Quantity and Quality: Not a Zero-Sum Game

Quantification of existing theories is a great challenge but also a great chance for the study of language in the brain. While quantification is necessary for the development of precise theories, it demands new methods and new perspectives. In light of this, four complementary methods were introduce...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
1. Verfasser: Alday, Phillip Marshel
Beteiligte: Bornkessel-Schlesewsky, Ina (Prof. Dr.) (BetreuerIn (Doktorarbeit))
Format: Dissertation
Sprache:Englisch
Veröffentlicht: Philipps-Universität Marburg 2015
Schlagworte:
Online Zugang:PDF-Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Quantification of existing theories is a great challenge but also a great chance for the study of language in the brain. While quantification is necessary for the development of precise theories, it demands new methods and new perspectives. In light of this, four complementary methods were introduced to provide a quantitative and computational account of the extended Argument Dependency Model from Bornkessel-Schlesewsky and Schlesewsky. First, a computational model of human language comprehension was introduced on the basis of dependency parsing. This model provided an initial comparison of two potential mechanisms for human language processing, the traditional "subject" strategy, based on grammatical relations, and the "actor" strategy based on prominence and adopted from the eADM. Initial results showed an advantage for the traditional subject" model in a restricted context; however, the "actor" model demonstrated behavior in a test run that was more similar to human behavior than that of the "subject" model. Next, a computational-quantitative implementation of the "actor" strategy as weighted feature comparison between memory units was used to compare it to other memory-based models from the literature on the basis of EEG data. The "actor" strategy clearly provided the best model, showing a better global fit as well as better match in all details. Building upon the success modeling EEG data, the feasibility of estimating free parameters from empirical data was demonstrated. Both the procedure for doing so and the necessary software were introduced and applied at the level of individual participants. Using empirically estimated parameters, the models from the previous EEG experiment were calculated again and yielded similar results, thus reinforcing the previous work. In a final experiment, the feasibility of analyzing EEG data from a naturalistic auditory stimulus was demonstrated, which conventional wisdom says is not possible. The analysis suggested a new perspective on the nature of event-related potentials (ERPs), which does not contradict existing theory yet nonetheless goes against previous intuition. Using this new perspective as a basis, a preliminary attempt at a parsimonious neurocomputational theory of cognitive ERP components was developed.
DOI:10.17192/z2015.0221