# Speculative Decoding and the Curse of Multilinguality

arXiv: https://arxiv.org/abs/2605.30580 · HTML: https://mlsanity.com/paper/2605.30580/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Speculative decoding, one of 12 directions in Efficient LLM inference and serving.
- Builds on 22 of its 38 references that are in the map.
- First posted on arXiv 28 May 2026.

## Placed in

- Efficient LLM inference and serving / [Speculative decoding](https://mlsanity.com/inference/speculative-decoding/)

## Builds on (most cited first, 22 of 22)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Unsupervised Cross-lingual Representation Learning at Scale](https://mlsanity.com/paper/1911.02116/) (2019)
- [Fast Inference from Transformers via Speculative Decoding](https://mlsanity.com/paper/2211.17192/) (2022)
- [Accelerating Large Language Model Decoding with Speculative Sampling](https://mlsanity.com/paper/2302.01318/) (2023)
- [Beto, Bentz, Becas: The Surprising Cross-Lingual Effectiveness of BERT](https://mlsanity.com/paper/1904.09077/) (2019)
- [Improving Massively Multilingual Neural Machine Translation and Zero-Shot Translation](https://mlsanity.com/paper/2004.11867/) (2020)
- [Language Model Tokenizers Introduce Unfairness Between Languages](https://mlsanity.com/paper/2305.15425/) (2023)
- [Are All Languages Created Equal in Multilingual BERT?](https://mlsanity.com/paper/2005.09093/) (2020)
- [Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning](https://mlsanity.com/paper/2402.06619/) (2024)
- [Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation](https://mlsanity.com/paper/2203.16487/) (2022)
- [The Tatoeba Translation Challenge -- Realistic Data Sets for Low Resource and Multilingual MT](https://mlsanity.com/paper/2010.06354/) (2020)
- [The Effect of Domain and Diacritics in Yor`ub'a-English Neural Machine Translation](https://mlsanity.com/paper/2103.08647/) (2021)
- [Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs](https://mlsanity.com/paper/2403.00858/) (2024)
- [OpenCSG Chinese Corpus: A Series of High-quality Chinese Datasets for LLM Training](https://mlsanity.com/paper/2501.08197/) (2025)
- [Language-Specific Latent Process Hinders Cross-Lingual Performance](https://mlsanity.com/paper/2505.13141/) (2025)
- [ChrEn: Cherokee-English Machine Translation for Endangered Language Revitalization](https://mlsanity.com/paper/2010.04791/) (2020)
- [Towards Fast Multilingual LLM Inference: Speculative Decoding and Specialized Drafters](https://mlsanity.com/paper/2406.16758/) (2024)
- [Revisiting Multilingual Data Mixtures in Language Model Pretraining](https://mlsanity.com/paper/2510.25947/) (2025)
- [Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation](https://mlsanity.com/paper/2602.20816/) (2026)
- [Development of Pre-Trained Transformer-based Models for the Nepali Language](https://mlsanity.com/paper/2411.15734/) (2024)
- [The Disparate Impacts of Speculative Decoding](https://mlsanity.com/paper/2510.02128/) (2025)

## Abstract (arXiv)

Speculative decoding is a popular technique for large language model (LLM) inference, enabling faster generation by drafting multiple tokens with a smaller draft model. However, the effectiveness of speculative decoding has mainly been studied for English. Motivated by the curse of multilinguality, we hypothesize that speculative decoding is far less effective for low-resource languages due to the limited multilingual capacities of smaller models. We test eleven languages under a standard speculative decoding setup and find strong evidence for our hypothesis. Next, we try to improve the multilingual capabilities of the smaller draft model via distillation from the larger model. We find, though, that distillation generalizes poorly across tasks in the same language, and we argue that assembling a task-agnostic, fully representative dataset is infeasible for low-resource languages. Finally, we propose weaker n-gram models as draft models; these provide moderate speed-ups due to their minuscule inference cost.