Author name cluster

Nanyun Peng

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

26 papers

1 author row

TMLR Journal 2026 Journal Article

From Preferences to Prejudice: The Role of Alignment Tuning in Shaping Social Bias in Video Diffusion Models

Zefan Cai
Haoyi Qiu
Haozhe Zhao
Ke Wan
Jiachen Li
Jiuxiang Gu
Wen Xiao
Nanyun Peng

Recent advances in video diffusion models have significantly enhanced text-to-video generation, particularly through alignment tuning using reward models trained on human preferences. While these methods improve visual quality, they can unintentionally encode and amplify social biases. To systematically trace how such biases evolve throughout the alignment pipeline, we introduce VideoBiasEval, a comprehensive diagnostic framework for evaluating social representation in video generation. Grounded in established social bias taxonomies, VideoBiasEval employs an event-based prompting strategy to disentangle semantic content (verbs and contexts) from actor attributes (gender and ethnicity). It further introduces multi-granular metrics to evaluate (1) overall ethnicity bias, (2) gender bias conditioned on ethnicity, (3) distributional shifts in social attributes across model variants, and (4) the temporal persistence of bias within videos. Using this framework, we conduct the first end-to-end analysis connecting biases in human preference datasets, their amplification in reward models, and their propagation through alignment-tuned video diffusion models. Our results reveal that alignment tuning not only strengthens representational biases but also makes them temporally stable, producing smoother yet more stereotyped portrayals. These findings highlight the need for bias-aware evaluation and mitigation throughout the alignment process to ensure fair and socially responsible video generation.

NeurIPS Conference 2025 Conference Paper

3DLLM-Mem: Long-Term Spatial-Temporal Memory for Embodied 3D Large Language Model

Wenbo Hu
Yining Hong
Yanjun Wang
Leison Gao
Zibu Wei
Xingcheng Yao
Nanyun Peng
Yonatan Bitton

Humans excel at performing complex tasks by leveraging long-term memory across temporal and spatial experiences. In contrast, current Large Language Models (LLMs) struggle to effectively plan and act in dynamic, multi-room 3D environments. We posit that part of this limitation is due to the lack of proper 3D spatial-temporal memory modeling in LLMs. To address this, we first introduce 3DMem-Bench, a comprehensive benchmark comprising over 26, 000 trajectories and 2, 892 embodied tasks, question-answering and captioning, designed to evaluate an agent's ability to reason over long-term memory in 3D environments. Second, we propose 3DLLM-Mem, a novel dynamic memory management and fusion model for embodied spatial-temporal reasoning and actions in LLMs. Our model uses working memory tokens, which represents current observations, as queries to selectively attend to and fuse the most useful spatial and temporal features from episodic memory, which stores past observations and interactions. Our approach allows the agent to focus on task-relevant information while maintaining memory efficiency in complex, long-horizon environments. Experimental results demonstrate that 3DLLM-Mem achieves state-of-the-art performance across various tasks, outperforming the strongest baselines by 16. 5\% in success rate on 3DMem-Bench's most challenging in-the-wild embodied tasks.

NeurIPS Conference 2025 Conference Paper

AI Debate Aids Assessment of Controversial Claims

Salman Rahman
Sheriff Issaka
Ashima Suvarna
Genglin Liu
James Shiffer
Jaeyoung Lee
Md Rizwan Parvez
Hamid Palangi

As AI grows more powerful, it will increasingly shape how we understand the world. But with this influence comes the risk of amplifying misinformation and deepening social divides—especially on consequential topics where factual accuracy directly impacts well-being. Scalable Oversight aims to ensure AI systems remain truthful even when their capabilities exceed those of their evaluators. Yet when humans serve as evaluators, their own beliefs and biases can impair judgment. We study whether AI debate can guide biased judges toward the truth by having two AI systems debate opposing sides of controversial factuality claims on COVID-19 and climate change where people hold strong prior beliefs. We conduct two studies. Study I recruits human judges with either mainstream or skeptical beliefs who evaluate claims through two protocols: debate (interaction with two AI advisors arguing opposing sides) or consultancy (interaction with a single AI advisor). Study II uses AI judges with and without human-like personas to evaluate the same protocols. In Study I, debate consistently improves human judgment accuracy and confidence calibration, outperforming consultancy by 4-10\% across COVID-19 and climate change claims. The improvement is most significant for judges with mainstream beliefs (up to +15. 2\% accuracy on COVID-19 claims), though debate also helps skeptical judges who initially misjudge claims move toward accurate views (+4. 7\% accuracy). In Study II, AI judges with human-like personas achieve even higher accuracy (78. 5\%) than human judges (70. 1\%) and default AI judges without personas (69. 8\%), suggesting their potential for supervising frontier AI models. These findings highlight AI debate as a promising path toward scalable, bias-resilient oversight in contested domains.

TMLR Journal 2025 Journal Article

Multimodal Cultural Safety: Evaluation Framework and Alignment Strategies

Haoyi Qiu
Kung-Hsiang Huang
Ruichen Zheng
Jiao Sun
Nanyun Peng

Large vision-language models (LVLMs) are increasingly deployed in globally distributed applications, such as tourism assistants, yet their ability to produce culturally appropriate responses remains underexplored. Existing multimodal safety benchmarks primarily focus on physical safety and overlook violations rooted in cultural norms, which can result in symbolic harm. For example, suggesting clocks as gifts for a baby’s birthday in China may invoke associations with death, leading to user discomfort and undermining trust. To address this gap, we introduce CROSS, a benchmark designed to assess the cultural safety reasoning capabilities of LVLMs. CROSS includes 1,284 multilingual visually grounded queries from 16 countries, three everyday domains (i.e., shopping, meal planning, and outdoor activities), and 14 languages, where cultural norm violations emerge only when images are interpreted in context. We propose CROSS-Eval, an intercultural theory-based framework that measures four key dimensions: cultural awareness, norm education, compliance, and helpfulness. Using this framework, we evaluate 21 leading LVLMs, including mixture-of-experts models (e.g., Llama-4-Maverick) and reasoning models (e.g., o1 and Gemini-2.5-Pro). Results reveal significant cultural safety gaps: the best-performing model achieves only 61.79% in awareness and 37.73% in compliance. While some open-source models achieve performance better or comparable to GPT-4o, they still fall notably short of proprietary models. Our results further show that increasing reasoning capacity improves cultural alignment but does not fully resolve the issue. To improve model performance, we develop two enhancement strategies: supervised fine-tuning with culturally grounded, open-ended data and preference tuning with contrastive response pairs that highlight safe versus unsafe behaviors. These methods substantially improve GPT-4o’s cultural awareness (+60.14%) and compliance (+55.2%), while preserving general multimodal capabilities with minimal performance reduction on general multimodal understanding benchmarks. This work establishes a framework for evaluating and improving cultural safety in vision-language systems across diverse global contexts.

NeurIPS Conference 2025 Conference Paper

On the Loss of Context Awareness in General Instruction Fine-tuning

Yihan Wang
Andrew Bai
Nanyun Peng
Cho-Jui Hsieh

Pre-trained Large Language Models (LLMs) require post-training methods such as supervised fine-tuning (SFT) on instruction-response pairs to enable instruction following. However, this process can cause forgetting in capabilities learned during pre-training. In this paper, we investigate the loss of context awareness after SFT, where context awareness is defined as the ability to extract and understand information from user-provided context. % and respond accordingly. Surprisingly, we discovered that the loss of context awareness occurs in instruction fine-tuned LLMs when the chat template is applied to input prompts. We identify that the performance decline is associated with a bias toward different roles learned during conversational instruction fine-tuning. The bias can be traced to training samples where the assistant response minimally relies on the user-provided instruction. Based on these observations, we propose a metric to identify context-dependent examples from general instruction fine-tuning datasets. We then apply conditional instruction fine-tuning with a context-dependency indicator, enabling the model to preserve context awareness after SFT. Experiments on four context-dependent downstream tasks and three pre-trained LLMs of different sizes show that our method effectively mitigates the loss of context awareness without compromising general instruction-following capabilities.

NeurIPS Conference 2025 Conference Paper

OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles

Yihe Deng
Hritik Bansal
Fan Yin
Nanyun Peng
Wei Wang
Kai-Wei Chang

We introduce OpenVLThinker, one of the first open-source large vision–language models (LVLMs) to exhibit sophisticated chain-of-thought reasoning, achieving notable performance gains on challenging visual reasoning tasks. While text-based reasoning models (e. g. , Deepseek R1) show promising results in text-only tasks, distilling their reasoning into LVLMs via supervised fine-tuning (SFT) often results in performance degradation due to imprecise visual grounding. Conversely, purely reinforcement learning (RL)-based methods face a large search space, hindering the emergence of reflective behaviors in smaller models (e. g. , 7B LVLMs). Surprisingly, alternating between SFT and RL ultimately results in significant performance improvements after a few iterations. Our analysis reveals that the base model rarely exhibits reasoning behaviors initially, but SFT effectively surfaces these latent actions and narrows the RL search space, accelerating the development of reasoning capabilities. Each subsequent RL stage further refines the model's reasoning skills, producing higher-quality SFT data for continued self-improvement. OpenVLThinker-7B consistently advances performance across six benchmarks demanding mathematical and general reasoning, notably improving MathVista by 3. 2\%, EMMA by 1. 4\%, and HallusionBench by 2. 7\%. Beyond demonstrating the synergy between SFT and RL for complex reasoning tasks, our findings provide early evidence towards achieving R1-style reasoning in multimodal contexts.

NeurIPS Conference 2025 Conference Paper

PARTONOMY: Large Multimodal Models with Part-Level Visual Understanding

Ansel Blume
Jeonghwan Kim
Hyeonjeong Ha
Elen Chatikyan
Xiaomeng Jin
Khanh Nguyen
Nanyun Peng
Kai-Wei Chang

Real-world objects are composed of distinctive, object-specific parts. Identifying these parts is key to performing fine-grained, compositional reasoning—yet, large multimodal models (LMMs) struggle to perform this seemingly straightforward task. In this work, we introduce PARTONOMY, an LMM benchmark designed for pixel-level part grounding. We construct PARTONOMY from existing part datasets and our own rigorously annotated set of images, encompassing 862 parts and 5346 objects for evaluation. Unlike existing datasets that simply ask models to identify generic parts, PARTONOMY utilizes highly technical concepts and challenges models to compare objects’ parts, consider part-whole relationships, and justify textual predictions with visual segmentations. Our experiments demonstrate significant limitations in state-of-the-art LMMs (e. g. , LISA-13B achieves only 5. 9% gIoU), highlighting a critical gap in their part grounding abilities. We note that existing segmentation-enabled LMMs (segmenting LMMs) have two key architectural shortcomings: they use special [SEG] tokens not seen during pretraining which induce distribution shift, and they discard predicted segmentations instead of using past predictions to guide future ones. To address these deficiencies, we train several part-centric LMMs and propose PLUM, a novel segmenting LMM that utilizes span tagging instead of segmentation tokens and that conditions on prior predictions in a feedback loop. We find that pretrained PLUM dominates existing segmenting LMMs on reasoning segmentation, VQA, and visual hallucination benchmarks. In addition, PLUM finetuned on our proposed Explanatory Part Segmentation task is competitive with segmenting LMMs trained on significantly more segmentation data. Our work opens up new avenues towards enabling fine-grained, grounded visual understanding in LMMs.

NeurIPS Conference 2024 Conference Paper

Adaptable Logical Control for Large Language Models

Honghua Zhang
Po-Nien Kung
Masahiro Yoshida
Guy Van den Broeck
Nanyun Peng

Despite the success of Large Language Models (LLMs) on various tasks following human instructions, controlling model generation to follow strict constraints at inference time poses a persistent challenge. In this paper, we introduce Ctrl-G, a neuro-symbolic framework that enables tractable and adaptable control of LLM generation to follow logical constraints reliably. Ctrl-G combines any production-ready LLM with a Hidden Markov Model (HMM), guiding LLM outputs to adhere to logical constraints represented as deterministic finite automata. We show that Ctrl-G, when a TULU2-7B model is coupled with a 2B-parameter HMM, outperforms GPT4 in text editing: on the task of generating text insertions/continuations following logical constraints, our approach achieves over 30% higher satisfaction rate in human evaluation. When applied to medium-size language models (e. g. , GPT2-large), Ctrl-G also beats its counterparts on standard benchmarks by large margins. Additionally, as a proof-of-concept study, we use Ctrl-G to assist LLM reasoning on the GSM benchmark, foreshadowing the application of Ctrl-G, as well as other constrained generation approaches, beyond traditional language generation tasks.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

DACO: Towards Application-Driven and Comprehensive Data Analysis via Code Generation

Xueqing Wu
Rui Zheng
Jingzhen Sha
Te-Lin Wu
Hanyu Zhou
Mohan Tang
Kai-Wei Chang
Nanyun Peng

Data analysis is a crucial analytical process essential for deriving insights from real-world databases. As shown in Figure 1, the need for data analysis typically arises from specific application scenarios, and requires diverse reasoning skills including mathematical reasoning, logical reasoning, and strategic reasoning. Existing work often focus on simple factual retrieval or arithmetic resolutions and thus are insufficient for addressing complex real-world queries. This work aims to propose new resources and benchmarks on this crucial yet challenging and under-explored task. Due to the prohibitively high cost of collecting expert annotations, we use large language models (LLMs) enhanced by code generation to automatically generate high-quality data analysis, which will later be refined by human annotators. We construct the DACO dataset, containing (1) 440 databases (of tabular data) collected from real-world scenarios, (2) ~2k automatically generated query-answer pairs that can serve as weak supervision for model training, and (3) a concentrated but high-quality test set with human refined annotations that serves as our main evaluation benchmark. Experiments show that while LLMs like GPT-4 exhibit promising data analysis capabilities, they are still evaluated as less helpful than human-written analysis on 58. 1% cases. Leveraging our weak supervision data, we experiment with various fine-tuning methods, including supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). Our trained model outperforms existing baselines for table question answering, and RLHF further boosts the helpfulness of generated analysis on 58. 5% cases. Data and code are released at https: //github. com/shirley-wu/daco.

PDF Details DOI

TMLR Journal 2024 Journal Article

FlexEControl: Flexible and Efficient Multimodal Control for Text-to-Image Generation

Xuehai He
Jian Zheng
Jacob Zhiyuan Fang
Robinson Piramuthu
Mohit Bansal
Vicente Ordonez
Gunnar A Sigurdsson
Nanyun Peng

Controllable text-to-image (T2I) diffusion models generate images conditioned on both text prompts and semantic inputs of other modalities like edge maps. Nevertheless, current controllable T2I methods commonly face challenges related to efficiency and faithfulness, especially when conditioning on multiple inputs from either the same or diverse modalities. In this paper, we propose a novel Flexible and Efficient method, FlexEControl, for controllable T2I generation. At the core of FlexEControl is a unique weight decomposition strategy, which allows for streamlined integration of various input types. This approach not only enhances the faithfulness of the generated image to the control, but also significantly reduces the computational overhead typically associated with multimodal conditioning. Our approach achieves a reduction of 41% in trainable parameters and 30% in memory usage compared with Uni-ControlNet. Moreover, it doubles data efficiency and can flexibly generate images under the guidance of multiple input conditions of various modalities.

NeurIPS Conference 2024 Conference Paper

Matryoshka Query Transformer for Large Vision-Language Models

Wenbo Hu
Zi-Yi Dou
Liunian H. Li
Amita Kamath
Nanyun Peng
Kai-Wei Chang

Large Vision-Language Models (LVLMs) typically encode an image into a fixed number of visual tokens (e. g. , 576) and process these tokens with a language model. Despite their strong performance, LVLMs face challenges in adapting to varying computational constraints. This raises the question: can we achieve flexibility in the number of visual tokens to suit different tasks and computational resources? We answer this with an emphatic yes. Inspired by Matryoshka Representation Learning, we introduce the Matryoshka Query Transformer (MQT), capable of encoding an image into $m$ visual tokens during inference, where $m$ can be any number up to a predefined maximum. This is achieved by employing a query transformer with $M$ latent query tokens to compress the visual embeddings. During each training step, we randomly select $m \leq M$ latent query tokens and train the model using only these first $m$ tokens, discarding the rest. Combining MQT with LLaVA, we train a single model once, and flexibly and drastically reduce the number of inference-time visual tokens while maintaining similar or better performance compared to training independent models for each number of tokens. Our model, MQT-LLaVA, matches LLaVA-1. 5 performance across 11 benchmarks using a maximum of 256 tokens instead of LLaVA’s fixed 576. Reducing to 16 tokens (8x less TFLOPs) only sacrifices the performance by 2. 4 points on MMBench. On certain tasks such as ScienceQA and MMMU, we can even go down to only 2 visual tokens with performance drops of just 3\% and 6\% each. Our exploration of the trade-off between the accuracy and computational cost brought about by the number of visual tokens facilitates future research to achieve the best of both worlds.

PDF Details DOI

AAAI Conference 2024 System Paper

MIDDAG: Where Does Our News Go? Investigating Information Diffusion via Community-Level Information Pathways

Mingyu Derek Ma
Alexander K. Taylor
Nuan Wen
Yanchen Liu
Po-Nien Kung
Wenna Qin
Shicheng Wen
Azure Zhou

We present MIDDAG, an intuitive, interactive system that visualizes the information propagation paths on social media triggered by COVID-19-related news articles accompanied by comprehensive insights including user/community susceptibility level, as well as events and popular opinions raised by the crowd while propagating the information. Besides discovering information flow patterns among users, we construct communities among users and develop the propagation forecasting capability, enabling tracing and understanding of how information is disseminated at a higher level. A demo video and more are available at https://info-pathways.github.io.

PDF Details DOI

NeurIPS Conference 2024 Conference Paper

SafeWorld: Geo-Diverse Safety Alignment

Da Yin
Haoyi Qiu
Kung-Hsiang Huang
Kai-Wei Chang
Nanyun Peng

In the rapidly evolving field of Large Language Models (LLMs), ensuring safety is a crucial and widely discussed topic. However, existing works often overlooks the geo-diversity of cultural and legal standards across the world. To reveal the chal5 lenges posed by geo-diverse safety standards, we introduce SafeWorld, a novel benchmark specifically designed to evaluate LLMs’ ability to generate responses that are not only helpful but also culturally sensitive and legally compliant across diverse global contexts. SafeWorld encompasses 2, 775 test user queries, each grounded in high-quality, human-verified cultural norms and legal policies from 50 countries and 493 regions/races. On top of it, we propose a multi-dimensional automatic safety evaluation framework that assesses the contextual appropriateness, accuracy, and comprehensiveness of responses. Our evaluations reveal that current LLMs struggle to meet these criteria effectively. To enhance LLMs’ alignment with geo-diverse safety standards, we synthesize helpful preference pairs for Direct Preference Optimization (DPO) alignment. The preference pair construction aims to encourage LLMs to behave appropriately and provide precise references to relevant cultural norms and policies when necessary. Our trained SafeWorldLM outperforms all competing models, including GPT-4o on all the three evaluation dimensions by a large margin. Global human evaluators also note a nearly 20% higher winning rate in helpfulness and harmfulness evaluation.

PDF Details DOI

AAAI Conference 2024 Conference Paper

STAR: Boosting Low-Resource Information Extraction by Structure-to-Text Data Generation with Large Language Models

Mingyu Derek Ma
Xiaoxuan Wang
Po-Nien Kung
P. Jeffrey Brantingham
Nanyun Peng
Wei Wang

Information extraction tasks such as event extraction require an in-depth understanding of the output structure and sub-task dependencies. They heavily rely on task-specific training data in the form of (passage, target structure) pairs to obtain reasonable performance. However, obtaining such data through human annotation is costly, leading to a pressing need for low-resource information extraction approaches that require minimal human labeling for real-world applications. Fine-tuning supervised models with synthesized training data would be a generalizable method, but the existing data generation methods either still rely on large-scale ground-truth data or cannot be applied to complicated IE tasks due to their poor performance. To address these challenges, we propose STAR, a data generation method that leverages Large Language Models (LLMs) to synthesize data instances given limited seed demonstrations, thereby boosting low-resource information extraction performance. Our approach involves generating target structures (Y) followed by generating passages (X), all accomplished with the aid of LLMs. We design fine-grained step-by-step instructions to obtain the initial data instances. We further reduce errors and improve data quality through self-reflection error identification and self-refinement with iterative revision. Our experiments show that the data generated by STAR significantly improve the performance of low-resource event extraction and relation extraction tasks, even surpassing the effectiveness of human-curated data. Human assessment of the data quality shows STAR-generated data exhibit higher passage quality and better align with the task definitions compared with the human-curated data.

PDF Details DOI

TMLR Journal 2023 Journal Article

Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models

Aarohi Srivastava
Abhinav Rastogi
Abhishek Rao
Abu Awal Md Shoeb
Abubakar Abid
Adam Fisch
Adam R. Brown
Adam Santoro

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG- bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood develop- ment, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google- internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

NeurIPS Conference 2023 Conference Paper

DesCo: Learning Object Recognition with Rich Language Descriptions

Liunian Li
Zi-Yi Dou
Nanyun Peng
Kai-Wei Chang

Recent development in vision-language approaches has instigated a paradigm shift in learning visual recognition models from language supervision. These approaches align objects with language queries (e. g. "a photo of a cat") and thus improve the models' adaptability to novel objects and domains. Recent studies have attempted to query these models with complex language expressions that include specifications of fine-grained details, such as colors, shapes, and relations. However, simply incorporating language descriptions into queries does not guarantee accurate interpretation by the models. In fact, our experiments show that GLIP, a state-of-the-art vision-language model for object detection, often disregards contextual information in the language descriptions and instead relies heavily on detecting objects solely by their names. To tackle the challenge, we propose a new description-conditioned (DesCo) paradigm of learning object recognition models with rich language descriptions consisting of two innovations: 1) we employ a large language model as a commonsense knowledge engine to generate rich language descriptions of objects; 2) we design context-sensitive queries to improve the model's ability in deciphering intricate nuances embedded within descriptions and enforce the model to focus on context rather than object names alone. On two novel object detection benchmarks, LVIS and OminiLabel, under the zero-shot detection setting, our approach achieves 34. 8 APr minival (+9. 1) and 29. 3 AP (+3. 6), respectively, surpassing the prior state-of-the-art models, GLIP and FIBER, by a large margin.

NeurIPS Conference 2022 Conference Paper

Coarse-to-Fine Vision-Language Pre-training with Fusion in the Backbone

Zi-Yi Dou
Aishwarya Kamath
Zhe Gan
Pengchuan Zhang
Jianfeng Wang
Linjie Li
Zicheng Liu
Ce Liu

Vision-language (VL) pre-training has recently received considerable attention. However, most existing end-to-end pre-training approaches either only aim to tackle VL tasks such as image-text retrieval, visual question answering (VQA) and image captioning that test high-level understanding of images, or only target region-level understanding for tasks such as phrase grounding and object detection. We present FIBER (Fusion-In-the-Backbone-based transformER), a new VL model architecture that can seamlessly handle both these types of tasks. Instead of having dedicated transformer layers for fusion after the uni-modal backbones, FIBER pushes multimodal fusion deep into the model by inserting cross-attention into the image and text backbones to better capture multimodal interactions. In addition, unlike previous work that is either only pre-trained on image-text data or on fine-grained data with box-level annotations, we present a two-stage pre-training strategy that uses both these kinds of data efficiently: (i) coarse-grained pre-training based on image-text data; followed by (ii) fine-grained pre-training based on image-text-box data. We conduct comprehensive experiments on a wide range of VL tasks, ranging from VQA, image captioning, and retrieval, to phrase grounding, referring expression comprehension, and object detection. Using deep multimodal fusion coupled with the two-stage pre-training, FIBER provides consistent performance improvements over strong baselines across all tasks, often outperforming methods using magnitudes more data. Code is released at https: //github. com/microsoft/FIBER.

NeurIPS Conference 2022 Conference Paper

Controllable Text Generation with Neurally-Decomposed Oracle

Tao Meng
Sidi Lu
Nanyun Peng
Kai-Wei Chang

We propose a general and efficient framework to control auto-regressive generation models with NeurAlly-Decomposed Oracle (NADO). Given a pre-trained base language model and a sequence-level boolean oracle function, we aim to decompose the oracle function into token-level guidance to steer the base model in text generation. Specifically, the token-level guidance is provided by NADO, a neural model trained with examples sampled from the base model, demanding no additional auxiliary labeled data. Based on posterior regularization, we present the close-form optimal solution to incorporate the decomposed token-level guidance into the base model for controllable generation. We further discuss how the neural approximation affects the quality of the solution. These experiments conducted on two different applications: (1) text generation with lexical constraints and (2) machine translation with formality control demonstrate that our framework efficiently guides the base model towards the given oracle while keeping high generation quality.

NeurIPS Conference 2022 Conference Paper

InsNet: An Efficient, Flexible, and Performant Insertion-based Text Generation Model

Sidi Lu
Tao Meng
Nanyun Peng

We propose InsNet, an expressive insertion-based text generator with efficient training and flexible decoding (parallel or sequential). Unlike most existing insertion-based text generation works that require re-encoding of the (decoding) context after each insertion operation and thus are inefficient to train, InsNet only requires one pass of context encoding for the entire insertion sequence during training by using a novel insertion-oriented position encoding to enable computation sharing. Furthermore, InsNet provides a controllable switch between parallel and sequential decoding, making it flexible to handle more parallelizable tasks such as machine translation to support efficient decoding, or less parallelizable tasks such as lexically constrained text generation to guarantee high-quality outputs. Experiments on two unsupervised lexically constrained text generation datasets and three machine translation datasets demonstrate InsNet’s advantages over previous insertion-based methods in terms of training speed, inference efficiency, and generation quality.

AAAI Conference 2022 Conference Paper

Zero-Shot Commonsense Question Answering with Cloze Translation and Consistency Optimization

Zi-Yi Dou
Nanyun Peng

Commonsense question answering (CQA) aims to test if models can answer questions regarding commonsense knowledge that everyone knows. Prior works that incorporate external knowledge bases have shown promising results, but knowledge bases are expensive to construct and are often limited to a fixed set of relations. In this paper, we instead focus on better utilizing the implicit knowledge stored in pretrained language models. While researchers have found that the knowledge embedded in pre-trained language models can be extracted by having them fill in the blanks of carefully designed prompts for relation extraction and text classification, it remains unclear if we can adopt this paradigm in CQA where the inputs and outputs take much more flexible forms. To this end, we investigate four translation methods that can translate natural questions into cloze-style sentences to better solicit commonsense knowledge from language models, including a syntactic-based model, an unsupervised neural model, and two supervised neural models. In addition, to combine the different translation methods, we propose to encourage consistency among model predictions on different translated questions with unlabeled data. We demonstrate the effectiveness of our methods on three CQA datasets in zeroshot settings. We show that our methods are complementary to a knowledge base improved model, and combining them can lead to state-of-the-art zero-shot performance. Analyses also reveal distinct characteristics of the different cloze translation methods and provide insights on why combining them can lead to great improvements. Code/dataset is available at https: //github. com/PlusLabNLP/zero shot cqa.

AAAI Conference 2021 Conference Paper

GATE: Graph Attention Transformer Encoder for Cross-lingual Relation and Event Extraction

Wasi Uddin Ahmad
Nanyun Peng
Kai-Wei Chang

Recent progress in cross-lingual relation and event extraction use graph convolutional networks (GCNs) with universal dependency parses to learn language-agnostic sentence representations such that models trained on one language can be applied to other languages. However, GCNs struggle to model words with long-range dependencies or are not directly connected in the dependency tree. To address these challenges, we propose to utilize the self-attention mechanism where we explicitly fuse structural information to learn the dependencies between words with different syntactic distances. We introduce GATE, a Graph Attention Transformer Encoder, and test its cross-lingual transferability on relation and event extraction tasks. We perform experiments on the ACE05 dataset that includes three typologically different languages: English, Chinese, and Arabic. The evaluation results show that GATE outperforms three recently proposed methods by a large margin. Our detailed analysis reveals that due to the reliance on syntactic dependencies, GATE produces robust representations that facilitate transfer across languages.

AAAI Conference 2021 Conference Paper

MELINDA: A Multimodal Dataset for Biomedical Experiment Method Classification

Te-Lin Wu
Shikhar Singh
Sayan Paul
Gully Burns
Nanyun Peng

We introduce a new dataset, MELINDA, for Multimodal biomEdicaL experImeNt methoD clAssification. The dataset is collected in a fully automated distant supervision manner, where the labels are obtained from an existing curated database, and the actual contents are extracted from papers associated with each of the records in the database. We benchmark various state-of-the-art NLP and computer vision models, including unimodal models which only take either caption texts or images as inputs, and multimodal models. Extensive experiments and analysis show that multimodal models, despite outperforming unimodal ones, still need improvements especially on a less-supervised way of grounding visual concepts with languages, and better transferability to low resource domains. We release our dataset and the benchmarks to facilitate future research in multimodal learning, especially to motivate targeted improvements for applications in scientific domains.

AAAI Conference 2020 Conference Paper

Predictive Engagement: An Efficient Metric for Automatic Evaluation of Open-Domain Dialogue Systems

Sarik Ghazarian
Ralph Weischedel
Aram Galstyan
Nanyun Peng

User engagement is a critical metric for evaluating the quality of open-domain dialogue systems. Prior work has focused on conversation-level engagement by using heuristically constructed features such as the number of turns and the total time of the conversation. In this paper, we investigate the possibility and efﬁcacy of estimating utterance-level engagement and deﬁne a novel metric, predictive engagement, for automatic evaluation of open-domain dialogue systems. Our experiments demonstrate that (1) human annotators have high agreement on assessing utterance-level engagement scores; (2) conversation-level engagement scores can be predicted from properly aggregated utterance-level engagement scores. Furthermore, we show that the utterance-level engagement scores can be learned from data. These scores can be incorporated into automatic evaluation metrics for open-domain dialogue systems to improve the correlation with human judgements. This suggests that predictive engagement can be used as a real-time feedback for training better dialogue models.

AAAI Conference 2019 Conference Paper

Plan-and-Write: Towards Better Automatic Storytelling

Lili Yao
Nanyun Peng
Ralph Weischedel
Kevin Knight
Dongyan Zhao
Rui Yan

Automatic storytelling is challenging since it requires generating long, coherent natural language to describes a sensible sequence of events. Despite considerable efforts on automatic story generation in the past, prior work either is restricted in plot planning, or can only generate stories in a narrow domain. In this paper, we explore open-domain story generation that writes stories given a title (topic) as input. We propose a plan-and-write hierarchical generation framework that first plans a storyline, and then generates a story based on the storyline. We compare two planning strategies. The dynamic schema interweaves story planning and its surface realization in text, while the static schema plans out the entire storyline before generating stories. Experiments show that with explicit storyline planning, the generated stories are more diverse, coherent, and on topic than those generated without creating a full plan, according to both automatic and human evaluations.

IJCAI Conference 2018 Conference Paper

Learning to Converse with Noisy Data: Generation with Calibration

Mingyue Shang
Zhenxin Fu
Nanyun Peng
Yansong Feng
Dongyan Zhao
Rui Yan

The availability of abundant conversational data on the Internet brought prosperity to the generation-based open domain conversation systems. In the training of the generation models, existing methods generally treat all the training data equivalently. However, the data crawled from the websites may contain many noises. Blindly training with the noisy data could harm the performance of the final generation model. In this paper, we propose a generation with calibration framework, that allows high- quality data to have more influences on the generation model and reduces the effect of noisy data. Specifically, for each instance in training set, we employ a calibration network to produce a quality score for it, then the score is used for the weighted update of the generation model parameters. Experiments show that the calibrated model outperforms baseline methods on both automatic evaluation metrics and human annotations.

AAAI Conference 2018 Conference Paper

Style Transfer in Text: Exploration and Evaluation

Zhenxin Fu
Xiaoye Tan
Nanyun Peng
Dongyan Zhao
Rui Yan

The ability to transfer styles of texts or images, is an important measurement of the advancement of artiﬁcial intelligence (AI). However, the progress in language style transfer is lagged behind other domains, such as computer vision, mainly because of the lack of parallel data and reliable evaluation metrics. In response to the challenge of lacking parallel data, we explore learning style transfer from non-parallel data. We propose two models to achieve this goal. The key idea behind the proposed models is to learn separate content representations and style representations using adversarial networks. Considering the problem of lacking principle evaluation metrics, we propose two novel evaluation metrics that measure two aspects of style transfer: transfer strength and content preservation. We benchmark our models and the evaluation metrics on two style transfer tasks: paper-news title transfer, and positive-negative review transfer. Results show that the proposed content preservation metric is highly correlate to human judgments, and the proposed models are able to generate sentences with similar content preservation score but higher style transfer strength comparing to autoencoder.