Hierarchic syntax improves reading time prediction

Abstract

Previous work has debated whether humans make use of hierarchic syntax when processing language (Frank and Bod, 2011; Fossum and Levy, 2012). This paper uses an eye-tracking corpus to demonstrate that hierarchic syntax significantly improves reading time prediction over a strong n-gram baseline. This study shows that an interpolated 5-gram baseline can be made stronger by combining n-gram statistics over entire eye-tracking regions rather than simply using the last n-gram in each region, but basic hierarchic syntactic measures are still able to achieve significant improvements over this improved baseline.

Abstract linguistic structure correlates with temporal activity during naturalistic comprehension

“…This finding corroborates experimental work showing early effects of syntactic structure on on-line processing (Xiang et al, 2009; Sturt & Lombardo, 2005; Yoshida et al, 2012; Kush et al, 2015; Phillips, 2006). Our results also align well with naturalistic eye-tracking studies that demonstrate sensitivity to expectations based on hierarchical structure (Fossum & Levy, 2012; van Schijndel & Schuler, 2015). …”

“…In these studies, word-category expectations based on hierarchical grammars have been found to predict eye-fixation measures (Boston et al, 2008; Demberg & Keller, 2008; Roark et al, 2009; Boston et al, 2011; Fossum & Levy, 2012; van Schijndel & Schuler, 2015). These findings are particularly relevant to the present study, since they indicate that hierarchical structure subserves every-day comprehension.…”

Evidence of syntactic working memory usage in MEG data

“…The cost associated with collecting MEG data may limit the immediate widespread application of the present findings, but since MEG and EEG signals are produced by electrical activity from the same underlying brain sources, this gives hope that anteriorposterior left hemisphere alpha coherence in EEG may be able to provide a similarly clear signal for future studies. The present data support findings like those of van Schijndel and Schuler (2015), who claim hierarchic structure must be used during linguistic processing because hierarchic structure improves the fit to reading times over competitive non-hierarchic models. A potential criticism of that finding is that humans may make use of linear sequences of part-ofspeech tags but not hierarchic structure during linguistic processing (Frank and Bod, 2011).…”

“…To obtain an accurate estimate of center-embedding depth, this study uses the van Schijndel et al (2013) leftcorner PCFG parser trained on the Penn Treebank (Marcus et al, 1993) reannotated into a Nguyen et al (2012) generalized categorial grammar (GCG), 6 which makes PCFG probabilities sensitive to fillergap propagation. This parser achieves a linguistic accuracy comparable to the Petrov and Klein (2007) parser, and the PCFG surprisal estimates it outputs using this grammar provide a state-of-the-art fit to psycholinguistic measures like self-paced reading times and eye-tracking fixation durations (van Schijndel and Schuler, 2015).…”

Modeling fMRI time courses with linguistic structure at various grain sizes

“…Recent work with eye-tracking has wrestled with just this question (Frank and Bod, 2011;Fossum and Levy, 2012;van Schijndel and Schuler, 2015). The argument concerns the strength of the fitted coefficients for different types of grammatical predictors.…”