Jeffrey Witzel

Deeper than shallow: Evidence for structure-based parsing biases in second-language sentence processing

This study examines the reading patterns of native speakers (NSs) and high-level (Chinese) nonnative speakers (NNSs) on three English sentence types involving temporarily ambiguous structural configurations. The reading patterns on each sentence type indicate that both NSs and NNSs were biased toward specific structural interpretations. These results are interpreted as evidence that both first-language and second-language (L2) sentence comprehension is guided (at least in part) by structure-based parsing strategies and, thus as counterevidence to the claim that NNSs are largely limited to rudimentary (or “shallow”) syntactic computation during online L2 sentence processing.

Comparisons of online reading paradigms: eye tracking, moving-window, and maze.

This study compares four methodologies used to examine online sentence processing during reading. Specifically, self-paced, non-cumulative, moving-window reading (Just et al. in J Exp Psychol Gen 111:228–238, 1982), eye tracking (see e.g., Rayner in Q J Exp Psychol 62:1457–1506, 2009), and two versions of the maze task (Forster et al. in Behav Res Methods 41:163–171, 2009)—the lexicality maze and the grammaticality maze—were used to investigate the processing of sentences containing temporary structural ambiguities. Of particular interest were (i) whether each task was capable of revealing processing differences on these sentences and (ii) whether these effects were indicated precisely at the predicted word/region. Although there was considerable overlap in the general pattern of results from the four tasks, there were also clear differences among them in terms of the strength and timing of the observed effects. In particular, excepting sentences that tap into clause-closure commitments, both maze task versions provided robust, “localized” indications of incremental sentence processing difficulty relative to self-paced reading and eye tracking.

Lexical co-occurrence and ambiguity resolution

This study investigates the possible influence of lexical co-occurrence on lexical ambiguity resolution in sentence contexts. Lexical co-occurrence refers to similarity between the co-occurrence vectors of words, such that if two words have similar profiles of occurrence with other words, they are said to have a strong co-occurrence relationship. The present study examines whether lexical ambiguity resolution can be biased by the prior presentation of a word that shares a strong co-occurrence relationship with an ambiguous word under one of its meanings, despite the absence of plausibility support. Two “maze” word-by-word reading experiments examined highly implausible/anomalous sentences with balanced homographs. In sentences in which the ambiguous word (e.g., bat) was preceded by a biasing word with which it shares a strong co-occurrence relationship (e.g., umpire), (1) response times (RTs) to the ambiguous word were facilitated, and (2) garden-path effects were found when subsequent (disambiguating) information was incongruent with the biased meaning (e.g., The umpire tried to swallow the bat but its wings got stuck in his throat). Additional experiments showed that these biasing effects resist explanation in terms of a passive process of spreading activation. Furthermore, an eye-tracking experiment revealed a pattern of results comparable to that of the maze task experiments, indicating that these effects are not artefacts of the maze procedure. These results are taken to support a heuristic for lexical ambiguity resolution that is driven by relatively low-level intralexical connections based on lexical co-occurrence.

Orthographic and Phonological Form Interference during Silent Reading.

This study reports 2 eye-tracking experiments investigating form interference during sentence-level silent reading. The items involved reduced and unreduced relative clauses (RCs) with words that were orthographically and phonologically similar (injection-infection; O+P+, Experiment 1) as well as with words that were orthographically similar, but phonologically dissimilar (laughter-daughter; O+P-, Experiment 2). Both experiments revealed syntactic processing disruptions for reduced RCs. Processing difficulty was also observed at the form-related word in both experiments under first-pass and second-pass reading measures. These form-interference effects did not interact with structural processing difficulty under first-pass measures in either experiment. Under second-pass time, there were larger processing disruptions for reduced RCs in O+P+ sentences relative to their controls. This was not the case, however, for O+P- sentences. These results suggest 2 components to form-interference effects during silent reading: (a) an early, low-level component that is driven in large part by visual form overlap and (b) a component that relates to late stages of interpretation and that is associated more closely with phonological form overlap.

The Processing of Japanese Control Sentences

This study investigates the processing of empty embedded clause subjects in Japanese Control sentences. In these sentences, the information necessary to specify the antecedent of the empty subject is often not available until after the embedded clause, at the Control verb. Therefore, the present study examines the following questions: (a) Is an antecedent provisionally assigned to the empty embedded clause subject in Japanese Control sentences before Control information becomes available? (b) If so, which noun phrase – the main clause subject or object – is assigned as this antecedent? The experiment consisted of a region-by-region, self-paced, non-cumulative moving-window reading task. Japanese native speakers were tested on four sentence types: (a) Subject Control sentences with empty embedded clause subjects (or SC/empty sentences), (b) Subject Control sentences with overt embedded clause subjects (or SC/overt sentences), (c) Object Control sentences with empty embedded clause subjects (or OC/empty sentences) and (d) Object Control sentences with overt embedded clause subjects (or OC/overt sentences). The reading time (RT) patterns on these sentences were consistent with an Object Control bias in the provisional assignment of an antecedent to the empty embedded clause subject. Specifically, at the Control verb, RTs were significantly longer for SC/empty sentences than for SC/overt sentences; however, there was no difference between the RTs for OC/empty and OC/overt sentences at this verb. These results suggest processing difficulty at the point of Control disambiguation only for SC/empty sentences. We interpret these findings as further support for incrementality in the parsing of Japanese sentences in general and in the processing of Japanese filler-gap structures in particular.

Incremental Sentence Processing in Japanese: A Maze Investigation into Scrambled and Control Sentences

This study investigates preverbal structural and semantic processing in Japanese, a head-final language, using the maze task. Two sentence types were tested—simple scrambled sentences (Experiment 1) and control sentences (Experiment 2). Experiment 1 showed that even for simple, mono-clausal Japanese sentences, (1) there are online processing costs associated with parsing noncanonical word orders and (2) these costs are incurred during the incremental integration of constituents into developing sentence representations. Experiment 2 indicated (1) that antecedents are provisionally assigned to empty subjects in Japanese control sentences before verb information becomes available and (2) that this process is guided by an object control bias. Taken together, these findings are interpreted to suggest an important role for preverbal analysis in the processing of displaced constituents and of referential properties for empty elements in head-final languages.

Script differences and masked translation priming: Evidence from Hindi-English bilinguals

This study reports on two experiments investigating the effects of script differences on masked translation priming in highly proficient early Hindi-English bilinguals. In Experiment 1 (the cross-script experiment), L1 Hindi was presented in the standard Devanagari script, while L2 English was presented in the Roman alphabet. In Experiment 2 (the same-script experiment), both L1 Hindi and L2 English were presented in the Roman alphabet. Both experiments revealed translation priming in the L1-L2 direction. However, L2-L1 priming was obtained in the same-script experiment, but not in the cross-script experiment. These findings are discussed in relation to the orthographic cue hypothesis as well as hypotheses that hold that script differences influence the distance between the L1 and L2 in lexical space and/or cross-language lateral inhibition. We also provide alternative accounts for these results in terms of how orthographic cues provided by L1 targets might lead to the discontinuation or disruption of processing for L2 primes.