Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Test-Time Code-Switching for Cross-lingual Aspect Sentiment Triplet Extraction

Dongming Sheng, Kexin Han, Hao Li, Yan Zhang, Yucheng Huang, Jun Lang, Wenqiang Liu

TL;DR

This work tackles cross-lingual aspect sentiment triplet extraction (ASTE) by introducing TT-CSW, a framework that couples boundary-aware code-switching during training with alignment-based test-time augmentation to improve term boundary detection and cross-language transfer. The training phase builds a bilingual code-switched dataset and trains a generative bilingual ASTE predictor alongside a bilingual alignment model, while the testing phase applies test-time code-switching and voting to produce language-consistent triplets. Empirical results on four datasets show TT-CSW yields an average weighted F1 improvement of 3.7% for backbone models, with additional gains from test-time augmentation; notably, fine-tuned smaller models with TT-CSW surpass ChatGPT and GPT-4 on average. The approach demonstrates practical impact by enabling stronger cross-lingual ASTE without resorting to large-language models, while highlighting trade-offs in translation-induced errors and computation cost.

Abstract

Aspect Sentiment Triplet Extraction (ASTE) is a thriving research area with impressive outcomes being achieved on high-resource languages. However, the application of cross-lingual transfer to the ASTE task has been relatively unexplored, and current code-switching methods still suffer from term boundary detection issues and out-of-dictionary problems. In this study, we introduce a novel Test-Time Code-SWitching (TT-CSW) framework, which bridges the gap between the bilingual training phase and the monolingual test-time prediction. During training, a generative model is developed based on bilingual code-switched training data and can produce bilingual ASTE triplets for bilingual inputs. In the testing stage, we employ an alignment-based code-switching technique for test-time augmentation. Extensive experiments on cross-lingual ASTE datasets validate the effectiveness of our proposed method. We achieve an average improvement of 3.7% in terms of weighted-averaged F1 in four datasets with different languages. Additionally, we set a benchmark using ChatGPT and GPT-4, and demonstrate that even smaller generative models fine-tuned with our proposed TT-CSW framework surpass ChatGPT and GPT-4 by 14.2% and 5.0% respectively.

Test-Time Code-Switching for Cross-lingual Aspect Sentiment Triplet Extraction

TL;DR

This work tackles cross-lingual aspect sentiment triplet extraction (ASTE) by introducing TT-CSW, a framework that couples boundary-aware code-switching during training with alignment-based test-time augmentation to improve term boundary detection and cross-language transfer. The training phase builds a bilingual code-switched dataset and trains a generative bilingual ASTE predictor alongside a bilingual alignment model, while the testing phase applies test-time code-switching and voting to produce language-consistent triplets. Empirical results on four datasets show TT-CSW yields an average weighted F1 improvement of 3.7% for backbone models, with additional gains from test-time augmentation; notably, fine-tuned smaller models with TT-CSW surpass ChatGPT and GPT-4 on average. The approach demonstrates practical impact by enabling stronger cross-lingual ASTE without resorting to large-language models, while highlighting trade-offs in translation-induced errors and computation cost.

Abstract

Aspect Sentiment Triplet Extraction (ASTE) is a thriving research area with impressive outcomes being achieved on high-resource languages. However, the application of cross-lingual transfer to the ASTE task has been relatively unexplored, and current code-switching methods still suffer from term boundary detection issues and out-of-dictionary problems. In this study, we introduce a novel Test-Time Code-SWitching (TT-CSW) framework, which bridges the gap between the bilingual training phase and the monolingual test-time prediction. During training, a generative model is developed based on bilingual code-switched training data and can produce bilingual ASTE triplets for bilingual inputs. In the testing stage, we employ an alignment-based code-switching technique for test-time augmentation. Extensive experiments on cross-lingual ASTE datasets validate the effectiveness of our proposed method. We achieve an average improvement of 3.7% in terms of weighted-averaged F1 in four datasets with different languages. Additionally, we set a benchmark using ChatGPT and GPT-4, and demonstrate that even smaller generative models fine-tuned with our proposed TT-CSW framework surpass ChatGPT and GPT-4 by 14.2% and 5.0% respectively.
Paper Structure (30 sections, 4 equations, 3 figures, 3 tables)

This paper contains 30 sections, 4 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Task Definition
  4. Training Phase
  5. Boundary-aware Code-switching
  6. Bilingual Structural Prediction
  7. Bilingual Alignment Prediction
  8. Testing Phase
  9. Alignment-based Code-switching
  10. Alignment on Candidate Triplets
  11. Experiments
  12. Datasets
  13. Implementation Details
  14. Compared Methods
  15. mT5-base xue2021mt5
  16. ...and 15 more sections

Figures (3)

  • Figure 1: An example of testing phase in cross-lingual ASTE task on Spanish dataset. Phrases with bold and underlined words represent aspect and opinion terms respectively. The substituted words are highlighted within the orange boxes. The diagram on the bottom right illustrates the pipeline of our proposed alignment-based code-switching method.
  • Figure 2: The overall architecture of our proposed TT-CSW framework.
  • Figure 3: Effect of maximum n-gram and number of candidates for code-switching in test phase on Spanish and Catalan datasets. "# candidates" refers to the number of augmented input sentence.