Papers by Mohit Bansal
Soft Layer-Specific Multi-Task Summarization with Entailment and Question Generation (P18-1)
Copied to clipboard
| Challenge: | Recent advances on abstractive summarization have allowed substantial improvements in the quality of the model, but there is still scope for improvement. |
| Approach: | They propose novel multi-task architectures with high-level layer-specific sharing across multiple encoder and decoder layers of the three tasks and soft-sharing mechanisms. |
| Outcome: | The proposed model improves on the CNN/DailyMail and Gigaword datasets and on the DUC-2002 transfer setup. |
Continuous Language Generative Flow (2021.acl-long)
Copied to clipboard
| Challenge: | Recent years have witnessed various types of generative models for natural language generation (NLG), especially RNNs or transformers. |
| Approach: | They propose a flow-based language generation model that adapts flow-derived generative models to language generation via continuous input embeddings, adapted affine coupling structures, and a novel architecture for autoregressive text generation. |
| Outcome: | The proposed model improves on QG and NMT and improves performance over baselines on SQuAD and TVQA and NML16. |
Closed-Book Training to Improve Summarization Encoder Memory (D18-1)
Copied to clipboard
| Challenge: | Extractive summarization models pick words, phrases, and sentences from the source text to form a summary, while abstractive models generate (samples) words from a fixed-size vocabulary instead of copying from text directly. |
| Approach: | They propose to add an additional ‘closed-book’ decoder without attention and pointer mechanisms to the model to improve its memorization capabilities. |
| Outcome: | The proposed model outperforms the baseline model on the CNN/Daily Mail dataset in terms of ROUGE and METEOR metrics, and achieves higher scores in a test-only setup. |
What Can We Learn from Collective Human Opinions on Natural Language Inference Data? (2020.emnlp-main)
Copied to clipboard
| Challenge: | Despite the subjective nature of many NLU evaluations, little attention has been paid to the distribution of human opinions. |
| Approach: | They use a dataset with 464,500 annotations to study Collective HumAn OpinionS . they argue that models lack the ability to recover the distribution over human labels . |
| Outcome: | The proposed dataset examines the distribution of human opinions in NLU evaluation datasets. |
FLAMES: Improving LLM Math Reasoning via a Fine-Grained Analysis of the Data Synthesis Pipeline (2025.findings-emnlp)
Copied to clipboard
Parker Seegmiller, Kartik Mehta, Soumya Saha, Chenyang Tao, Shereen Oraby, Arpit Gupta, Tagyoung Chung, Mohit Bansal, Nanyun Peng
| Challenge: | Recent work improving LLM math reasoning with synthetic data uses unique setups, making comparison of data synthesis strategies impractical. |
| Approach: | They propose a framework for LLM assessment of math reasoning with synthetic data . they use 10 existing data synthesis strategies and multiple other factors to study performance . |
| Outcome: | The proposed data synthesis strategies outperform public datasets on OlympiadBench, CollegeMath, GSMPlus and MATH. |
mTVR: Multilingual Moment Retrieval in Videos (2021.acl-short)
Copied to clipboard
| Challenge: | mTVR is a multilingual video moment retrieval dataset with 218K queries in English and Chinese . Various datasets have been proposed or adapted for the task, but they are all created for a single language (English). |
| Approach: | They propose a multilingual video moment retrieval dataset with 218K queries from 21.8K TV show video clips. |
| Outcome: | The proposed model outperforms strong monolingual baselines while using fewer parameters. |
Can Sequence-to-Sequence Transformers Naturally Understand Sequential Instructions? (2023.starsem-1)
Copied to clipboard
| Challenge: | Using a limited annotation budget, we can greatly improve the performance on intermediate steps with a drop in final-step performance. |
| Approach: | They propose to use a pre-supervised sequence-to-sequence transformer to provide training signals on intermediate steps with zero gold supervision instead of only final-step supervision to improve performance. |
| Outcome: | The proposed model significantly improves on intermediate steps with a drop in final-step performance on one subtask, but also shows decreased performance on another subtask. |
On Positional Bias of Faithfulness for Long-form Summarization (2025.naacl-long)
Copied to clipboard
| Challenge: | Large language models exhibit positional bias in long-context settings, under-attending to information in the middle. |
| Approach: | They compile eight human-annotated long-form summarization datasets to evaluate faithfulness . they find that LLMs faithfully summarize beginning and end of documents but neglect middle content . |
| Outcome: | The proposed methods show that LLMs under-attend to information in the middle of inputs. |
Explanation Graph Generation via Pre-trained Language Models: An Empirical Study with Contrastive Learning (2022.acl-long)
Copied to clipboard
| Challenge: | Pre-trained sequence-to-sequence language models generate structured outputs such as graphs with limited supervision. |
| Approach: | They propose to use pre-trained sequence-to-sequence language models to generate graphs . they propose to learn structural constraints and semantics of graphs with limited supervision . |
| Outcome: | The proposed models can learn structural constraints and semantics of graphs with limited supervision. |
Revealing Single Frame Bias for Video-and-Language Learning (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for video-and-language learning use multiple frames as inputs. |
| Approach: | They propose to use single-frame models for video-and-language learning to investigate temporality in video- and language tasks. |
| Outcome: | The proposed model does not take into account temporal information on video-and-language tasks. |
Learning and Analyzing Generation Order for Undirected Sequence Models (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Undirected neural sequence models generate monotonically from left to right in machine translation tasks. |
| Approach: | They train a policy that learns the generation order for a pre-trained, undirected translation model via reinforcement learning. |
| Outcome: | The proposed policy outperforms heuristic generation orders on three out of four language pairs. |
Multimodal Intent Discovery from Livestream Videos (2022.findings-naacl)
Copied to clipboard
Adyasha Maharana, Quan Tran, Franck Dernoncourt, Seunghyun Yoon, Trung Bui, Walter Chang, Mohit Bansal
| Challenge: | Existing models for instructional video understanding struggle to understand abstract intents . identifying procedural intent within instructional videos is a challenging task . |
| Approach: | They propose to extract instructional intent from software instructional livestreams by using a multimodal cascaded cross-attention model that integrates weaker and noisier video signals with more discriminative text signals. |
| Outcome: | The proposed model improves on baseline models and compares it to existing models. |
The Curse of Performance Instability in Analysis Datasets: Consequences, Source, and Suggestions (2020.emnlp-main)
Copied to clipboard
| Challenge: | Neural network models have significantly pushed forward performance on natural language processing benchmarks with the development of largescale language model pre-training. |
| Approach: | They find that models on natural language inference and reading comprehension are unstable . they propose to use a model-selection routine to analyze the model's instability . |
| Outcome: | The proposed models can perform poorly on two language-related tasks, the authors show . they also show that the model selection routine is unstable, and that it is not reliable . |
Fast Abstractive Summarization with Reinforce-Selected Sentence Rewriting (P18-1)
Copied to clipboard
| Challenge: | Empirically, we achieve the new state-of-the-art on all metrics (including human evaluation) on the CNN/Daily Mail dataset, as well as significantly higher abstractiveness scores. |
| Approach: | They propose a sentence-level policy gradient method that bridges computation between two neural networks in a hierarchical way while maintaining language fluency. |
| Outcome: | The proposed model achieves state-of-the-art on all metrics and higher abstractiveness scores on the CNN/Daily Mail dataset and faster training convergence than previous models. |
Continual Few-Shot Learning for Text Classification (2021.emnlp-main)
Copied to clipboard
| Challenge: | a large number of end-to-end systems are needed for many tasks in natural language processing. |
| Approach: | They propose a continual few-shot learning task where a system is asked to correct mistakes with a few training examples. |
| Outcome: | The proposed task compares two NLI and one sentiment analysis datasets with baselines from diverse paradigms. |
ALFRED-L: Investigating the Role of Language for Action Learning in Interactive Visual Environments (2022.emnlp-main)
Copied to clipboard
Arjun Akula, Spandana Gella, Aishwarya Padmakumar, Mahdi Namazifar, Mohit Bansal, Jesse Thomason, Dilek Hakkani-Tur
| Challenge: | Embodied Vision and Language Task Completion requires an embodied agent to interpret natural language instructions and egocentric visual observations to navigate through and interact with environments. |
| Approach: | They propose to use ALFRED to test whether models can adapt to tasks not seen during training that require the same types of language understanding as ALFred. |
| Outcome: | The proposed model can adapt to tasks that require the same types of language understanding as ALFRED-L. |
Improving Visual Question Answering by Referring to Generated Paragraph Captions (P19-1)
Copied to clipboard
| Challenge: | Empirical results show that paragraph captions help answer more visual questions . |
| Approach: | They propose a visual and textual question answering model which uses paragraph captions as input . they use cross-attention to extract related information, then consensus to fuse the inputs . |
| Outcome: | Empirical results show that paragraph captions help answer more visual questions . the proposed model significantly improves the baseline model . |
MeetingQA: Extractive Question-Answering on Meeting Transcripts (2023.acl-long)
Copied to clipboard
| Challenge: | Meeting transcripts are a promising domain for natural language tasks . lack of annotated data impedes research on other important tasks in this domain . |
| Approach: | They propose an extractive QA dataset comprising questions asked by meeting participants and corresponding responses. |
| Outcome: | The proposed dataset extracts questions asked by meeting participants and corresponding responses from transcripts. |
GRAVL-BERT: Graphical Visual-Linguistic Representations for Multimodal Coreference Resolution (2022.coling-1)
Copied to clipboard
Danfeng Guo, Arpit Gupta, Sanchit Agarwal, Jiun-Yu Kao, Shuyang Gao, Arijit Biswas, Chien-Wei Lin, Tagyoung Chung, Mohit Bansal
| Challenge: | Multimodal coreference resolution (MCR) is a crucial capability for building next-generation conversational agents. |
| Approach: | They propose a multimodal coreference resolution model that resolves coreferences made in multi-turn dialogues with scene images. |
| Outcome: | The proposed model resolves coreferences made in multi-turn dialogues with scene images. |
AutoSeM: Automatic Task Selection and Mixing in Multi-Task Learning (N19-1)
Copied to clipboard
| Challenge: | Multi-task learning is an inductive transfer mechanism that leverages information from related tasks to improve the primary model's generalization performance. |
| Approach: | They propose a multitask learning pipeline that finds relevant auxiliary tasks and learns their mixing ratio. |
| Outcome: | The proposed model can find relevant auxiliary tasks and learn their mixing ratio . the proposed model achieves significant performance boosts on several primary tasks . |
Teaching Models to Balance Resisting and Accepting Persuasion (2025.naacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) are susceptible to persuasion, which can pose risks when faced with an adversarial interlocutor. |
| Approach: | They propose a method to balance positive and negative persuasion by using recursive dialogue trees to train models to accept persulasion. |
| Outcome: | The proposed model-based training improves resistance to misinformation and resilience to being challenged while also resulting in the best overall performance on multi-agent debates across two domains. |
Masked Part-Of-Speech Model: Does Modeling Long Context Help Unsupervised POS-tagging? (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent Part-Of-Speech (POS) induction models assume certain independence assumptions that do not hold in real languages. |
| Approach: | They propose a Masked Part-of-Speech Model (MPoSM) that can model arbitrary tag dependency and perform POS induction through the objective of masked POS reconstruction. |
| Outcome: | The proposed model can model arbitrary tag dependency and perform POS induction through the objective of masked POS reconstruction. |
PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise (2026.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are prone to factually inconsistent statements, known as hallucinations. |
| Approach: | They propose to train a specialized model that detects inconsistencies over text prefixes to improve generation faithfulness by 5-14 F1 points. |
| Outcome: | The proposed model outperforms baseline models by 5-14 F1 points in prefix-level entailment. |
Dense-Caption Matching and Frame-Selection Gating for Temporal Localization in VideoQA (2020.acl-main)
Copied to clipboard
| Challenge: | Recent years have witnessed a paradigm shift in the way we get our information, and a lot of it. |
| Approach: | They propose a video question answering model which integrates multi-modal input sources and finds temporally relevant information to answer questions. |
| Outcome: | The proposed model outperforms the state-of-the-art on a TVQA dataset. |
LXMERT: Learning Cross-Modality Encoder Representations from Transformers (D19-1)
Copied to clipboard
| Challenge: | Existing models with better representations of visual content and language have been developed for visual-content understanding. |
| Approach: | They propose a framework to learn vision-and-language connections from Transformers models . they pre-train a large-scale Transformer model with large amounts of image-and sentence pairs . |
| Outcome: | The proposed model improves state-of-the-art on two visual-reasoning tasks by 22% . the proposed model is based on a large-scale Transformer model with three encoders . |
ADaPT: As-Needed Decomposition and Planning with Language Models (2024.findings-naacl)
Copied to clipboard
Archiki Prasad, Alexander Koller, Mareike Hartmann, Peter Clark, Ashish Sabharwal, Mohit Bansal, Tushar Khot
| Challenge: | Large Language Models (LLMs) are increasingly being used for interactive decision-making tasks requiring planning and adapting to the environment. |
| Approach: | They propose an approach that explicitly plans and decomposes complex sub-tasks when the LLM is unable to execute them. |
| Outcome: | The proposed approach significantly outperforms established strong baselines, achieving success rates up to 28.3% higher in ALFWorld, 27% in WebShop, and 33% in TextCraft. |
LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints (2024.findings-emnlp)
Copied to clipboard
Thomas Palmeira Ferraz, Kartik Mehta, Yu-Hsiang Lin, Haw-Shiuan Chang, Shereen Oraby, Sijia Liu, Vivek Subramanian, Tagyoung Chung, Mohit Bansal, Nanyun Peng
| Challenge: | Recent studies have shown that LLMs struggle with instructions containing multiple constraints. |
| Approach: | They propose a self-correction pipeline that decomposes the original instruction into a list of constraints and uses a Critic model to decide when and where the LLM’s response needs refinement. |
| Outcome: | The proposed model outperforms GPT-4 on RealInstruct and IFEval even with weak feedback. |
HistAlign: Improving Context Dependency in Language Generation by Aligning with History (2023.emnlp-main)
Copied to clipboard
| Challenge: | Language models (LMs) can generate hallucinations and incoherent outputs due to their weak context dependency. |
| Approach: | They propose a training approach to ensure good cache alignment so that the model receives useful signals from the history. |
| Outcome: | The proposed approach improves text coherence and faithfulness on diverse language generation tasks. |
Object Ordering with Bidirectional Matchings for Visual Reasoning (N18-2)
Copied to clipboard
| Challenge: | et al. (2016) proposed a novel end-to-end neural model for visual reasoning with compositional natural language instructions. |
| Approach: | They propose an end-to-end neural model for a visual reasoning task based on a newly-released Cornell dataset . they use joint bidirectional attention to build a two-way conditioning between visual information and language phrases . then they use an RL-based pointer network to sort and process the varying number of unordered objects in each image and pool over the three decisions . |
| Outcome: | The proposed model achieves 4-6% absolute improvements over the state-of-the-art model on the NLVR dataset. |
Video-RTS: Rethinking Reinforcement Learning and Test-Time Scaling for Efficient and Enhanced Video Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Despite advances in reinforcement learning, data collection and fine-tuning remain costly and hard to scale. |
| Approach: | They propose a video-adaptive test-time scaling strategy that combines RL with a supervised fine-tuning strategy to improve video reasoning capability. |
| Outcome: | The proposed method surpasses existing models by 2.4% in accuracy using only 3.6% training samples. |
SafeCity: Understanding Diverse Forms of Sexual Harassment Personal Stories (D18-1)
Copied to clipboard
| Challenge: | With the recent rise of #MeToo, an increasing number of personal stories about sexual harassment and sexual abuse have been shared online. |
| Approach: | They propose to use CNN-RNN model to automatically categorize and analyze sexual harassment data from SafeCity forums. |
| Outcome: | The proposed model achieves an accuracy of 86.5% for groping, ogling, and commenting, and 82.5% in multi-label models. |
DART: Leveraging Multi-Agent Disagreement for Tool Recruitment in Multimodal Reasoning (2026.eacl-long)
Copied to clipboard
Nithin Sivakumaran, Justin Chen, David Wan, Yue Zhang, Jaehong Yoon, Elias Stengel-Eskin, Mohit Bansal
| Challenge: | a key strength of human intelligence is the ability to debate and discuss reasoning with others. |
| Approach: | They propose a multi-agent framework that uses disagreements between visual agents to identify useful visual tools that can resolve inter-agency disagreement. |
| Outcome: | The proposed framework beats the strongest baseline on A-OKVQA and MMMU, respectively. |
Stabilizing Efficient Reasoning with Step-Level Advantage Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models generate long and verbose reasoning traces at inference time . short context post-training alone induces substantial reasoning compression . |
| Approach: | They propose a step-level advantage selection approach that reduces reasoning length by over 30% . they propose to use GRPO without any length-aware objective to train models in a shorter context window . |
| Outcome: | The proposed approach reduces average reasoning length by over 30% while improving Pass@1 accuracy by 3.79 points over the strongest length-aware baseline. |
REFINESUMM: Self-Refining MLLM for Generating a Multimodal Summarization Dataset (2024.acl-long)
Copied to clipboard
| Challenge: | generating accurate and faithful multimodal summaries is challenging due to lack of appropriate multimodal datasets . large language models excel at synthesizing key information from diverse sources, but lack of adequate multimodal data sets for fine-tuning . |
| Approach: | They propose a dataset specifically designed for image-text multimodal summarization . they generate summaries from Wikipedia sections and corresponding images and evaluate them . |
| Outcome: | The proposed dataset improves summary quality by training a critic model on human annotations and using its predictions to remove low-quality summaries. |
MART: Memory-Augmented Recurrent Transformer for Coherent Video Paragraph Captioning (2020.acl-main)
Copied to clipboard
| Challenge: | Existing methods for generating paragraph descriptions for videos require a coherent paragraph and a higher level of coherence. |
| Approach: | They propose a new method that generates a summarized memory state from video segments and sentence history to help better predict the next sentence. |
| Outcome: | The proposed method generates more coherent and less repetitive paragraph captions while maintaining relevance to the input video events. |
Crowdsourcing Lightweight Pyramids for Manual Summary Evaluation (N19-1)
Copied to clipboard
Ori Shapira, David Gabay, Yang Gao, Hadar Ronen, Ramakanth Pasunuru, Mohit Bansal, Yael Amsterdamer, Ido Dagan
| Challenge: | Manual evaluation methods are perceived as insufficient due to the high cost of the Pyramid method and the required expertise. |
| Approach: | They propose a crowdsourced method that compares system summaries to references and uses crowdsourced scripts to analyze the results. |
| Outcome: | The proposed method shows higher correlation relative to the original Pyramid method. |
DeepMaven: Deep Question Answering on Long-Distance Movie/TV Show Videos with Multimedia Knowledge Extraction and Synthesis (2023.eacl-main)
Copied to clipboard
| Challenge: | Long video content understanding poses a challenging set of research questions as it involves long-distance, cross-media reasoning and knowledge awareness. |
| Approach: | They propose a framework which extracts events, entities, and relations from the rich multimedia content in long videos to pre-construct movie knowledge graphs. |
| Outcome: | The proposed framework performs competitively for both the new DeepMovieQA and the pre-existing MovieQA dataset. |
GraDA: Graph Generative Data Augmentation for Commonsense Reasoning (2022.coling-1)
Copied to clipboard
| Challenge: | Recent advances in commonsense reasoning have been fueled by the availability of large-scale human annotated datasets. |
| Approach: | They propose a graph-generative data augmentation framework to synthesize factual data samples from knowledge graphs for commonsense reasoning. |
| Outcome: | The proposed framework improves SocialIQA, CODAH, HellaSwag and CommonsenseQA . it also performs well for generative tasks like ProtoQA proving its robustness to adversaries . |
Knowledge-Aware Reasoning over Multimodal Semi-structured Tables (2024.findings-emnlp)
Copied to clipboard
Suyash Mathur, Jainit Bafna, Kunal Kartik, Harshita Khandelwal, Manish Shrivastava, Vivek Gupta, Mohit Bansal, Dan Roth
| Challenge: | Existing datasets for tabular question answering focus on text within cells, but real-world data is multimodal, often blending images such as symbols, faces, icons, patterns, and charts with textual content. |
| Approach: | They propose a dataset to assess whether current AI models can perform knowledge-aware reasoning on multimodal structured data. |
| Outcome: | The proposed dataset is a robust benchmark for advancing AI’s comprehension and capabilities in analyzing multimodal structured data. |
Extending Multi-Document Summarization Evaluation to the Interactive Setting (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing approaches to interactive summarization are incomparable and divergent . a key gap in the development and adoption of interactive summaries is the lack of evaluation methodologies and benchmarks for meaningful comparison of systems. |
| Approach: | They propose an end-to-end evaluation framework for interactive summarization based on expansion-based interaction . framework includes procedure of collecting real user sessions, evaluation measures relying on summarizing standards, but adapted to reflect interaction. |
| Outcome: | The proposed evaluation framework is based on evaluations of baseline implementations and is available publicly as a benchmark. |
Efficiently Summarizing Text and Graph Encodings of Multi-Document Clusters (2021.naacl-main)
Copied to clipboard
| Challenge: | Abstractive multi-document summarization (MDS) is a task that has seen advances with the introduction of large-scale datasets and powerful Transformer-based models. |
| Approach: | They propose an efficient graph-enhanced approach to multi-document summarization with an encoder-decoder Transformer model. |
| Outcome: | The proposed model scales to large input documents and improves on a multi-document dataset. |
Reverse Thinking Makes LLMs Stronger Reasoners (2025.naacl-long)
Copied to clipboard
Justin Chen, Zifeng Wang, Hamid Palangi, Rujun Han, Sayna Ebrahimi, Long Le, Vincent Perot, Swaroop Mishra, Mohit Bansal, Chen-Yu Lee, Tomas Pfister
| Challenge: | Reverse-Enhanced Thinking (RevThink) is a framework for large language models to perform reverse thinking. |
| Approach: | They propose a framework for enhancing forward-backward reasoning by collecting data from a teacher model and employing three objectives to train a student model in a multi-task learning fashion. |
| Outcome: | The proposed framework outperforms a fine-tuning method trained on 10x more forward reasoning on 12 datasets covering commonsense, math, and logical reasoning. |
Evaluating and Improving Factuality in Multimodal Abstractive Summarization (2022.emnlp-main)
Copied to clipboard
| Challenge: | Current factuality metrics do not account for vision modality, thus are not adequate for vision-and-language summarization. |
| Approach: | They propose a weighted combination of CLIPScore and BERTScore to evaluate factuality for abstractive document summarization. |
| Outcome: | The proposed metric outperforms existing factuality metrics on four factuity metric-evaluation benchmarks and is robust to human judgments. |
Multi-Attribute Steering of Language Models via Targeted Intervention (2025.acl-long)
Copied to clipboard
| Challenge: | Existing approaches for steering large language models fail to scale to multi-attribute settings with conflicts, such as enhancing helpfulness while also reducing toxicity. |
| Approach: | They propose a steering framework for selective token-level intervention across multiple attributes that enforcing sparsity and orthogonality among vectors for different attributes. |
| Outcome: | The proposed framework outperforms existing ITI and parameter-efficient fine-tuning approaches across question answering tasks and generative tasks. |
Game-Based Video-Context Dialogue (D18-1)
Copied to clipboard
| Challenge: | Current dialogue systems focus more on textual and speech context knowledge and are usually based on two speakers. |
| Approach: | They propose to use live soccer game videos and Twitch.tv chats to develop visual-grounded dialogue models. |
| Outcome: | The proposed model can generate relevant temporal and spatial event language from live video and chat history while also being relevant to chat history. |
Data Factors for Better Compositional Generalization (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent diagnostic datasets on compositional generalization expose severe problems . state-of-the-art models trained on larger and more general datasets show better generalization ability . |
| Approach: | They conduct an empirical analysis by training Transformer models on a variety of training sets with different data factors including dataset scale, pattern complexity, example difficulty, etc. |
| Outcome: | The proposed model training on larger datasets improves on compositional generalization tasks. |
An Empirical Study of Multimodal Model Merging (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing studies have shown that model merging can generate a multi-task solution without synchronous training. |
| Approach: | They propose to merge vision, language, and cross-modal transformers of a modality-specific architecture to create a parameter-efficient architecture. |
| Outcome: | The proposed model merging outperforms naive models on various tasks with improvements of 3% on VQA, 7% on COCO retrieval, 25% on NLVR2, 14% on Flickr30k and 3% ADE20k. |
AdaCAD: Adaptively Decoding to Balance Conflicts between Contextual and Parametric Knowledge (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing contrastive methods that ignore the context of a large language model (LLM) fail to handle instances that vary in their amount of conflict, with static methods over-adjusting when conflict is absent. |
| Approach: | They propose a fine-grained, instance-level approach called AdaCAD which dynamically adjusts the degree of conflict based on the degree. |
| Outcome: | The proposed approach outperforms baselines and improves factuality of summaries by 6.19. |
Explaining and Improving Contrastive Decoding by Extrapolating the Probabilities of a Huge and Hypothetical LM (2024.emnlp-main)
Copied to clipboard
| Challenge: | Contrastive decoding (CD) improves the next-token distribution of a large expert language model (LM) using a small amateur LM. |
| Approach: | They propose a new unsupervised decoding method called Asymptotic Probability Decoding (APD) that extrapolates the probability curves from the LMs of different sizes to infer the asymptototic probabilities from an infinitely large LM. |
| Outcome: | The proposed method improves the next-token distribution of a large expert language model using a small amateur LM. |
Towards Robustifying NLI Models Against Lexical Dataset Biases (2020.acl-main)
Copied to clipboard
| Challenge: | Recent studies show that deep learning models exploit dataset biases without deep understanding of the language semantics. |
| Approach: | They propose two methods to debiase models against lexical dataset biases . they use contradiction-word bias and word-overlapping bias as examples . |
| Outcome: | The proposed method removes label bias at embedding level, while the other uses a bag-of-words sub-model to capture features likely to exploit the bias. |
SETSum: Summarization and Visualization of Student Evaluations of Teaching (2022.naacl-demo)
Copied to clipboard
| Challenge: | Student Evaluations of Teaching (SETs) are used in colleges and universities to assess student perceptions about their courses. |
| Approach: | They propose a system that leverages sentiment analysis, aspect extraction, summarization and visualization techniques to provide organized illustrations of SET findings to instructors and other reviewers. |
| Outcome: | The proposed system can be used by 10 professors from diverse departments to analyze SET results. |
Robustness Gym: Unifying the NLP Evaluation Landscape (2021.naacl-demos)
Copied to clipboard
| Challenge: | Existing tools cater to specialized set of evaluations and provide no clear way to leverage or share findings from prior evaluations. |
| Approach: | They propose a toolkit that unifies 4 evaluation paradigms to provide a common platform for evaluation. |
| Outcome: | The proposed evaluation toolkit unifies 4 evaluation paradigms and is under active development. |
PRInTS: Reward Modeling for Long-Horizon Information Seeking (2026.acl-long)
Copied to clipboard
| Challenge: | Existing PRMs cannot capture richer dimensions of information-seeking steps, such as tool interactions and reasoning over tool outputs. |
| Approach: | They propose a generative PRM trained with dual capabilities that compresses the growing context while preserving essential information for step evaluation. |
| Outcome: | PRInTS improves on FRAMES, GAIA, and WebWalkerQA models while preserving essential information for step evaluation. |
Hidden Biases in Unreliable News Detection Datasets (2021.eacl-main)
Copied to clipboard
| Challenge: | Recent studies show that automatic unreliable news detection models only use the article itself without resorting to fact-checking mechanisms. |
| Approach: | They propose to use a simple model as a difficulty/bias probe instead of a complex one . they observe a significant drop in accuracy for all models tested in a clean split . |
| Outcome: | The proposed model can achieve good performance by memorizing site-label mapping instead of modeling the real task. |
Explore, Propose, and Assemble: An Interpretable Model for Multi-Hop Reading Comprehension (P19-1)
Copied to clipboard
| Challenge: | Existing models for multi-hop reading comprehension only require a single-hop reasoning, meaning that the evidence needed to answer the question is scattered in a set of supporting documents. |
| Approach: | They propose an interpretable 3-module system called Explore-Propose-Assemble reader (EPAr) that explores and connects relevant information from multiple documents in order to answer a question about the context. |
| Outcome: | The proposed model approximates coarse-to-fine-grained comprehension behavior of human readers when facing multiple long documents. |
ExplaGraphs: An Explanation Graph Generation Task for Structured Commonsense Reasoning (2021.emnlp-main)
Copied to clipboard
| Challenge: | Current commonsense-reasoning tasks are discriminative in nature, where a model answers a multiple-choice question for a certain context. |
| Approach: | They propose a generative task that generates a commonsense-augmented graph for stance prediction by using a create-verify-and-refine graph collection framework. |
| Outcome: | The proposed model is able to generate a graph that serves as non-trivial, complete, and unambiguous explanation for the predicted stance. |
FactPEGASUS: Factuality-Aware Pre-training and Fine-tuning for Abstractive Summarization (2022.naacl-main)
Copied to clipboard
| Challenge: | Abstractive summarization models suffer from the problem of hallucinations, where a summary contains facts or entities not present in the original document. |
| Approach: | They propose an abstractive summarization model that addresses the problem of factuality during pre-training and fine-tuning. |
| Outcome: | Experiments on three downstream tasks show that FactPEGASUS significantly improves factuality compared to the original pre-training objective in zero-shot and few-shot settings. |
CoSIm: Commonsense Reasoning for Counterfactual Scene Imagination (2022.naacl-main)
Copied to clipboard
| Challenge: | a new dataset evaluates the ability of AI systems to reason about scene change imagination . a large human-model performance gap exists in the dataset . |
| Approach: | They propose a dataset to evaluate AI's ability to reason about scene change imagination . they use an image and a commonsense question to imagine a counterfactual scene change . |
| Outcome: | The proposed dataset evaluates the ability of AI systems to reason about scene change imagination. |
Multi-Reward Reinforced Summarization with Saliency and Entailment (N18-2)
Copied to clipboard
| Challenge: | Abstractive text summarization is the task of compressing and rewriting a long document into a short summary while maintaining saliency, directed logical entailment, and non-redundancy. |
| Approach: | They propose a novel reward function for ROUGESal and Entail to improve abstractive summarization . they use a coverage-based reward function to combine ROUGE and En Tail . |
| Outcome: | The proposed method achieves state-of-the-art results on CNN/Daily Mail dataset and strong improvements in a test-only transfer setup on DUC-2002. |
Routing with Generated Data: Annotation-Free LLM Skill Estimation and Expert Selection (2026.acl-long)
Copied to clipboard
Tianyi Niu, Justin Chen, Genta Indra Winata, Shi-Xiong Zhang, Supriyo Chakraborty, Sambit Sahu, Yue Zhang, Elias Stengel-Eskin, Mohit Bansal
| Challenge: | Existing approaches typically assume access to ground-truth labeled data . Existing methods require a classifier to select models given an input . |
| Approach: | They propose a routing setting where routers are trained exclusively on generated queries and answers from LLMs. |
| Outcome: | The proposed router outperforms the best query-answer router by 4.6% absolute accuracy when trained on weak generator data. |
Extractive is not Faithful: An Investigation of Broad Unfaithfulness Problems in Extractive Summarization (2023.acl-long)
Copied to clipboard
| Challenge: | Abstractive summarization is less prone to unfaithfulness issues than abstractive summaries . but, unfaitfulness problems, i.e., hallucinating new information, are still a problem in extractive summarisation . |
| Approach: | They propose a typology with five types of broad unfaithfulness problems that can appear in extractive summaries, including and beyond not-entailment. |
| Outcome: | The proposed metric shows that it detects unfaithful summaries faster than existing faithfulness evaluation metrics. |
Dynabench: Rethinking Benchmarking in NLP (2021.naacl-main)
Copied to clipboard
Douwe Kiela, Max Bartolo, Yixin Nie, Divyansh Kaushik, Atticus Geiger, Zhengxuan Wu, Bertie Vidgen, Grusha Prasad, Amanpreet Singh, Pratik Ringshia, Zhiyi Ma, Tristan Thrush, Sebastian Riedel, Zeerak Waseem, Pontus Stenetorp, Robin Jia, Mohit Bansal, Christopher Potts, Adina Williams
| Challenge: | Dynabench is an open-source platform for dynamic dataset creation and model benchmarking. |
| Approach: | They propose an open-source platform for dynamic dataset creation and model benchmarking. |
| Outcome: | The proposed platform can be used to create models that fail on simple challenges and falter in real-world scenarios. |
RESIN-11: Schema-guided Event Prediction for 11 Newsworthy Scenarios (2022.naacl-demo)
Copied to clipboard
Xinya Du, Zixuan Zhang, Sha Li, Pengfei Yu, Hongwei Wang, Tuan Lai, Xudong Lin, Ziqi Wang, Iris Liu, Ben Zhou, Haoyang Wen, Manling Li, Darryl Hannan, Jie Lei, Hyounghun Kim, Rotem Dror, Haoyu Wang, Michael Regan, Qi Zeng, Qing Lyu, Charles Yu, Carl Edwards, Xiaomeng Jin, Yizhu Jiao, Ghazaleh Kazeminejad, Zhenhailong Wang, Chris Callison-Burch, Mohit Bansal, Carl Vondrick, Jiawei Han, Dan Roth, Shih-Fu Chang, Martha Palmer, Heng Ji
| Challenge: | Existing methods for event prediction are incomplete and noisy. |
| Approach: | They propose to use news-related event schemas to extract newsworthy events . they build a demo website and include a video demonstrating the framework . |
| Outcome: | The proposed framework can be applied to a wide variety of newsworthy scenarios. |
Parsing Speech: a Neural Approach to Integrating Lexical and Acoustic-Prosodic Information (N18-1)
Copied to clipboard
| Challenge: | acoustic signals provide cues that help listeners disambiguate difficult parses . speech carries useful extra information associated with prosodic structure . |
| Approach: | They propose a model that integrates transcribed text and acoustic-prosodic features into a neural network that accepts text and prosodic feature. |
| Outcome: | The proposed model improves parse and disfluency detection scores over a strong text-only baseline. |
FactGraph: Evaluating Factuality in Summarization with Semantic Graph Representations (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that abstractive summarization approaches generate summaries that are not factually consistent with the source document. |
| Approach: | They propose a method that decomposes the document and summary into structured meaning representations (MRs) MRs describe core semantic concepts and their relations, aggregating the main content in both document and summary in a canonical form . |
| Outcome: | The proposed method outperforms existing methods on benchmarks for factuality evaluation. |
ReConcile: Round-Table Conference Improves Reasoning via Consensus among Diverse LLMs (2024.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) still struggle with natural language reasoning tasks. |
| Approach: | They propose a multi-model multi-agent framework that enhances collaborative reasoning between LLM agents via multiple rounds of discussion. |
| Outcome: | Experiments on seven benchmarks show that ReConcile significantly improves LLM reasoning on individual and team levels. |
A Simple LLM Framework for Long-Range Video Question-Answering (2024.emnlp-main)
Copied to clipboard
| Challenge: | a recent study has shown that short video understanding is not trivial due to the need for long-range temporal reasoning capabilities. |
| Approach: | They propose a language-based short- and long-range question-answering framework LLoVi . they propose 'multi-round summarization prompt' that asks the LLM to summarize the captions . |
| Outcome: | The proposed framework outperforms the state-of-the-art on the EgoSchema dataset and to grounded VideoQA. |
Self-Assembling Modular Networks for Interpretable Multi-Hop Reasoning (D19-1)
Copied to clipboard
| Challenge: | Existing models for multi-hop question answering require multiple pieces of evidence scattered in a given context. |
| Approach: | They propose an interpretable, controller-based self-assembling Neural Modular Network for multi-hop reasoning . their model can softly decompose a multi-step question into multiple single-hop sub-questions . |
| Outcome: | The proposed model improves on the static, single-hop model on regular and adversarial evaluations. |
Adversarial NLI: A New Benchmark for Natural Language Understanding (2020.acl-main)
Copied to clipboard
| Challenge: | a new large-scale NLI benchmark dataset is presented to test models on a variety of popular NLIs. |
| Approach: | They propose a large-scale NLI benchmark dataset that is iteratively compared with a human-and-model-in-the-loop procedure. |
| Outcome: | The proposed method can be applied in a never-ending learning scenario, becoming a moving target for NLU, rather than a static benchmark that will quickly saturate. |
Analyzing the Limits of Self-Supervision in Handling Bias in Language (2022.findings-emnlp)
Copied to clipboard
| Challenge: | a recent study shows that natural language models can perform tasks with little to no in-context supervision . a number of tasks are performed using self-supervised pre-training . |
| Approach: | They define and comprehensively evaluate how well natural language taskprompting captures the semantics of four tasks for bias: diagnosis, identification, extraction and rephrasing. |
| Outcome: | The proposed model performs to wide varying degrees across bias dimensions . the model is largely challenged when prompted to perform these tasks . |
Distributed NLI: Learning to Predict Human Opinion Distributions for Language Reasoning (2022.findings-acl)
Copied to clipboard
| Challenge: | Using distributed NLI, we show that models can capture human judgement distribution more effectively than the softmax baseline. |
| Approach: | They propose a new NLU task to predict the distribution of human judgements . they propose Monte Carlo, Deep Ensemble, Re-Calibration and Distribution Distillation methods to capture human judgement distributions. |
| Outcome: | The proposed methods perform better than the softmax baseline, but the results are still far below the estimated human upper-bound. |
DeCEMBERT: Learning from Noisy Instructional Videos via Dense Captions and Entropy Minimization (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing methods to train models on unlabeled web videos are noisy and temporally misaligned . authors propose a method that adds captions and constrained attention loss to improve performance . |
| Approach: | They propose a method that adds captions from video frames as auxiliary text input to provide visual cues for learning better video and language associations. |
| Outcome: | The proposed method outperforms state-of-the-art methods on video-and-language tasks . it adds captions and constrained attention loss to improve model performance . |
Soft Self-Consistency Improves Language Models Agents (2024.acl-short)
Copied to clipboard
| Challenge: | Current “sample and select” methods rely on majority voting to score answers . however, when tasks have many distinct and valid answers, selection by voting requires a large number of samples. |
| Approach: | They introduce a method that replaces SC's discontinuous scoring with a continuous score computed from model likelihoods to increase selection even when actions are sparsely distributed. |
| Outcome: | The proposed method improves performance and efficiency on long-horizon interactive tasks by replacing SC’s discontinuous scoring with a continuous score computed from model likelihoods. |
Leakage-Adjusted Simulatability: Can Models Generate Non-Trivial Explanations of Their Behavior in Natural Language? (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Existing models that generate NL explanations for tasks have been evaluated on the basis of surface-level similarities to human explanations, both through automatic metrics like BLEU and human evaluations. |
| Approach: | They propose to use a model as a proxy for a human observer to evaluate NL explanations from the model simulatability perspective. |
| Outcome: | The proposed model-generated explanations are evaluated on the basis of surface-level similarities to human explanations, both through automatic metrics like BLEU and human evaluations. |
FENAS: Flexible and Expressive Neural Architecture Search (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Recent approaches to architecture search have shown good improvements in terms of performance with reasonable training speed. |
| Approach: | They propose an algorithm with more activation functions, input edges, and atomic operations to search for architectures that are optimal for given task. |
| Outcome: | The proposed algorithm reproduces well-known LSTM and GRU architectures and initializes with them for finding architectures more efficiently. |
Vokenization: Improving Language Understanding with Contextualized, Visual-Grounded Supervision (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing language pretraining frameworks only take the language context as selfsupervision . current frameworks do not take grounding information from the external visual world . |
| Approach: | They propose a visually-supervised language model that extrapolates multimodal alignments to language-only data by contextually mapping language tokens to related images. |
| Outcome: | The proposed model improves on multiple pure-language tasks. |
ConjNLI: Natural Language Inference Over Conjunctive Sentences (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing stress tests do not consider non-boolean usages of conjunctions and use templates . large-scale pre-trained models do not understand conjunctive semantics well, we find . |
| Approach: | They propose a stress-test for natural language inference over conjunctive sentences where the premise differs from the hypothesis by conjunctions removed, added, or replaced. |
| Outcome: | The proposed stress-test for natural language inference over conjunctive sentences is challenging . it finds that pre-trained models do not understand conjunction semantics well . |
What is More Likely to Happen Next? Video-and-Language Future Event Prediction (2020.emnlp-main)
Copied to clipboard
| Challenge: | Existing models cannot make multimodal commonsense predictions of future events based on video and dialogue . |
| Approach: | They propose a task to predict which event is more likely to happen in a video clip . they use a dataset with 28,726 future event prediction examples from 10,234 videos . |
| Outcome: | The proposed model provides a good starting point but leaves room for future work. |
Expressing Visual Relationships via Language (P19-1)
Copied to clipboard
| Challenge: | Current studies on image captioning focus on single image, but there are no effective models for generating relational captions for two images. |
| Approach: | They propose a language-guided image editing dataset that contains real image pairs with corresponding editing instructions. |
| Outcome: | The proposed model outperforms baseline and existing methods on two datasets. |
On Curriculum Learning for Commonsense Reasoning (2022.naacl-main)
Copied to clipboard
| Challenge: | Recent research suggests that data order can have a significant impact on the performance of finetuned models for natural language understanding. |
| Approach: | They use paced curriculum learning to rank data and sample training mini-batches with increasing levels of difficulty during finetuning. |
| Outcome: | The proposed model improves performance for socialIQA, CosmosQA, CODAH, HellaSwag, WinoGrande in both tuning settings. |
Addressing Semantic Drift in Question Generation for Semi-Supervised Question Answering (D19-1)
Copied to clipboard
| Challenge: | Existing QG models suffer from a “semantic drift” problem, i.e., the semantics of the model-generated question drifts away from the given context and answer. |
| Approach: | They propose two semantics-enhanced rewards obtained from downstream question paraphrasing and question answering tasks to regularize the QG model to generate semantically valid questions. |
| Outcome: | The proposed method achieves state-of-the-art performance w.r.t. traditional evaluation metrics and performs best on QA-based evaluation metrics. |
Improving Cross-Modal Alignment in Vision Language Navigation via Syntactic Information (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing vision language navigation tasks require soft attention over words to locate instructions . a new approach uses syntax information to ground instructions with visual information . |
| Approach: | They propose a vision language navigation agent that utilizes syntax information to enhance alignment between the instruction and the current visual scenes. |
| Outcome: | The proposed agent outperforms the baseline model that does not use syntax information on the Room-to-Room dataset, especially in the unseen environment. |
MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | MEXA is a training-free framework that performs modality- and task-aware aggregation of multiple expert models to enable effective multimodal reasoning across diverse domains. |
| Approach: | MEXA is a training-free framework that performs modality- and task-aware aggregation of multiple expert models. |
| Outcome: | MEXA performs modality- and task-aware aggregation of multiple expert models . it generates interpretable textual reasoning outputs and reasons over them using a Large Reasoning Model (LRM) MEX A consistently delivers performance improvements over strong multimodal benchmarks . |
ChrEn: Cherokee-English Machine Translation for Endangered Language Revitalization (2020.emnlp-main)
Copied to clipboard
| Challenge: | Cherokee is a highly endangered Native American language spoken by the Cherokee people . there are only 2,000 fluent first language Cherokee speakers remaining in the world . |
| Approach: | They propose a Cherokee-English parallel dataset to facilitate machine translation between Cherokee and English. |
| Outcome: | The proposed dataset compares Cherokee-English and English-Cherokee machine translation systems . the results show that the datasets are low-resource and low-cost compared to other datasets . |
LAQuer: Localized Attribution Queries in Content-grounded Generation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing attribution methods associate entire sentences with source documents, which can be overwhelming for users seeking to fact-check specific claims. |
| Approach: | They propose a task that localizes selected spans of generated output to their corresponding source spans, allowing fine-grained and user-directed attribution. |
| Outcome: | The proposed task localizes selected spans of generated output to their corresponding source spans, allowing fine-grained and user-directed attribution. |
ACUEval: Fine-grained Hallucination Evaluation and Correction for Abstractive Summarization (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent-proposed evaluation metrics for large language models have a preference-bias . however, such metrics often lack interpretability and only offer a single score . |
| Approach: | They propose a metric that leverages the power of large language models to perform two sub-tasks: decomposing summaries into atomic content units and validating them against the source document. |
| Outcome: | The proposed metric improves faithfulness scores on three summarization evaluation benchmarks by 3% compared to the next-best metric. |
Integrating Visuospatial, Linguistic, and Commonsense Structure into Story Visualization (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing work on text-to-image synthesis does not explore the use of linguistic structure of the input text. |
| Approach: | They propose to use constituency parse trees to encode structured input and a Transformer-based recurrent architecture to augment commonsense information to generate visual story. |
| Outcome: | The proposed model improves visual quality and consistency of images from a target domain without fine-tuning. |
RACCooN: Versatile Instructional Video Editing with Auto-Generated Narratives (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent video generative models rely on detailed, labor-intensive text prompts for tasks, limiting adaptability for personal/raw videos. |
| Approach: | They propose a video-to-paragraph-to video editing method that supports diverse video editing capabilities, such as removal, addition, and modification, through a unified pipeline. |
| Outcome: | The proposed method supports diverse video editing capabilities, such as removal, addition, and modification, through a unified pipeline. |
Punny Captions: Witty Wordplay in Image Descriptions (N18-2)
Copied to clipboard
| Challenge: | Developing computational models that can produce contextually witty image descriptions is challenging because of the large corpus of sentences that are not available for large scale corpora. |
| Approach: | They propose to use linguistic wordplay, specifically puns, to generate witty image descriptions from large corpus of sentences or encode them via an encoder-decoder neural network architecture. |
| Outcome: | The proposed models perform better than baseline models using human data and show that they are slightly wittier than human-written witty descriptions. |
I like fish, especially dolphins: Addressing Contradictions in Dialogue Modeling (2021.acl-long)
Copied to clipboard
| Challenge: | Recent advances on neural approaches to natural language processing have triggered a resurgent interest on building intelligent open-domain chatbots. |
| Approach: | They propose a dialoguE COntradiction DEtection task and a conversational dataset . they show that their best contradiction detection model correlates well with human judgments . |
| Outcome: | The proposed model is more robust and generalizes well on analysis and out-of-distribution dialogues than standard (unstructured) Transformer models that explicitly hinge on utterance structures are more robust, the study shows . |
EmailSum: Abstractive Email Thread Summarization (2021.acl-long)
Copied to clipboard
| Challenge: | Recent years have brought about interest in the task of summarizing conversation threads. |
| Approach: | They develop an email thread summarization dataset that contains human-annotated short and long email threads over a wide variety of topics. |
| Outcome: | The proposed dataset contains human-annotated short (30 words) and long (100 words) summaries of 2,549 email threads over a wide variety of topics. |
InfoSurgeon: Cross-Media Fine-grained Information Consistency Checking for Fake News Detection (2021.acl-long)
Copied to clipboard
Yi Fung, Christopher Thomas, Revanth Gangi Reddy, Sandeep Polisetty, Heng Ji, Shih-Fu Chang, Kathleen McKeown, Mohit Bansal, Avi Sil
| Challenge: | a novel approach to detect fake news is needed due to training data scarcity . current methods focus on document-level fake news detection using lexical features and semantic embeddings . |
| Approach: | They propose a novel benchmark for fake news detection at the knowledge element level . they propose synthesis method which manipulates knowledge elements to generate noisy training data . |
| Outcome: | The proposed method outperforms the state-of-the-art in detecting misinformation . it yields fine-grained explanations and outperformed the current methods . |
Revealing the Importance of Semantic Retrieval for Machine Reading at Scale (D19-1)
Copied to clipboard
| Challenge: | Recent advances in representation learning have separated progress in both IR and MC . few studies have examined the relationship between retrieval and comprehension at different levels of granularity for development of MRS systems. |
| Approach: | They propose a simple yet effective pipeline system with consideration on hierarchical semantic retrieval at both paragraph and sentence level and their potential effects on the downstream task. |
| Outcome: | The proposed system achieves state-of-the-art on the leaderboard test sets of both FEVER and HOTPOTQA. |
Detecting Linguistic Characteristics of Alzheimer’s Dementia by Interpreting Neural Models (N18-2)
Copied to clipboard
| Challenge: | Current diagnoses often involve lengthy medical evaluations. |
| Approach: | They apply neural models based on CNNs, LSTM-RNNs, and their combination to classify AD and control language samples. |
| Outcome: | The proposed model achieves independent benchmark accuracy for the AD classification task. |
Faithfulness-Aware Decoding Strategies for Abstractive Summarization (2023.eacl-main)
Copied to clipboard
| Challenge: | Existing studies on faithfulness of abstractive summarization have focused on decoding strategies. |
| Approach: | They propose two faithfulness-aware generation methods to further improve faithfulness . they propose to use a distillation approach to generate faithful summaries with greedy decoding . |
| Outcome: | The proposed methods improve faithfulness across two datasets as evaluated by automatic faithfulness metrics and human evaluation. |
Finding a Balanced Degree of Automation for Summary Evaluation (2021.emnlp-main)
Copied to clipboard
| Challenge: | Automated summarization metrics are reliable but often poorly correlated with human judgment. |
| Approach: | They propose a semi-automatic to automatic summary evaluation metrics, following the Pyramid human evaluation method. |
| Outcome: | The proposed metrics are semi-automatic to automatic summary evaluation metrics, following the Pyramid human evaluation method. |
Exploring Continual Learning for Code Generation Models (2023.acl-short)
Copied to clipboard
Prateek Yadav, Qing Sun, Hantian Ding, Xiaopeng Li, Dejiao Zhang, Ming Tan, Parminder Bhatia, Xiaofei Ma, Ramesh Nallapati, Murali Krishna Ramanathan, Mohit Bansal, Bing Xiang
| Challenge: | Large-scale code generation models such as Copilot and CodeT5 are expensive to train and re-train. |
| Approach: | They propose a benchmark for Continual Learning (CL) that covers a wide range of tasks with different input and output programming languages. |
| Outcome: | The proposed method improves on Prompt Pooling with Teacher Forcing, which suffers catastrophic forgetting due to stark distribution shifts in coding tasks. |
Self-Correcting Text-to-Video Generation with Misalignment Detection and Localized Refinement (2026.findings-acl)
Copied to clipboard
| Challenge: | Recent text-to-video models struggle to faith-fully follow text prompts, authors say . authors propose a new refinement framework that detects fine-grained misalignments . |
| Approach: | They propose a video refinement framework that detects fine-grained misalignments . they propose preserving regions that should be preserved rather than regenerated . |
| Outcome: | The proposed framework detects fine-grained misalignments and performs targeted corrections . it preserves correctly generated entities, segments regions across frames, and regenerates problematic regions . |
Mementos: A Comprehensive Benchmark for Multimodal Large Language Model Reasoning over Image Sequences (2024.acl-long)
Copied to clipboard
Xiyao Wang, Yuhang Zhou, Xiaoyu Liu, Hongjin Lu, Yuancheng Xu, Feihong He, Jaehong Yoon, Taixi Lu, Fuxiao Liu, Gedas Bertasius, Mohit Bansal, Huaxiu Yao, Furong Huang
| Challenge: | Multimodal Large Language Models (MLLMs) have demonstrated proficiency in handling a variety of visual-language tasks, but their ability to extrapolate from image sequences has been less investigated. |
| Approach: | They propose a new benchmark to assess MLLMs’ sequential image reasoning abilities. |
| Outcome: | The proposed benchmark features 4,761 diverse image sequences with varying lengths. |
NDH-Full: Learning and Evaluating Navigational Agents on Full-Length Dialogue (2021.emnlp-main)
Copied to clipboard
| Challenge: | Vision-and-Dialogue Navigation is one of the tasks that evaluate the agent’s ability to interact with humans for assistance and navigate based on natural language responses. |
| Approach: | They propose a vision-and-dialogue navigation task which evaluates the agent's ability to interact with humans and navigate based on natural language responses. |
| Outcome: | The proposed model performs well on the Navigation from Dialogue History task, but it is not evaluated by the primary metric Goal Progress. |
Enriching Transformers with Structured Tensor-Product Representations for Abstractive Summarization (2021.naacl-main)
Copied to clipboard
Yichen Jiang, Asli Celikyilmaz, Paul Smolensky, Paul Soulos, Sudha Rao, Hamid Palangi, Roland Fernandez, Caitlin Smith, Mohit Bansal, Jianfeng Gao
| Challenge: | Abstractive summarization is the task of generating a concise summary of input documents . a middle-aged man and a young girl died after they were unable to avoid the plane . |
| Approach: | They propose a model that enriches the original Transformer with a Tensor Product Representation for abstractive summarization. |
| Outcome: | The proposed model outperforms the Transformer and the original TP-Transformer significantly on several datasets. |
Avoiding Reasoning Shortcuts: Adversarial Evaluation, Training, and Model Development for Multi-Hop QA (P19-1)
Copied to clipboard
| Challenge: | In multi-hop question answering, models need to connect multiple pieces of evidence scattered in a long context to answer the question. |
| Approach: | They propose to use a control unit that dynamically attends to the question at different reasoning hops to guide the model's multi-hop reasoning. |
| Outcome: | The proposed model outperforms baseline models but is limited on adversarial test. |
Disentangling Online Chats with DAG-structured LSTMs (2021.starsem-1)
Copied to clipboard
| Challenge: | a number of messaging systems allow fast and synchronous textual communication but they often have a more complicated structure in which independent sub-conversations are interwoven with one another. |
| Approach: | They propose a model that can handle directed acyclic dependencies and integrates structured information into the conversation. |
| Outcome: | The proposed model achieves state-of-the-art status on the task of recovering reply-to relations and is competitive on other disentanglement metrics. |
MAMM-Refine: A Recipe for Improving Faithfulness in Generation with Multi-Agent Collaboration (2025.naacl-long)
Copied to clipboard
| Challenge: | Multi-agent collaboration among models has shown promise in reasoning tasks but is underexplored in long-form generation tasks like summarization and question-answering. |
| Approach: | They propose a multi-agent multi-model reasoning recipe to improve faithfulness through refinement. |
| Outcome: | The proposed method improves faithfulness and error detection on three summarization datasets and on long-form question-answering tasks. |
Branch-Solve-Merge Improves Large Language Model Evaluation and Generation (2024.naacl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) are often used for multi-faceted language generation and evaluation tasks that require complex user constraints or taking into account multiple aspects and criteria. |
| Approach: | They propose a Large Language Model program that consists of branch, solve, and merge modules that are parameterized with specific prompts to the base LLM. |
| Outcome: | The proposed program improves the evaluation correctness and consistency for each LLM by up to 26%, reducing length and pairwise position biases by up 50%, and allowing LLaMA-2-chat to match or outperform GPT-4 on most domains. |
How can NLP Help Revitalize Endangered Languages? A Case Study and Roadmap for the Cherokee Language (2022.acl-long)
Copied to clipboard
| Challenge: | There are an estimated 6000 to 7000 spoken languages in the world, and at least 43% of them are endangered. |
| Approach: | They propose three principles that may help NLP practitioners foster mutual understanding and collaboration with language communities and three ways in which NLP can potentially assist in language education. |
| Outcome: | The proposed methods can be used to enrich Cherokee language resources with machine-in-the-loop processing and to provide language education. |
Evaluating the Factual Consistency of Large Language Models Through News Summarization (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing LLMs generally assign a higher score to factually consistent summaries than to factualally inconsistent summary. |
| Approach: | They propose a benchmark to measure whether large language models prefer factually consistent continuations of inputs. |
| Outcome: | The proposed benchmark compares the scores an LLM assigns to a factually consistent versus a inconsistent summary for an input news article. |
Inducing Transformer’s Compositional Generalization Ability via Auxiliary Sequence Prediction Tasks (2021.emnlp-main)
Copied to clipboard
| Challenge: | Existing neural models lack systematic compositionality in learning symbolic structures . existing models lack this ability in learning symbols, despite being able to understand complex structures. |
| Approach: | They propose to use auxiliary sequence prediction tasks to train a Transformer model to understand compositional symbolic structures of input data. |
| Outcome: | The proposed model improves on the SCAN compositionality challenge, with only 418 (5%) training instances, and achieves 97.8% accuracy on the MCD1 split. |
TVQA+: Spatio-Temporal Grounding for Video Question Answering (2020.acl-main)
Copied to clipboard
| Challenge: | Existing video QA datasets only contain QA pairs without labels for key clips or regions needed to answer the question. |
| Approach: | They propose a framework that grounds evidence in both spatial and temporal domains to answer questions about videos using bounding boxes. |
| Outcome: | The proposed framework can produce interpretable spatio-temporal attention visualizations. |
HoVer: A Dataset for Many-Hop Fact Extraction And Claim Verification (2020.findings-emnlp)
Copied to clipboard
| Challenge: | a dataset for many-hop evidence extraction and fact verification challenges models to reason with information from multiple Wikipedia articles. |
| Approach: | They propose a dataset for many-hop evidence extraction and fact verification . they challenge models to extract facts from Wikipedia articles relevant to a claim . |
| Outcome: | The proposed dataset shows that state-of-the-art models degrade as the number of reasoning hops increases. |
Prompting Vision-Language Models For Aspect-Controlled Generation of Referring Expressions (2024.findings-naacl)
Copied to clipboard
Danfeng Guo, Sanchit Agarwal, Arpit Gupta, Jiun-Yu Kao, Emre Barut, Tagyoung Chung, Jing Huang, Mohit Bansal
| Challenge: | Referring Expression Generation (REG) is the task of generating a descriptive caption that uniquely identifies a given target in the scene. |
| Approach: | They propose an Aspect-Controlled REG task which requires generating a referring expression conditioned on the input aspect(s) by changing the input input such as color, location, action etc. |
| Outcome: | The proposed model beats all prior works in the CIDEr score and achieves comparable performance to training with 100% of real data. |
Methods for Measuring, Updating, and Visualizing Factual Beliefs in Language Models (2023.eacl-main)
Copied to clipboard
Peter Hase, Mona Diab, Asli Celikyilmaz, Xian Li, Zornitsa Kozareva, Veselin Stoyanov, Mohit Bansal, Srinivasan Iyer
| Challenge: | Pretrained language models store a large amount of factual information that can be elicited by prompting or finetuning. |
| Approach: | They propose methods to measure model factual beliefs and update incorrect beliefs in models . they propose a new visualization tool that shows relationships between stored model beliefs . |
| Outcome: | The proposed methods improve models' consistency and accuracy, the authors show . their methods outperform existing methods in more difficult settings, the paper shows . |
Improving and Simplifying Pattern Exploiting Training (2021.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have shown that pre-trained language models can learn well when primed with only a few labeled examples. |
| Approach: | They propose a method that uses task-specific unlabeled data to provide denser supervision during fine-tuning. |
| Outcome: | The proposed approach outperforms GPT-3 on SuperGLUE without any unlabeled data. |
Proposition-Level Clustering for Multi-Document Summarization (2022.naacl-main)
Copied to clipboard
Ori Ernst, Avi Caciularu, Ori Shapira, Ramakanth Pasunuru, Mohit Bansal, Jacob Goldberger, Ido Dagan
| Challenge: | Existing methods focused on clustering sentences to indicate information saliency and avoid redundancy. |
| Approach: | They propose to group together sub-sentential propositions to generate a representative sentence for each cluster via text fusion. |
| Outcome: | The proposed method improves over the previous state-of-the-art method in the DUC 2004 and TAC 2011 datasets, both in automatic ROUGE scores and human preference. |
TVQA: Localized, Compositional Video Question Answering (D18-1)
Copied to clipboard
| Challenge: | Recent studies have focused on image-based question-answering (QA) tasks, but little has been done on video-based QA. |
| Approach: | They present a large-scale video QA dataset based on 6 popular TV shows . they provide analysis of the new dataset and trainable neural network framework . |
| Outcome: | The proposed dataset includes 152,545 QA pairs from 21,793 clips spanning over 460 hours of video. |
Video-Skill-CoT: Skill-based Chain-of-Thoughts for Domain-Adaptive Video Reasoning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for chain-of-thought reasoning fail to adapt to domain-specific skills over video content. |
| Approach: | They propose a framework that automatically constructs and leverages skill-aware CoT supervisions for domain-adaptive video reasoning. |
| Outcome: | The proposed framework outperforms strong baselines on three video understanding benchmarks. |
Fine-grained Image Captioning with CLIP Reward (2022.findings-naacl)
Copied to clipboard
| Challenge: | Modern image captioning models are usually trained with text similarity objectives . reference captions often describe only the most salient objects in images . |
| Approach: | They propose to use CLIP to calculate multi-modal similarity and use it as a reward function . they propose a simple finetuning strategy to improve grammar that does not require extra text annotation. |
| Outcome: | The proposed model generates more distinctive captions than the CIDEroptimized model on text-to-image retrieval and fineCapEval. |
RotBench: Evaluating Multi-modal Large Language Models on Identifying Image Rotation (2026.eacl-long)
Copied to clipboard
| Challenge: | Multimodal Large Language Models (MLLMs) can identify the orientation of input images rotated 0°, 90°, 180°, and 270°. |
| Approach: | They propose a manually-filtered benchmark to evaluate MLLMs' ability to accurately identify rotation in input images. |
| Outcome: | The proposed model improves on the 'rotational cues' of 360° and 180° images, but not 90° and 270° rotations. |
Continual and Multi-Task Architecture Search (P19-1)
Copied to clipboard
| Challenge: | Recent studies have shown that architecture search can improve performance on language modeling and image classification tasks with reasonable training speed. |
| Approach: | They propose a continual architecture search approach that continually evolves the model parameters during sequential training of several tasks without losing performance on previously learned tasks. |
| Outcome: | The proposed approach improves language modeling and image classification with reasonable training speed and a weight-sharing strategy. |
Inducing Systematicity in Transformers by Attending to Structurally Quantized Embeddings (2024.acl-long)
Copied to clipboard
| Challenge: | Empirically, we show SQ-Transformer achieves stronger compositional generalization than the vanilla Transformer on low-complexity datasets. |
| Approach: | They propose a Transformer that explicitly encourages systematicity in the embeddings and attention layers even with low-complexity data. |
| Outcome: | Empirically, the proposed model achieves stronger compositional generalization than the vanilla Transformer on low-complexity datasets. |
Hierarchical and Dynamic Prompt Compression for Efficient Zero-shot API Usage (2024.findings-eacl)
Copied to clipboard
| Challenge: | Long prompts present a significant challenge for LLM-based systems that learn to use unseen APIs directly in-context from their documentation. |
| Approach: | They propose a method that learns to compress a prompt into a few "gist token" activations during finetuning but is ineffective in compressing API documentation. |
| Outcome: | The proposed model keeps the simplicity, efficiency, and large compression factor (20x on SGD) of the gist token approach while achieving significantly better accuracy. |
Localizing Factual Inconsistencies in Attributable Text Generation (2026.tacl-1)
Copied to clipboard
Arie Cattan, Paul Roit, Shiyue Zhang, David Wan, Roee Aharoni, Idan Szpektor, Mohit Bansal, Ido Dagan
| Challenge: | Existing methods for detecting hallucinations in model-generated texts fail to pinpoint errors. |
| Approach: | They propose a formalism for localizing factual inconsistencies in attributable text generation . they propose to decompose the generated text into simple question-answer pairs . |
| Outcome: | The proposed method achieves substantial inter-annotator agreement while achieving a substantial consistency score. |
Language Models Identify Ambiguities and Exploit Loopholes (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing models that exploit loopholes identify and reason about ambiguity and conflicting goals, presenting a potential safety risk. |
| Approach: | They propose to study the responses of large language models to loopholes by examining ambiguity and pragmatics in LLMs. |
| Outcome: | The proposed models can identify ambiguities and exploit loopholes to satisfy their given goals as opposed to the goals of the user. |
Identify, Align, and Integrate: Matching Knowledge Graphs to Commonsense Reasoning Tasks (2021.eacl-main)
Copied to clipboard
| Challenge: | Empirically, we investigate KG matches for the SocialIQA, Physical IQA, and MCScript2.0 datasets with 3 diverse KGs: ATOMIC (SIQA), ConceptNet (Speer et al., 2017), and an automatically constructed instructional KG based on WikiHow (Ostermann e., 2019b). |
| Approach: | They propose a method to assess how well a candidate KG can fill in knowledge gaps for a given task by using commonsense probes. |
| Outcome: | Empirically, we show that the proposed KG-to-task match is a good match for socialIQA, physical IQA, and MCScript2.0 datasets with 3 diverse KGs: ATOMIC, ConceptNet, and an instructional KG based on WikiHow. |
The Unreasonable Effectiveness of Easy Training Data for Hard Tasks (2024.acl-long)
Copied to clipboard
| Challenge: | Existing pretrained language models perform well on hard data, but hard data is noisier and costlier to collect. |
| Approach: | They propose to use in-context learning, linear classifier heads, and QLoRA to show that pretrained language models generalize relatively well from easy to hard data. |
| Outcome: | The proposed model generalizes well from easy to hard data even better than oracle models finetuned on hard data. |
FastIF: Scalable Influence Functions for Efficient Model Interpretation and Debugging (2021.emnlp-main)
Copied to clipboard
| Challenge: | despite popularity of influence functions, their computational cost does not scale well with model and training data size. |
| Approach: | They propose a fast parallel variant that approximates the “influences” of training data-points for test predictions. |
| Outcome: | The proposed method achieves about 80X speedup while being highly correlated with the original influence values. |
GrAInS: Gradient-based Attribution for Inference-Time Steering of LLMs and VLMs (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for fine-tuning large language models often ignore token-level causal influence and underutilize model logits. |
| Approach: | They propose a novel approach that uses a gradient-based approach to identify influential tokens and construct directional steering vectors based on their contribution to preferred over dispreferred outputs. |
| Outcome: | The proposed approach outperforms fine-tuning and prior steering methods on both LLM and VLM tasks without degrading fluency or general capabilities. |
MAgICoRe: Multi-Agent, Iterative, Coarse-to-Fine Refinement for Reasoning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Excessive refinement can cause over-correction and reduce performance, authors say . they say MAgICoRe is a framework for multi-agent iteration for coarse-to-fine refinement . |
| Approach: | They propose a framework for multi-agent iteration for coarse-to-fine refinement that reduces excessive refinement by categorizing problems as easy or hard. |
| Outcome: | The proposed framework beats Self-Consistency by 3.4%, Best-of-k by 3.2%, and Self-Refine by 4.0% on Llama-3-8B and GPT- 3.5. |
MixCE: Training Autoregressive Language Models by Mixing Forward and Reverse Cross-Entropies (2023.acl-long)
Copied to clipboard
| Challenge: | Autoregressive language models are trained by minimizing the cross-entropy of the model distribution Q relative to the data distribution P. However, these systems still struggle in many openended generation settings, where they are asked to produce a long text following a short prompt. |
| Approach: | They propose to combine forward and reverse cross-entropy to train autoregressive language models by minimizing the cross-Entropy of the model distribution Q relative to the data distribution P. |
| Outcome: | The proposed model overgeneralizes and produces non-human-like text without complex decoding strategies. |
Interactive Query-Assisted Summarization via Deep Reinforcement Learning (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing systems that can perform interactive summarization cannot ingest the full document set or operate at sufficient speed for interactivity. |
| Approach: | They propose two deep reinforcement learning models for interactive summarization task . they use interactive session state and history to refrain from redundancy . |
| Outcome: | The proposed model improves informativeness while preserving positive user experience. |
Evaluating Very Long-Term Conversational Memory of LLM Agents (2024.acl-long)
Copied to clipboard
| Challenge: | Existing studies on long-term open-domain dialogues focus on evaluating responses within contexts spanning no more than five chat sessions. |
| Approach: | They propose a machine-human pipeline to generate very long-term dialogues by leveraging LLMs and retrieval augmented generation techniques. |
| Outcome: | The proposed pipeline generates very long-term dialogues using LLMs and RAGs . the generated conversations are verified and edited by human annotators for long-range consistency and grounding to the event graphs. |
Instruction Tuning with and without Context: Behavioral Shifts and Downstream Impact (2026.eacl-long)
Copied to clipboard
Hyunji Lee, Seunghyun Yoon, Yunjae Won, Hanseok Oh, Geewook Kim, Trung Bui, Franck Dernoncourt, Elias Stengel-Eskin, Mohit Bansal, Minjoon Seo
| Challenge: | Prior work on instruction tuning datasets combined these data types without examining their distinct effects. |
| Approach: | They investigate how training LLMs with or without context affects model behavior and performance . they find that using context-augmented data as the backbone for vision-language models reduces hallucination . |
| Outcome: | The proposed training with context-augmented data reduces hallucination and improves grounding in the visual domain. |
An Empirical Survey of Data Augmentation for Limited Data Learning in NLP (2023.tacl-1)
Copied to clipboard
| Challenge: | Existing methods for enhancing data efficiency in limited labeled data are limited. |
| Approach: | They propose to use data augmentation methods to increase the efficiency of limited data learning in NLP. |
| Outcome: | The proposed methods perform well on topics/news classification, inference tasks, paraphrasing tasks, and single-sentence tasks. |
Are Hard Examples also Harder to Explain? A Study with Human and Model-Generated Explanations (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent work on explainable NLP has shown that few-shot prompting can enable large pre-trained language models (LLMs) to generate grammatical and factual explanations for data labels. |
| Approach: | They propose to use few-shot prompting to generate grammatical and factual explanations for data labels by varying the hardness of the test samples and in-context samples to investigate the link between explainability and sample hardness. |
| Outcome: | The proposed model can generate grammatical and factual explanations for data labels with few-shot prompting. |
Automatically Learning Data Augmentation Policies for Dialogue Tasks (D19-1)
Copied to clipboard
| Challenge: | Recent advances in automatic data augmentation have focused on computer vision tasks where it is easy to apply imperceptible perturbations without changing an image’s semantic meaning. |
| Approach: | They adapt AutoAugment to automatically discover effective perturbation policies for natural language processing (NLP) tasks such as dialogue generation. |
| Outcome: | The proposed algorithm reduces data-level model bias by using a controller trained on the target task. |
Commonsense for Generative Multi-Hop Question Answering Tasks (D18-1)
Copied to clipboard
| Challenge: | Reading comprehension QA tasks have seen a recent surge in popularity, yet most work has focused on fact-finding extractive QA. |
| Approach: | They propose a multi-hop generative task that uses a pointer-generator decoder to synthesize disjoint pieces of information within the context to generate an answer. |
| Outcome: | The proposed model performs better than previous generative models and is competitive with current state-of-the-art span prediction models. |
ReCEval: Evaluating Reasoning Chains via Correctness and Informativeness (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods focus on whether the reasoning chain leads to the correct conclusion, but this view may confound reasoning quality with other spurious shortcuts to predict the answer. |
| Approach: | They propose a framework that evaluates reasoning chains via two key properties: (1) correctness, i.e., each step makes a valid inference based on information contained within the step, preceding steps, and input context, and (2) informativeness, respectively. |
| Outcome: | The proposed framework evaluates reasoning chains via two key properties: (1) correctness, i.e., each step makes a valid inference based on information contained within the step, preceding steps, and input context, and (2) informativeness, which is helpful towards deriving the generated answer. |
ArraMon: A Joint Navigation-Assembly Instruction Interpretation Task in Dynamic Environments (2020.findings-emnlp)
Copied to clipboard
| Challenge: | embodied agents are expected to perform specific tasks after reaching the destination . a novel vision-and-language navigation task is designed to support this task . |
| Approach: | They combine vision-and-language navigation, assembling objects and object referring expression comprehension to create a joint navigation-and assembly task. |
| Outcome: | The proposed task is based on vision-and-language navigation and assembly . it uses human-written navigation and assembling instructions and ground truth trajectories . the large model-human performance gap shows that the task is challenging and wide scope for future work. |
Unified Embeddings for Multimodal Retrieval via Frozen LLMs (2024.findings-eacl)
Copied to clipboard
| Challenge: | Unified Embeddings for Multimodal Retrieval (UniMuR) is a simple but effective approach that embeds multimodal inputs and retrieves visual and textual outputs via frozen Large Language Models (LLMs). |
| Approach: | They propose a method that embeds multimodal inputs and retrieves visual and textual outputs via frozen Large Language Models (LLMs). |
| Outcome: | The proposed method significantly reduces LLM’s modality bias towards generating text-only outputs and achieves strong image/text retrieval ability. |
Enhancing Multi-Document Summarization with Cross-Document Graph-based Information Extraction (2023.eacl-main)
Copied to clipboard
| Challenge: | Information extraction (IE) and summarization (summarization) are closely related, but both aims to abstract the most salient information into a generated text summary. |
| Approach: | They propose to use structured IE graphs to enhance the abstractive summarization task by using cross-document IE output to incorporate an alignment loss between IE nodes and their text spans to reduce inconsistencies. |
| Outcome: | The proposed model can generate summaries that are more factual while not losing abstractiveness. |
Exclusive Supermask Subnetwork Training for Continual Learning (2023.findings-acl)
Copied to clipboard
| Challenge: | Continual Learning (CL) methods focus on accumulating knowledge over time while preventing catastrophic forgetting. |
| Approach: | They propose a CL method that finds a supermask for each new task that keeps or removes each weight to produce a subnetwork. |
| Outcome: | The proposed method outperforms strong previous methods on NLP and Vision domains while preventing forgetting. |
Efficient Few-Shot Fine-Tuning for Opinion Summarization (2022.findings-naacl)
Copied to clipboard
| Challenge: | Abstractive summarization models are typically pre-trained on large amounts of generic texts . large annotated datasets of reviews paired with reference summaries are not available . |
| Approach: | They propose a few-shot method which uses adapters to store in-domain knowledge . they pre-train adapters on unannotated customer reviews and fine-tune them on annotated datasets . |
| Outcome: | The proposed method can store in-domain knowledge and improves on large annotated reviews . it improves coherence and redundancies on the Amazon and Yelp datasets . |
Analysis of Tree-Structured Architectures for Code Generation (2021.findings-acl)
Copied to clipboard
| Challenge: | Code generation is the task of generating code snippets from input user specifications written in natural language (NL). |
| Approach: | They evaluate the significance of input parse trees for code generation by using constituency-based parsers as input and an abstract syntax tree as the target. |
| Outcome: | The proposed models on a Python-based code generation dataset and a semantic parsing dataset show that constituency trees encoded using a structure-aware model improve performance. |
The Power of Summary-Source Alignments (2024.findings-acl)
Copied to clipboard
Ori Ernst, Ori Shapira, Aviv Slobodkin, Sharon Adar, Mohit Bansal, Jacob Goldberger, Ran Levy, Ido Dagan
| Challenge: | Multi-document summarization (MDS) is a challenging task, often decomposed to subtasks of salience and redundancy detection, followed by text generation. |
| Approach: | They propose to extend the summary-source alignment framework by applying it at the more fine-grained proposition span level and annotating alignment manually in a multi-document setup. |
| Outcome: | The proposed framework can yield several datasets for at least six different tasks. |
Generating Summaries with Controllable Readability Levels (2023.emnlp-main)
Copied to clipboard
| Challenge: | Current text generation approaches focus on a specific readability level, resulting in texts that are not customized to readers’ proficiency levels. |
| Approach: | They propose to generate summaries with fine-grained control over their readability by using instruction-based readability control, reinforcement learning and lookahead to estimate readability of upcoming decoding steps. |
| Outcome: | The generated summaries with different readability levels were compared with previous methods that focus on a specific readability level (e.g., lay summarization) and a lookahead approach significantly improved readability control on news summarizing. |
MURMUR: Modular Multi-Step Reasoning for Semi-Structured Data-to-Text Generation (2023.findings-acl)
Copied to clipboard
| Challenge: | MURMUR generates highly faithful and correct reasoning paths that lead to 26% more logically consistent summaries on LogicNLG compared to direct prompting. |
| Approach: | They propose a neuro-symbolic modular approach to text generation from semi-structured data with multi-step reasoning that generates reasoning paths using neural and symbolic modules with specific linguistic and logical skills. |
| Outcome: | The proposed method improves on two data-to-text generation tasks, while achieving comparable performance to fine-tuned GPT-2 on out-of-domain data. |
DORB: Dynamically Optimizing Multiple Rewards with Bandits (2020.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in end-to-end neural networks-based approaches have shown wide success in sequence generation tasks. |
| Approach: | They propose to optimize multiple metric rewards simultaneously using a multi-armed bandit approach . they empirically show the effectiveness of their approaches via various automatic metrics and human evaluation . |
| Outcome: | The proposed approach improves on question generation and data-to-text generation using a bandit approach. |
PaperRobot: Incremental Draft Generation of Scientific Ideas (P19-1)
Copied to clipboard
| Challenge: | a paper robot can read existing papers and create new nodes or links in the knowledge graphs. |
| Approach: | They propose to automate the creation of new ideas by predicting links from the background KGs. |
| Outcome: | The proposed paper automates three tasks: read existing papers, create new ideas, predict links . the paper generated abstracts, conclusion and future work sections, and new titles are chosen over human-written ones up to 30%, 24% and 12% of the time. |
Evaluating Explainable AI: Which Algorithmic Explanations Help Users Predict Model Behavior? (2020.acl-main)
Copied to clipboard
| Challenge: | a new study examines the impact of algorithmic explanations on simulatability of machine learning models . a model is simulatable when a person can predict its behavior on new inputs . |
| Approach: | They conduct human subject tests to isolate effect of algorithmic explanations on simulatability . they find ratings of explanations are not predictive of how helpful they are . |
| Outcome: | The results provide the first reliable estimates of how explanations influence simulatability . they show that ratings are not predictive of how helpful explanations are . |
Improving Generation and Evaluation of Visual Stories via Semantic Consistency (2021.naacl-main)
Copied to clipboard
| Challenge: | Story visualization is an underexplored task that requires a generative model to generate images . prior work has focused on image generation but there is room for improvement . |
| Approach: | They propose to add a dual learning framework to reinforce semantic alignment between story and generated images and a copy-transform mechanism to model sequentially-consistent story visualization. |
| Outcome: | The proposed models outperform text-to-image synthesis models on the story visualization task . the proposed models also improve visual quality, coherence and relevance . |
CLEAR: Improving Vision-Language Navigation with Cross-Lingual, Environment-Agnostic Representations (2022.findings-naacl)
Copied to clipboard
| Challenge: | Using multilingual instructions to learn a better cross-lingual representation is challenging for multilingual agents. |
| Approach: | They propose to use multilingual instructions to learn a shared cross-lingual language representation for the three languages in a Room-Across-Room dataset. |
| Outcome: | The proposed model improves on the room-Across-room and vision-and-dialogue navigation tasks by maximizing similarity between semantically aligned image pairs from different environments. |
multiPRover: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning (2021.naacl-main)
Copied to clipboard
| Challenge: | Existing work to generate proof graphs for formal reasoning over explicit knowledge is not unique and there may be multiple ways of reaching the correct answer. |
| Approach: | They propose to generate multiple proof graphs for reasoning over natural language rules and facts . they propose to combine all proofs and exploit correlations between them . |
| Outcome: | The proposed model outperforms PRover on multiple gold proofs on synthetic, zero-shot, and human-paraphrased datasets. |
iFacetSum: Coreference-based Interactive Faceted Summarization for Multi-Document Exploration (2021.emnlp-demo)
Copied to clipboard
Eran Hirsch, Alon Eirew, Ori Shapira, Avi Caciularu, Arie Cattan, Ori Ernst, Ramakanth Pasunuru, Hadar Ronen, Mohit Bansal, Ido Dagan
| Challenge: | iFS provides a faceted navigation scheme that provides abstractive summaries for the user’s selections. |
| Approach: | They propose a web application that integrates interactive summarization and faceted search to provide a faceted navigation scheme that yields abstractive summaries for the user's selections. |
| Outcome: | The proposed system provides a comprehensive overview as well as particular details regard-ing subtopics of interest. |
Learning to Navigate Unseen Environments: Back Translation with Environmental Dropout (N19-1)
Copied to clipboard
| Challenge: | Existing approaches perform significantly worse in unseen environments compared to seen ones. |
| Approach: | They propose to use a ‘environmental dropout’ method to generate unseen triplets to generate new paths and instructions to generalize the agent. |
| Outcome: | The proposed agent outperforms the state-of-the-art approaches on the private unseen test set and is ranked top on the leaderboard. |
Non-Sequential Graph Script Induction via Multimedia Grounding (2023.acl-long)
Copied to clipboard
| Challenge: | Existing scripts for everyday tasks are presented in a linear manner, which does not reflect the flexibility displayed by people executing tasks in real life. |
| Approach: | They propose to use loosely aligned videos to train a non-sequential graph script induction task by using a multimodal framework to ground procedural videos to WikiHow textual steps. |
| Outcome: | The proposed model outperforms the WikiHow linear baseline by 48.76% . it can predict future steps given a partial step sequence and generate explicit graph scripts . |
Enhancing Knowledge Selection for Grounded Dialogues via Document Semantic Graphs (2022.naacl-main)
Copied to clipboard
| Challenge: | Existing conversation models treat knowledge selection as a sentence ranking problem where each sentence is handled individually, ignoring the internal semantic connection between sentences. |
| Approach: | They propose to automatically convert background knowledge documents into document semantic graphs and perform knowledge selection over such graphs. |
| Outcome: | The proposed model improves on the knowledge selection task and the response generation task on HollE and generalizes on unseen topics in WoW. |
Dynamic Multi-Level Multi-Task Learning for Sentence Simplification (C18-1)
Copied to clipboard
| Challenge: | Sentence simplification is the task of improving readability and understandability of an input text. |
| Approach: | They propose a strong pointer-copy mechanism based sequence-to-sequence sentence simplification model and a novel ‘multi-level’ soft sharing approach where each auxiliary task shares different (higher versus lower) level layers of the model. |
| Outcome: | The proposed model outperforms competing simplification systems in SARI and FKGL automatic metrics, and human evaluation. |
Incorporating Background Knowledge into Video Description Generation (D18-1)
Copied to clipboard
| Challenge: | Existing methods for video captioning focus on generating generic descriptions that lack contextual knowledge. |
| Approach: | They propose a method that uses video meta-data to retrieve topically related news documents for a video and extracts the events and named entities from these documents. |
| Outcome: | The proposed model is based on a news video dataset and is evaluated on it. |
Debiasing Multimodal Models via Causal Information Minimization (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for debiasing multimodal models use approximate heuristics to represent the biases, such as shallow features from early stages of training or unimodal features for multimodal tasks like VQA, which may not be accurate. |
| Approach: | They propose a method that leverages causally-motivated information minimization to learn the confounder representations of a causal graph for multimodal data. |
| Outcome: | The proposed method improves out-of-distribution performance on multiple multimodal datasets without sacrificing in-distance performance. |
Adversarial Augmentation Policy Search for Domain and Cross-Lingual Generalization in Reading Comprehension (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Reading comprehension models often overfit to nuances of training datasets and fail at adversarial evaluation. |
| Approach: | They propose a method that introduces multiple points of confusion within the context and shows dependence on insertion location of the distractor. |
| Outcome: | The proposed methods improve robustness against adversarial evaluation but weak generalization to the source domain and new domains and languages. |
PRover: Proof Generation for Interpretable Reasoning over Rules (2020.emnlp-main)
Copied to clipboard
| Challenge: | Recent work shows that transformers can act as “soft theorem provers” by answering questions over explicitly provided knowledge in natural language. |
| Approach: | They propose a transformer-based model that answers binary questions over rule-bases and generates the corresponding proofs. |
| Outcome: | The proposed model generates proofs with an accuracy of 87% while maintaining or improving performance on the QA task. |
Mutual Exclusivity Training and Primitive Augmentation to Induce Compositionality (2022.emnlp-main)
Copied to clipboard
| Challenge: | Recent datasets expose the lack of systematic generalization ability in standard sequence-to-sequence models. |
| Approach: | They propose two techniques to address the lack of systematic generalization ability in standard sequence-to-sequence models by mutual exclusivity training and prim2primX data augmentation. |
| Outcome: | The proposed methods improve on two widely-used compositionality datasets. |
Robust Machine Comprehension Models via Adversarial Training (N18-2)
Copied to clipboard
| Challenge: | Existing models for the Stanford Question Answering Dataset suffer from a 50% decrease in F1 score during adversarial evaluation based on AddSent. |
| Approach: | They propose an alternative adversary-generation algorithm, AddSentDiverse, that significantly increases the variance within the adversarial training data by providing effective examples that punish the model for making certain superficial assumptions. |
| Outcome: | The proposed algorithm can achieve a 36.5% increase in F1 score while maintaining performance on the regular SQuAD task. |
ChrEnTranslate: Cherokee-English Machine Translation Demo with Quality Estimation and Corrective Feedback (2021.acl-demo)
Copied to clipboard
| Challenge: | a new study shows that machine translation models can translate fragments of the source sentence but make major mistakes. |
| Approach: | They propose an online machine translation demonstration system for translation between English and an endangered language Cherokee. |
| Outcome: | The proposed system achieves state-of-the-art translation performance and improves quality estimation . the proposed system can translate between English and an endangered language Cherokee . |
Social Commonsense for Explanation and Cultural Bias Discovery (2023.eacl-main)
Copied to clipboard
| Challenge: | Social commonsense contains many human biases due to social and cultural influence. |
| Approach: | They aim to identify cultural biases in data that strongly influence model decisions . they use social commonsense knowledge to augment large-scale language models . |
| Outcome: | The proposed method shows that social commonsense knowledge can explain model behavior on two social tasks. |
Glider: Global and Local Instruction-Driven Expert Router (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for routing-based expert models favor generalization over performance on held-in tasks. |
| Approach: | They propose a global and local instruction driven expert router that leverages recent LLMs' semantic reasoning capabilities to generate task-specific instructions from the input query. |
| Outcome: | The proposed method improves held-in performance while maintaining strong generalization on held-out tasks. |
GrIPS: Gradient-free, Edit-based Instruction Search for Prompting Large Language Models (2023.eacl-main)
Copied to clipboard
| Challenge: | Recent work aimed to improve task performance of large language models by rewriting or tuning them manually, but manual rewrite is time-consuming and requires subjective interpretation. |
| Approach: | They propose a gradient-free, edit-based search approach for improving task instructions for large language models. |
| Outcome: | The proposed approach outperforms manual rewriting and purely example-based prompts while allowing for API-based tuning. |