Idea-Gated Transformers: Enforcing Semantic Coherence via Differentiable Vocabulary Pruning
This addresses the issue of uncontrollable generation in language models for users needing more predictable outputs, representing an incremental improvement.
The paper tackled the problem of topic drift in autoregressive language models by introducing the Idea-Gated Transformer, which enforces semantic coherence through differentiable vocabulary pruning, resulting in significantly superior domain retention compared to a GPT-2 baseline on WikiText-103.
Autoregressive Language Models (LLMs) trained on Next-Token Prediction (NTP) often suffer from ``Topic Drift'' where the generation wanders away from the initial prompt due to a reliance on local associations rather than global planning \citep{holtzman2019curious}. While scaling model size mitigates this \citep{brown2020language}, the fundamental myopia of the NTP objective remains. In this work, we introduce the Idea-Gated Transformer, a novel architecture that separates semantic planning from syntactic generation. We introduce an auxiliary ``Idea Head'' trained to predict the bag-of-words distribution for a future context window, creating a latent ``Concept Vector'' that actively gates the main vocabulary during generation. We propose a differentiable gating mechanism that suppresses semantically irrelevant tokens, effectively pruning the search space in real-time. Experiments on WikiText-103 demonstrate that while the Idea-Gated model achieves comparable validation perplexity to a standard GPT-2 baseline, it exhibits significantly superior Domain Retention. Qualitative and quantitative analysis reveals that the gating mechanism successfully locks generation into specific semantic clusters (e.g., Finance, Science) and resists associative drift, offering a parameter-efficient path toward more controllable language modeling.