Woomin Song, Saket Dingliwal, Sai Muralidhar Jayanthi, Bhavana Ganesh, Jinwoo Shin, Aram Galstyan, Sravan Babu Bodapati. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

As large language models increasingly gain popularity in real-world applications, processing extremely long contexts, often exceeding the model’s pre-trained context limits, has emerged as a critical challenge. While existing approaches to efficient long-context processing show promise, recurrent compression-based methods struggle with information preservation, whereas random access approaches require substantial memory resources. We introduce REFORM, a novel inference framework that efficiently handles long contexts through a two-phase approach. First, it incrementally processes input chunks while maintaining a compressed KV cache, constructs cross-layer context embeddings, and utilizes early exit strategy for improved efficiency. Second, it identifies and gathers essential tokens via similarity matching and selectively recomputes the KV cache. Compared to baselines, REFORM achieves over 50% and 27% performance gains on RULER and BABILong respectively at 1M context length. It also outperforms baselines on ∞-Bench, RepoEval, and MM-NIAH, demonstrating flexibility across diverse tasks and domains. Additionally, REFORM reduces inference time by 30% and peak memory usage by 5%, achieving both efficiency and superior performance.

Fact-verification systems are well explored in the NLP literature with growing attention owing to shared tasks like FEVER.Though the task requires reasoning on extracted evidence to verify a claim's factuality, there is little work on understanding the reasoning process.In this work, we propose a new methodology for fact-verification, specifically FEVER, that enforces a closed-world reliance on extracted evidence.We present an extensive evaluation of state-of-the-art verification models under these constraints.

Meaningful comparisons between sets of speech-induced, dynamically evolving articulatory measurements require that the data be temporally aligned in a manner invariant to speech rate discrepancies. The best known approach to this problem is to apply dynamic time warping (DTW) to the corresponding audio signals. While the usefulness of DTW methods is well established in automatic speech recognition, they were never directly and quantitatively validated as a way of aligning and comparing signals from different speakers in a way that is useful for the study of speech as a biological process. This paper provides the first direct quantitative validation of such an audio-based temporal alignment algorithm, itself based on a new divide-and-warp strategy, using speaker invariant temporal landmarks from articulatory data. Results demonstrate that the proposed temporal alignment algorithm accurately brings these landmarks into correspondence between speakers (mean absolute delay of ~ 35ms).