id
stringlengths 1
12
| question
stringlengths 22
566
| golden_answers
dict | data_source
stringclasses 2
values | prompt
listlengths 1
1
| ability
stringclasses 6
values | reward_model
dict | extra_info
dict | metadata
dict |
|---|---|---|---|---|---|---|---|---|
exfever_5338
|
Metallica is an American heavy metal band that became one of the founding big four bands of thrash metal, along with an American heavy metal band, formed in Los Angeles in 1988.
|
{
"evidence": [
"Metallica",
"Megadeth"
],
"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:\nMetallica is an American heavy metal band that became one of the founding big four bands of thrash metal, along with an American heavy metal band, formed in Los Angeles in 1988.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Metallica",
"Megadeth"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2084,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29973
|
KTTI changed their format to jazz on October 1, 1981, and in 1983, Sun Country Broadcasting bought KTTI and KBLU (a commercial AM radio station in Yuma, Arizona) at the same time.
|
{
"evidence": [
"KTTI_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:\nKTTI changed their format to jazz on October 1, 1981, and in 1983, Sun Country Broadcasting bought KTTI and KBLU (a commercial AM radio station in Yuma, Arizona) at the same time.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"KTTI_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2085,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.711",
"value": "?search="
},
{
"operation": "search",
"time": "7.284",
"value": "KTTI"
},
{
"operation": "Now on",
"time": "7.932",
"value": "KTTI"
},
{
"operation": "Highlighting",
"time": "19.111",
"value": "KTTI_cell_0_6_1"
},
{
"operation": "Highlighting",
"time": "19.124",
"value": "KTTI_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "19.124",
"value": "KTTI_header_cell_0_5_0"
},
{
"operation": "Highlighting deleted",
"time": "28.657",
"value": "KTTI_cell_0_6_1"
},
{
"operation": "Highlighting deleted",
"time": "28.672",
"value": "KTTI_header_cell_0_6_0"
},
{
"operation": "Highlighting deleted",
"time": "28.674",
"value": "KTTI_header_cell_0_5_0"
},
{
"operation": "Highlighting",
"time": "29.996",
"value": "KTTI_sentence_7"
},
{
"operation": "finish",
"time": "39.947",
"value": "finish"
}
]
}
|
exfever_7362
|
Syco is an American media and entertainment company, formed in 2005 through a company founded in 1929 as the American Record Corporation.
|
{
"evidence": [
"Syco",
"Sony 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:\nSyco is an American media and entertainment company, formed in 2005 through a company founded in 1929 as the American Record Corporation.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Syco",
"Sony Music"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2086,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1105
|
How to Train Your Dragon is a 2011 American computer-animated action film that belong to the fantasy genre with fantastic themes, usually magic, supernatural events,
|
{
"evidence": [
"How to Train Your Dragon (film)",
"Fantasy film"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nHow to Train Your Dragon is a 2011 American computer-animated action film that belong to the fantasy genre with fantastic themes, usually magic, supernatural events,\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"How to Train Your Dragon (film)",
"Fantasy film"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2087,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
58389
|
Crazy for Daisy is about a boy named Ernest Valentine who is madly in love with Daisy (who has no feelings for him) and does anything to get her attention, and by the end of the strip she always gives in and goes out with him.
|
{
"evidence": [
"Crazy for Daisy_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:\nCrazy for Daisy is about a boy named Ernest Valentine who is madly in love with Daisy (who has no feelings for him) and does anything to get her attention, and by the end of the strip she always gives in and goes out with him.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Crazy for Daisy_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2088,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.212",
"value": "?search="
},
{
"operation": "search",
"time": "3.648",
"value": "Crazy for Daisy"
},
{
"operation": "Now on",
"time": "4.87",
"value": "Crazy for Daisy"
},
{
"operation": "Highlighting",
"time": "7.065",
"value": "Crazy for Daisy_sentence_4"
},
{
"operation": "Highlighting",
"time": "12.419",
"value": "Crazy for Daisy_sentence_5"
},
{
"operation": "Highlighting",
"time": "27.942",
"value": "Crazy for Daisy_sentence_6"
},
{
"operation": "Highlighting",
"time": "37.684",
"value": "Crazy for Daisy_sentence_7"
},
{
"operation": "Highlighting",
"time": "41.926",
"value": "Crazy for Daisy_sentence_7"
},
{
"operation": "finish",
"time": "46.543",
"value": "finish"
}
]
}
|
exfever_7962
|
Gerard Butler is a Scottish actor who has played small roles in productions such as a 1997 spy film produced by Eon Productions that stars Pierce Brosnan as the fictional MI6 agent James Bond, directed by Roger Spottiswoode.
|
{
"evidence": [
"Gerard Butler",
"Tomorrow Never Dies"
],
"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:\nGerard Butler is a Scottish actor who has played small roles in productions such as a 1997 spy film produced by Eon Productions that stars Pierce Brosnan as the fictional MI6 agent James Bond, directed by Roger Spottiswoode.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Gerard Butler",
"Tomorrow Never Dies"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2089,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
44259
|
The Kennesaw Mountain National Battlefield Park, a National Battlefield which consists of four battlefield areas, was listed on the National Register of Historic Places.
|
{
"evidence": [
"Kennesaw Mountain National Battlefield Park_sentence_11"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Kennesaw Mountain National Battlefield Park, a National Battlefield which consists of four battlefield areas, was listed on the National Register of Historic Places.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Kennesaw Mountain National Battlefield Park_sentence_11"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2090,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.026",
"value": "?search="
},
{
"operation": "search",
"time": "15.15",
"value": "Kennesaw Mountain National Battlefield Park"
},
{
"operation": "Now on",
"time": "15.708",
"value": "Kennesaw Mountain National Battlefield Park"
},
{
"operation": "Highlighting",
"time": "106.971",
"value": "Kennesaw Mountain National Battlefield Park_sentence_0"
},
{
"operation": "Highlighting",
"time": "161.569",
"value": "Kennesaw Mountain National Battlefield Park_sentence_8"
},
{
"operation": "Highlighting",
"time": "174.533",
"value": "Kennesaw Mountain National Battlefield Park_sentence_11"
},
{
"operation": "Highlighting deleted",
"time": "180.255",
"value": "Kennesaw Mountain National Battlefield Park_sentence_8"
},
{
"operation": "finish",
"time": "192.612",
"value": "finish"
}
]
}
|
exfever_7821
|
Breaking Bad is an American neo-Western crime drama television series, set and filmed in the most populous city in the U.S. state of New Mexico.
|
{
"evidence": [
"Breaking Bad",
"Albuquerque, New Mexico"
],
"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:\nBreaking Bad is an American neo-Western crime drama television series, set and filmed in the most populous city in the U.S. state of New Mexico.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Breaking Bad",
"Albuquerque, New Mexico"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2091,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_1147
|
Eric Singer is an American hard rock musician who has performed with artists such as an American singer, songwriter, and actor whose career spans over 54 years.
|
{
"evidence": [
"Eric Singer",
"Alice Cooper"
],
"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:\nEric Singer is an American hard rock musician who has performed with artists such as an American singer, songwriter, and actor whose career spans over 54 years.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Eric Singer",
"Alice Cooper"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2092,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
48279
|
Shoreham Airshow took place annually, in the late summer, at Brighton City Airport which is located in the parish of Lancing in West Sussex, England.
|
{
"evidence": [
"Shoreham Airshow_sentence_0",
"Brighton City Airport_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:\nShoreham Airshow took place annually, in the late summer, at Brighton City Airport which is located in the parish of Lancing in West Sussex, England.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Shoreham Airshow_sentence_0",
"Brighton City Airport_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2093,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.595",
"value": "?search="
},
{
"operation": "search",
"time": "30.662",
"value": "Shoreham Airshow"
},
{
"operation": "Now on",
"time": "31.683",
"value": "Shoreham Airshow"
},
{
"operation": "Highlighting",
"time": "35.405",
"value": "Shoreham Airshow_sentence_0"
},
{
"operation": "Highlighting",
"time": "74.417",
"value": "Shoreham Airshow_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "75.719",
"value": "Shoreham Airshow_cell_0_6_1"
},
{
"operation": "Highlighting",
"time": "75.725",
"value": "Shoreham Airshow_header_cell_0_6_0"
},
{
"operation": "Highlighting",
"time": "75.725",
"value": "Shoreham Airshow_header_cell_0_0_0"
},
{
"operation": "Page search",
"time": "98.076",
"value": "Sussex"
},
{
"operation": "Highlighting",
"time": "108.783",
"value": "Shoreham Airshow_header_cell_0_7_0"
},
{
"operation": "Highlighting",
"time": "109.672",
"value": "Shoreham Airshow_cell_0_7_1"
},
{
"operation": "Highlighting",
"time": "109.678",
"value": "Shoreham Airshow_header_cell_0_7_0"
},
{
"operation": "Page search",
"time": "116.936",
"value": "parish"
},
{
"operation": "Page search",
"time": "137.886",
"value": "lancing"
},
{
"operation": "Page search",
"time": "142.666",
"value": "lancing"
},
{
"operation": "hyperlink",
"time": "170.238",
"value": "Brighton City Airport"
},
{
"operation": "Now on",
"time": "171.854",
"value": "Brighton City Airport"
},
{
"operation": "Highlighting",
"time": "202.567",
"value": "Brighton City Airport_sentence_1"
},
{
"operation": "finish",
"time": "228.002",
"value": "finish"
}
]
}
|
exfever_1910
|
Aaron Charles Carter's Oh Aaron is his best-selling album.
|
{
"evidence": [
"Aaron Carter",
"Backstreet Boys"
],
"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:\nAaron Charles Carter's Oh Aaron is his best-selling album.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Aaron Carter",
"Backstreet Boys"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2094,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2465
|
New York City is known as the most popular city that has a lot of casinos and gambling games.
|
{
"evidence": [
"One World Trade Center",
"New York City"
],
"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:\nNew York City is known as the most popular city that has a lot of casinos and gambling games.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"One World Trade Center",
"New York City"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2095,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3267
|
Modern Family is an American family sitcom television series created by Christopher Lloyd and Steven Levitan for the American multinational commercial broadcast television network that serves as the flagship property of Walt Disney Television.
|
{
"evidence": [
"Modern Family",
"American Broadcasting Company"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nModern Family is an American family sitcom television series created by Christopher Lloyd and Steven Levitan for the American multinational commercial broadcast television network that serves as the flagship property of Walt Disney Television.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Modern Family",
"American Broadcasting Company"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2096,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
51931
|
The dorsal sepal is egg-shaped, 9–13 mm long, 4–6.5 mm wide and curves upwards; while the sepals are oblong, 9–14 mm long, 2–3 mm wide and turned downwards.
|
{
"evidence": [
"Crepidium fimbriatum_sentence_7",
"Dendrobium bigibbum_sentence_8",
"Prasophyllum caricetum_sentence_7"
],
"label": "NOT ENOUGH INFO"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe dorsal sepal is egg-shaped, 9–13 mm long, 4–6.5 mm wide and curves upwards; while the sepals are oblong, 9–14 mm long, 2–3 mm wide and turned downwards.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Crepidium fimbriatum_sentence_7",
"Dendrobium bigibbum_sentence_8",
"Prasophyllum caricetum_sentence_7"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2097,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.715",
"value": "?search="
},
{
"operation": "search",
"time": "20.512",
"value": "dorsal sepal"
},
{
"operation": "Now on",
"time": "24.31",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "27.201",
"value": "dorsal segal"
},
{
"operation": "Now on",
"time": "27.738",
"value": "undefined"
},
{
"operation": "search",
"time": "286.167",
"value": "Crepidium fimbriatum"
},
{
"operation": "Now on",
"time": "287.36",
"value": "Crepidium fimbriatum"
},
{
"operation": "Highlighting",
"time": "595.945",
"value": "Crepidium fimbriatum_sentence_7"
},
{
"operation": "search",
"time": "1007.632",
"value": "sepals oblong"
},
{
"operation": "Now on",
"time": "1010.019",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "1014.572",
"value": "Dendrobium bigibbum"
},
{
"operation": "Now on",
"time": "1015.419",
"value": "Dendrobium bigibbum"
},
{
"operation": "Highlighting",
"time": "1031.016",
"value": "Dendrobium bigibbum_sentence_8"
},
{
"operation": "search",
"time": "1078.541",
"value": "Contains...wide and curves upwards"
},
{
"operation": "Now on",
"time": "1080.838",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "1084.194",
"value": "Taxonomy and naming Prasophyllum caricetum_section_1"
},
{
"operation": "Now on",
"time": "1085.18",
"value": "Prasophyllum caricetum#Taxonomy and naming Prasophyllum caricetum section 1"
},
{
"operation": "Highlighting",
"time": "1095.894",
"value": "Prasophyllum caricetum_sentence_7"
},
{
"operation": "finish",
"time": "1115.649",
"value": "finish"
}
]
}
|
59597
|
Rönesans Holding was founded in Saint Petersburg (Established on May 27, 1703), Russia and is the 8th largest contracting company in Europe
|
{
"evidence": [
"Rönesans Holding_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:\nRönesans Holding was founded in Saint Petersburg (Established on May 27, 1703), Russia and is the 8th largest contracting company in Europe\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Rönesans Holding_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2098,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.39",
"value": "?search="
},
{
"operation": "search",
"time": "10.15",
"value": "Rönesans Holding"
},
{
"operation": "Now on",
"time": "10.966",
"value": "Rönesans Holding"
},
{
"operation": "Highlighting",
"time": "19.794",
"value": "Rönesans Holding_sentence_1"
},
{
"operation": "hyperlink",
"time": "38.521",
"value": "Saint Petersburg"
},
{
"operation": "Now on",
"time": "40.754",
"value": "Saint Petersburg"
},
{
"operation": "Highlighting",
"time": "48.344",
"value": "Saint Petersburg_cell_0_4_1"
},
{
"operation": "Highlighting",
"time": "48.345",
"value": "Saint Petersburg_title"
},
{
"operation": "Highlighting",
"time": "48.36",
"value": "Saint Petersburg_header_cell_0_4_0"
},
{
"operation": "Highlighting",
"time": "48.36",
"value": "Saint Petersburg_header_cell_0_0_0"
},
{
"operation": "back-button-clicked",
"time": "50.167",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "50.311",
"value": "Rönesans Holding"
},
{
"operation": "Page search",
"time": "81.102",
"value": "largest"
},
{
"operation": "Highlighting",
"time": "85.192",
"value": "Rönesans Holding_sentence_7"
},
{
"operation": "Highlighting deleted",
"time": "86.972",
"value": "Rönesans Holding_sentence_1"
},
{
"operation": "finish",
"time": "292.277",
"value": "finish"
}
]
}
|
exfever_6245
|
Fred Trump was a member of the Singaporean and German-American family originating in modern Germany.
|
{
"evidence": [
"Fred Trump",
"Trump family"
],
"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:\nFred Trump was a member of the Singaporean and German-American family originating in modern Germany.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Fred Trump",
"Trump family"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2099,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
53064
|
After World War I, the Treaty of Versaille left a Romanian minority of 75,223 people in 1910, 65,197 Romanian speakers in 1921, and 42,391 Romanians in 2011 (according to the Vojvodina census) still living inside the borders of Serbia.
|
{
"evidence": [
"Romanians of Serbia_sentence_6",
"Romanians of Serbia_sentence_7",
"Romanians of Serbia_sentence_8"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nAfter World War I, the Treaty of Versaille left a Romanian minority of 75,223 people in 1910, 65,197 Romanian speakers in 1921, and 42,391 Romanians in 2011 (according to the Vojvodina census) still living inside the borders of Serbia.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Romanians of Serbia_sentence_6",
"Romanians of Serbia_sentence_7",
"Romanians of Serbia_sentence_8"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2100,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.201",
"value": "?search="
},
{
"operation": "search",
"time": "33.26",
"value": "Treaty of Versailles"
},
{
"operation": "Now on",
"time": "35.932",
"value": "Treaty of Versailles"
},
{
"operation": "Page search",
"time": "72.697",
"value": "Romanian"
},
{
"operation": "search",
"time": "130.472",
"value": "Romanians of Serbia"
},
{
"operation": "Now on",
"time": "131.426",
"value": "Romanians of Serbia"
},
{
"operation": "Highlighting",
"time": "134.608",
"value": "Romanians of Serbia_sentence_6"
},
{
"operation": "Highlighting",
"time": "145.162",
"value": "Romanians of Serbia_sentence_7"
},
{
"operation": "Highlighting",
"time": "149.784",
"value": "Romanians of Serbia_sentence_8"
},
{
"operation": "finish",
"time": "164.927",
"value": "finish"
}
]
}
|
exfever_7460
|
The RMS Titanic was an American passenger liner designed by a shipyard that works primarily in ship repair, located in the capital and largest city of Northern Ireland.
|
{
"evidence": [
"RMS Titanic",
"Harland and Wolff",
"Belfast"
],
"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 RMS Titanic was an American passenger liner designed by a shipyard that works primarily in ship repair, located in the capital and largest city of Northern Ireland.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"RMS Titanic",
"Harland and Wolff",
"Belfast"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2101,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
12947
|
John Englart has worked in web design and web content administration since 1908, and worked for Telstra Corporation (an Australian telecommunications company) for 31 years before being made redundant in 1950.
|
{
"evidence": [
"John Englart_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:\nJohn Englart has worked in web design and web content administration since 1908, and worked for Telstra Corporation (an Australian telecommunications company) for 31 years before being made redundant in 1950.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"John Englart_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2102,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.527",
"value": "?search="
},
{
"operation": "search",
"time": "8.544",
"value": "John Englart"
},
{
"operation": "Now on",
"time": "9.333",
"value": "John Englart"
},
{
"operation": "Page search",
"time": "62.64",
"value": "web design"
},
{
"operation": "Highlighting",
"time": "65.161",
"value": "John Englart_sentence_5"
},
{
"operation": "finish",
"time": "87.597",
"value": "finish"
}
]
}
|
46942
|
The Oreta fulgens is from family Drepanidae (butterfly family) forewings are olive green with a paler yellowish costa with a few black scales while the hindwings deepen in tint towards the hindmargin.
|
{
"evidence": [
"Oreta fulgens_sentence_0",
"Drepanidae_sentence_3",
"Oreta fulgens_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Oreta fulgens is from family Drepanidae (butterfly family) forewings are olive green with a paler yellowish costa with a few black scales while the hindwings deepen in tint towards the hindmargin.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Oreta fulgens_sentence_0",
"Drepanidae_sentence_3",
"Oreta fulgens_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2103,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.268",
"value": "?search="
},
{
"operation": "search",
"time": "19.399",
"value": "Oreta fulgens"
},
{
"operation": "Now on",
"time": "20.304",
"value": "Oreta fulgens"
},
{
"operation": "Highlighting",
"time": "23.063",
"value": "Oreta fulgens_sentence_0"
},
{
"operation": "hyperlink",
"time": "24.525",
"value": "Drepanidae"
},
{
"operation": "Now on",
"time": "24.922",
"value": "Drepanidae"
},
{
"operation": "Highlighting",
"time": "26.618",
"value": "Drepanidae_sentence_3"
},
{
"operation": "back-button-clicked",
"time": "33.519",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "33.577",
"value": "Oreta fulgens"
},
{
"operation": "hyperlink",
"time": "67.893",
"value": "Drepanidae"
},
{
"operation": "Now on",
"time": "68.253",
"value": "Drepanidae"
},
{
"operation": "back-button-clicked",
"time": "75.113",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "75.169",
"value": "Oreta fulgens"
},
{
"operation": "Highlighting",
"time": "83.082",
"value": "Oreta fulgens_sentence_4"
},
{
"operation": "finish",
"time": "98.135",
"value": "finish"
}
]
}
|
exfever_6647
|
The Night Of is the first HBO produced television series that won an Emmy Award.
|
{
"evidence": [
"The Night Of",
"HBO"
],
"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 Night Of is the first HBO produced television series that won an Emmy Award.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Night Of",
"HBO"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2104,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
22369
|
Phillipes’ backup catcher Eddie Murphy sat in field by the team’s Baker Bowl lockers and offices and Murphy spotted the opposing catcher’s signals to the pitcher and signaled once for a fastball and twice for a breaking ball.
|
{
"evidence": [
"1900 Philadelphia Phillies season_sentence_6",
"1900 Philadelphia Phillies season_sentence_8"
],
"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:\nPhillipes’ backup catcher Eddie Murphy sat in field by the team’s Baker Bowl lockers and offices and Murphy spotted the opposing catcher’s signals to the pitcher and signaled once for a fastball and twice for a breaking ball.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"1900 Philadelphia Phillies season_sentence_6",
"1900 Philadelphia Phillies season_sentence_8"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2105,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.555",
"value": "?search="
},
{
"operation": "search",
"time": "38.228",
"value": "catcher Eddie Murphy"
},
{
"operation": "Now on",
"time": "39.568",
"value": "undefined"
},
{
"operation": "search",
"time": "55.363",
"value": "eddie murphy"
},
{
"operation": "Now on",
"time": "56.282",
"value": "Eddie Murphy"
},
{
"operation": "hyperlink",
"time": "61.631",
"value": "Eddie Murphy (disambiguation)"
},
{
"operation": "Now on",
"time": "62.161",
"value": "Eddie Murphy (disambiguation)"
},
{
"operation": "hyperlink",
"time": "65.698",
"value": "Eddie Murphy (baseball)"
},
{
"operation": "Now on",
"time": "66.249",
"value": "Eddie Murphy (baseball)"
},
{
"operation": "Page search",
"time": "93.515",
"value": "catcher"
},
{
"operation": "search",
"time": "135.46",
"value": "phillies backup catcher"
},
{
"operation": "Now on",
"time": "137.004",
"value": "undefined"
},
{
"operation": "search",
"time": "235.288",
"value": "Ed Murphy (rugby league)"
},
{
"operation": "Now on",
"time": "236.183",
"value": "Ed Murphy (rugby league)"
},
{
"operation": "search",
"time": "242.653",
"value": "Ed Murphy (first baseman)"
},
{
"operation": "Now on",
"time": "243.353",
"value": "Ed Murphy (first baseman)"
},
{
"operation": "Highlighting",
"time": "246.41",
"value": "Ed Murphy (first baseman)_sentence_0"
},
{
"operation": "Highlighting",
"time": "246.411",
"value": "Ed Murphy (first baseman)_title"
},
{
"operation": "search",
"time": "255.408",
"value": "Ed Murphy (pitcher)"
},
{
"operation": "Now on",
"time": "256.097",
"value": "Ed Murphy (pitcher)"
},
{
"operation": "Highlighting",
"time": "261.033",
"value": "Ed Murphy (pitcher)_sentence_3"
},
{
"operation": "Highlighting",
"time": "261.034",
"value": "Ed Murphy (pitcher)_title"
},
{
"operation": "search",
"time": "341.642",
"value": "1900 Philadelphia Phillies season"
},
{
"operation": "Now on",
"time": "342.752",
"value": "1900 Philadelphia Phillies season"
},
{
"operation": "Page search",
"time": "353.347",
"value": "murphy"
},
{
"operation": "Highlighting",
"time": "357.657",
"value": "1900 Philadelphia Phillies season_sentence_6"
},
{
"operation": "Highlighting",
"time": "362.321",
"value": "1900 Philadelphia Phillies season_sentence_8"
},
{
"operation": "finish",
"time": "374.481",
"value": "finish"
}
]
}
|
exfever_619
|
Black Sails is a historical adventure television series that debuted online for free on an American online video sharing and social media platform.
|
{
"evidence": [
"Black Sails (TV series)",
"YouTube"
],
"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:\nBlack Sails is a historical adventure television series that debuted online for free on an American online video sharing and social media platform.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Black Sails (TV series)",
"YouTube"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2106,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
13538
|
In 1962, Paul Rothchild, a friend of the band who had written their second album began working for Prestige Records, which reissued the album as Bluegrass And Old Timey Music (1962), and produced a further album on the label, Blue Grass Get Together (with Tex Logan, 1964).
|
{
"evidence": [
"Charles River Valley Boys_sentence_7"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn 1962, Paul Rothchild, a friend of the band who had written their second album began working for Prestige Records, which reissued the album as Bluegrass And Old Timey Music (1962), and produced a further album on the label, Blue Grass Get Together (with Tex Logan, 1964).\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Charles River Valley Boys_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2107,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.039",
"value": "?search="
},
{
"operation": "search",
"time": "45.909",
"value": "Contains...Blue Grass Get Together"
},
{
"operation": "Now on",
"time": "47.132",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "49.992",
"value": "Charles River Valley Boys"
},
{
"operation": "Now on",
"time": "50.762",
"value": "Charles River Valley Boys"
},
{
"operation": "Page search",
"time": "54.742",
"value": "Paul Rothchild"
},
{
"operation": "Highlighting",
"time": "72.6",
"value": "Charles River Valley Boys_sentence_7"
},
{
"operation": "finish",
"time": "99.816",
"value": "finish"
}
]
}
|
59229
|
In the 2009 Maccabiah Games, Jason Lezak won nine medals.
|
{
"evidence": [
"2009 Maccabiah Games_sentence_9",
"Jason Lezak_sentence_39"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn the 2009 Maccabiah Games, Jason Lezak won nine medals.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"2009 Maccabiah Games_sentence_9",
"Jason Lezak_sentence_39"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2108,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.945",
"value": "?search="
},
{
"operation": "search",
"time": "18.707",
"value": "2009 Maccabiah Games"
},
{
"operation": "Now on",
"time": "19.445",
"value": "2009 Maccabiah Games"
},
{
"operation": "Highlighting",
"time": "43.86",
"value": "2009 Maccabiah Games_sentence_9"
},
{
"operation": "hyperlink",
"time": "57.434",
"value": "Jason Lezak"
},
{
"operation": "Now on",
"time": "57.983",
"value": "Jason Lezak"
},
{
"operation": "Highlighting",
"time": "78.674",
"value": "Jason Lezak_sentence_39"
},
{
"operation": "Highlighting deleted",
"time": "85.922",
"value": "Jason Lezak_sentence_39"
},
{
"operation": "Highlighting",
"time": "88.154",
"value": "Jason Lezak_sentence_39"
},
{
"operation": "finish",
"time": "138.29",
"value": "finish"
}
]
}
|
exfever_5782
|
Christopher Frank Carandini Lee was an English actor with a film role as a fictional character in the Star Wars franchise that was created by George Lucas.
|
{
"evidence": [
"Christopher Lee",
"Count Dooku"
],
"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:\nChristopher Frank Carandini Lee was an English actor with a film role as a fictional character in the Star Wars franchise that was created by George Lucas.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Christopher Lee",
"Count Dooku"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2109,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
21153
|
Cameroon lists three definitely endangered languages, 13 severely endangered, and 16 critically endangered from among its at least 250 languages.
|
{
"evidence": [
"Languages of Cameroon_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:\nCameroon lists three definitely endangered languages, 13 severely endangered, and 16 critically endangered from among its at least 250 languages.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Languages of Cameroon_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2110,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.583",
"value": "?search="
},
{
"operation": "search",
"time": "21.234",
"value": "Contains...Cameroon endangered languages"
},
{
"operation": "Now on",
"time": "22.301",
"value": "undefined"
},
{
"operation": "search",
"time": "32.548",
"value": "cameroon languages"
},
{
"operation": "Now on",
"time": "33.527",
"value": "undefined"
},
{
"operation": "search",
"time": "44.154",
"value": "Cameroon"
},
{
"operation": "Now on",
"time": "46.028",
"value": "Cameroon"
},
{
"operation": "Page search",
"time": "49.106",
"value": "endangererd"
},
{
"operation": "Page search",
"time": "68.507",
"value": "languages"
},
{
"operation": "Highlighting",
"time": "83.036",
"value": "Cameroon_sentence_8"
},
{
"operation": "Highlighting",
"time": "83.038",
"value": "Cameroon_title"
},
{
"operation": "hyperlink",
"time": "103.304",
"value": "8.2 Languages Cameroon_section_18"
},
{
"operation": "hyperlink",
"time": "114.289",
"value": "Languages of Cameroon"
},
{
"operation": "Now on",
"time": "116.195",
"value": "Languages of Cameroon"
},
{
"operation": "Page search",
"time": "128.479",
"value": "endangered"
},
{
"operation": "Page search",
"time": "129.331",
"value": "endangered"
},
{
"operation": "Highlighting",
"time": "152.54",
"value": "Languages of Cameroon_sentence_0"
},
{
"operation": "Page search",
"time": "165.511",
"value": "critically"
},
{
"operation": "search",
"time": "214.581",
"value": "critically endangered cameroon"
},
{
"operation": "Now on",
"time": "215.915",
"value": "undefined"
},
{
"operation": "search",
"time": "234.177",
"value": "endangered languages cameroon"
},
{
"operation": "Now on",
"time": "234.926",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "253.319",
"value": "Bung language"
},
{
"operation": "Now on",
"time": "253.85",
"value": "Bung language"
},
{
"operation": "hyperlink",
"time": "261.26",
"value": "Cameroon"
},
{
"operation": "Now on",
"time": "261.681",
"value": "Cameroon"
},
{
"operation": "search",
"time": "281.776",
"value": "Languages of Cameroon"
},
{
"operation": "Now on",
"time": "282.304",
"value": "Languages of Cameroon"
},
{
"operation": "Page search",
"time": "295.73",
"value": "13"
},
{
"operation": "hyperlink",
"time": "325.048",
"value": "Demographics of Cameroon"
},
{
"operation": "Now on",
"time": "326.332",
"value": "Demographics of Cameroon"
},
{
"operation": "Page search",
"time": "330.483",
"value": "endangered"
},
{
"operation": "Page search",
"time": "347.474",
"value": "languages"
},
{
"operation": "hyperlink",
"time": "351.933",
"value": "4 Languages Demographics of Cameroon_section_5"
},
{
"operation": "search",
"time": "375.344",
"value": "Languages of Cameroon"
},
{
"operation": "Now on",
"time": "376.012",
"value": "Languages of Cameroon"
},
{
"operation": "Page search",
"time": "388.374",
"value": "critically"
},
{
"operation": "Page search",
"time": "429.013",
"value": "severely endangered language cameroon"
},
{
"operation": "search",
"time": "435.881",
"value": "severely endangered language cameroon"
},
{
"operation": "Now on",
"time": "437.334",
"value": "undefined"
},
{
"operation": "search",
"time": "453.057",
"value": "Contains...endangered language cameroon"
},
{
"operation": "Now on",
"time": "453.789",
"value": "undefined"
},
{
"operation": "search",
"time": "468.699",
"value": "Contains...lanugages of camerooon"
},
{
"operation": "Now on",
"time": "469.445",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "470.783",
"value": "Languages of Cameroon"
},
{
"operation": "Now on",
"time": "471.451",
"value": "Languages of Cameroon"
},
{
"operation": "Page search",
"time": "515.062",
"value": "threat"
},
{
"operation": "finish",
"time": "534.152",
"value": "finish"
}
]
}
|
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": 2111,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7494
|
Hamlet is a drama film based on the Shakespearean tragedy of the same name, directed by Franco Zeffirelli and starring an British producer known for his action hero roles.
|
{
"evidence": [
"Hamlet (1990 film)",
"Mel Gibson"
],
"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:\nHamlet is a drama film based on the Shakespearean tragedy of the same name, directed by Franco Zeffirelli and starring an British producer known for his action hero roles.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Hamlet (1990 film)",
"Mel Gibson"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2112,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4212
|
Kathleen Erin Walsh is an Australian actress known for her role as Olivia Baker on an American teen drama television series developed for Netflix by Brian Yorkey and based on the 2007 novel by Jay Asher.
|
{
"evidence": [
"Kate Walsh (actress)",
"13 Reasons Why"
],
"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:\nKathleen Erin Walsh is an Australian actress known for her role as Olivia Baker on an American teen drama television series developed for Netflix by Brian Yorkey and based on the 2007 novel by Jay Asher.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Kate Walsh (actress)",
"13 Reasons Why"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2113,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3562
|
The A Song of Ice and Fire series was the first epic fantasy novel created by GRRM.
|
{
"evidence": [
"Bran Stark",
"A Game of Thrones",
"George R. R. Martin"
],
"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 A Song of Ice and Fire series was the first epic fantasy novel created by GRRM.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Bran Stark",
"A Game of Thrones",
"George R. R. Martin"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2114,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
30733
|
Despite its textile industry, Pamunuwa has very few visitors throughout the year and its shops and stalls are rarely crowded.
|
{
"evidence": [
"Pamunuwa_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:\nDespite its textile industry, Pamunuwa has very few visitors throughout the year and its shops and stalls are rarely crowded.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Pamunuwa_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2115,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.685",
"value": "?search="
},
{
"operation": "search",
"time": "15.352",
"value": "Pamunuwa"
},
{
"operation": "Now on",
"time": "16.263",
"value": "Pamunuwa"
},
{
"operation": "Page search",
"time": "39.649",
"value": "shops and stalls"
},
{
"operation": "Highlighting",
"time": "48.093",
"value": "Pamunuwa_sentence_5"
},
{
"operation": "finish",
"time": "55.132",
"value": "finish"
}
]
}
|
exfever_5732
|
The phalanx was composed of heavy infantry armed with a pole weapon consisting of shaft wood or may be made of durable material, such as a chemical element with the symbol Fe and atomic number 26.
|
{
"evidence": [
"Phalanx",
"Spear",
"Iron"
],
"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 phalanx was composed of heavy infantry armed with a pole weapon consisting of shaft wood or may be made of durable material, such as a chemical element with the symbol Fe and atomic number 26.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Phalanx",
"Spear",
"Iron"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2116,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
67208
|
Ouyang Zhan's family had resided for several generations in China’s northernmost province (Harbin, Heilongjiang in China), then, Ouyang Zhan completed the Jinshi degree in 1988, in the year that came to know as the “Year of the dragon”.
|
{
"evidence": [
"Ouyang Zhan_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:\nOuyang Zhan's family had resided for several generations in China’s northernmost province (Harbin, Heilongjiang in China), then, Ouyang Zhan completed the Jinshi degree in 1988, in the year that came to know as the “Year of the dragon”.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Ouyang Zhan_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2117,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "12.452",
"value": "?search="
},
{
"operation": "search",
"time": "76.193",
"value": "Ouyang Zhan"
},
{
"operation": "Now on",
"time": "77.042",
"value": "Ouyang Zhan"
},
{
"operation": "Highlighting",
"time": "134.015",
"value": "Ouyang Zhan_sentence_4"
},
{
"operation": "finish",
"time": "228.949",
"value": "finish"
}
]
}
|
33525
|
Kolofata (a town and commune in Cameroon) was assaulted by Boko Haram on January 12, 2015 which was later repelled with only one Cameroonian officer killed and 143 Boko Haram insurgents killed by the Cameroonian Army.
|
{
"evidence": [
"Kolofata_sentence_0",
"Kolofata_sentence_1",
"Kolofata_sentence_2",
"Kolofata_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:\nKolofata (a town and commune in Cameroon) was assaulted by Boko Haram on January 12, 2015 which was later repelled with only one Cameroonian officer killed and 143 Boko Haram insurgents killed by the Cameroonian Army.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Kolofata_sentence_0",
"Kolofata_sentence_1",
"Kolofata_sentence_2",
"Kolofata_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2118,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.368",
"value": "?search="
},
{
"operation": "Now on",
"time": "9.586",
"value": "?search="
},
{
"operation": "search",
"time": "14.804",
"value": "Kolofata"
},
{
"operation": "Now on",
"time": "15.92",
"value": "Kolofata"
},
{
"operation": "Highlighting",
"time": "18.938",
"value": "Kolofata_sentence_0"
},
{
"operation": "Highlighting",
"time": "26.559",
"value": "Kolofata_sentence_1"
},
{
"operation": "Highlighting",
"time": "54.192",
"value": "Kolofata_sentence_2"
},
{
"operation": "Highlighting",
"time": "56.833",
"value": "Kolofata_sentence_3"
},
{
"operation": "finish",
"time": "164.08",
"value": "finish"
}
]
}
|
exfever_3759
|
The Ten Commandments is a 1956 American style of filmmaking with a large-scale, sweeping scope, and spectacle religious drama film.
|
{
"evidence": [
"The Ten Commandments (1956 film)",
"Epic film"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Ten Commandments is a 1956 American style of filmmaking with a large-scale, sweeping scope, and spectacle religious drama film.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Ten Commandments (1956 film)",
"Epic film"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2119,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
8329
|
Sophus Mads Jørgensen is one of the founders of coordination chemistry who is still alive.
|
{
"evidence": [
"Sophus Mads Jørgensen_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:\nSophus Mads Jørgensen is one of the founders of coordination chemistry who is still alive.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Sophus Mads Jørgensen_sentence_0"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2120,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.726",
"value": "?search="
},
{
"operation": "search",
"time": "19.744",
"value": "Sophus Mads Jørgensen"
},
{
"operation": "Now on",
"time": "20.777",
"value": "Sophus Mads Jørgensen"
},
{
"operation": "Highlighting",
"time": "28.135",
"value": "Sophus Mads Jørgensen_sentence_1"
},
{
"operation": "Highlighting deleted",
"time": "29.953",
"value": "Sophus Mads Jørgensen_sentence_1"
},
{
"operation": "Highlighting",
"time": "30.971",
"value": "Sophus Mads Jørgensen_sentence_0"
},
{
"operation": "finish",
"time": "36.498",
"value": "finish"
}
]
}
|
70190
|
St. Peter's Basilica is a tourist attraction in Rome, found in the Vatican City, and considered to be the "home of the Roman Catholic Church."
|
{
"evidence": [
"St. Peter's Basilica_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:\nSt. Peter's Basilica is a tourist attraction in Rome, found in the Vatican City, and considered to be the \"home of the Roman Catholic Church.\"\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"St. Peter's Basilica_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2121,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.075",
"value": "?search="
},
{
"operation": "search",
"time": "12.193",
"value": "St. Peter's Basilica"
},
{
"operation": "Now on",
"time": "14.521",
"value": "St. Peter's Basilica"
},
{
"operation": "Page search",
"time": "117.484",
"value": "tourisst"
},
{
"operation": "Page search",
"time": "119.731",
"value": "tourist"
},
{
"operation": "Highlighting",
"time": "268.809",
"value": "St. Peter's Basilica_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "279.998",
"value": "St. Peter's Basilica_sentence_3"
},
{
"operation": "Highlighting",
"time": "281.024",
"value": "St. Peter's Basilica_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "281.047",
"value": "St. Peter's Basilica_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "281.047",
"value": "St. Peter's Basilica_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "281.834",
"value": "St. Peter's Basilica_cell_0_2_1"
},
{
"operation": "Highlighting",
"time": "281.854",
"value": "St. Peter's Basilica_header_cell_0_2_0"
},
{
"operation": "Highlighting deleted",
"time": "289.763",
"value": "St. Peter's Basilica_cell_0_2_1"
},
{
"operation": "Highlighting deleted",
"time": "289.784",
"value": "St. Peter's Basilica_header_cell_0_2_0"
},
{
"operation": "Highlighting deleted",
"time": "290.532",
"value": "St. Peter's Basilica_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "290.554",
"value": "St. Peter's Basilica_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "290.555",
"value": "St. Peter's Basilica_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "291.357",
"value": "St. Peter's Basilica_sentence_3"
},
{
"operation": "Page search",
"time": "316.768",
"value": "home of the roman catholic church"
},
{
"operation": "Page search",
"time": "346.927",
"value": "attraction"
},
{
"operation": "hyperlink",
"time": "354.282",
"value": "Vatican City"
},
{
"operation": "Now on",
"time": "355.768",
"value": "Vatican City"
},
{
"operation": "Page search",
"time": "373.949",
"value": "home of the roman catholic church"
},
{
"operation": "search",
"time": "473.976",
"value": "home of the roman catholic church"
},
{
"operation": "Now on",
"time": "477.351",
"value": "undefined"
},
{
"operation": "finish",
"time": "494.054",
"value": "finish"
}
]
}
|
75458
|
Territorial police force is gaining increased official use, and the phrase Home Office (a ministerial department) Police is commonly used.
|
{
"evidence": [
"Territorial police force_sentence_9",
"Territorial police force_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:\nTerritorial police force is gaining increased official use, and the phrase Home Office (a ministerial department) Police is commonly used.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Territorial police force_sentence_9",
"Territorial police force_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2122,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.428",
"value": "?search="
},
{
"operation": "search",
"time": "28.153",
"value": "Territorial police force"
},
{
"operation": "Now on",
"time": "29.173",
"value": "Territorial police force"
},
{
"operation": "Highlighting",
"time": "54.906",
"value": "Territorial police force_sentence_9"
},
{
"operation": "Highlighting",
"time": "57.376",
"value": "Territorial police force_sentence_10"
},
{
"operation": "finish",
"time": "72.619",
"value": "finish"
}
]
}
|
exfever_5999
|
Penny Dreadful is a comedy-drama television series created for an American premium television network owned by Paramount Media Networks and Sky by John Logan.
|
{
"evidence": [
"Penny Dreadful (TV series)",
"Showtime (TV network)"
],
"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:\nPenny Dreadful is a comedy-drama television series created for an American premium television network owned by Paramount Media Networks and Sky by John Logan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Penny Dreadful (TV series)",
"Showtime (TV network)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2123,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
62412
|
Beita District is one of three urban districts which is used to refer to several unrelated political divisions in both ancient and modern China.
|
{
"evidence": [
"Beita District_sentence_0",
"District (China)_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:\nBeita District is one of three urban districts which is used to refer to several unrelated political divisions in both ancient and modern China.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Beita District_sentence_0",
"District (China)_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2124,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.725",
"value": "?search="
},
{
"operation": "search",
"time": "10.253",
"value": "Beita District"
},
{
"operation": "Now on",
"time": "10.921",
"value": "Beita District"
},
{
"operation": "Highlighting",
"time": "16.971",
"value": "Beita District_sentence_0"
},
{
"operation": "hyperlink",
"time": "23.938",
"value": "districts"
},
{
"operation": "Now on",
"time": "24.89",
"value": "District (China)"
},
{
"operation": "Highlighting",
"time": "28.978",
"value": "District (China)_sentence_3"
},
{
"operation": "finish",
"time": "37.385",
"value": "finish"
}
]
}
|
exfever_1933
|
Harry Potter is a film series based on the eponymous novels by J. K. Rowling and is distributed by an American film and entertainment company.
|
{
"evidence": [
"Harry Potter (film series)",
"Warner Bros."
],
"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:\nHarry Potter is a film series based on the eponymous novels by J. K. Rowling and is distributed by an American film and entertainment company.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Harry Potter (film series)",
"Warner Bros."
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2125,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2301
|
To the Bone was the first American drama film portrayed by Lily Collins.
|
{
"evidence": [
"To the Bone (film)",
"Marti Noxon",
"Bravo (U.S. TV network)"
],
"label": "NOT ENOUGH INFO"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nTo the Bone was the first American drama film portrayed by Lily Collins.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"To the Bone (film)",
"Marti Noxon",
"Bravo (U.S. TV network)"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2126,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
40219
|
Agnes Jebet Tirop, a Kenyan long-distance runner mainly competing in 5000-meter races and cross country running, won the Kenyan cross country junior title at the 2014 Kenyan Cross Country Championships after being second place during the 2013 African Cross Country Championships.
|
{
"evidence": [
"Agnes Jebet Tirop_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:\nAgnes Jebet Tirop, a Kenyan long-distance runner mainly competing in 5000-meter races and cross country running, won the Kenyan cross country junior title at the 2014 Kenyan Cross Country Championships after being second place during the 2013 African Cross Country Championships.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Agnes Jebet Tirop_sentence_9"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2127,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0",
"value": "?search="
},
{
"operation": "search",
"time": "0.023",
"value": "Agnes Jebet Tirop"
},
{
"operation": "Now on",
"time": "0.024",
"value": "Agnes Jebet Tirop"
},
{
"operation": "Highlighting",
"time": "0.028",
"value": "Agnes Jebet Tirop_sentence_0"
},
{
"operation": "Page search",
"time": "0.051",
"value": "2014"
},
{
"operation": "Highlighting",
"time": "0.086",
"value": "Agnes Jebet Tirop_sentence_12"
},
{
"operation": "Page search",
"time": "0.102",
"value": "2013"
},
{
"operation": "Highlighting",
"time": "0.111",
"value": "Agnes Jebet Tirop_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "0.14",
"value": "Agnes Jebet Tirop_sentence_9"
},
{
"operation": "Page search",
"time": "0.187",
"value": "2013"
},
{
"operation": "Highlighting",
"time": "0.202",
"value": "Agnes Jebet Tirop_sentence_9"
},
{
"operation": "Highlighting deleted",
"time": "0.211",
"value": "Agnes Jebet Tirop_sentence_12"
},
{
"operation": "Highlighting",
"time": "0.214",
"value": "Agnes Jebet Tirop_item_0_0"
},
{
"operation": "Highlighting",
"time": "0.214",
"value": "Agnes Jebet Tirop_section_1"
},
{
"operation": "Highlighting deleted",
"time": "0.215",
"value": "Agnes Jebet Tirop_item_0_0"
},
{
"operation": "Highlighting deleted",
"time": "0.215",
"value": "Agnes Jebet Tirop_section_1"
},
{
"operation": "Highlighting deleted",
"time": "0.219",
"value": "Agnes Jebet Tirop_sentence_0"
},
{
"operation": "finish",
"time": "0.26",
"value": "finish"
},
{
"operation": "Now on",
"time": "1620680513.5",
"value": "Agnes Jebet Tirop"
},
{
"operation": "finish",
"time": "1620680545.9",
"value": "finish"
}
]
}
|
20718
|
Interleukin-12 subunit alpha is a protein expressed by activated dendritic cells (DC), macrophages, and monocytes and plays an important role in sustaining IFNγ production and the generation and encoded by the IL12A gene which encodes a subunit of the cytokine Interleukin 12 (IL-12) that acts on T and natural killer cells and is required for the T-cell-dependent induction of interferon gamma .
|
{
"evidence": [
"IL12A_sentence_0",
"IL12A_sentence_1",
"IL12A_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:\nInterleukin-12 subunit alpha is a protein expressed by activated dendritic cells (DC), macrophages, and monocytes and plays an important role in sustaining IFNγ production and the generation and encoded by the IL12A gene which encodes a subunit of the cytokine Interleukin 12 (IL-12) that acts on T and natural killer cells and is required for the T-cell-dependent induction of interferon gamma .\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"IL12A_sentence_0",
"IL12A_sentence_1",
"IL12A_sentence_3"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2128,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.632",
"value": "?search="
},
{
"operation": "search",
"time": "6.974",
"value": "Interleukin-12 subunit alpha"
},
{
"operation": "Now on",
"time": "8.885",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "12.835",
"value": "IL12A"
},
{
"operation": "Now on",
"time": "13.502",
"value": "IL12A"
},
{
"operation": "Highlighting",
"time": "30.507",
"value": "IL12A_sentence_0"
},
{
"operation": "Highlighting",
"time": "45.042",
"value": "IL12A_sentence_1"
},
{
"operation": "Highlighting",
"time": "61.455",
"value": "IL12A_sentence_3"
},
{
"operation": "hyperlink",
"time": "80.767",
"value": "protein"
},
{
"operation": "Now on",
"time": "82.762",
"value": "Protein"
},
{
"operation": "Page search",
"time": "92.922",
"value": "dendritic"
},
{
"operation": "back-button-clicked",
"time": "115.271",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "115.373",
"value": "IL12A"
},
{
"operation": "finish",
"time": "133.731",
"value": "finish"
}
]
}
|
60965
|
Marzili Funicular had construction begin in March 1885 and the tracks had a length of 112 metres.
|
{
"evidence": [
"Marzili Funicular_sentence_3",
"Marzili Funicular_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:\nMarzili Funicular had construction begin in March 1885 and the tracks had a length of 112 metres.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Marzili Funicular_sentence_3",
"Marzili Funicular_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2129,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.715",
"value": "?search="
},
{
"operation": "search",
"time": "14.806",
"value": "Marzili Funicular"
},
{
"operation": "Now on",
"time": "15.461",
"value": "Marzili Funicular"
},
{
"operation": "Highlighting",
"time": "22.979",
"value": "Marzili Funicular_sentence_3"
},
{
"operation": "Highlighting",
"time": "31.306",
"value": "Marzili Funicular_sentence_4"
},
{
"operation": "finish",
"time": "43.442",
"value": "finish"
}
]
}
|
6465
|
Along the northwestern coast of Loch Indaal are the villages of Bruichladdich and Port Charlotte, while along its northeastern shore is the tiny village of Bridgend and on its southeastern shore is the island capital of Bowmore, a small town on the Scottish island of Islay.
|
{
"evidence": [
"Loch Indaal_sentence_2",
"Loch Indaal_sentence_3",
"Bowmore_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:\nAlong the northwestern coast of Loch Indaal are the villages of Bruichladdich and Port Charlotte, while along its northeastern shore is the tiny village of Bridgend and on its southeastern shore is the island capital of Bowmore, a small town on the Scottish island of Islay.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Loch Indaal_sentence_2",
"Loch Indaal_sentence_3",
"Bowmore_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2130,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.863",
"value": "?search="
},
{
"operation": "search",
"time": "19.006",
"value": "northwestern coast of Loch Indaal"
},
{
"operation": "Now on",
"time": "19.644",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "21.588",
"value": "Loch Indaal"
},
{
"operation": "Now on",
"time": "22.165",
"value": "Loch Indaal"
},
{
"operation": "Highlighting",
"time": "28.472",
"value": "Loch Indaal_sentence_2"
},
{
"operation": "Highlighting",
"time": "40.964",
"value": "Loch Indaal_sentence_3"
},
{
"operation": "hyperlink",
"time": "46.285",
"value": "Bowmore"
},
{
"operation": "Now on",
"time": "46.715",
"value": "Bowmore"
},
{
"operation": "Highlighting",
"time": "50.635",
"value": "Bowmore_sentence_2"
},
{
"operation": "finish",
"time": "66.949",
"value": "finish"
}
]
}
|
35220
|
Peter Fraize studied piano in the New England Conservatory of Music in Boston and later studied at the Royal Conservatory of The Hague under Dutch saxophonist Leo van Oostrom.
|
{
"evidence": [
"Peter Fraize_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nPeter Fraize studied piano in the New England Conservatory of Music in Boston and later studied at the Royal Conservatory of The Hague under Dutch saxophonist Leo van Oostrom.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Peter Fraize_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2131,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.309",
"value": "?search="
},
{
"operation": "search",
"time": "11.408",
"value": "Peter Fraize"
},
{
"operation": "Now on",
"time": "12.394",
"value": "Peter Fraize"
},
{
"operation": "Highlighting",
"time": "28.16",
"value": "Peter Fraize_sentence_1"
},
{
"operation": "Highlighting",
"time": "35.725",
"value": "Peter Fraize_sentence_2"
},
{
"operation": "Highlighting deleted",
"time": "40.521",
"value": "Peter Fraize_sentence_2"
},
{
"operation": "finish",
"time": "54.165",
"value": "finish"
}
]
}
|
exfever_3020
|
Dolly Sidhwani is the wife of Ritesh Sidhwani.
|
{
"evidence": [
"Zindagi Na Milegi Dobara",
"Ritesh Sidhwani"
],
"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:\nDolly Sidhwani is the wife of Ritesh Sidhwani.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Zindagi Na Milegi Dobara",
"Ritesh Sidhwani"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2132,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7711
|
Me Before You is a 2016 romantic drama film starring several artists, including an English actress who plays Hubert Page in a 2011 drama film directed by Rodrigo García.
|
{
"evidence": [
"Me Before You (film)",
"Janet McTeer",
"Albert Nobbs"
],
"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:\nMe Before You is a 2016 romantic drama film starring several artists, including an English actress who plays Hubert Page in a 2011 drama film directed by Rodrigo García.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Me Before You (film)",
"Janet McTeer",
"Albert Nobbs"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2133,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2043
|
Elizabeth Hughes Meriwether is known for creating an Asian television sitcom for the Fox Broadcasting Company that stars an American actor, comedian, film producer, and screenwriter.
|
{
"evidence": [
"Elizabeth Meriwether",
"New Girl",
"Jake Johnson"
],
"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:\nElizabeth Hughes Meriwether is known for creating an Asian television sitcom for the Fox Broadcasting Company that stars an American actor, comedian, film producer, and screenwriter.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Elizabeth Meriwether",
"New Girl",
"Jake Johnson"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2134,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_526
|
Andrew Jackson gained fame as a general in the United States Army and served in both houses of the legislature of the federal government of the United Kingdom.
|
{
"evidence": [
"Andrew Jackson",
"United States Congress"
],
"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:\nAndrew Jackson gained fame as a general in the United States Army and served in both houses of the legislature of the federal government of the United Kingdom.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Andrew Jackson",
"United States Congress"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2135,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
76571
|
The 1989–90 ACB season had eight teams, including Taugrés, Valvi Girona, and Caixa Ourense, and it was the first year using the current competition format.
|
{
"evidence": [
"1989–90 ACB season_sentence_1"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe 1989–90 ACB season had eight teams, including Taugrés, Valvi Girona, and Caixa Ourense, and it was the first year using the current competition format.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"1989–90 ACB season_sentence_1"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2136,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.036",
"value": "?search="
},
{
"operation": "search",
"time": "701.163",
"value": "1989–90 ACB season"
},
{
"operation": "Now on",
"time": "702.417",
"value": "1989–90 ACB season"
},
{
"operation": "Highlighting",
"time": "706.31",
"value": "1989–90 ACB season_cell_3_1_0"
},
{
"operation": "Highlighting",
"time": "706.312",
"value": "1989–90 ACB season_section_0"
},
{
"operation": "Highlighting",
"time": "706.313",
"value": "1989–90 ACB season_section_2"
},
{
"operation": "Highlighting",
"time": "706.313",
"value": "1989–90 ACB season_section_3"
},
{
"operation": "Highlighting",
"time": "706.314",
"value": "1989–90 ACB season_header_cell_3_0_0"
},
{
"operation": "Highlighting",
"time": "706.695",
"value": "1989–90 ACB season_cell_3_2_0"
},
{
"operation": "Highlighting",
"time": "707.013",
"value": "1989–90 ACB season_cell_3_3_0"
},
{
"operation": "Highlighting",
"time": "707.566",
"value": "1989–90 ACB season_cell_3_4_0"
},
{
"operation": "Highlighting",
"time": "707.853",
"value": "1989–90 ACB season_cell_3_5_0"
},
{
"operation": "Highlighting",
"time": "708.349",
"value": "1989–90 ACB season_cell_3_6_0"
},
{
"operation": "Highlighting",
"time": "708.893",
"value": "1989–90 ACB season_cell_3_7_0"
},
{
"operation": "Highlighting",
"time": "709.357",
"value": "1989–90 ACB season_cell_3_8_0"
},
{
"operation": "Highlighting",
"time": "723.782",
"value": "1989–90 ACB season_sentence_1"
},
{
"operation": "finish",
"time": "744.364",
"value": "finish"
}
]
}
|
64264
|
Skottie Young has never drawn covers for books, however he wrote and drew a Rocket Raccoon solo series for Marvel Comics (the brand name and primary imprint of Marvel Worldwide Inc.) starting in June 2016.
|
{
"evidence": [
"Skottie Young_sentence_6"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nSkottie Young has never drawn covers for books, however he wrote and drew a Rocket Raccoon solo series for Marvel Comics (the brand name and primary imprint of Marvel Worldwide Inc.) starting in June 2016.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Skottie Young_sentence_6"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2137,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.188",
"value": "?search="
},
{
"operation": "search",
"time": "11.706",
"value": "Skottie Young"
},
{
"operation": "Now on",
"time": "12.862",
"value": "Skottie Young"
},
{
"operation": "Page search",
"time": "44.609",
"value": "covers"
},
{
"operation": "Highlighting",
"time": "47.136",
"value": "Skottie Young_sentence_6"
},
{
"operation": "Highlighting",
"time": "89.438",
"value": "Skottie Young_sentence_6"
},
{
"operation": "finish",
"time": "148.202",
"value": "finish"
}
]
}
|
36041
|
In the 1973 Philadelphia Phillies season, Steve Carlton failed to recover from a spring training illness while Jim Lonborg and Ken Brett won 13 games each after being obtained from a trade with the Milwaukee Brewers, an American professional baseball team based in Milwaukee.
|
{
"evidence": [
"1973 Philadelphia Phillies season_sentence_3",
"1973 Philadelphia Phillies season_sentence_2",
"Milwaukee Brewers_sentence_2"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn the 1973 Philadelphia Phillies season, Steve Carlton failed to recover from a spring training illness while Jim Lonborg and Ken Brett won 13 games each after being obtained from a trade with the Milwaukee Brewers, an American professional baseball team based in Milwaukee.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"1973 Philadelphia Phillies season_sentence_3",
"1973 Philadelphia Phillies season_sentence_2",
"Milwaukee Brewers_sentence_2"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2138,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.734",
"value": "?search="
},
{
"operation": "search",
"time": "10.915",
"value": "1973 Philadelphia Phillies season"
},
{
"operation": "Now on",
"time": "11.95",
"value": "1973 Philadelphia Phillies season"
},
{
"operation": "Highlighting",
"time": "51.863",
"value": "1973 Philadelphia Phillies season_sentence_3"
},
{
"operation": "Highlighting",
"time": "52.941",
"value": "1973 Philadelphia Phillies season_sentence_2"
},
{
"operation": "hyperlink",
"time": "54.944",
"value": "Milwaukee Brewers"
},
{
"operation": "Now on",
"time": "56.065",
"value": "Milwaukee Brewers"
},
{
"operation": "Highlighting",
"time": "59.658",
"value": "Milwaukee Brewers_sentence_2"
},
{
"operation": "search",
"time": "70.291",
"value": "steve cARLTON"
},
{
"operation": "Now on",
"time": "71.25",
"value": "Steve Carlton"
},
{
"operation": "finish",
"time": "237.866",
"value": "finish"
}
]
}
|
exfever_3732
|
A glacier forms where the accumulation of rain exceeds its destruction, which includes ablative materials like a component of a structure that slows the spread of the effects of smoke without system activation.
|
{
"evidence": [
"Glacier",
"Ablation",
"Passive fire protection"
],
"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 glacier forms where the accumulation of rain exceeds its destruction, which includes ablative materials like a component of a structure that slows the spread of the effects of smoke without system activation.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Glacier",
"Ablation",
"Passive fire protection"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2139,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_2529
|
Star Wars is a 1977 American epic space opera film starring an Indian actress best known for playing Princess Leia in the Star Wars films.
|
{
"evidence": [
"Star Wars (film)",
"Carrie Fisher"
],
"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:\nStar Wars is a 1977 American epic space opera film starring an Indian actress best known for playing Princess Leia in the Star Wars films.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Star Wars (film)",
"Carrie Fisher"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2140,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_8611
|
Ultimate Spider-Man is an American animated television series with a genre of speculative fiction examining the adventures, personalities, and ethics of costumed crime fighters, who often possess superhuman powers.
|
{
"evidence": [
"Ultimate Spider-Man (TV series)",
"Superhero fiction"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nUltimate Spider-Man is an American animated television series with a genre of speculative fiction examining the adventures, personalities, and ethics of costumed crime fighters, who often possess superhuman powers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Ultimate Spider-Man (TV series)",
"Superhero fiction"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2141,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7409
|
The Fate of the Furious is a 2017 American action film directed by F. Gary Gray and written by an American producer who has also written the screenplay for Gang Related.
|
{
"evidence": [
"The Fate of the Furious",
"Chris Morgan (writer)"
],
"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 Fate of the Furious is a 2017 American action film directed by F. Gary Gray and written by an American producer who has also written the screenplay for Gang Related.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Fate of the Furious",
"Chris Morgan (writer)"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2142,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
90527
|
The Yonge Street facade will use the 1905 Canadian Bank of Commerce headquarters, and the Massey Tower will rise to 50 storeys and stand 208.3 metres (683 ft) in height.
|
{
"evidence": [
"Massey Tower_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe Yonge Street facade will use the 1905 Canadian Bank of Commerce headquarters, and the Massey Tower will rise to 50 storeys and stand 208.3 metres (683 ft) in height.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"Massey Tower_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2143,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.001",
"value": "?search="
},
{
"operation": "search",
"time": "25.243",
"value": "Massey Tower"
},
{
"operation": "Now on",
"time": "26.445",
"value": "Massey Tower"
},
{
"operation": "Page search",
"time": "61.087",
"value": "Yonge Street facade"
},
{
"operation": "Highlighting",
"time": "63.483",
"value": "Massey Tower_sentence_3"
},
{
"operation": "Highlighting",
"time": "85.544",
"value": "Massey Tower_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "88.008",
"value": "Massey Tower_sentence_3"
},
{
"operation": "finish",
"time": "93.847",
"value": "finish"
}
]
}
|
43887
|
The southeast of the village Rawu (has an altitude of over 3000 meters m) is a small chorten and a big temple overlooking the Ranwu River, and the second river is known as Rawak tso.
|
{
"evidence": [
"Rawu_sentence_4"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe southeast of the village Rawu (has an altitude of over 3000 meters m) is a small chorten and a big temple overlooking the Ranwu River, and the second river is known as Rawak tso.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Rawu_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2144,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.847",
"value": "?search="
},
{
"operation": "search",
"time": "33.589",
"value": "Rawu"
},
{
"operation": "Now on",
"time": "35.985",
"value": "Rawu"
},
{
"operation": "Highlighting",
"time": "46.003",
"value": "Rawu_sentence_4"
},
{
"operation": "finish",
"time": "86.986",
"value": "finish"
}
]
}
|
28272
|
The creation of the Congress for the Republic was declared in 2001, it was aimed to install a republican form of government but it was banned during the Ben Ali presidency.
|
{
"evidence": [
"Congress for the Republic_sentence_2",
"Congress for the Republic_sentence_6",
"Congress for the Republic_sentence_11"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nThe creation of the Congress for the Republic was declared in 2001, it was aimed to install a republican form of government but it was banned during the Ben Ali presidency.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Congress for the Republic_sentence_2",
"Congress for the Republic_sentence_6",
"Congress for the Republic_sentence_11"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2145,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.04",
"value": "?search="
},
{
"operation": "search",
"time": "27.804",
"value": "Contains...Ben Ali presidency"
},
{
"operation": "Now on",
"time": "29.264",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "39.59",
"value": "Zine El Abidine Ben Ali"
},
{
"operation": "Now on",
"time": "40.353",
"value": "Zine El Abidine Ben Ali"
},
{
"operation": "Page search",
"time": "61.172",
"value": "congress for the republic"
},
{
"operation": "search",
"time": "87.212",
"value": "Congress for the Republic"
},
{
"operation": "Now on",
"time": "87.883",
"value": "Congress for the Republic"
},
{
"operation": "Highlighting",
"time": "94.801",
"value": "Congress for the Republic_sentence_2"
},
{
"operation": "Highlighting",
"time": "115.118",
"value": "Congress for the Republic_sentence_6"
},
{
"operation": "Highlighting",
"time": "132.874",
"value": "Congress for the Republic_sentence_11"
},
{
"operation": "finish",
"time": "158.39",
"value": "finish"
}
]
}
|
exfever_8757
|
The Taj Mahal is an ivory-white marble mausoleum on the right bank of the river Yamuna in the Indian state of Uttar Pradesh, about 250 kilometers south of the national capital New Delhi and 335 km west of the state capital Lucknow.
|
{
"evidence": [
"Taj Mahal",
"Agra"
],
"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 Taj Mahal is an ivory-white marble mausoleum on the right bank of the river Yamuna in the Indian state of Uttar Pradesh, about 250 kilometers south of the national capital New Delhi and 335 km west of the state capital Lucknow.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Taj Mahal",
"Agra"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2146,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_4008
|
Harvey Kurtzman was an American cartoonist and editor who began to work on editing the anthology of war comics published bi-monthly by EC Comics in the early 1960s.
|
{
"evidence": [
"Harvey Kurtzman",
"Two-Fisted Tales"
],
"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:\nHarvey Kurtzman was an American cartoonist and editor who began to work on editing the anthology of war comics published bi-monthly by EC Comics in the early 1960s.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Harvey Kurtzman",
"Two-Fisted Tales"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2147,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
41966
|
Nangolo Ithete went into exile with SWAPO to Dar es Salaam, Tanzania in 1963 and studied in Russia until 1975.
|
{
"evidence": [
"Nangolo Ithete_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nNangolo Ithete went into exile with SWAPO to Dar es Salaam, Tanzania in 1963 and studied in Russia until 1975.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Nangolo Ithete_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2148,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.215",
"value": "?search="
},
{
"operation": "search",
"time": "10.023",
"value": "Nangolo Ithete"
},
{
"operation": "Now on",
"time": "10.727",
"value": "Nangolo Ithete"
},
{
"operation": "Highlighting",
"time": "157.653",
"value": "Nangolo Ithete_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "158.262",
"value": "Nangolo Ithete_sentence_0"
},
{
"operation": "Highlighting",
"time": "170.333",
"value": "Nangolo Ithete_sentence_2"
},
{
"operation": "Highlighting",
"time": "174.812",
"value": "Nangolo Ithete_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "175.89",
"value": "Nangolo Ithete_sentence_2"
},
{
"operation": "finish",
"time": "183.256",
"value": "finish"
}
]
}
|
exfever_325
|
Jordan James Spence is an English professional footballer who previously played for an English professional football club that plays its home matches in Stratford.
|
{
"evidence": [
"Jordan Spence",
"West Ham United F.C."
],
"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:\nJordan James Spence is an English professional footballer who previously played for an English professional football club that plays its home matches in Stratford.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Jordan Spence",
"West Ham United F.C."
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2149,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_3765
|
Princess Agents is a 2017 Chinese television series starring a Chinese actor who rose to fame for his role in Scarlet Heart.
|
{
"evidence": [
"Princess Agents",
"Lin Gengxin"
],
"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:\nPrincess Agents is a 2017 Chinese television series starring a Chinese actor who rose to fame for his role in Scarlet Heart.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Princess Agents",
"Lin Gengxin"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2150,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5161
|
My Everything is the second studio album by American singer Ariana Grande, released on August 25, 2014, by a New York City-based American record label owned by a Dutch–American multinational music corporation.
|
{
"evidence": [
"My Everything (Ariana Grande album)",
"Republic Records",
"Universal Music Group"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMy Everything is the second studio album by American singer Ariana Grande, released on August 25, 2014, by a New York City-based American record label owned by a Dutch–American multinational music corporation.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"My Everything (Ariana Grande album)",
"Republic Records",
"Universal Music Group"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2151,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
5006
|
Lying on the border of the intermediate and dry zones, Kandalama is nestled at the foot of Ereulagala and Dikkandahena and serves as a location of a 2700-year old irrigation tank.
|
{
"evidence": [
"Kandalama_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:\nLying on the border of the intermediate and dry zones, Kandalama is nestled at the foot of Ereulagala and Dikkandahena and serves as a location of a 2700-year old irrigation tank.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Kandalama_sentence_4"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2152,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "10.055",
"value": "Kandalama"
},
{
"operation": "Page search",
"time": "21.282",
"value": "border"
},
{
"operation": "Highlighting",
"time": "24.594",
"value": "Kandalama_sentence_6"
},
{
"operation": "Highlighting",
"time": "31.077",
"value": "Kandalama_sentence_5"
},
{
"operation": "Highlighting",
"time": "56.677",
"value": "Kandalama_sentence_4"
},
{
"operation": "Highlighting deleted",
"time": "58.379",
"value": "Kandalama_sentence_5"
},
{
"operation": "Highlighting deleted",
"time": "58.869",
"value": "Kandalama_sentence_6"
},
{
"operation": "finish",
"time": "73.785",
"value": "finish"
}
]
}
|
10180
|
SN 1994I is a supernova in the Whirlpool Galaxy that lies in the Canes Venatici constellation.
|
{
"evidence": [
"SN 1994I_sentence_0",
"Whirlpool Galaxy_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:\nSN 1994I is a supernova in the Whirlpool Galaxy that lies in the Canes Venatici constellation.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"SN 1994I_sentence_0",
"Whirlpool Galaxy_sentence_1"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2153,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.772",
"value": "?search="
},
{
"operation": "search",
"time": "10.663",
"value": "SN 1994I"
},
{
"operation": "Now on",
"time": "12",
"value": "SN 1994I"
},
{
"operation": "Highlighting",
"time": "20.978",
"value": "SN 1994I_sentence_0"
},
{
"operation": "hyperlink",
"time": "28.634",
"value": "Whirlpool Galaxy"
},
{
"operation": "Now on",
"time": "29.471",
"value": "Whirlpool Galaxy"
},
{
"operation": "Highlighting",
"time": "32.066",
"value": "Whirlpool Galaxy_sentence_1"
},
{
"operation": "finish",
"time": "39.156",
"value": "finish"
}
]
}
|
89345
|
James Akintunde has joined the Chester for the 2015–16 season.
|
{
"evidence": [
"James Akintunde_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:\nJames Akintunde has joined the Chester for the 2015–16 season.\n",
"role": "user"
}
] |
Numerical Reasoning
|
{
"ground_truth": {
"evidence": [
"James Akintunde_sentence_7"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2154,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.033",
"value": "?search="
},
{
"operation": "search",
"time": "6.333",
"value": "James Akintunde"
},
{
"operation": "Now on",
"time": "8.086",
"value": "James Akintunde"
},
{
"operation": "Highlighting",
"time": "18.825",
"value": "James Akintunde_cell_1_10_0"
},
{
"operation": "Highlighting",
"time": "18.827",
"value": "James Akintunde_section_1"
},
{
"operation": "Highlighting",
"time": "18.909",
"value": "James Akintunde_header_cell_1_0_0"
},
{
"operation": "Highlighting",
"time": "19.644",
"value": "James Akintunde_cell_1_10_1"
},
{
"operation": "Highlighting",
"time": "19.722",
"value": "James Akintunde_header_cell_1_5_0"
},
{
"operation": "Highlighting",
"time": "20.06",
"value": "James Akintunde_cell_1_11_0"
},
{
"operation": "Highlighting deleted",
"time": "28.928",
"value": "James Akintunde_cell_1_11_0"
},
{
"operation": "Highlighting deleted",
"time": "29.233",
"value": "James Akintunde_cell_1_10_1"
},
{
"operation": "Highlighting deleted",
"time": "29.307",
"value": "James Akintunde_header_cell_1_5_0"
},
{
"operation": "Highlighting deleted",
"time": "29.704",
"value": "James Akintunde_cell_1_10_0"
},
{
"operation": "Highlighting deleted",
"time": "29.709",
"value": "James Akintunde_section_1"
},
{
"operation": "Highlighting deleted",
"time": "29.781",
"value": "James Akintunde_header_cell_1_0_0"
},
{
"operation": "Highlighting",
"time": "31.66",
"value": "James Akintunde_sentence_6"
},
{
"operation": "Highlighting deleted",
"time": "37.692",
"value": "James Akintunde_sentence_6"
},
{
"operation": "Highlighting",
"time": "38.236",
"value": "James Akintunde_sentence_7"
},
{
"operation": "finish",
"time": "42.596",
"value": "finish"
}
]
}
|
exfever_7233
|
Murder on the Orient Express is a 2017 mystery film directed by a British actor and filmmaker.
|
{
"evidence": [
"Murder on the Orient Express (2017 film)",
"Kenneth Branagh"
],
"label": "SUPPORT"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nMurder on the Orient Express is a 2017 mystery film directed by a British actor and filmmaker.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Murder on the Orient Express (2017 film)",
"Kenneth Branagh"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2155,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_7003
|
Moscovium is one of 23 known chemical elements that do not occur naturally on Earth with the symbol Mc and atomic number 115.
|
{
"evidence": [
"Moscovium",
"Synthetic element"
],
"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:\nMoscovium is one of 23 known chemical elements that do not occur naturally on Earth with the symbol Mc and atomic number 115.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Moscovium",
"Synthetic element"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2156,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6455
|
Before We Go was the thirty-first film that had its world premiere in the Special Presentations section.
|
{
"evidence": [
"Before We Go",
"Chris Evans (actor)",
"Graphic novel"
],
"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:\nBefore We Go was the thirty-first film that had its world premiere in the Special Presentations section.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Before We Go",
"Chris Evans (actor)",
"Graphic novel"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2157,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
2858
|
McSherry's career, as a midfielder, started with Kilmarnock in 1970 where he had 118 appearances for the club and scored 8 goals.
|
{
"evidence": [
"John McSherry_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:\nMcSherry's career, as a midfielder, started with Kilmarnock in 1970 where he had 118 appearances for the club and scored 8 goals.\n",
"role": "user"
}
] |
Entity Disambiguation
|
{
"ground_truth": {
"evidence": [
"John McSherry_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2158,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "search",
"time": "11.875",
"value": "McSherry"
},
{
"operation": "hyperlink",
"time": "30.781",
"value": "John McSherry"
},
{
"operation": "search",
"time": "51.438",
"value": "kilmarnock"
},
{
"operation": "hyperlink",
"time": "66.201",
"value": "Kilmarnock F.C."
},
{
"operation": "hyperlink",
"time": "83.482",
"value": "7 Players Kilmarnock F.C._section_9"
},
{
"operation": "hyperlink",
"time": "109.038",
"value": "List of Kilmarnock F.C. seasons"
},
{
"operation": "search",
"time": "166.931",
"value": "kilmarnock F.C. season 1970"
},
{
"operation": "hyperlink",
"time": "173.988",
"value": "Stirling Albion F.C."
},
{
"operation": "hyperlink",
"time": "192.32",
"value": "1.3 The 1970–1980s Stirling Albion F.C._section_3"
},
{
"operation": "Page search",
"time": "205.699",
"value": "mcsherry"
},
{
"operation": "search",
"time": "219.196",
"value": "mcsherry"
},
{
"operation": "search",
"time": "232.555",
"value": "david mcsherry"
},
{
"operation": "back-button-clicked",
"time": "238.985",
"value": "back-button-clicked"
},
{
"operation": "hyperlink",
"time": "275.347",
"value": "John McSherry"
},
{
"operation": "Highlighting",
"time": "277.658",
"value": "John McSherry_sentence_0"
},
{
"operation": "finish",
"time": "291.469",
"value": "finish"
}
]
}
|
exfever_1412
|
The Economic Times is an Indian English-language business-focused weekly newspaper and is published in the capital city of the Indian state of Maharashtra.
|
{
"evidence": [
"The Economic Times",
"Mumbai"
],
"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 Economic Times is an Indian English-language business-focused weekly newspaper and is published in the capital city of the Indian state of Maharashtra.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Economic Times",
"Mumbai"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2159,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
18368
|
Nevada was the first state to sanction the use of the gas chamber with its first execution in May 1924.
|
{
"evidence": [
"Gas chamber_sentence_34"
],
"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:\nNevada was the first state to sanction the use of the gas chamber with its first execution in May 1924.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"Gas chamber_sentence_34"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2160,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.34",
"value": "?search="
},
{
"operation": "search",
"time": "17.326",
"value": "Nevada"
},
{
"operation": "Now on",
"time": "18.881",
"value": "Nevada"
},
{
"operation": "Page search",
"time": "25.723",
"value": "gas chamber"
},
{
"operation": "search",
"time": "55.167",
"value": "evada was the first state to sanction"
},
{
"operation": "Now on",
"time": "57.084",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "61.403",
"value": "evada was the first stage to sanctions"
},
{
"operation": "Now on",
"time": "61.818",
"value": "undefined"
},
{
"operation": "search",
"time": "78.429",
"value": "Death penalty US"
},
{
"operation": "Now on",
"time": "80.089",
"value": "Death penalty US"
},
{
"operation": "Page search",
"time": "90.151",
"value": "nevada"
},
{
"operation": "Page search",
"time": "137.183",
"value": "may 1924"
},
{
"operation": "Page search",
"time": "187.21",
"value": "gas chamber"
},
{
"operation": "Page search",
"time": "190.027",
"value": "gas chamber"
},
{
"operation": "search",
"time": "195.877",
"value": "Gas chamber"
},
{
"operation": "Now on",
"time": "196.968",
"value": "Gas chamber"
},
{
"operation": "search",
"time": "206.397",
"value": "may 1924"
},
{
"operation": "Now on",
"time": "207.341",
"value": "May 1924"
},
{
"operation": "search",
"time": "215.378",
"value": "nevada"
},
{
"operation": "Now on",
"time": "216.025",
"value": "Nevada"
},
{
"operation": "search",
"time": "230.617",
"value": "Gas chamber"
},
{
"operation": "Now on",
"time": "231.197",
"value": "Gas chamber"
},
{
"operation": "Page search",
"time": "280.261",
"value": "nevada"
},
{
"operation": "Highlighting",
"time": "303.824",
"value": "Gas chamber_sentence_34"
},
{
"operation": "hyperlink",
"time": "309.525",
"value": "Nevada"
},
{
"operation": "Now on",
"time": "310.096",
"value": "Nevada"
},
{
"operation": "Page search",
"time": "320.247",
"value": "gas chamber"
},
{
"operation": "finish",
"time": "358.889",
"value": "finish"
}
]
}
|
85563
|
Linggi River starts from Ampangan to Kg Jerlang, it flows through from Pantai to Ampangan and was not used as a route.
|
{
"evidence": [
"Linggi River_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:\nLinggi River starts from Ampangan to Kg Jerlang, it flows through from Pantai to Ampangan and was not used as a route.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Linggi River_sentence_5"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2161,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.308",
"value": "?search="
},
{
"operation": "search",
"time": "16.736",
"value": "Linggi River"
},
{
"operation": "Now on",
"time": "17.465",
"value": "Linggi River"
},
{
"operation": "Highlighting",
"time": "39.793",
"value": "Linggi River_sentence_3"
},
{
"operation": "Highlighting",
"time": "44.962",
"value": "Linggi River_sentence_4"
},
{
"operation": "Highlighting",
"time": "51.328",
"value": "Linggi River_sentence_5"
},
{
"operation": "finish",
"time": "55.665",
"value": "finish"
}
]
}
|
exfever_8727
|
Rebecca Maria Hall is a Canadian actress who got her breakthrough role in Christopher Nolan's 2006 mystery film, whose genre evokes excitement and suspense in the audience.
|
{
"evidence": [
"Rebecca Hall",
"The Prestige (film)",
"Thriller film"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRebecca Maria Hall is a Canadian actress who got her breakthrough role in Christopher Nolan's 2006 mystery film, whose genre evokes excitement and suspense in the audience.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Rebecca Hall",
"The Prestige (film)",
"Thriller film"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2162,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
29475
|
David Pingree, a professor at Brown University, believed that Antiochus of Athens lived during the end of the 2nd century AD.
|
{
"evidence": [
"Antiochus of Athens_sentence_2",
"David Pingree_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:\nDavid Pingree, a professor at Brown University, believed that Antiochus of Athens lived during the end of the 2nd century AD.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Antiochus of Athens_sentence_2",
"David Pingree_sentence_3"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2163,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0",
"value": "?search="
},
{
"operation": "search",
"time": "0.009",
"value": "David Pingree"
},
{
"operation": "Now on",
"time": "0.01",
"value": "David Pingree"
},
{
"operation": "Highlighting",
"time": "0.011",
"value": "David Pingree_sentence_0"
},
{
"operation": "Page search",
"time": "0.024",
"value": "Antiochus"
},
{
"operation": "Page search",
"time": "0.029",
"value": "Antiochus"
},
{
"operation": "Page search",
"time": "0.052",
"value": "athens"
},
{
"operation": "search",
"time": "0.165",
"value": "Antiochus of Athens lived during the end of the 2nd century AD"
},
{
"operation": "Now on",
"time": "0.168",
"value": "undefined"
},
{
"operation": "search",
"time": "0.183",
"value": "Antiochus of Athens"
},
{
"operation": "Now on",
"time": "0.184",
"value": "Antiochus of Athens"
},
{
"operation": "Page search",
"time": "0.19",
"value": "pingree"
},
{
"operation": "Highlighting",
"time": "0.194",
"value": "Antiochus of Athens_sentence_2"
},
{
"operation": "finish",
"time": "0.209",
"value": "finish"
},
{
"operation": "Now on",
"time": "1620222789.369",
"value": "David Pingree"
},
{
"operation": "Highlighting",
"time": "1620222798.755",
"value": "David Pingree_sentence_3"
},
{
"operation": "Highlighting deleted",
"time": "1620222799.975",
"value": "David Pingree_sentence_0"
},
{
"operation": "finish",
"time": "1620222803.723",
"value": "finish"
}
]
}
|
77534
|
Larry Santos released three albums in the 1970s and starred in the television show Hot Fudge, an American children's television series that was produced in Detroit by WXYZ-TV (owned and operated at the time by ABC) and distributed by the Lexington Broadcast Services Company, from 1976 to 1980.
|
{
"evidence": [
"Larry Santos_sentence_3",
"Larry Santos_sentence_4",
"Hot Fudge_sentence_3",
"Hot Fudge_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:\nLarry Santos released three albums in the 1970s and starred in the television show Hot Fudge, an American children's television series that was produced in Detroit by WXYZ-TV (owned and operated at the time by ABC) and distributed by the Lexington Broadcast Services Company, from 1976 to 1980.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Larry Santos_sentence_3",
"Larry Santos_sentence_4",
"Hot Fudge_sentence_3",
"Hot Fudge_sentence_4"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2164,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.109",
"value": "?search="
},
{
"operation": "search",
"time": "15.653",
"value": "Larry Santos"
},
{
"operation": "Now on",
"time": "16.348",
"value": "Larry Santos"
},
{
"operation": "Page search",
"time": "42.5",
"value": "albums"
},
{
"operation": "Highlighting",
"time": "45.748",
"value": "Larry Santos_sentence_3"
},
{
"operation": "Page search",
"time": "59.923",
"value": "Hot Fudge"
},
{
"operation": "Highlighting",
"time": "63.491",
"value": "Larry Santos_sentence_4"
},
{
"operation": "hyperlink",
"time": "76.627",
"value": "Hot Fudge"
},
{
"operation": "Now on",
"time": "77.362",
"value": "Hot Fudge"
},
{
"operation": "hyperlink",
"time": "78.802",
"value": "children's television series"
},
{
"operation": "Now on",
"time": "79.559",
"value": "Children's television series"
},
{
"operation": "back-button-clicked",
"time": "80.692",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "80.989",
"value": "Hot Fudge"
},
{
"operation": "Highlighting",
"time": "82.363",
"value": "Hot Fudge_sentence_3"
},
{
"operation": "Highlighting",
"time": "91.875",
"value": "Hot Fudge_sentence_4"
},
{
"operation": "finish",
"time": "101.194",
"value": "finish"
}
]
}
|
exfever_2021
|
John Fitzgerald Kennedy was an Australian politician whose majority of his work as president was concerned with relations with the Soviet Union and a country comprising the island of Isla de la Juventud and several minor archipelagos.
|
{
"evidence": [
"John F. Kennedy",
"Cuba"
],
"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:\nJohn Fitzgerald Kennedy was an Australian politician whose majority of his work as president was concerned with relations with the Soviet Union and a country comprising the island of Isla de la Juventud and several minor archipelagos.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"John F. Kennedy",
"Cuba"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2165,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5242
|
Mr. Freeze is a supervillain appearing in American comic books who was an expert in the production and behavior of materials at very low temperatures.
|
{
"evidence": [
"Mr. Freeze",
"Cryogenics"
],
"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:\nMr. Freeze is a supervillain appearing in American comic books who was an expert in the production and behavior of materials at very low temperatures.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Mr. Freeze",
"Cryogenics"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2166,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
70347
|
In 2002, Daniel Lalonde was hired by LVMH (Louis Vuitton Moët Hennessy) as president and CEO of LVMH Watches & Jewelry, and then he recruited celebrities, such as Leonardo DiCaprio and Uma Thurman to take part in watch campaigns.
|
{
"evidence": [
"Daniel Lalonde_sentence_1",
"LVMH_sentence_0",
"Daniel Lalonde_sentence_17"
],
"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 2002, Daniel Lalonde was hired by LVMH (Louis Vuitton Moët Hennessy) as president and CEO of LVMH Watches & Jewelry, and then he recruited celebrities, such as Leonardo DiCaprio and Uma Thurman to take part in watch campaigns.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Daniel Lalonde_sentence_1",
"LVMH_sentence_0",
"Daniel Lalonde_sentence_17"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2167,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.395",
"value": "?search="
},
{
"operation": "search",
"time": "26.926",
"value": "LVMH Watches & Jewelry"
},
{
"operation": "Now on",
"time": "28.221",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "36.331",
"value": "Daniel Lalonde"
},
{
"operation": "Now on",
"time": "37.598",
"value": "Daniel Lalonde"
},
{
"operation": "Page search",
"time": "47.93",
"value": "lvmh"
},
{
"operation": "Highlighting",
"time": "56.52",
"value": "Daniel Lalonde_sentence_1"
},
{
"operation": "hyperlink",
"time": "58.612",
"value": "LVMH"
},
{
"operation": "Now on",
"time": "59.404",
"value": "LVMH"
},
{
"operation": "Highlighting",
"time": "64.097",
"value": "LVMH_sentence_0"
},
{
"operation": "back-button-clicked",
"time": "66.67",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "66.846",
"value": "Daniel Lalonde"
},
{
"operation": "Page search",
"time": "87.866",
"value": "leonardo"
},
{
"operation": "Highlighting",
"time": "124.907",
"value": "Daniel Lalonde_sentence_17"
},
{
"operation": "search",
"time": "153.151",
"value": "leonardo ddicaprio lvmh"
},
{
"operation": "Now on",
"time": "153.975",
"value": "undefined"
},
{
"operation": "finish",
"time": "332.016",
"value": "finish"
}
]
}
|
exfever_8459
|
C.S Forester known for writing tales of naval warfare, such as the Hornblower novel including the seventh book in the series, which follows the first of the Horatio Hornblower novels.
|
{
"evidence": [
"C. S. Forester",
"A Ship of the Line",
"The Happy Return"
],
"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:\nC.S Forester known for writing tales of naval warfare, such as the Hornblower novel including the seventh book in the series, which follows the first of the Horatio Hornblower novels.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"C. S. Forester",
"A Ship of the Line",
"The Happy Return"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2168,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6134
|
Dangal is the second film produced by Aamir Khan.
|
{
"evidence": [
"Dangal (film)",
"Nitesh Tiwari",
"BRICS"
],
"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:\nDangal is the second film produced by Aamir Khan.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Dangal (film)",
"Nitesh Tiwari",
"BRICS"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2169,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6391
|
Stefanie Maria "Steffi" Graf ranked world No. 1 for being an American former professional player of a racket sport that can be played individually against a single opponent single or between two teams of two players each double.
|
{
"evidence": [
"Steffi Graf",
"Tennis"
],
"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:\nStefanie Maria \"Steffi\" Graf ranked world No. 1 for being an American former professional player of a racket sport that can be played individually against a single opponent single or between two teams of two players each double.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Steffi Graf",
"Tennis"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2170,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
24261
|
Isan people overran the Lao Wang and forced them into corvee (unpaid intermittent labour expected by landowners or monarchs) labour before they were integrated into Isan society; much of Isan was populated this way.
|
{
"evidence": [
"Isan people_sentence_38"
],
"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:\nIsan people overran the Lao Wang and forced them into corvee (unpaid intermittent labour expected by landowners or monarchs) labour before they were integrated into Isan society; much of Isan was populated this way.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Isan people_sentence_38"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2171,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.566",
"value": "?search="
},
{
"operation": "search",
"time": "27.593",
"value": "Isan"
},
{
"operation": "Now on",
"time": "28.716",
"value": "International Standard Audiovisual Number"
},
{
"operation": "search",
"time": "44.915",
"value": "Contains... Lao Wang"
},
{
"operation": "Now on",
"time": "46.041",
"value": "undefined"
},
{
"operation": "search",
"time": "93.094",
"value": "Contains...isan society"
},
{
"operation": "Now on",
"time": "94.226",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "107.265",
"value": "Isan language"
},
{
"operation": "Now on",
"time": "112.639",
"value": "Isan language"
},
{
"operation": "hyperlink",
"time": "124.272",
"value": "Isan"
},
{
"operation": "Now on",
"time": "125.233",
"value": "Isan people"
},
{
"operation": "Page search",
"time": "138.35",
"value": "lao wang"
},
{
"operation": "hyperlink",
"time": "257.245",
"value": "Thai Cultural Mandates"
},
{
"operation": "Now on",
"time": "258.051",
"value": "Thai cultural mandates"
},
{
"operation": "Highlighting",
"time": "277.058",
"value": "Thai cultural mandates_item_14_37"
},
{
"operation": "Highlighting",
"time": "277.059",
"value": "Thai cultural mandates_title"
},
{
"operation": "Highlighting",
"time": "277.059",
"value": "Thai cultural mandates_section_13"
},
{
"operation": "Highlighting deleted",
"time": "279.876",
"value": "Thai cultural mandates_item_14_37"
},
{
"operation": "Highlighting deleted",
"time": "279.878",
"value": "Thai cultural mandates_title"
},
{
"operation": "Highlighting deleted",
"time": "279.881",
"value": "Thai cultural mandates_section_13"
},
{
"operation": "back-button-clicked",
"time": "292.74",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "292.871",
"value": "Isan people"
},
{
"operation": "hyperlink",
"time": "299.184",
"value": "Holy Men's Rebellions"
},
{
"operation": "Now on",
"time": "299.715",
"value": "Holy Man's Rebellion"
},
{
"operation": "back-button-clicked",
"time": "316.432",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "316.567",
"value": "Isan people"
},
{
"operation": "Highlighting",
"time": "380.67",
"value": "Isan people_sentence_38"
},
{
"operation": "finish",
"time": "410.834",
"value": "finish"
}
]
}
|
69617
|
In the Boreham Circuit, the drivers who raced were Mike Hawthorn, Stirling Moss, and Ken Wharton, who won the Freiburg event outright in the E.R.A; then in 1955, the Ford Motor Company bought the airfield to use as a test facility for trucks, as well as the base for their Competitions Department where they prepared their cars.
|
{
"evidence": [
"Boreham Circuit_sentence_4",
"Boreham Circuit_sentence_5",
"Ken Wharton_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:\nIn the Boreham Circuit, the drivers who raced were Mike Hawthorn, Stirling Moss, and Ken Wharton, who won the Freiburg event outright in the E.R.A; then in 1955, the Ford Motor Company bought the airfield to use as a test facility for trucks, as well as the base for their Competitions Department where they prepared their cars.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Boreham Circuit_sentence_4",
"Boreham Circuit_sentence_5",
"Ken Wharton_sentence_13"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2172,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "2.358",
"value": "?search="
},
{
"operation": "search",
"time": "11.171",
"value": "Boreham Circuit"
},
{
"operation": "Now on",
"time": "22.259",
"value": "Boreham Circuit"
},
{
"operation": "Highlighting",
"time": "48.345",
"value": "Boreham Circuit_sentence_4"
},
{
"operation": "Highlighting",
"time": "80.194",
"value": "Boreham Circuit_sentence_5"
},
{
"operation": "hyperlink",
"time": "91.133",
"value": "Ken Wharton"
},
{
"operation": "Now on",
"time": "101.665",
"value": "Ken Wharton"
},
{
"operation": "Page search",
"time": "139.994",
"value": "freiburg"
},
{
"operation": "Highlighting",
"time": "160.45",
"value": "Ken Wharton_sentence_13"
},
{
"operation": "finish",
"time": "168.497",
"value": "finish"
}
]
}
|
exfever_4399
|
The Hurt Locker is a 2008 American comedy thriller film directed by an American film director who is the first woman to win an Academy Award for Best Director.
|
{
"evidence": [
"The Hurt Locker",
"Kathryn Bigelow"
],
"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 Hurt Locker is a 2008 American comedy thriller film directed by an American film director who is the first woman to win an Academy Award for Best Director.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"The Hurt Locker",
"Kathryn Bigelow"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2173,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
82072
|
Vitus Eicher has played professional football for awhile, though he has not scored any goals in that time.
|
{
"evidence": [
"Vitus Eicher_sentence_3"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nVitus Eicher has played professional football for awhile, though he has not scored any goals in that time.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"Vitus Eicher_sentence_3"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2174,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.498",
"value": "?search="
},
{
"operation": "Now on",
"time": "16.01",
"value": "?search="
},
{
"operation": "search",
"time": "19.371",
"value": "Vitus Eicher"
},
{
"operation": "Now on",
"time": "21.194",
"value": "Vitus Eicher"
},
{
"operation": "Highlighting",
"time": "50.955",
"value": "Vitus Eicher_sentence_3"
},
{
"operation": "Highlighting",
"time": "54.63",
"value": "Vitus Eicher_sentence_0Vitus Eicher_sentence_1Vitus Eicher_sentence_2"
},
{
"operation": "finish",
"time": "74.185",
"value": "finish"
}
]
}
|
exfever_4052
|
British Columbia is the westernmost province of Canada, situated between the Pacific Ocean and the continental divide of the Rocky Mountains that border the Canadian province of Alberta to the east and into the north was the Canadian territory of the smallest and westernmost of Canada that are still popular in usage and a Crown corporation that functions as the primary postal operator in Antarctica.
|
{
"evidence": [
"British Columbia",
"Yukon",
"Canada Post"
],
"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:\nBritish Columbia is the westernmost province of Canada, situated between the Pacific Ocean and the continental divide of the Rocky Mountains that border the Canadian province of Alberta to the east and into the north was the Canadian territory of the smallest and westernmost of Canada that are still popular in usage and a Crown corporation that functions as the primary postal operator in Antarctica.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"British Columbia",
"Yukon",
"Canada Post"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2175,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_6948
|
30 Seconds to Mars is the debut studio album distributed by a record label and music distributor originally founded as a British independent record label by entrepreneurs, including a British billionaire who controls more than 600 companies in various fields.
|
{
"evidence": [
"30 Seconds to Mars (album)",
"Virgin Records",
"Richard Branson"
],
"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:\n30 Seconds to Mars is the debut studio album distributed by a record label and music distributor originally founded as a British independent record label by entrepreneurs, including a British billionaire who controls more than 600 companies in various fields.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"30 Seconds to Mars (album)",
"Virgin Records",
"Richard Branson"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2176,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
exfever_5218
|
Maggie Gyllenhaal is an American actress and the older sister of the actor who made his acting debut in the movie City Slickers.
|
{
"evidence": [
"Maggie Gyllenhaal",
"Jake Gyllenhaal"
],
"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:\nMaggie Gyllenhaal is an American actress and the older sister of the actor who made his acting debut in the movie City Slickers.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Maggie Gyllenhaal",
"Jake Gyllenhaal"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2177,
"split": "train_online"
}
|
{
"annotator_operations": []
}
|
40371
|
Karen Karnes and her husband never had the chance to visit Black Mountain College nor meet any famous personality there.
|
{
"evidence": [
"Karen Karnes_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:\nKaren Karnes and her husband never had the chance to visit Black Mountain College nor meet any famous personality there.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Karen Karnes_sentence_10"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2178,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.826",
"value": "?search="
},
{
"operation": "search",
"time": "7.098",
"value": "Karen Karnes"
},
{
"operation": "Now on",
"time": "7.946",
"value": "Karen Karnes"
},
{
"operation": "hyperlink",
"time": "16.217",
"value": "2 Black Mountain College Karen Karnes_section_1"
},
{
"operation": "Highlighting",
"time": "36.13",
"value": "Karen Karnes_sentence_10"
},
{
"operation": "finish",
"time": "49.303",
"value": "finish"
}
]
}
|
78836
|
In February 2007 Domenici indicated his intention to run for re-election, changed his mind in October and in 2008 Pearce and Udall won their respective nomination contests.
|
{
"evidence": [
"2008 United States Senate election in New Mexico_sentence_7",
"2008 United States Senate election in New Mexico_sentence_8",
"2008 United States Senate election in New Mexico_sentence_10"
],
"label": "SUPPORT"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn February 2007 Domenici indicated his intention to run for re-election, changed his mind in October and in 2008 Pearce and Udall won their respective nomination contests.\n",
"role": "user"
}
] |
Search terms not in claim
|
{
"ground_truth": {
"evidence": [
"2008 United States Senate election in New Mexico_sentence_7",
"2008 United States Senate election in New Mexico_sentence_8",
"2008 United States Senate election in New Mexico_sentence_10"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2179,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.336",
"value": "?search="
},
{
"operation": "search",
"time": "72.833",
"value": "2008 United States Senate election in New Mexico"
},
{
"operation": "Now on",
"time": "74.1",
"value": "2008 United States Senate election in New Mexico"
},
{
"operation": "Highlighting",
"time": "89.248",
"value": "2008 United States Senate election in New Mexico_sentence_7"
},
{
"operation": "Highlighting",
"time": "92.597",
"value": "2008 United States Senate election in New Mexico_sentence_8"
},
{
"operation": "Highlighting",
"time": "102.209",
"value": "2008 United States Senate election in New Mexico_sentence_10"
},
{
"operation": "finish",
"time": "113.829",
"value": "finish"
}
]
}
|
94229
|
In the fourth book of the Gargantua and Pantagruel series, Pantagruel and his companions stay ashore.
|
{
"evidence": [
"Gargantua and Pantagruel_sentence_33"
],
"label": "REFUTE"
}
|
feverous
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nIn the fourth book of the Gargantua and Pantagruel series, Pantagruel and his companions stay ashore.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Gargantua and Pantagruel_sentence_33"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2180,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.004",
"value": "?search="
},
{
"operation": "search",
"time": "22.294",
"value": "Contains...Gargantua and Pantagruel series"
},
{
"operation": "Now on",
"time": "23.897",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "33.169",
"value": "Gargantua and Pantagruel"
},
{
"operation": "Now on",
"time": "34.191",
"value": "Gargantua and Pantagruel"
},
{
"operation": "Highlighting",
"time": "51.432",
"value": "Gargantua and Pantagruel_sentence_2"
},
{
"operation": "hyperlink",
"time": "52.571",
"value": "pentalogy"
},
{
"operation": "Now on",
"time": "53.359",
"value": "Pentalogy"
},
{
"operation": "back-button-clicked",
"time": "59.879",
"value": "back-button-clicked"
},
{
"operation": "Now on",
"time": "60.085",
"value": "Gargantua and Pantagruel"
},
{
"operation": "Highlighting deleted",
"time": "61.919",
"value": "Gargantua and Pantagruel_sentence_2"
},
{
"operation": "hyperlink",
"time": "65.386",
"value": "2.4 The Fourth Book Gargantua and Pantagruel_section_5"
},
{
"operation": "Highlighting",
"time": "75.679",
"value": "Gargantua and Pantagruel_sentence_33"
},
{
"operation": "finish",
"time": "86.797",
"value": "finish"
}
]
}
|
70103
|
Berkeley cardinals are certain large cardinals suggested by Hugh Woodin and are currently accepted without errors in mathematics.
|
{
"evidence": [
"Berkeley cardinal_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:\nBerkeley cardinals are certain large cardinals suggested by Hugh Woodin and are currently accepted without errors in mathematics.\n",
"role": "user"
}
] |
Other
|
{
"ground_truth": {
"evidence": [
"Berkeley cardinal_sentence_0"
],
"label": "NOT ENOUGH INFO"
},
"style": "rule"
}
|
{
"index": 2181,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "1.252",
"value": "?search="
},
{
"operation": "search",
"time": "39.316",
"value": "berkeley cardinals"
},
{
"operation": "Now on",
"time": "40.302",
"value": "undefined"
},
{
"operation": "hyperlink",
"time": "42.427",
"value": "Berkeley cardinal"
},
{
"operation": "Now on",
"time": "42.981",
"value": "Berkeley cardinal"
},
{
"operation": "Highlighting",
"time": "46.874",
"value": "Berkeley cardinal_sentence_0"
},
{
"operation": "Page search",
"time": "74.738",
"value": "error"
},
{
"operation": "Page search",
"time": "76.673",
"value": "errors"
},
{
"operation": "Page search",
"time": "84.201",
"value": "accepted"
},
{
"operation": "Page search",
"time": "87.378",
"value": "currently"
},
{
"operation": "Highlighting",
"time": "104.914",
"value": "Berkeley cardinal_sentence_2"
},
{
"operation": "hyperlink",
"time": "125.081",
"value": "axiom of countable choice"
},
{
"operation": "Now on",
"time": "125.498",
"value": "Axiom of countable choice"
},
{
"operation": "Page search",
"time": "138.737",
"value": "error"
},
{
"operation": "search",
"time": "156.498",
"value": "Hugh Woodin"
},
{
"operation": "Now on",
"time": "157.051",
"value": "W. Hugh Woodin"
},
{
"operation": "hyperlink",
"time": "193.225",
"value": "Mathematics"
},
{
"operation": "Now on",
"time": "194.259",
"value": "Mathematics"
},
{
"operation": "Page search",
"time": "199.493",
"value": "cardinal"
},
{
"operation": "search",
"time": "215.478",
"value": "Berkeley cardinal"
},
{
"operation": "Now on",
"time": "215.945",
"value": "Berkeley cardinal"
},
{
"operation": "hyperlink",
"time": "223.034",
"value": "large cardinals"
},
{
"operation": "Now on",
"time": "223.459",
"value": "Large cardinal"
},
{
"operation": "Page search",
"time": "231.29",
"value": "error"
},
{
"operation": "Page search",
"time": "248.186",
"value": "berkeley"
},
{
"operation": "hyperlink",
"time": "252.73",
"value": "List of large cardinal properties"
},
{
"operation": "Now on",
"time": "253.142",
"value": "List of large cardinal properties"
},
{
"operation": "hyperlink",
"time": "258.858",
"value": "Berkeley cardinal"
},
{
"operation": "Now on",
"time": "259.166",
"value": "Berkeley cardinal"
},
{
"operation": "Highlighting deleted",
"time": "277.939",
"value": "Berkeley cardinal_sentence_2"
},
{
"operation": "hyperlink",
"time": "280.738",
"value": "axiom of choice"
},
{
"operation": "Now on",
"time": "286.626",
"value": "Axiom of choice"
},
{
"operation": "Page search",
"time": "299.141",
"value": "berkeley"
},
{
"operation": "finish",
"time": "322.881",
"value": "finish"
}
]
}
|
77729
|
The 1790 Naval Air Squadron was formed in 1945 at RNAS Burscough, an FAA naval air station in Lancashire.
|
{
"evidence": [
"1790 Naval Air Squadron_sentence_0",
"RNAS Burscough (HMS Ringtail)_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 1790 Naval Air Squadron was formed in 1945 at RNAS Burscough, an FAA naval air station in Lancashire.\n",
"role": "user"
}
] |
Multi-hop Reasoning
|
{
"ground_truth": {
"evidence": [
"1790 Naval Air Squadron_sentence_0",
"RNAS Burscough (HMS Ringtail)_sentence_0"
],
"label": "SUPPORT"
},
"style": "rule"
}
|
{
"index": 2182,
"split": "train_online"
}
|
{
"annotator_operations": [
{
"operation": "start",
"time": "0",
"value": "start"
},
{
"operation": "Now on",
"time": "0.788",
"value": "?search="
},
{
"operation": "search",
"time": "8.965",
"value": "1790 Naval Air Squadron"
},
{
"operation": "Now on",
"time": "9.768",
"value": "1790 Naval Air Squadron"
},
{
"operation": "Highlighting",
"time": "227.232",
"value": "1790 Naval Air Squadron_cell_0_1_1"
},
{
"operation": "Highlighting",
"time": "227.236",
"value": "1790 Naval Air Squadron_header_cell_0_1_0"
},
{
"operation": "Highlighting",
"time": "227.236",
"value": "1790 Naval Air Squadron_header_cell_0_0_0"
},
{
"operation": "Highlighting",
"time": "230.387",
"value": "1790 Naval Air Squadron_sentence_0"
},
{
"operation": "Highlighting deleted",
"time": "231.507",
"value": "1790 Naval Air Squadron_cell_0_1_1"
},
{
"operation": "Highlighting deleted",
"time": "231.511",
"value": "1790 Naval Air Squadron_header_cell_0_1_0"
},
{
"operation": "Highlighting deleted",
"time": "231.512",
"value": "1790 Naval Air Squadron_header_cell_0_0_0"
},
{
"operation": "Page search",
"time": "405.003",
"value": "FAA"
},
{
"operation": "Page search",
"time": "408.878",
"value": "naval"
},
{
"operation": "hyperlink",
"time": "415.268",
"value": "RNAS Burscough"
},
{
"operation": "Now on",
"time": "415.917",
"value": "RNAS Burscough (HMS Ringtail)"
},
{
"operation": "Highlighting",
"time": "419.746",
"value": "RNAS Burscough (HMS Ringtail)_sentence_0"
},
{
"operation": "finish",
"time": "426.503",
"value": "finish"
}
]
}
|
exfever_5182
|
Richard Dreyfuss is known for starring in popular films, including a 1987 Canadian buddy cop action comedy film directed by John Badham.
|
{
"evidence": [
"Richard Dreyfuss",
"Stakeout (1987 film)"
],
"label": "REFUTE"
}
|
exfever
|
[
{
"content": "You are a claim-verification assistant. You MUST follow this protocol exactly:\n\n<plan>…</plan>\n- Once at the start: sketch your high-level strategy, such as claim decomposition, entity recognition, etc.\n\n<search>…</search>\n- When you need a fact: emit exactly this tag with your query.\n- To make the most of your search turns, don’t repeat identical queries.\n- You can search at most three times.\n\n<information>\n[[e_1]]: info1\n[[e_2]]: info2\n…\n</information>\n- You will be given claim related information in the format above.\n\n<think>…</think> \n- Use for every piece of reasoning; do not state your final verdict here.\n- You must conduct reasoning inside <think> and </think> first every time you get new information.\n\n<answer>\nLabel: SUPPORT / REFUTE / NOT ENOUGH INFO\nEvidence: [[e_1]], [[e_3]], ...\n</answer> \n- Emit exactly once at the end, no extra text or tags.\n- Evidence id such as e_1 will be replaced by real ids from the corpus. You must include useful real ids when answering\n- Evidence outputs must strictly enforce the format [[e_i]], [[e_j]]…\n- Answer Labels respectively stand for:\nSUPPORT: The claim is consistent with the cited evidence and the evidence is sufficient to confirm the claim.\nREFUTE: The claim contradicts the cited evidence and the evidence is sufficient to disprove the claim.\nNOT ENOUGH INFO: The available evidence is insufficient to determine whether the claim is true or false.\n\n- Note that the process should be: plan → (search → information → think) repeat until conclusion → answer\n\nVerify the claim:\nRichard Dreyfuss is known for starring in popular films, including a 1987 Canadian buddy cop action comedy film directed by John Badham.\n",
"role": "user"
}
] |
EX-FEVER
|
{
"ground_truth": {
"evidence": [
"Richard Dreyfuss",
"Stakeout (1987 film)"
],
"label": "REFUTE"
},
"style": "rule"
}
|
{
"index": 2183,
"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.