Confirming the generalizability of a chain-based animacy detector Conference

Jahan, L, Victor Yarlott, WH, Mittal, R et al. (2020). Confirming the generalizability of a chain-based animacy detector . 2794 43-46.

cited authors

  • Jahan, L; Victor Yarlott, WH; Mittal, R; Finlayson, MA



  • Animacy is the characteristic of a referent being able to independently carry out actions in a story world (e.g., movement, communication). It is a necessary property of characters in stories, and so detecting animacy is an important step in automatic story understanding; it is also potentially useful for many other natural language processing tasks such as word sense disambiguation, coreference resolution, character identification, and semantic role labeling. Recent work by Jahan et al. [2018] demonstrated a new approach to detecting animacy where animacy is considered a direct property of coreference chains (and referring expressions) rather than words. In Jahan et al., they combined hand-built rules and machine learning (ML) to identify the animacy of referring expressions and used majority voting to assign the animacy of coreference chains, and reported high performance of up to 0.90 F1. In this short report we verify that the approach generalizes to two different corpora (OntoNotes and the Corpus of English Novels) and we confirmed that the hybrid model performs best, with the rule-based model in second place. Our tests apply the animacy classifier to almost twice as much data as Jahan et al.'s initial study. Our results also strongly suggest, as would be expected, the dependence of the models on coreference chain quality. We release our data and code to enable reproducibility.

publication date

  • January 1, 2020

start page

  • 43

end page

  • 46


  • 2794