Leveraging LLMs for Synthesizing Training Data Across Many Languages in Multilingual Dense Retrieval

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Thakur, Nandan, Ni, Jianmo, Ábrego, Gustavo Hernández, Wieting, John, Lin, Jimmy, Cer, Daniel
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916207711485952
author Thakur, Nandan
Ni, Jianmo
Ábrego, Gustavo Hernández
Wieting, John
Lin, Jimmy
Cer, Daniel
author_facet Thakur, Nandan
Ni, Jianmo
Ábrego, Gustavo Hernández
Wieting, John
Lin, Jimmy
Cer, Daniel
contents There has been limited success for dense retrieval models in multilingual retrieval, due to uneven and scarce training data available across multiple languages. Synthetic training data generation is promising (e.g., InPars or Promptagator), but has been investigated only for English. Therefore, to study model capabilities across both cross-lingual and monolingual retrieval tasks, we develop SWIM-IR, a synthetic retrieval training dataset containing 33 (high to very-low resource) languages for fine-tuning multilingual dense retrievers without requiring any human supervision. To construct SWIM-IR, we propose SAP (summarize-then-ask prompting), where the large language model (LLM) generates a textual summary prior to the query generation step. SAP assists the LLM in generating informative queries in the target language. Using SWIM-IR, we explore synthetic fine-tuning of multilingual dense retrieval models and evaluate them robustly on three retrieval benchmarks: XOR-Retrieve (cross-lingual), MIRACL (monolingual) and XTREME-UP (cross-lingual). Our models, called SWIM-X, are competitive with human-supervised dense retrieval models, e.g., mContriever-X, finding that SWIM-IR can cheaply substitute for expensive human-labeled retrieval training data. SWIM-IR dataset and SWIM-X models are available at https://github.com/google-research-datasets/SWIM-IR.
format Preprint
id arxiv_https___arxiv_org_abs_2311_05800
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Leveraging LLMs for Synthesizing Training Data Across Many Languages in Multilingual Dense Retrieval
Thakur, Nandan
Ni, Jianmo
Ábrego, Gustavo Hernández
Wieting, John
Lin, Jimmy
Cer, Daniel
Information Retrieval
Artificial Intelligence
Computation and Language
There has been limited success for dense retrieval models in multilingual retrieval, due to uneven and scarce training data available across multiple languages. Synthetic training data generation is promising (e.g., InPars or Promptagator), but has been investigated only for English. Therefore, to study model capabilities across both cross-lingual and monolingual retrieval tasks, we develop SWIM-IR, a synthetic retrieval training dataset containing 33 (high to very-low resource) languages for fine-tuning multilingual dense retrievers without requiring any human supervision. To construct SWIM-IR, we propose SAP (summarize-then-ask prompting), where the large language model (LLM) generates a textual summary prior to the query generation step. SAP assists the LLM in generating informative queries in the target language. Using SWIM-IR, we explore synthetic fine-tuning of multilingual dense retrieval models and evaluate them robustly on three retrieval benchmarks: XOR-Retrieve (cross-lingual), MIRACL (monolingual) and XTREME-UP (cross-lingual). Our models, called SWIM-X, are competitive with human-supervised dense retrieval models, e.g., mContriever-X, finding that SWIM-IR can cheaply substitute for expensive human-labeled retrieval training data. SWIM-IR dataset and SWIM-X models are available at https://github.com/google-research-datasets/SWIM-IR.
title Leveraging LLMs for Synthesizing Training Data Across Many Languages in Multilingual Dense Retrieval
topic Information Retrieval
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2311.05800