v1.0

README.md

@@ -0,0 +1,146 @@
+---
+annotations_creators:
+- no-annotation
+language_creators:
+- found
+languages:
+- en,de
+licenses:
+- MIT
+multilinguality:
+- 2 languages
+size_categories:
+- 100K<n<1M
+source_datasets:
+- original
+task_categories:
+- syntactic-evaluation
+task_ids:
+- syntactic-transformations
+---
+
+# Dataset Card for syntactic_transformations
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-instances)
+  - [Data Splits](#data-instances)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+
+## Dataset Description
+
+- **Homepage:** [Needs More Information]
+- **Repository:** https://github.com/sebschu/multilingual-transformations
+- **Paper:** [Coloring the Blank Slate: Pre-training Imparts a Hierarchical Inductive Bias to Sequence-to-sequence Models](https://aclanthology.org/2022.findings-acl.106/)
+- **Leaderboard:** [Needs More Information]
+- **Point of Contact:** [Aaron Mueller](mailto:[email protected])
+
+### Dataset Summary
+
+This contains the the syntactic transformations datasets used in [Coloring the Blank Slate: Pre-training Imparts a Hierarchical Inductive Bias to Sequence-to-sequence Models](https://aclanthology.org/2022.findings-acl.106/). It consists of English and German question formation and passivization transformations. This dataset also contains zero-shot cross-lingual transfer training and evaluation data.
+
+### Supported Tasks and Leaderboards
+
+[Needs More Information]
+
+### Languages
+
+English and German.
+
+## Dataset Structure
+
+### Data Instances
+
+A typical data point consists of a source sequence ("src"), a target sequence ("tgt"), and a task prefix ("prefix"). The prefix indicates whether a given sequence should be kept the same in the target (indicated by the "decl:" prefix) or transformed into a question/passive ("quest:"/"passiv:", respectively). An example follows:
+
+{"src": "the yak has entertained the walruses that have amused the newt.",
+"tgt": "has the yak entertained the walruses that have amused the newt?",
+"prefix": "quest: "
+}
+
+### Data Fields
+
+- src: the original source sequence.
+- tgt: the transformed target sequence.
+- prefix: indicates which transformation to perform to map from the source to target sequences.
+
+### Data Splits
+
+The datasets are split into training, dev, test, and gen ("generalization") sets. The training sets are for fine-tuning the model. The dev and test sets are for evaluating model abilities on in-domain transformations. The generalization sets are for evaluating the inductive biases of the model.
+
+NOTE: for the zero-shot cross-lingual transfer datasets, the generalization sets are split into in-domain and out-of-domain syntactic structures. For in-domain transformations, use "gen_rc_o" for question formation or "gen_pp_o" for passivization. For out-of-domain transformations, use "gen_rc_s" for question formation or "gen_pp_s" for passivization.
+
+## Dataset Creation
+
+### Curation Rationale
+
+[Needs More Information]
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+[Needs More Information]
+
+#### Who are the source language producers?
+
+[Needs More Information]
+
+### Annotations
+
+#### Annotation process
+
+[Needs More Information]
+
+#### Who are the annotators?
+
+[Needs More Information]
+
+### Personal and Sensitive Information
+
+[Needs More Information]
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[Needs More Information]
+
+### Discussion of Biases
+
+[Needs More Information]
+
+### Other Known Limitations
+
+[Needs More Information]
+
+## Additional Information
+
+### Dataset Curators
+
+[Needs More Information]
+
+### Licensing Information
+
+[Needs More Information]
+
+### Citation Information
+
+[Needs More Information]

syntactic_transformations.py

@@ -0,0 +1,203 @@
+import dataclasses
+import datasets
+import json
+
+logger = datasets.logging.get_logger(__name__)
+
+_CITATION = """\
+@inproceedings{mueller-etal-2022-coloring,
+    title = "Coloring the Blank Slate: Pre-training Imparts a Hierarchical Inductive Bias to Sequence-to-sequence Models",
+    author = "Mueller, Aaron  and
+      Frank, Robert  and
+      Linzen, Tal  and
+      Wang, Luheng  and
+      Schuster, Sebastian",
+    booktitle = "Findings of the Association for Computational Linguistics: ACL 2022",
+    month = may,
+    year = "2022",
+    address = "Dublin, Ireland",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/2022.findings-acl.106",
+    doi = "10.18653/v1/2022.findings-acl.106",
+    pages = "1352--1368",
+}
+"""
+
+_DESCRIPTION = """\
+This is the dataset used for Coloring the Blank Slate: 
+Pre-training Imparts a Hierarchical Inductive Bias to
+Sequence-to-sequence Models.
+"""
+
+class SyntacticTransformationsConfig(datasets.BuilderConfig):
+    def __init__(self, description, features, data_url, citation, url, **kwargs):
+        super(SyntacticTransformationsConfig, self).__init__(version=datasets.Version("1.18.3"), **kwargs)
+        self.description = description
+        self.text_features = features
+        self.citation = citation
+        self.data_url = data_url
+        self.url = url
+    
+
+
+class SyntacticTransformations(datasets.GeneratorBasedBuilder):
+    standard_features = datasets.Features(
+        {
+            "src": datasets.Value("string"),
+            "tgt": datasets.Value("string"),
+            "prefix": datasets.Value("string")
+        }
+    )
+    
+    BUILDER_CONFIGS = [
+        SyntacticTransformationsConfig(
+            name="passiv-en-nps",
+            description="English passivization transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/passiv_en_nps/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        ),
+        SyntacticTransformationsConfig(
+            name="passiv-de-nps",
+            description="German passivization transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/passiv_de_nps/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        ),
+        SyntacticTransformationsConfig(
+            name="question-en",
+            description="English question formation transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/question_have-havent_en/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        ),
+        SyntacticTransformationsConfig(
+            name="question-de",
+            description="German question formation transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/question_have-can_withquest_de/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        ),
+        SyntacticTransformationsConfig(
+            name="passiv-en_de-nps",
+            description="Zero-shot English-to-German passivization transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/passiv_en-de_nps/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        ),
+        SyntacticTransformationsConfig(
+            name="question-en_de",
+            description="Zero-shot English-to-German question formation transformations.",
+            features=standard_features,
+            data_url="https://raw.githubusercontent.com/sebschu/multilingual-transformations/main/data/question_have-can_de/",
+            url="https://github.com/sebschu/multilingual-transformations/",
+            citation=_CITATION
+        )
+    ]
+    
+    def _split_generators(self, dl_manager):
+        if self.config.name == "passiv-en-nps":
+            template = "passiv_en_nps.{}.json"
+            _URLS = {
+                "train": self.config.data_url + template.format("train"),
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen": self.config.data_url + template.format("gen"),
+            }
+        elif self.config.name == "passiv-de-nps":
+            template = "passiv_de_nps.{}.json"
+            _URLS = {
+                "train": self.config.data_url + template.format("train"),
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen": self.config.data_url + template.format("gen"),
+            }
+        elif self.config.name == "question-en":
+            template = "question_have.{}.json"
+            _URLS = {
+                "train": self.config.data_url + template.format("train"),
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen": self.config.data_url + template.format("gen"),
+            }
+        elif self.config.name == "question-de":
+            template = "question_have_can.de.{}.json"
+            _URLS = {
+                "train": self.config.data_url + template.format("train"),
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen": self.config.data_url + template.format("gen"),
+            }
+        elif self.config.name == "question-en_de":
+            template = "question_have_can.de.{}.json"
+            _URLS = {
+                "train": self.config.data_url + "question_have_can.en-de.train.json",
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen_rc_s": self.config.data_url + template.format("gen_rc_s"),
+                "gen_rc_o": self.config.data_url + template.format("gen_rc_o"),
+            }
+        elif self.config.name == "passiv-en_de-nps":
+            template = "passiv_de_nps.{}.json"
+            _URLS = {
+                "train": self.config.data_url + "passiv_en-de_nps.train.json",
+                "dev": self.config.data_url + template.format("dev"),
+                "test": self.config.data_url + template.format("test"),
+                "gen_pp_s": self.config.data_url + template.format("gen_pp_s"),
+                "gen_pp_o": self.config.data_url + template.format("gen_pp_o"),
+            }
+
+        
+        data_files = dl_manager.download(_URLS)
+
+        if "en_de" not in self.config.name:
+            return [
+                datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_files["train"]}),
+                datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": data_files["dev"]}),
+                datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": data_files["test"]}),
+                datasets.SplitGenerator(name=datasets.NamedSplit('generalization'), gen_kwargs={"filepath": data_files["gen"]}),
+            ]
+        else:
+            gen_s = "gen_pp_s" if "passiv" in self.config.name else "gen_rc_s"
+            gen_o = "gen_pp_o" if "passiv" in self.config.name else "gen_rc_o"
+            return [
+                datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": data_files["train"]}),
+                datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": data_files["dev"]}),
+                datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": data_files["test"]}),
+                datasets.SplitGenerator(name=datasets.NamedSplit('generalization_s'), gen_kwargs={"filepath": data_files[gen_s]}),
+                datasets.SplitGenerator(name=datasets.NamedSplit('generalization_o'), gen_kwargs={"filepath": data_files[gen_o]}),
+            ]
+
+    def _info(self):
+        features = {text_feature: datasets.Value("string") for text_feature in self.config.text_features.keys()}
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(features),
+            homepage=self.config.url,
+            citation=_CITATION,
+        )
+    
+    def _generate_examples(self, filepath):
+        """This function returns the examples in the raw (text) form."""
+        logger.info("generating examples from = %s", filepath)
+        with open(filepath) as f:
+            id_ = 0
+            for line in f:
+                example = json.loads(line)
+                src = example["translation"]["src"]
+                tgt = example["translation"]["tgt"]
+                prefix = example["translation"]["prefix"]
+
+                # Features currently used are "context", "question", and "answers".
+                # Others are extracted here for the ease of future expansions.
+                yield id_, {
+                    "src": src,
+                    "tgt": tgt,
+                    "prefix": prefix,
+                }
+                id_ += 1