Datasets:
id
stringlengths 1
4
| tokens
sequence | ner_tags
sequence |
---|---|---|
0 | [
"References"
] | [
0
] |
1 | [
"Acknowledgments"
] | [
0
] |
2 | [
"We",
"hope",
"this",
"work",
"will",
"spur",
"more",
"research",
"in",
"how",
"to",
"better",
"use",
"pre-trained",
"encoder-decoders",
"for",
"not",
"only",
"MCQA,",
"but",
"also",
"beyond;",
"for",
"tasks",
"with",
"divergent",
"structures",
"from",
"the",
"pre-training,",
"a",
"smarter",
"use",
"of",
"PLMs",
"can",
"boost",
"the",
"performance",
"significantly."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
3 | [
"In",
"the",
"future,",
"we",
"will",
"focus",
"on",
"how",
"to",
"further",
"improve",
"the",
"clue",
"generation",
"quality,",
"which",
"remains",
"a",
"bottleneck",
"of",
"GenMC."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1
] |
4 | [
"Table",
"7:",
"Inference",
"time",
"for",
"answering",
"a",
"question",
"(seconds)."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
5 | [
"In",
"our",
"future",
"research,",
"we",
"will",
"focus",
"on",
"how",
"to",
"generate",
"more",
"helpful",
"clues",
"from",
"questions."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
6 | [
"This",
"suggests",
"a",
"great",
"room",
"for",
"improvement."
] | [
0,
0,
0,
0,
0,
0,
0
] |
7 | [
"Though",
"the",
"majority",
"of",
"our",
"clues",
"are",
"relevant",
"(i.e.,",
"76.4%",
"of",
"them",
"are",
"relevant",
"across",
"all",
"datasets),",
"which",
"seems",
"positive,",
"only",
"24%",
"of",
"the",
"clues",
"are",
"deemed",
"as",
"helpful."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
8 | [
"Figure",
"4",
"breaks",
"down",
"by",
"dataset."
] | [
0,
0,
0,
0,
0,
0
] |
9 | [
"Table",
"6",
"shows",
"the",
"percent",
"of",
"each",
"clue",
"type",
"across",
"all",
"datasets",
"with",
"an",
"example",
"for",
"each",
"type."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
10 | [
"If",
"all",
"three",
"students",
"annotate",
"differently",
"from",
"each",
"other",
"for",
"an",
"instance,",
"we",
"introduce",
"a",
"fourth",
"student",
"to",
"arbitrate."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
11 | [
"To",
"ensure",
"the",
"annotation",
"quality,",
"we",
"aggregate",
"annotated",
"results",
"from",
"three",
"students",
"for",
"every",
"dataset",
"using",
"majority",
"vote."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
12 | [
"to",
"answer",
"the",
"question."
] | [
0,
0,
0,
0
] |
13 | [
"The",
"clue",
"adds",
"helpful",
"information"
] | [
0,
0,
0,
0,
0
] |
14 | [
"•",
"Helpful:"
] | [
0,
0
] |
15 | [
"•",
"Relevant",
"but",
"unhelpful:",
"Though",
"relevant,",
"the",
"clue",
"makes",
"a",
"factually",
"incorrect",
"statement,",
"often",
"on",
"the",
"contrary",
"of",
"the",
"main",
"question,",
"or",
"the",
"clue",
"contributes",
"relevant",
"but",
"insufficient",
"knowledge",
"for",
"prediction,",
"such",
"as",
"repetition",
"of",
"the",
"question",
"or",
"other",
"distractors."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
16 | [
"understandable."
] | [
0
] |
17 | [
"5We",
"follow",
"a",
"similar",
"definition",
"by",
"Shwartz",
"et",
"al.",
"(2020)."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
18 | [
"They",
"know",
"and",
"agree",
"that",
"their",
"annotations",
"will",
"be",
"used",
"for",
"error",
"analysis",
"in",
"a",
"research",
"paper."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
19 | [
"4They",
"are",
"volunteers",
"recruited",
"from",
"the",
"contact",
"author’s",
"research",
"group."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
20 | [
"•",
"Irrelevant:",
"The",
"clue",
"is",
"off",
"topic",
"or",
"is",
"not"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
21 | [
"We",
"then",
"ask",
"them",
"to",
"categorize",
"clues",
"into",
"the",
"following",
"families:5"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
22 | [
"We",
"show",
"six",
"graduate",
"students",
"of",
"computer",
"science4",
"an",
"instance",
"along",
"with",
"the",
"generated",
"clue,",
"correct",
"answer,",
"and",
"predicted",
"answer."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
23 | [
"Specifically,",
"we",
"randomly",
"sample",
"50",
"negative",
"cases",
"from",
"T5LARGE",
"+",
"GenMC",
"for",
"each",
"dataset."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
1,
2,
2,
0,
0,
0
] |
24 | [
"By",
"studying",
"these",
"potentially",
"negative",
"clues,",
"we",
"can",
"gain",
"more",
"insights",
"into",
"how",
"GenMC",
"fails",
"and",
"discuss",
"venues",
"for",
"future",
"improvement."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0
] |
25 | [
"The",
"intuition",
"is",
"that",
"in",
"these",
"negative",
"cases,",
"the",
"clues",
"generated",
"by",
"GenMC",
"may",
"play",
"a",
"negative",
"role."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0
] |
26 | [
"We",
"analyze",
"the",
"clues",
"generated",
"by",
"GenMC",
"using",
"T5LARGE",
"with",
"a",
"focus",
"on",
"instances",
"that",
"are",
"correctly",
"predicted",
"by",
"the",
"baseline",
"in",
"our",
"main",
"experiments",
"(i.e.,",
"T5LARGE",
"+",
"Text2Textvanilla),",
"while",
"our",
"GenMC",
"fails."
] | [
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
2,
2,
0,
0,
1,
0
] |
27 | [
"5.4",
"Error",
"Analysis"
] | [
0,
0,
0
] |
28 | [
"demonstrates",
"that",
"naively",
"using",
"explicit",
"knowledge",
"in",
"plain",
"text,",
"instead",
"of",
"using",
"implicit",
"clues",
"from",
"the",
"decoder’s",
"hidden",
"state,",
"is",
"inferior",
"as",
"it",
"may",
"unnecessarily",
"bring",
"information",
"loss",
"and",
"noise."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
29 | [
"Instance",
"Which",
"would",
"you",
"likely",
"find",
"inside",
"a",
"beach",
"ball?"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
30 | [
"We",
"also",
"observe",
"that",
"the",
"performance",
"of",
"using",
"token-level",
"clues",
"lags",
"much",
"behind",
"GenMC."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1
] |
31 | [
"5.3.2",
"Results",
"Table",
"5",
"shows",
"that",
"masking",
"out",
"generation",
"loss",
"leads",
"to",
"substantial",
"performance",
"drops",
"across",
"all",
"datasets,",
"demonstrating",
"that",
"fine-tuning",
"the",
"decoder",
"GEN",
"helps",
"to",
"derive",
"useful",
"with",
"generation",
"loss",
"clues",
"from",
"pre-trained",
"encoder-decoder",
"models."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
32 | [
"(2020c),",
"which",
"also",
"adopts",
"a",
"pipeline",
"framework",
"to",
"first",
"generate",
"a",
"token-level",
"evidence",
"and",
"then",
"use",
"the",
"evidence",
"to",
"expand",
"the",
"question."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
33 | [
"This",
"variant",
"is",
"indeed",
"very",
"similar",
"to",
"Liu",
"et",
"al."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
34 | [
"We",
"then",
"directly",
"concatenate",
"C",
"with",
"Q",
"and",
"Oi",
"to",
"compute",
"a",
"score",
"for",
"Oi",
"using",
"the",
"model’s",
"encoder",
"part",
"stacked",
"with",
"an",
"MLP",
"layer."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
35 | [
"We",
"first",
"collect",
"the",
"generated",
"clue",
"text",
"C",
"(instead",
"of",
"its",
"representation)",
"from",
"the",
"decoder."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
36 | [
"In",
"this",
"setting,",
"we",
"separately",
"train",
"Token",
"Clue",
"a",
"clue",
"generator",
"and",
"a",
"reader."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
37 | [
"Intuitively,",
"under",
"this",
"setting,",
"the",
"generated",
"clue",
"is",
"weaker",
"than",
"GenMC",
"which",
"learns",
"how",
"to",
"generate",
"a",
"clue",
"with",
"supervision."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
38 | [
"We",
"train",
"this",
"variant",
"only",
"using",
"the",
"READ,",
"so",
"only",
"the",
"encoder",
"part",
"classification",
"loss",
"is",
"updated,",
"while",
"the",
"decoder",
"part",
"is",
"left",
"untouched",
"from",
"pre-training."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
39 | [
"5.3.1",
"Variants",
"of",
"GenMC",
"Weak",
"Clue"
] | [
0,
0,
0,
1,
0,
0
] |
40 | [
"To",
"better",
"understand",
"its",
"superior",
"results",
"and",
"the",
"influence",
"of",
"our",
"clue",
"generation,",
"we",
"compare",
"with",
"two",
"variants."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
41 | [
"Our",
"main",
"results",
"in",
"Section",
"5.1",
"have",
"demonstrated",
"the",
"effectiveness",
"of",
"our",
"model."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
42 | [
"As",
"a",
"fairer",
"comparison",
"in",
"Table",
"4,",
"by",
"unifying",
"the",
"training",
"sets",
"of",
"all",
"the",
"five",
"datasets,",
"our",
"GenMCT5-U",
"outperforms",
"UnifiedQAT5-FT",
"on",
"all",
"datasets",
"except",
"for",
"CSQA",
"with",
"large",
"models."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
1,
0,
0,
0,
0,
0,
13,
0,
0,
0
] |
43 | [
"The",
"promising",
"results",
"of",
"GenMC",
"further",
"reveals",
"that",
"our",
"model",
"can",
"learn",
"to",
"effectively",
"extract",
"knowledge",
"from",
"pre-trained",
"encoder-decoders",
"with",
"limited",
"training",
"data."
] | [
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
44 | [
"These",
"results",
"are",
"impressive",
"because",
"UnifiedQA",
"uses",
"more",
"datasets",
"(i.e.,",
"eight",
"different",
"QA",
"datasets)",
"for",
"training."
] | [
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
45 | [
"It",
"also",
"achieves",
"comparable",
"results",
"on",
"the",
"remaining",
"datasets."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
46 | [
"Moreover,",
"for",
"UnifiedQAT5-FT,",
"which",
"further",
"finetunes",
"the",
"model",
"on",
"the",
"training",
"set",
"of",
"the",
"target",
"dataset,",
"GenMCT5",
"outperforms",
"it",
"on",
"the",
"test",
"sets",
"of",
"CSQA,",
"OBQA,",
"and",
"ARC-Easy",
"for",
"the",
"base",
"models",
"and",
"ARC-Easy",
"for",
"the",
"large",
"models."
] | [
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
13,
13,
0,
13,
0,
0,
0,
0,
0,
13,
0,
0,
0,
0
] |
47 | [
"More",
"interestingly,",
"GenMCT5",
"also",
"performs",
"better",
"than",
"UnifiedQAT5",
"on",
"most",
"datasets."
] | [
0,
0,
1,
0,
0,
0,
0,
1,
0,
0,
0
] |
48 | [
"The",
"results",
"in",
"Table",
"3",
"show",
"that",
"GenMCT5"
] | [
0,
0,
0,
0,
0,
0,
0,
1
] |
49 | [
"significantly",
"(with",
"p-value",
"<",
"0.01)",
"outperforms",
"the",
"two",
"encoder-only",
"strong",
"baselines",
"RoBERTa",
"and",
"ALBERT."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
1
] |
50 | [
"5.2.2",
"Results"
] | [
0,
0
] |
51 | [
"All",
"models",
"are",
"of",
"comparable",
"model",
"size",
"to",
"ours."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
52 | [
"In",
"addition,",
"we",
"compare",
"with",
"two",
"encoder-only",
"models,",
"RoBERTa",
"(Liu",
"et",
"al.,",
"2019)",
"and",
"ALBERT",
"(Lan",
"et",
"al.,",
"2020),",
"which",
"have",
"served",
"as",
"the",
"basis",
"of",
"many",
"MCQA",
"models."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
12,
0
] |
53 | [
"5.3",
"Ablation",
"Study:",
"Influence",
"of",
"Clues"
] | [
0,
0,
0,
0,
0,
0
] |
54 | [
"Note",
"that",
"instead",
"of",
"training",
"on",
"each",
"dataset",
"separately,",
"UnifiedQA",
"converts",
"a",
"line",
"of",
"popular",
"QA",
"datasets",
"with",
"four",
"formats",
"(e.g.,",
"retrieval-based",
"QA,",
"MCQA)",
"into",
"a",
"unified",
"format,",
"and",
"trains",
"a",
"single",
"model",
"over",
"all",
"training",
"data,",
"while",
"GenMC",
"only",
"uses",
"each"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
55 | [
"While",
"UnifiedQA",
"reports",
"the",
"best",
"score",
"using",
"its",
"T5-11B",
"version,",
"since",
"for",
"T5",
"we",
"experiment",
"with",
"its",
"BASE",
"and",
"LARGE",
"versions,",
"we",
"only",
"report",
"and",
"compare",
"under",
"T5BASE",
"and",
"T5LARGE."
] | [
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
56 | [
"Among",
"these",
"models,",
"UnifiedQA",
"(Khashabi",
"et",
"al.,",
"2020)",
"is",
"the",
"current",
"best",
"model."
] | [
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
57 | [
"However,",
"to",
"enable",
"a",
"fair",
"comparison,",
"we",
"only",
"compare",
"with",
"models",
"that",
"adopt",
"the",
"same",
"setting",
"as",
"ours,",
"where",
"a",
"question",
"and",
"its",
"options",
"are",
"the",
"only",
"input",
"to",
"the",
"model."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
58 | [
"Existing",
"methods",
"that",
"rely",
"on",
"external",
"documents",
"or",
"corpora",
"have",
"achieved",
"state-ofthe-art",
"performance",
"on",
"several",
"MCQA",
"datasets."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
0
] |
59 | [
"5.2.1",
"Baselines",
"UnifiedQA"
] | [
0,
0,
1
] |
60 | [
"5.2",
"Comparison",
"with",
"Other",
"Models"
] | [
0,
0,
0,
0,
0
] |
61 | [
"This",
"suggests",
"that",
"the",
"embedded",
"knowledge",
"gained",
"from",
"pre-training",
"is",
"critical",
"to",
"MCQA",
"tasks,",
"strengthening",
"our",
"point",
"to",
"make",
"full",
"use",
"of",
"pre-trained",
"encoders",
"and",
"decoders."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
62 | [
"In",
"addition,",
"all",
"LARGE",
"models",
"significantly",
"outperform",
"their",
"BASE",
"counterparts."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
63 | [
"This",
"indicates",
"that",
"the",
"decoder’s",
"general",
"language",
"knowledge",
"gained",
"from",
"pre-training",
"is",
"largely",
"wasted",
"by",
"only",
"using",
"it",
"as",
"a",
"classifier,",
"which",
"may",
"further",
"explain",
"the",
"superior",
"performance",
"of",
"our",
"model",
"because",
"GenMC",
"can",
"exploit",
"the",
"pre-trained",
"decoder",
"more",
"effectively."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0
] |
64 | [
"Moreover,",
"we",
"interestingly",
"find",
"that",
"the",
"decoder-free",
"baseline",
"Text2Textenc",
"outperforms",
"Text2Textvanilla",
"on",
"over",
"half",
"of",
"the",
"experiments."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
1,
0,
0,
0,
0,
0,
0
] |
65 | [
"These",
"results",
"demonstrate",
"that",
"GenMC",
"is",
"a",
"more",
"effective",
"usage",
"of",
"pre-trained",
"encoder-decoder",
"models",
"than",
"existing",
"ones."
] | [
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
66 | [
"racy",
"of",
"23.69%",
"to",
"39.00%,",
"suggesting",
"a",
"relative",
"gain",
"of",
"around",
"65%."
] | [
0,
0,
9,
0,
9,
0,
0,
0,
0,
0,
0,
9
] |
67 | [
"For",
"example,",
"on",
"the",
"test",
"set",
"of",
"the",
"challenging",
"scientific",
"MCQA",
"dataset",
"ARC-Challenge,",
"T5BASE",
"+",
"GenMC",
"improves",
"T5BASE",
"+",
"Text2Textvanilla",
"from",
"an",
"accu"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
0,
13,
1,
2,
2,
0,
1,
2,
2,
0,
0,
0
] |
68 | [
"For",
"several",
"settings,",
"GenMC",
"even",
"obtains",
"an",
"absolute",
"gain",
"of",
"over",
"10%."
] | [
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
9
] |
69 | [
"The",
"main",
"results",
"(see",
"Table",
"2)",
"show",
"that",
"GenMC",
"consistently",
"and",
"significantly",
"(with",
"p-value",
"<",
"0.01)",
"outperforms",
"Text2Textvanilla",
"and",
"Text2Textenc",
"on",
"all",
"datasets."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
1,
0,
0,
0
] |
70 | [
"5.1.2",
"Results"
] | [
0,
0
] |
71 | [
"Though",
"Liu",
"et",
"al.",
"(2021)",
"find",
"that",
"their",
"encoder-only",
"model",
"performs",
"comparably",
"to",
"using",
"the",
"decoder",
"as",
"a",
"classifier,",
"we",
"argue",
"that",
"the",
"decoder",
"part",
"can",
"further",
"improve",
"the",
"performance,",
"if",
"being",
"properly",
"used."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
72 | [
"In",
"this",
"setting,",
"the",
"decoder",
"is",
"totally",
"unused."
] | [
0,
0,
0,
0,
0,
0,
0,
0
] |
73 | [
"The",
"model",
"then",
"predicts",
"the",
"option",
"with",
"the",
"highest",
"score."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
74 | [
"Then",
"the",
"representation",
"is",
"fed",
"into",
"a",
"scorer",
"(i.e.,",
"an",
"MLP)",
"to",
"obtain",
"a",
"matching",
"score",
"for",
"each",
"question-option",
"pair."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
75 | [
"Each",
"option",
"is",
"independently",
"paired",
"with",
"the",
"question",
"to",
"obtain",
"a",
"joint",
"representation",
"using",
"the",
"encoder."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
76 | [
"Text2Textenc",
"Similar",
"to",
"Liu",
"et",
"al.",
"(2021),",
"we",
"use",
"only",
"the",
"encoder",
"part",
"of",
"a",
"pre-trained",
"encoderdecoder",
"model."
] | [
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
77 | [
"In",
"this",
"setting,",
"the",
"decoder",
"is",
"basically",
"used",
"as",
"a",
"classifier."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
78 | [
"Based",
"on",
"the",
"joint",
"representation,",
"the",
"decoder",
"finally",
"outputs",
"an",
"option",
"ID."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
79 | [
"The",
"concatenated",
"sequence",
"is",
"fed",
"into",
"the",
"encoder",
"part",
"to",
"get",
"a",
"joint",
"representation",
"for",
"the",
"question",
"and",
"all",
"options."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
80 | [
"Specifically,",
"following",
"Raffel",
"et",
"al.",
"(2020),",
"we",
"concatenate",
"the",
"input",
"question",
"with",
"all",
"candidate",
"options,",
"where",
"each",
"option",
"is",
"also",
"preceded",
"by",
"its",
"option",
"ID,",
"and",
"then",
"prepend",
"the",
"sequence",
"with",
"a",
"dataset",
"name."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
81 | [
"The",
"vanilla",
"usage",
"of",
"pre-trained",
"encoder-decoders",
"for",
"MCQA",
"is",
"to",
"reform",
"the",
"input",
"and",
"output",
"in",
"a",
"way",
"that",
"can",
"be",
"directly",
"processed",
"by",
"a",
"encoder-decoder",
"model."
] | [
0,
0,
0,
0,
0,
0,
0,
11,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
82 | [
"5.1.1",
"Baselines",
"Text2Textvanilla"
] | [
0,
0,
1
] |
83 | [
"Table",
"2:",
"Comparison",
"with",
"text-to-text",
"models."
] | [
0,
0,
0,
0,
0,
0
] |
84 | [
"To",
"empirically",
"evaluate",
"GenMC",
"in",
"terms",
"of",
"whether",
"it",
"better",
"exploits",
"the",
"potential",
"of",
"pretrained",
"encoder-decoder",
"models",
"for",
"MCQA,",
"we",
"compare",
"GenMC",
"with",
"a",
"standard",
"text-to-text",
"implementation",
"and",
"with",
"a",
"variant",
"thereof",
"for",
"analysis."
] | [
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
11,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
85 | [
"Text-to-Text",
"Models"
] | [
0,
0
] |
86 | [
"5.1",
"Main",
"Results:",
"Comparison",
"with"
] | [
0,
0,
0,
0,
0
] |
87 | [
"5",
"Experimental",
"Results"
] | [
0,
0,
0
] |
88 | [
"For",
"each",
"model,",
"we",
"reported",
"its",
"proportion",
"of",
"correctly",
"answered",
"questions",
"in",
"each",
"dataset."
] | [
0,
0,
0,
0,
0,
0,
7,
8,
8,
8,
8,
0,
0,
0
] |
89 | [
"4.3",
"Evaluation",
"Metric"
] | [
0,
0,
0
] |
90 | [
"All"
] | [
0
] |
91 | [
"the",
"experiments",
"were",
"performed",
"on",
"a"
] | [
0,
0,
0,
0,
0,
0
] |
92 | [
"For",
"the",
"smallest",
"three",
"random",
"seeds",
"}",
"dataset",
"ARC-Challenge,",
"we",
"used",
"five",
"random",
"seeds",
"1,",
"10,",
"20,",
"30,",
"40",
"}",
"{"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
13,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
93 | [
"For",
"CSQA,",
"OBQA,",
"ARC-Easy,",
"and",
"QASC,",
"we",
"used",
"1,",
"10,",
"20",
"."
] | [
0,
13,
13,
13,
0,
13,
0,
0,
5,
5,
5,
0
] |
94 | [
"Because",
"neural",
"models",
"are",
"known",
"to",
"be",
"sensitive",
"to",
"different",
"random",
"seeds,",
"especially",
"when",
"the",
"training",
"set",
"is",
"small,",
"we",
"performed",
"multiple",
"experiments",
"for",
"all",
"models",
"with",
"different",
"random",
"seeds,",
"and",
"reported",
"the",
"mean",
"and",
"standard",
"deviation."
] | [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
95 | [
"5"
] | [
0
] |
96 | [
"−"
] | [
0
] |
97 | [
"4,",
"5e"
] | [
5,
5
] |
98 | [
"For",
"each",
"model,",
"we",
"searched",
"for",
"the",
"best",
"learning",
"rate",
"from",
"1e",
",",
"and",
"for",
"the",
"best",
"batch",
"size",
"{",
"}",
"−",
"out",
"of"
] | [
0,
0,
0,
0,
0,
0,
0,
0,
3,
4,
0,
5,
0,
0,
0,
0,
0,
3,
4,
0,
0,
0,
0,
0
] |
99 | [
"We",
"used",
"the",
"Adam",
"optimizer",
"and",
"set",
"warmup",
"fraction",
"=",
"0.1,",
"weight",
"decay",
"=",
"0.01,",
"maximum",
"source",
"length",
"=",
"64,",
"maximum",
"target",
"length",
"=",
"32,",
"epoch",
"=",
"30,",
"and",
"early",
"stop",
"training",
"when",
"there",
"was",
"no",
"better",
"result",
"on",
"the",
"dev",
"set",
"after",
"5",
"epochs."
] | [
0,
0,
0,
0,
0,
0,
0,
3,
4,
0,
5,
3,
4,
0,
5,
3,
4,
4,
0,
5,
3,
4,
4,
0,
5,
3,
0,
5,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0
] |
Dataset Card for [naacl2022]
Dataset Summary
This is a named entity recognition dataset annotated for the science entity recognition task, a project from the CMU 11-711 course.
Supported Tasks and Leaderboards
NER task.
Languages
English
Dataset Structure
Data Instances
A sample of the dataset {'id': '0', 'tokens': ['We', 'sample', '50', 'negative', 'cases', 'from', 'T5LARGE', '+', 'GenMC', 'for', 'each', 'dataset'], 'ner_tags':['O', 'O', 'O', 'O', 'O', 'O', 'B-MethodName', 'O', 'B-MethodName', 'O', 'O', 'O']}
Data Fields
id,tokens,ner_tags
id
: astring
feature give the sample index.tokens
: alist
ofstring
features give the sequence.ner_tags
: alist
of classification labels for each token in the sentence, with possible values includingO
(0),B-MethodName
(1),I-MethodName
(2),B-HyperparameterName
(3),I-HyperparameterName
(4),B-HyperparameterValue
(5),I-HyperparameterValue
(6),B-MetricName
(7),I-MetricName
(8),B-MetricValue
(9),I-MetricValue
(10),B-TaskName
(11),I-TaskName
(12),B-DatasetName
(13),I-DatasetName
(14).
Data Splits
Data split into train.txt dev.txt test.txt
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
[More Information Needed]
Who are the source language producers?
[More Information Needed]
Annotations
Annotation process
The data is annotated by using labelstudio, the papers are collected from TACL and ACL 2022 conferences.
Who are the annotators?
Xiaoyue Cui and Haotian Teng annotated the datasets.
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
[More Information Needed]
Licensing Information
[More Information Needed]
Citation Information
[More Information Needed]
Contributions
Thanks to @xcui297; @haotianteng for adding this dataset.
- Downloads last month
- 47