Research paper
Effects of animacy and sentence type on silent reading comprehension in aphasia: An eye-tracking study

https://doi.org/10.1016/j.jneuroling.2020.100950Get rights and content

Highlights

  • Eye tracking was used to examine reading of object and subject relative sentences with animate or inanimate subjects.

  • Persons with aphasia (PWA) and neurotypical older adults showed effects of structural frequency and syntactic complexity.

  • Processing disruptions were greatest in object relative sentences with animate subjects.

  • PWA showed delayed effects of syntactic complexity during online reading.

Abstract

The present study examined how healthy aging and aphasia influence the capacity for readers to generate structural predictions during online reading, and how animacy cues influence this process. Non-brain-damaged younger (n = 24) and older (n = 12) adults (Experiment 1) and individuals with aphasia (IWA; n = 11; Experiment 2) read subject relative and object relative sentences in an eye-tracking experiment. Half of the sentences included animate sentential subjects, and the other half included inanimate sentential subjects. All three groups used animacy information to mitigate effects of syntactic complexity. These effects were greater in older than younger adults. IWA were sensitive to structural frequency, with longer reading times for object relative than subject relative sentences. As in previous work, effects of structural complexity did not emerge on IWA's first pass through the sentence, but were observed when IWA reread critical segments of the sentences. Thus, IWA may adopt atypical reading strategies when they encounter low frequency or complex sentence structures, but they are able to use animacy information to reduce the processing disruptions associated with these structures.

Introduction

Many individuals with aphasia (IWA) have difficulty understanding sentences, even when they understand the content words contained in those sentences. Much of the research on sentence processing in aphasia has focused on auditory comprehension (e.g., Caplan, Waters, DeDe, Michaud, & Reddy, 2007; Shapiro, 1997; Thompson & Choy, 2009). In addition, most studies have focused on effects of syntactic complexity, with on-line and off-line experiments used to evaluate the relative difficulty associated with long distance dependencies (e.g., active vs. passive sentences, object relative vs. subject relative embedding). More recently, there has been interest in how IWA use other sources of knowledge, such as the relative frequency of different syntactic structures (Knilans & DeDe, 2015) or predictions based on real-world knowledge (e.g., Warren, Dickey, & Lei, 2016), during sentence processing. The present study focuses on how reading strategies used by IWA compare to online reading comprehension in unimpaired, older adults.

There is a lot of evidence that IWA have more difficulty understanding sentences that contain object than subject relative clauses (Caplan, Michaud, & Hufford, 2013; Caplan et al., 2007). Consider examples 1 & 2:

  • (1)

    It was the girl who the boy hugged. Complex Sentence: Object cleft

  • (2)

    It was the boy who hugged the girl. Simple Sentence: Subject cleft

Both sentences contain the meaning "The boy hugged the girl," yet object clefts such as (1) are more difficult to process than subject clefts such as (2). One reason for this difference is that object clefts do not follow canonical word order for English. The most frequent word order in English is subject-verb-object, where the agent of the verb precedes the patient. Subject clefts conform to this S–V–O word order. Conversely, in object clefts the patient (the girl) is encountered before the agent (the boy). Increased difficulty interpreting object-relative structures due to their non-canonical form can be described as an effect of structural frequency. Though both object and subject clefts are relatively rare, object clefts are more rare than subject clefts and have non-canonical word order.

In addition to differences associated with canonical versus non-canonical structure, sentences with subject and object relative clauses also differ in syntactic complexity. Structural differences include factors such as locality of filler and gap and number of intervening phrasal nodes (Gibson, Desmet, Grodner, Watson, & Ko, 2005; Gordon, Hendrick, & Johnson, 2001; Mak, Vonk, & Schriefers, 2002). For these reasons, sentences with object relative clauses have more complex syntactic structures than sentences with subject relative clauses.

Understanding normal reading comprehension processes, and how those processes change as a function of normal aging, are important precursors to determining the extent to which IWA show disordered processing. Neurotypical older and younger adults are sensitive to effects of structural frequency and complexity (Knilans & DeDe, 2015; Staub, 2010). Staub (2010) investigated effects of structural frequency and syntactic complexity on reading comprehension in neurotypical college age adults. He reported that neurotypical, college aged adults showed evidence of processing disruptions (i.e., longer reading times) on the second noun phrase in object relative sentences (Staub, 2010). Staub reasoned that the second noun phrase is the earliest point of an object relative clause sentence at which readers can realize that the sentence does not follow canonical word order. He interpreted longer reading times at this point in the sentence as an effect of structural frequency. Participants also showed longer reading times for the more complex object relative sentences at the embedded verb (e.g., hugged in the examples above), which was interpreted as an effect of syntactic complexity.

For neurotypical younger and older adults, the animacy of the sentential subject can mitigate the difficulty associated with parsing object relative clause sentences (DeDe, 2015; Mak, Vonk, & Schreifers, 2002; Traxler, Morris, & Seely, 2002; Traxler, Williams, Blozis, & Morris, 2005). In an eye tracking study with young adults, Traxler et al. (2002) examined whether the animacy of the sentential subject influenced effects of sentence type1 in sentences with subject and object relative clauses (example sentences 3–6). In their study, effects of sentence type were reduced or even eliminated in sentences with object relative clauses and inanimate sentential subjects (as in example 6). In contrast, typical effects of sentence type were observed in sentences with object relative clauses and animate sentential subjects (example 5). The findings were interpreted as evidence that constraints on animacy influenced how quickly readers were able to revise a mistaken first parse upon encountering components of a sentence incompatible with that parse.2 Thus, even though probabilistic, syntactic, and cognitive factors establish subject relative clauses as easier to process than object relative clauses, unimpaired younger adults use animacy cues to reduce or eliminate the processing cost associated with increased syntactic complexity, greater working memory demand, and less-canonical form.

  • (3)

    SR-A: The farmer that purchased the tractor arrived at the store late last night.

  • (4)

    SR-I: The tractor that impressed the farmer arrived at the store late last night.

  • (5)

    OR-A: The farmer that the tractor impressed arrived at the store late last night.

  • (6)

    OR-I: The tractor that the farmer purchased arrived at the store late last night.

DeDe (2015) asked whether normally aging older adults show similar interactions between animacy and sentence type in a self-paced listening study. Older adults experienced more disruption than younger adults when encountering object relative clauses with animate sentential subjects. However, when the sentential subject was inanimate, older adults did not show effects of sentence type. The results were interpreted as evidence that older adults exploit some aspects of their lexical knowledge to generate predictions about upcoming words in sentences, thus compensating for slowed parsing operations (DeDe, 2014, 2015). In sentences where the grammatical structure conforms to probability-based expectations, this strategy is beneficial. Conversely, when probabilistic cues lead older readers to predict the incorrect structure, the processing cost is greater than that observed in younger readers (DeDe, 2014, 2015).

In sum, there is evidence that neurotypical adults across the lifespan are sensitive to effects of sentence type, and that animacy of the sentential subject may reduce the magnitude of processing disruptions associated with object extracted relative clauses. For IWA, studies of auditory sentence comprehension demonstrate qualitatively similar effects for sentences with object and subject extracted relative clauses compared to neurotypical controls, except that effects of sentence type may be exaggerated or delayed in IWA (e.g., Caplan et al., 2013; Thompson & Choy, 2009). For example, studies of real-time auditory sentence comprehension have reported increased processing difficulty for the verb in object clefts compared to subject clefts in IWA (e.g., Caplan et al., 2007; Caplan et al., 2013; DeDe, 2013).

However, there is reason to believe that IWA show a qualitatively different pattern than healthy adults when silently reading for comprehension. Knilans and DeDe (2015) investigated how IWA and age-matched controls read object and subject cleft sentences using an eye-tracking paradigm. Similar to Staub (2010), Knilans and DeDe (2015) examined effects of both structural frequency and syntactic complexity. The effects of structural frequency in IWA and age-matched controls were consistent with the findings of Staub (2010), meaning that both groups showed processing disruptions at the second noun phrase in object compared to subject cleft sentences. Critically, only age-matched neurotypical controls showed effects of syntactic complexity in the first pass through the sentence. IWA did not show effects of syntactic complexity on the first pass through the sentence, in gaze durations or go-past times. For IWA the effects of syntactic complexity only emerged in rereading times.

Knilans and DeDe (2015) suggested that IWA might focus on lexical retrieval on the first pass through the sentence, and afterwards, use all available information to help build a syntactic representation during rereading. This claim was supported by a follow up study by DeDe (2017), which demonstrated that IWA show sensitivity to variables such as word frequency and word length on the first pass through the sentence. It is important to note that studies using self-paced reading have reported similar effects of syntactic complexity in IWA and age-matched controls (DeDe, 2013). Given that eye tracking during reading is more ecologically valid than self-paced reading, Knilans and DeDe's (2015) results are more likely to represent typical reading strategies used by IWA.

It is unknown whether IWA are able to use other information, such as the animacy of the subject, to mitigate the processing difficulty associated with object relative clauses. That is, no existing studies have examined whether IWA show similar effects of animacy and complexity in parsing compared to neurologically-healthy peers. It is possible that IWA would show more typical effects of syntactic complexity when animacy cues are supportive of an object relative clause reading. There is a large literature indicating that IWA are sensitive to animacy in offline measures of sentence comprehension (e.g., Zurif & Caramazza, 1976). However, it remains unclear whether IWA can use animacy cues to influence online parsing of complex sentences. Thus, the main purpose of the present study was to determine whether IWA are able to use the animacy of the first noun phrase to predict whether a relative clause is subject or object extracted, thus reducing effects of syntactic complexity and/or structural frequency.

Recent studies have examined predictive processing in IWA, and more specifically how IWA integrate morphosyntactic and lexical cues. Several studies have approached this question using visual world paradigms, which assess auditory sentence comprehension. Mack, Ji, and Thompson (2013) examined the role of lexical prediction in sentences such as “Susan will open/break the jar”. The visual displays were manipulated so that sometimes only one of four visual images was a plausible object for the verb. For example, in one condition, “the jar” might be the only pictured object that can be “opened”. Thus, the question was whether listeners integrated selectional restrictions associated with the verb and the visual images, in order to predict the second noun phrase. The results suggested that unimpaired older adults generate lexical predictions online based on verb meaning, whereas IWA showed delayed effects, indicating slower lexical predictions. Similarly, studies focused on verb agreement and nominal case markings have shown delayed integration of morphological cues in aphasia (Hanne, Burchert, De Bleser, & Vasishth, 2015; Schumacher et al., 2015). Hanne et al. (2015) investigated how IWA use verb agreement and nominal case marking morphology to process sentences with object and subject relative clauses. They reported that IWA were sensitive to both types of morphological cue but appeared to integrate the information on a delayed time course.

Studies focused on reading, rather than auditory, comprehension reported somewhat different findings (Huck, Thompson, & Marshall, 2017; Warren et al., 2016). Huck and colleagues used eye-tracking to examine reading of sentences with direct object/sentential complement ambiguities. Their stimuli varied with respect to the presence/absence of the relative clause pronoun “that,” and whether the verb was biased toward the direct object or sentential complement reading. The results showed that (1) IWA were sensitive to the presence of the relative clause pronoun and used it to reliably disambiguate the sentences and (2) IWA showed effects of verb bias at the same points in the sentence as controls. A self-paced reading study also suggested that IWA use lexical information to generate structural predictions online during sentence comprehension (Warren et al., 2016). In their study, the experimental sentences included disjunctions (“or”), which were or were not signaled by the word “either” (e.g. “The agency funded (either) a sociology study or a project to improve math skills.”) The question was whether IWA and controls used the word “either” to predict the upcoming disjunctive phrase. Both groups read critical regions significantly faster in trials where either was present, indicating that IWA used the cue to generate structural predictions on the same time course as controls.

One potential reason for the difference between the studies is the modality. As Knilans and DeDe (2015) suggested, it may be that IWA adopt different processing strategies when reading because they are able to reread portions of the sentence. Further, the IWA showed overall slower reading patterns in both Huck, Thompson, Cruice, and Marshall (2017) and Warren et al. (2016), which could have allowed extra time to process the cues. Another potential difference between the studies is the number of words that intervene between the critical stimulus and the cue. As Warren and colleagues pointed out, the presence of the first disjunct in between either and or provides readers time to generate structural predictions, even if those predictions are slowed. In contrast, in Hanne et al.’s (2015) study, the cues were inflectional morphemes that are attached to the critical words. In Mack et al.’s (2013) work, the noun phrase directly follows the critical verb, which may not permit enough time to access the full semantic representation and integrate it with the syntactic structure.

The present study investigated whether people with aphasia use animacy cues to mitigate effects of syntactic complexity and structural frequency during silent reading comprehension. In this study, the animacy cue was contained in the sentence-initial noun phrase and the relative clause pronoun “that” intervened before the reader encounters either the subordinate verb (indicating a subject relative clause) or the second noun phrase (indicating an object relative clause). Thus, in order to successfully use predictive processing, IWA would need to integrate the animacy of the noun phrase and the upcoming relative clause (as indicated by “that”). Both Huck et al.’s (2017) and Knilans and DeDe's (2015) work suggest that IWA will be sensitive to the presence of the relative clause pronoun, leaving open the question of whether they will also integrate the animacy of the first noun phrase.

To our knowledge, no existing studies have used eye tracking to examine whether IWA show similar effects of animacy and complexity in parsing compared to neurologically-healthy older adults. It is possible that IWA would show more typical effects of syntactic complexity when animacy cues are supportive of an object relative clause reading. However, effects of animacy and syntactic complexity have not been examined in older adults using an eye-tracking paradigm. For this reason, we first report a replication of DeDe's (2015) normal aging study in an eye-tracking paradigm (Experiment 1). The purpose of Experiment 1 was to document whether the effects of age observed in DeDe (2015) are also observed in the written modality; that is, in reading comprehension. Thus, Experiment 1 examines effects of animacy and sentence type in older and younger neurotypical adults. Experiment 2, which focuses on reading comprehension in aphasia, is the main focus of this study. Experiment 2 asks whether IWA show interactions between animacy and sentence type during reading comprehension.3 In the next section, we present methods common to both experiments. Next, we present experiments 1 and 2, followed by a general discussion.

Section snippets

Stimuli

Sentence stimuli were identical to those used by Traxler et al. (2002; Experiment 3). Stimuli were generated in sets of four. Half the sentences contained animate sentential subjects and the other half contained inanimate sentential subjects. Examples are given in (3)–(6) above. The quartets included verbs balanced for frequency and for length to control for confounds associated with lexical properties of the verb. Each sentence was followed by a yes/no comprehension question that required the

Overview & predictions

Experiment 1 examined whether older adults show effects of structural frequency and syntactic complexity when reading sentences with object and subject relative clauses, and whether the magnitude of those effects was modulated by the animacy of the sentential subject. Given that older adults were sensitive to the animacy manipulation during auditory comprehension (DeDe, 2015), we predicted similar effects during silent reading.

Effects of structural frequency. We predicted a three-way

Overview & predictions

Experiment 2 investigated whether individuals with aphasia (IWA) are sensitive to effects of structural frequency and syntactic complexity during silent reading comprehension, and whether the animacy of the sentential subject influences the magnitude of those results. In addition, this study provides a replication of Knilans and DeDe's (2015) results, which suggested that IWA delayed syntactic processing until rereading of the sentences.

Effects of structural frequency. Knilans and DeDe (2015)

General discussion

The present study examined silent reading comprehension of non-brain-damaged older and younger adults, as well as individuals with aphasia (IWA). The results showed that all three groups used the animacy of the sentential subject to mitigate effects of syntactic complexity. As a result, all three groups showed greater processing disruptions when the animacy cues were not predictive of the object relative interpretation of the sentence (as in the OR-A condition). Recall that our primary research

Declaration of competing interest

Gayle DeDe receives a salary from Temple University. There are no other financial or non-financial competing interests to disclose.

Acknowledgments

This work was supported by the National Institutes of Health [K23 DC010808]. The authors would like to thank the research participants and their families as well as the members of the Speech Language and Brain Lab who helped with data collection.

References (34)

  • T. Warren et al.

    Structural prediction in aphasia: Evidence from either

    Journal of Neurolinguistics

    (2016)
  • D. Bates et al.

    Ime4: Linear mixed effects models using S4 classes. R package version 0.999375 17

    (2008)
  • D. Caplan et al.

    Effects of age, speed of processing, and working memory on comprehension of sentences with relative clauses

    Psychology and Aging

    (2011)
  • S. Cho-Reyes et al.

    Verb and sentence production and comprehension in aphasia: Northwestern assessment of verbs and sentences (NAVS)

    Aphasiology

    (2012)
  • G. DeDe

    Effects of verb bias and syntactic ambiguity on reading in people with aphasia

    Aphasiology

    (2013)
  • G. DeDe

    Sentence comprehension in older adults: Evidence for risky processing strategies

    Experimental Aging Research

    (2014)
  • G. DeDe

    Effects of animacy on processing relative clauses in older and younger adults

    Quarterly Journal of Experimental Psychology

    (2015)
  • View full text