如何解释 Allen NLP 共指解析模型输出？

Question

我正在致力于从文本（多个句子）中提取人物和任务，并且需要一种方法来解决指代问题。我发现 this model，它看起来很有前途，但是一旦我安装了所需的库 allennlp 和 allennlp_models 并为自己测试了模型，我得到：

脚本：

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)

输出：

{'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]]]}

我无法解释此输出的格式。我期待

{entity_0_spans: [LIST_OF_INDEX_TUPLES],  # Paul Allen in this example
 entity_1_spans: [LIST_OF_INDEX_TUPLES],  # Seattle in this example
 ...}

或更类似于演示页面上提供的可视化效果的内容：

我已经浏览 https://demo.allennlp.org/coreference-resolution 但还没有找到关于如何使用模型输出的详细信息 - 谁能推荐一些对我有帮助的资源？非常感谢任何指点！

Answer 1

您要查找的信息在'clusters'中，其中每个列表对应一个实体。在每个实体列表中，您会发现提及的是同一实体。数字是标记每个指代提及的开始和结束的索引。例如。保罗艾伦 [0,1] 和艾伦 [24, 24].

如何解释 Allen NLP 共指解析模型输出？

How to interpret Allen NLP Coreference resolution model output?

python

allennlp

coreference-resolution