Rework AI-Drafted Nepali, Then Check Every Sentence
SupWriter's AI humanizer includes a Nepali setting for reworking AI-assisted drafts written in Devanagari. Our only Nepali test run so far came back hard to read, so treat any rewrite as a rough draft and check every sentence. For a thesis at Tribhuvan University (TU), the numbers that matter are in its anti-plagiarism regulation: an overall similarity index of up to 10% for a PhD, 15% for an MPhil and 20% for a Master's, with a certified report attached before the final viva.
Tribhuvan University's plagiarism regulation, level by level
The key document for a Nepali thesis writer is Tribhuvan University's Anti-Plagiarism Software Management and Operation Regulation, 2080, which the TU Executive Council approved with amendments at its meeting of Shrawan 22, 2082 BS (2025) [1].
- Checks are compulsory. PhD, MPhil and Master's theses and research reports must go through a plagiarism check, and the certified report has to be attached before the final viva [1].
- Each level has its own ceiling. The overall similarity index may reach 10% for a PhD thesis, 15% for an MPhil, 20% for a Master's, 15% for mini-research, 10% for other research and 15% for a research article [1].
- The central library runs the software. TU has handed its Central Library (TUCL) central authority over anti-plagiarism software and services, and TUCL says the tests are mandatory for all campuses, research units and any academic writing submitted to TU [2]. TUCL issues user IDs to TU units and affiliated units on request, and its 2024 user manual names the software as iThenticate [2][3].
- Faculties can go further. A Dean's Office letter from TU's Institute of Science and Technology, dated 2081/5/19 BS (2024), made checks compulsory for project work, field reports, term papers, dissertations and internship reports from bachelor's to PhD level, with iThenticate set to a similarity index below 20% for bachelor's and master's work and below 10% for PhD work, quotes and references excluded [4].
Before you submit, ask your department which of these settings applies to your programme.
- त्रिभुवन विश्वविद्यालय एण्टी प्लेजारिजम सफ्टवेयर व्यवस्थापन तथा सञ्चालन सम्बन्धी विनियम, २०८० (Anti-Plagiarism Software Management and Operation Regulation, 2080), as amended by Executive Council decision no. 52, 2082 BS, and decision no. 102, 2082 Bhadra 13 — Tribhuvan University. Accessed 2026-09-27.
- Plagiarism (anti-plagiarism software and services) — Tribhuvan University Central Library (TUCL). Accessed 2026-09-27.
- TUCL Anti-Plagiarism Software (iThenticate) User Manual 2024 — Tribhuvan University Central Library (TUCL). Accessed 2026-09-27.
- IOST plagiarism-check guideline: Dean's Office letter of 2081/5/19 BS (notice 4395) — Central Department of Geology, Tribhuvan University. Accessed 2026-09-27.
What TU's AI clause means when no detector scores Nepali
As far as we could find, no mainstream AI detector scores Nepali. Turnitin's AI-writing detection covers four languages in long-form prose (English, Spanish, Japanese and Modern Arabic), and a paper in any other language comes back with an empty or error indicator [1]. Nepali is also missing from Turnitin's Translated Matching list [2], from the 31 languages Originality.ai names [3] and from the five GPTZero fully supports [4]. In India, DrillBit, the checker supplied through the ShodhShuddhi programme, doesn't list Nepali either [5].
Section 10(3) of the regulation says documents in which “AI Detected Content” is found during plagiarism checking may not be accepted, and that the acceptable level will be as specified [6]. TU's Executive Council specified it on Bhadra 13, 2082 BS (decision no. 102, annexed to the same document): AI thresholds of up to 10% for a PhD thesis, 15% for an MPhil, 20% for a Master's, 15% for mini-research, 10% for other research and 15% for a research article, or as the relevant body specifies [6]. The same decision says TU runs its checks on Turnitin's iThenticate [6], and Turnitin's AI detection doesn't process Nepali [1], so these limits can only be measured on work written in a supported language such as English.
SupWriter's own detector is no help for Nepali. It gave a fully AI-written Nepali draft 0% AI and the rewrite 0% as well, which means it cannot evaluate the language [7]. We don't showcase a Nepali example yet, because our single test rewrite lost coherence after its opening sentence.
Ask your department which tool it uses and how it applies TU's AI threshold to your programme, keep your notes and drafts, and declare any AI help your programme permits.
- Turnitin's AI writing detection capabilities FAQ — Turnitin. Accessed 2026-09-26.
- Account settings for Turnitin Feedback Studio and Originality Check (translated matching languages) — Turnitin. Accessed 2026-09-26.
- Multiple Languages: How To Use Multi Language AI Detection — Originality.ai. Accessed 2026-09-26.
- GPTZero homepage: Which languages does GPTZero support? — GPTZero. Accessed 2026-09-26.
- Supported languages — DrillBit. Accessed 2026-09-27.
- त्रिभुवन विश्वविद्यालय एण्टी प्लेजारिजम सफ्टवेयर व्यवस्थापन तथा सञ्चालन सम्बन्धी विनियम, २०८० (Anti-Plagiarism Software Management and Operation Regulation, 2080), as amended by Executive Council decision no. 52, 2082 BS, and decision no. 102, 2082 Bhadra 13 — Tribhuvan University. Accessed 2026-09-27.
- How We Test AI Detectors (and What We Have Measured So Far) — SupWriter. Accessed 2026-09-27.
What Makes Nepali Different
Where AI-written Nepali slips
- •Three-tier honorific verb agreement (ta/timi/tapai)
- •Split-ergativity in the perfective aspect
- •Hindi calques in lexicon and postpositions
- •Gender-agreement patterns distinct from Hindi
What native Nepali writers do
- Honorific verb forms matched to social tier
- Nepali-specific lexicon over Hindi loans
- Particles 'ni', 'ra', 'hai', 'ta'
Writing notes
- Select Nepali in the editor, then set the rewrite beside your draft and go through it one sentence at a time, checking every figure and name. Keep only the sentences that still say what you meant.
- Work in Unicode Devanagari from the first draft. Section 17 of TU's anti-plagiarism regulation requires Devanagari theses to be prepared and submitted in Unicode, and theses not received in Unicode are left out of the plagiarism check, so convert anything typed in a legacy Nepali font before you paste it anywhere.
- End sentences in formal Nepali with the danda (।), as TU's own regulations do, which the Unicode Standard describes as similar to a full stop. After rewriting or pasting, make sure each danda is followed by a space before the next sentence.
- Set direct quotations apart and cite them. TU's regulation leaves quotes, bibliographies, acknowledgements, tables of contents and matches of up to ten words out of the similarity count, but a quotation that a tool has reworded into your own sentence no longer counts as a quote.
- Confirm the language before you draft. A review article in TU's Annual Day Publication 2026 notes that students may write exams in English or Nepali, especially at bachelor's level, yet a thesis or assignment may still require one of them, and translating at the last minute invites mistakes.
FAQ
Nepali questions
Humanize AI Text in More Languages
Transform AI-generated content into natural, human-like text in any language. Our AI humanizer supports all major world languages.
