Machine potentials in translation and post-editing: Morphosyntactic inadequacies in transference of children’s literature to morphologically rich languages

The aim of the present thesis was the study of morphosyntactic inadequacies in machine translation of literary texts when the target language is morphologically richer than the source language. Simple vocabulary and structures along with the lack of difficult to decode literary devices in children’s literature allowed for a scrutiny of morphosyntactic issues ignoring the potential effect of semantic or pragmatic choices. A rule-based translation system, SYSTRAN, and a neural one, DeepL, were chosen to compare performances. There are two levels on this evaluation—the human evaluating the machine and the machine evaluating the machine. ChatGPT 4.0 was used to assess the content of specific excerpts using the same criteria as in the human evaluation. The parameters that were examined touched upon issues, such as word order, constructions, agreement and inflection. Although not explicitly assigned as criteria, comments on the impact of morphosyntactic choices on the literariness of the text were expected to be made. The translation tends to occur on a local level often neglecting intra-textual cohesion and coherence. Inflection is overall unproblematic, yet issues of agreement are frequent, while the connection between elements (subject-verb, adjective-noun etc.) might be lost even intra-sententially. It is hard for these systems to cope with creativity, and compounds, creative words or diminutives often appear awkwardly translated, transliterated or untranslated, choices that can be disrupting to the reader. Neural-based systems outperformed rule-based ones in terms of natural language use, although they were not flawless. The Artificial Intelligence evaluator seemed to be aware of the ineffective choices machine translation systems made, although not always or to the extent the human agent was. There are also hints for machine post-editing of machine-translation.