A Pre-trained Transformer and CNN Model with Joint Language ID and Part-of-Speech Tagging for Code-Mixed Social-Media Text