Search

The conversational machine reading comprehension (CMRC) task aims to answer questions in conversations, which has been a hot research topic in recent years because of its wide applications. However, existing CMRC benchmarks in which each conversation is assigned a static passage are inconsistent with real scenarios. Thus, model's comprehension ability towards real scenarios are hard to evaluate reasonably. To this end, we propose the first Chinese CMRC benchmark Orca and further provide zero-shot/few-shot settings to evaluate model's generalization ability towards diverse domains. We collect 831 hot-topic driven conversations with 4,742 turns in total. Each turn of a conversation is assigned with a response-related passage, aiming to evaluate model's comprehension ability more reasonably. The topics of conversations are collected from social media platform and cover 33 domains, trying to be consistent with real scenarios. Importantly, answers in Orca are all well-annotated natural responses rather than the specific spans or short phrase in previous datasets. Besides, we implement three strong baselines to tackle the challenge in Orca. The results indicate the great challenge of our CMRC benchmark. Our datatset and checkpoints are available at https://github.com/nuochenpku/Orca., 14 pages

Nowadays, foundation models become one of fundamental infrastructures in artificial intelligence, paving ways to the general intelligence. However, the reality presents two urgent challenges: existing foundation models are dominated by the English-language community; users are often given limited resources and thus cannot always use foundation models. To support the development of the Chinese-language community, we introduce an open-source project, called Fengshenbang, which leads by the research center for Cognitive Computing and Natural Language (CCNL). Our project has comprehensive capabilities, including large pre-trained models, user-friendly APIs, benchmarks, datasets, and others. We wrap all these in three sub-projects: the Fengshenbang Model, the Fengshen Framework, and the Fengshen Benchmark. An open-source roadmap, Fengshenbang, aims to re-evaluate the open-source community of Chinese pre-trained large-scale models, prompting the development of the entire Chinese large-scale model community. We also want to build a user-centered open-source ecosystem to allow individuals to access the desired models to match their computing resources. Furthermore, we invite companies, colleges, and research institutions to collaborate with us to build the large-scale open-source model-based ecosystem. We hope that this project will be the foundation of Chinese cognitive intelligence., Added the Chinese version and is now a bilingual paper

A novel self-modulated superjunction lateral double-diffused MOSFET (SM-SJ LDMOS) is proposed and experimentally realized in this letter. When a SJ is introduced into a lightly doped N-drift region, the 3-D depletion regions appear surrounding the P-pillars. This 3-D junction field effect transistor (JFET) effect decreases the current paths in both N-pillars and N-drift, causing a largely increased specific on-resistance ${R} _{\text {on,sp}}$ especially for the SJ with a narrow cell pitch. To suppress the inherent 3-D JFET effect, the SM-SJ structure is proposed by replacing the P-pillars with a series of P-islands. These P-islands with linearly reduced lengths modulate the surface electric field to increase the breakdown voltage ${V} _{{\text {B}}}$ while the N-drift areas between P-islands weaken the 3-D JFET effect to reduce ${R} _{\text {on,sp}}$ . Compared with the conventional SJ device, the measured ${V} _{\text {B}}$ of the SM-SJ LDMOS is increased by 53.1% from 260 V to 398 V with ${R} _{on,sp}$ reduced by 22.3% from 35.4 $\text{m}\Omega ~\cdot $ cm2 to 27.5 $\text{m}\Omega ~\cdot $ cm2.

A novel lateral double-diffused metal–oxide semiconductor (LDMOS) with three-dimensional floating vertical field plate (3-D F-VFP) is proposed in this paper. The 3-D F-VFP LDMOS features a discrete pillar-type VFP array through the N-type drift region and the VFP pillars with the same distance from the source are connected to each other to realize the equal potential. In the off-state, the potential along the drift region are pinned by the series of equal-potential F-VFP rings and a new full-region depletion mode is introduced into the bulk of the device. Therefore, the highly doped drift region is depleted by the 3-D F-VFP, establishing a self-adaptive charge balance inside the drift region. In the on-state, the current flows through the gaps among all the F-VFP pillars, preventing the long current path and reducing the on resistance $\boldsymbol{R}_{\mathbf{on}}$ . A 3-D F-VFP LDMOS is experimentally implemented, which obtains a breakdown voltage $\boldsymbol{V}_{\mathbf{B}}$ of 630 V and a $\boldsymbol{R}_{\mathbf{on}}$ of $344.8\ \mathbf{\Omega}$ compared with 550 V and $722.5\ \mathbf{\Omega}$ of the device without the F-VFP. The measured saturation current of the 3-D F-VFP LDMOS is more than 4.5 times of that of the device without F-VFP.

In this paper, we introduce an approach to combining word embeddings and machine translation for multilingual semantic word similarity, the task2 of SemEval-2017. Thanks to the unsupervised transliteration model, our cross-lingual word embeddings encounter decreased sums of OOVs. Our results are produced using only monolingual Wikipedia corpora and a limited amount of sentence-aligned data. Although relatively little resources are utilized, our system ranked 3rd in the monolingual subtask and can be the 6th in the cross-lingual subtask.

Congeneric minority languages like Uyghur and Kazakh, are widely used in Middle East and Xinjiang province of China. They share the same characters even words, and can't be discriminated by encoding, which makes minority language processing difficult. Spoken texts such as short messages and online chatting messages are short, sparse, noisy and ungrammatical, regarded as the bottle neck of similar language identification. Here we propose a practical method to automatically distinguish Uyghur from Kazakh on short spoken texts. External online resource is used for data acquisition. Combining few knowledge with heuristic features, this method can work effectively without professional and elaborate rules. Utilizing the maximum entropy classifier, our novel system achieves performance of 98.7% recall and 94.6% precision on Uyghur, while reaching 96.5% precision on Kazakh, with each text contains less than 14 words. Considering that our system is the first to identify similar languages on short spoken texts, it can serve as a guidance to deal with analogous problems.