Javier Carnerero-Cano, Massimiliano Pronesti, Radu Marinescu, Tigran T. Tchrakian, James Barry, Jasmina Gajcin, Yufang Hou, Alessandra Pascale, Elizabeth M. Daly. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Retrieval-augmented generation (RAG) has emerged as a promising solution to mitigate the limitations of large language models (LLMs), such as hallucinations and outdated information. However, it remains unclear how LLMs handle knowledge conflicts arising from different augmented retrieved passages, especially when these passages originate from the same source and have equal trustworthiness. In this work, we conduct a comprehensive evaluation of LLM-generated answers to questions that have varying answers based on contradictory passages from Wikipedia, a dataset widely regarded as a high-quality pre-training resource for most LLMs. Specifically, we introduce WikiContradict, a benchmark consisting of 253 high-quality, human-annotated instances designed to assess the performance of LLMs in providing a complete perspective on conflicts from the retrieved documents, rather than choosing one answer over another, when augmented with retrieved passages containing real-world knowledge conflicts. We benchmark a diverse range of both closed and open-source LLMs under different QA scenarios, including RAG with a single passage, and RAG with 2 contradictory passages. Through rigorous human evaluations on a subset of WikiContradict instances involving 5 LLMs and over 3,500 judgements, we shed light on the behaviour and limitations of these models. For instance, when provided with two passages containing contradictory facts, all models struggle to generate answers that accurately reflect the conflicting nature of the context, especially for implicit conflicts requiring reasoning. Since human evaluation is costly, wealso introduce an automated model that estimates LLM performance using a strong open-source language model, achieving an F-score of 0.8. Using this automated metric, we evaluate more than 1,500 answers from seven LLMs across all WikiContradict instances.

The Kalman filter is a key tool for time-series forecasting and analysis. We show that the dependence of a prediction of Kalman filter on the past is decaying exponentially, whenever the process noise is non-degenerate. Therefore, Kalman filter may be approximated by regression on a few recent observations. Surprisingly, we also show that having some process noise is essential for the exponential decay. With no process noise, it may happen that the forecast depends on all of the past uniformly, which makes forecasting more difficult.Based on this insight, we devise an on-line algorithm for improper learning of a linear dynamical system (LDS), which considers only a few most recent observations. We use our decay results to provide the first regret bounds w.r.t. to Kalman filters within learning an LDS. That is, we compare the results of our algorithm to the best, in hindsight, Kalman filter for a given signal. Also, the algorithm is practical: its per-update run-time is linear in the regression depth.

In this paper, we propose a new framework for macroscopic traffic state estimation. Our approach is a robust “discretize” then “optimize” strategy, based on the Fourier-Galerkin projection method and minimax state estimation. We assign a Fourier-Galerkin reduced model to a macroscopic model of traffic flow, described by a hyperbolic partial differential equation. Taking into account a priori estimates for the projection error, we apply the minimax method to construct the state estimate for the reduced model that gives us, in turn, the estimate of the Fourier-Galerkin coefficients associated with a solution of the original macroscopic model. We illustrate our approach with a numerical example that demonstrates its shock capturing capability using only sparse measurements and under high uncertainty in initial conditions. We present implementation details for our algorithm, as well as a comparison of our method against the ensemble Kalman filter applied to a “local” discretization of the same traffic flow model.

An algorithm for the implementation of short-term prediction of traffic with real-time updating based on spectral analysis is described. The prediction is based on the characterization of the flow based on modal functions associated with a covariance matrix constructed from historical flow data. The number of these modal functions used for prediction depends on the local traffic characteristics. Although the method works well for the examples in this paper using the lower frequency modes, it can be adapted to include modes of higher frequency, as traffic conditions dictate. This paper describes the intended online implementation of the method that predicts within-day traffic flow using a forecasting horizon of 1 h 15 min with a 15-min step. Thus, every 15 min, the traffic flow for a further 1 h 15 min is predicted. As well as forecasting to this horizon, a second algorithm incorporating a weighted averaging technique is developed, which allows the prediction of one 15-min step ahead by using current and previous predictions of traffic flows at the given time instant while placing more weight on the more recent predictions. This technique combines the features of a time-series-based prediction with spectral analysis. The development of an algorithm for the real-time implementation is described, and results are presented for a number of different schemes.