Matching Structure for Dual Learning

Hao Fei1, Shengqiong Wu1, Yafeng Ren2, Meishan Zhang3,
1Sea-NExT Joint Lab, National University of Singapore,
2Guangdong University of Foreign Studies,
3Harbin Institute of Technology (Shenzhen)  
Paper Code Poster

Abstract

Many natural language processing (NLP) tasks appear in dual forms, which are generally solved by dual learning technique that models the dualities between the coupled tasks. In this work, we propose to further enhance dual learning with structure matching that explicitly builds structural connections in between. Starting with the dual text↔text generation, we perform duallysyntactic structure co-echoing of the region of interest (RoI) between the task pair, together with a syntax cross-reconstruction at the decoding side. We next extend the idea to a text↔non-text setup, making alignment between the syntactic-semantic structure. Over 2*14 tasks covering 5 dual learning scenarios, the proposed structure matching method shows its significant effectiveness in enhancing existing dual learning. Our method can retrieve the key RoIs that are highly crucial to the task performance. Besides NLP tasks, it is also revealed that our approach has great potential in facilitating more non-text↔non-text scenarios.


Presentation

Method

▶   Dually-Syntactic Structure Matching for Text↔Text Dual Learning:



▶   Syntactic-Semantic Structure Matching for Text↔Non-Text Dual Learning:


Experiment

• Text↔Text Dual Learning




• Text↔Non-Text Dual Learning



• Non-Text↔Non-Text Dual Learning


In-depth Analysis

Analysis Targets:


▶ Structure matching helps correctly retrieve and emphasize the key RoIs that are crucial to the task improvements.



▶ Our method strengthens the duality between two dual tasks by correctly aligning the RoIs.


▶ With syntactic structure co-echoing between the text↔text dual learning, the generated sentences are more diversified and grammarly correct.



▶ Non-text↔non-text dual learning can also benefit from structure matching.
▶ The dual tasks with richer structural information for the alignments will lead to better improvements.

Poster

Paper

BibTeX

@inproceedings{MSDual2022ICML,
  author    = {Hao Fei and
               Shengqiong Wu and
               Yafeng Ren and
               Meishan Zhang},
  title     = {Matching Structure for Dual Learning},
  booktitle = {Proceedings of the International Conference on Machine Learning, {ICML}},
  pages     = {6373--6391},
  year      = {2022},
}