Revisiting zero-shot cross-lingual topic identification: baselines, languages and evaluation