Scott Randal

The original report was just from trying every hour in isolation, and noticing that those three behaved differently from the other hours. The difference does persist in some contexts though....

SimpleConsole shows the same inconsistency for those sentences: Je dois être de retour au bureau à seize heures { "Text": "seize heures", "Start": 35, "End": 46, "TypeName": "datetimeV2.duration", "Resolution": {...

That's how Word Ninja treats words that aren't found in the language model. Wikipedia has articles about Patreon and Zelle, but it looks like those were not in the subset...

Originally the English dict had the same issue with strings that contained digits: If the input token contained a digit, all characters to the right of the digit were split...

For English possessive forms to be split correctly, the word list must include this entry: 's After it becomes a separate token, a post-processing step reattaches it to the preceding...