Published Mar 23, 2020

108 - Data-To-Text Generation, with Verena Rieser and Ondřej Dušek

This episode delves into the intricacies of data-to-text generation, featuring insights from experts Verena Rieser and Ondřej Dušek on overcoming data collection challenges, exploring dialogue system architectures, and the groundbreaking E2E NLG Challenge, highlighting the balance between diversity, accuracy, and system evaluation.

Episode Highlights

Topics covered

Episode Highlights

Related Episodes

61 - Neural Text Generation in Stories, with Elizabeth Clark and Yangfeng Ji
Answers 383 questions
120 - Evaluation of Text Generation, with Asli Celikyilmaz
Answers 383 questions
38 - A Corpus of Natural Language for Visual Reasoning, with Alane Suhr
Answers 383 questions
102 - Biomedical NLP research at the National Institute of Health with Dina Demner-Fushman
Answers 383 questions
114 - Behavioral Testing of NLP Models, with Marco Tulio Ribeiro
Answers 383 questions
44 - Truly Low Resource NLP, with Anders Søgaard
Answers 383 questions
79 - The glass ceiling in NLP, with Natalie Schluter
Answers 383 questions
119 - Social NLP, with Diyi Yang
Answers 383 questions
09 - Learning to Generate Reviews and Discovering Sentiment
Answers 383 questions
13 - Question Answering from Unstructured Text by Retrieval and Comprehension
Answers 383 questions
28 - Data Programming: Creating Large Training Sets, Quickly
Answers 383 questions
93 - NLP/ML for clinical data, with Alistair Johnson
Answers 383 questions
81 - BlackboxNLP, with Afra Alishahi and Tal Linzen
Answers 383 questions
128 - Dynamic Benchmarking, with Douwe Kiela
Answers 383 questions
04 - Recurrent Neural Network Grammars, with Chris Dyer
Answers 383 questions

Dexa/NLP Highlights

108 - Data-To-Text Generation, with Verena Rieser and Ondřej Dušek

Topics covered

Popular Clips

Natural Language Generation Challenge

Translation Evaluation Methods

Evaluating Information Coverage

Human Evaluation Insights

Conversational NLG Systems

Evolution of Data

Dialogue System Architectures

Generating Limited Domain Content

Advancements in NLP

Uncovered Topics

Structured Data Challenges

Diverse Data Creation

Faithfulness in Generation

Data Representation Challenges

Episode Highlights

Data Collection Challenges

System Architectures

E2E NLG Challenge

Related Episodes

61 - Neural Text Generation in Stories, with Elizabeth Clark and Yangfeng Ji

120 - Evaluation of Text Generation, with Asli Celikyilmaz

38 - A Corpus of Natural Language for Visual Reasoning, with Alane Suhr

102 - Biomedical NLP research at the National Institute of Health with Dina Demner-Fushman

114 - Behavioral Testing of NLP Models, with Marco Tulio Ribeiro

44 - Truly Low Resource NLP, with Anders Søgaard

79 - The glass ceiling in NLP, with Natalie Schluter

119 - Social NLP, with Diyi Yang

09 - Learning to Generate Reviews and Discovering Sentiment

13 - Question Answering from Unstructured Text by Retrieval and Comprehension

28 - Data Programming: Creating Large Training Sets, Quickly

93 - NLP/ML for clinical data, with Alistair Johnson

81 - BlackboxNLP, with Afra Alishahi and Tal Linzen

128 - Dynamic Benchmarking, with Douwe Kiela

04 - Recurrent Neural Network Grammars, with Chris Dyer

108 - Data-To-Text Generation, with Verena Rieser and Ondřej Dušek

Topics covered

Popular Clips

Episode Highlights

Data Collection ChallengesThe complexities of data representation and creation in neural networks are explored, highlighting the challenges of using structured data and the innovative techniques employed to ensure diversity and accuracy.

Data Collection Challenges

System ArchitecturesThe discussion shifts to dialogue system architectures, highlighting the differences between modular and end-to-end systems. The conversation also explores the challenges of evaluating these systems, focusing on maintaining faithfulness and accuracy in generated outputs.

System Architectures

E2E NLG ChallengeVerena Rieser and Ondřej Dušek discuss the E2E NLG Challenge, a pioneering effort in neural language generation. They explore its setup, the innovative data collection techniques, and the winning methods that emerged from this task.

E2E NLG Challenge

Related Episodes