Research map / Label-efficient and robust learning / Recognition and long-tail
On the Importance of Word Order Information in Cross-lingual Sequence Labeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace an order-sensitive encoder with ORT | Cross-lingual sequence labeling models · NER target-language average, frozen word-level embeddings | F1: 27.85 -> 30.67 | Helped |
| Replace an order-sensitive encoder with ORT | Cross-lingual sequence labeling models · SF target-language average, frozen word-level embeddings | F1: 44.02 -> 46.19 | Helped |
| Use ORT instead of RPT | POS cross-lingual labeling with word-level embeddings · POS target-language average | accuracy: 44.11 -> 53.10 | Helped |
| Initialize Transformer positional embeddings from M-BERT and freeze them | Transformer with frozen M-BERT embeddings · POS target-language average | accuracy: 64.72 -> 66.98 | Helped |
| Freeze M-BERT positional embeddings during fine-tuning | M-BERT · POS target-language average | accuracy: 75.12 -> 75.97 | Helped |
| Replace BiLSTM with ORT | Strong zero-shot NLU model · NLU, Spanish | SF F1: 65.79 -> 71.36 | Helped |
| Replace BiLSTM with ORT | Strong zero-shot NER model · NER target-language average | F1: 67.30 -> 68.73 | Helped |
| Replace Conv1d with a linear feed-forward layer in ORT (ablation) | ORT + CRF · SF, Spanish | F1: 66.84 -> 58.27 | Hurt |
| Replace Conv1d with a linear feed-forward layer in ORT (ablation) | ORT + CRF · SF, Thai | F1: 25.53 -> 20.35 | Hurt |
| Add Conv1d rather than a linear feed-forward layer (ablation) | ORT + CRF · SF, Spanish | F1: 58.27 -> 66.84 | Helped |
| Add the CRF layer (ablation) | ORT with Conv1d · SF, Spanish | F1: 61.76 -> 66.84 | Helped |
| Add the CRF layer (ablation) | ORT with Conv1d · SF, Thai | F1: 22.44 -> 25.53 | Helped |
Closest papers by meaning
- A Comparison of Architectures and Pretraining Methods for Contextualized Multilingual Word Embeddings (AAAI 2020 · similarity 0.87)
- Empower Sequence Labeling with Task-Aware Neural Language Model (AAAI 2018 · similarity 0.87)
- Constrained Decoding for Cross-lingual Label Projection (ICLR 2024 · similarity 0.86)
- Modeling Sequential Sentence Relation to Improve Cross-lingual Dense Retrieval (ICLR 2023 · similarity 0.86)
- Cross-lingual Alignment vs Joint Training: A Comparative Study and A Simple Unified Framework (ICLR 2020 · similarity 0.86)
- Implicit Word Reordering with Knowledge Distillation for Cross-Lingual Dependency Parsing (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.