Enhancing Character-Level Understanding in LLMs through Token Internal Structure Learning (2025.acl-long)
Copied to clipboard
Zhu Xu, Zhiqiang Zhao, Zihan Zhang, Yuchi Liu, Quanwei Shen, Fei Liu, Yu Kuang, Jian He, Conglin Liu
| Challenge: | Large language models (LLMs) use tokenization methods but often obscure internal character structures within tokens. |
| Approach: | They propose a method that improves models’ ability to capture character positions within tokens by training them on reverse character prediction tasks using the tokenizer’s vocabulary. |
| Outcome: | Experiments show that the proposed method improves position prediction accuracy in large language models, enabling more precise identification of target characters in original text. |
Similar Papers
Spelling-out is not Straightforward: LLMs’ Capability of Tokenization from Token to Characters (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) can spell out tokens character by character with high accuracy, yet struggle with more complex character-level tasks. |
| Approach: | They examine how large language models internally represent character-level information during the spelling-out process. |
| Outcome: | The embedding layer does not fully encode character-level information, especially beyond the first character. |
Token-Aware Editing of Internal Activations for Large Language Model Alignment (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods to optimize the behavior of large language models neglect misalignment discrepancies among tokens, resulting in deviant alignment direction and inflexible editing strength. |
| Approach: | They propose a token-aware editing approach to exploit the misalignment discrepancy among tokens to enhance activation probing and facilitate intervention. |
| Outcome: | Extensive experiments on three alignment capabilities demonstrate the efficacy of the proposed approach surpassing baseline by 25.8% on the primary metric of truthfulness with minimal cost. |
What do tokens know about their characters and how do they know it? (2022.naacl-main)
Copied to clipboard
| Challenge: | Pre-trained language models that use subword tokenization schemes can succeed at a variety of language tasks that require character-level information. |
| Approach: | They propose to use word tokenization schemes to probe what word pieces encode . they show that larger models can encode character-level information . |
| Outcome: | The proposed models can encode character-level information and perform better on non-Latin alphabets. |
Position-Aware Depth Decay Decoding (D3): Boosting Large Language Model Inference Efficiency (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent dynamic computation methods show that not all components are required for inference, enabling a training-free pipeline. |
| Approach: | They propose a token-position aware layer skipping framework to save 1.5x times operations efficiently while maintaining performance. |
| Outcome: | The proposed algorithm achieves 1.5x speedup on large language models with no retraining and with comparable performance on the GSM8K and BBH benchmarks. |
Empowering Character-level Text Infilling by Eliminating Sub-Tokens (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for character-level infilling relied on predicting sub-tokens, but this strategy was ineffective. |
| Approach: | They propose a method to fill-in-the-mid with Starting and Ending character constraints that avoids predicting sub-tokens in inference. |
| Outcome: | The proposed method surpasses existing methods and offers significant performance gains. |
Improve Decoding Factuality by Token-wise Cross Layer Entropy of Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) often struggle with the issue of generating inaccurate or fabricated content even when they possess correct knowledge. |
| Approach: | They propose a decoding method that mitigates hallucinations without extra training . they propose entropy eNhanced decoding that leverages inner probability changes . |
| Outcome: | The proposed method improves the truthfulness and informativeness of generation while maintaining robust QA accuracy. |
How Do Language Models Acquire Character-Level Information? (2026.eacl-long)
Copied to clipboard
| Challenge: | Language models (LMs) implicitly encode character-level information, despite not being explicitly provided during training. |
| Approach: | They analyze how language models acquire character-level knowledge by comparing them to standard settings. |
| Outcome: | The results show that LMs do not treat words as opaque tokens, but instead treat them as tokens. |
Language-Aware Token Boosting: LLM Language Confusion Reduction Without Tuning (2026.acl-short)
Copied to clipboard
| Challenge: | Existing approaches to reduce language confusion rely on fine-tuning . Existing methods rely only on fine tuning to mitigate this issue . |
| Approach: | They propose a tuning-free paradigm for reducing language confusion by applying targeted perturbations to tokens associated with the desired language and Adaptive Language-Aware Token Boosting. |
| Outcome: | The proposed methods improve multilingual alignment while maintaining the summarization quality without additional fine-tuning. |
Mitigating Tokenization-Induced Distance Distortion in Long-Context Multilingual Machine Translation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing positional encodings rely on fixed token indices and implicitly assume uniform semantic density, which breaks down for long-context inputs. |
| Approach: | They propose a tokenization-aware adaptive positional encoding that conditions relative positional bias on input-level sequence length and fragmentation statistics. |
| Outcome: | The proposed model improves long-context robustness and accuracy over baselines. |
Token Weighting for Long-Range Language Modeling (2025.findings-naacl)
Copied to clipboard
| Challenge: | Many applications of large language models (LLMs) require long-context understanding, but models still struggle with such tasks. |
| Approach: | They propose token-weighting schemes that assign different weights to each training token in the loss, generalizing existing works. |
| Outcome: | The proposed methods compare confidences of a long-context and short-concept model and show that non-uniform loss weights improve the long-constability of LLMs. |