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_3948
|
Adele is a singer and songwriter who signed a record deal with a British independent record label founded in 1979 by Tim Palmer and Nick Halkes.
|
{
"evidence": [
"Adele",
"XL Recordings"
],
"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:\nAdele is a singer and songwriter who signed a record deal with a British independent record label founded in 1979 by Tim Palmer and Nick Halkes.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Adele",
"XL Recordings"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1784,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7568
|
Little Mix is a music act featuring several male singers who generally harmonize together.
|
{
"evidence": [
"Little Mix",
"Girl group"
],
"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:\nLittle Mix is a music act featuring several male singers who generally harmonize together.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Little Mix",
"Girl group"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1785,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8058
|
Guru Nanak, the founder of Sikhism, was born in the village of Rāi Bhōi dī Talwandī.
|
{
"evidence": [
"Sikhism",
"Indian subcontinent",
"Indian Ocean"
],
"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:\nGuru Nanak, the founder of Sikhism, was born in the village of Rāi Bhōi dī Talwandī.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sikhism",
"Indian subcontinent",
"Indian Ocean"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1786,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
79733
|
Bee Building is found in Alabama, Canada.
|
{
"evidence": [
"Bee Building_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:\nBee Building is found in Alabama, Canada.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Bee Building_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1787,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.83",
"value": "?search="
},
{
"operation": "search",
"time": "20.384",
"value": "Bee Building"
},
{
"operation": "Now on",
"time": "21.545",
"value": "Bee Building"
},
{
"operation": "Highlighting",
"time": "26.79",
"value": "Bee Building_sentence_0"
},
{
"operation": "finish",
"time": "36.021",
"value": "finish"
}
]
}
|
exfever_5175
|
Mark Richard Hamill is an American actor best known for his portrayal of Luke Skywalker in the Star Wars film series as well as his appearances in the 1980 American adventure comedy film directed by Matthew Robbins.
|
{
"evidence": [
"Mark Hamill",
"Corvette Summer"
],
"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:\nMark Richard Hamill is an American actor best known for his portrayal of Luke Skywalker in the Star Wars film series as well as his appearances in the 1980 American adventure comedy film directed by Matthew Robbins.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mark Hamill",
"Corvette Summer"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1788,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8020
|
Shantel VanSanten played the wife of the lead character, Bob Lee Swagger, in the USA Network television series based on the 2007 film of the same name.
|
{
"evidence": [
"Shantel VanSanten",
"Shooter (TV series)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nShantel VanSanten played the wife of the lead character, Bob Lee Swagger, in the USA Network television series based on the 2007 film of the same name.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Shantel VanSanten",
"Shooter (TV series)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1789,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2333
|
The Diary of a Young Girl is the first book of writings that has since been translated into over 70 languages.
|
{
"evidence": [
"Anne Frank",
"The Holocaust",
"German-occupied Europe"
],
"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 Diary of a Young Girl is the first book of writings that has since been translated into over 70 languages.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Anne Frank",
"The Holocaust",
"German-occupied Europe"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1790,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4256
|
Karl Urban is a New Zealand actor who portrayed "Bones", a character in the American science-fiction franchise Star Trek.
|
{
"evidence": [
"Karl Urban",
"Leonard McCoy"
],
"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:\nKarl Urban is a New Zealand actor who portrayed \"Bones\", a character in the American science-fiction franchise Star Trek.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Karl Urban",
"Leonard McCoy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1791,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8552
|
Frederick Trump was a German–American barber who traveled to the Yukon Territory and made his fortune by operating a restaurant and a brothel for miners during the migration of an estimated 100,000 prospectors to the region of the Yukon Territory in northwestern Australia.
|
{
"evidence": [
"Frederick Trump",
"Klondike Gold Rush",
"Klondike, Yukon"
],
"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:\nFrederick Trump was a German–American barber who traveled to the Yukon Territory and made his fortune by operating a restaurant and a brothel for miners during the migration of an estimated 100,000 prospectors to the region of the Yukon Territory in northwestern Australia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Frederick Trump",
"Klondike Gold Rush",
"Klondike, Yukon"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1792,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1310
|
Detroict is nicknamed the 'Motor City.
|
{
"evidence": [
"Detroit",
"Michigan"
],
"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:\nDetroict is nicknamed the 'Motor City.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Detroit",
"Michigan"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1793,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7991
|
Felicity Jones is an actress who played the role of Emma Grundy on the national broadcaster of a sovereign country in north-western Europe, off the north--western coast of the European mainland.
|
{
"evidence": [
"Felicity Jones",
"BBC",
"United Kingdom"
],
"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:\nFelicity Jones is an actress who played the role of Emma Grundy on the national broadcaster of a sovereign country in north-western Europe, off the north--western coast of the European mainland.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Felicity Jones",
"BBC",
"United Kingdom"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1794,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
46886
|
In the 1960s and the 1970s, Germany and the Axis powers defeated the United States and the key American cities were destroyed by bombs.
|
{
"evidence": [
"Axis powers_sentence_3",
"World War II_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:\nIn the 1960s and the 1970s, Germany and the Axis powers defeated the United States and the key American cities were destroyed by bombs.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Axis powers_sentence_3",
"World War II_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1795,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.79",
"value": "?search="
},
{
"operation": "search",
"time": "40.649",
"value": "Axis powers"
},
{
"operation": "Now on",
"time": "43.728",
"value": "Axis powers"
},
{
"operation": "Highlighting",
"time": "58.777",
"value": "Axis powers_sentence_3"
},
{
"operation": "hyperlink",
"time": "60.132",
"value": "World War II"
},
{
"operation": "Now on",
"time": "62.531",
"value": "World War II"
},
{
"operation": "Highlighting",
"time": "65.057",
"value": "World War II_sentence_2"
},
{
"operation": "finish",
"time": "87.15",
"value": "finish"
}
]
}
|
45197
|
The town of Misaje is located in the Northwest Region of the land-locked country of Cameroon.
|
{
"evidence": [
"Misaje_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe town of Misaje is located in the Northwest Region of the land-locked country of Cameroon.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Misaje_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1796,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.381",
"value": "?search="
},
{
"operation": "search",
"time": "4.219",
"value": "Misaje"
},
{
"operation": "Now on",
"time": "4.941",
"value": "Misaje"
},
{
"operation": "Highlighting",
"time": "7.335",
"value": "Misaje_sentence_0"
},
{
"operation": "hyperlink",
"time": "9.049",
"value": "Cameroon"
},
{
"operation": "Now on",
"time": "11.001",
"value": "Cameroon"
},
{
"operation": "back-button-clicked",
"time": "23.844",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "23.942",
"value": "Misaje"
},
{
"operation": "hyperlink",
"time": "30.532",
"value": "Cameroon"
},
{
"operation": "Now on",
"time": "31.101",
"value": "Cameroon"
},
{
"operation": "back-button-clicked",
"time": "90.35",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "90.45",
"value": "Misaje"
},
{
"operation": "finish",
"time": "100.362",
"value": "finish"
}
]
}
|
exfever_940
|
Heather Graham played a supporting role in the television series that premiered on DBC.
|
{
"evidence": [
"Heather Graham",
"Twin Peaks"
],
"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:\nHeather Graham played a supporting role in the television series that premiered on DBC.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Heather Graham",
"Twin Peaks"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1797,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
34529
|
The display of Xerox 820 is a 24-line CRT, it has a compact 48-character keyboard and its communication ports are two 25-pin RS-232.
|
{
"evidence": [
"Xerox 820_sentence_11"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe display of Xerox 820 is a 24-line CRT, it has a compact 48-character keyboard and its communication ports are two 25-pin RS-232.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Xerox 820_sentence_11"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1798,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.733",
"value": "?search="
},
{
"operation": "search",
"time": "10.719",
"value": "Xerox 820"
},
{
"operation": "Now on",
"time": "12.979",
"value": "Xerox 820"
},
{
"operation": "Page search",
"time": "23.17",
"value": "display"
},
{
"operation": "Highlighting",
"time": "40.2",
"value": "Xerox 820_sentence_9"
},
{
"operation": "Page search",
"time": "63.236",
"value": "keyboard"
},
{
"operation": "Highlighting",
"time": "98.748",
"value": "Xerox 820_sentence_11"
},
{
"operation": "finish",
"time": "113.49",
"value": "finish"
}
]
}
|
exfever_6549
|
Seth Woodbury MacFarlane is an American actor and co-creator of the American adult animated sitcom created by an American producer and writer who was one of the creators of American Dad! along with Seth MacFarlane.
|
{
"evidence": [
"Seth MacFarlane",
"American Dad!",
"Matt Weitzman"
],
"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:\nSeth Woodbury MacFarlane is an American actor and co-creator of the American adult animated sitcom created by an American producer and writer who was one of the creators of American Dad! along with Seth MacFarlane.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Seth MacFarlane",
"American Dad!",
"Matt Weitzman"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1799,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8354
|
Alan Walbridge Ladd was often paired with Veronica Lake in film noir, such as in a film directed by Frank Tuttle, starring a Korean stage and film actor best known for his villainous performances in films such as I Wake Up Screaming.
|
{
"evidence": [
"Alan Ladd",
"This Gun for Hire",
"Laird Cregar"
],
"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:\nAlan Walbridge Ladd was often paired with Veronica Lake in film noir, such as in a film directed by Frank Tuttle, starring a Korean stage and film actor best known for his villainous performances in films such as I Wake Up Screaming.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alan Ladd",
"This Gun for Hire",
"Laird Cregar"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1800,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2624
|
Page Hamilton is an American guitarist, singer, songwriter, and record producer whose work has been in rock music whose subgenre is typified by heavy use of aggressive vocals, distorted electric guitars, bass guitar, and drums, sometimes accompanied by keyboards.
|
{
"evidence": [
"Page Hamilton",
"Hard rock"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPage Hamilton is an American guitarist, singer, songwriter, and record producer whose work has been in rock music whose subgenre is typified by heavy use of aggressive vocals, distorted electric guitars, bass guitar, and drums, sometimes accompanied by keyboards.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Page Hamilton",
"Hard rock"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1801,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
82724
|
Karen, an independent artist who currently works part-time as the receptionist at a cemetery, a place where the remains of dead people are buried or otherwise interred, wanted to apply for art college in 2006, but missed the deadline.
|
{
"evidence": [
"Planet Karen_sentence_22",
"Planet Karen_sentence_25",
"Cemetery_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:\nKaren, an independent artist who currently works part-time as the receptionist at a cemetery, a place where the remains of dead people are buried or otherwise interred, wanted to apply for art college in 2006, but missed the deadline.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Planet Karen_sentence_22",
"Planet Karen_sentence_25",
"Cemetery_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1802,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.107",
"value": "?search="
},
{
"operation": "search",
"time": "29.48",
"value": "independent artist receptionist at cemetary"
},
{
"operation": "Now on",
"time": "31.567",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "36.753",
"value": "Planet Karen"
},
{
"operation": "Now on",
"time": "37.556",
"value": "Planet Karen"
},
{
"operation": "Highlighting",
"time": "53.472",
"value": "Planet Karen_sentence_22"
},
{
"operation": "Highlighting",
"time": "58.883",
"value": "Planet Karen_sentence_25"
},
{
"operation": "hyperlink",
"time": "60.376",
"value": "cemetery"
},
{
"operation": "Now on",
"time": "62.092",
"value": "Cemetery"
},
{
"operation": "Highlighting",
"time": "72.423",
"value": "Cemetery_sentence_1"
},
{
"operation": "finish",
"time": "83.205",
"value": "finish"
}
]
}
|
exfever_6029
|
J. J. Cale was an African singer and songwriter who was one of the originators of a loose genre drawing on a music genre and musical form that originated in the Deep South of the United States by African-Americans from roots in African-American work songs and spirituals.
|
{
"evidence": [
"J. J. Cale",
"Blues"
],
"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:\nJ. J. Cale was an African singer and songwriter who was one of the originators of a loose genre drawing on a music genre and musical form that originated in the Deep South of the United States by African-Americans from roots in African-American work songs and spirituals.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"J. J. Cale",
"Blues"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1803,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
24421
|
Evie Hone, born on April 22, 1894 at Roebuck Grove, Ireland, was an Irish stained glass artist and painter who was considered to be an early pioneer of Dadaism.
|
{
"evidence": [
"Evie Hone_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:\nEvie Hone, born on April 22, 1894 at Roebuck Grove, Ireland, was an Irish stained glass artist and painter who was considered to be an early pioneer of Dadaism.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Evie Hone_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1804,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.624",
"value": "?search="
},
{
"operation": "search",
"time": "12.177",
"value": "Evie Hone"
},
{
"operation": "Now on",
"time": "12.881",
"value": "Evie Hone"
},
{
"operation": "Highlighting",
"time": "21.868",
"value": "Evie Hone_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "22.922",
"value": "Evie Hone_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "22.931",
"value": "Evie Hone_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "22.931",
"value": "Evie Hone_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "31.703",
"value": "Evie Hone_sentence_0"
},
{
"operation": "Highlighting",
"time": "42.644",
"value": "Evie Hone_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "53.173",
"value": "Evie Hone_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "55.229",
"value": "Evie Hone_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "55.232",
"value": "Evie Hone_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "55.233",
"value": "Evie Hone_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "63.37",
"value": "finish"
}
]
}
|
exfever_6364
|
As You Like It is a 2006 romance film written and directed by a British filmmaker who trained at the Royal Academy of Dramatic Arts in London.
|
{
"evidence": [
"As You Like It (2006 film)",
"Kenneth Branagh"
],
"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:\nAs You Like It is a 2006 romance film written and directed by a British filmmaker who trained at the Royal Academy of Dramatic Arts in London.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"As You Like It (2006 film)",
"Kenneth Branagh"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1805,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
89380
|
Jair Bernal was born on February 12, 1968 in Tunja, Central Boyacá Province, and is now a retired rider.
|
{
"evidence": [
"Jair Bernal_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:\nJair Bernal was born on February 12, 1968 in Tunja, Central Boyacá Province, and is now a retired rider.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Jair Bernal_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1806,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.099",
"value": "?search="
},
{
"operation": "search",
"time": "11.806",
"value": "Jair Bernal"
},
{
"operation": "Now on",
"time": "12.981",
"value": "Jair Bernal"
},
{
"operation": "Highlighting",
"time": "20.286",
"value": "Jair Bernal_sentence_1"
},
{
"operation": "finish",
"time": "71.061",
"value": "finish"
}
]
}
|
exfever_7353
|
Bedazzled was a comic retelling of the protagonist of a classic German legend.
|
{
"evidence": [
"Bedazzled (2000 film)",
"Faust"
],
"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:\nBedazzled was a comic retelling of the protagonist of a classic German legend.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bedazzled (2000 film)",
"Faust"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1807,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5652
|
The Incredible Hulk is a 2008 American superhero film directed by a Filipino film director and producer, whose films include the first two Transporter films.
|
{
"evidence": [
"The Incredible Hulk (film)",
"Louis Leterrier"
],
"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 Incredible Hulk is a 2008 American superhero film directed by a Filipino film director and producer, whose films include the first two Transporter films.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Incredible Hulk (film)",
"Louis Leterrier"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1808,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8445
|
Big Boi is an American rapper born on February 1, 1975, who released his second studio album on December 11, 2012, through Purple Ribbon Records and Def Jam Recordings.
|
{
"evidence": [
"Big Boi",
"Vicious Lies and Dangerous Rumors"
],
"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:\nBig Boi is an American rapper born on February 1, 1975, who released his second studio album on December 11, 2012, through Purple Ribbon Records and Def Jam Recordings.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Big Boi",
"Vicious Lies and Dangerous Rumors"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1809,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
49704
|
Robinson Crusoe Island (native name Isla Róbinson Crusoe) is part of the Juan Fernández Islands, a sparsely inhabited island group in the South Pacific Ocean reliant on tourism and fishing.
|
{
"evidence": [
"Robinson Crusoe Island_sentence_3",
"Robinson Crusoe Island_sentence_4",
"Robinson Crusoe Island_sentence_8",
"Robinson Crusoe Island_sentence_57"
],
"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:\nRobinson Crusoe Island (native name Isla Róbinson Crusoe) is part of the Juan Fernández Islands, a sparsely inhabited island group in the South Pacific Ocean reliant on tourism and fishing.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Robinson Crusoe Island_sentence_3",
"Robinson Crusoe Island_sentence_4",
"Robinson Crusoe Island_sentence_8",
"Robinson Crusoe Island_sentence_57"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1810,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.744",
"value": "?search="
},
{
"operation": "search",
"time": "48.298",
"value": "Robinson Crusoe Island"
},
{
"operation": "Now on",
"time": "49.292",
"value": "Robinson Crusoe Island"
},
{
"operation": "Highlighting",
"time": "99.179",
"value": "Robinson Crusoe Island_sentence_3"
},
{
"operation": "Highlighting",
"time": "100.655",
"value": "Robinson Crusoe Island_sentence_4"
},
{
"operation": "Highlighting",
"time": "103.409",
"value": "Robinson Crusoe Island_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "103.872",
"value": "Robinson Crusoe Island_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "103.889",
"value": "Robinson Crusoe Island_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "103.89",
"value": "Robinson Crusoe Island_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "103.89",
"value": "Robinson Crusoe Island_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "116.161",
"value": "Robinson Crusoe Island_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "116.171",
"value": "Robinson Crusoe Island_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "116.173",
"value": "Robinson Crusoe Island_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "116.175",
"value": "Robinson Crusoe Island_header_cell_0_0_0"
},
{
"operation": "Page search",
"time": "213.766",
"value": "fishing"
},
{
"operation": "Page search",
"time": "307.293",
"value": "fish"
},
{
"operation": "Highlighting",
"time": "314.372",
"value": "Robinson Crusoe Island_sentence_8"
},
{
"operation": "Highlighting",
"time": "347.613",
"value": "Robinson Crusoe Island_sentence_57"
},
{
"operation": "hyperlink",
"time": "376.495",
"value": "Juan Fernández Islands"
},
{
"operation": "Now on",
"time": "377.726",
"value": "Juan Fernández Islands"
},
{
"operation": "back-button-clicked",
"time": "410.222",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "410.343",
"value": "Robinson Crusoe Island"
},
{
"operation": "finish",
"time": "433.027",
"value": "finish"
}
]
}
|
68586
|
Brian Irr, a bronze medalist in karate at the 2015 Pan American Games, never won a gold medal.
|
{
"evidence": [
"Brian Irr_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:\nBrian Irr, a bronze medalist in karate at the 2015 Pan American Games, never won a gold medal.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Brian Irr_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1811,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.032",
"value": "?search="
},
{
"operation": "search",
"time": "10.629",
"value": "Contains...Brian Irr,"
},
{
"operation": "Now on",
"time": "12.163",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "16.487",
"value": "Brian Irr"
},
{
"operation": "Now on",
"time": "17.117",
"value": "Brian Irr"
},
{
"operation": "Highlighting",
"time": "19.892",
"value": "Brian Irr_sentence_1"
},
{
"operation": "finish",
"time": "31.897",
"value": "finish"
}
]
}
|
24520
|
Karim Allawi played in the 1986 Asian Games in March and the score was 5-0.
|
{
"evidence": [
"1986 Asian Games_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:\nKarim Allawi played in the 1986 Asian Games in March and the score was 5-0.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1986 Asian Games_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1812,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.735",
"value": "?search="
},
{
"operation": "search",
"time": "15.542",
"value": "Karim Allawi"
},
{
"operation": "Now on",
"time": "16.651",
"value": "Karim Allawi"
},
{
"operation": "Page search",
"time": "31.189",
"value": "1986"
},
{
"operation": "hyperlink",
"time": "63.089",
"value": "1986 Asian Games"
},
{
"operation": "Now on",
"time": "64",
"value": "Football at the 1986 Asian Games"
},
{
"operation": "Highlighting",
"time": "187.164",
"value": "Football at the 1986 Asian Games_sentence_0"
},
{
"operation": "Highlighting",
"time": "187.165",
"value": "Football at the 1986 Asian Games_title"
},
{
"operation": "hyperlink",
"time": "204.463",
"value": "1986 Asian Games"
},
{
"operation": "Now on",
"time": "205.905",
"value": "1986 Asian Games"
},
{
"operation": "Highlighting",
"time": "248.362",
"value": "1986 Asian Games_sentence_0"
},
{
"operation": "finish",
"time": "262.761",
"value": "finish"
}
]
}
|
81447
|
Nacereddine Khoualed has been capped for Algeria at the Under-20, he is an Algerian football defense player and was in 1986 in Biskra, Algeria.
|
{
"evidence": [
"Nacereddine Khoualed_sentence_0",
"Nacereddine Khoualed_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:\nNacereddine Khoualed has been capped for Algeria at the Under-20, he is an Algerian football defense player and was in 1986 in Biskra, Algeria.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Nacereddine Khoualed_sentence_0",
"Nacereddine Khoualed_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1813,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.331",
"value": "?search="
},
{
"operation": "search",
"time": "19.833",
"value": "Nacereddine Khoualed"
},
{
"operation": "Now on",
"time": "21.01",
"value": "Nacereddine Khoualed"
},
{
"operation": "Highlighting",
"time": "24.542",
"value": "Nacereddine Khoualed_sentence_0"
},
{
"operation": "Highlighting",
"time": "30.938",
"value": "Nacereddine Khoualed_sentence_1"
},
{
"operation": "finish",
"time": "74.218",
"value": "finish"
}
]
}
|
exfever_1928
|
Pulp Fiction is John Travolta's first black comedy crime film.
|
{
"evidence": [
"Pulp Fiction",
"Quentin Tarantino"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPulp Fiction is John Travolta's first black comedy crime film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pulp Fiction",
"Quentin Tarantino"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1814,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7616
|
Taal is a 1999 Indian Hindi musical romantic drama film directed by an Indian film director, known for his works predominantly in Thai cinema.
|
{
"evidence": [
"Taal (film)",
"Subhash Ghai"
],
"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:\nTaal is a 1999 Indian Hindi musical romantic drama film directed by an Indian film director, known for his works predominantly in Thai cinema.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Taal (film)",
"Subhash Ghai"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1815,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4132
|
The Bounty is a 1984 British historical drama film produced by Bernard Williams with an Italian-American executive producer who was one of the producers who brought Italian cinema to the international scene.
|
{
"evidence": [
"The Bounty (1984 film)",
"Dino De Laurentiis"
],
"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 Bounty is a 1984 British historical drama film produced by Bernard Williams with an Italian-American executive producer who was one of the producers who brought Italian cinema to the international scene.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Bounty (1984 film)",
"Dino De Laurentiis"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1816,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8393
|
Anderson Hays Cooper is an African broadcast journalist who is the primary anchor of an American television news show on CNN and CNN International, hosted by CNN journalist and news anchor Anderson Cooper.
|
{
"evidence": [
"Anderson Cooper",
"Anderson Cooper 360°"
],
"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:\nAnderson Hays Cooper is an African broadcast journalist who is the primary anchor of an American television news show on CNN and CNN International, hosted by CNN journalist and news anchor Anderson Cooper.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Anderson Cooper",
"Anderson Cooper 360°"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1817,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6055
|
Helen Hunt starred in the romantic comedy As Good as It Gets as Carol Connelly, where she won a prestigious award that was given to an actress who has delivered an outstanding performance in a leading role in a film that is given annually by the professional honorary organization with the stated goal of advancing the arts and sciences of motion pictures.
|
{
"evidence": [
"Helen Hunt",
"Academy Award for Best Actress",
"Academy of Motion Picture Arts and Sciences"
],
"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:\nHelen Hunt starred in the romantic comedy As Good as It Gets as Carol Connelly, where she won a prestigious award that was given to an actress who has delivered an outstanding performance in a leading role in a film that is given annually by the professional honorary organization with the stated goal of advancing the arts and sciences of motion pictures.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Helen Hunt",
"Academy Award for Best Actress",
"Academy of Motion Picture Arts and Sciences"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1818,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36343
|
SkyHigh Mount Dandenong (formerly Mount Dandenong Observatory) is a restaurant that can be found in the North America.
|
{
"evidence": [
"SkyHigh Mount Dandenong_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:\nSkyHigh Mount Dandenong (formerly Mount Dandenong Observatory) is a restaurant that can be found in the North America.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"SkyHigh Mount Dandenong_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1819,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.711",
"value": "?search="
},
{
"operation": "search",
"time": "14.717",
"value": "SkyHigh Mount Dandenong"
},
{
"operation": "Now on",
"time": "15.369",
"value": "SkyHigh Mount Dandenong"
},
{
"operation": "Page search",
"time": "23.906",
"value": "mount"
},
{
"operation": "Page search",
"time": "40.277",
"value": "observatory"
},
{
"operation": "Highlighting",
"time": "56.557",
"value": "SkyHigh Mount Dandenong_sentence_6"
},
{
"operation": "Highlighting",
"time": "56.559",
"value": "SkyHigh Mount Dandenong_section_1"
},
{
"operation": "Highlighting",
"time": "73.291",
"value": "SkyHigh Mount Dandenong_sentence_0"
},
{
"operation": "finish",
"time": "189.923",
"value": "finish"
}
]
}
|
exfever_2636
|
Jacki Weaver is an Australian actress who is known internationally for her performance in the 2010 Australian neo-noir crime drama film written and directed by David Michôd.
|
{
"evidence": [
"Jacki Weaver",
"Animal Kingdom (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:\nJacki Weaver is an Australian actress who is known internationally for her performance in the 2010 Australian neo-noir crime drama film written and directed by David Michôd.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jacki Weaver",
"Animal Kingdom (film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1820,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4847
|
Bonnie Hunt is an American actress known for her voice work in the 2006 American computer-animated action-adventure buddy cop comedy film produced by Walt Disney Animation Studios.
|
{
"evidence": [
"Bonnie Hunt",
"Zootopia"
],
"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:\nBonnie Hunt is an American actress known for her voice work in the 2006 American computer-animated action-adventure buddy cop comedy film produced by Walt Disney Animation Studios.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bonnie Hunt",
"Zootopia"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1821,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
80650
|
Battle of Dry Lake, located in the Lava Beds, in southeastern California, Siskiyou and Modoc counties, happened on May 10, 1873 resulting to United States' victory.
|
{
"evidence": [
"Lava Beds National Monument_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:\nBattle of Dry Lake, located in the Lava Beds, in southeastern California, Siskiyou and Modoc counties, happened on May 10, 1873 resulting to United States' victory.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Lava Beds National Monument_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1822,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.811",
"value": "?search="
},
{
"operation": "search",
"time": "12.816",
"value": "Battle of Dry Lake"
},
{
"operation": "Now on",
"time": "13.638",
"value": "Battle of Dry Lake"
},
{
"operation": "Highlighting",
"time": "27.525",
"value": "Battle of Dry Lake_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "27.527",
"value": "Battle of Dry Lake_title"
},
{
"operation": "Highlighting",
"time": "29.244",
"value": "Battle of Dry Lake_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "29.254",
"value": "Battle of Dry Lake_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "29.255",
"value": "Battle of Dry Lake_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "52.444",
"value": "Battle of Dry Lake_cell_0_1_0"
},
{
"operation": "hyperlink",
"time": "55.55",
"value": "Lava Beds"
},
{
"operation": "Now on",
"time": "56.339",
"value": "Lava Beds National Monument"
},
{
"operation": "Highlighting",
"time": "61.193",
"value": "Lava Beds National Monument_sentence_2"
},
{
"operation": "finish",
"time": "79.504",
"value": "finish"
}
]
}
|
exfever_5228
|
Kirk Douglas died at his home in Beverly Hills, Calif. was announced in a statement posted by his son Michael Douglas on his Facebook page.
|
{
"evidence": [
"Seven Days in May",
"Fredric March"
],
"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:\nKirk Douglas died at his home in Beverly Hills, Calif. was announced in a statement posted by his son Michael Douglas on his Facebook page.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Seven Days in May",
"Fredric March"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1823,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4170
|
Caitlyn Marie Jenner is an American media personality who won a silver medal in the decathlon in the leading international sporting events featuring summer and winter sports competitions, which are normally held every four years, alternating between the Winter and Games of the Olympiad, a major international multi-sport event normally held once every four years.
|
{
"evidence": [
"Caitlyn Jenner",
"Olympic Games",
"Summer Olympic Games"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCaitlyn Marie Jenner is an American media personality who won a silver medal in the decathlon in the leading international sporting events featuring summer and winter sports competitions, which are normally held every four years, alternating between the Winter and Games of the Olympiad, a major international multi-sport event normally held once every four years.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Caitlyn Jenner",
"Olympic Games",
"Summer Olympic Games"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1824,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
34584
|
Annie Dillard, who grew up in Pittsburgh, Pennsylvania, attended Hollins College in Virginia, where she not only received bachelor's and master's degree but also met her husband, creative writing professor Richard Henry Wilde Dillard.
|
{
"evidence": [
"Annie Dillard_sentence_18",
"Annie Dillard_sentence_24",
"Annie Dillard_sentence_5",
"R. H. W. Dillard_sentence_12"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAnnie Dillard, who grew up in Pittsburgh, Pennsylvania, attended Hollins College in Virginia, where she not only received bachelor's and master's degree but also met her husband, creative writing professor Richard Henry Wilde Dillard.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Annie Dillard_sentence_18",
"Annie Dillard_sentence_24",
"Annie Dillard_sentence_5",
"R. H. W. Dillard_sentence_12"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1825,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.893",
"value": "?search="
},
{
"operation": "search",
"time": "6.188",
"value": "Annie Dillard"
},
{
"operation": "Now on",
"time": "7.072",
"value": "Annie Dillard"
},
{
"operation": "Page search",
"time": "14.172",
"value": "grew up"
},
{
"operation": "Page search",
"time": "42.216",
"value": "Hollins"
},
{
"operation": "Highlighting",
"time": "48.436",
"value": "Annie Dillard_sentence_18"
},
{
"operation": "Highlighting",
"time": "90.302",
"value": "Annie Dillard_sentence_24"
},
{
"operation": "hyperlink",
"time": "114.719",
"value": "R. Annie Dillard_sentence_19\n\nH. W. Dillard"
},
{
"operation": "Now on",
"time": "115.223",
"value": "R. H. W. Dillard"
},
{
"operation": "back-button-clicked",
"time": "127.815",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "127.898",
"value": "Annie Dillard"
},
{
"operation": "Highlighting",
"time": "142.155",
"value": "Annie Dillard_sentence_19Annie Dillard_sentence_20"
},
{
"operation": "Highlighting",
"time": "151.065",
"value": "Annie Dillard_sentence_27"
},
{
"operation": "Highlighting deleted",
"time": "170.057",
"value": "Annie Dillard_sentence_27"
},
{
"operation": "Highlighting deleted",
"time": "177.548",
"value": "Annie Dillard_sentence_19Annie Dillard_sentence_20"
},
{
"operation": "hyperlink",
"time": "194.422",
"value": "R. Annie Dillard_sentence_19\n\nH. W. Dillard"
},
{
"operation": "Now on",
"time": "194.751",
"value": "R. H. W. Dillard"
},
{
"operation": "Page search",
"time": "213.118",
"value": "wife"
},
{
"operation": "Page search",
"time": "215.759",
"value": "married"
},
{
"operation": "Page search",
"time": "218.179",
"value": "annue"
},
{
"operation": "Page search",
"time": "219.239",
"value": "annie"
},
{
"operation": "back-button-clicked",
"time": "230.443",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "230.524",
"value": "Annie Dillard"
},
{
"operation": "Highlighting",
"time": "313.329",
"value": "Annie Dillard_sentence_5"
},
{
"operation": "hyperlink",
"time": "342.716",
"value": "R. Annie Dillard_sentence_19\n\nH. W. Dillard"
},
{
"operation": "Now on",
"time": "343.399",
"value": "R. H. W. Dillard"
},
{
"operation": "Highlighting",
"time": "346.959",
"value": "R. H. W. Dillard_sentence_12"
},
{
"operation": "finish",
"time": "371.65",
"value": "finish"
}
]
}
|
exfever_8179
|
Green Day is an American punk rock band from the East Bay of California, whose lead vocalist is Billie Joe Armstrong, and they released their second studio album through Lookout! Records on September 17, 1991.
|
{
"evidence": [
"Green Day",
"Kerplunk (album)"
],
"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:\nGreen Day is an American punk rock band from the East Bay of California, whose lead vocalist is Billie Joe Armstrong, and they released their second studio album through Lookout! Records on September 17, 1991.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Green Day",
"Kerplunk (album)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1826,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4286
|
Filmfare is an Indian fortnightly magazine that uses a West Germanic language of the Indo-European language family named after the main Germanic peoples who settled in Great Britain in the post-Roman period.
|
{
"evidence": [
"Filmfare",
"English language",
"Angles"
],
"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:\nFilmfare is an Indian fortnightly magazine that uses a West Germanic language of the Indo-European language family named after the main Germanic peoples who settled in Great Britain in the post-Roman period.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Filmfare",
"English language",
"Angles"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1827,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8158
|
Jamie Foxx is an American actor widely known for his role as Ray Charles in the movie Ray, for which he won the award presented annually by the Academy of Motion Picture Arts and Sciences.
|
{
"evidence": [
"Jamie Foxx",
"Academy Award for Best Actor"
],
"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 American actor widely known for his role as Ray Charles in the movie Ray, for which he won the award presented annually by the Academy of Motion Picture Arts and Sciences.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jamie Foxx",
"Academy Award for Best Actor"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1828,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
28259
|
John Nicholas Brown II was never a secretary.
|
{
"evidence": [
"John Nicholas Brown II_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:\nJohn Nicholas Brown II was never a secretary.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"John Nicholas Brown II_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1829,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.502",
"value": "?search="
},
{
"operation": "search",
"time": "7.352",
"value": "John Nicholas Brown II"
},
{
"operation": "Now on",
"time": "8.853",
"value": "John Nicholas Brown II"
},
{
"operation": "Highlighting",
"time": "12.976",
"value": "John Nicholas Brown II_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "14.057",
"value": "Assistant Secretary of the Navy"
},
{
"operation": "Now on",
"time": "15.477",
"value": "Assistant Secretary of the Navy (AIR)"
},
{
"operation": "back-button-clicked",
"time": "16.697",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "17.025",
"value": "John Nicholas Brown II"
},
{
"operation": "Highlighting deleted",
"time": "18.237",
"value": "John Nicholas Brown II_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "19.21",
"value": "John Nicholas Brown II_sentence_0"
},
{
"operation": "Page search",
"time": "26.301",
"value": "secretary"
},
{
"operation": "finish",
"time": "45.042",
"value": "finish"
}
]
}
|
3071
|
2009 Norwegian Football Cup Final was played between Molde FK and Aalesunds FK on 8 November 2009 on Ullevaal Stadion in Oslo, Norway.
|
{
"evidence": [
"2009 Norwegian Football Cup Final_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:\n2009 Norwegian Football Cup Final was played between Molde FK and Aalesunds FK on 8 November 2009 on Ullevaal Stadion in Oslo, Norway.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"2009 Norwegian Football Cup Final_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1830,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "10.784",
"value": "2009 Norwegian Football Cup Final"
},
{
"operation": "Highlighting",
"time": "13.805",
"value": "2009 Norwegian Football Cup Final_sentence_0"
},
{
"operation": "finish",
"time": "38.15",
"value": "finish"
}
]
}
|
73487
|
New South Wales Team won against South Canterbury on August 23, 1923 with a score of 23-10.
|
{
"evidence": [
"New South Wales cricket team_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNew South Wales Team won against South Canterbury on August 23, 1923 with a score of 23-10.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"New South Wales cricket team_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1831,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.659",
"value": "undefined"
},
{
"operation": "search",
"time": "19.889",
"value": "Contains...1923 New South Wales Team"
},
{
"operation": "Now on",
"time": "22.313",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "32.6",
"value": "New South Wales cricket team"
},
{
"operation": "Now on",
"time": "34.025",
"value": "New South Wales cricket team"
},
{
"operation": "Highlighting",
"time": "40.496",
"value": "New South Wales cricket team_sentence_0"
},
{
"operation": "search",
"time": "158.826",
"value": "New South Wales cricket team"
},
{
"operation": "Now on",
"time": "159.995",
"value": "New South Wales cricket team"
},
{
"operation": "Page search",
"time": "212.81",
"value": "1923"
},
{
"operation": "search",
"time": "216.451",
"value": "New South Wales Cricket Team"
},
{
"operation": "Now on",
"time": "218.576",
"value": "New South Wales cricket team"
},
{
"operation": "hyperlink",
"time": "249.36",
"value": "Sheffield Shield"
},
{
"operation": "Now on",
"time": "250.775",
"value": "Sheffield Shield"
},
{
"operation": "search",
"time": "279.96",
"value": "1923 Sheffield Shield"
},
{
"operation": "Now on",
"time": "281.728",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "284.974",
"value": "1923–24 Sheffield Shield season"
},
{
"operation": "Now on",
"time": "285.695",
"value": "1923–24 Sheffield Shield season"
},
{
"operation": "hyperlink",
"time": "301.98",
"value": "New South Wales"
},
{
"operation": "Now on",
"time": "303.027",
"value": "New South Wales cricket team"
},
{
"operation": "finish",
"time": "329.816",
"value": "finish"
}
]
}
|
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": 1832,
"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"
}
]
}
|
91406
|
Barry Brill was the youngest person to be appointed director of the New Zealand Broadcasting Corporation, and held many other positions such as student liaison in Victoria LAw School, Trustee of the Wellington Trustee Savings Bank,, and served on the Wellington Free Ambulance Board and Welling Hospital Board during his career.
|
{
"evidence": [
"Barry Brill_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:\nBarry Brill was the youngest person to be appointed director of the New Zealand Broadcasting Corporation, and held many other positions such as student liaison in Victoria LAw School, Trustee of the Wellington Trustee Savings Bank,, and served on the Wellington Free Ambulance Board and Welling Hospital Board during his career.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Barry Brill_sentence_10"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1833,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.296",
"value": "?search="
},
{
"operation": "search",
"time": "68.013",
"value": "Barry Brill"
},
{
"operation": "Now on",
"time": "68.848",
"value": "Barry Brill"
},
{
"operation": "Highlighting",
"time": "157.849",
"value": "Barry Brill_sentence_11"
},
{
"operation": "Highlighting",
"time": "318.471",
"value": "Barry Brill_sentence_10"
},
{
"operation": "Highlighting deleted",
"time": "319.639",
"value": "Barry Brill_sentence_11"
},
{
"operation": "finish",
"time": "399.335",
"value": "finish"
}
]
}
|
71333
|
The governmental learning spiral is a technique facilitated by a learning broker which include logistics, performance , drafting and implementation of the modules, moderation of the learning sessions.
|
{
"evidence": [
"Governmental learning spiral_sentence_0",
"Governmental learning spiral_sentence_4",
"Governmental learning spiral_sentence_5"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe governmental learning spiral is a technique facilitated by a learning broker which include logistics, performance , drafting and implementation of the modules, moderation of the learning sessions.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Governmental learning spiral_sentence_0",
"Governmental learning spiral_sentence_4",
"Governmental learning spiral_sentence_5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1834,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.746",
"value": "?search="
},
{
"operation": "search",
"time": "16.55",
"value": "Governmental learning spiral"
},
{
"operation": "Now on",
"time": "17.248",
"value": "Governmental learning spiral"
},
{
"operation": "Highlighting",
"time": "34.259",
"value": "Governmental learning spiral_sentence_0"
},
{
"operation": "Highlighting",
"time": "47.675",
"value": "Governmental learning spiral_sentence_4"
},
{
"operation": "Highlighting",
"time": "79.27",
"value": "Governmental learning spiral_sentence_5"
},
{
"operation": "Page search",
"time": "132.25",
"value": "performance"
},
{
"operation": "finish",
"time": "221.269",
"value": "finish"
}
]
}
|
82524
|
Commander Jaan Maide, chief of the Estonian Defence Forces, took office on September 18, 1944 and left office on October 24, 1944, with a total of 36 days in office in the Army Defence branch.
|
{
"evidence": [
"Jaan Maide_sentence_1",
"Jaan Maide_sentence_15",
"Jaan Maide_sentence_17",
"Jaan Maide_sentence_19"
],
"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:\nCommander Jaan Maide, chief of the Estonian Defence Forces, took office on September 18, 1944 and left office on October 24, 1944, with a total of 36 days in office in the Army Defence branch.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Jaan Maide_sentence_1",
"Jaan Maide_sentence_15",
"Jaan Maide_sentence_17",
"Jaan Maide_sentence_19"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1835,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.086",
"value": "?search="
},
{
"operation": "search",
"time": "29.044",
"value": "Jaan Maide"
},
{
"operation": "Now on",
"time": "29.684",
"value": "Jaan Maide"
},
{
"operation": "Highlighting",
"time": "36.611",
"value": "Jaan Maide_sentence_1"
},
{
"operation": "Page search",
"time": "64.866",
"value": "estonian defen"
},
{
"operation": "Highlighting",
"time": "81.48",
"value": "Jaan Maide_sentence_10"
},
{
"operation": "Highlighting",
"time": "113.024",
"value": "Jaan Maide_sentence_15"
},
{
"operation": "Highlighting deleted",
"time": "136.069",
"value": "Jaan Maide_sentence_10"
},
{
"operation": "Highlighting",
"time": "150.695",
"value": "Jaan Maide_sentence_17"
},
{
"operation": "Page search",
"time": "238.146",
"value": "october"
},
{
"operation": "Highlighting",
"time": "259.131",
"value": "Jaan Maide_sentence_19"
},
{
"operation": "Page search",
"time": "285.355",
"value": "days"
},
{
"operation": "finish",
"time": "359.852",
"value": "finish"
}
]
}
|
exfever_4201
|
Bob Riley is an American singer and a member of the party that was simultaneously strengthened by the collapse of a political party that espoused traditionalist conservatism in the United States.
|
{
"evidence": [
"Bob Riley",
"Republican Party (United States)",
"Whig Party (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:\nBob Riley is an American singer and a member of the party that was simultaneously strengthened by the collapse of a political party that espoused traditionalist conservatism in the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bob Riley",
"Republican Party (United States)",
"Whig Party (United States)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1836,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6467
|
Margaret Constance "Maisie" Williams's highest-grossing film role is Mary Shelley.
|
{
"evidence": [
"Maisie Williams",
"Game of Thrones"
],
"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:\nMargaret Constance \"Maisie\" Williams's highest-grossing film role is Mary Shelley.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Maisie Williams",
"Game of Thrones"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1837,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
6663
|
The Kalmunai Polling Division elections results never matched the country's results in the eight consecutive Sri Lankan presidential elections.
|
{
"evidence": [
"Kalmunai Polling Division_sentence_1",
"Kalmunai Polling Division_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Kalmunai Polling Division elections results never matched the country's results in the eight consecutive Sri Lankan presidential elections.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Kalmunai Polling Division_sentence_1",
"Kalmunai Polling Division_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1838,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.969",
"value": "undefined"
},
{
"operation": "search",
"time": "20.07",
"value": "Kalmunai Polling Division"
},
{
"operation": "Now on",
"time": "22.222",
"value": "Kalmunai Polling Division"
},
{
"operation": "Highlighting",
"time": "24.948",
"value": "Kalmunai Polling Division_sentence_1"
},
{
"operation": "Highlighting",
"time": "29.149",
"value": "Kalmunai Polling Division_sentence_3"
},
{
"operation": "finish",
"time": "35.4",
"value": "finish"
}
]
}
|
exfever_3666
|
The 2014 biographical drama film Pawn Sacrifice was Tobey Maguire's first feature.
|
{
"evidence": [
"Pawn Sacrifice",
"World Chess Championship 1972"
],
"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 2014 biographical drama film Pawn Sacrifice was Tobey Maguire's first feature.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pawn Sacrifice",
"World Chess Championship 1972"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1839,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
46831
|
Vasa övningsskola, established in 1684, was a public school catering preschool, 1-12, and IBDP (a two-year military programme primarily aimed at 16-to-19-year-olds); and headed by Principal Bernt Klockars.
|
{
"evidence": [
"IB Diploma Programme_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:\nVasa övningsskola, established in 1684, was a public school catering preschool, 1-12, and IBDP (a two-year military programme primarily aimed at 16-to-19-year-olds); and headed by Principal Bernt Klockars.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"IB Diploma Programme_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1840,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.987",
"value": "?search="
},
{
"operation": "search",
"time": "18.192",
"value": "Vasa övningsskola"
},
{
"operation": "Now on",
"time": "18.751",
"value": "Vasa övningsskola"
},
{
"operation": "Highlighting",
"time": "26.652",
"value": "Vasa övningsskola_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "26.654",
"value": "Vasa övningsskola_title"
},
{
"operation": "Highlighting",
"time": "27.573",
"value": "Vasa övningsskola_cell_0_5_1"
},
{
"operation": "Highlighting",
"time": "27.581",
"value": "Vasa övningsskola_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "27.581",
"value": "Vasa övningsskola_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "31.029",
"value": "Vasa övningsskola_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "31.99",
"value": "Vasa övningsskola_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "31.999",
"value": "Vasa övningsskola_header_cell_0_7_0"
},
{
"operation": "Page search",
"time": "54.031",
"value": "principal"
},
{
"operation": "Page search",
"time": "69.156",
"value": "Bernt Klockars"
},
{
"operation": "Highlighting",
"time": "74.178",
"value": "Vasa övningsskola_header_cell_0_10_0"
},
{
"operation": "Highlighting",
"time": "75.482",
"value": "Vasa övningsskola_cell_0_10_1"
},
{
"operation": "Highlighting",
"time": "75.49",
"value": "Vasa övningsskola_header_cell_0_10_0"
},
{
"operation": "Highlighting",
"time": "77.12",
"value": "Vasa övningsskola_header_cell_0_9_0"
},
{
"operation": "Highlighting",
"time": "77.945",
"value": "Vasa övningsskola_cell_0_9_1"
},
{
"operation": "Highlighting",
"time": "77.954",
"value": "Vasa övningsskola_header_cell_0_9_0"
},
{
"operation": "search",
"time": "104.744",
"value": "Bernt Klockars"
},
{
"operation": "Now on",
"time": "105.713",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "111.078",
"value": "Vasa övningsskola"
},
{
"operation": "Now on",
"time": "111.441",
"value": "Vasa övningsskola"
},
{
"operation": "Page search",
"time": "135.426",
"value": "Bernt"
},
{
"operation": "Highlighting",
"time": "152.998",
"value": "Vasa övningsskola_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "153.824",
"value": "Vasa övningsskola_cell_0_6_1"
},
{
"operation": "Highlighting",
"time": "153.835",
"value": "Vasa övningsskola_header_cell_0_6_0"
},
{
"operation": "Highlighting deleted",
"time": "157.647",
"value": "Vasa övningsskola_cell_0_9_1"
},
{
"operation": "Highlighting deleted",
"time": "157.653",
"value": "Vasa övningsskola_header_cell_0_9_0"
},
{
"operation": "Highlighting deleted",
"time": "158.337",
"value": "Vasa övningsskola_cell_0_10_1"
},
{
"operation": "Highlighting deleted",
"time": "158.345",
"value": "Vasa övningsskola_header_cell_0_10_0"
},
{
"operation": "Highlighting",
"time": "159.979",
"value": "Vasa övningsskola_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "160.753",
"value": "Vasa övningsskola_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "160.762",
"value": "Vasa övningsskola_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "161.675",
"value": "Vasa övningsskola_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "162.451",
"value": "Vasa övningsskola_cell_0_5_1"
},
{
"operation": "Highlighting",
"time": "162.458",
"value": "Vasa övningsskola_header_cell_0_5_0"
},
{
"operation": "hyperlink",
"time": "163.861",
"value": "IB DP"
},
{
"operation": "Now on",
"time": "164.547",
"value": "IB Diploma Programme"
},
{
"operation": "Highlighting",
"time": "174.119",
"value": "IB Diploma Programme_sentence_1"
},
{
"operation": "finish",
"time": "202.668",
"value": "finish"
}
]
}
|
exfever_6107
|
George H. W. Bush was a British politician, diplomat, and businessman who was raised in a town in southwestern Fairfield County, Connecticut, United States which is a home to many hedge funds and other financial services firms.
|
{
"evidence": [
"George H. W. Bush",
"Greenwich, Connecticut"
],
"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:\nGeorge H. W. Bush was a British politician, diplomat, and businessman who was raised in a town in southwestern Fairfield County, Connecticut, United States which is a home to many hedge funds and other financial services firms.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"George H. W. Bush",
"Greenwich, Connecticut"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1841,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2339
|
On Stranger Tides is a 1987 historical fantasy novel that takes place during the period between the 1650s and 1730s, where the history of piracy is often subdivided into three periods, including a sailing route followed by certain mainly English, in 1690.
|
{
"evidence": [
"On Stranger Tides",
"Golden Age of Piracy",
"Pirate Round"
],
"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:\nOn Stranger Tides is a 1987 historical fantasy novel that takes place during the period between the 1650s and 1730s, where the history of piracy is often subdivided into three periods, including a sailing route followed by certain mainly English, in 1690.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"On Stranger Tides",
"Golden Age of Piracy",
"Pirate Round"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1842,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8820
|
Paul Bettany is the first English actor who was nominated for BAFTA Award for Best Actor.
|
{
"evidence": [
"Paul Bettany",
"Marvel Cinematic Universe"
],
"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:\nPaul Bettany is the first English actor who was nominated for BAFTA Award for Best Actor.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Bettany",
"Marvel Cinematic Universe"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1843,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71903
|
Russian-based charter airline or air taxi SmartLynx Airlines has flight crews speaking multiple languages and captains having an average experience above 5300 block hours.
|
{
"evidence": [
"SmartLynx Airlines_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:\nRussian-based charter airline or air taxi SmartLynx Airlines has flight crews speaking multiple languages and captains having an average experience above 5300 block hours.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"SmartLynx Airlines_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1844,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.6",
"value": "?search="
},
{
"operation": "search",
"time": "14.298",
"value": "SmartLynx Airlines"
},
{
"operation": "Now on",
"time": "15.306",
"value": "SmartLynx Airlines"
},
{
"operation": "Highlighting",
"time": "24.767",
"value": "SmartLynx Airlines_sentence_0"
},
{
"operation": "finish",
"time": "51.992",
"value": "finish"
}
]
}
|
exfever_7242
|
Gabrielle Monique Union is an American actress who appeared on television sitcoms, before landing supporting roles in an American romantic comedy film starring an American actress who became a teen idol in the 1990s.
|
{
"evidence": [
"Gabrielle Union",
"10 Things I Hate About You",
"Larisa Oleynik"
],
"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:\nGabrielle Monique Union is an American actress who appeared on television sitcoms, before landing supporting roles in an American romantic comedy film starring an American actress who became a teen idol in the 1990s.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gabrielle Union",
"10 Things I Hate About You",
"Larisa Oleynik"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1845,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
57482
|
Enrique Jordá made several stereophonic recordings in San Francisco for RCA Victor in 1957 and 1958; then made several highly acclaimed recordings for Decca in the late 1940s and early 1950s of Spanish music.
|
{
"evidence": [
"Enrique Jordá_sentence_3",
"Enrique Jordá_sentence_4"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEnrique Jordá made several stereophonic recordings in San Francisco for RCA Victor in 1957 and 1958; then made several highly acclaimed recordings for Decca in the late 1940s and early 1950s of Spanish music.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Enrique Jordá_sentence_3",
"Enrique Jordá_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1846,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.981",
"value": "?search="
},
{
"operation": "search",
"time": "24.718",
"value": "Enrique Jorda"
},
{
"operation": "Now on",
"time": "25.649",
"value": "Enrique Jordá"
},
{
"operation": "Page search",
"time": "32.613",
"value": "record"
},
{
"operation": "Highlighting",
"time": "34.287",
"value": "Enrique Jordá_sentence_3"
},
{
"operation": "Highlighting",
"time": "41.167",
"value": "Enrique Jordá_sentence_4"
},
{
"operation": "finish",
"time": "52.489",
"value": "finish"
}
]
}
|
56729
|
The song Come Together by the Beatles, was written by John Lennon who was an English singer, songwriter and peace activist who was rebellious in nature and achieved worldwide fame as the founder, co-lead vocalist, and rhythm guitarist of the Beatles.
|
{
"evidence": [
"Come Together_sentence_2",
"John Lennon_sentence_4",
"John Lennon_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:\nThe song Come Together by the Beatles, was written by John Lennon who was an English singer, songwriter and peace activist who was rebellious in nature and achieved worldwide fame as the founder, co-lead vocalist, and rhythm guitarist of the Beatles.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Come Together_sentence_2",
"John Lennon_sentence_4",
"John Lennon_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1847,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.15",
"value": "?search="
},
{
"operation": "search",
"time": "16.442",
"value": "Come Together (disambiguation)"
},
{
"operation": "Now on",
"time": "17.583",
"value": "Come Together (disambiguation)"
},
{
"operation": "Page search",
"time": "43.003",
"value": "beatle"
},
{
"operation": "hyperlink",
"time": "45.915",
"value": "Come Together"
},
{
"operation": "Now on",
"time": "46.662",
"value": "Come Together"
},
{
"operation": "Highlighting",
"time": "52.003",
"value": "Come Together_sentence_2"
},
{
"operation": "hyperlink",
"time": "56.369",
"value": "John Lennon"
},
{
"operation": "Now on",
"time": "65.934",
"value": "John Lennon"
},
{
"operation": "Highlighting",
"time": "68.34",
"value": "John Lennon_sentence_4"
},
{
"operation": "Page search",
"time": "88.975",
"value": "rebel"
},
{
"operation": "Highlighting",
"time": "95.03",
"value": "John Lennon_sentence_11"
},
{
"operation": "finish",
"time": "246.143",
"value": "finish"
}
]
}
|
exfever_6173
|
Spencer Tracy is the youngest major stars of Hollywood's Golden Age.
|
{
"evidence": [
"Spencer Tracy",
"Broadway theatre"
],
"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:\nSpencer Tracy is the youngest major stars of Hollywood's Golden Age.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Spencer Tracy",
"Broadway theatre"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1848,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5364
|
The KGB was the main security agency for a socialist and communist state that spanned much of Eurasia from 1922 to 2022.
|
{
"evidence": [
"KGB",
"Soviet Union"
],
"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 KGB was the main security agency for a socialist and communist state that spanned much of Eurasia from 1922 to 2022.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"KGB",
"Soviet Union"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1849,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
46948
|
Anbirkku Alavillai, directed by Maryam Kassai & Anand S Babu and produced by Jim Sanjay, stars Johnny Lever (the only stand-up comedian in India), Sumeet Pathak, Neetu Chaudhury, Mansi Lal, and Rama Thapa.
|
{
"evidence": [
"Johnny Lever_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:\nAnbirkku Alavillai, directed by Maryam Kassai & Anand S Babu and produced by Jim Sanjay, stars Johnny Lever (the only stand-up comedian in India), Sumeet Pathak, Neetu Chaudhury, Mansi Lal, and Rama Thapa.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Johnny Lever_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1850,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.734",
"value": "?search="
},
{
"operation": "search",
"time": "9.435",
"value": "Anbirkku Alavillai"
},
{
"operation": "Now on",
"time": "10.337",
"value": "Anbirkku Alavillai"
},
{
"operation": "Highlighting",
"time": "28.219",
"value": "Anbirkku Alavillai_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "28.22",
"value": "Anbirkku Alavillai_title"
},
{
"operation": "Highlighting",
"time": "28.223",
"value": "Anbirkku Alavillai_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "28.223",
"value": "Anbirkku Alavillai_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "35.019",
"value": "Anbirkku Alavillai_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "35.023",
"value": "Anbirkku Alavillai_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "44.247",
"value": "Anbirkku Alavillai_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "44.252",
"value": "Anbirkku Alavillai_header_cell_0_4_0"
},
{
"operation": "hyperlink",
"time": "56.556",
"value": "Johnny Lever"
},
{
"operation": "Now on",
"time": "57.476",
"value": "Johnny Lever"
},
{
"operation": "Highlighting",
"time": "63.003",
"value": "Johnny Lever_sentence_1"
},
{
"operation": "Highlighting",
"time": "73.906",
"value": "Johnny Lever_sentence_1"
},
{
"operation": "finish",
"time": "96.786",
"value": "finish"
}
]
}
|
exfever_321
|
Leighton Meester is an American actress best known for starring in an Asian teen drama television series based on the novel series of the same name written by Cecily von Ziegesar.
|
{
"evidence": [
"Leighton Meester",
"Gossip Girl"
],
"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:\nLeighton Meester is an American actress best known for starring in an Asian teen drama television series based on the novel series of the same name written by Cecily von Ziegesar.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Leighton Meester",
"Gossip Girl"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1851,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
22999
|
Cullingworth is a village and civil parish 7 miles west of Bradford and 3 miles south of Keighley, 2 miles to the east of Haworth and is situated near the head of Harden valley in the town of Bradford, West Yorkshire, England.
|
{
"evidence": [
"Cullingworth_sentence_0",
"Cullingworth_sentence_1",
"Haworth_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:\nCullingworth is a village and civil parish 7 miles west of Bradford and 3 miles south of Keighley, 2 miles to the east of Haworth and is situated near the head of Harden valley in the town of Bradford, West Yorkshire, England.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Cullingworth_sentence_0",
"Cullingworth_sentence_1",
"Haworth_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1852,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.094",
"value": "?search="
},
{
"operation": "search",
"time": "8.19",
"value": "Cullingworth"
},
{
"operation": "Now on",
"time": "9.399",
"value": "Cullingworth"
},
{
"operation": "Highlighting",
"time": "13.511",
"value": "Cullingworth_sentence_0"
},
{
"operation": "Highlighting",
"time": "25.122",
"value": "Cullingworth_sentence_1"
},
{
"operation": "Page search",
"time": "57.953",
"value": "Haworth"
},
{
"operation": "hyperlink",
"time": "92.031",
"value": "South Pennines"
},
{
"operation": "Now on",
"time": "92.82",
"value": "South Pennines"
},
{
"operation": "back-button-clicked",
"time": "101.288",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "101.388",
"value": "Cullingworth"
},
{
"operation": "search",
"time": "108.034",
"value": "haworth"
},
{
"operation": "Now on",
"time": "109.137",
"value": "Haworth"
},
{
"operation": "Highlighting",
"time": "156.47",
"value": "Haworth_sentence_1"
},
{
"operation": "finish",
"time": "188.538",
"value": "finish"
}
]
}
|
exfever_2896
|
William Shakespeare is an English singer born in the market town of Warwickshire, situated in the southeast of a major city and metropolitan borough in the West Midlands, England.
|
{
"evidence": [
"William Shakespeare",
"Stratford-upon-Avon",
"Birmingham"
],
"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:\nWilliam Shakespeare is an English singer born in the market town of Warwickshire, situated in the southeast of a major city and metropolitan borough in the West Midlands, England.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"William Shakespeare",
"Stratford-upon-Avon",
"Birmingham"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1853,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
59509
|
CIL Reghin Albatros, a tandem seat, all wood glider designed and produced in small numbers at the CIL (Complexu Industrializare Lemnului - Reghin) in Romania in the 1950s, had wings that had a root chord of 1.60 m (5 ft 3 in), tapering to 0.66 m (2 ft 2 in) at the tip and carried 1.50° of dihedral, built around a single spar with a plywood covered leading edge torsion box in front of it and fabric covered behind.
|
{
"evidence": [
"CIL Reghin Albatros_sentence_0",
"CIL Reghin Albatros_sentence_2",
"CIL Reghin Albatros_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:\nCIL Reghin Albatros, a tandem seat, all wood glider designed and produced in small numbers at the CIL (Complexu Industrializare Lemnului - Reghin) in Romania in the 1950s, had wings that had a root chord of 1.60 m (5 ft 3 in), tapering to 0.66 m (2 ft 2 in) at the tip and carried 1.50° of dihedral, built around a single spar with a plywood covered leading edge torsion box in front of it and fabric covered behind.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"CIL Reghin Albatros_sentence_0",
"CIL Reghin Albatros_sentence_2",
"CIL Reghin Albatros_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1854,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.932",
"value": "?search="
},
{
"operation": "search",
"time": "23.997",
"value": "CIL Reghin Albatros"
},
{
"operation": "Now on",
"time": "25.019",
"value": "CIL Reghin Albatros"
},
{
"operation": "Highlighting",
"time": "97.768",
"value": "CIL Reghin Albatros_sentence_0"
},
{
"operation": "Highlighting",
"time": "132.387",
"value": "CIL Reghin Albatros_sentence_2"
},
{
"operation": "Highlighting",
"time": "145.612",
"value": "CIL Reghin Albatros_sentence_3"
},
{
"operation": "finish",
"time": "158.973",
"value": "finish"
}
]
}
|
62160
|
Cercamon composed a Plahn for William X of Aquitaine in 1137 and Lanfanc Cigala composed a Plahn for Raimbaut d'Aurenga in 1173.
|
{
"evidence": [
"Cercamon_sentence_0",
"Cercamon_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:\nCercamon composed a Plahn for William X of Aquitaine in 1137 and Lanfanc Cigala composed a Plahn for Raimbaut d'Aurenga in 1173.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cercamon_sentence_0",
"Cercamon_sentence_1"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1855,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.052",
"value": "?search="
},
{
"operation": "search",
"time": "32.724",
"value": "Cercamon"
},
{
"operation": "Now on",
"time": "33.403",
"value": "Cercamon"
},
{
"operation": "Page search",
"time": "61.832",
"value": "1137"
},
{
"operation": "Highlighting",
"time": "69.333",
"value": "Cercamon_sentence_0"
},
{
"operation": "Highlighting",
"time": "70.134",
"value": "Cercamon_sentence_1"
},
{
"operation": "finish",
"time": "80.115",
"value": "finish"
}
]
}
|
71812
|
Mason Pearson, who was called by a Yorkshireman in the mid 1860s to come to work at the British Steam Brush Works, invented the "pneumatic" rubber-cushion hairbrush.
|
{
"evidence": [
"Mason Pearson Brushes_sentence_1",
"Mason Pearson Brushes_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:\nMason Pearson, who was called by a Yorkshireman in the mid 1860s to come to work at the British Steam Brush Works, invented the \"pneumatic\" rubber-cushion hairbrush.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Mason Pearson Brushes_sentence_1",
"Mason Pearson Brushes_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1856,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.691",
"value": "undefined"
},
{
"operation": "search",
"time": "34.53",
"value": "Contains...Mason Pearson"
},
{
"operation": "Now on",
"time": "39.148",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "64.565",
"value": "Mason Pearson Brushes"
},
{
"operation": "Now on",
"time": "65.221",
"value": "Mason Pearson Brushes"
},
{
"operation": "Highlighting",
"time": "84.75",
"value": "Mason Pearson Brushes_sentence_1"
},
{
"operation": "Page search",
"time": "100.509",
"value": "british"
},
{
"operation": "Page search",
"time": "117.371",
"value": "pneumatic"
},
{
"operation": "Highlighting",
"time": "121.156",
"value": "Mason Pearson Brushes_sentence_2"
},
{
"operation": "Page search",
"time": "134.253",
"value": "rubber"
},
{
"operation": "finish",
"time": "383.314",
"value": "finish"
}
]
}
|
exfever_1790
|
Alexander Lukashenko retained an important Soviet-era symbol in a vertically bisected red-and-green with a white-and-red ornament pattern placed at the hoist end banner of Belarus.
|
{
"evidence": [
"Alexander Lukashenko",
"Flag of Belarus"
],
"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:\nAlexander Lukashenko retained an important Soviet-era symbol in a vertically bisected red-and-green with a white-and-red ornament pattern placed at the hoist end banner of Belarus.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alexander Lukashenko",
"Flag of Belarus"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1857,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_593
|
Charles Marie de La Condamine spent ten years measuring the length of a degree of latitude at the circle of latitude that divides the Earth.
|
{
"evidence": [
"Charles Marie de La Condamine",
"Equator"
],
"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:\nCharles Marie de La Condamine spent ten years measuring the length of a degree of latitude at the circle of latitude that divides the Earth.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Charles Marie de La Condamine",
"Equator"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1858,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_302
|
The Beatles were a rock band that helped shape their music?by spreading throughout various regions, consisting of the majority of Europe and Oceania.
|
{
"evidence": [
"The Beatles",
"Counterculture of the 1960s",
"Western world"
],
"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 Beatles were a rock band that helped shape their music?by spreading throughout various regions, consisting of the majority of Europe and Oceania.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Beatles",
"Counterculture of the 1960s",
"Western world"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1859,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2961
|
Inside Man is the third American heist thriller film directed by Spike Lee.
|
{
"evidence": [
"Inside Man",
"Denzel Washington",
"The New York Times"
],
"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:\nInside Man is the third American heist thriller film directed by Spike Lee.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Inside Man",
"Denzel Washington",
"The New York Times"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1860,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
55720
|
Voice of God, according to some religions, is a communication from God to human beings and. is not to be confused with Voice of the Gods which is a novel written by Trudi Canavan.
|
{
"evidence": [
"Voice of God_sentence_1",
"Voice of God_sentence_0",
"Age of the Five_sentence_4"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVoice of God, according to some religions, is a communication from God to human beings and. is not to be confused with Voice of the Gods which is a novel written by Trudi Canavan.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Voice of God_sentence_1",
"Voice of God_sentence_0",
"Age of the Five_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1861,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.845",
"value": "?search="
},
{
"operation": "search",
"time": "5.797",
"value": "Voice of God"
},
{
"operation": "Now on",
"time": "6.963",
"value": "Voice of God"
},
{
"operation": "Highlighting",
"time": "29.68",
"value": "Voice of God_sentence_1"
},
{
"operation": "Highlighting",
"time": "36.188",
"value": "Voice of God_sentence_0"
},
{
"operation": "hyperlink",
"time": "37.663",
"value": "Voice of the Gods"
},
{
"operation": "Now on",
"time": "38.505",
"value": "Voice of the Gods"
},
{
"operation": "Highlighting",
"time": "47.413",
"value": "Age of the Five_cell_2_0_1"
},
{
"operation": "Highlighting",
"time": "47.418",
"value": "Age of the Five_header_cell_2_0_0"
},
{
"operation": "Highlighting deleted",
"time": "48.033",
"value": "Age of the Five_cell_2_0_1"
},
{
"operation": "Highlighting deleted",
"time": "48.04",
"value": "Age of the Five_header_cell_2_0_0"
},
{
"operation": "Highlighting",
"time": "48.588",
"value": "Age of the Five_sentence_4"
},
{
"operation": "finish",
"time": "54.958",
"value": "finish"
}
]
}
|
56651
|
Richepanse, born in Metz as the son of an officer of the Conti-Dragooon Regiment, distinguished himself in the early battles of the French Revolutionary War, and by 1794, had been promoted to général de brigade.
|
{
"evidence": [
"Antoine Richepanse_sentence_1",
"Antoine Richepanse_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:\nRichepanse, born in Metz as the son of an officer of the Conti-Dragooon Regiment, distinguished himself in the early battles of the French Revolutionary War, and by 1794, had been promoted to général de brigade.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Antoine Richepanse_sentence_1",
"Antoine Richepanse_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1862,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.695",
"value": "?search="
},
{
"operation": "search",
"time": "9.453",
"value": "Contains...Richepanse"
},
{
"operation": "Now on",
"time": "10.082",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "13.561",
"value": "Antoine Richepanse"
},
{
"operation": "Now on",
"time": "14.093",
"value": "Antoine Richepanse"
},
{
"operation": "Highlighting",
"time": "25.601",
"value": "Antoine Richepanse_sentence_1"
},
{
"operation": "Highlighting",
"time": "31.205",
"value": "Antoine Richepanse_sentence_2"
},
{
"operation": "finish",
"time": "41.69",
"value": "finish"
}
]
}
|
exfever_6567
|
Howard Robard Hughes Jr. gained fame in Hollywood beginning in the late 1920s, when he produced big-budget and often controversial films such as a 1928 American silent crime drama film starring an American actor, of both stage and screen.
|
{
"evidence": [
"Howard Hughes",
"The Racket (1928 film)",
"Louis Wolheim"
],
"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:\nHoward Robard Hughes Jr. gained fame in Hollywood beginning in the late 1920s, when he produced big-budget and often controversial films such as a 1928 American silent crime drama film starring an American actor, of both stage and screen.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Howard Hughes",
"The Racket (1928 film)",
"Louis Wolheim"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1863,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4625
|
The former San Francisco Civic Auditorium was the first multi-purpose arena that holds 8,500 people.
|
{
"evidence": [
"Bill Graham Civic Auditorium",
"Bill Graham (promoter)"
],
"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 former San Francisco Civic Auditorium was the first multi-purpose arena that holds 8,500 people.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bill Graham Civic Auditorium",
"Bill Graham (promoter)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1864,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2962
|
The Beach Boys is the first American rock band that formed in 1961.
|
{
"evidence": [
"The Beach Boys",
"Rock music",
"Rock and roll"
],
"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 Beach Boys is the first American rock band that formed in 1961.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Beach Boys",
"Rock music",
"Rock and roll"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1865,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2008
|
Empress Matilda, the daughter of King Henry I of England, was controversially crowned in the building that stood from the 4th to 16th centuries in San Marino.
|
{
"evidence": [
"Empress Matilda",
"Old St. Peter's Basilica"
],
"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:\nEmpress Matilda, the daughter of King Henry I of England, was controversially crowned in the building that stood from the 4th to 16th centuries in San Marino.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Empress Matilda",
"Old St. Peter's Basilica"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1866,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4183
|
Jacinda Barrett makes an appearance in a 2006 American romantic comedy-drama film.
|
{
"evidence": [
"Jacinda Barrett",
"The Last Kiss (2006 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:\nJacinda Barrett makes an appearance in a 2006 American romantic comedy-drama film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jacinda Barrett",
"The Last Kiss (2006 film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1867,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4832
|
Radiohead is a Polish rock band that worked with an English record producer, recording engineer, and musician.
|
{
"evidence": [
"Radiohead",
"Nigel Godrich"
],
"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:\nRadiohead is a Polish rock band that worked with an English record producer, recording engineer, and musician.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Radiohead",
"Nigel Godrich"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1868,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
81246
|
Josh James (born June 28, 1973) is an American entrepreneur, founder and CEO of Domo, and a graduate of Brigham Young University.
|
{
"evidence": [
"Josh James_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:\nJosh James (born June 28, 1973) is an American entrepreneur, founder and CEO of Domo, and a graduate of Brigham Young University.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Josh James_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1869,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.111",
"value": "?search="
},
{
"operation": "search",
"time": "9.983",
"value": "Josh James"
},
{
"operation": "Now on",
"time": "10.874",
"value": "Josh James"
},
{
"operation": "Highlighting",
"time": "35.001",
"value": "Josh James_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "35.009",
"value": "Josh James_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "35.009",
"value": "Josh James_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "42.635",
"value": "Josh James_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "57.446",
"value": "Josh James_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "58.65",
"value": "Josh James_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "58.657",
"value": "Josh James_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "58.658",
"value": "Josh James_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "59.716",
"value": "Josh James_sentence_4"
},
{
"operation": "finish",
"time": "173.314",
"value": "finish"
}
]
}
|
81479
|
Jazz drummer Kaiser Marshall was active in chamber music during the 19th century.
|
{
"evidence": [
"Kaiser Marshall_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:\nJazz drummer Kaiser Marshall was active in chamber music during the 19th century.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Kaiser Marshall_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1870,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.018",
"value": "?search="
},
{
"operation": "search",
"time": "11.237",
"value": "Kaiser Marshall"
},
{
"operation": "Now on",
"time": "13.185",
"value": "Kaiser Marshall"
},
{
"operation": "hyperlink",
"time": "21.372",
"value": "Joe Marshall (musician)"
},
{
"operation": "Now on",
"time": "22.506",
"value": "Joe Marshall (musician)"
},
{
"operation": "hyperlink",
"time": "36.558",
"value": "Kaiser Marshall"
},
{
"operation": "Now on",
"time": "36.928",
"value": "Kaiser Marshall"
},
{
"operation": "search",
"time": "156.589",
"value": "Jazz drummer Kaiser Marshall"
},
{
"operation": "Now on",
"time": "159.384",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "203.526",
"value": "Joseph Marshall"
},
{
"operation": "Now on",
"time": "204.255",
"value": "Joseph Marshall"
},
{
"operation": "Highlighting",
"time": "243.601",
"value": "Joseph Marshall_item_0_5"
},
{
"operation": "Highlighting",
"time": "243.602",
"value": "Joseph Marshall_title"
},
{
"operation": "Highlighting deleted",
"time": "245.778",
"value": "Joseph Marshall_item_0_5"
},
{
"operation": "Highlighting deleted",
"time": "245.779",
"value": "Joseph Marshall_title"
},
{
"operation": "hyperlink",
"time": "247.151",
"value": "Kaiser Marshall"
},
{
"operation": "Now on",
"time": "247.546",
"value": "Kaiser Marshall"
},
{
"operation": "hyperlink",
"time": "431.676",
"value": "Joe Marshall (musician)"
},
{
"operation": "Now on",
"time": "432.404",
"value": "Joe Marshall (musician)"
},
{
"operation": "hyperlink",
"time": "464.712",
"value": "Kaiser Marshall"
},
{
"operation": "Now on",
"time": "465.151",
"value": "Kaiser Marshall"
},
{
"operation": "Highlighting",
"time": "508.028",
"value": "Kaiser Marshall_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "526.429",
"value": "Kaiser Marshall_sentence_2"
},
{
"operation": "Highlighting",
"time": "527.579",
"value": "Kaiser Marshall_sentence_0"
},
{
"operation": "finish",
"time": "544.545",
"value": "finish"
}
]
}
|
exfever_6507
|
The Georgia Institute of Technology is an institute of technology in Atlanta, Georgia, and a public university that is committed to research as a central part of its mission.
|
{
"evidence": [
"Georgia Institute of Technology",
"Research university"
],
"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 Georgia Institute of Technology is an institute of technology in Atlanta, Georgia, and a public university that is committed to research as a central part of its mission.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Georgia Institute of Technology",
"Research university"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1871,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
63831
|
The 1977 Virginia ballot measures took place on November 8, 1977, the same day as the Gubernatorial race between Mills E. Godwin Jr. vs John N. Dalton.
|
{
"evidence": [
"1977 Virginia gubernatorial election_sentence_1",
"1977 Virginia gubernatorial election_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 1977 Virginia ballot measures took place on November 8, 1977, the same day as the Gubernatorial race between Mills E. Godwin Jr. vs John N. Dalton.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1977 Virginia gubernatorial election_sentence_1",
"1977 Virginia gubernatorial election_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1872,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.587",
"value": "?search="
},
{
"operation": "search",
"time": "14.146",
"value": "1977 Virginia ballot measures"
},
{
"operation": "Now on",
"time": "14.813",
"value": "1977 Virginia ballot measures"
},
{
"operation": "Highlighting",
"time": "29.85",
"value": "1977 Virginia ballot measures_sentence_0"
},
{
"operation": "Highlighting",
"time": "29.85",
"value": "1977 Virginia ballot measures_title"
},
{
"operation": "hyperlink",
"time": "50.313",
"value": "Gubernatorial"
},
{
"operation": "Now on",
"time": "51.249",
"value": "1977 Virginia gubernatorial election"
},
{
"operation": "Highlighting",
"time": "78.391",
"value": "1977 Virginia gubernatorial election_sentence_1"
},
{
"operation": "back-button-clicked",
"time": "83.96",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "84.017",
"value": "1977 Virginia ballot measures"
},
{
"operation": "Highlighting deleted",
"time": "85.231",
"value": "1977 Virginia ballot measures_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "85.232",
"value": "1977 Virginia ballot measures_title"
},
{
"operation": "hyperlink",
"time": "88.703",
"value": "Gubernatorial"
},
{
"operation": "Now on",
"time": "89.192",
"value": "1977 Virginia gubernatorial election"
},
{
"operation": "Highlighting",
"time": "96.231",
"value": "1977 Virginia gubernatorial election_sentence_0"
},
{
"operation": "finish",
"time": "102.559",
"value": "finish"
}
]
}
|
exfever_3365
|
Tupac Shakur is an African rapper who released his debut album on November 12, 1991, on an American record label owned by Universal Music Group through its Interscope Geffen A&M imprint.
|
{
"evidence": [
"Tupac Shakur",
"2Pacalypse Now",
"Interscope Records"
],
"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:\nTupac Shakur is an African rapper who released his debut album on November 12, 1991, on an American record label owned by Universal Music Group through its Interscope Geffen A&M imprint.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tupac Shakur",
"2Pacalypse Now",
"Interscope Records"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1873,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7762
|
Naomi Grace Scott is a French actress who rose to prominence for her performances in an American teen musical drama television film that was based on the young adult novel by Mark Peter Hughes, published in 2007 by Delacorte Press.
|
{
"evidence": [
"Naomi Scott",
"Lemonade Mouth (film)",
"Lemonade Mouth"
],
"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:\nNaomi Grace Scott is a French actress who rose to prominence for her performances in an American teen musical drama television film that was based on the young adult novel by Mark Peter Hughes, published in 2007 by Delacorte Press.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Naomi Scott",
"Lemonade Mouth (film)",
"Lemonade Mouth"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1874,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3406
|
Hitman was the second action-thriller film directed by Xavier Gens.
|
{
"evidence": [
"Hitman (2007 film)",
"Agent 47"
],
"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:\nHitman was the second action-thriller film directed by Xavier Gens.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hitman (2007 film)",
"Agent 47"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1875,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76994
|
Taste the Sin is the third album by Black Tusk.
|
{
"evidence": [
"Taste the Sin_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:\nTaste the Sin is the third album by Black Tusk.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Taste the Sin_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1876,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.924",
"value": "?search="
},
{
"operation": "search",
"time": "15.728",
"value": "Taste the Sin"
},
{
"operation": "Now on",
"time": "17.029",
"value": "Taste the Sin"
},
{
"operation": "Highlighting",
"time": "20.477",
"value": "Taste the Sin_sentence_0"
},
{
"operation": "hyperlink",
"time": "27.146",
"value": "Black Tusk"
},
{
"operation": "Now on",
"time": "28.528",
"value": "Black Tusk (band)"
},
{
"operation": "Page search",
"time": "34.176",
"value": "taste the sin"
},
{
"operation": "back-button-clicked",
"time": "48.056",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "48.162",
"value": "Taste the Sin"
},
{
"operation": "finish",
"time": "112.004",
"value": "finish"
}
]
}
|
exfever_8970
|
At the 2008 Brit Awards, Kate Marie Nash won the award that was first presented in 1977, given by the British recorded music industry's Trade Association.
|
{
"evidence": [
"Kate Nash",
"Brit Award for British Female Solo Artist",
"British Phonographic Industry"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAt the 2008 Brit Awards, Kate Marie Nash won the award that was first presented in 1977, given by the British recorded music industry's Trade Association.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kate Nash",
"Brit Award for British Female Solo Artist",
"British Phonographic Industry"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1877,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8299
|
Simon & Garfunkel recorded albums together featuring songs mostly written by Paul Simon, including a song recorded by the American music duo for their fifth studio album.
|
{
"evidence": [
"Paul Simon",
"The Boxer"
],
"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:\nSimon & Garfunkel recorded albums together featuring songs mostly written by Paul Simon, including a song recorded by the American music duo for their fifth studio album.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Simon",
"The Boxer"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1878,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
79152
|
Vallée never graduated with a bachelor's degree in Management Information Systems from the University of Ottawa; then received the Ottawa Forty Under 40 award from the Ottawa Business Journal in 2011.
|
{
"evidence": [
"Paul Vallée_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:\nVallée never graduated with a bachelor's degree in Management Information Systems from the University of Ottawa; then received the Ottawa Forty Under 40 award from the Ottawa Business Journal in 2011.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Paul Vallée_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1879,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.149",
"value": "?search="
},
{
"operation": "search",
"time": "17.824",
"value": "Vallée"
},
{
"operation": "Now on",
"time": "19.579",
"value": "Vallée"
},
{
"operation": "search",
"time": "34.143",
"value": "\"Management Information Systems from the University of Ottawa\""
},
{
"operation": "Now on",
"time": "36.427",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "37.711",
"value": "Paul Vallée"
},
{
"operation": "Now on",
"time": "38.821",
"value": "Paul Vallée"
},
{
"operation": "Page search",
"time": "41.724",
"value": "Management Information Systems from the University of Ottawa"
},
{
"operation": "Highlighting",
"time": "43.653",
"value": "Paul Vallée_sentence_2"
},
{
"operation": "finish",
"time": "64.05",
"value": "finish"
}
]
}
|
exfever_2588
|
Conquest of the Planet of the Apes is a science fiction film followed by an American film that uses speculative, fictional science-based depictions of phenomena that are not accepted by mainstream science, such as extraterrestrial lifeforms spacecraft, robots, cyborgs, interstellar travel, or other technologies.
|
{
"evidence": [
"Conquest of the Planet of the Apes",
"Battle for the Planet of the Apes",
"Science fiction 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:\nConquest of the Planet of the Apes is a science fiction film followed by an American film that uses speculative, fictional science-based depictions of phenomena that are not accepted by mainstream science, such as extraterrestrial lifeforms spacecraft, robots, cyborgs, interstellar travel, or other technologies.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Conquest of the Planet of the Apes",
"Battle for the Planet of the Apes",
"Science fiction film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1880,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3124
|
Kurt Angle was the first Olympic gold medalist in American wrestling.
|
{
"evidence": [
"Kurt Angle",
"USA Wrestling",
"Freestyle wrestling"
],
"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:\nKurt Angle was the first Olympic gold medalist in American wrestling.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kurt Angle",
"USA Wrestling",
"Freestyle wrestling"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1881,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2454
|
William Shakespeare was an English playwright born and raised in a market town and civil parish in the Stratford-on-Avon district, in the county of Warwickshire.
|
{
"evidence": [
"William Shakespeare",
"Stratford-upon-Avon"
],
"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:\nWilliam Shakespeare was an English playwright born and raised in a market town and civil parish in the Stratford-on-Avon district, in the county of Warwickshire.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"William Shakespeare",
"Stratford-upon-Avon"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1882,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76262
|
Brian Finch longest relationship was with the ITV1 soap opera, Coronation Street, for which he wrote 120 scripts between 1980 and 1999 and he also helped the development of All Creatures Great and Small, The Tomorrow People, and Heartbeat.
|
{
"evidence": [
"Brian Finch_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:\nBrian Finch longest relationship was with the ITV1 soap opera, Coronation Street, for which he wrote 120 scripts between 1980 and 1999 and he also helped the development of All Creatures Great and Small, The Tomorrow People, and Heartbeat.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Brian Finch_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1883,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.997",
"value": "?search="
},
{
"operation": "search",
"time": "8.416",
"value": "Brian Finch"
},
{
"operation": "Now on",
"time": "9.142",
"value": "Brian Finch"
},
{
"operation": "Highlighting",
"time": "13.684",
"value": "Brian Finch_sentence_1"
},
{
"operation": "finish",
"time": "27.663",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.