# Unveiling Multilinguality in Transformer Models: Exploring Language Specificity in Feed-Forward Networks

arXiv: https://arxiv.org/abs/2310.15552 · HTML: https://mlsanity.com/paper/2310.15552/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 23 references that are in the map.
- First posted on arXiv 24 Oct 2023.

## Builds on (most cited first, 16 of 16)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [HuggingFace's Transformers: State-of-the-art Natural Language Processing](https://mlsanity.com/paper/1910.03771/) (2019)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [BERT Rediscovers the Classical NLP Pipeline](https://mlsanity.com/paper/1905.05950/) (2019)
- [Eliciting Latent Predictions from Transformers with the Tuned Lens](https://mlsanity.com/paper/2303.08112/) (2023)
- [Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space](https://mlsanity.com/paper/2203.14680/) (2022)
- [MoEfication: Transformer Feed-forward Layers are Mixtures of Experts](https://mlsanity.com/paper/2110.01786/) (2021)
- [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](https://mlsanity.com/paper/2205.06266/) (2022)
- [Investigating Multilingual NMT Representations at Scale](https://mlsanity.com/paper/1909.02197/) (2019)
- [How Language-Neutral is Multilingual BERT?](https://mlsanity.com/paper/1911.03310/) (2019)
- [When is BERT Multilingual? Isolating Crucial Ingredients for Cross-lingual Transfer](https://mlsanity.com/paper/2110.14782/) (2021)
- [Kformer: Knowledge Injection in Transformer Feed-Forward Layers](https://mlsanity.com/paper/2201.05742/) (2022)
- [Analyzing the Mono- and Cross-Lingual Pretraining Dynamics of Multilingual Language Models](https://mlsanity.com/paper/2205.11758/) (2022)
- [Rethinking the Value of Transformer Components](https://mlsanity.com/paper/2011.03803/) (2020)
- [Announcing CzEng 2.0 Parallel Corpus with over 2 Gigawords](https://mlsanity.com/paper/2007.03006/) (2020)

## Built on by (most cited first, 13 of 13)

- [Sharing Matters: Analysing Neurons Across Languages and Tasks in LLMs](https://mlsanity.com/paper/2406.09265/) (2024)
- [Exploring Multilingual Concepts of Human Value in Large Language Models: Is Value Alignment Consistent, Transferable and Controllable across Languages?](https://mlsanity.com/paper/2402.18120/) (2024)
- [Multilingual Knowledge Editing with Language-Agnostic Factual Neurons](https://mlsanity.com/paper/2406.16416/) (2024)
- [Language-specific Neurons Do Not Facilitate Cross-Lingual Transfer](https://mlsanity.com/paper/2503.17456/) (2025)
- [From Neurons to Semantics: Evaluating Cross-Linguistic Alignment Capabilities of Large Language Models via Neurons Alignment](https://mlsanity.com/paper/2507.14900/) (2025)
- [ReCoVeR the Target Language: Language Steering without Sacrificing Task Performance](https://mlsanity.com/paper/2509.14814/) (2025)
- [How Programming Concepts and Neurons Are Shared in Code Language Models](https://mlsanity.com/paper/2506.01074/) (2025)
- [Are Knowledge and Reference in Multilingual Language Models Cross-Lingually Consistent?](https://mlsanity.com/paper/2507.12838/) (2025)
- [Understanding the role of FFNs in driving multilingual behaviour in LLMs](https://mlsanity.com/paper/2404.13855/) (2024)
- [Unraveling Babel: Exploring Multilingual Activation Patterns of LLMs and Their Applications](https://mlsanity.com/paper/2402.16367/) (2024)
- [Layer-Targeted Multilingual Knowledge Erasure in Large Language Models](https://mlsanity.com/paper/2602.22562/) (2026)
- [Semantic Pivots Enable Cross-Lingual Transfer in Large Language Models](https://mlsanity.com/paper/2505.16385/) (2025)
- [An Analysis of Residual-Stream Geometry Across Transformer Depth](https://mlsanity.com/paper/2607.18348/) (2026)

## Abstract (arXiv)

Recent research suggests that the feed-forward module within Transformers can be viewed as a collection of key-value memories, where the keys learn to capture specific patterns from the input based on the training examples. The values then combine the output from the 'memories' of the keys to generate predictions about the next token. This leads to an incremental process of prediction that gradually converges towards the final token choice near the output layers. This interesting perspective raises questions about how multilingual models might leverage this mechanism. Specifically, for autoregressive models trained on two or more languages, do all neurons (across layers) respond equally to all languages? No! Our hypothesis centers around the notion that during pretraining, certain model parameters learn strong language-specific features, while others learn more language-agnostic (shared across languages) features. To validate this, we conduct experiments utilizing parallel corpora of two languages that the model was initially pretrained on. Our findings reveal that the layers closest to the network's input or output tend to exhibit more language-specific behaviour compared to the layers in the middle.