sentences
sequence | new_gt
sequence |
---|---|
[
"To",
"obtain",
"a",
"diverse",
"dataset",
",",
"we",
"considered",
"three",
"different",
"domains",
",",
"from",
"two",
"data",
"sources",
":",
"Yelp"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName"
] |
[
"This",
"dataset",
"includes",
"7",
"M",
"written",
"business",
"reviews",
",",
"where",
"each",
"business",
"may",
"be",
"classified",
"into",
"multiple",
"categories",
",",
"in",
"varying",
"levels",
"of",
"granular-",
"ity",
"We",
"apply",
"KPA",
"to",
"a",
"sample",
"of",
"businesses",
"that",
"include",
"at",
"least",
"one",
"of",
"the",
"following",
"categories",
":",
"RESTAURANTS",
",",
"HOTELS",
",",
"and",
"ART",
"&",
"E",
"NTER",
"-",
"TAINMENT",
",",
"and",
"had",
"at",
"least",
"1,000",
"reviews"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"For",
"the",
"KPH",
"annotation",
",",
"we",
"selected",
"four",
"restaurants",
"(",
"which",
"we",
"refer",
"to",
"as",
"the",
"RESTAURANTS",
"domain",
")",
",",
"and",
"four",
"businesses",
"categorized",
"as",
"ART",
"&",
"E",
"N-",
"TERTAINMENT",
",",
"out",
"of",
"which",
"three",
"were",
"hotels",
"(",
"hereafter",
",",
"the",
"Hotels",
"&",
"Entertainment",
"domain",
",",
"or",
"HOTELS",
"for",
"brevity",
")"
] | [
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Each",
"domain",
"includes",
"two",
"positive",
"and",
"two",
"negative",
"KPA",
"summaries"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O"
] |
[
"Amazon3.This",
"dataset",
"includes",
"over",
"130",
"M",
"cus-",
"tomer",
"reviews",
"for",
"a",
"huge",
"collection",
"of",
"products",
"inAmazon.com",
"across",
"a",
"wide",
"variety",
"of",
"domains",
"2Specifically",
",",
"our",
"implementation",
"follows",
"their",
"RKPA",
"-",
"FT",
"configuration",
",",
"except",
"that",
"we",
"extract",
"the",
"key",
"points",
"for",
"each",
"business",
"independently",
",",
"and",
"allow",
"each",
"sentence",
"to",
"match",
"multiple",
"key",
"points",
"Eventually",
",",
"we",
"annotated",
"a",
"KPH",
"for",
"three",
"positive",
"and",
"one",
"negative",
"KPA",
"summaries"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O"
] |
[
"4.2",
"KPH",
"Annotation",
"Annotating",
"complex",
"structures",
"such",
"as",
"KPHs",
"is",
"a",
"challenging",
"task",
",",
"since",
"it",
"involves",
"global",
",",
"inter-",
"dependent",
"decisions"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Furthermore",
",",
"the",
"annotator",
"needs",
"to",
"consider",
"different",
"types",
"of",
"hierarchical",
"relations",
"that",
"may",
"hold",
"between",
"the",
"key",
"points",
",",
"as",
"we",
"further",
"discuss",
"in",
"Section",
"4.3",
"Finally",
",",
"user",
"re-",
"views",
"make",
"extensive",
"use",
"of",
"informal",
"and",
"figurative",
"language",
"For",
"example",
",",
"β",
"The",
"food",
"is",
"outrageous",
"!",
"β",
"should",
"be",
"interpreted",
"as",
"great",
"food",
";",
"β",
"Elevators",
"should",
"go",
"up",
"and",
"down",
",",
"not",
"diagonal",
"β",
"means",
"that",
"the",
"elevators",
"were",
"scary",
"and",
"β",
"Internet",
"was",
"a",
"joke",
"to",
"get",
"to",
"work",
"β",
"indicates",
"a",
"poor",
"WiFi",
"signal",
"To",
"overcome",
"these",
"challenges",
"and",
"obtain",
"a",
"high-",
"quality",
"dataset",
",",
"three",
"annotators",
"individually",
"con-",
"structed",
"a",
"KPH",
"for",
"each",
"KPA",
"summary",
"(",
"Β§",
"4.2.1",
")",
";",
"The",
"annotators",
"then",
"met",
"to",
"resolve",
"their",
"disagree-",
"ments",
"and",
"reach",
"a",
"consolidated",
"KPH",
"(",
"Β§",
"4.2.2",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O"
] |
[
"4.2.1",
"Creating",
"an",
"Initial",
"KPH",
"To",
"construct",
"an",
"initial",
"KPH",
",",
"annotators",
"were",
"shown",
"the",
"key",
"points",
"one",
"by",
"one",
"in",
"a",
"descending",
"order",
"according",
"to",
"the",
"number",
"of",
"their",
"matched",
"sentences"
] | [
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"For",
"each",
"key",
"point",
",",
"they",
"first",
"decided",
"whether",
"it",
"conveys",
"the",
"same",
"idea",
"as",
"any",
"previously",
"seen",
"key",
"point",
",",
"in",
"which",
"case",
"it",
"was",
"added",
"to",
"an",
"existing",
"cluster",
"If",
"not",
",",
"a",
"new",
"node",
"was",
"added",
"to",
"the",
"KPH",
",",
"and",
"the",
"annotator",
"dragged",
"it",
"to",
"its",
"right",
"position",
"in",
"the",
"hierarchy"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Since",
"key",
"points",
"with",
"many",
"matches",
"tend",
"to",
"be",
"more",
"general",
",",
"the",
"key",
"point",
"ordering",
"facilitated",
"top",
"-",
"down",
"construction",
"of",
"the",
"KPH"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName"
] |
[
"At",
"any",
"point",
"in",
"the",
"annotation",
"process",
",",
"annotators",
"had",
"a",
"complete",
"view",
"of",
"the",
"KPH",
"constructed",
"so",
"far",
",",
"and",
"could",
"adjust",
"it",
"by",
"modifying",
"previous",
"decisions",
",",
"including",
"both",
"clustering",
"and",
"hierarchical",
"relations"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Each",
"KPH",
"was",
"annotated",
"separately",
"by",
"three",
"of",
"the",
"authors",
"and",
"took",
"about",
"one",
"hour",
"to",
"complete",
"per",
"annotator"
] | [
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Our",
"annotation",
"guidelines",
"are",
"detailed",
"in",
"Appendix",
"A.1",
"Since",
"the",
"key",
"points",
"were",
"extracted",
"automati-",
"cally",
",",
"some",
"of",
"them",
"did",
"not",
"satisfy",
"the",
"desired",
"prop-",
"erties",
"of",
"a",
"key",
"point",
"-",
"a",
"concise",
"and",
"self",
"contained",
"sentence",
"or",
"phrase",
"that",
"discusses",
"a",
"single",
"point",
"with",
"a",
"certain",
"polarity",
"(",
"Bar",
"-",
"Haim",
"et",
"al",
",",
"2021",
")",
"To",
"avoid",
"noise",
"in",
"THINK",
"P",
",",
"annotators",
"could",
"mark",
"such",
"badkey",
"points",
"as",
"candidates",
"for",
"removal",
"from",
"the",
"final",
"KPH"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName"
] |
[
"As",
"our",
"annotation",
"tool",
",",
"we",
"used",
"CoRefi",
"(",
"Born-",
"stein",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2020",
")",
",",
"an",
"interface",
"for",
"cross",
"-",
"document",
"coreference",
"annotation",
"with",
"Cattan",
"et",
"al",
"(",
"2021",
")",
"βs",
"extension",
"for",
"annotating",
"a",
"forest",
"of",
"clusters",
",",
"which",
"we",
"adapted",
"to",
"handle",
"key",
"points",
"(",
"see",
"Ap-",
"pendix",
"A.2",
")",
"4.2.2",
"KPH",
"Consolidation",
"To",
"obtain",
"the",
"final",
"KPHs",
",",
"the",
"three",
"annotators",
"met",
"to",
"discuss",
"and",
"resolve",
"the",
"differences",
"in",
"their",
"indi-",
"vidual",
"KPHs",
"annotations"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O"
] |
[
"This",
"is",
"a",
"complex",
"pro-",
"cess",
"because",
"both",
"clusters",
"and",
"the",
"relations",
"be-",
"tween",
"them",
"can",
"differ",
"We",
"therefore",
"separated",
"the",
"consolidation",
"process",
"into",
"two",
"subsequent",
"stages",
":",
"clustering",
"and",
"hierarchy",
"In",
"the",
"first",
"phase",
",",
"following",
"the",
"reviewer",
"mode",
"in",
"CoRefi",
"(",
"Bornstein",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O"
] |
[
",",
"2020",
")",
",",
"annotators",
"were",
"shown",
"one",
"key",
"point",
"at",
"a",
"time",
"with",
"their",
"origi-",
"nal",
"clustering",
"decisions",
"In",
"case",
"of",
"disagreement",
",",
"the",
"annotators",
"discussed",
"and",
"reached",
"a",
"joint",
"deci-",
"sion",
",",
"which",
"automatically",
"modified",
"their",
"original",
"KPH",
"accordingly"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O"
] |
[
"At",
"the",
"end",
"of",
"this",
"stage",
",",
"the",
"ini-",
"tial",
"KPH",
"of",
"each",
"of",
"the",
"annotators",
"was",
"modified",
"to",
"include",
"the",
"exact",
"same",
"nodes"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"In",
"the",
"second",
"phase",
",",
"since",
"each",
"key",
"point",
"has",
"a",
"single",
"parent",
",",
"we",
"could",
"easily",
"identify",
"the",
"remaining",
"disagreements",
"by",
"comparing",
"the",
"parent",
"of",
"each",
"node",
"across",
"the",
"different",
"annotators",
"To",
"support",
"this",
"consolidation",
"phase",
",",
"we",
"enhanced",
"CoRefi",
"with",
"the",
"ability",
"to",
"iden-",
"tify",
"and",
"highlight",
"both",
"clustering",
"and",
"hierarchy",
"disagreements",
"between",
"any",
"number",
"of",
"annotators",
"(",
"see",
"Appendix",
"A.3",
"for",
"more",
"details",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Consolidating",
"multiple",
"annotations",
"was",
"also",
"ef-",
"ficient",
"due",
"to",
"the",
"hierarchical",
"structure",
"of",
"the",
"KPH",
"and",
"took",
"about",
"an",
"hour",
"per",
"KPH"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName"
] |
[
"4.2.3",
"Dataset",
"Quality",
"Assessment",
"To",
"verify",
"the",
"quality",
"of",
"the",
"resulting",
"dataset",
",",
"we",
"asked",
"two",
"additional",
"annotators",
"to",
"annotate",
"and",
"consolidate",
"a",
"portion",
"of",
"THINK",
"P",
"(",
"3RESTAU",
"-",
"RANTS",
",",
"2HOTEL",
"and",
"2",
"PC",
")",
".4We",
"then",
"evaluated",
"their",
"individual",
"and",
"consolidated",
"KPHs",
"against",
"our",
"consolidated",
"annotation",
",",
"as",
"follows"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"In",
"each",
"do-",
"main",
",",
"we",
"compared",
"the",
"two",
"sets",
"of",
"annotated",
"KPHs",
"by",
"taking",
"the",
"union",
"of",
"the",
"KP",
"relations",
"induced",
"by",
"the",
"KPHs",
"in",
"each",
"set",
"(",
"Eq"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"1",
")",
",",
"and",
"computing",
"the",
"F1",
"score",
"over",
"the",
"two",
"resulting",
"sets",
"of",
"relations",
"The",
"4See",
"Appendix",
"A.4",
"for",
"more",
"details",
"about",
"annotators",
"train-",
"ing.915",
"REST",
"HOTEL",
"PC",
"Total",
"#",
"KPHs",
"4",
"4",
"4",
"12",
"#",
"Key",
"points",
"181",
"208",
"128",
"517",
"#",
"Filtered",
"KPs",
"21",
"17",
"48",
"86",
"#",
"R",
"(",
"H",
")",
"850",
"302",
"266",
"1,418",
"Table",
"1",
":",
"Statistics",
"of",
"THINK",
"P.R",
"(",
"H",
")",
"is",
"the",
"set",
"of",
"key",
"point",
"relations",
"that",
"can",
"be",
"derived",
"from",
"a",
"KPH",
"H",
"(",
"Β§3",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"O",
"B-MetricName",
"I-MetricName",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"O",
"B-MetricName",
"I-MetricName",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"O",
"B-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"B-MetricValue",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O"
] |
[
"final",
"F1",
"was",
"obtained",
"by",
"macro",
"-",
"averaging",
"over",
"the",
"three",
"domains"
] | [
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"The",
"annotators",
"β",
"performance",
"after",
"consolida-",
"tion",
"reached",
"an",
"F1",
"of",
"0.756",
",",
"indicating",
"substan-",
"tial",
"agreement.5Furthermore",
",",
"consolidation",
"was",
"shown",
"to",
"increase",
"individual",
"performance",
"by",
"5",
"-",
"6",
"points"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O",
"B-MetricValue",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"4.3",
"Dataset",
"Properties",
"Table",
"1",
"shows",
"some",
"statistics",
"for",
"the",
"THINK",
"P",
"dataset"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O"
] |
[
"Overall",
",",
"THINK",
"Pincludes",
"12",
"KPHs",
",",
"517",
"key",
"points",
",",
"and",
"1,418",
"key",
"points",
"relations",
"(",
"R",
"(",
"H",
")",
")",
"out",
"of",
"the",
"total",
"24,430",
"key",
"point",
"pairs"
] | [
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Due",
"to",
"its",
"size",
",",
"we",
"did",
"not",
"split",
"THINK",
"Pinto",
"development",
"and",
"test",
"sets",
",",
"but",
"rather",
"used",
"the",
"entire",
"dataset",
"for",
"evaluation"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"As",
"described",
"in",
"Section",
"4.2.1",
",",
"during",
"the",
"annotation",
",",
"we",
"filter",
"a",
"relatively",
"small",
"number",
"of",
"key",
"points",
"(",
"14",
"%",
")",
",",
"mostly",
"from",
"the",
"PCdomain",
"This",
"is",
"mainly",
"because",
"the",
"key",
"point",
"quality",
"model",
"that",
"we",
"used",
"was",
"not",
"trained",
"on",
"this",
"domain",
"From",
"a",
"qualitative",
"perspective",
",",
"THINK",
"Phas",
"sev-",
"eral",
"appealing",
"properties",
"that",
"make",
"it",
"a",
"valuable",
"benchmark",
"for",
"NLI"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"First",
",",
"recall",
"that",
"the",
"KPA",
"algo-",
"rithm",
"aims",
"to",
"remove",
"similar",
"key",
"points",
"to",
"avoid",
"re-",
"dundancy",
"in",
"the",
"summary",
"(",
"Bar",
"-",
"Haim",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2020b",
")",
"Hence",
",",
"remaining",
"equivalent",
"key",
"points",
"in",
"THINK",
"P",
"are",
"mostly",
"non",
"-",
"trivial",
"paraphrases",
"that",
"are",
"challeng-",
"ing",
"to",
"detect",
"(",
"e.g.",
",",
"β",
"Took",
"forever",
"to",
"get",
"our",
"room",
"β",
"ββLines",
"to",
"check",
"in",
"are",
"ridiculous",
"β",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"In",
"addition",
",",
"hierarchical",
"relations",
"between",
"key",
"points",
"represent",
"diverse",
"types",
"of",
"inferences",
"Table",
"2",
"shows",
"a",
"few",
"examples",
"of",
"common",
"relations",
"we",
"observed",
"by",
"ana-",
"lyzing",
"a",
"sample",
"from",
"the",
"dataset",
"Finally",
",",
"THINK",
"P",
"comprises",
"naturally",
"-",
"occurring",
"texts",
"and",
"relations",
",",
"coming",
"from",
"real",
"-",
"world",
"data"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"5",
"Automatic",
"KPH",
"Construction",
"We",
"use",
"a",
"two",
"-",
"step",
"approach",
"to",
"automatically",
"build",
"a",
"KPH",
"from",
"a",
"set",
"of",
"key",
"points"
] | [
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"In",
"the",
"first",
"step",
",",
"we",
"predict",
"directional",
"scores",
"between",
"all",
"pairs",
"of",
"5We",
"do",
"not",
"report",
"Kappa",
"because",
"decisions",
"are",
"mutually",
"dependent.key",
"points",
"(",
"Β§",
"5.1",
")",
"In",
"the",
"second",
"step",
",",
"we",
"construct",
"a",
"hierarchy",
"based",
"on",
"the",
"local",
"scores",
"(",
"Β§",
"5.2",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O"
] |
[
"5.1",
"Scoring",
"Pairwise",
"Key",
"Point",
"Relations",
"Given",
"a",
"pair",
"of",
"key",
"points",
"(",
"i",
",",
"j",
")",
",",
"we",
"aim",
"to",
"pre-",
"dict",
"whether",
"a",
"directional",
"relation",
"iβ",
"βjholds",
"between",
"iandj",
",",
"by",
"computing",
"a",
"likelihood",
"score",
"s",
"(",
"i",
",",
"j",
")",
"β",
"[",
"0,1",
"]"
] | [
"O",
"B-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"experimented",
"with",
"both",
"exist-",
"ing",
"baselines",
"and",
"new",
"methods",
"we",
"developed",
"for",
"this",
"task",
"Due",
"to",
"the",
"size",
"of",
"THINK",
"P",
",",
"it",
"was",
"not",
"used",
"to",
"fine",
"-",
"tune",
"the",
"scoring",
"models",
"(",
"Β§",
"4.3",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Baselines",
"Identifying",
"directional",
"relations",
"be-",
"tween",
"two",
"key",
"points",
"is",
"closely",
"related",
"to",
"two",
"exist-",
"ing",
"tasks",
":",
"Textual",
"Entailment",
",",
"also",
"known",
"as",
"Natu-",
"ral",
"Language",
"Inference",
"(",
"NLI",
")",
"(",
"Dagan",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O"
] |
[
",",
"2007",
")",
"and",
"matching",
"arguments",
"to",
"key",
"points",
"(",
"Bar",
"-",
"Haim",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2020a",
")",
"Accordingly",
",",
"we",
"implemented",
"two",
"baselines",
":",
"(",
"1",
")",
"NLI",
",",
"a",
"RoBERTa",
"model",
"(",
"Liu",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2019",
")",
"fine",
"-",
"tuned",
"on",
"the",
"MNLI",
"dataset",
"(",
"Williams",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O"
] |
[
",",
"2018",
")",
"to",
"predict",
"whether",
"ientails",
"j6and",
"(",
"2",
")",
"KPA",
"-",
"Match",
",",
"a",
"RoBERTa",
"model",
"trained",
"on",
"the",
"ArgKP",
"dataset",
"(",
"Bar",
"-",
"Haim",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2020a",
")",
"to",
"predict",
"whether",
"imatches",
"j",
",",
"following",
"(",
"Bar",
"-",
"Haim",
"et",
"al",
",",
"2021",
")",
"βs",
"implementation",
"Directional",
"Distributional",
"Similarity",
"Geffet",
"and",
"Dagan",
"(",
"2005",
")",
"introduced",
"the",
"distributional",
"inclusion",
"hypothesis",
"for",
"lexical",
"entailment",
"(",
"Gef-",
"fet",
"and",
"Dagan",
",",
"2004",
")",
",",
"which",
"suggests",
"that",
"the",
"context",
"surrounding",
"an",
"entailing",
"word",
"w1is",
"nat-",
"urally",
"expected",
"to",
"occur",
"also",
"with",
"the",
"entailed",
"word",
"w2",
"Specifically",
",",
"for",
"each",
"word",
"w",
",",
"they",
"built",
"a",
"sparse",
"feature",
"vector",
"where",
"the",
"value",
"of",
"the",
"i",
"-",
"th",
"entry",
"is",
"the",
"PMI",
"of",
"the",
"i",
"-",
"th",
"word",
"in",
"the",
"dictionary",
"with",
"w",
"Many",
"distributional",
"similar-",
"ity",
"metrics",
"have",
"been",
"proposed",
"to",
"predict",
"direc-",
"tional",
"relations",
"such",
"as",
"hyponymy",
"between",
"a",
"pair",
"of",
"words",
",",
"based",
"on",
"their",
"distributional",
"feature",
"vec-",
"tors",
"Among",
"these",
"methods",
"are",
"WeedsPrec",
"(",
"Weeds",
"and",
"Weir",
",",
"2003",
")",
",",
"BInc",
"(",
"Szpektor",
"and",
"Dagan",
",",
"2008",
")",
",",
"ClarkeDE",
"(",
"Clarke",
",",
"2009",
")",
"and",
"APinc",
"(",
"Kotlerman",
"et",
"al",
",",
"2009",
")",
"In",
"this",
"work",
",",
"we",
"argue",
"that",
"this",
"distributional",
"inclusion",
"hypothesis",
"may",
"be",
"extended",
"to",
"identify",
"directional",
"relations",
"between",
"two",
"key",
"points",
"In-",
"deed",
",",
"if",
"iβ",
"βj",
",",
"it",
"is",
"likely",
"that",
"an",
"input",
"sentence",
"that",
"matches",
"the",
"key",
"point",
"iwill",
"also",
"match",
"j",
"For",
"example",
",",
"the",
"sentence",
"β",
"The",
"beds",
"were",
"really",
"comfortable",
",",
"I",
"literally",
"knocked",
"out",
"as",
"soon",
"as",
"my",
"head",
"touched",
"the",
"pillow",
"β",
"matches",
"both",
"β",
"The",
"beds",
"6https",
":",
"/",
"/",
"huggingface.co",
"/",
"roberta-large-mnli916",
"Relation",
"Type",
"Examples",
"Support",
"/",
"ElaborationHousekeeping",
"needs",
"worked",
"on",
"β",
"βThe",
"beds",
"were",
"nβt",
"even",
"made",
"right",
"The",
"room",
"was",
"poorly",
"maintained",
"β",
"βThe",
"air",
"conditioning",
"was",
"not",
"functioning",
"right",
"The",
"device",
"itself",
"is",
"so",
"difficult",
"to",
"use",
"β",
"βTransferring",
"data",
"was",
"a",
"nightmare",
"!",
"Customer",
"service",
"is",
"a",
"joke",
"β",
"βNo",
"help",
"moving",
"rooms",
"Part",
"-",
"ofThehardware",
"is",
"fantastic",
"β",
"βSound",
"is",
"surprisingly",
"good",
"Thetheatre",
"is",
"great",
"β",
"βTheentrance",
"is",
"absolutely",
"beautiful",
"IS",
"-",
"AThetoiletries",
"they",
"offer",
"are",
"the",
"worst",
"β",
"βnot",
"even",
"good",
"shampoo",
"in",
"room",
"Food",
"varieties",
"was",
"very",
"limited",
"β",
"βDesert",
"selection",
"was",
"below",
"average",
"as",
"well",
"Table",
"2",
":",
"Examples",
"of",
"relations",
"between",
"key",
"points",
"in",
"T",
"HINK",
"P"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"I-DatasetName"
] |
[
"were",
"awesome",
"β",
"andβThe",
"rooms",
"are",
"comfortable",
"β",
"Therefore",
",",
"we",
"construct",
"a",
"feature",
"vector",
"for",
"each",
"key",
"point",
"k",
",",
"whose",
"length",
"is",
"equal",
"to",
"the",
"number",
"of",
"input",
"sentences",
"The",
"value",
"at",
"the",
"i",
"-",
"th",
"position",
"in",
"this",
"vector",
"is",
"the",
"likelihood",
"that",
"the",
"i",
"-",
"th",
"sen-",
"tence",
"matches",
"k",
",",
"as",
"predicted",
"by",
"the",
"KPA",
"match-",
"ing",
"model",
"(",
"Β§",
"4.1",
")",
"Then",
",",
"we",
"apply",
"the",
"aforemen-",
"tioned",
"distributional",
"similarity",
"metrics",
"to",
"predict",
"a",
"directional",
"score",
"s",
"(",
"i",
",",
"j",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"only",
"report",
"the",
"perfor-",
"mance",
"of",
"APinc",
"as",
"it",
"slightly",
"outperformed",
"other",
"metrics"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Additionally",
",",
"we",
"implemented",
"a",
"simple",
"variant",
"of",
"WeedsPrec",
",",
"in",
"which",
"the",
"entries",
"in",
"the",
"feature",
"vectors",
"are",
"binary",
"(",
"match",
"/",
"no",
"match",
")",
"This",
"metric",
",",
"termed",
"Binary",
"Inclusion",
"(",
"BinInc",
")",
",",
"com-",
"putes",
"the",
"ratio",
"between",
"the",
"number",
"of",
"sentences",
"matched",
"by",
"KPA",
"to",
"both",
"iandjand",
"the",
"number",
"sentences",
"matched",
"to",
"i"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Intuitively",
",",
"when",
"most",
"of",
"the",
"sentences",
"that",
"were",
"mapped",
"to",
"iwere",
"also",
"mapped",
"to",
"j",
",",
"it",
"is",
"a",
"strong",
"indication",
"that",
"iβ",
"βj",
"Combining",
"NLI",
"with",
"Distributional",
"Methods"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O"
] |
[
"As",
"further",
"discussed",
"in",
"Section",
"6",
",",
"we",
"empirically",
"found",
"that",
"the",
"NLI",
"model",
"and",
"the",
"distributional",
"methods",
"have",
"complementary",
"strengths",
"The",
"NLI",
"model",
"performs",
"better",
"on",
"RESTAURANTS",
",",
"whereas",
"the",
"distributional",
"methods",
"perform",
"better",
"on",
"the",
"HOTEL",
"andPCdomains",
"Furthermore",
",",
"even",
"within",
"each",
"domain",
",",
"those",
"two",
"methods",
"produce",
"very",
"dif-",
"ferent",
"rankings",
",",
"as",
"indicated",
"by",
"a",
"low",
"Spearman",
"correlation",
"between",
"their",
"output",
"scores",
"(",
"see",
"Ap-",
"pendix",
"C",
"for",
"more",
"details",
")",
"To",
"take",
"advantage",
"of",
"the",
"strengths",
"of",
"both",
"ap-",
"proaches",
",",
"we",
"explored",
"two",
"alternatives",
"for",
"com-",
"bining",
"BinInc",
",",
"the",
"best",
"-",
"performing",
"distributional",
"method",
"(",
"as",
"shown",
"in",
"Section",
"6",
")",
",",
"with",
"NLI",
":",
"1.Averaging",
"the",
"output",
"scores",
"of",
"NLI",
"and",
"Bin-",
"Inc",
"(",
"denoted",
"NLI+BinInc",
"-",
"Avg",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"2.Fine",
"-",
"tuning",
"the",
"NLI",
"model",
"on",
"weak",
"la-",
"bels",
"created",
"by",
"the",
"BinInc",
"model",
"(",
"denotedNLI+BinInc",
"-",
"WL",
")"
] | [
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O"
] |
[
"Specifically",
",",
"we",
"first",
"apply",
"theBinInc",
"method",
"to",
"a",
"large",
"number",
"of",
"unla-",
"beled",
"KPA",
"summaries",
"and",
"obtain",
"local",
"scores",
"between",
"all",
"pairs",
"of",
"key",
"points"
] | [
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"then",
"con-",
"vert",
"these",
"pairwise",
"scores",
"to",
"the",
"NLI",
"format",
",",
"where",
"we",
"consider",
"all",
"pairs",
"above",
"some",
"thresh-",
"old",
"as",
"entailment",
"and",
"the",
"others",
"as",
"neutral",
"Finally",
",",
"we",
"fine",
"-",
"tune",
"the",
"NLI",
"model",
"on",
"this",
"automatically",
"-",
"generated",
"training",
"data",
"and",
"use",
"the",
"resulting",
"model",
"to",
"predict",
"the",
"local",
"scores",
"s",
"(",
"i",
",",
"j",
")",
"onTHINK",
"P"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName"
] |
[
"Implementation",
"details",
"and",
"statistics",
"on",
"the",
"silver",
"data",
"are",
"detailed",
"in",
"Appendix",
"B",
"5.2",
"Hierarchy",
"Construction",
"We",
"proceed",
"to",
"construct",
"a",
"KPH",
"by",
"determining",
"its",
"semantic",
"clusters",
"and",
"the",
"hierarchical",
"relations",
"between",
"them"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Intuitively",
",",
"we",
"would",
"like",
"to",
"gener-",
"ate",
"a",
"KPH",
"such",
"that",
"the",
"set",
"of",
"pairwise",
"key",
"point",
"relations",
"induced",
"by",
"its",
"structure",
"are",
"consistent",
"with",
"the",
"local",
"directional",
"scores",
":",
"high",
"-",
"scoring",
"rela-",
"tions",
"should",
"be",
"included",
",",
"and",
"low",
"-",
"scoring",
"relations",
"should",
"be",
"excluded"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"explored",
"several",
"alter-",
"natives",
"for",
"constructing",
"a",
"KPH",
",",
"described",
"below"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O"
] |
[
"Each",
"of",
"these",
"methods",
"employs",
"a",
"decision",
"thresh-",
"oldΟover",
"the",
"local",
"scores",
",",
"which",
"needs",
"to",
"be",
"tuned",
"over",
"some",
"development",
"data"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"B-HyperparameterValue",
"I-HyperparameterValue",
"I-HyperparameterValue",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Reduced",
"Forest",
"(",
"2012",
")",
"described",
"a",
"simple",
"transformation",
"of",
"a",
"directed",
"graph",
"Ginto",
"a",
"forest",
"of",
"clusters",
"In",
"our",
"case",
",",
"we",
"start",
"with",
"a",
"graph",
"that",
"includes",
"the",
"key",
"points",
"as",
"nodes",
",",
"and",
"the",
"directional",
"edges",
"e",
"(",
"i",
",",
"j",
")",
"for",
"pairs",
"with",
"local",
"score",
"s",
"(",
"i",
",",
"j",
")",
">",
"Ο",
"The",
"reduced",
"forest",
"is",
"constructed",
"as",
"follows",
":",
"(",
"a",
")",
"the",
"condensation",
"of",
"Gis",
"computed",
"by",
"contracting",
"each",
"strongly",
"connected",
"component",
"into",
"a",
"single",
"vertex",
"that",
"represents",
"a",
"cluster",
"of",
"nodes",
"in",
"G",
"The",
"resulting",
"DAG",
"is",
"transformed",
"into",
"a",
"forest",
"by",
"(",
"b",
")",
"taking",
"its",
"transitive",
"reduction",
",",
"and",
"(",
"c",
")",
"heuristically917",
"selecting",
"a",
"single",
"parent",
"for",
"each",
"node",
"with",
"multi-",
"ple",
"parents",
"We",
"select",
"the",
"larger",
"cluster",
"as",
"a",
"parent",
",",
"and",
"as",
"a",
"tie",
"breaker",
",",
"we",
"use",
"the",
"mean",
"over",
"all",
"the",
"pairwise",
"scores",
"s",
"(",
"i",
",",
"j",
")",
"such",
"that",
"iis",
"in",
"the",
"child",
"cluster",
"and",
"jis",
"in",
"the",
"parent",
"cluster",
"As",
"defined",
"by",
"Berant",
"et",
"al",
",",
"Gis",
"aForest",
"Re-",
"ducible",
"Graph",
"(",
"FRG",
")",
"if",
"after",
"applying",
"step",
"babove",
",",
"none",
"of",
"the",
"nodes",
"has",
"multiple",
"parents",
"Tree",
"Node",
"and",
"Component",
"Fix",
"(",
"TNCF",
")",
"Given",
"a",
"directed",
"graph",
"with",
"local",
"edge",
"weights",
"that",
"are",
"either",
"positive",
"(",
"predicting",
"pairwise",
"entailment",
"be-",
"tween",
"connected",
"nodes",
")",
"or",
"negative",
"(",
"predicting",
"non",
"-",
"entailment",
")",
",",
"the",
"optimal",
"entailment",
"graph",
"may",
"be",
"defined",
"as",
"the",
"transitive",
"subgraph",
"in",
"which",
"the",
"sum",
"of",
"the",
"edge",
"weights",
"is",
"maximized",
"(",
"Berant",
"et",
"al",
",",
"2012",
")",
"showed",
"that",
"this",
"problem",
"is",
"NP-",
"Hard",
",",
"even",
"when",
"further",
"constraining",
"the",
"resulting",
"graph",
"to",
"be",
"forest",
"-",
"reducible",
"To",
"address",
"the",
"computational",
"complexity",
"of",
"find-",
"ing",
"an",
"exact",
"solution",
",",
"Berant",
"et",
"al",
"presented",
"an",
"ef-",
"ficient",
"approximation",
"algorithm",
",",
"termed",
"Tree",
"-",
"node-",
"fix",
"(",
"TNF",
")",
"that",
"generates",
"forest",
"-",
"reducible",
"entailment",
"graphs",
",",
"and",
"showed",
"empirically",
"that",
"the",
"quality",
"of",
"the",
"resulting",
"graphs",
"is",
"close",
"to",
"the",
"exact",
"solu-",
"tion",
"found",
"via",
"Integer",
"Linear",
"Programming",
"(",
"ILP",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName"
] |
[
"Starting",
"from",
"some",
"initial",
"FRG",
",",
"their",
"algorithm",
"it-",
"eratively",
"improves",
"the",
"graph",
"objective",
"function",
"by",
"removing",
"and",
"reattaching",
"one",
"node",
"at",
"a",
"time",
",",
"while",
"keeping",
"the",
"graph",
"forest",
"-",
"reducible",
"(",
"2015",
")",
"proposed",
"an",
"extension",
"for",
"this",
"algorithm",
",",
"termed",
"Tree",
"-",
"Node",
"-",
"and",
"-",
"Component-",
"Fix",
"(",
"TNCF",
")",
",",
"where",
"in",
"each",
"iteration",
"a",
"whole",
"cluster",
"may",
"be",
"re",
"-",
"attached",
",",
"in",
"addition",
"to",
"individual",
"nodes"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"found",
"this",
"extension",
"beneficial",
"Since",
"a",
"KPH",
"is",
"also",
"a",
"forest",
"of",
"clusters",
",",
"the",
"TNF",
"and",
"TNCF",
"algorithms",
"are",
"directly",
"applicable",
"to",
"our",
"setting"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"(",
"2012",
")",
"we",
"defined",
"the",
"edge",
"weights",
"as",
"wi",
",",
"j",
"=",
"s",
"(",
"i",
",",
"j",
")",
"βΟso",
"that",
"local",
"scores",
"below",
"the",
"threshold",
"Οare",
"considered",
"negative"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-HyperparameterName",
"O",
"O",
"O"
] |
[
"One",
"difference",
"between",
"the",
"original",
"TNF",
"imple-",
"mentation",
"and",
"ours",
"is",
"the",
"initialization",
":",
"while",
"they",
"used",
"(",
"Berant",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2011",
")",
"βs",
"exact",
"solution",
",",
"com-",
"puted",
"via",
"ILP",
"for",
"a",
"sparse",
"configuration",
",",
"we",
"take",
"a",
"simpler",
"approach",
"and",
"start",
"with",
"the",
"reduced",
"forest",
"described",
"above",
",",
"constructed",
"with",
"the",
"same",
"thresh-",
"oldΟ",
"Greedy",
"As",
"an",
"alternative",
"to",
"the",
"TNF",
"/",
"TNCF",
"al-",
"gorithms",
",",
"we",
"also",
"adapted",
"the",
"greedy",
"algorithm",
"proposed",
"by",
"Cattan",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"(",
"2021",
")",
"for",
"the",
"task",
"of",
"hi",
"-",
"erarchical",
"cross",
"-",
"document",
"coreference",
"resolution",
",",
"which",
"also",
"generates",
"a",
"forest",
"of",
"clusters",
"First",
",",
"key",
"point",
"clusters",
"are",
"obtained",
"by",
"agglomerative",
"clus-",
"tering",
"with",
"average",
"linkage",
"and",
"distance",
"threshold",
"of1βΟ",
",",
"where",
"the",
"distance",
"metric",
"between",
"two",
"key",
"points",
"iandjis",
"defined",
"as",
"1βmin",
"(",
"s",
"(",
"i",
",",
"j",
")",
",",
"s",
"(",
"j",
",",
"i",
")",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-HyperparameterValue",
"I-HyperparameterValue",
"I-HyperparameterValue",
"I-HyperparameterValue",
"I-HyperparameterValue",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Second",
",",
"we",
"define",
"the",
"score",
"of",
"the",
"directional",
"edge",
"between",
"two",
"clusters",
"(",
"C1",
",",
"C2",
")",
"as",
"the",
"average",
"of",
"the",
"s",
"(",
"i",
",",
"j",
")",
"scores",
"between",
"the",
"key",
"points",
"in",
"the",
"two",
"clusters",
":",
"S",
"(",
"C1",
",",
"C2",
")",
"=",
"1",
"|C1|",
"Β·",
"|C",
"2|",
"/",
"summationdisplay",
"iβC1",
"/",
"summationdisplay",
"jβC2s",
"(",
"i",
",",
"j",
")",
"(",
"2",
")",
"The",
"KPH",
"is",
"constructed",
"by",
"repeatedly",
"adding",
"the",
"highest",
"-",
"scoring",
"edge",
"(",
"if",
"the",
"score",
"is",
"above",
"the",
"Ο",
"threshold",
")",
",",
"skipping",
"edges",
"that",
"would",
"violate",
"the",
"definition",
"of",
"the",
"KPH",
"as",
"a",
"directed",
"forest",
"The",
"process",
"is",
"terminated",
"when",
"no",
"more",
"edges",
"can",
"be",
"added",
"Note",
"that",
"unlike",
"the",
"TNF",
"/",
"TNCF",
"algorithms",
",",
"the",
"Greedy",
"algorithm",
"does",
"not",
"modify",
"existing",
"clus-",
"ters",
"and",
"edges",
"in",
"each",
"iteration",
",",
"but",
"only",
"adds",
"new",
"edges"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"B-MethodName",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Greedy",
"with",
"Global",
"Score",
"(",
"Greedy",
"GS",
")"
] | [
"B-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName",
"I-MethodName"
] |
[
"One",
"limitation",
"of",
"the",
"Greedy",
"algorithm",
"is",
"that",
"the",
"edge",
"scoring",
"function",
"is",
"local",
"and",
"hence",
"ignores",
"indirect",
"relations",
"between",
"clusters",
"that",
"would",
"result",
"from",
"adding",
"the",
"edge"
] | [
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"For",
"example",
",",
"consider",
"a",
"KPH",
"with",
"three",
"clusters",
"{",
"A",
",",
"B",
",",
"C",
"}",
"such",
"that",
"Bβ",
"βA",
"The",
"criterion",
"to",
"add",
"the",
"edge",
"Cβ",
"βBwill",
"consider",
"onlyS",
"(",
"C",
",",
"B",
")",
"but",
"not",
"S",
"(",
"C",
",",
"A",
")",
",",
"which",
"corresponds",
"to",
"the",
"indirect",
"relation",
"C",
";",
"A",
"To",
"address",
"this",
"issue",
",",
"we",
"modified",
"the",
"algorithm",
"to",
"consider",
"the",
"relations",
"between",
"each",
"cluster",
"and",
"all",
"its",
"ancestors",
"in",
"the",
"resulting",
"KPH",
",",
"as",
"follows",
":",
"Ek+1",
"=",
"Ekβͺargmax",
"Ο΅βEβ\\EkO",
"(",
"V",
",",
"EkβͺΟ΅",
")",
"(",
"3",
")",
"O",
"(",
"V",
",",
"E",
")",
"=",
"/",
"summationdisplay",
"CiβV",
"/",
"summationdisplay",
"CjβAV",
",",
"E",
"(",
"Ci",
")",
"S",
"(",
"Ci",
",",
"Cj",
")",
"(",
"4",
")",
"where",
"Ekis",
"the",
"set",
"of",
"edges",
"in",
"the",
"resulting",
"KPH",
"afterkiterations",
",",
"Eβis",
"the",
"set",
"of",
"all",
"edges",
"scoring",
"above",
"ΟandAV",
",",
"E",
"(",
"C",
")",
"denotes",
"the",
"set",
"of",
"ancestors",
"ofCinH",
"(",
"V",
",",
"E",
")",
"6",
"Evaluation",
"Predicting",
"Local",
"Pairwise",
"Relations",
"Figure",
"2",
"compares",
"the",
"performance",
"of",
"the",
"different",
"local",
"scoring",
"methods",
"(",
"Β§",
"5.1",
")",
"For",
"each",
"domain",
",",
"we",
"con-",
"sider",
"all",
"the",
"key",
"point",
"pairs",
"in",
"the",
"dataset",
",",
"and",
"show918",
"Figure",
"2",
":",
"Precision",
"-",
"Recall",
"curves",
"of",
"local",
"scoring",
"methods",
"on",
"R",
"ESTAURANT",
",",
"HOTEL",
"and",
"PC",
"the",
"Precision",
"/",
"Recall",
"curve",
"and",
"the",
"Area",
"Under",
"the",
"Curve",
"(",
"AUC",
")",
"for",
"each",
"method"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O",
"B-MetricName",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"I-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"AUC",
"results",
"are",
"also",
"summarized",
"in",
"Table",
"3"
] | [
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"first",
"observe",
"that",
"applying",
"the",
"KPA",
"-",
"match",
"model",
"indirectly",
"via",
"the",
"distributional",
"methods",
"(",
"AP-",
"incandBinInc",
")",
"outperforms",
"its",
"direct",
"application",
"in",
"two",
"out",
"of",
"the",
"three",
"domains",
",",
"and",
"increases",
"the",
"average",
"AUC",
"from",
"0.237",
"to",
"0.277",
"/",
"0.288",
",",
"respec-",
"tively"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"B-MetricValue",
"O",
"B-MetricValue",
"I-MetricValue",
"I-MetricValue",
"O",
"O",
"O"
] |
[
"The",
"NLImodel",
"has",
"a",
"clear",
"advantage",
"over",
"the",
"distributional",
"methods",
"in",
"the",
"RESTAURANTS",
"domain",
",",
"but",
"is",
"much",
"worse",
"for",
"HOTEL",
"andPC"
] | [
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Both",
"NLI+BinInc",
"-",
"Avg",
"andNLI+BinInc",
"-",
"WL",
"models",
"are",
"able",
"to",
"combine",
"the",
"complementary",
"strengths",
"of",
"NLIandBinInc",
"and",
"outperform",
"all",
"the",
"stand",
"-",
"alone",
"models"
] | [
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Model",
"combination",
"via",
"weak",
"labeling",
"(",
"NLI+BinInc",
"-",
"WL",
")",
"achieves",
"the",
"best",
"performance",
"in",
"all",
"three",
"domains",
"by",
"a",
"large",
"margin",
"(",
"+0.11",
"aver-",
"age",
"AUC",
"improvement",
"over",
"the",
"best",
"stand",
"-",
"alone",
"method",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricValue",
"O",
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"To",
"further",
"assess",
"the",
"contribution",
"of",
"model",
"com-",
"bination",
"in",
"the",
"weak",
"labeling",
"setting",
",",
"we",
"also",
"tested",
"a",
"configuration",
"in",
"which",
"the",
"silver",
"data",
"is",
"labeled",
"by",
"the",
"NLI",
"model",
"(",
"denoted",
"NLI",
"-",
"WL",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O"
] |
[
"The",
"results",
"are",
"shown",
"on",
"the",
"last",
"row",
"of",
"Table",
"3",
"While",
"the",
"performance",
"is",
"better",
"than",
"NLI",
"alone",
"(",
"demonstrat-",
"ing",
"the",
"value",
"of",
"weak",
"labeling",
")",
",",
"it",
"is",
"still",
"far",
"be-",
"lowNLI+BinInc",
"-",
"WL"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName"
] |
[
"Overalll",
",",
"the",
"results",
"affirm",
"the",
"importance",
"of",
"both",
"model",
"combination",
"and",
"the",
"weakly",
"-",
"labeled",
"data",
"for",
"local",
"scoring",
"performance",
"Hierarchy",
"Construction",
"Next",
",",
"we",
"compare",
"dif-",
"ferent",
"methods",
"for",
"constructing",
"a",
"KPH",
"from",
"the",
"set",
"of",
"local",
"pairwise",
"scores",
"(",
"Β§",
"5.2",
")",
"We",
"use",
"the",
"scores",
"from",
"the",
"best",
"performing",
"local",
"method",
",",
"NLI+BinInc",
"-",
"WL",
",",
"as",
"found",
"in",
"the",
"previous",
"experi-",
"ment"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"use",
"the",
"F1",
"measure",
"as",
"defined",
"in",
"Sec-",
"tion",
"4.2.3",
"as",
"our",
"evaluation",
"metric",
",",
"similar",
"toKotlerman",
"et",
"al"
] | [
"O",
"O",
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Since",
"THINK",
"Phas",
"no",
"development",
"set",
"(",
"Β§",
"4.3",
")",
",",
"we",
"employ",
"a",
"leave",
"-",
"one",
"-",
"out",
"scheme",
"to",
"tune",
"the",
"threshold",
"Ο"
] | [
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-HyperparameterName",
"I-HyperparameterName"
] |
[
"Specifically",
",",
"for",
"each",
"KPA",
"summary",
"S",
",",
"we",
"find",
"the",
"threshold",
"that",
"maximizes",
"the",
"F1",
"score",
"of",
"the",
"three",
"other",
"KPHs",
"in",
"the",
"same",
"domain",
"and",
"predict",
"a",
"KPH",
"for",
"Susing",
"this",
"threshold"
] | [
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"then",
"compute",
"the",
"F1",
"score",
"for",
"the",
"predicted",
"KPHs",
"in",
"each",
"domain"
] | [
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"The",
"results",
"are",
"summarized",
"in",
"Table",
"4",
"TNCF",
"achieves",
"the",
"best",
"overall",
"performance",
"on",
"T",
"HINK",
"P",
"with",
"an",
"average",
"F1",
"of",
"0.526",
",",
"substantially",
"improv-",
"ing",
"the",
"Reduced",
"Forest",
"baseline"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"B-MetricName",
"O",
"B-MetricValue",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName"
] |
[
"The",
"Greedy",
"GS",
"algorithm",
"is",
"the",
"top",
"performer",
"in",
"the",
"Restaurants",
"domain",
"(",
"F1=0.641",
")"
] | [
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O"
] |
[
"Adding",
"a",
"global",
"scoring",
"func-",
"tion",
"to",
"the",
"greedy",
"algorithm",
"improves",
"the",
"perfor-",
"mance",
"by",
"0.059",
"(",
"from",
"0.45",
"to",
"0.509",
")"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricValue",
"O",
"O",
"B-MetricValue",
"O",
"B-MetricValue",
"O"
] |
[
"We",
"also",
"evaluated",
"the",
"quality",
"of",
"the",
"predicted",
"relations",
"using",
"only",
"the",
"local",
"scores",
",",
"with",
"a",
"thresh-",
"old",
"determined",
"via",
"leave",
"-",
"one",
"-",
"out",
",",
"as",
"before",
"(",
"last",
"row",
"in",
"Table",
"4",
")",
"While",
"the",
"resulting",
"set",
"of",
"relations",
"may",
"not",
"represent",
"a",
"valid",
"hierarchy",
",",
"it",
"still",
"provides",
"an",
"interesting",
"reference",
"point",
"for",
"comparison",
"with",
"the",
"various",
"KPH",
"construction",
"algorithms",
"We",
"can",
"see",
"that",
"both",
"Greedy",
"GS",
"andTNCF",
"improve",
"the",
"local",
"results",
"by",
"a",
"substantial",
"margin",
"(",
"+0.028",
"and",
"+0.045",
",",
"resp"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricValue",
"O",
"B-MetricValue",
"O",
"O"
] |
[
")",
"These",
"two",
"global",
"methods",
"not",
"only",
"satisfy",
"the",
"constraints",
"of",
"generating",
"a",
"valid",
"KPH",
",",
"but",
"also",
"improve",
"the",
"pairwise",
"relation",
"prediction",
"of",
"the",
"local",
"scorer",
"7",
"Conclusion",
"We",
"introduced",
"Key",
"Point",
"Hierarchies",
"as",
"a",
"novel",
"representation",
"for",
"structured",
",",
"expressive",
"opinion",
"summaries"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"We",
"explored",
"several",
"approaches",
"for",
"au-",
"tomatic",
"hierarchy",
"construction",
"from",
"a",
"given",
"set",
"of",
"key",
"points",
",",
"which",
"were",
"evaluated",
"on",
"a",
"new",
"bench-",
"mark",
"dataset",
"we",
"developed",
"for",
"this",
"task",
"We",
"also919",
"REST",
"HOTEL",
"PC",
"Avg",
"All",
"methods",
"use",
"the",
"NLI+BinInc",
"-",
"WL",
"local",
"scores"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName",
"I-MethodName",
"O",
"O"
] |
[
"proposed",
"a",
"novel",
"distributional",
"representation",
"for",
"key",
"points",
",",
"which",
"we",
"leveraged",
"via",
"weak",
"super-",
"vision",
"to",
"achieve",
"substantial",
"improvement",
"on",
"the",
"subtask",
"of",
"predicting",
"pairwise",
"key",
"point",
"relations",
"While",
"our",
"initial",
"results",
"are",
"promising",
",",
"there",
"is",
"still",
"much",
"room",
"for",
"improvement",
",",
"and",
"we",
"hope",
"that",
"releasing",
"our",
"dataset",
"would",
"encourage",
"the",
"commu-",
"nity",
"to",
"further",
"promote",
"this",
"line",
"of",
"research",
"Limitations",
"Key",
"Point",
"Hierarchies",
"may",
"be",
"valuable",
"for",
"sum-",
"marizing",
"opinions",
"and",
"views",
"in",
"multiple",
"domains",
",",
"including",
"reviews",
",",
"survey",
"responses",
",",
"customer",
"feed-",
"back",
",",
"political",
"debates",
"etc",
"However",
",",
"in",
"this",
"work",
",",
"we",
"only",
"demonstrated",
"their",
"value",
"for",
"business",
"and",
"product",
"reviews",
",",
"leaving",
"other",
"types",
"of",
"data",
"to",
"fu-",
"ture",
"work",
"Also",
",",
"we",
"only",
"attempted",
"to",
"create",
"KPHs",
"for",
"English",
"reviews",
",",
"for",
"which",
"an",
"abundance",
"of",
"resources",
"is",
"available",
",",
"including",
"a",
"huge",
"number",
"of",
"written",
"reviews",
"and",
"high",
"-",
"quality",
"trained",
"mod-",
"els",
",",
"e.g",
"for",
"NLI",
"and",
"key",
"point",
"matching"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName"
] |
[
"Apply-",
"ing",
"these",
"methods",
"to",
"low",
"-",
"resource",
"languages",
"is",
"expected",
"to",
"be",
"far",
"more",
"challenging",
"Finally",
",",
"the",
"quality",
"of",
"the",
"resulting",
"KPHs",
"depends",
"on",
"the",
"qual-",
"ity",
"of",
"the",
"extracted",
"key",
"points",
"provided",
"as",
"input",
",",
"which",
"may",
"vary",
"across",
"different",
"domains",
"To",
"alle-",
"viate",
"this",
"problem",
"in",
"THINK",
"P",
",",
"we",
"manually",
"filtered",
"out",
"problematic",
"key",
"points",
"from",
"the",
"dataset",
"(",
"Β§",
"4.2",
")",
".Acknowledgments",
"The",
"first",
"author",
"is",
"partially",
"supported",
"by",
"the",
"PBC",
"fellowship",
"for",
"outstanding",
"PhD",
"candidates",
"in",
"data",
"science"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"I-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Proceedings",
"of",
"the",
"61st",
"Annual",
"Meeting",
"of",
"the",
"Association",
"for",
"Computational",
"Linguistics",
"Volume",
"1",
":",
"Long",
"Papers",
",",
"pages",
"929β949",
"July",
"9",
"-",
"14",
",",
"2023",
"Β©",
"2023",
"Association",
"for",
"Computational",
"Linguistics",
"When",
"to",
"Use",
"What",
":",
"An",
"In",
"-",
"Depth",
"Comparative",
"Empirical",
"Analysis",
"of",
"OpenIE",
"Systems",
"for",
"Downstream",
"Applications",
"Kevin",
"Peia",
",",
"Ishan",
"Jindalb",
",",
"Kevin",
"Chen",
"-",
"Chuan",
"Changa",
",",
"Chengxiang",
"Zhaia",
",",
"Yunyao",
"Lic",
"aUniversity",
"of",
"Illinois",
"at",
"Urbana",
"-",
"Champaign",
",",
"bIBM",
"Research",
",",
"cApple",
"{",
"kspei2",
",",
"kcchang",
",",
"czhai",
"}",
"@",
"illinois.edu",
",",
"ishan.jindal",
"@",
"ibm.com",
",",
"yunyaoli",
"@",
"apple.com",
"Abstract",
"Open",
"Information",
"Extraction",
"(",
"OpenIE",
")",
"has",
"been",
"used",
"in",
"the",
"pipelines",
"of",
"various",
"NLP",
"tasks",
"Unfortunately",
",",
"there",
"is",
"no",
"clear",
"consensus",
"on",
"which",
"models",
"to",
"use",
"for",
"which",
"tasks",
"Muddy-",
"ing",
"things",
"further",
"is",
"the",
"lack",
"of",
"comparisons",
"that",
"take",
"differing",
"training",
"sets",
"into",
"account",
"In",
"this",
"paper",
",",
"we",
"present",
"an",
"application",
"-",
"focused",
"empirical",
"survey",
"of",
"neural",
"OpenIE",
"models",
",",
"training",
"sets",
",",
"and",
"benchmarks",
"in",
"an",
"effort",
"to",
"help",
"users",
"choose",
"the",
"most",
"suitable",
"OpenIE",
"systems",
"for",
"their",
"applications",
"We",
"find",
"that",
"the",
"different",
"assumptions",
"made",
"by",
"different",
"mod-",
"els",
"and",
"datasets",
"have",
"a",
"statistically",
"significant",
"effect",
"on",
"performance",
",",
"making",
"it",
"important",
"to",
"choose",
"the",
"most",
"appropriate",
"model",
"for",
"one",
"βs",
"applications",
"We",
"demonstrate",
"the",
"applicabil-",
"ity",
"of",
"our",
"recommendations",
"on",
"a",
"downstream",
"Complex",
"QA",
"application",
"1",
"Introduction",
"Open",
"Information",
"Extraction",
"(",
"OpenIE",
")",
"is",
"the",
"task",
"of",
"extracting",
"relation",
"tuples",
"from",
"plain",
"text",
"(",
"An-",
"geli",
"et",
"al",
",",
"2015",
")",
"In",
"its",
"simplest",
"form",
",",
"OpenIE",
"extracts",
"information",
"in",
"the",
"form",
"of",
"tuples",
"consist-",
"ing",
"of",
"subject",
"(",
"S",
")",
",",
"predicate",
"(",
"P",
")",
",",
"object",
"(",
"O",
")",
",",
"and",
"any",
"additional",
"arguments",
"(",
"A",
")",
"OpenIE",
"is",
"an",
"open",
"do-",
"main",
",",
"intended",
"to",
"be",
"easy",
"to",
"deploy",
"in",
"different",
"domains",
"without",
"fine",
"-",
"tuning",
",",
"with",
"all",
"relations",
"ex-",
"tracted",
"regardless",
"of",
"type"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"The",
"increasing",
"avail-",
"ability",
"of",
"semi",
"-",
"automatically",
"generated",
"training",
"datasets",
"(",
"Cui",
"et",
"al",
",",
"2018",
")",
"as",
"well",
"as",
"significant",
"ad-",
"vances",
"in",
"deep",
"learning",
"techniques",
"have",
"led",
"to",
"the",
"development",
"of",
"state",
"-",
"of",
"-",
"the",
"-",
"art",
"neural",
"models",
"(",
"Cui",
"et",
"al",
",",
"2018",
";",
"Garg",
"and",
"Kalai",
",",
"2018",
")",
"Since",
"its",
"introduction",
"in",
"Etzioni",
"et",
"al",
"(",
"2008",
")",
",",
"OpenIE",
"has",
"attracted",
"a",
"large",
"amount",
"of",
"attention",
"by",
"the",
"research",
"community",
"as",
"a",
"tool",
"for",
"a",
"wide",
"range",
"of",
"downstream",
"NLP",
"tasks",
"(",
"Mausam",
",",
"2016",
")",
"How-",
"ever",
",",
"there",
"is",
"no",
"real",
"consensus",
"on",
"which",
"OpenIE",
"model",
"is",
"best",
"for",
"each",
"application",
"One",
"example",
"of",
"this",
"lack",
"of",
"consensus",
"in",
"summarization",
",",
"where",
"dif-",
"ferent",
"papers",
"use",
"OLLIE",
"(",
"Christensen",
"et",
"al",
",",
"2014",
")",
",",
"Sentence",
"Bill",
"Gates",
",",
"former",
"CEO",
"of",
"Microsoft",
",",
"is",
"a",
"Harvard",
"dropout",
"OpenIE",
"Extractions",
"(",
"Bill",
"Gates",
",",
"was",
",",
"former",
"CEO",
"of",
"Microsoft",
")",
"(",
"Bill",
"Gates",
",",
"is",
",",
"a",
"Harvard",
"dropout",
")",
"Applications",
"QAWho",
"was",
"former",
"CEO",
"of",
"Microsoft",
"?"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"Where",
"did",
"Bill",
"Gates",
"dropout",
"of",
"?",
"Slot",
"Filling",
"(",
"?",
",",
"was",
",",
"former",
"CEO",
"of",
"Microsoft",
")",
"(",
"?",
",",
"is",
",",
"a",
"Harvard",
"dropout",
")",
"Table",
"1",
":",
"Sample",
"relation",
"tuples",
"and",
"examples",
"of",
"how",
"different",
"applications",
"use",
"OpenIE",
"extractions",
"MinIE",
"(",
"Ponza",
"et",
"al",
",",
"2018",
")",
",",
"and",
"Stanford",
"CoreNLP",
"(",
"Cao",
"et",
"al",
",",
"2018",
";",
"Zhang",
"et",
"al",
",",
"2021",
")",
"for",
"extraction",
"Different",
"applications",
"may",
"also",
"have",
"different",
"re-",
"quirements",
"As",
"an",
"example",
",",
"choosing",
"a",
"model",
"that",
"assumes",
"all",
"relations",
"only",
"have",
"a",
"subject",
"and",
"ob-",
"ject",
"may",
"not",
"be",
"suitable",
"for",
"event",
"schema",
"induction",
"since",
"that",
"excludes",
"any",
"event",
"schemas",
"with",
"more",
"than",
"two",
"entities",
"The",
"papers",
"that",
"introduce",
"new",
"OpenIE",
"models",
"and",
"datasets",
"do",
"not",
"specify",
"how",
"downstream",
"applications",
"would",
"be",
"impacted",
"by",
"the",
"different",
"assumptions",
"those",
"models",
"make",
"about",
"extracted",
"relations",
"We",
"find",
"that",
"prior",
"OpenIE",
"surveys",
"are",
"also",
"insuf-",
"ficient",
"to",
"find",
"the",
"best",
"OpenIE",
"model",
"for",
"a",
"given",
"application",
"The",
"only",
"previous",
"application",
"-",
"focused",
"OpenIE",
"survey",
"we",
"found",
"was",
"Mausam",
"(",
"2016",
")",
"However",
",",
"this",
"survey",
"does",
"not",
"identify",
"the",
"desired",
"properties",
"of",
"OpenIE",
"for",
"those",
"applications",
"or",
"pro-",
"vide",
"an",
"empirical",
"comparison",
"of",
"OpenIE",
"systems",
"Glauber",
"and",
"Claro",
"(",
"2018",
")",
",",
"Claro",
"et",
"al",
"(",
"2019",
")",
",",
"and",
"Zhou",
"et",
"al",
"(",
"2022",
")",
"also",
"do",
"not",
"provide",
"an",
"empirical",
"application",
"-",
"focused",
"survey",
"Another",
"obstacle",
"is",
"the",
"lack",
"of",
"apples",
"-",
"to",
"-",
"apples",
"comparisons",
"between",
"OpenIE",
"models",
"Compar-",
"isons",
"should",
"keep",
"the",
"training",
"set",
",",
"benchmark",
",",
"and",
"evaluation",
"metric",
"constant",
"to",
"eliminate",
"con-",
"founders",
"Unfortunately",
",",
"the",
"papers",
"that",
"intro-929",
"Question",
"AnsweringSlot",
"FillingEvent",
"Schema",
"InductionSummarizationKnowledge",
"Base",
"Population",
"HR",
":",
"Higher",
"Recall",
"β",
"β",
"β",
"β",
"β",
"HP",
":",
"Higher",
"Precision",
"β",
"β",
"N",
"-",
"ary",
":",
"N",
"-",
"ary",
"Relation",
"Extraction",
"β",
"β",
"β",
"IN",
":",
"Inferred",
"Relation",
"Extraction",
"β",
"β",
"β",
"β",
"FE",
":",
"Fast",
"Extraction",
"β",
"Table",
"2",
":",
"Properties",
"explicitly",
"mentioned",
"in",
"application",
"papers",
"as",
"motivation",
"for",
"choosing",
"a",
"particular",
"OpenIE",
"model",
"or",
"as",
"a",
"way",
"to",
"improve",
"performance",
"within",
"a",
"case",
"study",
"There",
"are",
"additional",
"desired",
"properties",
"we",
"omit",
"that",
"no",
"existing",
"OpenIE",
"models",
"or",
"datasets",
"possess",
",",
"such",
"as",
"the",
"canonicalization",
"of",
"extracted",
"relations",
"and",
"the",
"ability",
"to",
"extract",
"relations",
"from",
"imperative",
"sentences",
"(",
"Fader",
"et",
"al",
",",
"2013",
";",
"Khot",
"et",
"al",
",",
"2017",
";",
"Zhang",
"et",
"al",
",",
"2021",
")",
"duce",
"new",
"OpenIE",
"models",
"often",
"do",
"not",
"provide",
"this",
"apples",
"-",
"to",
"-",
"apples",
"comparison",
"For",
"example",
",",
"CopyAttention",
"(",
"Cui",
"et",
"al",
",",
"2018",
")",
",",
"SpanOIE",
"(",
"Zhan",
"and",
"Zhao",
",",
"2020",
")",
",",
"IMoJIE",
"(",
"Kolluru",
"et",
"al",
",",
"2020b",
")",
",",
"and",
"OpenIE6",
"(",
"Kolluru",
"et",
"al",
",",
"2020a",
")",
"all",
"compare",
"their",
"model",
"to",
"models",
"trained",
"on",
"different",
"training",
"sets",
"OpenIE6",
"reports",
"performance",
"on",
"the",
"WiRe57",
"benchmark",
"which",
"Multi2OIE",
"(",
"Ro",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"B-DatasetName",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O"
] |
[
",",
"2020",
")",
"does",
"not",
",",
"but",
"Multi2OIE",
"reports",
"performance",
"on",
"the",
"ReOIE2016",
"benchmark",
"which",
"OpenIE6",
"does",
"not"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"B-DatasetName",
"O",
"O",
"B-MethodName",
"O",
"O"
] |
[
"Because",
"the",
"training",
"set",
"can",
"greatly",
"affect",
"the",
"per-",
"formance",
"of",
"a",
"neural",
"model",
",",
"we",
"focus",
"on",
"selecting",
"both",
"the",
"appropriate",
"OpenIE",
"model",
"and",
"training",
"set",
",",
"which",
"we",
"refer",
"to",
"as",
"an",
"OpenIE",
"System",
"To",
"resolve",
"our",
"lack",
"of",
"understanding",
",",
"we",
"focus",
"on",
"one",
"particular",
"question",
":",
"How",
"do",
"I",
"choose",
"a",
"particular",
"OpenIE",
"system",
"for",
"a",
"given",
"application",
"?",
"Different",
"implicit",
"assumptions",
"about",
"OpenIE",
"may",
"have",
"a",
"significant",
"impact",
"on",
"the",
"performance",
"of",
"downstream",
"applications",
"such",
"as",
"the",
"assumptions",
"that",
"all",
"relations",
"are",
"verb",
"-",
"based",
"(",
"Zhan",
"and",
"Zhao",
",",
"2020",
")",
"or",
"that",
"all",
"relations",
"have",
"only",
"a",
"subject",
"and",
"object",
"(",
"Kolluru",
"et",
"al",
",",
"2020b",
")",
"To",
"answer",
"this",
"ques-",
"tion",
"an",
"apples",
"-",
"to",
"-",
"apples",
"comparison",
"must",
"be",
"con-",
"ducted",
"for",
"different",
"application",
"settings",
"Because",
"it",
"is",
"impractical",
"to",
"find",
"the",
"best",
"model",
"for",
"every",
"application",
"given",
"the",
"many",
"possible",
"appli-",
"cations",
"of",
"OpenIE",
",",
"we",
"instead",
"characterize",
"appli-",
"cations",
"based",
"on",
"what",
"properties",
"they",
"desire",
"from",
"OpenIE",
"such",
"as",
"the",
"desire",
"for",
"N",
"-",
"ary",
"relation",
"ex-",
"traction",
"by",
"event",
"schema",
"induction",
"We",
"provide",
"an",
"extensive",
"apples",
"-",
"to",
"-",
"apples",
"comparison",
"of",
"neural",
"OpenIE",
"models",
"such",
"that",
"a",
"practitioner",
"can",
"uti-",
"lize",
"our",
"practical",
"observations",
"to",
"effectively",
"select",
"a",
"neural",
"OpenIE",
"model",
"and",
"training",
"set",
"for",
"their",
"downstream",
"application",
"Finally",
",",
"we",
"apply",
"our",
"rec-",
"ommendations",
"to",
"a",
"downstream",
"Complex",
"QA",
"task",
"In",
"summary",
",",
"our",
"contributions",
"are",
"as",
"follows",
":",
"β’We",
"propose",
"a",
"taxonomy",
"that",
"covers",
"OpenIE",
"training",
"sets",
",",
"benchmarks",
",",
"and",
"neural",
"models",
"β’We",
"present",
"an",
"extensive",
"empirical",
"comparison",
"of",
"different",
"models",
"on",
"different",
"datasets",
"with",
"recommendations",
"based",
"on",
"the",
"results",
"β’We",
"perform",
"a",
"case",
"study",
"on",
"Complex",
"QA",
"to",
"show",
"the",
"efficacy",
"of",
"our",
"recommendations"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"To",
"the",
"best",
"of",
"our",
"knowledge",
",",
"our",
"survey",
"is",
"the",
"only",
"application",
"-",
"focused",
"empirical",
"survey",
"on",
"Ope-",
"nIE",
"datasets",
",",
"metrics",
",",
"and",
"neural",
"OpenIE",
"models",
"2",
"Motivating",
"Applications",
"In",
"this",
"section",
",",
"we",
"identify",
"the",
"properties",
"of",
"OpenIE",
"desired",
"by",
"5",
"downstream",
"applications",
":",
"Slot",
"Filling",
",",
"Question",
"Answering",
"(",
"QA",
")",
",",
"Summarization",
",",
"Event",
"Schema",
"Induction",
",",
"and",
"Knowledge",
"Base",
"Popula-",
"tion"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"B-TaskName",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName"
] |
[
"We",
"survey",
"how",
"OpenIE",
"is",
"used",
"and",
"the",
"prop-",
"erties",
"explicitly",
"desired",
"by",
"papers",
"corresponding",
"to",
"the",
"application",
",",
"either",
"as",
"motivation",
"for",
"choosing",
"a",
"given",
"OpenIE",
"model",
"or",
"within",
"a",
"case",
"study",
"as",
"a",
"property",
"that",
"would",
"improve",
"performance",
"The",
"desired",
"properties",
"we",
"observe",
"are",
"Higher",
"Recall",
",",
"Higher",
"Precision",
",",
"N",
"-",
"ary",
"Relation",
"Extrac-",
"tion",
",",
"Inferred",
"Relation",
"Extraction",
",",
"and",
"Fast",
"Ex-",
"traction",
"We",
"define",
"an",
"\"",
"Inferred",
"Relation",
"\"",
"(",
"IN",
")",
"to",
"be",
"a",
"relation",
"that",
"contains",
"words",
"that",
"are",
"not",
"in",
"the",
"original",
"sentence",
"For",
"example",
",",
"given",
"the",
"sentence",
"\"",
"Bill",
"Gates",
",",
"former",
"CEO",
"of",
"Microsoft",
",",
"is",
"a",
"Harvard",
"dropout",
"\"",
",",
"the",
"relation",
"(",
"Bill",
"Gates",
",",
"was",
",",
"former",
"CEO",
"of",
"Microsoft",
")",
"can",
"be",
"inferred",
"even",
"though",
"\"",
"was",
"\"",
"is",
"not",
"in",
"the",
"original",
"sentence",
"We",
"define",
"an",
"\"",
"N",
"-",
"ary",
"Relation",
"\"",
"(",
"N",
"-",
"ary",
")",
"to",
"be",
"a",
"relation",
"with",
"more",
"arguments",
"than",
"just",
"(",
"subject",
",",
"predicate",
",",
"object",
")",
"For",
"example",
",",
"the",
"relation",
"(",
"Alice",
",",
"went",
",",
"to",
"the",
"store",
",",
"today",
")",
"has",
"an",
"additional",
"argument",
"today",
"Table",
"2",
"provides",
"a",
"summary",
"the",
"explicitly",
"desired",
"properties",
"of",
"downstream",
"applications.930",
"Dataset",
"Creation",
"Method",
"Source",
"#",
"Extractions",
"#",
"IN",
"#",
"N",
"-",
"ary",
"Training",
"SetsSpanOIE",
"Weak",
"Labeling",
"Wikipedia",
"2,175",
"K",
"2",
"K",
"231",
"K",
"OIE4",
"Weak",
"Labeling",
"Wikipedia",
"181",
"K",
"3",
"K",
"34",
"K",
"IMoJIE",
"Weak",
"Labeling",
"Wikipedia",
"215",
"K",
"3",
"K",
"0",
"LSOIE",
"Weak",
"Labeling",
"QA",
"-",
"SRL",
"2.0",
"Wikipedia",
",",
"Science",
"101",
"K",
"0",
"32",
"K",
"Test",
"SetsOIE2016",
"Weak",
"Labeling",
"QA",
"-",
"SRL",
"1,730",
"359",
"708",
"WiRe57",
"Manual",
"Annotation",
"Wikipedia",
"and",
"Newswire",
"343",
"173",
"79",
"ReOIE2016",
"Manual",
"Annotation",
"OIE2016",
"1,508",
"155",
"611",
"CaRB",
"Crowdsourced",
"Annotation",
"OIE2016",
"5,263",
"736",
"683",
"LSOIE",
"Weak",
"Labeling",
"QA",
"-",
"SRL",
"2.0",
"Wikipedia",
",",
"Science",
"22,376",
"0",
"4,920",
"Table",
"3",
":",
"Comparison",
"of",
"the",
"attributes",
"of",
"different",
"datasets",
"#",
"Extractions",
":",
"Number",
"of",
"Extractions",
",",
"#",
"IN",
":",
"Number",
"of",
"inferred",
"relations",
",",
"#",
"N",
"-",
"ary",
":",
"Number",
"of",
"N",
"-",
"ary",
"Relations",
"Slot",
"Filling",
"Slot",
"Filling",
"is",
"a",
"task",
"where",
"an",
"incom-",
"plete",
"tuple",
"must",
"be",
"completed",
"using",
"information",
"from",
"a",
"given",
"corpus",
"(",
"Chen",
"et",
"al",
",",
"2019",
")",
"For",
"exam-",
"ple",
",",
"the",
"incomplete",
"tuple",
"(",
"Obama",
",",
"born",
"in",
",",
"?",
")",
"must",
"be",
"completed",
"as",
"(",
"Obama",
",",
"was",
"born",
"in",
",",
"Honolulu",
")",
"using",
"information",
"from",
"the",
"corpus",
"OpenIE",
"can",
"be",
"used",
"to",
"extract",
"complete",
"tuples",
"which",
"fill",
"slots",
"in",
"an",
"incomplete",
"tuple",
"using",
"entity",
"linking"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"I-MethodName"
] |
[
"Soderland",
"et",
"al",
"(",
"2013",
")",
",",
"Angeli",
"et",
"al",
"(",
"2015",
")",
",",
"Soderland",
"et",
"al",
"(",
"2015b",
")",
",",
"and",
"Soderland",
"et",
"al",
"(",
"2015a",
")",
"take",
"advan-",
"tage",
"of",
"how",
"correct",
"relations",
"often",
"appear",
"multiple",
"times",
"to",
"match",
"empty",
"slots",
"to",
"the",
"highest",
"precision",
"OpenIE",
"tuple"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O",
"O"
] |
[
"They",
"state",
"in",
"their",
"case",
"studies",
"they",
"would",
"benefit",
"from",
"INextraction",
"and",
"Soderland",
"et",
"al",
"(",
"2015b",
")",
"and",
"Soderland",
"et",
"al",
"(",
"2015a",
")",
"state",
"they",
"would",
"benefit",
"from",
"N",
"-",
"ary",
"extraction"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName"
] |
[
"These",
"two",
"properties",
"allow",
"more",
"relation",
"surface",
"forms",
"to",
"be",
"extracted",
",",
"which",
"increases",
"the",
"chance",
"an",
"in-",
"complete",
"tuple",
"can",
"be",
"linked",
"to",
"a",
"complete",
"tuple",
"Question",
"Answering",
"We",
"focus",
"on",
"two",
"subtasks",
"of",
"Question",
"Answering",
"(",
"QA",
")",
"that",
"utilize",
"OpenIE",
":",
"Open",
"-",
"domain",
"QA",
"(",
"OpenQA",
")",
"and",
"Complex",
"QA",
"OpenQA",
"involves",
"answering",
"questions",
"given",
"a",
"large",
"database",
"(",
"Fader",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2014a",
")",
"Complex",
"QA",
"involves",
"using",
"information",
"from",
"multiple",
"sen-",
"tences",
"to",
"find",
"answers",
"and",
"requires",
"inferring",
"re-",
"lationships",
"between",
"multiple",
"entities",
"(",
"Chali",
"et",
"al",
",",
"2009",
")",
"Fader",
"et",
"al",
"(",
"2013",
",",
"2014b",
")",
",",
"Yin",
"et",
"al",
"(",
"2015",
")",
",",
"and",
"Clark",
"et",
"al",
"(",
"2018",
")",
"are",
"OpenQA",
"methods",
"that",
"use",
"retrieval",
"-",
"based",
"methods",
"to",
"match",
"OpenIE",
"ex-",
"tractions",
"to",
"questions",
"By",
"rewriting",
"queries",
"into",
"incomplete",
"tuples",
",",
"such",
"as",
"rewriting",
"\"",
"Where",
"was",
"Obama",
"born",
"?",
"\"",
"into",
"(",
"Obama",
",",
"born",
"in",
",",
"?",
")",
",",
"it",
"is",
"possi-",
"ble",
"to",
"use",
"extracted",
"relations",
"to",
"answer",
"queries",
"by",
"filling",
"in",
"the",
"missing",
"slots",
"in",
"the",
"query",
"For",
"Com-",
"plexQA",
",",
"Khot",
"et",
"al",
"(",
"2017",
")",
"and",
"Lu",
"et",
"al",
"(",
"2019",
")",
"generate",
"graphs",
"from",
"extracted",
"relation",
"tuples",
",",
"then",
"reason",
"over",
"these",
"graphs",
"to",
"answer",
"questions",
"Inall",
"QA",
"applications",
"surveyed",
",",
"high",
"recall",
"(",
"HR",
")",
"is",
"desired",
",",
"with",
"Lu",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"(",
"2019",
")",
"using",
"a",
"custom",
"Ope-",
"nIE",
"method",
"specifically",
"for",
"higher",
"recall",
"Yin",
"et",
"al",
"(",
"2015",
")",
"βs",
"case",
"studies",
"state",
"that",
"N",
"-",
"ary",
"would",
"be",
"ben-",
"eficial",
"while",
"Lu",
"et",
"al",
"(",
"2019",
")",
"uses",
"a",
"custom",
"OpenIE",
"method",
"that",
"supports",
"IN",
"Summarization",
"OpenIE",
"addresses",
"the",
"problems",
"of",
"redundancy",
"and",
"fact",
"fabrication",
"in",
"summarization",
"Redundancy",
"is",
"when",
"a",
"fact",
"is",
"repeated",
"multiple",
"times",
"in",
"the",
"summary",
"To",
"combat",
"redundancy",
",",
"Ope-",
"nIE",
"is",
"used",
"to",
"ensure",
"that",
"the",
"generated",
"summary",
"does",
"not",
"have",
"repeated",
"relations",
"(",
"Christensen",
"et",
"al",
",",
"2014",
";",
"Zhang",
"et",
"al",
",",
"2021",
")",
"Fact",
"fabrication",
"is",
"when",
"a",
"fact",
"that",
"is",
"not",
"supported",
"by",
"the",
"text",
"being",
"summa-",
"rized",
"is",
"in",
"the",
"summary",
"To",
"combat",
"fact",
"fabrication",
",",
"OpenIE",
"is",
"used",
"to",
"ensure",
"that",
"the",
"generated",
"sum-",
"mary",
"only",
"contains",
"relations",
"from",
"the",
"original",
"text",
"(",
"Cao",
"et",
"al",
",",
"2018",
";",
"Zhang",
"et",
"al",
",",
"2021",
")",
"In",
"summa-",
"rization",
"tasks",
",",
"HRis",
"useful",
"to",
"ensure",
"summaries",
"contain",
"all",
"information",
",",
"with",
"Ponza",
"et",
"al",
"(",
"2018",
")",
"citing",
"greater",
"diversity",
"of",
"extractions",
"as",
"a",
"way",
"to",
"improve",
"performance",
"high",
"precision",
"(",
"HP",
")",
"is",
"also",
"desired",
"by",
"Zhang",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"I-MetricName",
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
"(",
"2021",
")",
"in",
"order",
"to",
"reduce",
"redundant",
"extractions",
"Event",
"Schema",
"Induction",
"Event",
"Schema",
"Induction",
"is",
"the",
"automatic",
"discovery",
"of",
"patterns",
"that",
"indicate",
"events",
",",
"agents",
",",
"and",
"the",
"agents",
"β",
"roles",
"within",
"that",
"event",
"Extracted",
"relations",
"can",
"be",
"used",
"to",
"find",
"sur-",
"face",
"forms",
"of",
"events",
",",
"with",
"redundant",
"tuples",
"being",
"used",
"to",
"induce",
"event",
"schemas",
"The",
"open",
"nature",
"of",
"OpenIE",
"allows",
"for",
"events",
"to",
"be",
"found",
"regardless",
"of",
"the",
"domain",
"or",
"surface",
"form",
"HRis",
"useful",
"for",
"Event",
"Schema",
"Induction",
"for",
"the",
"same",
"reason",
"it",
"is",
"useful",
"for",
"Slot",
"Filling",
":",
"finding",
"more",
"surface",
"forms",
"allows",
"for",
"more",
"event",
"schemas",
"to",
"be",
"induced",
"(",
"Bala-",
"subramanian",
"et",
"al"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MetricName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
[
",",
"2013",
";",
"Romadhony",
"et",
"al",
",",
"2019",
";",
"Sahnoun",
"et",
"al",
",",
"2020",
")",
"Sahnoun",
"et",
"al",
"(",
"2020",
")",
"also",
"specifically",
"desire",
"INso",
"that",
"more",
"event",
"schemas931",
"can",
"be",
"learned",
",",
"while",
"Balasubramanian",
"et",
"al",
"(",
"2013",
")",
"state",
"that",
"N",
"-",
"ary",
"would",
"improve",
"performance",
"Knowledge",
"Base",
"Population",
"The",
"relations",
"ex-",
"tracted",
"by",
"OpenIE",
"can",
"be",
"used",
"to",
"automatically",
"populate",
"knowledge",
"bases",
"(",
"KBs",
")",
",",
"creating",
"new",
"nodes",
"and",
"edges",
"Muhammad",
"et",
"al",
"(",
"2020",
")",
"and",
"Kroll",
"et",
"al",
"(",
"2021",
")",
"use",
"learning",
"-",
"based",
"OpenIE",
"mod-",
"els",
"because",
"of",
"their",
"ability",
"to",
"generalize",
"to",
"unseen",
"relations",
"and",
"achieve",
"HR",
"Kroll",
"et",
"al",
"(",
"2021",
")",
"also",
"explicitly",
"chooses",
"Stanford",
"CoreNLP",
"and",
"OpenIE6",
"for",
"their",
"fast",
"extraction",
"times",
"(",
"FE",
")",
"3",
"OpenIE",
"Datasets",
"In",
"this",
"section",
",",
"we",
"discuss",
"the",
"differences",
"between",
"different",
"OpenIE",
"training",
"sets",
"and",
"benchmarks",
"and",
"their",
"shortcomings",
"We",
"provide",
"statistics",
"about",
"different",
"datasets",
"in",
"Table",
"3",
"3.1",
"Training",
"Datasets",
"Given",
"how",
"data",
"-",
"hungry",
"deep",
"learning",
"models",
"are",
"and",
"how",
"costly",
"it",
"is",
"to",
"manually",
"label",
"OpenIE",
"datasets",
",",
"most",
"OpenIE",
"training",
"sets",
"are",
"weakly",
"la-",
"beled",
"using",
"high",
"confidence",
"extractions",
"from",
"prior",
"OpenIE",
"models",
"CopyAttention",
"(",
"Cui",
"et",
"al",
",",
"2018",
")",
",",
"SpanOIE",
"(",
"Zhan",
"and",
"Zhao",
",",
"2020",
")",
",",
"and",
"OIE4",
"(",
"Kolluru",
"et",
"al",
",",
"2020b",
")",
"are",
"training",
"sets",
"consisting",
"of",
"high",
"confidence",
"Ope-",
"nIE4",
"extractions",
"from",
"Wikipedia",
"SpanOIE",
"includes",
"extractions",
"of",
"all",
"confidences",
"unlike",
"CopyAttention",
"and",
"OIE4",
"which",
"only",
"contain",
"extractions",
"above",
"a",
"certain",
"confidence",
"threshold",
"The",
"IMoJIE",
"dataset",
"(",
"Kolluru",
"et",
"al",
",",
"2020b",
")",
"at-",
"tempts",
"to",
"get",
"higher",
"quality",
"labels",
"by",
"combining",
"Wikipedia",
"extractions",
"from",
"OpenIE4",
",",
"ClausIE",
",",
"and",
"RNNOIE",
",",
"using",
"a",
"common",
"scoring",
"metric",
"to",
"com-",
"bine",
"extractions",
"and",
"filter",
"out",
"repeated",
"extractions",
"The",
"LSOIE",
"training",
"set",
"(",
"Solawetz",
"and",
"Larson",
",",
"2021",
")",
"is",
"composed",
"of",
"automatically",
"converted",
"Se-",
"mantic",
"Role",
"Labeling",
"(",
"SRL",
")",
"extractions",
"with",
"high",
"inter",
"-",
"annotator",
"agreement",
"from",
"the",
"Wikipedia",
"and",
"Science",
"domain",
"of",
"the",
"crowdsourced",
"QA",
"-",
"SRL",
"Bank",
"2.0",
"dataset"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-DatasetName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"I-TaskName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-TaskName",
"I-TaskName",
"O",
"B-DatasetName",
"I-DatasetName",
"I-DatasetName",
"O"
] |
[
"Because",
"this",
"dataset",
"is",
"derived",
"from",
"SRL",
",",
"all",
"relations",
"are",
"assumed",
"to",
"be",
"verb-",
"based",
"and",
"none",
"are",
"inferred",
"Issues",
"with",
"existing",
"training",
"sets",
"Current",
"OpenIE",
"training",
"sets",
"are",
"limited",
"to",
"Wikipedia",
"and",
"Science",
"domains",
",",
"which",
"may",
"not",
"generalize",
"to",
"certain",
"other",
"domains"
] | [
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"B-MethodName",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O",
"O"
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.