Hugging Face's logo

Datasets:
Orange
/
csqa-sparqltotext

Tasks:
Question Answering
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
100K - 1M
Tags:
qa
knowledge-graph
sparql
multi-hop
Libraries:
Datasets
Dask
License:
Dataset card Viewer Files Community
1
csqa-sparqltotext / README.md
glecorve's picture
glecorve
Enriched README
8c8a754
preview code
|
raw
history blame contribute delete
14.2 kB
metadata
license: cc-by-sa-4.0
dataset_info:
  features:
    - name: id
      dtype: string
    - name: turns
      list:
        - name: id
          dtype: int64
        - name: ques_type_id
          dtype: int64
        - name: question-type
          dtype: string
        - name: description
          dtype: string
        - name: entities_in_utterance
          list: string
        - name: relations
          list: string
        - name: type_list
          list: string
        - name: speaker
          dtype: string
        - name: utterance
          dtype: string
        - name: all_entities
          list: string
        - name: active_set
          list: string
        - name: sec_ques_sub_type
          dtype: int64
        - name: sec_ques_type
          dtype: int64
        - name: set_op_choice
          dtype: int64
        - name: is_inc
          dtype: int64
        - name: count_ques_sub_type
          dtype: int64
        - name: count_ques_type
          dtype: int64
        - name: is_incomplete
          dtype: int64
        - name: inc_ques_type
          dtype: int64
        - name: set_op
          dtype: int64
        - name: bool_ques_type
          dtype: int64
        - name: entities
          list: string
        - name: clarification_step
          dtype: int64
        - name: gold_actions
          list:
            list: string
        - name: is_spurious
          dtype: bool
        - name: masked_verbalized_answer
          dtype: string
        - name: parsed_active_set
          list: string
        - name: sparql_query
          dtype: string
        - name: verbalized_all_entities
          list: string
        - name: verbalized_answer
          dtype: string
        - name: verbalized_entities_in_utterance
          list: string
        - name: verbalized_gold_actions
          list:
            list: string
        - name: verbalized_parsed_active_set
          list: string
        - name: verbalized_sparql_query
          dtype: string
        - name: verbalized_triple
          dtype: string
        - name: verbalized_type_list
          list: string
  splits:
    - name: train
      num_bytes: 6815016095
      num_examples: 152391
    - name: test
      num_bytes: 1007873839
      num_examples: 27797
    - name: validation
      num_bytes: 692344634
      num_examples: 16813
  download_size: 2406342185
  dataset_size: 8515234568
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
task_categories:
  - conversational
  - question-answering
tags:
  - qa
  - knowledge-graph
  - sparql
  - multi-hop
language:
  - en

Dataset Card for CSQA-SPARQLtoText

Table of Contents

Dataset Description

Dataset Summary

CSQA corpus (Complex Sequential Question-Answering, see https://amritasaha1812.github.io/CSQA/) is a large corpus for conversational knowledge-based question answering. The version here is augmented with various fields to make it easier to run specific tasks, especially SPARQL-to-text conversion.

The original data has been post-processing as follows:

  1. Verbalization templates were applied on the answers and their entities were verbalized (replaced by their label in Wikidata)

  2. Questions were parsed using the CARTON algorithm to produce a sequence of action in a specific grammar

  3. Sequence of actions were mapped to SPARQL queries and entities were verbalized (replaced by their label in Wikidata)

Supported tasks

  • Knowledge-based question-answering
  • Text-to-SPARQL conversion

Knowledge based question-answering

Below is an example of dialogue:

  • Q1: Which occupation is the profession of Edmond Yernaux ?
  • A1: politician
  • Q2: Which collectable has that occupation as its principal topic ?
  • A2: Notitia Parliamentaria, An History of the Counties, etc.

SPARQL queries and natural language questions 

SELECT DISTINCT ?x WHERE
{ ?x rdf:type ontology:occupation . resource:Edmond_Yernaux property:occupation ?x }

is equivalent to:

Which occupation is the profession of Edmond Yernaux ?

Languages

  • English

Dataset Structure

The corpus follows the global architecture from the original version of CSQA (https://amritasaha1812.github.io/CSQA/).

There is one directory of the train, dev, and test sets, respectively.

Dialogues are stored in separate directories, 100 dialogues per directory.

Finally, each dialogue is stored in a JSON file as a list of turns.

Types of questions

Comparison of question types compared to related datasets:

SimpleQuestions ParaQA LC-QuAD 2.0 CSQA WebNLQ-QA
Number of triplets in query 1 ✓ ✓ ✓ ✓ ✓
2 ✓ ✓ ✓ ✓
More ✓ ✓ ✓
Logical connector between triplets Conjunction ✓ ✓ ✓ ✓ ✓
Disjunction ✓ ✓
Exclusion ✓ ✓
Topology of the query graph Direct ✓ ✓ ✓ ✓ ✓
Sibling ✓ ✓ ✓ ✓
Chain ✓ ✓ ✓ ✓
Mixed ✓ ✓
Other ✓ ✓ ✓ ✓
Variable typing in the query None ✓ ✓ ✓ ✓ ✓
Target variable ✓ ✓ ✓ ✓
Internal variable ✓ ✓ ✓ ✓
Comparisons clauses None ✓ ✓ ✓ ✓ ✓
String ✓ ✓
Number ✓ ✓ ✓
Date ✓ ✓
Superlative clauses No ✓ ✓ ✓ ✓ ✓
Yes ✓
Answer type Entity (open) ✓ ✓ ✓ ✓ ✓
Entity (closed) ✓ ✓
Number ✓ ✓ ✓
Boolean ✓ ✓ ✓ ✓
Answer cardinality 0 (unanswerable) ✓ ✓
1 ✓ ✓ ✓ ✓ ✓
More ✓ ✓ ✓ ✓
Number of target variables 0 (⇒ ASK verb) ✓ ✓ ✓ ✓
1 ✓ ✓ ✓ ✓ ✓
2 ✓ ✓
Dialogue context Self-sufficient ✓ ✓ ✓ ✓ ✓
Coreference ✓ ✓
Ellipsis ✓ ✓
Meaning Meaningful ✓ ✓ ✓ ✓ ✓
Non-sense ✓

Data splits

Text verbalization is only available for a subset of the test set, referred to as challenge set. Other sample only contain dialogues in the form of follow-up sparql queries.

Train Validation Test
Questions 1.5M 167K 260K
Dialogues 152K 17K 28K
NL question per query 1
Characters per query 163 (± 100)
Tokens per question 10 (± 4)

JSON fields

Each turn of a dialogue contains the following fields:

Original fields

  • ques_type_id: ID corresponding to the question utterance

  • description: Description of type of question

  • relations: ID's of predicates used in the utterance

  • entities_in_utterance: ID's of entities used in the question

  • speaker: The nature of speaker: SYSTEM or USER

  • utterance: The utterance: either the question, clarification or response

  • active_set: A regular expression which identifies the entity set of answer list

  • all_entities: List of ALL entities which constitute the answer of the question

  • question-type: Type of question (broad types used for evaluation as given in the original authors' paper)

  • type_list: List containing entity IDs of all entity parents used in the question

New fields

  • is_spurious: introduced by CARTON,

  • is_incomplete: either the question is self-sufficient (complete) or it relies on information given by the previous turns (incomplete)

  • parsed_active_set:

  • gold_actions: sequence of ACTIONs as returned by CARTON

  • sparql_query: SPARQL query

Verbalized fields

Fields with verbalized in their name are verbalized versions of another fields, ie IDs were replaced by actual words/labels.

Format of the SPARQL queries

  • Clauses are in random order

  • Variables names are represented as random letters. The letters change from one turn to another.

  • Delimiters are spaced

Additional Information

Licensing Information

  • Content from original dataset: CC-BY-SA 4.0

  • New content: CC BY-SA 4.0

Citation Information

This version of the corpus (with SPARQL queries) 

@inproceedings{lecorve2022sparql2text,
  title={SPARQL-to-Text Question Generation for Knowledge-Based Conversational Applications},
  author={Lecorv\'e, Gw\'enol\'e and Veyret, Morgan and Brabant, Quentin and Rojas-Barahona, Lina M.},
  journal={Proceedings of the Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the International Joint Conference on Natural Language Processing (AACL-IJCNLP)},
  year={2022}
}

Original corpus (CSQA) 

@InProceedings{saha2018complex,
    title = {Complex {Sequential} {Question} {Answering}: {Towards} {Learning} to {Converse} {Over} {Linked} {Question} {Answer} {Pairs} with a {Knowledge} {Graph}},
    volume = {32},
    issn = {2374-3468},
    url = {https://ojs.aaai.org/index.php/AAAI/article/view/11332},
    booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
    author = {Saha, Amrita and Pahuja, Vardaan and Khapra, Mitesh and Sankaranarayanan, Karthik and Chandar, Sarath},
    month = apr,
    year = {2018}
}

CARTON 

@InProceedings{plepi2021context,
    author="Plepi, Joan and Kacupaj, Endri and Singh, Kuldeep and Thakkar, Harsh and Lehmann, Jens",
    editor="Verborgh, Ruben and Hose, Katja and Paulheim, Heiko and Champin, Pierre-Antoine and Maleshkova, Maria and Corcho, Oscar and Ristoski, Petar and Alam, Mehwish",
    title="Context Transformer with Stacked Pointer Networks for Conversational Question Answering over Knowledge Graphs",
    booktitle="Proceedings of The Semantic Web",
    year="2021",
    publisher="Springer International Publishing",
    pages="356--371",
    isbn="978-3-030-77385-4"
}