We study knowledge-grounded dialogue generation with pre-trained language models. To leverage the redundant external knowledge under capacity constraint, we propose equipping response defined by a model selection module, and an unsupervised approach to jointly optimizing unlabeled dialogues. Empirical results on two benchmarks indicate that our can significantly outperform state-of-the-art methods in both automatic evaluation human judgment.

Responding with knowledge has been recognized as an important capability for intelligent conversational agent. Yet knowledge-grounded dialogues, training data learning such a response generation model, are difficult to obtain. Motivated by the challenge in practice, we consider dialogue under natural assumption that only limited examples available. In low-resource setting, devise disentangled decoder order isolate parameters depend on dialogues from entire model. By this means, major part of...

While neural conversation models have shown great potentials towards generating informative and engaging responses via introducing external knowledge, learning such a model often requires knowledge-grounded dialogues that are difficult to obtain. To overcome the data challenge reduce cost of building dialogue system, we explore problem under zero-resource setting by assuming no context-knowledge-response triples needed for training. this end, propose representing knowledge bridges context...

Building an intelligent dialogue system with the ability to select a proper response according multi-turn context is great challenging task. Existing studies focus on building context-response matching model various neural architectures or pretrained language models (PLMs) and typically learning single prediction These approaches overlook many potential training signals contained in data, which might be beneficial for understanding produce better features prediction. Besides, retrieved from...

Recent years have witnessed a surge of interest in the field open-domain dialogue. Thanks to rapid development social media, large dialogue corpus from Internet builds up fundamental premise for data-driven model. The breakthrough neural network also brings new ideas researchers AI and NLP. A great number techniques methods therefore came into being. In this paper, we review some most representative works recent divide existing prevailing frameworks model three categories. We further analyze...

We present a document-grounded matching network (DGMN) for response selection that can power knowledge-aware retrieval-based chatbot system. The challenges of building such model lie in how to ground conversation contexts with background documents and recognize important information the matching. To overcome challenges, DGMN fuses document context into representations each other, dynamically determines if grounding is necessary importance different parts through hierarchical interaction at...

Recently, knowledge-grounded dialogue systems have gained increasing attention. Great efforts been made to build response matching models where all content and knowledge sentences are leveraged. However, redundancy distraction of irrelevant often exist in conversations, which may affect the process lead inferior performance. In addition, history excessive also hinder exploitation popular pre-trained language (PLMs) due limitation input length. To address these challenges, we propose a new...

With the ubiquity of mobile communication devices, people experiencing traffic jams share real-time information and interact with each other on social media sites, which provide new channels to monitor, estimate manage flows. In this paper, we use natural language processing data mining technologies extract jam related from Tianya.cn, analyze content people's talk discover "talking point" when facing jams, support for relevant authorities make successful effective decisions response management.

Building an intelligent dialogue system with the ability to select a proper response according multi-turn context is great challenging task. Existing studies focus on building context-response matching model various neural architectures or PLMs and typically learning single prediction These approaches overlook many potential training signals contained in data, which might be beneficial for understanding produce better features prediction. Besides, retrieved from existing systems supervised...

In the medical field, labeling of surgical video data requires Expert knowledge, collecting enough numbers marked is difficult and time-consuming. The insufficient (labeled data) leads to low generalization ability training model accuracy recognition. It has been recently shown that Generative Adversarial Networks (GANs) can produce synthetic images exceptional visual fidelity. this work, authors propose GAN-based method for automatic Surgical Workflow images. theory methodology paper are...

Video-grounded dialogue understanding is a challenging problem that requires machine to perceive, parse and reason over situated semantics extracted from weakly aligned video dialogues. Most existing benchmarks treat both modalities the same as frame-independent visual task, while neglecting intrinsic attributes in multimodal dialogues, such scene topic transitions. In this paper, we present Scene&Topic AwaRe (VSTAR) dataset, large scale video-grounded dataset based on 395 TV series. Based...

We present a document-grounded matching network (DGMN) for response selection that can power knowledge-aware retrieval-based chatbot system. The challenges of building such model lie in how to ground conversation contexts with background documents and recognize important information the matching. To overcome challenges, DGMN fuses document context into representations each other, dynamically determines if grounding is necessary importance different parts through hierarchical interaction at...

On July 10-12th, 2016, IEEE International Conference on Service Operations and Logistics, Informatics (SOLI), Vehicular Electronics Safety (ICVES), Forum Integrated Sustainable Transportation Systems (FISTS) were held together in Beijing, China. These three conferences organized by the State Key Laboratory for Management Control of Complex (SKL-MCCS), Institute Automation, Chinese Academy Sciences, China Association Automation. With continuous efforts improvement, SOLI, ICVES, FISTS have...

Video-grounded dialogue understanding is a challenging problem that requires machine to perceive, parse and reason over situated semantics extracted from weakly aligned video dialogues. Most existing benchmarks treat both modalities the same as frame-independent visual task, while neglecting intrinsic attributes in multimodal dialogues, such scene topic transitions. In this paper, we present Scene&Topic AwaRe (VSTAR) dataset, large scale video-grounded dataset based on 395 TV series. Based...

Logical data-to-text generation is a representative task in measuring the capabilities of both language and complex reasoning. Despite introduction reasoning skills generation, existing works still rely on neural models to output final table description. However, due inefficacy reasoning, these methods inevitably have difficulty working out key entities description might produce unfaithful descriptions. To alleviate issues, we propose dependency-aware symbolic framework that reasons each...