I am trying to use Stanza for Arabic PoS tagging. After analyzing the output it seems that both are giving different results.
It seems that CoreNLP is more accuracte than stanza model for Arabic.
Can anyone help why this is the case as both are managed by Stanford. Is there any way I can use CoreNLP server models in stanza by using stanza.Pipeline()
CoreNLP and Stanza are essentially separate tools, despite both being written at Stanford. Stanza is principally built around Universal Dependencies data and neural sequence model approaches, while CoreNLP is principally built around older (including LDC) data sources and probabilistic NLP approaches. For Arabic in particular, CoreNLP was worked on by people with good knowledge of Arabic. For different languages and situations, the output of CoreNLP and Stanza may be quite different, and one or the other may be considerably better. It's good to hear people's experiences.
