A Principled Approach to Feature Selection in Models of Sentence Processing

Abstract

Among theories of human language comprehension, cue‐based memory retrieval has proven to be a useful framework for understanding when and how processing difficulty arises in the resolution of long‐distance dependencies. Most previous work in this area has assumed that very general retrieval cues like [+subject] or [+singular] do the work of identifying (and sometimes misidentifying) a retrieval target in order to establish a dependency between words. However, recent work suggests that general, handpicked retrieval cues like these may not be enough to explain illusions of plausibility (Cunnings & Sturt, 2018), which can arise in sentences like The letter next to the porcelain plate shattered. Capturing such retrieval interference effects requires lexically specific features and retrieval cues, but handpicking the features is hard to do in a principled way and greatly increases modeler degrees of freedom. To remedy this, we use well‐established word embedding methods for creating distributed lexical feature representations that encode information relevant for retrieval using distributed retrieval cue vectors. We show that the similarity between the feature and cue vectors (a measure of plausibility) predicts total reading times in Cunnings and Sturt’s eye‐tracking data. The features can easily be plugged into existing parsing models (including cue‐based retrieval and self‐organized parsing), putting very different models on more equal footing and facilitating future quantitative comparisons.

Interference and filler-gap dependency formation in native and non-native language comprehension.

“…While many studies that have reported facilitatory interference tested sentence grammaticality, our results, along with Cunnings and Sturt (2018), indicate that similar effects arise because of plausibility. Cunnings and Sturt claimed that semantic interference arises because readers utilize the lexical properties of verbs, such as [drinkable] for the verb drink , as retrieval cues (see also Smith & Vasishth, 2020; Van Dyke & McElree, 2006). Our results are compatible with this claim for L1 language comprehension and extend it to L2 comprehension.…”

Features matter: the role of number and gender features during the online processing of subject- and object- relative clauses in Italian

“…Yet, which features can actually modulate processing and why, and whether the same set of features acts as cues cross-linguistically are still debated open questions (Smith & Vasishth, 2020; see also Mertzen et al, 2020). Finally, top-down models such as the one proposed by Chesi (2015) and Chesi and Canal (2019) also assign an important role to morphosyntactic features, since they can either increase or mitigate the cost involved to Biondo, N., Pagliarini, E., Moscati, V., Rizzi, L., & Belletti, A.…”

Parsing as a Cue‐Based Retrieval Model

“…Since we now deal with self-paced reading, in which readers have no look-ahead possibility, it is assumed that no upcoming words are used as context cues (see 18-a). For the wh-recall, only the syntactic category of the wh-dependent is used as a cue to increase spreading activation (more could be added; see Arnett & Wagers, 2017;Kush, 2013;Kush et al, 2015;Patil, Vasishth, & Lewis, 2016;Parker & Phillips, 2017;Smith & Vasishth, 2020 for investigations of what features are relevant for cue-based retrieval).…”