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

Download:

Current browse context:

cs.CL

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(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: Can Transformer Models Measure Coherence In Text? Re-Thinking the Shuffle Test

Abstract: The Shuffle Test is the most common task to evaluate whether NLP models can measure coherence in text. Most recent work uses direct supervision on the task; we show that by simply finetuning a RoBERTa model, we can achieve a near perfect accuracy of 97.8%, a state-of-the-art. We argue that this outstanding performance is unlikely to lead to a good model of text coherence, and suggest that the Shuffle Test should be approached in a Zero-Shot setting: models should be evaluated without being trained on the task itself. We evaluate common models in this setting, such as Generative and Bi-directional Transformers, and find that larger architectures achieve high-performance out-of-the-box. Finally, we suggest the k-Block Shuffle Test, a modification of the original by increasing the size of blocks shuffled. Even though human reader performance remains high (around 95% accuracy), model performance drops from 94% to 78% as block size increases, creating a conceptually simple challenge to benchmark NLP models. Code available: this https URL
Comments: Accepted at ACL-IJCNLP 2021 (short paper), 7 pages, 4 figures
Subjects: Computation and Language (cs.CL)
Journal reference: Association for Computational Linguistics (2021)
Cite as: arXiv:2107.03448 [cs.CL]
  (or arXiv:2107.03448v1 [cs.CL] for this version)

Submission history

From: Philippe Laban [view email]
[v1] Wed, 7 Jul 2021 19:19:35 GMT (5537kb,D)

Link back to: arXiv, form interface, contact.