Extremely Large Grammer (Every street in the country) RRS feed

  • Question

  • I am going to be building an application that will need to understand every street name in the country, along with every city and state.  For example, someone might say that they are near post st and Market street. 

    I might be able to limit it to every city and state in the country, but that in itself is a huge grammer. 

    Does anyone have any suggestions as to the best way to tackle this issue?



    Tuesday, July 17, 2007 9:00 PM


  • I don't recommend creating a single grammar containing all of this information. It would be both very expensive to run and rather inaccurate.


    You generally need to ask the user to specify a location in 2 steps:


    1) city+state or zip code. You could further break this into state, followed by city, if this proves problematic.


    2) Create a grammar for each zip code containing all the streets. If a user specifies a city, load all of the zip code grammars corresponding to that city.


    You also need to restrict the format of what they can say as much as possible: a single street name should be OK, but if you start allowing them to say "intersection of X and Y" in one go, I think recognition performance will suffer.


    Hope that helps! 



    Thursday, July 19, 2007 4:33 PM