customising dictionary for identifying proper nouns

[rmrbytes]

When using this wonderful library to identify pronouns to names, it works well for popular names like John and Jane. However, it does not recognise many other names (perhaps due to the dictionary having US names). For example:

Given text: John and Jane are neighbours. She admires him because he works hard.

Pronouns and their references:
She => Jane
him => John
he => John

But

Given text: Ram and Sita are neighbours. She admires him because he works hard.

Pronouns and their references:
him => Ram and Sita
he => Ram and Sita

I tried adding

add_dict={"Sita": ["woman", "girl"], "Ram": ["man", "boy"]}
coref = neuralcoref.NeuralCoref(nlp.vocab, conv_dict=add_dict)

yet

Given text: Ram and Sita are neighbours. She admires him because he works hard.

Pronouns and their references:
She => Ram and Sita
him => Ram and Sita
he => Ram and Sita

Is there any reason it is not accepting the added words? Thanks

[munaAchyuta]

yes, i'm too facing same problem. this is a common issue for new domain . what should be the solution ?
don't see training from beginning with new domain data(small size) is feasible. is there transfer learning possible ?

[svlandeg]

Thanks for the report! Merging with Issue #215.