Papers with translation capabilities
Action-Concentrated Embedding Framework: This Is Your Captain Sign-tokening (2024.lrec-main)
Copied to clipboard
| Challenge: | ACE is a new sign token embedding framework that tracks a signer’s actions based on human posture estimation and captures the token embeds using a short-time Fourier transform. |
| Approach: | They propose a novel sign token embedding framework that tracks a signer’s actions based on human posture estimation and a dedicated notation system tailored for sign language. |
| Outcome: | The proposed framework outperforms previous studies in translation performance against a disaster sign language dataset and improves by up to 5.79% for BLEU-4 and 5.46% for ROUGE-L metric. |
Searching for Needles in a Haystack: On the Role of Incidental Bilingualism in PaLM’s Translation Capability (2023.acl-long)
Copied to clipboard
| Challenge: | Large multilingual language models exhibit impressive zero- or few-shot machine translation capabilities, despite never having been explicitly and intentionally exposed to translation data. |
| Approach: | They propose a mixed-method approach to measure and understand incidental bilingualism at scale using the Pathways Language Model. |
| Outcome: | The proposed model is exposed to over 30 million translation pairs across at least 44 languages. |