Abstract

The WebNLG challenge consists in mapping sets of RDF triples to text. It provides a common benchmark on which to train, evaluate and compare “microplanners”, i.e. generation systems that verbalise a given content by making a range of complex interacting choices including referring expression generation, aggregation, lexicalisation, surface realisation and sentence segmentation. In this paper, we introduce the microplanning task, describe data preparation, introduce our evaluation methodology, analyse participant results and provide a brief description of the participating systems.

Highlights

  • Previous Natural Language Generation (NLG) challenges have focused on surface realisation (Banik et al, 2013; Belz et al, 2011), referring expression generation (Belz and Gatt, 2007; Gatt et al, 2008; Gatt et al, 2009; Belz et al, 2008; Belz et al, 2009; Belz et al, 2010) and content selection (BouayadAgha et al, 2013)

  • The WebNLG challenge focuses on microplanning, that subtask of NLG which consists in mapping a given content to a text verbalising this content

  • Despite a tight schedule, it generated a high level of interest among the NLG community: 62 groups from 18 countries6 downloaded the data, 6 groups submitted 8 systems and 3 groups developped a system but did not submit

Read more

Summary

Introduction

Previous Natural Language Generation (NLG) challenges have focused on surface realisation (Banik et al, 2013; Belz et al, 2011), referring expression generation (Belz and Gatt, 2007; Gatt et al, 2008; Gatt et al, 2009; Belz et al, 2008; Belz et al, 2009; Belz et al, 2010) and content selection (BouayadAgha et al, 2013). Given the WebNLG data unit shown in (1a), generating the text in (1b) involves choosing to lexicalise the JOHN E BLAHA entity only once (referring expression generation), lexicalising the OCCUPATION property as the phrase worked as (lexicalisation), using PP coordination to avoid repeating the word born (aggregation) and verbalising the three triples by a single complex sentence including an apposition, a PP coordination and a transitive verb construction (sentence segmentation and surface realisation). While triples sharing a subject (SIBLING configuration) are likely to induce a VP or a sentence coordination, a CHAIN configuration (where the object of one triple is the subject of the other) will more naturally give rise to object relative clauses or participials Another factor impacting syntactic variation is the set of properties (input patterns) cooccuring in a given input.

Participating Systems
Evaluation Methodology
Automatic Evaluation
Baseline System
Results
4–5 TILB-NMT
Conclusion

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call

Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.