Ziping Ye, Gourab Dey, Christos Christodoulopoulos, Charith Peris, Anil Ramakrishna, Weitong Ruan, Aram Galstyan, Kai-Wei Chang, Rahul Gupta, Ninareh Mehrabi. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recent years have seen significant advances in multi-turn Spoken Language Understanding (SLU), where dialogue contexts are used to guide intent classification and slot filling. However, how to selectively incorporate dialogue contexts, such as previous utterances and dialogue acts, into multi-turn SLU still remains a substantial challenge. In this work, we propose a novel contextual SLU model for multi-turn intent classification and slot filling tasks. We introduce an adaptive global-local context fusion mechanism to selectively integrate dialogue contexts into our model. The local context fusion aligns each dialogue context using multi-head attention, while the global context fusion measures overall context contribution to intent classification and slot filling tasks. Experiments show that on two benchmark datasets, our model achieves absolute F1 score improvements of 2.73% and 2.57% for the slot filling task on Sim-R and Sim M datasets, respectively. Additional experiments on a large-scale, de-identified, in-house dataset further verify the measurable accuracy gains of our proposed model.

In recent years, Smart Home Assistants have expanded into tens of thousands of devices and transformed from a voice only assistant to a much more versatile smart assistant, that uses a connected display to provide a multi-modal customer experience. In order to further improve on the multi-modality experience, comprehension systems need models that can work with multisensory inputs. We focus on the problem of visual grounding, which allows customers to interact with and manipulate items displayed on a screen via voice. We propose a novel learning approach that improves upon a lightweight single stream transformer architecture by adjusting it to better align the visual input features with the referring expressions. Our approach learns to cluster parts of the image along spatial and channel dimensions based on descriptive attributes in the query, and takes advantage of the information in separate clusters more efficiently, as demonstrated by a 1.32% absolute accuracy improvement on a public dataset over the baseline. Given that modern-day Smart Home Assistants have very stringent memory and latency requirements, we restrict our focus to a family of lightweight single stream transformer architectures - our focus is not to beat the ever improving state-of-the-art in visual grounding but to improve upon a lightweight transformer architecture which leads to a model that is easy to train and deploy while having improved semantic awareness.

Current spoken language understanding systems heavily rely on the best hypothesis (ASR 1-best) generated by automatic speech recognition, which is used as the input for downstream models such as natural language understanding (NLU) modules. However, the potential errors and misrecognition in ASR 1-best raise challenges to NLU. It is usually difficult for NLU models to recover from ASR errors without additional signals, which leads to suboptimal SLU performance. This paper proposes a fusion network to jointly consider ASR n-best hypotheses for enhanced robustness to ASR errors. Our experiments on Alexa data show that our model achieved 21.71% error reduction compared to baseline trained on transcription for domain classification.

Virtual adversarial training (VAT) is a powerful technique to improve model robustness in both supervised and semi-supervised settings. It is effective and can be easily adopted on lots of image classification and text classification tasks. However, its benefits to sequence labeling tasks such as named entity recognition (NER) have not been shown as significant, mostly, because the previous approach can not combine VAT with the conditional random field (CRF). CRF can significantly boost accuracy for sequence models by putting constraints on label transitions, which makes it an essential component in most state-of-the-art sequence labeling model architectures. In this paper, we propose SeqVAT, a method which naturally applies VAT to sequence labeling models with CRF. Empirical studies show that SeqVAT not only significantly improves the sequence labeling performance over baselines under supervised settings, but also outperforms state-of-the-art approaches under semi-supervised settings.

In recent years, CubeSats have emerged as a platform of intense interest for a wide range of applications, including remote sensing. Of specific interest in this paper are data processing challenges associated with the MIT's Microwave Atmospheric Satellite (MicroMAS). Due to the motion of MicroMAS and the geometry of the data acquisition process, measurements are not collected on a regular grid of spatial locations as required by most applications. Thus, a fundamental problem in processing these data is that of interpolation. The problem is further complicated by the fact that MicroMAS collects data from several frequencies at a single location. A baseline algorithm that can be used to solve this multi-frequency scattered data interpolation problem is to fit data from each frequency via independent Gaussian Process (GP) and apply standard GP regression to estimate unknown data on the regular grid for each frequency separately. However, this approach ignores the correlation between frequencies. From the covariance structure in the aforementioned Independent Multiple output GP Regression (IMGPR) algorithm, we proposed a Correlated Multiple output GP Regression (CMGPR) algorithm which replaces a set of delta vectors with parameterized weight vectors learned from the dataset. To test the effectiveness of our proposed algorithms, we use NOAA's ATMS temperature data. According to the experimental results, the CMGPR algorithm performs better than the IMGPR.

With the recent development of CubeSats, several ultracompact, low cost, and rapidly deployable satellites have been developed for earth observation missions. Because of the geometry of the acquisition process, measurements are irregularly sampled, whereas in meteorological applications, data are preferred on a regular grid. This problem is further complicated by the fact that, due to CubeSats' compact sizes and constraints, such as limited power, errors occur in geolocation calibration, resulting in positional errors. In this letter, we analyze how the commonly used triangulation-based linear data interpolation scheme behaves under probabilistic models for the positional errors. The derived distribution of interpolation error caused by positional error is intractable even under a Gaussian distribution for positional errors. To address this problem, we developed an analytical closed-form solution to the first two moments of the interpolation error. Using models for positional errors motivated by our prior work, experimental results show that, compared with the first-order linear model, the second-order one provides a better approximation in terms of the mean and variance, which is very close to that is obtained using more computationally intensive Monte Carlo simulations. This model also allows for the closed-form calculation of mean squared interpolation error, which can be of use in the context of system design where the impact of positional errors on remote sensing products must be considered.

With their greatly reduced sizes, low development cost and rapid construction times, CubeSats have emerged as a platform of intense interest for a wide range of applications, including remote sensing. However, due to their compact form factor, performance tradeoffs relative to larger existing platforms have been encountered. Of specific interest in this paper are data processing challenges associated with the Micro-MAS platform. In meteorological applications, the radiometer samples are preferred on a regularly spaced grid for generating subsequent scientific products such as vertical temperature and water vapor profiles, or fusing with other gridded datasets. However, in reality, MicroMAS radiometer samples are not regularly spaced, and are expected to have geolocation errors comparable in magnitude to the beam-width [10]. In this work, we present a joint maximum a posteriori (MAP) estimation approach to determine both sample locations as well as brightness temperature on a regular spatial grid given irregularly sampled data corrupted by noise and uncertainty in sample locations. The performance of this approach is tested on Advanced Technology Microwave Sounder (ATMS) data which demonstrates significant improvement both qualitatively and quantitatively compared with traditional estimation methods.