This folder contains the SpiderDec, the decomposed version of the Spider dev data set as the dev-decomposed.json.
It contains the following files:
- dev-decomposed.json
# Training Examples: 1034
# Databases: 20
- README.txt
For each example in the Spider dev set, we assign a JSON object including database_id, interaction, and final.
Within the interaction list, sub-questions are saved. The Final object is the question we apply decomposition on.
Here an example is shown.
Note that we name the execution result of each question as a table, tb1 is the execution result of sub-question 1, tb2 is the execution result of sub-question 2, and so on.
{
"database_id": "tvshow",
"interaction": [
{
"utterance": "what are the tv channels playing any cartoons directed by Ben Jones?",
"query": "SELECT channel FROM cartoon WHERE directed_by = 'Ben Jones'"
},
{
"utterance": "What are the package options of all tv channels that are not in tb1?",
"query": "SELECT package_option FROM TV_Channel WHERE id EXCEPT (SELECT channel FROM cartoon)"
}
],
"final": {
"query": "SELECT package_option FROM TV_Channel WHERE id NOT IN (SELECT channel FROM cartoon WHERE directed_by = 'Ben Jones')",
"utterance": "What are the package options of all tv channels that are not playing any cartoons directed by Ben Jones?"
}
}