DECIDER: A Dual-System Rule-Controllable Decoding Framework for Language Generation
This work addresses the challenge of making language model generation more controllable and natural for applications requiring rule-based constraints, though it is incremental in combining existing ideas from cognitive theory and logic.
The authors tackled the problem of controlling text generation in large language models by proposing DECIDER, a dual-system framework that integrates a First-Order Logic reasoner to enforce high-level rules, resulting in more human-like and logic-controlled outputs as demonstrated on CommonGen and PersonaChat datasets.
Constrained decoding approaches aim to control the meaning or style of text generated by the pre-trained large language models (LLMs or also PLMs) for various tasks at inference time. However, these methods often guide plausible continuations by greedily and explicitly selecting targets. Though fulfilling the task requirements, these methods may overlook certain general and natural logics that humans would implicitly follow towards such targets. Inspired by cognitive dual-process theory, in this work, we propose a novel decoding framework DECIDER where the base LLMs are equipped with a First-Order Logic (FOL) reasoner to express and evaluate the rules, along with a decision function that merges the outputs of both systems to guide the generation. Unlike previous constrained decodings, DECIDER transforms the encouragement of target-specific words into all words that satisfy several high-level rules, enabling us to programmatically integrate our logic into LLMs. Experiments on CommonGen and PersonaChat demonstrate that DECIDER effectively follows given FOL rules to guide LLMs in a more human-like and logic-controlled manner.