CALICO: Conversational Agent Localization via Synthetic Data Generation

TL;DR

CALICO is a novel method that fine-tunes large language models to generate localized conversational data across multiple languages, improving slot translation accuracy and creating more challenging test sets for multilingual conversational agents.

Contribution

CALICO introduces a new approach for multilingual slot localization using synthetic data generation and an iterative filtering mechanism, outperforming existing literal translation methods.

Findings

CALICO produces more accurate slot translations than state-of-the-art methods.

The human-localized test set is more challenging than the original.

CALICO improves downstream conversational agent performance.

Abstract

We present CALICO, a method to fine-tune Large Language Models (LLMs) to localize conversational agent training data from one language to another. For slots (named entities), CALICO supports three operations: verbatim copy, literal translation, and localization, i.e. generating slot values more appropriate in the target language, such as city and airport names located in countries where the language is spoken. Furthermore, we design an iterative filtering mechanism to discard noisy generated samples, which we show boosts the performance of the downstream conversational agent. To prove the effectiveness of CALICO, we build and release a new human-localized (HL) version of the MultiATIS++ travel information test set in 8 languages. Compared to the original human-translated (HT) version of the test set, we show that our new HL version is more challenging. We also show that CALICO out-performs state-of-the-art LINGUIST (which relies on literal slot translation out of context) both on the HT case, where CALICO generates more accurate slot translations, and on the HL case, where CALICO generates localized slots which are closer to the HL test set.