The core model of sentence processing used in the book is introduced and its empirical coverage relative to the existing reading time data is considered. Here, we also discuss the Approximate Bayesian Computation method for parameter estimation for model evaluation.

Reviews the different kinds of dependencies that have been investigated in sentence processing: subject–verb dependencies, reflexives and reciprocals, etc. This chapter also synthesizes the available empirical evidence by carrying out meta-analyses that provide estimates of the effect of interest in each dependency type.

This chapter discusses two extensions of the model presented in the previous chapter: the effect of prominence (through discourse prominence, etc.) and the effect of so-called multi-associative cues. The empirical coverage of the extended model is evaluated against benchmark data.

This chapter discusses three central phenomena of interest in sentence processing: reanalysis, underspecification, and capacity-based differences in sentence comprehension. The model's quantitative predictions are evaluated against two benchmark data-sets that investigate reanalysis and underspecification.

This final chapter discusses future directions that need to be pursued: we need common benchmark data-sets for model evaluation; larger-sample, properly powered studies that deliver accurate estimates of effects; and comprehensive model comparisons using a common benchmark data-set. A further gap in the literature is the need to understand the production–comprehension link; this link could shed further light on many aspects of sentence comprehension, but there are also several puzzles relating to the production–comprehension link (like the long-before-short preference in Japanese) that need to be investigated further.

This chapter presents another extension of the core model: an eye-movement control system is integrated with the parsing architecture, and this extended model is investigated using benchmark eyetracking data (the Potsdam Sentence Corpus).

This chapter investigates whether sentence comprehension difficulty in aphasia can be explained in terms of retrieval processes. By modelling individuals with aphasia (IWAs) separately, we show that different IWAs show impairments along different dimensions: slowed processing, intermittent deficiency, and resource reduction. The parameters in the cue-based retrieval model have a theoretical interpretation that allows these three theories to be implemented within the architecture. In a further investigation, we compare the relative predictive accuracy of the cue-based model with that of the direct-access model. The benchmark data here are from Caplan et al. (2015); k-fold cross-validation is used as in the preceding chapter. The cue-based retrieval model is shown to have a better predictive performance.

Reviews the role of working memory in theories of sentence comprehension, and reviews current theoretical positions in sentence processing. The chapter also identifies several gaps in current research: the relative scarcity of computational models, an excessive focus on average behavior, the absence of properly powered studies, and unclear criteria for identifying model fit. The chapter also summarizes the goals of the book: to provide open source code for facilitating reproducible analyses, to go beyond modelling average effects, and to provide a principled workflow for model evaluation and comparison.

This chapter presents a model comparison between two competing models of retrieval processes: the cue-based retrieval model presented in this book and the direct-access model. The two models are implemented in a Bayesian framework, and then model comparison is carried out using k-fold cross-validation. The benchmark data used for evaluation are from a previously published large-sample, self-paced reading study (181 participants). The results show that the direct-access model has a better performance on this benchmark data than the cue-based retrieval model.