Despite significant advances in large language models (LLMs), their knowledge memorization capabilities remain underexplored, due to the lack of standardized and high-quality testing grounds. In this paper, we introduce a novel, real-world and large-scale knowledge injection benchmark that continuously evolves without human intervention. Specifically, we propose WikiDYK, which leverages recently-added and expert-curated facts from Wikipedia’s “Did You Know...” entries. Each entry is converted into multiple question–answer pairs spanning diverse task formats from easy cloze prompts to complex multi-hop questions. WikiDYK currently contains 12,290 facts and 77,180 questions, and its design allows for seamless extension with future updates from Wikipedia editors. Through extensive experiments using continued pre-training, we reveal a surprising insight: despite their prevalence in modern LLMs, Causal Language Models (CLMs) demonstrate significantly weaker knowledge memorization capabilities compared to Bidirectional Language Models (BiLMs), exhibiting a 23% lower accuracy in terms of reliability. To compensate for the smaller scales of current BiLMs, we introduce a modular collaborative framework utilizing ensembles of BiLMs as external knowledge repositories to integrate with LLMs. Experiment shows that this framework further improves the reliability accuracy by up to 29.1%. Code: https://github.com/zhang-yu-wei/WikiDYK.

We present TRACE, a novel system for live common ground tracking in situated collaborative tasks. With a focus on fast, real-time performance, TRACE tracks the speech, actions, gestures, and visual attention of participants, uses these multimodal inputs to determine the set of task-relevant propositions that have been raised as the dialogue progresses, and tracks the group’s epistemic position and beliefs toward them as the task unfolds. Amid increased interest in AI systems that can mediate collaborations, TRACE represents an important step forward for agents that can engage with multiparty, multimodal discourse.

This project note describes challenges and procedures undertaken in annotating an audiovisual dataset capturing a multimodal situated collaborative construction task. In the task, all participants begin with different partial information, and must collaborate using speech, gesture, and action to arrive a solution that satisfies all individual pieces of private information. This rich data poses a number of annotation challenges, from small objects in a close space, to the implicit and multimodal fashion in which participants express agreement, disagreement, and beliefs. We discuss the data collection procedure, annotation schemas and tools, and future use cases.

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.

This paper discusses the challenges of annotating the predicate-argument structure of Chinese verb compounds in Uniform Meaning Representation (UMR), a recent meaning representation framework that extends Abstract Meaning Representation (AMR) to cross-linguistic settings. The key issue is to decide whether to annotate the argument structure of a verb compound as a whole, or to annotate the argument structure of their component verbs as well as the relations between them. We examine different types of Chinese verb compounds, and propose how to annotate them based on the principle of compositionality, level of grammaticalization, and productivity of component verbs. We propose a solution to the practical problem of having to define the semantic roles for Chinese verb compounds that are quite open-ended by separating compositional verb compounds from verb compounds that are non-compositional or have grammaticalized verb components. For compositional verb compounds, instead of annotating the argument structure of the verb compound as a whole, we annotate the argument structure of the component verbs as well as the semantic relations between them as creating an exhaustive list of such verb compounds is infeasible. Verb compounds with grammaticalized verb components also tend to be productive and we represent grammaticalized verb compounds as either attributes of the primary verb or as relations.