It has to do with the way text is tokenized into discrete groups and not individual characters. One way to try overcoming this problem is to ask the model to br

It has to do with the way text is tokenized into discrete groups and not individual characters. One way to try overcoming this problem is to ask the model to break each word into individual spaced characters as an intermediary step.