Papers by Sungwon Lyu
Sparse and Decorrelated Representations for Stable Zero-shot NMT (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Using a single encoder and decoder for all directions is a popular scheme for multilingual NMT. |
| Approach: | They propose a scheme that uses a single encoder and decoder for all directions . they show that enforcing sparsity and decorrelation on encoder intermediate representations mitigates this problem . |
| Outcome: | The proposed model degenerates when decoding non-English texts into English regardless of the target specifier token. |
AttnIO: Knowledge Graph Exploration with In-and-Out Attention Flow for Knowledge-Grounded Dialogue (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing models for retrieving proper knowledge relevant to conversational context use only KG structure . empirical evaluations present a marked performance improvement of AttnIO compared to all baselines in OpenDialKG dataset . |
| Approach: | They propose a dialog-conditioned path traversal model that makes full use of rich structural information in KG . they show a marked performance improvement compared to baselines in OpenDialKG a KG dataset . |
| Outcome: | The proposed model makes full use of rich structural information in KG structure . it can be trained to generate an adequate knowledge path even when paths are not available . |
Revisiting Modularized Multilingual NMT to Meet Industrial Demands (2020.emnlp-main)
Copied to clipboard
| Challenge: | Currently, the complete sharing of parameters for multilingual translation (1-1) is the most popular approach because of its compactness. |
| Approach: | They propose to use a multilingual neural machine translation model that only shares modules among the same languages as 1-1 to satisfy industrial requirements. |
| Outcome: | The proposed model can enjoy the benefits of multi-way training without the capacity bottleneck and low maintainability. |