How Language Directions Align with Token Geometry in Multilingual LLMs

• URL: http://arxiv.org/abs/2511.16693v1
• Date: Sun, 16 Nov 2025 16:36:56 GMT
• Title: How Language Directions Align with Token Geometry in Multilingual LLMs
• Authors: JaeSeong Kim, Suan Lee,
• Abstract summary: We conduct a comprehensive probing study on six multilingual LLMs, covering all 268 transformer layers.<n>Our results show that language information becomes sharply separated in the first transformer block.<n>Chinese-inclusive models achieve a ZH Match@Peak of 16.43%, whereas English-centric models achieve only 3.90%.
• Score: 1.0312968200748118
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multilingual LLMs demonstrate strong performance across diverse languages, yet there has been limited systematic analysis of how language information is structured within their internal representation space and how it emerges across layers. We conduct a comprehensive probing study on six multilingual LLMs, covering all 268 transformer layers, using linear and nonlinear probes together with a new Token--Language Alignment analysis to quantify the layer-wise dynamics and geometric structure of language encoding. Our results show that language information becomes sharply separated in the first transformer block (+76.4$\pm$8.2 percentage points from Layer 0 to 1) and remains almost fully linearly separable throughout model depth. We further find that the alignment between language directions and vocabulary embeddings is strongly tied to the language composition of the training data. Notably, Chinese-inclusive models achieve a ZH Match@Peak of 16.43\%, whereas English-centric models achieve only 3.90\%, revealing a 4.21$\times$ structural imprinting effect. These findings indicate that multilingual LLMs distinguish languages not by surface script features but by latent representational structures shaped by the training corpus. Our analysis provides practical insights for data composition strategies and fairness in multilingual representation learning. All code and analysis scripts are publicly available at: https://github.com/thisiskorea/How-Language-Directions-Align-with-Token-Geometry-in-Multilingual-LLM s.

Related papers

• Tracing Multilingual Representations in LLMs with Cross-Layer Transcoders [51.380449540006985]
  Large Language Models (LLMs) can process many languages, yet how they internally represent this diversity remains unclear.<n>Do they form shared multilingual representations with language-specific decoding, and if so, why does performance still favor the dominant training language?<n>We analyze their internal mechanisms using cross-layer transcoders (CLT) and attribution graphs.
  arXiv  Detail & Related papers  (2025-11-13T22:51:06Z)
• Crosslingual Capabilities and Knowledge Barriers in Multilingual Large Language Models [62.91524967852552]
  Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora.<n>But can these models relate corresponding concepts across languages, i.e., be crosslingual?<n>This study evaluates state-of-the-art LLMs on inherently crosslingual tasks.
  arXiv  Detail & Related papers  (2024-06-23T15:15:17Z)
• Hyperpolyglot LLMs: Cross-Lingual Interpretability in Token Embeddings [4.2243058640527575]
  Cross-lingual transfer learning is an important property of multilingual large language models (LLMs) Our research opens the door for investigations in 1) The effect of pre-training and model architectures on representations of languages and 2) The applications of cross-lingual representations embedded in language models.
  arXiv  Detail & Related papers  (2023-11-29T19:20:14Z)
• The Belebele Benchmark: a Parallel Reading Comprehension Dataset in 122 Language Variants [80.4837840962273]
  We present Belebele, a dataset spanning 122 language variants. This dataset enables the evaluation of text models in high-, medium-, and low-resource languages.
  arXiv  Detail & Related papers  (2023-08-31T17:43:08Z)
• Efficiently Aligned Cross-Lingual Transfer Learning for Conversational Tasks using Prompt-Tuning [98.60739735409243]
  Cross-lingual transfer of language models trained on high-resource languages like English has been widely studied for many NLP tasks. We introduce XSGD for cross-lingual alignment pretraining, a parallel and large-scale multilingual conversation dataset. To facilitate aligned cross-lingual representations, we develop an efficient prompt-tuning-based method for learning alignment prompts.
  arXiv  Detail & Related papers  (2023-04-03T18:46:01Z)
• Meta-X$_{NLG}$: A Meta-Learning Approach Based on Language Clustering for Zero-Shot Cross-Lingual Transfer and Generation [11.155430893354769]
  This paper proposes a novel meta-learning framework to learn shareable structures from typologically diverse languages. We first cluster the languages based on language representations and identify the centroid language of each cluster. A meta-learning algorithm is trained with all centroid languages and evaluated on the other languages in the zero-shot setting.
  arXiv  Detail & Related papers  (2022-03-19T05:22:07Z)
• A Massively Multilingual Analysis of Cross-linguality in Shared Embedding Space [61.18554842370824]
  In cross-lingual language models, representations for many different languages live in the same space. We compute a task-based measure of cross-lingual alignment in the form of bitext retrieval performance. We examine a range of linguistic, quasi-linguistic, and training-related features as potential predictors of these alignment metrics.
  arXiv  Detail & Related papers  (2021-09-13T21:05:37Z)
• Examining Cross-lingual Contextual Embeddings with Orthogonal Structural Probes [0.2538209532048867]
  A novel Orthogonal Structural Probe (Limisiewicz and Marevcek, 2021) allows us to answer this question for specific linguistic features. We evaluate syntactic (UD) and lexical (WordNet) structural information encoded inmBERT's contextual representations for nine diverse languages. We successfully apply our findings to zero-shot and few-shot cross-lingual parsing.
  arXiv  Detail & Related papers  (2021-09-10T15:03:11Z)
• FILTER: An Enhanced Fusion Method for Cross-lingual Language Understanding [85.29270319872597]
  We propose an enhanced fusion method that takes cross-lingual data as input for XLM finetuning. During inference, the model makes predictions based on the text input in the target language and its translation in the source language. To tackle this issue, we propose an additional KL-divergence self-teaching loss for model training, based on auto-generated soft pseudo-labels for translated text in the target language.
  arXiv  Detail & Related papers  (2020-09-10T22:42:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.