metadata

license: cc-by-2.0
task_categories:
  - text-classification
language:
  - en
size_categories:
  - 100K<n<1M
dataset_info:
  features:
    - name: seq
      dtype: string
    - name: label
      dtype: string
    - name: Adj_Class
      dtype: string
    - name: Adj
      dtype: string
    - name: Nn
      dtype: string
    - name: Hy
      dtype: string
  splits:
    - name: train
      num_bytes: 26047744
      num_examples: 300132
    - name: test
      num_bytes: 874524
      num_examples: 10080
  download_size: 4721262
  dataset_size: 26922268

Preprocessed from https://huggingface.co/datasets/lorenzoscottb/PLANE-ood/

df=pd.read_json('https://huggingface.co/datasets/lorenzoscottb/PLANE-ood/resolve/main/PLANE_trntst-OoV_inftype-all.json')
f = lambda df: pd.DataFrame(list(zip(*[df[c] for c in df.index])),columns=df.index)
ds=DatasetDict()
for split in ['train','test']:
    dfs=pd.concat([f(df[c]) for c in df.columns if split in c.lower()]).reset_index(drop=True)
    dfs['label']=dfs['label'].map(lambda x:{1:'entailment',0:'not-entailment'}[x])
    ds[split]=Dataset.from_pandas(dfs,preserve_index=False)
ds.push_to_hub('tasksource/PLANE-ood')

PLANE Out-of-Distribution Sets

PLANE (phrase-level adjective-noun entailment) is a benchmark to test models on fine-grained compositional inference. The current dataset contains five sampled splits, used in the supervised experiments of Bertolini et al., 22.

Features

Each entrance has 6 features: seq, label, Adj_Class, Adj, Nn, Hy

seq:test sequense
label: ground truth (1:entialment, 0:no-entailment)
Adj_Class: the class of the sequence adjectives
Adj: the adjective of the sequence (I: intersective, S: subsective, O: intensional)
Nn: the noun
Hy: the noun's hypericum

Each sample in seq can take one of three forms (or inference types, in paper):

An Adjective-Noun is a Noun (e.g. A red car is a car)
An Adjective-Noun is a Hypernym(Noun) (e.g. A red car is a vehicle)
An Adjective-Noun is a Adjective-Hypernym(Noun) (e.g. A red car is a red vehicle)

Please note that, as specified in the paper, the ground truth is automatically assigned based on the linguistic rule that governs the interaction between each adjective class and inference type – see the paper for more detail.

Cite

If you use PLANE for your work, please cite the main COLING 2022 paper.

@inproceedings{bertolini-etal-2022-testing,
    title = "Testing Large Language Models on Compositionality and Inference with Phrase-Level Adjective-Noun Entailment",
    author = "Bertolini, Lorenzo  and
      Weeds, Julie  and
      Weir, David",
    booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "International Committee on Computational Linguistics",
    url = "https://aclanthology.org/2022.coling-1.359",
    pages = "4084--4100",
}