It’s Harder to Break a Relationship When you Commit Long

Manabu Arai,Chie Nakamura

doi:10.1371/journal.pone.0156482

Abstract

Past research has produced evidence that parsing commitments strengthen over the processing of additional linguistic elements that are consistent with the commitments and undoing strong commitments takes more time than undoing weak commitments. It remains unclear, however, whether this so-called digging-in effect is exclusively due to the length of an ambiguous region or at least partly to the extra cost of processing these additional phrases. The current study addressed this issue by testing Japanese relative clause structure, where lexical content and sentence meaning were controlled for. The results showed evidence for a digging-in effect reflecting the strengthened commitment to an incorrect analysis caused by the processing of additional adjuncts. Our study provides strong support for the dynamical, self-organizing models of sentence processing but poses a problem for other models including serial two-stage models as well as frequency-based probabilistic models such as the surprisal theory.

Highlights

In the processing of temporarily ambiguous sentences, comprehenders tend to experience greater processing difficulty at disambiguating input following a long ambiguous region with additional adjuncts than following a short region without such adjuncts
Second-pass reading time is the sum of fixations made in a region after the region has already been exited to the right
The current study examined the effect of ambiguous phrase length in the comprehension of temporarily ambiguous Japanese relative clause sentences

Summary

Introduction

In the processing of temporarily ambiguous sentences, comprehenders tend to experience greater processing difficulty at disambiguating input following a long ambiguous region with additional adjuncts than following a short region without such adjuncts. This is called a length effect or a digging-in effect and has been documented in many studies (e.g., [1,2,3,4]). The idea that comprehenders immediately build a tentative structure based on any available cues in input is consistent with processing models such as lexicalist constraint-based models as well as the more recent surprisal theory [5,6,7]. These models assume word-by-word incremental processing of incoming input; at each word comprehenders estimate and update the PLOS ONE | DOI:10.1371/journal.pone.0156482 June 6, 2016

Methods

Results

Discussion

Conclusion