A computational study on accusativity and ergativity

Demirel, Seda
English is defined as an accusative language with Subject-Verb-Object (SVO) syntactic order. In this study, if children were exposed to hypothetical English, i.e. ergative English, rather than accusative English in the language acquisition process, what would happen was investigated by using a child-directed speech data set taken from the Eve fragment (Brown, 1973) of the Child Language Data Exchange System (CHILDES) database (MacWhinney, 2000). This process was modelled computationally by a training model. Based on the data set, the standard English grammar was constructed with the syntactic and semantic representations of the words. According to this grammar, correct pairs of sentences and their corresponding logical forms were generated. Subsequently, several models were developed to derive accusative sentences from the grammar. After training, the best model that prioritizes the correct pairs of sentences in the derivation results was obtained. Three experiments were conducted with this model: one exclusively employed accusative grammar, another used accusative grammar and ergative forms of transitive verbs, and the last focused only on syntactically ergative grammar. In these experiments, the trained model corresponded to the child acquiring the language, and the rank success represented whether the child successfully acquired the target language. The results of these experiments and the rank success of the model demonstrated that children can be assumed to have acquired accusative English when they are exposed to accusative English, and they can be assumed to have captured ergative English when they are exposed to hypothetical ergative English. These results indicated that each grammatical relation (accusative or ergative) system is equally likely for children in language acquisition, and the exposure to particular linguistic experiences decides which system takes precedence and which falls behind.
Citation Formats
S. Demirel, “A computational study on accusativity and ergativity,” M.S. - Master of Science, Middle East Technical University, 2024.