In medical service contexts, especially in migrant- and refugee-receiving countries, the increasing need for inter-lingual communication and the higher cost of human translators has driven the development of machine translation technologies and tools. However, these tools' reliability and efficiency are questioned (Patil et al. 2014; Bouillon et al. 2017), which calls for objective evaluation to ensure and improve the quality of translation results. The present study aims to provide a data-driven empirical evaluation of the linguistic similarity of the French source speech data and their English translations produced by machine translation and identify the main areas in which MT output deviates from natural oral English. Using corpus statistical methods, the evaluation of machine translation outputs can provide concrete and objective feedback for future machine translation improvement. For future research, this model can be trained with larger datasets.