GLUE基准数据集介绍

2023-11-07

图1：整篇文章的思维导图

一、简介

自然语言处理（NLP）主要自然语言理解（NLU）和自然语言生成（NLG）。为了让NLU任务发挥最大的作用，来自纽约大学、华盛顿大学等机构创建了一个多任务的自然语言理解基准和分析平台，也就是GLUE（General Language Understanding Evaluation）。

GLUE包含九项NLU任务，语言均为英语。GLUE九项任务涉及到自然语言推断、文本蕴含、情感分析、语义相似等多个任务。像BERT、XLNet、RoBERTa、ERINE、T5等知名模型都会在此基准上进行测试。目前，大家要把预测结果上传到官方的网站上，官方会给出测试的结果。

GLUE的论文为：GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding[1]

GLUE的官网为：https://gluebenchmark.com/

本文的目的在于针对GLUE的九个任务分别做一个相对详细的说明，给出一些样例，有一个相对整体确切的感受，同时提供一个可以方便下载GLUE数据集的链接，供读者使用。

二、任务介绍

GLUE共有九个任务，分别是CoLA、SST-2、MRPC、STS-B、QQP、MNLI、QNLI、RTE、WNLI。如下图图2所示，可以分为三类，分别是单句任务，相似性和释义任务，

图2：GLUE九大任务的描述和统计。所有任务都是单句或者句子对分类，除了STS-B是一个回归任务。MNLI有3个类别，所有其他分类任务都是2个类别。测试集中加粗的表示测试集中标签从未在公共论坛等场所展示过

2.1 CoLA

CoLA(The Corpus of Linguistic Acceptability，语言可接受性语料库)，单句子分类任务，语料来自语言理论的书籍和期刊，每个句子被标注为是否合乎语法的单词序列。本任务是一个二分类任务，标签共两个，分别是0和1，其中0表示不合乎语法，1表示合乎语法。

样本个数：训练集8, 551个，开发集1, 043个，测试集1, 063个。

任务：可接受程度，合乎语法与不合乎语法二分类。

评价准则：Matthews correlation coefficient。

标签为1（合乎语法）的样例：

She is proud.
she is the mother.
John thinks Mary left.
Yes, she did.
Will John not go to school?
Mary noticed John's excessive appreciation of himself.

标签为0（不合语法）的样例：

Mary sent.
Yes, she used.
Mary wonders for Bill to come.
They are intense of Bill.
Mary thinks whether Bill will come.
Mary noticed John's excessive appreciation of herself.

注意到，这里面的句子看起来不是很长，有些错误是性别不符，有些是缺词、少词，有些是加s不加s的情况，各种语法错误。但我也注意到，有一些看起来错误并没有那么严重，甚至在某些情况还是可以说的通的。

2.2 SST-2

SST-2(The Stanford Sentiment Treebank，斯坦福情感树库)，单句子分类任务，包含电影评论中的句子和它们情感的人类注释。这项任务是给定句子的情感，类别分为两类正面情感（positive，样本标签对应为1）和负面情感（negative，样本标签对应为0），并且只用句子级别的标签。也就是，本任务也是一个二分类任务，针对句子级别，分为正面和负面情感。

样本个数：训练集67, 350个，开发集873个，测试集1, 821个。

任务：情感分类，正面情感和负面情感二分类。

评价准则：accuracy。

标签为1（正面情感，positive）的样例：

two central performances
against shimmering cinematography that lends the setting the ethereal beauty of an asian landscape painting
the situation in a well-balanced fashion
a better movie
at achieving the modest , crowd-pleasing goals it sets for itself
a patient viewer

标签为0（负面情感，negative）的样例：

a transparently hypocritical work that feels as though it 's trying to set the women 's liberation movement back 20 years
so pat it makes your teeth hurt
blood work is laughable in the solemnity with which it tries to pump life into overworked elements from eastwood 's dirty harry period .
faced with the possibility that her life is meaningless , vapid and devoid of substance , in a movie that is definitely meaningless , vapid and devoid of substance
monotone
this new jangle of noise , mayhem and stupidity must be a serious contender for the title .

注意到，由于句子来源于电影评论，又有它们情感的人类注释，不同于CoLA的整体偏短，有些句子很长，有些句子很短，长短并不整齐划一。

2.3 MRPC

MRPC(The Microsoft Research Paraphrase Corpus，微软研究院释义语料库)，相似性和释义任务，是从在线新闻源中自动抽取句子对语料库，并人工注释句子对中的句子是否在语义上等效。类别并不平衡，其中68%的正样本，所以遵循常规的做法，报告准确率（accuracy）和F1值。

样本个数：训练集3, 668个，开发集408个，测试集1, 725个。

任务：是否释义二分类，是释义，不是释义两类。

评价准则：准确率（accuracy）和F1值。

标签为1（正样本，互为释义）的样例（每个样例是两句话，中间用tab隔开）：

The largest gains were seen in prices , new orders , inventories and exports . Sub-indexes measuring prices , new orders , inventories and exports increased .
Trading in Loral was halted yesterday ; the shares closed on Monday at $ 3.01 . The New York Stock Exchange suspended trading yesterday in Loral , which closed at $ 3.01 Friday .
He plans to have dinner with troops at Kosovo 's U.S. military headquarters , Camp Bondsteel . After that , he plans to have dinner at Camp Bondsteel with U.S. troops stationed there .
Retailers J.C. Penney Co . Inc . ( JCP ) and Walgreen Co . ( WAG ) kick things off on Monday . Retailers J.C. Penney Co . Inc . JCP.N and Walgreen Co . WAG.N kick things off on Monday .
Prosecutors filed a motion informing Lee they intend to seek the death penalty . He added that prosecutors will seek the death penalty .
Last year the court upheld Cleveland 's school voucher program , ruling 5-4 that vouchers are constitutional if they provide parents a choice of religious and secular schools . Last year , the court ruled 5-4 in an Ohio case that government vouchers are constitutional if they provide parents with choices among a range of religious and secular schools .

标签为0（负样本，不互为释义）的样例：

Earnings per share from recurring operations will be 13 cents to 14 cents . That beat the company 's April earnings forecast of 8 to 9 cents a share .
He beat testicular cancer that had spread to his lungs and brain . Armstrong , 31 , battled testicular cancer that spread to his brain .
Graves reported from Albuquerque , Villafranca from Austin and Ratcliffe from Laredo . Pete Slover reported from Laredo and Gromer Jeffers from Albuquerque .
The commission must work out the plan 's details , but the average residential customer paying $ 840 a year would get a savings of about $ 30 annually . An average residential customer paying $ 840 a year for electricity could see a savings of $ 30 annually .
A former teammate , Carlton Dotson , has been charged with the murder . His body was found July 25 , and former teammate Carlton Dotson has been charged in his shooting death .
The battles marked day four of a U.S. sweep to hunt down supporters of Saddam Hussein 's fallen regime . Twenty-seven Iraqis were killed , pushing the number of opposition deaths to about 100 in a U.S. operation to hunt down supporters of Saddam Hussein 's fallen regime .

本任务的数据集，包含两句话，每个样本的句子长度都非常长，且数据不均衡，正样本占比68%，负样本仅占32%。

2.4 STSB

STSB(The Semantic Textual Similarity Benchmark，语义文本相似性基准测试)，相似性和释义任务，是从新闻标题、视频标题、图像标题以及自然语言推断数据中提取的句子对的集合，每对都是由人类注释的，其相似性评分为0-5(大于等于0且小于等于5的浮点数，原始paper里写的是1-5，可能是作者失误）。任务就是预测这些相似性得分，本质上是一个回归问题，但是依然可以用分类的方法，可以归类为句子对的文本五分类任务。

样本个数：训练集5, 749个，开发集1, 379个，测试集1, 377个。

任务：回归任务，预测为1-5之间的相似性得分的浮点数。但是依然可以使用分类的方法，作为五分类。

评价准则：Pearson and Spearman correlation coefficients。

一些训练集中的样例句子对及其得分：

A plane is taking off. An air plane is taking off. 5.000
A man is playing a large flute. A man is playing a flute. 3.800
A dog rides a skateboard. A dog is riding a skateboard. 5.000
A woman is playing the flute. A man is playing the guitar. 1.000
A man is playing the guitar. A man is playing the drums. 1.556
A cat is playing a piano. A man is playing a guitar. 0.600
A group of people dance on a hill. A group of people are dancing. 3.200
A woman is sitting at a desk. A woman is riding a donkey. 0.400
Someone is slicing tortila's. Someone is riding a horse. 0.000
A man is playing the guitar. A man plays an acoustic guitar. 3.750

整体句子长度适中偏短，且均衡。

2.5 QQP

QQP(The Quora Question Pairs, Quora问题对数集)，相似性和释义任务，是社区问答网站Quora中问题对的集合。任务是确定一对问题在语义上是否等效。与MRPC一样，QQP也是正负样本不均衡的，不同是的QQP负样本占63%，正样本是37%，所以我们也是报告准确率和F1值。我们使用标准测试集，为此我们从作者那里获得了专用标签。我们观察到测试集与训练集分布不同。

样本个数：训练集363, 870个，开发集40, 431个，测试集390, 965个。

任务：判定句子对是否等效，等效、不等效两种情况，二分类任务。

评价准则：准确率（accuracy）和F1值。

标签为1（正样本，互为释义，等效）的样例（每个样例是两句话，中间用tab隔开）：

How can I improve my communication and verbal skills? What should we do to improve communication skills?
What has Hillary Clinton done that makes her trustworthy? Why do Democrats consider Hillary Clinton trustworthy?
What are the top countries worth visiting? What are the top ten countries you think are most worth visiting in your lifetime, and why?
hat would happen if all the people in the world didn't need to sleep? Where would humans be if we didn't need sleep?
Why is Russia improving relations with Israel so much in 2016? Why is Russia and Israel improving relations with each other so much as of 2016?
hy does the iPad Mini say not charging? Why is my iPad Mini not charging?

标签为0（负样本，不互为释义，不等效）的样例：

Why are you so sexy? How sexy are you?
Which programming languages are common to develop in the area of gamification? Who is the worst Director in the history of MNIT/MREC?
How do I solve 3^1/3? How do I solve (x^2-1) /(x-3) <0?
Why is the Mophie case charged by a micro-USB cable, and not a lightning cable? Which USB hub should I buy?
Can we do a mobile recharge using the BHIM app? How do I select state bank of Patiala in bhim app?
What is the feeling of love? What it feels to be loved?

类似于MRPC，句子对的释义问题。这里突出的除了样本不均衡、训练集测试集分布不一致外，还有这个训练集、测试集都非常大。这里的测试集比其他训练集都要多好几倍。

2.6 MNLI

MNLI(The Multi-Genre Natural Language Inference Corpus, 多类型自然语言推理数据库)，自然语言推断任务，是通过众包方式对句子对进行文本蕴含标注的集合。给定前提（premise）语句和假设（hypothesis）语句，任务是预测前提语句是否包含假设（蕴含, entailment），与假设矛盾（矛盾，contradiction）或者两者都不（中立，neutral）。前提语句是从数十种不同来源收集的，包括转录的语音，小说和政府报告。

样本个数：训练集392, 702个，开发集dev-matched 9, 815个，开发集dev-mismatched9, 832个，测试集test-matched 9, 796个，测试集test-dismatched9, 847个。因为MNLI是集合了许多不同领域风格的文本，所以又分为了matched和mismatched两个版本的数据集，matched指的是训练集和测试集的数据来源一致，mismached指的是训练集和测试集来源不一致。

任务：句子对，一个前提，一个是假设。前提和假设的关系有三种情况：蕴含（entailment），矛盾（contradiction），中立（neutral）。句子对三分类问题。

评价准则：matched accuracy/mismatched accuracy。

标签为蕴含（entailment）的句子对示例：

you know during the season and i guess at at your level uh you lose them to the next level if if they decide to recall the the parent team the Braves decide to call to recall a guy from triple A then a double A guy goes up to replace him and a single A guy goes up to replace him You lose the things to the following level if the people recall.
ow do you know? All this is their information again. This information belongs to them.
well you see that on television also You can see that on television, as well.
According to the Office of the Actuary at the Health Care Financing Administration, the estimated net present value of future additional resources needed to fund HI benefits alone over the 75 years is $4. The net present value of future additional resources for funding HI benefits was $4.

标签为矛盾（contradiction）的句子对示例：

They're made from a secret recipe handed down to the present-day villagers by their Mallorcan ancestors, who came here in the early 17th century as part of an official repopulation scheme. The recipe passed down from Mallorcan ancestors is known to everyone.
Felicia's Journey takes place behind the eyes of its central a young Irish girl, Felicia, who crosses the sea to England in a hopeful quest to find the father of her unborn child; and the fat, middle-aged catering manager, Hiditch, who takes a paternal interest in the lass when it becomes clear that her young man has caddishly given her the slip. The woman did not care where the man was as long as it was far.
Poirot, I exclaimed, with relief, and seizing him by both hands, I dragged him into the room. Poirot was now back and I was sorry that he would take over what I now considered my own investigation.
but that takes too much planning It doesn't take much planning.

标签为中立（neutral）的句子对示例：

Conceptually cream skimming has two basic dimensions - product and geography. Product and geography are what make cream skimming work.
hebes held onto power until the 12th Dynasty, when its first king, Amenemhet Iwho reigned between 1980 1951 b.c. established a capital near Memphis. The capital near Memphis lasted only half a century before its inhabitants abandoned it for the next capital.
When the trust fund begins running cash deficits in 2016, the government as a whole must come up with the cash to finance Social Security's cash deficit by reducing any projected non-Social Security surpluses, borrowing from the public, raising other taxes, or reducing other government spending. The public would generally prefer to see the government reduce its spending in other areas to finance Social Security.
She smiled back. She was so happy she couldn't stop smiling.

总体训练集很充足，GLUE论文作者使用并推荐SNLI数据集[2]作为辅助训练数据。

2.7 QNLI

QNLI(Qusetion-answering NLI，问答自然语言推断)，自然语言推断任务。QNLI是从另一个数据集The Stanford Question Answering Dataset(斯坦福问答数据集, SQuAD 1.0)[3]转换而来的。SQuAD 1.0是有一个问题-段落对组成的问答数据集，其中段落来自维基百科，段落中的一个句子包含问题的答案。这里可以看到有个要素，来自维基百科的段落，问题，段落中的一个句子包含问题的答案。通过将问题和上下文（即维基百科段落）中的每一句话进行组合，并过滤掉词汇重叠比较低的句子对就得到了QNLI中的句子对。相比原始SQuAD任务，消除了模型选择准确答案的要求；也消除了简化的假设，即答案适中在输入中并且词汇重叠是可靠的提示。

样本个数：训练集104, 743个，开发集5, 463个，测试集5, 461个。

任务：判断问题（question）和句子（sentence，维基百科段落中的一句）是否蕴含，蕴含和不蕴含，二分类。

评价准则：准确率（accuracy）。

标签为蕴含（entailment，正样本）的样例（每个样例是两句话，中间用tab隔开，第一句是问题，第二句是上下文中的一句）：

What did Arsenal consider the yellow and blue colors to be after losing a FA Cup final wearing red and white? Arsenal then competed in three consecutive FA Cup finals between 1978 and 1980 wearing their "lucky" yellow and blue strip, which remained the club's away strip until the release of a green and navy away kit in 1982–83.
Which collection of minor poems are sometimes attributed to Virgil? A number of minor poems, collected in the Appendix Vergiliana, are sometimes attributed to him.
What does confrontational scavenging involve doing to other predators after they've made a kill? Robert Blumenschine proposed the idea of confrontational scavenging, which involves challenging and scaring off other predators after they have made a kill, which he suggests could have been the leading method of obtaining protein-rich meat by early humans.
Why were dogs initially selected? Unlike other domestic species which were primarily selected for production-related traits, dogs were initially selected for their behaviors.
What is the term that refers to areas where an antennas radiation is zero? The radiation of many antennas shows a pattern of maxima or "lobes" at various angles, separated by "nulls", angles where the radiation falls to zero.
Why do people say KInseys work is not correct? Kinsey's methods have been criticized as flawed, particularly with regard to the randomness of his sample population, which included prison inmates, male prostitutes and those who willingly participated in discussion of previously taboo sexual topics.

标签为不蕴含（not_entailment，负样本）的样例（每个样例是两句话，中间用tab隔开，第一句是问题，第二句是上下文中的一句）：

When did the third Digimon series begin? Unlike the two seasons before it and most of the seasons that followed, Digimon Tamers takes a darker and more realistic approach to its story featuring Digimon who do not reincarnate after their deaths and more complex character development in the original Japanese.
While looking for bugs, what else can testing do? Although testing can determine the correctness of software under the assumption of some specific hypotheses (see hierarchy of testing difficulty below), testing cannot identify all the defects within software.
What was the highest order of species n land? The climate was much more humid than the Triassic, and as a result, the world was very tropical.
In what century was the church established at the location? Construction of the present church began in 1245, on the orders of King Henry III.
In what century was the church established at the location? Construction of the present church began in 1245, on the orders of King Henry III.
Do the roots and shoots need each other? Roots that spread out close to the surface, such as those of willows, can produce shoots and ultimately new plants.

总体就是问答句子组成的问答对，一个是问题，一个是句子信息，后者包含前者的答案就是蕴含，不包含就是不蕴含，是一个二分类。

2.8 RTE

RTE(The Recognizing Textual Entailment datasets，识别文本蕴含数据集)，自然语言推断任务，它是将一系列的年度文本蕴含挑战赛的数据集进行整合合并而来的，包含RTE1[4]，RTE2，RTE3[5]，RTE5等，这些数据样本都从新闻和维基百科构建而来。将这些所有数据转换为二分类，对于三分类的数据，为了保持一致性，将中立（neutral）和矛盾（contradiction）转换为不蕴含（not entailment）。

样本个数：训练集2, 491个，开发集277个，测试集3, 000个。

任务：判断句子对是否蕴含，句子1和句子2是否互为蕴含，二分类任务。

评价准则：准确率（accuracy）。

标签为蕴含（entailment，正样本）的样例（每个样例是两句话，中间用tab隔开）：

A place of sorrow, after Pope John Paul II died, became a place of celebration, as Roman Catholic faithful gathered in downtown Chicago to mark the installation of new Pope Benedict XVI. Pope Benedict XVI is the new leader of the Roman Catholic Church.
Herceptin was already approved to treat the sickest breast cancer patients, and the company said, Monday, it will discuss with federal regulators the possibility of prescribing the drug for more breast cancer patients. Herceptin can be used to treat breast cancer.
The name for the newest James Bond film has been announced today. The 22nd film, previously known only as "Bond 22", will be called "Quantum of Solace". EON Productions who are producing the film made the announcement today at Pinewood Studios, where production for the film has been under way since last year. The name of the film was inspired by a short story (of the same name) from For Your Eyes Only by Bond creator, Ian Fleming. James Bond was created by Ian Fleming.
The gastric bypass operation, also known as stomach stapling, has become the most common surgical procedure for treating obesity. Obesity is medically treated.

标签为不蕴含（not_entailment，正样本）的样例（每个样例是两句话，中间用tab隔开）：

No Weapons of Mass Destruction Found in Iraq Yet. Weapons of Mass Destruction Found in Iraq.
Sierra is likely to remain in jail at the Hillsborough County jail in her native Tampa until her next hearing on December 20, where she is being held without bail, which would prevent her attending the Washington event on Friday even if she still had permission to perform. Sierra has been in jail since the start of the month after an altercation with police officers outside a Tampa nightclub, which she had been ejected from. She is charged with disorderly intoxication and resisting arrest. Sierra once reached the finals of "American Idol".
Since 1987, however, Brazil has taken steps to dramatically reduce the destruction, including stepped-up enforcement and the elimination of tax incentives that led to large-scale land clearing. In the early 1990s Brazil began to take action to save the rainforest.
The ferry owner PT Nur Budi's spokesman blamed Indonesian port authorities for the tragedy. "The passenger capacity of the ferry is 205 people but the port administrator accepted more passengers as they thought it was possible," he said. The National Meteorological and Geophysics Agency, however, had published and raised an alert signal about high waves on Friday. It specifically stated that "Saturday 10th and Sunday 11th, Indonesian waters would have witnessed storm force waves," but despite the dire warnings KM Teratai set for the seas. An Indonesian ferry with 300 passengers sank.

2.9 WNLI

WNLI(Winograd NLI，Winograd自然语言推断)，自然语言推断任务，数据集来自于竞赛数据的转换。Winograd Schema Challenge[6]，该竞赛是一项阅读理解任务，其中系统必须读一个带有代词的句子，并从列表中找到代词的指代对象。这些样本都是都是手动创建的，以挫败简单的统计方法：每个样本都取决于句子中单个单词或短语提供的上下文信息。为了将问题转换成句子对分类，方法是通过用每个可能的列表中的每个可能的指代去替换原始句子中的代词。任务是预测两个句子对是否有关（蕴含、不蕴含）。训练集两个类别是均衡的，测试集是不均衡的，65%是不蕴含。

样本个数：训练集635个，开发集71个，测试集146个。

任务：判断句子对是否相关，蕴含和不蕴含，二分类任务。

评价准则：准确率（accuracy）。

标签为1（蕴含，entailment，正样本）的样例（每个样例是两句话，中间用tab隔开）：

The actress used to be named Terpsichore, but she changed it to Tina a few years ago, because she figured it was too hard to pronounce. Terpsichore was too hard to pronounce.
Since Chester was dependent on Uncle Vernon, he couldn't very well marry without his approval He couldn't very well marry without Uncle Vernon's approval
When they had eventually calmed down a bit, and had gotten home, Mr. Farley put the magic pebble in an iron safe. Some day they might want to use it , but really for now, what more could they wish for? Some day they might want to use the magic pebble.
The actress used to be named Terpsichore, but she changed it to Tina a few years ago, because she figured it was easier to pronounce. Tina was easier to pronounce.
Frank was upset with Tom because the toaster he had sold him didn't work. The toaster Tom had sold him didn't work.
My meeting started at 4:00 and I needed to catch the train at 4:30, so there wasn't much time. Luckily, it was delayed, so it worked out. The train was delayed, so it worked out.

标签为0（不蕴含，not_entailment，正样本）的样例（每个样例是两句话，中间用tab隔开）：

Bill passed the half-empty plate to John because he was hungry. Bill was hungry.
The donkey wished a wart on its hind leg would disappear, and it did. The donkey wished a wart on its hind leg would disappear, and leg did.
The man lifted the boy onto his shoulders. The man lifted the boy onto the boy's shoulders.
The delivery truck zoomed by the school bus because it was going so slow. The delivery truck was going so slow.
We went to the lake, because a shark had been seen at the ocean beach, so it was a safer place to swim. The ocean beach was a safer place to swim.
rchaeologists have concluded that humans lived in Laputa 20,000 years ago. They hunted for evidence on the river banks. Prehistoric humans hunted for evidence on the river banks.

可以看到，这个数据集是数量最少，训练集600多个，测试集才100多个。同时目前GLUE上这个数据集还有些问题。

三、榜单排名

图3 ： GLUE榜单

上图图3是目前（2020年4月）的榜单，我们熟知的各大模型，bert、xlnet、xlm、ELECTRA、T5等都榜上有名，不一一分析了。

四、数据集下载

使用以下几个官方的下载方法（需要科学上网）：

官方的下载链接：https://gluebenchmark.com/tasks
官方下载脚本：https://github.com/nyu-mll/jiant/blob/master/scripts/download_glue_data.py
下载数据的脚本：https://gist.github.com/W4ngatang/60c2bdb54d156a41194446737ce03e2e

祝大家使用愉快！

参考

本文内容由网友自发贡献，版权归原作者所有，本站不承担相应法律责任。如您发现有涉嫌抄袭侵权的内容，请联系:hwhale#tublm.com(使用前将#替换为@)

机器学习

深度学习