Rebuild from the original release with character offsets

README.md

@@ -1,24 +1,75 @@
 ---
-features:
+pretty_name: GAP
+license: apache-2.0
+language:
+- en
+dataset_info:
+- config_name: default
+  features:
   - name: input
     dtype: string
   - name: output
     dtype: json
   - name: schema
-    list:
-      - name: label
-        dtype: string
-      - name: description
-        dtype: string
-      - name: task
-        dtype: string
+    dtype: json
 configs:
-  - config_name: "default"
-    data_files:
-      - split: "train"
-        path: "data/gap_train.jsonl"
-      - split: "validation"
-        path: "data/gap_validation.jsonl"
-      - split: "test"
-        path: "data/gap_test.jsonl"
----
+- config_name: default
+  data_files:
+  - split: train
+    path: data/train.jsonl
+  - split: validation
+    path: data/validation.jsonl
+  - split: test
+    path: data/test.jsonl
+---
+
+# GAP
+
+GAP (relation extraction with given mentions (pronoun resolution)), rebuilt for MNEB from the original release with character-level offsets.
+
+**Source:** https://github.com/google-research-datasets/gap-coreference (commit 83135f26)
+
+| config | split | records |
+| --- | --- | --- |
+| default | train | 2001 |
+| default | validation | 454 |
+| default | test | 2002 |
+
+## Labels
+
+**entities:** `pronoun`, `name`
+
+**relations:** `refers to`
+
+## Format
+
+Each record has `input` (the text), `output` (gold annotations grouped by label) and `schema` (the label set with descriptions). Offsets are character offsets into `input`, half-open `[start, end)`.
+
+```
+"output": {
+  "entities":  {"<label>": [{"text", "start", "end"}]},
+  "relations": {"<label>": [{"head": {"text", "start", "end"},
+                             "tail": {"text", "start", "end"}}]},
+  "events":    {"<label>": [{"trigger": {"text", "start", "end"},
+                             "arguments": [{"role", "text", "start", "end"}]}]}
+}
+```
+
+Each schema item has a `description` and a `description_source`: `official` when adapted from the dataset's paper or guideline, `mneb` when written for MNEB. Relation datasets also give `setting` (`end_to_end` or `gold_entity`) and `negative_label`; relation items give `symmetric`. Event items list the `roles` valid for that event type.
+
+Within a record, identical annotations are kept once and outer whitespace is trimmed from spans; source annotations are otherwise kept as annotated.
+
+## Notes
+
+- Given mentions: the pronoun and the two candidate names; relation `refers to` from the pronoun to each coreferent name.
+- train = gap-development, validation = gap-validation, test = gap-test.
+
+## Build
+
+Built by `scripts/datasets/gap.py` in the MNEB repository.
+
+## Citation
+
+Webster et al. (2018), Mind the GAP: A Balanced Corpus of Gendered Ambiguous Pronouns. TACL.
+
+License: apache-2.0