Papers by Jason Zhang
Task-Oriented Dialogue as Dataflow Synthesis (2020.tacl-1)
Copied to clipboard
Jacob Andreas, John Bufe, David Burkett, Charles Chen, Josh Clausman, Jean Crawford, Kate Crim, Jordan DeLoach, Leah Dorner, Jason Eisner, Hao Fang, Alan Guo, David Hall, Kristin Hayes, Kellie Hill, Diana Ho, Wendy Iwaszuk, Smriti Jha, Dan Klein, Jayant Krishnamurthy, Theo Lanman, Percy Liang, Christopher H. Lin, Ilya Lintsbakh, Andy McGovern, Aleksandr Nisnevich, Adam Pauls, Dmitrij Petters, Brent Read, Dan Roth, Subhro Roy, Jesse Rusak, Beth Short, Div Slomin, Ben Snyder, Stephon Striplin, Yu Su, Zachary Tellman, Sam Thomson, Andrei Vorobev, Izabela Witoszko, Jason Wolfe, Abby Wray, Yuchen Zhang, Alexander Zotov
| Challenge: | Existing approaches to task-oriented dialogue represent dialogue state as a dataflow graph . microsoft's SMCalFlow dataset features complex dialogues about events, weather, places, and people . |
| Approach: | They propose a dataflow graph-based dialogue agent that maps each user utterance to a program that extends this graph. |
| Outcome: | The proposed framework improves representability and predictability in natural dialogues . it uses dataflow graphs and metacomputation to map user intents to a program . |
Rethinking Offensive Text Detection as a Multi-Hop Reasoning Problem (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing methods of offensive text detection perform poorly when asked to detect implicitly offensive statements . a dataset based on SLIGHT provides a framework for implicit offensive text identification . |
| Approach: | They propose a dataset to support the task of implicit offensive text detection in dialogues . they show that reasoning is crucial for understanding this broader class of offensive utterances - SLIGHT . |
| Outcome: | The proposed model achieves 11% accuracy in implicit offensive text detection tasks . the proposed model can be used to identify toxic speech in specific domains . |
Multi-Domain Goal-Oriented Dialogues (MultiDoGO): Strategies toward Curating and Annotating Large Scale Dialogue Data (D19-1)
Copied to clipboard
| Challenge: | a large number of goal-oriented dialogue datasets are limited in their size, linguistic diversity, domain coverage, or annotation granularity. |
| Approach: | They propose a multi-domain goal-oriented dialogue dataset that uses a crowd-sourced worker and a trained annotator to curate and annotate large scale data. |
| Outcome: | The proposed dataset is 8 times the size of the largest comparable dialogue dataset available to the public. |
Value-Agnostic Conversational Semantic Parsing (2021.acl-long)
Copied to clipboard
Emmanouil Antonios Platanios, Adam Pauls, Subhro Roy, Yuchen Zhang, Alexander Kyte, Alan Guo, Sam Thomson, Jayant Krishnamurthy, Jason Wolfe, Jacob Andreas, Dan Klein
| Challenge: | Existing models rely on rich representations of dialogue history that include all previously generated components of the output. |
| Approach: | They propose a model that abstracts over values to focus prediction on type- and function-level context. |
| Outcome: | The proposed model outperforms baseline models by 7.3% and 10.6% on SMCalFlow and TreeDST datasets. |
Intermediate-Task Transfer Learning with Pretrained Language Models: When and Why Does It Work? (2020.acl-main)
Copied to clipboard
Yada Pruksachatkun, Jason Phang, Haokun Liu, Phu Mon Htut, Xiaoyi Zhang, Richard Yuanzhe Pang, Clara Vania, Katharina Kann, Samuel R. Bowman
| Challenge: | Unsupervised pretraining has recently pushed the state of the art on many natural language understanding tasks. |
| Approach: | They perform a large-scale survey on a pretrained RoBERTa model with 110 intermediate-target task combinations and 25 probing tasks to reveal the specific skills that drive transfer. |
| Outcome: | The proposed model is trained on 110 intermediate-target task combinations and compared with 25 probing tasks to reveal the specific skills that drive transfer. |
Ask an Expert: Leveraging Language Models to Improve Strategic Reasoning in Goal-Oriented Dialogue Models (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing dialogue models may encounter scenarios which are not well-represented in the training data and produce unnatural, inappropriate, or unhelpful responses. |
| Approach: | They propose a framework in which a model is trained with access to an "expert" they propose to optimize the model to selectively utilize (or ignore) advice given context and dialogue history. |
| Outcome: | The proposed framework improves quality across all expert sizes and with fewer parameters than the dialogue model itself. |
Mind the Gap Between Conversations for Improved Long-Term Dialogue Generation (2023.findings-emnlp)
Copied to clipboard
| Challenge: | a gap between conversations can be weeks, months or years, and dialogue systems which do not explicitly model time may generate unnatural responses. |
| Approach: | They propose to model the passage of time between conversations by exposing time information to a multi-session dialogue dataset and comparing different representations of time and event progress. |
| Outcome: | The proposed model is based on a real-time dataset showing that it can predict topics and information gained from conversations over a long time span. |
TReMu: Towards Neuro-Symbolic Temporal Reasoning for LLM-Agents with Memory in Multi-Session Dialogues (2025.findings-acl)
Copied to clipboard
| Challenge: | Temporal reasoning in multi-session dialogues presents a significant challenge which has been under-studied in previous temporal reasoning benchmarks. |
| Approach: | They propose to augment LoCoMo dialogues and create multi-choice QAs to construct a temporal reasoning evaluation task and a framework to enhance temporal thinking capabilities of LLM-agents. |
| Outcome: | The proposed framework significantly improves temporal reasoning performance compared to baseline methods, raising from 29.83 on GPT-4o via standard prompting to 77.67 via the proposed framework. |
One Panel Does Not Fit All: Case-Adaptive Multi-Agent Deliberation for Clinical Prediction (2026.acl-srw)
Copied to clipboard
| Challenge: | Existing single-agent strategies sample from one role-conditioned distribution, and multi-agend frameworks use fixed roles with flat majority voting, discarding the diagnostic signal in disagreement. |
| Approach: | They propose a case-adaptive multi-agent panel where an attending-physician agent dynamically assembles a specialist panel tailored to each case’s diagnostic uncertainty. |
| Outcome: | The proposed model outperforms baseline models on diagnostic prediction and brief hospital course generation using MIMIC-IV. |
PAWS: Paraphrase Adversaries from Word Scrambling (N19-1)
Copied to clipboard
| Challenge: | Existing paraphrase identification datasets lack sentence pairs with high word overlap without being paraphrases. |
| Approach: | They propose a workflow for generating pairs of sentences with high word overlap . they use controlled word swapping and back translation followed by fluency and paraphrase judgments . |
| Outcome: | The proposed dataset has 108,463 well-formed paraphrase and non-paraphrase pairs with high lexical overlap. |
MemInsight: Autonomous Memory Augmentation for LLM Agents (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language model (LLM) agents have evolved to intelligently process information, make decisions, and interact with users or tools. |
| Approach: | They propose an autonomous memory augmentation approach to enhance semantic data representation and retrieval mechanisms by leveraging historical interactions. |
| Outcome: | The proposed approach outperforms a baseline RAG by 34% in recall for LoCoMo retrieval on three task scenarios and boosts persuasiveness of recommendations by 14%. |
LSSF: Safety Alignment for Large Language Models through Low-Rank Safety Subspace Fusion (2025.acl-long)
Copied to clipboard
| Challenge: | Existing safety alignment methods rely on fine-tuning, which inadvertently leads to the increased complexity and computational resources required. |
| Approach: | They propose a safety re-alignment framework with Low-Rank Safety Subspace Fusison that exploits low-rank safety characteristics of LLMs by constructing a low-ranked projection matrix to extract the principal components of safety vectors. |
| Outcome: | The proposed method exploits low-rank safety subspace of the LLMs and is stable during fine-tuning process and is isolated from the model’s general capabilities. |
The Structural Safety Generalization Problem (2025.findings-acl)
Copied to clipboard
Julius Broomfield, Tom Gibbs, George Ingebretsen, Ethan Kosak-Hine, Tia Nasir, Jason Zhang, Reihaneh Iranmanesh, Sara Pieri, Reihaneh Rabbany, Kellin Pelrine
| Challenge: | LLM jailbreaks are a widespread safety challenge. |
| Approach: | They propose a structure-rewriting guardrail that allows for more efficient safety assessment . single-turn attacks are the most extensively explored in the literature . |
| Outcome: | The proposed framework can be used to enable new defenses, the authors show . they show that the proposed framework reduces the risk of harmful inputs . |
Personalizing Dialogue Agents: I have a dog, do you have pets too? (P18-1)
Copied to clipboard
| Challenge: | chit-chat models lack specificity, do not display a consistent personality and are often not very captivating. |
| Approach: | They propose to train chit-chat models to condition on profile information and profile information about the interlocutors. |
| Outcome: | The proposed model can predict profile information about the interlocutors based on the data . the proposed model is able to generate meaningful responses in a chit-chat setting . |
WMT24++: Expanding the Language Coverage of WMT24 to 55 Languages & Dialects (2025.findings-acl)
Copied to clipboard
Daniel Deutsch, Eleftheria Briakou, Isaac Rayburn Caswell, Mara Finkelstein, Rebecca Galor, Juraj Juraska, Geza Kovacs, Alison Lui, Ricardo Rei, Jason Riesa, Shruti Rijhwani, Parker Riley, Elizabeth Salesky, Firas Trabelsi, Stephanie Winkler, Biao Zhang, Markus Freitag
| Challenge: | In order to evaluate large language models (LLMs), it is important to collect benchmark datasets in order to assess their multilingual performance. |
| Approach: | They extend the WMT24 dataset to cover 55 languages by collecting new human-written references and post-edits for 46 new languages/dialects. |
| Outcome: | The proposed dataset covers 55 languages and provides best-performing MT systems in all 55 languages. |
Learning Adaptive Axis Attentions in Fine-tuning: Beyond Fixed Sparse Attention Patterns (2022.findings-acl)
Copied to clipboard
Zihan Wang, Jiuxiang Gu, Jason Kuen, Handong Zhao, Vlad Morariu, Ruiyi Zhang, Ani Nenkova, Tong Sun, Jingbo Shang
| Challenge: | Adaptive Axis Attention learns different attention patterns for each task and model layer . sparse attention patterns do not improve the run time of the models but they reduce model memory requirements . |
| Approach: | They propose a method that learns different attention patterns for each Transformer layer . they propose 'adaptive axis attention' method that identifies important tokens . |
| Outcome: | The proposed method does not require pre-training to accommodate sparse attention patterns. |
A Critical Analysis of Document Out-of-Distribution Detection (2023.findings-emnlp)
Copied to clipboard
Jiuxiang Gu, Yifei Ming, Yi Zhou, Jason Kuen, Vlad Morariu, Handong Zhao, Ruiyi Zhang, Nikolaos Barmpalios, Anqi Liu, Yixuan Li, Tong Sun, Ani Nenkova
| Challenge: | Existing document understanding models focus on single-modal inputs such as images or texts. |
| Approach: | They propose to use a spatial-aware adapter to adapt transformer-based language models to document domain to exploit multi-modal information. |
| Outcome: | The proposed model significantly improves the OOD detection performance compared to using a standard language model and to competitive baselines. |
Mapping Natural Language Instructions to Mobile UI Action Sequences (2020.acl-main)
Copied to clipboard
| Challenge: | a new problem of grounding natural language instructions to mobile UI actions is emerging . we use a Transformer to extract action phrase tuples from long-range natural language instruction . |
| Approach: | They propose a dataset that pairs English instructions with actions performed by people on a mobile UI emulator. |
| Outcome: | The proposed model achieves 70.59% accuracy on predicting complete ground-truth action sequences in PixelHelp. |
Label Semantic Aware Pre-training for Few-shot Text Classification (2022.acl-long)
Copied to clipboard
| Challenge: | Existing models for text classification use label semantics but few studies have attempted to give models access to informative representations of labels. |
| Approach: | They propose to use label semantics to train generative models by performing secondary pre-training on labeled sentences from a variety of domains. |
| Outcome: | The proposed approach improves generalization and data efficiency of text classification systems while maintaining comparable performance to state-of-the-art models. |
Stanza: A Python Natural Language Processing Toolkit for Many Human Languages (2020.acl-demos)
Copied to clipboard
| Challenge: | Existing tools that support only a few major languages are under-optimized for accuracy due to a focus on efficiency or use of less powerful models. |
| Approach: | They introduce a Python natural language processing toolkit that supports 66 languages . they train Stanza on 112 datasets and show it generalizes well on all languages compared to other tools . |
| Outcome: | The proposed toolkit performs well on 112 datasets and is compatible with the popular Java CoreNLP software. |
PAWS-X: A Cross-lingual Adversarial Dataset for Paraphrase Identification (D19-1)
Copied to clipboard
| Challenge: | Existing work on adversarial data generation focuses on English . Existing multilingual datasets show effectiveness of deep, multilingual pre-training . |
| Approach: | They propose a dataset of 23,659 human translated PAWS evaluation pairs in six languages . they show the effectiveness of deep, multilingual pre-training while leaving considerable headroom . |
| Outcome: | The proposed model shows that multilingual training and evaluation regimes are more accurate than previous models. |
SAMULE: Self-Learning Agents Enhanced by Multi-level Reflection (2025.emnlp-main)
Copied to clipboard
| Challenge: | Modern AI agents rely on Large Language Models (LLMs) as their reasoning engines, but they still face the challenge of generating meaningful reflections due to inadequate error analysis and a reliance on rare successful trajectories. |
| Approach: | They propose a framework for self-learning agents powered by a retrospective language model that generates reflections during inference. |
| Outcome: | The proposed framework outperforms reflection-based baselines on three challenging benchmarks. |
A Fast, Compact, Accurate Model for Language Identification of Codemixed Text (D18-1)
Copied to clipboard
| Challenge: | a feed-forward network can label codemixed and monolingual text in 100 languages and 100 language pairs. |
| Approach: | They propose a feed-forward network that can provide a language code for every token in a sentence . they show that the model can label both codemixed and monolingual text in 100 languages . |
| Outcome: | The proposed model outperforms previous multilingual approaches in terms of accuracy and speed. |
Data Collection for Dialogue System: A Startup Perspective (N18-3)
Copied to clipboard
| Challenge: | Developing dialogue systems such as Apple Siri and Google Now requires high quality training data but data collection with crowdsourcing is largely an open question. |
| Approach: | They propose to use crowdsourcing to collect data for a user intent classification task in a dialogue system. |
| Outcome: | The proposed method improves the quality of the collected data and the model performance on real user queries. |