Document detail
ID

oai:arXiv.org:2404.00397

Topic
Computer Science - Computation and...
Author
Cognetta, Marco Hiraoka, Tatsuya Okazaki, Naoaki Sennrich, Rico Pinter, Yuval
Category

Computer Science

Year

2024

listing date

4/3/2024

Keywords
vocabulary trimming subwords
Metrics

Abstract

We explore threshold vocabulary trimming in Byte-Pair Encoding subword tokenization, a postprocessing step that replaces rare subwords with their component subwords.

The technique is available in popular tokenization libraries but has not been subjected to rigorous scientific scrutiny.

While the removal of rare subwords is suggested as best practice in machine translation implementations, both as a means to reduce model size and for improving model performance through robustness, our experiments indicate that, across a large space of hyperparameter settings, vocabulary trimming fails to improve performance, and is even prone to incurring heavy degradation.

;Comment: 15 pages

Cognetta, Marco,Hiraoka, Tatsuya,Okazaki, Naoaki,Sennrich, Rico,Pinter, Yuval, 2024, An Analysis of BPE Vocabulary Trimming in Neural Machine Translation

Document

Open

Share

Source

Articles recommended by ES/IODE AI

Box-Free Model Watermarks Are Prone to Black-Box Removal Attacks
watermarks egg computer attacks remover extractor box-free model
A meta-analysis of simulator sickness as a function of simulator fidelity
driving simulator motion sickness fidelity kinetosis relation static simulator sickness fidelity ss
Integrated analysis of single-cell RNA-seq and bulk RNA-seq unveils heterogeneity and establishes a novel signature for prognosis and tumor immune microenvironment in ovarian cancer
ovarian cancer single cell rna-seq differentiation trajectory signature tumor immune microenvironment single-cell microenvironment heterogeneity analysis tumor immune genes model cancer ovarian