Air Travel Information Service (ATIS) is a benchmark corpora/dataset in spoken language understanding field. Many papers make use of it to see how their algorithms perform.

There are many tagging methods defined on this corpora, and I hoped to have the corpora with one of those methods tagged but when I accessed it, I saw only the sentences (transcribed) and the output result of the system (the SQL query which generates the answer of the system to the user).

I thought there might be some codes to preprocess the corpora (or the processed data) available on the net, spend two days googling for it but nothing exists.

I am going to write a code to interpret the SQL query and generates the semantic frames representation but before that, I hope someone knows anything helpful and shares with me.

1 Answer 1


ATIS semantic frame annotation was done in Cambridge University under supervision of Prof. Steve Young. There are two types of annotations: Abstract and word level.

Abstract annotation is used to train Hidden Vector State (HVS) model introduced by Yulan He and is in the form of:

Show me flights from Boston to Washington : SS(FLIGHT(FROMLOC(CITY_NAME) TOLOC(CITY_NAME))

The annotation is parsed to a hierarchical tree explained in Yulan He's work.

Another annotation is the word-level one which is used in other parsers like TBED Parser is in the form of:

Show me flights from Boston to Washington <=> flight(fromloc.city_name=Boston, toloc.city_name=Washington)

flight is the frame name and shows the goal of the utterance and the slots (fromloc.city_name,...) shows the details of the goal.

There is no code available to convert SQL queries to this formats but, it can be easily induced from the queries (parsing the SELECT and WHERE clauses)

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.