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 |
|---|---|---|---|---|---|---|---|---|
31784
|
Johnson and Clark batted well before Clark got out leaving Australia with a healthy score of 343, then Dravid and Jaffer from the Indian team were sent into bat and batted very slowly before losing early wickets.
|
{
"evidence": [
"Indian cricket team in Australia in 2007–08_sentence_8",
"Indian cricket team in Australia in 2007–08_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:\nJohnson and Clark batted well before Clark got out leaving Australia with a healthy score of 343, then Dravid and Jaffer from the Indian team were sent into bat and batted very slowly before losing early wickets.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Indian cricket team in Australia in 2007–08_sentence_8",
"Indian cricket team in Australia in 2007–08_sentence_9"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1084,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.112",
"value": "?search="
},
{
"operation": "search",
"time": "24.173",
"value": "Contains...leaving Australia with a healthy score of 343"
},
{
"operation": "Now on",
"time": "26.346",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "30.793",
"value": "Indian cricket team in Australia in 2007–08"
},
{
"operation": "Now on",
"time": "31.555",
"value": "Indian cricket team in Australia in 2007–08"
},
{
"operation": "Highlighting",
"time": "43.491",
"value": "Indian cricket team in Australia in 2007–08_sentence_8"
},
{
"operation": "Highlighting",
"time": "52.897",
"value": "Indian cricket team in Australia in 2007–08_sentence_9"
},
{
"operation": "finish",
"time": "80.313",
"value": "finish"
}
]
}
|
55475
|
Laila Palau, a professional Spanish basketball player who is the top medalist, played 447 games during 1997.
|
{
"evidence": [
"Laia Palau_sentence_1",
"Laia Palau_sentence_4",
"Laia Palau_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLaila Palau, a professional Spanish basketball player who is the top medalist, played 447 games during 1997.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Laia Palau_sentence_1",
"Laia Palau_sentence_4",
"Laia Palau_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1085,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.647",
"value": "?search="
},
{
"operation": "search",
"time": "9.38",
"value": "Laila Palau"
},
{
"operation": "Now on",
"time": "10.721",
"value": "undefined"
},
{
"operation": "search",
"time": "29.092",
"value": "Laia Palau"
},
{
"operation": "Now on",
"time": "29.972",
"value": "Laia Palau"
},
{
"operation": "Highlighting",
"time": "33.556",
"value": "Laia Palau_sentence_1"
},
{
"operation": "Highlighting",
"time": "37.478",
"value": "Laia Palau_sentence_3"
},
{
"operation": "Highlighting",
"time": "66.29",
"value": "Laia Palau_cell_1_5_2"
},
{
"operation": "Highlighting",
"time": "66.292",
"value": "Laia Palau_section_1"
},
{
"operation": "Highlighting",
"time": "66.293",
"value": "Laia Palau_header_cell_1_0_2"
},
{
"operation": "Highlighting deleted",
"time": "66.863",
"value": "Laia Palau_cell_1_5_2"
},
{
"operation": "Highlighting deleted",
"time": "66.867",
"value": "Laia Palau_section_1"
},
{
"operation": "Highlighting deleted",
"time": "66.869",
"value": "Laia Palau_header_cell_1_0_2"
},
{
"operation": "Highlighting",
"time": "103.773",
"value": "Laia Palau_sentence_4"
},
{
"operation": "hyperlink",
"time": "122.859",
"value": "Spain women's national basketball team"
},
{
"operation": "Now on",
"time": "125.072",
"value": "Spain women's national basketball team"
},
{
"operation": "back-button-clicked",
"time": "201.124",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "201.206",
"value": "Laia Palau"
},
{
"operation": "Highlighting deleted",
"time": "203.651",
"value": "Laia Palau_sentence_3"
},
{
"operation": "Highlighting",
"time": "207.036",
"value": "Laia Palau_sentence_2"
},
{
"operation": "finish",
"time": "216.958",
"value": "finish"
}
]
}
|
exfever_1839
|
The Chair is the second series written and co-executive produced by Amanda Peet.
|
{
"evidence": [
"Amanda Peet",
"Changing Lanes"
],
"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 Chair is the second series written and co-executive produced by Amanda Peet.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Amanda Peet",
"Changing Lanes"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1086,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
91756
|
After graduating from the Szent László Gimnázium in 1978, Sándor Pörzse became a professional footballer and television news announcer.
|
{
"evidence": [
"Sándor Pörzse_sentence_10"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAfter graduating from the Szent László Gimnázium in 1978, Sándor Pörzse became a professional footballer and television news announcer.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sándor Pörzse_sentence_10"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1087,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.424",
"value": "?search="
},
{
"operation": "search",
"time": "79.116",
"value": "Sándor Pörzse"
},
{
"operation": "Now on",
"time": "80.105",
"value": "Sándor Pörzse"
},
{
"operation": "Highlighting",
"time": "220.515",
"value": "Sándor Pörzse_sentence_4"
},
{
"operation": "Highlighting",
"time": "236.745",
"value": "Sándor Pörzse_sentence_6"
},
{
"operation": "Highlighting",
"time": "326.634",
"value": "Sándor Pörzse_sentence_10"
},
{
"operation": "hyperlink",
"time": "563.049",
"value": "TV3"
},
{
"operation": "Now on",
"time": "563.973",
"value": "TV3 (Hungary)"
},
{
"operation": "back-button-clicked",
"time": "610.921",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "611.11",
"value": "Sándor Pörzse"
},
{
"operation": "Highlighting deleted",
"time": "615.568",
"value": "Sándor Pörzse_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "616.536",
"value": "Sándor Pörzse_sentence_4"
},
{
"operation": "finish",
"time": "646.585",
"value": "finish"
}
]
}
|
exfever_6783
|
Nicolas Cage is an actor and filmmaker who is also a musician.
|
{
"evidence": [
"Nicolas Cage",
"Peggy Sue Got Married"
],
"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:\nNicolas Cage is an actor and filmmaker who is also a musician.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Nicolas Cage",
"Peggy Sue Got Married"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1088,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
92534
|
Regis Canevin became a pastor of the mission in Canonsburg and later on served as chancellor of the Diocese of Pittsburgh—a Roman Catholic diocese with 188 parishes and 225 churches—which was before he became a pastor in Crafton.
|
{
"evidence": [
"Regis Canevin_sentence_10",
"Regis Canevin_sentence_11",
"Roman Catholic Diocese of Pittsburgh_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:\nRegis Canevin became a pastor of the mission in Canonsburg and later on served as chancellor of the Diocese of Pittsburgh—a Roman Catholic diocese with 188 parishes and 225 churches—which was before he became a pastor in Crafton.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Regis Canevin_sentence_10",
"Regis Canevin_sentence_11",
"Roman Catholic Diocese of Pittsburgh_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1089,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.01",
"value": "?search="
},
{
"operation": "search",
"time": "16.236",
"value": "Regis Canevin"
},
{
"operation": "Now on",
"time": "16.93",
"value": "Regis Canevin"
},
{
"operation": "Highlighting",
"time": "38.998",
"value": "Regis Canevin_sentence_10"
},
{
"operation": "Highlighting",
"time": "42.363",
"value": "Regis Canevin_sentence_11"
},
{
"operation": "hyperlink",
"time": "62.609",
"value": "Diocese of Pittsburgh"
},
{
"operation": "Now on",
"time": "63.932",
"value": "Roman Catholic Diocese of Pittsburgh"
},
{
"operation": "Highlighting",
"time": "66.035",
"value": "Roman Catholic Diocese of Pittsburgh_sentence_1"
},
{
"operation": "finish",
"time": "74.07",
"value": "finish"
}
]
}
|
59414
|
Living Marxism was a British magazine launched in 1988 that went on to critize how the media portrayed the civil wars in Rwanda and Bosnia.
|
{
"evidence": [
"Living Marxism_sentence_3",
"Living Marxism_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:\nLiving Marxism was a British magazine launched in 1988 that went on to critize how the media portrayed the civil wars in Rwanda and Bosnia.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Living Marxism_sentence_3",
"Living Marxism_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1090,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.634",
"value": "?search="
},
{
"operation": "search",
"time": "12.255",
"value": "Living Marxism"
},
{
"operation": "Now on",
"time": "13.041",
"value": "Living Marxism"
},
{
"operation": "Highlighting",
"time": "16.529",
"value": "Living Marxism_sentence_3"
},
{
"operation": "Highlighting",
"time": "31.64",
"value": "Living Marxism_sentence_10"
},
{
"operation": "Now on",
"time": "73.121",
"value": "Living Marxism"
},
{
"operation": "finish",
"time": "91.717",
"value": "finish"
}
]
}
|
exfever_8691
|
Tupac Shakur was a rapper and actor who also wrote haiku poetry.?
|
{
"evidence": [
"Tupac Shakur",
"West Coast hip hop"
],
"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:\nTupac Shakur was a rapper and actor who also wrote haiku poetry.?\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tupac Shakur",
"West Coast hip hop"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1091,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2815
|
Judith Therese Evans is the most famous American actress and comedian.
|
{
"evidence": [
"Judy Greer",
"Jawbreaker (film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nJudith Therese Evans is the most famous American actress and comedian.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Judy Greer",
"Jawbreaker (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1092,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
70879
|
CrossLink International was founded in November 1996, it was inspired when a Baptist Sunday school class visited a hospital in Moscow which was established in 1147.
|
{
"evidence": [
"CrossLink International_sentence_4",
"CrossLink International_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:\nCrossLink International was founded in November 1996, it was inspired when a Baptist Sunday school class visited a hospital in Moscow which was established in 1147.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"CrossLink International_sentence_4",
"CrossLink International_sentence_5"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1093,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.388",
"value": "?search="
},
{
"operation": "search",
"time": "15.335",
"value": "CrossLink International"
},
{
"operation": "Now on",
"time": "16.761",
"value": "CrossLink International"
},
{
"operation": "Highlighting",
"time": "24.137",
"value": "CrossLink International_sentence_4"
},
{
"operation": "Highlighting",
"time": "29.013",
"value": "CrossLink International_sentence_5"
},
{
"operation": "Page search",
"time": "70.659",
"value": "1147"
},
{
"operation": "finish",
"time": "88.68",
"value": "finish"
}
]
}
|
exfever_2405
|
The Middle East is the first continent in the world to develop civilizations and build large communities.
|
{
"evidence": [
"Assyrian people",
"Ethnic group"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Middle East is the first continent in the world to develop civilizations and build large communities.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Assyrian people",
"Ethnic group"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1094,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
66553
|
Athani Taluk's commercial crops include sugarcane, making it one of the largest sugarcane cultivation taluks in India.
|
{
"evidence": [
"Athani Taluk_sentence_15",
"Athani Taluk_sentence_0",
"Athani Taluk_sentence_14"
],
"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:\nAthani Taluk's commercial crops include sugarcane, making it one of the largest sugarcane cultivation taluks in India.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Athani Taluk_sentence_15",
"Athani Taluk_sentence_0",
"Athani Taluk_sentence_14"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1095,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.227",
"value": "?search="
},
{
"operation": "search",
"time": "13.805",
"value": "Athani Taluk"
},
{
"operation": "Now on",
"time": "14.746",
"value": "Athani Taluk"
},
{
"operation": "Highlighting",
"time": "25.739",
"value": "Athani Taluk_sentence_14"
},
{
"operation": "Highlighting",
"time": "33.11",
"value": "Athani Taluk_sentence_15"
},
{
"operation": "Highlighting",
"time": "94.322",
"value": "Athani Taluk_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "98.985",
"value": "Athani Taluk_sentence_0"
},
{
"operation": "Highlighting",
"time": "99.735",
"value": "Athani Taluk_sentence_0"
},
{
"operation": "Highlighting",
"time": "287.969",
"value": "Athani Taluk_sentence_14"
},
{
"operation": "finish",
"time": "299.576",
"value": "finish"
}
]
}
|
38590
|
Music engraving, the art of drawing music notation at high quality, communicates several different types of information simultaneously with no precision.
|
{
"evidence": [
"Music engraving_sentence_6",
"Music engraving_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:\nMusic engraving, the art of drawing music notation at high quality, communicates several different types of information simultaneously with no precision.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Music engraving_sentence_6",
"Music engraving_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1096,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.104",
"value": "?search="
},
{
"operation": "search",
"time": "13.09",
"value": "Music engraving"
},
{
"operation": "Now on",
"time": "14.212",
"value": "Music engraving"
},
{
"operation": "Highlighting",
"time": "28.301",
"value": "Music engraving_sentence_0"
},
{
"operation": "Highlighting",
"time": "51.576",
"value": "Music engraving_sentence_6"
},
{
"operation": "Highlighting",
"time": "52.803",
"value": "Music engraving_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "61.465",
"value": "Music engraving_sentence_0"
},
{
"operation": "finish",
"time": "67.992",
"value": "finish"
}
]
}
|
49683
|
The Mayor of Turin solely accountable for the government.
|
{
"evidence": [
"List of mayors of Turin_sentence_0"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Mayor of Turin solely accountable for the government.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"List of mayors of Turin_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1097,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.618",
"value": "?search="
},
{
"operation": "search",
"time": "16.281",
"value": "Mayor of Turin"
},
{
"operation": "Now on",
"time": "19.694",
"value": "List of mayors of Turin"
},
{
"operation": "Highlighting",
"time": "24.543",
"value": "List of mayors of Turin_sentence_0"
},
{
"operation": "finish",
"time": "38.659",
"value": "finish"
}
]
}
|
exfever_2265
|
Saskatchewan is located in the middle of the three prairie sub-national administrative divisions within the geographical areas of the second-largest country in South America.
|
{
"evidence": [
"Saskatchewan",
"Provinces and territories of Canada",
"Canada"
],
"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:\nSaskatchewan is located in the middle of the three prairie sub-national administrative divisions within the geographical areas of the second-largest country in South America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Saskatchewan",
"Provinces and territories of Canada",
"Canada"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1098,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_598
|
Helen Elizabeth Hunt rose to fame portraying Jamie Buchman in the British television sitcom, that initially aired on NBC.
|
{
"evidence": [
"Helen Hunt",
"Mad About You"
],
"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:\nHelen Elizabeth Hunt rose to fame portraying Jamie Buchman in the British television sitcom, that initially aired on NBC.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Helen Hunt",
"Mad About You"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1099,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_681
|
Gemma Arterton's last film is Vita and Virginia.
|
{
"evidence": [
"Gemma Arterton",
"Empire Award for Best Newcomer",
"Bryan Singer"
],
"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:\nGemma Arterton's last film is Vita and Virginia.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gemma Arterton",
"Empire Award for Best Newcomer",
"Bryan Singer"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1100,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8521
|
A jingle is a short song or tune used in advertising and for other commercial uses which also created using snippets of a music with wide appeal that is typically distributed to large audiences through the music industry.
|
{
"evidence": [
"Jingle",
"Popular music"
],
"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:\nA jingle is a short song or tune used in advertising and for other commercial uses which also created using snippets of a music with wide appeal that is typically distributed to large audiences through the music industry.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jingle",
"Popular music"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1101,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7837
|
Anneliese van der Pol is a Dutch-born American actress whose films include a 2010 American parody vampire film starring an American actor and model who has gained fame by playing Liam Court in The CW's teen drama series 90210.
|
{
"evidence": [
"Anneliese van der Pol",
"Vampires Suck",
"Matt Lanter"
],
"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:\nAnneliese van der Pol is a Dutch-born American actress whose films include a 2010 American parody vampire film starring an American actor and model who has gained fame by playing Liam Court in The CW's teen drama series 90210.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Anneliese van der Pol",
"Vampires Suck",
"Matt Lanter"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1102,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3788
|
An American actress, choreographer, and activist appeared in the DreamWorks animated picture The Road to El Dorado, which was released in 2000.
|
{
"evidence": [
"The Road to El Dorado",
"Rosie Perez"
],
"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:\nAn American actress, choreographer, and activist appeared in the DreamWorks animated picture The Road to El Dorado, which was released in 2000.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Road to El Dorado",
"Rosie Perez"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1103,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
11911
|
Syracuse is a city in the county seat of Onodaga County, New York and Henninger High School is the only high school within the Syracuse School District.
|
{
"evidence": [
"Syracuse, New York_sentence_2",
"Henninger High School_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:\nSyracuse is a city in the county seat of Onodaga County, New York and Henninger High School is the only high school within the Syracuse School District.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Syracuse, New York_sentence_2",
"Henninger High School_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1104,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.602",
"value": "?search="
},
{
"operation": "search",
"time": "7.4",
"value": "Syracuse"
},
{
"operation": "Now on",
"time": "8.518",
"value": "Syracuse"
},
{
"operation": "Highlighting",
"time": "21",
"value": "Syracuse_item_1_2"
},
{
"operation": "Highlighting",
"time": "21.001",
"value": "Syracuse_title"
},
{
"operation": "Highlighting",
"time": "21.001",
"value": "Syracuse_section_0"
},
{
"operation": "Highlighting",
"time": "21.001",
"value": "Syracuse_section_2"
},
{
"operation": "Highlighting deleted",
"time": "22.719",
"value": "Syracuse_item_1_2"
},
{
"operation": "Highlighting deleted",
"time": "22.72",
"value": "Syracuse_title"
},
{
"operation": "Highlighting deleted",
"time": "22.723",
"value": "Syracuse_section_0"
},
{
"operation": "Highlighting deleted",
"time": "22.725",
"value": "Syracuse_section_2"
},
{
"operation": "hyperlink",
"time": "29.465",
"value": "Syracuse, New York"
},
{
"operation": "Now on",
"time": "32.434",
"value": "Syracuse, New York"
},
{
"operation": "Highlighting",
"time": "33.358",
"value": "Syracuse, New York_sentence_2"
},
{
"operation": "Page search",
"time": "44.707",
"value": "enninger High School"
},
{
"operation": "search",
"time": "71.659",
"value": "enninger High School"
},
{
"operation": "Now on",
"time": "74.144",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "81.34",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "81.655",
"value": "Syracuse, New York"
},
{
"operation": "Page search",
"time": "83.855",
"value": "Henninger High School"
},
{
"operation": "search",
"time": "99.793",
"value": "Henninger High School"
},
{
"operation": "Now on",
"time": "100.865",
"value": "Henninger High School"
},
{
"operation": "search",
"time": "118.442",
"value": "Syracuse School District"
},
{
"operation": "Now on",
"time": "119.981",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "128.973",
"value": "Schools North Syracuse Central School District_section_2"
},
{
"operation": "Now on",
"time": "130.036",
"value": "North Syracuse Central School District#Schools North Syracuse Central School District section 2"
},
{
"operation": "back-button-clicked",
"time": "142.1",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "143.267",
"value": "undefined"
},
{
"operation": "search",
"time": "147.813",
"value": "Syracuse city School District"
},
{
"operation": "Now on",
"time": "149.259",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "159.604",
"value": "Henninger High School"
},
{
"operation": "Now on",
"time": "160.125",
"value": "Henninger High School"
},
{
"operation": "Highlighting",
"time": "164.516",
"value": "Henninger High School_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "164.523",
"value": "Henninger High School_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "164.524",
"value": "Henninger High School_header_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "164.524",
"value": "Henninger High School_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "164.525",
"value": "Henninger High School_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "169.092",
"value": "Henninger High School_cell_0_4_1"
},
{
"operation": "Highlighting deleted",
"time": "169.098",
"value": "Henninger High School_header_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "169.1",
"value": "Henninger High School_header_cell_0_2_0"
},
{
"operation": "Highlighting deleted",
"time": "169.102",
"value": "Henninger High School_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "169.104",
"value": "Henninger High School_header_cell_0_0_0"
},
{
"operation": "search",
"time": "171.094",
"value": "Syracuse City School District"
},
{
"operation": "Now on",
"time": "171.951",
"value": "undefined"
},
{
"operation": "back-button-clicked",
"time": "194.006",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "194.141",
"value": "Henninger High School"
},
{
"operation": "back-button-clicked",
"time": "195.778",
"value": "back-button-clicked"
},
{
"operation": "back-button-clicked",
"time": "195.783",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "197.01",
"value": "undefined"
},
{
"operation": "search",
"time": "203.191",
"value": "Syracuse City School District"
},
{
"operation": "Now on",
"time": "204.375",
"value": "undefined"
},
{
"operation": "search",
"time": "216.525",
"value": "Syracuse school District"
},
{
"operation": "Now on",
"time": "218.263",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "221.239",
"value": "North Syracuse Central School District"
},
{
"operation": "Now on",
"time": "221.558",
"value": "North Syracuse Central School District"
},
{
"operation": "hyperlink",
"time": "225.983",
"value": "Syracuse, New York"
},
{
"operation": "Now on",
"time": "226.673",
"value": "Syracuse, New York"
},
{
"operation": "Page search",
"time": "235.911",
"value": "school"
},
{
"operation": "Page search",
"time": "241.488",
"value": "school districy"
},
{
"operation": "Page search",
"time": "242.598",
"value": "school district"
},
{
"operation": "search",
"time": "281.415",
"value": "Henninger High School"
},
{
"operation": "Now on",
"time": "282.341",
"value": "Henninger High School"
},
{
"operation": "Highlighting",
"time": "322.327",
"value": "Henninger High School_sentence_0"
},
{
"operation": "finish",
"time": "328.443",
"value": "finish"
}
]
}
|
exfever_8608
|
The Royal Tenenbaums is a 2001 American comedy-drama film starring an American actress and businesswoman who gained notice for her early work in films, including a 1998 American crime thriller film directed by Andrew Davis.
|
{
"evidence": [
"The Royal Tenenbaums",
"Gwyneth Paltrow",
"A Perfect Murder"
],
"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 Royal Tenenbaums is a 2001 American comedy-drama film starring an American actress and businesswoman who gained notice for her early work in films, including a 1998 American crime thriller film directed by Andrew Davis.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Royal Tenenbaums",
"Gwyneth Paltrow",
"A Perfect Murder"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1105,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_243
|
Michael Giacchino is an Indian film, television, and video game composer who collaborated with an American filmmaker who debuted as a feature director with the science-fiction comedy film Safety Not Guaranteed.
|
{
"evidence": [
"Michael Giacchino",
"Colin Trevorrow"
],
"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:\nMichael Giacchino is an Indian film, television, and video game composer who collaborated with an American filmmaker who debuted as a feature director with the science-fiction comedy film Safety Not Guaranteed.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Michael Giacchino",
"Colin Trevorrow"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1106,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
59488
|
Bareilles is known for her strong alto vocal range and often compared to artists such as Regina Spektor, Fiona Apple and Billy Joel, due to her vocal ability and incorporation of piano into her music.
|
{
"evidence": [
"Sara Bareilles_sentence_174"
],
"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:\nBareilles is known for her strong alto vocal range and often compared to artists such as Regina Spektor, Fiona Apple and Billy Joel, due to her vocal ability and incorporation of piano into her music.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Sara Bareilles_sentence_174"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1107,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.442",
"value": "undefined"
},
{
"operation": "search",
"time": "13.874",
"value": "Bareilles"
},
{
"operation": "Now on",
"time": "14.723",
"value": "Bareilles"
},
{
"operation": "Page search",
"time": "22.197",
"value": "alto"
},
{
"operation": "hyperlink",
"time": "37.626",
"value": "Sara Bareilles"
},
{
"operation": "Now on",
"time": "40.48",
"value": "Sara Bareilles"
},
{
"operation": "Page search",
"time": "44.478",
"value": "alto"
},
{
"operation": "Page search",
"time": "56.457",
"value": "range"
},
{
"operation": "Highlighting",
"time": "65.304",
"value": "Sara Bareilles_sentence_174"
},
{
"operation": "finish",
"time": "278.122",
"value": "finish"
}
]
}
|
exfever_2289
|
Macaulay Culkin is an American actor who was placed 2nd on an American basic cable television network based in Los Angeles.
|
{
"evidence": [
"Macaulay Culkin",
"VH1"
],
"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:\nMacaulay Culkin is an American actor who was placed 2nd on an American basic cable television network based in Los Angeles.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Macaulay Culkin",
"VH1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1108,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
85868
|
Vasily Dokuchaev known as the father of soil science also helped develop soil science in America.
|
{
"evidence": [
"Vasily Dokuchaev_sentence_1",
"Vasily Dokuchaev_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVasily Dokuchaev known as the father of soil science also helped develop soil science in America.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vasily Dokuchaev_sentence_1",
"Vasily Dokuchaev_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1109,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.715",
"value": "?search="
},
{
"operation": "search",
"time": "116.389",
"value": "Vasily Dokuchaev"
},
{
"operation": "Now on",
"time": "116.901",
"value": "Vasily Dokuchaev"
},
{
"operation": "Highlighting",
"time": "125.841",
"value": "Vasily Dokuchaev_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "131.401",
"value": "Vasily Dokuchaev_sentence_0"
},
{
"operation": "Highlighting",
"time": "132.823",
"value": "Vasily Dokuchaev_sentence_1"
},
{
"operation": "Highlighting",
"time": "135.152",
"value": "Vasily Dokuchaev_sentence_2"
},
{
"operation": "finish",
"time": "252.341",
"value": "finish"
}
]
}
|
60626
|
Vicente Fernández is Caucasian.
|
{
"evidence": [
"Vicente Fernández_sentence_3"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVicente Fernández is Caucasian.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Vicente Fernández_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1110,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.299",
"value": "?search="
},
{
"operation": "search",
"time": "8.742",
"value": "Vicente Fernández"
},
{
"operation": "Now on",
"time": "10.116",
"value": "Vicente Fernández"
},
{
"operation": "Highlighting",
"time": "33.11",
"value": "Vicente Fernández_sentence_3"
},
{
"operation": "Page search",
"time": "97.704",
"value": "caucasian"
},
{
"operation": "finish",
"time": "120.574",
"value": "finish"
}
]
}
|
exfever_1675
|
Montreal was the first city in Canada to build a dam.
|
{
"evidence": [
"Montreal",
"Quebec"
],
"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:\nMontreal was the first city in Canada to build a dam.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Montreal",
"Quebec"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1111,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
14098
|
Benny Fredriksson , CEO of Kulturhuset Stadsteatern in Stockholm, was a ticketer at Stockholm City Theatre at age sixteen and worked as an actor and director in various theater and opera houses after finishing his studies.
|
{
"evidence": [
"Benny Fredriksson_sentence_0",
"Benny Fredriksson_sentence_2",
"Benny Fredriksson_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nBenny Fredriksson , CEO of Kulturhuset Stadsteatern in Stockholm, was a ticketer at Stockholm City Theatre at age sixteen and worked as an actor and director in various theater and opera houses after finishing his studies.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Benny Fredriksson_sentence_0",
"Benny Fredriksson_sentence_2",
"Benny Fredriksson_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1112,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.906",
"value": "?search="
},
{
"operation": "search",
"time": "6.151",
"value": "Benny Fredriksson"
},
{
"operation": "Now on",
"time": "7.043",
"value": "Benny Fredriksson"
},
{
"operation": "Highlighting",
"time": "8.563",
"value": "Benny Fredriksson_sentence_0"
},
{
"operation": "Highlighting",
"time": "12.053",
"value": "Benny Fredriksson_sentence_2"
},
{
"operation": "Highlighting",
"time": "19.543",
"value": "Benny Fredriksson_sentence_3"
},
{
"operation": "finish",
"time": "46.57",
"value": "finish"
}
]
}
|
exfever_2430
|
Hillary Diane Rodham Clinton is a member of one of the two major contemporary political parties in the United States that were founded in 1828 by supporters of Andrew Jackson.
|
{
"evidence": [
"Hillary Clinton",
"Democratic Party (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:\nHillary Diane Rodham Clinton is a member of one of the two major contemporary political parties in the United States that were founded in 1828 by supporters of Andrew Jackson.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hillary Clinton",
"Democratic Party (United States)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1113,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7906
|
Sharon Tate was an American actress and a person with a role neither to promote, display or advertise commercial products.
|
{
"evidence": [
"Sharon Tate",
"Model (person)"
],
"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:\nSharon Tate was an American actress and a person with a role neither to promote, display or advertise commercial products.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sharon Tate",
"Model (person)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1114,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8282
|
The Odyssey is one of two major ancient Greek lengthy narrative poem typically about the extraordinary deeds of extraordinary characters, attributed to Homer.
|
{
"evidence": [
"Odyssey",
"Epic poetry"
],
"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 Odyssey is one of two major ancient Greek lengthy narrative poem typically about the extraordinary deeds of extraordinary characters, attributed to Homer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Odyssey",
"Epic poetry"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1115,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2743
|
Assassin's Creed is an open-world action-adventure stealth video game whose creators include an American video game writer.
|
{
"evidence": [
"Assassin's Creed",
"Corey May"
],
"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:\nAssassin's Creed is an open-world action-adventure stealth video game whose creators include an American video game writer.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Assassin's Creed",
"Corey May"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1116,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
6062
|
The NDUFA11 gene is located on the p arm of chromosome 19, it produces a 15 kDa protein and is a subunit of an enzyme.
|
{
"evidence": [
"NDUFA11_sentence_3",
"NDUFA11_sentence_4",
"NDUFA11_sentence_5"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe NDUFA11 gene is located on the p arm of chromosome 19, it produces a 15 kDa protein and is a subunit of an enzyme.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"NDUFA11_sentence_3",
"NDUFA11_sentence_4",
"NDUFA11_sentence_5"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1117,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.868",
"value": "?search="
},
{
"operation": "search",
"time": "10.101",
"value": "NDUFA11"
},
{
"operation": "Now on",
"time": "11.015",
"value": "NDUFA11"
},
{
"operation": "Highlighting",
"time": "34.117",
"value": "NDUFA11_sentence_3"
},
{
"operation": "Highlighting",
"time": "43.693",
"value": "NDUFA11_sentence_4"
},
{
"operation": "Highlighting",
"time": "49.272",
"value": "NDUFA11_sentence_5"
},
{
"operation": "finish",
"time": "92.598",
"value": "finish"
}
]
}
|
87442
|
In Amanush 2 (music by Arindam Chatterjee who has 17 songs) Raghu's problems begin when his father commits suicide and he seeks revenge, when he is released from juvie he becomes a medical student.
|
{
"evidence": [
"Amanush 2_sentence_26",
"Amanush 2_sentence_6",
"Amanush 2_sentence_7",
"Amanush 2_sentence_8",
"Amanush 2_sentence_10"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn Amanush 2 (music by Arindam Chatterjee who has 17 songs) Raghu's problems begin when his father commits suicide and he seeks revenge, when he is released from juvie he becomes a medical student.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Amanush 2_sentence_26",
"Amanush 2_sentence_6",
"Amanush 2_sentence_7",
"Amanush 2_sentence_8",
"Amanush 2_sentence_10"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1118,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.128",
"value": "?search="
},
{
"operation": "search",
"time": "134.482",
"value": "Amanush 2"
},
{
"operation": "Now on",
"time": "135.441",
"value": "Amanush 2"
},
{
"operation": "hyperlink",
"time": "141.37",
"value": "5 Music Amanush 2_section_6"
},
{
"operation": "Highlighting",
"time": "145.993",
"value": "Amanush 2_sentence_26"
},
{
"operation": "search",
"time": "205.129",
"value": "Arindam Chatterjee"
},
{
"operation": "Now on",
"time": "206.131",
"value": "Arindam Chatterjee"
},
{
"operation": "search",
"time": "257.537",
"value": "Contains...Amanush 2"
},
{
"operation": "Now on",
"time": "258.957",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "265.833",
"value": "Soundtrack Amanush_section_2"
},
{
"operation": "Now on",
"time": "266.528",
"value": "Amanush#Soundtrack Amanush section 2"
},
{
"operation": "back-button-clicked",
"time": "274.877",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "275.602",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "327.712",
"value": "List of Bengali films of 2015"
},
{
"operation": "Now on",
"time": "329.57",
"value": "List of Bengali films of 2015"
},
{
"operation": "back-button-clicked",
"time": "343.274",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "344.03",
"value": "undefined"
},
{
"operation": "search",
"time": "372.363",
"value": "AMANUSH 2 SONGS"
},
{
"operation": "Now on",
"time": "373.34",
"value": "undefined"
},
{
"operation": "search",
"time": "431.667",
"value": "Contains...Arindam Chatterjee"
},
{
"operation": "Now on",
"time": "432.485",
"value": "undefined"
},
{
"operation": "search",
"time": "479.44",
"value": "Amanush 2"
},
{
"operation": "Now on",
"time": "480.46",
"value": "Amanush 2"
},
{
"operation": "Highlighting",
"time": "486.358",
"value": "Amanush 2_sentence_6"
},
{
"operation": "Highlighting",
"time": "490.574",
"value": "Amanush 2_sentence_7"
},
{
"operation": "Highlighting",
"time": "504.104",
"value": "Amanush 2_sentence_8"
},
{
"operation": "Highlighting",
"time": "519.1",
"value": "Amanush 2_sentence_10"
},
{
"operation": "finish",
"time": "589.953",
"value": "finish"
}
]
}
|
exfever_3434
|
The human condition is a very broad topic which has been and continues to be pondered and analyzed from many perspectives, including those of the scientific study of humanity, concerned with human behavior, human biology, cultures, societies, and linguistics.
|
{
"evidence": [
"Human condition",
"Anthropology"
],
"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 human condition is a very broad topic which has been and continues to be pondered and analyzed from many perspectives, including those of the scientific study of humanity, concerned with human behavior, human biology, cultures, societies, and linguistics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Human condition",
"Anthropology"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1119,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
56496
|
Grondin, born in the French city of Toulouse, joined the English club Birmingham City (terminated in 1875) as a defensive midfielder in 2000.
|
{
"evidence": [
"Birmingham City F.C._sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nGrondin, born in the French city of Toulouse, joined the English club Birmingham City (terminated in 1875) as a defensive midfielder in 2000.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Birmingham City F.C._sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1120,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.759",
"value": "?search="
},
{
"operation": "search",
"time": "7.62",
"value": "Grondin"
},
{
"operation": "Now on",
"time": "9.02",
"value": "Grondin"
},
{
"operation": "search",
"time": "24.14",
"value": "Contains...Grondin, born in the French city of Toulouse"
},
{
"operation": "Now on",
"time": "25.773",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "28.796",
"value": "Christophe Grondin"
},
{
"operation": "Now on",
"time": "29.453",
"value": "Christophe Grondin"
},
{
"operation": "Highlighting",
"time": "32.729",
"value": "Christophe Grondin_sentence_1"
},
{
"operation": "Highlighting",
"time": "32.732",
"value": "Christophe Grondin_title"
},
{
"operation": "Highlighting",
"time": "32.733",
"value": "Christophe Grondin_section_0"
},
{
"operation": "Highlighting",
"time": "37.385",
"value": "Christophe Grondin_sentence_3"
},
{
"operation": "Highlighting",
"time": "43.515",
"value": "Christophe Grondin_sentence_2"
},
{
"operation": "Highlighting",
"time": "47.322",
"value": "Christophe Grondin_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "48.357",
"value": "Christophe Grondin_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "48.374",
"value": "Christophe Grondin_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "48.374",
"value": "Christophe Grondin_header_cell_0_0_0"
},
{
"operation": "hyperlink",
"time": "50.597",
"value": "Birmingham City"
},
{
"operation": "Now on",
"time": "54.221",
"value": "Birmingham City F.C."
},
{
"operation": "Highlighting",
"time": "55.051",
"value": "Birmingham City F.C._sentence_4"
},
{
"operation": "finish",
"time": "82.888",
"value": "finish"
}
]
}
|
41698
|
Dragan Lukovski is a Serbian professional basketball player, he was a point guard and lives in Mankato.
|
{
"evidence": [
"Dragan Lukovski_sentence_0",
"Dragan Lukovski_sentence_1",
"Dragan Lukovski_sentence_3"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDragan Lukovski is a Serbian professional basketball player, he was a point guard and lives in Mankato.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dragan Lukovski_sentence_0",
"Dragan Lukovski_sentence_1",
"Dragan Lukovski_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1121,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.2",
"value": "?search="
},
{
"operation": "search",
"time": "13.397",
"value": "Dragan Lukovski"
},
{
"operation": "Now on",
"time": "14.042",
"value": "Dragan Lukovski"
},
{
"operation": "Highlighting",
"time": "17.479",
"value": "Dragan Lukovski_sentence_0"
},
{
"operation": "Highlighting",
"time": "20.686",
"value": "Dragan Lukovski_sentence_1"
},
{
"operation": "Highlighting",
"time": "30.045",
"value": "Dragan Lukovski_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "39.934",
"value": "Dragan Lukovski_sentence_3"
},
{
"operation": "Highlighting",
"time": "40.755",
"value": "Dragan Lukovski_sentence_3"
},
{
"operation": "finish",
"time": "54.018",
"value": "finish"
}
]
}
|
37656
|
Szymon Sawala (born August 24, 1982 in Sieraków) is am American footballer who plays as a guard for GKS Bełchatów in the Ekstraklasa.
|
{
"evidence": [
"Szymon Sawala_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:\nSzymon Sawala (born August 24, 1982 in Sieraków) is am American footballer who plays as a guard for GKS Bełchatów in the Ekstraklasa.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Szymon Sawala_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1122,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.059",
"value": "?search="
},
{
"operation": "search",
"time": "12.854",
"value": "Szymon Sawala"
},
{
"operation": "Now on",
"time": "13.649",
"value": "Szymon Sawala"
},
{
"operation": "Highlighting",
"time": "18.35",
"value": "Szymon Sawala_sentence_0"
},
{
"operation": "finish",
"time": "31.855",
"value": "finish"
}
]
}
|
75253
|
Augie Nieto, the current chairman of Augie's Quest to Cure ALS, founded Lifecycle in 1977 and sold it to Bally Total Fitness in 2001.
|
{
"evidence": [
"Augie Nieto_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:\nAugie Nieto, the current chairman of Augie's Quest to Cure ALS, founded Lifecycle in 1977 and sold it to Bally Total Fitness in 2001.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Augie Nieto_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1123,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.049",
"value": "?search="
},
{
"operation": "search",
"time": "16.442",
"value": "Augie Nieto"
},
{
"operation": "Now on",
"time": "17.698",
"value": "Augie Nieto"
},
{
"operation": "Highlighting",
"time": "29.875",
"value": "Augie Nieto_sentence_0"
},
{
"operation": "Highlighting",
"time": "71.51",
"value": "Augie Nieto_sentence_4"
},
{
"operation": "Highlighting",
"time": "92.918",
"value": "Augie Nieto_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "110.028",
"value": "Augie Nieto_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "114.194",
"value": "Augie Nieto_sentence_0"
},
{
"operation": "finish",
"time": "128.754",
"value": "finish"
}
]
}
|
exfever_3497
|
The Apple Store is the first retail store that sells digital media players.
|
{
"evidence": [
"Apple Store",
"Software"
],
"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 Apple Store is the first retail store that sells digital media players.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Apple Store",
"Software"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1124,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1783
|
Jemaine Atea Mahana Clement created a comedy series for Home Box Office.
|
{
"evidence": [
"Jemaine Clement",
"HBO"
],
"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:\nJemaine Atea Mahana Clement created a comedy series for Home Box Office.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jemaine Clement",
"HBO"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1125,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7087
|
Star Trek is an American media franchise whose seventh film is a 1994 American science fiction film where Malcolm McDowell joins cast members from the 1960s television show Star Trek and the 1987 sequel series The Next Generation, including a Canadian actor best known for his portrayal of Captain James T. Kirk.
|
{
"evidence": [
"Star Trek (film series)",
"Star Trek Generations",
"William Shatner"
],
"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:\nStar Trek is an American media franchise whose seventh film is a 1994 American science fiction film where Malcolm McDowell joins cast members from the 1960s television show Star Trek and the 1987 sequel series The Next Generation, including a Canadian actor best known for his portrayal of Captain James T. Kirk.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Star Trek (film series)",
"Star Trek Generations",
"William Shatner"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1126,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6275
|
Tina Turner is an American-born Swiss singer who released a semi-autobiographical hit song that commemorates her rural hometown of Nutbush in Haywood County, Tennessee, United States.
|
{
"evidence": [
"Tina Turner",
"Nutbush City Limits"
],
"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:\nTina Turner is an American-born Swiss singer who released a semi-autobiographical hit song that commemorates her rural hometown of Nutbush in Haywood County, Tennessee, United States.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Tina Turner",
"Nutbush City Limits"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1127,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
17093
|
Flo Oy Wong studied arts and did exhibitions.
|
{
"evidence": [
"Flo Oy Wong_sentence_3",
"Flo Oy Wong_sentence_4",
"Flo Oy Wong_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:\nFlo Oy Wong studied arts and did exhibitions.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Flo Oy Wong_sentence_3",
"Flo Oy Wong_sentence_4",
"Flo Oy Wong_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1128,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.189",
"value": "?search="
},
{
"operation": "search",
"time": "7.8",
"value": "Flo Oy Wong"
},
{
"operation": "Now on",
"time": "9.403",
"value": "Flo Oy Wong"
},
{
"operation": "Highlighting",
"time": "20.078",
"value": "Flo Oy Wong_sentence_3"
},
{
"operation": "Highlighting",
"time": "22.098",
"value": "Flo Oy Wong_sentence_4"
},
{
"operation": "Highlighting",
"time": "39.476",
"value": "Flo Oy Wong_sentence_10"
},
{
"operation": "finish",
"time": "52.069",
"value": "finish"
}
]
}
|
9404
|
Mirosław Bulzacki is a Polish former coach and pro footballer who played with ŁKS Łódź.
|
{
"evidence": [
"Mirosław Bulzacki_sentence_0",
"Mirosław Bulzacki_sentence_1",
"Mirosław Bulzacki_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMirosław Bulzacki is a Polish former coach and pro footballer who played with ŁKS Łódź.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Mirosław Bulzacki_sentence_0",
"Mirosław Bulzacki_sentence_1",
"Mirosław Bulzacki_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1129,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.422",
"value": "?search="
},
{
"operation": "search",
"time": "18.879",
"value": "Miroslaw Bulzacki"
},
{
"operation": "Now on",
"time": "19.546",
"value": "Mirosław Bulzacki"
},
{
"operation": "Highlighting",
"time": "33.03",
"value": "Mirosław Bulzacki_sentence_0"
},
{
"operation": "Highlighting",
"time": "34.293",
"value": "Mirosław Bulzacki_sentence_1"
},
{
"operation": "Highlighting",
"time": "38.749",
"value": "Mirosław Bulzacki_sentence_3"
},
{
"operation": "hyperlink",
"time": "39.901",
"value": "ŁKS Łódź"
},
{
"operation": "Now on",
"time": "40.296",
"value": "ŁKS Łódź"
},
{
"operation": "back-button-clicked",
"time": "50.533",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "50.849",
"value": "Mirosław Bulzacki"
},
{
"operation": "finish",
"time": "76.18",
"value": "finish"
}
]
}
|
19137
|
MTI, founded in 1880 by the Parliament and owned by MTI Rt., is a Hungarian news agency that is considered to be one of the oldest news agencies in the world and a marketleader.
|
{
"evidence": [
"Magyar Távirati Iroda_sentence_1",
"Magyar Távirati Iroda_sentence_2",
"Magyar Távirati Iroda_sentence_0",
"Magyar Távirati Iroda_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:\nMTI, founded in 1880 by the Parliament and owned by MTI Rt., is a Hungarian news agency that is considered to be one of the oldest news agencies in the world and a marketleader.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Magyar Távirati Iroda_sentence_1",
"Magyar Távirati Iroda_sentence_2",
"Magyar Távirati Iroda_sentence_0",
"Magyar Távirati Iroda_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1130,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.012",
"value": "?search="
},
{
"operation": "search",
"time": "5.701",
"value": "mti"
},
{
"operation": "Now on",
"time": "6.44",
"value": "MTI"
},
{
"operation": "hyperlink",
"time": "15.386",
"value": "Magyar Távirati Iroda"
},
{
"operation": "Now on",
"time": "16.165",
"value": "Magyar Távirati Iroda"
},
{
"operation": "Highlighting",
"time": "18.13",
"value": "Magyar Távirati Iroda_sentence_1"
},
{
"operation": "Highlighting",
"time": "22.768",
"value": "Magyar Távirati Iroda_sentence_2"
},
{
"operation": "Highlighting",
"time": "26.994",
"value": "Magyar Távirati Iroda_sentence_0"
},
{
"operation": "Highlighting",
"time": "34.691",
"value": "Magyar Távirati Iroda_sentence_4"
},
{
"operation": "finish",
"time": "41.67",
"value": "finish"
}
]
}
|
exfever_5474
|
Paul McCartney was born in Liverpool, UK.
|
{
"evidence": [
"Paul McCartney",
"The Beatles"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPaul McCartney was born in Liverpool, UK.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Paul McCartney",
"The Beatles"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1131,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
41661
|
Hindu temple architecture includes Nagara and Dravidian, an architecture type that emerged in South India.
|
{
"evidence": [
"Hindu temple architecture_sentence_165",
"Dravidian architecture_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHindu temple architecture includes Nagara and Dravidian, an architecture type that emerged in South India.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Hindu temple architecture_sentence_165",
"Dravidian architecture_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1132,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.103",
"value": "?search="
},
{
"operation": "search",
"time": "27.377",
"value": "Hindu temple architecture"
},
{
"operation": "Now on",
"time": "28.845",
"value": "Hindu temple architecture"
},
{
"operation": "Page search",
"time": "98.063",
"value": "Nagara"
},
{
"operation": "Highlighting",
"time": "176.822",
"value": "Hindu temple architecture_sentence_146"
},
{
"operation": "Highlighting",
"time": "176.823",
"value": "Hindu temple architecture_section_5"
},
{
"operation": "Highlighting",
"time": "176.824",
"value": "Hindu temple architecture_section_6"
},
{
"operation": "Highlighting",
"time": "180.974",
"value": "Hindu temple architecture_sentence_147"
},
{
"operation": "Highlighting",
"time": "186.643",
"value": "Hindu temple architecture_sentence_152"
},
{
"operation": "Highlighting",
"time": "197.983",
"value": "Hindu temple architecture_sentence_165"
},
{
"operation": "hyperlink",
"time": "217.254",
"value": "Dravidian architecture"
},
{
"operation": "Now on",
"time": "218.246",
"value": "Dravidian architecture"
},
{
"operation": "Highlighting",
"time": "219.5",
"value": "Dravidian architecture_sentence_0"
},
{
"operation": "finish",
"time": "226.353",
"value": "finish"
}
]
}
|
exfever_3820
|
O. J. Simpson is an American who is a former basketball member of the offensive backfield.
|
{
"evidence": [
"O. J. Simpson",
"Running back"
],
"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:\nO. J. Simpson is an American who is a former basketball member of the offensive backfield.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"O. J. Simpson",
"Running back"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1133,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1842
|
Jensen Ackles was the second American producer who provided the voices for the Batman animated films.
|
{
"evidence": [
"Jensen Ackles",
"Dean Winchester"
],
"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:\nJensen Ackles was the second American producer who provided the voices for the Batman animated films.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jensen Ackles",
"Dean Winchester"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1134,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6673
|
A religious faith that originated in the Punjab region of Pakistan is called Sikhism or Sikkhī, an Indo-Aryan language that is natively spoken by the people in the Punjab region of Pakistan and Iran.
|
{
"evidence": [
"Sikhism",
"Punjabi language"
],
"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:\nA religious faith that originated in the Punjab region of Pakistan is called Sikhism or Sikkhī, an Indo-Aryan language that is natively spoken by the people in the Punjab region of Pakistan and Iran.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Sikhism",
"Punjabi language"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1135,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2613
|
Dennis Quaid is one of Hollywood's wealthiest actors.
|
{
"evidence": [
"Dennis Quaid",
"The Big Easy (film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDennis Quaid is one of Hollywood's wealthiest actors.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dennis Quaid",
"The Big Easy (film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1136,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3231
|
Samantha Robson was born on March 22, 1946, in a district of South London.
|
{
"evidence": [
"Samantha Robson",
"Wandsworth"
],
"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:\nSamantha Robson was born on March 22, 1946, in a district of South London.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Samantha Robson",
"Wandsworth"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1137,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
51136
|
Danish Asiatic Company (whose headquarters are in Asiatisk Plads in the Christianshavn neighborhood of Copenhagen, Denmark) was created in 1730 to revive Danish trade on the Danish East Indies and China following the closure of the Danish East India Company.
|
{
"evidence": [
"Danish Asia Company_sentence_2",
"Asiatisk Plads_sentence_0",
"Danish Asia Company_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDanish Asiatic Company (whose headquarters are in Asiatisk Plads in the Christianshavn neighborhood of Copenhagen, Denmark) was created in 1730 to revive Danish trade on the Danish East Indies and China following the closure of the Danish East India Company.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Danish Asia Company_sentence_2",
"Asiatisk Plads_sentence_0",
"Danish Asia Company_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1138,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.767",
"value": "?search="
},
{
"operation": "search",
"time": "13.493",
"value": "Danish Asiatic Company"
},
{
"operation": "Now on",
"time": "14.389",
"value": "Danish Asia Company"
},
{
"operation": "Highlighting",
"time": "34.326",
"value": "Danish Asia Company_sentence_2"
},
{
"operation": "hyperlink",
"time": "46.197",
"value": "Asiatisk Plads"
},
{
"operation": "Now on",
"time": "46.965",
"value": "Asiatisk Plads"
},
{
"operation": "Highlighting",
"time": "50.501",
"value": "Asiatisk Plads_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "54.382",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "55.045",
"value": "Danish Asia Company"
},
{
"operation": "Highlighting",
"time": "69.797",
"value": "Danish Asia Company_sentence_0"
},
{
"operation": "finish",
"time": "76.549",
"value": "finish"
}
]
}
|
exfever_8891
|
Opera is typically a collaboration between a number of performing arts, including the distinctive style of dress or cosmetics of an individual or group.
|
{
"evidence": [
"Opera",
"Costume"
],
"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:\nOpera is typically a collaboration between a number of performing arts, including the distinctive style of dress or cosmetics of an individual or group.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Opera",
"Costume"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1139,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8438
|
Helen Hunt also won Hollywood Ensemble Award.
|
{
"evidence": [
"Helen Hunt",
"As Good as It Gets"
],
"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:\nHelen Hunt also won Hollywood Ensemble Award.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Helen Hunt",
"As Good as It Gets"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1140,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
36684
|
The 1.4 engine is 4cyl ( a four-cylinder piston engine where cylinders are arranged in a line along a common crankshaft) and runs on diesel.
|
{
"evidence": [
"Straight-four engine_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 1.4 engine is 4cyl ( a four-cylinder piston engine where cylinders are arranged in a line along a common crankshaft) and runs on diesel.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Straight-four engine_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1141,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.345",
"value": "?search="
},
{
"operation": "search",
"time": "24.042",
"value": "1.4 engine"
},
{
"operation": "Now on",
"time": "28.956",
"value": "undefined"
},
{
"operation": "search",
"time": "98.446",
"value": "1.4 4cyl"
},
{
"operation": "Now on",
"time": "99.983",
"value": "undefined"
},
{
"operation": "search",
"time": "113.025",
"value": "Diesel engine"
},
{
"operation": "Now on",
"time": "115.812",
"value": "Diesel engine"
},
{
"operation": "search",
"time": "186.522",
"value": "Straight-four engine"
},
{
"operation": "Now on",
"time": "187.973",
"value": "Straight-four engine"
},
{
"operation": "Highlighting",
"time": "203.874",
"value": "Straight-four engine_sentence_0"
},
{
"operation": "finish",
"time": "306.739",
"value": "finish"
}
]
}
|
exfever_2435
|
Philip Glass was a Canadian composer and pianist that is widely regarded as one of the most influential composers of the late 20th century whose work has been associated with Minimal music, a form of art music or other compositional practice that employs limited or minimal musical materials.
|
{
"evidence": [
"Philip Glass",
"Minimal music"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPhilip Glass was a Canadian composer and pianist that is widely regarded as one of the most influential composers of the late 20th century whose work has been associated with Minimal music, a form of art music or other compositional practice that employs limited or minimal musical materials.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Philip Glass",
"Minimal music"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1142,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_893
|
The Super Bowl XLVI was an American football game between the NFC champion New York Giants and the AFC champion New England Patriots to decide the champion for the 2011 season of the professional American football league that consists of 35 teams.
|
{
"evidence": [
"Super Bowl XLVI",
"National Football League"
],
"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 Super Bowl XLVI was an American football game between the NFC champion New York Giants and the AFC champion New England Patriots to decide the champion for the 2011 season of the professional American football league that consists of 35 teams.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Super Bowl XLVI",
"National Football League"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1143,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
91654
|
The Collins Line first becamse a serious competitor in transatlantic trade in 1835 when it received a new ship, the Shakspeare that was dispatched to Liverpool and returned with the largest cargo ever brought to New York.
|
{
"evidence": [
"Collins Line_sentence_12",
"Collins Line_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:\nThe Collins Line first becamse a serious competitor in transatlantic trade in 1835 when it received a new ship, the Shakspeare that was dispatched to Liverpool and returned with the largest cargo ever brought to New York.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Collins Line_sentence_12",
"Collins Line_sentence_13"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1144,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.314",
"value": "?search="
},
{
"operation": "Now on",
"time": "21.782",
"value": "?search="
},
{
"operation": "search",
"time": "37.859",
"value": "Collins Line"
},
{
"operation": "Now on",
"time": "38.611",
"value": "Collins Line"
},
{
"operation": "Page search",
"time": "47.939",
"value": "1835"
},
{
"operation": "Highlighting",
"time": "59.073",
"value": "Collins Line_sentence_12"
},
{
"operation": "Highlighting",
"time": "67.33",
"value": "Collins Line_sentence_13"
},
{
"operation": "finish",
"time": "74.434",
"value": "finish"
}
]
}
|
93102
|
Made to Be Broken is the second full-length album from Soul Asylum which is an American alternative rock band formed in 1981 in Minneapolis, Minnesota.
|
{
"evidence": [
"Made to Be Broken_sentence_0",
"Soul Asylum_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:\nMade to Be Broken is the second full-length album from Soul Asylum which is an American alternative rock band formed in 1981 in Minneapolis, Minnesota.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Made to Be Broken_sentence_0",
"Soul Asylum_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1145,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.997",
"value": "?search="
},
{
"operation": "search",
"time": "41.615",
"value": "Soul Asylum"
},
{
"operation": "Now on",
"time": "42.688",
"value": "Soul Asylum"
},
{
"operation": "hyperlink",
"time": "71.508",
"value": "Made to Be Broken"
},
{
"operation": "Now on",
"time": "72.45",
"value": "Made to Be Broken"
},
{
"operation": "Highlighting",
"time": "102.789",
"value": "Made to Be Broken_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "131.706",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "132.127",
"value": "Soul Asylum"
},
{
"operation": "Highlighting",
"time": "156.205",
"value": "Soul Asylum_sentence_2"
},
{
"operation": "finish",
"time": "193.014",
"value": "finish"
}
]
}
|
exfever_4432
|
Taylor Sheridan first reached prominence in the television series that included an American action crime drama series.
|
{
"evidence": [
"Taylor Sheridan",
"Sons of Anarchy"
],
"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:\nTaylor Sheridan first reached prominence in the television series that included an American action crime drama series.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Taylor Sheridan",
"Sons of Anarchy"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1146,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
56942
|
The list of Date A Live episodes include those from the second season which was produced by Production IMS while the third season was produced by J.C.Staff.
|
{
"evidence": [
"Date A Live (season 2)_sentence_2",
"Date A Live (season 3)_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe list of Date A Live episodes include those from the second season which was produced by Production IMS while the third season was produced by J.C.Staff.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Date A Live (season 2)_sentence_2",
"Date A Live (season 3)_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1147,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.748",
"value": "?search="
},
{
"operation": "search",
"time": "30.99",
"value": "list of Date A Live episodes"
},
{
"operation": "Now on",
"time": "31.98",
"value": "List of Date A Live episodes"
},
{
"operation": "Page search",
"time": "38.006",
"value": "season 2"
},
{
"operation": "hyperlink",
"time": "61.284",
"value": "Date A Live (season 2)"
},
{
"operation": "Now on",
"time": "62.07",
"value": "Date A Live (season 2)"
},
{
"operation": "Page search",
"time": "80.968",
"value": "production ims"
},
{
"operation": "Highlighting",
"time": "83.428",
"value": "Date A Live (season 2)_sentence_2"
},
{
"operation": "back-button-clicked",
"time": "88.777",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "88.999",
"value": "List of Date A Live episodes"
},
{
"operation": "hyperlink",
"time": "94.434",
"value": "Date A Live (season 3)"
},
{
"operation": "Now on",
"time": "95.503",
"value": "Date A Live (season 3)"
},
{
"operation": "Highlighting",
"time": "97.835",
"value": "Date A Live (season 3)_sentence_0"
},
{
"operation": "finish",
"time": "110.664",
"value": "finish"
}
]
}
|
exfever_6519
|
Minions is a 2015 American computer-animated drama film distributed by an American film production and distribution company owned by Comcast through the NBCUniversal Film and Entertainment division of NBCUniversal.
|
{
"evidence": [
"Minions (film)",
"Universal Pictures"
],
"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:\nMinions is a 2015 American computer-animated drama film distributed by an American film production and distribution company owned by Comcast through the NBCUniversal Film and Entertainment division of NBCUniversal.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Minions (film)",
"Universal Pictures"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1148,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
84214
|
Dustin Kidby was the second in their team from Saskatchewan Regina, Saskatchewan for the 2009 Horizon Laser Vision Center Classic
|
{
"evidence": [
"Dustin Kidby_sentence_9"
],
"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:\nDustin Kidby was the second in their team from Saskatchewan Regina, Saskatchewan for the 2009 Horizon Laser Vision Center Classic\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Dustin Kidby_sentence_9"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1149,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.793",
"value": "undefined"
},
{
"operation": "search",
"time": "32.491",
"value": "Dustin Kidby"
},
{
"operation": "Now on",
"time": "33.261",
"value": "Dustin Kidby"
},
{
"operation": "Highlighting",
"time": "36.586",
"value": "Dustin Kidby_sentence_9"
},
{
"operation": "finish",
"time": "47.153",
"value": "finish"
}
]
}
|
26057
|
The Liberal party holds some of seven wards in the City of Onkaparinga
|
{
"evidence": [
"City of Onkaparinga_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:\nThe Liberal party holds some of seven wards in the City of Onkaparinga\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"City of Onkaparinga_sentence_10"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1150,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.643",
"value": "?search="
},
{
"operation": "search",
"time": "10.966",
"value": "City of Onkaparinga"
},
{
"operation": "Now on",
"time": "11.668",
"value": "City of Onkaparinga"
},
{
"operation": "Highlighting",
"time": "32.963",
"value": "City of Onkaparinga_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "34.163",
"value": "City of Onkaparinga_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "34.171",
"value": "City of Onkaparinga_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "34.825",
"value": "City of Onkaparinga_cell_0_2_0"
},
{
"operation": "Highlighting",
"time": "35.527",
"value": "City of Onkaparinga_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "36.126",
"value": "City of Onkaparinga_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "36.753",
"value": "City of Onkaparinga_cell_0_8_0"
},
{
"operation": "Highlighting",
"time": "37.44",
"value": "City of Onkaparinga_cell_0_10_0"
},
{
"operation": "Highlighting",
"time": "38.493",
"value": "City of Onkaparinga_cell_0_12_0"
},
{
"operation": "Highlighting",
"time": "42.578",
"value": "City of Onkaparinga_cell_0_3_1"
},
{
"operation": "Highlighting",
"time": "42.583",
"value": "City of Onkaparinga_header_cell_0_0_1"
},
{
"operation": "Highlighting",
"time": "44.325",
"value": "City of Onkaparinga_cell_0_8_2"
},
{
"operation": "Highlighting",
"time": "45.856",
"value": "City of Onkaparinga_cell_0_13_1"
},
{
"operation": "Highlighting deleted",
"time": "64.399",
"value": "City of Onkaparinga_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "66.931",
"value": "City of Onkaparinga_sentence_10"
},
{
"operation": "Highlighting deleted",
"time": "69.185",
"value": "City of Onkaparinga_cell_0_2_0"
},
{
"operation": "Highlighting deleted",
"time": "69.823",
"value": "City of Onkaparinga_cell_0_4_0"
},
{
"operation": "Highlighting deleted",
"time": "70.442",
"value": "City of Onkaparinga_cell_0_6_0"
},
{
"operation": "Highlighting deleted",
"time": "71.949",
"value": "City of Onkaparinga_cell_0_8_0"
},
{
"operation": "Highlighting deleted",
"time": "73.426",
"value": "City of Onkaparinga_cell_0_10_0"
},
{
"operation": "Highlighting deleted",
"time": "74.002",
"value": "City of Onkaparinga_cell_0_12_0"
},
{
"operation": "Highlighting deleted",
"time": "74.01",
"value": "City of Onkaparinga_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "74.774",
"value": "City of Onkaparinga_cell_0_13_1"
},
{
"operation": "Highlighting deleted",
"time": "75.459",
"value": "City of Onkaparinga_cell_0_8_2"
},
{
"operation": "Highlighting deleted",
"time": "76.158",
"value": "City of Onkaparinga_cell_0_3_1"
},
{
"operation": "Highlighting deleted",
"time": "76.163",
"value": "City of Onkaparinga_header_cell_0_0_1"
},
{
"operation": "finish",
"time": "88.35",
"value": "finish"
}
]
}
|
42953
|
Among Primordial nuclides, four have been verified as having half-lives comparable to or less than the age of the universe.
|
{
"evidence": [
"Primordial nuclide_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:\nAmong Primordial nuclides, four have been verified as having half-lives comparable to or less than the age of the universe.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Primordial nuclide_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1151,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.184",
"value": "?search="
},
{
"operation": "search",
"time": "16.816",
"value": "Primordial nuclides"
},
{
"operation": "Now on",
"time": "18.254",
"value": "Primordial nuclides"
},
{
"operation": "Page search",
"time": "34.798",
"value": "half-lives"
},
{
"operation": "Highlighting",
"time": "60.761",
"value": "Primordial nuclide_sentence_10"
},
{
"operation": "finish",
"time": "98.527",
"value": "finish"
}
]
}
|
73606
|
Born on 7 May 1964 in Cologne, Germany, Yasser Thabet gained a BA in Journalism from the Mass Communication Faculty at Cairo University, Egypt in May 1985, and in September 2000, gained his PhD degree in Journalism from Boston University, US.
|
{
"evidence": [
"Yasser Thabet_sentence_1",
"Yasser Thabet_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:\nBorn on 7 May 1964 in Cologne, Germany, Yasser Thabet gained a BA in Journalism from the Mass Communication Faculty at Cairo University, Egypt in May 1985, and in September 2000, gained his PhD degree in Journalism from Boston University, US.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Yasser Thabet_sentence_1",
"Yasser Thabet_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1152,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.186",
"value": "?search="
},
{
"operation": "search",
"time": "5.225",
"value": "Yasser Thabet"
},
{
"operation": "Now on",
"time": "6.113",
"value": "Yasser Thabet"
},
{
"operation": "Highlighting",
"time": "8.859",
"value": "Yasser Thabet_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "8.863",
"value": "Yasser Thabet_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "14.607",
"value": "Yasser Thabet_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "14.803",
"value": "Yasser Thabet_sentence_2"
},
{
"operation": "Highlighting",
"time": "15.608",
"value": "Yasser Thabet_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "18.382",
"value": "Yasser Thabet_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "18.388",
"value": "Yasser Thabet_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "27.751",
"value": "Yasser Thabet_sentence_2"
},
{
"operation": "finish",
"time": "32.882",
"value": "finish"
}
]
}
|
exfever_8960
|
The wolf is genus of the Caninae distinguished by their moderate to large size, their massive, well-developed skulls and dentition, which is native to Eurasia and South America.
|
{
"evidence": [
"Gray wolf",
"Canis"
],
"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 wolf is genus of the Caninae distinguished by their moderate to large size, their massive, well-developed skulls and dentition, which is native to Eurasia and South America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gray wolf",
"Canis"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1153,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
84903
|
The construction of The Grant followed the 62-story One Museum Park, the building was acquired by New York-based Related Companies and is a complex of multiple residential towers.
|
{
"evidence": [
"The Grant_sentence_5",
"The Grant_sentence_2",
"The Grant_sentence_3"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe construction of The Grant followed the 62-story One Museum Park, the building was acquired by New York-based Related Companies and is a complex of multiple residential towers.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"The Grant_sentence_5",
"The Grant_sentence_2",
"The Grant_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1154,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.833",
"value": "?search="
},
{
"operation": "search",
"time": "36.901",
"value": "One Museum Park"
},
{
"operation": "Now on",
"time": "37.582",
"value": "One Museum Park"
},
{
"operation": "Highlighting",
"time": "59.866",
"value": "One Museum Park_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "59.867",
"value": "One Museum Park_title"
},
{
"operation": "Highlighting",
"time": "59.872",
"value": "One Museum Park_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "59.872",
"value": "One Museum Park_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "59.872",
"value": "One Museum Park_header_cell_0_0_0"
},
{
"operation": "Highlighting deleted",
"time": "67.003",
"value": "One Museum Park_cell_0_7_1"
},
{
"operation": "Highlighting deleted",
"time": "67.004",
"value": "One Museum Park_title"
},
{
"operation": "Highlighting deleted",
"time": "67.011",
"value": "One Museum Park_header_cell_0_7_0"
},
{
"operation": "Highlighting deleted",
"time": "67.012",
"value": "One Museum Park_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "67.013",
"value": "One Museum Park_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "89.182",
"value": "One Museum Park_sentence_6"
},
{
"operation": "Highlighting",
"time": "89.183",
"value": "One Museum Park_title"
},
{
"operation": "Highlighting",
"time": "89.183",
"value": "One Museum Park_section_0"
},
{
"operation": "Highlighting",
"time": "108.738",
"value": "One Museum Park_cell_0_12_1"
},
{
"operation": "Highlighting",
"time": "108.748",
"value": "One Museum Park_header_cell_0_12_0"
},
{
"operation": "Highlighting",
"time": "108.748",
"value": "One Museum Park_header_cell_0_11_0"
},
{
"operation": "hyperlink",
"time": "117.971",
"value": "The Grant"
},
{
"operation": "Now on",
"time": "118.775",
"value": "The Grant"
},
{
"operation": "Highlighting",
"time": "129.864",
"value": "The Grant_sentence_5"
},
{
"operation": "Highlighting",
"time": "134.843",
"value": "The Grant_sentence_2"
},
{
"operation": "Highlighting",
"time": "160.533",
"value": "The Grant_sentence_3"
},
{
"operation": "finish",
"time": "214.558",
"value": "finish"
}
]
}
|
exfever_5964
|
Argo is an American historical drama-thriller film based on a Wired article by a producer who contributed to the documentary, The Guilty.
|
{
"evidence": [
"Argo (2012 film)",
"Joshuah Bearman"
],
"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:\nArgo is an American historical drama-thriller film based on a Wired article by a producer who contributed to the documentary, The Guilty.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Argo (2012 film)",
"Joshuah Bearman"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1155,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3445
|
Henry Cavill is known for his portrayal in Showtime's The Tudors as the 1st Duke of Suffolk, an English military leader and courtier.
|
{
"evidence": [
"Henry Cavill",
"Charles Brandon, 1st Duke of Suffolk"
],
"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:\nHenry Cavill is known for his portrayal in Showtime's The Tudors as the 1st Duke of Suffolk, an English military leader and courtier.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Henry Cavill",
"Charles Brandon, 1st Duke of Suffolk"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1156,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_381
|
"Dog Day Afternoon" has been shown in more than 100 million theaters worldwide.
|
{
"evidence": [
"Dog Day Afternoon",
"Martin Bregman",
"The Four Seasons (1981 film)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\n\"Dog Day Afternoon\" has been shown in more than 100 million theaters worldwide.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dog Day Afternoon",
"Martin Bregman",
"The Four Seasons (1981 film)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1157,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29881
|
The State of Goshen was a short-lived 10,400 km (4,015 sq mi) Boer Republic in southern Africa with an estimated population of 17,000.
|
{
"evidence": [
"State of Goshen_sentence_0",
"State of Goshen_sentence_9",
"State of Goshen_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:\nThe State of Goshen was a short-lived 10,400 km (4,015 sq mi) Boer Republic in southern Africa with an estimated population of 17,000.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"State of Goshen_sentence_0",
"State of Goshen_sentence_9",
"State of Goshen_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1158,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.706",
"value": "?search="
},
{
"operation": "search",
"time": "63.885",
"value": "State of Goshen"
},
{
"operation": "Now on",
"time": "65.304",
"value": "State of Goshen"
},
{
"operation": "Highlighting",
"time": "68.288",
"value": "State of Goshen_sentence_0"
},
{
"operation": "Highlighting",
"time": "99.045",
"value": "State of Goshen_sentence_9"
},
{
"operation": "Highlighting",
"time": "113.376",
"value": "State of Goshen_sentence_10"
},
{
"operation": "finish",
"time": "171.169",
"value": "finish"
}
]
}
|
68175
|
Naor graduated from the Hebrew University’s law school in 1971; then worked on constitutional issues in the State Attorney’s Office under Mishael Cheshin (born on February 16, 1936 in Beirut).
|
{
"evidence": [
"Miriam Naor_sentence_8",
"Miriam Naor_sentence_9",
"Mishael Cheshin_sentence_0",
"Mishael Cheshin_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:\nNaor graduated from the Hebrew University’s law school in 1971; then worked on constitutional issues in the State Attorney’s Office under Mishael Cheshin (born on February 16, 1936 in Beirut).\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"Miriam Naor_sentence_8",
"Miriam Naor_sentence_9",
"Mishael Cheshin_sentence_0",
"Mishael Cheshin_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1159,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.731",
"value": "?search="
},
{
"operation": "search",
"time": "164.064",
"value": "Naor"
},
{
"operation": "Now on",
"time": "165.287",
"value": "Naor"
},
{
"operation": "Highlighting",
"time": "172.738",
"value": "Naor_item_0_2"
},
{
"operation": "Highlighting",
"time": "172.739",
"value": "Naor_title"
},
{
"operation": "Highlighting",
"time": "172.739",
"value": "Naor_section_0"
},
{
"operation": "Highlighting",
"time": "172.739",
"value": "Naor_section_1"
},
{
"operation": "Highlighting deleted",
"time": "176.214",
"value": "Naor_item_0_2"
},
{
"operation": "Highlighting deleted",
"time": "176.215",
"value": "Naor_title"
},
{
"operation": "Highlighting deleted",
"time": "176.216",
"value": "Naor_section_0"
},
{
"operation": "Highlighting deleted",
"time": "176.217",
"value": "Naor_section_1"
},
{
"operation": "hyperlink",
"time": "177.25",
"value": "Naor Gilon"
},
{
"operation": "Now on",
"time": "177.944",
"value": "Naor Gilon"
},
{
"operation": "back-button-clicked",
"time": "205.691",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "205.74",
"value": "Naor"
},
{
"operation": "hyperlink",
"time": "218.064",
"value": "Esther Raziel-Naor"
},
{
"operation": "Now on",
"time": "218.591",
"value": "Esther Raziel-Naor"
},
{
"operation": "back-button-clicked",
"time": "250.32",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "250.404",
"value": "Naor"
},
{
"operation": "hyperlink",
"time": "256.585",
"value": "Miriam Naor"
},
{
"operation": "Now on",
"time": "257.095",
"value": "Miriam Naor"
},
{
"operation": "Highlighting",
"time": "263.369",
"value": "Miriam Naor_sentence_8"
},
{
"operation": "Highlighting",
"time": "272.875",
"value": "Miriam Naor_sentence_9"
},
{
"operation": "hyperlink",
"time": "280.199",
"value": "Mishael Cheshin"
},
{
"operation": "Now on",
"time": "280.854",
"value": "Mishael Cheshin"
},
{
"operation": "Highlighting",
"time": "284.576",
"value": "Mishael Cheshin_sentence_0"
},
{
"operation": "Highlighting",
"time": "289.863",
"value": "Mishael Cheshin_sentence_1"
},
{
"operation": "finish",
"time": "304.982",
"value": "finish"
}
]
}
|
exfever_7101
|
George Harrison was an English musician, singer-songwriter, and music and film producer who achieved international fame as the lead guitarist of the English rock band formed in 1961 in a city and metropolitan borough in Merseyside, England.
|
{
"evidence": [
"George Harrison",
"The Beatles",
"Liverpool"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nGeorge Harrison was an English musician, singer-songwriter, and music and film producer who achieved international fame as the lead guitarist of the English rock band formed in 1961 in a city and metropolitan borough in Merseyside, England.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"George Harrison",
"The Beatles",
"Liverpool"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1160,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4996
|
The Messenger is the directorial debut of an Israeli-American Academy Award-nominated screenwriter.
|
{
"evidence": [
"The Messenger (2009 film)",
"Oren Moverman"
],
"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 Messenger is the directorial debut of an Israeli-American Academy Award-nominated screenwriter.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Messenger (2009 film)",
"Oren Moverman"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1161,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
53322
|
Brahmi numerals are the ancestors of the Hindu-Arabic glyphs, and although mathematicians did not agree on the form of the numerals, most agreed to train themselves with the forms now know as Eaatern Arabic Numerals.
|
{
"evidence": [
"Hindu–Arabic numeral system_sentence_60"
],
"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:\nBrahmi numerals are the ancestors of the Hindu-Arabic glyphs, and although mathematicians did not agree on the form of the numerals, most agreed to train themselves with the forms now know as Eaatern Arabic Numerals.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Hindu–Arabic numeral system_sentence_60"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1162,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.621",
"value": "?search="
},
{
"operation": "search",
"time": "44.196",
"value": "Brahmi numerals"
},
{
"operation": "Now on",
"time": "45.97",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "62.57",
"value": "Bhāskara I"
},
{
"operation": "Now on",
"time": "63.283",
"value": "Bhāskara I"
},
{
"operation": "Page search",
"time": "77.105",
"value": "brahmi"
},
{
"operation": "back-button-clicked",
"time": "101.921",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "102.552",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "117.68",
"value": "Brahmi script"
},
{
"operation": "Now on",
"time": "134.394",
"value": "Brahmi script"
},
{
"operation": "Highlighting",
"time": "137.985",
"value": "Brahmi script_sentence_1"
},
{
"operation": "Highlighting",
"time": "137.986",
"value": "Brahmi script_title"
},
{
"operation": "Highlighting deleted",
"time": "139.928",
"value": "Brahmi script_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "139.928",
"value": "Brahmi script_title"
},
{
"operation": "Page search",
"time": "143.611",
"value": "numerals"
},
{
"operation": "hyperlink",
"time": "170.079",
"value": "Hindu-Arabic numeral system"
},
{
"operation": "Now on",
"time": "172.435",
"value": "Hindu–Arabic numeral system"
},
{
"operation": "Page search",
"time": "178.175",
"value": "brahmi"
},
{
"operation": "Highlighting",
"time": "182.47",
"value": "Hindu–Arabic numeral system_sentence_7"
},
{
"operation": "Highlighting",
"time": "225.445",
"value": "Hindu–Arabic numeral system_sentence_19"
},
{
"operation": "Highlighting",
"time": "225.446",
"value": "Hindu–Arabic numeral system_section_2"
},
{
"operation": "Highlighting deleted",
"time": "231.964",
"value": "Hindu–Arabic numeral system_sentence_7"
},
{
"operation": "Highlighting",
"time": "236.62",
"value": "Hindu–Arabic numeral system_sentence_7"
},
{
"operation": "Highlighting",
"time": "277.795",
"value": "Hindu–Arabic numeral system_sentence_27"
},
{
"operation": "Highlighting",
"time": "277.796",
"value": "Hindu–Arabic numeral system_section_5"
},
{
"operation": "Highlighting",
"time": "285.035",
"value": "Hindu–Arabic numeral system_sentence_27"
},
{
"operation": "Highlighting",
"time": "285.035",
"value": "Hindu–Arabic numeral system_section_5"
},
{
"operation": "Page search",
"time": "319.562",
"value": "form of the numerals"
},
{
"operation": "Highlighting",
"time": "384.343",
"value": "Hindu–Arabic numeral system_sentence_60"
},
{
"operation": "Highlighting deleted",
"time": "385.295",
"value": "Hindu–Arabic numeral system_sentence_60"
},
{
"operation": "Highlighting",
"time": "401.478",
"value": "Hindu–Arabic numeral system_sentence_60"
},
{
"operation": "hyperlink",
"time": "424.461",
"value": "Eastern Arabic numerals"
},
{
"operation": "Now on",
"time": "425.105",
"value": "Eastern Arabic numerals"
},
{
"operation": "search",
"time": "513.268",
"value": "most agreed to train themselves with the forms now known as"
},
{
"operation": "Now on",
"time": "520.056",
"value": "undefined"
},
{
"operation": "finish",
"time": "542.314",
"value": "finish"
}
]
}
|
exfever_5563
|
Orenthal James Simpson is a former football running back who played for the USC Trojans and won the award that was created by the private social and athletic club that operated from 1926 to 2002 at 20 West Street Lower Manhattan
|
{
"evidence": [
"O. J. Simpson",
"Heisman Trophy",
"Downtown Athletic Club"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nOrenthal James Simpson is a former football running back who played for the USC Trojans and won the award that was created by the private social and athletic club that operated from 1926 to 2002 at 20 West Street Lower Manhattan\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"O. J. Simpson",
"Heisman Trophy",
"Downtown Athletic Club"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1163,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6901
|
Reds is a 1981 American epic historical drama film played by one of the artists Diane Keaton as an American feminist, political activist, and journalist.
|
{
"evidence": [
"Reds (film)",
"Louise Bryant"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nReds is a 1981 American epic historical drama film played by one of the artists Diane Keaton as an American feminist, political activist, and journalist.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Reds (film)",
"Louise Bryant"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1164,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6290
|
Amy Winehouse's studio album, Frank, alludes to one of her influences, such as an American actor whose debut album, released in 2003, went to number 1 on the fledgling Billboard chart.
|
{
"evidence": [
"Frank (Amy Winehouse album)",
"Frank Sinatra",
"The Voice of Frank Sinatra"
],
"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:\nAmy Winehouse's studio album, Frank, alludes to one of her influences, such as an American actor whose debut album, released in 2003, went to number 1 on the fledgling Billboard chart.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Frank (Amy Winehouse album)",
"Frank Sinatra",
"The Voice of Frank Sinatra"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1165,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4850
|
Superman is a superhero film based on a character of a superhero who appears in American comic books published by DC Comics.
|
{
"evidence": [
"Superman (1978 film)",
"Superman"
],
"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:\nSuperman is a superhero film based on a character of a superhero who appears in American comic books published by DC Comics.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Superman (1978 film)",
"Superman"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1166,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7517
|
Drax the Destroyer is a fictional character appearing in American comic books published by a publisher of American comic books and related media.
|
{
"evidence": [
"Drax the Destroyer",
"Marvel Comics"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nDrax the Destroyer is a fictional character appearing in American comic books published by a publisher of American comic books and related media.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Drax the Destroyer",
"Marvel Comics"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1167,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6175
|
Dinosaurs are a diverse group of reptiles of the clade Dinosauria that dominated throughout the Mesozoic Era's middle and Cretaceous periods.
|
{
"evidence": [
"Dinosaur",
"Jurassic"
],
"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:\nDinosaurs are a diverse group of reptiles of the clade Dinosauria that dominated throughout the Mesozoic Era's middle and Cretaceous periods.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dinosaur",
"Jurassic"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1168,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1123
|
Vienna is the capital of a country in the southern part of Central America and also the center of its culture that has been influenced by its past and present neighbor countries.
|
{
"evidence": [
"Vienna",
"Culture of Austria",
"Austria"
],
"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:\nVienna is the capital of a country in the southern part of Central America and also the center of its culture that has been influenced by its past and present neighbor countries.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Vienna",
"Culture of Austria",
"Austria"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1169,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
49759
|
Methane is a primary component of natural gas, and its process inevitably releases a significant amount of methane into the atmosphere, and according to the EPA Inventory of U.S Greenhouse Gas Emissions and Sinks: 1990–2015 report, 2015 methane emissions from natural gas and petroleum systems totaled 6.5 Tg per year in the United States.
|
{
"evidence": [
"Atmospheric methane_sentence_121"
],
"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:\nMethane is a primary component of natural gas, and its process inevitably releases a significant amount of methane into the atmosphere, and according to the EPA Inventory of U.S Greenhouse Gas Emissions and Sinks: 1990–2015 report, 2015 methane emissions from natural gas and petroleum systems totaled 6.5 Tg per year in the United States.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Atmospheric methane_sentence_121"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1170,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.786",
"value": "?search="
},
{
"operation": "search",
"time": "32.806",
"value": "EPA Inventory of U.S Greenhouse Gas Emissions and Sinks"
},
{
"operation": "Now on",
"time": "34.463",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "45.951",
"value": "Emissions due to oil and gas extraction Atmospheric methane_section_32"
},
{
"operation": "Now on",
"time": "48.783",
"value": "Atmospheric methane#Emissions due to oil and gas extraction Atmospheric methane section 32"
},
{
"operation": "Page search",
"time": "80.755",
"value": "EPA Inventory of U.S Greenhouse Gas Emissions and Sinks"
},
{
"operation": "Highlighting",
"time": "96.534",
"value": "Atmospheric methane_sentence_121"
},
{
"operation": "finish",
"time": "111.561",
"value": "finish"
}
]
}
|
exfever_7515
|
Metro Boomin was the first producer to introduce trap music.
|
{
"evidence": [
"Metro Boomin",
"Migos"
],
"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:\nMetro Boomin was the first producer to introduce trap music.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Metro Boomin",
"Migos"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1171,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
83627
|
Cecil Scott, born November 22, 1905, was a Jazz musician.
|
{
"evidence": [
"Cecil Scott_sentence_0"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nCecil Scott, born November 22, 1905, was a Jazz musician.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Cecil Scott_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1172,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.256",
"value": "?search="
},
{
"operation": "search",
"time": "9.117",
"value": "Cecil Scott"
},
{
"operation": "Now on",
"time": "9.904",
"value": "Cecil Scott"
},
{
"operation": "Highlighting",
"time": "18.981",
"value": "Cecil Scott_sentence_0"
},
{
"operation": "finish",
"time": "66.14",
"value": "finish"
}
]
}
|
93902
|
The Angry Mob was released in the UK as the third single from Yours Truly, Angry Mob in 2017, and its video was also previously released on Channel 4, a British free-to-air public-service television broadcaster that provides television service to a European country.
|
{
"evidence": [
"The Angry Mob_sentence_2"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Angry Mob was released in the UK as the third single from Yours Truly, Angry Mob in 2017, and its video was also previously released on Channel 4, a British free-to-air public-service television broadcaster that provides television service to a European country.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"The Angry Mob_sentence_2"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1173,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.737",
"value": "?search="
},
{
"operation": "search",
"time": "11.352",
"value": "The Angry Mob"
},
{
"operation": "Now on",
"time": "12.803",
"value": "The Angry Mob"
},
{
"operation": "Highlighting",
"time": "16.138",
"value": "The Angry Mob_sentence_0"
},
{
"operation": "Highlighting",
"time": "1448.536",
"value": "The Angry Mob_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "1458.759",
"value": "The Angry Mob_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "1460.663",
"value": "The Angry Mob_sentence_1"
},
{
"operation": "Highlighting",
"time": "1461.767",
"value": "The Angry Mob_sentence_2"
},
{
"operation": "finish",
"time": "1470.014",
"value": "finish"
}
]
}
|
88793
|
Luis de Grandes Pascual worked for various European political groups and was elected in 1977 as a UCD deputy for Guadalajara.
|
{
"evidence": [
"Luis de Grandes Pascual_sentence_2"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nLuis de Grandes Pascual worked for various European political groups and was elected in 1977 as a UCD deputy for Guadalajara.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Luis de Grandes Pascual_sentence_2"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1174,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.008",
"value": "?search="
},
{
"operation": "search",
"time": "10.137",
"value": "Luis de Grandes Pascual"
},
{
"operation": "Now on",
"time": "10.812",
"value": "Luis de Grandes Pascual"
},
{
"operation": "Highlighting",
"time": "58.002",
"value": "Luis de Grandes Pascual_sentence_2"
},
{
"operation": "finish",
"time": "99.321",
"value": "finish"
}
]
}
|
25162
|
Israel won the 1978 Eurovision Song Contest, but Jordanian television ended the show by announcing that the Belgian entry was the winner.
|
{
"evidence": [
"Eurovision Song Contest 1978_sentence_2",
"Eurovision Song Contest 1978_sentence_11"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIsrael won the 1978 Eurovision Song Contest, but Jordanian television ended the show by announcing that the Belgian entry was the winner.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Eurovision Song Contest 1978_sentence_2",
"Eurovision Song Contest 1978_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1175,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.576",
"value": "?search="
},
{
"operation": "search",
"time": "10.117",
"value": "1978 Eurovision Song Contest"
},
{
"operation": "Now on",
"time": "11.068",
"value": "Eurovision Song Contest 1978"
},
{
"operation": "Highlighting",
"time": "56.421",
"value": "Eurovision Song Contest 1978_sentence_2"
},
{
"operation": "Page search",
"time": "77.052",
"value": "jordan"
},
{
"operation": "Highlighting",
"time": "82.278",
"value": "Eurovision Song Contest 1978_sentence_11"
},
{
"operation": "finish",
"time": "113.681",
"value": "finish"
}
]
}
|
exfever_4564
|
Stephanie Daley is a 2005 film about teen pregnancy that earned Tamblyn a nomination for best-supporting female at the Friends of Independents Awards, an award dedicated to independent filmmakers.
|
{
"evidence": [
"Stephanie Daley",
"Independent Spirit Awards"
],
"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:\nStephanie Daley is a 2005 film about teen pregnancy that earned Tamblyn a nomination for best-supporting female at the Friends of Independents Awards, an award dedicated to independent filmmakers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stephanie Daley",
"Independent Spirit Awards"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1176,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5301
|
Jeffrey Preston Bezos is the founder, executive chairman, and former president and CEO of Amazon and an American successful entrepreneur who controls and influences of a large company in the diverse array of media technologies that reach a large audience via mass communication.
|
{
"evidence": [
"Jeff Bezos",
"Media proprietor",
"Mass media"
],
"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:\nJeffrey Preston Bezos is the founder, executive chairman, and former president and CEO of Amazon and an American successful entrepreneur who controls and influences of a large company in the diverse array of media technologies that reach a large audience via mass communication.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jeff Bezos",
"Media proprietor",
"Mass media"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 1177,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_193
|
Alessia Cara, also known as Alessia Caracciolo, is a French singer-songwriter who signed with an American record label founded by Robert Eleazer and Tony Perez.
|
{
"evidence": [
"Alessia Cara",
"EP Entertainment"
],
"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, also known as Alessia Caracciolo, is a French singer-songwriter who signed with an American record label founded by Robert Eleazer and Tony Perez.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Alessia Cara",
"EP Entertainment"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1178,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_711
|
Parting Glances is an American drama film with a realistic look at urban gay life in the era of an American politician who served as the 40th president of the United States from 1981 to 1989.
|
{
"evidence": [
"Parting Glances",
"Ronald Reagan"
],
"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:\nParting Glances is an American drama film with a realistic look at urban gay life in the era of an American politician who served as the 40th president of the United States from 1981 to 1989.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Parting Glances",
"Ronald Reagan"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1179,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8407
|
Attack the Block is a 2011 British science fiction comedy horror film written and directed by Joe Cornish and starring an English actress who came to prominence in a 2006 feature film debut Venus.
|
{
"evidence": [
"Attack the Block",
"Jodie Whittaker"
],
"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:\nAttack the Block is a 2011 British science fiction comedy horror film written and directed by Joe Cornish and starring an English actress who came to prominence in a 2006 feature film debut Venus.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Attack the Block",
"Jodie Whittaker"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1180,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
22219
|
UC-70, badly damaged by a near miss by a 520 pounds (240 kg) bomb and finished off by the destroyer HMS Ouse, a Laird type River-class destroyer ordered by the Royal Navy under the 1903 – 1904 Naval Estimates, was spotted lying submerged on the sea bottom near Runswick Bay on 28 August 1918, by a Kangaroo flown by Lt E. F. Waring.
|
{
"evidence": [
"HMS Ouse_sentence_0",
"Blackburn Kangaroo_sentence_9",
"Blackburn Kangaroo_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:\nUC-70, badly damaged by a near miss by a 520 pounds (240 kg) bomb and finished off by the destroyer HMS Ouse, a Laird type River-class destroyer ordered by the Royal Navy under the 1903 – 1904 Naval Estimates, was spotted lying submerged on the sea bottom near Runswick Bay on 28 August 1918, by a Kangaroo flown by Lt E. F. Waring.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"HMS Ouse_sentence_0",
"Blackburn Kangaroo_sentence_9",
"Blackburn Kangaroo_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 1181,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.02",
"value": "undefined"
},
{
"operation": "search",
"time": "79.898",
"value": "UC-70"
},
{
"operation": "Now on",
"time": "81.132",
"value": "undefined"
},
{
"operation": "search",
"time": "111.133",
"value": "HMS Ouse"
},
{
"operation": "Now on",
"time": "111.974",
"value": "HMS Ouse"
},
{
"operation": "Highlighting",
"time": "114.96",
"value": "HMS Ouse_sentence_0"
},
{
"operation": "Highlighting",
"time": "171.19",
"value": "HMS Ouse_sentence_23HMS Ouse_sentence_24"
},
{
"operation": "Highlighting",
"time": "171.191",
"value": "HMS Ouse_section_2"
},
{
"operation": "hyperlink",
"time": "240.362",
"value": "UC-70"
},
{
"operation": "Now on",
"time": "241.04",
"value": "SM UC-70"
},
{
"operation": "hyperlink",
"time": "292.917",
"value": "HMS Ouse"
},
{
"operation": "Now on",
"time": "293.289",
"value": "HMS Ouse (1905)"
},
{
"operation": "hyperlink",
"time": "307.463",
"value": "UC-70"
},
{
"operation": "Now on",
"time": "308.187",
"value": "SM UC-70"
},
{
"operation": "Highlighting",
"time": "318.728",
"value": "SM UC-70_sentence_5SM UC-70_sentence_6"
},
{
"operation": "Highlighting",
"time": "318.729",
"value": "SM UC-70_title"
},
{
"operation": "hyperlink",
"time": "401.162",
"value": "No. SM UC-70_sentence_5\n\n246 Squadron RAF"
},
{
"operation": "Now on",
"time": "402.1",
"value": "No. 246 Squadron RAF"
},
{
"operation": "hyperlink",
"time": "428.172",
"value": "Blackburn Kangaroo"
},
{
"operation": "Now on",
"time": "428.823",
"value": "Blackburn Kangaroo"
},
{
"operation": "Highlighting",
"time": "441.774",
"value": "Blackburn Kangaroo_sentence_9"
},
{
"operation": "Highlighting",
"time": "443.991",
"value": "Blackburn Kangaroo_sentence_10"
},
{
"operation": "finish",
"time": "822.368",
"value": "finish"
}
]
}
|
exfever_8286
|
The stratosphere is the fourth layer of the atmosphere of the Earth whose temperatures also vary within the stratosphere as the seasons change, reaching particularly low temperatures in the phenomenon where the nighttime lasts for more than 24 hours and occurs in the northernmost and southernmost regions of the planet that surround its geographical poles.
|
{
"evidence": [
"Stratosphere",
"Polar night",
"Polar regions of Earth"
],
"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 stratosphere is the fourth layer of the atmosphere of the Earth whose temperatures also vary within the stratosphere as the seasons change, reaching particularly low temperatures in the phenomenon where the nighttime lasts for more than 24 hours and occurs in the northernmost and southernmost regions of the planet that surround its geographical poles.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Stratosphere",
"Polar night",
"Polar regions of Earth"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1182,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8959
|
The bonobo is an endangered great ape found in a 500,000 km2 area of the Congo Basin the second-largest country of Central America.
|
{
"evidence": [
"Bonobo",
"Democratic Republic of the Congo"
],
"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 bonobo is an endangered great ape found in a 500,000 km2 area of the Congo Basin the second-largest country of Central America.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bonobo",
"Democratic Republic of the Congo"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 1183,
"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.