Answering complex questions that require making latent decisions is a challenging task, especially when limited supervision is available. Recent works leverage the capabilities of large language models (LMs) to perform complex question answering in a few-shot setting by demonstrating how to output intermediate rationalizations while solving the complex question in a single pass. We introduce ``Successive Prompting'', where we iteratively break down a complex task into a simple task, solve it, and then repeat the process until we get the final solution. Successive prompting decouples the supervision for decomposing complex questions from the supervision for answering simple questions, allowing us to (1) have multiple opportunities to query in-context examples at each reasoning step (2) learn question decomposition separately from question answering, including using synthetic data, and (3) use bespoke (fine-tuned) components for reasoning steps where a large LM does not perform well. The intermediate supervision is typically manually written, which can be expensive to collect. We introduce a way to generate a synthetic dataset which can be used to bootstrap a model's ability to decompose and answer intermediate questions. Our best model (with successive prompting) achieves an improvement of ~5% absolute F1 on a few-shot version of the DROP dataset when compared with a state-of-the-art model with the same supervision.

Question-answering datasets require a broad set of reasoning skills. We show how to use question decompositions to teach language models these broad reasoning skills in a robust fashion. Specifically, we use widely available QDMR representations to programmatically create hard-to-cheat synthetic contexts for real questions in six multi-step reasoning datasets. These contexts are carefully designed to avoid reasoning shortcuts prevalent in real contexts that prevent models from learning the right skills. This results in a pretraining dataset, named TeaBReaC, containing 525K multi-step questions (with associated formal programs) covering about 900 reasoning patterns. We show that pretraining standard language models (LMs) on TeaBReaC before fine-tuning them on target datasets improves their performance by up to 13 F1 points across 4 multi-step QA datasets, with up to 21 point gain on more complex questions. The resulting models also demonstrate higher robustness, with a 5-8 F1 point improvement on two contrast sets. Furthermore, TeaBReaC pretraining substantially improves model performance and robustness even when starting with numerate LMs pretrained using recent methods (e.g., PReasM, POET). Our work thus shows how to effectively use decomposition-guided contexts to robustly teach multi-step reasoning.

Retrieval-augmented in-context learning has emerged as a powerful approach for addressing knowledge-intensive tasks using frozen language models (LM) and retrieval models (RM). Existing work has combined these in simple "retrieve-then-read" pipelines in which the RM retrieves passages that are inserted into the LM prompt. To begin to fully realize the potential of frozen LMs and RMs, we propose Demonstrate-Search-Predict (DSP), a framework that relies on passing natural language texts in sophisticated pipelines between an LM and an RM. DSP can express high-level programs that bootstrap pipeline-aware demonstrations, search for relevant passages, and generate grounded predictions, systematically breaking down problems into small transformations that the LM and RM can handle more reliably. We have written novel DSP programs for answering questions in open-domain, multi-hop, and conversational settings, establishing in early evaluations new state-of-the-art in-context learning results and delivering 37-200%, 8-40%, and 80-290% relative gains against vanilla LMs, a standard retrieve-then-read pipeline, and a contemporaneous self-ask pipeline, respectively.

Multi-hop reading comprehension requires not only the ability to reason over raw text but also the ability to combine multiple evidence. We propose a novel learning approach that helps language models better understand difficult multi-hop questions and perform "complex, compositional" reasoning. Our model first learns to decompose each multi-hop question into several sub-questions by a trainable question decomposer. Instead of answering these sub-questions, we directly concatenate them with the original question and context, and leverage a reading comprehension model to predict the answer in a sequence-to-sequence manner. By using the same language model for these two components, our best seperate/unified t5-base variants outperform the baseline by 7.2/6.1 absolute F1 points on a hard subset of DROP dataset.

Recently, there has been significant progress in teaching language models to perform step-by-step reasoning to solve complex numerical reasoning tasks. Chain-of-thoughts prompting (CoT) is by far the state-of-art method for these tasks. CoT uses language models to perform both reasoning and computation in the multi-step `thought' process. To disentangle computation from reasoning, we propose `Program of Thoughts' (PoT), which uses language models (mainly Codex) to express the reasoning process as a program. The computation is relegated to an external computer, which executes the generated programs to derive the answer. We evaluate PoT on five math word problem datasets (GSM, AQuA, SVAMP, TabMWP, MultiArith) and three financial-QA datasets (FinQA, ConvFinQA, TATQA) for both few-shot and zero-shot setups. Under both few-shot and zero-shot settings, PoT can show an average performance gain over CoT by around 12\% across all the evaluated datasets. By combining PoT with self-consistency decoding, we can achieve SoTA performance on all math problem datasets and near-SoTA performance on financial datasets. All of our data and code are released in Github\footnote{\url{https://github.com/wenhuchen/Program-of-Thoughts}}.

This paper proposes a question-answering system that can answer questions whose supporting evidence is spread over multiple (potentially long) documents. The system, called Visconde, uses a three-step pipeline to perform the task: decompose, retrieve, and aggregate. The first step decomposes the question into simpler questions using a few-shot large language model (LLM). Then, a state-of-the-art search engine is used to retrieve candidate passages from a large collection for each decomposed question. In the final step, we use the LLM in a few-shot setting to aggregate the contents of the passages into the final answer. The system is evaluated on three datasets: IIRC, Qasper, and StrategyQA. Results suggest that current retrievers are the main bottleneck and that readers are already performing at the human level as long as relevant passages are provided. The system is also shown to be more effective when the model is induced to give explanations before answering a question. Code is available at \url{https://github.com/neuralmind-ai/visconde}.

Answering complex questions often requires multi-step reasoning in order to obtain the final answer. Most research into decompositions of complex questions involves open-domain systems, which have shown success in using these decompositions for improved retrieval. In the machine reading setting, however, work to understand when decompositions are helpful is understudied. We conduct experiments on decompositions in machine reading to unify recent work in this space, using a range of models and datasets. We find that decompositions can be helpful in the few-shot case, giving several points of improvement in exact match scores. However, we also show that when models are given access to datasets with around a few hundred or more examples, decompositions are not helpful (and can actually be detrimental). Thus, our analysis implies that models can learn decompositions implicitly even with limited data.

Powerful generative models have led to recent progress in question generation (QG). However, it is difficult to measure advances in QG research since there are no standardized resources that allow a uniform comparison among approaches. In this paper, we introduce QG-Bench, a multilingual and multidomain benchmark for QG that unifies existing question answering datasets by converting them to a standard QG setting. It includes general-purpose datasets such as SQuAD for English, datasets from ten domains and two styles, as well as datasets in eight different languages. Using QG-Bench as a reference, we perform an extensive analysis of the capabilities of language models for the task. First, we propose robust QG baselines based on fine-tuning generative language models. Then, we complement automatic evaluation based on standard metrics with an extensive manual evaluation, which in turn sheds light on the difficulty of evaluating QG models. Finally, we analyse both the domain adaptability of these models as well as the effectiveness of multilingual models in languages other than English. QG-Bench is released along with the fine-tuned models presented in the paper https://github.com/asahi417/lm-question-generation, which are also available as a demo https://autoqg.net/.

Recent work has shown that large language models are capable of generating natural language reasoning steps or Chains-of-Thoughts (CoT) to answer a multi-step question when prompted to do so. This is insufficient, however, when the necessary knowledge is not available or up-to-date within a model's parameters. A straightforward approach to address this is to retrieve text from an external knowledge source using the question as a query and prepend it as context to the model's input. This, however, is also insufficient for multi-step QA where \textit{what to retrieve} depends on \textit{what has already been derived}. To address this issue we propose IRCoT, a new approach that interleaves retrieval with CoT for multi-step QA, guiding the retrieval with CoT and in turn using retrieved results to improve CoT. Our experiments with GPT3 show substantial improvements in retrieval (up to 22 points) and downstream QA (up to 16 points) over the baselines on four datasets: HotpotQA, 2WikiMultihopQA, MuSiQue, and IIRC. Notably, our method also works well for much smaller models such as T5-Flan-large (0.7B) without any additional training.

Recent work has shown that fine-tuning large pre-trained language models on a collection of tasks described via instructions, a.k.a. instruction-tuning, improves their zero and few-shot generalization to unseen tasks. However, there is a limited understanding of the performance trade-offs of different decisions made during the instruction-tuning process. These decisions include the scale and diversity of the instruction-tuning benchmark, different task sampling strategies, fine-tuning with and without demonstrations, training using specialized datasets for reasoning and dialogue, and finally, the fine-tuning objectives themselves. In this paper, we characterize the effect of instruction-tuning decisions on downstream task performance when scaling both model and benchmark sizes. To this end, we create OPT-IML Bench: a large benchmark for Instruction Meta-Learning (IML) of 2000 NLP tasks consolidated into task categories from 8 existing benchmarks, and prepare an evaluation framework to measure three types of model generalizations: to tasks from fully held-out categories, to held-out tasks from seen categories, and to held-out instances from seen tasks. Through the lens of this framework, we first present insights about instruction-tuning decisions as applied to OPT-30B and further exploit these insights to train OPT-IML 30B and 175B, which are instruction-tuned versions of OPT. OPT-IML demonstrates all three generalization abilities at both scales on four different evaluation benchmarks with diverse tasks and input formats -- PromptSource, FLAN, Super-NaturalInstructions, and UnifiedSKG. Not only does it significantly outperform OPT on all benchmarks but is also highly competitive with existing models fine-tuned on each specific benchmark. We release OPT-IML at both scales, together with the OPT-IML Bench evaluation framework.

Open-Domain Question Answering (ODQA) requires models to answer factoid questions with no context given. The common way for this task is to train models on a large-scale annotated dataset to retrieve related documents and generate answers based on these documents. In this paper, we show that the ODQA architecture can be dramatically simplified by treating Large Language Models (LLMs) as a knowledge corpus and propose a Self-Prompting framework for LLMs to perform ODQA so as to eliminate the need for training data and external knowledge corpus. Concretely, we firstly generate multiple pseudo QA pairs with background passages and one-sentence explanations for these QAs by prompting LLMs step by step and then leverage the generated QA pairs for in-context learning. Experimental results show our method surpasses previous state-of-the-art methods by +8.8 EM averagely on three widely-used ODQA datasets, and even achieves comparable performance with several retrieval-augmented fine-tuned models.

Reasoning, as an essential ability for complex problem-solving, can provide back-end support for various real-world applications, such as medical diagnosis, negotiation, etc. This paper provides a comprehensive survey of cutting-edge research on reasoning with language model prompting. We introduce research works with comparisons and summaries and provide systematic resources to help beginners. We also discuss the potential reasons for emerging such reasoning abilities and highlight future research directions.

大型语言模型在各种任务上显示出令人印象深刻的几次结果。但是，当知识是此类结果的关键时，就像问题回答和事实检查之类的任务一样，似乎需要存储知识的大量参数计数。众所周知，检索增强模型可以在不需要多个参数的情况下在知识密集的任务上表现出色，但是目前尚不清楚它们是否在几个弹药设置中工作。在这项工作中，我们介绍了地图集，这是一个经过精心设计和预先训练的增强语言模型，能够通过很少的培训示例学习知识密集型任务。我们对包括MMLU，苏格兰短裙和归类等各种任务进行评估，并研究文档索引内容的影响，表明它可以很容易地进行更新。值得注意的是，在自然问题上仅使用64个示例在自然问题上达到超过42 \％的准确性，尽管参数少了50倍，但比540B参数模型的表现优于540b参数模型。

在维持预审预定序列模型的灵活性的同时，是否有利于常识性推理，这仍然是一个悬而未决的问题。为了调查这个问题，我们开发了生成的知识提示，该提示包括从语言模型中生成知识，然后在回答问题时提供知识作为附加输入。我们的方法不需要特定于任务的监督知识集成或访问结构化的知识库，但它可以提高四个常识性推理任务上的大规模，最先进的模型的性能，从而实现最先进-ART结果取决于数值常识（NumerSense），通用常识性（Commonsenseqa 2.0）和科学常识（QASC）基准。产生的知识促使大型语言模型是灵活的外部知识来源，以改善常识性推理。我们的代码可从https://github.com/liujch1998/gkp获得

Structured tabular data exist across nearly all fields. Reasoning task over these data aims to answer questions or determine the truthiness of hypothesis sentences by understanding the semantic meaning of a table. While previous works have devoted significant efforts to the tabular reasoning task, they always assume there are sufficient labeled data. However, constructing reasoning samples over tables (and related text) is labor-intensive, especially when the reasoning process is complex. When labeled data is insufficient, the performance of models will suffer an unendurable decline. In this paper, we propose a unified framework for unsupervised complex tabular reasoning (UCTR), which generates sufficient and diverse synthetic data with complex logic for tabular reasoning tasks, assuming no human-annotated data at all. We first utilize a random sampling strategy to collect diverse programs of different types and execute them on tables based on a "Program-Executor" module. To bridge the gap between the programs and natural language sentences, we design a powerful "NL-Generator" module to generate natural language sentences with complex logic from these programs. Since a table often occurs with its surrounding texts, we further propose novel "Table-to-Text" and "Text-to-Table" operators to handle joint table-text reasoning scenarios. This way, we can adequately exploit the unlabeled table resources to obtain a well-performed reasoning model under an unsupervised setting. Our experiments cover different tasks (question answering and fact verification) and different domains (general and specific), showing that our unsupervised methods can achieve at most 93% performance compared to supervised models. We also find that it can substantially boost the supervised performance in low-resourced domains as a data augmentation technique. Our code is available at https://github.com/leezythu/UCTR.

Despite the success of large language models (LLMs) in various natural language processing (NLP) tasks, the stored knowledge in these models may inevitably be incomplete, out-of-date, or incorrect. This motivates the need to utilize external knowledge to assist LLMs. Unfortunately, current methods for incorporating external knowledge often require additional training or fine-tuning, which can be costly and may not be feasible for LLMs. To address this issue, we propose a novel post-processing approach, rethinking with retrieval (RR), which retrieves relevant external knowledge based on the decomposed reasoning steps obtained from the chain-of-thought (CoT) prompting. This lightweight approach does not require additional training or fine-tuning and is not limited by the input length of LLMs. We evaluate the effectiveness of RR through extensive experiments with GPT-3 on three complex reasoning tasks: commonsense reasoning, temporal reasoning, and tabular reasoning. Our results show that RR can produce more faithful explanations and improve the performance of LLMs.

我们提出了一种系统推理的方法，该方法生产了基于事实基础的人类可解释的证明树。我们的解决方案类似于经典的基于序言的推理引擎的风格，在该引擎中，我们通过神经语言建模，指导生成和半磁头密集检索的结合来代替手工制作的规则。这款新颖的推理引擎Nellie动态实例化了可解释的推理规则，这些规则捕获和分数构成（DE）在自然语言陈述上。内莉（Nellie）在科学质量检查数据集上提供竞争性能，需要对多个事实进行结构化解释。

Step-by-step reasoning approaches like chain-of-thought (CoT) have proved to be a very effective technique to induce reasoning capabilities in large language models. However, the success of the CoT approach depends primarily on model size, and often billion parameter-scale models are needed to get CoT to work. In this paper, we propose a knowledge distillation approach, that leverages the step-by-step CoT reasoning capabilities of larger models and distils these reasoning abilities into smaller models. Our approach Decompositional Distillation learns a semantic decomposition of the original problem into a sequence of subproblems and uses it to train two models: a) a problem decomposer that learns to decompose the complex reasoning problem into a sequence of simpler sub-problems and b) a problem solver that uses the intermediate subproblems to solve the overall problem. On a multi-step math word problem dataset (GSM8K), we boost the performance of GPT-2 variants up to 35% when distilled with our approach compared to CoT. We show that using our approach, it is possible to train a GPT-2-large model (775M) that can outperform a 10X larger GPT-3 (6B) model trained using CoT reasoning. Finally, we also demonstrate that our approach of problem decomposition can also be used as an alternative to CoT prompting, which boosts the GPT-3 performance by 40% compared to CoT prompts.

Prompting large language models has enabled significant recent progress in multi-step reasoning over text. However, when applied to text generation from semi-structured data (e.g., graphs or tables), these methods typically suffer from low semantic coverage, hallucination, and logical inconsistency. We propose MURMUR, a neuro-symbolic modular approach to text generation from semi-structured data with multi-step reasoning. MURMUR is a best-first search method that generates reasoning paths using: (1) neural and symbolic modules with specific linguistic and logical skills, (2) a grammar whose production rules define valid compositions of modules, and (3) value functions that assess the quality of each reasoning step. We conduct experiments on two diverse data-to-text generation tasks like WebNLG and LogicNLG. These tasks differ in their data representations (graphs and tables) and span multiple linguistic and logical skills. MURMUR obtains significant improvements over recent few-shot baselines like direct prompting and chain-of-thought prompting, while also achieving comparable performance to fine-tuned GPT-2 on out-of-domain data. Moreover, human evaluation shows that MURMUR generates highly faithful and correct reasoning paths that lead to 26% more logically consistent summaries on LogicNLG, compared to direct prompting.

从头开始解决复杂问题通常是有挑战性的，但如果我们可以访问其解决方案的其他类似问题，则更容易 - 一种称为基于案例的推理（CBR）的范式。我们提出了一种神经象征性的CBR方法（CBR-KBQA），用于在大知识库上应答。 CBR-KBQA由非参数内存组成，该内存存储案例（问题和逻辑表单）和参数模型，该参数模型可以通过检索与其相关的案例来为新问题生成逻辑表单。在包含复杂问题的几个KBQA数据集上，CBR-KBQA实现了竞争性能。例如，在ComplexWebQuestions数据集上，CBR-KBQA以11 \％的准确度优于当前最新状态。此外，我们表明CBR-KBQA能够使用新案例\ EMPH {没有}任何进一步的培训：通过在案例存储器中纳入一些人类标记的示例，CBR-KBQA能够成功地生成包含未经看线KB实体的逻辑表格以及关系。