Conference Papers Year : 2020

Measuring Linguistic Complexity: Introducing a New Categorial Metric

Abstract

This paper provides a computable quantitative measure which accounts for the difficulty in human processing of sentences: why is a sentence harder to parse than another one? Why is some reading of a sentence easier than another one? We take for granted psycholinguistic results on human processing complexity like the ones by Gibson. We define a new metric which uses Categorial Proof Nets to correctly model Gibson's account in his Dependency Locality Theory. The proposed metric correctly predicts some performance phenomena such as structures with embedded pronouns, garden paths, unacceptable center embeddings, preference for lower attachment and passive paraphrases acceptability. Our proposal gets closer to the modern computational psycholinguistic theories, while it opens the door to include semantic complexity, because of the straightforward syntax-semantics interface in categorial grammars.

Fichier principal
Vignette du fichier
LACompLing2018_Chapter.pdf (1.14 Mo) Télécharger le fichier
Origin Files produced by the author(s)
Licence
Loading...

Dates and versions

hal-02146506 , version 1 (04-06-2019)

Licence

Identifiers

Cite

Mehdi Mirzapour, Jean-Philippe Prost, Christian Retoré. Measuring Linguistic Complexity: Introducing a New Categorial Metric. LACompLing 2018 - Symposium on Logic and Algorithms in Computational Linguistics, Aug 2018, Stockholm, Sweden. pp.95-123, ⟨10.1007/978-3-030-30077-7_5⟩. ⟨hal-02146506⟩
260 View
1375 Download

Altmetric

Share

  • More