Exploring the Role of Transliteration in In-Context Learning for Low-resource Languages Written in Non-Latin Scripts

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Ma, Chunlan, Liu, Yihong, Ye, Haotian, Schütze, Hinrich
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911940851269632
author Ma, Chunlan
Liu, Yihong
Ye, Haotian
Schütze, Hinrich
author_facet Ma, Chunlan
Liu, Yihong
Ye, Haotian
Schütze, Hinrich
contents Decoder-only large language models (LLMs) excel in high-resource languages across various tasks through few-shot or even zero-shot in-context learning (ICL). However, their performance often does not transfer well to low-resource languages, especially those written in non-Latin scripts. Inspired by recent work that leverages transliteration in encoder-only models, we investigate whether transliteration is also effective in improving LLMs' performance for low-resource languages written in non-Latin scripts. To this end, we propose three prompt templates, where the target-language text is represented in (1) its original script, (2) Latin script, or (3) both. We apply these methods to several representative LLMs of different sizes on various tasks including text classification and sequential labeling. Our findings show that the effectiveness of transliteration varies by task type and model size. For instance, all models benefit from transliterations for sequential labeling (with increases of up to 25%).
format Preprint
id arxiv_https___arxiv_org_abs_2407_02320
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Exploring the Role of Transliteration in In-Context Learning for Low-resource Languages Written in Non-Latin Scripts
Ma, Chunlan
Liu, Yihong
Ye, Haotian
Schütze, Hinrich
Computation and Language
Artificial Intelligence
Decoder-only large language models (LLMs) excel in high-resource languages across various tasks through few-shot or even zero-shot in-context learning (ICL). However, their performance often does not transfer well to low-resource languages, especially those written in non-Latin scripts. Inspired by recent work that leverages transliteration in encoder-only models, we investigate whether transliteration is also effective in improving LLMs' performance for low-resource languages written in non-Latin scripts. To this end, we propose three prompt templates, where the target-language text is represented in (1) its original script, (2) Latin script, or (3) both. We apply these methods to several representative LLMs of different sizes on various tasks including text classification and sequential labeling. Our findings show that the effectiveness of transliteration varies by task type and model size. For instance, all models benefit from transliterations for sequential labeling (with increases of up to 25%).
title Exploring the Role of Transliteration in In-Context Learning for Low-resource Languages Written in Non-Latin Scripts
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2407.02320