USER INTERFACE FOR DATA ACCESS AND
CROSS-REFERENCE TO RELATED
 This application claims priority from (i) U.S. Provisional Application No. 60/354,324, filed Feb. 7, 2002, and titled MOBILE APPLICATION ARCHITECTURE, and (ii)
U.S. application Ser. No. (Attorney Docket No.
13909-017001), filed Apr. 25, 2002, and titled MULTIMODAL SYNCHRONIZATION, both of which are hereby incorporated by reference in their entirety for all purposes.
 Certain implementations relate generally to a user interface and speech grammar, and more particularly to a user interface and speech grammar for voice-based data access and entry on a mobile device.
 A user interface may allow a user to gain access to data, such as, for example, products in a catalog database, or to enter data into a system, such as, for example, entering customer information into a customer database. User interfaces are used for applications residing on relatively stationary computing devices, such as desktop computers, as well as for applications residing on mobile computing devices, such as laptops, palmtops, and portable electronic organizers. A voice-activated user interface can be created to provide data access and entry to a system, and voice input may be particularly appealing for mobile devices.
 In various implementations, a grammar for speech recognition for a given voice-driven application, mobile or otherwise, can be written to enable accurate and efficient recognition. Particular implementations described below provide a user interface that allows a user to input data in one or more of a variety of different modes, including, for example, stylus and voice input. Output may also be in one or more of a variety of modes, such as, for example, display or voice. Particular implementations may be used with mobile devices, such as, for example, palmtops, and the combination of voice and stylus input with voice and display output may allow such mobile devices to be more useful to a user. Implementations may also be used with the multimodal synchronization system described in the incorporated provisional application.
 Implementations allow enhanced voice recognition accuracy and/or speed due in part to the use of a structured grammar that allows a grammar to be narrowed to a relevant part for a particular voice recognition operation. For example, narrowing of the grammar for a voice recognition operation on a full search string may be achieved by using the results of an earlier, or parallel, voice recognition operation on a component of the fall search string. Other implementations may narrow the grammar by accepting parameters of a search string in a particular order from a user, and, optionally, using the initial parameter(s) to narrow the grammar for subsequent parameters. Examples include (i) reversing the standard order of receiving street address information so that, for example, the country is received
before the state and the grammar used to recognize the state is narrowed to the states in the selected country, (ii) segmenting an electronic mail address or web site address so that a user supplies a domain identifier, such as, for example "com," separately, or (iii) automatically inserting the "at sign" and the "dot" into an electronic mail address and only prompting the user for the remaining terms, thus obviating the often complex process of recognizing these spoken characters.
 Implementations may also increase recognition accuracy and speed by augmenting a grammar with possible search strings, or utterances, thus decreasing the likelihood that a voice recognition system will need to identify an entry by its spelling. In such situations, the voice recognition system also obviates the need to ask the user to spell out a term that is not recognized when spoken. For example, after a user enters "com" as a domain identifier in an electronic mail address, the voice recognition system may include, for example, the names of all "Fortune 100" companies and a variety of popular commercial sites in the grammar for the server identifier of the electronic mail address. Thus, if the user then enters "amazon" as the server identifier, and if "amazon" has been included in the grammar, the system will recognize the entry without having to ask the user to spell it out.
 Implementations also allow enhanced database searching. This may be achieved, for example, by using a structured grammar and associating grammar entries with specific database entries. In this manner, when the structured grammar is used to recognize the search string, then particular database entries or relevant portions of the database may be identified at the same time.
 According to one general aspect, automatically searching for one or more matches to a search string includes accessing a first part and a second part of a search string. A first search space is searched for a match for the first part of the search string, and a second search space is limited based on a result of searching the first search space. The limited second search space is searched for a match for the second part of the search string.
 Searching the first search space may include searching a database. Searching the limited second search space may include searching at least part of the database. Limiting the second search space may include limiting the part of the database that is searched to database entries that include a match for the first part of the search string, thus allowing a quicker search compared to searching the full database. The second part of the search string may include a voice input or a manual input. Searching the first search space and searching the limited second search space may be performed at least partially in parallel. The search string may include an address.
 Accessing the first part of the search string may include accessing a voice input. Searching the first search space for the match may include performing voice recognition on the first part of the search string. Accessing at least the second part of the search string may include accessing the voice input. Limiting the second search space may include limiting the second search space to grammar entries associated with the first part of the search string. Searching the limited second search space may include performing voice recognition on at least the second part of the search