Jian-Guang Lou

Xiaohan Xu, Chongyang Tao, Tao Shen et al. · microsoft-research

To enhance the reasoning capabilities of off-the-shelf Large Language Models (LLMs), we introduce a simple, yet general and effective prompting method, Re2, i.e., \textbf{Re}-\textbf{Re}ading the question as input. Unlike most thought-eliciting prompting methods, such as Chain-of-Thought (CoT), which aim to elicit the reasoning process in the output, Re2 shifts the focus to the input by processing questions twice, thereby enhancing the understanding process. Consequently, Re2 demonstrates strong generality and compatibility with most thought-eliciting prompting methods, including CoT. Crucially, Re2 facilitates a "bidirectional" encoding in unidirectional decoder-only LLMs because the first pass could provide global information for the second pass. We begin with a preliminary empirical study as the foundation of Re2, illustrating its potential to enable "bidirectional" attention mechanisms. We then evaluate Re2 on extensive reasoning benchmarks across 14 datasets, spanning 112 experiments, to validate its effectiveness and generality. Our findings indicate that, with the exception of a few scenarios on vanilla ChatGPT, Re2 consistently enhances the reasoning performance of LLMs through a simple re-reading strategy. Further analyses reveal Re2's adaptability, showing how it can be effectively integrated with different LLMs, thought-eliciting prompting, and ensemble strategies. Our code is available at \url{https://github.com/Tebmer/Rereading-LLM-Reasoning/}

Fengji Zhang, Bei Chen, Yue Zhang et al.

The task of repository-level code completion is to continue writing the unfinished code based on a broader context of the repository. While for automated code completion tools, it is difficult to utilize the useful information scattered in different files. We propose RepoCoder, a simple, generic, and effective framework to address the challenge. It streamlines the repository-level code completion process by incorporating a similarity-based retriever and a pre-trained code language model in an iterative retrieval-generation pipeline. RepoCoder makes effective utilization of repository-level information for code completion and has the ability to generate code at various levels of granularity. Moreover, we propose a new benchmark RepoEval, which consists of the latest and high-quality real-world repositories covering line, API invocation, and function body completion scenarios. Experimental results indicate that RepoCoder significantly improves the In-File completion baseline by over 10% in all settings and consistently outperforms the vanilla retrieval-augmented code completion approach. Furthermore, we validate the effectiveness of RepoCoder through comprehensive analysis, providing valuable insights for future research. Our source code and benchmark are publicly available: https://github.com/microsoft/CodeT/tree/main/RepoCoder

Daoguang Zan, Bei Chen, Dejian Yang et al.

Code generation is a longstanding challenge, aiming to generate a code snippet based on a natural language description. Usually, expensive text-code paired data is essential for training a code generation model. Recently, thanks to the success of pre-training techniques, large language models are trained on large-scale unlabelled code corpora and perform well in code generation. In this paper, we investigate how to leverage an unlabelled code corpus to train a model for library-oriented code generation. Since it is a common practice for programmers to reuse third-party libraries, in which case the text-code paired data are harder to obtain due to the huge number of libraries. We observe that library-oriented code snippets are more likely to share similar code sketches. Hence, we present CERT with two steps: a sketcher generates the sketch, then a generator fills the details in the sketch. Both the sketcher and the generator are continually pre-trained upon a base model using unlabelled data. Furthermore, we craft two benchmarks named PandasEval and NumpyEval to evaluate library-oriented code generation. Experimental results demonstrate the impressive performance of CERT. For example, it surpasses the base model by an absolute 15.67% improvement in terms of pass@1 on PandasEval. Our work is available at https://github.com/microsoft/PyCodeGPT.

Zhaoyun Jiang, Jiaqi Guo, Shizhao Sun et al. · microsoft-research

Conditional graphic layout generation, which generates realistic layouts according to user constraints, is a challenging task that has not been well-studied yet. First, there is limited discussion about how to handle diverse user constraints flexibly and uniformly. Second, to make the layouts conform to user constraints, existing work often sacrifices generation quality significantly. In this work, we propose LayoutFormer++ to tackle the above problems. First, to flexibly handle diverse constraints, we propose a constraint serialization scheme, which represents different user constraints as sequences of tokens with a predefined format. Then, we formulate conditional layout generation as a sequence-to-sequence transformation, and leverage encoder-decoder framework with Transformer as the basic architecture. Furthermore, to make the layout better meet user requirements without harming quality, we propose a decoding space restriction strategy. Specifically, we prune the predicted distribution by ignoring the options that definitely violate user constraints and likely result in low-quality layouts, and make the model samples from the restricted distribution. Experiments demonstrate that LayoutFormer++ outperforms existing approaches on all the tasks in terms of both better generation quality and less constraint violation.

Shengnan An, Zexiong Ma, Zeqi Lin et al.

Large language models (LLMs) recently exhibited remarkable reasoning capabilities on solving math problems. To further improve their reasoning capabilities, this work explores whether LLMs can LEarn from MistAkes (LEMA), akin to the human learning process. Consider a human student who failed to solve a math problem, he will learn from what mistake he has made and how to correct it. Mimicking this error-driven learning process, LEMA incorporates mistake-correction data pairs during fine-tuning LLMs. Specifically, we first collect inaccurate reasoning paths from various LLMs, and then employ GPT-4 as a ''corrector'' to identify the mistake step, explain the reason for the mistake, correct the mistake and generate the final answer. In addition, we apply a correction-centric evolution strategy that effectively expands the question set for generating correction data. Experiments across various LLMs and reasoning tasks show that LEMA effectively improves CoT-alone fine-tuning. Our further ablations shed light on the non-homogeneous effectiveness between CoT data and correction data. These results suggest a significant potential for LLMs to improve through learning from their mistakes. Our code, models and prompts are publicly available at https://github.com/microsoft/LEMA.

Jiawei Lin, Jiaqi Guo, Shizhao Sun et al.

Conditional graphic layout generation, which automatically maps user constraints to high-quality layouts, has attracted widespread attention today. Although recent works have achieved promising performance, the lack of versatility and data efficiency hinders their practical applications. In this work, we propose LayoutPrompter, which leverages large language models (LLMs) to address the above problems through in-context learning. LayoutPrompter is made up of three key components, namely input-output serialization, dynamic exemplar selection and layout ranking. Specifically, the input-output serialization component meticulously designs the input and output formats for each layout generation task. Dynamic exemplar selection is responsible for selecting the most helpful prompting exemplars for a given input. And a layout ranker is used to pick the highest quality layout from multiple outputs of LLMs. We conduct experiments on all existing layout generation tasks using four public datasets. Despite the simplicity of our approach, experimental results show that LayoutPrompter can compete with or even outperform state-of-the-art approaches on these tasks without any model training or fine-tuning. This demonstrates the effectiveness of this versatile and training-free approach. In addition, the ablation studies show that LayoutPrompter is significantly superior to the training-based baseline in a low-data regime, further indicating the data efficiency of LayoutPrompter. Our project is available at https://github.com/microsoft/LayoutGeneration/tree/main/LayoutPrompter.

Xinyu Pi, Bing Wang, Yan Gao et al.

The robustness of Text-to-SQL parsers against adversarial perturbations plays a crucial role in delivering highly reliable applications. Previous studies along this line primarily focused on perturbations in the natural language question side, neglecting the variability of tables. Motivated by this, we propose the Adversarial Table Perturbation (ATP) as a new attacking paradigm to measure the robustness of Text-to-SQL models. Following this proposition, we curate ADVETA, the first robustness evaluation benchmark featuring natural and realistic ATPs. All tested state-of-the-art models experience dramatic performance drops on ADVETA, revealing models' vulnerability in real-world practices. To defend against ATP, we build a systematic adversarial training example generation framework tailored for better contextualization of tabular data. Experiments show that our approach not only brings the best robustness improvement against table-side perturbations but also substantially empowers models against NL-side perturbations. We release our benchmark and code at: https://github.com/microsoft/ContextualSP.

Jiawei Lin, Jiaqi Guo, Shizhao Sun et al. · deepmind

Creating layouts is a fundamental step in graphic design. In this work, we propose to use text as the guidance to create graphic layouts, i.e., Text-to-Layout, aiming to lower the design barriers. Text-to-Layout is a challenging task, because it needs to consider the implicit, combined, and incomplete layout constraints from text, each of which has not been studied in previous work. To address this, we present a two-stage approach, named parse-then-place. The approach introduces an intermediate representation (IR) between text and layout to represent diverse layout constraints. With IR, Text-to-Layout is decomposed into a parse stage and a place stage. The parse stage takes a textual description as input and generates an IR, in which the implicit constraints from the text are transformed into explicit ones. The place stage generates layouts based on the IR. To model combined and incomplete constraints, we use a Transformer-based layout generation model and carefully design a way to represent constraints and layouts as sequences. Besides, we adopt the pretrain-then-finetune strategy to boost the performance of the layout generation model with large-scale unlabeled layouts. To evaluate our approach, we construct two Text-to-Layout datasets and conduct experiments on them. Quantitative results, qualitative analysis, and user studies demonstrate the effectiveness of our approach.

Shengnan An, Yifei Li, Zeqi Lin et al. · pku

Recently the prompt-tuning paradigm has attracted significant attention. By only tuning continuous prompts with a frozen pre-trained language model (PLM), prompt-tuning takes a step towards deploying a shared frozen PLM to serve numerous downstream tasks. Although prompt-tuning shows good performance on certain natural language understanding (NLU) tasks, its effectiveness on natural language generation (NLG) tasks is still under-explored. In this paper, we argue that one of the factors hindering the development of prompt-tuning on NLG tasks is the unfamiliar inputs (i.e., inputs are linguistically different from the pretraining corpus). For example, our preliminary exploration reveals a large performance gap between prompt-tuning and fine-tuning when unfamiliar inputs occur frequently in NLG tasks. This motivates us to propose input-tuning, which fine-tunes both the continuous prompts and the input representations, leading to a more effective way to adapt unfamiliar inputs to frozen PLMs. Our proposed input-tuning is conceptually simple and empirically powerful. Experimental results on seven NLG tasks demonstrate that input-tuning is significantly and consistently better than prompt-tuning. Furthermore, on three of these tasks, input-tuning can achieve a comparable or even better performance than fine-tuning.

Bei Chen, Fengji Zhang, Anh Nguyen et al.

The task of generating code solutions for a given programming problem can benefit from the use of pre-trained language models such as Codex, which can produce multiple diverse samples. However, a major challenge for this task is to select the most appropriate solution from the multiple samples generated by the pre-trained language models. A natural way to evaluate the quality and correctness of a code solution is to run it against a set of test cases, but the manual creation of such test cases is often costly and time-consuming. In this paper, we propose a novel method, CodeT, that leverages the same pre-trained language models to automatically generate test cases for the code samples, thus reducing the human effort and increasing the coverage of the test scenarios. CodeT then executes the code samples using the generated test cases, and performs a dual execution agreement, which considers both the consistency of the outputs against the generated test cases and the agreement of the outputs with other code samples. We conduct comprehensive experiments on four benchmarks, HumanEval, MBPP, APPS and CodeContests, using five different pre-trained language models with varying sizes and capabilities. Our results show that CodeT can significantly improve the performance of code solution selection over previous methods, achieving remarkable and consistent gains across different models and benchmarks. For instance, CodeT improves the pass@1 metric on HumanEval to 65.8%, which represents an absolute improvement of 18.8% over the code-davinci-002 model, and an absolute improvement of more than 20% over the previous state-of-the-art results.

Bing Wang, Yan Gao, Zhoujun Li et al.

The task of text-to-SQL aims to convert a natural language question into its corresponding SQL query within the context of relational tables. Existing text-to-SQL parsers generate a "plausible" SQL query for an arbitrary user question, thereby failing to correctly handle problematic user questions. To formalize this problem, we conduct a preliminary study on the observed ambiguous and unanswerable cases in text-to-SQL and summarize them into 6 feature categories. Correspondingly, we identify the causes behind each category and propose requirements for handling ambiguous and unanswerable questions. Following this study, we propose a simple yet effective counterfactual example generation approach that automatically produces ambiguous and unanswerable text-to-SQL examples. Furthermore, we propose a weakly supervised DTE (Detecting-Then-Explaining) model for error detection, localization, and explanation. Experimental results show that our model achieves the best result on both real-world examples and generated examples compared with various baselines. We release our data and code at: \href{https://github.com/wbbeyourself/DTE}{https://github.com/wbbeyourself/DTE}.

Yifei Li, Zeqi Lin, Shizhuo Zhang et al.

Few-shot learning is a challenging task that requires language models to generalize from limited examples. Large language models like GPT-3 and PaLM have made impressive progress in this area, but they still face difficulties in reasoning tasks such as GSM8K, a benchmark for arithmetic problems. To improve their reasoning skills, previous work has proposed to guide the language model with prompts that elicit a series of reasoning steps before giving the final answer, achieving a significant improvement on GSM8K from 17.9% to 58.1% in problem-solving rate. In this paper, we present DIVERSE (Diverse Verifier on Reasoning Step), a novel approach that further enhances the reasoning capability of language models. DIVERSE has three main components: first, it generates diverse prompts to explore different reasoning paths for the same question; second, it uses a verifier to filter out incorrect answers based on a weighted voting scheme; and third, it verifies each reasoning step individually instead of the whole chain. We evaluate DIVERSE on the latest language model code-davinci-002 and show that it achieves new state-of-the-art results on six of eight reasoning benchmarks (e.g., GSM8K 74.4% to 83.2%).

Daoguang Zan, Bei Chen, Fengji Zhang et al.

The task of generating code from a natural language description, or NL2Code, is considered a pressing and significant challenge in code intelligence. Thanks to the rapid development of pre-training techniques, surging large language models are being proposed for code, sparking the advances in NL2Code. To facilitate further research and applications in this field, in this paper, we present a comprehensive survey of 27 existing large language models for NL2Code, and also review benchmarks and metrics. We provide an intuitive comparison of all existing models on the HumanEval benchmark. Through in-depth observation and analysis, we provide some insights and conclude that the key factors contributing to the success of large language models for NL2Code are "Large Size, Premium Data, Expert Tuning". In addition, we discuss challenges and opportunities regarding the gap between models and humans. We also create a website https://nl2code.github.io to track the latest progress through crowd-sourcing. To the best of our knowledge, this is the first survey of large language models for NL2Code, and we believe it will contribute to the ongoing development of the field.

Libo Qin, Qiguang Chen, Tianbao Xie et al.

Due to high data demands of current methods, attention to zero-shot cross-lingual spoken language understanding (SLU) has grown, as such approaches greatly reduce human annotation effort. However, existing models solely rely on shared parameters, which can only perform implicit alignment across languages. We present Global--Local Contrastive Learning Framework (GL-CLeF) to address this shortcoming. Specifically, we employ contrastive learning, leveraging bilingual dictionaries to construct multilingual views of the same utterance, then encourage their representations to be more similar than negative example pairs, which achieves to explicitly aligned representations of similar sentences across languages. In addition, a key step in GL-CLeF is a proposed Local and Global component, which achieves a fine-grained cross-lingual transfer (i.e., sentence-level Local intent transfer, token-level Local slot transfer, and semantic-level Global transfer across intent and slot). Experiments on MultiATIS++ show that GL-CLeF achieves the best performance and successfully pulls representations of similar sentences across languages closer.

Junyi Zhang, Jiaqi Guo, Shizhao Sun et al.

Creating graphic layouts is a fundamental step in graphic designs. In this work, we present a novel generative model named LayoutDiffusion for automatic layout generation. As layout is typically represented as a sequence of discrete tokens, LayoutDiffusion models layout generation as a discrete denoising diffusion process. It learns to reverse a mild forward process, in which layouts become increasingly chaotic with the growth of forward steps and layouts in the neighboring steps do not differ too much. Designing such a mild forward process is however very challenging as layout has both categorical attributes and ordinal attributes. To tackle the challenge, we summarize three critical factors for achieving a mild forward process for the layout, i.e., legality, coordinate proximity and type disruption. Based on the factors, we propose a block-wise transition matrix coupled with a piece-wise linear noise schedule. Experiments on RICO and PubLayNet datasets show that LayoutDiffusion outperforms state-of-the-art approaches significantly. Moreover, it enables two conditional layout generation tasks in a plug-and-play manner without re-training and achieves better performance than existing methods.

Longxu Dou, Yan Gao, Xuqi Liu et al.

In this paper, we study the problem of knowledge-intensive text-to-SQL, in which domain knowledge is necessary to parse expert questions into SQL queries over domain-specific tables. We formalize this scenario by building a new Chinese benchmark KnowSQL consisting of domain-specific questions covering various domains. We then address this problem by presenting formulaic knowledge, rather than by annotating additional data examples. More concretely, we construct a formulaic knowledge bank as a domain knowledge base and propose a framework (ReGrouP) to leverage this formulaic knowledge during parsing. Experiments using ReGrouP demonstrate a significant 28.2% improvement overall on KnowSQL.

Longxu Dou, Yan Gao, Mingyang Pan et al.

Existing text-to-SQL semantic parsers are typically designed for particular settings such as handling queries that span multiple tables, domains or turns which makes them ineffective when applied to different settings. We present UniSAr (Unified Structure-Aware Autoregressive Language Model), which benefits from directly using an off-the-shelf language model architecture and demonstrates consistently high performance under different settings. Specifically, UniSAr extends existing autoregressive language models to incorporate three non-invasive extensions to make them structure-aware: (1) adding structure mark to encode database schema, conversation context, and their relationships; (2) constrained decoding to decode well structured SQL for a given database schema; and (3) SQL completion to complete potential missing JOIN relationships in SQL based on database schema. On seven well-known text-to-SQL datasets covering multi-domain, multi-table and multi-turn, UniSAr demonstrates highly comparable or better performance to the most advanced specifically-designed text-to-SQL models. Importantly, our UniSAr is non-invasive, such that other core model advances in text-to-SQL can also adopt our extensions to further enhance performance.

Longxu Dou, Yan Gao, Mingyang Pan et al.

Text-to-SQL semantic parsing is an important NLP task, which greatly facilitates the interaction between users and the database and becomes the key component in many human-computer interaction systems. Much recent progress in text-to-SQL has been driven by large-scale datasets, but most of them are centered on English. In this work, we present MultiSpider, the largest multilingual text-to-SQL dataset which covers seven languages (English, German, French, Spanish, Japanese, Chinese, and Vietnamese). Upon MultiSpider, we further identify the lexical and structural challenges of text-to-SQL (caused by specific language properties and dialect sayings) and their intensity across different languages. Experimental results under three typical settings (zero-shot, monolingual and multilingual) reveal a 6.1% absolute drop in accuracy in non-English languages. Qualitative and quantitative analyses are conducted to understand the reason for the performance drop of each language. Besides the dataset, we also propose a simple schema augmentation framework SAVe (Schema-Augmentation-with-Verification), which significantly boosts the overall performance by about 1.8% and closes the 29.5% performance gap across languages.

Xinyu Pi, Wanjun Zhong, Yan Gao et al.

We present LogiGAN, an unsupervised adversarial pre-training framework for improving logical reasoning abilities of language models. Upon automatic identifying logical reasoning phenomena in massive text corpus via detection heuristics, we train language models to predict the masked-out logical statements. Inspired by the facilitation effect of reflective thinking in human learning, we analogically simulate the learning-thinking process with an adversarial Generator-Verifier architecture to assist logic learning. LogiGAN implements a novel sequential GAN approach that (a) circumvents the non-differentiable challenge of the sequential GAN by leveraging the Generator as a sentence-level generative likelihood scorer with a learning objective of reaching scoring consensus with the Verifier; (b) is computationally feasible for large-scale pre-training with arbitrary target length. Both base and large size language models pre-trained with LogiGAN demonstrate obvious performance improvement on 12 datasets requiring general reasoning abilities, revealing the fundamental role of logic in broad reasoning, as well as the effectiveness of LogiGAN. Ablation studies on LogiGAN components reveal the relative orthogonality between linguistic and logic abilities and suggest that reflective thinking's facilitation effect might also generalize to machine learning.

Daoguang Zan, Bei Chen, Zeqi Lin et al.

With the rapid development of pre-training techniques, a number of language models have been pre-trained on large-scale code corpora and perform well in code generation. In this paper, we investigate how to equip pre-trained language models with the ability of code generation for private libraries. In practice, it is common for programmers to write code using private libraries. However, this is a challenge for language models since they have never seen private APIs during training. Motivated by the fact that private libraries usually come with elaborate API documentation, we propose a novel framework with two modules: the APIRetriever finds useful APIs, and then the APICoder generates code using these APIs. For APIRetriever, we present a dense retrieval system and also design a friendly interaction to involve uses. For APICoder, we can directly use off-the-shelf language models, or continually pre-train the base model on a code corpus containing API information. Both modules are trained with data from public libraries and can be generalized to private ones. Furthermore, we craft three benchmarks for private libraries, named TorchDataEval, MonkeyEval, and BeatNumEval. Experimental results demonstrate the impressive performance of our framework.

Yucheng Zhou, Xiubo Geng, Tao Shen et al.

Large Language Models (LLMs) have ushered in a transformative era in the field of natural language processing, excelling in tasks related to text comprehension and generation. Nevertheless, they encounter difficulties when confronted with chaotic contexts (e.g., distractors rather than long irrelevant context), leading to the inadvertent omission of certain details within the chaotic context. In response to these challenges, we introduce the "Thread of Thought" (ThoT) strategy, which draws inspiration from human cognitive processes. ThoT systematically segments and analyzes extended contexts while adeptly selecting pertinent information. This strategy serves as a versatile "plug-and-play" module, seamlessly integrating with various LLMs and prompting techniques. In the experiments, we utilize the PopQA and EntityQ datasets, as well as a Multi-Turn Conversation Response dataset (MTCR) we collected, to illustrate that ThoT significantly improves reasoning performance compared to other prompting techniques.

Zhi Chen, Lu Chen, Bei Chen et al.

With the development of pre-trained language models, remarkable success has been witnessed in dialogue understanding (DU). However, current DU approaches usually employ independent models for each distinct DU task without considering shared knowledge across different DU tasks. In this paper, we propose a unified generative dialogue understanding framework, named {\em UniDU}, to achieve effective information exchange across diverse DU tasks. Here, we reformulate all DU tasks into a unified prompt-based generative model paradigm. More importantly, a novel model-agnostic multi-task training strategy (MATS) is introduced to dynamically adapt the weights of diverse tasks for best knowledge sharing during training, based on the nature and available data of each task. Experiments on ten DU datasets covering five fundamental DU tasks show that the proposed UniDU framework largely outperforms task-specific well-designed methods on all tasks. MATS also reveals the knowledge-sharing structure of these tasks. Finally, UniDU obtains promising performance in the unseen dialogue domain, showing the great potential for generalization.

Yuyan Chen, Qiang Fu, Ge Fan et al.

Recent years, Pre-trained Language models (PLMs) have swept into various fields of artificial intelligence and achieved great success. However, most PLMs, such as T5 and GPT3, have a huge amount of parameters, fine-tuning them is often expensive and time consuming, and storing them takes up a lot of space. Therefore, it is necessary to adopt a parameter-efficient approach to reduce parameters of PLMs in fine-tuning without compromising their performance in downstream tasks. In this paper, we design a novel adapter which only acts on self-attention outputs in PLMs. This adapter adopts element-wise linear transformation using Hadamard product, hence named as Hadamard adapter, requires the fewest parameters compared to previous parameter-efficient adapters. In addition, we also summarize some tuning patterns for Hadamard adapter shared by various downstream tasks, expecting to provide some guidance for further parameter reduction with shared adapters in future studies. The experiments conducted on the widely-used GLUE benchmark with several SOTA PLMs prove that the Hadamard adapter achieves competitive performance with only 0.033\% parameters compared with full fine-tuning, and it has the fewest parameters compared with other adapters. Moreover, we further find that there is also some redundant layers in the Hadamard adapter which can be removed to achieve more parameter efficiency with only 0.022\% parameters.

Zhi Chen, Jijia Bao, Lu Chen et al.

Building a universal conversational agent has been a long-standing goal of the dialogue research community. Most previous works only focus on a small set of dialogue tasks. In this work, we aim to build a unified dialogue foundation model (DFM) which can be used to solve massive diverse dialogue tasks. To achieve this goal, a large-scale well-annotated dialogue dataset with rich task diversity (DialogZoo) is collected. We introduce a framework to unify all dialogue tasks and propose novel auxiliary self-supervised tasks to achieve stable training of DFM on the highly diverse large scale DialogZoo corpus. Experiments show that, compared with models of the same size, DFM can achieve state-of-the-art or competitive performance on very rich cross-domain downstream dialogue tasks. This demonstrates that DFM largely extends the ability of unified dialogue pre-trained model.

Shengnan An, Zeqi Lin, Bei Chen et al.

Abstraction is a desirable capability for deep learning models, which means to induce abstract concepts from concrete instances and flexibly apply them beyond the learning context. At the same time, there is a lack of clear understanding about both the presence and further characteristics of this capability in deep learning models. In this paper, we introduce a systematic probing framework to explore the abstraction capability of deep learning models from a transferability perspective. A set of controlled experiments are conducted based on this framework, providing strong evidence that two probed pre-trained language models (PLMs), T5 and GPT2, have the abstraction capability. We also conduct in-depth analysis, thus shedding further light: (1) the whole training phase exhibits a "memorize-then-abstract" two-stage process; (2) the learned abstract concepts are gathered in a few middle-layer attention heads, rather than being evenly distributed throughout the model; (3) the probed abstraction capabilities exhibit robustness against concept mutations, and are more robust to low-level/source-side mutations than high-level/target-side ones; (4) generic pre-training is critical to the emergence of abstraction capability, and PLMs exhibit better abstraction with larger model sizes and data scales.

Shengnan An, Zexiong Ma, Zeqi Lin et al.

While many contemporary large language models (LLMs) can process lengthy input, they still struggle to fully utilize information within the long context, known as the lost-in-the-middle challenge. We hypothesize that it stems from insufficient explicit supervision during the long-context training, which fails to emphasize that any position in a long context can hold crucial information. Based on this intuition, our study presents information-intensive (IN2) training, a purely data-driven solution to overcome lost-in-the-middle. Specifically, IN2 training leverages a synthesized long-context question-answer dataset, where the answer requires (1) fine-grained information awareness on a short segment (~128 tokens) within a synthesized long context (4K-32K tokens), and (2) the integration and reasoning of information from two or more short segments. Through applying this information-intensive training on Mistral-7B, we present FILM-7B (FILl-in-the-Middle). To thoroughly assess the ability of FILM-7B for utilizing long contexts, we design three probing tasks that encompass various context styles (document, code, and structured-data context) and information retrieval patterns (forward, backward, and bi-directional retrieval). The probing results demonstrate that FILM-7B can robustly retrieve information from different positions in its 32K context window. Beyond these probing tasks, FILM-7B significantly improves the performance on real-world long-context tasks (e.g., 23.5->26.9 F1 score on NarrativeQA), while maintaining a comparable performance on short-context tasks (e.g., 59.3->59.2 accuracy on MMLU). Github Link: https://github.com/microsoft/FILM.

Lu Wang, Fangkai Yang, Chaoyun Zhang et al.

As AI continues to advance, there is a growing demand for systems that go beyond language-based assistance and move toward intelligent agents capable of performing real-world actions. This evolution requires the transition from traditional Large Language Models (LLMs), which excel at generating textual responses, to Large Action Models (LAMs), designed for action generation and execution within dynamic environments. Enabled by agent systems, LAMs hold the potential to transform AI from passive language understanding to active task completion, marking a significant milestone in the progression toward artificial general intelligence. In this paper, we present a comprehensive framework for developing LAMs, offering a systematic approach to their creation, from inception to deployment. We begin with an overview of LAMs, highlighting their unique characteristics and delineating their differences from LLMs. Using a Windows OS-based agent as a case study, we provide a detailed, step-by-step guide on the key stages of LAM development, including data collection, model training, environment integration, grounding, and evaluation. This generalizable workflow can serve as a blueprint for creating functional LAMs in various application domains. We conclude by identifying the current limitations of LAMs and discussing directions for future research and industrial deployment, emphasizing the challenges and opportunities that lie ahead in realizing the full potential of LAMs in real-world applications. The code for the data collection process utilized in this paper is publicly available at: https://github.com/microsoft/UFO/tree/main/dataflow, and comprehensive documentation can be found at https://microsoft.github.io/UFO/dataflow/overview/.

Junren Li, Lei Fang, Jian-Guang Lou

Computer-assisted methods have emerged as valuable tools for retrosynthesis analysis. However, quantifying the plausibility of generated retrosynthesis routes remains a challenging task. We introduce Retro-BLEU, a statistical metric adapted from the well-established BLEU score in machine translation, to evaluate the plausibility of retrosynthesis routes based on reaction template sequences analysis. We demonstrate the effectiveness of Retro-BLEU by applying it to a diverse set of retrosynthesis routes generated by state-of-the-art algorithms and compare the performance with other evaluation metrics. The results show that Retro-BLEU is capable of differentiating between plausible and implausible routes. Furthermore, we provide insights into the strengths and weaknesses of Retro-BLEU, paving the way for future developments and improvements in this field.

Xinyu Zhu, Cheng Yang, Bei Chen et al.

Question answering plays a pivotal role in human daily life because it involves our acquisition of knowledge about the world. However, due to the dynamic and ever-changing nature of real-world facts, the answer can be completely different when the time constraint in the question changes. Recently, Large Language Models (LLMs) have shown remarkable intelligence in question answering, while our experiments reveal that the aforementioned problems still pose a significant challenge to existing LLMs. This can be attributed to the LLMs' inability to perform rigorous reasoning based on surface-level text semantics. To overcome this limitation, rather than requiring LLMs to directly answer the question, we propose a novel approach where we reframe the $\textbf{Q}$uestion $\textbf{A}$nswering task $\textbf{a}$s $\textbf{P}$rogramming ($\textbf{QAaP}$). Concretely, by leveraging modern LLMs' superior capability in understanding both natural language and programming language, we endeavor to harness LLMs to represent diversely expressed text as well-structured code and select the best matching answer from multiple candidates through programming. We evaluate our QAaP framework on several time-sensitive question answering datasets and achieve decent improvement, up to $14.5$% over strong baselines. Our codes and data are available at https://github.com/TianHongZXY/qaap

Qian Liu, Bei Chen, Jiaqi Guo et al.

Recent progress in language model pre-training has achieved a great success via leveraging large-scale unstructured textual data. However, it is still a challenge to apply pre-training on structured tabular data due to the absence of large-scale high-quality tabular data. In this paper, we propose TAPEX to show that table pre-training can be achieved by learning a neural SQL executor over a synthetic corpus, which is obtained by automatically synthesizing executable SQL queries and their execution outputs. TAPEX addresses the data scarcity challenge via guiding the language model to mimic a SQL executor on the diverse, large-scale and high-quality synthetic corpus. We evaluate TAPEX on four benchmark datasets. Experimental results demonstrate that TAPEX outperforms previous table pre-training approaches by a large margin and achieves new state-of-the-art results on all of them. This includes the improvements on the weakly-supervised WikiSQL denotation accuracy to 89.5% (+2.3%), the WikiTableQuestions denotation accuracy to 57.5% (+4.8%), the SQA denotation accuracy to 74.5% (+3.5%), and the TabFact accuracy to 84.2% (+3.2%). To our knowledge, this is the first work to exploit table pre-training via synthetic executable programs and to achieve new state-of-the-art results on various downstream tasks. Our code can be found at https://github.com/microsoft/Table-Pretraining.

Chenyao Liu, Shengnan An, Zeqi Lin et al.

Neural sequence models exhibit limited compositional generalization ability in semantic parsing tasks. Compositional generalization requires algebraic recombination, i.e., dynamically recombining structured expressions in a recursive manner. However, most previous studies mainly concentrate on recombining lexical units, which is an important but not sufficient part of algebraic recombination. In this paper, we propose LeAR, an end-to-end neural model to learn algebraic recombination for compositional generalization. The key insight is to model the semantic parsing task as a homomorphism between a latent syntactic algebra and a semantic algebra, thus encouraging algebraic recombination. Specifically, we learn two modules jointly: a Composer for producing latent syntax, and an Interpreter for assigning semantic operations. Experiments on two realistic and comprehensive compositional generalization benchmarks demonstrate the effectiveness of our model. The source code is publicly available at https://github.com/microsoft/ContextualSP.

Qian Liu, Bei Chen, Jiaqi Guo et al.

Recently semantic parsing in context has received considerable attention, which is challenging since there are complex contextual phenomena. Previous works verified their proposed methods in limited scenarios, which motivates us to conduct an exploratory study on context modeling methods under real-world semantic parsing in context. We present a grammar-based decoding semantic parser and adapt typical context modeling methods on top of it. We evaluate 13 context modeling methods on two large complex cross-domain datasets, and our best model achieves state-of-the-art performances on both datasets with significant improvements. Furthermore, we summarize the most frequent contextual phenomena, with a fine-grained analysis on representative models, which may shed light on potential research directions. Our code is available at https://github.com/microsoft/ContextualSP.

Shengyue Guan, Haoyi Xiong, Jindong Wang et al.

This survey examines evaluation methods for large language model (LLM)-based agents in multi-turn conversational settings. Using a PRISMA-inspired framework, we systematically reviewed nearly 250 scholarly sources, capturing the state of the art from various venues of publication, and establishing a solid foundation for our analysis. Our study offers a structured approach by developing two interrelated taxonomy systems: one that defines \emph{what to evaluate} and another that explains \emph{how to evaluate}. The first taxonomy identifies key components of LLM-based agents for multi-turn conversations and their evaluation dimensions, including task completion, response quality, user experience, memory and context retention, as well as planning and tool integration. These components ensure that the performance of conversational agents is assessed in a holistic and meaningful manner. The second taxonomy system focuses on the evaluation methodologies. It categorizes approaches into annotation-based evaluations, automated metrics, hybrid strategies that combine human assessments with quantitative measures, and self-judging methods utilizing LLMs. This framework not only captures traditional metrics derived from language understanding, such as BLEU and ROUGE scores, but also incorporates advanced techniques that reflect the dynamic, interactive nature of multi-turn dialogues.

Chaoyun Zhang, He Huang, Chiming Ni et al.

Recent Computer-Using Agents (CUAs), powered by multimodal large language models (LLMs), offer a promising direction for automating complex desktop workflows through natural language. However, most existing CUAs remain conceptual prototypes, hindered by shallow OS integration, fragile screenshot-based interaction, and disruptive execution. We present UFO2, a multiagent AgentOS for Windows desktops that elevates CUAs into practical, system-level automation. UFO2 features a centralized HostAgent for task decomposition and coordination, alongside a collection of application-specialized AppAgent equipped with native APIs, domain-specific knowledge, and a unified GUI--API action layer. This architecture enables robust task execution while preserving modularity and extensibility. A hybrid control detection pipeline fuses Windows UI Automation (UIA) with vision-based parsing to support diverse interface styles. Runtime efficiency is further enhanced through speculative multi-action planning, reducing per-step LLM overhead. Finally, a Picture-in-Picture (PiP) interface enables automation within an isolated virtual desktop, allowing agents and users to operate concurrently without interference. We evaluate UFO2 across over 20 real-world Windows applications, demonstrating substantial improvements in robustness and execution accuracy over prior CUAs. Our results show that deep OS integration unlocks a scalable path toward reliable, user-aligned desktop automation.

Yurong Wu, Yan Gao, Bin Benjamin Zhu et al.

Prompt engineering is pivotal for harnessing the capabilities of large language models (LLMs) across diverse applications. While existing prompt optimization methods improve prompt effectiveness, they often lead to prompt drifting, where newly generated prompts can adversely impact previously successful cases while addressing failures. Furthermore, these methods tend to rely heavily on LLMs' intrinsic capabilities for prompt optimization tasks. In this paper, we introduce StraGo (Strategic-Guided Optimization), a novel approach designed to mitigate prompt drifting by leveraging insights from both successful and failed cases to identify critical factors for achieving optimization objectives. StraGo employs a how-to-do methodology, integrating in-context learning to formulate specific, actionable strategies that provide detailed, step-by-step guidance for prompt optimization. Extensive experiments conducted across a range of tasks, including reasoning, natural language understanding, domain-specific knowledge, and industrial applications, demonstrate StraGo's superior performance. It establishes a new state-of-the-art in prompt optimization, showcasing its ability to deliver stable and effective prompt improvements.

Sheng Yang, Yurong Wu, Yan Gao et al.

Prompt engineering is very important to enhance the performance of large language models (LLMs). When dealing with complex issues, prompt engineers tend to distill multiple patterns from examples and inject relevant solutions to optimize the prompts, achieving satisfying results. However, existing automatic prompt optimization techniques are only limited to producing single flow instructions, struggling with handling diverse patterns. In this paper, we present AMPO, an automatic prompt optimization method that can iteratively develop a multi-branched prompt using failure cases as feedback. Our goal is to explore a novel way of structuring prompts with multi-branches to better handle multiple patterns in complex tasks, for which we introduce three modules: Pattern Recognition, Branch Adjustment, and Branch Pruning. In experiments across five tasks, AMPO consistently achieves the best results. Additionally, our approach demonstrates significant optimization efficiency due to our adoption of a minimal search strategy.

Jinyang Li, Nan Huo, Yan Gao et al.

Interactive Data Analysis, the collaboration between humans and LLM agents, enables real-time data exploration for informed decision-making. The challenges and costs of collecting realistic interactive logs for data analysis hinder the quantitative evaluation of Large Language Model (LLM) agents in this task. To mitigate this issue, we introduce Tapilot-Crossing, a new benchmark to evaluate LLM agents on interactive data analysis. Tapilot-Crossing contains 1024 interactions, covering 4 practical scenarios: Normal, Action, Private, and Private Action. Notably, Tapilot-Crossing is constructed by an economical multi-agent environment, Decision Company, with few human efforts. We evaluate popular and advanced LLM agents in Tapilot-Crossing, which underscores the challenges of interactive data analysis. Furthermore, we propose Adaptive Interaction Reflection (AIR), a self-generated reflection strategy that guides LLM agents to learn from successful history. Experiments demonstrate that Air can evolve LLMs into effective interactive data analysis agents, achieving a relative performance improvement of up to 44.5%.

Zexiong Ma, Shengnan An, Zeqi Lin et al.

Large language models (LLMs) are susceptible to generating hallucinated information, despite the integration of retrieval-augmented generation (RAG). Parallel context extension (PCE) is a line of research attempting to effectively integrating parallel (unordered) contexts, while it still suffers from hallucinations when adapted to RAG scenarios. In this paper, we propose DePaC (Dehallucinating Parallel Context Extension), which alleviates the hallucination problem with context-aware negative training and information-calibrated aggregation. DePaC is designed to alleviate two types of in-context hallucination: fact fabrication (i.e., LLMs present claims that are not supported by the contexts) and fact omission (i.e., LLMs fail to present claims that can be supported by the contexts). Specifically, (1) for fact fabrication, we apply the context-aware negative training that fine-tunes the LLMs with negative supervisions, thus explicitly guiding the LLMs to refuse to answer when contexts are not related to questions; (2) for fact omission, we propose the information-calibrated aggregation which prioritizes context windows with higher information increment from their contexts. The experimental results on nine RAG tasks demonstrate that DePaC significantly alleviates the two types of hallucination and consistently achieves better performances on these tasks.

Weihao Zeng, Can Xu, Yingxiu Zhao et al.

Fine-tuning large pre-trained language models with Evol-Instruct has achieved encouraging results across a wide range of tasks. However, designing effective evolving methods for instruction evolution requires substantial human expertise. This paper proposes Auto Evol-Instruct, an end-to-end framework that evolves instruction datasets using large language models without any human effort. The framework automatically analyzes and summarizes suitable evolutionary strategies for the given instruction data and iteratively improves the evolving method based on issues exposed during the instruction evolution process. Our extensive experiments demonstrate that the best method optimized by Auto Evol-Instruct outperforms human-designed methods on various benchmarks, including MT-Bench, AlpacaEval, GSM8K, and HumanEval.

Zhenwen Li, Jian-Guang Lou, Tao Xie

In order to reduce the manual cost of designing ER models, recent approaches have been proposed to address the task of NL2ERM, i.e., automatically generating entity-relationship (ER) models from natural language (NL) utterances such as software requirements. These approaches are typically rule-based ones, which rely on rigid heuristic rules; these approaches cannot generalize well to various linguistic ways of describing the same requirement. Despite having better generalization capability than rule-based approaches, deep-learning-based models are lacking for NL2ERM due to lacking a large-scale dataset. To address this issue, in this paper, we report our insight that there exists a high similarity between the task of NL2ERM and the increasingly popular task of text-to-SQL, and propose a data transformation algorithm that transforms the existing data of text-to-SQL into the data of NL2ERM. We apply our data transformation algorithm on Spider, one of the most popular text-to-SQL datasets, and we also collect some data entries with different NL types, to obtain a large-scale NL2ERM dataset. Because NL2ERM can be seen as a special information extraction (IE) task, we train two state-of-the-art IE models on our dataset. The experimental results show that both the two models achieve high performance and outperform existing baselines.

Yan Liu, Yan Gao, Zhe Su et al.

Content Warning: This work contains examples that potentially implicate stereotypes, associations, and other harms that could be offensive to individuals in certain social groups.} Large pre-trained language models are acknowledged to carry social biases towards different demographics, which can further amplify existing stereotypes in our society and cause even more harm. Text-to-SQL is an important task, models of which are mainly adopted by administrative industries, where unfair decisions may lead to catastrophic consequences. However, existing Text-to-SQL models are trained on clean, neutral datasets, such as Spider and WikiSQL. This, to some extent, cover up social bias in models under ideal conditions, which nevertheless may emerge in real application scenarios. In this work, we aim to uncover and categorize social biases in Text-to-SQL models. We summarize the categories of social biases that may occur in structured data for Text-to-SQL models. We build test benchmarks and reveal that models with similar task accuracy can contain social biases at very different rates. We show how to take advantage of our methodology to uncover and assess social biases in the downstream Text-to-SQL task. We will release our code and data.

Yan Liu, Xiaokang Chen, Yan Gao et al.

With the popularity of automatic code generation tools, such as Copilot, the study of the potential hazards of these tools is gaining importance. In this work, we explore the social bias problem in pre-trained code generation models. We propose a new paradigm to construct code prompts and successfully uncover social biases in code generation models. To quantify the severity of social biases in generated code, we develop a dataset along with three metrics to evaluate the overall social bias and fine-grained unfairness across different demographics. Experimental results on three pre-trained code generation models (Codex, InCoder, and CodeGen) with varying sizes, reveal severe social biases. Moreover, we conduct analysis to provide useful insights for further choice of code generation models with low social bias. (This work contains examples that potentially implicate stereotypes, associations, and other harms that could be offensive to individuals in certain social groups.)

Jian Wu, Yicheng Xu, Yan Gao et al.

Hybrid Question-Answering (HQA), which targets reasoning over tables and passages linked from table cells, has witnessed significant research in recent years. A common challenge in HQA and other passage-table QA datasets is that it is generally unrealistic to iterate over all table rows, columns, and linked passages to retrieve evidence. Such a challenge made it difficult for previous studies to show their reasoning ability in retrieving answers. To bridge this gap, we propose a novel Table-alignment-based Cell-selection and Reasoning model (TACR) for hybrid text and table QA, evaluated on the HybridQA and WikiTableQuestions datasets. In evidence retrieval, we design a table-question-alignment enhanced cell-selection method to retrieve fine-grained evidence. In answer reasoning, we incorporate a QA module that treats the row containing selected cells as context. Experimental results over the HybridQA and WikiTableQuestions (WTQ) datasets show that TACR achieves state-of-the-art results on cell selection and outperforms fine-grained evidence retrieval baselines on HybridQA, while achieving competitive performance on WTQ. We also conducted a detailed analysis to demonstrate that being able to align questions to tables in the cell-selection stage can result in important gains from experiments of over 90\% table row and column selection accuracy, meanwhile also improving output explainability.

Shengnan An, Bo Zhou, Zeqi Lin et al.

In-context learning is the paradigm that adapts large language models to downstream tasks by providing a few examples. Few-shot selection -- selecting appropriate examples for each test instance separately -- is important for in-context learning. In this paper, we propose Skill-KNN, a skill-based few-shot selection method for in-context learning. The key advantages of Skill-KNN include: (1) it addresses the problem that existing methods based on pre-trained embeddings can be easily biased by surface natural language features that are not important for the target task; (2) it does not require training or fine-tuning of any models, making it suitable for frequently expanding or changing example banks. The key insight is to optimize the inputs fed into the embedding model, rather than tuning the model itself. Technically, Skill-KNN generates the skill-based descriptions for each test case and candidate example by utilizing a pre-processing few-shot prompting, thus eliminating unimportant surface features. Experimental results across five cross-domain semantic parsing datasets and six backbone models show that Skill-KNN significantly outperforms existing methods.

Shengnan An, Zeqi Lin, Qiang Fu et al.

Compositional generalization--understanding unseen combinations of seen primitives--is an essential reasoning capability in human intelligence. The AI community mainly studies this capability by fine-tuning neural networks on lots of training samples, while it is still unclear whether and how in-context learning--the prevailing few-shot paradigm based on large language models--exhibits compositional generalization. In this paper, we present CoFe, a test suite to investigate in-context compositional generalization. We find that the compositional generalization performance can be easily affected by the selection of in-context examples, thus raising the research question what the key factors are to make good in-context examples for compositional generalization. We study three potential factors: similarity, diversity and complexity. Our systematic experiments indicate that in-context examples should be structurally similar to the test case, diverse from each other, and individually simple. Furthermore, two strong limitations are observed: in-context compositional generalization on fictional words is much weaker than that on commonly used ones; it is still critical that the in-context examples should cover required linguistic structures, even though the backbone model has been pre-trained on large corpus. We hope our analysis would facilitate the understanding and utilization of in-context learning paradigm.

Xinyu Pi, Qian Liu, Bei Chen et al.

Reasoning over natural language is a long-standing goal for the research community. However, studies have shown that existing language models are inadequate in reasoning. To address the issue, we present POET, a novel reasoning pre-training paradigm. Through pre-training language models with programs and their execution results, POET empowers language models to harvest the reasoning knowledge possessed by program executors via a data-driven approach. POET is conceptually simple and can be instantiated by different kinds of program executors. In this paper, we showcase two simple instances POET-Math and POET-Logic, in addition to a complex instance, POET-SQL. Experimental results on six benchmarks demonstrate that POET can significantly boost model performance in natural language reasoning, such as numerical reasoning, logical reasoning, and multi-hop reasoning. POET opens a new gate on reasoning-enhancement pre-training, and we hope our analysis would shed light on the future research of reasoning like program executors.

Qi Shi, Qian Liu, Bei Chen et al.

Language-based environment manipulation requires agents to manipulate the environment following natural language instructions, which is challenging due to the huge space of the environments. To address this challenge, various approaches have been proposed in recent work. Although these approaches work well for their intended environments, they are difficult to generalize across environments. In this work, we propose LEMON, a general framework for language-based environment manipulation tasks. Specifically, we first specify a task-agnostic approach for language-based environment manipulation tasks, which can deal with various environments using the same generative language model. Then we propose an execution-guided pre-training strategy to inject prior knowledge of environments to the language model with a pure synthetic pre-training corpus. Experimental results on tasks including Alchemy, Scene, Tangrams, ProPara and Recipes demonstrate the effectiveness of LEMON: it achieves new state-of-the-art results on four of the tasks, and the execution-guided pre-training strategy brings remarkable improvements on all experimental tasks.

Lei Fang, Jian-Guang Lou

We study the problem of quantitative facts extraction for text with percentages. For example, given the sentence "30 percent of Americans like watching football, while 20% prefer to watch NBA.", our goal is to obtain a deep understanding of the percentage numbers ("30 percent" and "20%") by extracting their quantitative facts: part ("like watching football" and "prefer to watch NBA") and whole ("Americans). These quantitative facts can empower new applications like automated infographic generation. We formulate part and whole extraction as a sequence tagging problem. Due to the large gap between part/whole and its corresponding percentage, we introduce skip mechanism in sequence modeling, and achieved improved performance on both our task and the CoNLL-2003 named entity recognition task. Experimental results demonstrate that learning to skip in sequence tagging is promising.

Qian Liu, Dejian Yang, Jiahui Zhang et al.

Recent years pretrained language models (PLMs) hit a success on several downstream tasks, showing their power on modeling language. To better understand and leverage what PLMs have learned, several techniques have emerged to explore syntactic structures entailed by PLMs. However, few efforts have been made to explore grounding capabilities of PLMs, which are also essential. In this paper, we highlight the ability of PLMs to discover which token should be grounded to which concept, if combined with our proposed erasing-then-awakening approach. Empirical studies on four datasets demonstrate that our approach can awaken latent grounding which is understandable to human experts, even if it is not exposed to such labels during training. More importantly, our approach shows great potential to benefit downstream semantic parsing models. Taking text-to-SQL as a case study, we successfully couple our approach with two off-the-shelf parsers, obtaining an absolute improvement of up to 9.8%.

Zhoujun Cheng, Haoyu Dong, Zhiruo Wang et al.

Tables are often created with hierarchies, but existing works on table reasoning mainly focus on flat tables and neglect hierarchical tables. Hierarchical tables challenge existing methods by hierarchical indexing, as well as implicit relationships of calculation and semantics. This work presents HiTab, a free and open dataset to study question answering (QA) and natural language generation (NLG) over hierarchical tables. HiTab is a cross-domain dataset constructed from a wealth of statistical reports (analyses) and Wikipedia pages, and has unique characteristics: (1) nearly all tables are hierarchical, and (2) both target sentences for NLG and questions for QA are revised from original, meaningful, and diverse descriptive sentences authored by analysts and professions of reports. (3) to reveal complex numerical reasoning in statistical analyses, we provide fine-grained annotations of entity and quantity alignment. HiTab provides 10,686 QA pairs and descriptive sentences with well-annotated quantity and entity alignment on 3,597 tables with broad coverage of table hierarchies and numerical reasoning types. Targeting hierarchical structure, we devise a novel hierarchy-aware logical form for symbolic reasoning over tables, which shows high effectiveness. Targeting complex numerical reasoning, we propose partially supervised training given annotations of entity and quantity alignment, which helps models to largely reduce spurious predictions in the QA task. In the NLG task, we find that entity and quantity alignment also helps NLG models to generate better results in a conditional generation setting. Experiment results of state-of-the-art baselines suggest that this dataset presents a strong challenge and a valuable benchmark for future research.