Cataphora processing in agrammatic aphasia: Eye movement evidence for integration deficits

Introduction Individuals with agrammatic aphasia show selective impairments in comprehending syntactically complex sentences with long-distance dependencies such as WH-dependencies. In the literature, conflicting results have been reported with respect to the source of such deficits. Representational accounts claim that these deficits are attributed to impaired linguistic phenomenon critical to WH-dependency computation [1-3]; whereas, processing resource accounts suggest that general processing resource limitations such as impaired working memory capacity or generally slowed processing underlie comprehension difficulty [4-6]. Lexical Integration Deficit Hypothesis [LIDH, 7-8] suggests that the process of integrating lexical items into the syntax is impaired. Recent eye-tracking-while-listening studies show that agrammatic listeners process complex sentences normally in real time, but still are unable to understand sentence meaning. These findings suggest that comprehension difficulty arises from failure to integrate hierarchical grammatical structures and lexical elements. The present study further evaluates the LIDH by investigating patients’ processing of cataphoric dependencies, another type of long-distance dependency that relies on integration processes for sentence comprehension.


Introduction
Individuals with agrammatic aphasia show selective impairments in comprehending syntactically complex sentences with long-distance dependencies such as WH-dependencies.In the literature, conflicting results have been reported with respect to the source of such deficits.Representational accounts claim that these deficits are attributed to impaired linguistic phenomenon critical to WH-dependency computation [1][2][3]; whereas, processing resource accounts suggest that general processing resource limitations such as impaired working memory capacity or generally slowed processing underlie comprehension difficulty [4][5][6].Lexical Integration Deficit Hypothesis [LIDH, [7][8] suggests that the process of integrating lexical items into the syntax is impaired.Recent eye-tracking-while-listening studies show that agrammatic listeners process complex sentences normally in real time, but still are unable to understand sentence meaning.These findings suggest that comprehension difficulty arises from failure to integrate hierarchical grammatical structures and lexical elements.The present study further evaluates the LIDH by investigating patients' processing of cataphoric dependencies, another type of long-distance dependency that relies on integration processes for sentence comprehension.
Cataphoric Dependencies involve a long-distance relation between a pronoun and its antecedent, where the pronoun (dependent element) precedes the antecedent (licensing element), as illustrated in (2a).Structurally, the relation between the pronoun and the antecedent is governed by Binding Condition C [9]: the pronoun-antecedent relation is impossible when a pronoun ccommands it antecedent as in (2b).
After he i read the paper, John i went to the meeting.
b. *He i read the paper after John i went to the meeting.
Previous studies showed that cataphoric and WH-dependencies are processed similarly: in both structures, the parser actively searches for the licensing element, and the search processes respect grammatical constraints [10][11][12].That is, upon encountering the dependent element, the parser starts searching for its licensing element to establish the dependency as soon as possible.
Importantly, this search is grammatically constrained; therefore, in examples like (2), the parser tries to associate the pronoun and the antecedent in (2a), but not (2b).In previous studies, the active search effect is shown using the gender mismatch paradigm: during online reading of examples like (3), eye-gaze durations were longer when the gender specification of the pronoun (e.g., she-female) and the closest potential antecedent (e.g., the boy-male) mismatched (the socalled Gender Mismatch Effect (GMME)), indicating the parser's 'surprised' at failing to complete the dependency in the closest available antecedent position [11][12][13].
When he i /*she i was at the party, the boy i cruelly teased the girl.b.
He/she was at the party when the boy cruelly teased the girl.
Some recent studies on cataphora processing also have found that lexical information, such as stereotypical or definitional gender information is crucial for the GMME [13,14].These studies investigated cataphora processing in healthy young and older population, manipulating gender specification types of the antecedent.They found that the GMME is observed only with definitional gender nouns in (4a), where the gender of the antecedent is unambiguously specified, but not for stereotypical nouns in (4b) where the gender of the antecedent is ambiguously marked, but understood stereotypically either as male or female [13,14].Such results suggest that the gender interpretation of the antecedent is affected by the preceding pronoun/reflexive, i.e., when the pronoun/reflexive is male (himself), a stereotypical noun (e.g., minister) is interpreted as male, but when the pronoun/reflexive is female, the stereotypical noun is interpreted as female.This finding indicates that processing stereotypical nouns requires additional integration costs: readers have to integrate the gender specification of the antecedent based on the gender of the pronoun.In this way, cataphora involves dependency formation and lexical processes such as gender checking and integrating gender information into a sentential context.
After reminding himself i /*herself i about the letter the king i left London.b.
After reminding himself i /herself i about the letter the minister i left London.
This study examined the integration of grammatical and lexical information during cataphora processing in agrammatic and control participants using eye-tracking self-paced reading paradigm, since eye movements are preserved in agrammatic patients.Additionally, eyetracking enables us to examine patients' online cataphora processing, even when they fail to comprehend sentences offline.

Method
Participants: Six individuals with clinical diagnosis of agrammatic aphasia (WAB AQ: 70.8-89.6),and 15 young healthy control speakers participated in the study.All were native English speakers, and their vision and hearing were within normal limits.There was no reported history of neurological or psychological disorders.Table 1 lists the agrammatic patients' clinical and demographic information.
Materials: In an eye-tracking self-paced reading experiment, the c-command relation between the pronoun and antecedent (non-c-command/dependency vs. c-command/no-dependency), the gender between the pronoun and its potential antecedent (gender match vs. mismatch), and the gender specifications (stereotypical vs. definitional) were manipulated as in (5a-h).Eighty sets of experimental sentences were created and divided into two separate scripts, in addition to 160 fillers.Participants were tested using either script.
(5) a/b.Yesterday after he i /*she i read the paper, the king i happily went to the meeting.c/d.
Yesterday he/she read the paper after the king happily went to the meeting.e/f.Yesterday after he i /she i read the paper, the minister i happily went to the meeting.g/h.Yesterday he/she read the paper after the minister happily went to the meeting.
According to LIDH, it was anticipated that both healthy and agrammatic participants would show a GMME for the dependency-mismatch-definitional condition (as in (5b)), but not in any c-commanded no-dependency conditions.However, unlike controls, agrammatic participants were predicted to also show a GMME in the dependency-mismatch-stereotypical condition (as in (5f)), due to their deficits in integration processing, since processing such sentences requires revision of gender stereotype using prior contextual gender information in order to build a dependency between the pronoun and its gender mismatched stereotypical antecedent.Data analysis: A set of eyetracking measures (first fixation, first pass, regression-path and second pass reading times) recorded in the critical antecedent (the king/minister) and spillover (adverb) regions were analyzed using linear mixed effect models.Offline comprehension scores between two groups were analyzed using Mann-Whitney U test.

Results
Results are shown in Table 2-4.Importantly, the analyses of the regression-path reading times revealed a significant interaction of c-command x gender x noun type in the critical region for the healthy controls, associated with a GMME in the dependency/mismatched/ definitional condition, but not in the stereotypical condition (Figure 1).A similar, albeit marginally significant interaction was found in the spillover region for the patient group (Figure 2), followed by a highly significant interaction in the second-path reading time measure (Figure 3).However, GMMEs in both definitional and stereotypical c-command/mismatched conditions were found, with a greater and later GMME in the stereotypical compared to the definitional conditions.Additionally, the aphasic participants showed chance-level performance across all conditions.

Conclusions
These findings indicate that the patient group exhibited similar eye movement patterns to that of the control group when processing definitional nouns in cataphoric dependencies.However, unlike controls, they showed processing difficulties associated with gender ambiguity.These patterns support the LIDH.These findings will be discussed with regard to alternative theories of agrammatic sentence comprehension as well as their clinical applicability.(1165 words)

Table 2
Young Normals' reading time data in the critical region

Table 3
Patients' reading time data in the critical region

Table 4
Young Normals' eye data in the spillover region

Table 5
Patient's eyedata in the spillover region