paraphrasing a source with attribution

• Military & Veterans
• Transfer Students
• Education Partnerships
• COVID-19 Info
• 844-PURDUE-G
• Student Login
• Request Info
• Bachelor of Science
• Master of Science
• Associate of Applied Science
• Graduate Certificate
• Master of Business Administration
• ExcelTrack Master of Business Administration
• ExcelTrack Bachelor of Science
• Postbaccalaureate Certificate
• Certificate
• Associate of Applied Science (For Military Students)
• Programs and Courses
• Master of Public Administration
• Doctor of Education
• Postgraduate Certificate
• Bachelor of Science in Psychology
• Master of Health Care Administration
• Master of Health Informatics
• Doctor of Health Science
• Associate of Applied of Science (For Military Students)
• Associate of Science (For Military Students)
• Master of Public Health
• Executive Juris Doctor
• Juris Doctor
• Dual Master's Degrees
• ExcelTrack Master of Science
• Master of Science (DNP Path)
• Bachelor of Science (RN-to-BSN)
• ExcelTrack Bachelor of Science (RN-to-BSN)
• Associate of Science
• Doctor of Nursing Practice
• Master of Professional Studies

The average Purdue Global military student is awarded 54% of the credits needed for an associate's and 45% of the credits needed for a bachelor's.

• General Education Mobile (GEM) Program
• AAS in Health Science
• AS in Health Science
• BS in Organizational Management
• BS in Professional Studies
• AAS in Criminal Justice
• AAS in Small Group Management
• AAS Small Group Management
• Master's Degrees
• Bachelor's Degrees
• Associate's Degrees
• Certificate Programs
• Noncredit Courses
• Tuition and Financial Aid Overview
• Financial Aid Process
• Financial Aid Awards
• Financial Aid Resources
• Financial Aid Frequently Asked Questions
• Financial Aid Information Guide
• Tuition and Savings
• Aviation Degree Tuition and Fees
• Professional Studies Tuition and Fees
• Single Courses and Micro-Credentials
• Time and Tuition Calculator
• Net Price Calculator
• Military Benefits and Tuition Assistance
• Military Educational Resources
• Military Tuition Reductions
• Military Spouses
• Student Loans
• Student Grants
• Outside Scholarships
• Loan Management
• Financial Literacy Tools
• Academic Calendar
• General Requirements
• Technology Requirements
• Work and Life Experience Credit
• DREAMers Education Initiative
• Student Identity
• Student Experience
• Online Experience
• Student Life
• Alumni Engagement
• International Students
• Academic Support
• All Purdue Online Degrees
• Career Services
• COVID-19 FAQs
• Student Accessibility Services
• Student Resources
• Transcript Request
• About Purdue Global
• Accreditation
• Approach to Learning
• Career Opportunities
• Diversity Initiatives
• Purdue Global Commitment
• Cybersecurity Center
• Chancellor's Corner
• Purdue Global Moves
• Leadership and Board
• Facts and Statistics
• Researcher Request Intake Form

Most Commonly Searched:

• All Degree Programs
• Communication
• Criminal Justice
• Fire Science
• Health Sciences
• Human Services
• Information Technology
• Legal Studies
• Professional Studies
• Psychology and ABA
• Public Policy
• Military and Veterans
• Tuition and Fee Finder
• Financial Aid FAQs
• Military Benefits and Aid
• Admissions Overview
• Student Experience Overview
• Academic Support Overview
• Online Learning

A Guide to Plagiarism and Paraphrasing

According to the joint research efforts of Dr. Donald McCabe and the International Center for Academic Integrity , nearly 30% of university students admit to having cheated in some way on an exam.

Understanding how and when to cite sources is a critical skill for students to learn. Whether you borrow someone’s ideas from a textbook, blog post, or academic journal, you must give proper credit while representing the source’s ideas fairly and coherently.

This guide covers:

• Paraphrasing
• Plagiarism checkers, citation managers, and writing tools

The Purdue Global Writing Center defines plagiarism as “using another's words, ideas, results, or images without giving appropriate credit to that person, therefore, giving the impression that it is your own work.”

Types of Plagiarism

University of Oxford notes eight common forms of plagiarism:

• Verbatim plagiarism: Copying someone else’s work word for word.
• Cutting and pasting from web pages without clear acknowledgement: Pulling information off the internet without referencing it and without including it in the bibliography.
• Paraphrasing: Paraphrasing so closely so that the copy is almost an exact match to the original.
• Collusion: In group projects, or projects in which you received help, failing to properly attribute the assistance or failure to follow the project’s rules.
• Inaccurate citation: Failing to cite correctly, according to the conventions of your discipline.
• Failure to acknowledge assistance: Failing to clearly acknowledge all assistance that has contributed to your work (ordinary proofreading and help from a tutor or supervisor is excepted).
• Use of material written by professional agencies or other people: Using material that was written by a professional agency or another person, even if you have the consent of the person who wrote it.
• Auto-plagiarism (also known as self-plagiarism): Reusing work that you’ve previously submitted or published; presenting that information as new when you’ve already gotten credit for the work.

A new concern revolves around AI and copying directly from chat, composition, and visual tools. Using prompts to generate content for assignments and passing it off as your own contribution is considered plagiarism. Various organizations use AI software to check for submissions generated by a chatbot.

Also, keep in mind that AI tools may produce inaccurate and unreliable information. While there may be valid use cases for informal AI-generated brainstorming, this is a complex and evolving topic. Be sure to verify the policy expressed by your school, professors, or professional organizations for recent developments.

It’s important to note that plagiarism can be intentional or unintentional. Unintentional plagiarism occurs when a student unknowingly cites a source inaccurately or improperly. Intentional plagiarism, on the other hand, is when a student chooses not to cite a source or tries to pass off someone else’s ideas as their own.

Consequences of Plagiarism

The consequences of plagiarism vary by institution, but it could get you expelled or dropped from a course. In less severe instances, plagiarism — both intentional and unintentional — may result in a grade penalty, course failure, or suspension. Beyond the academic consequences, plagiarism also tarnishes your reputation and minimizes your integrity. Whether you’re in school or the working world, plagiarism is not a good look.

How to Avoid Plagiarism

The key to avoiding plagiarism is learning how to incorporate research into your writing. According to the Purdue Global Writing Center , you can do this in the following ways:

• Quoting: If you don’t want to alter a source, use quotation marks to enclose all verbatim phrases.
• Summarizing: If you find multiple relevant points in a lengthy text, simplify them into your own condensed synopsis.
• Paraphrasing: If you want to use a source’s information, restate it in your own words.

Whether you’re quoting, summarizing, or paraphrasing, don’t forget to cite all sources.

What Is Paraphrasing?

Paraphrasing is using your own words to convey the meaning of an excerpt. It shows your reader that you did your research and understand the content. While students may understand that they need to cite sources, many struggle with paraphrasing the ideas of others into their own words. However, like many aspects of writing, effective paraphrasing is a skill developed over time.

How to Approach Paraphrasing

The goal of paraphrasing is to translate the original work into your own wording and sentence structure. The best way to approach this is to focus on the meaning of the text, forcing you to interact with its purpose and context.

Paraphrasing Tips

A good way to judge your understanding of material is to see if you can explain it to someone else. Once you have this level of understanding, it’s easier to create effective paraphrases — changing the language and structure of a passage becomes more manageable.

Here are some tips to help you paraphrase:

• Reread the passage until you fully understand its meaning.
• Write your own summary of the passage without referencing the original.
• Check that your summary accurately captures the context of the original passage.
• Document the source information following your summary, whether it’s an endnote or footnote.

Remember that you still need to cite your paraphrases, but your follow-up analysis and discussion points belong to you.

What Requires Citation?

Any time you use information that isn’t common knowledge or you didn’t come up with yourself, you must cite it. The following requires citation, usually through in-text citation or a reference list entry:

• Quotes: If you are quoting the actual words someone said, put the words in quotation marks and cite the source.
• Information and ideas: If you obtain ideas or information from somewhere else, cite it — even if you paraphrase the original content.
• Illustrations: If you use someone else’s graphic, table, figure, or artwork, you must credit the source. These may also require permission and a copyright notice.
• Photographs: If you use your own photography or an image that allows use without attribution, no citation is required. In other cases, add a note below the image and a corresponding reference citation.

Common Knowledge Exception

You don’t need to cite information that’s considered common knowledge in the public domain — as long as you reword the well-known fact. According to the Purdue Global Writing Center , information must have the following traits to be considered common knowledge:

• The reader would already be aware of it.
• It’s a widely accepted fact; for example, there are 24 hours in a day.
• It’s accessible via common information sources.
• It originates from folklore or a well-known story.
• It’s commonly acknowledged in your field and known by your audience.

Why Citation Is Important

The importance of citation goes beyond the avoidance of plagiarism. According to the Purdue Global Writing Center’s Plagiarism Information page, citation:

• Distinguishes new ideas from existing information
• Reinforces arguments regarding a particular topic
• Allows readers to find your sources and conduct additional research
• Maintains ethical research and writing
• Ensures attribution of ideas, avoiding plagiarism

Additionally, proper citation enhances your credibility with readers, displays your critical thinking skills, and demonstrates your strong writing ability.

Plagiarism Prevention and Writing Resources

It takes time to develop strong writing and paraphrasing skills. Thinking of writing as more of a discussion than a report may help you develop your skills. Remember that it’s not about reporting and repeating information; it’s about expanding on ideas and making them your own.

Below are some tools to help you avoid plagiarism, accurately cite sources, and improve your writing as you develop your own unique voice.

Plagiarism Checkers

• DupliChecker
• Grammarly's Plagiarism Checker
• Plagiarism Detector

Citation Managers

• Academic Writer
• Grammarly’s Free Citation Generator

>> Read: Apps and Extensions to Help You With APA Citations

Writing Tools

Check out purdue global’s writing center resources.

The Purdue Global Writing Center can help guide students through the paper writing process — from avoiding plagiarism to proper paraphrasing to getting the right citations.

Students may access this resource from the Purdue Global campus homepage . Click “My Studies,” followed by “Academic Success Center.”

From there, students have several options:

• Ask a writing tutor
• Connect with a tutor for a one-on-one session
• Browse the Study Studio
• Watch webinars

Students can check out the Using Sources & APA Style page , which includes several resources to guide students through the process of formatting a document and citing sources in the American Psychological Association (APA) style. The Plagiarism Information page offers a tutorial designed to help students identify instances of plagiarism and understand how to avoid them.

See Notes and Conditions below for important information.

About the Author

Purdue Global

Earn a degree you're proud of and employers respect at Purdue Global, Purdue's online university for working adults. Accredited and online, Purdue Global gives you the flexibility and support you need to come back and move your career forward. Choose from 175+ programs, all backed by the power of Purdue.

• General Education
• Legal Studies & Public Policy

Your Path to Success Begins Here

Learn more about online programs at Purdue Global and download our program guide.

Connect with an Advisor to explore program requirements, curriculum, credit for prior learning process, and financial aid options.

Third-Party Products and Services: Links from the Purdue Global website to third-party sites do not constitute an endorsement by Purdue Global of the parties or their products and services. Purdue Global cannot guarantee that certain products will continue to be offered by their publishers for free. Users of third-party websites are responsible for reviewing the terms of use and being familiar with the privacy policy of such third-party websites.

5 Attribute All Sources

Kerry Benson

Learning Objectives

After reading this chapter, you will be able to:

• Distinguish between primary and secondary, and human and nonhuman sources.
• Explain why and when attribution is necessary.
• Use proper mechanics of attribution.
• Embed links to online sources in digital text.

Ladybug Rock, by Mark Caton

My father, a Presbyterian minister, rarely used the King James Version of the Bible, so I remember vividly when he referenced the Apostle Paul’s letter to the Romans from the KJV.

Romans 13:7 reads: “Render therefore to all their dues: tribute to whom tribute is due; custom to whom custom; fear to whom fear; honour to whom honour.”

Because I was a child, the language of the passage confused me, and I asked my dad what it meant. He laughed before briefly summing up Paul’s message.

“It pretty much means you give people recognition for what they’ve done,” my dad said. “Like when your mom and I really liked the ladybug rock you brought home from school and you told us Mark Caton was the one who painted it. You gave Mark what was due to him, the credit for being the painter, instead of telling us you’d done it.”

Not everyone needs a biblical lesson on giving credit where it’s due. And credit isn’t necessarily an acknowledgment of excellence, as Mark Caton could have done a poor job of painting a ladybug on stone, but in journalism and strategic communications attribution is like that ladybug: a rock. It’s one of the ethical (and often lawful) foundations of a news or feature story, a documentary, a company news release, a digital ad, or a marketing PowerPoint.

This chapter will help illuminate the concept of attribution, why it matters, who uses it, who benefits from its use, when it’s used, and why professionals may disagree on its use. This chapter also will address how to journalistically cite sources, how attribution can go wrong, and where to find more on the topic.

What is Attribution?

Reputable and engrossing writing, whether it’s journalism or strategic communication, starts with responsible and principled research and reporting. Attribution is vital to all ethical reporting because it identifies information sources.

The Accrediting Council on Education in Journalism and Mass Communications (ACEJMC), which accredits journalism schools, lists core values and competencies all graduates should be able to meet. Among those competencies is the ability to “demonstrate an understanding of professional ethical principles and work ethically in pursuit of truth, accuracy, fairness and diversity.” Attribution is key to the quest for veracity and transparency. Attribution’s job in journalism is to answer the “who” of a quotation, the “where” and “what” of background information, and –- sometimes –- the “how.”

Who said what? Where did reporters or editors get their data? What research was used to support an opinion? How did a human source provide those statistics?

Understanding attribution requires understanding sources.

Primary Sources

If a human contributes information for a story, whether it’s in-person, on the phone, or via email or text, that person is a source. The most credible human source is a primary one, a person with a direct connection to the information or situation pertinent to the story.

This first-hand relationship provides for an accurate telling of that person’s experience. Even though the source’s personal viewpoint can be an opinion, it can also provide a reporter with facts. It’s the reporter’s responsibility to confirm the facts. An exception to this is if the journalist is the witness to events. Journalists can’t name themselves as sources in articles.

Primary human sources also add what it sounds like –- humanness. They put a face to the facts and a person to the perspective. Often they can synthesize information in a way that makes it accessible and easy to understand for other people.

Any person who contributes any kind of information to a story is a human source, even if that material is never published or broadcast.

Primary human source examples:

Chadwick Boseman, star of the Disney and Marvel Studios film “Black Panther,” says in a 2018 interview with USA Today how respectful he is of the cinematic history the movie was about to make.

Infections, like the kind caused by what the Centers for Disease Control and Prevention call “nightmare bacteria,” are drug-resistant and “virtually untreatable with modern medicine,” CDC Principal Deputy Director Dr. Anne Schuchat said in a press briefing .

The reporter in each scenario above indicates to readers or viewers where the information originates. The importance of primary source credibility is clear. The main actor in a film will know about acting in that film. Schuchat, who served as acting director of the CDC twice, will know the agency’s public health concerns and alerts.

A journalist could probably get the same information from a nonhuman source, but Boseman and Schuchat put a trustworthy human face to the communication they’re sharing.

Primary sources also can be nonhuman. Government records, reports of original research studies, and polls are examples of primary sources because they are the original locations of the information they contain. A nonhuman source is primary if it provides original information that does not cite other sources.

Some sources, like research studies, often are both primary and secondary sources, because they both re-state information found elsewhere, and are the original sources of other information.

Primary nonhuman source examples:

April is designated as Alcohol Awareness Month by the federal government. A journalist developing a story about drug and alcohol trends among seniors, or in a specific geographical region, might use data published by the National Council on Alcoholism and Drug Dependence (NCADD), a primary nonhuman source.

Marketers at a major health care organization choose similarly to highlight the importance of alcohol awareness, and they also provide NCADD data in a story in their monthly e-zine or quarterly newsletter. NCADD serves as a primary nonhuman source. The marketers supplement their story with human primary sources from within their organization, such as physicians and counselors.

Journalists and strategic communicators should not leave their audience to question information or sources’ legitimacy. The exception is if something is a well-known –- or widely reported –- fact that’s reasonably indisputable.For example, it would not be necessary to cite a source for “Abraham Lincoln, the 16th president of the United States, delivered what would come to be known as ‘The Gettysburg Address’ in November 1863.”

Secondary Sources

Both human and nonhuman sources also can be secondary sources, or research material. A secondary source is information containing others’ reporting and data gathering, and it’s usually information used for other purposes as well as a journalist or strategic communicator’s purposes.

Journalists must determine if the secondary source information is fact or opinion, or both, which they usually do by cross-referencing the information with other verifiable sources.

If a reporter looks to a website for background information, or reads other media reports on a story, it’s the reporter’s responsibility to go to the information’s original, or primary, source.

Avoid quoting The New York Times or Fox News as a source from their stories on obesity in the United States. Go to the primary source those media reference. If they cite a study, or data released by the Centers for Disease Control and Prevention (CDC), go to that study or to the CDC website. After verifying the information, cite the study or the CDC.

For example, a reporter working on an article about border crossings along the United States’ southern border sees a CNN report on a similar story, using data about who is crossing and where. The reporter should look for the source of the data, not CNN’s information. If the numbers are from U.S. Customs and Border Protection (CBP), the reporter should go to CBP for its facts and figures, and cite it as the source.

Often journalists use secondary sources as a springboard to develop a story idea, including a single exposé, an in-depth series of articles or podcasts, or a documentary. From these secondary sources, they look for the primary sources of information, and use those in their reports.

When To Attribute?

The late journalist Steve Buttry, whose résumé included editor, reporter, newsroom trainer, and teacher of digital journalism, wrote the following in a blog post, “You can quote me on that: Advice on attribution for journalists” :

“Attribute any time that attribution strengthens the credibility of a story. Attribute any time you are using someone else’s words. Attribute when you are reporting information gathered by other journalists. Attribute when you are not certain of facts. Attribute statements of opinion. When you wonder whether you should attribute, you probably should attribute in some fashion.”

Buttry’s advice from the same post on when not to use attribution is shorter:

“Don’t attribute facts that the reporter observed first-hand: It was a sunny day. Don’t worry about attributing facts where the source is obvious and not particularly important and the fact is not in dispute.”

Journalists and strategic communicators who write or report factual information or opinions should attribute all those facts and opinions to a source. In some circumstances, attribution is particularly important. Attribute facts if controversy might surround them, such as when gun permit requests go up or down, or the number of middle-aged men addicted to opioids changes dramatically. Also, always attribute evaluative facts that depend on the rule of law, or facts that rely on an expert’s information.

In broadcast, reporters and podcasters should identify the source of any statement, particularly one of questionable accuracy. The source interviewed in a radio, podcast or videotaped segment must be identified at the start. The newscaster, reporter, or podcaster can identify with a sound bite before the source speaks.

With video, a source can be acknowledged verbally and with a lower third super, a graphic, usually the interviewee’s name and location, superimposed along the bottom of the screen.

Why Attribute?

Both journalists and strategic communicators use attribution to signal to their audiences that they’re reliable and sincere. It indicates that they’ve vetted the sources, which helps readers, listeners and viewers understand the information effortlessly, without having to stop and question the content’s accuracy and authenticity.

Journalists and strategic communicators benefit from using attribution, because the trust that their audience places in the sources they cite extends to the journalists and strat comm practitioners themselves.

Good attribution says to the audience, “You can trust me because the sources I use are trustworthy.”

Individual media companies underscore the importance of attribution in their values statements. According to The Associated Press , the goal of attribution is “to provide a reader with enough information to have full confidence in the story’s veracity.”

Attribution also lets the journalist or strat comm practitioner share or shift the responsibility for any information in a story. If a reader disagrees with something he or she sees in an article or report, attribution can take the heat off the journalist or strategic communicator who wrote the piece, and direct it toward the source of the information.

When a reader or viewer questions the veracity of some information, attribution says, “Blame the message source, not the messenger.”

Attribution also allows audience members to examine a topic further. By pointing to their sources, journalists and strat comm practitioners invite their readers and viewers to find those sources for themselves, and to take deeper dives into the topics they cover.

Attribution is like the entryway to Platform 9 3/4 in the Harry Potter books, from which readers can set off on their own journeys into the subjects that interest them.

Finally, attribution can be the antidote to journalism’s biggest transgressions of fabrication and plagiarism. A journalist or strat comm practitioner who points to his or her sources is less likely to have made up something, or taken credit for someone else’s words, than one whose sources are hidden.

There is sometimes a misguided perception that attribution is less important in strategic communication than it is in news and broadcast journalism.

The Public Relations Society of America, for one, opposes this view. It argues in its Ethical Standards Advisories — Best Practices , that despite the pressures of time and shortage of resources that all content creators face, public relations practitioners have a duty to disclose their sources:

“Public relations professionals may be … challenged when facing a deadline, an assignment in a new area or even the lack of a good idea and the easy solution may be to use someone else’s words or ideas. However, an ethical practitioner respects and protects information that comes into his or her possession and makes an effort to preserve the integrity of that information.

“An ethical practitioner also uses the works of others appropriately, with proper author/creator attribution. There are many ways to do this … including footnotes, parenthetical references to the original author or a reference to the original work within the text. When words are used verbatim, it is important that they be enclosed in quotation marks and the exact source of the quote be provided either within the text or in a reference section.”

These guidelines reflect the professional standards expected of all communications professionals.

How To Attribute?

How to select quotes is part of learning to build an article, newscast, or magazine story, but how to assign responsibility to quotes is part of understanding attribution.

Direct quotes

The following comes from guidelines used in the School of Journalism and Mass Communications at the University of Kansas.

A direct quote must be exactly what a source says. Direct quotes should add zest to the story. Don’t use quotes to deliver boring-but-necessary facts or use quotes that don’t drive the story forward.

Direct quotes are used also for precision. An accurate direct quote can add confirmation of controversial facts.

It can convey a person’s information and attitude, which adds character and flavor to a story.

Examples of direct quotes:

“It’s just a job. Grass grows, birds fly, waves pound the sand. I beat people up,” boxer Muhammad Ali said.

“I’ve missed more than 9,000 shots in my career. I’ve lost almost 300 games. Twenty-six times, I’ve been trusted to take the game winning shot and missed,” basketball legend Michael Jordan said. “I’ve failed over and over and over again in my life. And that is why I succeed.”

“That’s all I could ever hope for, to have a positive effect on women. ‘Cos women are powerful, powerful beings,” singer Rihanna said. “But they’re also the most doubtful beings. They’ll never know – we’ll never know – how powerful we are.”

Say “no” to quotes that add nothing, such as “we’re so excited,” and “we went out there and did our best.” Obvious. Goofy. This may be difficult for strategic communicators whose bosses or supervisors may press for hyperbole. Resist. It damages credibility.

But journalists and strategic communicators often include direct quotes from public officials or company executives, even if what’s said doesn’t push the story forward or add flavor, because readers and viewers see those figures as authorities who should know what’s going on.

Paraphrasing

A paraphrase, or indirect quote, is a re-wording of what a source says. It must reflect the source accurately, even though it’s not relayed word for word. An indirect quote must not alter the meaning of what someone said.

In incorporating quotes into their writing, journalists often mix direct and indirect quotes.

This is the direct quote:

“When I first started teaching J101, I like, was happy to have – wow, like, 450 students, but then I had doubts,” Benson said. “But I wanted to teach many students at once. I thought I could teach that many. But, wow, managing a huge class is like turning a cruise ship in a hurricane.”

This is the paraphrase, or indirect quote:

Benson said she is happy to teach J101, a course with 450 students, but initially had doubts.

The writer could then use a partial quote to support the paraphrase:

She compared managing a class that size to “turning a cruise ship in a hurricane.”

Handling human quotes

When referring to information given by specific human sources, the verb in print is “said,” even if a writer isn’t directly quoting a source. “Said” is best because it can’t be wrong. If a source said something, the source spoke and said it. “Said” doesn’t stop thought when a reader sees it.

Verbs such as “explained” or “disclosed” or “exclaimed” require a reader to process differently. Such verbs draw attention to themselves and away from the content that matters. Readers have to think about each verb because those have connotations that “said” does not.

Weird verbs of attribution, such as argued, claimed, concluded, warned, urged and remarked, are just that, weird. Writers don’t want to imply meaning that might alter the larger article’s credibility. “Said” as a verb is neutral. It doesn’t hint at any meaning beyond its action.

Handling nonhuman quotes

“According to” is used to attribute information to nonhuman sources. Journalists and strategic communicators should use “according to” for documents, news releases, studies, statistical abstracts, infographics, or secondary sources in general.

In journalism and strategic communication, writers do not use in-text citations. That is, in journalism, there is no MLA, APA, or Chicago citation style. Save that for English, history, and political science research papers. In journalism, for in-text source identification, if it’s not “said,” it is usually “according to.”

As with “said,” there is no need to come up with different terms.

Examples of nonhuman attributions:

Teachers in the district make at least three times as much per year as teachers in other area school districts, according to state employment records.

According to a World Health Organization report, this season’s flu strain may infect millions worldwide.

Student athletes are graduating at rates twice as high as they were a decade ago, according to NCAA findings.

But that’s so repetitive

Journalists and strategic communicators, particularly if they hear their English composition teachers in their heads, may resist “said” or “according to” for every attribution in a story. They fear the repetitive use will make their writing dull and unvaried. But readers appreciate the ease of reading, so they’re not usually troubled by “said” or “according to.”

Attribution terms may vary by news organization or publication. Some journalists have the option to use alternatives, such as “stated” for human sources. Magazine writers often have the editorial leeway to use “says” – using present tense even if they’re attributing content a source provided in an interview a day, a week, or a month prior to publication. But writers can’t be wrong with “said” when attributing a human source.

Order matters

Attribution can work at the beginning of a sentence, but often is even better at the end of the sentence. This places the emphasis on the information first, then on the source.

Starting with the quote or paraphrase, and then providing attribution, is more interesting for readers than the other way around. Presumably, what a person is saying is more interesting than who’s saying it. If it’s a well-chosen quote, the information is what’s important or relevant, and the attribution is just for context and credibility.

Grammar notes

In English, writers usually put the verb after the subject in declarative sentences. Not always, but it keeps the emphasis on the subject. Remember, “Jesus wept.”

The order of name and verb is, when possible, name, then verb.

Correct example:

“The scientist examining the evidence couldn’t conclude the origin of the DNA,” Fontaine said.

Incorrect example:

“The scientist examining the evidence couldn’t conclude the origin of the DNA,” said Fontaine.

Exception example: When there’s a title or description that makes it awkward:

“The scientist examining the evidence couldn’t conclude the origin of the DNA,” said Elliot Fontaine, Colorado Springs police spokesman in charge of the investigation.

Incorrect exception example:

“The scientist examining the evidence couldn’t conclude the origin of the DNA,” Elliot Fontaine, Colorado Springs police spokesman in charge of the investigation, said.

Broadcast specifics

Broadcast attribution differs from print in several ways. Direct quotations are rare. Radio, television and podcast writers prefer indirect quotes or statement summaries.

Direct quotes, if used, should be preceded by a phrase such as “in his words” or “what she called.” Quotation marks should also be shown. They give broadcasters a clue, or signpost, to change their vocal pattern.

Broadcast example:

President Donald Trump says he will roll back all policies and laws from what he called “Obama’s clown car of a presidency.”

If it’s critical that a source be quoted directly, a broadcaster or writer may use sound bites, or actualities, in the audio. Attribution is always given before sources speak. It must be clear from the start that the quote is not the broadcaster’s thoughts or opinions.

With radio or podcasts, because listeners use only their ears to absorb the information, they need to know right away who’s responsible for what’s being said. It’s too cumbersome to inject “quote and unquote” into broadcast to indicate to listeners what is and isn’t a direct quote.

An identifier, such as a title, always goes before the name in broadcast.

Broadcast identifier example:

New York City Mayor Bill DeBlasio says he will support the new NYPD policy change on overtime pay.

With TV or video, the visual shows who’s making the statement, and a character generator super (or lower third) identifies the person after three or four seconds of video. The anchor or reporter may or may not identify the source by name in the introduction, but usually provides an identifier for context.

Broadcast lead-in example:

If the KU chancellor says,

“The number of both undergraduate and graduate students suffering from food insecurity rose 48 percent between 2015 and today.”

The broadcaster might lead into the sound bite with a synopsis: Chancellor Doug Girod says food insecurity is on the increase among students at KU.

In broadcast, source attribution and identification should be written conversationally. Think of it as the difference between a formal, engraved invitation delivered by the postal service and an e-vite via email or social media.

Be careful with pronouns. Again, because listeners and viewers can’t refer back easily in a video or audio story, they may not remember, “She? Who is she?” It’s better to repeat a name, office or title to prevent confusion.

Use “says” and not said, as if it’s happening now. “Says” is present tense and describes an ongoing action. But when broadcasters are speaking to something a source said in the past, “said” makes more sense.

Broadcast says/said example:

Chancellor Doug Girod says food insecurity is on the rise among KU students. Before he became chancellor, Girod said he would address food scarcity across campus.

Broadcast style also may allow for “according to” when using human sources. It may be a matter of news organization policy.

Embedding Links: Digital Attribution

The Internet allows journalists and strat comm practitioners to elevate their attribution game by embedding links in their work to the sources that are available online.

If you are producing content for digital distribution, link, link, link. Linking goes hand in hand with attributing online content, whether it’s news or strategic communication. Readers can –- even if they don’t –- click on links that provide background and full context to the cited information.

Linking is about transparency and trust with readers. Linking to sources in articles and reports increases the transparency of the journalists’ and strategic communicators’ work. It brings readers closer to the sources, encouraging them to verify the veracity of the information they are reading.

If attribution is like giving your friend an address to a restaurant, embedding a source’s link is like holding the restaurant’s door open for your friend when they arrive.

Any source that can be linked in an online article, should be linked. Not doing so can raise questions in an audience’s mind about why the source isn’t linked.

What Embedded Sources Look Like

Here are two examples of what embedding looks like in professional publications.

The following screenshot shows a paragraph in a Lawrence Journal-World, July 13, 2018, article titled “ New Kansas AD Jeff Long addresses still-defunct KU-MU Border War .”

The link in the paragraph takes readers to an Oct. 22, 2017, article titled “ Bill Self on playing Mizzou: ‘I don’t think there’s been any change in our position .’”

The next screenshot shows two paragraphs from a July 14, 2018, article, “ IceCube: Unlocking the Secrets of Cosmic Rays ,” published on the website Space.com .

The link in the second paragraph leads to a FAQ page on the website of the University of Wisconsin’s South Pole Neutrino Observatory .

The URLs for the articles presented above are:

http://www2.ljworld.com/weblogs/tale-tait/2018/jul/13/new-kansas-ad-jeff-long-addresses-still/ , http://www2.kusports.com/news/2017/oct/22/bill-self-playing-mizzou-i-dont-think-theres-been-/

https://www.space.com/41170-icecube-neutrino-observatory.html , https://www.space.com/ , https://icecube.wisc.edu/about/faq

  https://icecube.wisc.edu/ .

But the professional examples do not show their readers these strips of URL code.

This is because a name or description that identifies exactly where readers are going when they click on the link is more welcoming than an incomprehensible string of code. A linked snippet of text gives readers the ability to choose their web source with confidence, and it looks much more professional than raw URL.

How to Embed Links to Online Sources

You probably already are familiar with inserting source links into documents, emails, social media posts, or presentations by copying and pasting the URLs of the sources. It takes eight steps to embed a link (also called a hyperlink) in text.

• Browse to the source’s webpage.
• At the top of the browser, locate the URL field (URL stands for “uniform resource locator”).
• Highlight the entire URL and copy it (Command+C, or Control+C, or Edit > Copy).
• In the document you are writing, write a statement that will serve as the link. It could be a descriptor, such as KU J-School Technology , or it could be more directive and fun, such as Start here to learn how best to use your technology.
• Highlight the text you just typed.
• Use the “insert hyperlink” tool in the platform you are using. Here are some visual examples of where to find these tools. The link tool often is represented graphically with two links of a little chain.

Google Doc:

Blackboard:

PowerPoint:

• In the dialog box that appears, paste the source URL into the appropriate field. Oftentimes, you will see the text you highlighted in this box as well.
• Test the link using a different browser or computer than you used originally. This is especially important for links that originate behind paywalls.

What If a Source Wants To Remain Anonymous?

Avoid using unidentified sources for news or strategic communication documents. But this might depend on newsroom or organization policy. It’s usually not acceptable, as trust and transparency are the agreement readers, viewers, and listeners have with media content providers.

Exceptions are sometimes made when the only way to get a story is to offer a source anonymity. It shouldn’t be given lightly and without understanding that the information must still be reliable and accurate.

Reasons to offer anonymity could include a situation where by providing a name, the source would suffer public humiliation, lose a job or position, or go to jail.

If an anonymous source must be used, offer as much detail as possible about the source and explain the reason for anonymity.

For example, name a source as “a university official with ties to the administration who requested anonymity because his superiors had ordered him not to speak publicly or he would lose his position.”

When a source requests anonymity, get the source’s name and contact information, just in case an editor needs it.

The following are examples of ethical codes and policies journalists follow when deciding to use anonymous sources or pseudonyms.

Under Associated Press rules , material from anonymous sources may be used only if:

• The material is information and not opinion or speculation, and is vital to the news report.
• The information is not available except under the conditions of anonymity imposed by the source.
• The source is reliable, and in a position to have accurate information.

The Society of Professional Journalists published  a position paper on anonymous sources:

• Identify sources whenever feasible. The public is entitled to as much information as possible on sources’ reliability.
• The most important professional possession of journalists is credibility. If the news consumers don’t have faith that the stories they are reading or watching are accurate and fair, if they suspect information attributed to an anonymous source has been made up, then the journalists are as useful as a parka at the equator.
• To protect their credibility and the credibility of their stories, reporters should use every possible avenue to confirm and attribute information before relying on unnamed sources. If the only way to publish a story that is of importance to the audience is to use anonymous sources, the reporter owes it to the readers to identify the source as clearly as possible without pointing a figure at the person who has been granted anonymity. If the investigating police officer confirms John Doe has been arrested, the officer is a “source in the police department” and not even a pronoun should point to the gender.

The Washington Post Standards and Ethics: Policy on Sources and Confidential Sources

• The Washington Post is committed to disclosing to its readers the sources of the information in its stories to the maximum possible extent. We want to make our reporting as transparent to the readers as possible so they may know how and where we got our information. Transparency is honest and fair, two values we cherish.
• Sources often insist that we agree not to name them before they agree to talk with us. We must be reluctant to grant their wish. When we use an unnamed source, we are asking our readers to take an extra step to trust the credibility of the information we are providing. We must be certain in our own minds that the benefit to readers is worth the cost in credibility.
• In some circumstances, we will have no choice but to grant confidentiality to sources. We recognize that there are situations in which we can give our readers better, fuller information by allowing sources to remain unnamed than if we insist on naming them. We realize that in many circumstances, sources will be unwilling to reveal to us information about corruption in their own organizations, or high-level policy disagreements, for example, if disclosing their identities could cost them their jobs or expose them to harm. Nevertheless, granting anonymity to a source should not be done casually or automatically.
• Named sources are vastly to be preferred to unnamed sources. Reporters should press to have sources go on the record. We have learned over the years that persistently pushing sources to identify themselves actually works—not always, of course, but more often than many reporters initially expect. If a particular source refuses to allow us to identify him or her, the reporter should consider seeking the information elsewhere.
• Editors have an obligation to know the identity of unnamed sources used in a story, so that editors and reporters can jointly assess the appropriateness of using them. Some sources may insist that a reporter not reveal their identity to her editors; we should resist this. When it happens, the reporter should make clear that information so obtained cannot be published. The source of anything that is published will be known to at least one editor.
• We prefer at least two sources for factual information in Post stories that depends on confidential informants, and those sources should be independent of each other. We prefer sources with firsthand or direct knowledge of the information. A relevant document can sometimes serve as a second source. There are situations in which we will publish information from a single source, but we should only do so after deliberations involving the executive editor, the managing editor and the appropriate department head. The judgment to use a single source depends on the source’s reliability and the basis for the source’s information.
• We must strive to tell our readers as much as we can about why our unnamed sources deserve our confidence. Our  obligation  is to serve readers, not sources. This means avoiding attributions to “sources” or “informed sources.” Instead we should try to give the reader something more, such as “sources familiar with the thinking of defense lawyers in the case,” or “sources whose work brings them into contact with the county executive,” or “sources on the governor’s staff who disagree with his policy.”

How To Attribute Information From an Email, a Text, or a Social Media Post?

If a credible source responds to an interview in an email, attribution should indicate this.

Email attribution example:

The CEO of Mosette Healthcare Group, Lana Dunham, wrote in an email that she plans to merge the group with St. Catherine’s Health Systems.

Social media posts are tricky and should serve primarily as story ideas to pursue.

According to National Public Radio’s ethics handbook, social platforms can serve as good newsgathering tools, but NPR said that it:

“requires the same diligence we exercise when reporting in other environments. When NPR bloggers post about breaking news, they do not cite anonymous posts on social media sites — though they may use information they find there to guide their reporting. They carefully attribute the information they cite and are clear about what NPR has and has not been able to confirm.”

Also, social media users aren’t always who they say they are, which poses a verification problem. If it’s reasonably possible to identify an account and the posts or tweets coming from it, use something like this to attribute:

Social media example

Illinois Senator Tammy Duckworth , who gave birth to her second child on April 9, tweeted on April 19: “May have to vote today. Maile’s outfit is prepped. Made sure she has a jacket so she doesn’t violate the Senate floor dress code requiring blazers.Not sure what the policy is on duckling onesies but I think we’re ready”

Tweeted, posted, shared. Use the appropriate attribution verbs for their social platforms.

How To Attribute a File, Archive or Stock Photo or Video?

It must be attributed to its original source. Include the title, author, source and date it was accessed.

The Miriam and Ira D. Wallach Division of Art, Prints and Photographs: Photography Collection, The New York Public Library. “Farmhouse and family of resettlement client. Waldo County, Maine.” New York Public Library Digital Collections. Accessed April 19, 2018. NYPL Digital Collections

Attribute any Creative Commons photos or video by identifying the title of the work, the author or creator, the source (where it’s found) and license type. All Creative Commons work has a license type, which must be acknowledged.

Find specifics about CC attribution best practices here: Creative Commons attribution guidelines

To identify the digital rights of an image, use a search, such as the one developed by the Visual Resources Association: Image search resource

News Releases

Reproducing news releases – either sent or gathered from a website – has been a lively topic in nearly all news centers that use releases and among all organizations and businesses that send or post them.

Raymond James attorney and KU alum Ellyn Angelotti Kamke wrote about attribution and its squishy spots for The Poynter Institute. In a 2013 article, Kamke addressed the sometimes-disputed issue of plagiarism-without-attribution in which some journalists view verbatim news release use.

In her article, Kamke raised the question many in the industry ask frequently, “How should journalists use and attribute information that comes from an official source via press release, a prepared statement an official social-media account or some other widely distributed avenue?”

Attribute. Attribute. Attribute. For transparency and credibility. Attributed material, Kamke wrote, “even when it comes from an official source, gives the audience more context about that information and how it was acquired by the writer.”

Strategic communicators, such as those specializing in public relations, want their material used and more often than not put the research and good writing into a news release so it’s fit for immediate publication with minimal editing. But even PR professionals see the value of readers knowing the sources and making their own decisions about their veracity.

Peer Tutorial: Attribution Review

In this video , Maggie Gould and Paige Moyer (JOUR 302, fall 2018) review the key types of attribution.

A Practitioner’s View

Mike Miller

B.S., KU Journalism, 2000

Senior Director, Editorial, NBC Sports Digital

NBC Sports Digital, including flagship sites NBCSports.com, Rotoworld and ProFootballTalk, serves a sports audience that craves sports news and analysis. How do we do that? We do some original reporting and we rely on extensive story aggregation.

Any story that isn’t reported by our writers is explicitly credited and linked to high up in the story, sometimes in the initial graf. Our editorial standard is that we don’t do lengthy excerpts or extensive quoting (why reproduce what the original story already has?), because we don’t want any confusion as to where the original story originates.

Excerpts are italicized, set off with quotes, or both. It should clearly be separated from the rest of the story.

For us, this places the onus on our writers to extend that aggregated story with a specific editorial take or analysis that drives the story forward. If we can’t break the story, we tell our audience why it’s important, which helps the original story’s credibility (we create awareness) and gives us some authority through analysis.

Activity 1: Search

Locate a recent news article or press release. Evaluate what source types — primary and secondary, and human and nonhuman sources — the author consulted and how the author cited these sources. Does the author follow the practices recommended in this chapter? If not, specifically how could the article or press release be improved? Summarize your suggestions in one or two paragraphs.

Activity 2: Credibility check

Find a published piece from the Associated Press or The Washington Post that cites an anonymous source. After reading the piece, consider if the use of an anonymous source followed the publication’s policy on granting anonymity to sources and if the use of an anonymous source affected the credibility of the author, the article, or the publication. Summarize your findings in one or two paragraphs.

Be Credible Copyright © 2018 by Kerry Benson is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License , except where otherwise noted.

Share This Book

Quoting and Paraphrasing

Download this Handout PDF

College writing often involves integrating information from published sources into your own writing in order to add credibility and authority–this process is essential to research and the production of new knowledge.

However, when building on the work of others, you need to be careful not to plagiarize : “to steal and pass off (the ideas and words of another) as one’s own” or to “present as new and original an idea or product derived from an existing source.”1 The University of Wisconsin–Madison takes this act of “intellectual burglary” very seriously and considers it to be a breach of academic integrity . Penalties are severe.

These materials will help you avoid plagiarism by teaching you how to properly integrate information from published sources into your own writing.

1. Merriam Webster’s Collegiate Dictionary, 10th ed. (Springfield, MA: Merriam-Webster, 1993), 888.

How to avoid plagiarism

When using sources in your papers, you can avoid plagiarism by knowing what must be documented.

Specific words and phrases

If you use an author’s specific word or words, you must place those words within quotation marks and you must credit the source.

Information and Ideas

Even if you use your own words, if you obtained the information or ideas you are presenting from a source, you must document the source.

Information : If a piece of information isn’t common knowledge (see below), you need to provide a source.

Ideas : An author’s ideas may include not only points made and conclusions drawn, but, for instance, a specific method or theory, the arrangement of material, or a list of steps in a process or characteristics of a medical condition. If a source provided any of these, you need to acknowledge the source.

Common Knowledge?

You do not need to cite a source for material considered common knowledge:

General common knowledge is factual information considered to be in the public domain, such as birth and death dates of well-known figures, and generally accepted dates of military, political, literary, and other historical events. In general, factual information contained in multiple standard reference works can usually be considered to be in the public domain.

Field-specific common knowledge is “common” only within a particular field or specialty. It may include facts, theories, or methods that are familiar to readers within that discipline. For instance, you may not need to cite a reference to Piaget’s developmental stages in a paper for an education class or give a source for your description of a commonly used method in a biology report—but you must be sure that this information is so widely known within that field that it will be shared by your readers.

If in doubt, be cautious and cite the source. And in the case of both general and field-specific common knowledge, if you use the exact words of the reference source, you must use quotation marks and credit the source.

Paraphrasing vs. Quoting — Explanation

Should i paraphrase or quote.

In general, use direct quotations only if you have a good reason. Most of your paper should be in your own words. Also, it’s often conventional to quote more extensively from sources when you’re writing a humanities paper, and to summarize from sources when you’re writing in the social or natural sciences–but there are always exceptions.

In a literary analysis paper , for example, you”ll want to quote from the literary text rather than summarize, because part of your task in this kind of paper is to analyze the specific words and phrases an author uses.

In research papers , you should quote from a source

• to show that an authority supports your point
• to present a position or argument to critique or comment on
• to include especially moving or historically significant language
• to present a particularly well-stated passage whose meaning would be lost or changed if paraphrased or summarized

You should summarize or paraphrase when

• what you want from the source is the idea expressed, and not the specific language used to express it
• you can express in fewer words what the key point of a source is

How to paraphrase a source

General advice.

• When reading a passage, try first to understand it as a whole, rather than pausing to write down specific ideas or phrases.
• Be selective. Unless your assignment is to do a formal or “literal” paraphrase, you usually don?t need to paraphrase an entire passage; instead, choose and summarize the material that helps you make a point in your paper.
• Think of what “your own words” would be if you were telling someone who’s unfamiliar with your subject (your mother, your brother, a friend) what the original source said.
• Remember that you can use direct quotations of phrases from the original within your paraphrase, and that you don’t need to change or put quotation marks around shared language.

Methods of Paraphrasing

• Look away from the source then write. Read the text you want to paraphrase several times until you feel that you understand it and can use your own words to restate it to someone else. Then, look away from the original and rewrite the text in your own words.
• Take notes. Take abbreviated notes; set the notes aside; then paraphrase from the notes a day or so later, or when you draft.

If you find that you can’t do A or B, this may mean that you don’t understand the passage completely or that you need to use a more structured process until you have more experience in paraphrasing.

The method below is not only a way to create a paraphrase but also a way to understand a difficult text.

Paraphrasing difficult texts

Consider the following passage from Love and Toil (a book on motherhood in London from 1870 to 1918), in which the author, Ellen Ross, puts forth one of her major arguments:

• Love and Toil maintains that family survival was the mother’s main charge among the large majority of London?s population who were poor or working class; the emotional and intellectual nurture of her child or children and even their actual comfort were forced into the background. To mother was to work for and organize household subsistence. (p. 9)

Children of the poor at the turn of the century received little if any emotional or intellectual nurturing from their mothers, whose main charge was family survival. Working for and organizing household subsistence were what defined mothering. Next to this, even the children’s basic comfort was forced into the background (Ross, 1995).

According to Ross (1993), poor children at the turn of the century received little mothering in our sense of the term. Mothering was defined by economic status, and among the poor, a mother’s foremost responsibility was not to stimulate her children’s minds or foster their emotional growth but to provide food and shelter to meet the basic requirements for physical survival. Given the magnitude of this task, children were deprived of even the “actual comfort” (p. 9) we expect mothers to provide today.

You may need to go through this process several times to create a satisfactory paraphrase.

Successful vs. unsuccessful paraphrases

Paraphrasing is often defined as putting a passage from an author into “your own words.” But what are your own words? How different must your paraphrase be from the original?

The paragraphs below provide an example by showing a passage as it appears in the source, two paraphrases that follow the source too closely, and a legitimate paraphrase.

The student’s intention was to incorporate the material in the original passage into a section of a paper on the concept of “experts” that compared the functions of experts and nonexperts in several professions.

The Passage as It Appears in the Source

Critical care nurses function in a hierarchy of roles. In this open heart surgery unit, the nurse manager hires and fires the nursing personnel. The nurse manager does not directly care for patients but follows the progress of unusual or long-term patients. On each shift a nurse assumes the role of resource nurse. This person oversees the hour-by-hour functioning of the unit as a whole, such as considering expected admissions and discharges of patients, ascertaining that beds are available for patients in the operating room, and covering sick calls. Resource nurses also take a patient assignment. They are the most experienced of all the staff nurses. The nurse clinician has a separate job description and provides for quality of care by orienting new staff, developing unit policies, and providing direct support where needed, such as assisting in emergency situations. The clinical nurse specialist in this unit is mostly involved with formal teaching in orienting new staff. The nurse manager, nurse clinician, and clinical nurse specialist are the designated experts. They do not take patient assignments. The resource nurse is seen as both a caregiver and a resource to other caregivers. . . . Staff nurses have a hierarchy of seniority. . . . Staff nurses are assigned to patients to provide all their nursing care. (Chase, 1995, p. 156)

Word-for-Word Plagiarism

Critical care nurses have a hierarchy of roles. The nurse manager hires and fires nurses. S/he does not directly care for patients but does follow unusual or long-term cases. On each shift a resource nurse attends to the functioning of the unit as a whole, such as making sure beds are available in the operating room , and also has a patient assignment . The nurse clinician orients new staff, develops policies, and provides support where needed . The clinical nurse specialist also orients new staff, mostly by formal teaching. The nurse manager, nurse clinician, and clinical nurse specialist , as the designated experts, do not take patient assignments . The resource nurse is not only a caregiver but a resource to the other caregivers . Within the staff nurses there is also a hierarchy of seniority . Their job is to give assigned patients all their nursing care .

Why this is plagiarism

Notice that the writer has not only “borrowed” Chase’s material (the results of her research) with no acknowledgment, but has also largely maintained the author’s method of expression and sentence structure. The phrases in red are directly copied from the source or changed only slightly in form.

Even if the student-writer had acknowledged Chase as the source of the content, the language of the passage would be considered plagiarized because no quotation marks indicate the phrases that come directly from Chase. And if quotation marks did appear around all these phrases, this paragraph would be so cluttered that it would be unreadable.

A Patchwork Paraphrase

Chase (1995) describes how nurses in a critical care unit function in a hierarchy that places designated experts at the top and the least senior staff nurses at the bottom. The experts — the nurse manager, nurse clinician, and clinical nurse specialist — are not involved directly in patient care. The staff nurses, in contrast, are assigned to patients and provide all their nursing care . Within the staff nurses is a hierarchy of seniority in which the most senior can become resource nurses: they are assigned a patient but also serve as a resource to other caregivers. The experts have administrative and teaching tasks such as selecting and orienting new staff, developing unit policies , and giving hands-on support where needed.

This paraphrase is a patchwork composed of pieces in the original author’s language (in red) and pieces in the student-writer’s words, all rearranged into a new pattern, but with none of the borrowed pieces in quotation marks. Thus, even though the writer acknowledges the source of the material, the underlined phrases are falsely presented as the student’s own.

A Legitimate Paraphrase

In her study of the roles of nurses in a critical care unit, Chase (1995) also found a hierarchy that distinguished the roles of experts and others. Just as the educational experts described above do not directly teach students, the experts in this unit do not directly attend to patients. That is the role of the staff nurses, who, like teachers, have their own “hierarchy of seniority” (p. 156). The roles of the experts include employing unit nurses and overseeing the care of special patients (nurse manager), teaching and otherwise integrating new personnel into the unit (clinical nurse specialist and nurse clinician), and policy-making (nurse clinician). In an intermediate position in the hierarchy is the resource nurse, a staff nurse with more experience than the others, who assumes direct care of patients as the other staff nurses do, but also takes on tasks to ensure the smooth operation of the entire facility.

Why this is a good paraphrase

The writer has documented Chase’s material and specific language (by direct reference to the author and by quotation marks around language taken directly from the source). Notice too that the writer has modified Chase’s language and structure and has added material to fit the new context and purpose — to present the distinctive functions of experts and nonexperts in several professions.

Shared Language

Perhaps you’ve noticed that a number of phrases from the original passage appear in the legitimate paraphrase: critical care, staff nurses, nurse manager, clinical nurse specialist, nurse clinician, resource nurse.

If all these phrases were in red, the paraphrase would look much like the “patchwork” example. The difference is that the phrases in the legitimate paraphrase are all precise, economical, and conventional designations that are part of the shared language within the nursing discipline (in the too-close paraphrases, they’re red only when used within a longer borrowed phrase).

In every discipline and in certain genres (such as the empirical research report), some phrases are so specialized or conventional that you can’t paraphrase them except by wordy and awkward circumlocutions that would be less familiar (and thus less readable) to the audience.

When you repeat such phrases, you’re not stealing the unique phrasing of an individual writer but using a common vocabulary shared by a community of scholars.

Some Examples of Shared Language You Don’t Need to Put in Quotation Marks

• Conventional designations: e.g., physician’s assistant, chronic low-back pain
• Preferred bias-free language: e.g., persons with disabilities
• Technical terms and phrases of a discipline or genre : e.g., reduplication, cognitive domain, material culture, sexual harassment

Chase, S. K. (1995). The social context of critical care clinical judgment. Heart and Lung, 24, 154-162.

How to Quote a Source

Introducing a quotation.

One of your jobs as a writer is to guide your reader through your text. Don’t simply drop quotations into your paper and leave it to the reader to make connections.

Integrating a quotation into your text usually involves two elements:

• A signal that a quotation is coming–generally the author’s name and/or a reference to the work
• An assertion that indicates the relationship of the quotation to your text

Often both the signal and the assertion appear in a single introductory statement, as in the example below. Notice how a transitional phrase also serves to connect the quotation smoothly to the introductory statement.

Ross (1993), in her study of poor and working-class mothers in London from 1870-1918 [signal], makes it clear that economic status to a large extent determined the meaning of motherhood [assertion]. Among this population [connection], “To mother was to work for and organize household subsistence” (p. 9).

The signal can also come after the assertion, again with a connecting word or phrase:

Illness was rarely a routine matter in the nineteenth century [assertion]. As [connection] Ross observes [signal], “Maternal thinking about children’s health revolved around the possibility of a child’s maiming or death” (p. 166).

Formatting Quotations

Short direct prose.

Incorporate short direct prose quotations into the text of your paper and enclose them in double quotation marks:

According to Jonathan Clarke, “Professional diplomats often say that trying to think diplomatically about foreign policy is a waste of time.”

Longer prose quotations

Begin longer quotations (for instance, in the APA system, 40 words or more) on a new line and indent the entire quotation (i.e., put in block form), with no quotation marks at beginning or end, as in the quoted passage from our Successful vs. Unsucessful Paraphrases page.

Rules about the minimum length of block quotations, how many spaces to indent, and whether to single- or double-space extended quotations vary with different documentation systems; check the guidelines for the system you’re using.

Quotation of Up to 3 Lines of Poetry

Quotations of up to 3 lines of poetry should be integrated into your sentence. For example:

In Julius Caesar, Antony begins his famous speech with “Friends, Romans, Countrymen, lend me your ears; / I come to bury Caesar, not to praise him” (III.ii.75-76).

Notice that a slash (/) with a space on either side is used to separate lines.

Quotation of More than 3 Lines of Poetry

More than 3 lines of poetry should be indented. As with any extended (indented) quotation, do not use quotation marks unless you need to indicate a quotation within your quotation.

Punctuating with Quotation Marks

Parenthetical citations.

With short quotations, place citations outside of closing quotation marks, followed by sentence punctuation (period, question mark, comma, semi-colon, colon):

Menand (2002) characterizes language as “a social weapon” (p. 115).

With block quotations, check the guidelines for the documentation system you are using.

Commas and periods

Place inside closing quotation marks when no parenthetical citation follows:

Hertzberg (2002) notes that “treating the Constitution as imperfect is not new,” but because of Dahl’s credentials, his “apostasy merits attention” (p. 85).

Semicolons and colons

Place outside of closing quotation marks (or after a parenthetical citation).

Question marks and exclamation points

Place inside closing quotation marks if the quotation is a question/exclamation:

Menand (2001) acknowledges that H. W. Fowler’s Modern English Usage is “a classic of the language,” but he asks, “Is it a dead classic?” (p. 114).

[Note that a period still follows the closing parenthesis.]

Place outside of closing quotation marks if the entire sentence containing the quotation is a question or exclamation:

How many students actually read the guide to find out what is meant by “academic misconduct”?

Quotation within a quotation

Use single quotation marks for the embedded quotation:

According to Hertzberg (2002), Dahl gives the U. S. Constitution “bad marks in ‘democratic fairness’ and ‘encouraging consensus'” (p. 90).

[The phrases “democratic fairness” and “encouraging consensus” are already in quotation marks in Dahl’s sentence.]

Indicating Changes in Quotations

Quoting only a portion of the whole.

Use ellipsis points (. . .) to indicate an omission within a quotation–but not at the beginning or end unless it’s not obvious that you’re quoting only a portion of the whole.

Adding Clarification, Comment, or Correction

Within quotations, use square brackets [ ] (not parentheses) to add your own clarification, comment, or correction.

Use [sic] (meaning “so” or “thus”) to indicate that a mistake is in the source you’re quoting and is not your own.

Additional information

Information on summarizing and paraphrasing sources.

American Heritage Dictionary of the English Language (4th ed.). (2000). Retrieved January 7, 2002, from http://www.bartleby.com/61/ Bazerman, C. (1995). The informed writer: Using sources in the disciplines (5th ed). Boston: Houghton Mifflin. Leki, I. (1995). Academic writing: Exploring processes and strategies (2nd ed.) New York: St. Martin?s Press, pp. 185-211.

Leki describes the basic method presented in C, pp. 4-5.

Spatt, B. (1999). Writing from sources (5th ed.) New York: St. Martin?s Press, pp. 98-119; 364-371.

Information about specific documentation systems

The Writing Center has handouts explaining how to use many of the standard documentation systems. You may look at our general Web page on Documentation Systems, or you may check out any of the following specific Web pages.

If you’re not sure which documentation system to use, ask the course instructor who assigned your paper.

• American Psychological Assoicaion (APA)
• Modern Language Association (MLA)
• Chicago/Turabian (A Footnote or Endnote System)
• American Political Science Association (APSA)
• Council of Science Editors (CBE)
• Numbered References

You may also consult the following guides:

• American Medical Association, Manual for Authors and Editors
• Council of Science Editors, CBE style Manual
• The Chicago Manual of Style
• MLA Handbook for Writers of Research Papers
• Publication Manual of the American Psychological Association

Academic and Professional Writing

This is an accordion element with a series of buttons that open and close related content panels.

Analysis Papers

Reading Poetry

A Short Guide to Close Reading for Literary Analysis

Using Literary Quotations

Play Reviews

Writing a Rhetorical Précis to Analyze Nonfiction Texts

Incorporating Interview Data

Grant Proposals

Planning and Writing a Grant Proposal: The Basics

Additional Resources for Grants and Proposal Writing

Job Materials and Application Essays

Writing Personal Statements for Ph.D. Programs

• Before you begin: useful tips for writing your essay
• Guided brainstorming exercises
• Get more help with your essay
• Frequently Asked Questions

Resume Writing Tips

CV Writing Tips

Cover Letters

Business Letters

Proposals and Dissertations

Resources for Proposal Writers

Resources for Dissertators

Research Papers

Planning and Writing Research Papers

Writing Annotated Bibliographies

Creating Poster Presentations

Writing an Abstract for Your Research Paper

Thank-You Notes

Advice for Students Writing Thank-You Notes to Donors

Reading for a Review

Critical Reviews

Writing a Review of Literature

Scientific Reports

Scientific Report Format

Sample Lab Assignment

Writing for the Web

Writing an Effective Blog Post

Writing for Social Media: A Guide for Academics

• Prospective Students
• Current Students
• Faculty & Staff

Citation Guide

• Citation Basics

Quoting, Paraphrasing, Summarizing & Patchwriting

Quotes, paraphrases, and summaries are different methods of incorporating other people's ideas and words into your research. You use quotes, paraphrases, and summaries to provide evidence of having researched your topic, which shows you have a thorough understanding of the topic you are discussing, and to show support for your arguments.

Below you will find specific information on each of these different methods of incorporating sources into your research including what they are and best practices for including them in your project.

Quoting is when you use the exact words of another person.

Using Quotes

• Quotes should be used sparingly in your research since they do not show that you have digested and understood the material or showcase your writing ability.  They only show that you've read information related to your topic.
• Quotes should only be used when the exact wording is important, or you are unable to paraphrase the author's words.
• It is best to integrate quotes into your sentences rather than use them as stand-alone sentences. (See the example below for how to incorporate a quote into your sentence.)
• Try not to quote an entire sentence, unless absolutely necessary.  Only quote the most important words or information.
• It is a good idea to  start or end a sentence containing a quote  with your own words to tie the quote back into your assignment. This shows you are applying the quote to support your own ideas and are adding value to the quote.
• Consider using a signal phrase to introduce your quote to the reader.

Formatting Quotes

Short quotations , usually those under 4 lines of text, are enclosed in quotation marks and include an in-text citation at the end. (Check the style guide assigned by your professor for information on formatting the citation correctly.)

Example: Graphic novels have educational and literary value and can be used in science classes to " engage students and support literacy skill development, " (Haroldson, p. 37).

Haroldson, Rachelle. “PICTURE THIS! The Versatility of Graphic Novels in Science Class.” Science Teacher , vol. 89, no. 2, Nov. 2021, p. 37-43. EBSCOhos t, search.ebscohost.com/login.aspx?direct=true&AuthType=cookie,cpid&custid=norwalk&db=sch&AN=153491142&site=eds-live .

Long quotations , usually those at or over 4 lines of text, are often called "block quotes" and have specific formatting that differs depending on the citation style being used.  Check the style guide assigned by your professor for direction on how to properly format a block quote.

(largely adapted from content on James Cook University Library " Writing Guide " licensed under a  Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 International License  and with permission from  Purdue University Online Writing Lab )

Summarizing

Summarizing is when you take a large portion of the material (paragraphs, chapters, or the entire contents) and condense it down to the main points using your own words.  A summary is very short overview of the resource, or portion of the resource, focusing on the key concepts.

Using Summaries

• Summaries must be written entirely using  your own words
• Summaries should be used when you need to quickly introduce background information or another person's ideas into your work to provide context or help set-up your analysis for your reader
• Try not to use summaries as "stand-alone evidence". You should always try to provide your own commentary, opinions, and/or analysis on the content of the summary.

Formatting Summaries

There is no special formatting for including summaries in your writing, except that you must still include an in-text citation , citing where you got the information provided in the summary. (Check the style guide assigned by your professor for information on formatting the in-text citation correctly.)

Example: In her article, "Picture This!" Rachelle Haroldson discusses the benefits of using graphic novels in science classrooms and provides suggestions for teachers on how to incorporate them into their lessons (pp. 37-43).

(adapted from content on James Cook University Library " Writing Guide " licensed under a  Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 International License )

Paraphrasing

Paraphrasing is when you restate a concept or passage from someone else using your own words. In addition to changing the words used, correct paraphrasing also changes the sentence structure. Unlike summaries, which are much condensed representations of the original work, paraphrases are typically similar in length to the original text. Paraphrases often include your own thoughts, interpretations, and understanding of the information being conveyed.

Using Paraphrases

• Shows that you've done research into your topic
• Shows understanding of the topic and source being cited, and adds value to the conversation surrounding the topic being discussed
• Provides supporting evidence for your arguments, adding credibility to your ideas and work
• Improves the flow of your paper, by retaining your own tone and voice rather than that of your sources
• It is not  simply swapping out words or phrases with synonyms, or reordering the phrases within the sentence (this is considered "patchwriting" and is a form of plagiarism - see the section on Patchwriting below for more information)
• Paraphrasing should be the most frequently utilized method of incorporating sources into your research since it shows a deeper understanding of the material and that you have developed your own thoughts on the topic
• When paraphrasing you should seek to include your own thoughts, interpretations, and/or analysis of the information being paraphrased

Formatting Paraphrases

There is no special formatting for including paraphrases in your writing, except that you must still include an in-text citation , citing where you got the information being paraphrased. (Check the style guide assigned by your professor for information on formatting the in-text citation correctly.)

Example: Haroldson suggests that science teachers incorporate graphic novels into their lessons because the pictorial format encourages student interest and therefore engagement in scientific concepts and supports literacy acquisition (pp. 37-38).

(adapted from content on James Cook University Library " Writing Guide " licensed under a  Creative Commons Attribution-ShareAlike (CC BY-SA) 4.0 International License  and Academic Integrity ,  by Ulrike Kestler, licensed under a  Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License )

Patchwriting

Patchwriting is when you only change a few words or the sentence structure without incorporating your own ideas or voice in an attempt to paraphrase. Patchwriting mimics the language and structure of the original source and therefore cannot be considered a true paraphrase, which also requires your own thoughts and understanding of the content be included. As such, patchwriting is often an unintentional form of plagiarism.

Patchwriting vs. Paraphrasing

(from Academic Integrity , by Ulrike Kestler, licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License )

Patchwriting Example

Original Text

Graphic novels by their nature are intriguing. They are colorful, full of pictures, word bubbles, and funny sounds coming out of the characters’ mouths. Many recent publications are culturally diverse, offering traditionally underrepresented students the opportunity to see themselves in the texts and majority students the opportunity to connect with different characters and perspectives. (Haroldson, p. 39)

Source: Haroldson, Rachelle. “PICTURE THIS! The Versatility of Graphic Novels in Science Class.”  Science Teacher , vol. 89, no. 2, Nov. 2021, p. 37-43.  EBSCOhos t,  search.ebscohost.com/login.aspx?direct=true&AuthType=cookie,cpid&custid=norwalk&db=sch&AN=153491142&site=eds-live .

Graphic novels are inherently captivating due to their vibrant visuals, use of illustrations, word bubbles, and sound effects. Many contemporary graphic novels embrace cultural diversity, providing traditionally marginalized students with a chance to identify with characters who resemble them, while also allowing students in majority groups to engage with diverse perspectives and characters (Haroldson, p. 39).

Graphic novels have a unique ability to capture student interest in scientific concepts due to their inherently entertaining method of conveying information through visual imagery. Moreover, these novels offer a valuable opportunity for students from traditionally underrepresented groups to identify with and relate to the characters and storylines presented within the narrative, making students more likely to engage with the material (Haroldson, p. 39).

Additional Resources

• Quoting, Paraphrasing, and Summarizing - Purdue OWL A guide from Purdue OWL explaining what quotes, paraphrases, and summaries are, what they're used for, and how to use them.
• The Writing Guide - James Cook University Library A guide detailing the research and writing process with an entire page dedicated to using sources appropriately.
• Reading and Writing with Sources PowerPoint - High School
• Reading and Writing with Sources PowerPoint - College
• Paraphrasing vs. Patchwriting from KPU's Academic Integrity eBook Explanation of the difference between paraphrasing and patchwriting with an example and activity to check your understanding.
• Paraphrasing vs. Patchwriting - Ch. 13 of CS 050 Academic Writing and Grammar eBook Chapter explaining the difference between paraphrasing and patchwriting with videos, examples, and activities to test your knowledge, by the Confederation College Communications Department and Paterson Library Commons.
• ‘Patchwriting’ is more common than plagiarism, just as dishonest, by Kelly McBride - Poynter (2012) A Poynter article discussing what patchwriting is, how common it is, the ethics of patchwriting and its implications, with a particular focus on its use in journalism.
• The Citation Project The Citation Project is a series of research studies on source use. Their purpose is to provide data and analyses that can help with educators’ questions about plagiarism, information literacy, and the teaching of source-based writing. more... less... By collecting data and replicating or adapting the methods of other studies to analyze it, ongoing Citation Project research builds on and extends the work of other scholars, generating deeper and more nuanced understanding of source-based writing. (description from website)
• Paraphrasing - Penn State Academic Integrity Tutorial A page on the Penn State Academic Integrity tutorial with information on correct and incorrect attempts at paraphrasing.
• How to Quote | Citing Quotes in APA, MLA & Chicago A Scribbr guide on using and citing quotes in academic writing, complete with examples, videos, and FAQs.
• << Previous: Citation Basics
• Next: MLA >>
• Last Updated: May 16, 2024 1:52 PM
• URL: https://library.ctstate.edu/citations

How to Use Attribution Correctly in Journalism

And Why It's Important

Mihajlo Maricic/Getty Images

• Writing Essays
• Writing Research Papers
• English Grammar
• M.S., Journalism, Columbia University
• B.A., Journalism, University of Wisconsin-Madison

To a journalist, attribution simply means telling your readers where the information in your story comes from, as well as who is being quoted.

Generally, attribution means using a source’s full name and job title if that's relevant. Information from sources can be paraphrased or quoted directly, but in both cases, it should be attributed.

Attribution Style

Keep in mind that on-the-record attribution—meaning a source's full name and job title are given—should be used whenever possible. On-the-record attribution is inherently more credible than any other type of attribution for the simple reason that the ​ source has put their name on the line with the information they've provided.

But there are some cases where a source might not be willing to give full on-the-record attribution.

Let's say you're an investigative journalist looking into allegations of corruption in city government. You have a source in the mayor's office who is willing to give you information, but they're worried about repercussions if their name is revealed. In that case, you as the reporter would talk to this source about what kind of attribution they are willing to commit to. You are compromising on full on-the-record attribution because the story is worth getting for the public good.

Here are some examples of different kinds of attribution.

Source – Paraphrase

Jeb Jones, a resident of the trailer park, said the sound of the tornado was terrifying.

Source – Direct Quote

“It sounded like a giant locomotive train coming through. I’ve never heard anything like it,” said Jeb Jones, who lives in the trailer park.

Journalists often use both paraphrases and direct quotes from a source. Direct quotes provide immediacy and a more connected, human element to the story. They tend to draw the reader in.

Source – Paraphrase and Quote

“It sounded like a giant locomotive train coming through. I’ve never heard anything like it,” Jones said.

(Notice that in Associated Press style , a source’s full name is used on the first reference, then just the last name on all subsequent references. If your source has a specific title or rank, use the title before their full name on the first reference, then just the last name after that.)

When to Attribute

Any time the information in your story comes from a source and not from your own firsthand observations or knowledge, it must be attributed. A good rule of thumb is to attribute once per paragraph if you are telling the story mainly through comments from an interview or eyewitnesses to an event. It might seem repetitive, but it’s important for journalists to be clear about where their information originates.

Example: The suspect escaped from the police van on Broad Street, and officers captured him about a block away on Market Street, said Lt. Jim Calvin.

Different Types of Attribution

In his book News Reporting and Writing , journalism professor Melvin Mencher outlines four distinct types of attribution:

1. On the record: All statements are directly quotable and attributable, by name and title, to the person making the statement. This is the most valuable type of attribution.

Example: "The U.S. has no plans to invade Iran," said White House press secretary Jim Smith.

2. On Background: All statements are directly quotable but can't be attributed by name or specific title to the person commenting.

Example: "The U.S. has no plans to invade Iran," a White House spokesman said.

3. On Deep Background: Anything that is said in the interview is usable but not in a  direct quotation and not for attribution. The reporter writes it in their own words. 

Example: Invading Iran is not in the cards for the U.S. 

4. Off the Record: Information is for the reporter's use only and is not to be published. The information also is not to be taken to another source in hopes of getting confirmation. 

You probably don’t need to get into all of Mencher’s categories when you’re interviewing a source. But you should clearly establish how the information your source gives you can be attributed.

• How to Work with Anonymous Sources
• Here's How to Use Attribution to Avoid Plagiarism in Your News Stories
• 10 Important Steps for Producing a Quality News Story
• What Is an Anonymous Source?
• Definition and Examples of Direct Quotations
• Avoid the Common Mistakes That Beginning Reporters Make
• What Is Plagiarism?
• How to Conduct Interviews for News Stories
• Everything You Need to Know About Interviewing
• 15 News Writing Rules for Beginning Journalism Students
• These Are Frequently Used Journalism Terms You Need to Know
• What Is a Citation?
• How to Cite Genealogy Sources
• How to Write a News Article That's Effective
• How to Use Shakespeare Quotes
• How and When to Paraphrase Quotations

• Library Catalogue

Techniques for Paraphrasing

When you write a paraphrase, you restate other’s ideas in your own words. That is, you write the meaning of the author’s ideas. You use some of the author’s key terms, but you use many of your own words and sentence structures. You include in-text citation, including the author’s last name and (for APA style) the year of publication.

An effective paraphrase includes more than one of the following techniques. If you use only one of these techniques when paraphrasing, you have not paraphrased effectively.  

• Change a word from one part of speech to another

Original:  Medical professor John Swanson says that global changes are influencing the spread of disease.

Paraphrase:  According to John Swanson, a professor of medicine, changes across the globe are causing diseases to spread (James, 2004).

• Use synonyms

Original:  The U.S. government declared that the AIDS crisis poses a national security threat.  The announcement followed an intelligence report that found high rates of HIV infection could lead to widespread political destabilization.

Paraphrase: The government of the United States announced that AIDS could harm the nation's security.  The government warned the population after an important governmental study concluded that political problems could result from large numbers of people infected with HIV (Snell, 2005).

• Change numbers and percentages to different forms

Original: Minority groups in the United States have been hit hardest by the epidemic.  African Americans, who make up 13 percent of the U.S. population, accounted for 46 percent of the AIDS cases diagnosed in 1998.

Paraphrase: The AIDS epidemic has mostly affected minorities in the United States.  For example, in 1998, less than 15 percent of the total population was African, but almost half of the people diagnosed with AIDS in the United States that year were African America (Jenson, 2000).

• Change word order:  this might include changing from active to passive voice or moving modifiers to different positions.  

Original: Angier (2001) reported that malaria kills more than one million people annually, the overwhelming majority of them children in sub-Saharan Africa.

Paraphrase: Every year, more than a million people are killed by malaria, and most of the victims are children who live in sub-Saharan Africa (Angier, 2001).

    5. Use different definition structures

Original: Lyme disease is an inflammatory disease caused by a bacterium transmitted by ticks (small bloodsucking arachnids that attach themselves to larger animals). The disease is usually characterized by a rash followed by flu-like symptoms, including fever, joint pain, and headache.

Paraphrase: Lyme disease-a disease that causes swelling and redness-is caused by a bacterium carried by a small arachnid known as a tick. The ticks attach to and suck the blood of animals and humans, transferring some of the Lyme disease bacteria into their hosts and causing symptoms similar to the flu (Wald, 2005).

    6. Use different attribution signals

Original: “That’s because there are so many different ways the diseases could have arrived,” veterinarian Mark Walters declared in his recent book, Six Modern Plagues.

Paraphrase: According to Mark Walters, a veterinarian who wrote Six Modern Plagues, the disease could have arrived in numerous ways (Peterson, 2004).

   7. Change the sentence structure and use different connecting words

Original:  Although only about one-tenth of the world’s population lives there, sub-Saharan Africa remains the hardest hit region, accounting for 72 percent of the people infected with HIV during 2000.

Paraphrase: Approximately 10 percent of the world’s population resides in sub-Saharan Africa.  However, this area of the world has the highest percentage of AIDS-related illnesses.  In fact, in 2000, almost three-fourths of the population had the HIV virus (Bunting, 2004).

Caution:  When paraphrasing, do not change key terms or proper nouns.

Original: In the northeastern United States, people are building homes on the edge of woods, where ticks that carry Lyme disease hitch rides on deer.  In addition, in Africa, hunters bring back the meat of animals that scientists think may transmit Ebola, a usually fatal disease that causes massive hemorrhaging in its victims.

Paraphrase: In the United States, residential areas are being built near wooded areas in the northeast. These areas are also the homes of ticks carrying Lyme disease.  Also, according to scientists, hunters in Africa kill animals that may carry the Ebola virus (an often fatal virus that causes massive hemorrhaging) (Yaya, 2004).

Schuemann, C., Bryd, P., & Reid, J. (2006). College Writing 4 (1st ed.). USA: Heinle/ELT. Reproduced by permission .

• school Campus Bookshelves
• menu_book Bookshelves
• perm_media Learning Objects
• login Login
• how_to_reg Request Instructor Account
• hub Instructor Commons

Margin Size

• Download Page (PDF)
• Download Full Book (PDF)
• Periodic Table
• Physics Constants
• Scientific Calculator
• Reference & Cite
• Tools expand_more
• Readability

selected template will load here

This action is not available.

12.6: Quoting and Paraphrasing

• Last updated
• Save as PDF
• Page ID 120102

• Amy Guptill
• The College at Brockport, SUNY

\( \newcommand{\vecs}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \)

\( \newcommand{\vecd}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash {#1}}} \)

\( \newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\)

( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\)

\( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\)

\( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\)

\( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\)

\( \newcommand{\Span}{\mathrm{span}}\)

\( \newcommand{\id}{\mathrm{id}}\)

\( \newcommand{\kernel}{\mathrm{null}\,}\)

\( \newcommand{\range}{\mathrm{range}\,}\)

\( \newcommand{\RealPart}{\mathrm{Re}}\)

\( \newcommand{\ImaginaryPart}{\mathrm{Im}}\)

\( \newcommand{\Argument}{\mathrm{Arg}}\)

\( \newcommand{\norm}[1]{\| #1 \|}\)

\( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\AA}{\unicode[.8,0]{x212B}}\)

\( \newcommand{\vectorA}[1]{\vec{#1}}      % arrow\)

\( \newcommand{\vectorAt}[1]{\vec{\text{#1}}}      % arrow\)

\( \newcommand{\vectorB}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \)

\( \newcommand{\vectorC}[1]{\textbf{#1}} \)

\( \newcommand{\vectorD}[1]{\overrightarrow{#1}} \)

\( \newcommand{\vectorDt}[1]{\overrightarrow{\text{#1}}} \)

\( \newcommand{\vectE}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash{\mathbf {#1}}}} \)

Media Alternative

Listen to an audio version of this page (15 min, 22 sec):

We have seen that paragraphs need supporting sentences, but how specifically can we bring in quotations and paraphrases of other sources into our own essay?

Listen to your sources

Have you ever had the maddening experience of arguing with someone who twisted your words to make it seem like you were saying something you weren’t? Novice writers sometimes inadvertently misrepresent their sources when they quote very minor points from an article or even positions that the authors of an article disagree with. It often happens when students approach their sources with the goal of finding snippets that align with their own opinion. For example, the passage above contains the phrase “measuring teachers’ performance by student test scores is the best way to improve education.” An inexperienced writer might include that quote in a paper without making it clear that the author(s) of the source actually dispute that very claim. Doing so is not intentionally fraudulent, but it reveals that the paper-writer isn’t really thinking about and responding to claims and arguments made by others. In that way, it harms his or her credibility.

Academic journal articles are especially likely to be misrepresented by student writers because their literature review sections often summarize a number of contrasting viewpoints. For example, sociologists Jennifer C. Lee and Jeremy Staff wrote a paper in which they note that high-schoolers who spend more hours at a job are more likely to drop out of school. 1 However, Lee and Staff’s analysis finds that working more hours doesn’t actually make a student more likely to drop out. Instead, the students who express less interest in school are both more likely to work a lot of hours and more likely to drop out. In short, Lee and Staff argue that disaffection with school causes students to drop-out, not working at a job. In reviewing prior research about the impact of work on dropping out, Lee and Staff write “Paid work, especially when it is considered intensive, reduces grade point averages, time spent on homework, educational aspirations, and the likelihood of completing high school” 2 . If you included that quote without explaining how it fits into Lee and Staff’s actual argument, you would be misrepresenting that source.

Provide context

Another error beginners often make is to drop in a quote without any context. If you simply quote, “Students begin preschool with a set of self-regulation skills that are a product of their genetic inheritance and their family environment” (Willingham, 2011, p.24), your reader is left wondering who Willingham is, why he or she is included here, and where this statement fits into his or her larger work. The whole point of incorporating sources is to situate your own insights in the conversation. As part of that, you should provide some kind of context the first time you use that source. Some examples:

Willingham, a cognitive scientist, claims that …

Research in cognitive science has found that … (Willingham, 2011).

Willingham argues that “Students begin preschool with a set of self-regulation skills that are a product of their genetic inheritance and their family environment” (Willingham, 2011, 24). Drawing on findings in cognitive science, he explains “…”

As the first example above shows, providing a context doesn’t mean writing a brief biography of every author in your bibliography—it just means including some signal about why that source is included in your text.

Quoted material that does not fit into the flow of the text baffles the reader even more. For example, a novice student might write,

Schools and parents shouldn’t set limits on how much teenagers are allowed to work at jobs. “We conclude that intensive work does not affect the likelihood of high school dropout among youths who have a high propensity to spend long hours on the job” (Lee and Staff, 2007, p. 171). Teens should be trusted to learn how to manage their time.

The reader is thinking, who is this sudden, ghostly “we”? Why should this source be believed? If you find that passages with quotes in your draft are awkward to read out loud, that’s a sign that you need to contextualize the quote more effectively. Here’s a version that puts the quote in context:

Schools and parents shouldn’t set limits on how much teenagers are allowed to work at jobs. Lee and Staff’s carefully designed study found that “intensive work does not affect the likelihood of high school dropout among youths who have a high propensity to spend long hours on the job” (2007, p. 171). Teens should be trusted to learn how to manage their time.

In this latter example, it’s now clear that Lee and Staff are scholars and that their empirical study is being used as evidence for this argumentative point. Using a source in this way invites the reader to check out Lee and Staff’s work for themselves if they doubt this claim.

Many writing instructors encourage their students to contextualize their use of sources by making a “ quotation sandwich ”; that is, introduce the quote in some way and then follow it up with your own words. If you’ve made a bad habit of dropping in unintroduced quotes, the quotation sandwich idea may help you improve your skills, but in general you don’t need to approach every quote or paraphrase as a three-part structure to have well-integrated sources. You should, however, avoid ending a paragraph with a quotation. If you’re struggling to figure out what to write after a quote or close paraphrase, it may be that you haven’t yet figured out what role the quote is playing in your own analysis. If that happens to you a lot, try writing the whole first draft in your own words and then incorporate material from sources as you revise with “They Say/I Say” in mind.

Use sources efficiently

Some student writers are in a rut of only quoting whole sentences. Some others, like myself as a student, get overly enamored of extended block quotes and the scholarly look they give to the page. 7 These aren’t the worst sins of academic writing, but they get in the way of one of the key principles of writing with sources: shaping quotes and paraphrases efficiently. Efficiency follows from the second principle, because when you fully incorporate sources into your own explicit argument, you zero in on the phrases, passages, and ideas that are relevant to your points. It’s a very good sign for your paper when most quotes are short (key terms, phrases, or parts of sentences) and the longer quotes (whole sentences and passages) are clearly justified by the discussion in which they’re embedded. Every bit of every quote should feel indispensable to the paper. An overabundance of long quotes usually means that your own argument is undeveloped. The most incandescent quotes will not hide that fact from your professor.

Also, some student writers forget that quoting is not the only way to incorporate sources. Paraphasing and summarizing are sophisticated skills that are often more appropriate to use than direct quoting. The first two paragraphs of the example passage above do not include any quotations, even though they are both clearly focused on presenting the work of others. Student writers may avoid paraphrasing out of fear of plagiarizing, and it’s true that a poorly executed paraphrase will make it seem like the student writer is fraudulently claiming the wordsmithing work of others as his or her own. Sticking to direct quotes seems safer. However, it is worth your time to master paraphasing because it often helps you be more clear and concise, drawing out only those elements that are relevant to the thread of your analysis.

For example, here’s a passage from a hypothetical paper with a block quote that is fully relevant to the argument but, nevertheless, inefficient:

Drawing on a lifetime of research, Kahneman concludes our brains are prone to error: System 1 registers the cognitive ease with which it processes information, but it does not generate a warning signal when it becomes unreliable. Intuitive answers come to mind quickly and confidently, whether they originate from skills or from heuristics. There is no simple way for System 2 to distinguish between a skilled and a heuristic response. Its only recourse is to slow down and attempt to construct an answer on its own, which it is reluctant to do because it is indolent. Many suggestions of System 1 are casually endorsed with minimal checking, as in the bat-and-ball problem.

While people can get better at recognizing and avoiding these errors, Kahneman suggests, the more robust solutions involve developing procedures within organizations to promote careful, effortful thinking in making important decisions and judgments.

Even a passage that is important to reference and is well contextualized in the flow of the paper will be inefficient if it introduces terms and ideas that aren’t central to the analysis within the paper. Imagine, for example, that other parts of this hypothetical paper use Kahneman’s other terms for System 1 (fast thinking) and System 2 (slow thinking); the sudden encounter of “System 1” and “System 2” would be confusing and tedious for your reader. Similarly, the terms “heuristics” and “bat-and-ball problem” might be unfamiliar to your reader. Their presence in the block quote just muddies the waters. In this case, a paraphrase is a much better choice. Here’s an example passage that uses a paraphrase to establish the same points more clearly and efficiently:

Drawing on a lifetime of research, Kahneman summarizes that our brains are prone to error because they necessarily rely on cognitive shortcuts that may or may not yield valid judgments. 4 We have the capacity to stop and examine our assumptions, Kahneman points out, but we often want to avoid that hard work. As a result, we tend to accept our quick, intuitive responses. While people can get better at recognizing and avoiding these errors, Kahneman suggests that the more robust solutions involve developing procedures within organizations to promote careful, effortful thinking in making important decisions and judgments.

Not only is the paraphrased version shorter (97 words versus 151), it is clearer and more efficient because it highlights the key ideas, avoiding specific terms and examples that aren’t used in the rest of the paper. If other parts of your paper did refer to Kahneman’s System 1 and System 2, then you might choose to include some quoted phrases to make use of some of Kahneman’s great language. Perhaps something like this:

Drawing on a lifetime of research, Kahneman summarizes that our brains are prone to error because they necessarily rely on cognitive shortcuts that may or may not yield valid judgments. 5 System 1, Kahneman explains, “does not generate a warning signal when it becomes unreliable.” 6 System 2 can stop and examine these assumptions, but it usually wants to avoid that hard work. As a result, our quick, intuitive responses are “casually endorsed with minimal checking.” 7 While people can get better at recognizing and avoiding these errors, Kahneman suggests, the more robust solutions involve developing procedures within organizations to promote careful, effortful thinking in making important decisions and judgments.

Whether you choose a long quote, short quote, paraphrase or summary depends on the role that the source is playing in your analysis. The trick is to make deliberate, thoughtful decisions about how to incorporate ideas and words from others.

Paraphrasing, summarizing, and the mechanical conventions of quoting take a lot of practice to master. Numerous other resources (like those listed at the end of this chapter) explain these practices clearly and succinctly. Bookmark some good sources and refer to them as needed. If you suspect that you’re in a quoting rut, try out some new ways of incorporating sources.

Choose precise signal phrases

It’s time to get beyond the all-purpose “says.” And please don’t look up “says” in the thesaurus and substitute verbs like “proclaim” (unless there was actually a proclamation) or “pronounce” (unless there was actually a pronouncement). Here’s a list of 15 useful alternatives:

• Suggests (if the author is speculating or hypothesizing)
• Contests (disagrees)

More precise choices like these carry a lot more information than “says”, enabling you to relate more with fewer words. For one thing, they can quickly convey what kind of idea you’re citing: a speculative one (“postulates”)? A conclusive one (“determines”)? A controversial one (“counters”)? You can further show how you’re incorporating these sources into your own narrative. For example, if you write that an author “claims” something, you’re presenting yourself as fairly neutral about that claim. If you instead write that the author “shows” something, then you signal to your reader that you find that evidence more convincing. “Suggests” on the other hand is a much weaker endorsement.

1 Jennifer C. Lee, J.C. and Jeremy Staff, “When Work Matters: The Varying Impact of Work Intensity on High School Drop Out,” Sociology of Education 80, no. 2 (2007): 158-178.

2 Ibid. , 159.

3 It took me a long time to stop abusing block quotes. They made me feel like my paper was an unassailable fortress of citation! With the friendly but pointed feedback of my professors, I gradually came to see how they took too much space away from my own argument.

4 Kahneman, Thinking, Fast and Slow, 416-7.

7 Ibid, 416.

8 Ibid , 417.

9 Robert B. Marks, The Origins of the Modern World: A Global and Ecological Narrative from the Fifteenth to the Twenty-first Century (Lanham, MD: Rowman & Littlefield, 2007), 95.

Attribution

Adapted by Anna Mills from Writing in College: From Competence to Excellence by Amy Guptill, published by Open SUNY Textbooks , licensed CC BY NC SA 4.0 .

Want to create or adapt books like this? Learn more about how Pressbooks supports open publishing practices.

VI. Joining the Academic Conversation

6.3 Using Sources in Your Paper

John Lanning; Amanda Lloyd; Robin Jeffrey; Melanie Gagich; Terri Pantuso; Sarah LeMire; and Kalani Pattison

Academic writing requires the use of signal phrases to properly embed quoted material and document information. While basic signal phrases require the use of the author’s name and a strong verb, attribution tags emphasize different types of information related to the source in order to set up the quoted material and can help shape your reader’s response to the information presented. In grammatical terms, an attribution tag can be viewed as an appositive, an adjectival clause following a noun that modifies the noun and provides contextual information. In the following examples, the signal phrases (appositives) are italicized.

As Benjamin Franklin, one of the Founding Fathers , states, “Well done is better than well said.” [1]

The slogan “Just Do It” was highly successful for Nike, an athletic apparel company .

What you have to say is more important than the passage you are citing, so you want the information leading into your evidence/ support to work to your advantage. A basic signal phrase is a device used to smoothly integrate quotations and paraphrases into your essay and consists of an author’s name and an active verb indicating how the author is presenting the material. It is important for beginning academic writers to use signal phrases to clearly attribute textual evidence to an author and to avoid interrupting the flow of an essay.

Referring to the Author within a Signal Phrase

In most instances, a signal phrase should contain only the last name of the author or authors of the source text (as opposed to the author’s first and last name). APA style guidelines require no reference to a first name at any point in an essay and few if any gender specific pronouns. But in MLA papers, if you are referring to an author for the first time in your essay, you should include that author’s first name. Any future signal phrase should refer to the author by last name only or with a pronoun when it’s perfectly clear to whom the pronoun refers.

Ellen J. Langer observes, “For us to pay attention to something for any amount of time, the image must be varied” (39). [2]

Langer continues, “Thus, for students who have trouble paying attention the problem may be that they are following the wrong instructions” (39).

She then states, “To pay constant, fixed attention to a thought or an image may be a kind of oxymoron” (39).

Notice how each signal phrase verb is followed by a comma, which is then followed by one space before the opening quotation mark.

Varying Your Verbs

While it’s important to use signal phrase verbs, you’ll want to make sure that you vary them to avoid repetition (rather than simply using “states” throughout your entire essay for example) in order to maintain your readers’ interest and to indicate the author’s intended use of the excerpted material. See below for examples of strong signal phrase verbs.

Table 6.3.1: Strong Signal Phrase Verbs

Why use signal phrases and attributive tags.

While many students may see attributive tags as filler, they can provide the audience with valuable insight into how you, the writer, intend the quoted material to be read/viewed. In addition to setting up the source evidence, attribution tags can also be used as meaningful transitions moving your readers between your ideas and those of your support.

In most instances, the first time the author is mentioned in an MLA style essay, it is a good idea to provide an attributive tag as well as the author’s first and last name. When using APA style, list the author’s first initial and last name. Style will vary with studies including multiple authors.

While providing the author’s credentials and title of the source are the most common attributions used, there are others we should be aware of.

Types of Attributive Tags (attributive tag is underlined in each example)

Type : Author’s credentials are indicated.

Grace Chapmen, Curator of Human Health & Evolutionary Medicine at the Springfield Natural History Museum , explains…

Purpose: Presenting an author’s credentials should help build credibility for the passage you are about to present. Including the author’s credentials gives your readers a reason to consider your sources.

Type : Author’s lack of credentials is indicated.

Matthew Spencer, whose background is in marriage counseling, not foreign policy , claims…

Purpose: Identifying an author’s lack of credentials in a given area can help illustrate a lack of authority on the subject matter and persuade the audience not to adopt the author’s ideas. Pointing to an author’s lack of credentials can be beneficial when developing your response to counterarguments.

Type : Author’s social or political stance, if necessary to the content, is explained.

Ted Cruz, the Republican Senator from Texas , claims… Debbie Dingell, the Democrat representing Michigan’s 6th district , spoke today about…

Purpose: Explaining the author’s social or political stance can help a reader to understand why that author expresses a particular view. This understanding can positively or negatively influence an audience. Be careful to avoid engaging in logical fallacies such as loaded language or genetic fallacy .

Type : Publisher of the source is identified.

According to a recent Gallup poll…

Purpose: Identifying the publisher of the passage can help reinforce the credibility of the information presented and you can capitalize on the reputation/ credibility of the publisher of the source material.

Type : Title of the source is included.

In “ Understanding Human Behavior ,” Riley argues …

Purpose: Informs the reader where the cited passage is being pulled from.

Type : Information that establishes context is presented.

In a speech presented during a Free Speech rally , Elaine Wallace encourages …

Purpose: Presenting the context that the original information was presented can help the audience understand the author’s purpose more clearly

What are Direct Quotes?

Direct quotes are portions of a text taken word for word and placed inside of a work. Readers know when an author is using a direct quote because it is denoted by the use of quotation marks and an in-text citation. [3]

Example of Direct Quote

In his seminal work, David Bartholomae argues that “Every time a student sits down to write for us, he has to invent the university for the occasion-invent the university ”(4).

Direct quotes might also be formatted as a “block quote” which occurs if the borrowed language is longer than four (4) lines of text in MLA formatting, or more than 40 words in APA formatting. In MLA, A block quote requires the author to indent the borrowed language by 1/2 an inch, place the citation at the end of the block, and remove quotation marks.

Example of Block Quote

In his seminal work, David Bartholomae argues that

Every time a student sits down to write for us, he has to invent the university for the occasion-invent the university, that is, or a branch of it, like History or Anthropology or Economics or English. He has to learn to speak our language, to speak as we do, to try on the peculiar ways of knowing, selecting, evaluating, reporting, concluding, and arguing that define the discourse of our community. (4).

Be careful when using direct quotes because failing to write the text exactly as it appears in the original is not an ethical use of direct quotes. Also, failing to bracket the quote with quotation marks and/or cite it inside the text is also unethical. Both mistakes are a form of plagiarism.

When Should I Use Direct Quotes?

Generally speaking, direct quotes should be used sparingly because you want to rely on your own understanding of material and avoid over-relying on another’s words. You want your voice to be the dominant one in an argument. Over quoting does not reinforce your credibility as an author; however, according to the Purdue Online Writing Lab (OWL) you should use direct quotes when “the author you are quoting has coined a term unique to her or his research and relevant within your own paper.” [4]

The Basics of Directly Quoting

• All quoted material should be enclosed in quotation marks to set it off from the rest of the text. The exception to this is block quotes which require different formatting.
• Quoted material should be an accurate word-for-word reproduction from the author’s original text. You cannot alter any wording or any spelling. If you must do so, you must use a bracket or an ellipsis.
• A clear signal phrase/attribution tag should precede each quotation.
• A parenthetical citation should follow each quotation. For information about formatting parenthetical citations, see section 8.6 (APA) and 8.7 (MLA) .

The Hard Part of Directly Quoting: Integrating Quotes into Your Writing

As the author of your essay, you should explain the significance of each quotation to your reader. This goes far beyond simply including a signal phrase. Explaining the significance means indicating how the quoted material supports the point you are making in that paragraph. Remember, just because you add a quote does not mean that you have made your point. Quotes never speak for themselves. When quoting material, ask yourself how and why does that quoted material make the point you think it does? Then, follow the quote with a sentence that adds clarity for your insertion of that quoted material. Table 6.3.2 contains some helpful phrases for explaining quoted materials where “X” represents the author’s last name.

Table 6.3.2. Phrases for Explaining Quoted Material

Sometimes, in order to smoothly integrate quoted material into your paper, you may need to remove a word or add a word to make the quote make sense. If you make any change to quoted material, it must be formatted correctly using an ellipsis or brackets. In the following, a portion of Hamlet’s “To Be, or Not To Be” soliloquy is used as the exemplar:

Original quote: “To be, or not to be, that is the question”

• Use brackets [these are brackets] to change a word or add additional information.

As Hamlet states, “To be, or not to be, that is the [essential] question.”

• Use an ellipsis (this is an ellipsis…) to indicate omissions in the middle of a quote, not at the beginning or ending of quoted material.

As Hamlet states, “To be, or not to be … is the question.”

When in doubt, strive to allow your voice – not a quote from a source – to begin each paragraph, precede each quote, follow each quote, and end each paragraph. Quotes that are integrated well into a paper allow you to control the paper. That is what a reader wants to see: your ideas and the way that you engage sources to shape and discuss your ideas.

Paraphrasing and Summarizing

While quoting may be the first thing that many people think of when they think about integrating sources, paraphrasing, summarizing, and citing data are also ways to incorporate information from outside materials into your essays or projects.

Paraphrasing

• Paraphrases allow you to describe specific information from a source (ideas from a paragraph or several consecutive paragraphs) in your own words .
• Paraphrases are like translations of an author’ original idea. You retain the detail of the original thought, but you express it in your own way.
• Paraphrases of the text should be expressed in your own words, with your own sentence structure, in your own way. You should not simply “word swap”; that is, replace a few words from the original with synonyms.
• If you must use a few of the author’s words within your paraphrase, they must have quotation marks around them.
• Paraphrases often include attributive tags (or signal phrases) to let your readers know where the paraphrased material begins.
• Paraphrased material should be followed by a parenthetical citation. For information about formatting parenthetical citations, see section 8.6 (APA) and 8.7 (MLA) .
• As with a quote, you need to explain to your reader why the paraphrased material is significant to the point you are making in your paper.

Summarizing

• Summaries allow you to describe general ideas from a source. You do not express detailed information as you would with a paraphrase.
• Summaries are shorter than the original text.
• Any summaries of the text should not include direct wording from the original source. All text should be in your words, though the ideas are those of the original author.
• A signal phrase should let your readers know where the summarized material begins. Depending on what information you include in your signal phrase, you may still need to include a parenthetical citation. For information about formatting parenthetical citations, see section 8.6 (APA) and 8.7 (MLA) .
• If you are offering a general summary of an entire article, there is no need to cite a specific page number.

Referring to AI-Generated Content

In some ways, incorporating information from an AI source looks very similar to incorporating information from any other type of source.

• Are you using language directly from the AI source? Quotations from AI sources need quotation marks and parenthetical citations just like quotations from other sources.
• Are you paraphrasing or summarizing information from an AI source? Although you don’t need quotation marks, you still need parenthetical citations.

For more information on citing AI sources, see sections 8.6 (APA) and 8.7 (MLA) .

Unlike other sources, however, a reader can’t easily look up an AI source and read the original text. For that reason, it is important to be very transparent about where your information comes from. Both APA and MLA recommend including your prompt as part of your attribution. APA recommends introducing the prompt within the body of your text, [5] while MLA includes that information within the citation itself. [6]

Example (APA format)

ChatGPT, when prompted to answer “What are the ethical implications of the death penalty,” responded that the “disproportionate impact of capital punishment on people of color” is an important factor (OpenAI, 2023).

Example (MLA format)

When considering the ethical implications of the death penalty, the “disproportionate impact of capital punishment on people of color” is an important factor (“What are the ethical implications”).

In order to maximize transparency, you may wish to include the full transcript of your AI-generated content in an appendix to your paper. [7] In this way, you can ensure that your source remains fully accessible to you and to your audience. Such transparency and documentation also greatly contributes to maintaining your ethos — your credibility.

Practice Activity

This section contains material from:

Lanning, John, and Amanda Lloyd. “Signal Phrases.” In A Guide to Rhetoric, Genre, and Success in First-Year Writing , by Melanie Gagich and Emilie Zickel. Cleveland: MSL Academic Endeavors. Accessed July 2019. https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/apa-signal-phrases/ . Licensed under a Creative Commons Attribution 4.0 International License . Archival link: https://web.archive.org/web/20201027005526/https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/apa-signal-phrases/

Gagich, Melanie. “Quoting.” In A Guide to Rhetoric, Genre, and Success in First-Year Writing , by Melanie Gagich and Emilie Zickel. Cleveland: MSL Academic Endeavors. Accessed July 2019. https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/quoting-paraphrasing-and-summarizing-to-avoid-plagiarism/ . Licensed under a Creative Commons Attribution 4.0 International License . Archival link: https://web.archive.org/web/20201027012338/https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/quoting-paraphrasing-and-summarizing-to-avoid-plagiarism/

Jeffrey, Robin. “Paraphrasing and Summarizing.” In A Guide to Rhetoric, Genre, and Success in First-Year Writing , by Melanie Gagich and Emilie Zickel. Cleveland: MSL Academic Endeavors. Accessed July 2019. https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/9-3-paraphrasing-summarizing-and-integrating-data/ . Licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License . Archival link: https://web.archive.org/web/20201027011135/https://pressbooks.ulib.csuohio.edu/csu-fyw-rhetoric/chapter/9-3-paraphrasing-summarizing-and-integrating-data/

OER credited in the texts above includes:

Jeffrey, Robin. About Writing: A Guide . Portland, OR: Open Oregon Educational Resources. Accessed December 18, 2020. https://openoregon.pressbooks.pub/aboutwriting/ . Licensed under a Creative Commons Attribution 4.0 International License . Archival link: https://web.archive.org/web/20230711210756/https://openoregon.pressbooks.pub/aboutwriting/

• Benjamin Franklin, Poor Richard, 1737, Founders Online, National Archives, accessed December 18, 2020, https://founders.archives.gov/documents/Franklin/01-02-02-0028 . ↵
• Ellen J. Langer, The Power of Mindful Learning (Boston: Da Capo Press, 1997). ↵
• The following examples come from: David Bartholomae, “Inventing the University,” Journal of Basic Writing 5, no. 3 (1986): 4-23. ↵
• “How to Use Quotation Marks,” Purdue Online Writing Lab , accessed May 8, 2020, https://owl.purdue.edu/owl/general_writing/punctuation/quotation_marks/index.html . ↵
• Tim McAdoo, "How to Cite ChatGPT," APA Style, accessed August 17, 2023, https://apastyle.apa.org/blog/how-to-cite-chatgpt ↵
• "How Do I Cite Generative AI in MLA Style?," MLA Style Center, accessed August 17, 2023, https://style.mla.org/citing-generative-ai/ ↵

Loaded language is related to the fallacy of "begging the question." Sometimes begging the question includes using "loaded language" in which word choice has strong connotations or extra meanings that attempt to sway the audience.

This fallacy claims that an idea or fact or argument is incorrect because the speaker is someone one usually disagrees with -- that the idea is wrong because of the origin of the idea or argument, rather than because it is incorrect or invalid on its own merits. For instance, just because you generally disagree with a politician's policies or ideas doesn't mean that every single thing they say is incorrect or invalid just because it was them who said it. (The idiom often used is "even a stopped clock is right twice a day).

An ambiguous or amorphous quality to writing comprising the vocabulary, word choice, tone, point of view, syntax, attitude, emotion, and style of a writer. Because writing is a personal and individual exercise, every writer has their own unique voice.

6.3 Using Sources in Your Paper Copyright © 2023 by John Lanning; Amanda Lloyd; Robin Jeffrey; Melanie Gagich; Terri Pantuso; Sarah LeMire; and Kalani Pattison is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License , except where otherwise noted.

Using Sources

Paraphrasing.

So why paraphrase?  Paraphrasing offers a way to maintain your own writing style and voice throughout the writing.  It helps cut down on the number of different styles from different sources, creating a sleeker, easier reading experience for your reader.  Most of all, though, paraphrasing is a means of helping you understand what your sources are saying, in order to incorporate that information into your own writing.  You have to understand the source’s ideas fully in order to rewrite them clearly.

When you paraphrase, make sure not to simply substitute one word for another, retaining the same sentence structure.  Paraphrasing requires you to use your own sentence structures as well as words, so that you are not inadvertently plagiarizing the source.

• There is no good reason to use a quote to refer to your evidence. If the author’s exact words are not especially important to the point you are trying to make, you are usually better off paraphrasing the evidence.
• You are trying to explain technical information or complicated language to a more general reading audience.
• You are trying to explain a particular a piece of evidence in order to explain or interpret it in more detail. This might be particularly true in writing projects like critiques.
• You need to balance a direct quote in your writing. You need to be careful about directly quoting your research too much because it can sometimes make for awkward and difficult to read prose. So, one of the reasons to use a paraphrase instead of a quote is to create balance within your writing.

Writing a Paraphrase

Make sure that you understand the original text that you intend to paraphrase.  Rewrite that text at least twice, in your own words.  After the first rewriting, set the paraphrase aside for a short time.  When you go back to it, you’ll most likely see that you’ve tended to retain some of the original text’s wording and sentence structure.  On a second (or third, or fourth) rewriting, try to make the language and sentence structure your own, while retaining the meaning of the original text. If you find that the original text uses a key word or phrase that you don’t want to rewrite, know that you can always include it in quotation marks within your paraphrase.  Finally, make sure to attribute the paraphrase at the start (e.g., “According to…”) and include a citation at the end.  Your readers should be able to distinguish your own information from paraphrased information, and the attribution and citation signal the beginning and end of the paraphrase.

Paraphrasing Example

Mr. President, I confess that I do not entirely approve of this Constitution at present; but, Sir, I am not sure I shall never approve it; for, having lived long, I have experienced many instances of being obliged, by better information or fuller consideration, to change my opinions even on important subjects, which I once thought right, but found to be otherwise.

Benjamin Franklin tells the president of the Constitutional Convention that he does  not entirely  approve of the Constitution at the  present  time, but that he is not sure  he will never approve it . He points out that he has lived a long time, and in his experience there have been  many instances  when  better information of fuller consideration  of a topic have made him change his opinions  on important subjects  that he had originally thought to be correct. He points out that he finds himself more likely to doubt his own judgment the older he gets, and contrasts his knowledge of his own fallibility with other people’s conviction of their infallibility.

Benjamin Franklin tells the president of the Constitutional Convention that although he is currently uncertain about the Constitution they have created, he may eventually acknowledge its effectiveness. This is due, he explains, to new information or a different understanding of similarly important topics that have caused him to change his mind in the past.

Paraphrase Checklist

• Have you used your own words and sentence structures?
• Even though the wording is your own, have you carefully retained the meaning of the original text
• Did you attribute the paraphrase at the start, using language in some way that explains that you’re paraphrasing another’s text? (e.g., “Smith states that…”)
• Did you cite the paraphrase correctly at the end, using a standard citation format for in-text citations?
• Did you cite the paraphrased source in the Works Cited list at the end of the essay?

Paraphrasing Practice #1

Paraphrasing is a skill that takes time to develop. One way of becoming familiar with paraphrasing is by examining successful and unsuccessful attempts at paraphrasing. Read the quote below from page 179 of Howard Gardner’s book titled Multiple Intelligences and then examine the two attempts at paraphrasing that follow [1] .

Paraphrasing Attempt 1: America has now gone too far toward formal testing, without realizing the costs and limitations of exclusively emphasizing that approach (Gardner 179).

Paraphrasing Attempt 2: In the United States, the education system places too much emphasis on formal testing, overlooking the limitations and expenses imposed when that assessment strategy is employed exclusively (Gardner 179).

Paraphrasing practice #2

The original passage:

“Scientists and policymakers generally agree that the likelihood of flooding in the UK will increase as a result of climate change. It is also accepted that sensible land use and development planning plays a role in the management of flood risk, while allowing necessary development to continue” (Dept…Government, 2006, Evans et al., 2004; Thorne et al., 2007).

The paraphrase:

Scientists and policymakers agree that climate change means that the likelihood of UK flooding will increase. It is also agreed that the role of sensible land use and development planning are important in the management of flood risk, also allowing necessary development to continue (Dept…Government, 2006; Evans et al., 2004; Thorne et al., 2007).

Answer these questions about the paraphrase example from above to decide if it is a good or bad paraphrase.

1. Has the student changed a lot of the words from the original passage?

2. Has the student changed the word order and structure compared to the original passage?

3. Has the student included a citation for the information?

4. Overall, do you think this is a good paraphrase?

View the Writing with Sources: Paraphrasing & Quotation , an interactive resource for a review and self-test on paraphrasing and quoting sources.

• Gardner, Howard. Multiple Intelligences: New Horizons in Theory and Practice. BasicBooks, 2006. ↵
• Paraphrasing. Revision and adaptation of the page Using Sources in Your Writing at https://courses.lumenlearning.com/wm-englishcomposition1/chapter/text-using-sources-in-your-writing/, the page Paraphrasing at https://courses.lumenlearning.com/wm-englishcomposition1/chapter/text-paraphrasing/, and the page Using Sources in Your Writing at https://courses.lumenlearning.com/introtocollegecomp/chapter/text-using-sources-in-your-writing/ which are revisions and adptations of the sources listed below. Authored by : Susan Oaks. Provided by : Empire State College, SUNY OER Services. Project : College Writing. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Using Sources in Your Writing. Provided by : Lumen Learning. Located at : https://courses.lumenlearning.com/wm-englishcomposition1/chapter/text-using-sources-in-your-writing/ . Project : English Composition I. License : CC BY-SA: Attribution-ShareAlike
• Paraphrasing. Provided by : Lumen Learning. Located at : https://courses.lumenlearning.com/wm-englishcomposition1/chapter/text-paraphrasing/ . Project : English Composition I. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Using Sources in Your Writing . Provided by : Lumen Learning. Located at : https://courses.lumenlearning.com/introtocollegecomp/chapter/text-using-sources-in-your-writing/ . Project : Introduction to College Composition. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Information on Quoting and Paraphrasing from the Academic Integrity Tutorial. Authored by : DiMenna-Nyselius Library. Provided by : Fairfield University. Located at : http://librarybestbets.fairfield.edu/c.php?g=476878&p=3335282 . License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Paraphrasing. Provided by : Texas A&M University Writing Center. Located at : http://writingcenter.tamu.edu/Students/Handouts-Guides/Handouts-(Get-It-Written)/Citing/Paraphrasing . Project : Grounds for Argument. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Using Information Ethically, Avoiding Plagiarism Paraphrasing Example. Provided by : Loyola Marymount University, William H. Hannon Library. Located at : http://electra.lmu.edu/LGRL/UIE2014/ . Project : Lion's Guide to Research and the Library. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Quoting Examples from Paraphrasing, and Avoiding Plagiarism. Authored by : Steven D. Krause. Located at : http://www.stevendkrause.com/tprw/chapter3.html . Project : The Process of Research Writing. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• Benjamin Franklin example from Chapter 5: Using Materials from Sources. Authored by : Denise Snee, Kristin Houlton, Nancy Heckel. Edited by Kimberly Jacobs. Located at : http://lgdata.s3-website-us-east-1.amazonaws.com/docs/679/734444/Snee_2012_Research_Analysis_and_Writing.pdf . Project : Research, Analysis, and Writing. License : CC BY-NC-SA: Attribution-NonCommercial-ShareAlike
• image of man writing on notepad, with open laptop on desk. Authored by : StartupStock Photos. Provided by : Pixabay. Located at : https://pixabay.com/en/write-plan-desk-notes-pen-writing-593333/ . License : CC0: No Rights Reserved
• interactive video Writing with Sources: Paraphrasing & Quotations. Authored by : Kelsey Foote, Brett Sherman, Dan McCrea. Provided by : SUNY Empire State College. Located at : https://escoer.sunyempirefaculty.net/iitg_2017/iitg_paraphrasing/story_html5.html . License : CC BY-NC: Attribution-NonCommercial

Want to create or adapt books like this? Learn more about how Pressbooks supports open publishing practices.

3.4: Using Source Text: Quoting, Paraphrasing, and Summarizing

Learning objectives.

Once you have a collection of credible sources as part of a formal secondary research project such as a report, your next step is to build that report around those sources, using them as anchors of evidence around your own arguments. If you began with an hypothesis and you’re using the sources as evidence to support it, or if you realize that your hypothesis is wrong because all the credible sources you’ve found poked holes in it, you should at this point be able to draft a thesis—your whole point in a nutshell. From there, you can arrange your sources in an order that follows a logical sequence such as general to specific or advantages versus disadvantages. We will examine organizational structures in the next chapter ( Ch. 4 ) on drafting, but we are now going to focus on how to incorporate source material into usable evidence.

You essentially have four ways of using source material available to you, three of them involving text, and one media:

• Quoting text: copying the source’s exact words and marking them off with quotation marks
• Paraphrasing text: representing the source’s ideas in your own words (without quotation marks)
• Summarizing text: representing the source’s main ideas in your own words (without quotation marks)
• Reproducing media: embedding pictures, videos, audio, graphic elements, etc. into your document

In each case, acknowledging your source with a citation at the point of use and follow-up bibliographical reference at the end of your document (see §3.5 below) is essential to avoid a charge of plagiarism. Let’s now look at each of these in turn.

3.4.1: Quoting Sources

3.4.2: paraphrasing sources, 3.4.3: summarizing sources.

Quoting is the easiest way to use sources in a research document, but it also requires care in using it properly so that you don’t accidentally plagiarize, misquote, or overquote. At its simplest, quoting takes source text exactly as it is and puts quotation marks (“ ”) around that text to set it off from your own words. The following points represent conventions and best practices when quoting:

• You may have seen single quotation marks and think that they’re also acceptable to use, but that’s only true in the UK and some other Commonwealth countries, not in Canada; some European countries use << >> to set off quotations instead.
• Also use double quotation marks for putting a single word or two in “scare quotes” when you’re drawing attention to how people use certain words and phrases—again, not single quotation marks since there is no such thing as quotation marks “lite.”
• Use single quotation marks only for reported speech when you have a quotation within a quotation, as in, “The minister responded to say, ‘No comment at this time’ regarding the allegations of wrongdoing.”
• If no parenthetical citation follows immediately after the closing quotation marks, the sentence-ending period falls to the left of those closing quotation marks (between the final letter and the “99”); a common mistake is to place the period to the right of the closing quotation marks ( . . . wrongdoing”.).
• According to researchers Tblisky and Darion (2003), “. . .”
• As Vice President of Operations Rhonda Rendell has noted, “. . .”
• John Rucker, the first responder who pulled Mr. Warren from the wreckage, said that “. . .”
• Spokespersons Gloria and Tom Grady clarified the new regulations: “. . .”
• “. . . ,” confirmed the minister responsible for the initiative.
• “. . . ,” writes Eva Hess, “. . .”
• Quote purposefully: Quote only when the original wording is important. When we quote famous thinkers like Albert Einstein or Marshall McLuhan, we use their exact words because no one could say it better or more interestingly than they did. Also quote when you want your audience to see wording exactly as it appeared in the source text or as it was said in speech so that they can be sure that you’re not distorting the words as you might if you paraphrased instead. But if there’s nothing special about the original wording, then you’re better to paraphrase properly (see §3.4.2 below) than to quote.

Students frequently overuse direct quotation [when] taking notes, and as a result they overuse quotations in the final [research] paper. Probably only about 10% of your final manuscript should appear as directly quoted matter. Therefore, you should strive to limit the amount of exact transcribing of source materials while taking notes. (Lester, 1976, pp. 46-47)

• Don’t overquote: As the above source says, a good rule of thumb is that your completed document should contain no more than 10% quoted material. Much above that will look lazy because it appears that you’re getting quotation to write your document for you. Quote no more than a sentence or two at a time if you quote at all.
• To avoid introducing spelling mistakes or other transcription errors, best practice (if your source is electronic) is to highlight the text you want to quote, copy it (ctrl. + c), and paste it (ctrl. + v) into your document so that it matches the formatting of the rest of your document (i.e., with the same font type, size, etc.). To match the formatting, use the Paste Options drop-down menu that appears beside pasted text as soon as you drop it in and disappears as soon as you perform any operation other than clicking on the drop-down menu.
• Though many people mistakenly refer to parentheses ( ) as “brackets,” brackets are squared [ ] and are used mainly to indicate changes to quoted words, whereas parentheses follow the quotation and mark off the citation. If you were to clarify and streamline the final sentence of the block quotation a few points above, for instance, you could say something like: Lester (1976) recommended “limit[ing] the amount of exact transcribing . . . while taking notes” (p. 47). Here, the verb “limit” in the source text needs to be converted into its participle form (having an -ing ending) to follow the past-tense verb in the sentence framing the quotation grammatically. Sneakily adding the “ing” to “limit” without using brackets would be misquotation because “limiting” appears nowhere in the original.
• Notice that the ellipsis above is three spaced periods (not three stuck together, as in “…”) and that one doesn’t appear at the beginning of the quotation to represent the words in the original prior to “limit” nor at the end to represent source text following the quoted words (“… limit …”). Use the ellipsis only to show that you’re skipping over unnecessary words within a quotation.
• Be careful not to use brackets and ellipses in a way that distorts or obscures the meaning of the original text. For instance, omitting “Probably” and changing “should” to “[can]” in the Lester quotation above will turn his soft guideline into a hard rule, which are not the same.
• When you said in the class discussion forum, “No one cares about grammer, [ sic ] it dont [ sic ] really matter,” you tend to undermine your credibility on the topic with poor spelling and a comma splice.
• Capitalize as in the original, even if it seems strange to start a quotation with a capital (because it was the first word in the original) though it’s no longer the first word because it follows a signal phrase in your sentence. See the example in the point above, for instance.
• Quotation is a powerful tool in the arsenal of any writer needing to support a point with evidence. Capturing the source’s words exactly as they were written or spoken is an honest way of presenting research. For more on quotation, consult Purdue OWL ’s series of modules starting with the How to Use Quotation Marks page and ending with their Quotation Mark Exercise and Answers (2024).

Lester, J. D. (1976). Writing research papers: A complete guide (2nd ed.). Scott, Foresman.

Miller-Wilson, K. (2020, May 27). Examples of signal phrases and how to use them . YourDictionary. https://www.yourdictionary.com/articles/examples-signal-phrases

Purdue OWL. (2024). How to use quotation marks . Purdue University. https://owl.purdue.edu/owl/general_writing/punctuation/quotation_marks/index.html

Paraphrasing or “indirect quotation” is putting source text in your own words and altering the sentence structure to avoid using the quotation marks required in direct quotation. Paraphrasing is the preferred way of using a source when the original wording isn’t important. This way, you can incorporate the source’s ideas so they’re stylistically consistent with the rest of your document and thus better tailored to the needs of your audience (presuming the original was tailored for a different audience with different needs). Also, paraphrasing a source into your own words proves your advanced understanding of the source text.

A paraphrase must faithfully represent the source text by containing the same ideas as in the original in about the same length. As a matter of good writing, however, you should try to streamline your paraphrase so that it tallies fewer words than the source passage while still preserving the original meaning. An accurate paraphrase of the Lester (1976) passage block-quoted in the section above, for instance, can reduce a five-line passage to three lines without losing or distorting any of the original points:

Lester (1976) advises against exceeding 10% quotation in your written work. Since students writing research reports often quote excessively because of copy-cut-and-paste note-taking, try to minimize using sources word for word (pp. 46-47).

Notice that using a few isolated words from the original (“research,” “students,” “10%”) is fine, but also that this paraphrase doesn’t repeat any two-word sequence from the original because it changes the order of the content details along with most of the words. Properly paraphrasing without distorting, slanting, adding to, or deleting ideas from the source passage takes skill. The stylistic versatility required to paraphrase can be especially challenging to EAL learners and native English users whose general writing skills are still developing.

A common mistake that students make when paraphrasing is to go only part way towards paraphrasing by substituting-out major words (nouns, verbs, and adjectives) here and there while leaving the source passage’s basic sentence structure intact. This inevitably leaves strings of words from the original untouched in the “paraphrased” version, which can be dangerous because including such direct quotation without quotation marks will be caught by the plagiarism-detecting software that college instructors use these days. Consider, for instance, the following botched attempt at a paraphrase of the Lester (1976) passage that substitutes words selectively (lazily):

Students often overuse quotations when taking notes, and thus overuse them in research reports. About 10% of your final paper should be direct quotation. You should thus attempt to reduce the exact copying of source materials while note taking (pp. 46-47).

Let’s look at the same botched paraphrasing attempt, but set it under the original Lester passage given above in §3.4.1 (in the point advising to “Block-quote sparingly if at all”) so that we can compare the two. We’ll colour the unchanged words red in the botched paraphrase to see exactly how unsuccessful the paraphraser was in rephrasing the original passage in their own words (given in blue):

As you can see, several strings of words from the original are left untouched because the writer didn’t go the distance in changing the structure of the original; the order in which the information is presented is the same, which means several connected words are left unchanged. The Originality Report from plagiarism-catching software such as Turnitin would indicate that the passage is 64% plagiarized because it retains 25 of the original words (out of 39 in this “paraphrase”) but without quotation marks around them. Correcting this by simply adding quotation marks around passages like “when taking notes, and” would be unacceptable because those words aren’t important enough on their own to warrant direct quotation. The fix would just be to paraphrase more thoroughly by altering the words and the order of information, as shown in the paraphrase a few paragraphs above, the block-quote that begins “Lester (1976) advises . . . .” But how do you go about doing this?

Paraphrase easily by breaking down the task into these seven steps:

• Read and re-read the source-text passage so that you thoroughly understand each point it makes. If it’s a long passage, you might want to break it up into digestible chunks. If you’re unsure of the meaning of any of the words, look them up in a dictionary; you can even just type the word into the Google search bar, hit Enter , and a definition will appear, along with results of other online dictionary pages that define the same word.
• Look away and get your mind off the target passage. Process some different information for a while (e.g., a few minutes of gaming or social media—but just a few!)
• Without looking back at the source text, repeat its main points as you understood them—not from memorizing the exact words, but as you would explain the same ideas in different words out loud to a friend.
• Still without looking back at the source text, jot down that spoken wording and tailor the language so that it’s stylistically appropriate for your audience; edit and proofread your written version to make it grammatically correct in a way that perhaps your spoken-word version wasn’t.
• Now compare your written paraphrase version to the original to ensure that:
• Deleting any of the original points
• Adding any points of your own
• Distorting any of the ideas so they mean something substantially different from those in the original, or even take on a different character because you use words that, say, put a positive spin on something neutral or negative in the original
• You haven’t repeated any two identical words from the original in a row
• If any two words from the original remain, go further in changing those expressions by using a thesaurus in combination with a dictionary. When you enter a word into a thesaurus, it gives you a list of synonyms, which are different words that mean the same thing as the word you enter into it.
• For instance, the noun party can mean a group that is involved in something serious (e.g., a third-party software company in a data-collection process), but the verb party means something you do on a wild Saturday night out with friends; it can also function as an adjective related to the verb (e.g., party trick , meaning a trick performed at a party).
• Whenever you see synonymous words listed in a thesaurus and they look like something you want to use but you don’t know what they mean exactly, always look them up to ensure that they mean what you hope they mean; if not, move on to the next synonym until you find one that captures the meaning you intend. Doing this can save your reader the confusion and you the embarrassment of obvious thesaurus-driven diction problems (poor word choices).
• Cite your source. Just because you didn’t put quotation marks around the words doesn’t mean that you don’t have to cite your source. For more on citing, see §3.5.2 below).

For more on paraphrasing, consult the Purdue OWL Paraphrase learning module, Exercise , and Possible Answers .

Purdue OWL. (2024). Paraphrase: Write it in your own words. Purdue University. https://owl.purdue.edu/owl/research_and_citation/using_research/quoting_paraphrasing_and_summarizing/paraphrasing.html

Summarizing is one of the most important skills in communications because professionals of every kind must explain to non-expert customers, managers, and even co-workers the complex concepts on which they are experts, but in a way that those non-experts can understand. Adapting the message to such audiences requires brevity but also translating jargon-heavy technical details into plain, accessible language.

Summarizing is thus paraphrasing only the highlights of a source text or speech. Like paraphrasing, a summary is indirect quotation that re-casts the source in your own words; unlike a paraphrase, however, a summary is a fraction of the source length—anywhere from less than 1% to a quarter depending on the source length and length of the summary required. A summary can reduce a whole novel or film to a single-sentence blurb, for instance, or it could reduce a 50-word paragraph to a 15-word sentence. It can be as casual as a spoken run-down of a meeting your colleague was absent from and wanted to know what they missed, or an elevator pitch selling a project idea to a manager. It can also be as formal as a memo report on a conference you attended on behalf of your organization so your colleagues there can learn in a few minutes of reading the highlights of what you learned in a few days of attending the conference, saving them time and money.

Now, you may be wondering: “Summarizing is what generative AI like ChatGPT does especially well, so why do I have to bother learning how to do it myself? I ’ m just going to get AI to summarize things for me.” There are two problems with this objection, however. First, both paraphrasing and summarizing are necessary steps in the learning process. If you ’ re going to become an expert at anything, you must prove you know your subject by explaining it to someone assessing your true understanding, not just your ability to repeat things you memorized. Re-phrasing ideas in your own terms is how you prove true understanding, just like you would if you accurately translated what’s said in one language into another. Second, you won’t have time to get AI to summarize things for you when your job requires you to do this yourself on the spot with a message tailored to the needs of a particular audience. There’s nothing more frustrating than dealing with a chatbot that doesn’t seem to understand how to deliver exactly the information you require. This is why customers prefer dealing with knowledgeable human experts who quickly assess their needs and deliver on them. If your plan is to be a go-between for such customers and AI, their frustrations will be directed at you rather than at the technology, and you will make yourself easily disposable. Given how often you must prove your expertise on the job by summarizing what you know, it’s worth your while to get good at this essential skill.

The procedure for summarizing text to prove you understand its most important content is much like that of paraphrasing except it involves the extra step of pulling out highlights from the source. Altogether, this can be done in six steps, one of which includes the seven steps of paraphrasing, making this a twelve-step procedure:

• Determine how big your summary should be (according to your audience’s needs) so that you have a sense of how much material you should collect from the source.
• Read and re-read the source text so that you thoroughly understand it.
• Disregard detail such as supporting evidence and examples.
• If you have an electronic copy of the source, copy and paste the main points into your notes; for a print source that you can mark up, use a highlighter then transcribe those main points into your electronic notes.
• How many points you collect depends on how big your summary should be (according to audience needs).
• Paraphrase those main points following the seven-step procedure for paraphrasing outlined in §3.4.2 above.
• Edit your draft to make it coherent, clear, and especially concise.
• Ensure that your summary meets the needs of your audience and that your source is cited. Again, not having quotation marks around words doesn’t mean that you are off the hook for documenting your source(s).

Building a research assignment around a collection of summarized, paraphrased, and quoted passages from credible sources requires good organizational skills. We’ll focus more on this next step of the drafting process in the following chapter ( Ch. 4 ) , but basically it involves arranging your integrated research material in a coherent fashion, with main points up front and supporting points below proceeding in a logical sequence towards a convincing conclusion. Throughout this chapter, however, we’ve frequently encountered the requirement to document sources by citing and referencing, as in the last steps of both summarizing and paraphrasing indicated above. After reinforcing our quoting, paraphrasing, and summarizing skills, we can turn our focus on how to document sources.

Key Takeaway

• If you’ve already pulled out the main points as part of the previous exercise, practice including them as properly punctuated quotations in your document with smooth signal phrases introducing them.
• Paraphrase those same main-point sentences following the seven-step procedure outlined in §3.4.2 above. In other words, if Exercise 1 above was direct quotation, now try indirect quotation for each passage.
• Following the six-step procedure outlined in §3.4.3 above, summarize the entire source article, webpage, or whatever document you chose by reducing it to a single coherent paragraph of no more than 10 lines on your page.

Communication at Work Copyright © 2019 by Jordan Smith is licensed under a Creative Commons Attribution 4.0 International License , except where otherwise noted.

Share This Book

Want to create or adapt books like this? Learn more about how Pressbooks supports open publishing practices.

Write … and More

21 Citation vs. Attribution

Even though they share characteristics, citations and attributions play different roles and appear in different places. This chapter defines citation and attribution, explains how and when they should be used in an open textbook, and discusses their purposes, similarities, and differences.

Outside ideas and information provide evidence that build an argument or lay the foundation for a textbook’s topic. A strong textbook will appropriately reference these sources, showing the student reader where information and ideas that do not originate with the open textbook author come from. This should be done for both restricted and open works through citations and attribution statements. Use this as an opportunity to show students by example how a scholar respects and shares information from other sources.

A citation allows authors to provide the source of any quotations, ideas, and information that they include in their own work based on the copyrighted works of other authors. The Oxford Living Dictionary defines it as a “…quotation from or reference to a book, paper, or author, especially in a scholarly work.” [1]   To exclude citation of a resource referenced in your own work places you at risk of plagiarizing . Plagiarizing is the act of passing another individual’s ideas or work as your own. (Also see Concerns About Plagiarism .)

Citation is a common and long-time practice among scholars used to indicate where a resource is from and who the author is. Unlike an attribution, citation is typically used for copyrighted works with restricted rights or “all rights reserved.” In other words, it is used in works for which broad permissions have not been granted.

As a scholar and potential author of an open textbook, we assume that you are familiar with the rules around citation. However, the article  Warning: When You Must Cite from the Yale Center for Teaching and Learning provides some guidance about how, what, and the amount of a work that can be cited. (See  Textbook Citation .)

Attribution

Attribution is the cornerstone condition when using a resource or text released with an open-copyright licence. This legal requirement states that users must attribute — give credit — to the creator of the work. (See Copyright and Open Licences .)

In a CC BY licence, the “CC” stands for “Creative Commons” and the “BY” stands for “Attribution,” or who the work is “by.”

An attribution statement  is used to provide credit to the original creator; its purpose is similar to a citation. Best practice says that the statement should include the title of the work, name of the creator, and licence type (with links to each). When using text from another open educational resource, be clear in your attribution statement what section of your textbook contains this information.

A useful tool to help create attribution statements is the OPEN Attribution Builder by Open Washington. (See also Resources: Captions and Attributions .)

Differences

Citation and attribution serve different purposes.

• Citation is used for academic reasons in order to give credit to a colleague for their work as part of academic integrity. It’s also used for legal reasons. Attributing an open work fulfills the legal requirement of the open-copyright licence, which requires you to give credit to the creator of the work.
• Citation is used for restricted works where the copyright holder does not share the rights of the copy with the general public. The opposite is true for cases where attribution is used.
• Citation legally protects an author who wants to refer to someone else’s work and to avoid plagiarism and copyright infringement. The author of an open work has given advanced permission for others to use their work. (See Concerns About Plagiarism and Copyright and Open Licences .)
• When referencing a restricted work with a citation, one must be careful about the amount referenced. Both direct quotations and paraphrasing are permitted. All of an open work may be used with no limitations; attribution is used to give the author of this work credit.
• The closest one can come to altering a restricted work is to paraphrase the original author’s ideas and expression of these ideas. Whereas the author of an open work has provided advanced permission to use AND change their work (except in cases where ND — NoDerivatives —  has been applied).
• Citation styles are varied and established. They dictate how to cite or reference a paraphrase or quotation within text (e.g., with an in-text citation or footnote) and how and where to provide the full reference, whether it be in a reference list, a works cited, or a bibliography and the end of a book.
• The styles for attribution statements are still emerging. Current best practice for an attribution statement states it should reside on the same page (digital or printed) as the resource it refers to. Statements can stand alone, e.g., within the caption of an image, or in a list at the bottom of the page.

The following table summarizes the differences between citations and attributions.

Similarities

There are also similarities between a citation and attribution.

• Both can be — and often are — copyrighted. (See Copyright and Open Licences .)
• Both give credit to the creator of the original work
• For both restricted and open works, the author or creator of a work might be different from the copyright holder. For example, if a faculty member writes an open textbook, their institution might hold copyright. However, it’s standard practice to attribute the creator – not the copyright holder – in the attribution statement.
• Both can be used for either a newly created work or a revised work
• Both can be used when referring to a portion of another work, though the amount that can be cited from a fully copyrighted work is substantially less than what can be used from an open work
• Both can be used when building an argument or the foundation of a textbook

Special cases

Citing and attributing a closed online resource.

Resources kept in a closed system, such as password-protected platform, can still be cited and attributed in an OER. Below are templates and examples showing how based on APA guidelines. (For other examples, see  style guides to your preferred citation style.)

Citation – APA

When a resource cannot be accessed publicly, APA citation style states it should be treated as a personal communication. Additionally, only a parenthetical in-text citation is required, without inclusion in the reference list. See below for a template and example that can be used for the in-text citation.

• APA citation template (in-text citation): Author, personal communication, Date of access.

Example of a citation (APA) for a closed online resource

• J. Doe, personal communication, May 10, 2022.

If you wish to clarify that the personal communication is in fact a closed online system, here is a suggestion for how the in-text citation might read:

• J. Doe, Anywhere College SharePoint (internal access only), May 10, 2022.

If you plan to publish a resource currently stored in a restricted system, then it can be cited as an unpublished manuscript using this template.

• APA citation template : Author. (Year published). Title of manuscript [Unpublished manuscript]. Faculty department, Name of post-secondary institution.

Example of a citation (APA) for a closed online unpublished resource

• Doe, J., Smith, M., & McDonald, P. (2022). The effects of sugar on children’s health [Unpublished manuscript]. Department of Biology, Anywhere College.

Citation – MLA

If your preference is the MLA citation style then treating a resource within a closed online system as an unpublished document makes sense. Below are the template and an example of how this can be done.

• MLA citation template : Author. Title of Manuscript/Document . date of composition (at least year), along with “the name and location of the library, research institution, or personal collection housing the material.”

Example of a citation (MLA) for a closed online resource

• Henderson, George Wylie. Baby Lou and the Angel Bud . Collection of Roslyn Kirkland Allen, New York.

The legal code for Creative Commons licences states that “When a URI or hyperlink to the Licensed Material to the extent reasonably practicable (emphasis added).” In other words, providing a link to the OER to be attributed is not legally required. You must just do your best to provide one if available.

Therefore, if you wish to include some or all of a private or non-accessible online resource (that is openly-licensed) in another OER, here are suggested templates for the attribution statement for those closed system OER.

Example of an attribution statement for a closed online resource

This chapter is an adaptation of Natural Disasters and Human Impacts (in Anywhere College SharePoint, internal access only) by R. Adam Dastrup and Maura Hahnenberger, and is used under a CC BY-SA 4.0 licence.

When BCcampus Open Education began publishing open textbooks, we discovered that there were few openly licenced tables that our authors could use. So, with the help of our copy editors, we developed a way to present information in a table format without violating copyright.

We learned during our research that a table is comprised of two parts:

• The style or layout of the table, which displays the information. These elements can include the size, placement, and colour of the cells; the style of fonts; and the wording and placement of column and row headers.
• The data or information contained within the table

Our solution was to instruct authors to create an original table, and then cite the data added to that table. As you can see in the below example from Introduction to Tourism and Hospitality in B.C. , we provided the source for the data in the last row in the table. For clarity, we labelled this in-text citation as “Data source.” [2] Alternatively, you could add the source information to a footnote.

The original table, created by the author or a designer working with the author, is an original creation. Because of this, no attribution statement is required.  The table design is copyrighted by the author (or designer). However, as the data comes from an external source, it requires a citation. This same process can be applied to charts and graphs.

Citation-Attribution Fusion

The libraries at Thompson Rivers University in Kamloops have come up with a clever solution to the citation versus attribution dilemma. In the Crediting Images found Online section of their APA Citation Style web page, they suggest modifying the APA citation style so it incorporates open licence or public domain information for the image’s caption and reference. Here is an example.

The image and its caption would appear like this:

The corresponding reference would be laid out like this:

Vernon, A. (Photographer). (2007). Yellow-bellied marmot pups – Kamloops, BC [digital image]. Retrieved from Wikimedia Commons website: http://commons.wikimedia.org/wiki/File:Yellow-bellied_Marmot_pups_-_Kamloops,_BC..jpg

Attributions

• The concepts and portions of this text have been taken from Quill West’s presentation for Pierce College called Citations vs. Attributions. And how to deal with them in your work and is used under a CC BY 4.0 Licence .
• Table 2.1 by Morgan Westcott is used under a CC BY 4.0 Licence .
• "citation," Oxford Dictionary , https://en.oxforddictionaries.com/definition/citation (accessed February 6, 2018). ↵
• Morgan Westcott, "Chapter 2: Transportation," in Introduction to Tourism and Hospitality in B.C., ed. Morgan Westcott, (Victoria, B.C.: BCcampus, 2015). https://opentextbc.ca/introtourism/chapter/chapter-2-transportation/. ↵

Self-Publishing Guide Copyright © 2018 by Lauri M. Aesoph is licensed under a Creative Commons Attribution 4.0 International License , except where otherwise noted.

Share This Book

Purdue Online Writing Lab Purdue OWL® College of Liberal Arts

Frequently Asked Questions Regarding Plagiarism

Welcome to the Purdue OWL

This page is brought to you by the OWL at Purdue University. When printing this page, you must include the entire legal notice.

Copyright ©1995-2018 by The Writing Lab & The OWL at Purdue and Purdue University. All rights reserved. This material may not be published, reproduced, broadcast, rewritten, or redistributed without permission. Use of this site constitutes acceptance of our terms and conditions of fair use.

This page outlines just some of many frequently asked questions regarding plagiarism. For more information about how to avoid plagiarism, as well as best practices, visit some of our additional plagiarism resources.  

When is it Plagiarism? 

There are instances when something is clearly  intentional  plagiarism : buying, stealing, or borrowing a paper from someone else. This includes:  

• C opying a blog post or stealing an article from online.   
• Hiring someone to write your paper for you.   
• Copying a large section of text from a source without making it clear it comes from somewhere else through quotation marks or proper citation.   
• I ntentionally  failing to  cite someone else’s work,  to claim  that the ideas and words belong to you .  
• It is possible to plagiarize from yourself . In academia, if you repurpose a paper from previous class or write one paper for two classes without the instructor’s permission  this is plagiarism .   

Writers may also  unintentionally  plagiarize. This usually happens for a few common reasons:  

• The writer doesn’t fully understand the citation system they are using and ends up missing key elements of the source attribution.  
• The writer thinks they are paraphrasing (restating a source’s point in their own words) and ends up accidentally directly quoting words or phrases without realizing; in this case there is usually some attribution to the source, but not the right kind (paraphrasing vs quoting).  
• The writer misattributes a quote or idea to the wrong source; this is especially common in larger research projects where the writer is dealing with a lot of source material.

What could happen if I plagiarize?  

Students who unintentionally plagiarize often (but not always) have a meeting with their teacher, who usually goes over the issue with the student and explains how to avoid it in the future. Many teachers understand that citation practices are complex and can be difficult for beginners to learn, and therefore treat one instance of accidental plagiarism as a learning opportunity.    

However, this is not always the case. Program or department policy may force teachers to begin disciplinary proceedings about plagiarism immediately; you should always be aware that this issue is serious and not to be brushed off or ignored. The best defense against disciplinary proceedings is knowledge —  fully understanding citation practices and employing them in your writing so that you don’t plagiarize is always going to be your best option. You can find our extensive citation resources here.

The corporate world and universit ie s in particular have very strict guidelines for those accused of plagiarism. It is important to familiarize yourself with your university policy, because the consequences can be serious, which includes:  

• Automatic failure of the paper  
• Automatic failure of the class  
• A report to the Dean of Students  
• Some universities make it their policy for it to appear on your Official Transcript, which are usually needed when applying for jobs or graduate school   
• Some programs will dismiss you; for example, the English department at your university may tell you to apply to a different area of study  
• The university may dismiss you completely   
• In the corporate world, you may face litigation depending on who the victim is  
• Your job may fire you in order to avoid the company being associated with a plagiarist

Some of these consequences may seem harsh or extreme, but they are meant to stop students from plagiarizing. This is a crime and it is also an ethical dilemma. You are in school to learn and produce original work; you are doing yourself a disservice by paying someone else to write your work or reuse an old paper.   

If you find yourself in a situation when you are out of time to work on a research project, which is usually one of the common reasons why well-meaning students plagiarize,  it is always better to have an open communication with your professor  before you turn to plagiarism. Professors are there and want to help students, but they cannot help you if they do not know you need it.   

When should I cite a source to avoid plagiarizing?  

  Always give credit where credit is due. If the words that you are including in your research belong to someone else, give credit.   

Here is  a brief list of what needs to be credited or documented :  

• Words or ideas presented in a magazine, book, newspaper, song, TV program, movie, website, computer program, letter, advertisement, or any other medium  
• Information you gain through interviewing or conversing with another person, face to face, over the phone, or in writing  
• When you copy the exact words or a unique phrase  
• When you reprint any diagrams, illustrations, charts, pictures, or other visual materials  
• When you reuse or repost any digital media, including images, audio, video, or other media  

There are certain things that  do not need documentation or credit, including :  

• Writing your own lived experiences, your own observations and insights, your own thoughts, and your own conclusions about a subject  
• When you are writing up your own results obtained through lab or field experiments  
• When you use your own artwork, digital photographs, video, audio, etc.  
• When you are using "common knowledge," things like folklore, common sense observations, myths, urban legends, and historical events (but  not  historical documents)  
• When you are using generally accepted facts (e.g., pollution is bad for the environment) including facts that are accepted within particular discourse communities (e.g., in the field of composition studies, "writing is a process" is a generally accepted fact).  

What if I am accused of plagiarism, but I didn’t plagiarize?

It is important to keep notes and previous drafts of your paper, especially if you become a victim of theft yourself. This can happen in many different ways (a roommate copying your files, using a computer from a lab that still has someone else’s log in, etc).

There are some simple and easy tips to keep your intellectual property safe:

• You might have research_paper001.doc, research_paper002.doc, research_paper003.doc as you progress.
• Do the same thing for any online files you are working with. Having multiple draft versions may help prove that the work is yours (assuming you are being ethical in how you cite ideas in your work)
• Do not rely only on your hard drive, it can crash.
• Do not rely only on a USB drive, you can lose it.
• Do not rely only on cloud services, you might not be able to gain access for whatever reason.
• Many students typically use one or more of the methods above to secure their files, including emailing themselves or emailing a friend, with their permission.
• If you have to leave the computer lab for a bathroom break, lock or log out of your station
• This is possible in all sorts of programs, from Adobe Acrobat to Microsoft Word
• Just don’t forget your password
• If you have never tried this method before but would like to, use a different file for a test run to see how it works.
• If this is a habit you currently, you can make sure you have a Master Copy that you will eventually turn in, and open or create new documents and save them in the filing method mentioned above.

Press ESC to close

Or check our popular categories..., what is paraphrasing plagiarism and how to avoid it, key takeaways:.

• To prevent duplication of content, always ensure your paraphrasing significantly alters the original text while retaining the core message.
• Crafting effective content involves using your own words and a distinct sentence structure to convey ideas for assignment writing clearly and originally.
• An acceptable paraphrase requires more than just changing a few words; it demands a complete rewrite to avoid any hints of copying.
• Online paraphraser tools can aid in creating unique content but must be used responsibly to maintain originality and avoid content replication.
• An efficient writing experience incorporates thorough understanding and rephrasing of source material, ensuring the new text is original and free from content duplication.

Have you ever struggled to put someone else’s ideas into your own words? We’ve all been there.

Paraphrasing is a valuable writing skill, but the line between effective paraphrasing and plagiarism can be blurry. Inadvertent copying, or paraphrasing, can happen because of this.

By following these tips, you’ll be able to steer clear of plagiarism and comfortably overcome the paraphrasing arena. Here’s what we’ll cover:

• What is plagiarism , and why is it important to avoid?
• What is paraphrasing, and how can it benefit your writing?
• Meaning and example of paraphrasing plagiarism .
• How can seemingly good paraphrasing practices turn into plagiarism ?
• How to paraphrase correctly, including clear steps and examples.
• The potential consequences of plagiarism, both in school and beyond.

What is Plagiarism?

Plagiarism is basically stealing someone else’s work and pretending it’s your own. It applies to written stuff like copying parts of a book without quotation marks or rewriting someone’s ideas without mentioning them. But it’s not limited to writing!

Imagine you’re giving a presentation and you use your friend’s research without saying it’s theirs. That’s plagiarism too. It can also involve music, art, or even inventions if you copy someone else’s work and claim it as your own.

The important thing is to be honest about where your information comes from, whether it’s writing, a presentation, or anything else you create.

What is Paraphrasing?

Paraphrasing is like putting on a new outfit for an idea. You take the same core message from someone else’s work and express it in your own words, with different phrasing and maybe even sentence structure.

It’s key to keep the original meaning intact while making it your own.

Here’s the spectrum of paraphrasing:

• Close Paraphrase: This stays very similar to the original text, swapping synonyms for some words but keeping the core sentence structure.
• Mid-Range Paraphrase: This rearranges the sentence structure and uses more synonyms, making it a bit more distant from the original wording.
• Loose Paraphrase: This takes the central idea and expresses it in a completely new way, potentially even changing the tone or emphasis.

It’s important to remember that paraphrasing is different from summarising. Summarising condenses the main points of a longer piece, while paraphrasing focuses on restating a specific idea or passage differently.

Benefits of Paraphrasing

Paraphrasing offers a treasure trove of benefits that go beyond just avoiding plagiarism . Here’s a breakdown of some key advantages:

• Boosts Comprehension: The act of paraphrasing forces you to truly grasp the meaning of the original text. As you search for synonyms and restructure individual sentences, you solidify your understanding of the core ideas.
• Enhances Writing Skills: Paraphrasing strengthens your vocabulary and refines your ability to express yourself clearly and concisely. It’s like a workout for your writing process!
• Avoids Plagiarism: This is a big one! Paraphrasing helps you ethically integrate information from sources without simply copying.
• Improves Communication: Communication is a two-way street, and paraphrasing helps you bridge the gap with your audience. Imagine explaining a complex scientific concept to your grandma. Through paraphrasing, you can tailor the information to her level of understanding, ensuring your message resonates and lands.
• Fosters Critical Thinking: Paraphrasing isn’t passive note-taking; it’s an active dialogue with the information. As you grapple with how best to rephrase, you’re forced to analyse the core concepts, identify underlying assumptions, and potentially even challenge the original ideas. This critical thinking fosters a deeper understanding that goes beyond surface-level memorisation.
• Original Slant: Paraphrasing allows you to inject your own perspective into the information. You can choose words and sentence structures that resonate with your own voice and style.
• Clarity and Concision: Sometimes the original text might be overly complex or wordy. Paraphrasing lets you present the ideas in a clearer and more concise way.

By incorporating paraphrasing into your learning and communication strategies, you’ll unlock a world of benefits that will strengthen your writing, improve your understanding, and allow you to effectively share information.

What is Paraphrasing Plagiarism?

Paraphrasing plagiarism occurs when someone rephrases another person’s ideas or text without proper attribution, making it appear as if the rephrased content piece is their own original work. This type of plagiarism is deceptive because, while the words and sentence structures may be different, the core ideas and information remain the same as the original source.

Example of Paraphrasing Plagiarism:

Consider a student who reads an article on the effects of social media on mental health. If the student takes a paragraph from the article, changes a few words and the order of the sentences, but does not credit the original author, this is paraphrasing plagiarism.

For instance, if the original text states, “Social media usage can lead to increased feelings of anxiety and depression among teenagers,” and the student writes, “Using social media can raise anxiety and depression levels in teens,” without citation, it is still plagiarism.

Proper paraphrasing involves not only changing the wording and structure but also citing the source. Even if the information is reworded, the original ideas must be attributed to the original author to maintain academic and professional integrity.

How Paraphrasing can Lead to Plagiarism?

Here’s how paraphrasing can lead to plagiarism:

• Misunderstanding vs. Originality: Sometimes, when paraphrasing, we might misinterpret the original idea. We unintentionally rephrase a core concept believing it’s our own original thought. This unintentional plagiarism highlights the importance of thorough comprehension before paraphrasing.
• Overreliance and Patchwork: Relying too heavily on the original source’s structure or phrasing can lead to plagiarism. Imagine piecing together a text by paraphrasing sentences from a single source without weaving in your own analysis or changing the core structure significantly. This creates a patchwork of paraphrased sentences that still belongs to the original author.
• Incomplete Paraphrasing: A common pitfall is focusing solely on synonyms while keeping the sentence structure and overall flow very similar to the original. Swapping a few words doesn’t constitute true paraphrasing, and it fails to demonstrate your understanding of the concept.

How to Paraphrase Correctly to Avoid Plagiarism?

Paraphrasing correctly is an essential skill to avoid plagiarism and maintain the integrity of your work. Here are key steps to ensure you paraphrase properly:

• Fully Understand the Source Material : Before attempting to paraphrase, read the original text several times until you thoroughly understand the main ideas and details. This comprehension is crucial for effective paraphrasing.
• Use Your Own Words and Sentence Structure : When paraphrasing, aim to completely rephrase the text using your own vocabulary and sentence structure. Avoid simply replacing words with synonyms. The goal is to rewrite the entire content in a way that reflects your unique understanding.
• Avoid Copying the Original Structure : Ensure that the structure of your paraphrased text is different from the original. This means not only changing words but also reorganising the information in a new format.
• Include Proper Citation : Even when paraphrasing, it is essential to credit the original source. Include an in-text citation and a corresponding entry in your reference list to acknowledge the original author’s work.
• Compare with the Original : After paraphrasing, compare your version with the original text to ensure you have sufficiently transformed the wording and structure while accurately conveying the same meaning.
• Use Plagiarism Detection Tools : Employ plagiarism detection tools to check your work for any unintentional similarities with the source material. These AI-based technology can help you identify areas that may need further revision.

Paraphrasing Plagiarism in Various Fields

Academic writing.

Paraphrasing plagiarism in academic writing is particularly common. Students and researchers often need to use information from various sources to support their arguments and ideas. When they fail to paraphrase correctly or forget to cite their sources, it results in plagiarism.

Academic institutions take it very seriously, and students caught plagiarising can face severe consequences, including failing grades, suspension, or even expulsion.

Paraphrasing plagiarism in journalism occurs when reporters or writers use information from other articles, reports, or sources without proper attribution.

Journalists are expected to present original content and give credit to the sources they use. Failure to do so can damage their credibility and career, and lead to legal issues for their publications.

Content Creation

Content writers, such as bloggers, social media influencers, and web writers, often paraphrase information to produce engaging and informative content.

Paraphrasing plagiarism happens when they do not sufficiently alter the original text or fail to credit the original author. This can harm their reputation and lead to issues with copyright infringement .

Business and Marketing

In business and marketing, professionals often create reports, presentations, and marketing materials that include information from various sources.

Paraphrasing plagiarism occurs when they do not properly reword the information or provide citations. This can lead to loss of trust, legal problems, and damage to the company’s reputation.

Scientific Research

Scientists and researchers frequently paraphrase findings from other studies in their papers and reports.

Paraphrasing plagiarism happens when they do not sufficiently change the wording or structure of the original text or neglect to cite the original study. This can result in retraction of published papers, loss of credibility, and ethical issues in the scientific community.

Teachers and educators sometimes paraphrase materials from textbooks, research papers, or other educational resources. When they fail to properly rephrase the engaging content or credit the original authors, it results in paraphrasing plagiarism. This sets a poor example for students and undermines the integrity of the educational process.

The Sting of Stealing Knowledge: Consequences of Paraphrasing Plagiarism

Paraphrasing plagiarism, the act of rephrasing someone else’s ideas without proper citation, might seem like a harmless act. However, the consequences can sting just as much as blatant plagiarism. Here’s why you should tread carefully:

• Academic Repercussions: In schools and universities, it can be a serious offense. Depending on the severity, it could lead to a failing grade on the assignment, a mark on your academic record, or even suspension or expulsion.
• Loss of Credibility: In professional settings, it undermines your credibility. It shows a lack of respect for intellectual property and raises questions about your research and writing skills.
• Legal Issues: In extreme cases, especially with copyrighted material , it could lead to legal action. Copyright infringement can result in fines and even lawsuits.
• Damage to Reputation: A reputation for plagiarism can be difficult to shake off. It can impact your future academic or professional opportunities.

Beyond Penalties:

• Hinders Learning: It prevents you from truly understanding the material. By simply rephrasing without critical analysis, you miss out on the opportunity to learn and integrate the information effectively.
• Unethical Behavior: At its core, it is a form of intellectual theft. It takes credit away from the original author and undermines the value of their work.

What’s Next?

Paraphrasing is a valuable skill that helps you understand and explain information in your own words. However, it’s important to do it correctly to avoid rewording plagiarism .

Remember to always give credit to the original author and change the words and structure of the text completely. By practicing good paraphrasing habits, you can show respect for others’ work and maintain your own integrity.

Plagiarism has serious consequences, so it’s crucial to avoid it at all costs. By employing Bytescare Plagiarism checker, you can ensure original content. Book a demo to see first hand, how it helps you to write plagiarism free content.

What is the difference between paraphrasing and word-for-word plagiarism?

Paraphrasing involves rewording someone else’s ideas into your own words while maintaining the original meaning.

Word-for-word copying is when you copy text exactly as it appears in the original source without any changes or crediting the author.

Paraphrasing requires significant changes to the wording and structure, whereas word-for-word plagiarism is direct copying.

How can paraphrasing lead to plagiarism?

Paraphrasing can lead to plagiarism if the rephrased text is too similar to the original or if the source is not properly cited. Even if the words are changed, maintaining the same sentence structure and failing to give credit to the original author is considered plagiarism.

Is it possible to paraphrase without committing plagiarism?

Yes, it is possible to paraphrase online without committing plagiarism. To do this, you must completely rewrite the original text in your own words and sentence structure. Additionally, you must cite the original source to give proper credit to the author.

What is an example of word-for-word plagiarism?

An example of word-for-word imitating is copying a sentence directly from a book or website and pasting it into your work without quotation marks or a citation. For instance, taking the sentence “The Eiffel Tower is one of the most recognisable structures in the world” and using it verbatim in your essay without crediting the source.

Can paraphrasing be too close to the original text?

Yes, paraphrasing can be too close to the original text if only minor changes are made, such as replacing a few words with synonyms or slightly rearranging the sentences. This is often called “patchwriting” and is still considered plagiarism because the rephrased text closely mirrors the original.

How do I avoid paraphrasing plagiarism?

To avoid the examples of plagiarism in paraphrasing, you should thoroughly understand the original text and then write it in your own words, significantly changing the wording and structure. Additionally, always cite the original source to give proper credit to the author.

Why is citing sources important even when paraphrasing?

Citing sources is important when paraphrasing because it gives credit to the original author for their ideas and work. Proper citation shows respect for intellectual property, maintains academic integrity, and helps avoid accusations of plagiarism.

Categorized in:

Share Article:

About the Author

Manish jindal.

Manish Jindal is a Co-Founder and COO of Bytescare, with expertise in investment banking and a CFA Charterholder. He actively advises startups, offering guidance in fundraising, team setup, and growth strategies.

Check latest articles from this author:

What is plagiarism in academic writing, plagiarism in australian universities, previous article, plagiarism in journalism: a threat to truth and trust, next article.

Peering into the Mind of Language Models: An Approach for Attribution in Contextual Question Answering

With the enhancement in the field of generative artificial intelligence (AI), contextual question answering has become extremely relevant. Attributing model generations to the input source document is essential to ensure trustworthiness and reliability. We observe that when large language models (LLMs) are used for contextual question answering, the output answer often consists of text copied verbatim from the input prompt which is linked together with "glue text" generated by the LLM. Motivated by this, we propose that LLMs have an inherent awareness from where the text was copied, likely captured in the hidden states of the LLM. We introduce a novel method for attribution in contextual question answering, leveraging the hidden state representations of LLMs. Our approach bypasses the need for extensive model retraining and retrieval model overhead, offering granular attributions and preserving the quality of generated answers. Our experimental results demonstrate that our method performs on par or better than GPT-4 at identifying verbatim copied segments in LLM generations and in attributing these segments to their source. Importantly, our method shows robust performance across various LLM architectures, highlighting its broad applicability. Additionally, we present Verifiability-granular 1 1 1 Dataset is available at https://github.com/Anirudh-Phukan/verifiability-granular . , an attribution dataset which has token level annotations for LLM generations in the contextual question answering setup.

1 Introduction

The surge in the capabilities of Large Language Models (LLMs) has revolutionized natural language understanding. Their ability to comprehend and generate human-like text has resulted in their widespread adoption across various industries. A prominent and pivotal application of these models is question answering, particularly in contextual settings. The ability of LLMs to parse, interpret, and respond to queries within a given context has facilitated efficient information retrieval and comprehension.

Despite the remarkable strides made in contextual question answering, challenges persist within LLMs. While LLMs excel in generating informative responses, they often fall short in providing explicit references or attributions to the specific sections or sources within the context from which their answers derive Liu et al. ( 2023b ) . The absence of this attribution impedes the ability to verify the authenticity and accuracy of the generated information. Attribution not only enables verification Rashkin et al. ( 2023 ) but also increases user trust and confidence in the responses generated by these models, thus fostering their broader acceptance and utilization across diverse domains Bohnet et al. ( 2022 ) .

Existing methods for attribution typically align within three primary categories Li et al. ( 2023 ) : Systems fine-tuned or trained explicitly for attribution tasks Weller et al. ( 2023 ); Gao et al. ( 2023b ) , retrieve-then-read Chen et al. ( 2017 ); Lee et al. ( 2019 ) and approaches relying on post-generation attribution Gao et al. ( 2023a ); Huo et al. ( 2023 ) . However, each of these approaches encounters substantial challenges in achieving accurate and granular attribution, thereby limiting their effectiveness.

The initial category, which consists of systems tailored or explicitly trained for attribution tasks, poses a significant challenge Huang and Chang ( 2023 ) . This challenge primarily arises because the training process demands extensive resources, including time, computational power, and a large corpus of annotated data.

Furthermore, the continuous development of more sophisticated models requires retraining, perpetuating an unending cycle of adaptation. Additionally, the need for regression testing to ensure answer quality amplifies the burden, creating a considerable bottleneck in practical application and scalability.

On the other hand, the second category—post-generation attribution—offers an alternative but not without its own set of drawbacks. This approach relies on leveraging retrieval models to trace the sources of the generated answers. However, this method suffers from the overhead associated with these retrieval models, which often demand substantial computational resources and time. Furthermore, the granularity of attribution is constrained by the chunk size at which retrieval is performed, limiting the precision of attribution.

Lastly, the third category—retrieve then read methods—also grapples with its unique set of complications. Similar to the post-generation attribution approach, this method faces the problem of retrieval method overhead and chunking issues. The concept here is to utilize retrieved evidence as a basis for generating answers and consequently, attributions. However, as outlined in Gao et al. ( 2023a ) , retrieval does not equate to attribution due to the potential integration of external knowledge.

Our proposed method aims to surmount the limitations posed by existing attribution approaches by adopting a distinct strategy that eliminates the need for additional training while offering granular attributions. By delving into the inner workings of the LLM during its generation process, we sidestep the resource-intensive training requirements.

The crux of our approach lies in accessing the hidden state representations of tokens produced by the LLM when generating responses. Leveraging the contextual cues provided by the input—context, question, and generated answer—we extract these hidden state representations via a forward pass through the model. These hidden representations which happen to be contextual embeddings of tokens are then matched to perform attribution. An overview of this process is shown in Figure 1 .

Operating at a token-level granularity mirrors the natural generation process of LLMs. Unlike conventional post generation attribution methods that grapple with decisions on chunking at paragraph or sentence levels, our approach bypasses this dilemma. By steering clear of chunk-based attributions, we are liberated from arbitrary segmentation of the context, which often leads to imprecise referencing and diluted contextual connections.

The existing public datasets curated for the task of attribution (attribution of LLM generations) have annotations only at the response level or at the sentence level (for every sentence within the response). Kamalloo et al. ( 2023 ); Liu et al. ( 2023b ); Malaviya et al. ( 2023 ) . To assess our method’s effectiveness, token-level annotations are essential. Consequently, we process the data collected by Liu et al. ( 2023b ) to generate token-level annotations. We release this dataset to facilitate further research on token-level attribution.

In essence, our approach offers a unique and efficient solution to the problem of attribution in contextual question answering. Additionally, we also introduce a dataset that facilitates token-level attribution. With this in mind, we are excited to share our primary contributions in this paper.

We introduce a pioneering method for attribution in contextual question answering. The strengths of our approach include its token-level granularity ensuring precise attributions, lack of additional retrieval model overhead and its training-free nature which ensures consistent answer quality.

Our experimental findings demonstrate the efficacy of our method across various model families. This indicates that leveraging hidden layer representations for attribution can be broadly applied across different LLM architectures, highlighting the wide-ranging applicability of our approach.

Additionally, we release the Verifiability-granular dataset, which contains token-level attributions for LLM generations in the contextual question answering setup.

2 Related Work

Researchers have employed different approaches for the attribution of generated or identified text spans. A shared task was organized by Thorne et al. ( 2018 ) to encourage researchers to build systems capable of performing fact verification and attribution to the source texts. Evans et al. ( 2021 ) emphasized the role of source attribution in fostering truthful and responsible AI. The growing concern of fake news detection in AI-based news generation has also been considered as an attribution task Pomerleau and Rao ( 2017 ); Ferreira and Vlachos ( 2016 ) . In addition to automatic attribution, studies on manual attribution have been performed by domain specific individuals Borel ( 2023 ); Li et al. ( 2015 ) . While we discuss about fact verification as an attribution task, it is important to note that user interactions have also been found to require attribution Dziri et al. ( 2022 ) . Petroni et al. ( 2022 ) has highlighted the fact that wikipedia articles needs validation and that the citations need to be attributed. On the other hand Sarti et al. ( 2023b ) introduced a python library based on GPT-2 capable of identifying feature attributions generated from the Insequence model; capable of performing in multilingual settings Sarti et al. ( 2023a ) . Recently, researchers performed attribution task on multimodal systems as well Ancona et al. ( 2017 ); Holzinger et al. ( 2021 ); Zhao et al. ( 2023 ) .

However, what is missing in prior art is the utilisation of the contextual guidance inherently present during generation for the attribution task. In our work, we tackle this problem by utilising the contextual information encoded during generation by the LLM and attribute spans to semantically relevant parts of the source document.

3 Problem Statement

We observe a phenomenon in contextual question answering using LLMs which results in a comprehensive answer typically characterized by factual spans, replicated verbatim from various segments of the provided context, interwoven with "glue text". An example of this phenomenon is shown in Figure 2 . Yang et al. ( 2023 ) also note this pattern and use references to losslessly speed up LLM inference.

Building upon the aforementioned observations, we dissect the challenge of attribution in the contextual generation setting into two distinct yet interconnected sub-problems. The first sub-problem involves the identification of tokens within the output that have been directly copied from the provided context.

The second sub-problem delves deeper into the attribution of these identified tokens. This step involves mapping these tokens back to their original positions within the document. In essence, this implies tracing back the tokens to the exact sections within the document from which they were copied verbatim.

By doing this, we aim to establish a clear flow of information from the source document to the generated output, thereby enabling a more nuanced understanding of the context generation process.

The task at hand involves two distinct subtasks within the context of attributing the output of a large language model (LLM). We are given a document D 𝐷 D italic_D represented as a sequence of tokens ( d 1 , d 2 , … , d n ) subscript 𝑑 1 subscript 𝑑 2 … subscript 𝑑 𝑛 (d_{1},d_{2},\ldots,d_{n}) ( italic_d start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT , italic_d start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT , … , italic_d start_POSTSUBSCRIPT italic_n end_POSTSUBSCRIPT ) , a question Q 𝑄 Q italic_Q , and an answer A 𝐴 A italic_A represented as a sequence of tokens ( a 1 , a 2 , … , a m ) subscript 𝑎 1 subscript 𝑎 2 … subscript 𝑎 𝑚 (a_{1},a_{2},\ldots,a_{m}) ( italic_a start_POSTSUBSCRIPT 1 end_POSTSUBSCRIPT , italic_a start_POSTSUBSCRIPT 2 end_POSTSUBSCRIPT , … , italic_a start_POSTSUBSCRIPT italic_m end_POSTSUBSCRIPT ) . Our objective comprises two interconnected subtasks:

Subtask 1: Token Attribution Identification. This subtask involves identifying a subset of tokens in A 𝐴 A italic_A , denoted as A attr ⊆ A subscript 𝐴 attr 𝐴 A_{\text{attr}}\subseteq A italic_A start_POSTSUBSCRIPT attr end_POSTSUBSCRIPT ⊆ italic_A , that require attribution to tokens in D 𝐷 D italic_D . Mathematically, this can be expressed as selecting tokens a i ∈ A subscript 𝑎 𝑖 𝐴 a_{i}\in A italic_a start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ∈ italic_A that are directly influenced or copied from D 𝐷 D italic_D . In our current study, we restrict the scope to tokens that are verbatim copied from D 𝐷 D italic_D .

Subtask 2: Corresponding Token Mapping. For each token in A attr subscript 𝐴 attr A_{\text{attr}} italic_A start_POSTSUBSCRIPT attr end_POSTSUBSCRIPT , this subtask aims to find a mapping function f : A attr → D : 𝑓 → subscript 𝐴 attr 𝐷 f:A_{\text{attr}}\rightarrow D italic_f : italic_A start_POSTSUBSCRIPT attr end_POSTSUBSCRIPT → italic_D such that for every token a i ∈ A attr subscript 𝑎 𝑖 subscript 𝐴 attr a_{i}\in A_{\text{attr}} italic_a start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ∈ italic_A start_POSTSUBSCRIPT attr end_POSTSUBSCRIPT , there exists a corresponding token or sequence of tokens in D 𝐷 D italic_D to which a i subscript 𝑎 𝑖 a_{i} italic_a start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT can be attributed.

Formally, the mapping function can be expressed as:

In this formulation, the function f 𝑓 f italic_f establishes a connection between the tokens in the answer requiring attribution and their corresponding tokens or token sequences in the document D 𝐷 D italic_D .

4 Proposed Work

4.1 motivation.

Humans easily identify which parts of a document they use to answer questions. This skill to trace information back to its source is key for effective communication, especially when discussing complex ideas. Similarly, we propose that Large Language Models (LLMs) have an inherent awareness of the document parts they use while generating answers. This awareness is likely captured within the hidden states of the LLM, which encode token information during the generation process.

Our work is based on the idea that if LLMs can generate responses combining copied segments and self-generated "glue text", they must inherently differentiate between copied content and self-generated content. This differentiation is what we aim to uncover. By accessing and analyzing the hidden states of an LLM, we seek to reveal how it uses and attributes source information in its responses.

This approach allows us to trace how the LLM processes and uses the provided context at a detailed, token-level granularity. We believe that fully understanding how LLMs generate responses requires examining the process at this fundamental level - where each token’s role in the response is clarified and attributed to its original source. Not only does this give us an understanding of a LLM’s inner workings, but also accomplishes the useful task of attributing outputs, which helps humans easily verify/navigate the LLM outputs.

4.2 Methodology

𝐷 𝑄 𝐴 P=D+Q+A italic_P = italic_D + italic_Q + italic_A , where ‘+‘ denotes the concatenation operation. The prompt P 𝑃 P italic_P is then passed through the model M 𝑀 M italic_M in a forward pass to obtain the hidden layer representations for each token in P 𝑃 P italic_P . These representations capture the contextual information encoded by the model for each token. An illustrative example of this process is provided below.

Let’s denote the hidden layer representation of each token t i subscript 𝑡 𝑖 t_{i} italic_t start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT for a specific layer l 𝑙 l italic_l as h i l superscript subscript ℎ 𝑖 𝑙 h_{i}^{l} italic_h start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_l end_POSTSUPERSCRIPT .

4.2.1 Identifying extractive output tokens

For the first sub-task of identifying tokens in answer A 𝐴 A italic_A that are directly copied from document D 𝐷 D italic_D , we perform the following operation. For any specific layer l 𝑙 l italic_l and for all tokens a i ∈ A subscript 𝑎 𝑖 𝐴 a_{i}\in A italic_a start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT ∈ italic_A , we say that a token a i subscript 𝑎 𝑖 a_{i} italic_a start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT comes from D 𝐷 D italic_D if there exists a token d j ∈ D subscript 𝑑 𝑗 𝐷 d_{j}\in D italic_d start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT ∈ italic_D such that the cosine similarity between h i l superscript subscript ℎ 𝑖 𝑙 h_{i}^{l} italic_h start_POSTSUBSCRIPT italic_i end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_l end_POSTSUPERSCRIPT and h j l superscript subscript ℎ 𝑗 𝑙 h_{j}^{l} italic_h start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_l end_POSTSUPERSCRIPT is greater than a threshold θ 𝜃 \theta italic_θ .

This can be formally represented as:

4.2.2 Attributing Extractive Spans

Next, we use h S subscript ℎ 𝑆 h_{S} italic_h start_POSTSUBSCRIPT italic_S end_POSTSUBSCRIPT to identify anchor tokens in D 𝐷 D italic_D . For each token d j ∈ D subscript 𝑑 𝑗 𝐷 d_{j}\in D italic_d start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT ∈ italic_D , we compute the cosine similarity between h S subscript ℎ 𝑆 h_{S} italic_h start_POSTSUBSCRIPT italic_S end_POSTSUBSCRIPT and h j l superscript subscript ℎ 𝑗 𝑙 h_{j}^{l} italic_h start_POSTSUBSCRIPT italic_j end_POSTSUBSCRIPT start_POSTSUPERSCRIPT italic_l end_POSTSUPERSCRIPT and select tokens with the highest similarities as anchor tokens, denoted as D T subscript 𝐷 𝑇 D_{T} italic_D start_POSTSUBSCRIPT italic_T end_POSTSUBSCRIPT .

For each anchor token d a ∈ D T subscript 𝑑 𝑎 subscript 𝐷 𝑇 d_{a}\in D_{T} italic_d start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT ∈ italic_D start_POSTSUBSCRIPT italic_T end_POSTSUBSCRIPT , we explore windows of tokens around d a subscript 𝑑 𝑎 d_{a} italic_d start_POSTSUBSCRIPT italic_a end_POSTSUBSCRIPT , up to a maximum length L 𝐿 L italic_L . We calculate the average hidden layer representation h W subscript ℎ 𝑊 h_{W} italic_h start_POSTSUBSCRIPT italic_W end_POSTSUBSCRIPT for each window W 𝑊 W italic_W and identify the window with the highest similarity to h S subscript ℎ 𝑆 h_{S} italic_h start_POSTSUBSCRIPT italic_S end_POSTSUBSCRIPT . The highest-ranked window is considered the final attribution for the span S 𝑆 S italic_S .

In cases where D 𝐷 D italic_D is segmented into evidence spans e ∈ E 𝑒 𝐸 e\in E italic_e ∈ italic_E , the score for each e 𝑒 e italic_e is the similarity between h S subscript ℎ 𝑆 h_{S} italic_h start_POSTSUBSCRIPT italic_S end_POSTSUBSCRIPT and the best window within e 𝑒 e italic_e .

5 Experimental Setup

In this section, we outline the details of our experimental setup; the dataset, the evaluation metrics; and the baselines.

5.1 Datasets

QuoteSum Schuster et al. ( 2023 ) : 2 2 2 Downloaded from https://github.com/google-research-datasets/QuoteSum/tree/main . The dataset is licensed under CC-BY-SA-4.0 license. This dataset consists of questions, relevant passages, and human-written semi-extractive answers. In the process of dataset construction, the human annotators were tasked to answer multiple source questions by combining information from various sources. They were instructed to explicitly extract factual spans and weave them together into a coherent, well-grounded passage. The dataset comprises of 4,009 semi-extractive answers to 1,376 unique questions in total and is split into train, validation and test sets with ratios 60%, 7%, 33%, respectively.

Verifiability-Granular: We also curate a dataset with token-level attributions called Verifiability-granular (Veri-gran) by processing the dataset introduced by Liu et al. ( 2023b ) 3 3 3 Downloaded from https://tinyurl.com/verifiability . The dataset is licensed under MIT license. . The original dataset contains commercial generative search engines generated responses to input queries along with the retrieved content used for the generation. Annotations map sentences in the response to portions of the source text.

We split the source text using nltk.sent_tokenize 4 4 4 https://github.com/nltk/nltk and filter the samples that map a response statement to only one sentence of the source text. Characters are matched using diff_match_patch 5 5 5 https://github.com/google/diff-match-patch and tokens are annotated based on if all characters in the corresponding tokens match. Some additional post-processing (removing spans containing only punctuation and stop words) is done to obtain annotations in the same format as the QuoteSum dataset. The dataset contains 170, 197 annotated statements and 272, 320 annotated spans in the dev and test set respectively.

5.2 Metrics

The evaluation metrics for the two sub-tasks are described separately below.

5.2.1 Metrics for Sub-task 1

For the first sub-task, which involves the identification of spans that have been directly copied from the context, we use token-level precision, recall, and F1 score as our metrics. The ground truth for this task is the set of tokens marked as spans explicitly extracted from the context.

5.2.2 Metrics for Sub-task 2

For the second sub-task, which entails attributing the identified spans to their original positions within the document, we use accuracy as our metric. Accuracy is computed as the fraction of instances where the system correctly predicts the paragraph from which the span was extracted. This allows us to measure how effectively the system can trace the origin of the spans within the document.

5.3 Baselines

In the context of our study, we identify and utilize a number of baselines for both sub-tasks.

5.3.1 Baselines for Sub-task 1

For the first sub-task, the identification of copied spans, it is worth noting that, to our best knowledge, no existing system can perform this task without being explicitly trained which alters the quality of the answers. Common methods such as few-shot prompting and fine-tuning tend to modify the answer. Thus, we resort to a modified version of the SEMQA prompt Schuster et al. ( 2023 ) , whereby instead of prompting the system to generate an answer via few-shot prompting, we instruct it to identify spans within a provided answer (prompt included in Appendix A.7 , Figure 13 ). We prompt GPT 3.5 6 6 6 https://openai.com/blog/chatgpt and GPT-4 7 7 7 https://openai.com/research/gpt-4 using this modified prompt and they serve as our baselines for the first sub-task.

5.3.2 Baselines for Sub-task 2

GPT 3.5 and GPT-4 prompted with the modified SEMQA prompt (prompt included in Appendix A.7 , Figure 14 ) are used as baseline for the second sub-task as well, i.e., attributing the identified spans to their original source paragraph. In addition, we also employ retrieval methods such as BM25 (sparse) Robertson et al. ( 2009 ) , GTR (dense) Ni et al. ( 2021 ) , and MonoT5 Nogueira et al. ( 2020 ) as baselines for this task.

By comparing the performance of our system against these baselines, we aim to evaluate its effectiveness in tracing the origin of the spans.

6 Results and Analysis

We perform experiments for our approach (§  4.2 ) with the following LLMs: Llama-7b Touvron et al. ( 2023 ) , Llama-70b Touvron et al. ( 2023 ) , Mistral-7b Jiang et al. ( 2023 ) , Yi-6b Young et al. ( 2024 ) and OPT-350m Zhang et al. ( 2022 ) . Note that OPT-350m has been excluded from comparision on Verifiability-granular dataset due to its limitation of 2048 token context length and the dataset contains samples upto 4096 tokens. All of our experiments were run on a A100 machine with 4 80GB GPUs.

6.1 Results for Sub-task 1

The performance of our method and the baselines on sub-task 1, i.e., identifying the copied spans, is summarized in Table 1 and Figure 3 for QuoteSum and Verifiability-granular datasets respectively. We observe that our proposed method performs better than GPT models on the QuoteSum dataset and performs on par with GPT-4 on Verifiability-granular .

As seen in Table 1 , while trying to optimize for F1 on the Verifiability-Granular dataset the scores we obtain indicate that the optimal solution is to mark all tokens as copied. Therefore, for a better analysis of model performance we plot the PR curves as shown in Figure 3 .

The consistency in performance metrics across all models for this sub-task suggests that the capability to identify extracted tokens via hidden representations is not confined to any particular model family or size. This observation underscores the broader applicability and versatility of our proposed methodology, affirming its effectiveness regardless of the underlying architecture or capacity of the language model being used.

The selection of the best hyperparameters (layer, threshold) for each of our models was based on the F1 performance metric. We share more details in the Appendix (Section A.3 ). The results presented in Table 1 and Figure 3 are derived from the test set.

The results indicate a considerable shortfall in the performance of GPT-3.5, particularly in terms of recall. This suggests that the model is unable to accurately identify all tokens considered by the annotators as being extracted from the document. A representative example of this failure case is illustrated in the Appendix (Figure 6 , Section A.4 ).

As can be seen from the example, GPT-3.5 tends to identify entities rather than the specific spans that have been directly copied from the document. Even GPT-4, which demonstrates a better understanding of the task, occasionally overlooks certain components like "The" that are parts of the directly copied spans.

An additional challenge with the GPT models is their tendency to hallucinate, leading to the introduction or elimination of content that doesn’t exist in the original answer. We try to recover from the failure by taking spans marked on the modified response and superimposing it on the original response.

To elaborate our observations, we delve into the performance variation across different model families and layers in this sub-task, as illustrated in Figure 4 . For more precise comparisons among closely located numbers, the later layers from the Yi-6b model have been excluded due to their substantially lower performance compared to others.

The smoothest graph among all the model families belongs to OPT-350m which shows a gradual increase in performance until the middle layer, followed by a similarly gradual decrease. This performance peak at the middle layers aligns with the findings of Zou et al. ( 2023 ) , suggesting that earlier layers are often dedicated to low-level tasks, while later layers tend to be excessively focused on next token prediction.

Interestingly, the larger models, despite their varied architectures, exhibit a shared performance trend, with a peak at the earlier layers followed by a subsequent decrease. This pattern suggests that for these larger models, the sub-task of identifying extracted tokens tends to be a relatively low-level or straightforward task. Consequently, the best performance for these models is typically observed in the earlier layers.

6.2 Results for Sub-task 2

Table 2 presents the performance results for the second sub-task. For each model utilizing our methodology in this task, the sole hyperparameter is the layer. The results provided in the table are derived from evaluations conducted on the test set of QuoteSum and Verifiability-granular datasets.

The results indicate that our methods perform on par with GPT-4 ( ∼ 90 % similar-to absent percent 90 \sim 90\% ∼ 90 % ) on the QuoteSum dataset, while they outperform GPT-4 by ∼ 15 % similar-to absent percent 15 \sim 15\% ∼ 15 % on Verifiability-granular . Note that the human performance on Veri-gran is 92.04%, calculated as the mean performance of 4 NLP practitioners on the dataset.

The reason for this is the much harder nature of the later dataset due to the larger number of passages (3.38 vs 69.1 passages on average). Prompting GPT to attribute already generated answers is essentially a retrieval task. It has been observed by Liu et al. ( 2023a ) that LLM retrieval performance degrades with increasing context length. Our method benefits from the contextual representations of LLM embeddings without relying on the LLM for retrieval and matching representations directly.

Two noteworthy observations from the dataset and results are:

The specific nature of the datasets necessitate a delicate balance between two competing tasks: directly matching the substring in the span to be attributed, and disambiguating among multiple instances of the substring. The former is a low-level task, while the latter requires a more nuanced, high-level understanding of context-based matching.

The successful performance of models across different families and capacities suggests that approach of matching based on hidden representations is broadly applicable.

Similar to sub-task 1, we carry out a detailed analysis of performance on sub-task 2 across various model configurations and layers, as depicted in Figure 5 .

A fundamental difference between the OPT-350m and larger models lies in the influence of token positions on the layer 0 embedding. For OPT-350m, the token’s position directly impacts its layer 0 representation. This differs significantly from larger models, where the layer 0 representation depends solely on the token ID like Word2Vec Mikolov et al. ( 2013 ) , with the token’s position not having a direct influence. Instead, in larger models, the token position is a part of the transformer’s learning process Su et al. ( 2023 ) .

In light of this distinction, our methodology, observed in Table 2 , excludes layer 0 for the larger models while choosing the best layer. Because its representation, which is solely dependent on the token ID, tends to prioritize exact substring matches.

Therefore, OPT-350m begins with lower performance that gradually at the middle and the final layers. On the other hand, all larger models across different families exhibit a similar trend where they start with the highest performance at layer 0. This high initial performance is largely attributable to the dataset’s characteristics, which contain numerous examples that can be unambiguously matched using exact substring match.

However, the dataset also comprises examples with multiple similar substrings, necessitating disambiguation. This requirement for contextual understanding is well catered to by the middle and later layers, as our ablation study in the Appendix (Section 9 ) reveals. Thus, the slight upturns in performance experienced by the larger models in the middle and later layers indicates the trade-off between exact and contextual matching.

7 Discussion & Conclusion

Through this study, we introduce a novel, efficient solution to the challenge of attribution in contextual question answering. Our method leverages the hidden state representations of LLMs, providing detailed, granular attributions without requiring extensive model retraining. Our approach’s ability to perform token-level attribution easily lends itself to end-user applications, and we have included an actual system input-output in the Appendix (Section A.6 ) depicting the same.

Our experimental results, across various model configurations and layers, demonstrate the efficacy of our approach. Notably, our method performs on par or better than training-free baselines in both identifying tokens in an answer that are directly copied from the context and attributing these tokens to their original positions in the document. Additionally we make available Verifiability-granular dataset, which contains token-level attributions for LLM generations in the contextual question answering setup.

Perhaps one of the most interesting findings from our study is that the ability to identify extracted tokens and attribute them back to the source is not exclusive to a specific model family. This suggests that our approach can be applied broadly across different LLM architectures, underscoring its versatility.

8 Limitations & Future Work

One limitation of our work stems from the specific nature of the QuoteSum and Verifiability-granular datasets we utilized for our experiments. The datasets are constructed such that only verbatim spans from the document are annotated and attributed in the answers. Our method, thus, is primarily evaluated on verbatim spans. However, our method is not specifically designed for verbatim spans only; it could potentially work with paraphrased information as indicated in our experiments with a synthetically paraphrased version of Quotesum presented in the Appendix (Section A.5 ).

In the future, we intend to employ datasets that include paraphrased spans. Testing on such data will allow us to assess the method’s effectiveness in attributing paraphrased information, identify potential challenges, and adapt our methodology accordingly.

We also see potential for our method to be useful beyond the realm of contextual question answering. We tried mapping an LLM generated text span in Spanish to its source document in English (Figure 12 ).

Interestingly, our method shows potential for this use case too. We envision applying our approach to other tasks that involve information extraction and attribution to a given context. This will not only enhance our understanding of the generation process of LLMs but also extend the applicability and value of our method. We are currently limited by the availability of datasets to perform detailed analyses of other applications.

• Ancona et al. (2017) Marco Ancona, Enea Ceolini, A. Cengiz Öztireli, and Markus H. Gross. 2017. A unified view of gradient-based attribution methods for deep neural networks . CoRR , abs/1711.06104.
• Bohnet et al. (2022) Bernd Bohnet, Vinh Q Tran, Pat Verga, Roee Aharoni, Daniel Andor, Livio Baldini Soares, Jacob Eisenstein, Kuzman Ganchev, Jonathan Herzig, Kai Hui, et al. 2022. Attributed question answering: Evaluation and modeling for attributed large language models. arXiv preprint arXiv:2212.08037 .
• Borel (2023) Brooke Borel. 2023. The Chicago guide to fact-checking . University of Chicago Press.
• Chen et al. (2017) Danqi Chen, Adam Fisch, Jason Weston, and Antoine Bordes. 2017. Reading wikipedia to answer open-domain questions. arXiv preprint arXiv:1704.00051 .
• Dziri et al. (2022) Nouha Dziri, Hannah Rashkin, Tal Linzen, and David Reitter. 2022. Evaluating attribution in dialogue systems: The BEGIN benchmark . Transactions of the Association for Computational Linguistics , 10:1066–1083.
• Evans et al. (2021) Owain Evans, Owen Cotton-Barratt, Lukas Finnveden, Adam Bales, Avital Balwit, Peter Wills, Luca Righetti, and William Saunders. 2021. Truthful AI: developing and governing AI that does not lie . CoRR , abs/2110.06674.
• Ferreira and Vlachos (2016) William Ferreira and Andreas Vlachos. 2016. Emergent: a novel data-set for stance classification . In NAACL HLT 2016, The 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, San Diego California, USA, June 12-17, 2016 , pages 1163–1168. The Association for Computational Linguistics.
• Gao et al. (2023a) Luyu Gao, Zhuyun Dai, Panupong Pasupat, Anthony Chen, Arun Tejasvi Chaganty, Yicheng Fan, Vincent Zhao, Ni Lao, Hongrae Lee, Da-Cheng Juan, et al. 2023a. Rarr: Researching and revising what language models say, using language models. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages 16477–16508.
• Gao et al. (2023b) Tianyu Gao, Howard Yen, Jiatong Yu, and Danqi Chen. 2023b. Enabling large language models to generate text with citations. arXiv preprint arXiv:2305.14627 .
• Holzinger et al. (2021) Andreas Holzinger, Bernd Malle, Anna Saranti, and Bastian Pfeifer. 2021. Towards multi-modal causability with graph neural networks enabling information fusion for explainable AI . Inf. Fusion , 71:28–37.
• Huang and Chang (2023) Jie Huang and Kevin Chen-Chuan Chang. 2023. Citation: A key to building responsible and accountable large language models. arXiv preprint arXiv:2307.02185 .
• Huo et al. (2023) Siqing Huo, Negar Arabzadeh, and Charles Clarke. 2023. Retrieving supporting evidence for generative question answering. In Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region , pages 11–20.
• Jiang et al. (2023) Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. 2023. Mistral 7b. arXiv preprint arXiv:2310.06825 .
• Kamalloo et al. (2023) Ehsan Kamalloo, Aref Jafari, Xinyu Zhang, Nandan Thakur, and Jimmy Lin. 2023. Hagrid: A human-llm collaborative dataset for generative information-seeking with attribution. arXiv preprint arXiv:2307.16883 .
• Lee et al. (2019) Kenton Lee, Ming-Wei Chang, and Kristina Toutanova. 2019. Latent retrieval for weakly supervised open domain question answering. arXiv preprint arXiv:1906.00300 .
• Li et al. (2023) Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, Ziyang Chen, Baotian Hu, Aiguo Wu, and Min Zhang. 2023. A survey of large language models attribution . ArXiv , abs/2311.03731.
• Li et al. (2015) Yaliang Li, Jing Gao, Chuishi Meng, Qi Li, Lu Su, Bo Zhao, Wei Fan, and Jiawei Han. 2015. A survey on truth discovery . SIGKDD Explor. , 17(2):1–16.
• Liu et al. (2023a) Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023a. Lost in the middle: How language models use long contexts. arXiv preprint arXiv:2307.03172 .
• Liu et al. (2023b) Nelson F Liu, Tianyi Zhang, and Percy Liang. 2023b. Evaluating verifiability in generative search engines. arXiv preprint arXiv:2304.09848 .
• Malaviya et al. (2023) Chaitanya Malaviya, Subin Lee, Sihao Chen, Elizabeth Sieber, Mark Yatskar, and Dan Roth. 2023. Expertqa: Expert-curated questions and attributed answers . ArXiv , abs/2309.07852.
• Mikolov et al. (2013) Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 .
• Ni et al. (2021) Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo Hernández Ábrego, Ji Ma, Vincent Y Zhao, Yi Luan, Keith B Hall, Ming-Wei Chang, et al. 2021. Large dual encoders are generalizable retrievers. arXiv preprint arXiv:2112.07899 .
• Nogueira et al. (2020) Rodrigo Nogueira, Zhiying Jiang, and Jimmy Lin. 2020. Document ranking with a pretrained sequence-to-sequence model. arXiv preprint arXiv:2003.06713 .
• Petroni et al. (2022) Fabio Petroni, Samuel Broscheit, Aleksandra Piktus, Patrick S. H. Lewis, Gautier Izacard, Lucas Hosseini, Jane Dwivedi-Yu, Maria Lomeli, Timo Schick, Pierre-Emmanuel Mazaré, Armand Joulin, Edouard Grave, and Sebastian Riedel. 2022. Improving wikipedia verifiability with AI . CoRR , abs/2207.06220.
• Pomerleau and Rao (2017) Dean Pomerleau and Delip Rao. 2017. The fake news challenge: Exploring how artificial intelligence technologies could be leveraged to combat fake news. Fake news challenge .
• Rashkin et al. (2023) Hannah Rashkin, Vitaly Nikolaev, Matthew Lamm, Lora Aroyo, Michael Collins, Dipanjan Das, Slav Petrov, Gaurav Singh Tomar, Iulia Turc, and David Reitter. 2023. Measuring attribution in natural language generation models. Computational Linguistics , pages 1–64.
• Robertson et al. (2009) Stephen Robertson, Hugo Zaragoza, et al. 2009. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends® in Information Retrieval , 3(4):333–389.
• Sarti et al. (2023a) Gabriele Sarti, Grzegorz Chrupala, Malvina Nissim, and Arianna Bisazza. 2023a. Quantifying the plausibility of context reliance in neural machine translation . CoRR , abs/2310.01188.
• Sarti et al. (2023b) Gabriele Sarti, Nils Feldhus, Ludwig Sickert, and Oskar van der Wal. 2023b. Inseq: An interpretability toolkit for sequence generation models . In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) , pages 421–435, Toronto, Canada. Association for Computational Linguistics.
• Schuster et al. (2023) Tal Schuster, Adam D Lelkes, Haitian Sun, Jai Gupta, Jonathan Berant, William W Cohen, and Donald Metzler. 2023. Semqa: Semi-extractive multi-source question answering. arXiv preprint arXiv:2311.04886 .
• Su et al. (2023) Jianlin Su, Murtadha Ahmed, Yu Lu, Shengfeng Pan, Wen Bo, and Yunfeng Liu. 2023. Roformer: Enhanced transformer with rotary position embedding. Neurocomputing , page 127063.
• Thorne et al. (2018) James Thorne, Andreas Vlachos, Oana Cocarascu, Christos Christodoulopoulos, and Arpit Mittal. 2018. The fact extraction and verification (FEVER) shared task . CoRR , abs/1811.10971.
• Touvron et al. (2023) Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. 2023. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971 .
• Weller et al. (2023) Orion Weller, Marc Marone, Nathaniel Weir, Dawn Lawrie, Daniel Khashabi, and Benjamin Van Durme. 2023. " according to…" prompting language models improves quoting from pre-training data. arXiv preprint arXiv:2305.13252 .
• Yang et al. (2023) Nan Yang, Tao Ge, Liang Wang, Binxing Jiao, Daxin Jiang, Linjun Yang, Rangan Majumder, and Furu Wei. 2023. Inference with reference: Lossless acceleration of large language models .
• Young et al. (2024) Alex Young, Bei Chen, Chao Li, Chengen Huang, Ge Zhang, Guanwei Zhang, Heng Li, Jiangcheng Zhu, Jianqun Chen, Jing Chang, et al. 2024. Yi: Open foundation models by 01. ai. arXiv preprint arXiv:2403.04652 .
• Zhang et al. (2022) Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, et al. 2022. Opt: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068 .
• Zhao et al. (2023) Ruochen Zhao, Hailin Chen, Weishi Wang, Fangkai Jiao, Do Xuan Long, Chengwei Qin, Bosheng Ding, Xiaobao Guo, Minzhi Li, Xingxuan Li, and Shafiq Joty. 2023. Retrieving multimodal information for augmented generation: A survey . In Findings of the Association for Computational Linguistics: EMNLP 2023, Singapore, December 6-10, 2023 , pages 4736–4756. Association for Computational Linguistics.
• Zou et al. (2023) Andy Zou, Long Phan, Sarah Chen, James Campbell, Phillip Guo, Richard Ren, Alexander Pan, Xuwang Yin, Mantas Mazeika, Ann-Kathrin Dombrowski, et al. 2023. Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405 .

Appendix A Appendix

A.1 performance over target span position.

In this section, we analyze if the position of the target span (the span that needs to be attributed) within the answer influences the efficacy of our methodology. The start character of each span is normalized with respect to the length of the answer and the accuracy is measured for all the models on Sub-task2. As shown in Figure 8 , the accuracy improves with increasing span position for all models. In decoder-only models each step can access information from previously generated or processed tokens only. As the model progresses through the text, it accumulates more information and context about the text it is analyzing. We believe that this behaviour is the reason why accuracy increases when we attribute later text spans.

A.2 Performance of layers in the disambiguation task setting

Given a span that needs to be attributed, there are instances in the dataset where ground truth attribution sub-string is present multiple times over different sources. During inference, our approach has to both identify the correct attribution sub-string and also disambiguate between the multiple occurances of the sub-string. Disambiguating over these multiple occurences, requires contextual understanding of the span to be attributed in order to choose between the occurences. Data points that require disambiguation are collected from the test set of the QuoteSum dataset.

We measure performance across all layers for different models on this subset (Figure 9 ). We compute the performance when one of the multiple occurrences of the sub-string is chosen randomly. This is highlighted as the random baseline in the figure.

Figure 9 shows an interesting trend, where early layers perform worse than the random baseline on the disambiguation task. Early layers capture low level, basic features of the input text, making them less suitable for a complex task such as disambiguation. Performance improves as we use later layers, with the middle layers yielding the best performance for Llama-70b and Mistral-7b, and later layers yielding the best performance for Llama-7b and OPT-350m.

A.3 Effect of hyper-parameters on Sub-task 1 performance

We present graphs for precision, recall and F1 in Figure 7 generated during our experiments on the QuoteSum train set, which show the impact of the threshold θ 𝜃 \theta italic_θ , and the choice of layer on the performance of our method (Llama-7b) for Sub-task 1. θ 𝜃 \theta italic_θ was chosen for each layer-model setup to achieve the best f1 score. Interestingly we note that in the earlier layers, higher thresholds are preferred. We hypothesize that there’s a higher chance of low-level information overlapping, necessitating more stringent filtering at the earlier layers.

A.4 Limitations of GPT models for Sub-task 1

Figure 6 depicts a qualitative example where GPT models fail in identifying directly copied spans from a source document on the QuoteSum dataset. In this example GPT-3.5 is only attributing entities like Lincoln Castle, Battle of Lincoln and the Second Battle of Lincoln. GPT-4 has a better understanding of the task but leaves out tokens like ’The’ which were copied verbatim from the source.

A.5 Synthetic Paraphrasing of QuoteSum

We paraphrase the QuoteSum dataset by prompting GPT-4 as shown below. A sample from the dataset is shown in Figure 10 .

We computed the sub-task 2 performance of the various methods on the paraphrased test set and summarize the results in Table 3 .

We observe that the performance of all the methods drop on the paraphrased version of QuoteSum by nearly the same amount(1-3%). This probably implying that synthetic paraphrasing setup while harder than the original dataset is not difficult enough to offer additional insights.

A.6 Qualitative Example for Sub-task 1 and Sub-task 2

To illustrate the utility of our method, a qualitative example is presented in Figure 11 .

A.7 Prompts for GPT baselines

We present the prompts used for our GPT-based baselines for Sub-task 1 and Sub-task 2 in Figure 13 and Figure 14 respectively. For Sub-task 1, the LLM has to respond with copied spans along with the source paragraph number. For Sub-task 2, we mark the verbatim copied spans in "[]" and prompt the LLM to identify the source paragraph only.