I. Core Legal Principles

Translations as Derivative Works

Under most copyright systems, a translation is a derivative work.

Permission from the original copyright holder is generally required.

The translator’s creative input may itself be protected.

AI as Translator

AI cannot hold copyright in most jurisdictions.

Human involvement is required for copyright protection in AI-assisted translations.

Training Data Legality

Using copyrighted material to train AI may be allowed under fair use (U.S.) or Text & Data Mining (EU).

Unauthorized reproduction of copyrighted text may constitute infringement.

Market Substitution Risk

AI translations that replace licensed human translations can harm copyright holders’ market rights.

II. Key Case Laws

1. Authors Guild v. Google, Inc.

Background:
Google Books scanned millions of copyrighted books, making snippets searchable.

Holding:

The U.S. Court of Appeals ruled that Google’s use was fair use.

Transformative purposes (search, analysis) justify reproduction.

Implications for AI Translation:

Training AI on copyrighted literary works may be considered fair use in the U.S.

However, direct reproduction of expressive translations may not be allowed.

2. Authors Guild v. HathiTrust

Background:
Digital library project scanned copyrighted books for accessibility and search.

Holding:

Fair use was affirmed for search and accessibility purposes.

AI Translation Implication:

Text mining for AI training can be fair use if the AI output is transformative and does not substitute the original work.

3. Infopaq International A/S v. Danske Dagblades Forening

Background:
A Danish newspaper extracted 11-word text snippets for indexing.

Holding:

Even small portions can be protected if they reflect the author’s intellectual creation.

AI Translation Implication:

In the EU, using copyrighted text—even temporarily—for AI training may infringe unless covered by Text & Data Mining exceptions.

AI translations that closely mirror protected expression can violate copyright.

4. Naruto v. Slater

Holding:

Only humans can be authors.

AI-generated works without meaningful human input cannot be copyrighted.

AI Translation Implication:

Purely AI-generated translations without human editing are unlikely to be protected.

Human oversight or post-editing is essential for copyright protection.

5. Thaler v. Perlmutter

Background:
Stephen Thaler sought copyright for AI-generated artwork.

Holding:

Human authorship is required.

AI-alone cannot create copyrightable works.

AI Translation Implication:

AI translations require human creative input for copyright protection.

Autonomous AI output falls into public domain, raising licensing and infringement concerns if based on copyrighted texts.

6. Andy Warhol Foundation v. Goldsmith

Holding:

Transformative use doctrine does not automatically allow commercial exploitation if it substitutes for the original market.

AI Translation Implication:

AI translations that replace licensed translations may not qualify as fair use, even if transformative.

Market harm is a key factor in U.S. copyright evaluation.

7. SAS Institute Inc. v. World Programming Ltd

Holding:

Functionality, ideas, and methods of software are not copyrightable.

The expressive elements of software are protected.

AI Translation Implication:

Linguistic rules learned by AI are not copyrightable.

But reproducing a copyrighted translation verbatim may infringe.

8. Feist Publications, Inc. v. Rural Telephone Service Co.

Holding:

Originality is required.

Mere factual compilation or automated output without human creative input is not protected.

AI Translation Implication:

Machine-generated translations without creative human input may lack copyright protection.

Human post-editing or adaptation is necessary to claim authorship.

III. Practical Governance Considerations

Training Data Licensing

Secure licenses for copyrighted material in AI training datasets, especially outside U.S. fair use.

Derivative Work Clearance

AI-generated translations may be derivative works.

Obtaining permission from original authors is prudent.

Human Creative Input

Post-editing or curated human supervision is required to create copyrightable output.

Market Substitution Risk

AI translations that compete with licensed translations may face liability.

International Jurisdiction

U.S.: Fair use is broader.

EU: TDM exceptions are narrow and require legal authorization.

UK: TDM and originality standards similar to EU.

IV. Summary Table of Key Cases

V. Conclusion

AI in automated translation publishing operates at a crossroads of copyright law:

Training phase: Can sometimes be covered under fair use (U.S.) or TDM (EU).

Output phase: AI-generated translations are derivative works; human creative input is required for copyright.

Market substitution: Commercial AI translations may infringe if they replace authorized translations.

International divergence: U.S. courts more permissive; EU stricter on temporary copying and derivative works.

The landscape is still evolving, with ongoing debates over AI authorship, fair use, and cross-border licensing.