Knjižnica Filozofskog fakulteta
Sveučilišta u Zagrebu
Faculty of Humanities and Social Sciences Institutional Repository

Fine-grained human evaluation of an English to Croatian hybrid machine translation system

Downloads

Downloads per month over past year

Klubička, Filip. (2017). Fine-grained human evaluation of an English to Croatian hybrid machine translation system. Diploma Thesis. Filozofski fakultet u Zagrebu, Department of English Language and Literature
Department of Information Science. [mentor Stanojević, Mateusz-Milan and Ljubešić, Nikola].

[img]
Preview
PDF (English)
Download (2MB) | Preview

Abstract

This research compares two approaches to statistical machine translation - pure phrasebased and factored phrase-based - by performing a fine-grained manual evaluation via error annotation of the systems’ outputs, and subsequently analysing the results. The error types considered in the annotation are compliant with the multidimensional quality metrics (MQM), and the annotation is performed by two annotators. Inter-annotator agreement is relatively high for such a task, and the results show that the factored system, i.e. hybrid system, performs much better, reducing the amount of errors produced by the pure phrase-based system by 20%.

Item Type: Diploma Thesis
Uncontrolled Keywords: machine translation, human evaluation, PBMT, factored PBMT, MQM, inter-annotator agreement, statistical machine translation, error annotation
Subjects: English language and literature
Information sciences
Departments: Department of English Language and Literature
Department of Information Science
Supervisor: Stanojević, Mateusz-Milan and Ljubešić, Nikola
Date Deposited: 20 Sep 2017 12:49
Last Modified: 20 Sep 2017 12:49
URI: http://darhiv.ffzg.unizg.hr/id/eprint/9032

Actions (login required)

View Item View Item