Demonstrate ethical conduct in research with adherence to academic integrity standards.
Lectures 1–3 covered how to find, appraise and synthesise literature. This lecture covers the principal way that process goes ethically wrong, and the machinery that detects it.
Two things in this lecture are examined most often: the boundaries between the four types (Section 4), and the ability to interpret a similarity report rather than read the percentage as a verdict (Section 6.3).
Note throughout that the definitions name processes, results, data and software — not merely words. This is the textual basis for code and data plagiarism being plagiarism proper.
1. Learning Objectives
- Define plagiarism in academic and research contexts with precision.
- Identify and differentiate the major types: direct, self, mosaic and accidental.
- Apply recognised citation frameworks (APA, IEEE, ACM, MLA) to mitigate attribution errors.
- Operate commonly used detection tools and interpret similarity reports critically.
- Recognise the ethical, legal and professional consequences of plagiarism in computer science research.
- Formulate institutional and personal strategies to uphold academic integrity.
2. Why This Matters Now
The proliferation of digital repositories, open-access journals, generative AI tools and code-sharing platforms has transformed how knowledge is produced and consumed in computer science. While these advances accelerate innovation, they simultaneously expand the surface area for academic misconduct.
A 2023 analysis by Retraction Watch indicated that plagiarism-related retractions in computer science rose by approximately 38 per cent over the previous decade, with a significant proportion attributed to inadequate paraphrasing and improper reuse of prior work.
The growth is concentrated in poor paraphrase and self-reuse — not in brazen copying. These are precisely the two categories students believe are not plagiarism, which means the realistic risk to a careful student is negligence rather than dishonesty.
For computer scientists, whose research outputs include source code, algorithms, datasets and documentation, the boundaries of acceptable reuse require careful definition. Reuse is normal engineering practice; what distinguishes legitimate from illegitimate reuse is disclosure and licence compliance.
3. Defining Plagiarism
3.1 Etymology and Authoritative Definitions
The term derives from the Latin plagiarius, meaning “kidnapper”, reflecting the historical framing of appropriated ideas as abducted property.
| Source | Definition |
|---|---|
| Office of Research Integrity (USA) | “Plagiarism is the appropriation of another person's ideas, processes, results, or words without giving appropriate credit.” |
| Council of Science Editors | Using the work, ideas or words of others without proper attribution, irrespective of intent. |
| IEEE Publication Policy | Plagiarism includes “copying ideas, text, data, software, or any other material without proper citation.” |
| UGC (India) Regulations 2018 | The practice of taking someone else's work or ideas and passing them off as one's own. |
3.2 Four Core Constituents
- Originality of the source. The appropriated material must originate from a verifiable source.
- Lack of attribution. The borrower fails to provide credit commensurate with the borrowed content.
- Intent or negligence. Either deliberate deception or reckless disregard for citation norms.
- Scope of appropriation. The extent may range from a phrase to an entire manuscript.
Irrespective of intent. Negligence suffices. Most institutional policies, including the UGC Regulations, are constructed on a negligence standard rather than an intent standard — which is why the accidental category in Section 4.4 is not an excuse but a category of offence.
4. Typology of Plagiarism
4.1 Direct Plagiarism
The verbatim copying of text, code, data or visual content from a source without quotation marks, citation, or any indication of the original author.
- Identical reproduction of sentences or paragraphs.
- No use of quotation marks.
- Complete absence of citations.
- Often the most blatant and most easily detected form.
A thesis paragraph reading: “Deep learning is a subset of machine learning that employs neural networks with many layers. These networks can learn complex patterns in large datasets, enabling breakthroughs in image recognition, natural language processing, and autonomous systems.”
If taken word-for-word from a published article without quotation marks or citation, this is direct plagiarism — notwithstanding that the content is textbook-level and may feel like common knowledge. The specific wording belongs to its author.
Textual overlap can be expressed conceptually as:
A directly copied paragraph typically yields an overlap ratio exceeding 85–90 per cent within that segment. Note that this is a segment-level figure, not a document-level one: a whole-document index of 12 per cent can still conceal a wholly copied paragraph. Severity is highest, particularly where the copied material forms a substantial portion of the submission.
4.2 Self-Plagiarism (Auto-Plagiarism)
Reuse of one's own previously published work, or portions of it, without proper disclosure or citation. Although the author retains ownership of the ideas, academic publishing ethics require transparency about prior dissemination.
Manifestations in computer science research:
- Submitting the same research paper to multiple journals — duplicate publication.
- Reusing substantial portions of methodology sections across papers without citation.
- Recycling previously published datasets, code or algorithms in new publications without acknowledgement.
Three harms, each of which exists independently of who wrote the text:
- Misrepresentation of novelty. Reviewers and readers may believe the work is original.
- Copyright infringement. Publishers often hold exclusive rights to the published version.
- Violation of peer review. Multiple submissions waste reviewer effort and editorial resources.
A researcher publishes in Journal A in 2022 on “Efficient Graph Neural Networks for Social Networks”. In 2024 they submit a substantially similar paper to Conference B without citing the 2022 publication. This is self-plagiarism even though the author owns both works.
By contrast, a conference paper extended into a journal article is entirely acceptable when the extension is disclosed, the earlier paper cited, and the added contribution stated — which is exactly what most venues require.
Mitigation: disclose prior related work in the cover letter and the manuscript; use quotation marks and citations when reusing exact text; obtain permission from the original publisher where required; limit textual overlap across manuscripts to cited methodology and standard references.
4.3 Mosaic (Patchwork) Plagiarism
Combining text, ideas or data from multiple sources while making superficial modifications — synonym substitution, sentence restructuring, light paraphrase — without proper citation.
- No single source contributes a dominant share; the document is a “mosaic”.
- Modifications are superficial and do not reflect genuine understanding.
- Citations are either absent or selectively applied.
- Detection requires sophisticated semantic analysis rather than simple string matching.
Original: “Convolutional neural networks have revolutionized image classification by automatically learning hierarchical features from raw pixel data.”
Mosaic version: “CNNs have transformed image categorization by autonomously extracting multi-level representations directly from unprocessed pixel inputs.”
Every content word is substituted, yet the structure, idea and intellectual contribution remain unchanged. This is plagiarism.
Modern tools model meaning rather than surface form. Semantic overlap between an original and a paraphrased passage is quantified by cosine similarity over embedding vectors:
where u and v are vector representations of the original and submitted passages. A value above approximately 0.80 typically signals significant semantic overlap. Near-parallel vectors mean near-identical meaning regardless of surface wording, which is precisely why the “change every word” strategy now fails.
Genuine paraphrase requires comprehension, restructuring of the argument, and citation. Synonym substitution alone is disguise, not scholarship. This is the fastest-growing category in the retraction data and therefore the highest-risk form for this cohort.
4.4 Accidental Plagiarism
Plagiarism that occurs without intent to deceive, arising from ignorance of citation conventions, careless note-taking, or structural imitation of a source.
- Forgotten citations. Reproducing information from a source without having recorded the reference.
- Improper paraphrasing. Failing to transform the original wording sufficiently while neglecting quotation marks.
- Unintentional structural replication. Following the organisational outline of a source too closely.
- Cryptomnesia. Subconscious reproduction of material encountered previously but believed to be original.
- Misattribution. Citing a secondary source rather than the original, or the reverse.
A student summarises a paragraph from a Wikipedia article, omits the citation because they “thought it was common knowledge”, and submits the text in an assignment. Although unintentional, this constitutes accidental plagiarism under most institutional policies.
The absence of malicious intent does not absolve the author. Most academic integrity policies treat accidental plagiarism as a punishable offence, albeit typically with lesser sanctions on a first offence.
Because the causes are procedural, the remedies are procedural: a literature log, a reference manager, and citation capture at the moment of reading rather than at the moment of writing. Cryptomnesia is a documented cognitive phenomenon; it explains, but does not excuse.
4.5 Further Recognised Forms
| Form | Description |
|---|---|
| Idea plagiarism | Appropriating concepts or theoretical frameworks without attribution |
| Code plagiarism | Copying source code, algorithms or software components without credit — particularly relevant in CS |
| Data plagiarism | Reusing datasets without permission or acknowledgement |
| Image plagiarism | Reproducing figures, diagrams or screenshots without citation |
| Translation plagiarism | Translating a foreign-language source and presenting it as original |
| AI-assisted plagiarism | Submitting AI-generated text as one's own, or paraphrasing AI output without disclosure |
Open-source licences — MIT, GPL, Apache, BSD — impose specific attribution formats. Copying a Stack Overflow snippet or a GitHub function without the required notice is simultaneously an academic and a licensing violation, and the licensing obligation is legally binding rather than merely academic.
On AI-assisted work the norms are still settling. Venues differ, but the emerging expectation is disclosure of AI assistance plus author responsibility for verification; unverified AI output additionally risks fabricated citations. Lecture 5 covers the UPES declaration requirement.
5. How Detection Works
| Approach | Methodology | Strengths | Limitations |
|---|---|---|---|
| String matching | Exact or near-exact text comparison | Fast, transparent | Fails on paraphrased content |
| Fingerprinting | Hash-based indexing of n-grams | Efficient for large corpora | Sensitive to word-order changes |
| Stylometry | Statistical analysis of writing style | Detects ghostwriting | Requires substantial reference text |
| Semantic similarity | NLP-based embedding comparison | Identifies paraphrasing | Computationally intensive |
| Citation analysis | Examination of citation patterns | Detects missing citations | Limited to structured documents |
String matching catches direct plagiarism; fingerprinting scales that comparison to web-sized corpora; semantic similarity is what catches mosaic plagiarism; stylometry addresses authorship rather than overlap. Be able to explain n-gram fingerprinting and cosine similarity qualitatively — hash overlapping n-grams and compare hash sets, so comparison cost becomes independent of document length.
Source code needs its own algorithms: identifier renaming, statement reordering and formatting changes defeat text comparison entirely, which is why token-based and structural methods exist.
6. Detection Tools
6.1 Text and Manuscript Tools
- Turnitin — the most widely deployed platform for student submissions. Compares against internet sources, student-paper repositories and publications; outputs a similarity index with highlighted matching passages; integrates with learning management systems. Does not detect well-paraphrased text or ideas, and thresholds vary by institution.
- iThenticate — preferred by journal publishers for research manuscripts; built on the CrossCheck database of published scholarly content; integrated with editorial workflows. Subscription-based and primarily text-based.
- Grammarly — lightweight checking integrated with writing assistance; convenient for early drafts, with a limited database.
- Copyscape — effective for detecting unauthorised online republication; less suitable for academic manuscript review.
- PlagScan — enterprise and institutional deployment; configurable sensitivity and multi-format support; smaller database than Turnitin.
- Urkund (Ouriginal) — European institutional deployments; multilingual support; LMS integration.
6.2 Source-Code and AI Detectors
- MOSS (Measure of Software Similarity) — designed specifically for source code, using winnowing and fingerprinting algorithms adapted for programming structure. The standard tool for programming assignments.
- JPlag — open-source, multi-language, token-based comparison.
- AI-generated text detectors — GPTZero, Originality.ai, Turnitin AI Writing Detection, Copyleaks AI Content Detector. These analyse perplexity, burstiness and stylistic patterns to estimate the likelihood of machine generation.
Reliability remains contested, with documented false-positive rates. Non-native English writers and heavily edited prose are disproportionately flagged. Such output is a prompt for human inquiry, never evidence of misconduct on its own. Note also that no tool detects idea plagiarism, which is why human review remains irreplaceable.
6.3 Interpreting Similarity Reports
A high similarity index does not automatically imply plagiarism, and a low index does not prove its absence. Proper interpretation requires five steps:
- Source identification. Determine whether matches originate from properly cited sources.
- Quoted material recognition. Exclude legitimately quoted passages with quotation marks.
- Common knowledge exclusion. Phrases or terminology widely accepted as common knowledge.
- Methodology overlap. Standard methods or equations appear legitimately across multiple papers.
- Reference list exclusions. Most tools exclude bibliographies — verify that this is configured correctly.
| Similarity index | Typical interpretation |
|---|---|
| Below 10% | Generally acceptable |
| 10–20% | Review required |
| 20–40% | Significant overlap; likely requires revision |
| Above 40% | High concern; requires major restructuring or proper citation |
These thresholds are indicative and are institution- and journal-specific. IEEE permits up to 30 per cent similarity for review articles under certain conditions, and the UGC Regulations 2018 set graded similarity levels for theses and dissertations with escalating consequences, beginning with a tolerance of up to 10 per cent. Confirm your own programme's threshold rather than relying on the indicative bands.
Treating a 32 per cent index composed entirely of quoted, cited and bibliographic matches as misconduct — and treating a 6 per cent index as clearance when that 6 per cent is an uncited verbatim paragraph. Both are failures of interpretation, and avoiding them is the skill being assessed.
7. Case Studies
7.1 The Hwang Woo-suk Affair (2004–2005)
The South Korean researcher published landmark papers in Science claiming successful cloning of human embryonic stem cells. Investigations revealed that data had been fabricated and key figures duplicated from earlier publications; the incident included image reuse and uncredited appropriation of research methodologies. Papers were retracted, funding revoked, and a criminal conviction followed, prompting reform of research-integrity oversight.
Lesson: detection tools applied rigorously to figures and data can reveal inconsistencies that textual review misses.
7.2 Retraction of a CS Paper for Self-Plagiarism (2019)
A paper in a Springer journal was retracted after editors identified that the authors had published substantially overlapping content in two separate venues without cross-citation.
Lesson: self-plagiarism is a serious ethical violation even when the author owns both works. Maintain a careful publication record and disclose prior related work.
7.3 Code Plagiarism in Programming Courses
Universities report widespread code plagiarism in introductory programming courses. Studies using MOSS have shown that in some cohorts 15–25 per cent of submitted programming assignments contain significant code overlap with other submissions.
Mitigation strategies: use MOSS or JPlag for automated detection; design assignments with unique parameters per student; conduct viva voce examinations to verify understanding; and educate students about academic integrity policies from the outset of the course.
8. Consequences
| Level | Consequences |
|---|---|
| Academic | Failure or zero grade; course failure or academic probation; suspension or expulsion; revocation of degrees in severe cases |
| Professional | Retraction of published papers; blacklisting from journals and conferences; loss of research funding; termination of employment |
| Legal | Copyright infringement lawsuits; breach of publisher agreements; damages and injunctions in civil proceedings |
| Reputational | Long-term damage to professional credibility; difficulty securing collaborations; public retraction notices affecting future publications |
Note the asymmetry: the reputational consequence outlasts every other category, because a retraction notice is itself a permanent, indexed, citable record attached to the original paper.
9. Best Practices
- Begin research with a structured literature log, recording quotations and paraphrases with their sources.
- Use reference management tools — Zotero, Mendeley or EndNote — to organise citations systematically.
- Master the citation style mandated by your target venue (APA, IEEE, ACM, MLA, Chicago).
- Paraphrase responsibly: comprehend the original, restructure the argument, and cite the source.
- Use quotation marks for any verbatim text beyond standard terminology.
- Attribute ideas, not just text. Even paraphrased concepts require citation.
- Disclose prior work when extending your own publications.
- For code, attribute origins. Open-source licences (MIT, GPL, Apache) require specific attribution formats.
- Use AI tools responsibly. Disclose assistance where the venue requires it, and verify generated content against sources.
- Run pre-submission checks using institutional access to Turnitin or iThenticate.
Capture the citation at the moment of reading, not at the moment of writing. Nine of these ten safeguards fail silently if deferred to the drafting stage. Note also that a pre-submission check is diagnostic, not exculpatory: it finds negligence before a reviewer does, but running it does not license whatever it fails to flag.
10. Summary
- Plagiarism is the uncredited appropriation of another's intellectual work, encompassing ideas, text, data and code.
- The principal types are direct, self, mosaic and accidental, each with distinct characteristics and severities.
- Modern detection employs fingerprinting, NLP-based semantic similarity and stylometric analysis.
- Similarity indices must be interpreted contextually; quoted material, references and common knowledge all affect interpretation.
- Computer science introduces unique challenges, particularly regarding code plagiarism and AI-assisted authorship.
- Rigorous citation practices, reference management and pre-submission checks are the essential safeguards.
11. Exam Preparation
- Distinguish definitional nuances. Be precise about the differences between direct, self, mosaic and accidental plagiarism. Examiners test comprehension of the boundaries.
- Memorise institutional definitions — UGC, IEEE and ACM.
- Understand detection algorithms. Explain fingerprinting, n-gram analysis and cosine similarity qualitatively.
- Know the tools and their primary use cases — MOSS for code, iThenticate for manuscripts, Turnitin for student work.
- Practise applied scenarios. Expect case-study questions requiring you to identify the type and recommend remediation.
- Cite correctly in answers, even under examination conditions.
11.1 Short-Answer Questions
- Define plagiarism as per the Office of Research Integrity. 2 marks
- Differentiate between direct plagiarism and mosaic plagiarism, with one example of each. 3 marks
- What is self-plagiarism? Why is it unethical even though the author owns the original work? 3 marks
- List any four causes of accidental plagiarism. 2 marks
- Name two plagiarism detection tools designed specifically for source code. 2 marks
11.2 Long-Answer Questions
- Discuss the four major types of plagiarism in detail, providing a real-world or hypothetical example for each. 10 marks
- Explain the working principles of plagiarism detection tools. Compare string-matching approaches with semantic similarity methods. 10 marks
- Describe the consequences of plagiarism at academic, professional and legal levels. What policies has the UGC framed to address plagiarism in higher education? 10 marks
- MOSS and JPlag are used for source-code plagiarism detection. Compare their methodologies and explain why specialised tools are needed for code. 8 marks
- With the rise of generative AI tools, how has the definition and detection of plagiarism evolved? Discuss the ethical responsibilities of researchers. 10 marks
11.3 Application Question
You are reviewing a manuscript in which the authors have reused four paragraphs verbatim from their own 2021 conference paper without citation or quotation marks, and have paraphrased two additional paragraphs from a 2019 journal article without attribution.
Identify each instance of plagiarism, classify its type, and recommend the corrective action in each case.
12. Further Reading
- Roig, M. (2015). Avoiding Plagiarism, Self-Plagiarism, and Other Questionable Writing Practices: A Guide to Ethical Writing. Office of Research Integrity.
- Anderson, M. S., & Steneck, N. H. (2011). The Problem of Plagiarism. Urologic Oncology.
- Wager, E. (2014). Defining and responding to plagiarism. Learned Publishing, 27(1), 33–42.
- IEEE Publication Services and Products Board Operations Manual — sections on author ethics and plagiarism.
- UGC (India) (2018). University Grants Commission (Promotion of Academic Integrity and Prevention of Plagiarism in Higher Educational Institutions) Regulations, 2018.