[PDF] HC2L: Hybrid and Cooperative Contrastive Learning for Cross-lingual Spoken Language Understanding. | Semantic Scholar (2024)

[PDF] Semantic Reader

Global-Local Contrastive Learning Framework (GL-CLeF) employs contrastive learning, leveraging bilingual dictionaries to construct multilingual views of the same utterance, then encourages their representations to be more similar than negative example pairs, which achieves to explicitly align representations of similar sentences across languages.

A new language representation model, BERT, designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers, which can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks.

A data augmentation framework to generate multi-lingual code-switching data to fine-tune mBERT, which encourages model to align representations from source and multiple target languages once by mixing their context information.

This paper proposes a novel cross-lingual SLU framework termed DiffSLU, which leverages powerful diffusion model to enhance the mutual guidance and also utilizes knowledge distillation to facilitate knowledge transfer.

A novel framework termed FC-MTLF is proposed, which applies a multi-task learning by introducing an auxiliary multilingual neural machine translation (NMT) task to compensate for the short-comings of code-switching.

A speaker-aware temporal graph (SATG) and a dual-task relational temporalGraph (DRTG) is proposed to facilitate relational temporal modeling in dialog understanding and dual- task reasoning and outperform state-of-the-art models by a large margin.

A novel model termed ReLa-Net, which surpasses the previous best model by over 20% in terms of overall accuracy on MixATIS dataset and also proposes the label-aware inter-dependent decoding mechanism to further exploit the label correlations for decoding.

A novel model termed Co-guiding Net is proposed, which implements a two-stage framework achieving the mutual guidances between the two tasks, and outperforms existing models by a large margin.

This paper proposes to model the utterance-slot-word structure by a multi-level contrastive learning framework at the utterances, slot and word levels to facilitate explicit alignment and develops a label-aware joint model leveraging label semantics to enhance the implicit alignment and feed to Contrastive learning.

Hierarchical text classification is a challenging subtask of multi-label classification due to its complex label hierarchy and Hierarchy-guided Contrastive Learning (HGCLR) is proposed to directly embed the hierarchy into a text encoder to dispense with the redundant hierarchy.

...

...