--- language: - en license: mit task_categories: - question-answering pretty_name: bAbI divided by task description: "These tasks measure understanding in several ways: whether a system is able to answer questions via chaining facts, simple induction, deduction and many more. The tasks are designed to be prerequisites for any system that aims to be capable of conversing with a human." configs: - config_name: task_1-single_supporting_fact data_files: - split: train path: "data/task_1-single_supporting_fact/babi-train.parquet" - split: test path: "data/task_1-single_supporting_fact/babi-test.parquet" - split: validation path: "data/task_1-single_supporting_fact/babi-validation.parquet" - config_name: task_2-two_supporting_facts data_files: - split: train path: "data/task_2-two_supporting_facts/babi-train.parquet" - split: test path: "data/task_2-two_supporting_facts/babi-test.parquet" - split: validation path: "data/task_2-two_supporting_facts/babi-validation.parquet" - config_name: task_3-three_supporting_facts data_files: - split: train path: "data/task_3-three_supporting_facts/babi-train.parquet" - split: test path: "data/task_3-three_supporting_facts/babi-test.parquet" - split: validation path: "data/task_3-three_supporting_facts/babi-validation.parquet" - config_name: task_4-two_argument_relations data_files: - split: train path: "data/task_4-two_argument_relations/babi-train.parquet" - split: test path: "data/task_4-two_argument_relations/babi-test.parquet" - split: validation path: "data/task_4-two_argument_relations/babi-validation.parquet" - config_name: task_5-three_argument_relations data_files: - split: train path: "data/task_5-three_argument_relations/babi-train.parquet" - split: test path: "data/task_5-three_argument_relations/babi-test.parquet" - split: validation path: "data/task_5-three_argument_relations/babi-validation.parquet" - config_name: task_6-yes_no_questions data_files: - split: train path: "data/task_6-yes_no_questions/babi-train.parquet" - split: test path: "data/task_6-yes_no_questions/babi-test.parquet" - split: validation path: "data/task_6-yes_no_questions/babi-validation.parquet" - config_name: task_7-counting data_files: - split: train path: "data/task_7-counting/babi-train.parquet" - split: test path: "data/task_7-counting/babi-test.parquet" - split: validation path: "data/task_7-counting/babi-validation.parquet" - config_name: task_8-lists_sets data_files: - split: train path: "data/task_8-lists_sets/babi-train.parquet" - split: test path: "data/task_8-lists_sets/babi-test.parquet" - split: validation path: "data/task_8-lists_sets/babi-validation.parquet" - config_name: task_9-simple_negation data_files: - split: train path: "data/task_9-simple_negation/babi-train.parquet" - split: test path: "data/task_9-simple_negation/babi-test.parquet" - split: validation path: "data/task_9-simple_negation/babi-validation.parquet" - config_name: task_10-indefinite_knowledge data_files: - split: train path: "data/task_10-indefinite_knowledge/babi-train.parquet" - split: test path: "data/task_10-indefinite_knowledge/babi-test.parquet" - split: validation path: "data/task_10-indefinite_knowledge/babi-validation.parquet" - config_name: task_11-basic_coreference data_files: - split: train path: "data/task_11-basic_coreference/babi-train.parquet" - split: test path: "data/task_11-basic_coreference/babi-test.parquet" - split: validation path: "data/task_11-basic_coreference/babi-validation.parquet" - config_name: task_12-conjunction data_files: - split: train path: "data/task_12-conjunction/babi-train.parquet" - split: test path: "data/task_12-conjunction/babi-test.parquet" - split: validation path: "data/task_12-conjunction/babi-validation.parquet" - config_name: task_13-compound_coreference data_files: - split: train path: "data/task_13-compound_coreference/babi-train.parquet" - split: test path: "data/task_13-compound_coreference/babi-test.parquet" - split: validation path: "data/task_13-compound_coreference/babi-validation.parquet" - config_name: task_14-time_reasoning data_files: - split: train path: "data/task_14-time_reasoning/babi-train.parquet" - split: test path: "data/task_14-time_reasoning/babi-test.parquet" - split: validation path: "data/task_14-time_reasoning/babi-validation.parquet" - config_name: task_15-basic_deduction data_files: - split: train path: "data/task_15-basic_deduction/babi-train.parquet" - split: test path: "data/task_15-basic_deduction/babi-test.parquet" - split: validation path: "data/task_15-basic_deduction/babi-validation.parquet" - config_name: task_16-basic_induction data_files: - split: train path: "data/task_16-basic_induction/babi-train.parquet" - split: test path: "data/task_16-basic_induction/babi-test.parquet" - split: validation path: "data/task_16-basic_induction/babi-validation.parquet" - config_name: task_17-positional_reasoning data_files: - split: train path: "data/task_17-positional_reasoning/babi-train.parquet" - split: test path: "data/task_17-positional_reasoning/babi-test.parquet" - split: validation path: "data/task_17-positional_reasoning/babi-validation.parquet" - config_name: task_18-size_reasoning data_files: - split: train path: "data/task_18-size_reasoning/babi-train.parquet" - split: test path: "data/task_18-size_reasoning/babi-test.parquet" - split: validation path: "data/task_18-size_reasoning/babi-validation.parquet" - config_name: task_19-path_finding data_files: - split: train path: "data/task_19-path_finding/babi-train.parquet" - split: test path: "data/task_19-path_finding/babi-test.parquet" - split: validation path: "data/task_19-path_finding/babi-validation.parquet" - config_name: task_20-agents_motivations data_files: - split: train path: "data/task_20-agents_motivations/babi-train.parquet" - split: test path: "data/task_20-agents_motivations/babi-test.parquet" - split: validation path: "data/task_20-agents_motivations/babi-validation.parquet" --- # bAbI Dataset : Per-Task Splits The examples in this dataset were obtained from "Muennighoff/babi". This dataset offer the same examples but it can be loaded using a per-task basis: ```python from datasets import load_dataset ds = load_dataset('RawthiL/babi_tasks', 'task_1-single_supporting_fact') ``` The available tasks are: | Task ID | Task Name | Split Name| |---------|-----------|----| | 1 | Single Supporting Fact | task_1-single_supporting_fact | | 2 | Two Supporting Facts | task_2-two_supporting_facts | | 3 | Three Supporting Facts | task_3-three_supporting_facts | | 4 | Two Argument Relations | task_4-two_argument_relations | | 5 | Three Argument Relations | task_5-three_argument_relations | | 6 | Yes/No Questions | task_6-yes_no_questions | | 7 | Counting | task_7-counting | | 8 | Lists/Sets | task_8-lists_sets | | 9 | Simple Negation | task_9-simple_negation | | 10 | Indefinite Knowledge | task_10-indefinite_knowledge | | 11 | Basic Coreference | task_11-basic_coreference | | 12 | Conjunction | task_12-conjunction | | 13 | Compound Coreference | task_13-compound_coreference | | 14 | Time Reasoning | task_14-time_reasoning | | 15 | Basic Deduction | task_15-basic_deduction | | 16 | Basic Induction | task_16-basic_induction | | 17 | Positional Reasoning | task_17-positional_reasoning | | 18 | Size Reasoning | task_18-size_reasoning | | 19 | Path Finding | task_19-path_finding | | 20 | Agent's Motivations | task_20-agents_motivations | ### TOWARDS AI-COMPLETE QUESTION ANSWERING: A SET OF PREREQUISITE TOY TASKS @article{weston2015towards, title={Towards ai-complete question answering: A set of prerequisite toy tasks}, author={Weston, Jason and Bordes, Antoine and Chopra, Sumit and Rush, Alexander M and Van Merri{"e}nboer, Bart and Joulin, Armand and Mikolov, Tomas}, journal={arXiv preprint arXiv:1502.05698}, year={2015} }