It's very common to sort on an analyzed field (typically, we put the text all lowercase, remove accents, remove special characters and so on).

The issue is more the tokenizer. We should check if the field is using a tokenizer different from the KeywordTokenizer.