Learning from use: an error-driven approach to Polish aspect

Borowski, Maciej (2023). Learning from use: an error-driven approach to Polish aspect. University of Birmingham. Ph.D.

Preview

Borowski2023PhD.pdf
Text
Available under License All rights reserved.
Download (3MB) | Preview

Abstract

This thesis investigates how speakers learn to use grammatical categories of their native language. While most linguistic categories are useful for descriptive purposes, the questions of their learnability and psychological plausibility are often ignored. Taking Polish grammatical aspect as a case study, we investigate if and how speakers learn to distinguish such general categories as well as master their use based on the usage patterns available in the input.

Grounding our research in usage-based theory of language and the principles of error-driven learning, we conducted a number of computational learning simulations based on a manually annotated corpus sample. These simulations, corroborated with the results of behavioural study show that patterns of usage explain the linguistic behaviour of speakers better than the abstract semantic dimensions that are traditionally used to describe the aspectual classes. We also demonstrate that the patterns of use learned by our models contain enough information to correctly classify verbs into their respective aspectual classes. The results of the studies also indicate an important relationship between tense and aspect and suggest that certain tense-aspect combinations could be considered default. This issue is investigated in more detail, again using a combination of learning simulations and experimental methods.

We argue that the studies presented in this dissertation force us to reflect on the relevance of traditional linguistic distinctions for language cognition and acquisition as well as points us towards a usage-based explanation of the aspectual choice. In addition, we also discuss methodological implications that follow from the work presented here. In particular, we highlight the importance of combining different sources of evidence and the need to corroborate the corpus-based models with behavioural data.

Type of Work:

Thesis (Doctorates > Ph.D.)

Award Type:

Doctorates > Ph.D.

Supervisor(s):