Procesamiento de frases e información léxica

  1. Demestre, Josep
Supervised by:
  1. José Eugenio García-Albea Ristol Director

Defence university: Universitat Rovira i Virgili

Fecha de defensa: 17 May 2004

Committee:
  1. Manuel Francisco Carreiras Valiña Chair
  2. Rosa Maria Sánchez-Casas Padilla Secretary
  3. Elizabeth Gilboy Committee member
  4. José Manuel Igoa González Committee member
  5. Josep Maria Sopena Sisquella Committee member

Type: Thesis

Teseo: 106792 DIALNET lock_openTDX editor

Abstract

The main aim of this thesis was to examine the influence verb-specific information has on sentence processing. We studied two particular sources of information that are stored at a verbs lexical entry: (i) the control properties of a verb that determine the assignment of an antecedent to the null subject of an infinitival complement, and (ii) the information that specifies that a verb subcategorizes for a tensed complement in a particular verbal mood. We have tested the predictions of the garden-path model (Frazier, 1987) and the predictions of constraint-based lexicalist models (MacDonald, et al., 1994a, b; Trueswell & Tanenhaus, 1994). The central question we wanted to address is whether the sentence processor initially ignores verb-specific information as the garden-path model assumes or whether such lexical information has a rapid influence on the early stages of parsing as assumed by constraint-based lexicalist parsing models. We run nine experiments a cross modal naming experiment, two ERPs experiments and six self-paced reading experiments to examine the role of control and mood information on the early stages of parsing. There are two novel aspects in this thesis that we would like to highlight. On the one hand, the relation between verb control information and on-line sentence comprehension had received little attention in Spanish. Moreover, in the six experiments designed to examine the influence of this type of information on sentence processing we manipulated the agreement patterns between the null subject of the infinitive and an adjective predicated of it. To date, nobody had used such a manipulation to study the assignment of an antecedent to the null subject of an infinitival complement. On the other hand, the influence of mood information on sentence processing had not been studied yet. The results of the experiments show that the sentence processor accesses and uses control- and mood information very rapidly. The experiments that examined the influence of control information have shown that the processor interprets the null subject of an infinitival complement very rapidly. Moreover, we have shown that the processor, rather than initially associating the null subject with the most recent filler as predicted by the garden-path model rapidly uses control information to assign an antecedent to this null subject as predicted by constraint-based lexicalist models. The ERP experiments, in addition to show the early influence of control information on parsing, have provided further evidence about the brain responses to syntactic anomalies in Spanish. We examined a type of agreement, that between an NP and an adjective predicated of it, that had not been previously studied. The waves elicited by agreement violations in Spanish have been shown to be similar to those reported in previous work. Furthermore, we have contributed to the demonstration of the feasibility of doing ERP studies using continuous, natural speech as the stimulus materials. The experiments that examined the role of mood information on parsing have shown that this particular type of lexical information is rapidly used by the sentence processor. Moreover, we have shown that this type of information is made available as soon as the system recognizes a verb in the input string. In the light of the results, we conclude that, in contrast to the predictions of the garden-path model, the information stored at a verbs lexical entry plays an important role at the early stages of parsing. The early use of such verb-specific information is in accordance with the predictions of constraint-based lexicalist parsing models.