CoheSentia: A Novel Benchmark of Incremental versus Holistic Assessment of Coherence in Generated Texts

TL;DR

This paper introduces CoheSentia, a new benchmark for assessing the coherence of generated texts from both holistic and incremental perspectives, revealing challenges in current models' performance.

Contribution

It presents a novel annotation protocol and benchmark for human-perceived coherence, highlighting the advantages of incremental assessment and exposing limitations of existing models.

Findings

Incremental coherence scoring has higher inter-annotator agreement.

Standard language models perform poorly on coherence detection.

Incremental assessment helps identify specific incoherence factors.

Abstract

Coherence is a linguistic term that refers to the relations between small textual units (sentences, propositions), which make the text logically consistent and meaningful to the reader. With the advances of generative foundational models in NLP, there is a pressing need to automatically assess the human-perceived coherence of automatically generated texts. Up until now, little work has been done on explicitly assessing the coherence of generated texts and analyzing the factors contributing to (in)coherence. Previous work on the topic used other tasks, e.g., sentence reordering, as proxies of coherence, rather than approaching coherence detection heads on. In this paper, we introduce {\sc CoheSentia}, a novel benchmark of human-perceived coherence of automatically generated texts. Our annotation protocol reflects two perspectives; one is global, assigning a single coherence score, and the other is incremental, scoring sentence by sentence. The incremental method produces an (in)coherence score for each text fragment and also pinpoints reasons for incoherence at that point. Our benchmark contains 500 automatically-generated and human-annotated paragraphs, each annotated in both methods, by multiple raters. Our analysis shows that the inter-annotator agreement in the incremental mode is higher than in the holistic alternative, and our experiments show that standard LMs fine-tuned for coherence detection show varied performance on the different factors contributing to (in)coherence. All in all, these models yield unsatisfactory performance, emphasizing the need for developing more reliable methods for coherence assessment.