Machine Translation and Artificial Intelligence Applied to the Translation of Puns and Wordplay in "Alice's adventures in Wonderland"

The present work focused on the use of Machine Translation and Artificial Intelligence applied to the translation of puns, wordplay and other humorous content in literary texts. These textual instances tend to pose several problems to translators; however, technological tools such as Machine Translation and Artificial Intelligence have been proved to be helpful to translators when it comes to tackle these elements. Even though these tools can be helpful, their performance could be improved. The present study tried to identify methods directly applicable by translators in order to obtain better translations when using the NMT providers ModernMT and RWS Language Weaver, and when using the general Artificial Intelligence interfaces ChatGPT 3.5 and Microsoft Edge Copilot. The present dissertation is introduced by a detailed literature review focusing in the history and development of Machine Translation and Artificial Intelligence. Moreover, puns and wordplay are addressed in order to provide a definition of this linguistic phenomenon and in order to present to the readers some possible translation strategies that can be adopted when tackling these textual instance. The literature review also focuses on studies already conducted on MT and AI applied to the translation of puns and wordplay (an humorous language in general). Then, the actual study is described. In order to carry ou the research, five instances of humorous language were extracted from the novel “Alice’s Adventures in Wonderland”. These textual instances were translated using the above-mentioned tool; after that, the translations obtained were evaluated using both a manual evaluation system, and two automatic evaluation metrics (BLEU and TER). Once the scores were assigned, the same instances were translated a second time using the same tools. This time, improvement strategies were applied: NMT providers were trained for the translation of puns and wordplay using TMs specifically created for this purpose, and AI interfaces were fed prompts suitable for the type of translation result needed. After the second translation, the results were evaluated a second time and compared to the results obtained after the first translation, in order to control whether or not the improvement strategies worked. The results of the study showed that while prompting engineering proved to be useful, the same cannot be said for the training of NMT providers. The instances of puns and wordplay were not translated in a satisfactory way even after the training. However, some improvements as far as syntax was concerned could be observed, especially in relation to RWS Language Weaver. Generative AI and Neural Machine Translation could sure be of great help to translators when it comes to the translation of puns and wordplay. However, some work on the part of the translator is required. For example the translator needs to feed the AI provider with suitable prompts in order to obtain a satisfactory result.