Papers by Ryo Sato
Efficient Vocabulary Reduction for Small Language Models (2025.coling-industry)
Copied to clipboard
| Challenge: | Large language models (LLMs) have high computational costs and energy consumption, making their deployment in industrial settings difficult. |
| Approach: | They propose a small language model that compresses the embedding layer and reduces model size without significant loss of performance. |
| Outcome: | The proposed model reduces the embedding layer while maintaining performance while improving accuracy and performance. |
Multi-Perspective Document Revision (2022.coling-1)
Copied to clipboard
| Challenge: | a novel document revision task that revises multiple perspectives is proposed . grammatical error correction tasks have been studied in the natural language processing field . |
| Approach: | They propose a Japanese multi-perspective document revision task that revises multiple perspectives to improve the readability and clarity of a document. |
| Outcome: | The proposed model can be used to improve the readability and clarity of a document. |
Exploring the Influence of Spelling Errors on Lexical Variation Measures (C18-1)
Copied to clipboard
| Challenge: | Lexical richness measures such as Type-Token Ratio and Yule's K are often used for learner English analysis and assessment but are unstable because of spelling errors. |
| Approach: | They propose to use a dictionary to calculate the difference between TTR and Yule’s K caused by spelling errors and to deepen the understanding of the influence of spelling errors on them. |
| Outcome: | The proposed measures are based on English learner corpora of three groups and estimate their values before and after spelling errors are manually corrected. |
OptiPrune: Effective Pruning Approach for Every Target Sparsity (2025.coling-main)
Copied to clipboard
| Challenge: | Existing methods for model pruning only perform optimally within specific sparsity ranges. |
| Approach: | They propose a pruning method that reduces model size by eliminating redundant parameters . they compare it with OptiPrune, which adapts non-uniform sparsity with adaptive deviation . |
| Outcome: | The proposed method reduces model size and maintains performance despite large size and high computational demands. |