Challenge: Using pretraining data, we find that a designated monolingual tokenizer plays an equally important role in the downstream performance of the model.
Approach: They propose to compare pretrained multilingual models with their monolingual counterparts on a set of five diverse monolingual downstream tasks.
Outcome: The proposed models offer previously unmatched performance in all NLP tasks.

Similar Papers

Tokenizer Choice For LLM Training: Negligible or Crucial? (2024.findings-naacl)

Copied to clipboard

Challenge: Recent success of large language models has been driven by curating the training dataset composition, scaling of model architectures and advancements in pretraining objectives, leaving tokenizer influence as a blind spot.
Approach: They conduct a comprehensive study on the influence of tokenizer choice on LLM downstream performance by training 24 mono- and multilingual LLMs at a 2.6B parameter scale.
Outcome: The proposed model can significantly impact the model's downstream performance and training costs.
A Multi-dimensional Evaluation of Tokenizer-free Multilingual Pretrained Models (2023.findings-eacl)

Copied to clipboard

Challenge: Recent work on tokenizer-free models shows promising results in cross-lingual transfer . previous work focused on reporting accuracy on a limited set of tasks and data settings .
Approach: They compare tokenizer-free and subword-based models using various dimensions . they find subword models are still the most practical choice in many settings .
Outcome: The proposed model improves cross-lingual transfer and reduces engineering overhead.
How do different tokenizers perform on downstream tasks in scriptio continua languages?: A case study in Japanese (2023.acl-srw)

Copied to clipboard

Challenge: Existing studies on scriptio continua languages lack comprehensiveness of tokenizers . authors use Byte-Pair-Encoding or Unigram instead of WordPiece for subword tokenizer .
Approach: They investigate the effect of tokenizers on the downstream performance of pretrained language models in scriptio continua languages where no explicit spaces exist between words.
Outcome: The proposed tokenizers perform better on a wide range of tasks compared with other tokenizer methods . the results show that each task has an optimal morphological analyzer .
Tokenization Impacts Multilingual Language Modeling: Assessing Vocabulary Allocation and Overlap Across Languages (2023.findings-acl)

Copied to clipboard

Challenge: Multilingual language models perform surprisingly well in a variety of NLP tasks for diverse languages.
Approach: They propose to evaluate the quality of lexical representation and vocabulary overlap observed in sub-word tokenizers.
Outcome: The proposed criteria show that the overlap of vocabulary across languages can be detrimental to certain downstream tasks.
A Comparison of Language Modeling and Translation as Multilingual Pretraining Objectives (2024.emnlp-main)

Copied to clipboard

Challenge: Pretrained language models (PLMs) display impressive performances and have captured the attention of the NLP community.
Approach: They propose to compare multilingual pretraining objectives in a controlled methodological environment with multilingual models.
Outcome: The proposed model outperforms existing models in 6 languages and demonstrates that multilingual translation is an effective pretraining objective under the right conditions.
Beyond Text Compression: Evaluating Tokenizers Across Scales (2025.acl-long)

Copied to clipboard

Challenge: Language models rely on tokenizers to convert text into machine-interpretable tokens, which shape the statistical patterns that language models learn to estimate.
Approach: They propose to use Zipf's law to measure tokenizer performance by combining several metrics to capture multiple aspects of tokenizer behavior.
Outcome: The proposed metrics correlate more strongly with downstream performance than text compression when modeling unseen languages.
One Tokenizer To Rule Them All: Emergent Language Plasticity via Multilingual Tokenizers (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to train multilingual large language models for many languages at once are limited due to limited model capacity, scarce high-quality data, and compute constraints.
Approach: They propose to use a universal tokenizer to improve language plasticity and adaptability to new languages by up to 20%.
Outcome: The proposed tokenizer improves language plasticity and improves plasticity towards languages that are completely unseen in the tokenizer and pretraining, by up to 5% win rate gain.
MonoByte: A Pool of Monolingual Byte-level Language Models (2022.coling-1)

Copied to clipboard

Challenge: Existing studies have shown that multilingual models can achieve zero-shot cross-lingual performance on various NLP tasks, but due to the cost of pretraining, they often use public models with limited budgets.
Approach: They propose to use tokenized models to test cross-lingual ability in multilingual and monolingual corpora.
Outcome: The results show that models pretrained on multilingual and even monolingual corpora perform better than models pre-trained on SOTA models.
Multilingual Word Segmentation: Training Many Language-Specific Tokenizers Smoothly Thanks to the Universal Dependencies Corpus (L18-1)

Copied to clipboard

Challenge: Towards language scalability, major progress has been achieved in multilingual language technology in recent years.
Approach: They propose a tokenizer that can be trained from any Universal Dependencies corpus dataset . they argue that tokenization should be seen as a supervised task and scalability requires a software engineering process across languages.
Outcome: The proposed tokenizer can be trained from any dataset in the corpus UD2 . the proposed software tool relies on elephant to perform the training .
What is the best recipe for character-level encoder-only modelling? (2023.acl-long)

Copied to clipboard

Challenge: aims to benchmark recent progress in language understanding models that output contextualised representations at the character level.
Approach: They aim to find the best way to build and train character-level BERT-like models by comparing architectural innovations with pretraining objectives.
Outcome: The proposed model outperforms a token-based model on a set of evaluation tasks with a fixed training procedure.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations