URI.extract is too strong, not limited to URLs, matched real text. Same issue was present in language detector.