Open-domain Dialogue Generation: What We Can Do, Cannot Do, And Should Do Next

Kann, Katharina; Ebrahimi, Abteen; Koh, Joewie J.; Dudy, Shiran; Roncone, Alessandro

doi:10.18653/v1/2022.nlp4convai-1.13

Cited by 14 publications

(12 citation statements)

References 69 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Difficulty of Open-domain dialogue generation. Open domain dialogue dataset in languages besides English and Chinese are difficult to find [7]. Users usually do not have any specific goals, and the content of the dialogue is more casual, which makes the model need to guess the meaning of the user's dialogue.…”

Section: Discussionmentioning

confidence: 99%

“…As the definition of open-domain dialogue generation in [7], the system must output a fluent, engaging, and meaningful natural language response based on the given previous dialogue turns. Different from task-based dialogue generation, open-domain dialogue generation does not need to have a clear goal, and the content of the interlocutor is often very casual hence open-domain dialogue will appear with unseen entities.…”

Section: Open-domain Dialogue Generationmentioning

confidence: 99%

See 1 more Smart Citation

GE-Blender: Graph-Based Knowledge Enhancement for Blender

Lian¹,

Tang²,

Wang³

2023

Preprint

View full text Add to dashboard Cite

Although the great success of open-domain dialogue generation, unseen entities can have a large impact on the dialogue generation task. It leads to performance degradation of the model in the dialog generation. Previous researches used retrieved knowledge of seen entities as the auxiliary data to enhance the representation of the model. Nevertheless, logical explanation of unseen entities remains unexplored, such as possible co-occurrence or semantically similar words of them and their entity category. In this work, we propose an approach to address the challenge above. We construct a graph by extracting entity nodes in them, enhancing the representation of the context of the unseen entity with the entity's 1-hop surrounding nodes. Furthermore, We added the named entity tag prediction task to apply the problem that the unseen entity does not exist in the graph. We conduct our experiments on an open dataset Wizard of Wikipedia and the empirical results indicate that our approach outperforms the state-of-the-art approaches on Wizard of Wikipedia.

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Open-domain Dialogue Generationmentioning

confidence: 99%

GE-Blender: Graph-Based Knowledge Enhancement for Blender

Lian¹,

Tang²,

Wang³

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Using conversational interfaces to complete tasks is the basis of task-oriented dialogue (Chen et al, 2017;Zhang et al, 2020b). End-to-end solutions have shown promising results (Zhang et al, 2020a;Kann et al, 2022), but the use of LLMs remains under scrutiny (Hudeček & Dušek, 2023). For real-world services, Dialog2API (Shu et al, 2022) proposed an interface for interacting with API-based services, whereas META-GUI (Sun et al, 2022) introduced a dataset focused on automating actions in mobile apps rather than general websites.…”

Section: Conversational Interfacesmentioning

confidence: 99%

UAS Navigation in the Real World Using Visual Observation

2022

View full text Add to dashboard Cite

We propose the problem of conversational web navigation, where a digital agent controls a web browser and follows user instructions to solve realworld tasks in a multi-turn dialogue fashion. To support this problem, we introduce WEBLINX -a large-scale benchmark of 100K interactions across 2300 expert demonstrations of conversational web navigation. Our benchmark covers a broad range of patterns on over 150 real-world websites and can be used to train and evaluate agents in diverse scenarios. Due to the magnitude of information present, Large Language Models (LLMs) cannot process entire web pages in real-time. To solve this bottleneck, we design a retrieval-inspired model that efficiently prunes HTML pages by ranking relevant elements. We use the selected elements, along with screenshots and action history, to assess a variety of models for their ability to replicate human behavior when navigating the web. Our experiments span from small text-only to proprietary multimodal LLMs. We find that smaller finetuned decoders surpass the best zero-shot LLMs (including GPT-4V), but also larger finetuned multimodal models which were explicitly pretrained on screenshots. However, all finetuned models struggle to generalize to unseen websites. Our findings highlight the need for large multimodal models that can generalize to novel settings. Our code, data and models are available for research: https://mcgillnlp.github.io/weblinx.

show abstract

“…Generally speaking, dialogue systems aim to generate appropriate responses to user input, and it is an important research direction not only due to its practical applications but also for its direct connection to artificial intelligence (AI) [ 16 , 17 ]. Dialogue systems can be categorized into two classes according to their applications: (1) open-domain dialogue systems that aim at synthesizing human-like conversations with users [ 18 , 19 , 20 ], and (2) task-oriented dialogue (TOD) systems of which the goal is to help human users to complete certain tasks such as virtual assistant [ 21 , 22 , 23 ].…”

Section: Introductionmentioning

confidence: 99%

IvCDS: An End-to-End Driver Simulator for Personal In-Vehicle Conversational Assistant

Yin

Peng

et al. 2022

IJERPH

View full text Add to dashboard Cite

An advanced driver simulator methodology facilitates a well-connected interaction between the environment and drivers. Multiple traffic information environment language processing aims to help drivers accommodate travel demand: safety prewarning, destination navigation, hotel/restaurant reservation, and so on. Task-oriented dialogue systems generally aim to assist human users in achieving these specific goals by a conversation in the form of natural language. The development of current neural network based dialogue systems relies on relevant datasets, such as KVRET. These datasets are generally used for training and evaluating a dialogue agent (e.g., an in-vehicle assistant). Therefore, a simulator for the human user side is necessarily required for assessing an agent system if no real person is involved. We propose a new end-to-end simulator to operate as a human driver that is capable of understanding and responding to assistant utterances. This proposed driver simulator enables one to interact with an in-vehicle assistant like a real person, and the diversity of conversations can be simply controlled by changing the assigned driver profile. Results of our experiment demonstrate that this proposed simulator achieves the best performance on all tasks compared with other models.

show abstract

Open-domain Dialogue Generation: What We Can Do, Cannot Do, And Should Do Next

Cited by 14 publications

References 69 publications

GE-Blender: Graph-Based Knowledge Enhancement for Blender

GE-Blender: Graph-Based Knowledge Enhancement for Blender

UAS Navigation in the Real World Using Visual Observation

IvCDS: An End-to-End Driver Simulator for Personal In-Vehicle Conversational Assistant

Contact Info

Product

Resources

About