| This thesis presents two connectionist models, which can learn the thematic roles of words in sentences by receiving aspects of real world situations to which the sentences are referring, and exhibit strong systematicity without prior syntactic knowledge. The models are intended towards cognitively and biologically plausible connectionist models. Current models could be parts of the larger network to represent the meaning of a whole sentence. The first model, closest, of the two models, to being purely connectionist, attains an acceptable result (98.31% of the roles correctly identified). The second one, not purely connectionist, achieves a perfect result. It could be argued that humans learn the thematic roles, as an emergent property of learning the relationship between the words/sentences and the real world situations. However, it is not claimed that the models are the human learning mechanism for language acquisition.;Keywords: Computational Cognitive Science; Connectionism; Systematicity; Thematic Roles; Neural Networks; Semantic Role Labeling. |
