AI Navigate

One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations

arXiv cs.CL / 3/11/2026

Ideas & Deep AnalysisModels & Research

Key Points

  • The study investigates whether Sparse Autoencoders (SAEs) learn features representing abstract meaning or are tied to the text's orthographic form by using Serbian digraphia, which employs two scripts (Latin and Cyrillic) with a perfect character mapping.
  • Analysis across the Gemma models (270M-27B parameters) shows that identical sentences in different scripts activate highly overlapping features, indicating that the models prioritize meaning over script form.
  • Representational divergence caused by script changes is smaller than divergence caused by paraphrasing within the same script, suggesting semantic abstraction beyond surface tokenization.
  • Cross-script and cross-paraphrase feature overlaps provide evidence against mere memorization since these combinations rarely occur in training data.
  • The degree of script invariance in learned representations strengthens with increasing model size, and Serbian digraphia is proposed as a benchmark to evaluate the abstraction level of learned features.

Computer Science > Computation and Language

arXiv:2603.08869 (cs)
[Submitted on 9 Mar 2026]

Title:One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations

Authors:Sripad Karne
View a PDF of the paper titled One Language, Two Scripts: Probing Script-Invariance in LLM Concept Representations, by Sripad Karne
View PDF HTML (experimental)
Abstract:Do the features learned by Sparse Autoencoders (SAEs) represent abstract meaning, or are they tied to how text is written? We investigate this question using Serbian digraphia as a controlled testbed: Serbian is written interchangeably in Latin and Cyrillic scripts with a near-perfect character mapping between them, enabling us to vary orthography while holding meaning exactly constant. Crucially, these scripts are tokenized completely differently, sharing no tokens whatsoever. Analyzing SAE feature activations across the Gemma model family (270M-27B parameters), we find that identical sentences in different Serbian scripts activate highly overlapping features, far exceeding random baselines. Strikingly, changing script causes less representational divergence than paraphrasing within the same script, suggesting SAE features prioritize meaning over orthographic form. Cross-script cross-paraphrase comparisons provide evidence against memorization, as these combinations rarely co-occur in training data yet still exhibit substantial feature overlap. This script invariance strengthens with model scale. Taken together, our findings suggest that SAE features can capture semantics at a level of abstraction above surface tokenization, and we propose Serbian digraphia as a general evaluation paradigm for probing the abstractness of learned representations.
Comments:
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2603.08869 [cs.CL]
  (or arXiv:2603.08869v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2603.08869
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Sripad Karne [view email]
[v1] Mon, 9 Mar 2026 19:31:20 UTC (1,038 KB)
Full-text links:

Access Paper:

Current browse context:
cs.CL
< prev   |   next >
Change to browse by:
cs

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.