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_7120
|
Jaime Lannister is a fictional character in a series whose genre of speculative fiction involves magical elements in the novel by American author George R. R. Martin.
|
{
"evidence": [
"Jaime Lannister",
"Fantasy"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJaime Lannister is a fictional character in a series whose genre of speculative fiction involves magical elements in the novel by American author George R. R. Martin.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jaime Lannister",
"Fantasy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3984,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
52849
|
Liver X receptor alpha is one of three liver X receptors, which activate thyroid hormones and play a role in regulating the expression of target genes.
|
{
"evidence": [
"Liver X receptor alpha_sentence_4",
"Liver X receptor_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:\nLiver X receptor alpha is one of three liver X receptors, which activate thyroid hormones and play a role in regulating the expression of target genes.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Liver X receptor alpha_sentence_4",
"Liver X receptor_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 3985,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.235",
"value": "?search="
},
{
"operation": "search",
"time": "9.712",
"value": "Liver X receptor alpha"
},
{
"operation": "Now on",
"time": "10.627",
"value": "Liver X receptor alpha"
},
{
"operation": "Page search",
"time": "83.24",
"value": "receptor"
},
{
"operation": "Page search",
"time": "94.317",
"value": "liver x receptor"
},
{
"operation": "Page search",
"time": "95.464",
"value": "liver x receptor"
},
{
"operation": "Page search",
"time": "119.467",
"value": "thyroid"
},
{
"operation": "hyperlink",
"time": "126.646",
"value": "liver X receptors"
},
{
"operation": "Now on",
"time": "127.626",
"value": "Liver X receptor"
},
{
"operation": "Page search",
"time": "264.11",
"value": "alpha"
},
{
"operation": "Page search",
"time": "264.826",
"value": "alpha"
},
{
"operation": "Page search",
"time": "289.935",
"value": "thyroid"
},
{
"operation": "back-button-clicked",
"time": "301.073",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "301.139",
"value": "Liver X receptor alpha"
},
{
"operation": "Page search",
"time": "306.408",
"value": "thyroid"
},
{
"operation": "Highlighting",
"time": "309.631",
"value": "Liver X receptor alpha_sentence_4"
},
{
"operation": "hyperlink",
"time": "367.541",
"value": "liver X receptors"
},
{
"operation": "Now on",
"time": "369.21",
"value": "Liver X receptor"
},
{
"operation": "Highlighting",
"time": "372.211",
"value": "Liver X receptor_sentence_0"
},
{
"operation": "finish",
"time": "426.3",
"value": "finish"
}
]
}
|
exfever_7950
|
Lady Gaga is a singer who signed a deal with a record label founded by Akon that served as a joint-venture deal between an American record executive and media proprietor and Interscope Records.
|
{
"evidence": [
"Lady Gaga",
"KonLive Distribution",
"Jimmy Iovine"
],
"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:\nLady Gaga is a singer who signed a deal with a record label founded by Akon that served as a joint-venture deal between an American record executive and media proprietor and Interscope Records.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lady Gaga",
"KonLive Distribution",
"Jimmy Iovine"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3986,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6365
|
Cleopatra is an American epic historical drama film directed by Joseph L. Mankiewicz that starred a Welsh actor noted for his mellifluous baritone voice.
|
{
"evidence": [
"Cleopatra (1963 film)",
"Richard Burton"
],
"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:\nCleopatra is an American epic historical drama film directed by Joseph L. Mankiewicz that starred a Welsh actor noted for his mellifluous baritone voice.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Cleopatra (1963 film)",
"Richard Burton"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3987,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
93657
|
Pierre Amandry was born at Troyes in 1912, he joined the École Normale Supérieure in 1933, he carried out excavations which led him to be interested in oriental art.
|
{
"evidence": [
"Pierre Amandry_sentence_1",
"Pierre Amandry_sentence_3",
"Pierre Amandry_sentence_4",
"Pierre Amandry_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPierre Amandry was born at Troyes in 1912, he joined the École Normale Supérieure in 1933, he carried out excavations which led him to be interested in oriental art.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Pierre Amandry_sentence_1",
"Pierre Amandry_sentence_3",
"Pierre Amandry_sentence_4",
"Pierre Amandry_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3988,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.348",
"value": "?search="
},
{
"operation": "search",
"time": "11.951",
"value": "Pierre Amandry"
},
{
"operation": "Now on",
"time": "12.753",
"value": "Pierre Amandry"
},
{
"operation": "Highlighting",
"time": "18.715",
"value": "Pierre Amandry_sentence_1"
},
{
"operation": "Page search",
"time": "28.913",
"value": "normale"
},
{
"operation": "Highlighting",
"time": "31.328",
"value": "Pierre Amandry_sentence_3"
},
{
"operation": "Highlighting",
"time": "40.448",
"value": "Pierre Amandry_sentence_4"
},
{
"operation": "Page search",
"time": "52.124",
"value": "oriental"
},
{
"operation": "Highlighting",
"time": "54.896",
"value": "Pierre Amandry_sentence_5"
},
{
"operation": "finish",
"time": "212.623",
"value": "finish"
}
]
}
|
exfever_7750
|
Howard Allan Stern is an American radio and television personality who developed his on-air personality through morning positions at WRNW in a suburban village in Westchester County, New York, United States.
|
{
"evidence": [
"Howard Stern",
"Briarcliff Manor, New York"
],
"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 Allan Stern is an American radio and television personality who developed his on-air personality through morning positions at WRNW in a suburban village in Westchester County, New York, United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Howard Stern",
"Briarcliff Manor, New York"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3989,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7283
|
Gary Kasparov is the current Chairman of Human Rights Foundation.
|
{
"evidence": [
"Garry Kasparov",
"Magnus Carlsen"
],
"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:\nGary Kasparov is the current Chairman of Human Rights Foundation.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Garry Kasparov",
"Magnus Carlsen"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 3990,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3491
|
The Blacklist is an American crime thriller television series starring known for playing Ryan Matthews on the CW American teen drama series, developed by Rob Thomas, Gabe Sachs, and Jeff Judah.
|
{
"evidence": [
"The Blacklist (TV series)",
"Ryan Eggold",
"90210 (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:\nThe Blacklist is an American crime thriller television series starring known for playing Ryan Matthews on the CW American teen drama series, developed by Rob Thomas, Gabe Sachs, and Jeff Judah.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Blacklist (TV series)",
"Ryan Eggold",
"90210 (TV series)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3991,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
81152
|
Davies Chisopa defeated the incumbent MP, Sydney Chisanga of the Movement for Multi-Party Democracy during the 2011 general elections and contested the results at the Court of Appeals.
|
{
"evidence": [
"Davies Chisopa_sentence_2",
"Davies Chisopa_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:\nDavies Chisopa defeated the incumbent MP, Sydney Chisanga of the Movement for Multi-Party Democracy during the 2011 general elections and contested the results at the Court of Appeals.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Davies Chisopa_sentence_2",
"Davies Chisopa_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 3992,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.655",
"value": "?search="
},
{
"operation": "search",
"time": "92.768",
"value": "Davies Chisopa"
},
{
"operation": "Now on",
"time": "93.526",
"value": "Davies Chisopa"
},
{
"operation": "Highlighting",
"time": "106.891",
"value": "Davies Chisopa_sentence_2"
},
{
"operation": "Highlighting",
"time": "167.916",
"value": "Davies Chisopa_sentence_3"
},
{
"operation": "finish",
"time": "214.79",
"value": "finish"
}
]
}
|
exfever_2620
|
Iggy Azalea is a Nigerian model whose debut studio album was released on April 21, 2014, by Def Jam Recordings, has peaked among the top five on several charts worldwide.
|
{
"evidence": [
"Iggy Azalea",
"The New Classic"
],
"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:\nIggy Azalea is a Nigerian model whose debut studio album was released on April 21, 2014, by Def Jam Recordings, has peaked among the top five on several charts worldwide.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Iggy Azalea",
"The New Classic"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 3993,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
25609
|
An American physicist and chemist born in Mount Enterprise, Texas, Elise Harmon, despite the fact that she was not an active student at the University of North Texas, she was able to earn a Bachelor’s degree in Chemistry in 1931 and she was also considered as an influential contributor to the miniaturization of computers.
|
{
"evidence": [
"Elise Harmon_sentence_17",
"Elise Harmon_sentence_18"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAn American physicist and chemist born in Mount Enterprise, Texas, Elise Harmon, despite the fact that she was not an active student at the University of North Texas, she was able to earn a Bachelor’s degree in Chemistry in 1931 and she was also considered as an influential contributor to the miniaturization of computers.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Elise Harmon_sentence_17",
"Elise Harmon_sentence_18"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 3994,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.1",
"value": "?search="
},
{
"operation": "search",
"time": "24.697",
"value": "Elise Harmon"
},
{
"operation": "Now on",
"time": "25.399",
"value": "Elise Harmon"
},
{
"operation": "Highlighting",
"time": "27.397",
"value": "Elise Harmon_sentence_0"
},
{
"operation": "Page search",
"time": "39.822",
"value": "chem"
},
{
"operation": "page-search-reset",
"time": "43.566",
"value": "page-search-reset"
},
{
"operation": "Page search",
"time": "59.339",
"value": "North Texas"
},
{
"operation": "Page search",
"time": "59.623",
"value": "North Texas"
},
{
"operation": "Highlighting",
"time": "97.323",
"value": "Elise Harmon_sentence_17"
},
{
"operation": "Highlighting",
"time": "97.873",
"value": "Elise Harmon_sentence_18"
},
{
"operation": "Highlighting",
"time": "101.973",
"value": "Elise Harmon_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "101.975",
"value": "Elise Harmon_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "101.975",
"value": "Elise Harmon_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "102.029",
"value": "Elise Harmon_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "102.03",
"value": "Elise Harmon_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "102.031",
"value": "Elise Harmon_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "102.959",
"value": "Elise Harmon_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "102.96",
"value": "Elise Harmon_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "102.96",
"value": "Elise Harmon_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "103.282",
"value": "Elise Harmon_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "103.283",
"value": "Elise Harmon_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "103.283",
"value": "Elise Harmon_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "131.72",
"value": "Elise Harmon_sentence_0"
},
{
"operation": "finish",
"time": "133.456",
"value": "finish"
}
]
}
|
exfever_161
|
The South Island, also officially named Te Waipounamu, is bordered on the north by Cook Strait, the south and east by the Pacific Ocean, and the west by the marginal sea of the South Pacific Ocean, situated between Australia and an island country in the southwestern Pacific Ocean.
|
{
"evidence": [
"South Island",
"Tasman Sea",
"New Zealand"
],
"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 South Island, also officially named Te Waipounamu, is bordered on the north by Cook Strait, the south and east by the Pacific Ocean, and the west by the marginal sea of the South Pacific Ocean, situated between Australia and an island country in the southwestern Pacific Ocean.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"South Island",
"Tasman Sea",
"New Zealand"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3995,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8044
|
Wonder Woman is the most famous superheroine appearing in American comic books published by DC Comics.
|
{
"evidence": [
"Wonder Woman",
"William Moulton Marston",
"Elizabeth Holloway Marston"
],
"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:\nWonder Woman is the most famous superheroine appearing in American comic books published by DC Comics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Wonder Woman",
"William Moulton Marston",
"Elizabeth Holloway Marston"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 3996,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7632
|
Dracula is a novel by an Irish author who became a personal assistant of an English stage actor in the Victorian era.
|
{
"evidence": [
"Dracula",
"Bram Stoker",
"Henry Irving"
],
"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:\nDracula is a novel by an Irish author who became a personal assistant of an English stage actor in the Victorian era.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dracula",
"Bram Stoker",
"Henry Irving"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 3997,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2445
|
Muhammad Ali won a gold medal at the "Giochi Olimpici Estivi del 1960", in lingua italiana.
|
{
"evidence": [
"Muhammad Ali",
"1960 Summer Olympics",
"Italian language"
],
"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:\nMuhammad Ali won a gold medal at the \"Giochi Olimpici Estivi del 1960\", in lingua italiana.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Muhammad Ali",
"1960 Summer Olympics",
"Italian language"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 3998,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
51555
|
Beau Purple was withdrawn from the Kentucky Derby, a Grade I stakes race for three-year-old Thoroughbreds, after his 1960 Derby Trial Stakes win and has ran only five times in 1961.
|
{
"evidence": [
"Beau Purple_sentence_9"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBeau Purple was withdrawn from the Kentucky Derby, a Grade I stakes race for three-year-old Thoroughbreds, after his 1960 Derby Trial Stakes win and has ran only five times in 1961.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Beau Purple_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 3999,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.007",
"value": "?search="
},
{
"operation": "search",
"time": "9.02",
"value": "Beau Purple"
},
{
"operation": "Now on",
"time": "9.701",
"value": "Beau Purple"
},
{
"operation": "Page search",
"time": "30.908",
"value": "withdrawn"
},
{
"operation": "Highlighting",
"time": "33.096",
"value": "Beau Purple_sentence_6"
},
{
"operation": "hyperlink",
"time": "39.777",
"value": "Kentucky Derby"
},
{
"operation": "Now on",
"time": "41.51",
"value": "Kentucky Derby"
},
{
"operation": "Highlighting",
"time": "43.94",
"value": "Kentucky Derby_sentence_3"
},
{
"operation": "Highlighting",
"time": "43.941",
"value": "Kentucky Derby_title"
},
{
"operation": "back-button-clicked",
"time": "58.639",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "58.716",
"value": "Beau Purple"
},
{
"operation": "Highlighting",
"time": "76.509",
"value": "Beau Purple_sentence_9"
},
{
"operation": "Page search",
"time": "89.214",
"value": "1961"
},
{
"operation": "Highlighting deleted",
"time": "98.63",
"value": "Beau Purple_sentence_9"
},
{
"operation": "Highlighting",
"time": "99.717",
"value": "Beau Purple_sentence_9"
},
{
"operation": "finish",
"time": "127.116",
"value": "finish"
}
]
}
|
exfever_6882
|
Frederick Seibert was the richest American television producer.
|
{
"evidence": [
"Fred Seibert",
"YouTube Next Lab and Audience Development Group"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFrederick Seibert was the richest American television producer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Fred Seibert",
"YouTube Next Lab and Audience Development Group"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4000,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7124
|
The Hills Have Eyes is a 1991 American psychological horror film directed by Jonathan Demme and adapted from Thomas Harris's psychological horror novel featuring the cannibalistic serial killer Dr. Hannibal Lecter.
|
{
"evidence": [
"The Silence of the Lambs (film)",
"The Silence of the Lambs (novel)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Hills Have Eyes is a 1991 American psychological horror film directed by Jonathan Demme and adapted from Thomas Harris's psychological horror novel featuring the cannibalistic serial killer Dr. Hannibal Lecter.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Silence of the Lambs (film)",
"The Silence of the Lambs (novel)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4001,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
94460
|
Bavaria is connected to Hungary through a bicycle path, as well as Budapest to Serbia.
|
{
"evidence": [
"EV6 The Rivers Route_sentence_0",
"EV6 The Rivers Route_sentence_27"
],
"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:\nBavaria is connected to Hungary through a bicycle path, as well as Budapest to Serbia.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"EV6 The Rivers Route_sentence_0",
"EV6 The Rivers Route_sentence_27"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4002,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.452",
"value": "?search="
},
{
"operation": "search",
"time": "265.529",
"value": "Budapest"
},
{
"operation": "Now on",
"time": "269.88",
"value": "Budapest"
},
{
"operation": "search",
"time": "1023.468",
"value": "Danube Cycle Path"
},
{
"operation": "Now on",
"time": "1024.529",
"value": "Danube Cycle Path"
},
{
"operation": "hyperlink",
"time": "1193.146",
"value": "holiday cycle route"
},
{
"operation": "Now on",
"time": "1194.089",
"value": "Bicycle touring"
},
{
"operation": "back-button-clicked",
"time": "1300.994",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "1301.56",
"value": "EV6 The Rivers Route"
},
{
"operation": "Highlighting",
"time": "1547.891",
"value": "EV6 The Rivers Route_sentence_27"
},
{
"operation": "Highlighting",
"time": "1664.087",
"value": "EV6 The Rivers Route_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "1702.626",
"value": "EV6 The Rivers Route_sentence_27"
},
{
"operation": "Highlighting deleted",
"time": "1706.71",
"value": "EV6 The Rivers Route_sentence_1"
},
{
"operation": "Highlighting",
"time": "1708.27",
"value": "EV6 The Rivers Route_sentence_0"
},
{
"operation": "Highlighting",
"time": "1720.582",
"value": "EV6 The Rivers Route_sentence_1"
},
{
"operation": "Highlighting",
"time": "1736.59",
"value": "EV6 The Rivers Route_sentence_27"
},
{
"operation": "Highlighting deleted",
"time": "1815.526",
"value": "EV6 The Rivers Route_sentence_1"
},
{
"operation": "finish",
"time": "1833.447",
"value": "finish"
}
]
}
|
78846
|
Bartłomiej Nowodworski was born 1552 in Poland, in recognition of his contribution the school became known as Collegium Nowodworskiego and in 1898 the school relocated from the university area.
|
{
"evidence": [
"Bartłomiej Nowodworski High School_sentence_6",
"Bartłomiej Nowodworski High School_sentence_7",
"Bartłomiej Nowodworski High School_sentence_9"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBartłomiej Nowodworski was born 1552 in Poland, in recognition of his contribution the school became known as Collegium Nowodworskiego and in 1898 the school relocated from the university area.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Bartłomiej Nowodworski High School_sentence_6",
"Bartłomiej Nowodworski High School_sentence_7",
"Bartłomiej Nowodworski High School_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4003,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.049",
"value": "?search="
},
{
"operation": "search",
"time": "10.805",
"value": "Bartłomiej Nowodworski High School"
},
{
"operation": "Now on",
"time": "12.273",
"value": "Bartłomiej Nowodworski High School"
},
{
"operation": "Highlighting",
"time": "17.251",
"value": "Bartłomiej Nowodworski High School_sentence_1"
},
{
"operation": "back-button-clicked",
"time": "29.337",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "29.454",
"value": "?search="
},
{
"operation": "search",
"time": "44.721",
"value": "Bartłomiej Nowodworski High School"
},
{
"operation": "Now on",
"time": "45.596",
"value": "Bartłomiej Nowodworski High School"
},
{
"operation": "Highlighting",
"time": "54.063",
"value": "Bartłomiej Nowodworski High School_sentence_6"
},
{
"operation": "Highlighting",
"time": "67.879",
"value": "Bartłomiej Nowodworski High School_sentence_7"
},
{
"operation": "Highlighting",
"time": "74.235",
"value": "Bartłomiej Nowodworski High School_sentence_9"
},
{
"operation": "finish",
"time": "91.012",
"value": "finish"
}
]
}
|
exfever_7274
|
Scooby-Doo is an American animated franchise comprising much animated television series produced in 1969 featuring a fictional character in the American animated series Scooby-Doo, portrayed by voice actor Frank Welker.
|
{
"evidence": [
"Scooby-Doo",
"Fred Jones (Scooby-Doo)"
],
"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:\nScooby-Doo is an American animated franchise comprising much animated television series produced in 1969 featuring a fictional character in the American animated series Scooby-Doo, portrayed by voice actor Frank Welker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Scooby-Doo",
"Fred Jones (Scooby-Doo)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4004,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4746
|
Jerome was the first priest to be considered a saint in the Eastern Orthodox Church.
|
{
"evidence": [
"Jerome",
"Emona"
],
"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:\nJerome was the first priest to be considered a saint in the Eastern Orthodox Church.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jerome",
"Emona"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4005,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8064
|
Jerome was a Christian priest who was born in a town in the Roman province of Dalmatia that was destroyed by the Germanic people who played a major role in the fall of the Western Roman Empire and the emergence of medieval Europe.
|
{
"evidence": [
"Jerome",
"Stridon",
"Goths"
],
"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:\nJerome was a Christian priest who was born in a town in the Roman province of Dalmatia that was destroyed by the Germanic people who played a major role in the fall of the Western Roman Empire and the emergence of medieval Europe.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jerome",
"Stridon",
"Goths"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4006,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
2232
|
The United Nations Security Council Resolution 442 was adopted in December 6, 1978 that accepted the application of the Commonwealth of Dominica (officially the Commonwealth of Dominica, is an island country in the Caribbean, and not to be confused with the Dominican Republic) as a member of the Uniter Nations.
|
{
"evidence": [
"United Nations Security Council Resolution 442_sentence_0",
"Dominica_sentence_0",
"Dominica_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe United Nations Security Council Resolution 442 was adopted in December 6, 1978 that accepted the application of the Commonwealth of Dominica (officially the Commonwealth of Dominica, is an island country in the Caribbean, and not to be confused with the Dominican Republic) as a member of the Uniter Nations.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"United Nations Security Council Resolution 442_sentence_0",
"Dominica_sentence_0",
"Dominica_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4007,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "12.88",
"value": "united nations security council resoulution 442"
},
{
"operation": "hyperlink",
"time": "20.377",
"value": "united nations security council resolution 442"
},
{
"operation": "hyperlink",
"time": "32.301",
"value": "United Nations Security Council Resolution 442"
},
{
"operation": "Highlighting",
"time": "43.589",
"value": "United Nations Security Council Resolution 442_sentence_0"
},
{
"operation": "hyperlink",
"time": "60.886",
"value": "Commonwealth of Dominica"
},
{
"operation": "Highlighting",
"time": "70.295",
"value": "Dominica_sentence_0"
},
{
"operation": "Highlighting",
"time": "72.303",
"value": "Dominica_sentence_1"
},
{
"operation": "finish",
"time": "91.569",
"value": "finish"
}
]
}
|
31933
|
Świnoujście, a city on the Baltic Sea which was established in the twelfth century became part of Poland after the Soviet Union imposed new borders, which led to the entire population being expelled.
|
{
"evidence": [
"Świnoujście_sentence_0",
"Świnoujście_sentence_34",
"Świnoujście_sentence_35",
"Świnoujście_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:\nŚwinoujście, a city on the Baltic Sea which was established in the twelfth century became part of Poland after the Soviet Union imposed new borders, which led to the entire population being expelled.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Świnoujście_sentence_0",
"Świnoujście_sentence_34",
"Świnoujście_sentence_35",
"Świnoujście_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4008,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.83",
"value": "?search="
},
{
"operation": "search",
"time": "14.538",
"value": "Świnoujście"
},
{
"operation": "Now on",
"time": "16.171",
"value": "Świnoujście"
},
{
"operation": "Highlighting",
"time": "22",
"value": "Świnoujście_sentence_0"
},
{
"operation": "Highlighting",
"time": "48.241",
"value": "Świnoujście_sentence_34"
},
{
"operation": "Highlighting",
"time": "59.834",
"value": "Świnoujście_sentence_35"
},
{
"operation": "Highlighting",
"time": "166.306",
"value": "Świnoujście_sentence_11"
},
{
"operation": "finish",
"time": "200.997",
"value": "finish"
}
]
}
|
exfever_8987
|
Sue Sylvester is a fictional character that appears in Glee's?first TV airtime, premiered on the ABC network on May 19, 2009.
|
{
"evidence": [
"Sue Sylvester",
"Pilot (Glee)"
],
"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:\nSue Sylvester is a fictional character that appears in Glee's?first TV airtime, premiered on the ABC network on May 19, 2009.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sue Sylvester",
"Pilot (Glee)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4009,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7559
|
Eric Overmyer,?best known for his work on The Wire, is also known for co-writing The Corner: A Year in the Life of an Inner-City Neighborhood with a novelist who was also a?former Baltimore police detective for the homicide and narcotics divisions.
|
{
"evidence": [
"David Simon",
"Ed Burns"
],
"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:\nEric Overmyer,?best known for his work on The Wire, is also known for co-writing The Corner: A Year in the Life of an Inner-City Neighborhood with a novelist who was also a?former Baltimore police detective for the homicide and narcotics divisions.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"David Simon",
"Ed Burns"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4010,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6425
|
Brooks Wackerman is an Indian musician and songwriter who performed as a touring member of an American funk metal supergroup led by Suicidal Tendencies frontman Mike Muir.
|
{
"evidence": [
"Brooks Wackerman",
"Infectious Grooves"
],
"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:\nBrooks Wackerman is an Indian musician and songwriter who performed as a touring member of an American funk metal supergroup led by Suicidal Tendencies frontman Mike Muir.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Brooks Wackerman",
"Infectious Grooves"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4011,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7857
|
Jon Snow is the youngest fictional character in the A Song of Ice and Fire series.
|
{
"evidence": [
"Jon Snow (character)",
"Kit Harington",
"Golden Globe Award"
],
"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:\nJon Snow is the youngest fictional character in the A Song of Ice and Fire series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jon Snow (character)",
"Kit Harington",
"Golden Globe Award"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4012,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
88229
|
Sani Monastery is a monastery of the Kagyu sect of Tibetan Buddhism and oldest religious site in the whole region of Ladakh and Zanskar, dedicated to the deity Avalokiteshvara that is located in northern India
|
{
"evidence": [
"Sani Monastery_sentence_3"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSani Monastery is a monastery of the Kagyu sect of Tibetan Buddhism and oldest religious site in the whole region of Ladakh and Zanskar, dedicated to the deity Avalokiteshvara that is located in northern India\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sani Monastery_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4013,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.094",
"value": "?search="
},
{
"operation": "search",
"time": "19.112",
"value": "Sani Monastery"
},
{
"operation": "Now on",
"time": "20.036",
"value": "Sani Monastery"
},
{
"operation": "Highlighting",
"time": "68.48",
"value": "Sani Monastery_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "69.566",
"value": "Sani Monastery_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "69.586",
"value": "Sani Monastery_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "69.587",
"value": "Sani Monastery_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "69.588",
"value": "Sani Monastery_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "81.823",
"value": "Sani Monastery_sentence_2"
},
{
"operation": "Highlighting",
"time": "91.37",
"value": "Sani Monastery_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "101.39",
"value": "Sani Monastery_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "104.107",
"value": "Sani Monastery_cell_0_3_1"
},
{
"operation": "Highlighting deleted",
"time": "104.112",
"value": "Sani Monastery_header_cell_0_3_0"
},
{
"operation": "Highlighting deleted",
"time": "104.115",
"value": "Sani Monastery_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "104.116",
"value": "Sani Monastery_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "134.545",
"value": "finish"
}
]
}
|
exfever_2543
|
Dan Aykroyd was an original member of the "Not Ready for Prime Time Players" on Chappelle's Show, where he performed with his friend, an American comedian, actor, and musician.
|
{
"evidence": [
"Dan Aykroyd",
"John Belushi"
],
"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:\nDan Aykroyd was an original member of the \"Not Ready for Prime Time Players\" on Chappelle's Show, where he performed with his friend, an American comedian, actor, and musician.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dan Aykroyd",
"John Belushi"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4014,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
23484
|
Rosa Mariani was the older sister of bass Luciano Mariani—an Italian operatic bass born in Cremona—who she toured Italy as well as performed in London’s King’s Theatre with in 1832.
|
{
"evidence": [
"Rosa Mariani_sentence_3",
"Luciano Mariani_sentence_0",
"Luciano Mariani_sentence_5",
"Rosa Mariani_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:\nRosa Mariani was the older sister of bass Luciano Mariani—an Italian operatic bass born in Cremona—who she toured Italy as well as performed in London’s King’s Theatre with in 1832.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Rosa Mariani_sentence_3",
"Luciano Mariani_sentence_0",
"Luciano Mariani_sentence_5",
"Rosa Mariani_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4015,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.526",
"value": "?search="
},
{
"operation": "search",
"time": "13.022",
"value": "Rosa Mariani"
},
{
"operation": "Now on",
"time": "13.526",
"value": "Rosa Mariani"
},
{
"operation": "Page search",
"time": "18.226",
"value": "sister"
},
{
"operation": "Highlighting",
"time": "21.002",
"value": "Rosa Mariani_sentence_3"
},
{
"operation": "hyperlink",
"time": "24.889",
"value": "Luciano Mariani"
},
{
"operation": "Now on",
"time": "25.261",
"value": "Luciano Mariani"
},
{
"operation": "Highlighting",
"time": "29.719",
"value": "Luciano Mariani_sentence_0"
},
{
"operation": "Highlighting",
"time": "35.471",
"value": "Luciano Mariani_sentence_5"
},
{
"operation": "back-button-clicked",
"time": "43.164",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "43.222",
"value": "Rosa Mariani"
},
{
"operation": "Page search",
"time": "52.156",
"value": "tour"
},
{
"operation": "Page search",
"time": "60.116",
"value": "king"
},
{
"operation": "Highlighting",
"time": "62.094",
"value": "Rosa Mariani_sentence_4"
},
{
"operation": "finish",
"time": "103.146",
"value": "finish"
}
]
}
|
exfever_6543
|
The Caspian Sea lies west of the broad steppes of a subregion of Asia that stretches from Afghanistan and Iran in the south to Russia in the north.
|
{
"evidence": [
"Caspian Sea",
"Central Asia"
],
"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 Caspian Sea lies west of the broad steppes of a subregion of Asia that stretches from Afghanistan and Iran in the south to Russia in the north.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Caspian Sea",
"Central Asia"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4016,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76251
|
Procambarus Natchitoches, also known as the Red River creek crayfish, don't prefer to live with the water current and are located in all the countries in the world.
|
{
"evidence": [
"Procambarus natchitochae_sentence_9",
"Procambarus natchitochae_sentence_10"
],
"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:\nProcambarus Natchitoches, also known as the Red River creek crayfish, don't prefer to live with the water current and are located in all the countries in the world.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Procambarus natchitochae_sentence_9",
"Procambarus natchitochae_sentence_10"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4017,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.694",
"value": "?search="
},
{
"operation": "search",
"time": "11.574",
"value": "Contains...Procambarus Natchitoches"
},
{
"operation": "Now on",
"time": "12.517",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "20.618",
"value": "Procambarus natchitochae"
},
{
"operation": "Now on",
"time": "21.141",
"value": "Procambarus natchitochae"
},
{
"operation": "Highlighting",
"time": "25.483",
"value": "Procambarus natchitochae_sentence_0"
},
{
"operation": "Highlighting",
"time": "62.181",
"value": "Procambarus natchitochae_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "64.691",
"value": "Procambarus natchitochae_sentence_0"
},
{
"operation": "Highlighting",
"time": "97.053",
"value": "Procambarus natchitochae_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "112.695",
"value": "Procambarus natchitochae_sentence_0"
},
{
"operation": "Highlighting",
"time": "113.785",
"value": "Procambarus natchitochae_sentence_10"
},
{
"operation": "finish",
"time": "123.506",
"value": "finish"
}
]
}
|
exfever_8946
|
Ancient transcontinental country was a civilization of ancient Africa.
|
{
"evidence": [
"Ancient Egypt",
"Egypt"
],
"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:\nAncient transcontinental country was a civilization of ancient Africa.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ancient Egypt",
"Egypt"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4018,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7978
|
The Punisher is an American television series derived from the already existing work of the first Marvel Netflix series, Daredevil.
|
{
"evidence": [
"The Punisher (TV series)",
"Spin-off (media)"
],
"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 Punisher is an American television series derived from the already existing work of the first Marvel Netflix series, Daredevil.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Punisher (TV series)",
"Spin-off (media)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4019,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76890
|
In Dunster Dovecote the priory was abolished in 1540 and was sold to the family from Dunster Castle (now a country house in Somerset, England), and in the 18th century the floor and door level was raised.
|
{
"evidence": [
"Dunster Dovecote_sentence_9"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn Dunster Dovecote the priory was abolished in 1540 and was sold to the family from Dunster Castle (now a country house in Somerset, England), and in the 18th century the floor and door level was raised.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dunster Dovecote_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4020,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.759",
"value": "?search="
},
{
"operation": "search",
"time": "23.474",
"value": "Dunster Dovecote"
},
{
"operation": "Now on",
"time": "24.164",
"value": "Dunster Dovecote"
},
{
"operation": "Highlighting",
"time": "87.766",
"value": "Dunster Dovecote_sentence_9"
},
{
"operation": "finish",
"time": "107.052",
"value": "finish"
}
]
}
|
34701
|
Helen Troy was a young woman who had engaged Massage as a celebrant for her wedding to Roy Francis some four weeks previously and some 400 people attended.
|
{
"evidence": [
"Funeral celebrant_sentence_26"
],
"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:\nHelen Troy was a young woman who had engaged Massage as a celebrant for her wedding to Roy Francis some four weeks previously and some 400 people attended.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Funeral celebrant_sentence_26"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4021,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.667",
"value": "?search="
},
{
"operation": "search",
"time": "10.693",
"value": "Helen Troy"
},
{
"operation": "Now on",
"time": "12.016",
"value": "Helen Troy"
},
{
"operation": "search",
"time": "64.006",
"value": "Roy Francis"
},
{
"operation": "Now on",
"time": "64.854",
"value": "Roy Francis"
},
{
"operation": "hyperlink",
"time": "76.596",
"value": "Roy Francis (musician)"
},
{
"operation": "Now on",
"time": "77.117",
"value": "Roy Francis (musician)"
},
{
"operation": "search",
"time": "91.264",
"value": "Helen Troy"
},
{
"operation": "Now on",
"time": "92.165",
"value": "Helen of Troy"
},
{
"operation": "search",
"time": "118.582",
"value": "Massage as a celebrant for her wedding"
},
{
"operation": "Now on",
"time": "120.6",
"value": "undefined"
},
{
"operation": "search",
"time": "140.899",
"value": "Helen Roy Francis"
},
{
"operation": "Now on",
"time": "142.695",
"value": "undefined"
},
{
"operation": "search",
"time": "193.207",
"value": "Roy Francis"
},
{
"operation": "Now on",
"time": "193.756",
"value": "Roy Francis"
},
{
"operation": "hyperlink",
"time": "196.077",
"value": "Roy Francis (rugby)"
},
{
"operation": "Now on",
"time": "196.603",
"value": "Roy Francis (rugby)"
},
{
"operation": "back-button-clicked",
"time": "200.997",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "201.081",
"value": "Roy Francis"
},
{
"operation": "hyperlink",
"time": "204.564",
"value": "Roy Francis (naval officer)"
},
{
"operation": "Now on",
"time": "205.081",
"value": "Roy Francis (Royal Navy officer)"
},
{
"operation": "back-button-clicked",
"time": "207.511",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "207.592",
"value": "Roy Francis"
},
{
"operation": "search",
"time": "231.256",
"value": "Contains...engaged Massage"
},
{
"operation": "Now on",
"time": "232.529",
"value": "undefined"
},
{
"operation": "search",
"time": "257.47",
"value": "Contains...wedding to Roy Francis"
},
{
"operation": "Now on",
"time": "258.739",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "265.962",
"value": "Funeral celebrant"
},
{
"operation": "Now on",
"time": "266.644",
"value": "Funeral celebrant"
},
{
"operation": "Highlighting",
"time": "270.086",
"value": "Funeral celebrant_sentence_26"
},
{
"operation": "finish",
"time": "282.38",
"value": "finish"
}
]
}
|
exfever_6156
|
The Martian is a science fiction film adapted by an American screenwriter, director, and producer who made his directorial debut with an American horror comedy film starring Chris Hemsworth.
|
{
"evidence": [
"The Martian (film)",
"Drew Goddard",
"The Cabin in the Woods"
],
"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 Martian is a science fiction film adapted by an American screenwriter, director, and producer who made his directorial debut with an American horror comedy film starring Chris Hemsworth.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Martian (film)",
"Drew Goddard",
"The Cabin in the Woods"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4022,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29951
|
The "Society to Build a Mosque in Yenişehir, Ankara,” formed on December 8, 1944, by Ahmet Hamdi Akseki, along with seventy-two founding members was the group that rejected the idea of building the Kocatepe Mosque.
|
{
"evidence": [
"Kocatepe Mosque_sentence_3",
"Kocatepe Mosque_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:\nThe \"Society to Build a Mosque in Yenişehir, Ankara,” formed on December 8, 1944, by Ahmet Hamdi Akseki, along with seventy-two founding members was the group that rejected the idea of building the Kocatepe Mosque.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Kocatepe Mosque_sentence_3",
"Kocatepe Mosque_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4023,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.035",
"value": "?search="
},
{
"operation": "search",
"time": "21.525",
"value": "The Society to Build a Mosque"
},
{
"operation": "Now on",
"time": "24.105",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "40.529",
"value": "Kocatepe Mosque"
},
{
"operation": "Now on",
"time": "41.116",
"value": "Kocatepe Mosque"
},
{
"operation": "Highlighting",
"time": "56.676",
"value": "Kocatepe Mosque_sentence_3"
},
{
"operation": "Highlighting",
"time": "203.677",
"value": "Kocatepe Mosque_sentence_4"
},
{
"operation": "finish",
"time": "248.39",
"value": "finish"
}
]
}
|
44636
|
James Chilton (a passenger on the Mayflower–a ship that transported passengers to the New World in 1620) had three daughters who lived to maturity: Mary, Ingel, and Isabella (who was baptized in 1586/87).
|
{
"evidence": [
"James Chilton_sentence_24",
"James Chilton_sentence_25",
"James Chilton_sentence_28",
"James Chilton_sentence_11"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJames Chilton (a passenger on the Mayflower–a ship that transported passengers to the New World in 1620) had three daughters who lived to maturity: Mary, Ingel, and Isabella (who was baptized in 1586/87).\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"James Chilton_sentence_24",
"James Chilton_sentence_25",
"James Chilton_sentence_28",
"James Chilton_sentence_11"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4024,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.268",
"value": "?search="
},
{
"operation": "search",
"time": "10.022",
"value": "James Chilton"
},
{
"operation": "Now on",
"time": "10.954",
"value": "James Chilton"
},
{
"operation": "Highlighting",
"time": "63.934",
"value": "James Chilton_sentence_0"
},
{
"operation": "hyperlink",
"time": "69.047",
"value": "Mayflower"
},
{
"operation": "Now on",
"time": "69.935",
"value": "Mayflower"
},
{
"operation": "Highlighting",
"time": "72.248",
"value": "Mayflower_sentence_2"
},
{
"operation": "Highlighting",
"time": "72.249",
"value": "Mayflower_title"
},
{
"operation": "back-button-clicked",
"time": "78.412",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "78.479",
"value": "James Chilton"
},
{
"operation": "hyperlink",
"time": "120.265",
"value": "freeman"
},
{
"operation": "Now on",
"time": "120.982",
"value": "Freeman (Colonial)"
},
{
"operation": "back-button-clicked",
"time": "134.446",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "134.514",
"value": "James Chilton"
},
{
"operation": "Highlighting",
"time": "188.946",
"value": "James Chilton_sentence_24"
},
{
"operation": "Highlighting",
"time": "189.79",
"value": "James Chilton_sentence_25"
},
{
"operation": "Highlighting",
"time": "207.757",
"value": "James Chilton_sentence_28"
},
{
"operation": "Highlighting deleted",
"time": "212.571",
"value": "James Chilton_sentence_0"
},
{
"operation": "Highlighting",
"time": "214.135",
"value": "James Chilton_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "216.085",
"value": "James Chilton_sentence_0"
},
{
"operation": "hyperlink",
"time": "216.733",
"value": "Mayflower"
},
{
"operation": "Now on",
"time": "217.915",
"value": "Mayflower"
},
{
"operation": "Highlighting deleted",
"time": "224.289",
"value": "Mayflower_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "224.291",
"value": "Mayflower_title"
},
{
"operation": "back-button-clicked",
"time": "226.358",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "226.432",
"value": "James Chilton"
},
{
"operation": "Highlighting",
"time": "272.667",
"value": "James Chilton_sentence_11"
},
{
"operation": "finish",
"time": "284.692",
"value": "finish"
}
]
}
|
exfever_7118
|
Fergie Duhamel was the only female member of the group Black Eyed Peas.
|
{
"evidence": [
"Fergie (singer)",
"Hip hop music"
],
"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:\nFergie Duhamel was the only female member of the group Black Eyed Peas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Fergie (singer)",
"Hip hop music"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4025,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
66611
|
The 24 Sextantis can only be viewed from rural skies under good seeing conditions and previously it was an A-type main sequence star before using up the hydrogen at its core.
|
{
"evidence": [
"24 Sextantis_sentence_1",
"24 Sextantis_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 24 Sextantis can only be viewed from rural skies under good seeing conditions and previously it was an A-type main sequence star before using up the hydrogen at its core.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"24 Sextantis_sentence_1",
"24 Sextantis_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4026,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.121",
"value": "?search="
},
{
"operation": "search",
"time": "13.234",
"value": "24 Sextantis"
},
{
"operation": "Now on",
"time": "14.506",
"value": "24 Sextantis"
},
{
"operation": "Highlighting",
"time": "31.873",
"value": "24 Sextantis_sentence_1"
},
{
"operation": "Highlighting",
"time": "38.978",
"value": "24 Sextantis_sentence_3"
},
{
"operation": "finish",
"time": "54.287",
"value": "finish"
}
]
}
|
81220
|
Chill point is original market of the Stop & Shop supermarket chain was on the corner of Northern Artery and McGrath Highway but in 2008, it would be dismantled and sold to Daewoo Shipbuilding & Marine Engineering of South Korea and relocated to Mangalia, Romania.
|
{
"evidence": [
"DSME_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:\nChill point is original market of the Stop & Shop supermarket chain was on the corner of Northern Artery and McGrath Highway but in 2008, it would be dismantled and sold to Daewoo Shipbuilding & Marine Engineering of South Korea and relocated to Mangalia, Romania.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"DSME_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4027,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.721",
"value": "?search="
},
{
"operation": "search",
"time": "32.981",
"value": "Chill point"
},
{
"operation": "Now on",
"time": "34.304",
"value": "undefined"
},
{
"operation": "search",
"time": "67.288",
"value": "Stop & Shop"
},
{
"operation": "Now on",
"time": "68.441",
"value": "Stop & Shop"
},
{
"operation": "search",
"time": "406.548",
"value": "Daewoo Shipbuilding & Marine Engineering"
},
{
"operation": "Now on",
"time": "407.541",
"value": "DSME"
},
{
"operation": "search",
"time": "500.645",
"value": "Contains...Chill point"
},
{
"operation": "Now on",
"time": "501.526",
"value": "undefined"
},
{
"operation": "search",
"time": "1742.601",
"value": "Chill point stop & shop"
},
{
"operation": "Now on",
"time": "1744.534",
"value": "undefined"
},
{
"operation": "search",
"time": "1769.106",
"value": "Northern Artery and McGrath Highway"
},
{
"operation": "Now on",
"time": "1769.823",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "1794.583",
"value": "Massachusetts Route 38"
},
{
"operation": "Now on",
"time": "1795.223",
"value": "Massachusetts Route 38"
},
{
"operation": "back-button-clicked",
"time": "1815.683",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "1816.182",
"value": "undefined"
},
{
"operation": "search",
"time": "1856.498",
"value": "Daewoo Shipbuilding and Marine Engineering"
},
{
"operation": "Now on",
"time": "1857.07",
"value": "DSME"
},
{
"operation": "hyperlink",
"time": "1880.213",
"value": "Daewoo dissolution and corruption scandal"
},
{
"operation": "Now on",
"time": "1880.656",
"value": "Daewoo dissolution and corruption scandal"
},
{
"operation": "hyperlink",
"time": "1888.494",
"value": "Daewoo Shipbuilding & Marine Engineering"
},
{
"operation": "Now on",
"time": "1888.809",
"value": "Daewoo Shipbuilding & Marine Engineering"
},
{
"operation": "Highlighting",
"time": "1926.318",
"value": "DSME_sentence_0"
},
{
"operation": "finish",
"time": "1938.142",
"value": "finish"
}
]
}
|
83429
|
Keetmanshoop was an ancient, lawless village.
|
{
"evidence": [
"Keetmanshoop_sentence_19"
],
"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:\nKeetmanshoop was an ancient, lawless village.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Keetmanshoop_sentence_19"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4028,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.165",
"value": "?search="
},
{
"operation": "search",
"time": "432.357",
"value": "Keetmanshoop"
},
{
"operation": "Now on",
"time": "433.296",
"value": "Keetmanshoop"
},
{
"operation": "Page search",
"time": "443.485",
"value": "ancient"
},
{
"operation": "Page search",
"time": "499.85",
"value": "villae"
},
{
"operation": "Page search",
"time": "502.026",
"value": "village"
},
{
"operation": "Highlighting",
"time": "621.348",
"value": "Keetmanshoop_sentence_19"
},
{
"operation": "finish",
"time": "629.242",
"value": "finish"
}
]
}
|
59421
|
Claude Sheppard was a stipendary magistrate for ten years and a deputy speaker for three years.
|
{
"evidence": [
"Claude Sheppard_sentence_5",
"Claude Sheppard_sentence_7"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nClaude Sheppard was a stipendary magistrate for ten years and a deputy speaker for three years.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Claude Sheppard_sentence_5",
"Claude Sheppard_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4029,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.487",
"value": "?search="
},
{
"operation": "search",
"time": "34.66",
"value": "Claude Sheppard"
},
{
"operation": "Now on",
"time": "35.799",
"value": "Claude Sheppard"
},
{
"operation": "Highlighting",
"time": "71.18",
"value": "Claude Sheppard_sentence_5"
},
{
"operation": "Highlighting",
"time": "76.06",
"value": "Claude Sheppard_sentence_7"
},
{
"operation": "finish",
"time": "203.314",
"value": "finish"
}
]
}
|
exfever_5477
|
Paul Samuel Feig known for directing films including a 2011 American film which emphasizes humor.
|
{
"evidence": [
"Paul Feig",
"Bridesmaids (2011 film)",
"Comedy 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:\nPaul Samuel Feig known for directing films including a 2011 American film which emphasizes humor.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul Feig",
"Bridesmaids (2011 film)",
"Comedy film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4030,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6766
|
Vantage Point is the most prominent film of 2008.?
|
{
"evidence": [
"Vantage Point (film)",
"Thriller (genre)",
"Mood (psychology)"
],
"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:\nVantage Point is the most prominent film of 2008.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Vantage Point (film)",
"Thriller (genre)",
"Mood (psychology)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4031,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
13076
|
Among the A roads in Zone 5 of the Great Britain, Road A55 is the least used because it leads to a dead-end.
|
{
"evidence": [
"A55 road_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:\nAmong the A roads in Zone 5 of the Great Britain, Road A55 is the least used because it leads to a dead-end.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"A55 road_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4032,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.633",
"value": "?search="
},
{
"operation": "search",
"time": "18.957",
"value": "Contains...Road A55"
},
{
"operation": "Now on",
"time": "20.23",
"value": "undefined"
},
{
"operation": "search",
"time": "40.884",
"value": "A roads in Zone 5 of the Great Britain numbering scheme"
},
{
"operation": "Now on",
"time": "42.328",
"value": "A roads in Zone 5 of the Great Britain numbering scheme"
},
{
"operation": "Page search",
"time": "52.937",
"value": "a55"
},
{
"operation": "hyperlink",
"time": "135.194",
"value": "A55"
},
{
"operation": "Now on",
"time": "135.986",
"value": "A55 road (Great Britain)"
},
{
"operation": "search",
"time": "211.663",
"value": "Contains...is the least used because it leads to a dead-end"
},
{
"operation": "Now on",
"time": "216.066",
"value": "undefined"
},
{
"operation": "search",
"time": "230.133",
"value": "A55 road"
},
{
"operation": "Now on",
"time": "230.954",
"value": "A55 road"
},
{
"operation": "Highlighting",
"time": "294.905",
"value": "A55 road_sentence_0"
},
{
"operation": "finish",
"time": "311.264",
"value": "finish"
}
]
}
|
70518
|
At the 122nd Illinois Infantry Regiment continued on duty at Trenton till November 12, 1862, when Colonel George W. Deitzler was ordered to take the right wing of the Regiment and proceed to Humboldt, Tennessee and take command of the post and troops there, while the remainder of the Regiment was left under command of Lieutenant Colonel J. F. Drish.,
|
{
"evidence": [
"122nd Illinois Infantry Regiment_sentence_7"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAt the 122nd Illinois Infantry Regiment continued on duty at Trenton till November 12, 1862, when Colonel George W. Deitzler was ordered to take the right wing of the Regiment and proceed to Humboldt, Tennessee and take command of the post and troops there, while the remainder of the Regiment was left under command of Lieutenant Colonel J. F. Drish.,\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"122nd Illinois Infantry Regiment_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4033,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.824",
"value": "?search="
},
{
"operation": "search",
"time": "24.47",
"value": "\"Colonel George W. Deitzler was ordered\""
},
{
"operation": "Now on",
"time": "25.28",
"value": "undefined"
},
{
"operation": "search",
"time": "35.396",
"value": "\"Colonel George W. Deitzler \""
},
{
"operation": "Now on",
"time": "35.889",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "37.608",
"value": "122nd Illinois Infantry Regiment"
},
{
"operation": "Now on",
"time": "39.007",
"value": "122nd Illinois Infantry Regiment"
},
{
"operation": "Page search",
"time": "49.875",
"value": "Colonel George W. Deitzler"
},
{
"operation": "Page search",
"time": "78.804",
"value": "\"Colonel George W. Deitzler \""
},
{
"operation": "Page search",
"time": "99.141",
"value": "trenton"
},
{
"operation": "Highlighting",
"time": "102.661",
"value": "122nd Illinois Infantry Regiment_sentence_7"
},
{
"operation": "finish",
"time": "135.841",
"value": "finish"
}
]
}
|
exfever_4175
|
Saved is a play that was originally refused a license without severe cuts by the most senior officer of the Royal Household of the United States.
|
{
"evidence": [
"Saved (play)",
"Lord Chamberlain"
],
"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:\nSaved is a play that was originally refused a license without severe cuts by the most senior officer of the Royal Household of the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Saved (play)",
"Lord Chamberlain"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4034,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36582
|
The Hyundai i20 R5 is a rally car built to R5 regulations.
|
{
"evidence": [
"Hyundai i20 R5_sentence_0",
"Hyundai i20 R5_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Hyundai i20 R5 is a rally car built to R5 regulations.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hyundai i20 R5_sentence_0",
"Hyundai i20 R5_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4035,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.968",
"value": "?search="
},
{
"operation": "search",
"time": "13.356",
"value": "Hyundai i20 R5"
},
{
"operation": "Now on",
"time": "14.48",
"value": "Hyundai i20 R5"
},
{
"operation": "Highlighting",
"time": "22.411",
"value": "Hyundai i20 R5_sentence_0"
},
{
"operation": "Highlighting",
"time": "24.048",
"value": "Hyundai i20 R5_sentence_1"
},
{
"operation": "finish",
"time": "44.648",
"value": "finish"
}
]
}
|
45278
|
Tm is a 32.7 kDa protein composed of amino acids, and Tm is a protein homodimer or heterodimer composed of alpha-helical chains.
|
{
"evidence": [
"TPM1_sentence_2",
"TPM1_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:\nTm is a 32.7 kDa protein composed of amino acids, and Tm is a protein homodimer or heterodimer composed of alpha-helical chains.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"TPM1_sentence_2",
"TPM1_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4036,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.141",
"value": "?search="
},
{
"operation": "search",
"time": "12.067",
"value": "Tm"
},
{
"operation": "Now on",
"time": "14.158",
"value": "Tm"
},
{
"operation": "hyperlink",
"time": "21.849",
"value": "2 Science and technology TM_section_1"
},
{
"operation": "hyperlink",
"time": "68.316",
"value": "Tm ligands"
},
{
"operation": "Now on",
"time": "69.246",
"value": "Tm ligands"
},
{
"operation": "back-button-clicked",
"time": "99.201",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "100.072",
"value": "TM#Science and technology TM section 1"
},
{
"operation": "search",
"time": "124.211",
"value": "inpage"
},
{
"operation": "Now on",
"time": "125.12",
"value": "undefined"
},
{
"operation": "search",
"time": "155.761",
"value": "\"3.27 kDa protein\""
},
{
"operation": "Now on",
"time": "157.268",
"value": "undefined"
},
{
"operation": "search",
"time": "171.193",
"value": "TM protein"
},
{
"operation": "Now on",
"time": "172.459",
"value": "Transmembrane protein"
},
{
"operation": "search",
"time": "215.529",
"value": "insource"
},
{
"operation": "Now on",
"time": "217.305",
"value": "undefined"
},
{
"operation": "search",
"time": "247.71",
"value": "insource"
},
{
"operation": "Now on",
"time": "248.879",
"value": "undefined"
},
{
"operation": "search",
"time": "273.5",
"value": "intitle"
},
{
"operation": "Now on",
"time": "275.026",
"value": "undefined"
},
{
"operation": "search",
"time": "295.451",
"value": "intitle"
},
{
"operation": "Now on",
"time": "296.583",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "325.631",
"value": "Transmembrane protein"
},
{
"operation": "Now on",
"time": "326.589",
"value": "Transmembrane protein"
},
{
"operation": "search",
"time": "350.551",
"value": "kda protein"
},
{
"operation": "Now on",
"time": "351.536",
"value": "undefined"
},
{
"operation": "search",
"time": "368.066",
"value": "Proteins"
},
{
"operation": "Now on",
"time": "369.556",
"value": "Protein"
},
{
"operation": "hyperlink",
"time": "376.891",
"value": "3 Biochemistry Protein_section_2"
},
{
"operation": "search",
"time": "403.732",
"value": "Tm"
},
{
"operation": "Now on",
"time": "404.626",
"value": "Tm"
},
{
"operation": "hyperlink",
"time": "410.231",
"value": "2.1 Biology and medicine TM_section_2"
},
{
"operation": "hyperlink",
"time": "421.282",
"value": "Transverse myelitis"
},
{
"operation": "Now on",
"time": "422.197",
"value": "Transverse myelitis"
},
{
"operation": "back-button-clicked",
"time": "446.035",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "446.082",
"value": "TM#Biology and medicine TM section 2"
},
{
"operation": "search",
"time": "505.833",
"value": "tmp1"
},
{
"operation": "Now on",
"time": "506.793",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "511.73",
"value": "tmp"
},
{
"operation": "Now on",
"time": "512.728",
"value": "undefined"
},
{
"operation": "search",
"time": "526.353",
"value": "tpm1"
},
{
"operation": "Now on",
"time": "527.921",
"value": "TPM1"
},
{
"operation": "Highlighting",
"time": "538.323",
"value": "TPM1_sentence_2"
},
{
"operation": "Highlighting",
"time": "548.639",
"value": "TPM1_sentence_3"
},
{
"operation": "finish",
"time": "588.231",
"value": "finish"
}
]
}
|
exfever_1856
|
As You Like It is a 2006 film starring an American actor who received an Academy Award and three Tony Awards.
|
{
"evidence": [
"As You Like It (2006 film)",
"Kevin Kline"
],
"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 film starring an American actor who received an Academy Award and three Tony Awards.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"As You Like It (2006 film)",
"Kevin Kline"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4037,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
40150
|
The Colour of My Love was released in Canada on 9 November 1993 by Columbia Records (owned by Sony Music Entertainment), and contained 15 tracks.
|
{
"evidence": [
"The Colour of My Love_sentence_1",
"Columbia Records_sentence_3"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Colour of My Love was released in Canada on 9 November 1993 by Columbia Records (owned by Sony Music Entertainment), and contained 15 tracks.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"The Colour of My Love_sentence_1",
"Columbia Records_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4038,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.713",
"value": "?search="
},
{
"operation": "search",
"time": "14.649",
"value": "The Color of my Love"
},
{
"operation": "Now on",
"time": "15.416",
"value": "The Colour of My Love"
},
{
"operation": "Highlighting",
"time": "24.095",
"value": "The Colour of My Love_sentence_1"
},
{
"operation": "Page search",
"time": "47.349",
"value": "Canada"
},
{
"operation": "hyperlink",
"time": "100.92",
"value": "Columbia Records"
},
{
"operation": "Now on",
"time": "101.705",
"value": "Columbia Records"
},
{
"operation": "Highlighting",
"time": "107.771",
"value": "Columbia Records_sentence_3"
},
{
"operation": "back-button-clicked",
"time": "147.79",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "148.359",
"value": "The Colour of My Love"
},
{
"operation": "finish",
"time": "155.273",
"value": "finish"
}
]
}
|
exfever_556
|
Amy Winehouse was Blake Shelton's mentor.
|
{
"evidence": [
"Frank (Amy Winehouse album)",
"Island Records"
],
"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:\nAmy Winehouse was Blake Shelton's mentor.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Frank (Amy Winehouse album)",
"Island Records"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4039,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
18810
|
According to the Yongbieocheonga, Yi Seong-gye was born in 1336 and would later become one of Korea's greatest army generals and kings.
|
{
"evidence": [
"Yongbieocheonga_sentence_9"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAccording to the Yongbieocheonga, Yi Seong-gye was born in 1336 and would later become one of Korea's greatest army generals and kings.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Yongbieocheonga_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4040,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.056",
"value": "?search="
},
{
"operation": "search",
"time": "81.297",
"value": "Contains...Yi Seong-gye"
},
{
"operation": "Now on",
"time": "82.54",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "99.407",
"value": "People surrounding Yi Seong-gye Jeong Do-jeon (TV series)_section_5"
},
{
"operation": "Now on",
"time": "100.252",
"value": "Jeong Do-jeon (TV series)#People surrounding Yi Seong-gye Jeong Do-jeon .28TV series.29 section 5"
},
{
"operation": "Page search",
"time": "117.958",
"value": "yi seong"
},
{
"operation": "Highlighting",
"time": "129.072",
"value": "Jeong Do-jeon (TV series)_sentence_7"
},
{
"operation": "Highlighting",
"time": "129.074",
"value": "Jeong Do-jeon (TV series)_title"
},
{
"operation": "Highlighting",
"time": "129.074",
"value": "Jeong Do-jeon (TV series)_section_0"
},
{
"operation": "search",
"time": "143.326",
"value": "King Taejo"
},
{
"operation": "Now on",
"time": "144.037",
"value": "Taejo"
},
{
"operation": "hyperlink",
"time": "163.959",
"value": "King Taejo"
},
{
"operation": "Now on",
"time": "164.822",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "166.149",
"value": "Taejo"
},
{
"operation": "Now on",
"time": "166.498",
"value": "Taejo"
},
{
"operation": "hyperlink",
"time": "168.316",
"value": "Taejo of Goguryeo"
},
{
"operation": "Now on",
"time": "168.886",
"value": "Taejodae of Goguryeo"
},
{
"operation": "back-button-clicked",
"time": "196.55",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "196.663",
"value": "Taejo"
},
{
"operation": "hyperlink",
"time": "198.898",
"value": "Taejo of Goryeo"
},
{
"operation": "Now on",
"time": "199.786",
"value": "Taejo of Goryeo"
},
{
"operation": "search",
"time": "274.055",
"value": "Yongbieocheonga"
},
{
"operation": "Now on",
"time": "275.023",
"value": "Yongbieocheonga"
},
{
"operation": "Page search",
"time": "290.17",
"value": "yi seong"
},
{
"operation": "Highlighting",
"time": "302.214",
"value": "Yongbieocheonga_sentence_9"
},
{
"operation": "finish",
"time": "321.838",
"value": "finish"
}
]
}
|
17147
|
In 2016, Rochester Rhino, which had eleven out of the 33 players from the United States and two players from Canada, were eliminated in the conference semi-finals by the New York Red Bulls II, an American professional soccer team based in Harrison, New Jersey.
|
{
"evidence": [
"2016 U.S. Open Cup_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn 2016, Rochester Rhino, which had eleven out of the 33 players from the United States and two players from Canada, were eliminated in the conference semi-finals by the New York Red Bulls II, an American professional soccer team based in Harrison, New Jersey.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"2016 U.S. Open Cup_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4041,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.469",
"value": "?search="
},
{
"operation": "search",
"time": "14.172",
"value": "Rochester Rhino"
},
{
"operation": "Now on",
"time": "15.252",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "25.692",
"value": "rochester rhinos"
},
{
"operation": "Now on",
"time": "26.491",
"value": "undefined"
},
{
"operation": "search",
"time": "76.913",
"value": "Rochester Rhinos 2016"
},
{
"operation": "Now on",
"time": "78.323",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "84.344",
"value": "2016 U.S. Open Cup"
},
{
"operation": "Now on",
"time": "85.831",
"value": "2016 U.S. Open Cup"
},
{
"operation": "Page search",
"time": "103.09",
"value": "rhinos"
},
{
"operation": "Highlighting",
"time": "142.649",
"value": "2016 U.S. Open Cup_sentence_2"
},
{
"operation": "finish",
"time": "167.323",
"value": "finish"
}
]
}
|
exfever_3549
|
The American television series, aired on CBS, is based on the DC Comics superhero with superhuman speed.
|
{
"evidence": [
"The Flash (1990 TV series)",
"Flash (Barry Allen)"
],
"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 American television series, aired on CBS, is based on the DC Comics superhero with superhuman speed.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Flash (1990 TV series)",
"Flash (Barry Allen)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4042,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6875
|
Gone with the Wind is a novel by writer Margaret Mitchell that was set during the war in the United States between the Union and the Confederacy.
|
{
"evidence": [
"Gone with the Wind (novel)",
"American Civil War"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nGone with the Wind is a novel by writer Margaret Mitchell that was set during the war in the United States between the Union and the Confederacy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gone with the Wind (novel)",
"American Civil War"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4043,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
25980
|
According to mythology, the first chariot race took place in the Hippodrome of Olympia between King Tantalus and King Oenomaus of Pisa, where the horses are declared winners and enjoy their whole lives in the castle.
|
{
"evidence": [
"Hippodrome of Olympia_sentence_5",
"Hippodrome of Olympia_sentence_7"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAccording to mythology, the first chariot race took place in the Hippodrome of Olympia between King Tantalus and King Oenomaus of Pisa, where the horses are declared winners and enjoy their whole lives in the castle.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hippodrome of Olympia_sentence_5",
"Hippodrome of Olympia_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4044,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.089",
"value": "?search="
},
{
"operation": "search",
"time": "14.403",
"value": "Chariot race"
},
{
"operation": "Now on",
"time": "15.278",
"value": "Chariot racing"
},
{
"operation": "hyperlink",
"time": "74.408",
"value": "Oenomaus"
},
{
"operation": "Now on",
"time": "75.151",
"value": "Oenomaus"
},
{
"operation": "Highlighting",
"time": "116.975",
"value": "Oenomaus_sentence_10"
},
{
"operation": "Highlighting",
"time": "116.977",
"value": "Oenomaus_title"
},
{
"operation": "Highlighting",
"time": "116.979",
"value": "Oenomaus_section_1"
},
{
"operation": "Highlighting",
"time": "116.98",
"value": "Oenomaus_section_2"
},
{
"operation": "Highlighting",
"time": "124.534",
"value": "Oenomaus_sentence_13"
},
{
"operation": "Highlighting",
"time": "124.536",
"value": "Oenomaus_section_3"
},
{
"operation": "hyperlink",
"time": "153.541",
"value": "Tantalus"
},
{
"operation": "Now on",
"time": "154.507",
"value": "Tantalus"
},
{
"operation": "search",
"time": "174.912",
"value": "hippodrome"
},
{
"operation": "Now on",
"time": "175.633",
"value": "undefined"
},
{
"operation": "search",
"time": "184.074",
"value": "Hippodrome of Olympia"
},
{
"operation": "Now on",
"time": "184.941",
"value": "Hippodrome of Olympia"
},
{
"operation": "Highlighting",
"time": "196.478",
"value": "Hippodrome of Olympia_sentence_5"
},
{
"operation": "Highlighting",
"time": "209.567",
"value": "Hippodrome of Olympia_sentence_7"
},
{
"operation": "finish",
"time": "219.23",
"value": "finish"
}
]
}
|
37551
|
A part of the Realism movement and an avid realist painter, Lhermitte's The Gleaners (1887, Philadelphia Museum of Art - an art museum originally chartered in 1876 for the Centennial Exposition in Philadelphia) depicts the working class poverty in Italy where he took very obvious inspiration from Millet, a painting of the same name; when comparing his work to Millet's, even the poses of the women are very similar if not the same.
|
{
"evidence": [
"Léon Augustin Lhermitte_sentence_9"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nA part of the Realism movement and an avid realist painter, Lhermitte's The Gleaners (1887, Philadelphia Museum of Art - an art museum originally chartered in 1876 for the Centennial Exposition in Philadelphia) depicts the working class poverty in Italy where he took very obvious inspiration from Millet, a painting of the same name; when comparing his work to Millet's, even the poses of the women are very similar if not the same.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Léon Augustin Lhermitte_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4045,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.063",
"value": "?search="
},
{
"operation": "search",
"time": "16.304",
"value": "Contains...Lhermitte's The Gleaners"
},
{
"operation": "Now on",
"time": "16.878",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "18.395",
"value": "Léon Augustin Lhermitte"
},
{
"operation": "Now on",
"time": "19.098",
"value": "Léon Augustin Lhermitte"
},
{
"operation": "Page search",
"time": "40.685",
"value": "Gleaners"
},
{
"operation": "Page search",
"time": "41.043",
"value": "Gleaners"
},
{
"operation": "Page search",
"time": "41.314",
"value": "Gleaners ="
},
{
"operation": "Page search",
"time": "42.1",
"value": "Gleaners ="
},
{
"operation": "Page search",
"time": "42.539",
"value": "Gleaners"
},
{
"operation": "Page search",
"time": "42.766",
"value": "Gleaners"
},
{
"operation": "Page search",
"time": "42.969",
"value": "Gleaners"
},
{
"operation": "Highlighting",
"time": "46.411",
"value": "Léon Augustin Lhermitte_sentence_8"
},
{
"operation": "Page search",
"time": "55.857",
"value": "realism"
},
{
"operation": "Highlighting",
"time": "57.715",
"value": "Léon Augustin Lhermitte_sentence_9"
},
{
"operation": "hyperlink",
"time": "72",
"value": "Philadelphia Museum of Art"
},
{
"operation": "Now on",
"time": "73.007",
"value": "Philadelphia Museum of Art"
},
{
"operation": "Highlighting",
"time": "94.811",
"value": "Philadelphia Museum of Art_sentence_0"
},
{
"operation": "Highlighting",
"time": "94.811",
"value": "Philadelphia Museum of Art_title"
},
{
"operation": "back-button-clicked",
"time": "100.451",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "100.532",
"value": "Léon Augustin Lhermitte"
},
{
"operation": "Highlighting deleted",
"time": "110.39",
"value": "Léon Augustin Lhermitte_sentence_8"
},
{
"operation": "Highlighting",
"time": "113.37",
"value": "Léon Augustin Lhermitte_sentence_9"
},
{
"operation": "finish",
"time": "162.008",
"value": "finish"
}
]
}
|
40512
|
With French neoclassic, Gothic, and Renaissance elements, the Florence Public Library was built with Beaux Arts design influences.
|
{
"evidence": [
"Florence Public Library_sentence_1",
"Beaux-Arts architecture_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:\nWith French neoclassic, Gothic, and Renaissance elements, the Florence Public Library was built with Beaux Arts design influences.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Florence Public Library_sentence_1",
"Beaux-Arts architecture_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4046,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.05",
"value": "?search="
},
{
"operation": "search",
"time": "16.086",
"value": "Florence Public Library"
},
{
"operation": "Now on",
"time": "16.943",
"value": "Florence Public Library"
},
{
"operation": "Page search",
"time": "38.363",
"value": "neo"
},
{
"operation": "Highlighting",
"time": "59.954",
"value": "Florence Public Library_sentence_1"
},
{
"operation": "Page search",
"time": "78.93",
"value": "gothic"
},
{
"operation": "Page search",
"time": "90.971",
"value": "french"
},
{
"operation": "hyperlink",
"time": "97.795",
"value": "Beaux Arts"
},
{
"operation": "Now on",
"time": "98.589",
"value": "Beaux-Arts architecture"
},
{
"operation": "Highlighting",
"time": "104.08",
"value": "Beaux-Arts architecture_sentence_0"
},
{
"operation": "Highlighting",
"time": "109.347",
"value": "Beaux-Arts architecture_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "111.751",
"value": "Beaux-Arts architecture_sentence_0"
},
{
"operation": "finish",
"time": "192.986",
"value": "finish"
}
]
}
|
exfever_7843
|
The House of Lusignan was a royal house of French, which ruled several principalities in the Eastern Mediterranean region of Western Asia.
|
{
"evidence": [
"House of Lusignan",
"Levant"
],
"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 House of Lusignan was a royal house of French, which ruled several principalities in the Eastern Mediterranean region of Western Asia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"House of Lusignan",
"Levant"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4047,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1586
|
Erich von Manstein was a German Field Marshal of the unified armed forces of Nazi Germany from 1935 to 1945.
|
{
"evidence": [
"Erich von Manstein",
"Wehrmacht"
],
"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:\nErich von Manstein was a German Field Marshal of the unified armed forces of Nazi Germany from 1935 to 1945.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Erich von Manstein",
"Wehrmacht"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4048,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
68636
|
The Bradykinin receptor B2 is a T protein-coupled receptor (also known as 7TM receptors), probably coupled to Tq, which stimulates phospholipase C to increase intracellular free calcium, and to Ti, which inhibits adenylate cyclase.
|
{
"evidence": [
"Bradykinin receptor B2_sentence_0",
"Bradykinin receptor B2_sentence_1",
"Bradykinin receptor B2_sentence_2"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Bradykinin receptor B2 is a T protein-coupled receptor (also known as 7TM receptors), probably coupled to Tq, which stimulates phospholipase C to increase intracellular free calcium, and to Ti, which inhibits adenylate cyclase.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Bradykinin receptor B2_sentence_0",
"Bradykinin receptor B2_sentence_1",
"Bradykinin receptor B2_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4049,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.348",
"value": "?search="
},
{
"operation": "search",
"time": "13.295",
"value": "Bradykinin receptor B2"
},
{
"operation": "Now on",
"time": "13.829",
"value": "Bradykinin receptor B2"
},
{
"operation": "Highlighting",
"time": "95.989",
"value": "Bradykinin receptor B2_sentence_0"
},
{
"operation": "Highlighting",
"time": "110.413",
"value": "Bradykinin receptor B2_sentence_1"
},
{
"operation": "Highlighting",
"time": "122.653",
"value": "Bradykinin receptor B2_sentence_2"
},
{
"operation": "hyperlink",
"time": "131.437",
"value": "G-protein coupled receptor"
},
{
"operation": "Now on",
"time": "140.078",
"value": "G protein-coupled receptor"
},
{
"operation": "finish",
"time": "182.62",
"value": "finish"
}
]
}
|
exfever_4395
|
Anna Cooke Kendrick is an American actress and singer who received a nomination for The Antoinette Perry Award for Excellence in Broadway Theatre, which recognizes excellence in live Broadway theatre.
|
{
"evidence": [
"Anna Kendrick",
"Tony Award"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAnna Cooke Kendrick is an American actress and singer who received a nomination for The Antoinette Perry Award for Excellence in Broadway Theatre, which recognizes excellence in live Broadway theatre.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Anna Kendrick",
"Tony Award"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4050,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7330
|
Music is the art of arranging sounds in time through the elements, including the process by which individual sounds are joined together or composed into whole units or compositions.
|
{
"evidence": [
"Music",
"Harmony"
],
"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:\nMusic is the art of arranging sounds in time through the elements, including the process by which individual sounds are joined together or composed into whole units or compositions.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Music",
"Harmony"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4051,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
78444
|
The first match at Estadi Palamós Costa Brava was a draw against FC Barcelona and Levante UD.
|
{
"evidence": [
"Estadi Palamós Costa Brava_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 first match at Estadi Palamós Costa Brava was a draw against FC Barcelona and Levante UD.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Estadi Palamós Costa Brava_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4052,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.831",
"value": "?search="
},
{
"operation": "search",
"time": "19.766",
"value": "Estadi Palamós Costa Brava"
},
{
"operation": "Now on",
"time": "20.703",
"value": "Estadi Palamós Costa Brava"
},
{
"operation": "Highlighting",
"time": "29.976",
"value": "Estadi Palamós Costa Brava_sentence_3"
},
{
"operation": "hyperlink",
"time": "55.315",
"value": "friendly"
},
{
"operation": "Now on",
"time": "56.101",
"value": "Exhibition game"
},
{
"operation": "back-button-clicked",
"time": "62.976",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "63.092",
"value": "Estadi Palamós Costa Brava"
},
{
"operation": "finish",
"time": "76.733",
"value": "finish"
}
]
}
|
10236
|
A lawnmower may require SAE 0W-20 multigrade engine oil, while cars may only require SAE 30 monograde engine oils.
|
{
"evidence": [
"SAE J300_sentence_5"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nA lawnmower may require SAE 0W-20 multigrade engine oil, while cars may only require SAE 30 monograde engine oils.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"SAE J300_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4053,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.947",
"value": "?search="
},
{
"operation": "search",
"time": "22.557",
"value": "Contains...SAE 0W-20 multigrade engine oil"
},
{
"operation": "Now on",
"time": "23.435",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "45.178",
"value": "Viscosity grades SAE J300_section_0"
},
{
"operation": "Now on",
"time": "45.708",
"value": "SAE J300#Viscosity grades SAE J300 section 0"
},
{
"operation": "Highlighting",
"time": "54.252",
"value": "SAE J300_sentence_5"
},
{
"operation": "Highlighting",
"time": "74.549",
"value": "SAE J300_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "116.551",
"value": "SAE J300_sentence_6"
},
{
"operation": "finish",
"time": "133.293",
"value": "finish"
}
]
}
|
exfever_885
|
Judith Therese Evans is an Japanese actress who expanded into other genres, with roles in such films as an American period thriller film written, produced, and directed by M. Night Shyamalan.
|
{
"evidence": [
"Judy Greer",
"The Village (2004 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJudith Therese Evans is an Japanese actress who expanded into other genres, with roles in such films as an American period thriller film written, produced, and directed by M. Night Shyamalan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Judy Greer",
"The Village (2004 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4054,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2391
|
Diane Ruggiero-Wright is an American television writer and producer whose credits include an American supernatural procedural crime drama television series that is a loose adaptation of the comic book series of the same name created by an American science fiction author and publisher based in Portland, Oregon.
|
{
"evidence": [
"Diane Ruggiero",
"IZombie (TV series)",
"Chris Roberson (author)"
],
"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:\nDiane Ruggiero-Wright is an American television writer and producer whose credits include an American supernatural procedural crime drama television series that is a loose adaptation of the comic book series of the same name created by an American science fiction author and publisher based in Portland, Oregon.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Diane Ruggiero",
"IZombie (TV series)",
"Chris Roberson (author)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4055,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2029
|
The Avengers is a 2012 American superhero film by a publisher of American comic books and related media.
|
{
"evidence": [
"The Avengers (2012 film)",
"Marvel Comics"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Avengers is a 2012 American superhero film by a publisher of American comic books and related media.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Avengers (2012 film)",
"Marvel Comics"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4056,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4931
|
The Disaster Artist was the third film written by Scott Neustadter.
|
{
"evidence": [
"The Disaster Artist (film)",
"James Franco",
"Pineapple Express (film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Disaster Artist was the third film written by Scott Neustadter.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Disaster Artist (film)",
"James Franco",
"Pineapple Express (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4057,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8590
|
Billy Bob Thornton is an American actor, writer, and director of Sling Blade, who starred in a 1998 American science fiction disaster film produced and directed by Jerry Bruckheimer
|
{
"evidence": [
"Billy Bob Thornton",
"Armageddon (1998 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBilly Bob Thornton is an American actor, writer, and director of Sling Blade, who starred in a 1998 American science fiction disaster film produced and directed by Jerry Bruckheimer\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Billy Bob Thornton",
"Armageddon (1998 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4058,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4922
|
Quran is the equivalent of the Bible.
|
{
"evidence": [
"Islam",
"Abrahamic religions"
],
"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:\nQuran is the equivalent of the Bible.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Islam",
"Abrahamic religions"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4059,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_567
|
Kumar Sangakkara is perhaps the best keeper-batsman in the current cricketing world.
|
{
"evidence": [
"MS Dhoni",
"2007 ICC World Twenty20"
],
"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:\nKumar Sangakkara is perhaps the best keeper-batsman in the current cricketing world.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"MS Dhoni",
"2007 ICC World Twenty20"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4060,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1964
|
Kendrick Lamar is an American rapper, songwriter, and record producer who is also known as a member of the supergroup Black Hippy, alongside Ab-Soul, Schoolboy Q, and a rapper from Watts, Texas.
|
{
"evidence": [
"Kendrick Lamar",
"Jay Rock"
],
"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:\nKendrick Lamar is an American rapper, songwriter, and record producer who is also known as a member of the supergroup Black Hippy, alongside Ab-Soul, Schoolboy Q, and a rapper from Watts, Texas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kendrick Lamar",
"Jay Rock"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4061,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_822
|
Winona Ryder is an American actress who rose to prominence with major roles in a 1990 American fantasy film directed by Bruce Lee.
|
{
"evidence": [
"Winona Ryder",
"Edward Scissorhands"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWinona Ryder is an American actress who rose to prominence with major roles in a 1990 American fantasy film directed by Bruce Lee.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Winona Ryder",
"Edward Scissorhands"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4062,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1981
|
Alessia Cara is the youngest Canadian singer who signed with EP Entertainment.
|
{
"evidence": [
"Alessia Cara",
"Def Jam Recordings"
],
"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:\nAlessia Cara is the youngest Canadian singer who signed with EP Entertainment.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alessia Cara",
"Def Jam Recordings"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4063,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8085
|
The Golden Girls is an American sitcom aired on CBS, starring Bea Arthur, Betty White, Estelle Getty, and an American actress and comedian best known for her role as Vivian Harmon on Maude.
|
{
"evidence": [
"The Golden Girls",
"Rue McClanahan"
],
"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 Golden Girls is an American sitcom aired on CBS, starring Bea Arthur, Betty White, Estelle Getty, and an American actress and comedian best known for her role as Vivian Harmon on Maude.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Golden Girls",
"Rue McClanahan"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4064,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6804
|
Fergie is a singer who sang a song that serviced mainstream radios through A&M Records, together with a vanity label founded by a music artist of Black Eyed Peas.
|
{
"evidence": [
"Fergie (singer)",
"Glamorous (Fergie song)",
"Will.i.am Music Group"
],
"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:\nFergie is a singer who sang a song that serviced mainstream radios through A&M Records, together with a vanity label founded by a music artist of Black Eyed Peas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Fergie (singer)",
"Glamorous (Fergie song)",
"Will.i.am Music Group"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4065,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
94377
|
Giorgio Mainerio sought to work as a chaplain at the church of Santa Maria Annunziata in Udine, a city in north-eastern Italy.
|
{
"evidence": [
"Giorgio Mainerio_sentence_4",
"Udine_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:\nGiorgio Mainerio sought to work as a chaplain at the church of Santa Maria Annunziata in Udine, a city in north-eastern Italy.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Giorgio Mainerio_sentence_4",
"Udine_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4066,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.492",
"value": "?search="
},
{
"operation": "search",
"time": "10.985",
"value": "Giorgio Mainerio"
},
{
"operation": "Now on",
"time": "11.765",
"value": "Giorgio Mainerio"
},
{
"operation": "Page search",
"time": "25.254",
"value": "chaplain"
},
{
"operation": "Highlighting",
"time": "32.204",
"value": "Giorgio Mainerio_sentence_4"
},
{
"operation": "Page search",
"time": "53.637",
"value": "north-eastern"
},
{
"operation": "Highlighting",
"time": "73.038",
"value": "Giorgio Mainerio_sentence_4"
},
{
"operation": "hyperlink",
"time": "102.506",
"value": "Udine"
},
{
"operation": "Now on",
"time": "103.465",
"value": "Udine"
},
{
"operation": "Highlighting",
"time": "110.597",
"value": "Udine_sentence_2"
},
{
"operation": "finish",
"time": "148.859",
"value": "finish"
}
]
}
|
28991
|
Ujjal Dosanjh won the vote becoming Canada's first Indo-Canadian provincial leader.
|
{
"evidence": [
"Ujjal Dosanjh_sentence_7",
"Ujjal Dosanjh_sentence_8"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nUjjal Dosanjh won the vote becoming Canada's first Indo-Canadian provincial leader.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ujjal Dosanjh_sentence_7",
"Ujjal Dosanjh_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4067,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.76",
"value": "?search="
},
{
"operation": "search",
"time": "10.659",
"value": "Ujjal Dosanjh"
},
{
"operation": "Now on",
"time": "11.68",
"value": "Ujjal Dosanjh"
},
{
"operation": "Highlighting",
"time": "23.025",
"value": "Ujjal Dosanjh_sentence_8"
},
{
"operation": "Highlighting",
"time": "28.145",
"value": "Ujjal Dosanjh_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "28.857",
"value": "Ujjal Dosanjh_sentence_8"
},
{
"operation": "Highlighting",
"time": "29.193",
"value": "Ujjal Dosanjh_sentence_8"
},
{
"operation": "finish",
"time": "42.673",
"value": "finish"
}
]
}
|
55039
|
Souha Hatono, a Japanese physician specializing in surgery, was known for treating both sides at the Satsuma Rebellion equally, in accord with the spirit of International Red Cross.
|
{
"evidence": [
"Souha Hatono_sentence_0",
"Souha Hatono_sentence_1",
"Souha Hatono_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSouha Hatono, a Japanese physician specializing in surgery, was known for treating both sides at the Satsuma Rebellion equally, in accord with the spirit of International Red Cross.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Souha Hatono_sentence_0",
"Souha Hatono_sentence_1",
"Souha Hatono_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4068,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.492",
"value": "?search="
},
{
"operation": "search",
"time": "17.941",
"value": "Souha Hatono"
},
{
"operation": "Now on",
"time": "18.665",
"value": "Souha Hatono"
},
{
"operation": "Highlighting",
"time": "20.684",
"value": "Souha Hatono_sentence_0"
},
{
"operation": "Highlighting",
"time": "32.146",
"value": "Souha Hatono_sentence_1"
},
{
"operation": "Highlighting",
"time": "44.692",
"value": "Souha Hatono_sentence_3"
},
{
"operation": "finish",
"time": "53.137",
"value": "finish"
}
]
}
|
73730
|
The Norwegian Kven Organization, which has local branches in Tromsø, which is located in Northern Norway, and Østlandet, which is found in Eastern Norway, aims to improve the media coverage on Kven issues and to establish a Kven kindergarten.
|
{
"evidence": [
"Norwegian Kven Organization_sentence_1",
"Tromsø_sentence_4",
"Eastern Norway_sentence_0",
"Norwegian Kven Organization_sentence_2",
"Norwegian Kven Organization_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Norwegian Kven Organization, which has local branches in Tromsø, which is located in Northern Norway, and Østlandet, which is found in Eastern Norway, aims to improve the media coverage on Kven issues and to establish a Kven kindergarten.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Norwegian Kven Organization_sentence_1",
"Tromsø_sentence_4",
"Eastern Norway_sentence_0",
"Norwegian Kven Organization_sentence_2",
"Norwegian Kven Organization_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4069,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.965",
"value": "?search="
},
{
"operation": "search",
"time": "18.883",
"value": "Norwegian Kven Organization"
},
{
"operation": "Now on",
"time": "20.172",
"value": "Norwegian Kven Organization"
},
{
"operation": "Highlighting",
"time": "26.774",
"value": "Norwegian Kven Organization_sentence_1"
},
{
"operation": "hyperlink",
"time": "32.401",
"value": "Tromsø"
},
{
"operation": "Now on",
"time": "37.481",
"value": "Tromsø"
},
{
"operation": "Highlighting",
"time": "40.429",
"value": "Tromsø_sentence_4"
},
{
"operation": "back-button-clicked",
"time": "45.041",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "45.164",
"value": "Norwegian Kven Organization"
},
{
"operation": "hyperlink",
"time": "47.308",
"value": "Østlandet"
},
{
"operation": "Now on",
"time": "48.836",
"value": "Eastern Norway"
},
{
"operation": "Highlighting",
"time": "56.681",
"value": "Eastern Norway_sentence_0"
},
{
"operation": "Highlighting",
"time": "59.49",
"value": "Eastern Norway_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "67.242",
"value": "Eastern Norway_sentence_2"
},
{
"operation": "back-button-clicked",
"time": "69.575",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "69.665",
"value": "Norwegian Kven Organization"
},
{
"operation": "Highlighting",
"time": "87.086",
"value": "Norwegian Kven Organization_sentence_2"
},
{
"operation": "Highlighting",
"time": "88.29",
"value": "Norwegian Kven Organization_sentence_3"
},
{
"operation": "finish",
"time": "112.782",
"value": "finish"
}
]
}
|
exfever_3269
|
John Williams was an American composer who won numerous awards, including five for artistic merit in the film industry, for which the winners are given a copy of a golden statuette rendered in a style of visual art that first appeared in France.
|
{
"evidence": [
"John Williams",
"Academy Awards",
"Art Deco"
],
"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:\nJohn Williams was an American composer who won numerous awards, including five for artistic merit in the film industry, for which the winners are given a copy of a golden statuette rendered in a style of visual art that first appeared in France.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Williams",
"Academy Awards",
"Art Deco"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4070,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
41261
|
Bird Girl's parent's want to marry her off, but she escapes as she does not want to get pregnant in Bird Girl and the Man Who Followed the Sun, a 1996 novel by Velma Wallis.
|
{
"evidence": [
"Bird Girl and the Man Who Followed the Sun_sentence_0",
"Bird Girl and the Man Who Followed the Sun_sentence_17",
"Bird Girl and the Man Who Followed the Sun_sentence_18"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBird Girl's parent's want to marry her off, but she escapes as she does not want to get pregnant in Bird Girl and the Man Who Followed the Sun, a 1996 novel by Velma Wallis.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Bird Girl and the Man Who Followed the Sun_sentence_0",
"Bird Girl and the Man Who Followed the Sun_sentence_17",
"Bird Girl and the Man Who Followed the Sun_sentence_18"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4071,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.734",
"value": "?search="
},
{
"operation": "search",
"time": "22.148",
"value": "Bird Girl and the Man Who Followed the Sun"
},
{
"operation": "Now on",
"time": "22.874",
"value": "Bird Girl and the Man Who Followed the Sun"
},
{
"operation": "Highlighting",
"time": "29.267",
"value": "Bird Girl and the Man Who Followed the Sun_sentence_0"
},
{
"operation": "Highlighting",
"time": "39.864",
"value": "Bird Girl and the Man Who Followed the Sun_sentence_17"
},
{
"operation": "Highlighting",
"time": "49.446",
"value": "Bird Girl and the Man Who Followed the Sun_sentence_18"
},
{
"operation": "finish",
"time": "107.329",
"value": "finish"
}
]
}
|
exfever_7916
|
Ryuk s a fictional character in the manga series Death Note, an adaptation of hand-drawn and computer-generated animation originating from Japan.
|
{
"evidence": [
"Ryuk (Death Note)",
"Anime"
],
"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:\nRyuk s a fictional character in the manga series Death Note, an adaptation of hand-drawn and computer-generated animation originating from Japan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ryuk (Death Note)",
"Anime"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4072,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2244
|
Aeneas is the son of the Greek goddess Aphrodite and the Trojan prince Anchises, the first cousin of the last king of Troy during the Trojan War.
|
{
"evidence": [
"Aeneas",
"Priam"
],
"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:\nAeneas is the son of the Greek goddess Aphrodite and the Trojan prince Anchises, the first cousin of the last king of Troy during the Trojan War.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Aeneas",
"Priam"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4073,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4933
|
Lewis Friedman is an American speechwriter who has credits in an American satirical comedy film directed by David Zucker.
|
{
"evidence": [
"Lewis Friedman",
"An American Carol"
],
"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:\nLewis Friedman is an American speechwriter who has credits in an American satirical comedy film directed by David Zucker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Lewis Friedman",
"An American Carol"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4074,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21280
|
A genus of flowering plants belong to a distinct kingdom, order and tribe.
|
{
"evidence": [
"Flowering plant_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nA genus of flowering plants belong to a distinct kingdom, order and tribe.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Flowering plant_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4075,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.571",
"value": "?search="
},
{
"operation": "search",
"time": "35.762",
"value": "a genus of flowering plants belong to a distinct kingdom, order and tribe"
},
{
"operation": "Now on",
"time": "38.09",
"value": "undefined"
},
{
"operation": "search",
"time": "52.929",
"value": "Contains...\"distinct kingdom, order, and tribe\""
},
{
"operation": "Now on",
"time": "53.484",
"value": "undefined"
},
{
"operation": "search",
"time": "62.538",
"value": "Contains...\"a genus of flowering plants\""
},
{
"operation": "Now on",
"time": "64.635",
"value": "undefined"
},
{
"operation": "search",
"time": "93.702",
"value": "Contains...\"a genus of flowering plants belongs\""
},
{
"operation": "Now on",
"time": "94.088",
"value": "undefined"
},
{
"operation": "search",
"time": "105.449",
"value": "Contains...\"distinct kingdom\""
},
{
"operation": "Now on",
"time": "106.043",
"value": "undefined"
},
{
"operation": "search",
"time": "114.269",
"value": "Contains... a genus of flowering plants"
},
{
"operation": "Now on",
"time": "115.289",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "116.095",
"value": "Aechmea"
},
{
"operation": "Now on",
"time": "116.902",
"value": "Aechmea"
},
{
"operation": "Page search",
"time": "135.288",
"value": "tribe"
},
{
"operation": "search",
"time": "152.699",
"value": "Contains... a genus of flowering plants"
},
{
"operation": "Now on",
"time": "153.448",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "156.085",
"value": "Schefflera"
},
{
"operation": "Now on",
"time": "156.911",
"value": "Schefflera"
},
{
"operation": "search",
"time": "193.654",
"value": "Contains... a genus of flowering plants"
},
{
"operation": "Now on",
"time": "194.884",
"value": "undefined"
},
{
"operation": "search",
"time": "223.52",
"value": "Flowering plants"
},
{
"operation": "Now on",
"time": "224.958",
"value": "Flowering plants"
},
{
"operation": "Highlighting",
"time": "237.22",
"value": "Flowering plant_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "237.876",
"value": "Flowering plant_sentence_1"
},
{
"operation": "Highlighting",
"time": "239.152",
"value": "Flowering plant_sentence_2"
},
{
"operation": "Page search",
"time": "255.352",
"value": "kingdom order"
},
{
"operation": "Page search",
"time": "264.782",
"value": "order"
},
{
"operation": "Highlighting deleted",
"time": "325.655",
"value": "Flowering plant_sentence_2"
},
{
"operation": "Highlighting",
"time": "326.334",
"value": "Flowering plant_sentence_2"
},
{
"operation": "Page search",
"time": "350.466",
"value": "distinct"
},
{
"operation": "page-search-reset",
"time": "376.428",
"value": "page-search-reset"
},
{
"operation": "Highlighting deleted",
"time": "378.305",
"value": "Flowering plant_sentence_2"
},
{
"operation": "Highlighting",
"time": "378.866",
"value": "Flowering plant_sentence_2"
},
{
"operation": "hyperlink",
"time": "414.114",
"value": "genera"
},
{
"operation": "Now on",
"time": "414.919",
"value": "Genera"
},
{
"operation": "Page search",
"time": "420.623",
"value": "distinct"
},
{
"operation": "Page search",
"time": "436.798",
"value": "order"
},
{
"operation": "Page search",
"time": "471.439",
"value": "plants"
},
{
"operation": "hyperlink",
"time": "491.365",
"value": "List of the largest genera of flowering plants"
},
{
"operation": "Now on",
"time": "492.955",
"value": "List of the largest genera of flowering plants"
},
{
"operation": "Page search",
"time": "499.65",
"value": "distinct"
},
{
"operation": "search",
"time": "543.073",
"value": "Flowering plant"
},
{
"operation": "Now on",
"time": "544.84",
"value": "Flowering plant"
},
{
"operation": "search",
"time": "558.232",
"value": "Contains...genera of flowering plants"
},
{
"operation": "Now on",
"time": "559.941",
"value": "undefined"
},
{
"operation": "finish",
"time": "631.507",
"value": "finish"
}
]
}
|
exfever_4896
|
Jack Black is an American actor, comedian, musician, and songwriter who starred in the 2011 American biographical black rom-com film.
|
{
"evidence": [
"Jack Black",
"Bernie (2011 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJack Black is an American actor, comedian, musician, and songwriter who starred in the 2011 American biographical black rom-com film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jack Black",
"Bernie (2011 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4076,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
7249
|
Tennant McVea, a center back footballer, was born January 31, 1978, in Newtownards, Northern Ireland where he began as a youth and continued up through college until currently, he is playing for Lionsbridge FC.
|
{
"evidence": [
"Tennant McVea_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:\nTennant McVea, a center back footballer, was born January 31, 1978, in Newtownards, Northern Ireland where he began as a youth and continued up through college until currently, he is playing for Lionsbridge FC.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Tennant McVea_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 4077,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.739",
"value": "?search="
},
{
"operation": "search",
"time": "13.651",
"value": "Tennant McVea"
},
{
"operation": "Now on",
"time": "14.66",
"value": "Tennant McVea"
},
{
"operation": "Highlighting",
"time": "36.792",
"value": "Tennant McVea_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "37.104",
"value": "Tennant McVea_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "37.117",
"value": "Tennant McVea_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "39.232",
"value": "Tennant McVea_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "43.578",
"value": "Tennant McVea_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "43.894",
"value": "Tennant McVea_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "43.924",
"value": "Tennant McVea_header_cell_0_0_0"
},
{
"operation": "finish",
"time": "57.84",
"value": "finish"
}
]
}
|
exfever_179
|
STS-5 was the first operational flight of the NASA Space Shuttle mission, followed by STS-6, which was also a success.
|
{
"evidence": [
"STS-96",
"Space Shuttle",
"STS-5"
],
"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:\nSTS-5 was the first operational flight of the NASA Space Shuttle mission, followed by STS-6, which was also a success.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"STS-96",
"Space Shuttle",
"STS-5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 4078,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
57413
|
The band was transformed in October 1944 as a training band unit of the Soviet Navy; then in February 1993, in honor of the 150th birthday of Russian composer Nikolai Rimsky-Korsakov, the orchestra was given its current honorary name of the Central Concert Band of the Navy Nikolai Rimsky-Korsakov.
|
{
"evidence": [
"Central Navy Band of Russia_sentence_8",
"Central Navy Band of Russia_sentence_9"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe band was transformed in October 1944 as a training band unit of the Soviet Navy; then in February 1993, in honor of the 150th birthday of Russian composer Nikolai Rimsky-Korsakov, the orchestra was given its current honorary name of the Central Concert Band of the Navy Nikolai Rimsky-Korsakov.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Central Navy Band of Russia_sentence_8",
"Central Navy Band of Russia_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4079,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.299",
"value": "undefined"
},
{
"operation": "search",
"time": "84.544",
"value": "Central Navy Band of Russia"
},
{
"operation": "Now on",
"time": "85.895",
"value": "Central Navy Band of Russia"
},
{
"operation": "Highlighting",
"time": "108.596",
"value": "Central Navy Band of Russia_sentence_8"
},
{
"operation": "Highlighting",
"time": "115.454",
"value": "Central Navy Band of Russia_sentence_9"
},
{
"operation": "finish",
"time": "362.631",
"value": "finish"
}
]
}
|
77045
|
Yadollah Royaee’s work, Sea Songs is full of symbolism and heavily influenced by Persian mysticism.
|
{
"evidence": [
"Yadollah Royaee_sentence_4",
"Yadollah Royaee_sentence_6"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nYadollah Royaee’s work, Sea Songs is full of symbolism and heavily influenced by Persian mysticism.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Yadollah Royaee_sentence_4",
"Yadollah Royaee_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4080,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.255",
"value": "?search="
},
{
"operation": "search",
"time": "89.609",
"value": "Yadollah Royaee"
},
{
"operation": "Now on",
"time": "90.733",
"value": "Yadollah Royaee"
},
{
"operation": "Page search",
"time": "149.374",
"value": "sea songs"
},
{
"operation": "Highlighting",
"time": "224.314",
"value": "Yadollah Royaee_sentence_4"
},
{
"operation": "Highlighting",
"time": "240.928",
"value": "Yadollah Royaee_sentence_6"
},
{
"operation": "finish",
"time": "325.716",
"value": "finish"
}
]
}
|
exfever_2456
|
Akira Toriyama is a character designer for a 2007 role-playing video game developed by Mistwalker and Artoon.
|
{
"evidence": [
"Akira Toriyama",
"Blue Dragon (video game)"
],
"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:\nAkira Toriyama is a character designer for a 2007 role-playing video game developed by Mistwalker and Artoon.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Akira Toriyama",
"Blue Dragon (video game)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4081,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2351
|
The Lion King is a 1994 musical drama film produced by an American film producer of most successful animated films in recent history.
|
{
"evidence": [
"The Lion King",
"Don Hahn"
],
"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 Lion King is a 1994 musical drama film produced by an American film producer of most successful animated films in recent history.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Lion King",
"Don Hahn"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4082,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
31271
|
The 1937 Michigan State Spartans football team, under head coach Charlie Bachman, had only two losses out of the ten games scheduled for the season.
|
{
"evidence": [
"1937 Michigan State Spartans football team_sentence_1"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 1937 Michigan State Spartans football team, under head coach Charlie Bachman, had only two losses out of the ten games scheduled for the season.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"1937 Michigan State Spartans football team_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 4083,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.591",
"value": "?search="
},
{
"operation": "search",
"time": "12.854",
"value": "1937 Michigan State Spartans football team"
},
{
"operation": "Now on",
"time": "13.574",
"value": "1937 Michigan State Spartans football team"
},
{
"operation": "Highlighting",
"time": "41.824",
"value": "1937 Michigan State Spartans football team_sentence_1"
},
{
"operation": "finish",
"time": "71.669",
"value": "finish"
}
]
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.