Combination of the ranking method and time reversal asymmetry for evaluating the quality of translations: a case study on passages from works by Edgar Allan Poe and Kenzaburo Oe

The complex systems approach to natural language has become a fascinating topic that is closely related to physics and linguistics. We assess the quality of translations of English literary passages into Japanese and perform a comparative study on several translations. First, translated writings are expressed with 45 syllabics, and subsequently, the ranking of their frequencies is analyzed statistically with the combined use of a nonlinear regression and the Durbin–Watson ratio, which allows one to detect the serial correlation in a sequence. To examine the correlations of syllabic sequences, regressions are made on both a long-tailed (LT) and a short-tailed function. The validity of our method for revealing correlations in the Markovian sequences of Japanese syllabics is confirmed by comparing the results of two original examples: popularity ranking of boy names and passages from a Japanese novel. Subsequently, the method is applied, for the first time to our knowledge, to sixteen translations of passages from Edgar Allan Poe (1809–49), and five backtranslations of those from Kenzaburo Oe (1935–2023), a Japanese Nobel laureate for literature in 1994. A diachronic analysis of computed results for the latter translations by three humans and two artificial intelligences (AIs) shows that while the quality of the translation by Google Translate (as of April 2020) falls far short of the human translator next in rank, the updated version (as of October 2024) is comparable to the top, but it still might be too soon to speculate on its unconditional potentiality. Both our methodology and results are not only novel but are also expected to make a contribution toward an interdisciplinary study between physics and linguistics.