Is reanalysis selective when regressions are consciously controlled?

Abstract

The selective reanalysis hypothesis of Frazier and Rayner (1982) states that readers direct their eyes towards critical words in the sentence when faced with garden-path structures (e.g., Since Jay always jogs a mile seems like a short distance to him). Given the mixed evidence for this proposal in the literature, we investigated the possibility that selective reanalysis is tied to conscious awareness of the garden-path effect. To this end, we adapted the well-known self-paced reading paradigm to allow for regressive as well as progressive key presses. Assuming that regressions in such a paradigm are consciously controlled, we found no evidence for selective reanalysis, but rather for occasional extensive, heterogeneous rereading of garden-path sentences. We discuss the implications of our findings for the selective reanalysis hypothesis, the role of awareness in sentence processing, as well as the usefulness of the bidirectional self-paced reading method for sentence processing research.

Estimating the True Cost of Garden Pathing: A Computational Model of Latent Cognitive Processes

“…The fact that surprisal alone cannot account for the magnitude of garden-path effects suggests that Deep Learning models may need to be augmented with additional mechanisms if one wants to use them as realistic models of human sentence processing. One important factor in this regard may be the involvement of conscious awareness in garden-path processing, which has often been invoked in the classical psycholinguistic literature (e.g., Frazier & Rayner, 1982;Marcus, 1980;Pritchett, 1992): In human readers, strong garden paths like the one caused by the NP/Z ambiguity often lead to a conscious experience of processing difficulty, as well as temporary or permanent processing breakdown, and may require deliberate rereading and reanalysis (e.g., Gibson, 1991;Paape & Vasishth, 2022). Language models lack consciousness, which may be part of the reason why they do not fully capture the human data in this specific domain.…”

“…7 After having been garden‐pathed, the parser can either apply triage or carry out covert reanalysis, which is very costly (579–690 ms). A third alternative, namely making a regression, is also very costly (494–617 ms), which however may be due to the more deliberate nature of regressive rereading in the BSPR paradigm compared to natural reading (Paape & Vasishth, 2022).…”

“…(1999) with 24 participants was 86–218 ms. The NP/Z effect observed by Prasad and Linzen (2019) with 267 participants was 19–37 ms. By comparison, the estimated mean difference for NP/Z sentences in the data of Paape and Vasishth (2022) is 151–209 ms, which may be partly driven by the judgment task.…”

“…The NP/Z effect inGrodner et al (2003) with 35 participants was 53-78 ms, but the NP/Z effect inSturt et al (1999) with 24 participants was 86-218 ms. The NP/Z effect observed byPrasad and Linzen (2019) with 267 participants was 19-37 ms. By comparison, the estimated mean difference for NP/Z sentences in the data ofPaape and Vasishth (2022) is 151-209 ms, which may be partly driven by the judgment task.PAAPE & VASISHTH: THE TRUE COST OF GARDEN-PATHING…”

“…The data of Paape and Vasishth (2021) were collected using a web‐based bidirectional self‐paced reading (BSPR) paradigm. The BSPR paradigm extends regular moving‐window self‐paced reading by giving participants the opportunity to move backward in addition to forward through the sentence in a region‐by‐region fashion (Paape & Vasishth, 2022). The right arrow key reveals the next word or region (and masks the previous one), while the left arrow key reveals the previous word (and masks the next one).…”