Work

How Do Language Models Acquire Character-Level Information?

Year: 2026

Type: conference-paper

Abstract: Language models (LMs) have been reported to implicitly encode character-level information, despite not being explicitly provided during training.However, the mechanisms underlying this phenomenon remain largely unexplored.To reveal the mechanisms, we analyze how models acquire character-level knowledge by comparing LMs trained under controlled settings, such as specifying the pre-training dataset or tokenizer, with those trained under standard settings.We categorize the contributing factors into those independent of tokenization.Our analysis reveals that merge rules and orthographic constraints constitute primary factors arising from tokenization, whereas semantic associations of substrings and syntactic information function as key factors independent of tokenization.

Source: Conference of the European Chapter of the Association for Computational Linguistics

DOI: 10.18653/v1/2026.eacl-long.282

Authors: Soma Sato, Ryohei Sasano

Institution: Nagoya University

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funder: Japan Society for the Promotion of Science

Award: Evolution of Large Language Models into Large Knowledge Models through Integration with Human Knowledge