Copyright Implications Of AI-Based Book Translations In Swahili.
📌 1) Authors Guild v. Google Books (U.S. 2015) – Automated Processing vs. Unauthorized Reproduction
Facts:
Google scanned millions of copyrighted books to create a searchable database, allowing users to view limited text snippets.
Holding:
The U.S. District Court (and reaffirmed on appeal) held this was a fair use because the transformation (searchability) was novel and did not harm the market for the original works.
Legal Insight for AI Translation:
Automated transformation of text — even if derived from copyrighted works — can be fair use if the use is highly transformative, and does not serve as a market substitute.
A machine translation (e.g., from English to Swahili) that adds commentary, analysis, or is used for research/educational purposes may lean toward fair use under similar reasoning.
However, mere translation without significant transformation (a literal translation that readers can consume instead of the original) may not qualify as fair use.
Key Principles from the Case:
Transformative use increases fair use likelihood.
Amount and substantiality matter.
Effect on market is critical.
📌 2) Thaler / DABUS (U.S. 2023) – AI Authorship Rule
Facts:
An AI system called DABUS was used to create "works," and its owner sought copyright registrations for them. The Copyright Office rejected the registrations.
Holding:
The court confirmed only human authors can hold copyright; machine‑generated works are ineligible if human creative contribution is insufficient.
Implications for AI Translation:
An AI translation alone — without meaningful human editorial control — cannot itself hold copyright.
If a news organization, publisher, or translator wants copyright in an AI‑assisted Swahili translation, there must be significant human editorial input: choices about style, phrasing, nuance, cultural adjustments, etc.
Machine alone cannot own rights and may weaken protection for the translated text.
Key Principles from the Case:
Human creativity is essential in derivative creations.
AI functions as a tool; it does not generate protectable authorship.
📌 3) Zarya of the Dawn / AI Artwork (U.S. 2023) – Human Input Threshold
Facts:
The U.S. Copyright Office denied copyright registration for artwork generated mostly by AI (Midjourney) with minimal prompting from humans.
Holding:
If human contribution is limited to basic prompts and minimal editing, the resulting work does not have sufficient human authorship for copyright.
Implication for Translation:
Simple selection of an AI system and minimal oversight (e.g., clicking “translate”) is not enough to create a copyrightable Swahili translation.
Publishers must demonstrate substantive human editing and creative input (choices about idioms, cultural context, reader audience, etc.) to claim copyright.
Key Principles:
Quantity and quality of human contribution are critical.
Courts look at how much creative decision‑making humans add.
📌 4) Klinger v. Google (Australia, 2018) – Automated Processing and Copyright
Facts:
The plaintiff sued Google for automated copying in text mining (keyword indexing) of newspaper archives.
Holding:
Australian courts held that automated processing can infringe if it reproduces substantial parts of copyrighted works and does not qualify for fair dealing exceptions.
Implications for AI Translation:
Automated translation that outputs large passages of a copyrighted book into Swahili without authorization may constitute unauthorized reproduction, similar to automated copying.
Merely transforming language does not avoid reproduction if the substantial expressive content remains intact.
Key Principles:
Automated tools matter, but the output and how it reproduces copyrighted elements is decisive.
📌 5) Andersen, McKernan & Ortiz v. AI Art Platforms (U.S. 2025) – Training Data Liability
Facts:
Artists sued AI art generators alleging their works were used without permission to train models. The court allowed claims to proceed.
Relevance:
Although about art, the reasoning applies to text models: if an AI system was trained on copyrighted books without permission, generating translations could implicate liability related to training data usage.
Implications for AI Translation:
If a text translation AI was trained on copyrighted books without licenses, translators and publishers could risk liability not just for translation output but based on the training process itself.
Legal risk attaches even if the translation is novel.
Key Principles:
Training on copyrighted materials can create liability if models generate recognizably copyrighted derivatives.
Copyright isn’t only about output — how a model is trained matters legally.
📌 6) Feist Publications v. Rural Telephone (U.S. 1991) – Facts vs. Expression
Facts:
Feist concerned whether a phone directory’s listings were copyrightable (they were not, because facts alone lack originality).
Relevance to Translation:
Individual facts in a book (e.g., historical facts, data) cannot be copyrighted.
But literary expression and creative elements (plot, style, metaphors) are protected.
An AI translating facts (e.g., research citations) into Swahili may be risk‑free, but translating creative expression carries stronger copyright implications.
Key Principles:
Factual content is not protected, but expression is.
Copyright prohibits copying expression, not facts.
📌 7) Donald v. Beckett (UK 1774) – Derivative Rights
Note:
While a much older case, Donald v. Beckett is foundational in affirming that translation is a derivative right reserved exclusively to the original author or rights holder unless permission is granted.
Legal Principle:
Translation (even human translation) is a derivative work that requires authorization from the original copyright holder.
For AI translation into Swahili, this holds: permission is needed unless an exception (like fair use/fair dealing) applies.
🔍 Key Copyright Issues in AI-Based Book Translations into Swahili
🧠1) AI Translation Generates a Derivative Work
Translation is explicitly a derivative right under copyright laws worldwide (U.S., EU, India).
AI‑generated translations qualify as derivative works since they convert expression from one language (English/French/etc.) into another (Swahili).
Unauthorized translation is infringement.
🧑‍🎨 2) Human Authorship & Copyright Eligibility
AI alone cannot hold copyright.
To claim copyright in a Swahili translation, the human translator/editor must exercise meaningful creative judgment, including:
Cultural adjustment of phrases,
Editing for style and audience,
Revision beyond raw AI output.
đź“Š 3) Fair Use / Fair Dealing & Context
Fair Use Factors (U.S.):
Purpose & character (transformative, educational, noncommercial stronger),
Nature of original work (fiction harder),
Amount used,
Effect on market.
AI translation for education or scholarship may be more defensible than commercial publishing without permission.
📦 4) Training Data & Secondary Liability
If AI systems are trained on copyrighted books without authorization, liability may attach:
To the model’s output,
To the publisher or user relying on that output,
Even if the translation appears novel.
đź“Ś 5) Market Harm & Substitute Effect
An AI‑generated Swahili translation — if published widely — may substitute for an authorized translation, undermining the market for licensed Swahili editions, increasing risk of infringement claims.
🛠Practical Takeaways for AI‑Based Swahili Translations
âś” Obtain permission from copyright holders before translating entire books.
âś” Use human editors to meaningfully revise and localize AI output.
âś” Favor public domain works or works with explicit translation rights.
✔ Carefully assess training data origins — don’t assume AI models are free from prior licensing issues.
âś” Document editorial decisions to support copyright claims.
🧠Conclusion
AI‑based translation into Swahili raises multiple copyright issues:
Derivative works require authorization.
AI is not an author; human creative input matters.
Training data origins can create liability.
Fair use offers limited defenses — stronger in noncommercial and transformative contexts.
RELATED Blog
comments