allennlp.data.dataset_readers.coreference_resolution

class allennlp.data.dataset_readers.coreference_resolution.conll.ConllCorefReader(max_span_width: int, token_indexers: typing.Dict[str, allennlp.data.token_indexers.token_indexer.TokenIndexer] = None, lazy: bool = False) → None[source]

Bases: allennlp.data.dataset_readers.dataset_reader.DatasetReader

Reads a single CoNLL-formatted file. This is the same file format as used in the SrlReader, but is preprocessed to dump all documents into a single file per train, dev and test split. See scripts/compile_coref_data.sh for more details of how to pre-process the Ontonotes 5.0 data into the correct format.

Returns a Dataset where the Instances have four fields: text, a TextField containing the full document text, spans, a ListField[SpanField] of inclusive start and end indices for span candidates, and metadata, a MetadataField that stores the instance’s original text. For data with gold cluster labels, we also include the original clusters (a list of list of index pairs) and a SequenceLabelField of cluster ids for every span candidate.

Parameters:
max_span_width: ``int``, required.

The maximum width of candidate spans to consider.

token_indexers : Dict[str, TokenIndexer], optional

This is used to index the words in the document. See TokenIndexer. Default is {"tokens": SingleIdTokenIndexer()}.

classmethod from_params(params: allennlp.common.params.Params) → allennlp.data.dataset_readers.coreference_resolution.conll.ConllCorefReader[source]
text_to_instance(sentences: typing.List[typing.List[str]], gold_clusters: typing.Union[typing.List[typing.List[typing.Tuple[int, int]]], NoneType] = None) → allennlp.data.instance.Instance[source]
Parameters:
sentences : List[List[str]], required.

A list of lists representing the tokenised words and sentences in the document.

gold_clusters : Optional[List[List[Tuple[int, int]]]], optional (default = None)

A list of all clusters in the document, represented as word spans. Each cluster contains some number of spans, which can be nested and overlap, but will never exactly match between clusters.

Returns:
An ``Instance`` containing the following ``Fields``:
text : TextField

The text of the full document.

spans : ListField[SpanField]

A ListField containing the spans represented as SpanFields with respect to the document text.

span_labels : SequenceLabelField, optional
The id of the cluster which each possible span belongs to, or -1 if it does

not belong to a cluster. As these labels have variable length (it depends on how many spans we are considering), we represent this a as a SequenceLabelField with respect to the spans ``ListField.

allennlp.data.dataset_readers.coreference_resolution.conll.canonicalize_clusters(clusters: typing.DefaultDict[int, typing.List[typing.Tuple[int, int]]]) → typing.List[typing.List[typing.Tuple[int, int]]][source]

The CONLL 2012 data includes 2 annotatated spans which are identical, but have different ids. This checks all clusters for spans which are identical, and if it finds any, merges the clusters containing the identical spans.

class allennlp.data.dataset_readers.coreference_resolution.winobias.WinobiasReader(max_span_width: int, token_indexers: typing.Dict[str, allennlp.data.token_indexers.token_indexer.TokenIndexer] = None, lazy: bool = False) → None[source]

Bases: allennlp.data.dataset_readers.dataset_reader.DatasetReader

TODO(Mark): Add paper reference.

Winobias is a dataset to analyse the issue of gender bias in co-reference resolution. It contains simple sentences with pro/anti stereotypical gender associations with which to measure the bias of a coreference system trained on another corpus. It is effectively a toy dataset and as such, uses very simplistic language; it has little use outside of evaluating a model for bias.

The dataset is formatted with a single sentence per line, with a maximum of 2 non-nested coreference clusters annotated using either square or round brackets. For example:

[The salesperson] sold (some books) to the librarian because [she] was trying to sell (them).

Returns a list of Instances which have four fields: text, a TextField containing the full sentence text, spans, a ListField[SpanField] of inclusive start and end indices for span candidates, and metadata, a MetadataField that stores the instance’s original text. For data with gold cluster labels, we also include the original clusters (a list of list of index pairs) and a SequenceLabelField of cluster ids for every span candidate in the metadata also.

Parameters:
max_span_width: ``int``, required.

The maximum width of candidate spans to consider.

token_indexers : Dict[str, TokenIndexer], optional

This is used to index the words in the sentence. See TokenIndexer. Default is {"tokens": SingleIdTokenIndexer()}.

classmethod from_params(params: allennlp.common.params.Params) → allennlp.data.dataset_readers.coreference_resolution.winobias.WinobiasReader[source]
text_to_instance(sentence: typing.List[allennlp.data.tokenizers.token.Token], gold_clusters: typing.Union[typing.List[typing.List[typing.Tuple[int, int]]], NoneType] = None) → allennlp.data.instance.Instance[source]
Parameters:
sentences : List[Token], required.

The already tokenised sentence to analyse.

gold_clusters : Optional[List[List[Tuple[int, int]]]], optional (default = None)

A list of all clusters in the sentence, represented as word spans. Each cluster contains some number of spans, which can be nested and overlap, but will never exactly match between clusters.

Returns:
An ``Instance`` containing the following ``Fields``:
text : TextField

The text of the full sentence.

spans : ListField[SpanField]

A ListField containing the spans represented as SpanFields with respect to the sentence text.

span_labels : SequenceLabelField, optional
The id of the cluster which each possible span belongs to, or -1 if it does

not belong to a cluster. As these labels have variable length (it depends on how many spans we are considering), we represent this a as a SequenceLabelField with respect to the spans ``ListField.