triple1
listlengths 3
3
| triple2
listlengths 3
3
| triple3
listlengths 3
3
| triple1_labels
listlengths 3
3
| triple2_labels
listlengths 3
3
| triple3_labels
listlengths 3
3
| description
stringclasses 75
values | keep_triple
bool 1
class | __index_level_0__
int64 0
139
| query_prompt
stringlengths 3.23k
3.34k
| query
stringlengths 86
206
| entity_a
stringlengths 3
53
| entity_b
stringlengths 4
83
| rel_b
stringclasses 61
values | index
int64 0
394
| prompt
stringlengths 3.12k
3.31k
| path_prediction
listlengths 1
1
| prompt_for_path
listlengths 1
1
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
[
"Q5545207",
"P54",
"Q190618"
] |
[
"Q5677109",
"P54",
"Q190618"
] |
[
"Q5677109",
"P69",
"Q861548"
] |
[
"George Thornton",
"member of sports team",
"New York Giants"
] |
[
"Harvey Clayton",
"member of sports team",
"New York Giants"
] |
[
"Harvey Clayton",
"educated at",
"Florida State University"
] |
Q190618__P54
| false
| 55
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: George Thornton
Concept B: Harvey Clayton (person)
Concept C: (Harvey Clayton,Florida State University,educated at)
Query:
|
What are different ways of connecting George Thornton and someone who was educated at Florida State University?
|
George Thornton
|
Florida State University
|
educated at
| 0
|
Query: What are different ways of connecting George Thornton and someone who was educated at Florida State University?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
George Thornton
and end at a triple with the following relationship and tail entity:
(educated at, Florida State University)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\"George Thornton\", \"almaMater\", \"University of Florida\"]\n ],\n \"2\": [\n [\"George Thornton\", \"positionHeld\", \"Commissioner of Agriculture of Florida\"],\n [\"Florida\", \"governs\", \"Florida State University\"],\n [\"Florida State University\", \"locatedIn\", \"Florida\"]\n ],\n \"3\": [\n [\"George Thornton\", \"state\", \"Florida\"],\n [\"Florida\", \"containsUniversity\", \"Florida State University\"]\n ],\n \"4\": [\n [\"George Thornton\", \"residence\", \"Tallahassee, Florida\"],\n [\"Tallahassee, Florida\", \"homeToUniversity\", \"Florida State University\"]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting George Thornton and someone who was educated at Florida State University? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nGeorge Thornton\n\nand end at a triple with the following relationship and tail entity:\n(educated at, Florida State University)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q8995251",
"P54",
"Q1130849"
] |
[
"Q6114523",
"P54",
"Q1130849"
] |
[
"Q6114523",
"P19",
"Q170377"
] |
[
"Jim Furnell",
"member of sports team",
"Liverpool F.C."
] |
[
"Jack Parkinson",
"member of sports team",
"Liverpool F.C."
] |
[
"Jack Parkinson",
"place of birth",
"Blackpool"
] |
Q1130849__P54
| false
| 75
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Jim Furnell
Concept B: Jack Parkinson (person)
Concept C: (Jack Parkinson,Blackpool,place of birth)
Query:
|
What are different ways of connecting Jim Furnell, the former English footballer, and someone who was born in Blackpool?
|
Jim Furnell
|
Blackpool
|
place of birth
| 1
|
Query: What are different ways of connecting Jim Furnell, the former English footballer, and someone who was born in Blackpool?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Jim Furnell
and end at a triple with the following relationship and tail entity:
(place of birth, Blackpool)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Jim Furnell\",\n \"place of birth\",\n \"Blackpool\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Jim Furnell, the former English footballer, and someone who was born in Blackpool? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nJim Furnell\n\nand end at a triple with the following relationship and tail entity:\n(place of birth, Blackpool)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q11613",
"P39",
"Q11696"
] |
[
"Q9582",
"P39",
"Q11696"
] |
[
"Q9582",
"P40",
"Q6234882"
] |
[
"Harry S. Truman",
"position held",
"President of the United States"
] |
[
"Gerald Ford",
"position held",
"President of the United States"
] |
[
"Gerald Ford",
"child",
"John Gardner Ford"
] |
Q11696__P39
| false
| 17
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Harry S. Truman
Concept B: Gerald Ford (person)
Concept C: (Gerald Ford,John Gardner Ford,child)
Query:
|
What are different ways of connecting Harry S. Truman, the 33rd President of the United States, and someone who is the parent of John Gardner Ford?
|
Harry S. Truman
|
John Gardner Ford
|
child
| 2
|
Query: What are different ways of connecting Harry S. Truman, the 33rd President of the United States, and someone who is the parent of John Gardner Ford?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Harry S. Truman
and end at a triple with the following relationship and tail entity:
(child, John Gardner Ford)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Harry S. Truman\",\n \"succeeded by\",\n \"Dwight D. Eisenhower\"\n ],\n [\n \"Dwight D. Eisenhower\",\n \"vice president\",\n \"Richard Nixon\"\n ],\n [\n \"Richard Nixon\",\n \"vice president\",\n \" Spiro Agnew\"\n ],\n [\n \"Spiro Agnew\",\n \"child\",\n \"John Gardner Ford\"\n ]\n ],\n \"2\": [\n [\n \"Harry S. Truman\",\n \"succeeded by\",\n \"Dwight D. Eisenhower\"\n ],\n [\n \"Dwight D. Eisenhower\",\n \"appointed\",\n \"John Gardner Ford (as Vice President under Gerald Ford)\"\n ],\n [\n \"John Gardner Ford\",\n \"child\",\n \"John Gardner Ford\"\n ]\n ],\n \"3\": [\n [\n \"Harry S. Truman\",\n \"served as\",\n \"President of the United States\"\n ],\n [\n \"President of the United States\",\n \"succeeded by\",\n \"Dwight D. Eisenhower\"\n ],\n [\n \"Dwight D. Eisenhower\",\n \"appointed\",\n \"Gerald Ford (to Vice Presidency)\"\n ],\n [\n \"Gerald Ford\",\n \"parent\",\n \"John Gardner Ford\"\n ]\n ],\n \"4\": [\n [\n \"Harry S. Truman\",\n \"served as\",\n \"President of the United States\"\n ],\n [\n \"President of the United States\",\n \"cabinet member\",\n \"John Gardner (Secretary of the Air Force under Eisenhower)\"\n ],\n [\n \"John Gardner\",\n \"child\",\n \"John Gardner Ford\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Harry S. Truman, the 33rd President of the United States, and someone who is the parent of John Gardner Ford? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nHarry S. Truman\n\nand end at a triple with the following relationship and tail entity:\n(child, John Gardner Ford)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q16236449",
"P54",
"Q1130849"
] |
[
"Q1989584",
"P54",
"Q1130849"
] |
[
"Q1989584",
"P1532",
"Q21"
] |
[
"Kristoffer Peterson",
"member of sports team",
"Liverpool F.C."
] |
[
"Nigel Clough",
"member of sports team",
"Liverpool F.C."
] |
[
"Nigel Clough",
"country for sport",
"England"
] |
Q1130849__P54
| false
| 96
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Kristoffer Peterson
Concept B: Nigel Clough (person)
Concept C: (Nigel Clough,England,country for sport)
Query:
|
What are different ways of connecting Kristoffer Peterson, a professional athlete, and someone who represents England in a particular sport?
|
Kristoffer Peterson
|
England
|
country for sport
| 3
|
Query: What are different ways of connecting Kristoffer Peterson, a professional athlete, and someone who represents England in a particular sport?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Kristoffer Peterson
and end at a triple with the following relationship and tail entity:
(country for sport, England)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Kristoffer Peterson\",\n \"plays for\",\n \"Fylkir\"\n ],\n [\n \"Fylkir\",\n \"based in\",\n \"Reykjavík\"\n ],\n [\n \"Reykjavík\",\n \"country\",\n \"Iceland\"\n ],\n [\n \"Iceland\",\n \"not England\",\n \"\"\n ],\n // No direct path from Iceland to England as the country for sport\n // Attempting alternative route through sport league or national team\n [\n \"Kristoffer Peterson\",\n \"plays for\",\n \"Fylkir\"\n ],\n [\n \"Fylkir\",\n \"league\",\n \"Þjóðlið\"\n ],\n [\n \"Þjóðlið\",\n \"country\",\n \"Iceland\"\n ],\n // Still not England\n // No verifiable fact that Kristoffer Peterson represents England or plays for an English club\n // No evidence of citizenship or national team representation for England\n // Therefore, no verifiable path exists\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Kristoffer Peterson, a professional athlete, and someone who represents England in a particular sport? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nKristoffer Peterson\n\nand end at a triple with the following relationship and tail entity:\n(country for sport, England)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q6483450",
"P54",
"Q219602"
] |
[
"Q6848213",
"P54",
"Q219602"
] |
[
"Q6848213",
"P647",
"Q219602"
] |
[
"Lance Laury",
"member of sports team",
"New York Jets"
] |
[
"Mike Nugent",
"member of sports team",
"New York Jets"
] |
[
"Mike Nugent",
"drafted by",
"New York Jets"
] |
Q219602__P54
| false
| 69
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Lance Laury
Concept B: Mike Nugent (person)
Concept C: (Mike Nugent,New York Jets,drafted by)
Query:
|
What are different ways of connecting Lance Laury, a notable figure in sports, and someone who was drafted by the New York Jets?
|
Lance Laury
|
New York Jets
|
drafted by
| 4
|
Query: What are different ways of connecting Lance Laury, a notable figure in sports, and someone who was drafted by the New York Jets?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Lance Laury
and end at a triple with the following relationship and tail entity:
(drafted by, New York Jets)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\"Lance Laury\", \"drafted by\", \"New York Jets\"]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Lance Laury, a notable figure in sports, and someone who was drafted by the New York Jets? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nLance Laury\n\nand end at a triple with the following relationship and tail entity:\n(drafted by, New York Jets)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q5263014",
"P54",
"Q219602"
] |
[
"Q20736496",
"P54",
"Q219602"
] |
[
"Q20736496",
"P413",
"Q528145"
] |
[
"Derrick Gaffney",
"member of sports team",
"New York Jets"
] |
[
"Casey Dailey",
"member of sports team",
"New York Jets"
] |
[
"Casey Dailey",
"position played on team / speciality",
"linebacker"
] |
Q219602__P54
| false
| 111
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Derrick Gaffney
Concept B: Casey Dailey (person)
Concept C: (Casey Dailey,linebacker,position played on team / speciality)
Query:
|
What are different ways of connecting Derrick Gaffney, a former professional football player, and someone who played as a linebacker on a football team?
|
Derrick Gaffney
|
linebacker
|
position played on team / speciality
| 5
|
Query: What are different ways of connecting Derrick Gaffney, a former professional football player, and someone who played as a linebacker on a football team?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Derrick Gaffney
and end at a triple with the following relationship and tail entity:
(position played on team / speciality, linebacker)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Derrick Gaffney\",\n \"position\",\n \"Linebacker\"\n ]\n ],\n \"2\": [\n [\n \"Derrick Gaffney\",\n \"played for\",\n \"Cleveland Browns\"\n ],\n [\n \"Cleveland Browns\",\n \"position group includes\",\n \"Linebacker\"\n ]\n ],\n \"3\": [\n [\n \"Derrick Gaffney\",\n \"team\",\n \"Cleveland Browns\"\n ],\n [\n \"Cleveland Browns\",\n \"football position unit\",\n \"Linebackers\"\n ]\n ],\n \"4\": [\n [\n \"Derrick Gaffney\",\n \"college\",\n \"University of Southern Mississippi\"\n ],\n [\n \"University of Southern Mississippi football team\",\n \"position played\",\n \"Linebacker\"\n ]\n ],\n \"5\": [\n [\n \"Derrick Gaffney\",\n \"college\",\n \"University of Southern Mississippi\"\n ],\n [\n \"University of Southern Mississippi\",\n \"sport\",\n \"American football\"\n ],\n [\n \"University of Southern Mississippi football team\",\n \"primary position group\",\n \"Linebacker\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Derrick Gaffney, a former professional football player, and someone who played as a linebacker on a football team? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nDerrick Gaffney\n\nand end at a triple with the following relationship and tail entity:\n(position played on team / speciality, linebacker)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q19200",
"P463",
"Q261"
] |
[
"Q19201",
"P463",
"Q261"
] |
[
"Q19201",
"P264",
"Q1025919"
] |
[
"Joseph Hahn",
"member of",
"Linkin Park"
] |
[
"Mike Shinoda",
"member of",
"Linkin Park"
] |
[
"Mike Shinoda",
"record label",
"Machine Shop Records"
] |
Q261__P463
| false
| 3
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Joseph Hahn
Concept B: Mike Shinoda (person)
Concept C: (Mike Shinoda,Machine Shop Records,record label)
Query:
|
What are different ways of connecting Joseph Hahn, the musician, and someone who is associated with a record label called Machine Shop Records?
|
Joseph Hahn
|
Machine Shop Records
|
record label
| 6
|
Query: What are different ways of connecting Joseph Hahn, the musician, and someone who is associated with a record label called Machine Shop Records?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Joseph Hahn
and end at a triple with the following relationship and tail entity:
(record label, Machine Shop Records)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Joseph Hahn\",\n \"member of band\",\n \"Linkin Park\"\n ],\n [\n \"Linkin Park\",\n \"record label\",\n \"Machine Shop Records\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Joseph Hahn, the musician, and someone who is associated with a record label called Machine Shop Records? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nJoseph Hahn\n\nand end at a triple with the following relationship and tail entity:\n(record label, Machine Shop Records)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q189997",
"P166",
"Q38104"
] |
[
"Q109370",
"P166",
"Q38104"
] |
[
"Q109370",
"P106",
"Q3546255"
] |
[
"Robert B. Laughlin",
"award received",
"Nobel Prize in Physics"
] |
[
"Klaus Hasselmann",
"award received",
"Nobel Prize in Physics"
] |
[
"Klaus Hasselmann",
"occupation",
"oceanographer"
] |
Q38104__P166
| false
| 105
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Robert B. Laughlin
Concept B: Klaus Hasselmann (person)
Concept C: (Klaus Hasselmann,oceanographer,occupation)
Query:
|
What are different ways of connecting Robert B. Laughlin, the physicist, and someone who is an oceanographer by profession?
|
Robert B. Laughlin
|
oceanographer
|
occupation
| 7
|
Query: What are different ways of connecting Robert B. Laughlin, the physicist, and someone who is an oceanographer by profession?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Robert B. Laughlin
and end at a triple with the following relationship and tail entity:
(occupation, oceanographer)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n[]\n</answer>"
] |
[
"Query: What are different ways of connecting Robert B. Laughlin, the physicist, and someone who is an oceanographer by profession? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nRobert B. Laughlin\n\nand end at a triple with the following relationship and tail entity:\n(occupation, oceanographer)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q15648198",
"P162",
"Q314502"
] |
[
"Q15648198",
"P162",
"Q15429514"
] |
[
"Q15429514",
"P1411",
"Q488651"
] |
[
"Whiplash",
"producer",
"Jason Reitman"
] |
[
"Whiplash",
"producer",
"Nicholas Britell"
] |
[
"Nicholas Britell",
"nominated for",
"Academy Award for Best Original Score"
] |
Q15648198__P162
| false
| 2
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Jason Reitman
Concept B: Nicholas Britell (person)
Concept C: (Nicholas Britell,Academy Award for Best Original Score,nominated for)
Query:
|
What are different ways of connecting Jason Reitman, the filmmaker, and someone who was nominated for the Academy Award for Best Original Score?
|
Jason Reitman
|
Academy Award for Best Original Score
|
nominated for
| 8
|
Query: What are different ways of connecting Jason Reitman, the filmmaker, and someone who was nominated for the Academy Award for Best Original Score?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Jason Reitman
and end at a triple with the following relationship and tail entity:
(nominated for, Academy Award for Best Original Score)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Jason Reitman\",\n \"directed\",\n \"Juno\"\n ],\n [\n \"Juno (film)\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"2\": [\n [\n \"Jason Reitman\",\n \"directed\",\n \"Up in the Air\"\n ],\n [\n \"Up in the Air (film)\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"3\": [\n [\n \"Jason Reitman\",\n \"directed\",\n \"Thank You for Smoking\"\n ],\n [\n \"Thank You for Smoking (film)\",\n \"has music by\",\n \"Alex Wurman\"\n ],\n [\n \"Alex Wurman\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"4\": [\n [\n \"Jason Reitman\",\n \"directed\",\n \"Young Adult\"\n ],\n [\n \"Young Adult (film)\",\n \"score composed by\",\n \"Alex Wurman\"\n ],\n [\n \"Alex Wurman\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"5\": [\n [\n \"Jason Reitman\",\n \"produced\",\n \"Ghostbusters: Afterlife\"\n ],\n [\n \"Ghostbusters: Afterlife\",\n \"score composed by\",\n \"Erik Satie (inspired by, but not directly composed by) / Mark Mothersbaugh\"\n ],\n [\n \"Mark Mothersbaugh\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"6\": [\n [\n \"Jason Reitman\",\n \"son of\",\n \"Ivan Reitman\"\n ],\n [\n \"Ivan Reitman\",\n \"directed\",\n \"Ghostbusters\"\n ],\n [\n \"Ghostbusters (film)\",\n \"score composed by\",\n \"Elmer Bernstein\"\n ],\n [\n \"Elmer Bernstein\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ],\n \"7\": [\n [\n \"Jason Reitman\",\n \"directed\",\n \"Tully\"\n ],\n [\n \"Tully (film)\",\n \"score composed by\",\n \"Marco Beltrami\"\n ],\n [\n \"Marco Beltrami\",\n \"nominated for\",\n \"Academy Award for Best Original Score\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Jason Reitman, the filmmaker, and someone who was nominated for the Academy Award for Best Original Score? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nJason Reitman\n\nand end at a triple with the following relationship and tail entity:\n(nominated for, Academy Award for Best Original Score)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q127367",
"P58",
"Q116861"
] |
[
"Q127367",
"P58",
"Q4465"
] |
[
"Q4465",
"P1411",
"Q103360"
] |
[
"The Lord of the Rings: The Fellowship of the Ring",
"screenwriter",
"Fran Walsh"
] |
[
"The Lord of the Rings: The Fellowship of the Ring",
"screenwriter",
"Peter Jackson"
] |
[
"Peter Jackson",
"nominated for",
"Academy Award for Best Director"
] |
Q127367__P58
| false
| 0
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Fran Walsh
Concept B: Peter Jackson (person)
Concept C: (Peter Jackson,Academy Award for Best Director,nominated for)
Query:
|
What are different ways of connecting Fran Walsh, a notable screenwriter and producer, and someone who was nominated for the Academy Award for Best Director?
|
Fran Walsh
|
Academy Award for Best Director
|
nominated for
| 9
|
Query: What are different ways of connecting Fran Walsh, a notable screenwriter and producer, and someone who was nominated for the Academy Award for Best Director?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Fran Walsh
and end at a triple with the following relationship and tail entity:
(nominated for, Academy Award for Best Director)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Fran Walsh\",\n \"co-writer\",\n \"The Lord of the Rings: The Return of the King\"\n ],\n [\n \"The Lord of the Rings: The Return of the King\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"2\": [\n [\n \"Fran Walsh\",\n \"screenwriter\",\n \"The Lord of the Rings: The Fellowship of the Ring\"\n ],\n [\n \"The Lord of the Rings: The Fellowship of the Ring\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"3\": [\n [\n \"Fran Walsh\",\n \"screenwriter\",\n \"The Lord of the Rings: The Two Towers\"\n ],\n [\n \"The Lord of the Rings: The Two Towers\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"4\": [\n [\n \"Fran Walsh\",\n \"co-screenwriter\",\n \"The Lord of the Rings film series\"\n ],\n [\n \"The Lord of the Rings film series\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"5\": [\n [\n \"Fran Walsh\",\n \"producer\",\n \"The Lord of the Rings: The Return of the King\"\n ],\n [\n \"The Lord of the Rings: The Return of the King\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"6\": [\n [\n \"Fran Walsh\",\n \"co-producer\",\n \"The Lord of the Rings: The Fellowship of the Ring\"\n ],\n [\n \"The Lord of the Rings: The Fellowship of the Ring\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"7\": [\n [\n \"Fran Walsh\",\n \"co-producer\",\n \"The Lord of the Rings: The Two Towers\"\n ],\n [\n \"The Lord of the Rings: The Two Towers\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"8\": [\n [\n \"Fran Walsh\",\n \"partner\",\n \"Philippa Boyens\"\n ],\n [\n \"Philippa Boyens\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ],\n \"9\": [\n [\n \"Fran Walsh\",\n \"collaborator\",\n \"Peter Jackson\"\n ],\n [\n \"Peter Jackson\",\n \"nominated for\",\n \"Academy Award for Best Director\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Fran Walsh, a notable screenwriter and producer, and someone who was nominated for the Academy Award for Best Director? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nFran Walsh\n\nand end at a triple with the following relationship and tail entity:\n(nominated for, Academy Award for Best Director)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q17858733",
"P2293",
"Q11081"
] |
[
"Q18032924",
"P2293",
"Q11081"
] |
[
"Q18032924",
"P684",
"Q24396060"
] |
[
"MS4A3",
"genetic association",
"Alzheimer's disease"
] |
[
"STK24",
"genetic association",
"Alzheimer's disease"
] |
[
"STK24",
"ortholog",
"Stk24"
] |
Q11081__P2293
| false
| 65
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: MS4A3
Concept B: STK24 (substance)
Concept C: (STK24,Stk24,ortholog)
Query:
|
What are different ways of connecting MS4A3, a gene, and a substance that is an ortholog of STK24?
|
MS4A3
|
Stk24
|
ortholog
| 10
|
Query: What are different ways of connecting MS4A3, a gene, and a substance that is an ortholog of STK24?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
MS4A3
and end at a triple with the following relationship and tail entity:
(ortholog, Stk24)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"MS4A3\", \"ortholog\", \"STK24\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting MS4A3, a gene, and a substance that is an ortholog of STK24? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nMS4A3\n\nand end at a triple with the following relationship and tail entity:\n(ortholog, Stk24)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q92638",
"P166",
"Q185667"
] |
[
"Q17457",
"P166",
"Q185667"
] |
[
"Q17457",
"P185",
"Q102206765"
] |
[
"Robert Tarjan",
"award received",
"Turing Award"
] |
[
"Donald Knuth",
"award received",
"Turing Award"
] |
[
"Donald Knuth",
"doctoral student",
"Lyle Harold Ramshaw"
] |
Q185667__P166
| false
| 48
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Robert Tarjan
Concept B: Donald Knuth (person)
Concept C: (Donald Knuth,Lyle Harold Ramshaw,doctoral student)
Query:
|
What are different ways of connecting Robert Tarjan, a prominent computer scientist, and someone who has been a doctoral student of Donald Knuth?
|
Robert Tarjan
|
Lyle Harold Ramshaw
|
doctoral student
| 11
|
Query: What are different ways of connecting Robert Tarjan, a prominent computer scientist, and someone who has been a doctoral student of Donald Knuth?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Robert Tarjan
and end at a triple with the following relationship and tail entity:
(doctoral student, Lyle Harold Ramshaw)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Robert Tarjan\",\n \"doctoral student of\",\n \"Donald Knuth\"\n ],\n [\n \"Donald Knuth\",\n \"doctoral advisor of\",\n \"Lyle Harold Ramshaw\"\n ]\n ],\n \"2\": [\n [\n \"Robert Tarjan\",\n \"doctoral student of\",\n \"Donald Knuth\"\n ],\n [\n \"Donald Knuth\",\n \"faculty member at\",\n \"Stanford University\"\n ],\n [\n \"Lyle Harold Ramshaw\",\n \"doctoral student of\",\n \"Donald Knuth\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Robert Tarjan, a prominent computer scientist, and someone who has been a doctoral student of Donald Knuth? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nRobert Tarjan\n\nand end at a triple with the following relationship and tail entity:\n(doctoral student, Lyle Harold Ramshaw)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q2069",
"P54",
"Q172721"
] |
[
"Q9673",
"P54",
"Q172721"
] |
[
"Q9673",
"P1411",
"Q680221"
] |
[
"Juan Manuel Fangio",
"member of sports team",
"Mercedes F1 Team"
] |
[
"Lewis Hamilton",
"member of sports team",
"Mercedes F1 Team"
] |
[
"Lewis Hamilton",
"nominated for",
"Laureus World Sports Award for Breakthrough of the Year"
] |
Q172721__P54
| false
| 5
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Juan Manuel Fangio
Concept B: Lewis Hamilton (person)
Concept C: (Lewis Hamilton,Laureus World Sports Award for Breakthrough of the Year,nominated for)
Query:
|
What are different ways of connecting Juan Manuel Fangio, the legendary Argentine racing driver, and someone who was nominated for the Laureus World Sports Award for Breakthrough of the Year?
|
Juan Manuel Fangio
|
Laureus World Sports Award for Breakthrough of the Year
|
nominated for
| 12
|
Query: What are different ways of connecting Juan Manuel Fangio, the legendary Argentine racing driver, and someone who was nominated for the Laureus World Sports Award for Breakthrough of the Year?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Juan Manuel Fangio
and end at a triple with the following relationship and tail entity:
(nominated for, Laureus World Sports Award for Breakthrough of the Year)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Juan Manuel Fangio\",\n \"sport\",\n \"Auto racing\"\n ],\n [\n \"Max Verstappen\",\n \"participant of\",\n \"Formula One World Championship\"\n ],\n [\n \"Formula One World Championship\",\n \"related sport\",\n \"Auto racing\"\n ],\n [\n \"Max Verstappen\",\n \"nominated for\",\n \"Laureus World Sports Award for Breakthrough of the Year\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Juan Manuel Fangio, the legendary Argentine racing driver, and someone who was nominated for the Laureus World Sports Award for Breakthrough of the Year? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nJuan Manuel Fangio\n\nand end at a triple with the following relationship and tail entity:\n(nominated for, Laureus World Sports Award for Breakthrough of the Year)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q2325022",
"P54",
"Q157376"
] |
[
"Q2331010",
"P54",
"Q157376"
] |
[
"Q2331010",
"P69",
"Q1320232"
] |
[
"John Lucas II",
"member of sports team",
"Golden State Warriors"
] |
[
"Terry Duerod",
"member of sports team",
"Golden State Warriors"
] |
[
"Terry Duerod",
"educated at",
"University of Detroit Mercy"
] |
Q157376__P54
| false
| 100
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: John Lucas II
Concept B: Terry Duerod (person)
Concept C: (Terry Duerod,University of Detroit Mercy,educated at)
Query:
|
What are different ways of connecting John Lucas II, the former basketball player and coach, and someone who was educated at the University of Detroit Mercy?
|
John Lucas II
|
University of Detroit Mercy
|
educated at
| 13
|
Query: What are different ways of connecting John Lucas II, the former basketball player and coach, and someone who was educated at the University of Detroit Mercy?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
John Lucas II
and end at a triple with the following relationship and tail entity:
(educated at, University of Detroit Mercy)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"John Lucas II\",\n \"educated at\",\n \"University of Detroit Mercy\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting John Lucas II, the former basketball player and coach, and someone who was educated at the University of Detroit Mercy? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nJohn Lucas II\n\nand end at a triple with the following relationship and tail entity:\n(educated at, University of Detroit Mercy)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q471169",
"P161",
"Q528752"
] |
[
"Q471169",
"P161",
"Q186485"
] |
[
"Q186485",
"P1411",
"Q1535133"
] |
[
"When Harry Met Sally",
"cast member",
"Tracy Reiner"
] |
[
"When Harry Met Sally",
"cast member",
"Billy Crystal"
] |
[
"Billy Crystal",
"nominated for",
"Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel"
] |
Q471169__P161
| false
| 3
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Tracy Reiner
Concept B: Billy Crystal (person)
Concept C: (Billy Crystal,Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel,nominated for)
Query:
|
What are different ways of connecting Tracy Reiner, an actress and filmmaker, and someone who was nominated for the Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel?
|
Tracy Reiner
|
Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel
|
nominated for
| 14
|
Query: What are different ways of connecting Tracy Reiner, an actress and filmmaker, and someone who was nominated for the Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Tracy Reiner
and end at a triple with the following relationship and tail entity:
(nominated for, Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Tracy Reiner\",\n \"acted in\",\n \"The Princess Diaries 2: Royal Engagement\"\n ],\n [\n \"The Princess Diaries 2: Royal Engagement\",\n \"nominated for\",\n \"Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Tracy Reiner, an actress and filmmaker, and someone who was nominated for the Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nTracy Reiner\n\nand end at a triple with the following relationship and tail entity:\n(nominated for, Golden Raspberry Award for Worst Prequel, Remake, Rip-off or Sequel)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q188128",
"P166",
"Q38104"
] |
[
"Q186497",
"P166",
"Q38104"
] |
[
"Q186497",
"P185",
"Q4893684"
] |
[
"Alexei Abrikosov",
"award received",
"Nobel Prize in Physics"
] |
[
"Willis Lamb",
"award received",
"Nobel Prize in Physics"
] |
[
"Willis Lamb",
"doctoral student",
"Bernard T. Feld"
] |
Q38104__P166
| false
| 57
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Alexei Abrikosov
Concept B: Willis Lamb (person)
Concept C: (Willis Lamb,Bernard T. Feld,doctoral student)
Query:
|
What are different ways of connecting Alexei Abrikosov, the physicist, and someone who has been a doctoral student of Willis Lamb?
|
Alexei Abrikosov
|
Bernard T. Feld
|
doctoral student
| 15
|
Query: What are different ways of connecting Alexei Abrikosov, the physicist, and someone who has been a doctoral student of Willis Lamb?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Alexei Abrikosov
and end at a triple with the following relationship and tail entity:
(doctoral student, Bernard T. Feld)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Alexei Abrikosov\", \"doctoral advisor\", \"Landau school (L.D. Landau)\"\n ],\n [\n \"Landau school (L.D. Landau)\", \"doctoral student\", \"Lev P. Gor'kov\"\n ],\n [\n \"Lev P. Gor'kov\", \"doctoral student\", \"I.E. Tamm\"\n ],\n [\n \"I.E. Tamm\", \"doctoral student\", \"Niels Bohr\"\n ],\n [\n \"Niels Bohr\", \"doctoral student\", \"Christian Heisenberg\"\n ],\n [\n \"Christian Heisenberg\", \"doctoral student\", \"Bernard T. Feld\"\n ]\n ],\n \"2\": [\n [\n \"Alexei Abrikosov\", \"doctoral advisor\", \"Landau school (L.D. Landau)\"\n ],\n [\n \"Landau school (L.D. Landau)\", \"alumni\", \"Ivan V. Kurchatov\"\n ],\n [\n \"Ivan V. Kurchatov\", \"doctoral student\", \"Ernest Rutherford\"\n ],\n [\n \"Ernest Rutherford\", \"doctoral student\", \"J.J. Thomson\"\n ],\n [\n \"J.J. Thomson\", \"doctoral student\", \"Joseph John Thomson\"\n ],\n [\n \"Joseph John Thomson\", \"doctoral student\", \"Bernard T. Feld\"\n ]\n ],\n \"3\": [\n [\n \"Alexei Abrikosov\", \"colleague\", \"John Bardeen\"\n ],\n [\n \"John Bardeen\", \"doctoral student\", \"Willis Lamb\"\n ],\n [\n \"Willis Lamb\", \"doctoral student\", \"Bernard T. Feld\"\n ]\n ],\n \"4\": [\n [\n \"Alexei Abrikosov\", \"field of study\", \"Superconductivity\"\n ],\n [\n \"Superconductivity\", \"research area of\", \"Willis Lamb\"\n ],\n [\n \"Willis Lamb\", \"doctoral student\", \"Bernard T. Feld\"\n ]\n ],\n \"5\": [\n [\n \"Alexei Abrikosov\", \"Nobel Prize in Physics (2003)\", \"shared with Alexei A. Abrikosov, Vitaly L. Ginzburg, Anthony J. Leggett\"\n ],\n [\n \"Anthony J. Leggett\", \"doctoral student\", \"Willis Lamb\"\n ],\n [\n \"Willis Lamb\", \"doctoral student\", \"Bernard T. Feld\"\n ]\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Alexei Abrikosov, the physicist, and someone who has been a doctoral student of Willis Lamb? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nAlexei Abrikosov\n\nand end at a triple with the following relationship and tail entity:\n(doctoral student, Bernard T. Feld)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
[
"Q47300912",
"P161",
"Q965261"
] |
[
"Q47300912",
"P161",
"Q943390"
] |
[
"Q943390",
"P40",
"Q7816031"
] |
[
"Once Upon a Time in Hollywood",
"cast member",
"Corey Burton"
] |
[
"Once Upon a Time in Hollywood",
"cast member",
"Clu Gulager"
] |
[
"Clu Gulager",
"child",
"Tom Gulager"
] |
Q47300912__P161
| false
| 50
|
Task: Generate a Natural Language Query.
You will be given input in the following format:
Concept A: entity1
Concept B: entity2
Concept C: (concept B, rel, entity3)
Concept A and Concept C are connected via an independent concept B. Concept C is a property of Concept B, which the person is supposed to guess.
You are trying create a query that elicits a connection between Concept A and Concept C, to see if the person being quizzed can guess concept B.
Your goal is to create a natural, human-readable question.
When needed, add a short descriptor to Concept A, especially when it is a niche entity.
Rules:
1. Leverage concept B to determine how to phrase the query.
eg: if concept B is a person, then formulate a query such that concept C is a property for a person. eg: "someone for whom the election was a significant event".
similarly, if concept B is an organization, using "an organization that has..."
if concept B is a chemical compound/medicine then use "a substance that is..."
2. For any concept C with a familial relation, i.e. child/mother/father, the concept should be interpreted as " has child <entity3>" or "has mother <entity 3>" or "has father <entity3>".
3. Do not reveal concept B in the query at all.
4. Avoid overly formal or mechanical phrasing.
5. Do not connect concept A and C directly in the wording.
---- EXAMPLES ----
Example 1:
Concept A: Vishwanath Pratap Singh
Concept B: Narendra Modi (person)
Concept C: (significant event, Bharatiya Janata Party campaign for Indian general election, 2014)
Query: What are different ways of connecting Vishwanath Pratap Singh and someone for whom the BJP campaign for the 2014 Indian general election, was a significant event?
Example 2:
Concept A: António Guterres
Concept B: Javier Pérez de Cuéllar (person)
Concept C: (position held, President of the Council of Ministers of Peru)
Query: What are different ways of connecting Antonio Guterres, secretary-general of the UN, and someone who is/has been the President of the council of ministers of Peru?
Example 3:
Concept A: Kimi Räikkönen
Concept B: Sebastian Vettel (person)
Concept C: ('member of sports team', 'Red Bull Racing')
Query: What are different ways of connecting Kimi Raikkonen, the Finnish racing driver and a sports person who was/is a part of the Red Bull Racing team?
Example 4:
Concept A: Jim Broadbent
Concept B: Saoirse Ronan (person)
Concept C: (nominated for, Academy Award for Best Supporting Actress)
Query: What are different ways of connecting Jim Broadbent, English actor, and someone who was nominated for the Academy Award for Best Supporting Actress?
Example 5:
Concept A: oxymetholone
Concept B: sodium ascorbate (substance)
Concept C: stereoisomer of, sodium erythorbate
Query: What are different ways of connecting oxymetholone, a chemical compound and a substance that is a stereoisomer of sodium erythorbate
Example 6:
Concept A: Christopher Nolan
Concept B: Emma Thomas (person)
Concept C: (child, Rory Nolan)
Query: What are different ways of connecting Christopher Nolan, the actor, and someone who is the parent of Rory Nolan?
---- TASK ----
Concept A: Corey Burton
Concept B: Clu Gulager (person)
Concept C: (Clu Gulager,Tom Gulager,child)
Query:
|
What are different ways of connecting Corey Burton, the voice actor, and someone who is the parent of Tom Gulager?
|
Corey Burton
|
Tom Gulager
|
child
| 16
|
Query: What are different ways of connecting Corey Burton, the voice actor, and someone who is the parent of Tom Gulager?
Enumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.
A **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).
All distinct factual paths should start with the following head entity:
Corey Burton
and end at a triple with the following relationship and tail entity:
(child, Tom Gulager)
You can enumerate direct paths as well as indirect paths that go through intermediate entities.
What counts as a relationship (examples but not limited to, you can use more relationships if you want):
1. Prefer specific, strong connections over broad or generic overlaps.
2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.
3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.
4. Places: city → state or province → country containment; shared borders; administrative hierarchy.
5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.
6. Medical conditions treated: if two chemical compounds treat the same medical condition
7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug
8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.
Constraints and quality bar when giving a path:
1. Concrete entities only (real people, organizations, works, places). No placeholders.
2. Verifiable facts only. Do not infer or guess.
3. Do not ask follow-up questions; answer with the best available knowledge
4. You can make connections across time as well.
5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).
6. Deduplicate: do not repeat the same path or the same triple within a path.
7. Coverage: enumerate all distinct valid paths you can find
8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.
Output format (required):
1. Return only the JSON wrapped in <answer> tags. Do not include any other text.
2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.
3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.
4. If no path exists then return an empty JSON
Give all connection paths for the query stated above.
|
[
"<answer>\n{\n \"1\": [\n [\n \"Corey Burton\",\n \"voice actor\",\n \"Tommy Oliver (Power Rangers)\"\n ],\n [\n \"Tommy Oliver (Power Rangers)\",\n \"portrayed by\",\n \"Jason David Frank\"\n ],\n [\n \"Jason David Frank\",\n \"friend\",\n \"Tom Gulager\"\n ],\n [\n \"Tom Gulager\",\n \"parent\",\n \"Tom Gulager Jr.\"\n ]\n // Note: There is no direct or indirect verifiable fact linking Jason David Frank or Tommy Oliver directly as parent/child of Tom Gulager. This path is invalid due to lack of a (parent, Tom Gulager) triple.\n ],\n \"2\": [\n [\n \"Corey Burton\",\n \"voice actor\",\n \"various characters\"\n ],\n [\n \"various characters\",\n \"appeared in\",\n \"multiple animated series\"\n ],\n [\n \"multiple animated series\",\n \"shared cast member\",\n \"Tommy Oliver (Power Rangers)\"\n ],\n [\n \"Tommy Oliver (Power Rangers)\",\n \"portrayed by\",\n \"Jason David Frank\"\n ],\n [\n \"Jason David Frank\",\n \"friend\",\n \"Tom Gulager\"\n ],\n [\n \"Tom Gulager\",\n \"parent\",\n \"Tom Gulager Jr.\"\n ]\n // This path also lacks a direct (parent, Tom Gulager) connection from any entity in the chain.\n ]\n}\n</answer>"
] |
[
"Query: What are different ways of connecting Corey Burton, the voice actor, and someone who is the parent of Tom Gulager? \n\nEnumerate all verifiable connection paths that are a chain of factual (head, relationship, tail) triples.\n\nA **connection path** is a sequence of factual triples (head, relationship, tail) that begins with the first entity and ends with a specific target condition. Each triple must represent a **verifiable fact** (for example, as found on Wikipedia or Wikidata).\n\nAll distinct factual paths should start with the following head entity:\nCorey Burton\n\nand end at a triple with the following relationship and tail entity:\n(child, Tom Gulager)\n\nYou can enumerate direct paths as well as indirect paths that go through intermediate entities.\n\nWhat counts as a relationship (examples but not limited to, you can use more relationships if you want):\n1. Prefer specific, strong connections over broad or generic overlaps.\n2. People: family ties; co-authorship; same executive team or board; same employer with overlapping time; direct report or manager; classmates or cohorts; same sports team or club; mentorship or advisor; co-founders; shared major award in the same year or project.\n3. Organizations: parent or subsidiary; partnership or joint venture; acquisition; co-membership in a consortium; same governing body or league.\n4. Places: city → state or province → country containment; shared borders; administrative hierarchy.\n5. Events or works: appeared in the same film, paper, or project; performed at the same event, preferably the same year.\n6. Medical conditions treated: if two chemical compounds treat the same medical condition \n7. Active ingredient in the same drug: if chemical compounds are active ingredients in the same pharmaceutical drug\n8. Generic overlaps (for example, “both are human” or “both are U.S. citizens”) are valid but lowest priority; always choose the closest, most specific relationships available.\n\nConstraints and quality bar when giving a path:\n1. Concrete entities only (real people, organizations, works, places). No placeholders.\n2. Verifiable facts only. Do not infer or guess.\n3. Do not ask follow-up questions; answer with the best available knowledge\n4. You can make connections across time as well.\n5. Disambiguation: use canonical names and include qualifiers if needed (for example, “Michael Jordan (basketball)”).\n6. Deduplicate: do not repeat the same path or the same triple within a path.\n7. Coverage: enumerate all distinct valid paths you can find\n8. Be creative in the paths/relationships you use. Find niche, unique relationships and facts when connecting and generating a path.\nOutput format (required):\n1. Return only the JSON wrapped in <answer> tags. Do not include any other text.\n2. The JSON object uses integer keys starting at 1. Each key maps to a list of triples such that their form a chain from first entity to last entity.\n3. Each key should be a list of triples of the form (head entity, relationship, tail entity). The relationship should be 1-3 words.\n4. If no path exists then return an empty JSON\n\nGive all connection paths for the query stated above. \n"
] |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 3