I am working on extracting people and tasks from texts (multiple sentences) and need a way to resolve coreferencing. I found this model, and it seems very promising, but once I installed the required libraries allennlp and allennlp_models and testing the model out for myself I got:
Script:
predictor = Predictor.from_path("https://storage.googleapis.com/allennlp-public-models/coref-spanbert-large-2021.03.10.tar.gz")
prediction = predictor.predict(
document="Paul Allen was born on January 21, 1953, in Seattle, Washington, to Kenneth Sam Allen and Edna Faye Allen. Allen attended Lakeside School, a private school in Seattle, where he befriended Bill Gates, two years younger, with whom he shared an enthusiasm for computers.")
print(prediction)
Output:
{'top_spans': [[0, 1], [3, 3], [5, 8], [5, 14], [8, 8], [11, 13], [11, 14], [13, 13], [16, 18], [16, 22], [20, 22], [24, 24], [26, 52], [33, 33], [36, 36], [37, 37], [38, 52], [41, 42], [47, 47], [48, 48], [49, 52]],
'antecedent_indices': [[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20], [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20]],
'predicted_antecedents': [-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 0, -1, 5, 11, -1, -1, -1, 11, -1, -1],
'document': ['Paul', 'Allen', 'was', 'born', 'on', 'January', '21', ',', '1953', ',', 'in', 'Seattle', ',', 'Washington', ',', 'to', 'Kenneth', 'Sam', 'Allen', 'and', 'Edna', 'Faye', 'Allen', '.', 'Allen', 'attended', 'Lakeside', 'School', ',', 'a', 'private', 'school', 'in', 'Seattle', ',', 'where', 'he', 'befriended', 'Bill', 'Gates', ',', 'two', 'years', 'younger', ',', 'with', 'whom', 'he', 'shared', 'an', 'enthusiasm', 'for', 'computers', '.'],
'clusters': [[[0, 1], [24, 24], [36, 36], [47, 47]], [[11, 13], [33, 33]]]}
I'm having trouble interpreting the format of this output. I was expecting something like
{entity_0_spans: [LIST_OF_INDEX_TUPLES], # Paul Allen in this example
entity_1_spans: [LIST_OF_INDEX_TUPLES], # Seattle in this example
...}
or something that more closely resembles the visualisation available on the demo page:
I've looked through https://demo.allennlp.org/coreference-resolution but couldn't find a breakdown of how to use the model output yet - can anyone suggest some resources that will help me? Any pointers are much appreciated!
