Abstract: We train language models to automate the diagnosis of OTN configuration errors, and the diagnostic accuracy is up to 97.56%. We additionally demonstrate the effectiveness of the models on a ...