Search

In Grammatical Error Correction (GEC), it is crucial to ensure the user's comprehension of a reason for correction. Existing studies present tokens, examples, and hints as to the basis for correction but do not directly explain the reasons for corrections. Although methods that use Large Language Models (LLMs) to provide direct explanations in natural language have been proposed for various tasks, no such method exists for GEC. Generating explanations for GEC corrections involves aligning input and output tokens, identifying correction points, and presenting corresponding explanations consistently. However, it is not straightforward to specify a complex format to generate explanations, because explicit control of generation is difficult with prompts. This study introduces a method called controlled generation with Prompt Insertion (PI) so that LLMs can explain the reasons for corrections in natural language. In PI, LLMs first correct the input text, and then we automatically extract the correction points based on the rules. The extracted correction points are sequentially inserted into the LLM's explanation output as prompts, guiding the LLMs to generate explanations for the correction points. We also create an Explainable GEC (XGEC) dataset of correction reasons by annotating NUCLE, CoNLL2013, and CoNLL2014. Although generations from GPT-3 and ChatGPT using original prompts miss some correction points, the generation control using PI can explicitly guide to describe explanations for all correction points, contributing to improved performance in generating correction reasons., Comment: Work in progress

We study the large $N$ and large representation limits of the Schur line defect correlators of the Wilson line operators transforming in the (anti)symmetric, hook and rectangular representations for $\mathcal{N}=4$ $U(N)$ super Yang-Mills theory. By means of the factorization property, the large $N$ correlators of the Wilson line operators in arbitrary representations can be exactly calculated in principle. In the large representation limit they turn out to be expressible in terms of certain infinite series such as Ramanujan's general theta functions and the $q$-analogues of multiple zeta values ($q$-MZVs). Several generating functions for combinatorial objects, including partitions with non-negative cranks and conjugacy classes of general linear groups over finite fields, emerge from the large $N$ correlators. Also we find conjectured properties of the automorphy and the hook-length expansion satisfied by the large $N$ correlators., Comment: 68 pages, 1 figure

Discriminatory gender biases have been found in Pre-trained Language Models (PLMs) for multiple languages. In Natural Language Inference (NLI), existing bias evaluation methods have focused on the prediction results of one specific label out of three labels, such as neutral. However, such evaluation methods can be inaccurate since unique biased inferences are associated with unique prediction labels. Addressing this limitation, we propose a bias evaluation method for PLMs, called NLI-CoAL, which considers all the three labels of NLI task. First, we create three evaluation data groups that represent different types of biases. Then, we define a bias measure based on the corresponding label output of each data group. In the experiments, we introduce a meta-evaluation technique for NLI bias measures and use it to confirm that our bias measure can distinguish biased, incorrect inferences from non-biased incorrect inferences better than the baseline, resulting in a more accurate bias evaluation. We create the datasets in English, Japanese, and Chinese, and successfully validate the compatibility of our bias measure across multiple languages. Lastly, we observe the bias tendencies in PLMs of different languages. To our knowledge, we are the first to construct evaluation datasets and measure PLMs' bias from NLI in Japanese and Chinese., Comment: LREC-COLING 2024

Pre-trained language models trained on large-scale data have learned serious levels of social biases. Consequently, various methods have been proposed to debias pre-trained models. Debiasing methods need to mitigate only discriminatory bias information from the pre-trained models, while retaining information that is useful for the downstream tasks. In previous research, whether useful information is retained has been confirmed by the performance of downstream tasks in debiased pre-trained models. On the other hand, it is not clear whether these benchmarks consist of data pertaining to social biases and are appropriate for investigating the impact of debiasing. For example in gender-related social biases, data containing female words (e.g. ``she, female, woman''), male words (e.g. ``he, male, man''), and stereotypical words (e.g. ``nurse, doctor, professor'') are considered to be the most affected by debiasing. If there is not much data containing these words in a benchmark dataset for a target task, there is the possibility of erroneously evaluating the effects of debiasing. In this study, we compare the impact of debiasing on performance across multiple downstream tasks using a wide-range of benchmark datasets that containing female, male, and stereotypical words. Experiments show that the effects of debiasing are consistently \emph{underestimated} across all tasks. Moreover, the effects of debiasing could be reliably evaluated by separately considering instances containing female, male, and stereotypical words than all of the instances in a benchmark dataset., Comment: IJCNLP-AACL 2023

Charged carriers in solids diffuse from hot to cold sides under temperature gradient to induce the thermoelectric voltage. Carrier filtering effect, which only passes either electrons or holes for the conduction process, is an efficient method to enhance such voltage, although it is challenging to experimentally realize it especially in conventional metals with weak energy dependence of the density of states near the Fermi level. Here we measure the in-plane and out-of-plane thermopower of the layered perovskite Sr$_2$RuO$_4$ single crystals above room temperature, and find that the out-of-plane thermopower is largely enhanced with increasing temperature, while the in-plane one seems to remain a temperature-independent constant value which is expected from the Heikes formula. The observed large out-of-plane thermopower may originate from the recently proposed intriguing hole filtering effect in the body-centered tetragonal system, in which the carrier hopping through the centered atom is essential. Thus, the present carrier filtering effect may be a universal property to be applicable in various materials belonging to such crystal system., Comment: 7 pages, 5 figures. Accepted for publication in Phys. Rev. Materials

We find boundary confining dualities of 3d $\mathcal{N}=2$ supersymmetric gauge theories with exceptional gauge groups. The half-indices which enumerate the boundary BPS local operators in the presence of Neumann and Dirichlet boundary conditions for gauge fields are identified with the Askey-Wilson type $q$-beta integrals and Macdonald type sums respectively. New conjectural identities of $E_6$ and $E_7$ integrals and sums are derived from the boundary confining dualities., Comment: 38 pages

We give a generalization of weighted Weil heights. These heights generalize both Weil's heights and Dobrowolski's height. We study Northcott numbers for our heights. Our results generalize the authors' former work on Vidaux and Videla's question about the Northcott number. As an application, we evaluate Northcott numbers for Talamanca's spectral height on matrices., Comment: 28 pages

Transparent conducting oxides (TCOs) are essential components of optoelectronic devices and various materials have been explored for highly efficient TCOs having a combination of high transmittance and low sheet resistance. Here, we focus on a misfit thermoelectric oxide [Ca$_2$CoO$_3$]$_{0.62}$[CoO$_2$] and fabricate the transparent low-layered crystals by a mechanical tape-peeling method using the single-crystalline samples. From the transmittance measurement, we find that the thickness of low-layered samples is several orders of hundred nanometers, which is comparable with the estimation from the scanning electron microscopy images. Compared to the previous results on the polycrystalline and $c$-axis oriented transparent films, the electrical resistivity is reduced owing to the single-crystalline nature. The figure of merit for the transparent conducting materials in the present low-layered samples is then evaluated to be higher than the values in the previous reports. The present results on the low-layered single-crystalline [Ca$_2$CoO$_3$]$_{0.62}$[CoO$_2$] may offer a unique class of multi-functional transparent thermoelectric oxides., Comment: 6 pages, 4 figures, 1 table. This article may be downloaded for personal use only. Any other use requires prior permission of the author and AIP Publishing

We investigate the electronic structure of $2H$-$\mathrm{Nb}\mathrm{S}_2$ and $h$-$\mathrm{BN}$ by angle-resolved photoemission spectroscopy (ARPES) and photoemission intensity calculations. Although in bulk form, these materials are expected to exhibit band degeneracy in the $k_z=\pi/c$ plane due to screw rotation and time-reversal symmetries, we observe gapped band dispersion near the surface. We extract from first-principles calculations the near-surface electronic structure probed by ARPES and find that the calculated photoemission spectra from the near-surface region reproduce the gapped ARPES spectra. Our results show that the near-surface electronic structure can be qualitatively different from the bulk one due to partially broken nonsymmorphic symmetries., Comment: 6+13 pages, 4+15 figures

Large Language Models (LLMs) have achieved human-level fluency in text generation, making it difficult to distinguish between human-written and LLM-generated texts. This poses a growing risk of misuse of LLMs and demands the development of detectors to identify LLM-generated texts. However, existing detectors lack robustness against attacks: they degrade detection accuracy by simply paraphrasing LLM-generated texts. Furthermore, a malicious user might attempt to deliberately evade the detectors based on detection results, but this has not been assumed in previous studies. In this paper, we propose OUTFOX, a framework that improves the robustness of LLM-generated-text detectors by allowing both the detector and the attacker to consider each other's output. In this framework, the attacker uses the detector's prediction labels as examples for in-context learning and adversarially generates essays that are harder to detect, while the detector uses the adversarially generated essays as examples for in-context learning to learn to detect essays from a strong attacker. Experiments in the domain of student essays show that the proposed detector improves the detection performance on the attacker-generated texts by up to +41.3 points F1-score. Furthermore, the proposed detector shows a state-of-the-art detection performance: up to 96.9 points F1-score, beating existing detectors on non-attacked texts. Finally, the proposed attacker drastically degrades the performance of detectors by up to -57.0 points F1-score, massively outperforming the baseline paraphrasing method for evading detection., Comment: AAAI 2024 camera ready. Code and dataset available at https://github.com/ryuryukke/OUTFOX

We study the asymptotic growth of the degeneracy of the BPS local operators with scaling dimension $n/2$ in the three-dimensional superconformal field theories describing $N$ M2-branes. From the large $N$ supersymmetric indices we obtain the asymptotic formulas for degeneracies of the M2-brane SCFTs according to the Meinardus theorem. We observe an intriguing universal $n^{2/3}$ growth of the degeneracies in various theories of M2-brane SCFTs. We also determine the coefficients of $n^{2/3}$ growth as well as further corrections in these theories explicitly., Comment: 51 pages, v2: published version in CIMP