Papers with next-token prediction paradigm
The Mystery of the Pathological Path-star Task for Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Language models have become increasingly capable of solving a variety of complex tasks. |
| Approach: | They propose a path-star task where multiple arms radiate from a single starting node and each node is unique. |
| Outcome: | The proposed task is learnable using teacher-forcing in alternative settings and improves results across a variety of model types. |
LLM-ForcedAligner: A Non-Autoregressive and Accurate LLM-Based Forced Aligner for Multilingual and Long-Form Speech (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for forcing alignment are language-specific and prone to temporal shifts. |
| Approach: | They propose a slot-filling paradigm that uses time indices to predict slot positions. |
| Outcome: | The proposed method reduces accumulated temporal shifts by 69% compared with prior methods. |