18.083",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "25.633",
"value": "Vibhor Sogani"
},
{
"operation": "Now on",
"time": "26.781",
"value": "Vibhor Sogani"
},
{
"operation": "Highlighting",
"time": "45.178",
"value": "Vibhor Sogani_sentence_7"
},
{
"operation": "finish",
"time": "66.77",
"value": "finish"
}
]
}
|
exfever_7976
|
RuPaul built a fashion school in New York.
|
{
"evidence": [
"RuPaul",
"New York City",
"Combined statistical area"
],
"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:\nRuPaul built a fashion school in New York.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"RuPaul",
"New York City",
"Combined statistical area"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7724,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
30552
|
On January 17, 2015, Kavita signed a homegrown contract with Real Salt Lake then with Puerto Rico FC on April 28, 2017.
|
{
"evidence": [
"Phanuel Kavita_sentence_3",
"Phanuel Kavita_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:\nOn January 17, 2015, Kavita signed a homegrown contract with Real Salt Lake then with Puerto Rico FC on April 28, 2017.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Phanuel Kavita_sentence_3",
"Phanuel Kavita_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7725,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.614",
"value": "undefined"
},
{
"operation": "search",
"time": "21.142",
"value": "Puerto Rico FC"
},
{
"operation": "Now on",
"time": "22.915",
"value": "Puerto Rico FC"
},
{
"operation": "search",
"time": "46.965",
"value": "Contains...Kavita"
},
{
"operation": "Now on",
"time": "48.397",
"value": "undefined"
},
{
"operation": "search",
"time": "91.301",
"value": "Kavita Real Salt Lake"
},
{
"operation": "Now on",
"time": "92.879",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "98.931",
"value": "Phanuel Kavita"
},
{
"operation": "Now on",
"time": "99.97",
"value": "Phanuel Kavita"
},
{
"operation": "Page search",
"time": "109.171",
"value": "Real Salt Lake"
},
{
"operation": "Highlighting",
"time": "112.833",
"value": "Phanuel Kavita_sentence_3"
},
{
"operation": "Page search",
"time": "123.506",
"value": "rico"
},
{
"operation": "Highlighting",
"time": "128.39",
"value": "Phanuel Kavita_sentence_5"
},
{
"operation": "finish",
"time": "138.445",
"value": "finish"
}
]
}
|
38476
|
Ozzie Cobb Lake is located in Oklahoma, it is a lake which is localized water in a basin surrounded by land.
|
{
"evidence": [
"Ozzie Cobb Lake_sentence_0",
"Lake_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:\nOzzie Cobb Lake is located in Oklahoma, it is a lake which is localized water in a basin surrounded by land.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Ozzie Cobb Lake_sentence_0",
"Lake_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7726,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.931",
"value": "?search="
},
{
"operation": "search",
"time": "12.238",
"value": "Ozzie Cobb Lake"
},
{
"operation": "Now on",
"time": "13.007",
"value": "Ozzie Cobb Lake"
},
{
"operation": "Highlighting",
"time": "17.668",
"value": "Ozzie Cobb Lake_sentence_0"
},
{
"operation": "search",
"time": "43.894",
"value": "Lake"
},
{
"operation": "Now on",
"time": "45.162",
"value": "Lake"
},
{
"operation": "Highlighting",
"time": "49.628",
"value": "Lake_sentence_1"
},
{
"operation": "Page search",
"time": "71.345",
"value": "locali"
},
{
"operation": "Highlighting deleted",
"time": "74.411",
"value": "Lake_sentence_1"
},
{
"operation": "Highlighting",
"time": "75.532",
"value": "Lake_sentence_1"
},
{
"operation": "finish",
"time": "88.403",
"value": "finish"
}
]
}
|
58355
|
Akira Watanabe holds three titles in Japanese Shogi, which includes Ōshō, Kiō, and Meijin, which is the most prestigious title and only open to professional shogi players that are members of the Meijin tournament system.
|
{
"evidence": [
"Akira Watanabe (shogi)_sentence_1",
"Meijin (shogi)_sentence_3",
"Meijin (shogi)_sentence_12"
],
"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:\nAkira Watanabe holds three titles in Japanese Shogi, which includes Ōshō, Kiō, and Meijin, which is the most prestigious title and only open to professional shogi players that are members of the Meijin tournament system.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Akira Watanabe (shogi)_sentence_1",
"Meijin (shogi)_sentence_3",
"Meijin (shogi)_sentence_12"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7727,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.78",
"value": "?search="
},
{
"operation": "search",
"time": "8.043",
"value": "akira watanabe"
},
{
"operation": "Now on",
"time": "9.078",
"value": "Akira Watanabe"
},
{
"operation": "hyperlink",
"time": "13.163",
"value": "Akira Watanabe (shogi)"
},
{
"operation": "Now on",
"time": "14.028",
"value": "Akira Watanabe (shogi)"
},
{
"operation": "Highlighting",
"time": "41.482",
"value": "Akira Watanabe (shogi)_sentence_1"
},
{
"operation": "hyperlink",
"time": "44.623",
"value": "Meijin"
},
{
"operation": "Now on",
"time": "45.949",
"value": "Meijin (shogi)"
},
{
"operation": "Highlighting",
"time": "47.369",
"value": "Meijin (shogi)_sentence_3"
},
{
"operation": "Page search",
"time": "72.726",
"value": "meijin tournament system"
},
{
"operation": "Highlighting",
"time": "77.467",
"value": "Meijin (shogi)_sentence_12"
},
{
"operation": "finish",
"time": "83.321",
"value": "finish"
}
]
}
|
exfever_4833
|
Jamie Foxx is an actor who became widely known for his portrayal of an American singer regarded as one of the most iconic and influential singers ever in the 2004 biographical film Ray.
|
{
"evidence": [
"Jamie Foxx",
"Ray Charles"
],
"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:\nJamie Foxx is an actor who became widely known for his portrayal of an American singer regarded as one of the most iconic and influential singers ever in the 2004 biographical film Ray.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jamie Foxx",
"Ray Charles"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7728,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
49610
|
Dudhi or Duddhi is a town and a nagar panchayat in Sonbhadra district in the state of Chennai, India
|
{
"evidence": [
"Dudhi_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:\nDudhi or Duddhi is a town and a nagar panchayat in Sonbhadra district in the state of Chennai, India\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dudhi_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7729,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.013",
"value": "?search="
},
{
"operation": "search",
"time": "16.955",
"value": "Dudhi"
},
{
"operation": "Now on",
"time": "17.644",
"value": "Dudhi"
},
{
"operation": "Page search",
"time": "26.874",
"value": "Dudd"
},
{
"operation": "Highlighting",
"time": "29.159",
"value": "Dudhi_sentence_2"
},
{
"operation": "hyperlink",
"time": "103.328",
"value": "Uttar Pradesh"
},
{
"operation": "Now on",
"time": "105.016",
"value": "Uttar Pradesh"
},
{
"operation": "back-button-clicked",
"time": "109.629",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "109.712",
"value": "Dudhi"
},
{
"operation": "finish",
"time": "148.164",
"value": "finish"
}
]
}
|
exfever_6013
|
The Big Country is a 1958 American epic Western film directed by William Wyler that starred the British actress and singer who appeared in the movie Hamlet.
|
{
"evidence": [
"The Big Country",
"Jean Simmons"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Big Country is a 1958 American epic Western film directed by William Wyler that starred the British actress and singer who appeared in the movie Hamlet.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Big Country",
"Jean Simmons"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7730,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
91340
|
KGRW began broadcasting as Grinnell College's first radio station, it broadcasts classical music and eighty-five students provided enough staffing to allow for 121 hours a week of programming.
|
{
"evidence": [
"KDIC_sentence_7",
"KDIC_sentence_8",
"KDIC_sentence_11"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nKGRW began broadcasting as Grinnell College's first radio station, it broadcasts classical music and eighty-five students provided enough staffing to allow for 121 hours a week of programming.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"KDIC_sentence_7",
"KDIC_sentence_8",
"KDIC_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7731,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.974",
"value": "?search="
},
{
"operation": "search",
"time": "8.093",
"value": "KGRW"
},
{
"operation": "Now on",
"time": "8.816",
"value": "KGRW"
},
{
"operation": "Highlighting",
"time": "12.765",
"value": "KGRW_sentence_0"
},
{
"operation": "Highlighting",
"time": "12.766",
"value": "KGRW_title"
},
{
"operation": "Highlighting",
"time": "61.027",
"value": "KGRW_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "61.028",
"value": "KGRW_title"
},
{
"operation": "Highlighting",
"time": "61.035",
"value": "KGRW_header_cell_0_4_0"
},
{
"operation": "back-button-clicked",
"time": "91.744",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "91.825",
"value": "?search="
},
{
"operation": "search",
"time": "99.473",
"value": "Grinnell College"
},
{
"operation": "Now on",
"time": "100.426",
"value": "Grinnell College"
},
{
"operation": "Highlighting",
"time": "185.075",
"value": "Grinnell College_sentence_171"
},
{
"operation": "Highlighting",
"time": "185.077",
"value": "Grinnell College_title"
},
{
"operation": "Highlighting",
"time": "185.079",
"value": "Grinnell College_section_10"
},
{
"operation": "hyperlink",
"time": "188.395",
"value": "KDIC"
},
{
"operation": "Now on",
"time": "189.201",
"value": "KDIC"
},
{
"operation": "Highlighting",
"time": "198.128",
"value": "KDIC_sentence_1"
},
{
"operation": "Highlighting",
"time": "199.532",
"value": "KDIC_sentence_2"
},
{
"operation": "Highlighting",
"time": "227.817",
"value": "KDIC_sentence_7"
},
{
"operation": "Highlighting",
"time": "236.081",
"value": "KDIC_sentence_8"
},
{
"operation": "Highlighting",
"time": "249.195",
"value": "KDIC_sentence_11"
},
{
"operation": "Highlighting",
"time": "292.231",
"value": "KDIC_sentence_7"
},
{
"operation": "Highlighting",
"time": "292.886",
"value": "KDIC_sentence_8"
},
{
"operation": "Highlighting",
"time": "297.465",
"value": "KDIC_sentence_11"
},
{
"operation": "finish",
"time": "309.076",
"value": "finish"
}
]
}
|
exfever_5012
|
The Hangover Part III is a 2013 American action drama film starring an actor who received an award that recognizes excellence in live Broadway theatre.
|
{
"evidence": [
"The Hangover Part III",
"Bradley Cooper",
"Tony Award"
],
"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 Hangover Part III is a 2013 American action drama film starring an actor who received an award that recognizes excellence in live Broadway theatre.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Hangover Part III",
"Bradley Cooper",
"Tony Award"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7732,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8173
|
Dangal is an Indian Hindi-language biographical sports drama film produced by an Indian actor who works in Hindi films and is also the recipient of numerous awards.
|
{
"evidence": [
"Dangal (film)",
"Aamir Khan"
],
"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:\nDangal is an Indian Hindi-language biographical sports drama film produced by an Indian actor who works in Hindi films and is also the recipient of numerous awards.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dangal (film)",
"Aamir Khan"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7733,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7612
|
Ronald Ulysses Swanson is a fictional character portrayed by Nick Offerman in the American political satire and mockumentary television series created by Greg Daniels and Michael Schur.
|
{
"evidence": [
"Ron Swanson",
"Parks and Recreation"
],
"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:\nRonald Ulysses Swanson is a fictional character portrayed by Nick Offerman in the American political satire and mockumentary television series created by Greg Daniels and Michael Schur.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ron Swanson",
"Parks and Recreation"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7734,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
38272
|
Operational databases can track warehouse/stock quantities as customers order products from an online web store.
|
{
"evidence": [
"Operational database_sentence_13",
"Operational database_sentence_14"
],
"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:\nOperational databases can track warehouse/stock quantities as customers order products from an online web store.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Operational database_sentence_13",
"Operational database_sentence_14"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7735,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.584",
"value": "?search="
},
{
"operation": "Page search",
"time": "19.381",
"value": "Operational databases can track warehouse/stock quantities"
},
{
"operation": "search",
"time": "24.153",
"value": "Operational databases can track warehouse/stock quantities"
},
{
"operation": "Now on",
"time": "25.689",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "36.859",
"value": "Operational database"
},
{
"operation": "Now on",
"time": "38.19",
"value": "Operational database"
},
{
"operation": "Page search",
"time": "54.423",
"value": "stock"
},
{
"operation": "Highlighting",
"time": "93.444",
"value": "Operational database_sentence_12"
},
{
"operation": "Highlighting",
"time": "100.116",
"value": "Operational database_sentence_13"
},
{
"operation": "Highlighting",
"time": "110.706",
"value": "Operational database_sentence_14"
},
{
"operation": "Highlighting deleted",
"time": "128.475",
"value": "Operational database_sentence_12"
},
{
"operation": "finish",
"time": "154.319",
"value": "finish"
}
]
}
|
exfever_3852
|
Atlantic City was the seventeenth film directed by Louis Malle.
|
{
"evidence": [
"Atlantic City (1980 film)",
"Kate Reid"
],
"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:\nAtlantic City was the seventeenth film directed by Louis Malle.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Atlantic City (1980 film)",
"Kate Reid"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7736,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
2618
|
It is low lying in the northern half of China's coastline, while in the southern half, it's more of an irregular where in the South China Sea, it is neighbored by the Indochinese Peninsula, Taiwan and the northern Philippines.
|
{
"evidence": [
"Coastline of China_sentence_1",
"Coastline of China_sentence_2",
"South China Sea_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:\nIt is low lying in the northern half of China's coastline, while in the southern half, it's more of an irregular where in the South China Sea, it is neighbored by the Indochinese Peninsula, Taiwan and the northern Philippines.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Coastline of China_sentence_1",
"Coastline of China_sentence_2",
"South China Sea_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7737,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "51.155",
"value": "Coastline of China"
},
{
"operation": "Page search",
"time": "64.288",
"value": "low lying"
},
{
"operation": "Highlighting",
"time": "66.64",
"value": "Coastline of China_sentence_1"
},
{
"operation": "Highlighting",
"time": "74.722",
"value": "Coastline of China_sentence_2"
},
{
"operation": "Page search",
"time": "105.75",
"value": "Indochinese Peninsula"
},
{
"operation": "Page search",
"time": "129.865",
"value": "taiwan"
},
{
"operation": "Page search",
"time": "130.816",
"value": "taiwan"
},
{
"operation": "Highlighting",
"time": "149.303",
"value": "Coastline of China_sentence_1"
},
{
"operation": "Highlighting",
"time": "150.425",
"value": "Coastline of China_sentence_2"
},
{
"operation": "hyperlink",
"time": "173.442",
"value": "South China Sea"
},
{
"operation": "Highlighting",
"time": "181.264",
"value": "South China Sea_sentence_1"
},
{
"operation": "finish",
"time": "190.836",
"value": "finish"
}
]
}
|
41411
|
Teldenia alba is a moth in the family Drepanidae, It is found in Sri Lanka and was described by Moore in 1882.
|
{
"evidence": [
"Teldenia alba_sentence_0",
"Teldenia alba_sentence_1",
"Teldenia alba_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:\nTeldenia alba is a moth in the family Drepanidae, It is found in Sri Lanka and was described by Moore in 1882.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Teldenia alba_sentence_0",
"Teldenia alba_sentence_1",
"Teldenia alba_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7738,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.765",
"value": "?search="
},
{
"operation": "search",
"time": "336.154",
"value": "Teldenia alba"
},
{
"operation": "Now on",
"time": "337.063",
"value": "Teldenia alba"
},
{
"operation": "Highlighting",
"time": "347.379",
"value": "Teldenia alba_sentence_0"
},
{
"operation": "Highlighting",
"time": "348.084",
"value": "Teldenia alba_sentence_1"
},
{
"operation": "Highlighting",
"time": "348.726",
"value": "Teldenia alba_sentence_2"
},
{
"operation": "finish",
"time": "370.827",
"value": "finish"
}
]
}
|
29349
|
The A647 road has nine major junctions in the city of Leeds, the cultural, financial and commercial heart of the West Yorkshire Built-up Area (2011 census classification).
|
{
"evidence": [
"A647 road_sentence_0",
"A647 road_sentence_2",
"A647 road_sentence_3",
"Leeds_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 A647 road has nine major junctions in the city of Leeds, the cultural, financial and commercial heart of the West Yorkshire Built-up Area (2011 census classification).\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"A647 road_sentence_0",
"A647 road_sentence_2",
"A647 road_sentence_3",
"Leeds_sentence_4"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7739,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.611",
"value": "?search="
},
{
"operation": "search",
"time": "125.415",
"value": "A647 road"
},
{
"operation": "Now on",
"time": "126.509",
"value": "A647 road"
},
{
"operation": "Highlighting",
"time": "149.882",
"value": "A647 road_sentence_0"
},
{
"operation": "Highlighting",
"time": "157.784",
"value": "A647 road_sentence_2"
},
{
"operation": "Highlighting",
"time": "159.321",
"value": "A647 road_sentence_3"
},
{
"operation": "hyperlink",
"time": "188.903",
"value": "Leeds"
},
{
"operation": "Now on",
"time": "192.072",
"value": "Leeds"
},
{
"operation": "Highlighting",
"time": "193.704",
"value": "Leeds_sentence_4"
},
{
"operation": "finish",
"time": "198.658",
"value": "finish"
}
]
}
|
49134
|
Charlemagne, the monarch responsible for uniting western and central Europe in the middle ages, has a connection to the vita of Amalberga according to her cult.
|
{
"evidence": [
"Charlemagne_sentence_1",
"Charlemagne_sentence_2",
"Charlemagne_sentence_508"
],
"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:\nCharlemagne, the monarch responsible for uniting western and central Europe in the middle ages, has a connection to the vita of Amalberga according to her cult.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Charlemagne_sentence_1",
"Charlemagne_sentence_2",
"Charlemagne_sentence_508"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7740,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.114",
"value": "?search="
},
{
"operation": "search",
"time": "8.832",
"value": "Charlemagne"
},
{
"operation": "Now on",
"time": "10.878",
"value": "Charlemagne"
},
{
"operation": "Highlighting",
"time": "45.373",
"value": "Charlemagne_sentence_1"
},
{
"operation": "Highlighting",
"time": "47.029",
"value": "Charlemagne_sentence_2"
},
{
"operation": "Page search",
"time": "53.562",
"value": "vita"
},
{
"operation": "Page search",
"time": "112.829",
"value": "amalberga"
},
{
"operation": "Page search",
"time": "137.045",
"value": "vita of"
},
{
"operation": "Page search",
"time": "156.369",
"value": "vita"
},
{
"operation": "Highlighting",
"time": "182.735",
"value": "Charlemagne_sentence_508"
},
{
"operation": "search",
"time": "210.416",
"value": "vita of amalberga"
},
{
"operation": "Now on",
"time": "211.123",
"value": "undefined"
},
{
"operation": "finish",
"time": "233.857",
"value": "finish"
}
]
}
|
exfever_6679
|
Toy Story is a 1995 American computer-animated film written by the founder of Mutant Enemy Productions.
|
{
"evidence": [
"Toy Story",
"Joss Whedon"
],
"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:\nToy Story is a 1995 American computer-animated film written by the founder of Mutant Enemy Productions.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Toy Story",
"Joss Whedon"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7741,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5533
|
Wilhelmina Slater is a fictional character played by Vanessa Williams, for whom she received a supporting actress nomination at the ceremony that honors the best in the U.S. block of broadcast programming taking place during the middle of the evening for a television show.
|
{
"evidence": [
"Wilhelmina Slater",
"59th Primetime Emmy Awards",
"Prime time"
],
"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:\nWilhelmina Slater is a fictional character played by Vanessa Williams, for whom she received a supporting actress nomination at the ceremony that honors the best in the U.S. block of broadcast programming taking place during the middle of the evening for a television show.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wilhelmina Slater",
"59th Primetime Emmy Awards",
"Prime time"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7742,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
66515
|
Turk never released any mixtapes in his career.
|
{
"evidence": [
"Turk (rapper)_sentence_18"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTurk never released any mixtapes in his career.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Turk (rapper)_sentence_18"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7743,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.234",
"value": "undefined"
},
{
"operation": "search",
"time": "13.028",
"value": "Turk"
},
{
"operation": "Now on",
"time": "16.609",
"value": "Turk"
},
{
"operation": "Highlighting",
"time": "24.83",
"value": "Turk_item_1_10"
},
{
"operation": "Highlighting",
"time": "24.832",
"value": "Turk_title"
},
{
"operation": "Highlighting",
"time": "24.833",
"value": "Turk_section_1"
},
{
"operation": "hyperlink",
"time": "32.38",
"value": "Turk (rapper)"
},
{
"operation": "Now on",
"time": "35.816",
"value": "Turk (rapper)"
},
{
"operation": "Highlighting",
"time": "53.948",
"value": "Turk (rapper)_sentence_18"
},
{
"operation": "finish",
"time": "60.192",
"value": "finish"
}
]
}
|
41705
|
Matsya is an avatar of Vishnu who appears a half fish/half man; he saves the world from cosmic deluge with the help of a boat made from knowledge, on which he rescues all living beings.
|
{
"evidence": [
"Matsya_sentence_1",
"Matsya_sentence_2",
"Matsya_sentence_4",
"Matsya_sentence_3",
"Matsya_sentence_72",
"Matsya_sentence_73",
"Matsya_sentence_74"
],
"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:\nMatsya is an avatar of Vishnu who appears a half fish/half man; he saves the world from cosmic deluge with the help of a boat made from knowledge, on which he rescues all living beings.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Matsya_sentence_1",
"Matsya_sentence_2",
"Matsya_sentence_4",
"Matsya_sentence_3",
"Matsya_sentence_72",
"Matsya_sentence_73",
"Matsya_sentence_74"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7744,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.713",
"value": "?search="
},
{
"operation": "search",
"time": "9.396",
"value": "Matsya"
},
{
"operation": "Now on",
"time": "10.544",
"value": "Matsya"
},
{
"operation": "Highlighting",
"time": "12.953",
"value": "Matsya_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "14.458",
"value": "Matsya_sentence_2"
},
{
"operation": "Highlighting",
"time": "15.755",
"value": "Matsya_sentence_1"
},
{
"operation": "Highlighting",
"time": "16.302",
"value": "Matsya_sentence_2"
},
{
"operation": "Highlighting",
"time": "27.351",
"value": "Matsya_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "33.119",
"value": "Matsya_sentence_4"
},
{
"operation": "Page search",
"time": "55.18",
"value": "half"
},
{
"operation": "Highlighting",
"time": "80.87",
"value": "Matsya_sentence_159"
},
{
"operation": "Highlighting",
"time": "80.872",
"value": "Matsya_section_7"
},
{
"operation": "Page search",
"time": "130.69",
"value": "cosmic"
},
{
"operation": "Page search",
"time": "163.089",
"value": "boat"
},
{
"operation": "Highlighting",
"time": "172.159",
"value": "Matsya_sentence_39"
},
{
"operation": "Highlighting",
"time": "172.161",
"value": "Matsya_section_2"
},
{
"operation": "Highlighting deleted",
"time": "173.069",
"value": "Matsya_sentence_39"
},
{
"operation": "Highlighting deleted",
"time": "173.074",
"value": "Matsya_section_2"
},
{
"operation": "Page search",
"time": "242.63",
"value": "knowledge"
},
{
"operation": "Highlighting",
"time": "309.126",
"value": "Matsya_sentence_97"
},
{
"operation": "Highlighting",
"time": "309.128",
"value": "Matsya_section_4"
},
{
"operation": "Highlighting",
"time": "313.732",
"value": "Matsya_sentence_98"
},
{
"operation": "Highlighting",
"time": "334.714",
"value": "Matsya_sentence_100"
},
{
"operation": "Highlighting",
"time": "336.728",
"value": "Matsya_sentence_101"
},
{
"operation": "Highlighting deleted",
"time": "338.07",
"value": "Matsya_sentence_101"
},
{
"operation": "Highlighting",
"time": "356.038",
"value": "Matsya_sentence_102"
},
{
"operation": "Highlighting",
"time": "357.525",
"value": "Matsya_sentence_101"
},
{
"operation": "Page search",
"time": "361.709",
"value": "deluge"
},
{
"operation": "Highlighting",
"time": "427.424",
"value": "Matsya_sentence_54"
},
{
"operation": "Highlighting",
"time": "433.834",
"value": "Matsya_sentence_59"
},
{
"operation": "Highlighting",
"time": "452.637",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting deleted",
"time": "454.721",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting",
"time": "459.648",
"value": "Matsya_sentence_60"
},
{
"operation": "Highlighting",
"time": "460.875",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting",
"time": "470.124",
"value": "Matsya_sentence_62"
},
{
"operation": "Highlighting deleted",
"time": "490.566",
"value": "Matsya_sentence_54"
},
{
"operation": "Highlighting deleted",
"time": "491.564",
"value": "Matsya_sentence_59"
},
{
"operation": "Highlighting deleted",
"time": "492.126",
"value": "Matsya_sentence_60"
},
{
"operation": "Highlighting deleted",
"time": "492.647",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting deleted",
"time": "494.14",
"value": "Matsya_sentence_62"
},
{
"operation": "Highlighting",
"time": "495.092",
"value": "Matsya_sentence_59"
},
{
"operation": "Highlighting",
"time": "503.117",
"value": "Matsya_sentence_72"
},
{
"operation": "Highlighting",
"time": "517.723",
"value": "Matsya_sentence_73"
},
{
"operation": "Highlighting",
"time": "520.173",
"value": "Matsya_sentence_74"
},
{
"operation": "Highlighting",
"time": "526.445",
"value": "Matsya_sentence_75"
},
{
"operation": "Page search",
"time": "650.245",
"value": "ship"
},
{
"operation": "Highlighting",
"time": "727.074",
"value": "Matsya_sentence_46"
},
{
"operation": "Highlighting",
"time": "727.075",
"value": "Matsya_section_2"
},
{
"operation": "search",
"time": "855.901",
"value": "help of a boat made from knowledge"
},
{
"operation": "Now on",
"time": "860.064",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "868.651",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "868.743",
"value": "Matsya"
},
{
"operation": "hyperlink",
"time": "872.749",
"value": "Manu"
},
{
"operation": "Now on",
"time": "873.414",
"value": "Manu"
},
{
"operation": "back-button-clicked",
"time": "883.756",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "883.836",
"value": "Matsya"
},
{
"operation": "hyperlink",
"time": "893.456",
"value": "Shatapatha Brahmana"
},
{
"operation": "Now on",
"time": "894.199",
"value": "Shatapatha Brahmana"
},
{
"operation": "back-button-clicked",
"time": "901.514",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "901.599",
"value": "Matsya"
},
{
"operation": "Page search",
"time": "915.116",
"value": "knowledge"
},
{
"operation": "Page search",
"time": "915.87",
"value": "knowledge"
},
{
"operation": "Highlighting",
"time": "1002.637",
"value": "Matsya_sentence_1"
},
{
"operation": "Highlighting",
"time": "1003.238",
"value": "Matsya_sentence_2"
},
{
"operation": "Highlighting",
"time": "1009.223",
"value": "Matsya_sentence_4"
},
{
"operation": "Highlighting",
"time": "1017.775",
"value": "Matsya_sentence_3"
},
{
"operation": "Highlighting",
"time": "1059.193",
"value": "Matsya_sentence_29"
},
{
"operation": "Highlighting",
"time": "1059.194",
"value": "Matsya_section_2"
},
{
"operation": "Highlighting",
"time": "1101.39",
"value": "Matsya_sentence_33"
},
{
"operation": "Highlighting deleted",
"time": "1113.667",
"value": "Matsya_sentence_29"
},
{
"operation": "Highlighting deleted",
"time": "1114.421",
"value": "Matsya_sentence_33"
},
{
"operation": "Highlighting deleted",
"time": "1114.425",
"value": "Matsya_section_2"
},
{
"operation": "Highlighting",
"time": "1139.383",
"value": "Matsya_sentence_59"
},
{
"operation": "Highlighting",
"time": "1143.42",
"value": "Matsya_sentence_60"
},
{
"operation": "Highlighting",
"time": "1145.104",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting",
"time": "1146.482",
"value": "Matsya_sentence_62"
},
{
"operation": "Highlighting deleted",
"time": "1157.521",
"value": "Matsya_sentence_62"
},
{
"operation": "Highlighting deleted",
"time": "1158.425",
"value": "Matsya_sentence_61"
},
{
"operation": "Highlighting deleted",
"time": "1159.845",
"value": "Matsya_sentence_59"
},
{
"operation": "Highlighting deleted",
"time": "1160.319",
"value": "Matsya_sentence_60"
},
{
"operation": "Highlighting",
"time": "1170.715",
"value": "Matsya_sentence_72"
},
{
"operation": "Highlighting",
"time": "1184.992",
"value": "Matsya_sentence_73"
},
{
"operation": "Highlighting",
"time": "1188.396",
"value": "Matsya_sentence_74"
},
{
"operation": "finish",
"time": "1236.039",
"value": "finish"
}
]
}
|
exfever_5408
|
Backstreet Boys are an American vocal group that includes an American singer, songwriter, actor, and model.
|
{
"evidence": [
"Backstreet Boys",
"Kevin Richardson (musician)"
],
"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:\nBackstreet Boys are an American vocal group that includes an American singer, songwriter, actor, and model.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Backstreet Boys",
"Kevin Richardson (musician)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7745,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3416
|
The Fate of the Furious is a 2018 American action film starring an American actress and screenwriter who is best known for portraying a troubled boxer in the independent sports drama film Girlfight.
|
{
"evidence": [
"The Fate of the Furious",
"Michelle Rodriguez"
],
"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 Fate of the Furious is a 2018 American action film starring an American actress and screenwriter who is best known for portraying a troubled boxer in the independent sports drama film Girlfight.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Fate of the Furious",
"Michelle Rodriguez"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7746,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8722
|
The Prestige is a 2006 thriller film directed by Christopher Nolan whose genre revolves around the solution of a problem or a crime.
|
{
"evidence": [
"The Prestige (film)",
"Mystery film"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Prestige is a 2006 thriller film directed by Christopher Nolan whose genre revolves around the solution of a problem or a crime.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Prestige (film)",
"Mystery film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7747,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4190
|
Pineapple Express is a 2008 American stoner comedy film directed by an American filmmaker who directed the drama, Goodfellas.
|
{
"evidence": [
"Pineapple Express (film)",
"David Gordon Green"
],
"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:\nPineapple Express is a 2008 American stoner comedy film directed by an American filmmaker who directed the drama, Goodfellas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pineapple Express (film)",
"David Gordon Green"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7748,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
73056
|
Pedro Itriago Chacín was a Venezuelan lawyer, professor, politican and diplomat who was the Minister of Foreign Affairs of Venezuela till 1921.
|
{
"evidence": [
"Pedro Itriago Chacín_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:\nPedro Itriago Chacín was a Venezuelan lawyer, professor, politican and diplomat who was the Minister of Foreign Affairs of Venezuela till 1921.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Pedro Itriago Chacín_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7749,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.126",
"value": "?search="
},
{
"operation": "search",
"time": "8.662",
"value": "Pedro Itriago Chacín"
},
{
"operation": "Now on",
"time": "9.458",
"value": "Pedro Itriago Chacín"
},
{
"operation": "Highlighting",
"time": "15.226",
"value": "Pedro Itriago Chacín_sentence_0"
},
{
"operation": "Highlighting",
"time": "23.009",
"value": "Pedro Itriago Chacín_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "26.712",
"value": "Pedro Itriago Chacín_sentence_0"
},
{
"operation": "finish",
"time": "38.048",
"value": "finish"
}
]
}
|
exfever_3188
|
Steffi Graf is a former German player in a racket sport, a modern game that originated in a country that is part of the United States.
|
{
"evidence": [
"Steffi Graf",
"Tennis",
"England"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSteffi Graf is a former German player in a racket sport, a modern game that originated in a country that is part of the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Steffi Graf",
"Tennis",
"England"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7750,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7028
|
David Dhawan is an American director who has directed several films, including a 1997 remake of the Telugu film Hello Brother.
|
{
"evidence": [
"David Dhawan",
"Judwaa"
],
"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:\nDavid Dhawan is an American director who has directed several films, including a 1997 remake of the Telugu film Hello Brother.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"David Dhawan",
"Judwaa"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7751,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_312
|
Vedic Sanskrit was an ancient language of the Indo-Aryan that is attested in the large body of religious texts originating in ancient India composed of Rigveda, the Yajurveda, the Samaveda, and the Atharvaveda.
|
{
"evidence": [
"Vedic Sanskrit",
"Vedas"
],
"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:\nVedic Sanskrit was an ancient language of the Indo-Aryan that is attested in the large body of religious texts originating in ancient India composed of Rigveda, the Yajurveda, the Samaveda, and the Atharvaveda.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Vedic Sanskrit",
"Vedas"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7752,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
14043
|
The Perry Hall Mansion from which Gough ran his vast plantation operation, is a historical structure situated on a hill above the Gunpowder River Valley.
|
{
"evidence": [
"Perry Hall Mansion_sentence_5",
"Perry Hall Mansion_sentence_0",
"Perry Hall Mansion_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Perry Hall Mansion from which Gough ran his vast plantation operation, is a historical structure situated on a hill above the Gunpowder River Valley.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Perry Hall Mansion_sentence_5",
"Perry Hall Mansion_sentence_0",
"Perry Hall Mansion_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7753,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.976",
"value": "?search="
},
{
"operation": "search",
"time": "5.876",
"value": "Perry Hall Mansion"
},
{
"operation": "Now on",
"time": "6.527",
"value": "Perry Hall Mansion"
},
{
"operation": "Highlighting",
"time": "20.232",
"value": "Perry Hall Mansion_sentence_5"
},
{
"operation": "Page search",
"time": "34.04",
"value": "historical"
},
{
"operation": "Page search",
"time": "35.695",
"value": "historical"
},
{
"operation": "Highlighting",
"time": "41.665",
"value": "Perry Hall Mansion_sentence_21"
},
{
"operation": "Highlighting",
"time": "41.665",
"value": "Perry Hall Mansion_section_1"
},
{
"operation": "Page search",
"time": "58.106",
"value": "historic"
},
{
"operation": "Highlighting",
"time": "60.21",
"value": "Perry Hall Mansion_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "61.381",
"value": "Perry Hall Mansion_sentence_21"
},
{
"operation": "Highlighting deleted",
"time": "61.382",
"value": "Perry Hall Mansion_section_1"
},
{
"operation": "Highlighting",
"time": "85.074",
"value": "Perry Hall Mansion_sentence_1"
},
{
"operation": "finish",
"time": "103.091",
"value": "finish"
}
]
}
|
exfever_2283
|
Alicia Amanda Vikander is an American actress trained as a ballet dancer at the Swedish Ballet in Sweden's largest and capital city.
|
{
"evidence": [
"Alicia Vikander",
"Stockholm"
],
"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:\nAlicia Amanda Vikander is an American actress trained as a ballet dancer at the Swedish Ballet in Sweden's largest and capital city.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alicia Vikander",
"Stockholm"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7754,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8804
|
Lumen Pierce is a fictional character in the Showtime television series Dexter who leaves in the fifth season finale, which originally aired on Showtime on December 12, 2010.
|
{
"evidence": [
"Lumen Pierce",
"The Big One (Dexter)"
],
"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:\nLumen Pierce is a fictional character in the Showtime television series Dexter who leaves in the fifth season finale, which originally aired on Showtime on December 12, 2010.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lumen Pierce",
"The Big One (Dexter)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7755,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3838
|
Bryan Cranston portrays Hal in the sitcom Malcolm in the Middle, released on an American commercial broadcast television network.
|
{
"evidence": [
"Bryan Cranston",
"Fox 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:\nBryan Cranston portrays Hal in the sitcom Malcolm in the Middle, released on an American commercial broadcast television network.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bryan Cranston",
"Fox Broadcasting Company"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7756,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4000
|
This Is Me is Demi Lovato’s debut single and duet with an artist who rose to fame as a member of the popular American pop rock band, formed in 2005.
|
{
"evidence": [
"Demi Lovato",
"Joe Jonas",
"Jonas Brothers"
],
"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:\nThis Is Me is Demi Lovato’s debut single and duet with an artist who rose to fame as a member of the popular American pop rock band, formed in 2005.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Demi Lovato",
"Joe Jonas",
"Jonas Brothers"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7757,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5451
|
Stephanie Lynn Nicks is an American singer-songwriter and producer best known for her work with Fleetwood Mac on their eleventh studio album, produced by an American record producer who engineered the album, Tusk.
|
{
"evidence": [
"Stevie Nicks",
"Rumours (album)",
"Ken Caillat"
],
"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:\nStephanie Lynn Nicks is an American singer-songwriter and producer best known for her work with Fleetwood Mac on their eleventh studio album, produced by an American record producer who engineered the album, Tusk.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stevie Nicks",
"Rumours (album)",
"Ken Caillat"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7758,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4630
|
Fiona is the fictional goddess that Pallums used to worship before the gods came down to the lower world.
|
{
"evidence": [
"Jack Black",
"DreamWorks Animation",
"Shrek 2"
],
"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:\nFiona is the fictional goddess that Pallums used to worship before the gods came down to the lower world.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jack Black",
"DreamWorks Animation",
"Shrek 2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7759,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6487
|
Audrey Horne is a fictional character from the television series of an American multinational commercial broadcast television network that serves as the flagship property of Walt Disney Television.
|
{
"evidence": [
"Audrey Horne",
"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:\nAudrey Horne is a fictional character from the television series of 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": [
"Audrey Horne",
"American Broadcasting Company"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7760,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5788
|
Cougar Town is a television sitcom whose filming took place at a movie studio created by Thomas H. Ince, where many of Hollywood's Golden Age were filmed, including a 1939 American epic historical romance film adapted from the 1936 novel by Margaret Mitchell.
|
{
"evidence": [
"Cougar Town",
"Culver Studios",
"Gone with the Wind (film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCougar Town is a television sitcom whose filming took place at a movie studio created by Thomas H. Ince, where many of Hollywood's Golden Age were filmed, including a 1939 American epic historical romance film adapted from the 1936 novel by Margaret Mitchell.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Cougar Town",
"Culver Studios",
"Gone with the Wind (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7761,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
81651
|
Tim Heubach is a German professional footballer which is also known as soccer and is a team sport played with a spherical ball between two teams of 11 players.
|
{
"evidence": [
"Tim Heubach_sentence_0",
"Association football_sentence_6"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTim Heubach is a German professional footballer which is also known as soccer and is a team sport played with a spherical ball between two teams of 11 players.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Tim Heubach_sentence_0",
"Association football_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7762,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.168",
"value": "?search="
},
{
"operation": "search",
"time": "13.095",
"value": "Tim Heubach"
},
{
"operation": "Now on",
"time": "14.106",
"value": "Tim Heubach"
},
{
"operation": "Highlighting",
"time": "21.529",
"value": "Tim Heubach_sentence_0"
},
{
"operation": "hyperlink",
"time": "27.452",
"value": "footballer"
},
{
"operation": "Now on",
"time": "28.835",
"value": "Association football"
},
{
"operation": "Highlighting",
"time": "46.435",
"value": "Association football_sentence_6"
},
{
"operation": "finish",
"time": "62.697",
"value": "finish"
}
]
}
|
31033
|
Ethnoscape deals with perspectival constructs related to the location of communities and their history, politics, and linguistics, the scientific study of language..
|
{
"evidence": [
"Ethnoscape_sentence_1"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEthnoscape deals with perspectival constructs related to the location of communities and their history, politics, and linguistics, the scientific study of language..\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ethnoscape_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7763,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.091",
"value": "?search="
},
{
"operation": "search",
"time": "16.01",
"value": "Ethnoscape"
},
{
"operation": "Now on",
"time": "16.937",
"value": "Ethnoscape"
},
{
"operation": "Page search",
"time": "28.52",
"value": "perspectival"
},
{
"operation": "Page search",
"time": "61.687",
"value": "language"
},
{
"operation": "Highlighting",
"time": "77.474",
"value": "Ethnoscape_sentence_1"
},
{
"operation": "finish",
"time": "158.632",
"value": "finish"
}
]
}
|
exfever_8090
|
Evil Empire is Rage Against the Machine's best-selling album.
|
{
"evidence": [
"Rage Against the Machine",
"Zack de la Rocha",
"Big business"
],
"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:\nEvil Empire is Rage Against the Machine's best-selling album.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Rage Against the Machine",
"Zack de la Rocha",
"Big business"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7764,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
44178
|
PPP2R2A (An enzyme regulator in humans) consist of a common heteromeric core enzyme, and it is implicated in the negative control of cell growth.
|
{
"evidence": [
"PPP2R2A_sentence_0",
"PPP2R2A_sentence_3",
"PPP2R2A_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:\nPPP2R2A (An enzyme regulator in humans) consist of a common heteromeric core enzyme, and it is implicated in the negative control of cell growth.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"PPP2R2A_sentence_0",
"PPP2R2A_sentence_3",
"PPP2R2A_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7765,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.736",
"value": "?search="
},
{
"operation": "search",
"time": "5.421",
"value": "PPP2R2A"
},
{
"operation": "Now on",
"time": "6.308",
"value": "PPP2R2A"
},
{
"operation": "Page search",
"time": "15.085",
"value": "enzyme regulator"
},
{
"operation": "Page search",
"time": "15.319",
"value": "enzyme regulator"
},
{
"operation": "Highlighting",
"time": "18.132",
"value": "PPP2R2A_sentence_0"
},
{
"operation": "Page search",
"time": "28.615",
"value": "heteromeric core"
},
{
"operation": "Page search",
"time": "28.815",
"value": "heteromeric core"
},
{
"operation": "Highlighting",
"time": "30.344",
"value": "PPP2R2A_sentence_3"
},
{
"operation": "Highlighting",
"time": "39.784",
"value": "PPP2R2A_sentence_2"
},
{
"operation": "finish",
"time": "45.701",
"value": "finish"
}
]
}
|
46674
|
Metabolic gene clusters are tightly linked sets of genes with an often coregulated expression.
|
{
"evidence": [
"Metabolic gene cluster_sentence_1",
"Metabolic gene cluster_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:\nMetabolic gene clusters are tightly linked sets of genes with an often coregulated expression.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Metabolic gene cluster_sentence_1",
"Metabolic gene cluster_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7766,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.394",
"value": "?search="
},
{
"operation": "search",
"time": "7.868",
"value": "Metabolic gene clusters"
},
{
"operation": "Now on",
"time": "9.778",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "17.152",
"value": "Metabolic gene cluster"
},
{
"operation": "Now on",
"time": "17.87",
"value": "Metabolic gene cluster"
},
{
"operation": "Highlighting",
"time": "21.298",
"value": "Metabolic gene cluster_sentence_1"
},
{
"operation": "Highlighting",
"time": "26.368",
"value": "Metabolic gene cluster_sentence_2"
},
{
"operation": "finish",
"time": "30.833",
"value": "finish"
}
]
}
|
72314
|
The 1958 Eastern Michigan Hurons football team played under head coach Fred Trosko and lost five games that season.
|
{
"evidence": [
"1958 Eastern Michigan Hurons football team_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 1958 Eastern Michigan Hurons football team played under head coach Fred Trosko and lost five games that season.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1958 Eastern Michigan Hurons football team_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7767,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.024",
"value": "?search="
},
{
"operation": "search",
"time": "9.595",
"value": "1958 Eastern Michigan Hurons football team"
},
{
"operation": "Now on",
"time": "10.393",
"value": "1958 Eastern Michigan Hurons football team"
},
{
"operation": "Highlighting",
"time": "24.983",
"value": "1958 Eastern Michigan Hurons football team_sentence_1"
},
{
"operation": "finish",
"time": "51.079",
"value": "finish"
}
]
}
|
exfever_4542
|
Rick and Morty is an American young animated science fiction sitcom that follows the misadventures of a cynical stock character of a scientist who is described as "mad" or "insane" owing to a combination of unusual or unsettling personality traits.
|
{
"evidence": [
"Rick and Morty",
"Mad scientist"
],
"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:\nRick and Morty is an American young animated science fiction sitcom that follows the misadventures of a cynical stock character of a scientist who is described as \"mad\" or \"insane\" owing to a combination of unusual or unsettling personality traits.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Rick and Morty",
"Mad scientist"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7768,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
74071
|
Teldenia geminata is a moth that was not accounted for and is not found anywhere near Australian mainland.
|
{
"evidence": [
"New Guinea_sentence_5",
"Teldenia geminata_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:\nTeldenia geminata is a moth that was not accounted for and is not found anywhere near Australian mainland.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"New Guinea_sentence_5",
"Teldenia geminata_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7769,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.609",
"value": "?search="
},
{
"operation": "search",
"time": "11.761",
"value": "Teldenia geminata"
},
{
"operation": "Now on",
"time": "12.959",
"value": "Teldenia geminata"
},
{
"operation": "hyperlink",
"time": "32.851",
"value": "New Guinea"
},
{
"operation": "Now on",
"time": "36.346",
"value": "New Guinea"
},
{
"operation": "Highlighting",
"time": "45.088",
"value": "New Guinea_sentence_5"
},
{
"operation": "back-button-clicked",
"time": "50.126",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "50.262",
"value": "Teldenia geminata"
},
{
"operation": "Highlighting",
"time": "52.019",
"value": "Teldenia geminata_sentence_2"
},
{
"operation": "finish",
"time": "87.156",
"value": "finish"
}
]
}
|
exfever_637
|
San Francisco is the most visited city in California.
|
{
"evidence": [
"San Francisco",
"U.S. 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:\nSan Francisco is the most visited city in California.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"San Francisco",
"U.S. state"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7770,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
34645
|
Three large organizations were part of the feminist movement in Chile, including the Club de Señoras of Santiago, founded by Amanda Labarca.
|
{
"evidence": [
"Feminism in Chile_sentence_17",
"Feminism in Chile_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:\nThree large organizations were part of the feminist movement in Chile, including the Club de Señoras of Santiago, founded by Amanda Labarca.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Feminism in Chile_sentence_17",
"Feminism in Chile_sentence_19"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7771,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.657",
"value": "?search="
},
{
"operation": "search",
"time": "16.566",
"value": "Feminist movement Chile"
},
{
"operation": "Now on",
"time": "19.549",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "25.127",
"value": "Feminism in Chile"
},
{
"operation": "Now on",
"time": "25.839",
"value": "Feminism in Chile"
},
{
"operation": "Page search",
"time": "96.727",
"value": "Santiago"
},
{
"operation": "search",
"time": "138.994",
"value": "Club de Senoras of Santiago"
},
{
"operation": "Now on",
"time": "140.485",
"value": "undefined"
},
{
"operation": "search",
"time": "154.175",
"value": "Amanda Labarca"
},
{
"operation": "Now on",
"time": "155.06",
"value": "Amanda Labarca"
},
{
"operation": "Highlighting",
"time": "220.771",
"value": "Amanda Labarca_sentence_9"
},
{
"operation": "Highlighting",
"time": "220.774",
"value": "Amanda Labarca_title"
},
{
"operation": "Highlighting",
"time": "220.776",
"value": "Amanda Labarca_section_1"
},
{
"operation": "Highlighting",
"time": "221.804",
"value": "Amanda Labarca_sentence_11"
},
{
"operation": "search",
"time": "254.411",
"value": "Club de Senoras de Santiago"
},
{
"operation": "Now on",
"time": "255.152",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "267.305",
"value": "club de senora de santiago"
},
{
"operation": "Now on",
"time": "268.203",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "278.216",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "278.9",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "280.441",
"value": "Feminism in Chile"
},
{
"operation": "Now on",
"time": "281.033",
"value": "Feminism in Chile"
},
{
"operation": "Highlighting",
"time": "314.311",
"value": "Feminism in Chile_sentence_17"
},
{
"operation": "Highlighting",
"time": "339.204",
"value": "Feminism in Chile_sentence_19"
},
{
"operation": "finish",
"time": "396.65",
"value": "finish"
}
]
}
|
exfever_845
|
Jane Austen's Emma is an adaptation of the 1815 novel of the same name, starring an English actress and model who has appeared in British costume dramas, including a 2000 period romantic drama film directed by James Ivory.
|
{
"evidence": [
"Emma (1996 TV film)",
"Kate Beckinsale",
"The Golden Bowl (film)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJane Austen's Emma is an adaptation of the 1815 novel of the same name, starring an English actress and model who has appeared in British costume dramas, including a 2000 period romantic drama film directed by James Ivory.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Emma (1996 TV film)",
"Kate Beckinsale",
"The Golden Bowl (film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7772,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7970
|
Black Panther is a superhero who became one of the best in the film Avengers in the Marvel Universe.
|
{
"evidence": [
"Black Panther (film)",
"Ryan Coogler"
],
"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:\nBlack Panther is a superhero who became one of the best in the film Avengers in the Marvel Universe.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Black Panther (film)",
"Ryan Coogler"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7773,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8439
|
30 Seconds to Mars is the 12th debut studio album distributed by Virgin Records.
|
{
"evidence": [
"30 Seconds to Mars (album)",
"Concept album"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n30 Seconds to Mars is the 12th debut studio album distributed by Virgin Records.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"30 Seconds to Mars (album)",
"Concept album"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7774,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
82727
|
Archer J.P. Martin quit working at the Wool Industries Research Association in Leeds a year before Richard L.M. Synge began.
|
{
"evidence": [
"Archer Martin_sentence_3",
"Archer Martin_sentence_11",
"Archer Martin_sentence_10"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nArcher J.P. Martin quit working at the Wool Industries Research Association in Leeds a year before Richard L.M. Synge began.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Archer Martin_sentence_3",
"Archer Martin_sentence_11",
"Archer Martin_sentence_10"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7775,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.667",
"value": "?search="
},
{
"operation": "search",
"time": "16.649",
"value": "Archer J.P. Martin"
},
{
"operation": "Now on",
"time": "17.422",
"value": "Archer Martin"
},
{
"operation": "Page search",
"time": "36.908",
"value": "wool"
},
{
"operation": "Highlighting",
"time": "51.178",
"value": "Archer Martin_sentence_3"
},
{
"operation": "Page search",
"time": "98.331",
"value": "wool"
},
{
"operation": "Page search",
"time": "148.357",
"value": "richard"
},
{
"operation": "Highlighting",
"time": "318.802",
"value": "Archer Martin_sentence_11"
},
{
"operation": "Highlighting",
"time": "341.021",
"value": "Archer Martin_sentence_10"
},
{
"operation": "finish",
"time": "351.534",
"value": "finish"
}
]
}
|
exfever_8858
|
Reds is a 1981 American historical drama film whose style of filmmaking is characterized by an extravagant setting and lavish costumes, accompanied by an expansive musical score.
|
{
"evidence": [
"Reds (film)",
"Epic film"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nReds is a 1981 American historical drama film whose style of filmmaking is characterized by an extravagant setting and lavish costumes, accompanied by an expansive musical score.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Reds (film)",
"Epic film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7776,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
55843
|
Alexandra Close is a journalist connected with Ethnic Media Services.
|
{
"evidence": [
"Sandy Close_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:\nAlexandra Close is a journalist connected with Ethnic Media Services.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Sandy Close_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7777,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.781",
"value": "?search="
},
{
"operation": "search",
"time": "75.659",
"value": "Alexandra Close"
},
{
"operation": "Now on",
"time": "76.844",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "80.404",
"value": "Sandy Close"
},
{
"operation": "Now on",
"time": "81.156",
"value": "Sandy Close"
},
{
"operation": "Highlighting",
"time": "84.459",
"value": "Sandy Close_sentence_0"
},
{
"operation": "finish",
"time": "99.158",
"value": "finish"
}
]
}
|
81899
|
Polish-German architect, socialist, and feminist Karola Bloch lived in the left-wing in Wilmersdorf and she finished her studies at the ETH Zurich.
|
{
"evidence": [
"Karola Bloch_sentence_0",
"Karola Bloch_sentence_11",
"Karola Bloch_sentence_13"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPolish-German architect, socialist, and feminist Karola Bloch lived in the left-wing in Wilmersdorf and she finished her studies at the ETH Zurich.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Karola Bloch_sentence_0",
"Karola Bloch_sentence_11",
"Karola Bloch_sentence_13"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7778,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.79",
"value": "?search="
},
{
"operation": "search",
"time": "8.34",
"value": "Karola Bloch"
},
{
"operation": "Now on",
"time": "9.201",
"value": "Karola Bloch"
},
{
"operation": "Highlighting",
"time": "14.936",
"value": "Karola Bloch_sentence_0"
},
{
"operation": "Highlighting",
"time": "33.235",
"value": "Karola Bloch_sentence_11"
},
{
"operation": "Highlighting",
"time": "39.89",
"value": "Karola Bloch_sentence_13"
},
{
"operation": "finish",
"time": "48.322",
"value": "finish"
}
]
}
|
exfever_1429
|
To Kill a Mockingbird was the first novel adapted into a Broadway stage play.
|
{
"evidence": [
"To Kill a Mockingbird",
"Atticus Finch"
],
"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:\nTo Kill a Mockingbird was the first novel adapted into a Broadway stage play.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"To Kill a Mockingbird",
"Atticus Finch"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 7779,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36776
|
Bibhuti Bhushan Nayak (born 1965) is an Indonesian journalist, who holds seven Guinness World Records, including the most number of double alternate squat thrusts, the most number of one-legged martial arts kicks, and the most number of squats in one minute performed on June 24, 2015.
|
{
"evidence": [
"Bibhuti Bhushan Nayak_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:\nBibhuti Bhushan Nayak (born 1965) is an Indonesian journalist, who holds seven Guinness World Records, including the most number of double alternate squat thrusts, the most number of one-legged martial arts kicks, and the most number of squats in one minute performed on June 24, 2015.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Bibhuti Bhushan Nayak_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7780,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.085",
"value": "?search="
},
{
"operation": "search",
"time": "31.017",
"value": "Bibhuti Bhushan Nayak"
},
{
"operation": "Now on",
"time": "31.848",
"value": "Bibhuti Bhushan Nayak"
},
{
"operation": "Highlighting",
"time": "37.534",
"value": "Bibhuti Bhushan Nayak_sentence_0"
},
{
"operation": "finish",
"time": "50.94",
"value": "finish"
}
]
}
|
58344
|
Arena Naucalpan 36th Anniversary Show is IWRG's longest-running show, and the main event was a Triangular de la Muerte match.
|
{
"evidence": [
"Arena Naucalpan 36th Anniversary Show_sentence_2",
"Arena Naucalpan 36th Anniversary Show_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:\nArena Naucalpan 36th Anniversary Show is IWRG's longest-running show, and the main event was a Triangular de la Muerte match.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Arena Naucalpan 36th Anniversary Show_sentence_2",
"Arena Naucalpan 36th Anniversary Show_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7781,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.98",
"value": "?search="
},
{
"operation": "search",
"time": "22.229",
"value": "Arena Naucalpan 36th Anniversary Show"
},
{
"operation": "Now on",
"time": "23.402",
"value": "Arena Naucalpan 36th Anniversary Show"
},
{
"operation": "Highlighting",
"time": "35.479",
"value": "Arena Naucalpan 36th Anniversary Show_sentence_2"
},
{
"operation": "Highlighting",
"time": "42.165",
"value": "Arena Naucalpan 36th Anniversary Show_sentence_3"
},
{
"operation": "finish",
"time": "201.135",
"value": "finish"
}
]
}
|
exfever_485
|
Lyme disease's most common sign of infection is an expanding red change in the human skin that affects its appearance.
|
{
"evidence": [
"Lyme disease",
"Rash"
],
"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:\nLyme disease's most common sign of infection is an expanding red change in the human skin that affects its appearance.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lyme disease",
"Rash"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 7782,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
73741
|
Frederick Gilbert Bourne was the first president of the Singer Manufacturing Company until today.
|
{
"evidence": [
"Frederick Gilbert Bourne_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:\nFrederick Gilbert Bourne was the first president of the Singer Manufacturing Company until today.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Frederick Gilbert Bourne_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 7783,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "13.421",
"value": "Ang Iibigin ay Ikaw Pa Rin"
},
{
"operation": "search",
"time": "22.772",
"value": "Frederick Gilbert Bourne"
},
{
"operation": "Now on",
"time": "24.313",
"value": "Frederick Gilbert Bourne"
},
{
"operation": "Highlighting",
"time": "29.019",
"value": "Frederick Gilbert Bourne_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "38.844",
"value": "Frederick Gilbert Bourne_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "40.062",
"value": "Frederick Gilbert Bourne_sentence_1"
},
{
"operation": "finish",
"time": "62.805",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.