parkervg/destt5-schema-prediction

Fine-tuned weights for the schema prediction model described in Correcting Semantic Parses with Natural Language through Dynamic Schema Encoding, based on t5-large.

Training Data

The model has been fine-tuned on the 7,481 training examples in the SPLASH interactive semantic parsing dataset.

Training Objective

This model was initialized with t5-large and fine-tuned with the text-to-text generation objective.

As this model works in the interactive setting, we utilize the standard text2sql features such as question and db_schema, in addition to feedback and incorrect_parse.

[question] || [incorrect_parse] || [db_id] | [table] : [column] ( [content] , [content] ) , [column] ( ... ) , [...] | [table] : ... | ... || [feedback]

The model then attempts to predict those schema items that appear in the final gold SQL query, prefaced by the db_id.

[db_id] | [table] : [column] ( [content] , [content] ) , [column] ( ... ) , [...] | [table] : ...

Performance

This model achieves 88.98% F1 score in identifying schema items on the SPLASH test set.

When combined with the destt5-text2sql model, it achieves 53.43% correction accuracy (exact-match) on the SPLASH test set.

References

Citation

@inproceedings{glenn2023correcting,
  author = {Parker Glenn, Parag Pravin Dakle, Preethi Raghavan},
  title = "Correcting Semantic Parses with Natural Language through Dynamic Schema Encoding",
  booktitle = "Proceedings of the 5th Workshop on NLP for Conversational AI",
  publisher = "Association for Computational Linguistics",
  year = "2023"
}