research references meaning

• Privacy Policy

Home » References in Research – Types, Examples and Writing Guide

References in Research – Types, Examples and Writing Guide

Table of Contents

References in Research

Definition:

References in research are a list of sources that a researcher has consulted or cited while conducting their study. They are an essential component of any academic work, including research papers, theses, dissertations, and other scholarly publications.

Types of References

There are several types of references used in research, and the type of reference depends on the source of information being cited. The most common types of references include:

References to books typically include the author’s name, title of the book, publisher, publication date, and place of publication.

Example: Smith, J. (2018). The Art of Writing. Penguin Books.

Journal Articles

References to journal articles usually include the author’s name, title of the article, name of the journal, volume and issue number, page numbers, and publication date.

Example: Johnson, T. (2021). The Impact of Social Media on Mental Health. Journal of Psychology, 32(4), 87-94.

Web sources

References to web sources should include the author or organization responsible for the content, the title of the page, the URL, and the date accessed.

Example: World Health Organization. (2020). Coronavirus disease (COVID-19) advice for the public. Retrieved from https://www.who.int/emergencies/disease/novel-coronavirus-2019/advice-for-public

Conference Proceedings

References to conference proceedings should include the author’s name, title of the paper, name of the conference, location of the conference, date of the conference, and page numbers.

Example: Chen, S., & Li, J. (2019). The Future of AI in Education. Proceedings of the International Conference on Educational Technology, Beijing, China, July 15-17, pp. 67-78.

References to reports typically include the author or organization responsible for the report, title of the report, publication date, and publisher.

Example: United Nations. (2020). The Sustainable Development Goals Report. United Nations.

Formats of References

Some common Formates of References with their examples are as follows:

APA (American Psychological Association) Style

The APA (American Psychological Association) Style has specific guidelines for formatting references used in academic papers, articles, and books. Here are the different reference formats in APA style with examples:

Author, A. A. (Year of publication). Title of book. Publisher.

Example : Smith, J. K. (2005). The psychology of social interaction. Wiley-Blackwell.

Journal Article

Author, A. A., Author, B. B., & Author, C. C. (Year of publication). Title of article. Title of Journal, volume number(issue number), page numbers.

Example : Brown, L. M., Keating, J. G., & Jones, S. M. (2012). The role of social support in coping with stress among African American adolescents. Journal of Research on Adolescence, 22(1), 218-233.

Author, A. A. (Year of publication or last update). Title of page. Website name. URL.

Example : Centers for Disease Control and Prevention. (2020, December 11). COVID-19: How to protect yourself and others. https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html

Magazine article

Author, A. A. (Year, Month Day of publication). Title of article. Title of Magazine, volume number(issue number), page numbers.

Example : Smith, M. (2019, March 11). The power of positive thinking. Psychology Today, 52(3), 60-65.

Newspaper article:

Author, A. A. (Year, Month Day of publication). Title of article. Title of Newspaper, page numbers.

Example: Johnson, B. (2021, February 15). New study shows benefits of exercise on mental health. The New York Times, A8.

Edited book

Editor, E. E. (Ed.). (Year of publication). Title of book. Publisher.

Example : Thompson, J. P. (Ed.). (2014). Social work in the 21st century. Sage Publications.

Chapter in an edited book:

Author, A. A. (Year of publication). Title of chapter. In E. E. Editor (Ed.), Title of book (pp. page numbers). Publisher.

Example : Johnson, K. S. (2018). The future of social work: Challenges and opportunities. In J. P. Thompson (Ed.), Social work in the 21st century (pp. 105-118). Sage Publications.

MLA (Modern Language Association) Style

The MLA (Modern Language Association) Style is a widely used style for writing academic papers and essays in the humanities. Here are the different reference formats in MLA style:

Author’s Last name, First name. Title of Book. Publisher, Publication year.

Example : Smith, John. The Psychology of Social Interaction. Wiley-Blackwell, 2005.

Journal article

Author’s Last name, First name. “Title of Article.” Title of Journal, volume number, issue number, Publication year, page numbers.

Example : Brown, Laura M., et al. “The Role of Social Support in Coping with Stress among African American Adolescents.” Journal of Research on Adolescence, vol. 22, no. 1, 2012, pp. 218-233.

Author’s Last name, First name. “Title of Webpage.” Website Name, Publication date, URL.

Example : Centers for Disease Control and Prevention. “COVID-19: How to Protect Yourself and Others.” CDC, 11 Dec. 2020, https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html.

Author’s Last name, First name. “Title of Article.” Title of Magazine, Publication date, page numbers.

Example : Smith, Mary. “The Power of Positive Thinking.” Psychology Today, Mar. 2019, pp. 60-65.

Newspaper article

Author’s Last name, First name. “Title of Article.” Title of Newspaper, Publication date, page numbers.

Example : Johnson, Bob. “New Study Shows Benefits of Exercise on Mental Health.” The New York Times, 15 Feb. 2021, p. A8.

Editor’s Last name, First name, editor. Title of Book. Publisher, Publication year.

Example : Thompson, John P., editor. Social Work in the 21st Century. Sage Publications, 2014.

Chapter in an edited book

Author’s Last name, First name. “Title of Chapter.” Title of Book, edited by Editor’s First Name Last name, Publisher, Publication year, page numbers.

Example : Johnson, Karen S. “The Future of Social Work: Challenges and Opportunities.” Social Work in the 21st Century, edited by John P. Thompson, Sage Publications, 2014, pp. 105-118.

Chicago Manual of Style

The Chicago Manual of Style is a widely used style for writing academic papers, dissertations, and books in the humanities and social sciences. Here are the different reference formats in Chicago style:

Example : Smith, John K. The Psychology of Social Interaction. Wiley-Blackwell, 2005.

Author’s Last name, First name. “Title of Article.” Title of Journal volume number, no. issue number (Publication year): page numbers.

Example : Brown, Laura M., John G. Keating, and Sarah M. Jones. “The Role of Social Support in Coping with Stress among African American Adolescents.” Journal of Research on Adolescence 22, no. 1 (2012): 218-233.

Author’s Last name, First name. “Title of Webpage.” Website Name. Publication date. URL.

Example : Centers for Disease Control and Prevention. “COVID-19: How to Protect Yourself and Others.” CDC. December 11, 2020. https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html.

Author’s Last name, First name. “Title of Article.” Title of Magazine, Publication date.

Example : Smith, Mary. “The Power of Positive Thinking.” Psychology Today, March 2019.

Author’s Last name, First name. “Title of Article.” Title of Newspaper, Publication date.

Example : Johnson, Bob. “New Study Shows Benefits of Exercise on Mental Health.” The New York Times, February 15, 2021.

Example : Thompson, John P., ed. Social Work in the 21st Century. Sage Publications, 2014.

Author’s Last name, First name. “Title of Chapter.” In Title of Book, edited by Editor’s First Name Last Name, page numbers. Publisher, Publication year.

Example : Johnson, Karen S. “The Future of Social Work: Challenges and Opportunities.” In Social Work in the 21st Century, edited by John P. Thompson, 105-118. Sage Publications, 2014.

Harvard Style

The Harvard Style, also known as the Author-Date System, is a widely used style for writing academic papers and essays in the social sciences. Here are the different reference formats in Harvard Style:

Author’s Last name, First name. Year of publication. Title of Book. Place of publication: Publisher.

Example : Smith, John. 2005. The Psychology of Social Interaction. Oxford: Wiley-Blackwell.

Author’s Last name, First name. Year of publication. “Title of Article.” Title of Journal volume number (issue number): page numbers.

Example: Brown, Laura M., John G. Keating, and Sarah M. Jones. 2012. “The Role of Social Support in Coping with Stress among African American Adolescents.” Journal of Research on Adolescence 22 (1): 218-233.

Author’s Last name, First name. Year of publication. “Title of Webpage.” Website Name. URL. Accessed date.

Example : Centers for Disease Control and Prevention. 2020. “COVID-19: How to Protect Yourself and Others.” CDC. https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html. Accessed April 1, 2023.

Author’s Last name, First name. Year of publication. “Title of Article.” Title of Magazine, month and date of publication.

Example : Smith, Mary. 2019. “The Power of Positive Thinking.” Psychology Today, March 2019.

Author’s Last name, First name. Year of publication. “Title of Article.” Title of Newspaper, month and date of publication.

Example : Johnson, Bob. 2021. “New Study Shows Benefits of Exercise on Mental Health.” The New York Times, February 15, 2021.

Editor’s Last name, First name, ed. Year of publication. Title of Book. Place of publication: Publisher.

Example : Thompson, John P., ed. 2014. Social Work in the 21st Century. Thousand Oaks, CA: Sage Publications.

Author’s Last name, First name. Year of publication. “Title of Chapter.” In Title of Book, edited by Editor’s First Name Last Name, page numbers. Place of publication: Publisher.

Example : Johnson, Karen S. 2014. “The Future of Social Work: Challenges and Opportunities.” In Social Work in the 21st Century, edited by John P. Thompson, 105-118. Thousand Oaks, CA: Sage Publications.

Vancouver Style

The Vancouver Style, also known as the Uniform Requirements for Manuscripts Submitted to Biomedical Journals, is a widely used style for writing academic papers in the biomedical sciences. Here are the different reference formats in Vancouver Style:

Author’s Last name, First name. Title of Book. Edition number. Place of publication: Publisher; Year of publication.

Example : Smith, John K. The Psychology of Social Interaction. 2nd ed. Oxford: Wiley-Blackwell; 2005.

Author’s Last name, First name. Title of Article. Abbreviated Journal Title. Year of publication; volume number(issue number):page numbers.

Example : Brown LM, Keating JG, Jones SM. The Role of Social Support in Coping with Stress among African American Adolescents. J Res Adolesc. 2012;22(1):218-233.

Author’s Last name, First name. Title of Webpage. Website Name [Internet]. Publication date. [cited date]. Available from: URL.

Example : Centers for Disease Control and Prevention. COVID-19: How to Protect Yourself and Others [Internet]. 2020 Dec 11. [cited 2023 Apr 1]. Available from: https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html.

Author’s Last name, First name. Title of Article. Title of Magazine. Year of publication; month and day of publication:page numbers.

Example : Smith M. The Power of Positive Thinking. Psychology Today. 2019 Mar 1:32-35.

Author’s Last name, First name. Title of Article. Title of Newspaper. Year of publication; month and day of publication:page numbers.

Example : Johnson B. New Study Shows Benefits of Exercise on Mental Health. The New York Times. 2021 Feb 15:A4.

Editor’s Last name, First name, editor. Title of Book. Edition number. Place of publication: Publisher; Year of publication.

Example: Thompson JP, editor. Social Work in the 21st Century. 1st ed. Thousand Oaks, CA: Sage Publications; 2014.

Author’s Last name, First name. Title of Chapter. In: Editor’s Last name, First name, editor. Title of Book. Edition number. Place of publication: Publisher; Year of publication. page numbers.

Example : Johnson KS. The Future of Social Work: Challenges and Opportunities. In: Thompson JP, editor. Social Work in the 21st Century. 1st ed. Thousand Oaks, CA: Sage Publications; 2014. p. 105-118.

Turabian Style

Turabian style is a variation of the Chicago style used in academic writing, particularly in the fields of history and humanities. Here are the different reference formats in Turabian style:

Author’s Last name, First name. Title of Book. Place of publication: Publisher, Year of publication.

Example : Smith, John K. The Psychology of Social Interaction. Oxford: Wiley-Blackwell, 2005.

Author’s Last name, First name. “Title of Article.” Title of Journal volume number, no. issue number (Year of publication): page numbers.

Example : Brown, LM, Keating, JG, Jones, SM. “The Role of Social Support in Coping with Stress among African American Adolescents.” J Res Adolesc 22, no. 1 (2012): 218-233.

Author’s Last name, First name. “Title of Webpage.” Name of Website. Publication date. Accessed date. URL.

Example : Centers for Disease Control and Prevention. “COVID-19: How to Protect Yourself and Others.” CDC. December 11, 2020. Accessed April 1, 2023. https://www.cdc.gov/coronavirus/2019-ncov/prevent-getting-sick/prevention.html.

Author’s Last name, First name. “Title of Article.” Title of Magazine, Month Day, Year of publication, page numbers.

Example : Smith, M. “The Power of Positive Thinking.” Psychology Today, March 1, 2019, 32-35.

Author’s Last name, First name. “Title of Article.” Title of Newspaper, Month Day, Year of publication.

Example : Johnson, B. “New Study Shows Benefits of Exercise on Mental Health.” The New York Times, February 15, 2021.

Editor’s Last name, First name, ed. Title of Book. Place of publication: Publisher, Year of publication.

Example : Thompson, JP, ed. Social Work in the 21st Century. Thousand Oaks, CA: Sage Publications, 2014.

Author’s Last name, First name. “Title of Chapter.” In Title of Book, edited by Editor’s Last name, First name, page numbers. Place of publication: Publisher, Year of publication.

Example : Johnson, KS. “The Future of Social Work: Challenges and Opportunities.” In Social Work in the 21st Century, edited by Thompson, JP, 105-118. Thousand Oaks, CA: Sage Publications, 2014.

IEEE (Institute of Electrical and Electronics Engineers) Style

IEEE (Institute of Electrical and Electronics Engineers) style is commonly used in engineering, computer science, and other technical fields. Here are the different reference formats in IEEE style:

Author’s Last name, First name. Book Title. Place of Publication: Publisher, Year of publication.

Example : Oppenheim, A. V., & Schafer, R. W. Discrete-Time Signal Processing. Upper Saddle River, NJ: Prentice Hall, 2010.

Author’s Last name, First name. “Title of Article.” Abbreviated Journal Title, vol. number, no. issue number, pp. page numbers, Month year of publication.

Example: Shannon, C. E. “A Mathematical Theory of Communication.” Bell System Technical Journal, vol. 27, no. 3, pp. 379-423, July 1948.

Conference paper

Author’s Last name, First name. “Title of Paper.” In Title of Conference Proceedings, Place of Conference, Date of Conference, pp. page numbers, Year of publication.

Example: Gupta, S., & Kumar, P. “An Improved System of Linear Discriminant Analysis for Face Recognition.” In Proceedings of the 2011 International Conference on Computer Science and Network Technology, Harbin, China, Dec. 2011, pp. 144-147.

Author’s Last name, First name. “Title of Webpage.” Name of Website. Date of publication or last update. Accessed date. URL.

Example : National Aeronautics and Space Administration. “Apollo 11.” NASA. July 20, 1969. Accessed April 1, 2023. https://www.nasa.gov/mission_pages/apollo/apollo11.html.

Technical report

Author’s Last name, First name. “Title of Report.” Name of Institution or Organization, Report number, Year of publication.

Example : Smith, J. R. “Development of a New Solar Panel Technology.” National Renewable Energy Laboratory, NREL/TP-6A20-51645, 2011.

Author’s Last name, First name. “Title of Patent.” Patent number, Issue date.

Example : Suzuki, H. “Method of Producing Carbon Nanotubes.” US Patent 7,151,019, December 19, 2006.

Standard Title. Standard number, Publication date.

Example : IEEE Standard for Floating-Point Arithmetic. IEEE Std 754-2008, August 29, 2008

ACS (American Chemical Society) Style

ACS (American Chemical Society) style is commonly used in chemistry and related fields. Here are the different reference formats in ACS style:

Author’s Last name, First name; Author’s Last name, First name. Title of Article. Abbreviated Journal Title Year, Volume, Page Numbers.

Example : Wang, Y.; Zhao, X.; Cui, Y.; Ma, Y. Facile Preparation of Fe3O4/graphene Composites Using a Hydrothermal Method for High-Performance Lithium Ion Batteries. ACS Appl. Mater. Interfaces 2012, 4, 2715-2721.

Author’s Last name, First name. Book Title; Publisher: Place of Publication, Year of Publication.

Example : Carey, F. A. Organic Chemistry; McGraw-Hill: New York, 2008.

Author’s Last name, First name. Chapter Title. In Book Title; Editor’s Last name, First name, Ed.; Publisher: Place of Publication, Year of Publication; Volume number, Chapter number, Page Numbers.

Example : Grossman, R. B. Analytical Chemistry of Aerosols. In Aerosol Measurement: Principles, Techniques, and Applications; Baron, P. A.; Willeke, K., Eds.; Wiley-Interscience: New York, 2001; Chapter 10, pp 395-424.

Author’s Last name, First name. Title of Webpage. Website Name, URL (accessed date).

Example : National Institute of Standards and Technology. Atomic Spectra Database. https://www.nist.gov/pml/atomic-spectra-database (accessed April 1, 2023).

Author’s Last name, First name. Patent Number. Patent Date.

Example : Liu, Y.; Huang, H.; Chen, H.; Zhang, W. US Patent 9,999,999, December 31, 2022.

Author’s Last name, First name; Author’s Last name, First name. Title of Article. In Title of Conference Proceedings, Publisher: Place of Publication, Year of Publication; Volume Number, Page Numbers.

Example : Jia, H.; Xu, S.; Wu, Y.; Wu, Z.; Tang, Y.; Huang, X. Fast Adsorption of Organic Pollutants by Graphene Oxide. In Proceedings of the 15th International Conference on Environmental Science and Technology, American Chemical Society: Washington, DC, 2017; Volume 1, pp 223-228.

AMA (American Medical Association) Style

AMA (American Medical Association) style is commonly used in medical and scientific fields. Here are the different reference formats in AMA style:

Author’s Last name, First name. Article Title. Journal Abbreviation. Year; Volume(Issue):Page Numbers.

Example : Jones, R. A.; Smith, B. C. The Role of Vitamin D in Maintaining Bone Health. JAMA. 2019;321(17):1765-1773.

Author’s Last name, First name. Book Title. Edition number. Place of Publication: Publisher; Year.

Example : Guyton, A. C.; Hall, J. E. Textbook of Medical Physiology. 13th ed. Philadelphia, PA: Saunders; 2015.

Author’s Last name, First name. Chapter Title. In: Editor’s Last name, First name, ed. Book Title. Edition number. Place of Publication: Publisher; Year: Page Numbers.

Example: Rajakumar, K. Vitamin D and Bone Health. In: Holick, M. F., ed. Vitamin D: Physiology, Molecular Biology, and Clinical Applications. 2nd ed. New York, NY: Springer; 2010:211-222.

Author’s Last name, First name. Webpage Title. Website Name. URL. Published date. Updated date. Accessed date.

Example : National Cancer Institute. Breast Cancer Prevention (PDQ®)–Patient Version. National Cancer Institute. https://www.cancer.gov/types/breast/patient/breast-prevention-pdq. Published October 11, 2022. Accessed April 1, 2023.

Author’s Last name, First name. Conference presentation title. In: Conference Title; Conference Date; Place of Conference.

Example : Smith, J. R. Vitamin D and Bone Health: A Meta-Analysis. In: Proceedings of the Annual Meeting of the American Society for Bone and Mineral Research; September 20-23, 2022; San Diego, CA.

Thesis or dissertation

Author’s Last name, First name. Title of Thesis or Dissertation. Degree level [Doctoral dissertation or Master’s thesis]. University Name; Year.

Example : Wilson, S. A. The Effects of Vitamin D Supplementation on Bone Health in Postmenopausal Women [Doctoral dissertation]. University of California, Los Angeles; 2018.

ASCE (American Society of Civil Engineers) Style

The ASCE (American Society of Civil Engineers) style is commonly used in civil engineering fields. Here are the different reference formats in ASCE style:

Author’s Last name, First name. “Article Title.” Journal Title, volume number, issue number (year): page numbers. DOI or URL (if available).

Example : Smith, J. R. “Evaluation of the Effectiveness of Sustainable Drainage Systems in Urban Areas.” Journal of Environmental Engineering, vol. 146, no. 3 (2020): 04020010. https://doi.org/10.1061/(ASCE)EE.1943-7870.0001668.

Example : McCuen, R. H. Hydrologic Analysis and Design. 4th ed. Upper Saddle River, NJ: Pearson Education; 2013.

Author’s Last name, First name. “Chapter Title.” In: Editor’s Last name, First name, ed. Book Title. Edition number. Place of Publication: Publisher; Year: page numbers.

Example : Maidment, D. R. “Floodplain Management in the United States.” In: Shroder, J. F., ed. Treatise on Geomorphology. San Diego, CA: Academic Press; 2013: 447-460.

Author’s Last name, First name. “Paper Title.” In: Conference Title; Conference Date; Location. Place of Publication: Publisher; Year: page numbers.

Example: Smith, J. R. “Sustainable Drainage Systems for Urban Areas.” In: Proceedings of the ASCE International Conference on Sustainable Infrastructure; November 6-9, 2019; Los Angeles, CA. Reston, VA: American Society of Civil Engineers; 2019: 156-163.

Author’s Last name, First name. “Report Title.” Report number. Place of Publication: Publisher; Year.

Example : U.S. Army Corps of Engineers. “Hurricane Sandy Coastal Risk Reduction Program, New York and New Jersey.” Report No. P-15-001. Washington, DC: U.S. Army Corps of Engineers; 2015.

CSE (Council of Science Editors) Style

The CSE (Council of Science Editors) style is commonly used in the scientific and medical fields. Here are the different reference formats in CSE style:

Author’s Last name, First Initial. Middle Initial. “Article Title.” Journal Title. Year;Volume(Issue):Page numbers.

Example : Smith, J.R. “Evaluation of the Effectiveness of Sustainable Drainage Systems in Urban Areas.” Journal of Environmental Engineering. 2020;146(3):04020010.

Author’s Last name, First Initial. Middle Initial. Book Title. Edition number. Place of Publication: Publisher; Year.

Author’s Last name, First Initial. Middle Initial. “Chapter Title.” In: Editor’s Last name, First Initial. Middle Initial., ed. Book Title. Edition number. Place of Publication: Publisher; Year:Page numbers.

Author’s Last name, First Initial. Middle Initial. “Paper Title.” In: Conference Title; Conference Date; Location. Place of Publication: Publisher; Year.

Example : Smith, J.R. “Sustainable Drainage Systems for Urban Areas.” In: Proceedings of the ASCE International Conference on Sustainable Infrastructure; November 6-9, 2019; Los Angeles, CA. Reston, VA: American Society of Civil Engineers; 2019.

Author’s Last name, First Initial. Middle Initial. “Report Title.” Report number. Place of Publication: Publisher; Year.

Bluebook Style

The Bluebook style is commonly used in the legal field for citing legal documents and sources. Here are the different reference formats in Bluebook style:

Case citation

Case name, volume source page (Court year).

Example : Brown v. Board of Education, 347 U.S. 483 (1954).

Statute citation

Name of Act, volume source § section number (year).

Example : Clean Air Act, 42 U.S.C. § 7401 (1963).

Regulation citation

Name of regulation, volume source § section number (year).

Example: Clean Air Act, 40 C.F.R. § 52.01 (2019).

Book citation

Author’s Last name, First Initial. Middle Initial. Book Title. Edition number (if applicable). Place of Publication: Publisher; Year.

Example: Smith, J.R. Legal Writing and Analysis. 3rd ed. New York, NY: Aspen Publishers; 2015.

Journal article citation

Author’s Last name, First Initial. Middle Initial. “Article Title.” Journal Title. Volume number (year): first page-last page.

Example: Garcia, C. “The Right to Counsel: An International Comparison.” International Journal of Legal Information. 43 (2015): 63-94.

Website citation

Author’s Last name, First Initial. Middle Initial. “Page Title.” Website Title. URL (accessed month day, year).

Example : United Nations. “Universal Declaration of Human Rights.” United Nations. https://www.un.org/en/universal-declaration-human-rights/ (accessed January 3, 2023).

Oxford Style

The Oxford style, also known as the Oxford referencing system or the documentary-note citation system, is commonly used in the humanities, including literature, history, and philosophy. Here are the different reference formats in Oxford style:

Author’s Last name, First name. Book Title. Place of Publication: Publisher, Year of Publication.

Example : Smith, John. The Art of Writing. New York: Penguin, 2020.

Author’s Last name, First name. “Article Title.” Journal Title volume, no. issue (year): page range.

Example: Garcia, Carlos. “The Role of Ethics in Philosophy.” Philosophy Today 67, no. 3 (2019): 53-68.

Chapter in an edited book citation

Author’s Last name, First name. “Chapter Title.” In Book Title, edited by Editor’s Name, page range. Place of Publication: Publisher, Year of Publication.

Example : Lee, Mary. “Feminism in the 21st Century.” In The Oxford Handbook of Feminism, edited by Jane Smith, 51-69. Oxford: Oxford University Press, 2018.

Author’s Last name, First name. “Page Title.” Website Title. URL (accessed day month year).

Example : Jones, David. “The Importance of Learning Languages.” Oxford Language Center. https://www.oxfordlanguagecenter.com/importance-of-learning-languages/ (accessed 3 January 2023).

Dissertation or thesis citation

Author’s Last name, First name. “Title of Dissertation/Thesis.” PhD diss., University Name, Year of Publication.

Example : Brown, Susan. “The Art of Storytelling in American Literature.” PhD diss., University of Oxford, 2020.

Newspaper article citation

Author’s Last name, First name. “Article Title.” Newspaper Title, Month Day, Year.

Example : Robinson, Andrew. “New Developments in Climate Change Research.” The Guardian, September 15, 2022.

AAA (American Anthropological Association) Style

The American Anthropological Association (AAA) style is commonly used in anthropology research papers and journals. Here are the different reference formats in AAA style:

Author’s Last name, First name. Year of Publication. Book Title. Place of Publication: Publisher.

Example : Smith, John. 2019. The Anthropology of Food. New York: Routledge.

Author’s Last name, First name. Year of Publication. “Article Title.” Journal Title volume, no. issue: page range.

Example : Garcia, Carlos. 2021. “The Role of Ethics in Anthropology.” American Anthropologist 123, no. 2: 237-251.

Author’s Last name, First name. Year of Publication. “Chapter Title.” In Book Title, edited by Editor’s Name, page range. Place of Publication: Publisher.

Example: Lee, Mary. 2018. “Feminism in Anthropology.” In The Oxford Handbook of Feminism, edited by Jane Smith, 51-69. Oxford: Oxford University Press.

Author’s Last name, First name. Year of Publication. “Page Title.” Website Title. URL (accessed day month year).

Example : Jones, David. 2020. “The Importance of Learning Languages.” Oxford Language Center. https://www.oxfordlanguagecenter.com/importance-of-learning-languages/ (accessed January 3, 2023).

Author’s Last name, First name. Year of Publication. “Title of Dissertation/Thesis.” PhD diss., University Name.

Example : Brown, Susan. 2022. “The Art of Storytelling in Anthropology.” PhD diss., University of California, Berkeley.

Author’s Last name, First name. Year of Publication. “Article Title.” Newspaper Title, Month Day.

Example : Robinson, Andrew. 2021. “New Developments in Anthropology Research.” The Guardian, September 15.

AIP (American Institute of Physics) Style

The American Institute of Physics (AIP) style is commonly used in physics research papers and journals. Here are the different reference formats in AIP style:

Example : Johnson, S. D. 2021. “Quantum Computing and Information.” Journal of Applied Physics 129, no. 4: 043102.

Example : Feynman, Richard. 2018. The Feynman Lectures on Physics. New York: Basic Books.

Example : Jones, David. 2020. “The Future of Quantum Computing.” In The Handbook of Physics, edited by John Smith, 125-136. Oxford: Oxford University Press.

Conference proceedings citation

Author’s Last name, First name. Year of Publication. “Title of Paper.” Proceedings of Conference Name, date and location: page range. Place of Publication: Publisher.

Example : Chen, Wei. 2019. “The Applications of Nanotechnology in Solar Cells.” Proceedings of the 8th International Conference on Nanotechnology, July 15-17, Tokyo, Japan: 224-229. New York: AIP Publishing.

Example : American Institute of Physics. 2022. “About AIP Publishing.” AIP Publishing. https://publishing.aip.org/about-aip-publishing/ (accessed January 3, 2023).

Patent citation

Author’s Last name, First name. Year of Publication. Patent Number.

Example : Smith, John. 2018. US Patent 9,873,644.

References Writing Guide

Here are some general guidelines for writing references:

• Follow the citation style guidelines: Different disciplines and journals may require different citation styles (e.g., APA, MLA, Chicago). It is important to follow the specific guidelines for the citation style required.
• Include all necessary information : Each citation should include enough information for readers to locate the source. For example, a journal article citation should include the author(s), title of the article, journal title, volume number, issue number, page numbers, and publication year.
• Use proper formatting: Citation styles typically have specific formatting requirements for different types of sources. Make sure to follow the proper formatting for each citation.
• Order citations alphabetically: If listing multiple sources, they should be listed alphabetically by the author’s last name.
• Be consistent: Use the same citation style throughout the entire paper or project.
• Check for accuracy: Double-check all citations to ensure accuracy, including correct spelling of author names and publication information.
• Use reputable sources: When selecting sources to cite, choose reputable and authoritative sources. Avoid sources that are biased or unreliable.
• Include all sources: Make sure to include all sources used in the research, including those that were not directly quoted but still informed the work.
• Use online tools : There are online tools available (e.g., citation generators) that can help with formatting and organizing references.

Purpose of References in Research

References in research serve several purposes:

• To give credit to the original authors or sources of information used in the research. It is important to acknowledge the work of others and avoid plagiarism.
• To provide evidence for the claims made in the research. References can support the arguments, hypotheses, or conclusions presented in the research by citing relevant studies, data, or theories.
• To allow readers to find and verify the sources used in the research. References provide the necessary information for readers to locate and access the sources cited in the research, which allows them to evaluate the quality and reliability of the information presented.
• To situate the research within the broader context of the field. References can show how the research builds on or contributes to the existing body of knowledge, and can help readers to identify gaps in the literature that the research seeks to address.

Importance of References in Research

References play an important role in research for several reasons:

• Credibility : By citing authoritative sources, references lend credibility to the research and its claims. They provide evidence that the research is based on a sound foundation of knowledge and has been carefully researched.
• Avoidance of Plagiarism : References help researchers avoid plagiarism by giving credit to the original authors or sources of information. This is important for ethical reasons and also to avoid legal repercussions.
• Reproducibility : References allow others to reproduce the research by providing detailed information on the sources used. This is important for verification of the research and for others to build on the work.
• Context : References provide context for the research by situating it within the broader body of knowledge in the field. They help researchers to understand where their work fits in and how it builds on or contributes to existing knowledge.
• Evaluation : References provide a means for others to evaluate the research by allowing them to assess the quality and reliability of the sources used.

Advantages of References in Research

There are several advantages of including references in research:

• Acknowledgment of Sources: Including references gives credit to the authors or sources of information used in the research. This is important to acknowledge the original work and avoid plagiarism.
• Evidence and Support : References can provide evidence to support the arguments, hypotheses, or conclusions presented in the research. This can add credibility and strength to the research.
• Reproducibility : References provide the necessary information for others to reproduce the research. This is important for the verification of the research and for others to build on the work.
• Context : References can help to situate the research within the broader body of knowledge in the field. This helps researchers to understand where their work fits in and how it builds on or contributes to existing knowledge.
• Evaluation : Including references allows others to evaluate the research by providing a means to assess the quality and reliability of the sources used.
• Ongoing Conversation: References allow researchers to engage in ongoing conversations and debates within their fields. They can show how the research builds on or contributes to the existing body of knowledge.

About the author

Muhammad Hassan

Researcher, Academic Writer, Web developer

You may also like

Data Analysis – Process, Methods and Types

Research Questions – Types, Examples and Writing...

Limitations in Research – Types, Examples and...

Research Approach – Types Methods and Examples

Dissertation – Format, Example and Template

Research Methodology – Types, Examples and...

References provide the information necessary for readers to identify and retrieve each work cited in the text .

Check each reference carefully against the original publication to ensure information is accurate and complete. Accurately prepared references help establish your credibility as a careful researcher and writer.

Consistency in reference formatting allows readers to focus on the content of your reference list, discerning both the types of works you consulted and the important reference elements (who, when, what, and where) with ease. When you present each reference in a consistent fashion, readers do not need to spend time determining how you organized the information. And when searching the literature yourself, you also save time and effort when reading reference lists in the works of others that are written in APA Style.

Academic Writer ®

Master academic writing with APA’s essential teaching and learning resource

Course Adoption

Teaching APA Style? Become a course adopter of the 7th edition Publication Manual

Instructional Aids

Guides, checklists, webinars, tutorials, and sample papers for anyone looking to improve their knowledge of APA Style

• Study and research support
• Referencing

Referencing explained

Why and when to reference.

Referencing is an important part of academic work. It puts your work in context, demonstrates the breadth and depth of your research, and acknowledges other people’s work. You should reference whenever you use someone else’s idea.

View video using Microsoft Stream (link opens in a new window, available for University members only)

These webpages explain what referencing is, why it is important and give an overview of the main elements of how to reference. Our Referencing made simple tutorial opens in a new window and covers how to identify your source and create a reference with interactive examples.

Why reference?

Referencing correctly:

• helps you to avoid plagiarism by making it clear which ideas are your own and which are someone else’s
• shows your understanding of the topic
• gives supporting evidence for your ideas, arguments and opinions
• allows others to identify the sources you have used.

When to reference

Whenever you use an idea from someone else's work, for example from a journal article, textbook or website, you should cite the original author to make it clear where that idea came from. This is the case regardless of whether you have paraphrased, summarised or directly quoted their work. This is a key part of good practice in academic writing.

Read more on:

• academic integrity
• quoting, summarising, paraphrasing, and synthesising
• citing direct quotations in Leeds Harvard or citing direct quotations in Leeds Numeric styles.

University and school policies

The University referencing policy (PDF) sets out the referencing requirements that all taught students and tutors are expected to follow.

Each school in the University requires students to use a specific style of referencing. Check the referencing style used in your school before you begin.

All your citations and references should match the style you are using exactly, including any punctuation, capitalisation, italics and bold, and you should use the same referencing style throughout your assignment.

✨ Enrol by 3 July to get access to our Summer Writing Accelerator at no additional cost! 💰 ✨

The Ultimate Guide to References and Citations

Learn which studies to choose as a reference in your paper, what biases to check for, if self-citation is okay, and the ultimate way to reference studies in the text.

We build knowledge through citing other researchers’ work. In this way, science can move forward. When someone reads your study, they don’t only want to know about your results, but they also want an objective introduction to the context of your research: What is your field about, what are the issues and how do your findings relate to other studies? Although references to literature are crucial in scientific publishing, there seems to be little education about the topic. As a consequence, sloppy citation practices have become all too common.

This post provides some guidance on how to select studies to reference in your research and how to cite them in order to write a paper that gets accepted in a high-impact journal .

1. How many and which studies should I cite?

You’re likely overwhelmed with studies you could mention in the introduction section. When selecting literature to cite, the first step is to read them. Sounds obvious, right? Fact is that some scientific myths have spread for decades because authors weren’t careful enough when referencing other studies. So, if you see a reference in another paper that seems important to include, make sure the study actually makes the points the author said it would. Many people tend to cite secondary sources, meaning papers that refer to yet another source to support a statement. Therefore, make sure to trace back the original work. This ensures both that your reference is correct and that your citation honours the right author.

Once you have identified some candidates you want to cite, shortlist even further. Choose only those references that are crucial to understand the context of your study. You don’t need to go too broad. For a general introduction to your field, the way to go might be a recent review paper, even though a secondary source. That could spare you ten or so separate references.

2.  How do I make sure my selection is objective?

When you think about the studies to cite in your paper, be aware of your own bias. You may want to ask yourself, whether you only want to cite a certain study because the author cited yours last time? Or perhaps a personal relationship with the author makes you favour their paper?

That’s one type of unconscious bias you could have. Another is geographical. Computer scientists found that authors have a location bias: They prefer to cite studies from research groups nearby. Therefore, make sure to check journals that might be more popular on other continents or countries.  

You can also have a confirmation bias. It is easy to neglect studies that disagree with your approach or findings. That’s a mistake! It makes you far more credible if you can show that you are aware of disagreements and have an answer to them. There might also be a study in the literature jungle that has been neglected because its findings contradict the common belief in a field. Make sure unpopular opinions don’t get lost – because who knows who’s right?

3. Can I cite myself?

Do definitely cite your own work if you have contributed to the field. Your reader, particularly if she’s your editor, will want to know about your previous experience of the topic area. One thing I’ve noticed when I edit my clients’ manuscripts is that many authors use a passive voice even when they talk about their own work, such as “it has been previously observed…”. But you are missing a chance here! Why not write: “Previously, we have observed…” or “In one of our previous studies, we found that…”? When you choose some of your own studies to cite, the same principle applies as for the studies of others: Be selective. Excessive self-citation puts readers off and is just unfair.

4. What’s the best way to cite studies?

Think about your reader when you describe the literature. They want to know what these studies found, not what their aim or approach was. While you talk about other people’s findings, do check how certain they are. Be critical. If a study only measured one data point, it’s hardly a fact, and you shouldn’t present it as such. Here it comes down to choosing the right verb. In some cases, “indicate”, “speculate” or “assume” are more accurate than “is”.

I know, reference lists are something you sometimes leave to the evening before a submission deadline. Creating a reference list can be dreadful. I recommend to use a reference manager software during your whole paper writing process, which can speed up the process at the end.

A little pro tip: You save your editors time if you have made an effort to adjust your references to their style. Every journal uses slightly varying reference styles, so make sure you know what yours requires. If you are using a reference manager, this isn’t even a hard task: Journals usually supply you with a style file that you can integrate in your software to generate the requested reference list. 

If the reference style allows, do include the title of papers. That’ll save your reader’s time. Do you know what else does? Checking and double checking that all letters and numbers are correct in your reference list.

There you have it. Apart from having your references in check, there are some other elements a high-impact paper needs, such as story-telling , simple and easy-to-understand figures , a concise abstract and an informative title .

What’s your experience with citing and references? How much thought are you giving them? What part do you struggle with most? Please let me know in a comment below or send me a message. 

Further resources:

• My blog post on unethical citation practices published on Elevate Science
• Ethical publication guidelines by the International Committee of Medical Journal Editors

Before you go… Register for a private viewing session of this free training for science researchers.

Share article

© Copyright 2018-2024 by Anna Clemens. All Rights Reserved. 

Photography by Alice Dix

• Locations and Hours
• UCLA Library
• Research Guides
• Research Tips and Tools

Articles, Books and . . . ? Understanding the Many Types of Information Found in Libraries

• Reference Sources

Encyclopedias

Dictionaries, almanacs and yearbooks, handbooks and manuals.

• Documents and Reports
• Non-Text Content
• Archival Materials

Reference sources are generally the place to begin your research, especially when you're starting out with an unfamiliar field. But they're also where you return when you need to look up formulas, facts, definitions, and other standard details; they tend to pack a lot of information into simple, easy-to-use packages.

Physical Media

Many reference works are available online and are accessible through links from the Library Catalog and from subject or course guides , but many valuable reference resources are still available only in print, and a few highly specialized tools are on microform or CD. Because print-only reference books are in high demand, they are kept in separate, non-circulating reference collections in most UCLA libraries.

Scholarly Sources

Reference sources are rarely peer-reviewed. In fact, because they mostly contain established, factual information, they're sometimes not even cited in academic works, unless directly quoted. Check your style manual for best guidelines.

Primary or Secondary Sources

As compilations of existing information, reference works are decisively in the category of secondary sources... to the point that some people call them tertiary sources .

Encyclopedias attempt to provide comprehensive summaries of knowledge in either a specific field (subject encyclopedias) or "everything" (general encyclopedias). Encyclopedias are typically divided into a collection of articles on discrete topics. Academically oriented encyclopedias will often include short bibliographies, making them a good resource for identifying key books and articles on a topic.

• Online Encyclopedias
• Finding Print Encyclopedias
• Subject dictionaries: define technical terms in specific fields, sometimes in as much detail as an encyclopedia
• Bilingual dictionaries: define words in a different language
• Thesauri: provide synonyms
• Rhyming dictionariess
• Major Online Dictionaries
• Finding Dictionaries (both online and in print)
• World Almanac and Book of Facts
• Statistical Yearbooks

Both "handbook" and "manual" refer to the traditional small size of the volumes, designed to fit in one hand for ease of use. Despite this origin, many modern handbooks are quite hefty!

• Databases containing online handbooks
• Find print handbooks in the Library Catalog Search "[subject] handbooks" within Subject List.
• << Previous: Home
• Next: Books >>
• Last Updated: May 16, 2024 10:30 AM
• URL: https://guides.library.ucla.edu/content-types

Citing and referencing

What is referencing.

• What is Plagiarism?
• In-text Citations
• Periodicals
• Reference Works
• Theses and Dissertations
• Conference Papers and Proceedings
• Chicago Style
• Vancouver Style
• Styles by UOB Departments

• Graduate Services Librarian

Phone ext. 4044

[email protected]

• Public Services Librarian

Phone ext. 4142

[email protected]

Referencing is a standardized method of formatting the sources you have used in your written works. Any given referencing style serves many purposes: 

• Establishing credibility by citing reliable sources.
• Preventing plagiarism by giving credit to original creators. 
• Building on and connecting with existing knowledge. 
• Enabling verification and reproducibility of research. 
• Abiding and demonstrating intellectual property rights and standards.
• Provide evidence to support your arguments. 
• Acknowledging diverse perspectives and sources in your work. 

1. In-text citations

In-text citations are important, they consist of mentioning a specific source used in the body of the work. The format of the citation may change depending on the style you are using (e.g. MLA, APA, Chicago, etc.), yet there are some basic elements you need to include, which are: 

• Name of the author(s)
• Year of publication

You must add the exact page number in your citation if you used a direct quote from a source. 

2. List of references

This is  a list of the sources you have cited in the text  at the end of your paper. It is not a list of “works consulted”. Every source listed in your references list must also be cited in the body of your work and vice versa. 

• Next: What is Plagiarism? >>
• Last Updated: Jun 4, 2024 11:33 AM
• URL: https://balamand.libguides.com/citing

Educational resources and simple solutions for your research journey

How to Write References in Research Papers: Navigating the Maze (Part 2)

You truly are navigating a maze when it comes to ci tations and the question of how to write references in research pa per s. In part 1 of this article, we touched upon citations, which are pointers embedded in the text of a research paper, to sources of information or to other research relevant to that being described in the research paper. Those pointers lead to references in research papers , which typically appear at the end of the text. Whereas citations merely point us to sources in research papers, references describe those sources in sufficient detail for readers (1) to know the title of each source, who is responsible for its content, and when it was published; (2) to look up those sources; and (3) to obtain the documents in question if required.  

Table of Contents

Types of references in research papers

In writing a research paper, a researcher draws upon many sources of information, knowledge, opinions, and so on. One of the the most common type s of reference s  in research papers is other research papers published in journals; other common sources include technical reports, handbooks, presentations at conferences, and books. Increasingly, the sources in research papers are digital and include web pages, databases, blog posts, and even tweets and emails.  

Not all sources are considered equally credible , and some may not be accessible to all because they are behind paywalls or available only to members of a network (company intranets, for example) or because they are personal exchanges.  

How to write references in research papers

If the citations follow the Harvard system, references in a research paper s are sorted alphabetically by the last name of the first author; if the citations follow the Vancouver system, the references are arranged by numbers: the reference corresponding to the first numbered citation is numbered 1, and so on. If a source is cited again, its allocated number does not change.  

Some additional conventions govern the alphabetic sorting of references in research papers . For instance, when authors have some papers in which they are the only author and others in which they have one or more co-authors or when the same author or authors have papers published in different years or even within the same year.  

Some publishers make even greater demands of references in research papers : authors are expected to sort the list of references alphabetically, as in the Harvard system; then number the sorted list serially; and then renumber all the citations within the text so that each corresponds to its new number!  

How to add references in a research paper: Key elements

For a source of information to be described accurately, some minimum details are required. Here’s one example of w rit ing references in research paper s – ‘ Nature 171 : 737’ is a code that, if you know how to decipher it, tells you that it means an article published in Nature (a weekly journal published from the UK) that begins on page 737 of volume 171 of that journal. However, it does not tell you what the article was about, who wrote it, when it was published, or even how long it is. A complete reference in research paper s (Fig. 1), however, tells you that the title of the article was ‘Molecular structure of nucleic acids: a structure for deoxyribose nucleic acid’, that it was written by J. D. Watson and F. H. C. Crick, that it was published in 1953, and that it ran to no more than two pages.  

When thinking about how to write research references , remember that the elements that make up a reference to an article published in a journal are different from those that make up a reference to a book (edition if not the first, the publisher, and the place of publication, although the last is no longer considered essential in today’s globalized publishing). The elements that make up a reference to a technical report include the name of the organization issuing that report and the report number, if any, and that to a conference presentation gives the title of the conference, the date(s) on which it was held and the place, the name of the organizer(s) of the conference, and so on.  

Note that journals or publishers differ in the elements they expect authors to include when they state how to put references in research papers ; for example, some journals give only minimal information and exclude the titles of articles and some use the ‘elided’ form of page numbers (737–38 instead of 737–738, for example).  

Then there is the question of abbreviated names of journals: some publishers abbreviate journal titles and some don’t ( Annals of Applied Biology or Ann. Appl. Biol.). And those who do, often disagree on the correct abbreviation—and on whether the abbreviations should end in dots (whether the word ‘Journal’ should be given as J. or J or Jnl or Jnl.).  

Sequence of the elements that make up reference s in research papers  

Publishers and journals also differ in the order or sequence in which they present the elements or components of reference s in research papers : usually, British and European publishers put the year of publication after the names of authors whereas US publishers move the year closer to the volume number of the journal.  

Even within an element, the sequence of references in research paper s can have subtle differences. In Harvard system, because the last name of the first author is using for sorting, the name is ‘inverted’, that is the last name is given first, followed by initials (Watson J D instead of J D Watson). However, some journals invert the names of all the authors whereas some invert the name of only the first author. In Vancouver system, the names are seldom inverted because the sequence is not alphabetical.  

Punctuation to separate the elements that make up references in research papers

The many exasperating details that go into formatting references include punctuation marks (or their absence). In giving the initials of authors, some journals use dots, some journals use space, some use both, and some use neither (Watson J.D. or Watson J D or Watson J. D. or Watson JD). Some use a comma between the last name and the initials whereas some reserve the comma only to separate one name from the next (Watson, J D and Crick, F H C or Watson J D, Crick F H C). Some use ‘and’ some don’t, even when there are only two authors, and some use ‘&’ instead which makes it even more confusing for those struggling with how to write references in a research paper.

When the place of publication was a required element in the case of books, some publishers used the colon and some used the comma (and also changed the order, as in New York: Harper & Row or Harper & Row, New York). Some publishers end each reference with a full stop (period) and some don’t.

Typography of references in research papers  

As if the variations mentioned above were not enough, when figuring out how to add references in a research paper , you also have to contend with the differences in typography as well: journal titles in italics or in normal type, volume numbers in bold or in normal type, hyphens or en dashes between page numbers (737-738 or 737–738), and so on.  

All is not lost, however, if you despair of ever getting the references in a research paper right. For example, some publishers now insist on correct formatting only after a paper has been accepted for publication. Also, ICMJE, the International Committee of Medical Journal Editors, recommends a set of uniform requirements for manuscripts (the requirements include the formatting of citations and references), and hundreds of medical journals ( www.icmje.org/journals-following-the-icmje-recommendations/ ) have agreed that as long as authors adhere to those recommendations on how to mention references for research papers , any changes to the formatting any journal wants to make will be made by the journal in question.  

Lastly, several software packages help authors to automate this mundane task of consistent formatting of references in research paper s—but that is another article and another day.  

The details involved in using citations and references correctly can be overwhelming for some of us. While this article covers the key tips to help you understand how to give reference s in research paper s , be sure to check out article 1 of this two-part series for more on what, when and how to cite in a research paper. One way to check whether these are handled correctly in your manuscript is to use Researcher.Life’s AI powered manuscript optimizer , which can flag any discrepancies, departures from standard style, and mismatches between citations and references in research paper s.  

R Discovery is a literature search and research reading platform that accelerates your research discovery journey by keeping you updated on the latest, most relevant scholarly content. With 250M+ research articles sourced from trusted aggregators like CrossRef, Unpaywall, PubMed, PubMed Central, Open Alex and top publishing houses like Springer Nature, JAMA, IOP, Taylor & Francis, NEJM, BMJ, Karger, SAGE, Emerald Publishing and more, R Discovery puts a world of research at your fingertips.  

Try R Discovery Prime FREE for 1 week or upgrade at just US$72 a year to access premium features that let you listen to research on the go, read in your language, collaborate with peers, auto sync with reference managers, and much more. Choose a simpler, smarter way to find and read research – Download the app and start your free 7-day trial today !  

Related Posts

How R Discovery is Transforming Research Reading Habits for Academics in Brazil

Article Processing Charges: Impact on Open Access Publishing

References: How to Cite and List Correctly

• First Online: 25 February 2021

Cite this chapter

• C. George Thomas 2  

4797 Accesses

When we write an essay, research paper, thesis, or book, it is normal to include information from the work of others or support our arguments by reference to other published works. All such academic documents draw heavily on the ideas and findings of previous and current researchers available through various sources such as books, journals, theses, newspapers, magazines, government reports, or Internet sources. In all these cases, proper referencing is essential in order to ensure easy retrieval of information. Referencing is the name given to the method of showing and acknowledging the sources from which the author has obtained ideas or information.

Everything deep is also simple and can be reproduced simply as long as its reference to the whole truth is maintained. But what matters is not what is witty but what is true. Albert Schweitzer (1875–1965)

This is a preview of subscription content, log in via an institution to check access.

Access this chapter

• Available as PDF
• Read on any device
• Instant download
• Own it forever
• Available as EPUB and PDF
• Compact, lightweight edition
• Dispatched in 3 to 5 business days
• Free shipping worldwide - see info
• Durable hardcover edition

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Bibliography

AMA [American Medical Association] 2007. AMA Manual of Style: A Guide for Authors and Editors (10 th Ed.). Oxford University Press, New York, 1024p.

Google Scholar  

APA [American Psychological Association] 2013. Publication Manual of the American Psychological Association (6 th Ed.). American Psychological Association, Washington DC, 272p.

Barrass, R. 2002. Scientists Must Write: A Guide to Better Writing for Scientists, Engineers and Students (2 nd Ed.). Routledge, London, 204p.

BSI [British Standards Institution] 1990. Recommendations for Citing and Referencing Published Material (2 nd Ed.). British Standards Institution (BS 5605:1990), 8p.

Cargill, M. and O’Connor, P. 2009. Writing Scientific Research Articles: Strategy and Steps. Wiley-Blackwell, 173p.

COA [Commonwealth of Australia] 2002. Style Manual for Authors, Editors and Printers (6 th Ed.). John Wiley & Sons, Brisbane, Australia, 550 p.

Cousens, R. 1987. Theory and reality of weed control thresholds. Plant Prot. Q. 2: 13–20

Craswell, G. 2004. Writing for Academic Success: A Postgraduate Guide. Sage Publications, London, 271p.

CSE [Council of Science Editors] 2014. Scientific Style and Format: The CSE Manual for Authors, Editors, and Publishers (8 th Ed.): The University of Chicago Press, 722p.

Davis, M. 2005. Scientific Papers and Presentations. Academic Press, Massachusetts, 356p.

Day, R.A. and Gastel, B. 2006. How to Write and Publish a Scientific Paper (6 th Ed.). Cambridge University Press, Cambridge, 320p.

Gustavii, B. 2008. How to Write and Illustrate a Scientific Paper (2 nd Ed.). Cambridge University press, Cambridge, 168p.

Hakansson, S. 1983. Competition and Production in Short-lived Crop Weed Stands: Density Effects. Report 127, Swedish University of Agricultural Sciences, Uppasala, 85p.

ISSN [International Standards Serial Number].2017. The List of title word abbreviations [on line]. Available: http://www.issn.org/services/onlineservices/access-to-the-ltwa/ [Accessed 06 May 2018].

Leggett, G., Mead, C.D., Kramer, M.G., and Beal, R.S. 1985. Handbook for Writers. Prentice- Hall, New Jersey, 558p.

Lester, J.D. and Lester, J.D. Jr. 2009. Writing Research Papers: A Complete Guide (13 th Ed.), Longman, 416p.

MLA [Modern Language Association of America] 2016. MLA Handbook for Writers of Research Papers (8 th Ed.) Modern Language Association of America, 146p

Neville, C. 2010. The Complete Guide to Referencing and Avoiding Plagiarism (2 nd Ed.). McGraw-Hill Education, 288 p.

Renu, S. and Thomas, C.G. 2000. Stale seedbed technique for the management of Sacciolepis interrupta in semi-dry rice. Indian J. Weed Sci. 32(3&4):146-149

Rubens, P. 2004. Science and Technical writing: A Manual of Style (2 nd Ed.). Routledge, New York, 427p.

Sheng, T.C. 1989. Soil Conservation for Small Farmers in the Humid Tropics. FAO Soils Bulletin No. 60. Food and Agriculture Organization of the United Nations, Rome, 104p

Thomas, C. G. 2008. Forage Crop Production in the Tropics. Kalyani Publishers, Ludhiana, 333p.

Thomas, C. G., Abraham, C. T., and Sreedevi, P. 1997. Weed flora and their relative dominance in semi-dry rice culture. J. Trop. Agric. 35:51–53.

Thomas, C. G. 2010. Land Husbandry and Watershed Management. Kalyani Publishers, Ludhiana, 716p.

Turabian, K.L. 2007. A Manual for Writers of Term Papers, Theses, and Dissertations (9 th Ed.). The University of Chicago Press, 461p.

UChicago [University of Chicago] 2017. Chicago Manual of Style [CMOS].

University of Chicago Press, 1144p.

AMA style- http://www.amamanualofstyle.com/oso/public/index.html .

APA style- https://www.apastyle.org/manual

BibSonomy- https://www.bibsonomy.org/

Bibus - http://www.sourceforage.net/projects/bibus-biblio

Bookends-www. sonnysoftware.com/

CAS Source index search tool- http://cassi.cas.org/search.jsp

Citavi- https://citavi.com/

CiteULike- http://www.citeulike.org

EndNote- https://www.endnote.com/

FAO [Food and Agriculture Organization of the United Nations]. 2018. The State of Food Security and Nutrition in the World 2018. Building Climate Resilience for Food Security and Nutrition. FAO, Rome, 181p. Available: http://www.fao.org/3/I9553EN/i9553en.pdf [Accessed 10 Oct. 2018].

Genamics JournalSeek- http://journalseek.net/

JabRef - http://jabref.sourceforge.net/

Journal database, NCBI- http://www.ncbi.nlm.nih.gov/nlmcatalog/journals ISSN, LTWA- http://www.issn.org/2-22661-LTWA-online.php .

Mendeley- https://www.mendeley.com

Papers- http://www.readcube.com/papers/

Qiqqa - http://www.qiqqa.com/

Refbase- http://www.refbase.net

RefWorks- https://www.proquest.com/

Scientific Style and Format (CSE) - https://www.scientificstyleandformat.org/Home.html

SciRef- http://sci-progs.com/

Web of Science journal abbreviations- https://images.webofknowledge.com/ images/help/WOS/A_abrvjt.html

Wikindx - https://sourceforge.net/ projects/wikindx/

Zotero- https://www.zotero.org/

Download references

Author information

Authors and affiliations.

Kerala Agricultural University, Thrissur, Kerala, India

C. George Thomas

You can also search for this author in PubMed   Google Scholar

Corresponding author

Correspondence to C. George Thomas .

Rights and permissions

Reprints and permissions

Copyright information

© 2021 The Author(s)

About this chapter

Thomas, C.G. (2021). References: How to Cite and List Correctly. In: Research Methodology and Scientific Writing . Springer, Cham. https://doi.org/10.1007/978-3-030-64865-7_15

Download citation

DOI : https://doi.org/10.1007/978-3-030-64865-7_15

Published : 25 February 2021

Publisher Name : Springer, Cham

Print ISBN : 978-3-030-64864-0

Online ISBN : 978-3-030-64865-7

eBook Packages : Education Education (R0)

Share this chapter

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Publish with us

Policies and ethics

• Find a journal
• Track your research

Library Services

UCL LIBRARY SERVICES

• Guides and databases
• Library skills

References, citations and avoiding plagiarism

• Understanding a reference
• Getting Started
• Assignments
• Independent research
• Managing your references
• How to reference
• Acknowledging and referencing AI
• Harvard referencing
• Vancouver referencing
• APA referencing
• Chicago referencing
• OSCOLA referencing
• MHRA referencing
• MLA referencing
• Avoiding plagiarism
• Further help

Understanding a reference: Working out what it is

When conducting library research, you will often need to find material based on the information in a reference list or reading list. In order to then locate that source, you will need to understand what the different elements of the reference mean and identify what type of source it is.

A reference is made up of some key details/elements about the source, that can broadly be broken down into the following:

• Who has ‘made’ the item (the author, creator, compiler…)?
• What is it called?
• If part of a larger work, what is that called?
• Where was (is) it disseminated/published?
• Who is responsible for the dissemination / publishing?
• When was it disseminated / published?
• A direct quotation, or allusion, should always include the page number(s).

You can normally identify the type of source by the information presented in the reference. Here are some tips for identifying some of the key types of source that you might come across on a reading/reference list:

Cooke, A. (2001) A guide to finding quality information on the Internet: selection and evaluation strategies. 2nd ed. London: Library Association Publishing.

Note the edition information, the publisher location (London) and publisher name (Library Association Publishing) which are included in the reference.

Journal article

De Pinto, M., Jelacic, J., Edwards, W.T. (2008) Very-low-dose ketamine for the management of pain and sedation in the ICU. Acute pain [online], 10 (2), p. 100. Available at: doi:10.1016/j.acpain.2008.05.023 [Accessed 8 September 2008].

Note that the title of both the article ('Very low dose ketamine ... ') and the journal ('Acute pain') are included. The journal volume (10) and issue number (2) are also included.

• << Previous: Independent research
• Next: Managing your references >>
• Last Updated: Apr 18, 2024 5:20 PM
• URL: https://library-guides.ucl.ac.uk/referencing-plagiarism

University Library

Start your research.

• Research Process
• Find Background Info
• Find Sources through the Library
• Evaluate Your Info
• Cite Your Sources
• Evaluate, Write & Cite

• is the right thing to do  to give credit to those who had the idea
• shows that you have read and understand  what experts have had to say about your topic
• helps people find the sources  that you used in case they want to read more about the topic
• provides   evidence  for your arguments
• is professional and  standard practice   for students and scholars

What is a Citation?

A citation identifies for the reader the original source for an idea, information, or image that is referred to in a work.

• In the body of a paper, the  in-text citation  acknowledges the source of information used.
• At the end of a paper, the citations are compiled on a  References  or  Works Cited  list. A basic citation includes the author, title, and publication information of the source. 

From:  Lemieux  Library,  University  of Seattle 

Why Should You Cite?

Quoting Are you quoting two or more consecutive words from a source? Then the original source should be cited and the words or phrase placed in quotes. 

Paraphrasing If an idea or information comes from another source,  even if you put it in your own words , you still need to credit the source.  General vs. Unfamiliar Knowledge You do not need to cite material which is accepted common knowledge. If in doubt whether your information is common knowledge or not, cite it. Formats We usually think of books and articles. However, if you use material from web sites, films, music, graphs, tables, etc. you'll also need to cite these as well.

Plagiarism is presenting the words or ideas of someone else as your own without proper acknowledgment of the source. When you work on a research paper and use supporting material from works by others, it's okay to quote people and use their ideas, but you do need to correctly credit them. Even when you summarize or paraphrase information found in books, articles, or Web pages, you must acknowledge the original author.

Citation Style Help

Helpful links:

• MLA ,  Works Cited : A Quick Guide (a template of core elements)
• CSE  (Council of Science Editors)

For additional writing resources specific to styles listed here visit the  Purdue OWL Writing Lab

Citation and Bibliography Resources

• How to Write an Annotated Bibliography
• << Previous: Evaluate Your Info
• Next: Evaluate, Write & Cite >>

Creative Commons Attribution 3.0 License except where otherwise noted.

Land Acknowledgement

The land on which we gather is the unceded territory of the Awaswas-speaking Uypi Tribe. The Amah Mutsun Tribal Band, comprised of the descendants of indigenous people taken to missions Santa Cruz and San Juan Bautista during Spanish colonization of the Central Coast, is today working hard to restore traditional stewardship practices on these lands and heal from historical trauma.

The land acknowledgement used at UC Santa Cruz was developed in partnership with the Amah Mutsun Tribal Band Chairman and the Amah Mutsun Relearning Program at the UCSC Arboretum .

• Directories
• What are citations and why should I use them?
• When should I use a citation?
• Why are there so many citation styles?
• Which citation style should I use?
• Chicago Notes Style
• Chicago Author-Date Style
• AMA Style (medicine)
• Bluebook (law)
• Additional Citation Styles
• Built-in Citation Tools
• Quick Citation Generators
• Citation Management Software
• Start Your Research
• Research Guides
• University of Washington Libraries
• Library Guides
• UW Libraries
• Citing Sources

Citing Sources: What are citations and why should I use them?

What is a citation.

Citations are a way of giving credit when certain material in your work came from another source. It also gives your readers the information necessary to find that source again-- it provides an important roadmap to your research process. Whenever you use sources such as books, journals or websites in your research, you must give credit to the original author by citing the source. 

Why do researchers cite?

Scholarship is a conversation  and scholars use citations not only to  give credit  to original creators and thinkers, but also to  add strength and authority  to their own work.  By citing their sources, scholars are  placing their work in a specific context  to show where they “fit” within the larger conversation.  Citations are also a great way to  leave a trail  intended to help others who may want to explore the conversation or use the sources in their own work.

In short, citations

(1) give credit

(2) add strength and authority to your work

(3) place your work in a specific context

(4) leave a trail for other scholars

"Good citations should reveal your sources, not conceal them. They should honeslty reflect the research you conducted." (Lipson 4)

Lipson, Charles. "Why Cite?"  Cite Right: A Quick Guide to Citation Styles--MLA, APA, Chicago, the Sciences, Professions, and More . Chicago: U of Chicago, 2006. Print.

What does a citation look like?

Different subject disciplines call for citation information to be written in very specific order, capitalization, and punctuation. There are therefore many different style formats. Three popular citation formats are MLA Style (for humanities articles) and APA or Chicago (for social sciences articles).

MLA style (print journal article):  

Whisenant, Warren A. "How Women Have Fared as Interscholastic Athletic Administrators Since the Passage of Title IX." Sex Roles Vol. 49.3 (2003): 179-182.

APA style (print journal article):

Whisenant, W. A. (2003) How Women Have Fared as Interscholastic Athletic Administrators Since the Passage of Title IX. Sex Roles , 49 (3), 179-182.

Chicago style (print journal article):

Whisenant, Warren A. "How Women Have Fared as Interscholastic Athletic Administrators Since the Passage of Title IX." Sex Roles 49, no. 3 (2003): 179-182.

No matter which style you use, all citations require the same basic information:

• Author or Creator
• Container (e.g., Journal or magazine, website, edited book)
• Date of creation or publication
• Publisher 

You are most likely to have easy access to all of your citation information when you find it in the first place. Take note of this information up front, and it will be much easier to cite it effectively later.

• << Previous: Basics of Citing
• Next: When should I use a citation? >>
• Last Updated: May 1, 2024 12:48 PM
• URL: https://guides.lib.uw.edu/research/citations

Referencing

Referencing is one of the most important aspects of any academic research and poor or lack of referencing will not only diminish your marks, but such practices may also be perceived as plagiarism by your university and disciplinary actions may follow that may even result in expulsion from the course.

Difference between References and Bibliography

It is very important to be able to distinguish between References and Bibliography. Under References you list resources that you referred to within the body of the work that also include quotations.  For example,

It has been noted that “time and the management of time is an important issue, and the supply of time management products – books, articles, CDs, workshops, etc. – reflects the huge demand for these products” (Walsh, 2007, p.3).

Interchangeability of identical parts and a high level of straightforwardness of attaching these parts through the assembly line can be considered as revolutionary components of Fordism for the first part of the 20 th century (Nolan, 2008).

Under Bibliography, on the other hand, you need to list resources that you have read during the research process in order to widen your knowledge about the research area , but specific piece of information from these resources have not been used in your research in the direct manner. You do not need to refer to Bibliography within the body of the text.

There are various methods of referencing such as Harvard, APA and Vancouver referencing systems. You should check with your dissertation handbook for the exact type of referencing required and follow this requirement thoroughly.

John Dudovskiy

• Walden University
• Faculty Portal

Reference List: Common Reference List Examples

Article (with doi).

Alvarez, E., & Tippins, S. (2019). Socialization agents that Puerto Rican college students use to make financial decisions. Journal of Social Change , 11 (1), 75–85. https://doi.org/10.5590/JOSC.2019.11.1.07

Laplante, J. P., & Nolin, C. (2014). Consultas and socially responsible investing in Guatemala: A case study examining Maya perspectives on the Indigenous right to free, prior, and informed consent. Society & Natural Resources , 27 , 231–248. https://doi.org/10.1080/08941920.2013.861554

Use the DOI number for the source whenever one is available. DOI stands for "digital object identifier," a number specific to the article that can help others locate the source. In APA 7, format the DOI as a web address. Active hyperlinks for DOIs and URLs should be used for documents meant for screen reading. Present these hyperlinks in blue and underlined text (the default formatting in Microsoft Word), although plain black text is also acceptable. Be consistent in your formatting choice for DOIs and URLs throughout your reference list. Also see our Quick Answer FAQ, "Can I use the DOI format provided by library databases?"

Jerrentrup, A., Mueller, T., Glowalla, U., Herder, M., Henrichs, N., Neubauer, A., & Schaefer, J. R. (2018). Teaching medicine with the help of “Dr. House.” PLoS ONE , 13 (3), Article e0193972. https://doi.org/10.1371/journal.pone.0193972

For journal articles that are assigned article numbers rather than page ranges, include the article number in place of the page range.

For more on citing electronic resources, see  Electronic Sources References .

Article (Without DOI)

Found in a common academic research database or in print.

Casler , T. (2020). Improving the graduate nursing experience through support on a social media platform. MEDSURG Nursing , 29 (2), 83–87.

If an article does not have a DOI and you retrieved it from a common academic research database through the university library, there is no need to include any additional electronic retrieval information. The reference list entry looks like the entry for a print copy of the article. (This format differs from APA 6 guidelines that recommended including the URL of a journal's homepage when the DOI was not available.) Note that APA 7 has additional guidance on reference list entries for articles found only in specific databases or archives such as Cochrane Database of Systematic Reviews, UpToDate, ProQuest Dissertations and Theses Global, and university archives. See APA 7, Section 9.30 for more information.

Found on an Open Access Website

Eaton, T. V., & Akers, M. D. (2007). Whistleblowing and good governance. CPA Journal , 77 (6), 66–71. http://archives.cpajournal.com/2007/607/essentials/p58.htm

Provide the direct web address/URL to a journal article found on the open web, often on an open access journal's website. In APA 7, active hyperlinks for DOIs and URLs should be used for documents meant for screen reading. Present these hyperlinks in blue and underlined text (the default formatting in Microsoft Word), although plain black text is also acceptable. Be consistent in your formatting choice for DOIs and URLs throughout your reference list.

Weinstein, J. A. (2010).  Social change  (3rd ed.). Rowman & Littlefield.

If the book has an edition number, include it in parentheses after the title of the book. If the book does not list any edition information, do not include an edition number. The edition number is not italicized.

American Nurses Association. (2015). Nursing: Scope and standards of practice (3rd ed.).

If the author and publisher are the same, only include the author in its regular place and omit the publisher.

Lencioni, P. (2012). The advantage: Why organizational health trumps everything else in business . Jossey-Bass. https://amzn.to/343XPSJ

As a change from APA 6 to APA 7, it is no longer necessary to include the ebook format in the title. However, if you listened to an audiobook and the content differs from the text version (e.g., abridged content) or your discussion highlights elements of the audiobook (e.g., narrator's performance), then note that it is an audiobook in the title element in brackets. For ebooks and online audiobooks, also include the DOI number (if available) or nondatabase URL but leave out the electronic retrieval element if the ebook was found in a common academic research database, as with journal articles. APA 7 allows for the shortening of long DOIs and URLs, as shown in this example. See APA 7, Section 9.36 for more information.

Chapter in an Edited Book

Poe, M. (2017). Reframing race in teaching writing across the curriculum. In F. Condon & V. A. Young (Eds.), Performing antiracist pedagogy in rhetoric, writing, and communication (pp. 87–105). University Press of Colorado.

Include the page numbers of the chapter in parentheses after the book title.

Christensen, L. (2001). For my people: Celebrating community through poetry. In B. Bigelow, B. Harvey, S. Karp, & L. Miller (Eds.), Rethinking our classrooms: Teaching for equity and justice (Vol. 2, pp. 16–17). Rethinking Schools.

Also include the volume number or edition number in the parenthetical information after the book title when relevant.

Freud, S. (1961). The ego and the id. In J. Strachey (Ed.),  The standard edition of the complete psychological works of Sigmund Freud  (Vol. 19, pp. 3-66). Hogarth Press. (Original work published 1923)

When a text has been republished as part of an anthology collection, after the author’s name include the date of the version that was read. At the end of the entry, place the date of the original publication inside parenthesis along with the note “original work published.” For in-text citations of republished work, use both dates in the parenthetical citation, original date first with a slash separating the years, as in this example: Freud (1923/1961). For more information on reprinted or republished works, see APA 7, Sections 9.40-9.41.

Classroom Resources

Citing classroom resources.

If you need to cite content found in your online classroom, use the author (if there is one listed), the year of publication (if available), the title of the document, and the main URL of Walden classrooms. For example, you are citing study notes titled "Health Effects of Exposure to Forest Fires," but you do not know the author's name, your reference entry will look like this:

Health effects of exposure to forest fires [Lecture notes]. (2005). Walden University Canvas. https://waldenu.instructure.com

If you do know the author of the document, your reference will look like this:

Smith, A. (2005). Health effects of exposure to forest fires [PowerPoint slides]. Walden University Canvas. https://waldenu.instructure.com  

A few notes on citing course materials:

• [Lecture notes]
• [Course handout]
• [Study notes]
• It can be difficult to determine authorship of classroom documents. If an author is listed on the document, use that. If the resource is clearly a product of Walden (such as the course-based videos), use Walden University as the author. If you are unsure or if no author is indicated, place the title in the author spot, as above.
• If you cannot determine a date of publication, you can use n.d. (for "no date") in place of the year.

Note:  The web location for Walden course materials is not directly retrievable without a password, and therefore, following APA guidelines, use the main URL for the class sites: https://class.waldenu.edu.

Citing Tempo Classroom Resources

Clear author: 

Smith, A. (2005). Health effects of exposure to forest fires [PowerPoint slides]. Walden University Brightspace. https://mytempo.waldenu.edu

Unclear author:

Health effects of exposure to forest fires [Lecture notes]. (2005). Walden University Brightspace. https://mytempo.waldenu.edu

Conference Sessions and Presentations

Feinman, Y. (2018, July 27). Alternative to proctoring in introductory statistics community college courses [Poster presentation]. Walden University Research Symposium, Minneapolis, MN, United States. https://scholarworks.waldenu.edu/symposium2018/23/

Torgerson, K., Parrill, J., & Haas, A. (2019, April 5-9). Tutoring strategies for online students [Conference session]. The Higher Learning Commission Annual Conference, Chicago, IL, United States. http://onlinewritingcenters.org/scholarship/torgerson-parrill-haas-2019/

Dictionary Entry

Merriam-Webster. (n.d.). Leadership. In Merriam-Webster.com dictionary . Retrieved May 28, 2020, from https://www.merriam-webster.com/dictionary/leadership

When constructing a reference for an entry in a dictionary or other reference work that has no byline (i.e., no named individual authors), use the name of the group—the institution, company, or organization—as author (e.g., Merriam Webster, American Psychological Association, etc.). The name of the entry goes in the title position, followed by "In" and the italicized name of the reference work (e.g., Merriam-Webster.com dictionary , APA dictionary of psychology ). In this instance, APA 7 recommends including a retrieval date as well for this online source since the contents of the page change over time. End the reference entry with the specific URL for the defined word.

Discussion Board Post

Osborne, C. S. (2010, June 29). Re: Environmental responsibility [Discussion post]. Walden University Canvas.  https://waldenu.instructure.com  

Dissertations or Theses

Retrieved From a Database

Nalumango, K. (2019). Perceptions about the asylum-seeking process in the United States after 9/11 (Publication No. 13879844) [Doctoral dissertation, Walden University]. ProQuest Dissertations and Theses.

Retrieved From an Institutional or Personal Website

Evener. J. (2018). Organizational learning in libraries at for-profit colleges and universities [Doctoral dissertation, Walden University]. ScholarWorks. https://scholarworks.waldenu.edu/cgi/viewcontent.cgi?article=6606&context=dissertations

Unpublished Dissertation or Thesis

Kirwan, J. G. (2005). An experimental study of the effects of small-group, face-to-face facilitated dialogues on the development of self-actualization levels: A movement towards fully functional persons [Unpublished doctoral dissertation]. Saybrook Graduate School and Research Center.

For further examples and information, see APA 7, Section 10.6.

Legal Material

For legal references, APA follows the recommendations of The Bluebook: A Uniform System of Citation , so if you have any questions beyond the examples provided in APA, seek out that resource as well.

Court Decisions

Reference format:

Name v. Name, Volume Reporter Page (Court Date). URL

Sample reference entry:

Brown v. Board of Education, 347 U.S. 483 (1954). https://www.oyez.org/cases/1940-1955/347us483

Sample citation:

In Brown v. Board of Education (1954), the Supreme Court ruled racial segregation in schools unconstitutional.

Note: Italicize the case name when it appears in the text of your paper.

Name of Act, Title Source § Section Number (Year). URL

Sample reference entry for a federal statute:

Individuals With Disabilities Education Act, 20 U.S.C. § 1400 et seq. (2004). https://www.congress.gov/108/plaws/publ446/PLAW-108publ446.pdf

Sample reference entry for a state statute:

Minnesota Nurse Practice Act, Minn. Stat. §§ 148.171 et seq. (2019). https://www.revisor.mn.gov/statutes/cite/148.171

Sample citation: Minnesota nurses must maintain current registration in order to practice (Minnesota Nurse Practice Act, 2010).

Note: The § symbol stands for "section." Use §§ for sections (plural). To find this symbol in Microsoft Word, go to "Insert" and click on Symbol." Look in the Latin 1-Supplement subset. Note: U.S.C. stands for "United States Code." Note: The Latin abbreviation " et seq. " means "and what follows" and is used when the act includes the cited section and ones that follow. Note: List the chapter first followed by the section or range of sections.

Unenacted Bills and Resolutions

(Those that did not pass and become law)

Title [if there is one], bill or resolution number, xxx Cong. (year). URL

Sample reference entry for Senate bill:

Anti-Phishing Act, S. 472, 109th Cong. (2005). https://www.congress.gov/bill/109th-congress/senate-bill/472

Sample reference entry for House of Representatives resolution:

Anti-Phishing Act, H.R. 1099, 109th Cong. (2005). https://www.congress.gov/bill/109th-congress/house-bill/1099

The Anti-Phishing Act (2005) proposed up to 5 years prison time for people running Internet scams.

These are the three legal areas you may be most apt to cite in your scholarly work. For more examples and explanation, see APA 7, Chapter 11.

Magazine Article

Clay, R. (2008, June). Science vs. ideology: Psychologists fight back about the misuse of research. Monitor on Psychology , 39 (6). https://www.apa.org/monitor/2008/06/ideology

Note that for citations, include only the year: Clay (2008). For magazine articles retrieved from a common academic research database, leave out the URL. For magazine articles from an online news website that is not an online version of a print magazine, follow the format for a webpage reference list entry.

Newspaper Article (Retrieved Online)

Baker, A. (2014, May 7). Connecticut students show gains in national tests. New York Times . http://www.nytimes.com/2014/05/08/nyregion/national-assessment-of-educational-progress-results-in-Connecticut-and-New-Jersey.html

Include the full date in the format Year, Month Day. Do not include a retrieval date for periodical sources found on websites. Note that for citations, include only the year: Baker (2014). For newspaper articles retrieved from a common academic research database, leave out the URL. For newspaper articles from an online news website that is not an online version of a print newspaper, follow the format for a webpage reference list entry.

OASIS Resources

Oasis webpage.

OASIS. (n.d.). Common reference list examples . Walden University. https://academicguides.waldenu.edu/writingcenter/apa/references/examples

For all OASIS content, list OASIS as the author. Because OASIS webpages do not include publication dates, use “n.d.” for the year.

Interactive Guide

OASIS. (n.d.). Embrace iterative research and writing [Interactive guide]. Walden University. https://academics.waldenu.edu/oasis/iterative-research-writing-web

For OASIS multimedia resources, such as interactive guides, include a description of the resource in brackets after the title.

Online Video/Webcast

Walden University. (2013).  An overview of learning  [Video]. Walden University Canvas.  https://waldenu.instructure.com  

Use this format for online videos such as Walden videos in classrooms. Most of our classroom videos are produced by Walden University, which will be listed as the author in your reference and citation. Note: Some examples of audiovisual materials in the APA manual show the word “Producer” in parentheses after the producer/author area. In consultation with the editors of the APA manual, we have determined that parenthetical is not necessary for the videos in our courses. The manual itself is unclear on the matter, however, so either approach should be accepted. Note that the speaker in the video does not appear in the reference list entry, but you may want to mention that person in your text. For instance, if you are viewing a video where Tobias Ball is the speaker, you might write the following: Tobias Ball stated that APA guidelines ensure a consistent presentation of information in student papers (Walden University, 2013). For more information on citing the speaker in a video, see our page on Common Citation Errors .

Taylor, R. [taylorphd07]. (2014, February 27). Scales of measurement [Video]. YouTube. https://www.youtube.com/watch?v=PDsMUlexaMY

OASIS. (2020, April 15). One-way ANCOVA: Introduction [Video]. YouTube. https://youtu.be/_XnNDQ5CNW8

For videos from streaming sites, use the person or organization who uploaded the video in the author space to ensure retrievability, whether or not that person is the speaker in the video. A username can be provided in square brackets. As a change from APA 6 to APA 7, include the publisher after the title, and do not use "Retrieved from" before the URL. See APA 7, Section 10.12 for more information and examples.

See also reference list entry formats for TED Talks .

Technical and Research Reports

Edwards, C. (2015). Lighting levels for isolated intersections: Leading to safety improvements (Report No. MnDOT 2015-05). Center for Transportation Studies. http://www.cts.umn.edu/Publications/ResearchReports/reportdetail.html?id=2402

Technical and research reports by governmental agencies and other research institutions usually follow a different publication process than scholarly, peer-reviewed journals. However, they present original research and are often useful for research papers. Sometimes, researchers refer to these types of reports as gray literature , and white papers are a type of this literature. See APA 7, Section 10.4 for more information.

Reference list entires for TED Talks follow the usual guidelines for multimedia content found online. There are two common places to find TED talks online, with slightly different reference list entry formats for each.

TED Talk on the TED website

If you find the TED Talk on the TED website, follow the format for an online video on an organizational website:

Owusu-Kesse, K. (2020, June). 5 needs that any COVID-19 response should meet [Video]. TED Conferences. https://www.ted.com/talks/kwame_owusu_kesse_5_needs_that_any_covid_19_response_should_meet

The speaker is the author in the reference list entry if the video is posted on the TED website. For citations, use the speaker's surname.

TED Talk on YouTube

If you find the TED Talk on YouTube or another streaming video website, follow the usual format for streaming video sites:

TED. (2021, February 5). The shadow pandemic of domestic violence during COVID-19 | Kemi DaSilvalbru [Video]. YouTube. https://www.youtube.com/watch?v=PGdID_ICFII

TED is the author in the reference list entry if the video is posted on YouTube since it is the channel on which the video is posted. For citations, use TED as the author.

Walden University Course Catalog

To include the Walden course catalog in your reference list, use this format:

Walden University. (2020). 2019-2020 Walden University catalog . https://catalog.waldenu.edu/index.php

If you cite from a specific portion of the catalog in your paper, indicate the appropriate section and paragraph number in your text:

...which reflects the commitment to social change expressed in Walden University's mission statement (Walden University, 2020, Vision, Mission, and Goals section, para. 2).

And in the reference list:

Walden University. (2020). Vision, mission, and goals. In 2019-2020 Walden University catalog. https://catalog.waldenu.edu/content.php?catoid=172&navoid=59420&hl=vision&returnto=search

Vartan, S. (2018, January 30). Why vacations matter for your health . CNN. https://www.cnn.com/travel/article/why-vacations-matter/index.html

For webpages on the open web, include the author, date, webpage title, organization/site name, and URL. (There is a slight variation for online versions of print newspapers or magazines. For those sources, follow the models in the previous sections of this page.)

American Federation of Teachers. (n.d.). Community schools . http://www.aft.org/issues/schoolreform/commschools/index.cfm

If there is no specified author, then use the organization’s name as the author. In such a case, there is no need to repeat the organization's name after the title.

In APA 7, active hyperlinks for DOIs and URLs should be used for documents meant for screen reading. Present these hyperlinks in blue and underlined text (the default formatting in Microsoft Word), although plain black text is also acceptable. Be consistent in your formatting choice for DOIs and URLs throughout your reference list.

Related Resources

Knowledge Check: Common Reference List Examples

Didn't find what you need? Email us at [email protected] .

• Previous Page: Reference List: Overview
• Next Page: Common Military Reference List Examples
• Office of Student Disability Services

Walden Resources

Departments.

• Academic Residencies
• Academic Skills
• Career Planning and Development
• Customer Care Team
• Field Experience
• Military Services
• Student Success Advising
• Writing Skills

Centers and Offices

• Center for Social Change
• Office of Academic Support and Instructional Services
• Office of Degree Acceleration
• Office of Research and Doctoral Services
• Office of Student Affairs

Student Resources

• Doctoral Writing Assessment
• Form & Style Review
• Quick Answers
• ScholarWorks
• SKIL Courses and Workshops
• Walden Bookstore
• Walden Catalog & Student Handbook
• Student Safety/Title IX
• Legal & Consumer Information
• Website Terms and Conditions
• Cookie Policy
• Accessibility
• Accreditation
• State Authorization
• Net Price Calculator
• Contact Walden

Walden University is a member of Adtalem Global Education, Inc. www.adtalem.com Walden University is certified to operate by SCHEV © 2024 Walden University LLC. All rights reserved.

Have a language expert improve your writing

Run a free plagiarism check in 10 minutes, generate accurate citations for free.

• Knowledge Base
• Citing sources
• Citation examples for common sources types

Citation Examples | Books, Articles, Websites & More

Published on April 9, 2021 by Jack Caulfield . Revised on January 17, 2024.

The most common citation styles are APA and MLA . To cite a source in these styles, you need a brief in-text citation and a full reference.

Use the interactive tool to understand how a citation is structured and see examples for common source types .

As well as switching between styles and sources, you can explore how a citation looks when there are multiple authors, different editions, no publication date, and other common scenarios.

If you’re still not sure how to cite your source correctly, dive into our comprehensive articles. For each source type, we provide a range of examples in APA , MLA , and Chicago style .

Cite this Scribbr article

If you want to cite this source, you can copy and paste the citation or click the “Cite this Scribbr article” button to automatically add the citation to our free Citation Generator.

Caulfield, J. (2024, January 17). Citation Examples | Books, Articles, Websites & More. Scribbr. Retrieved June 25, 2024, from https://www.scribbr.com/citing-sources/citation-example/

Is this article helpful?

Jack Caulfield

Other students also liked, citation styles guide | examples for all major styles, apa vs. mla | the key differences in format & citation, how to avoid plagiarism | tips on citing sources, scribbr apa citation checker.

An innovative new tool that checks your APA citations with AI software. Say goodbye to inaccurate citations!

• Walden University

What is a reference list?

• Student-Facing Quick Answers
• Walden College of Nursing | Continuing Education
• zPop Up Widgets
• 119 Academic Residencies
• 267 Academic Skills Center
• 16 Accommodations
• 5 APA: in-text citations
• 3 APA: references
• 5 APA: Word formatting
• 12 Appointments
• 106 Archive
• 23 Blackboard
• 1 Blackboard App
• 5 Blackboard: Assignments
• 2 Blackboard: Basics
• 1 Blackboard: Collaborate Ultra
• 1 Blackboard: Course Materials
• 1 Blackboard: Discussion Posts
• 1 Blackboard: Getting Started
• 4 Blackboard: Videos
• 7 Capstone Intensive
• 9 Career Management
• 27 Career Planning & Development
• 13 Career Services Center
• 5 Center for Global, Professional, and Applied Learning
• 3 Commencement & Graduation
• 18 Course Materials
• 63 Course-Level Statistics
• 123 Customer Care Team
• 15 Disability Services
• 48 Dissertation
• 61 Doctoral Capstone
• 12 Doctoral Degree Coach
• 29 Doctoral Peer Mentors
• 48 Doctoral Study
• 28 Doctoral Writing Assessment
• 19 Doctoral Writing Workshops
• 7 ePortfolio
• 39 Field Experience
• 35 Financial Aid
• 18 Financial Services
• 18 Full Text
• 18 Google Scholar
• 2 Grammarly
• 8 Institutional Review Board (IRB)
• 31 International
• 1 International Student Finance Portal
• 15 Job Search
• 211 Library
• 39 Library Databases
• 34 Library Research
• 161 Library Skills
• 72 Literature Review
• 36 Mechanics/Punctuation
• 12 Methodology
• 29 Military Services
• 2 Military Spouses & Dependants
• 11 MS PowerPoint
• 10 MyWalden
• 7 Networking
• 12 New Students
• 5 Office 365
• 11 Office of Degree Acceleration
• 40 Office of Research and Doctoral Services
• 69 Policies
• 15 Practicum
• 32 Project Study
• 7 Reading Skills
• 19 Registration
• 12 School-Life Balance
• 29 Skills Courses
• 89 Software/Technology
• 4 Statistical Tests: Multivariate Methods
• 17 Statistical Tests: Probability and Regression
• 15 Statistical Tests: Tests of Mean Differences
• 48 Statistical Tools
• 61 Statistics
• 4 Student Organizations
• 4 Student Records
• 46 Student Success Advising
• 1 Student Wellness & Disability Services
• 6 Study Abroad
• 17 Study Skills
• 7 Taskstream
• 21 Textbooks & Course Materials
• 4 Transcripts
• 15 Tutoring
• 28 Veterans
• 84 Videos & Webinars
• 20 Virtual Residency
• 86 WriteCast
• 363 Writing Center
• 3 Writing: Academic Writing

Answered By: Paul Lai Last Updated: Jul 17, 2023     Views: 54469

A reference list is a list of the publication information for the sources you’ve cited in your paper and is intended to give your readers all the information they need to find those sources. In other publication styles, this list may be called a bibliography or a works cited page, but APA uses the term reference list. Here are a few things to keep in mind about reference lists:

• Only list sources you cite in your text . Do not include sources you read but did not cite.
• Double-space your list and use a hanging indent for each reference. ( View resources on creating hanging indents and double-spacing your work on the Academic Skills Center's website .)
• Organize your references alphabetically by the author's last name.
• Include a reference list in every paper in which you cite outside material.
• Do note bold the title of the list.

Additional Resources:

• Check out more reference list tips  on the Writing Center's website. 
• Note that there are specific reference entry formatting requirements for different source types .
• You can also download the  Course Paper template  and scroll down to the bottom for a correctly formatted sample reference list.

Further Questions?

Would you like a current or future assignment to be reviewed by the Writing Center?  If so please visit  the Writing Center's Paper Review Website  and make an appointment with us! 

Do you have other general writing questions? Ask OASIS !

Other questions about your doctoral capstone or the Form & Style review? E-mail the Dissertation Editors at  [email protected] .

Want to peruse other writing resources?  Go to the Writing Center’s homepage . 

• Share on Facebook

Help us do better. Was this helpful?

Related Topics

• Writing Center

More Information

Need more information? Ask us !

Or browse Quick Answers by Topic .

• Office of Student Disability Services

Walden Resources

Departments.

• Academic Residencies
• Academic Skills
• Career Planning and Development
• Customer Care Team
• Field Experience
• Military Services
• Student Success Advising
• Writing Skills

Centers and Offices

• Center for Social Change
• Office of Academic Support and Instructional Services
• Office of Student Affairs
• Office of Degree Acceleration
• Office of Research and Doctoral Services

Student Resources

• Doctoral Writing Assessment
• Form & Style Review
• Quick Answers
• ScholarWorks
• SKIL Courses and Workshops
• Walden Bookstore
• Walden Catalog & Student Handbook
• Student Safety/Title IX
• Legal & Consumer Information
• Website Terms and Conditions
• Cookie Policy
• Accessibility
• Accreditation
• State Authorization
• Net Price Calculator
• Contact Walden

Walden University is a member of Adtalem Global Education, Inc. www.adtalem.com Walden University is certified to operate by SCHEV © 2024 Walden University LLC. All rights reserved.

• Search This Site All UCSD Sites Faculty/Staff Search Term
• Contact & Directions
• Climate Statement
• Cognitive Behavioral Neuroscience
• Cognitive Psychology
• Developmental Psychology
• Social Psychology
• Adjunct Faculty
• Non-Senate Instructors
• Researchers
• Psychology Grads
• Affiliated Grads
• New and Prospective Students
• Honors Program
• Experiential Learning
• Programs & Events
• Psi Chi / Psychology Club
• Prospective PhD Students
• Current PhD Students
• Area Brown Bags
• Colloquium Series
• Anderson Distinguished Lecture Series
• Speaker Videos
• Undergraduate Program
• Academic and Writing Resources

Writing Research Papers

• What Types of References Are Appropriate?

When writing a research paper, there are many different types of sources that you might consider citing.  Which are appropriate?  Which are less appropriate?  Here we discuss the different types of sources that you may wish to use when working on a research paper.   

Please note that the following represents a general set of recommended guidelines that is not specific to any class and does not represent department policy.  The types of allowable sources may vary by course and instructor.

Highly appropriate: peer-reviewed journal articles

In general, you should primarily cite peer-reviewed journal articles in your research papers.  Peer-reviewed journal articles are research papers that have been accepted for publication after having undergone a rigorous editorial review process.  During that review process, the article was carefully evaluated by at least one journal editor and a group of reviewers (usually scientists that are experts in the field or topic under investigation).  Often the article underwent revisions before it was judged to be satisfactory for publication. 

Most articles submitted to high quality journals are not accepted for publication.  As such, research that is successfully published in a respected peer-reviewed journal is generally regarded as higher quality than research that is not published or is published elsewhere, such as in a book, magazine, or on a website.  However, just because a study was published in a peer-reviewed journal does not mean that it is free from error or that its conclusions are correct.  Accordingly, it is important to critically read and carefully evaluate all sources, including peer-reviewed journal articles.

Tips for finding and using peer-reviewed journal articles:

• Many databases, such as PsycINFO, can be set to only search for peer-reviewed journal articles. Other search engines, such as Google Scholar, typically include both peer-reviewed and not peer-reviewed articles in search results, and thus should be used with greater caution. 
• Even though a peer-reviewed journal article is, by definition, a source that has been carefully vetted through an editorial process, it should still be critically evaluated by the reader. 

Potentially appropriate: books, encyclopedias, and other scholarly works

Another potential source that you might use when writing a research paper is a book, encyclopedia, or an official online source (such as demographic data drawn from a government website).  When relying on such sources, it is important to carefully consider its accuracy and trustworthiness.  For example, books vary in quality; most have not undergone any form of review process other than basic copyediting.  In many cases, a book’s content is little more than the author’s informed or uninformed opinion. 

However, there are books that have been edited prior to publication, as is the case with many reputable encyclopedias; also, many books from academic publishers are comprised of multiple chapters, each written by one or more researchers, with the entire volume carefully reviewed by one or more editors.  In those cases, the book has undergone a form of peer review, albeit often not as rigorous as that for a peer-reviewed journal article.

Tips for using books, encyclopedias, and other scholarly works:

• When using books, encyclopedias, and other scholarly works (that is, works written or produced by researchers, official agencies, or corporations), it is important to very carefully evaluate the quality of that source.
• If the source is an edited volume (in which case in the editor(s) will be listed on the cover), is published by a reputable source (such as Academic Press, MIT Press, and others), or is written by a major expert in the field (such as a researcher with a track record of peer-reviewed journal articles on the subject), then it is more likely to be trustworthy.
• For online encyclopedias such as Wikipedia, an instructor may or may not consider that an acceptable source (by default, don’t assume that a non-peer reviewed source will be considered acceptable). It is best to ask the instructor for clarification. 1

Usually inappropriate: magazines, blogs, and websites  

Most research papers can be written using only peer-reviewed journal articles as sources.  However, for many topics it is possible to find a plethora of sources that have not been peer-reviewed but also discuss the topic.  These may include articles in popular magazines or postings in blogs, forums, and other websites.  In general, although these sources may be well-written and easy to understand, their scientific value is often not as high as that of peer-reviewed articles.  Exceptions include some magazine and newspaper articles that might be cited in a research paper to make a point about public awareness of a given topic, to illustrate beliefs and attitudes about a given topic among journalists, or to refer to a news event that is relevant to a given topic. 

Tips for using magazines, blogs, and websites:

• Avoid such references if possible. You should primarily focus on peer-reviewed journal articles as sources for your research paper.  High quality research papers typically do not rely on non-academic and not peer-reviewed sources.
• Refer to non-academic, not peer-reviewed sources sparingly, and if you do, be sure to carefully evaluate the accuracy and scientific merit of the source.

Downloadable Resources

• How to Write APA Style Research Papers (a comprehensive guide) [ PDF ]
• Tips for Writing APA Style Research Papers (a brief summary) [ PDF ]

Further Resources

How-To Videos     

• Writing Research Paper Videos

Databases and Search Engines (may require connection to UCSD network)

• Google Scholar
• PubMed (NIH/NLM)
• Web of Science  

UCSD Resources on Finding and Evaluating Sources

• UCSD Library Databases A-Z
• UCSD Library Psychology Research Guide: Start Page
• UCSD Library Psychology Research Guide : Finding Articles
• UCSD Library Psychology Research Guide : Evaluating Sources

External Resources

• Critically Reading Journal Articles from PSU/ Colby College
• How to Seriously Read a Journal Article from Science Magazine
• How to Read Journal Articles from Harvard University
• How to Read a Scientific Paper Infographic from Elsevier Publishing
• Tips for searching PsycINFO from UC Berkeley Library
• Tips for using PsycINFO effectively from the APA Student Science Council

1 Wikipedia articles vary in quality; the site has a peer review system and the very best articles ( Featured Articles ), which go through a multi-stage review process, rival those in traditional encyclopedias and are considered the highest quality articles on the site.

Prepared by s. c. pan for ucsd psychology, graphic adapted from  t-x-generic-apply.svg , a public domain creation by the tango desktop project..

Back to top

• Research Paper Structure
• Formatting Research Papers
• Using Databases and Finding References
• Evaluating References and Taking Notes
• Citing References
• Writing a Literature Review
• Writing Process and Revising
• Improving Scientific Writing
• Academic Integrity and Avoiding Plagiarism
• Writing Research Papers Videos

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 19 June 2024

Detecting hallucinations in large language models using semantic entropy

• Sebastian Farquhar   ORCID: orcid.org/0000-0002-9185-6415 1   na1 ,
• Jannik Kossen 1   na1 ,
• Lorenz Kuhn 1   na1 &
• Yarin Gal   ORCID: orcid.org/0000-0002-2733-2078 1  

Nature volume  630 ,  pages 625–630 ( 2024 ) Cite this article

64k Accesses

1457 Altmetric

Metrics details

• Computer science
• Information technology

Large language model (LLM) systems, such as ChatGPT 1 or Gemini 2 , can show impressive reasoning and question-answering capabilities but often ‘hallucinate’ false outputs and unsubstantiated answers 3 , 4 . Answering unreliably or without the necessary information prevents adoption in diverse fields, with problems including fabrication of legal precedents 5 or untrue facts in news articles 6 and even posing a risk to human life in medical domains such as radiology 7 . Encouraging truthfulness through supervision or reinforcement has been only partially successful 8 . Researchers need a general method for detecting hallucinations in LLMs that works even with new and unseen questions to which humans might not know the answer. Here we develop new methods grounded in statistics, proposing entropy-based uncertainty estimators for LLMs to detect a subset of hallucinations—confabulations—which are arbitrary and incorrect generations. Our method addresses the fact that one idea can be expressed in many ways by computing uncertainty at the level of meaning rather than specific sequences of words. Our method works across datasets and tasks without a priori knowledge of the task, requires no task-specific data and robustly generalizes to new tasks not seen before. By detecting when a prompt is likely to produce a confabulation, our method helps users understand when they must take extra care with LLMs and opens up new possibilities for using LLMs that are otherwise prevented by their unreliability.

Similar content being viewed by others

Testing theory of mind in large language models and humans

Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT

ThoughtSource: A central hub for large language model reasoning data

‘Hallucinations’ are a critical problem 9 for natural language generation systems using large language models (LLMs), such as ChatGPT 1 or Gemini 2 , because users cannot trust that any given output is correct.

Hallucinations are often defined as LLMs generating “content that is nonsensical or unfaithful to the provided source content” 9 , 10 , 11 but they have come to include a vast array of failures of faithfulness and factuality. We focus on a subset of hallucinations which we call ‘confabulations’ 12 for which LLMs fluently make claims that are both wrong and arbitrary—by which we mean that the answer is sensitive to irrelevant details such as random seed. For example, when asked a medical question “What is the target of Sotorasib?” an LLM confabulates by sometimes answering KRASG12 ‘C’ (correct) and other times KRASG12 ‘D’ (incorrect) despite identical instructions. We distinguish this from cases in which a similar ‘symptom’ is caused by the following different mechanisms: when LLMs are consistently wrong as a result of being trained on erroneous data such as common misconceptions 13 ; when the LLM ‘lies’ in pursuit of a reward 14 ; or systematic failures of reasoning or generalization. We believe that combining these distinct mechanisms in the broad category hallucination is unhelpful. Our method makes progress on a portion of the problem of providing scalable oversight 15 by detecting confabulations that people might otherwise find plausible. However, it does not guarantee factuality because it does not help when LLM outputs are systematically bad. Nevertheless, we significantly improve question-answering accuracy for state-of-the-art LLMs, revealing that confabulations are a great source of error at present.

We show how to detect confabulations by developing a quantitative measure of when an input is likely to cause an LLM to generate arbitrary and ungrounded answers. Detecting confabulations allows systems built on LLMs to avoid answering questions likely to cause confabulations, to make users aware of the unreliability of answers to a question or to supplement the LLM with more grounded search or retrieval. This is essential for the critical emerging field of free-form generation in which naive approaches, suited to closed vocabulary and multiple choice, fail. Past work on uncertainty for LLMs has focused on simpler settings, such as classifiers 16 , 17 and regressors 18 , 19 , whereas the most exciting applications of LLMs relate to free-form generations.

The term hallucination in the context of machine learning originally comes from filling in ungrounded details, either as a deliberate strategy 20 or as a reliability problem 4 . The appropriateness of the metaphor has been questioned as promoting undue anthropomorphism 21 . Although we agree that metaphor must be used carefully with LLMs 22 , the widespread adoption of the term hallucination reflects the fact that it points to an important phenomenon. This work represents a step towards making that phenomenon more precise.

To detect confabulations, we use probabilistic tools to define and then measure the ‘semantic’ entropy of the generations of an LLM—an entropy that is computed over meanings of sentences. High entropy corresponds to high uncertainty 23 , 24 , 25 —so semantic entropy is one way to estimate semantic uncertainties. Semantic uncertainty, the broader category of measures we introduce, could be operationalized with other measures of uncertainty, such as mutual information, instead. Entropy in free-form generation is normally hard to measure because answers might mean the same thing (be semantically equivalent) despite being expressed differently (being syntactically or lexically distinct). This causes naive estimates of entropy or other lexical variation scores 26 to be misleadingly high when the same correct answer might be written in many ways without changing its meaning.

By contrast, our semantic entropy moves towards estimating the entropy of the distribution of meanings of free-form answers to questions, insofar as that is possible, rather than the distribution over the ‘tokens’ (words or word-pieces) which LLMs natively represent. This can be seen as a kind of semantic consistency check 27 for random seed variation. An overview of our approach is provided in Fig. 1 and a worked example in Supplementary Table 1 .

a , Naive entropy-based uncertainty measures variation in the exact answers, treating ‘Paris’, ‘It’s Paris’ and ‘France’s capital Paris’ as different. But this is unsuitable for language tasks for which sometimes different answers mean the same things. Our semantic entropy clusters answers which share meanings before computing the entropy. A low semantic entropy shows that the LLM is confident about the meaning. b , Semantic entropy can also detect confabulations in longer passages. We automatically decompose a long generated answer into factoids. For each factoid, an LLM generates questions to which that factoid might have been the answer. The original LLM then samples  M possible answers to these questions. Finally, we compute the semantic entropy over the answers to each specific question, including the original factoid. Confabulations are indicated by high average semantic entropy for questions associated with that factoid. Here, semantic entropy classifies Fact 1 as probably not a confabulation because generations often mean the same thing, despite very different wordings, which a naive entropy would have missed.

Intuitively, our method works by sampling several possible answers to each question and clustering them algorithmically into answers that have similar meanings, which we determine on the basis of whether answers in the same cluster entail each other bidirectionally 28 . That is, if sentence A entails that sentence B is true and vice versa, then we consider them to be in the same semantic cluster. We measure entailment using both general-purpose LLMs and natural language inference (NLI) tools developed specifically for detecting entailment for which we show direct evaluations in Supplementary Tables 2 and 3 and Supplementary Fig. 1 . Textual entailment has previously been shown to correlate with faithfulness 10 in the context of factual consistency 29 as well as being used to measure factuality in abstractive summarization 30 , especially when applied at the right granularity 31 .

Semantic entropy detects confabulations in free-form text generation across a range of language models and domains, without previous domain knowledge. Our evaluations cover question answering in trivia knowledge (TriviaQA 32 ), general knowledge (SQuAD 1.1; ref. 33 ), life sciences (BioASQ 34 ) and open-domain natural questions (NQ-Open 35 ) derived from actual queries to Google Search 36 . In addition, semantic entropy detects confabulations in mathematical word problems (SVAMP 37 ) and in a biography-generation dataset, FactualBio, accompanying this paper.

Our results for TriviaQA, SQuAD, BioASQ, NQ-Open and SVAMP are all evaluated context-free and involve sentence-length answers (96 ± 70 characters, mean ± s.d.) and use LLaMA 2 Chat (7B, 13B and 70B parameters) 38 , Falcon Instruct (7B and 40B) 39 and Mistral Instruct (7B) 40 . In the Supplementary Information , we further consider short-phrase-length answers. Results for FactualBio (442 ± 122 characters) use GPT-4 (ref. 1 ). At the time of writing, GPT-4 (ref. 1 ) did not expose output probabilities 41 or hidden states, although it does now. As a result, we propose a discrete approximation of our estimator for semantic entropy which allows us to run experiments without access to output probabilities, which we use for all GPT-4 results in this paper and which performs similarly well.

Our confabulation detection with semantic entropy is more robust to user inputs from previously unseen domains than methods which aim to ‘learn’ how to detect confabulations from a set of example demonstrations. Our method is unsupervised, meaning that we do not need labelled examples of confabulations. By contrast, supervised methods detect confabulations by learning patterns behind examples of confabulations, assuming that future questions preserve these patterns. But this assumption is often untrue in new situations or with confabulations that human overseers are unable to identify (compare Fig. 17 of ref. 24 ). As a strong supervised baseline, we compare to an embedding regression method inspired by ref. 24 which trains a logistic regression classifier to predict whether the model correctly answered a question on the basis of the final ‘embedding’ (hidden state) of the LLM. We also use the P (True) method 24 which looks at the probability with which an LLM predicts that the next token is ‘True’ when few-shot prompted to compare a main answer with ‘brainstormed’ alternatives.

Confabulations contribute substantially to incorrect answers given by language models. We show that semantic entropy can be used to predict many incorrect model answers and to improve question-answering accuracy by refusing to answer those questions the model is uncertain about. Corresponding to these two uses, we evaluate two main metrics. First, the widely used area under the receiver operating characteristic (AUROC) curve for the binary event that a given answer is incorrect. This measure captures both precision and recall and ranges from 0 to 1, with 1 representing a perfect classifier and 0.5 representing an un-informative classifier. We also show a new measure, the area under the ‘rejection accuracy’ curve (AURAC). This studies the case in which the confabulation detection score is used to refuse to answer the questions judged most likely to cause confabulations. Rejection accuracy is the accuracy of the answers of the model on the remaining questions and the area under this curve is a summary statistic over many thresholds (representative threshold accuracies are provided in Supplementary Material ). The AURAC captures the accuracy improvement which users would experience if semantic entropy was used to filter out questions causing the highest entropy.

Detecting confabulations in QA and math

In Fig. 2 , we show that both semantic entropy and its discrete approximation outperform our best baselines for sentence-length generations. These results are averaged across datasets and provide the actual scores on the held-out evaluation dataset. We report the raw average score across held-out evaluation datasets without standard error because the distributional characteristics are more a property of the models and datasets selected than the method. Consistency of relative results across different datasets is a stronger indicator of variation in this case.

Semantic entropy outperforms leading baselines and naive entropy. AUROC (scored on the y -axes) measures how well methods predict LLM mistakes, which correlate with confabulations. AURAC (likewise scored on the y -axes) measures the performance improvement of a system that refuses to answer questions which are judged likely to cause confabulations. Results are an average over five datasets, with individual metrics provided in the Supplementary Information .

Semantic entropy greatly outperforms the naive estimation of uncertainty using entropy: computing the entropy of the length-normalized joint probability of the token sequences. Naive entropy estimation ignores the fact that token probabilities also express the uncertainty of the model over phrasings that do not change the meaning of an output.

Our methods also outperform the supervised embedding regression method both in- and out-of-distribution. In pale-yellow bars we show that embedding regression performance deteriorates when its training data do not match the deployment distribution—which mirrors the common real-world case in which there is a distribution shift between training and deployment 42 —the plotted value is the average metric for embedding regression trained on one of the four ‘off-distribution’ datasets for that evaluation. This is critical because reliable uncertainty is most important when the data distribution shifts. Semantic entropy also outperforms P (True) which is supervised ‘in-context’; that is, it is adapted to the deployment task with a few training examples provided in the LLM prompt itself. The discrete variant of semantic entropy performs similarly to our standard estimator, despite not requiring exact output probabilities.

Averaged across the 30 combinations of tasks and models we study, semantic entropy achieves the best AUROC value of 0.790 whereas naive entropy (0.691), P (True) (0.698) and the embedding regression baseline (0.687) lag behind it. Semantic entropy performs well consistently, with stable performance (between 0.78 and 0.81 AUROC) across the different model families (LLaMA, Falcon and Mistral) and scales (from 7B to 70B parameters) which we study (we report summary statistics for each dataset and model as before). Although semantic entropy outperforms the baselines across all model sizes, P (True) seems to improve with model size, suggesting that it might become more competitive for very capable honest models in settings that the model understands well (which are, however, not the most important cases to have good uncertainty). We use ten generations to compute entropy, selected using analysis in Supplementary Fig. 2 . Further results for short-phrase generations are described in Supplementary Figs. 7 – 10 .

The results in Fig. 2 offer a lower bound on the effectiveness of semantic entropy at detecting confabulations. These evaluations determine whether semantic entropy and baseline methods can detect when the answers of the model are incorrect (which we validate against human correctness evaluations in Supplementary Table 4 ). In addition to errors from confabulations (arbitrary incorrectness), this also includes other types of mistakes for which semantic entropy is not suited, such as consistent errors learned from the training data. The fact that methods such as embedding regression are able to spot other kinds of errors, not just confabulations, but still are outperformed by semantic entropy, suggests that confabulations are a principal category of errors for actual generations.

Examples of questions and answers from TriviaQA, SQuAD and BioASQ, for LLaMA 2 Chat 70B, are shown in Table 1 . These illustrate how only semantic entropy detects when the meaning is constant but the form varies (the first row of the table) whereas semantic entropy and naive entropy both correctly predict the presence of confabulations when the form and meaning vary together (second row) and predict the absence of confabulations when the form and meaning are both constant across several resampled generations (third row). In the final row, we give an example in which semantic entropy is erroneously high as a result of overly sensitive semantic clustering relative to the reference answer. Our clustering method distinguishes the answers which provide a precise date from those which only provide a year. For some contexts that would have been correct but in this context the distinction between the specific day and the year is probably irrelevant. This highlights the importance of context and judgement in clustering, especially in subtle cases, as well as the shortcomings of evaluating against fixed reference answers which do not capture the open-ended flexibility of conversational deployments of LLMs.

Detecting confabulations in biographies

Semantic entropy is most natural for sentences that express a single proposition but the idea of semantic equivalence is trickier to apply to longer passages which express many propositions which might only agree partially 43 . Nevertheless, we can use semantic entropy to detect confabulations in longer generations, such as entire paragraphs of text. To show this, we develop a dataset of biographical generations from GPT-4 (v.0613) for 21 individuals notable enough to have their own Wikipedia page but without extensive online biographies. From each biography generated by GPT-4, we automatically extract propositional factual claims about the individual (150 factual claims in total), which we manually label as true or false.

Applying semantic entropy to this problem is challenging. Naively, one might simply regenerate each sentence (conditioned on the text so far) and then compute semantic entropy over these regenerations. However, the resampled sentences often target different aspects of the biography: for example, one time describing family and the next time profession. This is analogous to the original problem semantic entropy was designed to resolve: the model is uncertain about the right ordering of facts, not about the facts themselves. To address this, we break down the entire paragraph into factual claims and reconstruct questions which might have been answered by those claims. Only then do we apply semantic entropy (Fig. 1 ) by generating three new answers to each question (selected with analysis in Supplementary Figs. 3 and 4 ) and computing the semantic entropy over those generations plus the original factual claim. We aggregate these by averaging the semantic entropy over all the questions to get an uncertainty score for each proposition, which we use to detect confabulations. Unaggregated results are shown in Supplementary Figs. 5 and 6 .

As GPT-4 did not allow access to the probability of the generation at the time of writing, we use a discrete variant of semantic entropy which makes the further approximation that we can infer a discrete empirical distribution over semantic meaning clusters from only the generations ( Methods ). This allows us to compute semantic entropy using only the black-box outputs of an LLM. However, we were unable to compute the naive entropy baseline, the standard semantic entropy estimator or the embedding regression baseline for GPT-4 without output probabilities and embeddings.

In Fig. 3 we show that the discrete variant of semantic entropy effectively detects confabulations on this dataset. Its AUROC and AURAC are higher than either a simple ‘self-check’ baseline—which just asks the LLM whether the factoid is likely to be true—or a variant of P (True) which has been adapted to work for the paragraph-length setting. Discrete semantic entropy has better rejection accuracy performance until 20% of the questions have been rejected at which point P (True) has a narrow edge. This indicates that the questions predicted to cause confabulations are indeed more likely to be wrong.

The discrete variant of our semantic entropy estimator outperforms baselines both when measured by AUROC and AURAC metrics (scored on the y -axis). The AUROC and AURAC are substantially higher than for both baselines. At above 80% of questions being answered, semantic entropy has the highest accuracy. Only when the top 20% of answers judged most likely to be confabulations are rejected does the answer accuracy on the remainder for the P (True) baseline exceed semantic entropy.

Our probabilistic approach, accounting for semantic equivalence, detects an important class of hallucinations: those that are caused by a lack of LLM knowledge. These are a substantial portion of the failures at present and will continue even as models grow in capabilities because situations and cases that humans cannot reliably supervise will persist. Confabulations are a particularly noteworthy failure mode for question answering but appear in other domains too. Semantic entropy needs no previous domain knowledge and we expect that algorithmic adaptations to other problems will allow similar advances in, for example, abstractive summarization. In addition, extensions to alternative input variations such as rephrasing or counterfactual scenarios would allow a similar method to act as a form of cross-examination 44 for scalable oversight through debate 45 .

The success of semantic entropy at detecting errors suggests that LLMs are even better at “knowing what they don’t know” than was argued by ref. 24 —they just don’t know they know what they don’t know. Our method explicitly does not directly address situations in which LLMs are confidently wrong because they have been trained with objectives that systematically produce dangerous behaviour, cause systematic reasoning errors or are systematically misleading the user. We believe that these represent different underlying mechanisms—despite similar ‘symptoms’—and need to be handled separately.

One exciting aspect of our approach is the way it makes use of classical probabilistic machine learning methods and adapts them to the unique properties of modern LLMs and free-form language generation. We hope to inspire a fruitful exchange of well-studied methods and emerging new problems by highlighting the importance of meaning when addressing language-based machine learning problems.

Semantic entropy as a strategy for overcoming confabulation builds on probabilistic tools for uncertainty estimation. It can be applied directly to any LLM or similar foundation model without requiring any modifications to the architecture. Our ‘discrete’ variant of semantic uncertainty can be applied even when the predicted probabilities for the generations are not available, for example, because access to the internals of the model is limited.

In this section we introduce background on probabilistic methods and uncertainty in machine learning, discuss how it applies to language models and then discuss our contribution, semantic entropy, in detail.

Uncertainty and machine learning

We aim to detect confabulations in LLMs, using the principle that the model will be uncertain about generations for which its output is going to be arbitrary.

One measure of uncertainty is the predictive entropy of the output distribution, which measures the information one has about the output given the input 25 . The predictive entropy (PE) for an input sentence x is the conditional entropy ( H ) of the output random variable Y with realization y given x ,

A low predictive entropy indicates an output distribution which is heavily concentrated whereas a high predictive entropy indicates that many possible outputs are similarly likely.

Aleatoric and epistemic uncertainty

We do not distinguish between aleatoric and epistemic uncertainty in our analysis. Researchers sometimes separate aleatoric uncertainty (uncertainty in the underlying data distribution) from epistemic uncertainty (caused by having only limited information) 46 . Further advances in uncertainty estimation which separate these kinds of uncertainty would enhance the potential for our semantic uncertainty approach by allowing extensions beyond entropy.

Joint probabilities of sequences of tokens

Generative LLMs produce strings of text by selecting tokens in sequence. Each token is a wordpiece that often represents three or four characters (though especially common sequences and important words such as numbers typically get their own token). To compute entropies, we need access to the probabilities the LLM assigns to the generated sequence of tokens. The probability of the entire sequence, s , conditioned on the context, x , is the product of the conditional probabilities of new tokens given past tokens, whose resulting log-probability is \(\log P({\bf{s}}| {\boldsymbol{x}})={\sum }_{i}\log P({s}_{i}| {{\bf{s}}}_{ < i},{\boldsymbol{x}})\) , where s i is the i th output token and s < i denotes the set of previous tokens.

Length normalization

When comparing the log-probabilities of generated sequences, we use ‘length normalization’, that is, we use an arithmetic mean log-probability, \(\frac{1}{N}{\sum }_{i}^{N}\log P({s}_{i}| {{\bf{s}}}_{ < i},{\boldsymbol{x}})\) , instead of the sum. In expectation, longer sequences have lower joint likelihoods because of the conditional independence of the token probabilities 47 . The joint likelihood of a sequence of length N shrinks exponentially in N . Its negative log-probability therefore grows linearly in N , so longer sentences tend to contribute more to entropy. We therefore interpret length-normalizing the log-probabilities when estimating the entropy as asserting that the expected uncertainty of generations is independent of sentence length. Length normalization has some empirical success 48 , including in our own preliminary experiments, but little theoretical justification in the literature.

Principles of semantic uncertainty

If we naively calculate the predictive entropy directly from the probabilities of the generated sequence of tokens, we conflate the uncertainty of the model over the meaning of its answer with the uncertainty over the exact tokens used to express that meaning. For example, even if the model is confident in the meaning of a generation, there are still usually many different ways for phrasing that generation without changing its meaning. For the purposes of detecting confabulations, the uncertainty of the LLM over meanings is more important than the uncertainty over the exact tokens used to express those meanings.

Our semantic uncertainty method therefore seeks to estimate only the uncertainty the LLM has over the meaning of its generation, not the choice of words. To do this, we introduce an algorithm that clusters model generations by meaning and subsequently calculates semantic uncertainty. At a high level this involves three steps:

Generation: sample output sequences of tokens from the predictive distribution of a LLM given a context x .

Clustering: cluster sequences by their meaning using our clustering algorithm based on bidirectional entailment.

Entropy estimation: estimate semantic entropy by summing probabilities of sequences that share a meaning following equation ( 2 ) and compute their entropy.

Generating a set of answers from the model

Given some context x as input to the LLM, we sample M sequences, { s (1) , …,  s ( M ) } and record their token probabilities, { P ( s (1) ∣ x ), …,  P ( s ( M ) ∣ x )}. We sample all our generations from a single model, varying only the random seed used for sampling from the token probabilities. We do not observe the method to be particularly sensitive to details of the sampling scheme. In our implementation, we sample at temperature 1 using nucleus sampling ( P  = 0.9) (ref. 49 ) and top- K sampling ( K  = 50) (ref. 50 ). We also sample a single generation at low temperature (0.1) as an estimate of the ‘best generation’ of the model to the context, which we use to assess the accuracy of the model. (A lower sampling temperature increases the probability of sampling the most likely tokens).

Clustering by semantic equivalence

To estimate semantic entropy we need to cluster generated outputs from the model into groups of outputs that mean the same thing as each other.

This can be described using ‘semantic equivalence’ which is the relation that holds between two sentences when they mean the same thing. We can formalize semantic equivalence mathematically. Let the space of tokens in a language be \({\mathcal{T}}\) . The space of all possible sequences of tokens of length N is then \({{\mathcal{S}}}_{N}\equiv {{\mathcal{T}}}^{N}\) . Note that N can be made arbitrarily large to accommodate whatever size of sentence one can imagine and one of the tokens can be a ‘padding’ token which occurs with certainty for each token after the end-of-sequence token. For some sentence \({\bf{s}}\in {{\mathcal{S}}}_{N}\) , composed of a sequence of tokens, \({s}_{i}\in {\mathcal{T}}\) , there is an associated meaning. Theories of meaning are contested 51 . However, for specific models and deployment contexts many considerations can be set aside. Care should be taken comparing very different models and contexts.

Let us introduce a semantic equivalence relation, E (  ⋅  ,  ⋅  ), which holds for any two sentences that mean the same thing—we will operationalize this presently. Recall that an equivalence relation is any reflexive, symmetric and transitive relation and that any equivalence relation on a set corresponds to a set of equivalence classes. Each semantic equivalence class captures outputs that can be considered to express the same meaning. That is, for the space of semantic equivalence classes \({\mathcal{C}}\) the sentences in the set \(c\in {\mathcal{C}}\) can be regarded in many settings as expressing a similar meaning such that \(\forall {\bf{s}},{{\bf{s}}}^{{\prime} }\in c:E({\bf{s}},{{\bf{s}}}^{{\prime} })\) . So we can build up these classes of semantically equivalent sentences by checking if new sentences share a meaning with any sentences we have already clustered and, if so, adding them into that class.

We operationalize E (  ⋅  ,  ⋅  ) using the idea of bidirectional entailment, which has a long history in linguistics 52 and natural language processing 28 , 53 , 54 . A sequence, s , means the same thing as a second sequence, s ′, only if the sequences entail (that is, logically imply) each other. For example, ‘The capital of France is Paris’ entails ‘Paris is the capital of France’ and vice versa because they mean the same thing. (See later for a discussion of soft equivalence and cases in which bidirectional entailment does not guarantee equivalent meanings).

Importantly, we require that the sequences mean the same thing with respect to the context—key meaning is sometimes contained in the context. For example, ‘Paris’ does not entail ‘The capital of France is Paris’ because ‘Paris’ is not a declarative sentence without context. But in the context of the question ‘What is the capital of France?’, the one-word answer does entail the longer answer.

Detecting entailment has been the object of study of a great deal of research in NLI 55 . We rely on language models to predict entailment, such as DeBERTa-Large-MNLI 56 , which has been trained to predict entailment, or general-purpose LLMs such as GPT-3.5 (ref. 57 ), which can predict entailment given suitable prompts.

We then cluster sentences according to whether they bidirectionally entail each other using the algorithm presented in Extended Data Fig. 1 . Note that, to check if a sequence should be added to an existing cluster, it is sufficient to check if the sequence bidirectionally entails any of the existing sequences in that cluster (we arbitrarily pick the first one), given the transitivity of semantic equivalence. If a sequence does not share meaning with any existing cluster, we assign it its own cluster.

Computing the semantic entropy

Having determined the classes of generated sequences that mean the same thing, we can estimate the likelihood that a sequence generated by the LLM belongs to a given class by computing the sum of the probabilities of all the possible sequences of tokens which can be considered to express the same meaning as

Formally, this treats the output as a random variable whose event-space is the space of all possible meaning-classes, C , a sub- σ -algebra of the standard event-space S . We can then estimate the semantic entropy (SE) as the entropy over the meaning-distribution,

There is a complication which prevents direct computation: we do not have access to every possible meaning-class c . Instead, we can only sample c from the sequence-generating distribution induced by the model. To handle this, we estimate the expectation in equation ( 3 ) using a Rao–Blackwellized Monte Carlo integration over the semantic equivalence classes C ,

where \(P({C}_{i}| {\boldsymbol{x}})=\frac{P({c}_{i}| {\boldsymbol{x}})}{{\sum }_{c}P(c| {\boldsymbol{x}})}\) estimates a categorical distribution over the cluster meanings, that is, ∑ i P ( C i ∣ x ) = 1. Without this normalization step cluster ‘probabilities’ could exceed one because of length normalization, resulting in degeneracies. Equation ( 5 ) is the estimator giving our main method that we refer to as semantic entropy throughout the text.

For scenarios in which the sequence probabilities are not available, we propose a variant of semantic entropy which we call ‘discrete’ semantic entropy. Discrete semantic entropy approximates P ( C i ∣ x ) directly from the number of generations in each cluster, disregarding the token probabilities. That is, we approximate P ( C i ∣ x ) as \({\sum }_{1}^{M}\frac{{I}_{c={C}_{i}}}{M}\) , the proportion of all the sampled answers which belong to that cluster. Effectively, this just assumes that each output that was actually generated was equally probable—estimating the underlying distribution as the categorical empirical distribution. In the limit of M the estimator converges to equation ( 5 ) by the law of large numbers. We find that discrete semantic entropy results in similar performance empirically.

We provide a worked example of the computation of semantic entropy in Supplementary Note  1 .

Semantic entropy is designed to detect confabulations, that is, model outputs with arbitrary meaning. In our experiments, we use semantic uncertainty to predict model accuracy, demonstrating that confabulations make up a notable fraction of model mistakes. We further show that semantic uncertainty can be used to improve model accuracy by refusing to answer questions when semantic uncertainty is high. Last, semantic uncertainty can be used to give users a way to know when model generations are probably unreliable.

We use the datasets BioASQ 34 , SQuAD 33 , TriviaQA 32 , SVAMP 37 and NQ-Open 35 . BioASQ is a life-sciences question-answering dataset based on the annual challenge of the same name. The specific dataset we use is based on the QA dataset from Task B of the 2023 BioASQ challenge (11B). SQuAD is a reading comprehension dataset whose context passages are drawn from Wikipedia and for which the answers to questions can be found in these passages. We use SQuAD 1.1 which excludes the unanswerable questions added in v.2.0 that are deliberately constructed to induce mistakes so they do not in practice cause confabulations to occur. TriviaQA is a trivia question-answering dataset. SVAMP is a word-problem maths dataset containing elementary-school mathematical reasoning tasks. NQ-Open is a dataset of realistic questions aggregated from Google Search which have been chosen to be answerable without reference to a source text. For each dataset, we use 400 train examples and 400 test examples randomly sampled from the original larger dataset. Note that only some of the methods require training, for example semantic entropy does not use the training data. If the datasets themselves are already split into train and test (or validation) samples, we sample our examples from within the corresponding split.

All these datasets are free-form, rather than multiple choice, because this better captures the opportunities created by LLMs to produce free-form sentences as answers. We refer to this default scenario as our ‘sentence-length’ experiments. In Supplementary Note  7 , we also present results for confabulation detection in a ‘short-phrase’ scenario, in which we constrain model answers on these datasets to be as concise as possible.

To make the problems more difficult and induce confabulations, we do not provide the context passages for any of the datasets. When the context passages are provided, the accuracy rate is too high for these datasets for the latest generations of models to meaningfully study confabulations.

For sentence-length generations we use: Falcon 39 Instruct (7B and 40B), LLaMA 2 Chat 38 (7B, 13B and 70B) and Mistral 40 Instruct (7B).

In addition to reporting results for semantic entropy, discrete semantic entropy and naive entropy, we consider two strong baselines.

Embedding regression is a supervised baseline inspired by the P (IK) method 24 . In that paper, the authors fine-tune their proprietary LLM on a dataset of questions to predict whether the model would have been correct. This requires access to a dataset of ground-truth answers to the questions. Rather than fine-tuning the entire LLM in this way, we simply take the final hidden units and train a logistic regression classifier to make the same prediction. By contrast to their method, this is much simpler because it does not require fine-tuning the entire language model, as well as being more reproducible because the solution to the logistic regression optimization problem is not as seed-dependent as the fine-tuning procedure. As expected, this supervised approach performs well in-distribution but fails when the distribution of questions is different from that on which the classifier is trained.

The second baseline we consider is the P (True) method 24 , in which the model first samples M answers (identically to our semantic entropy approach) and then is prompted with the list of all answers generated followed by the highest probability answer and a question whether this answer is “(a) True” or “(b) False”. The confidence score is then taken to be the probability with which the LLM responds with ‘a’ to the multiple-choice question. The performance of this method is boosted with a few-shot prompt, in which up to 20 examples from the training set are randomly chosen, filled in as above, but then provided with the actual ground truth of whether the proposed answer was true or false. In this way, the method can be considered as supervised ‘in-context’ because it makes use of some ground-truth training labels but can be used without retraining the model. Because of context-size constraints, this method cannot fit a full 20 few-shot examples in the context when input questions are long or large numbers of generations are used. As a result, we sometimes have to reduce the number of few-shot examples to suit the context size and we note this in the  Supplementary Material .

Entailment estimator

Any NLI classification system could be used for our bidirectional entailment clustering algorithm. We consider two different kinds of entailment detector.

One option is to use an instruction-tuned LLM such as LLaMA 2, GPT-3.5 (Turbo 1106) or GPT-4 to predict entailment between generations. We use the following prompt:

We are evaluating answers to the question {question} Here are two possible answers: Possible Answer 1: {text1} Possible Answer 2: {text2} Does Possible Answer 1 semantically entail Possible Answer 2? Respond with entailment, contradiction, or neutral.

Alternatively, we consider using a language model trained for entailment prediction, specifically the DeBERTa-large model 56 fine-tuned on the NLI dataset MNLI 58 . This builds on past work towards paraphrase identification based on embedding similarity 59 , 60 and BERT-style models 61 , 62 . We template more simply, checking if DeBERTa predicts entailment between the concatenation of the question and one answer and the concatenation of the question and another answer. Note that DeBERTa-large is a relatively lightweight model with only 1.5B parameters which is much less powerful than most of the LLMs under study.

In Supplementary Note 2 , we carefully evaluate the benefits and drawbacks of these methods for entailment prediction. We settle on using GPT-3.5 with the above prompt, as its entailment predictions agree well with human raters and lead to good confabulation detection performance.

In Supplementary Note  3 , we provide a discussion of the computational cost and choosing the number of generations for reliable clustering.

Prompting templates

We use a simple generation template for all sentence-length answer datasets:

Answer the following question in a single brief but complete sentence. Question: {question} Answer:

Metrics and accuracy measurements

We use three main metrics to evaluate our method: AUROC, rejection accuracy and AURAC. Each of these is grounded in an automated factuality estimation measurement relative to the reference answers provided by the datasets that we use.

AUROC, rejection accuracy and AURAC

First, we use the AUROC curve, which measures the reliability of a classifier accounting for both precision and recall. The AUROC can be interpreted as the probability that a randomly chosen correct answer has been assigned a higher confidence score than a randomly chosen incorrect answer. For a perfect classifier, this is 1.

Second, we compute the ‘rejection accuracy at X %’, which is the question-answering accuracy of the model on the most-confident X % of the inputs as identified by the respective uncertainty method. If an uncertainty method works well, predictions on the confident subset should be more accurate than predictions on the excluded subset and the rejection accuracy should increase as we reject more inputs.

To summarize this statistic we compute the AURAC—the total area enclosed by the accuracies at all cut-off percentages X %. This should increase towards 1 as given uncertainty method becomes more accurate and better at detecting likely-inaccurate responses but it is more sensitive to the overall accuracy of the model than the AUROC metric.

In Supplementary Note  5 , we provide the unaggregated rejection accuracies for sentence-length generations.

Assessing accuracy

For the short-phrase-length generation setting presented in Supplementary Note  7 , we simply assess the accuracy of the generations by checking if the F1 score of the commonly used SQuAD metric exceeds 0.5. There are limitations to such simple scoring rules 63 but this method is widely used in practice and its error is comparatively small on these standard datasets.

For our default scenario, the longer sentence-length generations, this measure fails, as the overlap between the short reference answer and our long model answer is invariably too small. For sentence-length generations, we therefore automatically determine whether an answer to the question is correct or incorrect by using GPT-4 to compare the given answer to the reference answer. We use the template:

We are assessing the quality of answers to the following question: {question} The expected answer is: {reference answer} The proposed answer is: {predicted answer} Within the context of the question, does the proposed answer mean the same as the expected answer? Respond only with yes or no.

We make a small modification for datasets with several reference answers: line two becomes “The following are expected answers to this question:” and the final line asks “does the proposed answer mean the same as any of the expected answers?”.

In Supplementary Note 6 , we check the quality of our automated ground-truth evaluations against human judgement by hand. We find that GPT-4 gives the best results for determining model accuracy and thus use it in all our sentence-length experiments.

In this section we describe the application of semantic entropy to confabulation detection in longer model generations, specifically paragraph-length biographies.

We introduce a biography-generation dataset—FactualBio—available alongside this paper. FactualBio is a collection of biographies of individuals who are notable enough to have Wikipedia pages but not notable enough to have large amounts of detailed coverage, generated by GPT-4 (v.0613). To generate the dataset, we randomly sampled 21 individuals from the WikiBio dataset 64 . For each biography, we generated a list of factual claims contained in each biography using GPT-4, with 150 total factual claims (the total number is only coincidentally a round number). For each of these factual claims, we manually determined whether the claim was correct or incorrect. Out of 150 claims, 45 were incorrect. As before, we apply confabulation detection to detect incorrect model predictions, even though there may be model errors which are not confabulations.

Prompting and generation

Given a paragraph-length piece of LLM-generated text, we apply the following sequence of steps:

Automatically decompose the paragraph into specific factual claims using an LLM (not necessarily the same as the original).

For each factual claim, use an LLM to automatically construct Q questions which might have produced that claim.

For each question, prompt the original LLM to generate M answers.

For each question, compute the semantic entropy of the answers, including the original factual claim.

Average the semantic entropies over the questions to arrive at a score for the original factual claim.

We pursue this slightly indirect way of generating answers because we find that simply resampling each sentence creates variation unrelated to the uncertainty of the model about the factual claim, such as differences in paragraph structure.

We decompose the paragraph into factual claims using the following prompt:

Please list the specific factual propositions included in the answer above. Be complete and do not leave any factual claims out. Provide each claim as a separate sentence in a separate bullet point.

We found that we agreed with the decompositions in all cases in the dataset.

We then generate six questions for each of the facts from the decomposition. We generate these questions by prompting the model twice with the following:

Following this text: {text so far} You see the sentence: {proposition} Generate a list of three questions, that might have generated the sentence in the context of the preceding original text, as well as their answers. Please do not use specific facts that appear in the follow-up sentence when formulating the question. Make the questions and answers diverse. Avoid yes-no questions. The answers should not be a full sentence and as short as possible, e.g. only a name, place, or thing. Use the format “1. {question} – {answer}”.

These questions are not necessarily well-targeted and the difficulty of this step is the main source of errors in the procedure. We generate three questions with each prompt, as this encourages diversity of the questions, each question targeting a different aspect of the fact. However, we observed that the generated questions will sometimes miss obvious aspects of the fact. Executing the above prompt twice (for a total of six questions) can improve coverage. We also ask for brief answers because the current version of GPT-4 tends to give long, convoluted and highly hedged answers unless explicitly told not to.

Then, for each question, we generate three new answers using the following prompt:

We are writing an answer to the question “{user question}”. So far we have written: {text so far} The next sentence should be the answer to the following question: {question} Please answer this question. Do not answer in a full sentence. Answer with as few words as possible, e.g. only a name, place, or thing.

We then compute the semantic entropy over these answers plus the original factual claim. Including the original fact ensures that the estimator remains grounded in the original claim and helps detect situations in which the question has been interpreted completely differently from the original context. We make a small modification to handle the fact that GPT-4 generations often include refusals to answer questions. These refusals were not something we commonly observe in our experiments with LLaMA 2, Falcon or Mistral models. If more than half of the answers include one of the strings ‘not available’, ‘not provided’, ‘unknown’ or ‘unclear’ then we treat the semantic uncertainty as maximal.

We then average the semantic entropies for each question corresponding to the factual claim to get an entropy for this factual claim.

Despite the extra assumptions and complexity, we find that this method greatly outperforms the baselines.

To compute semantic entailment between the original claim and regenerated answers, we rely on the DeBERTa entailment prediction model as we find empirically that DeBERTa predictions result in higher train-set AUROC than other methods. Because DeBERTa has slightly lower recall than GPT-3.5/4, we use a modified set-up for which we say the answers mean the same as each other if at least one of them entails the other and neither is seen to contradict the other—a kind of ‘non-defeating’ bidirectional entailment check rather than true bidirectional entailment. The good performance of DeBERTa in this scenario is not surprising as both factual claims and regenerated answers are relatively short. We refer to Supplementary Notes 2 and 3 for ablations and experiments regarding our choice of entailment estimator for paragraph-length generations.

We implement two baselines. First, we implement a variant of the P (True) method, which is adapted to the new setting. For each factoid, we generate a question with answers in the same way as for semantic entropy. We then use the following prompt:

Question: {question} Here are some brainstormed ideas: {list of regenerated answers} Possible answer: {original answer} Is the possible answer true? Respond with “yes” or “no”.

As we cannot access the probabilities GPT-4 assigns to predicting ‘yes’ and ‘no’ as the next token, we approximate this using Monte Carlo samples. Concretely, we execute the above prompt ten times (at temperature 1) and then take the fraction of answers which was ‘yes’ as our unbiased Monte Carlo estimate of the token probability GPT-4 assigns to ‘yes’.

As a second, simpler, baseline we check if the model thinks the answer is true. We simply ask:

Following this text: {text so far} You see this statement: {proposition} Is it likely that the statement is true? Respond with ‘yes’ or ‘no’.

It is interesting that this method ought to perform very well if we think that the model has good ‘self-knowledge’ (that is, if “models mostly know what they don’t know” 24 ) but in fact semantic entropy is much better at detecting confabulations.

Data availability

The data used for the short-phrase and sentence-length generations are publicly available and the released code details how to access it. We release a public version of the FactualBio dataset as part of the code base for reproducing the paragraph-length experiments.

Code availability

We release all code used to produce the main experiments. The code for short-phrase and sentence-length experiments can be found at github.com/jlko/semantic_uncertainty and https://doi.org/10.5281/zenodo.10964366 (ref. 65 ). The code for paragraph-length experiments can be found at github.com/jlko/long_hallucinations and https://doi.org/10.5281/zenodo.10964366 (ref. 65 ).

GPT-4 technical report. Preprint at https://arxiv.org/abs/2303.08774 (2023).

Gemini: a family of highly capable multimodal models. Preprint at https://arxiv.org/abs/2312.11805 (2023).

Xiao, Y. & Wang, W. Y. On hallucination and predictive uncertainty in conditional language generation. In Proc. 16th Conference of the European Chapter of the Association for Computational Linguistics 2734–2744 (Association for Computational Linguistics, 2021).

Rohrbach, A., Hendricks, L. A., Burns, K., Darrell, T. & Saenko, K. Object hallucination in image captioning. In Proc. 2018 Conference on Empirical Methods in Natural Language Processing (eds Riloff, E., Chiang, D., Hockenmaier, J. & Tsujii, J.) 4035–4045 (Association for Computational Linguistics, 2018).

Weiser, B. Lawyer who used ChatGPT faces penalty for made up citations. The New York Times (8 Jun 2023).

Opdahl, A. L. et al. Trustworthy journalism through AI. Data Knowl. Eng . 146 , 102182 (2023).

Shen, Y. et al. ChatGPT and other large language models are double-edged swords. Radiology 307 , e230163 (2023).

Article   PubMed   Google Scholar  

Schulman, J. Reinforcement learning from human feedback: progress and challenges. Presented at the Berkeley EECS Colloquium. YouTube www.youtube.com/watch?v=hhiLw5Q_UFg (2023).

Ji, Z. et al. Survey of hallucination in natural language generation. ACM Comput. Surv. 55 , 248 (2023).

Maynez, J., Narayan, S., Bohnet, B. & McDonald, R. On faithfulness and factuality in abstractive summarization. In Proc. 58th Annual Meeting of the Association for Computational Linguistics (eds Jurafsky, D., Chai, J., Schluter, N. & Tetreault, J.) 1906–1919 (Association for Computational Linguistics, 2020).

Filippova, K. Controlled hallucinations: learning to generate faithfully from noisy data. In Findings of the Association for Computational Linguistics: EMNLP 2020 (eds Webber, B., Cohn, T., He, Y. & Liu, Y.) 864–870 (Association for Computational Linguistics, 2020).

Berrios, G. Confabulations: a conceptual history. J. Hist. Neurosci. 7 , 225–241 (1998).

Article   CAS   PubMed   Google Scholar  

Lin, S., Hilton, J. & Evans, O. Teaching models to express their uncertainty in words. Transact. Mach. Learn. Res. (2022).

Evans, O. et al. Truthful AI: developing and governing AI that does not lie. Preprint at https://arxiv.org/abs/2110.06674 (2021).

Amodei, D. et al. Concrete problems in AI safety. Preprint at https://arxiv.org/abs/1606.06565 (2016).

Jiang, Z., Araki, J., Ding, H. & Neubig, G. How can we know when language models know? On the calibration of language models for question answering. Transact. Assoc. Comput. Linguist. 9 , 962–977 (2021).

Article   Google Scholar  

Desai, S. & Durrett, G. Calibration of pre-trained transformers. In Proc. 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (eds Webber, B., Cohn, T., He, Y. & Liu, Y.) 295–302 (Association for Computational Linguistics, 2020).

Glushkova, T., Zerva, C., Rei, R. & Martins, A. F. Uncertainty-aware machine translation evaluation. In Findings of the Association for Computational Linguistics: EMNLP 2021 (eds Moens, M-F., Huang, X., Specia, L. & Yih, S.) 3920–3938 (Association for Computational Linguistics, 2021).

Wang, Y., Beck, D., Baldwin, T. & Verspoor, K. Uncertainty estimation and reduction of pre-trained models for text regression. Transact. Assoc. Comput. Linguist. 10 , 680–696 (2022).

Baker, S. & Kanade, T. Hallucinating faces. In Proc. Fourth IEEE International Conference on Automatic Face and Gesture Recognition . 83–88 (IEEE, Catalogue no PR00580, 2002).

Eliot, L. AI ethics lucidly questioning this whole hallucinating AI popularized trend that has got to stop. Forbes Magazine (24 August 2022).

Shanahan, M. Talking about large language models. Commun. Assoc. Comp. Machinery 67 , 68–79 (2024).

MacKay, D. J. C. Information-based objective functions for active data selection. Neural Comput. 4 , 590–604 (1992).

Kadavath, S. et al. Language models (mostly) know what they know. Preprint at https://arxiv.org/abs/2207.05221 (2022).

Lindley, D. V. On a measure of the information provided by an experiment. Ann. Math. Stat. 27 , 986–1005 (1956).

Article   MathSciNet   Google Scholar  

Xiao, T. Z., Gomez, A. N. & Gal, Y. Wat zei je? Detecting out-of-distribution translations with variational transformers. In Workshop on Bayesian Deep Learning at the Conference on Neural Information Processing Systems (NeurIPS, Vancouver, 2019).

Christiano, P., Cotra, A. & Xu, M. Eliciting Latent Knowledge (Alignment Research Center, 2021); https://docs.google.com/document/d/1WwsnJQstPq91_Yh-Ch2XRL8H_EpsnjrC1dwZXR37PC8/edit .

Negri, M., Bentivogli, L., Mehdad, Y., Giampiccolo, D. & Marchetti, A. Divide and conquer: crowdsourcing the creation of cross-lingual textual entailment corpora. In Proc. 2011 Conference on Empirical Methods in Natural Language Processing 670–679 (Association for Computational Linguistics, 2011).

Honovich, O. et al. TRUE: Re-evaluating factual consistency evaluation. In Proc. Second DialDoc Workshop on Document-grounded Dialogue and Conversational Question Answering 161–175 (Association for Computational Linguistics, 2022).

Falke, T., Ribeiro, L. F. R., Utama, P. A., Dagan, I. & Gurevych, I. Ranking generated summaries by correctness: an interesting but challenging application for natural language inference. In Proc. 57th Annual Meeting of the Association for Computational Linguistics 2214–2220 (Association for Computational Linguistics, 2019).

Laban, P., Schnabel, T., Bennett, P. N. & Hearst, M. A. SummaC: re-visiting NLI-based models for inconsistency detection in summarization. Trans. Assoc. Comput. Linguist. 10 , 163–177 (2022).

Joshi, M., Choi, E., Weld, D. S. & Zettlemoyer, L. TriviaQA: a large scale distantly supervised challenge dataset for reading comprehension. In Proc. 55th Annual Meeting of the Association for Computational Linguistics 1601–1611 (Association for Computational Linguistics. 2017).

Rajpurkar, P., Zhang, J., Lopyrev, K. & Liang, P. SQuAD: 100,000+ questions for machine compression of text. In Proc. 2016 Conference on Empirical Methods in Natural Language Processing (eds Su, J., Duh, K. & Carreras, X.) 2383–2392 (Association for Computational Linguistics, 2016).

Tsatsaronis, G. et al. An overview of the BIOASQ large-scale biomedical semantic indexing and question answering competition. BMC Bioinformatics 16 , 138 (2015).

Article   PubMed   PubMed Central   Google Scholar  

Lee, K., Chang, M.-W. & Toutanova, K. Latent retrieval for weakly supervised open domain question answering. In Proc. 57th Annual Meeting of the Association for Computational Linguistics 6086–6096 (Association for Computational Linguistics, 2019).

Kwiatkowski, T. et al. Natural questions: a benchmark for question answering research. Transact. Assoc. Comput. Linguist. 7 , 452–466 (2019).

Patel, A., Bhattamishra, S. & Goyal, N. Are NLP models really able to solve simple math word problems? In Proc. 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (eds Toutanova, K. et al.) 2080–2094 (Assoc. Comp. Linguistics, 2021).

Touvron, H. et al. Llama 2: open foundation and fine-tuned chat models. Preprint at https://arxiv.org/abs/2307.09288 (2023).

Penedo, G. et al. The RefinedWeb dataset for Falcon LLM: outperforming curated corpora with web data, and web data only. In Proc. 36th Conference on Neural Information Processing Systems (eds Oh, A. et al.) 79155–79172 (Curran Associates, 2023)

Jiang, A. Q. et al. Mistral 7B. Preprint at https://arxiv.org/abs/2310.06825 (2023).

Manakul, P., Liusie, A. & Gales, M. J. F. SelfCheckGPT: Zero-Resource Black-Box hallucination detection for generative large language models. In Findings of the Association for Computational Linguistics: EMNLP 2023 (eds Bouamor, H., Pino, J. & Bali, K.) 9004–9017 (Assoc. Comp. Linguistics, 2023).

Mukhoti, J., Kirsch, A., van Amersfoort, J., Torr, P. H. & Gal, Y. Deep deterministic uncertainty: a new simple baseline. In IEEE/CVF Conference on Computer Vision and Pattern Recognition 24384–24394 (Computer Vision Foundation, 2023).

Schuster, T., Chen, S., Buthpitiya, S., Fabrikant, A. & Metzler, D. Stretching sentence-pair NLI models to reason over long documents and clusters. In Findings of the Association for Computational Linguistics: EMNLP 2022 (eds Goldberg, Y. et al.) 394–412 (Association for Computational Linguistics, 2022).

Barnes, B. & Christiano, P. Progress on AI Safety via Debate. AI Alignment Forum www.alignmentforum.org/posts/Br4xDbYu4Frwrb64a/writeup-progress-on-ai-safety-via-debate-1 (2020).

Irving, G., Christiano, P. & Amodei, D. AI safety via debate. Preprint at https://arxiv.org/abs/1805.00899 (2018).

Der Kiureghian, A. & Ditlevsen, O. Aleatory or epistemic? Does it matter? Struct. Saf. 31 , 105–112 (2009).

Malinin, A. & Gales, M. Uncertainty estimation in autoregressive structured prediction. In Proceedings of the International Conference on Learning Representations https://openreview.net/forum?id=jN5y-zb5Q7m (2021).

Murray, K. & Chiang, D. Correcting length bias in neural machine translation. In Proc. Third Conference on Machine Translation (eds Bojar, O. et al.) 212–223 (Assoc. Comp. Linguistics, 2018).

Holtzman, A., Buys, J., Du, L., Forbes, M. & Choi, Y. The curious case of neural text degeneration. In Proceedings of the International Conference on Learning Representations https://openreview.net/forum?id=rygGQyrFvH (2020).

Fan, A., Lewis, M. & Dauphin, Y. Hierarchical neural story generation. In Proc. 56th Annual Meeting of the Association for Computational Linguistics (eds Gurevych, I. & Miyao, Y.) 889–898 (Association for Computational Linguistics, 2018).

Speaks, J. in The Stanford Encyclopedia of Philosophy (ed. Zalta, E. N.) (Metaphysics Research Lab, Stanford Univ., 2021).

Culicover, P. W. Paraphrase generation and information retrieval from stored text. Mech. Transl. Comput. Linguist. 11 , 78–88 (1968).

Google Scholar  

Padó, S., Cer, D., Galley, M., Jurafsky, D. & Manning, C. D. Measuring machine translation quality as semantic equivalence: a metric based on entailment features. Mach. Transl. 23 , 181–193 (2009).

Androutsopoulos, I. & Malakasiotis, P. A survey of paraphrasing and textual entailment methods. J. Artif. Intell. Res. 38 , 135–187 (2010).

MacCartney, B. Natural Language Inference (Stanford Univ., 2009).

He, P., Liu, X., Gao, J. & Chen, W. Deberta: decoding-enhanced BERT with disentangled attention. In International Conference on Learning Representations https://openreview.net/forum?id=XPZIaotutsD (2021).

Brown, T. et al. Language models are few-shot learners. Adv. Neural Inf. Process. Syst. 33 , 1877–1901 (2020).

Williams, A., Nangia, N. & Bowman, S. R. A broad-coverage challenge corpus for sentence understanding through inference. In Proc. 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (eds Walker, M. et al.) 1112–1122 (Assoc. Comp. Linguistics, 2018).

Yu, L., Hermann, K. M., Blunsom, P. & Pulman, S. Deep learning for answer sentence selection. Preprint at https://arxiv.org/abs/1412.1632 (2014).

Socher, R., Huang, E., Pennin, J., Manning, C. D. & Ng, A. Dynamic pooling and unfolding recursive autoencoders for paraphrase detection. In Proceedings of the 24th Conference on Neural Information Processing Systems (eds Shawe-Taylor, J. et al.) (2011)

He, R., Ravula, A., Kanagal, B. & Ainslie, J. Realformer: Transformer likes residual attention. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 (eds Zhong, C., et al.) 929–943 (Assoc. Comp. Linguistics, 2021).

Tay, Y. et al. Charformer: fast character transformers via gradient-based subword tokenization. In Proceedings of the International Conference on Learning Representations https://openreview.net/forum?id=JtBRnrlOEFN (2022).

Kane, H., Kocyigit, Y., Abdalla, A., Ajanoh, P. & Coulibali, M. Towards neural similarity evaluators. In Workshop on Document Intelligence at the 32nd conference on Neural Information Processing (2019).

Lebret, R., Grangier, D. & Auli, M. Neural text generation from structured data with application to the biography domain. In Proc. 2016 Conference on Empirical Methods in Natural Language Processing (eds Su, J. et al.) 1203–1213 (Association for Computational Linguistics, 2016).

Kossen, J., jlko/semantic_uncertainty: Initial release v.1.0.0. Zenodo https://doi.org/10.5281/zenodo.10964366 (2024).

Download references

Acknowledgements

We thank G. Irving, K. Perlin, J. Richens, L. Rimell and M. Turpin for their comments or discussion related to this work. We thank K. Handa for his help with the human evaluation of our automated accuracy assessment. We thank F. Bickford Smith and L. Melo for their code review. Y.G. is supported by a Turing AI Fellowship funded by the UK government’s Office for AI, through UK Research and Innovation (grant reference EP/V030302/1), and delivered by the Alan Turing Institute.

Author information

These authors contributed equally: Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn

Authors and Affiliations

OATML, Department of Computer Science, University of Oxford, Oxford, UK

Sebastian Farquhar, Jannik Kossen, Lorenz Kuhn & Yarin Gal

You can also search for this author in PubMed   Google Scholar

Contributions

S.F. led the work from conception to completion and proposed using bidirectional entailment to cluster generations as a way of computing entropy in LLMs. He wrote the main text, most of the Methods and Supplementary Information and prepared most of the figures. J.K. improved the mathematical formalization of semantic entropy; led the extension of semantic entropy to sentence- and paragraph-length generations; wrote the code for, and carried out, all the experiments and evaluations; wrote much of the Methods and Supplementary Information and prepared drafts of many figures; and gave critical feedback on the main text. L.K. developed the initial mathematical formalization of semantic entropy; wrote code for, and carried out, the initial experiments around semantic entropy and its variants which demonstrated the promise of the idea and helped narrow down possible research avenues to explore; and gave critical feedback on the main text. Y.G. ideated the project, proposing the idea to differentiate semantic and syntactic diversity as a tool for detecting hallucinations, provided high-level guidance on the research and gave critical feedback on the main text; he runs the research laboratory in which the work was carried out.

Corresponding author

Correspondence to Sebastian Farquhar .

Ethics declarations

Competing interests.

S.F. is currently employed by Google DeepMind and L.K. by OpenAI. For both, this paper was written under their University of Oxford affiliation. The remaining authors declare no competing interests.

Peer review

Peer review information.

Nature thanks Mirella Lapata and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data figures and tables

Extended data fig. 1 algorithm outline for bidirectional entailment clustering..

Given a set of outputs in response to a context, the bidirectional entailment answer returns a set of sets of outputs which have been classified as sharing a meaning.

Supplementary information

Supplementary information.

Supplementary Notes 1–7, Figs. 1–10, Tables 1–4 and references. Includes, worked example for semantic entropy calculation, discussion of limitations and computational cost of entailment clustering, ablation of entailment prediction and clustering methods, discussion of automated accuracy assessment, unaggregated results for sentence-length generations and further results for short-phrase generations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Farquhar, S., Kossen, J., Kuhn, L. et al. Detecting hallucinations in large language models using semantic entropy. Nature 630 , 625–630 (2024). https://doi.org/10.1038/s41586-024-07421-0

Download citation

Received : 17 July 2023

Accepted : 12 April 2024

Published : 19 June 2024

Issue Date : 20 June 2024

DOI : https://doi.org/10.1038/s41586-024-07421-0

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

By submitting a comment you agree to abide by our Terms and Community Guidelines . If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.

Definition of the UMaine VolturnUS-S Reference Platform Developed for the IEA Wind 15-Megawatt Offshore Reference Wind Turbine: IEA Wind TCP Task 37

• National Wind Technology Center
• University of Maine

Research output : NREL › Technical Report

NREL Publication Number

• NREL/TP-5000-76773
• floating wind
• reference design
• reference wind turbine
• semisubmersible

Access to Document

• https://www.nrel.gov/docs/fy20osti/76773.pdf

SME definition

Small and medium-sized enterprises (SMEs) represent 99% of all businesses in the EU. The definition of an SME is important for access to finance and EU support programmes targeted specifically at these enterprises.

What is an SME?

Small and medium-sized enterprises (SMEs) are defined in the EU recommendation 2003/361 .

The main factors determining whether an enterprise is an SME are

• staff headcount
• either turnover or balance sheet total

These ceilings apply to the figures for individual firms only. A firm that is part of a larger group may need to include staff headcount/turnover/balance sheet data from that group too.

Further details include

• The revised user guide to the SME definition (2020) (2 MB, available in all EU languages)
• Declaring your enterprise to be an SME (the form is available in all languages as an annex in the revised user guide)
• The SME self-assessment tool which you can use to determine whether your organisation qualifies as a small and medium-sized enterprise

What help can SMEs get?

There are 2 broad types of potential benefit for an enterprise if it meets the criteria

• eligibility for support under many EU business-support programmes targeted specifically at SMEs: research funding, competitiveness and innovation funding and similar national support programmes that could otherwise be banned as unfair government support ('state aid' – see block exemption regulation )
• fewer requirements or reduced fees for EU administrative compliance

Monitoring of the implementation of the SME definition

The Commission monitors the implementation of the SME definition and reviews it in irregular intervals. Pursuant to the latest evaluation, the Commission concluded that there is no need for a revision.

On 25 October 2021, we informed stakeholders by holding a webinar with presentations on the SME evaluation's results and next steps.

Supporting documents

• Study to map, measure and portray the EU mid-cap landscape (2022)
• Staff working document on the evaluation of the SME definition  (2021)
• Executive summary on the evaluation of the SME definition  (2021)
• Q&A on the evaluation of the SME definition  (2021)
• Final report on evaluation of the SME definition  (2018) (10 MB)
• Final report on evaluation of the SME definition (2012)  (1.8 MB)
• Executive summary on evaluation of the SME definition (2012)  (345 kB)
• Implementing the SME definition (2009)  (50 kB)
• Implementing the SME definition (2006)  (40 kB)

Share this page