We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Show and Write: Entity-aware Article Generation with Image Information

Abstract: Many vision-language applications contain long articles of text paired with images (e.g., news or Wikipedia articles). Prior work learning to encode and/or generate these articles has primarily focused on understanding the article itself and some related metadata like the title or date it was written. However, the images and their captions or alt-text often contain crucial information such as named entities that are difficult to be correctly recognized and predicted by language models. To address this shortcoming, this paper introduces an ENtity-aware article Generation method with Image iNformation, ENGIN, to incorporate an article's image information into language models. ENGIN represents articles that can be conditioned on metadata used by prior work and information such as captions and named entities extracted from images. Our key contribution is a novel Entity-aware mechanism to help our model better recognize and predict the entity names in articles. We perform experiments on three public datasets, GoodNews, VisualNews, and WikiText. Quantitative results show that our approach improves generated article perplexity by 4-5 points over the base models. Qualitative results demonstrate the text generated by ENGIN is more consistent with embedded article images. We also perform article quality annotation experiments on the generated articles to validate that our model produces higher-quality articles. Finally, we investigate the effect ENGIN has on methods that automatically detect machine-generated articles.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2112.05917 [cs.CL]
  (or arXiv:2112.05917v2 [cs.CL] for this version)

Submission history

From: Zhongping Zhang [view email]
[v1] Sat, 11 Dec 2021 05:32:09 GMT (17382kb,D)
[v2] Thu, 24 Mar 2022 04:49:39 GMT (16731kb,D)

Link back to: arXiv, form interface, contact.