Research map / Label-efficient and robust learning / Recognition and long-tail
Minimally-Constrained Multilingual Embeddings via Artificial Code-Switching
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 |
|---|---|---|---|
| add artificial code-switching to multilingual CBOW training | CBOW multilingual embedding · Word-pair cosine | avg cosine similarity: 0.286 -> 0.439 | Helped |
| add dictionary constraints to multilingual CBOW | CBOW multilingual embedding · Word-pair cosine | avg cosine similarity: 0.286 -> 0.422 | Helped |
| add artificial code-switching to multilingual CBOW | CBOW multilingual embedding · Mixed En+Fr word analogy | accuracy: 39.3% -> 52.6 | Helped |
| add dictionary constraints to multilingual CBOW | CBOW multilingual embedding · Mixed En+Fr word analogy | accuracy: 39.3% -> 43.6 | Helped |
| train ACS multilingual embedding instead of monolingual French | CBOW French embedding · French word analogy | accuracy: 45.2 -> 53.3 | Helped |
| train ACS multilingual embedding instead of monolingual English | CBOW English embedding · English word analogy | accuracy: 81.1 -> 66.9 | Hurt |
| add dictionary constraints to multilingual CBOW | CBOW multilingual embedding · English word analogy | accuracy: 77.5% -> 52.8 | Hurt |
| add dictionary constraints to multilingual CBOW | CBOW multilingual embedding · French word analogy | accuracy: 47.8% -> 40.4 | Hurt |
| use ACS embeddings instead of no-constraint CBOW for cross-lingual sentiment | English-trained sentiment classifier · en-de bilingual sentiment, all docs | % of target baseline accuracy: 58.5 -> 74.7 | Helped |
| use ACS embeddings instead of no-constraint CBOW for cross-lingual sentiment | English-trained sentiment classifier · en-no multilingual sentiment, all docs | % of target baseline accuracy: 94.4 -> 88.8 | Hurt |
| use multilingual embeddings (ACS) instead of no embedding features for cross-lingual sentiment | English-trained sentiment classifier · en-fr bilingual sentiment, Twitter | % of target baseline accuracy: 53.1 -> 84.0 | Helped |
Closest papers by meaning
- Word translation without parallel data (ICLR 2018 · similarity 0.89)
- A Comparison of Architectures and Pretraining Methods for Contextualized Multilingual Word Embeddings (AAAI 2020 · similarity 0.89)
- Normalization of Language Embeddings for Cross-Lingual Alignment (ICLR 2022 · similarity 0.89)
- MMTEB: Massive Multilingual Text Embedding Benchmark (ICLR 2025 · similarity 0.88)
- Repurposing Language Models into Embedding Models: Finding the Compute-Optimal Recipe (NeurIPS 2024 · similarity 0.87)
- DEPT: Decoupled Embeddings for Pre-training Language Models (ICLR 2025 · similarity 0.87)
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.