A influência de informações temáticas na previsão de argumentos verbais

Abstract

This study aims at investigating whether the thematic properties of the verbs can help predicting the semantic features of names which frequently occupy the position of internal argument during language processing. Prediction during language comprehension can be defined as the use of high-level internal representations - the speaker's real-world knowledge interacting with information coming from sentence/discourse contexts - to preactivate lower levels of linguistic representations (e.g. phonological, semantic, syntactic) during the processing of an input. Kuperberg and Jaeger (2016), the work on which this dissertation is based, posits that language prediction is a gradual and probabilistic phenomenon. According to this perspective, the interaction between context and finegrained knowledge about events may facilitate the comprehension of words related to the structure of the activated event. However, little is known about the contextual strength necessary for this interaction to occur. We intend to investigate whether the verbs with a well-defined thematic role (according to MCRAE; FERRETTI; AMYOTE, 1997) have enough strength to pre-activate more refined semantic representations in fairly simple context formed only by a proper noun and a verb (e.g. spilled - milk). Our aim is to verify if such preactivation of semantic information facilitates the processing of the internal argument. Furthermore, we intend to observe if there are any differences between predictive behavior of verbs with well-defined thematic roles and verbs with poorly defined thematic roles. In order to understand such questions, we initially built a corpus containing data from cloze tests. This corpus allowed for an observation of the predictive behavior of 520 verbs collected from the judgements of at least 50 participants. Subsequently, we conducted a thematic role assignment task (FERRETTI; MCRAE; HATHERELL, 2001; MCRAE; FERRETTI; AMYOTE, 1997) in which we manipulated the subselection strength of selected verbs and the previsibility of the internal arguments. We aim at observing whether the ability of subselection of arguments caused divergent thematic behavior between two groups of verbs. The analysis of judgments suggests that verbs with greater subselection strength have well-defined thematic roles. When compared to verbs with lower subselection strength, the grades attributed to the events expressed by the verbs with greater subselection strength are higher when the argument is more predictable and lower when the argument is implausible. Finally, aiming to verify the behavior of the verbs during online processing in simple contexts, we have run two lexical decision tests. We have verified whether the thematic characteristics of the verbs could help the most expected arguments prediction (e.g., spill milk) in comparison to implausible internal arguments (e.g., spill house). The tests have also been applied with the objective of investigating whether there were any correspondences between results obtained from both the thematic role assignment test and the lexical decision test. Evidence was found shows that the thematic information from strong subselection verbs could contribute to verbal arguments’ semantic features prediction. For the most part, the results correlate with the evidence obtained from thematic role assignment test.