Miriam Ellert

The pdf-file contains the full linear mixed-effect models which were calculated on the visual-world eye-tracking data of the thesis entitled "Ambiguous Pronoun Resolution in L1 and L2 German and Dutch" (chapters 3 and 4). The studies of this thesis investigated how German and Dutch personal (German: "er", Dutch "hij") and d-pronouns (German: "der", Dutch "die") are resolved when two potential antecedents are available. This question was investigated in L1 listeners of German and Dutch (chapter 3) and Dutch L2 learners of German and German L2 learners of Dutch (chapter 4). In both groups, two experiments were conducted manipulating the sentence structure of the antecedent clause. While experiment 1 (subchapter 1) presented both pronouns after canonical antecedent structures ("The doctor is friendlier than the cook. He (p/d)..."), in experiment 2 (subchapter 2) the pronouns followed non-canonical antecedent structures ("Friendlier than the cook is the doctor. He (p/d)..."). Additional analyses were conducted on the data of experiment 1, including animacy (subchapter 3) as an additional factor in the model (12 items presented two animate NPs, 12 items two inanimate NPs), or evaluating whether inter-individual differences across participants (subchapter 4) had an effect on the on-line eye-tracking data (L1 data: antecedent choice in a subsequent off-line questionnaire/ L2 data: proficiency in the target language). The mixed models contained participants and items as a crossed-random factor. The analyses were conducted on ten different time windows ranging from 0-2000 ms with the pronoun onset at 0ms. For the data presented in chapters 3.1, 3.2., 4.1. and 4.2. of the thesis, two linear mixed models were calculated for each time window. In a first model, pronoun type (called condition in the analysis with 2 levels: personal vs. d-pronoun) and order of mention (called mention in the analysis with 2 levels: 1st vs. 2nd) were entered as the predictors of interest (simple model), and in a second model, their interaction was added (interaction model). The d-pronoun condition and the 1st-mentioned level were mapped onto the intercept. Thus, a positive beta coefficient for condition indicated more looks to the personal pronoun. A positive beta coefficient for mention indicated more looks to the second-mentioned entity. Formulas in R for both models: Simple model = lmer(looks ~ condition + mention + (1 | pp) + (1 | item)) Interaction model = lmer(looks~ condition*mention+ (1 | pp) + (1 | item)) A stepwise approach to model building was used, applying a forward elimination method, which ensures that only the minimum number of predictors accounting for the variance in the data enter the model. The simple model containing both main effects of order of mention and type of pronoun was chosen as the most basic model, since we were interested in the estimates for order of mention and the two types of pronouns. The second model, the interaction model, investigated whether the looks to both types of target pictures were affected differently by the two pronouns. Thus, for the analyses, first both the simple model and the interaction model were calculated for a specific time window, and subsequently a loglikelihood test was calculated on the fit of the models. The interaction model was only chosen when it significantly better explained the variance in the data. This procedure was repeated for the other time windows. This pdf-file, presents - the loglikelihood analyses - the fixed effects of the mixed models - and if there was a significant interaction, the individual analyses for each type of pronoun. The additional analyses in chapters 3.3., 3.4., 4.3., and 4.4. tested whether adding an additional predictor to the models (data of experiment 1) would significantly better predict the data than the simple/interaction model. - animacy model: the three-way interaction term between order of mention, type of pronoun and animacy was added (intercept = condition: d-pronoun, mention: 1st, animacy: animate) - off-line model: the three-way interaction term, namely condition x mention x off-line was added (intercept = condition: d-pronoun, mention: 1st, off-line: 1st) - proficiency model: the three-way interaction term between order of mention, type of pronoun and proficiency was added (intercept = condition: d-pronoun, mention: 1st, proficiency: B-level) In case the interactions were significant, individual analyses were conducted.


[author(s)] (2010). Collection "Miriam Ellert". The Language Archive. https://hdl.handle.net/1839/00-0000-0000-0011-EBA2-C. (Accessed 2024-05-19)

Note: This citation was extracted automatically from the available metadata and may contain inaccuracies. In case of multiple authors, the ordering is arbitrary. Please contact the archive staff in case you need help on how to cite this resource. Author information could not be extracted automatically for this resource.