Julia Hockenmaier

Yekaterina Yegorova, Argyrios Gerogiannis, Haolong Zheng et al.

Speech-aware large language models often generalize poorly to out-of-domain settings. We propose SALSA (Speech-Aware LLM Adaptation via Learned Steering Activations), a lightweight adaptation method that learns layer-wise steering vectors. Unlike commonly used steering approaches that rely on contrastive activation differences, SALSA directly optimizes steering vectors using a supervised objective. Across children's speech, multilingual speech, and Mandarin-English code-switching benchmarks, SALSA substantially improves performance over zero-shot inference and speech in-context learning baselines, achieving up to 46.8% relative improvements over zero-shot. Analysis further demonstrates that steering the encoder, particularly the later layers, is more effective than steering the LLM backbone. These findings suggest that steering improves downstream ASR performance by adapting higher-level acoustic and phonetic representations to better align with the pretrained language model representation space, rather than by modifying the decoder itself.

Jinu Lee, Shivam Agarwal, Amruta Parulekar et al.

Large reasoning models (LRMs) produce reasoning traces with non-linear structures, such as backtracking and self-correction, that complicate the evaluation and monitoring of the reasoning process. We introduce ReasoningFlow, a framework that captures the discourse structures of LRM reasoning traces into fine-grained directed acyclic graphs (DAGs). We develop and validate our annotation schema through careful manual annotation of 31 traces (2.1k steps), achieving high inter-annotator agreement, then scale to automatic annotation of 1,260 traces (247.7k steps) spanning three tasks (math, science, argumentation) and five models (Qwen2.5-32B-Inst, QwQ-32B, DeepSeek-V3, DeepSeek-R1, GPT-oss-120B). By analyzing ReasoningFlow graphs, we find: (1) LRMs exhibit structurally similar traces, despite being trained from different base models and potentially non-overlapping post-training data. (2) ReasoningFlow reveals diverse fine-grained reasoning behaviors (e.g., local verification, self-reflection, and assumptions) that can be used for better reasoning trace monitorability. (3) In LRMs, most of the erroneous steps are not used to derive final answers. (4) Mechanistic causal dependencies between steps do not reflect the language-level discourse structure. We release the dataset and code in: https://github.com/jinulee-v/reasoningflow.

Harsha Kokel, Mayukh Das, Rakibul Islam et al. · ibm-research

We consider the problem of human-machine collaborative problem solving as a planning task coupled with natural language communication. Our framework consists of three components -- a natural language engine that parses the language utterances to a formal representation and vice-versa, a concept learner that induces generalized concepts for plans based on limited interactions with the user, and an HTN planner that solves the task based on human interaction. We illustrate the ability of this framework to address the key challenges of collaborative problem solving by demonstrating it on a collaborative building task in a Minecraft-based blocksworld domain. The accompanied demo video is available at https://youtu.be/q1pWe4aahF0.

Qingyun Wang, Manling Li, Hou Pong Chan et al.

Goal-oriented generative script learning aims to generate subsequent steps to reach a particular goal, which is an essential task to assist robots or humans in performing stereotypical activities. An important aspect of this process is the ability to capture historical states visually, which provides detailed information that is not covered by text and will guide subsequent steps. Therefore, we propose a new task, Multimedia Generative Script Learning, to generate subsequent steps by tracking historical states in both text and vision modalities, as well as presenting the first benchmark containing 5,652 tasks and 79,089 multimedia steps. This task is challenging in three aspects: the multimedia challenge of capturing the visual states in images, the induction challenge of performing unseen tasks, and the diversity challenge of covering different information in individual steps. We propose to encode visual state changes through a selective multimedia encoder to address the multimedia challenge, transfer knowledge from previously observed tasks using a retrieval-augmented decoder to overcome the induction challenge, and further present distinct information at each step by optimizing a diversity-oriented contrastive learning objective. We define metrics to evaluate both generation and inductive quality. Experiment results demonstrate that our approach significantly outperforms strong baselines.

Marc Canby, Adam Davies, Chirag Rastogi et al.

Causal probing aims to analyze foundation models by examining how intervening on their representation of various latent properties impacts their outputs. Recent works have cast doubt on the theoretical basis of several leading causal probing methods, but it has been unclear how to systematically evaluate the effectiveness of these methods in practice. To address this, we define two key causal probing desiderata: completeness (how thoroughly the representation of the target property has been transformed) and selectivity (how little non-targeted properties have been impacted). We find that there is an inherent tradeoff between the two, which we define as reliability, their harmonic mean. We introduce an empirical analysis framework to measure and evaluate these quantities, allowing us to make the first direct comparisons between different families of leading causal probing methods (e.g., linear vs. nonlinear, or concept removal vs. counterfactual interventions). We find that: (1) all methods show a clear tradeoff between completeness and selectivity; (2) more complete and reliable methods have a greater impact on LLM behavior; and (3) nonlinear interventions are almost always more reliable than linear interventions. Our project webpage is available at: https://ahdavies6.github.io/causal_probing_reliability/

Jinu Lee, Kyoung-Woon On, Simeng Han et al.

Evaluating the quality of LLM-generated reasoning traces in expert domains (e.g., law) is essential for ensuring credibility and explainability, yet remains challenging due to the inherent complexity of such reasoning tasks. We introduce LEGIT (LEGal Issue Trees), a novel large-scale (24K instances) expert-level legal reasoning dataset with an emphasis on reasoning trace evaluation. We convert court judgments into hierarchical trees of opposing parties' arguments and the court's conclusions, which serve as rubrics for evaluating the issue coverage and correctness of the reasoning traces. We verify the reliability of these rubrics via human expert annotations and comparison with coarse, less informative rubrics. Using the LEGIT dataset, we show that (1) LLMs' legal reasoning ability is seriously affected by both legal issue coverage and correctness, and that (2) retrieval-augmented generation (RAG) and RL with rubrics bring complementary benefits for legal reasoning abilities, where RAG improves overall reasoning capability, whereas RL improves correctness albeit with reduced coverage.

Rajarshi Haldar, Julia Hockenmaier

As Natural Language Generation (NLG) continues to be widely adopted, properly assessing it has become quite difficult. Lately, using large language models (LLMs) for evaluating these generations has gained traction, as they tend to align more closely with human preferences than conventional n-gram or embedding-based metrics. In our experiments, we show that LLM judges have low intra-rater reliability in their assigned scores across different runs. This variance makes their ratings inconsistent, almost arbitrary in the worst case, making it difficult to measure how good their judgments actually are. We quantify this inconsistency across different NLG tasks and benchmarks and see if judicious use of LLM judges can still be useful following proper guidelines.

Sewoong Lee, Adam Davies, Marc E. Canby et al.

Sparse autoencoders (SAEs) are widely used in mechanistic interpretability research for large language models; however, the state-of-the-art method of using $k$-sparse autoencoders lacks a theoretical grounding for selecting the hyperparameter $k$ that represents the number of nonzero activations, often denoted by $\ell_0$. In this paper, we reveal a theoretical link that the $\ell_2$-norm of the sparse feature vector can be approximated with the $\ell_2$-norm of the dense vector with a closed-form error, which allows sparse autoencoders to be trained without the need to manually determine $\ell_0$. Specifically, we validate two applications of our theoretical findings. First, we introduce a new methodology that can assess the feature activations of pre-trained SAEs by computing the theoretically expected value from the input embedding, which has been overlooked by existing SAE evaluation methods and loss functions. Second, we introduce a novel activation function, top-AFA, which builds upon our formulation of approximate feature activation (AFA). This function enables top-$k$ style activation without requiring a constant hyperparameter $k$ to be tuned, dynamically determining the number of activated features for each input. By training SAEs on three intermediate layers to reconstruct GPT2 hidden embeddings for over 80 million tokens from the OpenWebText dataset, we demonstrate the empirical merits of this approach and compare it with current state-of-the-art $k$-sparse autoencoders. Our code is available at: https://github.com/SewoongLee/top-afa-sae.

Jiacheng Liu, Julia Hockenmaier

The phrase grounding task aims to ground each entity mention in a given caption of an image to a corresponding region in that image. Although there are clear dependencies between how different mentions of the same caption should be grounded, previous structured prediction methods that aim to capture such dependencies need to resort to approximate inference or non-differentiable losses. In this paper, we formulate phrase grounding as a sequence labeling task where we treat candidate regions as potential labels, and use neural chain Conditional Random Fields (CRFs) to model dependencies among regions for adjacent mentions. In contrast to standard sequence labeling tasks, the phrase grounding task is defined such that there may be multiple correct candidate regions. To address this multiplicity of gold labels, we define so-called Soft-Label Chain CRFs, and present an algorithm that enables convenient end-to-end training. Our method establishes a new state-of-the-art on phrase grounding on the Flickr30k Entities dataset. Analysis shows that our model benefits both from the entity dependencies captured by the CRF and from the soft-label training regime. Our code is available at \url{github.com/liujch1998/SoftLabelCCRF}

Yonatan Bisk, Siva Reddy, John Blitzer et al.

We compare the effectiveness of four different syntactic CCG parsers for a semantic slot-filling task to explore how much syntactic supervision is required for downstream semantic analysis. This extrinsic, task-based evaluation provides a unique window to explore the strengths and weaknesses of semantics captured by unsupervised grammar induction systems. We release a new Freebase semantic parsing dataset called SPADES (Semantic PArsing of DEclarative Sentences) containing 93K cloze-style questions paired with answers. We evaluate all our models on this dataset. Our code and data are available at https://github.com/sivareddyg/graph-parser.

Risham Sidhu, Julia Hockenmaier

We introduce GSU, a text-only grid dataset to evaluate the spatial reasoning capabilities of LLMs over 3 core tasks: navigation, object localization, and structure composition. By forgoing visual inputs, isolating spatial reasoning from perception, we show that while most models grasp basic grid concepts, they struggle with frames of reference relative to an embodied agent and identifying 3D shapes from coordinate lists. We also find that exposure to a visual modality does not provide a generalizable understanding of 3D space that VLMs are able to utilize for these tasks. Finally, we show that while the very latest frontier models can solve the provided tasks (though harder variants may still stump them), fully fine-tuning a small LM or LORA fine-tuning a small LLM show potential to match frontier model performance, suggesting an avenue for specialized embodied agents.

Rajarshi Haldar, Julia Hockenmaier

Large language models (LLMs) such as Llama 2 perform very well on tasks that involve both natural language and source code, particularly code summarization and code generation. We show that for the task of code summarization, the performance of these models on individual examples often depends on the amount of (subword) token overlap between the code and the corresponding reference natural language descriptions in the dataset. This token overlap arises because the reference descriptions in standard datasets (corresponding to docstrings in large code bases) are often highly similar to the names of the functions they describe. We also show that this token overlap occurs largely in the function names of the code and compare the relative performance of these models after removing function names versus removing code structure. We also show that using multiple evaluation metrics like BLEU and BERTScore gives us very little additional insight since these metrics are highly correlated with each other.

Jerry Huang, Siddarth Madala, Risham Sidhu et al.

Retrieval-augmented generation (RAG) systems rely on retrieval models for identifying relevant contexts and answer generation models for utilizing those contexts. However, retrievers exhibit imperfect recall and precision, limiting downstream performance. We introduce RAG-RL, an answer generation model trained not only to produce answers but also to identify and cite relevant information from larger sets of retrieved contexts, shifting some of the burden of identifying relevant documents from the retriever to the answer generator. Our approach uses curriculum learning, where the model is first trained on easier examples that include only relevant contexts. Our experiments show that these training samples enable models to acquire citation and reasoning skills with greater sample efficiency and generalizability, demonstrating strong model performance even as the number of irrelevant passages increases. We benchmark our methods on three open-domain multi-hop question answering datasets and report significant gains in answer and citation accuracy. Our experiments provide empirical insights into how easier training samples can give models stronger signals for learning specific skills (e.g., citation generation) and how different components of post-training (e.g., training set construction, rule-based rewards, training sample ordering, etc.) impact final model performance.

Seungone Kim, Ian Wu, Jinu Lee et al. · cmu

As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.

Jinu Lee, Julia Hockenmaier

Step-by-step reasoning is widely used to enhance the reasoning ability of large language models (LLMs) in complex problems. Evaluating the quality of reasoning traces is crucial for understanding and improving LLM reasoning. However, existing evaluation practices are highly inconsistent, resulting in fragmented progress across evaluator design and benchmark development. To address this gap, this survey provides a comprehensive overview of step-by-step reasoning evaluation, proposing a taxonomy of evaluation criteria with four top-level categories (factuality, validity, coherence, and utility). Based on the taxonomy, we review different datasets, evaluator implementations, and recent findings, leading to promising directions for future research.

Yoonhwa Jung, Ikhyun Cho, Shun-Hsiang Hsu et al.

With growing concerns surrounding privacy and regulatory compliance, the concept of machine unlearning has gained prominence, aiming to selectively forget or erase specific learned information from a trained model. In response to this critical need, we introduce a novel approach called Attack-and-Reset for Unlearning (ARU). This algorithm leverages meticulously crafted adversarial noise to generate a parameter mask, effectively resetting certain parameters and rendering them unlearnable. ARU outperforms current state-of-the-art results on two facial machine-unlearning benchmark datasets, MUFAC and MUCAC. In particular, we present the steps involved in attacking and masking that strategically filter and re-initialize network parameters biased towards the forget set. Our work represents a significant advancement in rendering data unexploitable to deep learning models through parameter re-initialization, achieved by harnessing adversarial noise to craft a mask.

Jinu Lee, Sagnik Mukherjee, Dilek Hakkani-Tur et al.

Large reasoning models (LRMs) generate complex reasoning traces with planning, reflection, verification, and backtracking. In this work, we introduce ReasoningFlow, a unified schema for analyzing the semantic structures of these complex traces. ReasoningFlow parses traces into directed acyclic graphs, enabling the characterization of distinct reasoning patterns as subgraph structures. This human-interpretable representation offers promising applications in understanding, evaluating, and enhancing the reasoning processes of LRMs.

Jinu Lee, Qi Liu, Runzhi Ma et al.

First-order logic (FOL) can represent the logical entailment semantics of natural language (NL) sentences, but determining natural language entailment using FOL remains a challenge. To address this, we propose the Entailment-Preserving FOL representations (EPF) task and introduce reference-free evaluation metrics for EPF, the Entailment-Preserving Rate (EPR) family. In EPF, one should generate FOL representations from multi-premise natural language entailment data (e.g. EntailmentBank) so that the automatic prover's result preserves the entailment labels. Experiments show that existing methods for NL-to-FOL translation struggle in EPF. To this extent, we propose a training method specialized for the task, iterative learning-to-rank, which directly optimizes the model's EPR score through a novel scoring function and a learning-to-rank objective. Our method achieves a 1.8-2.7% improvement in EPR and a 17.4-20.6% increase in EPR@16 compared to diverse baselines in three datasets. Further analyses reveal that iterative learning-to-rank effectively suppresses the arbitrariness of FOL representation by reducing the diversity of predicate signatures, and maintains strong performance across diverse inference types and out-of-domain data.

Prashant Jayannavar, Liliang Ren, Marisa Hudspeth et al.

Developing interactive agents that can understand language, perceive their surroundings, and act within the physical world is a long-standing goal of AI research. The Minecraft Collaborative Building Task (MCBT) (Narayan-Chen, Jayannavar, and Hockenmaier 2019), a two-player game in which an Architect (A) instructs a Builder (B) to construct a target structure in a simulated 3D Blocks World environment, offers a rich platform to work towards this goal. In this work, we focus on the Builder Action Prediction (BAP) subtask: predicting B's actions in a multimodal game context (Jayannavar, Narayan-Chen, and Hockenmaier 2020) - a challenging testbed for grounded instruction following, with limited training data. We holistically re-examine this task and introduce BAP v2 to address key challenges in evaluation, training data, and modeling. Specifically, we define an enhanced evaluation benchmark, featuring a cleaner test set and fairer, more insightful metrics that also reveal spatial reasoning as the primary performance bottleneck. To address data scarcity and to teach models basic spatial skills, we generate different types of synthetic MCBT data. We observe that current, LLM-based SOTA models trained on the human BAP dialogues fail on these simpler, synthetic BAP ones, but show that training models on this synthetic data improves their performance across the board. We also introduce a new SOTA model, Llama-CRAFTS, which leverages richer input representations, and achieves an F1 score of 53.0 on the BAP v2 task and strong performance on the synthetic data. While this result marks a notable 6 points improvement over previous work, it also underscores the task's remaining difficulty, establishing BAP v2 as a fertile ground for future research, and providing a useful measure of the spatial capabilities of current text-only LLMs in such embodied tasks.

Ikhyun Cho, Changyeon Park, Julia Hockenmaier

Machine unlearning (MUL) is an arising field in machine learning that seeks to erase the learned information of specific training data points from a trained model. Despite the recent active research in MUL within computer vision, the majority of work has focused on ResNet-based models. Given that Vision Transformers (ViT) have become the predominant model architecture, a detailed study of MUL specifically tailored to ViT is essential. In this paper, we present comprehensive experiments on ViTs using recent MUL algorithms and datasets. We anticipate that our experiments, ablation studies, and findings could provide valuable insights and inspire further research in this field.

Marc E. Canby, Julia Hockenmaier

Transformer-based encoder-decoder models that generate outputs in a left-to-right fashion have become standard for sequence-to-sequence tasks. In this paper, we propose a framework for decoding that produces sequences from the "outside-in": at each step, the model chooses to generate a token on the left, on the right, or join the left and right sequences. We argue that this is more principled than prior bidirectional decoders. Our proposal supports a variety of model architectures and includes several training methods, such as a dynamic programming algorithm that marginalizes out the latent ordering variable. Our model sets state-of-the-art (SOTA) on the 2022 and 2023 shared tasks, beating the next best systems by over 4.7 and 2.7 points in average accuracy respectively. The model performs particularly well on long sequences, can implicitly learn the split point of words composed of stem and affix, and performs better relative to the baseline on datasets that have fewer unique lemmas (but more examples per lemma).

Liliang Ren, Chenkai Sun, Heng Ji et al.

Text-to-Graph extraction aims to automatically extract information graphs consisting of mentions and types from natural language texts. Existing approaches, such as table filling and pairwise scoring, have shown impressive performance on various information extraction tasks, but they are difficult to scale to datasets with longer input texts because of their second-order space/time complexities with respect to the input length. In this work, we propose a Hybrid Span Generator (HySPA) that invertibly maps the information graph to an alternating sequence of nodes and edge types, and directly generates such sequences via a hybrid span decoder which can decode both the spans and the types recurrently in linear time and space complexities. Extensive experiments on the ACE05 dataset show that our approach also significantly outperforms state-of-the-art on the joint entity and relation extraction task.

Rajarshi Haldar, Lingfei Wu, Jinjun Xiong et al.

The ability to match pieces of code to their corresponding natural language descriptions and vice versa is fundamental for natural language search interfaces to software repositories. In this paper, we propose a novel multi-perspective cross-lingual neural framework for code--text matching, inspired in part by a previous model for monolingual text-to-text matching, to capture both global and local similarities. Our experiments on the CoNaLa dataset show that our proposed model yields better performance on this cross-lingual text-to-code matching task than previous approaches that map code and text to a single joint embedding space.

Alice Lai, Yonatan Bisk, Julia Hockenmaier

We define a novel textual entailment task that requires inference over multiple premise sentences. We present a new dataset for this task that minimizes trivial lexical inferences, emphasizes knowledge of everyday events, and presents a more challenging setting for textual entailment. We evaluate several strong neural baselines and analyze how the multiple premise task differs from standard textual entailment.

Bryan A. Plummer, Arun Mallya, Christopher M. Cervantes et al.

This paper presents a framework for localization or grounding of phrases in images using a large collection of linguistic and visual cues. We model the appearance, size, and position of entity bounding boxes, adjectives that contain attribute information, and spatial relationships between pairs of entities connected by verbs or prepositions. Special attention is given to relationships between people and clothing or body part mentions, as they are useful for distinguishing individuals. We automatically learn weights for combining these cues and at test time, perform joint inference over all phrases in a caption. The resulting system produces state of the art performance on phrase localization on the Flickr30k Entities dataset and visual relationship detection on the Stanford VRD dataset.

Bryan A. Plummer, Liwei Wang, Chris M. Cervantes et al.

The Flickr30k dataset has become a standard benchmark for sentence-based image description. This paper presents Flickr30k Entities, which augments the 158k captions from Flickr30k with 244k coreference chains, linking mentions of the same entities across different captions for the same image, and associating them with 276k manually annotated bounding boxes. Such annotations are essential for continued progress in automatic image description and grounded language understanding. They enable us to define a new benchmark for localization of textual entity mentions in an image. We present a strong baseline for this task that combines an image-text embedding, detectors for common objects, a color classifier, and a bias towards selecting larger objects. While our baseline rivals in accuracy more complex state-of-the-art models, we show that its gains cannot be easily parlayed into improvements on such tasks as image-sentence retrieval, thus underlining the limitations of current methods and the need for further research.

Hannaneh Hajishirzi, Julia Hockenmaier, Erik T. Mueller et al.

This paper presents an approach for learning to translate simple narratives, i.e., texts (sequences of sentences) describing dynamic systems, into coherent sequences of events without the need for labeled training data. Our approach incorporates domain knowledge in the form of preconditions and effects of events, and we show that it outperforms state-of-the-art supervised learning systems on the task of reconstructing RoboCup soccer games from their commentaries.