The Importance of Generation Order in Language Modeling
This work addresses a fundamental issue in neural language modeling for tasks like machine translation and summarization, though it is incremental as it builds on existing autoregressive approaches.
The paper tackled the problem of token generation order in autoregressive language models by proposing a novel two-pass model that creates partially-filled sentence templates and then fills in missing tokens, finding that generating function words first followed by content words yields the most effective strategy with surprisingly large variations in model quality.
Neural language models are a critical component of state-of-the-art systems for machine translation, summarization, audio transcription, and other tasks. These language models are almost universally autoregressive in nature, generating sentences one token at a time from left to right. This paper studies the influence of token generation order on model quality via a novel two-pass language model that produces partially-filled sentence "templates" and then fills in missing tokens. We compare various strategies for structuring these two passes and observe a surprisingly large variation in model quality. We find the most effective strategy generates function words in the first pass followed by content words in the second. We believe these experimental results justify a more extensive investigation of generation order for neural language models.