Depends on a lot of factors of course, like how well documented is the schema and how large is the semantic gap between the surface text and the actual structure you want to extract. That is exactly what we want to measure here. But technically you can ask a model to generate any unseen schema by passing the JSON schema specification so at least from a syntactic perspective the problem is somewhat solved. Semantics is the big problem of course, and the fun one to tackle!
Depends on a lot of factors of course, like how well documented is the schema and how large is the semantic gap between the surface text and the actual structure you want to extract. That is exactly what we want to measure here. But technically you can ask a model to generate any unseen schema by passing the JSON schema specification so at least from a syntactic perspective the problem is somewhat solved. Semantics is the big problem of course, and the fun one to tackle!
Hey, great read. How zero-shot handles unseen schemas?
Depends on a lot of factors of course, like how well documented is the schema and how large is the semantic gap between the surface text and the actual structure you want to extract. That is exactly what we want to measure here. But technically you can ask a model to generate any unseen schema by passing the JSON schema specification so at least from a syntactic perspective the problem is somewhat solved. Semantics is the big problem of course, and the fun one to tackle!
Depends on a lot of factors of course, like how well documented is the schema and how large is the semantic gap between the surface text and the actual structure you want to extract. That is exactly what we want to measure here. But technically you can ask a model to generate any unseen schema by passing the JSON schema specification so at least from a syntactic perspective the problem is somewhat solved. Semantics is the big problem of course, and the fun one to tackle!