Please use this identifier to cite or link to this item:
Probabilistic and Predictive Parsing in Tagalog Voice Alternations
|Title:||Probabilistic and Predictive Parsing in Tagalog Voice Alternations|
|Authors:||Bondoc, Ivan Paul|
|Contributors:||Schafer, Amy J (advisor)|
show 4 moreintegration
|Publisher:||University of Hawai'i at Manoa|
|Abstract:||People tend to predict upcoming elements in language as speech unfolds. Prediction, or the generation of expectations about upcoming input, is argued to be the unifying principle of the human information processing system. Despite the lack of consensus among scholars on its definition and mechanisms, the majority of the extensive work on prediction has focused on generating expectations for a specific lexical item and its lower-level features, giving less attention to the ability of the parser to generate gradient expectations beyond the upcoming word, such as the unfolding linear word order of a sentence. |
This dissertation extends our knowledge on prediction by investigating how an understudied type of linguistic information, called voice morphology, is used to develop probabilistic expectations about a sentence’s particular word order. Five experiments examine whether comprehenders use verb and voice morphology early in the sentence to anticipate its likely arguments and predict the likely order of the phrases that follow. I test the hypothesis that Tagalog comprehenders use voice morphology to develop probabilistic syntactic expectations to predict specific sentential word order patterns.
Experiments 1 and 2 investigated the role of agentivity and pivothood in Tagalog word order preferences and on the anticipation of the verb’s likely arguments. A sentence continuation task (Experiment 1) demonstrated the role of two probabilistic and equally-weighted constraints, namely agent-first and pivot-second, in shaping word order preferences across voices, and highlighted the strong link between voice and word order. However, despite this link, comprehenders were not observed to anticipate the agent and the pivot as the likely arguments of the verb in a visual world eyetracking study (Experiment 2); instead, gaze patterns illustrated preferential looks to animates and undifferentiated looks across voices. I speculated these results were a product of comprehenders’ engagement in a wide range of predictions across voices.
Three reading experiments examined whether voice morphology is used by comprehenders to predict the specific linear word order pattern of the sentence, following a gradient cline of patterns. In a self-paced reading study (Experiment 3), comprehenders showed limited predictive effects of voice, as it was only at a later sentence region that they showed the hypothesized gradience of patterns. These limited effects were subsequently verified in two experiments with induced cognitive load. When faced with time pressure, predicted gradience diminished, and comprehenders only generated coarse-grained predictions of the unfolding sentence in a rapid serial visual presentation task (Experiment 4). Correspondingly, limited effects of generating gradient expectations of linear word order were demonstrated in a self-paced reading task with induced memory load (Experiment 5), replicating results from Experiment 3.
In contrast to the major claim that prediction is immediate and ubiquitous (e.g., DeLong et al., 2014; Kuperberg & Jaeger, 2016), these findings reveal the limits of prediction and the variability of prediction in sentence processing. I argue that these results are best captured by a dynamic account that describes the trajectory of probabilistic activation for the unfolding linguistic structures in the course of processing as shaped by gradient prediction and integration.
|Rights:||All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.|
|Appears in Collections:||
Ph.D. - Linguistics|
Please email email@example.com if you need this content in ADA-compliant format.
Items in ScholarSpace are protected by copyright, with all rights reserved, unless otherwise indicated.