Instability of LLM Pre-Pretraining: It Doesn't Always Help. An Investigation on Multiple Languages

• URL: http://arxiv.org/abs/2608.08800v1
• Date: Sun, 09 Aug 2026 16:26:29 GMT
• Title: Instability of LLM Pre-Pretraining: It Doesn't Always Help. An Investigation on Multiple Languages
• Abstract summary: Pretraining LLMs on artificial languages ("pre-pretraining") is a technique that could reportedly increase token efficiency by 33%.<n>We validate this prior result for English on a larger set of natural languages across four language families.<n>We also relate the observed gains (or losses) in token efficiency to quantified linguistic properties of the languages.
• Score: 0.3441021278275805
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Pretraining LLMs on artificial languages ("pre-pretraining") is a technique that could reportedly increase token efficiency by 33%, i.e., save up to 33% of training tokens needed to reach a certain performance. We validate this prior result for English on a larger set of natural languages across four language families, using two different tokenizers and varying model sizes. We also relate the observed gains (or losses) in token efficiency to quantified linguistic properties of the languages, such as sentence length, morphological richness, and features of dependency syntactic trees (tree depth, number of children, number of crossing dependencies). Our empirical results indicate that the reported gains depend heavily on the experiment setup and the choice of random seed, although we can confirm the trend of stable gains with 128-Dyck pretraining of small models with the Llama tokenizer for most of the examined languages. On a general note, we argue that multiple training runs should be carried out at least for a subset of experiments to avoid the community adopting unstable approaches.

Related papers

• Language Models are not Equally Robust to Non-Canonical Tokenization across Languages [29.183380954421242]
  We study the behavior of language models under non-canonical tokenizations across diverse languages.<n>We show that tokenization robustness is not a universal property of language models, but depends strongly on the language and its interaction with the tokenizer.
  arXiv  Detail & Related papers  (2026-07-29T12:23:15Z)
• The Token Tax: Systematic Bias in Multilingual Tokenization [0.8820808252713569]
  Tokenization inefficiency imposes structural disadvantages on morphologically complex, low-resource languages.<n>We show that fertility (tokens/word) reliably predicts accuracy.<n>We also find that reasoning models (DeepSeek, o1) consistently outperform non-reasoning peers.
  arXiv  Detail & Related papers  (2025-09-05T20:20:51Z)
• Cross-Lingual Pitfalls: Automatic Probing Cross-Lingual Weakness of Multilingual Large Language Models [55.14276067678253]
  This paper introduces a novel methodology for efficiently identifying inherent cross-lingual weaknesses in Large Language Models (LLMs)<n>We construct a new dataset of over 6,000 bilingual pairs across 16 languages using this methodology, demonstrating its effectiveness in revealing weaknesses even in state-of-the-art models.<n>Further experiments investigate the relationship between linguistic similarity and cross-lingual weaknesses, revealing that linguistically related languages share similar performance patterns.
  arXiv  Detail & Related papers  (2025-05-24T12:31:27Z)
• Investigating Language-Specific Calibration For Pruning Multilingual Large Language Models [11.421452042888523]
  We compare different calibration languages for pruning multilingual models across diverse languages, tasks, models, and SotA pruning techniques. Our results offer practical suggestions, for example, calibrating in the target language can efficiently retain the language modeling capability but does not necessarily benefit downstream tasks.
  arXiv  Detail & Related papers  (2024-08-26T16:29:13Z)
• The Role of Language Imbalance in Cross-lingual Generalisation: Insights from Cloned Language Experiments [57.273662221547056]
  In this study, we investigate an unintuitive novel driver of cross-lingual generalisation: language imbalance. We observe that the existence of a predominant language during training boosts the performance of less frequent languages. As we extend our analysis to real languages, we find that infrequent languages still benefit from frequent ones, yet whether language imbalance causes cross-lingual generalisation there is not conclusive.
  arXiv  Detail & Related papers  (2024-04-11T17:58:05Z)
• Tokenizer Choice For LLM Training: Negligible or Crucial? [30.33170936148845]
  We study the influence of tokenizer choice on Large Language Models (LLMs) downstream performance by training 24 mono- and multilingual LLMs. We find that the tokenizer choice can significantly impact the model's downstream performance and training costs. We show that multilingual tokenizers trained on the five most frequent European languages require vocabulary size increases of factor three in comparison to English.
  arXiv  Detail & Related papers  (2023-10-12T22:44:19Z)
• How Robust is Neural Machine Translation to Language Imbalance in Multilingual Tokenizer Training? [86.48323488619629]
  We analyze how translation performance changes as the data ratios among languages vary in the tokenizer training corpus. We find that while relatively better performance is often observed when languages are more equally sampled, the downstream performance is more robust to language imbalance than we usually expected.
  arXiv  Detail & Related papers  (2022-04-29T17:50:36Z)
• Probing Structured Pruning on Multilingual Pre-trained Models: Settings, Algorithms, and Efficiency [62.0887259003594]
  This work investigates three aspects of structured pruning on multilingual pre-trained language models: settings, algorithms, and efficiency. Experiments on nine downstream tasks show several counter-intuitive phenomena. We present Dynamic Sparsification, a simple approach that allows training the model once and adapting to different model sizes at inference.
  arXiv  Detail & Related papers  (2022-04-06T06:29:52Z)
• From Good to Best: Two-Stage Training for Cross-lingual Machine Reading Comprehension [51.953428342923885]
  We develop a two-stage approach to enhance the model performance. The first stage targets at recall: we design a hard-learning (HL) algorithm to maximize the likelihood that the top-k predictions contain the accurate answer. The second stage focuses on precision: an answer-aware contrastive learning mechanism is developed to learn the fine difference between the accurate answer and other candidates.
  arXiv  Detail & Related papers  (2021-12-09T07:31:15Z)
• Analysing The Impact Of Linguistic Features On Cross-Lingual Transfer [3.299672391663527]
  We analyze a state-of-the-art multilingual model and try to determine what impacts good transfer between languages. We show that looking at particular syntactic features is 2-4 times more helpful in predicting the performance than an aggregated syntactic similarity.
  arXiv  Detail & Related papers  (2021-05-12T21:22:58Z)
• Emergent Communication Pretraining for Few-Shot Machine Translation [66.48990742411033]
  We pretrain neural networks via emergent communication from referential games. Our key assumption is that grounding communication on images---as a crude approximation of real-world environments---inductively biases the model towards learning natural languages.
  arXiv  Detail & Related papers  (2020-11-02T10:57:53Z)
• Knowledge Distillation for Multilingual Unsupervised Neural Machine Translation [61.88012735215636]
  Unsupervised neural machine translation (UNMT) has recently achieved remarkable results for several language pairs. UNMT can only translate between a single language pair and cannot produce translation results for multiple language pairs at the same time. In this paper, we empirically introduce a simple method to translate between thirteen languages using a single encoder and a single decoder.
  arXiv  Detail & Related papers  (2020-04-21T17:26:16Z)

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.