Evaluating Cross-Lingual Transfer Learning Approaches in Multilingual Conversational Agent Models

TL;DR

This paper proposes a general multilingual NLU framework for conversational agents that achieves comparable or better performance than monolingual models, reducing development effort for new languages.

Contribution

It introduces a versatile multilingual model framework and evaluates how different architectures impact performance, facilitating faster deployment of voice assistants in new languages.

Findings

Multilingual models match or outperform monolingual models in accuracy.

The proposed framework reduces feature creation and maintenance effort.

Different deep learning architectures influence model performance.

Abstract

With the recent explosion in popularity of voice assistant devices, there is a growing interest in making them available to user populations in additional countries and languages. However, to provide the highest accuracy and best performance for specific user populations, most existing voice assistant models are developed individually for each region or language, which requires linear investment of effort. In this paper, we propose a general multilingual model framework for Natural Language Understanding (NLU) models, which can help bootstrap new language models faster and reduce the amount of effort required to develop each language separately. We explore how different deep learning architectures affect multilingual NLU model performance. Our experimental results show that these multilingual models can reach same or better performance compared to monolingual models across language-specific test data while require less effort in creating features and model maintenance.