Abstract This study introduces the educational assistant robots that we developed for foreign language learning and explores effectiveness of robot-assisted (RALL) which is in its early stages. To achieve this purpose, a course was designed students have meaningful interactions with intelligent an immersive environment. A total 24 elementary students, ranging age from ten to twelve, were enrolled English lessons. pre-test/post-test design used investigate cognitive effects RALL approach on...

The encoder-decoder dialog model is one of the most prominent methods used to build systems in complex domains. Yet it limited because cannot output interpretable actions as traditional systems, which hinders humans from understanding its generation process. We present an unsupervised discrete sentence representation learning method that can integrate with any existing models for response generation. Building upon variational autoencoders (VAEs), we two novel models, DI-VAE and DI-VST...

Generative encoder-decoder models offer great promise in developing domain-general dialog systems. However, they have mainly been applied to open-domain conversations. This paper presents a practical and novel framework for building task-oriented systems based on models. enables accomplish slot-value independent decision-making interact with external databases. Moreover, this shows the flexibility of proposed method by interleaving chatting capability slot-filling system better out-of-domain...

We introduce SPARTA, a novel neural retrieval method that shows great promise in performance, generalization, and interpretability for open-domain question answering. Unlike many ranking methods use dense vector nearest neighbor search, SPARTA learns sparse representation can be efficiently implemented as an Inverted Index. The resulting enables scalable does not require expensive approximate search leads to better performance than its counterpart. validated our approaches on 4 answering...

Tiancheng Zhao, Tianqi Zhang, Mingwei Zhu, Haozhan Shen, Kyusong Lee, Xiaopeng Lu, Jianwei Yin. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing: System Demonstrations. 2022.

Vision-Language Pretraining (VLP) models have recently successfully facilitated many cross-modal downstream tasks. Most existing works evaluated their systems by comparing the fine-tuned task performance. However, only average accuracy provides little information about pros and cons of each VLP method, let alone insights on how community can improve in future. Inspired CheckList for testing natural language processing, we exploit VL-CheckList, a novel framework to understand capabilities...

Xiaopeng Lu, Tiancheng Zhao, Kyusong Lee. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Object detection (OD) in computer vision has made significant progress recent years, transitioning from closed-set labels to open-vocabulary (OVD) based on large-scale vision-language pre-training (VLP). However, current evaluation methods and datasets are limited testing generalization over object types referral expressions, which do not provide a systematic, fine-grained, accurate benchmark of OVD models' abilities. In this paper, we propose new named OVDEval, includes 9 sub-tasks...

This study examines the dialog-based language learning game (DB-LLG) realized in a 3D environment built with contents. We designed DB-LLG to communicate users who can conduct interactive conversations characters various immersive environments. From pilot test, we found that several technologies were identified as essential construction of such dialog management, hint generation, and grammar error detection feedback. describe technical details our system POSTECH English (Pomy). evaluated...

The encoder-decoder dialog model is one of the most prominent methods used to build systems in complex domains. Yet it limited because cannot output interpretable actions as traditional systems, which hinders humans from understanding its generation process. We present an unsupervised discrete sentence representation learning method that can integrate with any existing models for response generation. Building upon variational autoencoders (VAEs), we two novel models, DI-VAE and DI-VST...

When creating a dialog system, developers need to test each version ensure that it is performing correctly. Recently the trend has been on large datasets or ask many users try out system. Crowdsourcing solved issue of finding users, but presents new challenges such as how use crowdsourcing platform and what type appropriate. DialCrowd designed make system assessment easier quality result. This paper describes DialCrowd, specific needs fulfills works. It then relates by group developer.

This corpus-based study presents how English relative clauses are used in science and engineering journal papers. Relative ensure semantic clarity textual variety but they cause difficulty to non-native speakers of due their diverse uses functions. With pedagogical purposes mind, this research investigates frequently what context employed three representative journals, namely CELL, Journal American Chemical Society, IEEE Solid-State Circuits. In addition, papers for Specific Purposes...

Abstract The advancement of object detection (OD) in open‐vocabulary and open‐world scenarios is a critical challenge computer vision. OmDet, novel language‐aware architecture an innovative training mechanism that harnesses continual learning multi‐dataset vision‐language pre‐training introduced. Leveraging natural language as universal knowledge representation, OmDet accumulates “visual vocabularies” from diverse datasets, unifying the task language‐conditioned framework. multimodal network...

End-to-end transformer-based detectors (DETRs) have shown exceptional performance in both closed-set and open-vocabulary object detection (OVD) tasks through the integration of language modalities. However, their demanding computational requirements hindered practical application real-time (OD) scenarios. In this paper, we scrutinize limitations two leading models OVDEval benchmark, OmDet Grounding-DINO, introduce OmDet-Turbo. This novel OVD model features an innovative Efficient Fusion Head...

This paper describes a new spoken dialog portal that connects systems produced by the academic research community and gives them access to real users. We introduce distributed, multi-modal, multi-agent prototype framework affords easy integration with various remote resources, ranging from end-to-end external knowledge APIs. The provides seamless passage one system another. To date, DialPort has successfully connected multi-domain at Cambridge University, NOAA (National Oceanic Atmospheric...

The advancement of object detection (OD) in open-vocabulary and open-world scenarios is a critical challenge computer vision. This work introduces OmDet, novel language-aware architecture, an innovative training mechanism that harnesses continual learning multi-dataset vision-language pre-training. Leveraging natural language as universal knowledge representation, OmDet accumulates "visual vocabulary" from diverse datasets, unifying the task language-conditioned framework. Our multimodal...

Generative encoder-decoder models offer great promise in developing domain-general dialog systems. However, they have mainly been applied to open-domain conversations. This paper presents a practical and novel framework for building task-oriented systems based on models. enables accomplish slot-value independent decision-making interact with external databases. Moreover, this shows the flexibility of proposed method by interleaving chatting capability slot-filling system better out-of-domain...

We introduce Talk to Papers, which exploits the recent open-domain question answering (QA) techniques improve current experience of academic search. It's designed enable researchers use natural language queries find precise answers and extract insights from a massive amount papers. present large improvement over classic search engine baseline on several standard QA datasets provide community collaborative data collection tool curate first processing research dataset via effort.

When implementing a conversational educational teaching agent, user-intent understanding and dialog management in system are not sufficient to give users information.In this paper, we propose agent that gives some information or triggers interests on contents.The proposed only converses with user but also answer questions the asked asks by integrating knowledge base.We used Wikipedia corpus learn weights between two entities embedding of properties calculate similarities for selection answers.

This paper describes the POSTECH grammatical error correction system. Various methods are proposed to correct errors such as rule-based, probability n-gram vector approaches and router-based approach. Google N-gram count corpus is used mainly resource. Correction candidates extracted from NUCLE training data each candidate evaluated with development extract high precision rules frames. Out of 13 participating teams, our system ranked 4 th on both original revised annotation.

The demand for computer-assisted language learning systems that can provide corrective feedback on learners’ speaking has increased. However, it is not a trivial task to detect grammatical errors in oral conversations because of the unavoidable automatic speech recognition systems. To feedback, novel method performance proposed. proposed consists two sub-models: grammaticality-checking model and error-type classification model. We automatically generate learners are likely commit construct...