A novel task of automatically constructing Recent breakthroughs in generative AI have raised "deliberation chains" of probing questions in a the possibility of systems that follow and interact dialogue and with their causal utterances; with multiparty dialogue. Inherent in group dialogues A formal graphical framework for deliberation are utterance sequences that deliberate on chains derived from formal semantics of the same information. Modeling these is particularly situated conversation (Hunter et al., 2018); challenging; while such utterances have a linear order and overlapping information, they may A unique adaptation of methods from coreference be distantly separated in time and the same information resolution to this new task; may be expressed very differently. In this paper, we construct deliberation chains Baseline evaluation on two challenging collaborative in dialogue: turn sequences that surface pieces of dialogue datasets--DeliData and the evidence or questions under discussion that culminate Weights Task Dataset--and a novel method of in a "probing utterance," or explicit elicitation jointly modeling probing and causal interventions of input that does not introduce new information.

Event coreference resolution (ECR) is the task of determining whether distinct mentions of events within a multi-document corpus are actually linked to the same underlying occurrence. Images of the events can help facilitate resolution when language is ambiguous. Here, we propose a multimodal cross-document event coreference resolution method that integrates visual and textual cues with a simple linear map between vision and language models. As existing ECR benchmark datasets rarely provide images for all event mentions, we augment the popular ECB+ dataset with event-centric images scraped from the internet and generated using image diffusion models. We establish three methods that incorporate images and text for coreference: 1) a standard fused model with finetuning, 2) a novel linear mapping method without finetuning and 3) an ensembling approach based on splitting mention pairs by semantic and discourse-level difficulty. We evaluate on 2 datasets: the augmented ECB+, and AIDA Phase 1. Our ensemble systems using cross-modal linear mapping establish an upper limit (91.9 CoNLL F1) on ECB+ ECR performance given the preprocessing assumptions used, and establish a novel baseline on AIDA Phase 1. Our results demonstrate the utility of multimodal information in ECR for certain challenging coreference problems, and highlight a need for more multimodal resources in the coreference resolution space.

Within Dialogue Modeling research in AI and NLP, considerable attention has been spent on ``dialogue state tracking'' (DST), which is the ability to update the representations of the speaker's needs at each turn in the dialogue by taking into account the past dialogue moves and history. Less studied but just as important to dialogue modeling, however, is ``common ground tracking'' (CGT), which identifies the shared belief space held by all of the participants in a task-oriented dialogue: the task-relevant propositions all participants accept as true. In this paper we present a method for automatically identifying the current set of shared beliefs and ``questions under discussion'' (QUDs) of a group with a shared goal. We annotate a dataset of multimodal interactions in a shared physical space with speech transcriptions, prosodic features, gestures, actions, and facets of collaboration, and operationalize these features for use in a deep neural model to predict moves toward construction of common ground. Model outputs cascade into a set of formal closure rules derived from situated evidence and belief axioms and update operations. We empirically assess the contribution of each feature type toward successful construction of common ground relative to ground truth, establishing a benchmark in this novel, challenging task.

Collaborative problem solving (CPS) in teams is tightly coupled with the creation of shared meaning between participants in a situated, collaborative task. In this work, we assess the quality of different utterance segmentation techniques as an aid in annotating CPS. We (1) manually transcribe utterances in a dataset of triads collaboratively solving a problem involving dialogue and physical object manipulation, (2) annotate collaborative moves according to these gold-standard transcripts, and then (3) apply these annotations to utterances that have been automatically segmented using toolkits from Google and OpenAI's Whisper. We show that the oracle utterances have minimal correspondence to automatically segmented speech, and that automatically segmented speech using different segmentation methods is also inconsistent. We also show that annotating automatically segmented speech has distinct implications compared with annotating oracle utterances--since most annotation schemes are designed for oracle cases, when annotating automatically-segmented utterances, annotators must invoke other information to make arbitrary judgments which other annotators may not replicate. We conclude with a discussion of how future annotation specs can account for these needs.

Neural networks are vulnerable to a wide range of erroneous inputs such as adversarial, corrupted, out-of-distribution, and misclassified examples. In this work, we train a linear SVM classifier to detect these four types of erroneous data using hidden and softmax feature vectors of pre-trained neural networks. Our results indicate that these faulty data types generally exhibit linearly separable activation properties from correct examples, giving us the ability to reject bad inputs with no extra training or overhead. We experimentally validate our findings across a diverse range of datasets, domains, pre-trained models, and adversarial attacks.

Previous work has shown that a neural network with the rectified linear unit (ReLU) activation function leads to a convex polyhedral decomposition of the input space. These decompositions can be represented by a dual graph with vertices corresponding to polyhedra and edges corresponding to polyhedra sharing a facet, which is a subgraph of a Hamming graph. This paper illustrates how one can utilize the dual graph to detect and analyze adversarial attacks in the context of digital images. When an image passes through a network containing ReLU nodes, the firing or non-firing at a node can be encoded as a bit ($1$ for ReLU activation, $0$ for ReLU non-activation). The sequence of all bit activations identifies the image with a bit vector, which identifies it with a polyhedron in the decomposition and, in turn, identifies it with a vertex in the dual graph. We identify ReLU bits that are discriminators between non-adversarial and adversarial images and examine how well collections of these discriminators can ensemble vote to build an adversarial image detector. Specifically, we examine the similarities and differences of ReLU bit vectors for adversarial images, and their non-adversarial counterparts, using a pre-trained ResNet-50 architecture. While this paper focuses on adversarial digital images, ResNet-50 architecture, and the ReLU activation function, our methods extend to other network architectures, activation functions, and types of datasets.

We present evidence that many common convolutional neural networks (CNNs) trained for face verification learn functions that are nearly equivalent under rotation. More specifically, we demonstrate that one face verification model's embeddings (i.e. last--layer activations) can be compared directly to another model's embeddings after only a rotation or linear transformation, with little performance penalty. This finding is demonstrated using IJB-C 1:1 verification across the combinations of ten modern off-the-shelf CNN-based face verification models which vary in training dataset, CNN architecture, way of using angular loss, or some combination of the 3, and achieve a mean true accept rate of 0.96 at a false accept rate of 0.01. When instead evaluating embeddings generated from two CNNs, where one CNN's embeddings are mapped with a linear transformation, the mean true accept rate drops to 0.95 using the same verification paradigm. Restricting these linear maps to only perform rotation produces a mean true accept rate of 0.91. These mappings' existence suggests that a common representation is learned by models with variation in training or structure. A discovery such as this likely has broad implications, and we provide an application in which face embeddings can be de-anonymized using a limited number of samples.

Autonomous driving consists of a multitude of interacting modules, where each module must contend with errors from the others. Typically, the motion prediction module depends on a robust tracking system to capture each agent's past movement. In this work, we systematically explore the importance of the tracking module for the motion prediction task and ultimately conclude that the tracking module is detrimental to overall motion prediction performance when the module is imperfect (with as low as 1% error). We explicitly compare models that use tracking information to models that do not across multiple scenarios and conditions. We find that the tracking information only improves performance in noise-free conditions. A noise-free tracker is unlikely to remain noise-free in real-world scenarios, and the inevitable noise will subsequently negatively affect performance. We thus argue future work should be mindful of noise when developing and testing motion/tracking modules, or that they should do away with the tracking component entirely.