Seeing frames in order does not mean representing time. Modern VideoLMs receive ordered video streams, yet their main supervision acts on generated text rather than video-token representations where event dynamics should first emerge. This mismatch allows models to learn temporal answers from shortcuts such as objects, scenes, and language priors, without requiring internal video representations to capture event progression. To address this, we propose VT-Contrast, a representation-level temporal counterfactual objective for VideoLMs. Its design asks where temporal supervision should act and what temporal differences it should expose. VT-Contrast supervises selected late-layer last-frame video tokens, where temporal information is expected to be integrated before language generation, and contrasts order-preserving views with same-video reordered counterfactuals graded by Kendall tau distance. It requires no architectural changes, is compatible with diverse VideoLM training tasks, and improves overall performance across temporal understanding benchmarks. Our code is available at https://github.com/ANDgate99/VT-Contrast.
Wenqi Pei, Henry Hengyuan Zhao, Yilai Liu +4cs.CV cs.AI
Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video-LLMs. Given the beginning and ending clips of a video, models must identify the true missing middle from four candidates. TempCloze contains 1,521 carefully filtered videos from seven sources, mainly long-take and egocentric videos. We construct same-source distractors along three dimensions: Semantic asks what event should happen, Alignment probes when it should occur, and Progression tests how it should unfold, while shared scenes and objects reduce appearance cues. Our evaluation of 10 proprietary and 21 open-source Video-LLMs reveals Alignment as the primary bottleneck: models often recognize plausible semantic content and local event progression but struggle with temporal alignment. We further conduct error pattern and behavioral sensitivity analyses on TempCloze-Mixed and TempCloze-Hard with four representative models to examine where errors arise and how candidate order, context direction, visible span, frame density, and test-time scaling influence model choices.
Video-text models adapted from image-text architectures (e.g., CLIP) frequently exhibit temporal blindness, the inability to perceive fundamental cues like order, direction, and motion dynamics. Standard datasets mask this limitation by enabling models to exploit static spatial shortcuts. To systematically evaluate this, we introduce XTE-Bench, a diagnostic probe revealing that even large-scale video-language models struggle with basic temporal reasoning, indicating that parameter scaling alone is insufficient to resolve this flaw. To address this, we propose Cross-Modal Temporal Edits (XTE), a self-supervised framework that injects precise temporal supervision. By performing synchronized video-text transformations, XTE generates hard temporal negatives without manual annotation. We instantiate this with ViTAL-X, a lightweight model that equips frozen image-text backbones with temporal awareness while preserving their foundational spatial knowledge. Across six temporal benchmarks, ViTAL-X achieves state-of-the-art performance. Utilizing only 0.4B parameters and 1M training clips, ViTAL-X outperforms 7B-parameter models and surpasses baselines trained on 600x more data. These results demonstrate that targeted, high-quality temporal alignment provides a highly efficient alternative to pure scaling.
Temporal inconsistencies, such as mandates attributed outside their real interval, events presented as past before they occurred, or inverted causal sequences, are a form of political disinformation that evades style-based fake news detectors: a well-written article with a single wrong date carries no lexical signal of falsehood. This paper introduces the Temporal Coherence Score (TCS), a continuous, intrinsically interpretable metric that quantifies the temporal coherence of a news article, computed by a four-stage pipeline: extraction of temporal facts, construction of a temporal knowledge graph, hierarchical verification against internal consistency rules and external reference sources, and score aggregation with automatically generated explanations. Verification combines eight internal checkers derived from Allen's interval algebra with a five-level external hierarchy ranging from a locally stored reference knowledge base of 1{,}256 curated political facts to live Wikidata SPARQL queries. On a benchmark of 100 political news articles with injected temporal errors, the system reaches a precision of 0.909 at the selected operating threshold, with a single residual false positive, a profile deliberately tuned for human-in-the-loop fact-checking assistance, where false alarms are costlier than missed detections. Unlike lexical baselines that output only a binary label, every flagged article is accompanied by the inconsistency type, the entities involved, and the reference source that contradicts the claim.
History is not preserved in complete, continuous form. Accounts of a person's activities, relationships and historical contexts are scattered across texts, chapters and narrative perspectives; historians must retrieve, identify and compare these materials to reconstruct temporal sequences and verify them against sources. Here we present AI Historian (AIH), an AI agent system that helps historians organize person-time evidence from dispersed biographical narratives. It takes source sentences as evidence units, identifies people and temporal cues, verifies candidate cross-text associations and infers comparable temporal ranges while preserving traceable source-text evidence. We evaluated AIH on six Shiji cases concerning Liu Bang, Xiang Yu and Xiao He. AIH Agent achieved a temporal-localization MicroIoU of 86.2%, compared with 81.3% for human-only annotation and 17.1% for direct large-language-model prompting; it required about 14 min, versus 1 h 32 min for human-only annotation. We further applied AIH to the Twenty-Four Histories and other ancient Chinese histories, ancient Japanese and Korean histories, and modern and contemporary historical materials, and released the results through Westlake Historian. These results indicate that AIH can reduce the cost of organizing historical materials at scale while turning connections obscured by chapter-based narration into traceable, revisable research questions for collaborative testing.
Longitudinal radiology report generation (LRRG) requires identifying both current findings and their changes relative to a prior study. Existing methods jointly model diagnosis, attribute estimation, temporal comparison, and language generation within implicit representations, which can cause task interference, obscure the evidence underlying each decision, and limit error traceability. They also model progression states as independent labels, ignoring their ordered structure and thus treating missed changes and direction reversals equally. We present STRIVE, Multi-Agent Structured Temporal Reasoning with Integrated Verification for LRRG, which decomposes clinical reasoning into specialized Diagnosis, Attribute, and Temporal Change Agents that produce explicit intermediate evidence. In particular, the Temporal Change Agent is further post-trained using Progression-Aware GRPO, a verifiable, shaped reward that assigns partial credit to direction-preserving errors while scoring direction reversals lowest. STRIVE performs verification at two stages: a deterministic Consistency Gate reconciles the agent outputs before report generation, and a Validation Agent checks whether the generated report is supported by the aggregated clinical evidence. On Longitudinal-MIMIC, STRIVE attains the best clinical efficacy among recent methods and more than doubles Longitudinal Change Concordance (LCC), a measure of temporal agreement with the reference report, over the strongest baseline.
Marek Hradil, Danae Sánchez Villegascs.CL cs.AI cs.CV
Vision-language models (VLMs) achieve strong performance on video and image-sequence benchmarks, yet it remains unclear whether they capture temporal structure. To study this question, we formulate temporal grounding as an anomaly detection problem, providing a simple and controlled evaluation that directly tests sensitivity to temporal consistency. We introduce TimeCatch, where temporal anomalies are created by swapping consecutive frames and frame-level anomalies by replacing a frame with Gaussian noise. Models are evaluated on anomaly detection and localization tasks across four synthetic and real-world datasets, alongside a human study. Our evaluation reveals a substantial gap between frame-level and temporal anomaly detection. While VLMs consistently detect frame-level anomalies and often localize them accurately, they perform near chance on temporal anomaly detection and only modestly above chance on localization. Humans, in contrast, achieve near-ceiling performance on both tasks. Additional analyses across model scales, prompting strategies, sequence lengths, and visual similarity suggest that these failures cannot be explained solely by limitations in perception or model capacity. Together, these findings indicate that current VLMs can identify anomalies within individual frames but struggle to integrate information across frames to reason about temporal consistency. TimeCatch provides a controlled benchmark for evaluating temporal grounding in vision-language models.
In long-term collaboration spanning multiple meetings, factual states such as decisions and risks are continually revised, overturned, and replaced. Existing long-context methods typically stack the entire history, while many RAG and structured-memory methods organize knowledge as static or append-only facts and rely on semantic relevance at read time. Without explicit modeling of knowledge lifecycles, these approaches may retain conflicting old and new states simultaneously or discard history, leading to stale retrieval and answers that are difficult to verify. We present EvoWiki, an incremental question-answering architecture for dynamic long-form text. EvoWiki decouples offline incremental construction (BUILD) from online structured reading (READ). BUILD captures the intra-meeting micro-evolution from proposal to decision and uses entity version chains and a fine-grained State-Overwrite Protocol to explicitly distinguish current valid states from superseded history while preserving meeting-level provenance anchors. READ bypasses relevance-based Top-k retrieval and performs deterministic entity addressing, temporal resolution, and cross-entity multi-hop aggregation over the complete Wiki to produce grounded and traceable answers. We further introduce CrossMeet, a high-fidelity bilingual benchmark designed to simulate long-term state evolution, covering factual consistency, temporal reasoning, and cross-meeting multi-hop reasoning. Across six datasets and two reader models, EvoWiki improves macro-average Judge Accuracy over the strongest baselines by 9.72 and 10.00 percentage points, respectively. Human evaluation shows that EvoWiki is more robust and factually faithful under frequent state flips, validating valid-state-oriented reading as a reliable approach to cross-meeting knowledge evolution.
Chenghua Zhu, Zhaolu Kang, Qifan Shi +8cs.CV cs.CL cs.LG
Video multimodal large language models have advanced significantly, yet fine-grained motion-temporal understanding remains fragile. The core bottleneck is not only sparse frame sampling, but also the lack of a complete temporal modeling pipeline for explicitly representing frame-to-frame change, enabling appearance-motion interaction, and optimizing temporal direction sensitivity. We propose COMET, a temporally grounded framework that systematically strengthens video MLLMs through explicit temporal representation, appearance-motion fusion, and direction-aware optimization. Architecturally, COMET introduces a temporal motion branch built on Taylor frame differences and injects its motion evidence into the appearance stream via temporal attention bias-enhanced cross-attention. For optimization, COMET combines temporal prior distillation with a forward-reverse TC-GRPO stage that turns temporal order into a direct learning signal and strengthens the model's use of directional motion patterns encoded by the temporal motion branch. The method achieves consistent overall improvements with a pronounced motion-temporal bias: on Qwen3-VL-8B, action-centric tasks (STAR, SSv2) improve by 4.9% on average, temporal reasoning tasks (NExT-QA, CLEVRER, LLaVA-178K) by 2.1% over BL-GRPO, while static perception tasks (PerceptionTest) remain on par. The same gain pattern also transfers to InternVL2.5-8B, indicating that COMET generalizes across model families.
While multimodal large language models (MLLMs) excel in medical applications, most of them favor static images or short-term signals. In the critical field of dynamic electrocardiograms (ECG), models struggle with complex temporal reasoning and diagnostic report generation due to a lack of high-quality datasets and benchmarks. To address this, we introduce (i) Holtercare-23K, a large-scale multimodal dynamic ECG dataset comprising 22,980 QA pairs derived from 788 clinical Holter records and featuring a novel signal-video-text tri-modal alignment. Based on this dataset, we present (ii) Holtercare-Bench, a multimodal benchmark that evaluates models on temporal localization, clinical diagnosis, and global summarization. Zero-shot evaluations of leading MLLMs reveal a significant performance gap in processing ultra-long pathological sequences. However, fine-tuning representative models yields substantial improvements. This work illuminates the limitations of current MLLMs in electrophysiology and provides a foundational benchmark for long-term medical MLLMs. Our project is available at https://github.com/ZJU4HealthCare/Holtercare-Bench.
Xinlong Dai, Jinchuan Zhang, Lei Gao +3cs.CL cs.AI
By leveraging large-scale pretraining, LLMs can interpret diverse temporal expressions and question formulations without task-specific training. However, existing prompt-based neuro-symbolic systems continue to rely on LLMs for both semantic interpretation and exact temporal inference. Consequently, discrete decisions regarding intervals, time anchors, and ordered states remain vulnerable to probabilistic errors and difficult to verify. We present STAIR, a \textbf{S}emantic-\textbf{T}emporal \textbf{A}utomaton for \textbf{I}nterpretable \textbf{R}easoning. STAIR separates semantic interpretation from precise temporal inference: an answer-free LLM adapter maps complex question formulations to normalized temporal intents, while a deterministic temporal automaton with finite control and guarded transitions executes the corresponding policies over canonicalized evidence. Following a rule-first design, STAIR resolves standard questions without invoking an LLM and applies semantic adaptation only when the rule path fails to produce an executable intent. This approach reduces free-form reasoning, making temporal decisions verifiable and interpretable. Specifically, guarded execution supports precise point-time containment and before/after selection, while semantic adaptation handles non-exact intervals and time-anchored queries. Across the TimeQA-Easy, TimeQA-Hard, TempReason-L2, and TempReason-L3 datasets, STAIR consistently outperforms strong baselines in the TQA task using matched model settings, achieving average F1 improvements of 16.57\% and 3.10\% when utilizing the Qwen2.5-7B and GPT-4o-mini models, respectively. Furthermore, ablations and diagnostic analyses demonstrate that STAIR excels at handling both boundary-sensitive and order-sensitive queries, while its guarded execution and semantic adaptation ensure precise point-time reasoning and inexact intervals, respectively.
Suze van Adrichem, Aditi Bhaskar, Diyi Yang +2cs.CL cs.LG
A consistent concept of the current time is important for temporal reasoning, yet how language models represent the current time is not well understood. We contribute two tasks that probe the current year in conceptually distinct ways: an associative task, which infers the current year from verb tense, and a declarative task, which directly queries for the current year. Both tasks estimate current years within one year of the post-training data cutoff of instruction-tuned language models. For base models, predictions on the associative task serve as a strong proxy for the pre-training data cutoff, with an average error of only 10 months across 13 models. However, their internal mechanisms diverge: the associative task uses mechanisms similar to factual recall, while the declarative task lacks consistent causal pathways. This divergence poses a challenge for updating the current year in language models. None of prompting, SFT, or weight editing succeed in shifting the associative and declarative years simultaneously. Prompting updates the declarative year (94.6% success across 351 target years) but leaves the associative year nearly unchanged (1.7% success). Year-shifted SFT also fails to shift the associative year, matching the target year in only one of eight models. Weight editing, while effective for both tasks individually, does not generalize across both. Overall, our results show that the current year is not consistently encoded in language models: The associative notion, deeply ingrained in linguistic structures learned in pre-training, uses different causal mechanisms and resists the same modifications that easily shift the declarative notion learned in post-training.
Human perception of time is subjective. Well-documented temporal illusions show that the brain relies on context and relational cues for judging duration instead of tracking elapsed time directly. Prior studies established these effects with visual and auditory stimuli. Existing LLM evaluations of temporal perception focus on estimating event durations or multi-step temporal reasoning. In this work, we investigate whether written narratives alone can evoke human temporal illusions, using a new benchmark of 6,684 narrative pairs spanning five illusions. We find that human readers (60 participants) prefer expected scenarios in only two of the five illusions, those where the manipulation is directly visible in text rather than requiring readers to internally simulate duration. We evaluate 14 LLMs on the same benchmark. Surprisingly, we find that models pick the literature-predicted scenario across four of the five illusions, diverging from human behavior. Reasoning traces show that ~70% of responses explicitly evoke psychology research, suggesting that this alignment is consistent with retrieval of published findings rather than human-like temporal biases.
Miss a filing deadline by one day and the claim is barred, however strong the case. Computing that deadline is rarely simple: the period runs from a triggering event, is counted by a statutory convention, and may be suspended by a mandatory conciliation window. We ask whether a language model should answer such questions directly, or read the document and leave the arithmetic to code. We extract dated facts and their dependencies into a temporal dependency graph and compute deadlines from it with a calendar-correct engine. On UK Employment Appeal Tribunal judgments the engine reproduces six of seven timeliness rulings, and matches the judges' own dates to the day. The strongest of four language models, asked the same cases, gets the arithmetic right and the answer wrong: in six of twenty-one responses its stated verdict contradicts its own thinking, and every contradiction runs the same way, calling a late claim timely. To test the systems at scale we move the dismissal date across the statutory boundary, generating 427 cases whose answers are computed rather than annotated. On the cases both systems answer, the pipeline is right 90.2% of the time against 61.2% for direct answering. The limit is extraction: on contracts the errors are almost never in the arithmetic, but in choosing which event the period starts from.
Graph RAG connects facts no single passage states, but implementations pay three times: in infrastructure, keeping vector store, graph database and document store in sync; in quality, because a pipeline that never refuses extractor output stores edges that assert nothing; and over time, because a graph that only accumulates treats superseded and current facts alike. post-graph-rag is an open-source engine addressing all three. Chunks with embeddings, a canonical entity graph and community summaries live in one PostgreSQL database, with pgvector for search and edge tables for traversal. Extraction output is validated before writing: vague predicates, pronominal names and bare quantities are rejected, predicates normalise onto an optional vocabulary, entities resolve to one vertex per canonical name, and denials keep the positive predicate under a negation flag. A bi-temporal layer records when a relation held and when the system believed it, superseding incompatible earlier assertions from document order. Against LightRAG on three corpora with extraction and embedding models fixed, it builds a denser graph everywhere, up to $2.4\times$ the relations per entity, and a more queryable one: 0.46-0.58 distinct edge labels per relation against 0.77-1.33. It supersedes 13 and 8 relationships where the baseline, having no temporal model, supersedes none. On LongMemEval, 500 questions of long-horizon chat memory, it scores 85.8 percent with gemini-3.6-flash against 71.2 for Zep's gpt-4o and 60.2 for a full-context baseline, leading on all six question types. The largest single contribution is temporal grounding in the prompt: carrying each relation's validity period through to synthesis moves temporal reasoning from 0.496 to 0.881, ablated paired on one graph per instance. Code: post-graph-rag https://github.com/crajah/post-graph-rag; post-graph https://github.com/crajah/post-graph
Vision-Language-Action (VLA) models have recently achieved promising performance in robotic manipulation. However, existing benchmarks mainly evaluate generalization on static manipulation tasks and largely overlook dynamic interaction scenarios. To address this gap, we present ReflexBench, a benchmark for reaction-critical manipulation. ReflexBench contains six dynamic tasks and introduces an evaluation framework that decouples simulator stepping from robot control while supporting configurable latency under synchronous and asynchronous inference. Building upon ReflexBench, we propose ReflexVLA, an efficient VLA model designed for reaction-critical manipulation without large-scale robot-data pretraining. ReflexVLA enhances temporal reasoning through latent future prediction and multi-frame temporal fusion within the vision backbone, while reducing deployment latency through batched visual encoding and CUDA Graph replay. Experiments show that ReflexVLA consistently improves dynamic manipulation performance while maintaining competitive accuracy on standard static manipulation benchmarks, and real-world experiments further demonstrate its effectiveness under practical deployment conditions. Project website: https://reflexvla.github.io
Manufacturing knowledge graphs that integrate data from heterogeneous industrial systems face a trust deficit: consumers cannot determine whether queried data is valid, whether it was valid when a decision was made, where it originated, or how it was acted upon. We argue that four trust capabilities -- SHACL validation, PROV-O provenance, domain-aware bi-temporal versioning, and graph-native decision objects -- compose through shared correlation identifiers to produce emergent trust properties that no single capability delivers alone. We present a composable trust infrastructure that integrates these four capabilities into a unified RDF architecture. Capabilities compose through shared entity URIs, ingestion activity identifiers, and temporal correlation keys, enabling compound queries spanning all four dimensions. An experimental ablation confirms that removing any single capability causes exactly three of six composition queries to fail, demonstrating that all four are equally load-bearing. Analysis of higher-order compositions reveals four emergent three-way properties and one irreducible four-way property (full-chain auditability, 31ms execution). The infrastructure is validated on a testbed integrating eleven industrial sources -- OPC UA, TIA Portal, eClass, AAS, ISA-95, ISA-18.2, SAP S/4HANA, Teamcenter, Opcenter EX, Insights Hub, and SCM -- under an 89-class ISA-95-aligned ontology. The unified graph contains 8,743 triples across five named graphs, stitched by 81 owl:sameAs identity edges. Evaluation uses simulated but structurally realistic data from purpose-built emulators; data structures and cross-system linkage patterns are representative of real industrial installations.
Chengyang He, Tahreem Arif, Marko Zivkovic +3cs.CL cs.AI cs.IR
Understanding the temporal progression of symptoms in clinical narratives is critical for disease monitoring, safety surveillance, and causality assessment. Clinical narratives, however, rarely provide explicit temporal anchors. Current approaches to temporal information reasoning focus predominantly on pairwise relation classification across multi-visit and timestamp-rich records, leaving the reconstruction of structured symptom trajectories from individual anchor-sparse reports largely unaddressed. We propose CRAFT, an LLM framework that pairs a generator with a constraint-based verifier to iteratively produce and refine stage-wise symptom timelines through targeted feedback. We conduct evaluation on MedTempo, a new benchmark of 5,347 vaccine adverse-event narratives spanning three COVID-19 vaccine types, with expert-validated temporal stage annotations for 3,166 reports. Experiments across four LLM backbones demonstrate that CRAFT consistently improves temporal ordering accuracy, with ablation analysis isolating the contribution of generator and verifier components across model capability levels.
Gist-based context compression---summarising older conversation history into compact representations---is a common approach in long-horizon language model agents, yet its effect on different types of memory retrieval is poorly understood. We use Salience-Weighted Consolidation (SWC), a biologically-inspired compression framework motivated by sleep-based memory consolidation, as a diagnostic probe to study when gist compression helps and when it hurts. SWC scores conversation history by salience, partitions it into priority tiers, and applies structured gist abstraction to mid-priority content. Evaluating four conditions on all ten LoCoMo conversations---1,935 matched text-only questions in total, 1,501 used in the primary aggregate after excluding Category 5 (adversarial) questions---at temperature 0, we find a consistent task-type interaction: gist compression substantially outperforms truncation on multi-hop reasoning and single-hop factual questions, but temporal questions remain substantially harder under compression, with compressed conditions scoring well below the full-context reference on the conversations where both are evaluated. We trace this failure to a specific mechanism: the gist abstraction prompt preserves relational and event structure while discarding dates and times. A preservation analysis across all ten conversations confirms the mechanism: an approximately 20-fold increase in temporal expression preservation (3.05% to 62.39%) with a one-sentence prompt modification, while named entity and event preservation rates barely change (x1.02 and x1.11), demonstrating that the fix is a precision instrument. The prompt modification recovers +0.314 [0.254, 0.375] judge accuracy on category-2 (temporal) questions in the matched set. Code and results: https://github.com/kyrkewood/sleeping-agent.
Logic Tensor Networks (LTN) provide a neurosymbolic framework in which first-order logic is interpreted through tensor operations, enabling logical constraints to be integrated with differentiable learning. However, the original formulation of LTN is primarily suited to data represented as flat collections of individuals, and does not explicitly capture structural organization such as temporal order, sequential position, or graph connectivity. We introduce sLTN, an extension of LTN that makes structural dimensions first-class elements of the language. Structural dimensions represent named tensor axes associated with domain-specific organization, such as time steps, sequence positions, or graph nodes. They can be quantified explicitly, related through structural relations, and used to express temporal, sequential, and relational constraints directly at the logical level. We formalize the syntax and fuzzy tensor semantics of sLTN and show that, in the absence of structural dimensions, the framework recovers the original LTN semantics as a special case. We further describe a PyTorch implementation based on a declarative signature, formula parsing, and tensorial interpretation. The framework is illustrated on representative temporal and sequential reasoning examples. This paper serves as a companion to the sltn library, available at https://github.com/logictensornetworks/sltn.
Martina Ianaro, Guilherme Fernandes, Maurizio Gabbrielli +1cs.CV cs.CL
As generative multimedia evolves from static image synthesis to complex, interleaved visual narratives, a foundational bottleneck has emerged: the judgment crisis. While human perception naturally synthesizes the temporal and logical flow of a story, automated evaluation systems remain largely "blind" to sequential continuity, often failing to distinguish between a coherent narrative and a semantically shuffled or contradictory sequence. This work identifies a critical structural gap in current multimodal evaluation paradigms, arguing that the reliance on Large Vision-Language Models (LVLMs) as judges is fundamentally limited by architectural biases. Our analysis reveals a profound performance dichotomy: while models may appear competent in isolated pointwise scoring, they suffer a catastrophic collapse when required to perform pairwise discrimination of temporal order. We demonstrate that this is not merely a data-scarcity issue but a structural one. Through a series of diagnostic probes, we uncover systematic positional asymmetries, specifically primacy and recency effects, where a model's judgment of a story is significantly influenced by the placement of a frame, often more than by its semantic consistency. These biases, potentially rooted in causal masking and rotary embeddings, suggest that current transformer-based judges are inherently ill-equipped for long-form visual reasoning. By exposing these blind spots, we challenge the multimedia community to move beyond snapshot-centric metrics and instead pioneer Temporally-Aware Evaluation paradigms that treat visual sequences as unified logical structures rather than unordered collections of frames.
Rose Cymbler, Daniel Guez, Laurent Fabrecs.CL cs.AI cs.IR
We identify and quantify temporal misgrounding: the systematic retrieval and citation of the currently in-force version of a legal article when the applicable version is an earlier or future one. Standard legal RAG treats the corpus as static; we argue legal question answering is a temporally-indexed retrieval problem. We introduce FiscalQA Pro, pairing a versioned corpus of 32,436 article-versions of the French tax code (93 years, 1938-2031) with an all-model-hard temporal-reasoning track: 209 scored, expert-reviewed questions across 33 CGI articles (221 released; twelve flagged out of the answerable scope). At selection time, no evaluated model recovered its date-applicable answer closed-book in any of four sampling draws, and the currently in-force text lacks the gold value for all but one of the scored questions. Answers are scored deterministically via atomic ground-truth "nuggets" (regex and numeric-with-tolerance), never LLM-as-judge: an LLM judge would inherit the temporal bias it is meant to score. Across eleven models (five frontier closed-API systems plus Gemini 2.5 Pro as a substitute entry, and five open-weight), parametric knowledge yields 3.0% mean strict accuracy and RAG over a static current-version corpus 2.7%. Static RAG retrieves the date-applicable version 0% of the time, confidently citing a real but inapplicable version. Our end-to-end retriever over a multi-version index, with no oracle, reaches 98.3% mean strict; an oracle-article ablation reaches 99.1%, locating the residual gap in first-stage recall, not version selection. We additionally release a version-aware jurisprudence dataset of 69,208 citation links, together with the corpus, benchmark, model responses, and pipeline code.
Sam Siavoshian, Omar Ramadan, Amir K. Saeed +3cs.AI
Temporal decisions in language-model systems often depend on both symbolic task state and elapsed wall-clock time, such as cache expiration, job completion, quota resets, deadlines, or stale sessions. We study whether elapsed time can be supplied as a non-token, system-side scalar and composed with visible symbolic state by a frozen-backbone language model. We introduce ChronoState, a compositional temporal-state benchmark in which symbolic state appears in the prompt, elapsed seconds tau are supplied through a hidden chronometric-injection channel, and the model selects a forced-choice temporal action. Here, "hidden" means hidden from the user-visible token sequence, not from model computation. Using Qwen2.5-3B-Instruct as a frozen bf16 backbone with a 31-dimensional sinusoidal-plus-log time encoding, gated FiLM residual modulation, and a rank-8 LoRA action surface, hidden-time CI reaches 0.9305 +/- 0.0134 accuracy and 0.9410 +/- 0.0103 balanced accuracy. No-time and shuffled-time controls fall to 0.5511 +/- 0.0042 and 0.3323 +/- 0.0097, respectively, with high shuffled-time wrong-state consistency supporting causal dependence on the injected scalar within the trained distribution. Generalization remains strong for held-out templates, durations, and multi-constraint compositions, but held-out quota-family transfer is weak at 0.5065 +/- 0.0559, while a fair prompt+LoRA timestamp baseline reaches 0.9893 +/- 0.0052. Thus, ChronoState supports a narrow conclusion: hidden elapsed time can be composed with symbolic task state under direct supervision, but does not establish autonomous time tracking, broad unseen-family abstraction, or superiority over prompt-injected timestamps.
Large language models (LLMs) have demonstrated strong performance across a wide range of legal tasks. In legal practice, time is a critical concept that governs the validity of statutes, the progression of legal cases, and the enforcement of procedural deadlines. However, legal temporal capabilities remain underexplored in existing legal AI benchmarks. To address this gap, we propose LexKairos, a comprehensive benchmark for evaluating the temporal capabilities of LLMs in the Chinese legal context across three dimensions: statutory temporal knowledge, case temporal modeling, and statute-case temporal reasoning. LexKairos comprises nine sub-tasks drawn from real-world Chinese judicial cases and statutes. We conduct systematic evaluations of eight LLMs under multiple inference settings, including vanilla, Chain-of-Thought (CoT), and thinking modes. Our results show that Gemini-3-Flash achieves the strongest overall performance, yet even the best-performing model exhibits notable limitations on tasks demanding precise time-sensitive statutory metadata recall or complex reasoning in time limits, indicating that legal temporal knowledge and reasoning remain open challenges for current LLMs. Data and code are available at https://github.com/thunlp/LexKairos.
Large language model (LLM) agents that assist users over weeks of conversation must remember what is currently true, not merely what was once said. Flat RAG diaries and Markdown logs optimize needle retrieval but under-serve currency, provenance, and ordered temporal reasoning (Maharana et al. 2024; Wu et al. 2024; Packer et al. 2023; Chhikara et al. 2025). We present SodaMem, an evidence-grounded temporal graph memory that (i) extracts typed FactEvents with mandatory provenance spans, (ii) persists mention time, occurrence time, and validity with SUPERSEDES/CONTRADICTS/UPDATES edges under hybrid lexical-dense indexing, and (iii) answers via a planner-reader loop that gathers citable evidence before composing a final response. On LongMemEval-S, our store-of-record configuration reaches 92.8% accuracy (464/500; best of N=3) at mean $0.00161/question (approximately 18.3k tokens; median $0.00111 / approximately 14.6k) with deepseek-v4-flash. We compile public systems with estimable API cost into a cost table and cost-accuracy map; under these estimates SodaMem sits near the accuracy frontier at Flash-tier spend and strictly dominates several higher-cost, lower-accuracy points. Accuracy uses the same Flash model as reader and judge (self-grading); costs exclude ingest/judge and cross-system comparisons are compiled estimates rather than a single-harness bake-off.Our code is available at https://github.com/SodaMem/SodaMem
Real-world video benchmarks provide broad coverage, but their fixed clips entangle event count, rate, duration, and visual complexity, making failure modes hard to isolate. While existing programmatic benchmarks offer better control, they score only the final answer rather than auditing reported events against executable ground truth. To bridge this gap, we introduce trace-grounded parametric profiling for event counting in three controlled video tasks: bouncing-ball wall contacts, visual blinks, and categorical state transitions. Across 2,190 videos, we vary event count N and frequency F while holding rendering fixed. Each video includes an executable event trace for capability-surface estimation and timestamp-level evaluation. Our results reveal a staged temporal failure. At an 80% reliability threshold, Gemini 3.6 Flash reliably counts persistent state transitions up to 12 events at 0.5 and 1.0 Hz, yet demonstrates no reliable positive-count region for transient blinking events. Thus, event representation dictates whether a model initially accesses evidence -- a limitation that compounds as count and frequency increase. In the high-count, high-frequency regime, only 0.2% of final counts are correct and the model recovers just 18.1% of true events. To test if visual access is the primary bottleneck, we increase sampling rate. Although this boosts Bounce Ball accuracy from 19.6% to 29.3%, the reported sequence agrees with ground truth only 3.7% of the time. Extra frames can therefore inflate final scores without producing faithful event recovery. Different prompting strategies yield similarly limited gains, and real-world video evaluations show the same concentration of success at low event counts. Ultimately, trace-grounded profiling shifts video evaluation from aggregate accuracy metrics to a detailed diagnostic of where temporal reasoning fails.
Multimodal large language models excel at passive perception but struggle with complex visual cognitive tasks requiring multi-step temporal reasoning. This degradation largely stems from the inherent ambiguity of language-based reasoning, which often fails to accurately articulate continuous visual transformations. To address this, we propose ChronoVision, a multimodal framework designed to align visual logic with latent imagery. During supervised fine-tuning, a Reconstructive Visual Head predicts the latent representation of the final transformed state, while an ROI Attention Locating module focuses the model on key visual evidence via semantic span queries. In post-training, we apply reinforcement learning with an implicit process grounding mechanism, guided by a composite reward function that evaluates outcome correctness, latent process alignment, and unsupervised visual focus. Furthermore, we introduce Vbvr-VQA, a novel dataset that evaluates temporal tracking by reformulating video reasoning into a strict image-ordering task. Experiments demonstrate that ChronoVision achieves state-of-the-art performance on Vbvr-VQA with 74.8% in-domain and 71.6% out-of-domain accuracy, alongside a strong 55.0% accuracy on IntPhys2, a highly challenging cross-domain benchmark.
Acting in a physical scene requires knowing its real later state, not a plausible one. Current evaluations often accept words or a realistic-looking image, so the predicted state is never checked against the true one. We introduce DynaPix (Dynamic Pixels), a benchmark that makes prediction checkable. Given a video clip that stops before a key event and a question about a later moment, a model must pick the true future image from close candidates or a large gallery. The scenes come from a physics simulator, so the correct image and its time are known exactly and the wrong options are deliberately similar. Models often succeed when a visible event marks the target moment, but are near chance when only elapsed time marks it. Gallery search is harder still, as the true image rarely ranks first. People handle the elapsed-time items well, so the difficulty lies with the models, not the questions. Training on scene accounts drawn from the simulator's true record, not a teacher's guess, repairs much of this but not the longer elapsed time case. DynaPix thus exposes a temporal-anchoring gap: models attach a prediction to an event far better than to time itself.
Understanding complex surgical scenes requires recognizing multiple interdependent entities, such as instruments, actions, and targets, while maintaining their relational consistency across time. Existing surgical triplet recognition methods struggle to jointly model intra-frame label dependencies and inter-frame temporal semantics in a unified manner. To address these limitations, we propose a unified framework that integrates spatial, relational, and temporal cues for robust surgical triplet recognition. Specifically, class-specific spatial priors are first extracted through a multi-scale encoder. These priors are then refined by a Label Correlation Modeling module with multi-scale class activation map-guided relational extraction (MS-CAMRE), enabling the model to capture both static co-occurrence patterns and dynamic contextual dependencies among triplet components. Furthermore, a Bidirectional Temporal-Relational Fusion Attention (BTRFA) module harmonizes temporal and relational representations to achieve coherent temporal reasoning. We also introduce a new evaluation metric, the Triplet Consistency Error Rate (TCER), which quantitatively measures the model's ability to preserve causal and semantic consistency across triplets. Extensive experiments on the CholecT45 and ProstaTD datasets show that our method achieves state-of-the-art performance, improving AP_IVT by 5.1 percent and 7.8 percent, respectively. Moreover, according to TCER, our approach achieves relative reductions of more than 36 percent and 25 percent on the two datasets, respectively, demonstrating the effectiveness of our framework in temporal-relational co-reasoning.
Vision-language models for autonomous driving primarily rely on cameras and LiDAR, leaving 4D radar largely unexplored as a standalone perceptual modality despite its robustness to adverse visibility and direct measurement of radial velocity. We introduce Radar4D-VLM, a radar-only temporal vision-language model that reasons from ten consecutive 4D-radar point-cloud sweeps without camera or LiDAR input. Radar4D-VLM extracts geometrically grounded object proposals and organizes radar evidence into a compact hierarchy of object, scene, and kinematic tokens. A parameter-efficient projector maps these tokens into frozen language backbones, while auditable prediction heads jointly model object count, spatial distribution, motion state, collision risk, semantic category, and radial velocity. Radar4D-VLM combines proposal-grounded temporal object tokenization, global scene context, and explicit kinematic tokens within a unified frozen-backbone interface. On sequence-isolated K-Radar development validation, its Top-64 proposal recall reaches 98.13% at 4 m, exceeding fixed-lattice and uniform-random controls by 6.40 and 22.83 percentage points, respectively. We further evaluate 24 matched runs spanning eight frozen Qwen, Phi, Mistral, Llama, and Gemma backbones under an identical adaptation budget. The radar-token interface remains compatible across all five language-model families, while matched aligned, permuted, and no-language controls show sensor dependence but no stable direct-head gain from aligned language supervision. These results establish a reproducible foundation for radar-only multimodal scene and motion reasoning while separating interface compatibility from the benefit of language supervision.