Saved in:
Bibliographic Details
Main Authors: Arnett, Catherine, Rivière, Pamela D., Chang, Tyler A., Trott, Sean
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2403.13754
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911805761126400
author Arnett, Catherine
Rivière, Pamela D.
Chang, Tyler A.
Trott, Sean
author_facet Arnett, Catherine
Rivière, Pamela D.
Chang, Tyler A.
Trott, Sean
contents The relationship between language model tokenization and performance is an open area of research. Here, we investigate how different tokenization schemes impact number agreement in Spanish plurals. We find that morphologically-aligned tokenization performs similarly to other tokenization schemes, even when induced artificially for words that would not be tokenized that way during training. We then present exploratory analyses demonstrating that language model embeddings for different plural tokenizations have similar distributions along the embedding space axis that maximally distinguishes singular and plural nouns. Our results suggest that morphologically-aligned tokenization is a viable tokenization approach, and existing models already generalize some morphological patterns to new items. However, our results indicate that morphological tokenization is not strictly required for performance.
format Preprint
id arxiv_https___arxiv_org_abs_2403_13754
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Different Tokenization Schemes Lead to Comparable Performance in Spanish Number Agreement
Arnett, Catherine
Rivière, Pamela D.
Chang, Tyler A.
Trott, Sean
Computation and Language
The relationship between language model tokenization and performance is an open area of research. Here, we investigate how different tokenization schemes impact number agreement in Spanish plurals. We find that morphologically-aligned tokenization performs similarly to other tokenization schemes, even when induced artificially for words that would not be tokenized that way during training. We then present exploratory analyses demonstrating that language model embeddings for different plural tokenizations have similar distributions along the embedding space axis that maximally distinguishes singular and plural nouns. Our results suggest that morphologically-aligned tokenization is a viable tokenization approach, and existing models already generalize some morphological patterns to new items. However, our results indicate that morphological tokenization is not strictly required for performance.
title Different Tokenization Schemes Lead to Comparable Performance in Spanish Number Agreement
topic Computation and Language
url https://arxiv.org/abs/2403.13754