Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MetaXCR: Reinforcement-Based Meta-Transfer Learning for Cross-Lingual Commonsense Reasoning

Jie He, Yu Fu

TL;DR

MetaXCR tackles cross-lingual low-resource commonsense reasoning by integrating multi-source adapters with cross-task and cross-lingual meta-learning, governed by a reinforcement-based source-task sampling strategy. The framework trains adapters across English source datasets while adapting to target languages, using CTML and CLML to generalize across tasks and languages. Empirical results on XCOPA show state-of-the-art performance in zero- and few-shot settings while using only 1.6% of the training parameters, and analyses highlight the benefits of multi-task adapters and RL-based sampling. This work enables more effective cross-language CR with limited labeled data, reducing annotation costs and expanding practical deployment potential.

Abstract

Commonsense reasoning (CR) has been studied in many pieces of domain and has achieved great progress with the aid of large datasets. Unfortunately, most existing CR datasets are built in English, so most previous work focus on English. Furthermore, as the annotation of commonsense reasoning is costly, it is impossible to build a large dataset for every novel task. Therefore, there are growing appeals for Cross-lingual Low-Resource Commonsense Reasoning, which aims to leverage diverse existed English datasets to help the model adapt to new cross-lingual target datasets with limited labeled data. In this paper, we propose a multi-source adapter for cross-lingual low-resource Commonsense Reasoning (MetaXCR). In this framework, we first extend meta learning by incorporating multiple training datasets to learn a generalized task adapters across different tasks. Then, we further introduce a reinforcement-based sampling strategy to help the model sample the source task that is the most helpful to the target task. Finally, we introduce two types of cross-lingual meta-adaption methods to enhance the performance of models on target languages. Extensive experiments demonstrate MetaXCR is superior over state-of-the-arts, while being trained with fewer parameters than other work.

MetaXCR: Reinforcement-Based Meta-Transfer Learning for Cross-Lingual Commonsense Reasoning

TL;DR

MetaXCR tackles cross-lingual low-resource commonsense reasoning by integrating multi-source adapters with cross-task and cross-lingual meta-learning, governed by a reinforcement-based source-task sampling strategy. The framework trains adapters across English source datasets while adapting to target languages, using CTML and CLML to generalize across tasks and languages. Empirical results on XCOPA show state-of-the-art performance in zero- and few-shot settings while using only 1.6% of the training parameters, and analyses highlight the benefits of multi-task adapters and RL-based sampling. This work enables more effective cross-language CR with limited labeled data, reducing annotation costs and expanding practical deployment potential.

Abstract

Commonsense reasoning (CR) has been studied in many pieces of domain and has achieved great progress with the aid of large datasets. Unfortunately, most existing CR datasets are built in English, so most previous work focus on English. Furthermore, as the annotation of commonsense reasoning is costly, it is impossible to build a large dataset for every novel task. Therefore, there are growing appeals for Cross-lingual Low-Resource Commonsense Reasoning, which aims to leverage diverse existed English datasets to help the model adapt to new cross-lingual target datasets with limited labeled data. In this paper, we propose a multi-source adapter for cross-lingual low-resource Commonsense Reasoning (MetaXCR). In this framework, we first extend meta learning by incorporating multiple training datasets to learn a generalized task adapters across different tasks. Then, we further introduce a reinforcement-based sampling strategy to help the model sample the source task that is the most helpful to the target task. Finally, we introduce two types of cross-lingual meta-adaption methods to enhance the performance of models on target languages. Extensive experiments demonstrate MetaXCR is superior over state-of-the-arts, while being trained with fewer parameters than other work.

Paper Structure

This paper contains 24 sections, 8 equations, 4 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. MetaXCR
  4. Commonsense Reasoning Framework
  5. Cross-Lingual and Cross-Task Meta-Transfer Learning
  6. Cross-Task Meta-Learning
  7. Cross-Lingual Meta-Learning
  8. Target-adapted meta sampling
  9. Heuristic Sampling and Motivation for Reinforced Sampling
  10. RL algorithm
  11. Our Sampling Approach
  12. Experiments
  13. Datasets
  14. Source Datasets
  15. Target Dataset
  16. ...and 9 more sections

Figures (4)

  • Figure 1: Our MetaXCR framework.
  • Figure 2: The test accuracy curve of the COPA for three sampling strategries
  • Figure 3: The accuracy of different Sampling strategies
  • Figure 4: The accuracy results for Target-only, mono-lingual, cross-lingual training method.