Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning

By Antonia Creswell, Murray Shanahan, Irina Higgins. 2022

482 citations recorded by the source corpus.

Large language models (LLMs) have been shown to be capable of impressive few-shot generalisation to new tasks. However, they still tend to perform poorly on multi-step logical reasoning problems. Here we carry out a comprehensive evaluation of LLMs on 50 tasks that probe different aspects of logical reasoning. We show that language models tend to perform fairly well at single step inference or entailment tasks, but struggle to chain together multiple reasoning steps to solve more complex problems. In light of this, we propose a Selection-Inference (SI) framework that exploits pre-trained LLMs as general processing modules, and alternates between selection and inference to generate a series of interpretable, casual reasoning steps leading to the final answer. We show that a 7B parameter LLM used within the SI framework in a 5-shot generalisation setting, with no fine-tuning, yields a performance improvement of over 100% compared to an equivalent vanilla baseline on a suite of 10 logical reasoning tasks. The same model in the same setting even outperforms a significantly larger 280B parameter baseline on the same suite of tasks. Moreover, answers produced by the SI framework are accompanied by a causal natural-language-based reasoning trace, which has important implications for the safety and trustworthiness of the system.

Subfields: Artificial Intelligence, Natural Language Processing, Prompting & Reasoning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

In Prompting & Reasoning

Citations in the corpus

ReAct: Synergizing Reasoning and Acting in Language Models (2022, neutral): “For example Selection-Inference (Creswell et al., 2022) divides the reasoning process into two steps of “selection” and “inference”.”
Augmented Language Models: a Survey (2023, background): “Previous work has shown that LLMs can solve simple reasoning problems but fail at complex reasoning (Creswell et al., 2022): hence, this section focuses on various strategies to augment LM’s reasoning skills.”
Improving alignment of dialogue agents via targeted human judgements (2022, background): “We believe these limitations are best addressed via multistep reasoning (Creswell et al., 2022; Dohan et al., 2022; Lewkowycz et al., 2022), with the further benefit of producing interpretable reasoning traces.”
Faithful Reasoning Using Large Language Models (2022, background): “While contemporary language models (LMs) are good at many natural language tasks, they often struggle with logical reasoning (Betz et al., 2021; Creswell et al., 2022; Dasgupta et al., 2022; Rae et al., 2021; Zhang et al., 2022).”
Rationale-Augmented Ensembles in Language Models (2022, background): “…pretrained large language models has been achieved by expanding prompt exemplars with rationales, delivering successes in a variety of natural language reasoning tasks (Wei et al., 2022; Chowdhery et al., 2022; Lampinen et al., 2022; Creswell et al., 2022; Kojima et al., 2022; Jung et al., 2022).”
Language Model Cascades (2022, background): “Inference can be implemented by ancestral sampling.”
Let's Verify Step by Step (2023, background): “Within the domain of logical reasoning, models trained with outcome supervision regularly use incorrect reasoning to reach the correct final answer (Zelikman et al., 2022; Creswell et al., 2022).”
Decomposed Prompting: A Modular Approach for Solving Complex Tasks (2022, background): “, 2022), selection inference (Creswell et al., 2022), coupling in-context learning with sparse fine-tuning (Zelikman et al.”
Large language models in health care: Development, applications, and challenges (2023, methodology): “Proposed methods to improve interpretability include a selection inference multi ‐ step reasoning framework by Creswell et al. to generate a series of casual reasoning steps toward the final generated response [58].”
Can Language Models Solve Graph Problems in Natural Language? (2023, background): “In multi-hop question answering, LLMs implicitly find connections and paths among a vast network of entities and concepts [Creswell et al., 2022].”
Cognitive Architectures for Language Agents (2023, background): “, 2020) or pre-defined sequences of prompts to structure reasoning (Creswell et al., 2023; Wu et al., 2022b).”
From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge (2024, unknown): “Creswell et al. (2023) decompose the reasoning process into Selection and Inference.”
Large Language Models Can Be Easily Distracted by Irrelevant Context (2023, background): “However, previous work largely focuses on designing models that require extra training, and prompting alone still hardly achieves the same level of performance as ﬁnetuned models for these tasks (Han et al., 2022; Creswell et al., 2022).”
Making Retrieval-Augmented Language Models Robust to Irrelevant Context (2023, background): “Amongst others, Jia & Liang (2017); Petroni et al. (2020); Creswell et al. (2023) show that adding random or irrelevant context can decrease QA performance.”
See, Think, Confirm: Interactive Prompting Between Vision and Language Models for Knowledge-based Visual Reasoning (2023, unknown): “Some prompting techniques [16,47,60,73] have also been developed for more effective or transparent reasoning in NLP.”
Solving math word problems with process- and outcome-based feedback (2022, unknown): “For example, recent work on natural-language-based reasoning (Zelikman et al., 2022; Creswell et al., 2022) suggests that models optimized exclusively for ﬁnal-answer correctness can produce the correct ﬁnal answer, even when their generated reasoning traces are incorrect.”
Alphazero-like Tree-Search can Guide Large Language Model Decoding and Training (2023, unknown): “…al., 2023; Gunasekar et al., 2023; Feng et al., 2023; Taylor et al., 2022), more sophisticated prompt design (Wei et al., 2022; Zhou et al., 2022; Creswell et al., 2022), or better training algorithms with Supervised Learning or Reinforcement Learning (RL) (Dong et al., 2023; Gulcehre et al.,…”