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_6605
|
Emmanuel Adebayor is a Togolese professional footballer who previously played for English clubs, including an English basketball club which moved to Etihad Stadium after having played at a football stadium in Moss Side.
|
{
"evidence": [
"Emmanuel Adebayor",
"Manchester City F.C.",
"Maine Road"
],
"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:\nEmmanuel Adebayor is a Togolese professional footballer who previously played for English clubs, including an English basketball club which moved to Etihad Stadium after having played at a football stadium in Moss Side.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Emmanuel Adebayor",
"Manchester City F.C.",
"Maine Road"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2584,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
84449
|
The longest-reigning in the list of premiers or the first minister for Alberta was Ernest Manning with seven mandates counting 25 years and 195 days in office, followed by Peter Lougheed with four mandates serving 14 years and 52 days.
|
{
"evidence": [
"List of premiers of Alberta_sentence_0",
"Ernest Manning_sentence_1",
"Ernest Manning_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe longest-reigning in the list of premiers or the first minister for Alberta was Ernest Manning with seven mandates counting 25 years and 195 days in office, followed by Peter Lougheed with four mandates serving 14 years and 52 days.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"List of premiers of Alberta_sentence_0",
"Ernest Manning_sentence_1",
"Ernest Manning_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2585,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.469",
"value": "?search="
},
{
"operation": "Now on",
"time": "13.521",
"value": "?search="
},
{
"operation": "search",
"time": "214.313",
"value": "List of premiers of Alberta"
},
{
"operation": "Now on",
"time": "216.607",
"value": "List of premiers of Alberta"
},
{
"operation": "Highlighting",
"time": "218.76",
"value": "List of premiers of Alberta_sentence_0"
},
{
"operation": "search",
"time": "462.689",
"value": "Contains...first minister for Alberta"
},
{
"operation": "Now on",
"time": "464.711",
"value": "undefined"
},
{
"operation": "search",
"time": "477.235",
"value": "Ernest Manning"
},
{
"operation": "Now on",
"time": "478.424",
"value": "Ernest Manning"
},
{
"operation": "Highlighting",
"time": "486.609",
"value": "Ernest Manning_sentence_1"
},
{
"operation": "Highlighting",
"time": "493.013",
"value": "Ernest Manning_sentence_0"
},
{
"operation": "finish",
"time": "548.006",
"value": "finish"
}
]
}
|
exfever_3500
|
Peter Edward "Ginger" Baker was an English drummer, composer, songwriter, and the co-founder and occasional vocalist of the rock band formed in London in 1966, consisting of Eric Clapton, Ginger Baker, and a Scottish bassist and singer-songwriter that joined the Graham Bond Organisation in the early 1960s.
|
{
"evidence": [
"Ginger Baker",
"Cream (band)",
"Jack Bruce"
],
"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:\nPeter Edward \"Ginger\" Baker was an English drummer, composer, songwriter, and the co-founder and occasional vocalist of the rock band formed in London in 1966, consisting of Eric Clapton, Ginger Baker, and a Scottish bassist and singer-songwriter that joined the Graham Bond Organisation in the early 1960s.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ginger Baker",
"Cream (band)",
"Jack Bruce"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2586,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36246
|
Mika Häkkinen was the champion of the 1998 British Grand Prix held at the Silverstone Circuit, England with a lap time of 1:23.271; followed by Michael Schumacher from Germany with a lap time of 1:23.720.
|
{
"evidence": [
"1998 British Grand Prix_sentence_2",
"1998 British Grand Prix_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:\nMika Häkkinen was the champion of the 1998 British Grand Prix held at the Silverstone Circuit, England with a lap time of 1:23.271; followed by Michael Schumacher from Germany with a lap time of 1:23.720.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1998 British Grand Prix_sentence_2",
"1998 British Grand Prix_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2587,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.637",
"value": "?search="
},
{
"operation": "search",
"time": "10.249",
"value": "Mika Häkkinen"
},
{
"operation": "Now on",
"time": "11.481",
"value": "Mika Häkkinen"
},
{
"operation": "Highlighting",
"time": "15.263",
"value": "Mika Häkkinen_sentence_1"
},
{
"operation": "Highlighting",
"time": "15.264",
"value": "Mika Häkkinen_title"
},
{
"operation": "Highlighting deleted",
"time": "17.322",
"value": "Mika Häkkinen_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "17.324",
"value": "Mika Häkkinen_title"
},
{
"operation": "Page search",
"time": "69.484",
"value": "1998 British Grand Prix"
},
{
"operation": "search",
"time": "72.328",
"value": "1998 British Grand Prix"
},
{
"operation": "Now on",
"time": "73.374",
"value": "1998 British Grand Prix"
},
{
"operation": "Highlighting",
"time": "78.836",
"value": "1998 British Grand Prix_sentence_3"
},
{
"operation": "Highlighting",
"time": "88.995",
"value": "1998 British Grand Prix_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "89.838",
"value": "1998 British Grand Prix_sentence_3"
},
{
"operation": "Highlighting",
"time": "90.03",
"value": "1998 British Grand Prix_sentence_3"
},
{
"operation": "finish",
"time": "104.407",
"value": "finish"
}
]
}
|
exfever_6693
|
Twitter is an American online broadcast medium that may include text messaging or an electronic representation of moving visual images in the form of encoded digital data.
|
{
"evidence": [
"Twitter",
"Microblogging",
"Digital video"
],
"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:\nTwitter is an American online broadcast medium that may include text messaging or an electronic representation of moving visual images in the form of encoded digital data.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Twitter",
"Microblogging",
"Digital video"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2588,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
74901
|
Families who value piety control multiple aspects of the members' lives such as social status, religion, clothing, and individual freedom.
|
{
"evidence": [
"Piety_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:\nFamilies who value piety control multiple aspects of the members' lives such as social status, religion, clothing, and individual freedom.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Piety_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2589,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.517",
"value": "?search="
},
{
"operation": "search",
"time": "12.064",
"value": "Contains...Families who value piety control multiple aspects of the members' lives"
},
{
"operation": "Now on",
"time": "14.948",
"value": "undefined"
},
{
"operation": "search",
"time": "54.699",
"value": "Contains...multiple aspects of the members' lives such as social status, religion, clothing, and individual freedom"
},
{
"operation": "Now on",
"time": "58.019",
"value": "undefined"
},
{
"operation": "search",
"time": "86.447",
"value": "Contains...Families who value piety"
},
{
"operation": "Now on",
"time": "87.968",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "96.016",
"value": "Samurai"
},
{
"operation": "Now on",
"time": "99.786",
"value": "Samurai"
},
{
"operation": "Page search",
"time": "105.303",
"value": "Families who value piety"
},
{
"operation": "back-button-clicked",
"time": "125.689",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "126.66",
"value": "undefined"
},
{
"operation": "search",
"time": "142.767",
"value": "Contains...control multiple aspects of the members"
},
{
"operation": "Now on",
"time": "145.812",
"value": "undefined"
},
{
"operation": "search",
"time": "171.656",
"value": "Contains... value piety control"
},
{
"operation": "Now on",
"time": "172.861",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "178.515",
"value": "Confucianism"
},
{
"operation": "Now on",
"time": "180.465",
"value": "Confucianism"
},
{
"operation": "Page search",
"time": "195.994",
"value": "value piety control"
},
{
"operation": "search",
"time": "225.948",
"value": "Contains...control social status, religion, clothing"
},
{
"operation": "Now on",
"time": "227.215",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "248.132",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "248.891",
"value": "Confucianism"
},
{
"operation": "search",
"time": "260.743",
"value": "Contains...Families who value piety control multiple aspects of the members' lives such as social status, religion, clothing, and individual freedom"
},
{
"operation": "Now on",
"time": "261.609",
"value": "undefined"
},
{
"operation": "search",
"time": "325.994",
"value": "Families piety members social status, religion, clothing, individual freedom"
},
{
"operation": "Now on",
"time": "327.465",
"value": "undefined"
},
{
"operation": "search",
"time": "373.755",
"value": "Contains...members' lives"
},
{
"operation": "Now on",
"time": "374.619",
"value": "undefined"
},
{
"operation": "search",
"time": "417.788",
"value": "Piety"
},
{
"operation": "Now on",
"time": "419.179",
"value": "Piety"
},
{
"operation": "Page search",
"time": "433.47",
"value": "social status"
},
{
"operation": "Highlighting",
"time": "443.67",
"value": "Piety_sentence_0"
},
{
"operation": "Page search",
"time": "454.5",
"value": "individual freedom"
},
{
"operation": "finish",
"time": "461.411",
"value": "finish"
}
]
}
|
exfever_2396
|
Daniel Michael DeVito Jr. is an American actor known for his roles in a 1983 comedy-drama film directed by the co-founder of Gracie Films.
|
{
"evidence": [
"Danny DeVito",
"Terms of Endearment",
"James L. Brooks"
],
"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:\nDaniel Michael DeVito Jr. is an American actor known for his roles in a 1983 comedy-drama film directed by the co-founder of Gracie Films.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Danny DeVito",
"Terms of Endearment",
"James L. Brooks"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2590,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
23942
|
Sola, born in Pamplona, Navarre, made his senior debuts with UDC Chantrea in the summer of 2012; then made a deal with CA Osasuna on February 16, 2013.
|
{
"evidence": [
"Kike Sola_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:\nSola, born in Pamplona, Navarre, made his senior debuts with UDC Chantrea in the summer of 2012; then made a deal with CA Osasuna on February 16, 2013.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Kike Sola_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2591,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.609",
"value": "?search="
},
{
"operation": "search",
"time": "14.829",
"value": "Sola"
},
{
"operation": "Now on",
"time": "15.675",
"value": "Sola"
},
{
"operation": "hyperlink",
"time": "20.711",
"value": "Sola (disambiguation)"
},
{
"operation": "Now on",
"time": "21.224",
"value": "Sola (disambiguation)"
},
{
"operation": "hyperlink",
"time": "25.252",
"value": "2 People Sola (disambiguation)_section_1"
},
{
"operation": "hyperlink",
"time": "31.442",
"value": "Enrique Sola"
},
{
"operation": "Now on",
"time": "32.108",
"value": "Kike Sola"
},
{
"operation": "Highlighting",
"time": "61.272",
"value": "Kike Sola_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "114.43",
"value": "Kike Sola_sentence_3"
},
{
"operation": "back-button-clicked",
"time": "117.103",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "117.204",
"value": "Sola (disambiguation)#People Sola .28disambiguation.29 section 1"
},
{
"operation": "hyperlink",
"time": "121.475",
"value": "Guillermo Solá"
},
{
"operation": "Now on",
"time": "122.173",
"value": "Guillermo Solá"
},
{
"operation": "back-button-clicked",
"time": "131.151",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "131.251",
"value": "Sola (disambiguation)#People Sola .28disambiguation.29 section 1"
},
{
"operation": "hyperlink",
"time": "136.096",
"value": "Enrique Sola"
},
{
"operation": "Now on",
"time": "136.826",
"value": "Kike Sola"
},
{
"operation": "Page search",
"time": "144.697",
"value": "chantrea"
},
{
"operation": "Page search",
"time": "160.689",
"value": "ocasuna"
},
{
"operation": "search",
"time": "174.829",
"value": "sola ocasune"
},
{
"operation": "Now on",
"time": "175.605",
"value": "undefined"
},
{
"operation": "search",
"time": "183.696",
"value": "sola ocasuna"
},
{
"operation": "Now on",
"time": "184.421",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "189.187",
"value": "sola ocasuna"
},
{
"operation": "Now on",
"time": "189.683",
"value": "undefined"
},
{
"operation": "search",
"time": "199.968",
"value": "ca ocasuna"
},
{
"operation": "Now on",
"time": "200.797",
"value": "undefined"
},
{
"operation": "search",
"time": "217.998",
"value": "udc chantrea"
},
{
"operation": "Now on",
"time": "218.885",
"value": "UDC Txantrea"
},
{
"operation": "back-button-clicked",
"time": "228.596",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "228.676",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "229.352",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "229.361",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "229.415",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "230.07",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "230.076",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "230.662",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "231.371",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "231.377",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "231.437",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "232.538",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "232.545",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "233.177",
"value": "Kike Sola"
},
{
"operation": "back-button-clicked",
"time": "234.431",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "234.437",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "234.523",
"value": "Sola (disambiguation)#People Sola .28disambiguation.29 section 1"
},
{
"operation": "hyperlink",
"time": "238.684",
"value": "Enrique Sola"
},
{
"operation": "Now on",
"time": "239.108",
"value": "Enrique Sola"
},
{
"operation": "Highlighting",
"time": "270.01",
"value": "Kike Sola_sentence_3"
},
{
"operation": "finish",
"time": "291.079",
"value": "finish"
}
]
}
|
15629
|
Dragon Island is the island referred to in chapter eight of the historical novel Master and Commander and also in the fictional world of Narnia in the book The Voyage of the Dawn Treader.
|
{
"evidence": [
"Dragon Island_sentence_6",
"Dragon Island_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:\nDragon Island is the island referred to in chapter eight of the historical novel Master and Commander and also in the fictional world of Narnia in the book The Voyage of the Dawn Treader.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dragon Island_sentence_6",
"Dragon Island_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2592,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.787",
"value": "?search="
},
{
"operation": "search",
"time": "16.013",
"value": "Dragon Island"
},
{
"operation": "Now on",
"time": "17.268",
"value": "Dragon Island"
},
{
"operation": "Highlighting",
"time": "152.106",
"value": "Dragon Island_sentence_6"
},
{
"operation": "Highlighting",
"time": "155.111",
"value": "Dragon Island_sentence_7"
},
{
"operation": "finish",
"time": "162.528",
"value": "finish"
}
]
}
|
exfever_4296
|
Eva Longoria is most known for her role as Gabrielle Solis in the American comedy-drama television series whose genre is where the nature of an event is usually a murder or other crime.
|
{
"evidence": [
"Eva Longoria",
"Desperate Housewives",
"Mystery fiction"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nEva Longoria is most known for her role as Gabrielle Solis in the American comedy-drama television series whose genre is where the nature of an event is usually a murder or other crime.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Eva Longoria",
"Desperate Housewives",
"Mystery fiction"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2593,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5428
|
"Dirty Diana" is the ninth track on Jackson's seventh studio album released on August 31, 1987, by a Canadian record label owned by Sony Music Entertainment.
|
{
"evidence": [
"Dirty Diana",
"Bad (album)",
"Epic Records"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n\"Dirty Diana\" is the ninth track on Jackson's seventh studio album released on August 31, 1987, by a Canadian record label owned by Sony Music Entertainment.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dirty Diana",
"Bad (album)",
"Epic Records"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2594,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2674
|
Kumkum Bhagya is an Indonesian Hindi-language drama television series starring an actress who won the Indian Telly Award for Best Actress in a Leading Role.
|
{
"evidence": [
"Kumkum Bhagya",
"Sriti Jha"
],
"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:\nKumkum Bhagya is an Indonesian Hindi-language drama television series starring an actress who won the Indian Telly Award for Best Actress in a Leading Role.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kumkum Bhagya",
"Sriti Jha"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2595,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
31798
|
Marko Ciciliani's was born in 1970 and his parents emigrated to Germany, he studied composition and spent a year in New York City.
|
{
"evidence": [
"Marko Ciciliani_sentence_0",
"Marko Ciciliani_sentence_2",
"Marko Ciciliani_sentence_3",
"Marko Ciciliani_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:\nMarko Ciciliani's was born in 1970 and his parents emigrated to Germany, he studied composition and spent a year in New York City.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Marko Ciciliani_sentence_0",
"Marko Ciciliani_sentence_2",
"Marko Ciciliani_sentence_3",
"Marko Ciciliani_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2596,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.742",
"value": "?search="
},
{
"operation": "search",
"time": "23.067",
"value": "Marko Ciciliani"
},
{
"operation": "Now on",
"time": "24.092",
"value": "Marko Ciciliani"
},
{
"operation": "Highlighting",
"time": "28.867",
"value": "Marko Ciciliani_sentence_0"
},
{
"operation": "Page search",
"time": "36.264",
"value": "parent"
},
{
"operation": "Highlighting",
"time": "52.932",
"value": "Marko Ciciliani_sentence_2"
},
{
"operation": "Highlighting",
"time": "64.78",
"value": "Marko Ciciliani_sentence_3"
},
{
"operation": "Highlighting",
"time": "66.457",
"value": "Marko Ciciliani_sentence_4"
},
{
"operation": "finish",
"time": "185.333",
"value": "finish"
}
]
}
|
exfever_4453
|
Sammy Cahn is best known for his romantic lyrics to films and Broadway songs, as well as stand-alone songs premiered by recording companies in the second-largest metropolitan region in the United States.
|
{
"evidence": [
"Sammy Cahn",
"Greater Los Angeles Area"
],
"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:\nSammy Cahn is best known for his romantic lyrics to films and Broadway songs, as well as stand-alone songs premiered by recording companies in the second-largest metropolitan region in the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sammy Cahn",
"Greater Los Angeles Area"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2597,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
83166
|
Frank Grant was born in Pittsfield, Massachusetts and played semipro baseball in Pittsfield and in Plattsburgh, New York.
|
{
"evidence": [
"Frank Grant_sentence_8",
"Frank Grant_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:\nFrank Grant was born in Pittsfield, Massachusetts and played semipro baseball in Pittsfield and in Plattsburgh, New York.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Frank Grant_sentence_8",
"Frank Grant_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2598,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.133",
"value": "?search="
},
{
"operation": "search",
"time": "43.51",
"value": "Frank Grant"
},
{
"operation": "Now on",
"time": "44.139",
"value": "Frank Grant"
},
{
"operation": "Highlighting",
"time": "77.487",
"value": "Frank Grant_sentence_8"
},
{
"operation": "Highlighting",
"time": "100.729",
"value": "Frank Grant_sentence_9"
},
{
"operation": "finish",
"time": "110.602",
"value": "finish"
}
]
}
|
56236
|
Lassana Fané was born October 12, 1987 in Bamako, Mali (the capital and smallest city of Mali) and he played the Midfielder position.
|
{
"evidence": [
"Lassana Fané_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:\nLassana Fané was born October 12, 1987 in Bamako, Mali (the capital and smallest city of Mali) and he played the Midfielder position.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Lassana Fané_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2599,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.07",
"value": "?search="
},
{
"operation": "search",
"time": "11.149",
"value": "Lassana Fané"
},
{
"operation": "Now on",
"time": "12.095",
"value": "Lassana Fané"
},
{
"operation": "Highlighting",
"time": "21.738",
"value": "Lassana Fané_sentence_0"
},
{
"operation": "finish",
"time": "119.577",
"value": "finish"
}
]
}
|
90089
|
Kaufmann Kohler protested against Reform Judaism when he moved to America in 1869, and in 1886 wrote for a weekly magazine with interests in Orthodox Judaism.
|
{
"evidence": [
"Kaufmann Kohler_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:\nKaufmann Kohler protested against Reform Judaism when he moved to America in 1869, and in 1886 wrote for a weekly magazine with interests in Orthodox Judaism.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Kaufmann Kohler_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2600,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.197",
"value": "?search="
},
{
"operation": "search",
"time": "5.892",
"value": "Kaufmann Kohler"
},
{
"operation": "Now on",
"time": "6.647",
"value": "Kaufmann Kohler"
},
{
"operation": "Highlighting",
"time": "30.493",
"value": "Kaufmann Kohler_sentence_10"
},
{
"operation": "Highlighting",
"time": "30.982",
"value": "Kaufmann Kohler_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "32.38",
"value": "Kaufmann Kohler_sentence_10"
},
{
"operation": "finish",
"time": "38.762",
"value": "finish"
}
]
}
|
exfever_5606
|
Liv Rundgren Tyler, the daughter of Bebe Buell and Steven Tyler, made her film debut in a 1994 American psychological thriller directed by Bruce Beresford.
|
{
"evidence": [
"Liv Tyler",
"Silent Fall"
],
"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:\nLiv Rundgren Tyler, the daughter of Bebe Buell and Steven Tyler, made her film debut in a 1994 American psychological thriller directed by Bruce Beresford.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Liv Tyler",
"Silent Fall"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2601,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2153
|
Marilyn Monroe was the first American model who became famous for playing the "blonde bombshell".?
|
{
"evidence": [
"Marilyn Monroe",
"AFI's 100 Years...100 Stars",
"AFI 100 Years... series"
],
"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:\nMarilyn Monroe was the first American model who became famous for playing the \"blonde bombshell\".?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Marilyn Monroe",
"AFI's 100 Years...100 Stars",
"AFI 100 Years... series"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2602,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71339
|
Hans Kockelmans composes for unusual instruments such as Phantastron, a type of electron tube oscillator or timer circuit, and has collaborated with Gerard Kockelmans on a few experimental compositions.
|
{
"evidence": [
"Hans Kockelmans_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:\nHans Kockelmans composes for unusual instruments such as Phantastron, a type of electron tube oscillator or timer circuit, and has collaborated with Gerard Kockelmans on a few experimental compositions.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Hans Kockelmans_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2603,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.891",
"value": "?search="
},
{
"operation": "search",
"time": "14.616",
"value": "Hans Kockelmans"
},
{
"operation": "Now on",
"time": "15.625",
"value": "Hans Kockelmans"
},
{
"operation": "Highlighting",
"time": "24.28",
"value": "Hans Kockelmans_sentence_4"
},
{
"operation": "hyperlink",
"time": "26.808",
"value": "Phantastron"
},
{
"operation": "Now on",
"time": "27.45",
"value": "Phantastron"
},
{
"operation": "Highlighting",
"time": "31.199",
"value": "Phantastron_sentence_0"
},
{
"operation": "Highlighting",
"time": "31.2",
"value": "Phantastron_title"
},
{
"operation": "back-button-clicked",
"time": "32.849",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "32.934",
"value": "Hans Kockelmans"
},
{
"operation": "Highlighting",
"time": "42.247",
"value": "Hans Kockelmans_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "43.598",
"value": "Hans Kockelmans_sentence_4"
},
{
"operation": "finish",
"time": "54.891",
"value": "finish"
}
]
}
|
63197
|
Ryō Hirohashi remembered a radio show she used to listen to in junior high school and became interested in voice acting which persuaded her to become a voice actor herself.
|
{
"evidence": [
"Ryō Hirohashi_sentence_5",
"Ryō Hirohashi_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:\nRyō Hirohashi remembered a radio show she used to listen to in junior high school and became interested in voice acting which persuaded her to become a voice actor herself.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ryō Hirohashi_sentence_5",
"Ryō Hirohashi_sentence_6"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2604,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.916",
"value": "?search="
},
{
"operation": "search",
"time": "11.892",
"value": "Ryō Hirohashi"
},
{
"operation": "Now on",
"time": "13.9",
"value": "Ryō Hirohashi"
},
{
"operation": "Highlighting",
"time": "18.922",
"value": "Ryō Hirohashi_sentence_5"
},
{
"operation": "Highlighting",
"time": "26.818",
"value": "Ryō Hirohashi_sentence_6"
},
{
"operation": "finish",
"time": "41.786",
"value": "finish"
}
]
}
|
195
|
Hart to Hart had 24 episodes featured for Season 3 from 1981 to 1982.
|
{
"evidence": [
"Stefanie Powers_sentence_52"
],
"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:\nHart to Hart had 24 episodes featured for Season 3 from 1981 to 1982.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Stefanie Powers_sentence_52"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2605,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "search",
"time": "0",
"value": "Hart to Hart"
},
{
"operation": "hyperlink",
"time": "12.709",
"value": "Stefanie Powers"
},
{
"operation": "Highlighting",
"time": "88.038",
"value": "Stefanie Powers_sentence_52"
},
{
"operation": "search",
"time": "133.827",
"value": "Season 3 Hart to Hart"
},
{
"operation": "finish",
"time": "166.092",
"value": "finish"
}
]
}
|
exfever_6575
|
The 2013 NBA draft on June 27, 2013, was the first draft for New Orleans under their new names often referred as the Pels, a Canadian professional basketball team based in New Orleans.
|
{
"evidence": [
"2013 NBA draft",
"New Orleans Pelicans"
],
"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 2013 NBA draft on June 27, 2013, was the first draft for New Orleans under their new names often referred as the Pels, a Canadian professional basketball team based in New Orleans.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"2013 NBA draft",
"New Orleans Pelicans"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2606,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3552
|
The Republican Party is one of the two major contemporary political parties in the United States that was strengthened by the collapse of a political party that espoused traditionalist conservatism in the United States during the middle of the 17th century.
|
{
"evidence": [
"Republican Party (United States)",
"Whig Party (United States)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Republican Party is one of the two major contemporary political parties in the United States that was strengthened by the collapse of a political party that espoused traditionalist conservatism in the United States during the middle of the 17th century.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Republican Party (United States)",
"Whig Party (United States)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2607,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5066
|
The Who are an English rock band consisting of lead singer with a hit song with the Who including the song that was named as the 11th greatest song by Rolling Stone on its list of the 500 Greatest Songs of All Time.
|
{
"evidence": [
"The Who",
"Roger Daltrey",
"My Generation"
],
"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 Who are an English rock band consisting of lead singer with a hit song with the Who including the song that was named as the 11th greatest song by Rolling Stone on its list of the 500 Greatest Songs of All Time.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Who",
"Roger Daltrey",
"My Generation"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2608,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2452
|
The Northeast megapolis has the longest railroad in the United States.
|
{
"evidence": [
"Northeast megalopolis",
"Northeast Corridor"
],
"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 Northeast megapolis has the longest railroad in the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Northeast megalopolis",
"Northeast Corridor"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2609,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5051
|
William Richard Tolbert Jr. was the 20th President of Liberia who was killed in a coup d'état led by a Liberian politician who served as the Liberian leader from 1980 to 1997.
|
{
"evidence": [
"William R. Tolbert Jr.",
"Samuel Doe"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nWilliam Richard Tolbert Jr. was the 20th President of Liberia who was killed in a coup d'état led by a Liberian politician who served as the Liberian leader from 1980 to 1997.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"William R. Tolbert Jr.",
"Samuel Doe"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2610,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2865
|
Philippine Airlines, the second-largest UK-based carrier, is headquartered in the capital and largest city in England and the United Kingdom.
|
{
"evidence": [
"British Airways",
"London"
],
"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:\nPhilippine Airlines, the second-largest UK-based carrier, is headquartered in the capital and largest city in England and the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"British Airways",
"London"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2611,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2908
|
Jacqueline Kennedy Onassis was attracted to Kennedy's physical appearance, wit and wealth.
|
{
"evidence": [
"Jacqueline Kennedy Onassis",
"John F. Kennedy",
"Cuba"
],
"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:\nJacqueline Kennedy Onassis was attracted to Kennedy's physical appearance, wit and wealth.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jacqueline Kennedy Onassis",
"John F. Kennedy",
"Cuba"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2612,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2670
|
Quincy Jones is a record producer who produced some of Michael Jackson's most successful albums, including the seventh studio album that was released on August 31, 1985.
|
{
"evidence": [
"Quincy Jones",
"Bad (album)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nQuincy Jones is a record producer who produced some of Michael Jackson's most successful albums, including the seventh studio album that was released on August 31, 1985.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Quincy Jones",
"Bad (album)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2613,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
72033
|
Mercy Nku is a retired Nigerian sprinter who specialized in the 100 metres which is a sprint race in track and field competitions.
|
{
"evidence": [
"Mercy Nku_sentence_0",
"100 metres_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:\nMercy Nku is a retired Nigerian sprinter who specialized in the 100 metres which is a sprint race in track and field competitions.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Mercy Nku_sentence_0",
"100 metres_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2614,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.618",
"value": "?search="
},
{
"operation": "search",
"time": "11.801",
"value": "Mercy Nku"
},
{
"operation": "Now on",
"time": "12.921",
"value": "Mercy Nku"
},
{
"operation": "Highlighting",
"time": "99.432",
"value": "Mercy Nku_sentence_0"
},
{
"operation": "hyperlink",
"time": "105.178",
"value": "100 metres"
},
{
"operation": "Now on",
"time": "105.999",
"value": "100 metres"
},
{
"operation": "Highlighting",
"time": "107.514",
"value": "100 metres_sentence_2"
},
{
"operation": "finish",
"time": "112.606",
"value": "finish"
}
]
}
|
45781
|
Russia Tower was proposed to be built on plots 2 and 3 of the Moscow International Business Centre in 1994, it was designed by the Chicago-based architecture firm Skidmore and building site was moved to plot 14.
|
{
"evidence": [
"Russia Tower_sentence_16"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRussia Tower was proposed to be built on plots 2 and 3 of the Moscow International Business Centre in 1994, it was designed by the Chicago-based architecture firm Skidmore and building site was moved to plot 14.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Russia Tower_sentence_16"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2615,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.238",
"value": "?search="
},
{
"operation": "search",
"time": "91.588",
"value": "Russia Tower"
},
{
"operation": "Now on",
"time": "92.617",
"value": "Russia Tower"
},
{
"operation": "Page search",
"time": "166.041",
"value": "plot"
},
{
"operation": "Highlighting",
"time": "168.928",
"value": "Russia Tower_sentence_15"
},
{
"operation": "Highlighting",
"time": "180.393",
"value": "Russia Tower_sentence_16"
},
{
"operation": "Page search",
"time": "210.648",
"value": "skidmore"
},
{
"operation": "Highlighting deleted",
"time": "229.55",
"value": "Russia Tower_sentence_15"
},
{
"operation": "Highlighting",
"time": "230.667",
"value": "Russia Tower_sentence_15"
},
{
"operation": "Highlighting deleted",
"time": "231.53",
"value": "Russia Tower_sentence_16"
},
{
"operation": "Highlighting",
"time": "232.332",
"value": "Russia Tower_sentence_16"
},
{
"operation": "Highlighting",
"time": "238.891",
"value": "Russia Tower_sentence_17"
},
{
"operation": "Highlighting deleted",
"time": "269.687",
"value": "Russia Tower_sentence_16"
},
{
"operation": "search",
"time": "275.665",
"value": "Skidmore, Owings and Merrill."
},
{
"operation": "Now on",
"time": "277.246",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "290.876",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "290.98",
"value": "Russia Tower"
},
{
"operation": "hyperlink",
"time": "293.054",
"value": "Moscow International Business Centre"
},
{
"operation": "Now on",
"time": "295.137",
"value": "Moscow International Business Centre"
},
{
"operation": "back-button-clicked",
"time": "302.079",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "302.178",
"value": "Russia Tower"
},
{
"operation": "Highlighting",
"time": "304.218",
"value": "Russia Tower_sentence_16"
},
{
"operation": "Highlighting deleted",
"time": "304.64",
"value": "Russia Tower_sentence_15"
},
{
"operation": "Highlighting deleted",
"time": "305.432",
"value": "Russia Tower_sentence_17"
},
{
"operation": "finish",
"time": "347.385",
"value": "finish"
}
]
}
|
75000
|
Sho Dozono is a Japanese-American Business owner and CEO who has never been to Portland, Oregon.
|
{
"evidence": [
"Sho Dozono_sentence_0",
"Sho Dozono_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:\nSho Dozono is a Japanese-American Business owner and CEO who has never been to Portland, Oregon.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sho Dozono_sentence_0",
"Sho Dozono_sentence_9"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2616,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.596",
"value": "?search="
},
{
"operation": "search",
"time": "292.373",
"value": "Sho Dozono"
},
{
"operation": "Now on",
"time": "294.82",
"value": "Sho Dozono"
},
{
"operation": "Highlighting",
"time": "305.04",
"value": "Sho Dozono_sentence_0"
},
{
"operation": "Highlighting",
"time": "373.476",
"value": "Sho Dozono_sentence_9"
},
{
"operation": "finish",
"time": "564.295",
"value": "finish"
}
]
}
|
24854
|
Hillhead, is falsely believed to be north of Kelvingrove Park and have a population of 500 and is located in Scotland.
|
{
"evidence": [
"Hillhead_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:\nHillhead, is falsely believed to be north of Kelvingrove Park and have a population of 500 and is located in Scotland.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hillhead_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2617,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.068",
"value": "?search="
},
{
"operation": "search",
"time": "7.185",
"value": "Hillhead"
},
{
"operation": "Now on",
"time": "7.73",
"value": "Hillhead"
},
{
"operation": "Highlighting",
"time": "25.526",
"value": "Hillhead_sentence_3"
},
{
"operation": "finish",
"time": "30.635",
"value": "finish"
}
]
}
|
exfever_1768
|
Usher Raymond IV was born in the tenth most populous city in the U.S. and the third-largest in Texas after Houston and San Antonio.
|
{
"evidence": [
"Usher (musician)",
"Dallas"
],
"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:\nUsher Raymond IV was born in the tenth most populous city in the U.S. and the third-largest in Texas after Houston and San Antonio.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Usher (musician)",
"Dallas"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2618,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
39812
|
In the 1998 Lebanese presidential election, Emile Lahoud was elected to be President of the Lebanese Republic; in the Lebanese Constitution, a majority of two-thirds of the 128-seat Parliament would be required to elect him; every deput in attendance voted for him.
|
{
"evidence": [
"1998 Lebanese presidential election_sentence_6",
"1998 Lebanese presidential election_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:\nIn the 1998 Lebanese presidential election, Emile Lahoud was elected to be President of the Lebanese Republic; in the Lebanese Constitution, a majority of two-thirds of the 128-seat Parliament would be required to elect him; every deput in attendance voted for him.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1998 Lebanese presidential election_sentence_6",
"1998 Lebanese presidential election_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2619,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.702",
"value": "?search="
},
{
"operation": "search",
"time": "12.803",
"value": "1998 Lebanese presidential election"
},
{
"operation": "Now on",
"time": "13.718",
"value": "1998 Lebanese presidential election"
},
{
"operation": "Highlighting",
"time": "29.115",
"value": "1998 Lebanese presidential election_sentence_6"
},
{
"operation": "Highlighting",
"time": "55.814",
"value": "1998 Lebanese presidential election_sentence_2"
},
{
"operation": "hyperlink",
"time": "62.837",
"value": "qualified majority"
},
{
"operation": "Now on",
"time": "63.802",
"value": "Supermajority"
},
{
"operation": "back-button-clicked",
"time": "78.923",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "79.019",
"value": "1998 Lebanese presidential election"
},
{
"operation": "finish",
"time": "117.291",
"value": "finish"
}
]
}
|
71407
|
New Renaissance Pictures is a music production and distribution company in La Habra, California, that was founded in July 2005.
|
{
"evidence": [
"New Renaissance Pictures_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:\nNew Renaissance Pictures is a music production and distribution company in La Habra, California, that was founded in July 2005.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"New Renaissance Pictures_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2620,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.772",
"value": "?search="
},
{
"operation": "search",
"time": "18.748",
"value": "New Renaissance Pictures"
},
{
"operation": "Now on",
"time": "19.868",
"value": "New Renaissance Pictures"
},
{
"operation": "Highlighting",
"time": "63.051",
"value": "New Renaissance Pictures_sentence_0"
},
{
"operation": "finish",
"time": "94.071",
"value": "finish"
}
]
}
|
exfever_5839
|
The film Basic Instinct follows San Francisco police detective Nick Curran played by an American actor and producer who has received numerous accolades, including an award bestowed by the Academy of Television Arts & Sciences in recognition of excellence in African primetime television programming.
|
{
"evidence": [
"Basic Instinct",
"Michael Douglas",
"Primetime Emmy Award"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe film Basic Instinct follows San Francisco police detective Nick Curran played by an American actor and producer who has received numerous accolades, including an award bestowed by the Academy of Television Arts & Sciences in recognition of excellence in African primetime television programming.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Basic Instinct",
"Michael Douglas",
"Primetime Emmy Award"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2621,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5664
|
Florida is located in a country located in North America.
|
{
"evidence": [
"Florida",
"United States"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nFlorida is located in a country located in North America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Florida",
"United States"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2622,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29306
|
Alex Caskey, who willingly resigned from 2011 MLS SuperDraft, signed with the USL Pro League and made his professional debut on April 2011.
|
{
"evidence": [
"Alex Caskey_sentence_6",
"Alex Caskey_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:\nAlex Caskey, who willingly resigned from 2011 MLS SuperDraft, signed with the USL Pro League and made his professional debut on April 2011.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Alex Caskey_sentence_6",
"Alex Caskey_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2623,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.095",
"value": "?search="
},
{
"operation": "search",
"time": "18.413",
"value": "Alex Caskey"
},
{
"operation": "Now on",
"time": "19.451",
"value": "Alex Caskey"
},
{
"operation": "Highlighting",
"time": "32.649",
"value": "Alex Caskey_sentence_6"
},
{
"operation": "Highlighting",
"time": "39.247",
"value": "Alex Caskey_sentence_7"
},
{
"operation": "finish",
"time": "75.443",
"value": "finish"
}
]
}
|
exfever_1139
|
Kaakha Kaakha has also done a remake in Telugu.
|
{
"evidence": [
"Kaakha Kaakha",
"Harris Jayaraj"
],
"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:\nKaakha Kaakha has also done a remake in Telugu.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kaakha Kaakha",
"Harris Jayaraj"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2624,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
74469
|
Art Hirahara started piano at age four, later studying at the Oberlin Conservatory of Music and, afterwards, the California Institute of the Arts.
|
{
"evidence": [
"Art Hirahara_sentence_1",
"Art Hirahara_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:\nArt Hirahara started piano at age four, later studying at the Oberlin Conservatory of Music and, afterwards, the California Institute of the Arts.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Art Hirahara_sentence_1",
"Art Hirahara_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2625,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.832",
"value": "?search="
},
{
"operation": "search",
"time": "10.987",
"value": "Art Hirahara"
},
{
"operation": "Now on",
"time": "11.81",
"value": "Art Hirahara"
},
{
"operation": "Highlighting",
"time": "14.968",
"value": "Art Hirahara_sentence_1"
},
{
"operation": "Highlighting",
"time": "17.921",
"value": "Art Hirahara_sentence_2"
},
{
"operation": "finish",
"time": "24.727",
"value": "finish"
}
]
}
|
exfever_7512
|
River Jude Phoenix grew up in an itinerant family, as the older brother of an American actress, musician, and singer born on November 21, 1972.
|
{
"evidence": [
"River Phoenix",
"Rain Phoenix"
],
"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:\nRiver Jude Phoenix grew up in an itinerant family, as the older brother of an American actress, musician, and singer born on November 21, 1972.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"River Phoenix",
"Rain Phoenix"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2626,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
65353
|
Rachid Baba Ahmed (20 August 1946 – 15 February 1995) was a Algerian record producer, composer, and singerinvolved in the regional genre known as raï and dubbed as the ‘Jean-Michel Jarre of raï’
|
{
"evidence": [
"Rachid Baba Ahmed_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:\nRachid Baba Ahmed (20 August 1946 – 15 February 1995) was a Algerian record producer, composer, and singerinvolved in the regional genre known as raï and dubbed as the ‘Jean-Michel Jarre of raï’\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Rachid Baba Ahmed_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2627,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.904",
"value": "undefined"
},
{
"operation": "search",
"time": "14.416",
"value": "Rachid Baba Ahmed"
},
{
"operation": "Now on",
"time": "16.815",
"value": "Rachid Baba Ahmed"
},
{
"operation": "Highlighting",
"time": "21.439",
"value": "Rachid Baba Ahmed_sentence_0"
},
{
"operation": "hyperlink",
"time": "45.303",
"value": "raï"
},
{
"operation": "Now on",
"time": "49.172",
"value": "Raï"
},
{
"operation": "Page search",
"time": "54.848",
"value": "dubbed"
},
{
"operation": "Page search",
"time": "76.101",
"value": "jarre"
},
{
"operation": "back-button-clicked",
"time": "88.798",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "88.957",
"value": "Rachid Baba Ahmed"
},
{
"operation": "finish",
"time": "239.129",
"value": "finish"
}
]
}
|
400
|
Rugby tournament, the 2011 Safari Sevens, was played for the first time in RFUEA Ground, built to hold 30,000 spectators, and currently used primarily for football matches.
|
{
"evidence": [
"2011 Safari Sevens_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRugby tournament, the 2011 Safari Sevens, was played for the first time in RFUEA Ground, built to hold 30,000 spectators, and currently used primarily for football matches.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"2011 Safari Sevens_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2628,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "61.259",
"value": "2011 Safari Sevens"
},
{
"operation": "Highlighting",
"time": "113.629",
"value": "2011 Safari Sevens_sentence_1"
},
{
"operation": "hyperlink",
"time": "128.128",
"value": "RFUEA Ground"
},
{
"operation": "back-button-clicked",
"time": "131.015",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "132.373",
"value": "Nyayo National Stadium"
},
{
"operation": "back-button-clicked",
"time": "142.33",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "147.003",
"value": "Nyayo National Stadium"
},
{
"operation": "finish",
"time": "155.609",
"value": "finish"
}
]
}
|
82269
|
Dicoronylene is a by-products in petroleum processing formed when three or more coronene molecules fuse.
|
{
"evidence": [
"Dicoronylene_sentence_14"
],
"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:\nDicoronylene is a by-products in petroleum processing formed when three or more coronene molecules fuse.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dicoronylene_sentence_14"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2629,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.781",
"value": "?search="
},
{
"operation": "search",
"time": "11.025",
"value": "Dicoronylene"
},
{
"operation": "Now on",
"time": "11.707",
"value": "Dicoronylene"
},
{
"operation": "Highlighting",
"time": "55.308",
"value": "Dicoronylene_sentence_14"
},
{
"operation": "finish",
"time": "69.638",
"value": "finish"
}
]
}
|
exfever_2539
|
World War I began on the 19th of July, 1914, in which the fighting expanded to the world's second-largest and second-most populous continents, after the Earth's largest and most populous continent and shares the continental landmass of Eurasia.
|
{
"evidence": [
"World War I",
"Africa",
"Asia"
],
"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:\nWorld War I began on the 19th of July, 1914, in which the fighting expanded to the world's second-largest and second-most populous continents, after the Earth's largest and most populous continent and shares the continental landmass of Eurasia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"World War I",
"Africa",
"Asia"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2630,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2036
|
Anneliese Louise van der Pol is a Dutch-born American actress and singer who has recorded several songs for a Canadian multinational entertainment and media conglomerate headquartered in Burbank, California.
|
{
"evidence": [
"Anneliese van der Pol",
"The Walt Disney Company"
],
"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:\nAnneliese Louise van der Pol is a Dutch-born American actress and singer who has recorded several songs for a Canadian multinational entertainment and media conglomerate headquartered in Burbank, California.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Anneliese van der Pol",
"The Walt Disney Company"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2631,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
61622
|
Kym Whitley played Auntie Suga in Next Friday in 2000, then after three years, he played Ormandy in Deliver Us from Eva (a 2003 British romantic comedy film starring LL Cool J and Gabrielle Union).
|
{
"evidence": [
"Deliver Us from Eva_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:\nKym Whitley played Auntie Suga in Next Friday in 2000, then after three years, he played Ormandy in Deliver Us from Eva (a 2003 British romantic comedy film starring LL Cool J and Gabrielle Union).\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Deliver Us from Eva_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2632,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.325",
"value": "?search="
},
{
"operation": "search",
"time": "40.523",
"value": "Kym Whitley"
},
{
"operation": "Now on",
"time": "42.086",
"value": "Kym Whitley"
},
{
"operation": "hyperlink",
"time": "46.347",
"value": "5 Filmography Kym Whitley_section_4"
},
{
"operation": "Highlighting",
"time": "50.689",
"value": "Kym Whitley_cell_1_2_2"
},
{
"operation": "Highlighting",
"time": "50.69",
"value": "Kym Whitley_title"
},
{
"operation": "Highlighting",
"time": "50.691",
"value": "Kym Whitley_section_4"
},
{
"operation": "Highlighting",
"time": "50.693",
"value": "Kym Whitley_header_cell_1_0_2"
},
{
"operation": "Highlighting",
"time": "51.568",
"value": "Kym Whitley_cell_1_2_1"
},
{
"operation": "Highlighting",
"time": "51.572",
"value": "Kym Whitley_header_cell_1_0_1"
},
{
"operation": "Highlighting",
"time": "52.948",
"value": "Kym Whitley_cell_1_2_0"
},
{
"operation": "Highlighting",
"time": "52.95",
"value": "Kym Whitley_header_cell_1_0_0"
},
{
"operation": "Highlighting",
"time": "59.478",
"value": "Kym Whitley_cell_1_6_2"
},
{
"operation": "Highlighting",
"time": "61.678",
"value": "Kym Whitley_cell_1_6_1"
},
{
"operation": "Highlighting",
"time": "63.106",
"value": "Kym Whitley_cell_1_6_0"
},
{
"operation": "hyperlink",
"time": "66.762",
"value": "Deliver Us from Eva"
},
{
"operation": "Now on",
"time": "67.812",
"value": "Deliver Us from Eva"
},
{
"operation": "Highlighting",
"time": "75.947",
"value": "Deliver Us from Eva_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "81.188",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "81.425",
"value": "Kym Whitley#Filmography Kym Whitley section 4"
},
{
"operation": "Highlighting deleted",
"time": "82.563",
"value": "Kym Whitley_cell_1_2_1"
},
{
"operation": "Highlighting deleted",
"time": "83.077",
"value": "Kym Whitley_cell_1_2_2"
},
{
"operation": "Highlighting deleted",
"time": "84.327",
"value": "Kym Whitley_cell_1_2_0"
},
{
"operation": "Highlighting deleted",
"time": "85.092",
"value": "Kym Whitley_cell_1_6_0"
},
{
"operation": "Highlighting deleted",
"time": "85.097",
"value": "Kym Whitley_header_cell_1_0_0"
},
{
"operation": "Highlighting deleted",
"time": "85.655",
"value": "Kym Whitley_cell_1_6_1"
},
{
"operation": "Highlighting deleted",
"time": "85.66",
"value": "Kym Whitley_header_cell_1_0_1"
},
{
"operation": "Highlighting deleted",
"time": "86.148",
"value": "Kym Whitley_cell_1_6_2"
},
{
"operation": "Highlighting deleted",
"time": "86.15",
"value": "Kym Whitley_title"
},
{
"operation": "Highlighting deleted",
"time": "86.152",
"value": "Kym Whitley_section_4"
},
{
"operation": "Highlighting deleted",
"time": "86.154",
"value": "Kym Whitley_header_cell_1_0_2"
},
{
"operation": "finish",
"time": "100.249",
"value": "finish"
}
]
}
|
43483
|
The passenger building of Sondrio railway station, following a renovation, consists of the following public spaces: ticketing, bar, kiosk and waiting room, while the station's building containing the public conveniences is now Milanese in style.
|
{
"evidence": [
"Sondrio railway station_sentence_12",
"Sondrio railway station_sentence_15"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe passenger building of Sondrio railway station, following a renovation, consists of the following public spaces: ticketing, bar, kiosk and waiting room, while the station's building containing the public conveniences is now Milanese in style.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sondrio railway station_sentence_12",
"Sondrio railway station_sentence_15"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2633,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.116",
"value": "?search="
},
{
"operation": "search",
"time": "35.244",
"value": "Sondrio railway station"
},
{
"operation": "Now on",
"time": "35.979",
"value": "Sondrio railway station"
},
{
"operation": "hyperlink",
"time": "59.949",
"value": "passenger building"
},
{
"operation": "Now on",
"time": "60.666",
"value": "Station building"
},
{
"operation": "back-button-clicked",
"time": "63.977",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "64.004",
"value": "Sondrio railway station"
},
{
"operation": "Page search",
"time": "71.763",
"value": "passenger"
},
{
"operation": "Highlighting",
"time": "76.38",
"value": "Sondrio railway station_sentence_12"
},
{
"operation": "Page search",
"time": "96.291",
"value": "milanese"
},
{
"operation": "Highlighting",
"time": "100.697",
"value": "Sondrio railway station_sentence_15"
},
{
"operation": "finish",
"time": "122.922",
"value": "finish"
}
]
}
|
20652
|
Christopher Clark was educated at the University of Sydney where he studied History, then received his PhD at the University of Cambridge and is now a Professor in Modern European History there.
|
{
"evidence": [
"Christopher Clark_sentence_6",
"Christopher Clark_sentence_5",
"Christopher Clark_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:\nChristopher Clark was educated at the University of Sydney where he studied History, then received his PhD at the University of Cambridge and is now a Professor in Modern European History there.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Christopher Clark_sentence_6",
"Christopher Clark_sentence_5",
"Christopher Clark_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2634,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.051",
"value": "?search="
},
{
"operation": "search",
"time": "70.102",
"value": "Christopher Clark"
},
{
"operation": "Now on",
"time": "70.95",
"value": "Christopher Clark"
},
{
"operation": "Highlighting",
"time": "133.492",
"value": "Christopher Clark_sentence_6"
},
{
"operation": "Highlighting",
"time": "138.651",
"value": "Christopher Clark_sentence_5"
},
{
"operation": "Highlighting",
"time": "142.193",
"value": "Christopher Clark_sentence_4"
},
{
"operation": "finish",
"time": "169.092",
"value": "finish"
}
]
}
|
87204
|
Holkham railway station has only one platform and is located at Holkham, a village and civil parish in the South Norfolk.
|
{
"evidence": [
"Holkham_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHolkham railway station has only one platform and is located at Holkham, a village and civil parish in the South Norfolk.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Holkham_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2635,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.13",
"value": "?search="
},
{
"operation": "search",
"time": "10.684",
"value": "Holkham railway station"
},
{
"operation": "Now on",
"time": "11.922",
"value": "Holkham railway station"
},
{
"operation": "Highlighting",
"time": "24.777",
"value": "Holkham railway station_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "24.781",
"value": "Holkham railway station_title"
},
{
"operation": "Highlighting",
"time": "25.707",
"value": "Holkham railway station_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "25.743",
"value": "Holkham railway station_header_cell_0_3_0"
},
{
"operation": "Highlighting",
"time": "25.753",
"value": "Holkham railway station_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "33.771",
"value": "Holkham railway station_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "33.811",
"value": "Holkham railway station_header_cell_0_1_0"
},
{
"operation": "hyperlink",
"time": "35.616",
"value": "Holkham"
},
{
"operation": "Now on",
"time": "36.673",
"value": "Holkham"
},
{
"operation": "Highlighting",
"time": "38.668",
"value": "Holkham_sentence_1"
},
{
"operation": "Highlighting",
"time": "57.058",
"value": "Holkham_sentence_1"
},
{
"operation": "finish",
"time": "85.693",
"value": "finish"
}
]
}
|
exfever_2175
|
Julie Christie is a British actress known for her performance in a 1965 French film directed by John Schlesinger.
|
{
"evidence": [
"Julie Christie",
"Darling (1965 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:\nJulie Christie is a British actress known for her performance in a 1965 French film directed by John Schlesinger.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Julie Christie",
"Darling (1965 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2636,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
22340
|
The model ALCO RS-3, a 1,600 hp (1.2 MW), B-B road switcher diesel-electric locomotive, fleet number 92-99 were acquired by Norfolk and Western Railways in 1956.
|
{
"evidence": [
"ALCO RS-3_sentence_0"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe model ALCO RS-3, a 1,600 hp (1.2 MW), B-B road switcher diesel-electric locomotive, fleet number 92-99 were acquired by Norfolk and Western Railways in 1956.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"ALCO RS-3_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2637,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.296",
"value": "?search="
},
{
"operation": "search",
"time": "52.747",
"value": "model ALCO RS-3"
},
{
"operation": "Now on",
"time": "54.98",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "60.255",
"value": "ALCO RS-3"
},
{
"operation": "Now on",
"time": "61.361",
"value": "ALCO RS-3"
},
{
"operation": "Highlighting",
"time": "86.964",
"value": "ALCO RS-3_sentence_0"
},
{
"operation": "Page search",
"time": "95.283",
"value": "92-99"
},
{
"operation": "Page search",
"time": "117.681",
"value": "fleet"
},
{
"operation": "Page search",
"time": "119.081",
"value": "fleet"
},
{
"operation": "Page search",
"time": "119.24",
"value": "fleet"
},
{
"operation": "Page search",
"time": "145.699",
"value": "Norfolk"
},
{
"operation": "Page search",
"time": "146.994",
"value": "Norfolk"
},
{
"operation": "Page search",
"time": "147.137",
"value": "Norfolk"
},
{
"operation": "Page search",
"time": "169.565",
"value": "western"
},
{
"operation": "Page search",
"time": "169.931",
"value": "western"
},
{
"operation": "Page search",
"time": "170.155",
"value": "western"
},
{
"operation": "Page search",
"time": "170.323",
"value": "western"
},
{
"operation": "finish",
"time": "191.595",
"value": "finish"
}
]
}
|
55123
|
Piątnica is a village founded by Janusz from Zaborowo which played an important role during World War 1.
|
{
"evidence": [
"Piątnica_sentence_0",
"Piątnica_sentence_4",
"Piątnica_sentence_6",
"Piątnica_sentence_7"
],
"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:\nPiątnica is a village founded by Janusz from Zaborowo which played an important role during World War 1.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Piątnica_sentence_0",
"Piątnica_sentence_4",
"Piątnica_sentence_6",
"Piątnica_sentence_7"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2638,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.098",
"value": "?search="
},
{
"operation": "search",
"time": "11.573",
"value": "Piątnica"
},
{
"operation": "Now on",
"time": "12.237",
"value": "Piątnica"
},
{
"operation": "Highlighting",
"time": "19.017",
"value": "Piątnica_sentence_0"
},
{
"operation": "Highlighting",
"time": "28.478",
"value": "Piątnica_sentence_4"
},
{
"operation": "Highlighting",
"time": "36.162",
"value": "Piątnica_sentence_6"
},
{
"operation": "Highlighting",
"time": "43.215",
"value": "Piątnica_sentence_7"
},
{
"operation": "finish",
"time": "84.006",
"value": "finish"
}
]
}
|
67751
|
Juan Dies, who taught at Earlham College and Indiana University, is the co-founder and executive director of Sones de Mexico Ensemble, a Chicago folk music group that specializes in the Mexican musical tradition known as son.
|
{
"evidence": [
"Juan Dies_sentence_2",
"Juan Dies_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:\nJuan Dies, who taught at Earlham College and Indiana University, is the co-founder and executive director of Sones de Mexico Ensemble, a Chicago folk music group that specializes in the Mexican musical tradition known as son.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Juan Dies_sentence_2",
"Juan Dies_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2639,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.426",
"value": "?search="
},
{
"operation": "search",
"time": "492.415",
"value": "Juan Dies"
},
{
"operation": "Now on",
"time": "493.531",
"value": "Juan Dies"
},
{
"operation": "Highlighting",
"time": "515.131",
"value": "Juan Dies_sentence_2"
},
{
"operation": "Highlighting",
"time": "532.579",
"value": "Juan Dies_sentence_0"
},
{
"operation": "finish",
"time": "572.403",
"value": "finish"
}
]
}
|
exfever_1643
|
Kafka is a mystery film directed by Steven Soderbergh, whose genre is characterized and defined by giving viewers heightened feelings of a state of mental uncertainty, anxiety, being undecided, or being doubtful.
|
{
"evidence": [
"Kafka (film)",
"Thriller (genre)",
"Suspense"
],
"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:\nKafka is a mystery film directed by Steven Soderbergh, whose genre is characterized and defined by giving viewers heightened feelings of a state of mental uncertainty, anxiety, being undecided, or being doubtful.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kafka (film)",
"Thriller (genre)",
"Suspense"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2640,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
71873
|
Born in 1997, Ferdinand Zvonimir von Habsburg, an Austrian motor racing driver and heir apparent to the headship of the House of Habsburg as eldest son of Karl von Habsburg, was the youngest individual in 2017 to be conferred the knightly Order of the Golden Fleece.
|
{
"evidence": [
"Ferdinand Zvonimir von Habsburg_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:\nBorn in 1997, Ferdinand Zvonimir von Habsburg, an Austrian motor racing driver and heir apparent to the headship of the House of Habsburg as eldest son of Karl von Habsburg, was the youngest individual in 2017 to be conferred the knightly Order of the Golden Fleece.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ferdinand Zvonimir von Habsburg_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2641,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.765",
"value": "?search="
},
{
"operation": "search",
"time": "38.079",
"value": "Ferdinand Zvonimir von Habsburg"
},
{
"operation": "Now on",
"time": "39.769",
"value": "Ferdinand Zvonimir von Habsburg"
},
{
"operation": "Highlighting",
"time": "47.612",
"value": "Ferdinand Zvonimir von Habsburg_sentence_0"
},
{
"operation": "hyperlink",
"time": "304.276",
"value": "Karl von Habsburg"
},
{
"operation": "Now on",
"time": "305.422",
"value": "Karl von Habsburg"
},
{
"operation": "back-button-clicked",
"time": "320.899",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "321.714",
"value": "Ferdinand Zvonimir von Habsburg"
},
{
"operation": "search",
"time": "329.976",
"value": "Contains...knightly Order of the Golden Fleece"
},
{
"operation": "Now on",
"time": "332.114",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "362.219",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "363.019",
"value": "Ferdinand Zvonimir von Habsburg"
},
{
"operation": "finish",
"time": "389.378",
"value": "finish"
}
]
}
|
exfever_110
|
The Crown is a historical drama television series that was developed from a play by the Canadian playwright and screenwriter Peter Morgan.
|
{
"evidence": [
"The Crown (TV series)",
"The Audience (2013 play)"
],
"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 Crown is a historical drama television series that was developed from a play by the Canadian playwright and screenwriter Peter Morgan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Crown (TV series)",
"The Audience (2013 play)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2642,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
46597
|
In 1999, Hickey opened her own agency known for offering full-service she worked for another company.
|
{
"evidence": [
"Lisa Hickey_sentence_8",
"Lisa Hickey_sentence_9",
"Lisa Hickey_sentence_10"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn 1999, Hickey opened her own agency known for offering full-service she worked for another company.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Lisa Hickey_sentence_8",
"Lisa Hickey_sentence_9",
"Lisa Hickey_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2643,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "3.247",
"value": "undefined"
},
{
"operation": "search",
"time": "14.556",
"value": "Hickey 1999"
},
{
"operation": "Now on",
"time": "17.228",
"value": "undefined"
},
{
"operation": "search",
"time": "37.213",
"value": "1999 Hickey agency"
},
{
"operation": "Now on",
"time": "40.002",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "56.814",
"value": "Lisa Hickey"
},
{
"operation": "Now on",
"time": "57.662",
"value": "Lisa Hickey"
},
{
"operation": "Highlighting",
"time": "65.601",
"value": "Lisa Hickey_sentence_8"
},
{
"operation": "Highlighting",
"time": "68.885",
"value": "Lisa Hickey_sentence_9"
},
{
"operation": "Highlighting",
"time": "87.313",
"value": "Lisa Hickey_sentence_10"
},
{
"operation": "finish",
"time": "91.734",
"value": "finish"
}
]
}
|
exfever_4068
|
John Lennon was the first English singer to star in an international film.
|
{
"evidence": [
"John Lennon",
"In His Own Write",
"James Thurber"
],
"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 Lennon was the first English singer to star in an international film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John Lennon",
"In His Own Write",
"James Thurber"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2644,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
62766
|
In 1993, Huschka and his cousin formed the hip-hop dup God's Favorite Dog, and they released an album on Sony Music's Epic label.
|
{
"evidence": [
"Hauschka_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:\nIn 1993, Huschka and his cousin formed the hip-hop dup God's Favorite Dog, and they released an album on Sony Music's Epic label.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Hauschka_sentence_10"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2645,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.358",
"value": "?search="
},
{
"operation": "search",
"time": "13.232",
"value": "Contains...God's Favorite Dog"
},
{
"operation": "Now on",
"time": "15.441",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "18.975",
"value": "Hauschka"
},
{
"operation": "Now on",
"time": "20.093",
"value": "Hauschka"
},
{
"operation": "Page search",
"time": "28.507",
"value": "God's Favorite Dog"
},
{
"operation": "Page search",
"time": "48.764",
"value": "God's Favorite Dog"
},
{
"operation": "Highlighting",
"time": "66.194",
"value": "Hauschka_sentence_10"
},
{
"operation": "Highlighting",
"time": "72.67",
"value": "Hauschka_sentence_11"
},
{
"operation": "Highlighting deleted",
"time": "84.336",
"value": "Hauschka_sentence_11"
},
{
"operation": "finish",
"time": "95.903",
"value": "finish"
}
]
}
|
exfever_922
|
Jim Henson was an American puppeteer who achieved worldwide notice as the creator of an American ensemble cast of puppet characters known for an absurdist, burlesque, and self-referential style of variety-sketch comedy.
|
{
"evidence": [
"Jim Henson",
"The Muppets"
],
"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:\nJim Henson was an American puppeteer who achieved worldwide notice as the creator of an American ensemble cast of puppet characters known for an absurdist, burlesque, and self-referential style of variety-sketch comedy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jim Henson",
"The Muppets"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2646,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_170
|
The Strain is the last American horror drama television series created by Guillermo del Toro.
|
{
"evidence": [
"The Strain (TV series)",
"FX (TV channel)",
"Pay television"
],
"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 Strain is the last American horror drama television series created by Guillermo del Toro.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Strain (TV series)",
"FX (TV channel)",
"Pay television"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2647,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3358
|
Tropic Thunder is a 2008 action-comedy film starring a Canadian actor, director, stand-up comedian, and screenwriter known for the 2010 American action-fantasy film produced by Jerry Bruckheimer, directed by Jon Turteltaub, and released by Walt Disney Pictures,
|
{
"evidence": [
"Tropic Thunder",
"Jay Baruchel",
"The Sorcerer's Apprentice (2010 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:\nTropic Thunder is a 2008 action-comedy film starring a Canadian actor, director, stand-up comedian, and screenwriter known for the 2010 American action-fantasy film produced by Jerry Bruckheimer, directed by Jon Turteltaub, and released by Walt Disney Pictures,\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tropic Thunder",
"Jay Baruchel",
"The Sorcerer's Apprentice (2010 film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2648,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21552
|
In 2012, Small Planet Airlines (a Lithuanian leisure airline based at Vilnius Airport) had 20 planes and catered to 635,118 passengers.
|
{
"evidence": [
"Small Planet Airlines_sentence_31"
],
"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 2012, Small Planet Airlines (a Lithuanian leisure airline based at Vilnius Airport) had 20 planes and catered to 635,118 passengers.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Small Planet Airlines_sentence_31"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2649,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.219",
"value": "?search="
},
{
"operation": "search",
"time": "12.038",
"value": "Small Planet Airlines"
},
{
"operation": "Now on",
"time": "13.56",
"value": "Small Planet Airlines"
},
{
"operation": "Highlighting",
"time": "19.789",
"value": "Small Planet Airlines_sentence_1"
},
{
"operation": "Page search",
"time": "47.834",
"value": "plane"
},
{
"operation": "Page search",
"time": "50.87",
"value": "planes"
},
{
"operation": "Page search",
"time": "71.211",
"value": "passenger"
},
{
"operation": "Highlighting",
"time": "122.121",
"value": "Small Planet Airlines_sentence_31"
},
{
"operation": "finish",
"time": "135.048",
"value": "finish"
}
]
}
|
32676
|
Ontario Highway 88, is wrongly believed to have been established in 1938 and goes through two locations, one of which is Bond Head.
|
{
"evidence": [
"Ontario Highway 88_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOntario Highway 88, is wrongly believed to have been established in 1938 and goes through two locations, one of which is Bond Head.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ontario Highway 88_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2650,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.683",
"value": "undefined"
},
{
"operation": "search",
"time": "15.079",
"value": "Ontario Highway 88"
},
{
"operation": "Now on",
"time": "15.949",
"value": "Ontario Highway 88"
},
{
"operation": "Highlighting",
"time": "30.99",
"value": "Ontario Highway 88_sentence_1"
},
{
"operation": "finish",
"time": "42.23",
"value": "finish"
}
]
}
|
exfever_306
|
Nicole Scherzinger rose to fame as the lead singer of a Latvian girl group and dance ensemble.
|
{
"evidence": [
"Nicole Scherzinger",
"The Pussycat Dolls"
],
"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:\nNicole Scherzinger rose to fame as the lead singer of a Latvian girl group and dance ensemble.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nicole Scherzinger",
"The Pussycat Dolls"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2651,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21111
|
Heinrich Ries married two wives, had two sons and was the president of the Geological Society of America in 1929 which was founded in Boulder, Colorado.
|
{
"evidence": [
"Geological Society of America_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHeinrich Ries married two wives, had two sons and was the president of the Geological Society of America in 1929 which was founded in Boulder, Colorado.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Geological Society of America_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2652,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.568",
"value": "?search="
},
{
"operation": "search",
"time": "9.995",
"value": "Heinrich Ries"
},
{
"operation": "Now on",
"time": "10.658",
"value": "Heinrich Ries"
},
{
"operation": "Highlighting",
"time": "53.077",
"value": "Heinrich Ries_sentence_3"
},
{
"operation": "Highlighting",
"time": "53.078",
"value": "Heinrich Ries_title"
},
{
"operation": "Highlighting",
"time": "58.122",
"value": "Heinrich Ries_sentence_5"
},
{
"operation": "Highlighting",
"time": "86.442",
"value": "Heinrich Ries_sentence_4"
},
{
"operation": "Highlighting",
"time": "112.195",
"value": "Heinrich Ries_sentence_6"
},
{
"operation": "hyperlink",
"time": "145.152",
"value": "Geological Society of America"
},
{
"operation": "Now on",
"time": "146.166",
"value": "Geological Society of America"
},
{
"operation": "Highlighting",
"time": "154.246",
"value": "Geological Society of America_sentence_1"
},
{
"operation": "finish",
"time": "193.503",
"value": "finish"
}
]
}
|
exfever_2451
|
Boris Pasternak's 1957 novel Doctor Zhivago is a novel named after its protagonist and title character.
|
{
"evidence": [
"Doctor Zhivago (novel)",
"Yuri Zhivago"
],
"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:\nBoris Pasternak's 1957 novel Doctor Zhivago is a novel named after its protagonist and title character.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Doctor Zhivago (novel)",
"Yuri Zhivago"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2653,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
85279
|
Blues hooker Royce Simmons typified the newfound New South Wales' spirit when early in the opening game he was knocked senseless; he staggered back into position, falling over and regaining his feet then waving trainers away, as he took his place in the defensive formation, before pulling off a try-saving tackle on his own tryline soon after.
|
{
"evidence": [
"Royce Simmons_sentence_6",
"Royce Simmons_sentence_8"
],
"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:\nBlues hooker Royce Simmons typified the newfound New South Wales' spirit when early in the opening game he was knocked senseless; he staggered back into position, falling over and regaining his feet then waving trainers away, as he took his place in the defensive formation, before pulling off a try-saving tackle on his own tryline soon after.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Royce Simmons_sentence_6",
"Royce Simmons_sentence_8"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2654,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.155",
"value": "?search="
},
{
"operation": "search",
"time": "156.878",
"value": "Royce Simmons"
},
{
"operation": "Now on",
"time": "157.799",
"value": "Royce Simmons"
},
{
"operation": "Page search",
"time": "216.325",
"value": "blues"
},
{
"operation": "Highlighting",
"time": "228.679",
"value": "Royce Simmons_sentence_6"
},
{
"operation": "Page search",
"time": "255.277",
"value": "knocked senseless"
},
{
"operation": "Page search",
"time": "271.476",
"value": "knocked"
},
{
"operation": "Page search",
"time": "317.724",
"value": "stagger"
},
{
"operation": "Page search",
"time": "326.132",
"value": "staggered"
},
{
"operation": "search",
"time": "453.721",
"value": "Contains...Royce Simmons"
},
{
"operation": "Now on",
"time": "455.268",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "473.28",
"value": "1991 NSWRL season"
},
{
"operation": "Now on",
"time": "474.21",
"value": "1991 NSWRL season"
},
{
"operation": "Page search",
"time": "495.894",
"value": "staggered back"
},
{
"operation": "back-button-clicked",
"time": "559.884",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "560.443",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "582.822",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "582.825",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "582.87",
"value": "Royce Simmons"
},
{
"operation": "Highlighting",
"time": "592.306",
"value": "Royce Simmons_sentence_8"
},
{
"operation": "finish",
"time": "637.105",
"value": "finish"
}
]
}
|
7204
|
Huka Falls has small waterfalls at the top along with a six meter drop at the end of it.
|
{
"evidence": [
"Huka Falls_sentence_6",
"Huka Falls_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:\nHuka Falls has small waterfalls at the top along with a six meter drop at the end of it.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Huka Falls_sentence_6",
"Huka Falls_sentence_7"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2655,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.702",
"value": "?search="
},
{
"operation": "search",
"time": "13.461",
"value": "Huka Falls"
},
{
"operation": "Now on",
"time": "14.471",
"value": "Huka Falls"
},
{
"operation": "Highlighting",
"time": "33.543",
"value": "Huka Falls_sentence_0"
},
{
"operation": "Highlighting",
"time": "52.802",
"value": "Huka Falls_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "61.456",
"value": "Huka Falls_sentence_0"
},
{
"operation": "Highlighting",
"time": "68.124",
"value": "Huka Falls_sentence_7"
},
{
"operation": "finish",
"time": "82.717",
"value": "finish"
}
]
}
|
75279
|
The Indonesian army aviation started in 1959 when the doctrine of Heliborne Troops or Air Mobile was being adopted in other countries, such as in Sarawak (a state in Malaysia located in northwest Kalimantan) the British SAS launched an air assault operation with Westland Wessex Mk.
|
{
"evidence": [
"Indonesian Army Aviation Command_sentence_3",
"Indonesian Army Aviation Command_sentence_4"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Indonesian army aviation started in 1959 when the doctrine of Heliborne Troops or Air Mobile was being adopted in other countries, such as in Sarawak (a state in Malaysia located in northwest Kalimantan) the British SAS launched an air assault operation with Westland Wessex Mk.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Indonesian Army Aviation Command_sentence_3",
"Indonesian Army Aviation Command_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2656,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.709",
"value": "?search="
},
{
"operation": "search",
"time": "46.329",
"value": "Indonesian army aviation"
},
{
"operation": "Now on",
"time": "52.325",
"value": "Indonesian Army Aviation"
},
{
"operation": "Page search",
"time": "126.072",
"value": "1959"
},
{
"operation": "Page search",
"time": "143.932",
"value": "heliborne"
},
{
"operation": "search",
"time": "212.319",
"value": "Heliborne Troops"
},
{
"operation": "Now on",
"time": "218.551",
"value": "undefined"
},
{
"operation": "search",
"time": "280.15",
"value": "Air Mobile was being adopted in other countries"
},
{
"operation": "Now on",
"time": "283.517",
"value": "undefined"
},
{
"operation": "search",
"time": "311.947",
"value": "British SAS launched an air assault operation with Westland Wessex Mk"
},
{
"operation": "Now on",
"time": "313.904",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "319.357",
"value": "Indonesian Army Aviation Command"
},
{
"operation": "Now on",
"time": "320.159",
"value": "Indonesian Army Aviation Command"
},
{
"operation": "Highlighting",
"time": "328.765",
"value": "Indonesian Army Aviation Command_sentence_3"
},
{
"operation": "Highlighting",
"time": "338.635",
"value": "Indonesian Army Aviation Command_sentence_4"
},
{
"operation": "finish",
"time": "387.332",
"value": "finish"
}
]
}
|
39627
|
During the 1931 General election in Aberdeen North Electorate, William Wedgwood Benn won fewer votes than John George Burnett who was born in 1876.
|
{
"evidence": [
"William Wedgwood Benn, 1st Viscount Stansgate_sentence_15",
"John George Burnett_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:\nDuring the 1931 General election in Aberdeen North Electorate, William Wedgwood Benn won fewer votes than John George Burnett who was born in 1876.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"William Wedgwood Benn, 1st Viscount Stansgate_sentence_15",
"John George Burnett_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2657,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.654",
"value": "?search="
},
{
"operation": "search",
"time": "18.357",
"value": "1931 General election in Aberdeen North Electorate"
},
{
"operation": "Now on",
"time": "20.356",
"value": "undefined"
},
{
"operation": "search",
"time": "40.652",
"value": "General election in Aberdeen North Electorate"
},
{
"operation": "Now on",
"time": "42.059",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "48.496",
"value": "general election in aberdeen north electoral"
},
{
"operation": "Now on",
"time": "49.863",
"value": "undefined"
},
{
"operation": "search",
"time": "63.292",
"value": "William Wedgwood Benn"
},
{
"operation": "Now on",
"time": "64.907",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "66.796",
"value": "William Wedgwood Benn"
},
{
"operation": "Now on",
"time": "67.998",
"value": "William Wedgwood Benn, 1st Viscount Stansgate"
},
{
"operation": "Page search",
"time": "72.818",
"value": "election"
},
{
"operation": "Page search",
"time": "73.047",
"value": "election"
},
{
"operation": "Highlighting",
"time": "82.722",
"value": "William Wedgwood Benn, 1st Viscount Stansgate_sentence_15"
},
{
"operation": "hyperlink",
"time": "101.699",
"value": "John George Burnett"
},
{
"operation": "Now on",
"time": "102.889",
"value": "John George Burnett"
},
{
"operation": "Highlighting",
"time": "106.021",
"value": "John George Burnett_sentence_1"
},
{
"operation": "finish",
"time": "115.191",
"value": "finish"
}
]
}
|
45655
|
Tom Mees was an American sportscaster who died in an accidental drowning ate age 46.
|
{
"evidence": [
"Tom Mees_sentence_1",
"Tom Mees_sentence_16",
"Tom Mees_sentence_17"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTom Mees was an American sportscaster who died in an accidental drowning ate age 46.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Tom Mees_sentence_1",
"Tom Mees_sentence_16",
"Tom Mees_sentence_17"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2658,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.058",
"value": "?search="
},
{
"operation": "search",
"time": "9.435",
"value": "Tom Mees"
},
{
"operation": "Now on",
"time": "10.419",
"value": "Tom Mees"
},
{
"operation": "Highlighting",
"time": "12.813",
"value": "Tom Mees_sentence_1"
},
{
"operation": "Highlighting",
"time": "35.227",
"value": "Tom Mees_sentence_16"
},
{
"operation": "Highlighting deleted",
"time": "37.015",
"value": "Tom Mees_sentence_16"
},
{
"operation": "Highlighting",
"time": "37.548",
"value": "Tom Mees_sentence_16"
},
{
"operation": "Highlighting",
"time": "38.4",
"value": "Tom Mees_sentence_17"
},
{
"operation": "finish",
"time": "171.708",
"value": "finish"
}
]
}
|
37651
|
Claude Turmes (of America) was born in the USA in the 40s.
|
{
"evidence": [
"Claude Turmes_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:\nClaude Turmes (of America) was born in the USA in the 40s.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Claude Turmes_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2659,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.745",
"value": "?search="
},
{
"operation": "search",
"time": "236.443",
"value": "Claude Turmes"
},
{
"operation": "Now on",
"time": "237.479",
"value": "Claude Turmes"
},
{
"operation": "Highlighting",
"time": "248.235",
"value": "Claude Turmes_sentence_0"
},
{
"operation": "finish",
"time": "256.261",
"value": "finish"
}
]
}
|
exfever_4143
|
The Maverick boys consist of Bret, Bart, Beau and Brent.
|
{
"evidence": [
"Roger Moore",
"Maverick (TV series)"
],
"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 Maverick boys consist of Bret, Bart, Beau and Brent.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Roger Moore",
"Maverick (TV series)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2660,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5912
|
Chris Hardwick is an American comedian who is affiliated with the network's zombie drama series, an American post-apocalyptic horror drama for an American basic cable television channel that is the flagship property.
|
{
"evidence": [
"Chris Hardwick",
"Fear the Walking Dead",
"AMC (TV channel)"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nChris Hardwick is an American comedian who is affiliated with the network's zombie drama series, an American post-apocalyptic horror drama for an American basic cable television channel that is the flagship property.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Chris Hardwick",
"Fear the Walking Dead",
"AMC (TV channel)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2661,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
78207
|
After graduating from high school, Brumbaugh attended University of Florida in Columbus, Ohio where he played for the Florida Gators football team, then was later inducted into the University of Florida Athletic Hall of Fame as a "Gator Great."
|
{
"evidence": [
"Carl Brumbaugh_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:\nAfter graduating from high school, Brumbaugh attended University of Florida in Columbus, Ohio where he played for the Florida Gators football team, then was later inducted into the University of Florida Athletic Hall of Fame as a \"Gator Great.\"\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Carl Brumbaugh_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2662,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.798",
"value": "?search="
},
{
"operation": "search",
"time": "37.615",
"value": "Brumbaugh"
},
{
"operation": "Now on",
"time": "38.722",
"value": "Brumbaugh"
},
{
"operation": "hyperlink",
"time": "44.8",
"value": "Carl Brumbaugh"
},
{
"operation": "Now on",
"time": "45.33",
"value": "Carl Brumbaugh"
},
{
"operation": "back-button-clicked",
"time": "67.257",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "67.335",
"value": "Brumbaugh"
},
{
"operation": "search",
"time": "107.67",
"value": "Brumbaugh gators"
},
{
"operation": "Now on",
"time": "109.225",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "145.734",
"value": "1929 Florida Gators football team"
},
{
"operation": "Now on",
"time": "146.671",
"value": "1929 Florida Gators football team"
},
{
"operation": "Page search",
"time": "184.122",
"value": "Brumbaugh"
},
{
"operation": "hyperlink",
"time": "191.23",
"value": "Carl Brumbaugh"
},
{
"operation": "Now on",
"time": "191.636",
"value": "Carl Brumbaugh"
},
{
"operation": "Highlighting",
"time": "222.155",
"value": "Carl Brumbaugh_sentence_3"
},
{
"operation": "finish",
"time": "238.588",
"value": "finish"
}
]
}
|
71473
|
Ludwig Adolf Wilhelm von Lützow (18 May 1782 – 6 December 1834) came from the House of Mecklenburg and first entered the Prussian Army in 1785, and eleven years later as a lieutenant took part in the disastrous battle of Auerstadt.
|
{
"evidence": [
"Ludwig Adolf Wilhelm von Lützow_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:\nLudwig Adolf Wilhelm von Lützow (18 May 1782 – 6 December 1834) came from the House of Mecklenburg and first entered the Prussian Army in 1785, and eleven years later as a lieutenant took part in the disastrous battle of Auerstadt.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ludwig Adolf Wilhelm von Lützow_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2663,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.086",
"value": "?search="
},
{
"operation": "search",
"time": "8.918",
"value": "Ludwig Adolf Wilhelm von Lützow"
},
{
"operation": "Now on",
"time": "10.325",
"value": "Ludwig Adolf Wilhelm von Lützow"
},
{
"operation": "Highlighting",
"time": "136.73",
"value": "Ludwig Adolf Wilhelm von Lützow_sentence_2"
},
{
"operation": "Highlighting",
"time": "146.252",
"value": "Ludwig Adolf Wilhelm von Lützow_sentence_4"
},
{
"operation": "Highlighting",
"time": "150.545",
"value": "Ludwig Adolf Wilhelm von Lützow_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "157.531",
"value": "Ludwig Adolf Wilhelm von Lützow_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "159.496",
"value": "Ludwig Adolf Wilhelm von Lützow_sentence_2"
},
{
"operation": "finish",
"time": "164.518",
"value": "finish"
}
]
}
|
exfever_7318
|
Northwestern University is one of Illinois' top universities.
|
{
"evidence": [
"Northwestern University",
"Northwest Territory"
],
"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:\nNorthwestern University is one of Illinois' top universities.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Northwestern University",
"Northwest Territory"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2664,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5240
|
Alessia Cara is a British singer whose debut single was the lead single for her musical recording that contains more tracks than a single but fewer than an album.
|
{
"evidence": [
"Alessia Cara",
"Here (Alessia Cara song)",
"Extended play"
],
"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:\nAlessia Cara is a British singer whose debut single was the lead single for her musical recording that contains more tracks than a single but fewer than an album.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alessia Cara",
"Here (Alessia Cara song)",
"Extended play"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2665,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1336
|
The Boeing 737 is a wide-body aircraft with airbreathing jet engines used in the aircraft propulsion and a blend of words in which parts of multiple words are combined to create a new word as smog, coined by blending smoke and fog, or motel, from motor and hotel.
|
{
"evidence": [
"Boeing 737",
"Turbofan",
"Portmanteau"
],
"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 Boeing 737 is a wide-body aircraft with airbreathing jet engines used in the aircraft propulsion and a blend of words in which parts of multiple words are combined to create a new word as smog, coined by blending smoke and fog, or motel, from motor and hotel.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Boeing 737",
"Turbofan",
"Portmanteau"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2666,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2052
|
Pompeii was an ancient city located in what is now the comune of Pompei, near Naples in region of Italy, which has a population of around 5,820,000 people making it Italy's third most populous region.
|
{
"evidence": [
"Pompeii",
"Campania"
],
"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:\nPompeii was an ancient city located in what is now the comune of Pompei, near Naples in region of Italy, which has a population of around 5,820,000 people making it Italy's third most populous region.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Pompeii",
"Campania"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2667,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
26606
|
Karl Casper, born on May 22, 1893, was a Generalleutnant in the Army of Nazi Germany and was awarded the Knight's Cross of the Iron Cross.
|
{
"evidence": [
"Carl Casper_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:\nKarl Casper, born on May 22, 1893, was a Generalleutnant in the Army of Nazi Germany and was awarded the Knight's Cross of the Iron Cross.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Carl Casper_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2668,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.119",
"value": "?search="
},
{
"operation": "search",
"time": "24.977",
"value": "Karl Casper"
},
{
"operation": "Now on",
"time": "26.027",
"value": "Carl Casper"
},
{
"operation": "Highlighting",
"time": "37.276",
"value": "Carl Casper_sentence_0"
},
{
"operation": "finish",
"time": "74.799",
"value": "finish"
}
]
}
|
88904
|
Before the Bavarian Forest Nature Park and Bavarian Forest National Park, the Bavarian Forest Club was the only institution that placed the culture and nature of the Bavarian Forest at the centre of activties, with its goal of development of the town for hiking and holiday, and marking and maintaining the hiking trails.
|
{
"evidence": [
"Bavarian Forest Club_sentence_14",
"Bavarian Forest Club_sentence_12",
"Bavarian Forest Club_sentence_13"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBefore the Bavarian Forest Nature Park and Bavarian Forest National Park, the Bavarian Forest Club was the only institution that placed the culture and nature of the Bavarian Forest at the centre of activties, with its goal of development of the town for hiking and holiday, and marking and maintaining the hiking trails.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Bavarian Forest Club_sentence_14",
"Bavarian Forest Club_sentence_12",
"Bavarian Forest Club_sentence_13"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2669,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.954",
"value": "?search="
},
{
"operation": "search",
"time": "20.975",
"value": "Contains...Before the Bavarian Forest Nature Park"
},
{
"operation": "Now on",
"time": "23.455",
"value": "undefined"
},
{
"operation": "search",
"time": "48.4",
"value": "Contains...the Bavarian Forest Club was the only institution"
},
{
"operation": "Now on",
"time": "50.323",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "55.576",
"value": "Bavarian Forest Club"
},
{
"operation": "Now on",
"time": "56.374",
"value": "Bavarian Forest Club"
},
{
"operation": "Highlighting",
"time": "76.231",
"value": "Bavarian Forest Club_sentence_14"
},
{
"operation": "Highlighting",
"time": "100.215",
"value": "Bavarian Forest Club_sentence_12"
},
{
"operation": "Highlighting",
"time": "117.509",
"value": "Bavarian Forest Club_sentence_13"
},
{
"operation": "finish",
"time": "134.275",
"value": "finish"
}
]
}
|
58816
|
Labor Party representative Merv Hunter was the member for Lake Macquarie from 1969 to 1991.
|
{
"evidence": [
"Merv Hunter_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:\nLabor Party representative Merv Hunter was the member for Lake Macquarie from 1969 to 1991.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Merv Hunter_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2670,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.812",
"value": "?search="
},
{
"operation": "search",
"time": "6.499",
"value": "Merv Hunter"
},
{
"operation": "Now on",
"time": "7.409",
"value": "Merv Hunter"
},
{
"operation": "Highlighting",
"time": "112.971",
"value": "Merv Hunter_cell_0_8_1"
},
{
"operation": "Highlighting",
"time": "112.986",
"value": "Merv Hunter_header_cell_0_8_0"
},
{
"operation": "Highlighting",
"time": "112.987",
"value": "Merv Hunter_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "117.386",
"value": "Merv Hunter_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "118.965",
"value": "Merv Hunter_cell_0_8_1"
},
{
"operation": "Highlighting deleted",
"time": "118.983",
"value": "Merv Hunter_header_cell_0_8_0"
},
{
"operation": "Highlighting deleted",
"time": "118.987",
"value": "Merv Hunter_header_cell_0_4_0"
},
{
"operation": "finish",
"time": "132.007",
"value": "finish"
}
]
}
|
exfever_1934
|
Nerrukku Ner is a 1997 Indian Tamil-language action rom-com film starring Suriya and an Indian actor, dancer, playback singer, and philanthropist who works predominantly in Tamil cinema.
|
{
"evidence": [
"Nerrukku Ner",
"Vijay (actor)"
],
"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:\nNerrukku Ner is a 1997 Indian Tamil-language action rom-com film starring Suriya and an Indian actor, dancer, playback singer, and philanthropist who works predominantly in Tamil cinema.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nerrukku Ner",
"Vijay (actor)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2671,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
10910
|
Frederick Nolting was appointed in 1955 before being appointed again in 1957.
|
{
"evidence": [
"Frederick Nolting_sentence_7",
"Frederick Nolting_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:\nFrederick Nolting was appointed in 1955 before being appointed again in 1957.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Frederick Nolting_sentence_7",
"Frederick Nolting_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2672,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.512",
"value": "?search="
},
{
"operation": "search",
"time": "9.728",
"value": "Frederick Nolting"
},
{
"operation": "Now on",
"time": "10.43",
"value": "Frederick Nolting"
},
{
"operation": "Highlighting",
"time": "24.466",
"value": "Frederick Nolting_sentence_7"
},
{
"operation": "Highlighting",
"time": "25.607",
"value": "Frederick Nolting_sentence_8"
},
{
"operation": "finish",
"time": "44.788",
"value": "finish"
}
]
}
|
exfever_4956
|
Oz the Great and Powerful is a 2013 American fantasy adventure film based on an American author's early 20th century Oz novels.
|
{
"evidence": [
"Oz the Great and Powerful",
"L. Frank Baum"
],
"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:\nOz the Great and Powerful is a 2013 American fantasy adventure film based on an American author's early 20th century Oz novels.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Oz the Great and Powerful",
"L. Frank Baum"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2673,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
62466
|
Carlos Amezcua served as a co-anchor at KUSI-TV San Diego from 2015 to 2019 and is currently the co-host of the Carlos & Lisa show on BEONDTV, which aims to create a platform that delivers compelling and entertaining content worldwide.
|
{
"evidence": [
"Carlos Amezcua_sentence_4",
"Carlos Amezcua_sentence_5"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCarlos Amezcua served as a co-anchor at KUSI-TV San Diego from 2015 to 2019 and is currently the co-host of the Carlos & Lisa show on BEONDTV, which aims to create a platform that delivers compelling and entertaining content worldwide.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Carlos Amezcua_sentence_4",
"Carlos Amezcua_sentence_5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2674,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.673",
"value": "?search="
},
{
"operation": "search",
"time": "16.098",
"value": "Carlos Amezcua"
},
{
"operation": "Now on",
"time": "16.623",
"value": "Carlos Amezcua"
},
{
"operation": "Highlighting",
"time": "79.577",
"value": "Carlos Amezcua_sentence_4"
},
{
"operation": "Highlighting",
"time": "90.304",
"value": "Carlos Amezcua_sentence_5"
},
{
"operation": "search",
"time": "197.057",
"value": "BEONDTV"
},
{
"operation": "Now on",
"time": "197.591",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "202.032",
"value": "Carlos Amezcua"
},
{
"operation": "Now on",
"time": "202.26",
"value": "Carlos Amezcua"
},
{
"operation": "search",
"time": "260.577",
"value": "which aims to create a platform that delivers compelling and entertaining content worldwide"
},
{
"operation": "Now on",
"time": "262.925",
"value": "undefined"
},
{
"operation": "finish",
"time": "365.686",
"value": "finish"
}
]
}
|
exfever_7580
|
Jerry Lewis was an American comedian, actor, singer, director, producer, writer, and humanitarian who made his debut in 1946 as part of the famous duo with an American singer, actor, and comedian of Ivorian descent.
|
{
"evidence": [
"Jerry Lewis",
"Dean Martin"
],
"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:\nJerry Lewis was an American comedian, actor, singer, director, producer, writer, and humanitarian who made his debut in 1946 as part of the famous duo with an American singer, actor, and comedian of Ivorian descent.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jerry Lewis",
"Dean Martin"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2675,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3508
|
My Bloody Valentine 3D serves as a remake of the 1981 Canadian slasher film of the same name, starring an American actress and model.
|
{
"evidence": [
"My Bloody Valentine 3D",
"Jaime King"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMy Bloody Valentine 3D serves as a remake of the 1981 Canadian slasher film of the same name, starring an American actress and model.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"My Bloody Valentine 3D",
"Jaime King"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2676,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3110
|
Odin is one of the oldest gods in Germanic mythology.
|
{
"evidence": [
"Odin",
"Old English",
"Anglo-Norman 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:\nOdin is one of the oldest gods in Germanic mythology.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Odin",
"Old English",
"Anglo-Norman language"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2677,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8715
|
Shraddha Kapoor had her first leading role in a 2011 Hindi comedy-drama film directed by Bumpy.
|
{
"evidence": [
"Shraddha Kapoor",
"Luv Ka The End"
],
"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:\nShraddha Kapoor had her first leading role in a 2011 Hindi comedy-drama film directed by Bumpy.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Shraddha Kapoor",
"Luv Ka The End"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2678,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3671
|
Katherine Marie Heigl is an American actress, producer and former fashion model who appeared in a 1994 French-American comedy-drama directed by Steve Miner.
|
{
"evidence": [
"Katherine Heigl",
"My Father the Hero (1994 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:\nKatherine Marie Heigl is an American actress, producer and former fashion model who appeared in a 1994 French-American comedy-drama directed by Steve Miner.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Katherine Heigl",
"My Father the Hero (1994 film)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2679,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3047
|
The Murder of Roger Ackroyd is a work of detective fiction that is the third novel to feature a fictional Belgian detective that appears in a play produced initially in 1940.
|
{
"evidence": [
"The Murder of Roger Ackroyd",
"Hercule Poirot",
"Black Coffee (play)"
],
"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 Murder of Roger Ackroyd is a work of detective fiction that is the third novel to feature a fictional Belgian detective that appears in a play produced initially in 1940.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Murder of Roger Ackroyd",
"Hercule Poirot",
"Black Coffee (play)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2680,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3956
|
SoCal is a geographic and cultural region that generally comprises the southern portion of a state in the Western United States.
|
{
"evidence": [
"Southern California",
"California"
],
"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:\nSoCal is a geographic and cultural region that generally comprises the southern portion of a state in the Western United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Southern California",
"California"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2681,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3912
|
Tom Hiddleston is an English actor who has appeared in a 2017 American superhero film, produced by Marvel Studios.
|
{
"evidence": [
"Tom Hiddleston",
"Thor: Ragnarok"
],
"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:\nTom Hiddleston is an English actor who has appeared in a 2017 American superhero film, produced by Marvel Studios.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tom Hiddleston",
"Thor: Ragnarok"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2682,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6651
|
All My Children is an American television soap opera, created by Will Smith, set in Pine Valley, Pennsylvania, a fictional suburb of the largest city in the Commonwealth of Pennsylvania in the United States.
|
{
"evidence": [
"All My Children",
"Philadelphia"
],
"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:\nAll My Children is an American television soap opera, created by Will Smith, set in Pine Valley, Pennsylvania, a fictional suburb of the largest city in the Commonwealth of Pennsylvania in the United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"All My Children",
"Philadelphia"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2683,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.