Papers by Honglak Lee
GRACE: Discriminator-Guided Chain-of-Thought Reasoning (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing language models (LMs) can assign a high likelihood to incorrect steps . Existing models (LLMs), however, struggle with complex multi-step reasoning. |
| Approach: | They propose a stepwise decoding approach that steers the decoding process towards producing correct reasoning steps. |
| Outcome: | The proposed approach outperforms existing methods on math and symbolic reasoning tasks. |
Auto-Intent: Automated Intent Discovery and Self-Exploration for Large Language Model Web Agents (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing work to improve LLMs' planning and reasoning capabilities has shown that they have strong priors to reason about intermediate subgoals, but their accuracy is limited especially in complex environments such as web navigation. |
| Approach: | They propose a method to adapt a pre-trained large language model (LLM) as an agent for a target domain without direct fine-tuning by extracting intents from target domain demonstrations in a highly compact form. |
| Outcome: | The proposed method significantly improves the performance of GPT-3.5, 4 and Llama-3.1-70B, 405B agents on large-scale real-website navigation benchmarks from Mind2Web and online navigation tasks with its cross-benchmark generalization from Mind2. |
Code Models are Zero-shot Precondition Reasoners (2024.naacl-long)
Copied to clipboard
Lajanugen Logeswaran, Sungryull Sohn, Yiwei Lyu, Anthony Liu, Dong-Ki Kim, Dongsub Shim, Moontae Lee, Honglak Lee
| Challenge: | Existing methods to reason about action preconditions are lacking for agent to complete tasks. |
| Approach: | They propose a method to reason about action preconditions using pre-trained code models. |
| Outcome: | The proposed approach improves few-shot policy learning approaches across task-oriented dialog and embodied textworld benchmarks. |
From Heuristic to Analytic: Cognitively Motivated Strategies for Coherent Physical Commonsense Reasoning (2023.emnlp-main)
Copied to clipboard
| Challenge: | Pre-trained language models have shown impressive performance in various language tasks, but are prone to spurious correlations and illusory information. |
| Approach: | They propose to use pre-trained language models to justify decisions with formalized, coherent reasoning chains. |
| Outcome: | The proposed strategies improve coherence of rationalizations yielding state-of-the-art results on Tiered Reasoning for Intuitive Physics (TRIP). |
Revisiting LLM Value Probing Strategies: Are They Robust and Expressive? (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing value probing methods that capture in-context information and predict models’ real-world actions are limited and lack systematic comparisons. |
| Approach: | They compare three widely used value probing methods: token likelihood, sequence perplexity, and text generation. |
| Outcome: | The proposed methods exhibit large variances under non-semantic perturbations in prompts and option formats, with sequence perplexity being the most robust overall. |
Beyond Blind Following: Evaluating Robustness of LLM Agents under Imperfect Guidance (2026.eacl-long)
Copied to clipboard
Yao Fu, Ran Qiu, Xinhe Wang, Jacob Sansom, Sathvika Ayyappa Prabhu, Huijie Tang, Jaekyeom Kim, Sungryull Sohn, Honglak Lee
| Challenge: | Large language models (LLMs) have shown strong capabilities as task-solving agents across interactive domains, but in complex environments, auxiliary guidance may be imperfect. |
| Approach: | They propose a benchmark to measure the robustness of large language models under imperfect guidance. |
| Outcome: | The proposed benchmark compared LLM agents in navigation, cooking, and gaming in a variety of environments with auxiliary guidance and noisy or underspecified instructions extracted from demonstrations. |
Deep Exploration of Cross-Lingual Zero-Shot Generalization in Instruction Tuning (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent studies have focused on instruction tuning to show cross-lingual generalization . a novel non-English meta-dataset is used to study instruction tuning . |
| Approach: | They perform instruction tuning individually for two distinct language meta-datasets and assess the performance on unseen tasks in a non-English language. |
| Outcome: | The proposed model outperforms baseline training in English and Korean by 20.7% and 13.6%. |
Prospector: Improving LLM Agents with Self-Asking and Trajectory Ranking (2024.findings-emnlp)
Copied to clipboard
Byoungjip Kim, Youngsoo Jang, Lajanugen Logeswaran, Geon-Hyeong Kim, Yu Jin Kim, Honglak Lee, Moontae Lee
| Challenge: | Existing LLMs are limited in their ability to incorporate feedback from an environment. |
| Approach: | They propose an LLM agent that consists of an Actor and a Critic. |
| Outcome: | The proposed agent outperforms existing LLMs on benchmark environments and shows that it can generate diverse trajectories and pick the most rewarding trajectory. |
Few-shot Reranking for Multi-hop QA via Language Model Prompting (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for multi-hop QA with open-domain questions require a large number of labeled question-document pairs for retrieval. |
| Approach: | They propose a language-based prompt for multi-hop path reranking that relies on language model prompting to generate a relevance score between a question and the path. |
| Outcome: | The proposed method yields strong retrieval performance on HotpotQA with only 128 training examples compared to state-of-the-art methods trained on thousands of examples. |
Mitigating Biases for Instruction-following Language Models via Bias Neurons Elimination (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods to mitigate undesirable biases in instruction-following language models are not effective in accelerating instruction-based learning. |
| Approach: | They propose a method to eliminate bias neurons of language models in instruction-following settings by defining the bias neuron and prove its existence empirically. |
| Outcome: | The proposed method dramatically increases the task performance of language models under zero-shot instruction-following settings without losing the model’s knowledge. |
Zero-Shot Entity Linking by Reading Entity Descriptions (P19-1)
Copied to clipboard
| Challenge: | Existing approaches to link entities to unseen entities require in-domain labeled data. |
| Approach: | They propose a zero-shot entity linking task where mentions must be linked to unseen entities without in-domain labeled data. |
| Outcome: | The proposed task can generalize to unseen entities without metadata or alias tables . the proposed system improves over baselines, including BERT, on a new dataset . |
A Picture is Worth a Thousand Words: Language Models Plan from Pixels (2023.emnlp-main)
Copied to clipboard
| Challenge: | Recent work uses pre-trained language models to reason about plans from text instructions in embodied visual environments. |
| Approach: | They propose to use pre-trained language models to reason about plan sequences from text instructions in embodied visual environments. |
| Outcome: | The proposed approach outperforms previous approaches on the ALFWorld and VirtualHome benchmarks. |
Unsupervised Task Graph Generation from Instructional Video Transcripts (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to task graph generation use instruction-tuned language models to generate accurate task graphs. |
| Approach: | They propose an unsupervised approach that combines the reasoning capabilities of instruction-tuned language models with clustering and ranking components to generate accurate task graphs. |
| Outcome: | The proposed approach generates more accurate task graphs than a supervised learning approach on tasks from the ProceL and CrossTask datasets. |
Instruction Matters: A Simple yet Effective Task Selection for Optimized Instruction Tuning of Specific Tasks (2024.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results show that instruction tuning improves zero-shot generalization across various tasks and improves performance of specific tasks. |
| Approach: | They propose a task selection method that leverages instruction information alone to identify relevant tasks and optimize instruction tuning for specific tasks. |
| Outcome: | The proposed method is significantly more efficient than traditional approaches, which require complex measurements of pairwise transferability between tasks or the creation of data samples for the target task. |
Few-shot Subgoal Planning with Language Models (2022.naacl-main)
Copied to clipboard
| Challenge: | Pre-trained language models have shown successful progress in many text understanding benchmarks. |
| Approach: | They propose a strategy to re-rank language model predictions based on interaction and feedback from the environment. |
| Outcome: | The proposed approach shows competitive performance on subgoal prediction and task completion in the ALFRED benchmark compared to prior methods that assume more subgoals supervision. |
Understanding the Capabilities and Limitations of Large Language Models for Cultural Commonsense (2024.naacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated substantial commonsense understanding through numerous benchmark evaluations. |
| Approach: | They conduct a comprehensive examination of the capabilities and limitations of several state-of-the-art LLMs in the context of cultural commonsense tasks. |
| Outcome: | The language used to query the LLMs can impact their performance on cultural-related tasks. |
Interactive and Expressive Code-Augmented Planning with Large Language Models (2025.acl-long)
Copied to clipboard
Anthony Zhe Liu, Xinhe Wang, Jacob Sansom, Yao Fu, Jongwook Choi, Sungryull Sohn, Jaekyeom Kim, Honglak Lee
| Challenge: | Large Language Models (LLMs) have strong abilities in common-sense reasoning and interactive decision-making, but struggle with complex, long-horizon planning tasks. |
| Approach: | They propose a code-based LLM planning approach that is code-expressive while also dynamically adapting from errors. |
| Outcome: | The proposed approach can be error-prone and insufficient for handling ambiguous or unstructured data. |
TOD-Flow: Modeling the Structure of Task-Oriented Dialogues (2023.emnlp-main)
Copied to clipboard
Sungryull Sohn, Yiwei Lyu, Anthony Liu, Lajanugen Logeswaran, Dong-Ki Kim, Dongsub Shim, Honglak Lee
| Challenge: | Recent advances in task-oriented dialogue systems have limitations regarding transparency and controllability. |
| Approach: | They propose to infer the TOD-flow graph from dialog data annotated with dialog acts and integrate it with any dialogue model to improve its prediction performance, transparency, and controllability. |
| Outcome: | The proposed approach improves dialog act classification and response generation performance in the MultiWOZ and SGD benchmarks. |
MASSW: A New Dataset and Benchmark Tasks for AI-Assisted Scientific Workflows (2025.findings-naacl)
Copied to clipboard
Xingjian Zhang, Yutong Xie, Jin Huang, Jinge Ma, Zhaoying Pan, Qijia Liu, Ziyang Xiong, Tolga Ergen, Dongsub Shim, Honglak Lee, Qiaozhu Mei
| Challenge: | Scientific innovation is driven by detailed workflows, which include critical steps such as contextualizing literature, generating ideas, validating ideas, and planning new research. |
| Approach: | They propose to use large language models to extract five key aspects from scientific publications to optimize scientific workflows. |
| Outcome: | The proposed dataset includes more than 152,000 peer-reviewed publications from 17 leading computer science conferences spanning the past 50 years. |
IRPO: Implicit Policy Regularized Preference Optimization (2026.findings-eacl)
Copied to clipboard
| Challenge: | Recent DPOs introduce additional hyperparameters, reducing feasibility for LLM fine-tuning. |
| Approach: | They propose an algorithm that regularizes the reward against a reference policy without extra hyperparameters to address suboptimal outcomes. |
| Outcome: | The proposed algorithm outperforms baseline algorithms with the same hyperparameter complexity while maintaining training simplicity. |
Merging Generated and Retrieved Knowledge for Open-Domain QA (2023.emnlp-main)
Copied to clipboard
| Challenge: | Open-domain question answering systems often have retrieval modules but retrieving passages from external knowledge sources is known to suffer from insufficient knowledge coverage. |
| Approach: | They propose a Compatibility-Oriented knowledge Merging framework to leverage both sources of information by matching LLM-generated passages with retrieved counterparts into compatible pairs. |
| Outcome: | The proposed framework outperforms baselines on three out of four tested open-domain QA benchmarks. |
Small Language Models Need Strong Verifiers to Self-Correct Reasoning (2024.findings-acl)
Copied to clipboard
Yunxiang Zhang, Muhammad Khalifa, Lajanugen Logeswaran, Jaekyeom Kim, Moontae Lee, Honglak Lee, Lu Wang
| Challenge: | Existing studies show that large language models can self-correct their outputs by generating a critique and revising it based on the critique. |
| Approach: | They propose a pipeline that prompts small language models to collect self-correction data that supports the training of self-refinement abilities. |
| Outcome: | The proposed pipeline improves the self-correction abilities of two models on five datasets spanning math and commonsense reasoning. |