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

arXiv: https://arxiv.org/abs/2603.08869 · HTML: https://mlsanity.com/paper/2603.08869/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in SAE features and their applications, one of 13 directions in Mechanistic interpretability.
- Builds on 9 of its 13 references that are in the map.
- First posted on arXiv 9 Mar 2026.

## Placed in

- Mechanistic interpretability / [SAE features and their applications](https://mlsanity.com/mechinterp/sae-features-and-their-applications/)

## Builds on (most cited first, 9 of 9)

- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [How multilingual is Multilingual BERT?](https://mlsanity.com/paper/1906.01502/) (2019)
- [Language-agnostic BERT Sentence Embedding](https://mlsanity.com/paper/2007.01852/) (2020)
- [Beto, Bentz, Becas: The Surprising Cross-Lingual Effectiveness of BERT](https://mlsanity.com/paper/1904.09077/) (2019)
- [Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders](https://mlsanity.com/paper/2407.14435/) (2024)
- [Does Transliteration Help Multilingual Language Modeling?](https://mlsanity.com/paper/2201.12501/) (2022)
- [Breaking the Script Barrier in Multilingual Pre-Trained Language Models with Transliteration-Based Post-Training Alignment](https://mlsanity.com/paper/2406.19759/) (2024)

## Built on by (most cited first, 1 of 1)

- [The Latin Substrate: How Language Models Represent and Mediate Script Choice](https://mlsanity.com/paper/2605.31363/) (2026)

## Abstract (arXiv)

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.