Prominence of gender cues in the assignment of thematic roles in German

YULIA ESAULOVA; CHIARA REALI; LISA VON STOCKHAUSEN

doi:10.1017/S014271641700008X

Prominence of gender cues in the assignment of thematic roles in German

Published online by Cambridge University Press: 09 May 2017

YULIA ESAULOVA ,

CHIARA REALI and

LISA VON STOCKHAUSEN

Show author details

YULIA ESAULOVA*: Affiliation:
University of Potsdam
CHIARA REALI: Affiliation:
University of Duisburg–Essen
LISA VON STOCKHAUSEN: Affiliation:
University of Duisburg–Essen
*: ADDRESS FOR CORRESPONDENCE Yulia Esaulova, Potsdam Research Institute for Multilingualism, University of Potsdam, Karl-Liebknecht-Strasse 24–25, Potsdam 14476, Germany. E-mail: esaulova@uni-potsdam.de

Article contents

Abstract
THEMATIC STRUCTURE AND PROMINENCE HIERARCHIES
GENDER PROCESSING AND AGENCY
HYPOTHESES
OVERVIEW OF THE PRESENT RESEARCH
EXPERIMENT 1
EXPERIMENT 2
GENERAL DISCUSSION
Footnotes
References

Rights & Permissions

Abstract

Two eye-tracking experiments examined influences of grammatical and stereotypical gender cues on the assignment of thematic roles in German. Participants (N 1 = 32, N 2 = 40) read sentences with subject- and object-extracted relative clauses, where thematic agents and patients remained ambiguous until the end of the relative clause. The results reveal a linguistic gender bias: agent roles are assigned more easily to grammatically masculine than feminine role nouns and stereotypically neutral than female ones. The opposite pattern is observed in the assignment of patient roles for stereotypical but not grammatical gender. The findings are discussed within the framework of situation model theories as well as in constraint-based and similarity-based interference accounts, while gender is viewed as a dimension of prominence.

Information

Type: Articles
Information: Applied Psycholinguistics , Volume 38 , Issue 5 , September 2017 , pp. 1133 - 1172

DOI: https://doi.org/10.1017/S014271641700008X [Opens in a new window]
Copyright: Copyright © Cambridge University Press 2017

Resolving linguistic ambiguities can reveal tendencies in the use of provided information, or biases toward one of the possible interpretations, that can be easily overlooked otherwise. In this paper we use temporarily ambiguous sentences to examine whether gender markings may function as cues moderating the assignment of thematic roles in complex relative clause constructions. The processing of sentence (1) may appear as difficult as that of sentence (2), as they both contain object-extracted relative clauses (ORC). Sentences (3) and (4) contain subject-extracted relative clauses (SRC), and therefore (3) may seem equally difficult to comprehend as (4).

(1) The beautician, whom both designers recognized, is experienced.
(2) The artist, whom both designers recognized, is experienced.
(3) The beautician, who recognized both designers, is experienced.
(4) The artist, who recognized both designers, is experienced.

However, there is evidence that certain features shared by nouns or noun phrases (e.g., animacy) facilitate the assignment of specific thematic roles, such as when agent roles are assigned more easily to animate and patient roles to inanimate nouns (e.g., Bornkessel-Schlesewsky & Schlesewsky, Reference Bornkessel-Schlesewsky and Schlesewsky2009; MacDonald, Reference MacDonald1994; Trueswell, Tanenhaus, & Kello, Reference Trueswell, Tanenhaus and Kello1993). While claims about animate nouns as better agents (Gennari & MacDonald, Reference Gennari and MacDonald2008) and inanimate nouns as poorer agents (Clifton et al., Reference Clifton, Traxler, Mohamed, Williams, Morris and Rayner2003; Just & Carpenter, Reference Just and Carpenter1992) have already been supported by empirical evidence, the claim about gender cues as predictors of good or poor agents has not yet been demonstrated experimentally. At the same time, similarly to animacy when described as a “supralinguistic” concept (Yamamoto, Reference Yamamoto1991) affecting linguistic phenomena (e.g., word order), gender can be regarded as a fundamental semantic dimension expressed on biological and social levels (Ridgeway, Reference Ridgeway2001) with various linguistic representations (e.g., grammaticalFootnote ¹ [der Musiker/die Musikerin “the musician_{masculine/feminine}” in German], natural [woman/man], definitional [mother/father], and stereotypical [nurse/surgeon] gender). Based on psycholinguistic and sociocognitive research on gender that will be discussed in more detail below, this study attempts to clarify whether gender, along with animacy, may affect the identification of thematic roles. Previous research has reported gender-related observations, such as case syncretism of feminines in German (Krifka, Reference Krifka, Steinkrüger and Krifka2009) and the strong link between agency and masculinity (Koenig, Mitchell, Eagly, & Ristikari, Reference Koenig, Mitchell, Eagly and Ristikari2011), suggesting that entities possessing female characteristics may be less agentive and therefore more likely to serve as thematic patients. If this is the case, then stereotypically female beautician may be perceived as a better patient compared to neutral artist (i.e., receiving an action) in the ORC sentences, making the comprehension of (1) easier than (2). Similarly, neutral artist may be perceived as a better agent compared to stereotypically female beautician (i.e., producing an action) in the SRC sentences, making (4) easier than (3). The experiments reported in the present paper use relative clause structures in German as a tool to address this issue.

THEMATIC STRUCTURE AND PROMINENCE HIERARCHIES

A number of studies invoked thematic structure to explain biases that influence the interpretation of complex linguistic constructions, such as relative clauses (e.g., Boland, Tanenhaus, Garnsey, & Carlson, Reference Boland, Tanenhaus, Garnsey and Carlson1995; Pickering & Traxler, Reference Pickering and Traxler1998; Pickering, Traxler, & Crocker, Reference Pickering, Traxler and Crocker2000; Traxler, Morris, & Seely, Reference Traxler, Morris and Seely2002). Evaluating the role of various factors in the comprehension of relative clauses, previous research has repeatedly shown that ORCs are more difficult to process than SRCs (Gordon, Hendrick, & Johnson, Reference Gordon, Hendrick and Johnson2001; Staub, Reference Staub2010; Traxler, Williams, Blozis, & Morris, Reference Traxler, Williams, Blozis and Morris2005). However, Mak, Vonk, and Schriefers (Reference Mak, Vonk and Schriefers2002, Reference Mak, Vonk and Schriefers2006) and Traxler et al. (Reference Traxler, Morris and Seely2002) demonstrated that ORCs with inanimate heads, such as The movie that the director watched received the prize, were almost as easy to comprehend as SRCs of the type The director that watched the movie received a prize. Thus, the feature of animacy has been shown to modulate the difficulty in the interpretation of relative clauses, showing that linguistic characteristics are a significant factor modulating the likelihood of the assignment of an agent role to one of the two nouns in a sentence.

The expectations readers have about entities possessing certain characteristics to occupy syntactically prominent positions in a sentence can be seen within the framework of the thematic hierarchy hypothesis (e.g., Grimshaw, Reference Grimshaw1990; Jackendoff, Reference Jackendoff1987). This hypothesis lists the ordering of thematic roles by prominence, with the agent role ranking the highest on the hierarchy of semantic features. At the same time, prominence can be assessed along several dimensions other than thematic agency, such as animacy, definiteness, or person, with animate entities ranking over inanimate, definite over indefinite, and first and second person over third (Lamers & de Swart, Reference Lamers, de Swart, Lamers and de Swart2012). The tripartite animacy hierarchy proposed by Croft (Reference Croft1990) offers a similar ordering by person (first and second over third), NP-type (pronouns over common nouns), and animacy itself (human over nonhuman, animate over inanimate). The definition of the agentive case, given by Fillmore as “the typically animate perceived instigator of the action identified by the verb” (Reference Fillmore, Bach and Harms1968, p. 24), indicates the relatedness of the two concepts: agency and animacy. In line with this definition, Yamamoto (Reference Yamamoto1991) suggests that agency presupposes animacy, considering that previous research had named such conceptual properties of agency as intentionality (Davidson, Reference Davidson, Binkley, Bronaugh and Marras1971), dynamicity, and control (Dik, Reference Dik1989). These properties are not purely linguistic, which contributes to Yamamoto's understanding of animacy as an “extra-” or “supralinguistic” concept, which nevertheless relates to such linguistic phenomena as case marking, word order, subject selection, and gender.

Even though the interaction of different prominence dimensions remains a subject of debate (e.g., Klein, Guntsetseg, & von Heusinger, Reference Klein, Guntsetseg, von Heusinger, Lamers and Swart2012; Primus, Reference Primus, Lamers and de Swart2012), the principle of harmonic alignment suggests that hierarchies within separate dimensions map onto one another, so that the hierarchy within the dimension of animacy, for instance, correlates with that of thematic roles (Lamers, Reference Lamers, Lamers and de Swart2012). The processing is facilitated when rankings on different hierarchies point to the same argument in a sentence as being more prominent (Bornkessel-Schlesewsky & Schlesewsky, Reference Bornkessel-Schlesewsky and Schlesewsky2008). As a result of such alignment, animacy information can be used in the assignment of thematic roles during language comprehension. Thus, readers seem to have expectations about high-ranked animate entities to produce actions represented by the verb (i.e., serve as agents that are high ranked on a thematic role hierarchy), while low-ranked inanimate entities are expected to rather receive those actions (i.e., serve as patients that are low ranked on a thematic role hierarchy). This tendency can be regarded as a bias moderating difficulties in the interpretation of syntactically complex sentences.

GENDER PROCESSING AND AGENCY

The present investigation extends the current knowledge about biases in linguistic structures by examining the role of grammatical and stereotypical gender in the resolution of relative clauses. In the literature on reference resolution, the integration of grammatical and stereotypical gender cues is widely discussed in terms of mismatch effects that are reflected in longer processing times when stereotypically male (e.g., electrician) or female (e.g., beautician) role nouns corefer with mismatching information, such as gender suffixes, gender-specific pronouns, or noun phrases (e.g., Cacciari, Corradini, Padovani, & Carreiras, Reference Cacciari, Corradini, Padovani and Carreiras2011; Esaulova, Reali, & von Stockhausen, Reference Esaulova, Reali and von Stockhausen2014; Irmen, Reference Irmen2007; Kreiner, Sturt, & Garrod, Reference Kreiner, Sturt and Garrod2008; Reali, Esaulova, & von Stockhausen, in press). Grammatical and stereotypical gender cues have been shown to affect readers’ interpretation of role nouns in highly automatized ways and to strongly influence the comprehension of sentences (e.g., Banaji & Hardin, Reference Banaji and Hardin1996; Cacciari & Padovani, Reference Cacciari and Padovani2007; Carreiras, Garn- ham, Oakhill, & Cain, Reference Carreiras, Garnham, Oakhill and Cain1996; Esaulova et al., Reference Esaulova, Reali and von Stockhausen2014; Irmen, Holt, & Weisbrod, Reference Irmen, Holt and Weisbrod2010).

In the literature mentioned above, the effects of both grammatical features and stereotype-based connotations of roles and contexts are analyzed. On the one hand, this research clearly points at the fact that both gender representations affect reference resolution, and on the other hand, it relates the linguistic and the social psychological understanding of the term gender on a conceptual level. This is in line with social cognition research (e.g., Stahlberg, Braun, Irmen, & Sczesny, Reference Stahlberg, Braun, Irmen, Sczesny and Fiedler2007), which indicates the association between conceptual and formal gender representations, where the former are expressed through gender stereotypes and the latter through grammatical features, such as gender suffixes. However, neither research on thematic structures (e.g., Clifton et al., Reference Clifton, Traxler, Mohamed, Williams, Morris and Rayner2003; Trueswell & Tanenhaus, Reference Trueswell, Tanenhaus, Clifton, Frazier and Rayner1994) nor research on gender processing (e.g., Cacciari & Padovani, Reference Cacciari and Padovani2007; Carreiras et al., Reference Carreiras, Garnham, Oakhill and Cain1996; Esaulova et al., Reference Esaulova, Reali and von Stockhausen2014; Irmen, Reference Irmen2007) and social cognition (Stahlberg et al., Reference Stahlberg, Braun, Irmen, Sczesny and Fiedler2007) have ever supposed the link between gender and agency or considered gender a relevant factor in the assignment of thematic roles. At the same time, some approaches in historical linguistics indicate such a link arguing that an initial two-way distinction in gender in Germanic languages may have been based on animacy, agentivity, or the capability of subject marking (Kürschner & Nübling, Reference Kürschner and Nübling2011; Meier-Brügger, Reference Meier-Brügger2002). Bringing the research from these different areas together, we supposed that gender characteristics of nouns should be examined as constituting another dimension along which the prominence of thematic role nouns can be assessed. This perspective draws largely on McRae, Hare, Elman, and Ferretti's (Reference McRae, Hare, Elman and Ferretti2005) account of thematic roles, which views them as concepts formed through everyday experience rather than simply lexical information. This account, in turn, is based on the situation model theories (Sanford & Garrod, Reference Sanford and Garrod1981), which emphasize the role of long-term memory in the representation of a linguistically described event. Following Yamamoto, we suggest that, similar to agency, gender presupposes animacy, most certainly in those cases when it points at the sex of the referent. The evidence of animacy-based role assignments (Wang, Schlesewsky, Philipp, & Bornkessel-Schlesewsky, Reference Wang, Schlesewsky, Philipp, Bornkessel-Schlesewsky, Lamers and de Swart2012) encourages the assumption that gender cues may influence readers’ implicit beliefs about good or poor agents in a similar way. This should be reflected through readers’ expectations about nouns of certain gender to produce or receive actions represented by the verb in a sentence. In social psychology, male roles are associated with higher status and power and are described as more agentive and less communal than female ones (e.g., Koenig et al., Reference Koenig, Mitchell, Eagly and Ristikari2011). Agency and communion are fundamental dimensions of social categorization (e.g., Fiske, Cuddy, & Glick, Reference Fiske, Cuddy and Glick2007), in which the former comprises such characteristics as assertion, competence, and independence, while the latter is associated with cooperation, warmth, and empathy. An extensive body of research in this field indicates a possible association between agency and gender representations (e.g., masculine and feminine sex roles, see Bem, Reference Bem1981; sex-role socialization, see Cross & Madson, Reference Cross and Madson1997; Helgeson, Reference Helgeson1994; masculinity, see Koenig et al., Reference Koenig, Mitchell, Eagly and Ristikari2011; Spence & Buckner, Reference Spence and Buckner2000).

HYPOTHESES

In this study we examine gender represented through grammatical and conceptual characteristics, since research on language-based representation of women and men points at the commonalities in their processing. If gender constitutes an important factor in the assessment of agency, as we would like to argue, we should be able to observe its influences on agency when it is expressed through both gender-marking suffixes and through gender stereotypes, such as in typically male (/neutral)/female occupational role nouns. The hypotheses below specify this general proposition by making specific predictions about each of these two types of gender representations. These predictions are tested in experimental designs that manipulate the morphological marking of gender in role nouns and the stereotypical gender of role nouns independently in order to provide converging evidence about the relationship between gender processing and the identification of thematic roles.

The phenomenon of differential object marking described in functional/typo- logical literature (e.g., Aissen, Reference Aissen2003; Stiebels, Reference Stiebels, Stiebels and Wunderlich2000) offers a theoretical frame for considering grammatical gender in relation to agency. Differential object marking defines the likelihood of an object to be overtly case-marked as a function of prominence ranking. Thus, Aissen (Reference Aissen2003) describes preferred case marking patterns relating them to prominence hierarchies and shows that aligned agent > patient and animate > inanimate hierarchies lead to unmarked animate agents and inanimate patients, whereas the violation of this alignment leads to the overt case marking of entities. Krifka (Reference Krifka, Steinkrüger and Krifka2009) further considers the phenomenon of differential object marking in relation to the gender and case system in Germanic languages in order to explain syncretism of nominatives and accusatives observed for feminines in German. In German, the case marking of singular masculine determiners is expressed overtly for nominative and accusative (der _Nominative; den _Accusative), while feminine determiners lack any morphological distinction in these cases (die _{Nominative/Accusative}). Similar syncretism for the neuter has been explained by the fact that the distinction between nominative and accusative is functionally important for animates, as neuter is, originally, a class of inanimates. Since there is no reason why the masculine and the feminine should be unequal in terms of animacy, Krifka (Reference Krifka, Steinkrüger and Krifka2009) evaluates alternative explanations (e.g., Becker, Reference Becker and Köpcke1994; Bittner, Reference Bittner2003) suggesting that this asymmetry may result from the evolution of language within the speech community where feminine NPs occur less frequently in the agent role than masculine ones. According to the concept of harmonic alignment, the hierarchy of thematic roles corresponds to the underlying mapping of prominence hierarchies, which would indicate that masculine entities should be more prominent, since they are more likely to align with agent roles than feminine entities. This prediction is expressed in Hypothesis 1, which concerns grammatical gender: if grammatical gender constitutes a prominence dimension and feminine is ranked lower on the hierarchy than masculine, then readers should have expectations about grammatically feminine role nouns to rather function as patients than agents (to receive rather than produce actions) compared to masculine role nouns, which would be reflected in longer processing times for feminine agents than masculine ones.

Social cognition research indicates theoretical grounds to relate agency and stereotypical gender. If the association between agency and masculinity (e.g., Koenig et al., Reference Koenig, Mitchell, Eagly and Ristikari2011) described above can be carried over to linguistic terms, then stereotypically male (/neutral) nouns should be good agents and poor patients compared to stereotypically female nouns. Hypothesis 2 is based on these considerations and regards stereotypical gender: if stereotypical gender constitutes a prominence dimension and female entities are ranked lower on the hierarchy than neutral, then readers should have expectations about stereotypically female role nouns to function as patients rather than agents compared to neutral role nouns, which would become evident through longer processing times for female agents than neutral ones.

In terms of eye-tracking measures, both hypotheses translate into the prediction that longer fixation times and more regressions should occur in sentences where feminine or female role nouns are agents and shorter fixations and fewer regressions in sentences where masculine or neutral role nouns are agents.

OVERVIEW OF THE PRESENT RESEARCH

The influence of thematic role characteristics on the production and comprehension of particular syntactic structures (e.g., SRCs and ORCs) does not seem to be restricted to a particular language (e.g., English, see McDonald, Bock, & Kelly, Reference McDonald, Bock and Kelly1993; Spanish, see Prat-Sala, Reference Prat-Sala1997; German, see Van Nice & Dietrich, Reference Van Nice and Dietrich2003). The standard finding that ORCs are interpreted with greater difficulty than SRCs mentioned above also extends to the case of German (e.g., Friederici, Steinhauer, Mecklinger, & Meyer, Reference Friederici, Steinhauer, Mecklinger and Meyer1998).

Since German is one of the languages in which the verb often appears in clause-final position, the integration of information to predict the temporarily unavailable or ambiguous input must be critical to efficiently interpret the sentence. Previous studies have shown that both syntactic and semantic information is used to predict the subsequent input (e.g., Kamide, Altmann, & Haywood, Reference Kamide, Altmann and Haywood2003). According to McRae et al.’s (Reference McRae, Hare, Elman and Ferretti2005) account on thematic roles, conceptual/world knowledge information from thematic role characteristics is computed and used immediately in online language processing. Our research question consisted of clarifying whether gender cues are relevant indicators of prominence in readers’ expectations about thematic agents and patients, which contribute to the rapid integration of information in a sentence. In German, certain combinations of gender and number in nouns of the main and the relative clause make it possible to construct sentences where ORCs and SRCs can only be identified as such by the form of the auxiliary verb at the end of the relative clause but are otherwise identical in structure, as in the following Examples (5) and (6):

Such temporary ambiguity in the thematic structure allowed us to vary grammatical and stereotypical gender of role nouns in German sentences containing ORCs and SRCs to test whether gender information is used in the assignment of thematic agent and patient roles. Previous research has shown that the analysis of subject–object ambiguity is influenced by the relative ranking of the arguments on prominence hierarchies (Haupt, Reference Haupt2008). If gender information constitutes a relevant dimension that contributes to the prominence of thematic roles in a sentence, then it should be reflected in the processing of syntactically ambiguous structures, such as in the German sentences containing SRCs and ORCs mentioned above.

In both experiments reported in this paper, we examined the empirical validity of our theoretical assumptions about gender as a dimension of prominence. The hypotheses were tested in two experiments designed using temporarily ambiguous sentences containing SRC and ORC structures, as provided in Examples (5) and (6). The identification of role nouns as agents and patients in these sentences was not possible until the auxiliary verb of relative clauses had been reached. Experiment 1 examined the effects of grammatical gender by varying grammatical cues (masculine and feminine) in role nouns that were neutral with regard to stereotypical gender. Experiment 2 extended the focus to stereotypical gender influences and included the variation of grammatical (masculine and feminine) and stereotypical (neutral and female) gender cues of involved role nouns.Footnote ²

EXPERIMENT 1

In Experiment 1, we investigated to what extent grammatical gender functions as a cue to agency and affects the resolution of temporarily ambiguous relative clauses.

Method

Participants

Thirty-two students at the University of Duisburg-Essen (15 male, 17 female, mean age 26.3 years, SD = 4.7) were paid to participate. All of them were native speakers of German who did not report to be highly proficient in other languages and had normal or corrected to normal vision.

Materials

EXPERIMENTAL STIMULI

Twenty-four experimental sentences consisted of a main and a relative clause connected by the relative pronoun die “who/whom_{feminine sg/masculine or feminine plural}” that can be interpreted either as feminine singular or as masculine or feminine plural. Main clauses contained plural forms of 24 role nouns (RN1), which varied in grammatical gender (feminine and masculine, feminine marked by the feminine plural suffix –innen Footnote ³ ) but were neutral with regard to stereotypical gender (e.g., Student-en/-innen “students_{Neutral+masculine/feminine}”). Relative clauses contained singular forms of 24 role nouns (RN2), all of which were grammatically feminine and neutral with regard to stereotypical gender (according to the previously published normsFootnote ⁴ ). The verb in the relative clause was an action verb and necessarily involved two arguments, while the verb of the main clause was a state verb (see Appendix A for experimental stimuli with masculine RN1). Both main and relative clauses of each sentence were presented simultaneously in one line. In sentences with SRCs, such as (7), RN1 served as agents and RN2 served as patients. According to Hypothesis 1, masculine RN1 agents were expected to facilitate comprehension compared to feminine ones. In sentences with ORCs, such as (8), RN2 served as agents and RN1 served as patients. Therefore feminine RN1 patients were expected to facilitate comprehension compared to masculine ones.

All of the sentences had the following fixed structure: determiner + RN1 + relative pronoun + determiner + RN2 + action verb + auxiliary verb + main clause verb + adjective. The identification of a relative clause as subject-extracted (die = “who”) or object-extracted (die = “whom”) was not possible until its last word (the auxiliary verb hat “has” or haben “have”) had been reached.

VERB AND ADJECTIVE PRETESTS

To exclude potential confounding effects resulting from the context, a series of pretests was conducted to ensure that verbs and adjectives used in experimental items did not contain any gender information. The verb pretest consisted of 24 transitive verbs requiring a two-argument structure (e.g., erkennen “to recognize”), while 16 verbs reflecting stereotypically female and male characteristics (8 verbs like kochen “to cook” and 8 verbs like boxen “to box”), as well as 8 verbs allowing a different number of arguments (e.g., versprechen “to promise”), served as pretest fillers. The adjective pretest consisted of 82 items that were structurally similar to the main clause in the experimental items, except that role nouns were replaced with X (e.g., X ist aufmerksam “X is thoughtful”). Adjectives that reflected stereotypically male and stereotypically female characteristics (22 adjectives like wetteifernd “competitive” and 21 adjectives like liebevoll “affectionate”) were used as pretest fillers. A total of 37 participants were asked to rate pretest items (i.e., verbs and adjectives) for gender stereotypicality on a scale from 1 (stereotypically male) to 7 (stereotypically female). All verbs and adjectives around the scale midpoint, that is, with ratings from 3.5 to 4.5, were considered neutral with regard to gender stereotypicality and used in the study. In addition, the selected RN pairs were controlled for hierarchical relations, so that RNs that differed in status (e.g., professor/student) were not used in the same sentence.

FILLERS

To prevent participants from developing reading strategies based on the gender characteristics of role nouns and on temporarily ambiguous relative clause structures of experimental items, 24 filler items were constructed. Filler sentences consisted of a main clause and a relative clause connected by the relative pronoun die, which was followed by an unambiguous nominative or accusative masculine determiner der/den (each occurring in 50% of all fillers). All fillers had a fixed structure that imitated the experimental sentences. Plural forms of 24 neutral role nouns served as RN1 in main clauses; they were either grammatically masculine or nominalized participles, which do not express grammatical gender. The singular forms for another 24 neutral role nouns served as RN2 in relative clauses.

Design

The experimental design included two factors: grammatical gender of RN1 (masculine vs. feminine; within subjects and within items), and type of relative clause (SRC vs. ORC; within subjects and within items). Four randomized lists presented each item in one of the four conditions: masculine RN1 + SRC; masculine RN1 + ORC; feminine RN1 + SRC; and feminine RN1 + ORC. Across lists, each item occurred equally often in each condition. Participants were presented with one of the lists (24 experimental items); that is, they received 6 items in each of the four conditions and encountered each item only once. A quarter of the sentences (including fillers) were followed by a yes/no comprehension question to ensure that participants read materials carefully enough and understood their content.

Procedure

Eye movements were recorded using an Eyelink 1000 eye-tracker with a sampling rate of 1000 Hz and angular resolution of 10–30 min of arc (about 0.15° to 0.5°). Participants were seated 70 cm from the computer screen, at which distance three characters subtended 1° of visual arc. All sentences in the experiment were presented in 22-point Lucida Console font and displayed on a single line. Viewing was binocular, but only the dominant eye was recorded. A chin rest was used to minimize head movements.

Participants were tested individually. Before the experiment began, they were instructed to read for comprehension in their normal reading speed, pressing corresponding buttons on a response pad to move on to the next sentence, and to answer questions. Then a calibration procedure with a 9-point grid was performed. Each trial started with the presentation of a fixation point located at the beginning of the sentence to be triggered. Whenever the experimenter judged fixation on the point as inaccurate, recalibration was carried out. The first four sentences with two questions served as practice trials. The eye-tracking session lasted approximately 20 min.

Results

Data analysis

For the analysis of the eye movement data, the experimental sentences were divided into the following regions (marked with angle brackets in the example below and in italics in the following text): Die Student-en/-innen, <die die> <Fahrradfahrerin> <übersehen> <hat/haben,> <sind verletzt> “The students_{Neutral+masculine/feminine}, whom the cyclist_{Neutral+feminine} has overlooked / who have overlooked the cyclist_{Neutral+feminine}, are hurt.” We refer to the analyzed regions of the relative clause as relative pronoun (the relative pronoun with the following determiner), RN2, action verb, and auxiliary verb.Footnote ⁵

For each region, five reading time measures were computed: first fixation duration (the duration of the very first eye fixation on a region entered from the left), first-pass reading time (the sum of fixation times from first entering a region from the left until leaving it either to the right or to the left), regression path (the sum of fixation times from first entering a region from the left until leaving it to the right, including the time spent regressing to the left of the region), total fixation time (the sum of all fixation times on a region), and regressions into a region (the number of times the region was entered from the right; see Staub & Rayner, Reference Staub, Rayner and Gaskell2007).

The initial stages of data analysis consisted of merging fixations shorter than 70 ms with neighboring fixations within one character and removing fixations below 70 ms and above 600 ms (2.13% of the data), for previous research on reading had shown that such fixations are not representative of normal acquisition of information (Breen & Clifton, Reference Breen and Clifton2011; Rayner, Sereno, Morris, Schmauder, & Clifton, Reference Rayner, Sereno, Morris, Schmauder and Clifton1989). Trials that exceeded the reading time range of the total reading time mean plus 3 SD were considered outliers (1.30% of all trials) and were excluded from the analyses.Footnote ⁶ The data were subjected to repeated measures analyses of variance (ANOVAs) with the RN1 grammatical gender (masculine vs. feminine) and the relative clause type (SRC vs. ORC) treated as within-subjects and within-items factors. Computations based on the data averaged across participants and across items are referred to as F ₁ and F ₂ analyses, respectively. The analyses were based on residual fixation times after region-length correction (Trueswell, Tanenhaus, & Garnsey, Reference Trueswell, Tanenhaus and Garnsey1994).Footnote ⁷ Table 1 provides means and standard deviations of residual fixation times for all measures and regions and probabilities of regressions, and Table 2 presents the results of ANOVAs.

Table 1.

Means (standard deviations) of residual fixation times and probabilities of regressions for Experiment 1

Note: RC, relative clause; FF, first fixation durations; FP, first-pass reading time; RP, regression path; TT, total fixation time; RI, regressions into the region; ORC, object-extracted relative clauses; SRC, subject-extracted relative clauses.

Table 2.

Results of analyses of variance for all regions of interest for Experiment 1

Note: FF, first fixation durations; GG, RN1 grammatical gender; RC, relative clause type; FP, first-pass reading time; RP, regression path; TT, total fixation time; RI, regressions into the region.

†p ≤ .1. *p ≤ .05. **p ≤ .01.

Pairwise contrast analyses were performed based on F ₁ only in cases when patterns of mean differences were similar and significant in either both F ₁ and F ₂, or in one (p ≤ .05) and marginally significant (p ≤ .1) in the other analysis. Corresponding t test results are reported and interpreted below region by region (left to right), starting from measures reflecting earlier processing to those reflecting later processing.Footnote ⁸ Note that it is the interaction term between the RN1 grammatical gender and the relative clause type that is of critical interest to our Hypothesis 1.

First fixation durations

The ANOVA revealed a main effect of the RN1 grammatical gender on the RN2 region with shorter fixations after masculine than feminine RN1, M _masc = −11.40; M _fem = −1.66, t (31) = −2.05, SEM = 4.76, p = .049, r = .35.

Regression path

A main effect of grammatical gender emerged on the action verb with shorter fixations after masculine than feminine RN1, M _masc = −475.30; M _fem = −416.95, t (31) = −2.15, SEM = 27.18, p = .040, r = .36. The ANOVA also revealed an interaction between the RN1 grammatical gender and the relative clause type on the auxiliary verb. A t test showed shorter fixations of the auxiliary verb in SRCs after masculine compared to feminine RN1 agents, M _mascSRC = −198.99; M_femSRC = −104.80, t (31) = −3.00, SEM = 31.45, p = .005, r = .47 (see Figure 1).

Figure 1.

Regression path residuals on the auxiliary verb region after grammatically masculine and feminine plural role noun agents in subject-extracted relative clauses. Error bars indicate standard errors (Experiment 1).

Total fixation time

A main effect of grammatical gender occurred on the RN2 showing shorter fixations after masculine than feminine RN1, M _masc = 22.23; M _fem = 104.34, t (31) = −2.22, SEM = 36.96, p = .034, r = .37.

Regression into a region

A main effect of grammatical gender showed a tendency for fewer regressions into the RN2 after masculine than feminine RN1, M _masc = 0.60; M _fem = 0.72, t (31) = −1.88, SEM = 0.06, p = .070, r = .32.

Response accuracy

The average accuracy in answering comprehension questions during the experiment was 81.25%.

Discussion

As expected, the grammatical gender of RN1 affected the resolution of temporarily ambiguous relative clauses in Experiment 1, which was represented by the interaction between the grammatical gender and the relative clause type. In SRCs, feminine RN1 agents caused more difficulties in processing than masculine. This finding suggests that grammatical gender may function as a cue to agency in that masculine role nouns are more expected to serve as agents in relative clauses than feminine role nouns. However, the extent to which this result can be generalized remains limited at this point since no such pattern was observed in ORCs and, despite a reliable effect in regression path residuals, the effect in SRCs did not reach significance across measures. One plausible reason for why the pattern observed in SRCs was absent in ORCs may be the greater difficulty of ORCs compared to SRCs requiring higher processing costs that could have washed out rather subtle effects of gender. In addition, it should be noted that the agent role is more central for comprehension than the patient role (e.g., Cohn & Paczynski, Reference Cohn and Paczynski2013), which could be another explanation of the effect of grammatical gender, which was observed for agents (in SRCs) and not for patients (in ORCs). These possibilities will be considered in more detail in the General Discussion.

As to the main effect of grammatical gender, most of the examined measures reliably showed that feminine RN1 caused more difficulties in processing compared to masculine RN1. Since agents are more likely to precede patients (e.g., Bornkessel, Zysset, Friederici, Von Cramon, & Schlesewsky, 2005), this may indicate a general tendency in readers to expect masculine rather than feminine agents to be mentioned in a sentence first.

EXPERIMENT 2

In Experiment 2 we extended the research question to the stereotypical gender of role nouns. Based on our previous reasoning, not only grammatical but also stereotypical gender of role nouns should contribute to thematic role assignment as gender typicality has repeatedly been shown to prompt the representation of gender congruent referents. Theoretically this can be explained by the situation model approach (Sanford & Garrod, Reference Sanford and Garrod1981, Reference Sanford and Garrod1998), which claims that updating the representation of an event that is being described from long-term memory is a constant process running in parallel to the unfolding of linguistic input. Stereotypical gender information represented linguistically through role nouns likely affects the upgrading process, in which it is mapped onto the world knowledge about typical gender roles. Research on gender processing has shown that both stereotypical and grammatical gender information is relevant for the interpretation of sentences, which manifests itself through mismatch effects when gender cues are incongruent (e.g., Carreiras et al., Reference Carreiras, Garnham, Oakhill and Cain1996; Esaulova et al., Reference Esaulova, Reali and von Stockhausen2014). Furthermore, gender effects are even stronger when gender cues are made more salient such as through gender markings and/or through varying both grammatical and stereotypical gender (e.g., Irmen, Reference Irmen2007). Making gender information more salient through the use of grammatical and stereotypical gender in Experiment 2 should result in both functioning as cues to agency, as predicted in Hypotheses 1 and 2.

Therefore, in Experiment 2 we examined to which extent both grammatical gender and stereotypical gender function as cues to agency and affect the resolution of temporarily ambiguous relative clauses. For this purpose, we varied the stereotypical gender of RN1 and the grammatical gender of RN2.